COHERENCE AND GROUNDING IN DISCOURSE

TYPOLOGICAL STUDIES IN L A N G U A G E (TSL)

A companion series to the journal "STUDIES IN L A N G U A G E "

Honorary Editor: Joseph H. Greenberg

General Editor: T. Givón

Editorial Board:
Alton Becker (Michigan) Paul Hopper (Binghamton)
Wallace Chafe (Santa Barbara) Margaret Langdon (San Diego)
Bernard Comrie (Los Angeles) Charles Li (Santa Barbara)
Scott DeLancey (Oregon) Johanna Nichols (Berkeley)
Gerard Diffloth (Chicago) Andrew Pawley (Auckland)
R.M.W.Dixon (Canberra) Frans Plank (Konstanz)
John Haiman (Winnipeg) Gillian Sankoff (Philadelphia)
Kenneth Hale (Cambridge, Mass.) Dan Slobin (Berkeley)
Bernd Heine (Köln) Sandra Thompson (Santa Barbara)

Volumes in this series will be functionally and typologically oriented, cove­

ring specific topics in language by collecting together data from a wide variety
of languages and language typologies. The orientation of the volumes will be
substantive rather than formal, with the aim of investigating universals of
human language via as broadly defined a data base as possible, leaning
toward cross-linguistic, diachronic, developmental and live-discourse data.
The series is, in spirit as well as in fact, a continuation of the tradition initiated
by C. Li (Word Order and Word Order Change, Subject and Topic,
Mechanisms for Syntactic Change) and continued by T. Givón (Discourse and
Syntax) and P. Hopper (Tense-Aspect: Between Semantics and Pragmatics).

Volume 11

Russell S. Tomlin (ed.)

COHERENCE AND GROUNDING IN DISCOURSE

COHERENCE AND GROUNDING
IN DISCOURSE

Outcome of a Symposium,
Eugene, Oregon, June 1984

edited by

RUSSELL S.TOMLIN
University of Oregon

JOHN BENJAMINS PUBLISHING COMPANY

Amsterdam/Philadelphia

1987
 TM
The paper used in this publication meets the minimum requirements of
8

the American National Standard for Information Sciences – Permanence

of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data

Tomlin, Russell S.
Coherence and grounding in discourse : outcome of a symposium, Eugene, Oregon, June
1984 / Russell S. Tomlin.
p. cm. (Typological Studies in Language, issn 0167-7373 ; v. 11)
Includes bibliographical references and index.
1. Discourse analysis -- Congresses. 2. Grammar, Comparative and general -- Syntax
-- Congresses. 3. Psycholinguistics -- Congresses. 4. Cohesion (Linguistics) --
Congresses. I. Tomlin, Russell S. II. Series.
PS302.G75 1987
401’.41 87000898
isbn 978 90 272 2882 6 (EUR) / 978 0 915027 85 9 (US) (Hb ; alk. paper)
isbn 978 90 272 2881 9 (EUR) / 978 0 915027 86 6 (US) (Pb ; alk. paper)
isbn 978 90 272 8627 7 (Eb)

© 1987 – John Benjamins B.V.

No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands
John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa
 CONTENTS

Preface vii
Narrative connectives in Sùpyìré 1
Robert Carlson
Cognitive constraints on information flow 21
Wallace Chafe
Transitivity in grammar and cognition 53
Scott DeLancey
Actions and procedural directions 69
Peter Dixon
Pragmatic functions of word order in Sesotho acquisition 91
Katherine Demuth
Psycholinguistic evidence for foregrounding and backgrounding 109
Mary S. Erbaugh
The grammatical marking of theme in oral Polish narrative 131
Vanessa Flashner
Anaphora in popular written English narratives 157
Barbara A. Fox
Beyond foreground and background 175
T. Givón
The use of pitch phenomena in the structuring of stories 189
Lorraine E. Kumpf
On the status of SVO sentences in French discourse 217
Knud Lambrecht
On the role of conditionals in Godié procedural discourse . . . . . . . . . 263
Lynell Marchese
Is basic word order universal? 281
Marianne Mithun
Encoding events in Kalam and English: different logics for reporting
experience . 329
Andrew Pawley
Word order in intransitive clauses in High and Low Malay of the late 19th
century 361
Ellen Rafferty
VI CONTENTS

The functional distribution of preposed and postposed 'if and 'when'

clauses in written discourse 383
Violeta Ramsey
"Vividness" and "narrative events" in Japanese conversational
narratives 409
Polly E. Szatrowski
"Subordination" and narrative event structure 435
Sandra A. Thompson
Linguistic reflections of cognitive events 455
Russell S. Tomlin
Cross-clause relations and temporal sequence in narrative and
beyond 481
Benji Wald
 PREFACE

In the spring of 1984 a symposium was held at the University of Oregon

on the relationships holding among discourse relations like foreground and
background information, linguistic or cognitive units like events, and linguis­
tic coding. The basic themes of this symposium derived from early research
by Grimes (1972), Longacre (1968, 1974, 1977), Hopper (1979, 1980), Jones
and Jones (1979), and others. Their work proposed that some of the informa­
tion in discourse is more central or significant than other, simply elaborative,
information. For narrative discourse, this discourse relation has been argued
to be related to the cognitive unit of event, and coded by numerous syntactic
devices, including tense-aspect, word order, subordination, transitivity, par­
ticipant coding and voice.
It was our feeling that this research, while an important beginning, was
also too simplistic, depending for its insight too much on introspection and
argument by example. The purpose of the symposium was to expand our
understanding of the relation holding between discourse relations, cognitive
units, and linguistic coding and to increase the empirical rigor of our efforts
to do so.
Each of the twenty papers in this volume explores one or more of the fol­
lowing themes.
1. How point of view, or the salience of information in discourse, affects
the organizational coherence of text and discourse (Carlson, Erbaugh, Flash-
ner, Fox, Givón, Mithun)
2. The concept of cognitive and linguistic event and how events are
reflected in text and discourse organization (Dixon, Pawley, Szatrowski,
Thompson, Tomlin, Wald)
3. The nature of the linguistic coding of events and other kinds of signif­
icant information (Carlson, Demuth, Erbaugh, Flashner, Fox, Kumpf, Lam-
brecht, Marchese, Rafferty, Ramsey, Thompson)
4) The cognitive bases or cognitive correlates of the linguistic organiza­
tion of discourse (Chafe, DeLancey, Dixon, Pawley, Tomlin)
Support for the Symposium was provided by the University of Oregon
and by the Shao Lin West Foundation.
viii PREFACE

We hope that colleagues interested in problems of discourse organiza­

tion, discourse processing, and functional grammar will find useful the con­
tributions included here.

Russell S. Tomlin
 NARRATIVE CONNECTIVES IN SÙPYÌRÉ

ROBERT CARLSON
University of Oregon

Background
Sùpyîré1 narrative makes heavy use of sequential connectives and a
sequential aspect marker. Generally only the first clause of a narrative is
marked for tense. All subsequent main line (ML) clauses2 are marked with
one of the two sequential connectives kà and ma, which mean roughly "and
then", and most are marked with the sequential aspect marker sí,3 without
any other tense or aspect marking. 4 The two sequential connectives also func­
tion as switch reference markers. As a first approximation, one might say that
kà, which must occur with an overt subject, signals change of subject.5 ma,
which must occur with a zero subject (zero anaphora) 6 signals continuity of
subject. While kà only occurs in ML clauses in narrative, ma also occurs fre­
quently in process texts, where the usual aspect is habitual, and the actions
are encoded in chronological order, ma also signals same subject in these
texts.
The use of these connectives is illustrated in (1) below, the first four
clauses from a folk story. Note that when the sequential aspect (SEQ) marker
si follows a pronoun or the connective ma, it loses its [s] and assimilates in
quality to the preceding vowel.
(1) a. ceè-ŋi wà u ma?a pyà
woman-DEF/Gls IND/Gls PN/Gls PAST child
si
give birth to
A certain woman gave birth to a child.
b. kà u ú faa
PN/Gls SEQ wilt
And she became paralysed.
2 ROBERT CARLSON

c. kà u û nkârâ a sà ù wà
PN/Gls SEQ go go PN/Gls throw
dú-gé na
stréam-DEF/G2s mouth-DEF/G2s on
And she went and threw her away (= exposed her) at the edge
of the stream,
d. ma a
SEQ toad take
and took a toad (in her place).
In the above example, the tense is set for the entire narrative in the initial
clause, (la). Subsequent clauses are marked only with si. Although both par­
ticipants are gender 1 singular, and are thus both referred to with the pronoun
u, there is no ambiguity of reference. The kà in b and c signals that the subject
of its clause is different from that of the previous clause. The ma in d signals
that the (understood) subject of its clause is the same as that of the previous
clause.
The use of ma after a clause with habitual aspect is illustrated in the fol­
lowing examples, which is a parenthetical meta-comment embedded in a nar­
rative. Exactly parallel examples could be gotten from a process text.
(2) a.
youth-Gls HAB woman-stranger-Gls see
Whenever a young man sees a stranger-woman
b. ma â jwó
SEQ say
he says. . .
A careful examination of a Sùpyîré narrative will quickly lead to the con­
clusion that the above picture of the function of kà and ma is too simple. The
switch reference function of kà and ma works consistently only in sequences
of ML clauses. As soon as a non-ML clause intervenes, the situation becomes
much more complicated. The most obvious problem for the switch reference
system is what'is to be the point of reference for the change or the continuity.
Will the change/continuity be marked relative to the immediately preceding,
non-ML clause, or will it be relative to the previous ML clause, skipping over
the intervening non-ML clause? There are, of course, different sorts of non-
ML clauses, and it appears that they have different effects on the functioning
of the switch reference system. But there are a number of examples where the
alternation of kà and ma does not seem to function as a switch reference sys-
 NARRATIVE CONNECTIVES IN SUPYIRE 3

tern at all. The hypothesis put forward here, which I will try to substantiate in
the rest of this paper, may be stated as follows:
(3) a. kà marks thematic discontinuity (i.e. begins a new section/
thematic paragraph);
b. ma marks thematic continuity.
The above hypothesis is intended not only to cover the problem cases
mentioned above (i.e. following a non-ML clause), but also the switch-refer­
ence function of kà and ma. That is, I am making the explicit claim that, in
Sùpyìré at least, switch of subject is merely one kind of thematic discon­
tinuity. As well be seen in the final section of the paper, there are some indi­
cations that different subject marking in at least some other languages may
behave in a similar fashion.
Before continuing, it is necessary to point out that ma may occur in three
different forms: 1) together with the sequential aspect marker sí (sí becomes
[a] in this environment, yielding [maá]), as in 1d above; 2) alone, as in exam­
ple (4) below7 (note that with this form of ma, the sequential aspect is signal­
led by zero); or 3) a reduced form à, as in the middle of lc above.
(4) a. kà u û si
PN/Gls SEQ born
Then she was born,
b. ma lye
old
and grew up,
c. mà sa no kwo-mo
go arrive cut-G5
and arrived at the age of marriage.
The reduced form à may only occur when it directly follows a verb, and
when that verb and the verb following form a closely-knit pair semantically.
One of the verbs may be an "adverbial" verb, meaning, e.g. "again", "fi­
nally", "quickly", "softly". Or the two verbs may express one complex event,
as in "take and come" for "bring", or as in "go to throw away" in (lc) above.
These pairs are thus equivalent to the serial verb constructions found in so
many West African languages. The subject of the verb following à is always
the same as the subject of the preceeding verb. It's occurrence can therefore
tell us nothing about a function of ma beyond same subject marking, and I
have chosen to leave it out of the present study. Thus in all discussion below,
ma refers to the unreduced forms of the conjunction.
4 ROBERT CARLSON

Method
The data base for the study reported here consisted of 25 relatively plan­
ned narratives (= 2,288 clauses), and a lengthy, spontaneous conversation ( =
1,770) clauses). The planned narratives include folk stories, personal narra­
tives, and histories. They were "planned" in the sense that there was ample
time between the request for a text and the actual taping for the narrators to
think about what they would say. The conversation was between three men,
and consists of about two-thirds narrative.
Each kà and ma clause was tabulated according to the type of clause
immediately preceding, and according to whether on not its subject was same
or different than the subject of the preceding clause and of the preceding ML
clause.
A serious problem for a study of this sort is how to operationalize such
notions as thematic discontinuity. I chose to follow the suggestion in Givón
(1983b) that a change in time, place, participant, or action is a reliable guide
to identifying a break in thematicity. These criteria are not entirely satisfac­
tory: such changes only tend to coincide with a thematic break, they are not
the break itself. It is not difficult to find instances where there seems to be a
thematic break, yet none of the above changes is present. And conversely,
there are instances where one or more of the changes are present, yet there
seems to be no thematic break. "Seems to be", however, is not a notion that
will take one very far in analysis, and so one is forced back to the use of
criteria which, although not completely satisfactory, are nevertheless rela­
tively objective.
The above considerations mean that the conclusions presented in this
paper must be regarded as tentative. The final test of the hypothesis in (3)
above must await an independent, reliable method of determining thematic
boundaries, perhaps something like the experimental techniques described in
Tomlin (this volume).
All occurrences of kà and ma which could not be explained by the initial
hypothesis of a switch reference function were examined to see if they coin­
cided with any of the changes noted above. The hypothesis in (3) predicts that
kà should so coincide, whereas ma should not.

Findings
As pointed out above, the evidence for the switch reference function of
kà and ma comes mainly from sequences of ML clauses. Table 1 presents this
 NARRATIVE CONNECTIVES IN SUPYIRE 5

evidence. In all the following tables, DS and SS mean different subject and
same subject from the following kà or ma clause.

ML clause kà maá má
DS 345 = 98.6% 2 = .7% 0
SS 5 = 1.4% 287 = 99.3% 176 = 100%
TABLE 1
Occurrences following a ML clause

It is clear from this table that in the great majority of cases directly fol­
lowing a ML clause, the use of kà may be explained simply as marking a
change of subject.
Of the five cases of kà following a SS ML clause, three co-occur with a
change in place. In the following example, translated fairly literally, there are
other indications as well that there is a thematic break before the kà in ques­
tion:
(5)a. About midnight, kà a python came into the compound.
b. This found my father
c. he was with a male dog (= he had a male dog)
d. This dog had lain down at the door of my father's house.
e. kà he (= the dog) heard the sound of the snake coming
—► f. kà he got up and went and barked at it.
g. then return and bark at the door
i. ma scratch it
j. ma return to where the snake was . . .
Here the anterior, background section (5b-d) is followed by a ML clause
which relates it to the previous ML clause (5a) (for the use of kà in (5e) after
the anterior in (5d), see below). The following clause, (5f), in addition to
encoding a change in place, serves to begin a new section talking about the
dog's repeated barking at the python.
A similar case is seen in the example below:
(6) a. kà I finally stopped
b. ma said
c. "aa, I had better return from here.
d. If I don't return from here,
e. I'm not with anyone here;
6 ROBERT CARLSON

f. if it (= the lion) attacks me

g. I'll remain here (concerned)
h. no-one will know where I am."
i. kà I stopped with my loaded gun.
—►j. kà I returned while the night was coming
k. When the night had gotten dark,
1. kà it (= the lion) afterwards came . . .
Here, (6i) essentially repeats (6a). The reported speech in (6b-h) gives
the reason for the stopping in (6a), and (6i) simply repeats (6a) with an added
adverbial phrase. (6j) then begins with kà, although it has the same subject.
It is obvious that (6i), as a repetition, is"lessML" than a normal ML clause.
In any event, it is certain that a change in time and/or place is not in itself
sufficient to trigger the use of kà: there are many examples where ma is used,
as in the following:
(7) a. kà they killed it (= the python)
b. kà my father took it and went to his place.
—►c. In the morning, ma grilled and cooked it
d. (for) them to eat . . .
In (7c) ma is used in spite of the change of place in (7b) and the time lapse in
(7c). Another example of the use of ma in spite of place change is (5j) above.
The two cases of ma following a different subject ML clause are given
below:
(8) a. kà they (= a mother with her infant on her back) bathed
b. ma went and put the beds down
c. ma said . . .
(9) a. kà the one from Serei said
b. the one from Fantere j should go with the magic powder,
c. he would stay with the land:
d. he. was the elder brother,
i

e. he. would stay with the magic powder,

f. but hei would stay with the land.
g. kà the land became the possession of the people of Sere
—►h. ma gave the magic powder to the people of Fantere.
i. It is for this reason that they go to (help) bury the people of
Sere.
In (8a), the subject includes the mother and her obstinate child who
 NARRATIVE CONNECTIVES IN SÙPYíRÉ 7

refuses to get down off her back. But the subject of (8c) is the mother only.
Although both the mother and child are included in the subject of (8a), only
the mother would be actively involved as agent in the action of bathing. The
use of ma acknowledges this dominance, and continues reference to the
mother only. The case in (9) is similar: the elder brother from Sere is the
agent in the passage. This agenthood lies behind the event recorded in (9g).
What really happened was that the man from Sere kept the land for himself
(and thus it became the inheritance of his descendants), and gave the magic
powder to his younger brother, the ancestor of the people of Fantere. It
seems that in both (8) and (9), ma may be used to indicate an agent as subject,
although it is not the subject of the previous clause, as long as the reference
is easily recoverable from the context. It is to be noted that this is not a very
common strategy, only occurring twice in the corpus.
With the addition of intervening non-ML material, the proportion of kà
clauses which do not simply signal different subject increases. Table 2 gives
the numbers of occurrences after indirect reported speech (RS) and Table 3
those after direct RS.

previous reported
ML clause speech kà maá má
DS 36 = 80% 0 0
SS _ 9 = 20% 4 = 100% 0
TABLE 2
Occurrences after indirect reported speech

previous reported
ML clause speech kà maá má
DS 165 = 96.5% 0 0=
SS 6 = 3.5% 26 = 100% 5 = 100%
TABLE 3
Occurrences after direct reported speech

A comparison of Tables 2 and 3 shows that indirect RS is more disruptive

than direct RS: a much higher proportion of ka clauses have SS as the previ­
ous ML clause when indirect RS intervenes. This is in part due to a special use
of indirect RS. In five out of the nine cases of SS kà from Table 2, the indirect
RS functions not primarily as RS, but as a means of encoding intention. (10)
8 ROBERT CARLSON

below is an example:
(10) a. Once Monkey said
b. that he would look for and eat a person.
c. kà he went and found a girl . . .
These are the first three clauses of a folk tale. (10a) and (b) introduce the
major character and state his intention. In (10c) the story proper begins, with
the putting into execution of the intention. Note that the kà clause is SS with
both the previous ML clause and with the intervening RS clause.
In the other four of the nine cases, the kà clause coincides with a change
in time in two cases, as in (11) below:
(11) a. kà she said
b. "Okay, take her."
c. kà the man took out 100 francs
d. má gave (them to her)
e. (saying) that he had taken her.
f. The next day, kà you ended up going
g. má entering into né gociations for (obtaining) her in (marriage)
h. kà it became a marriage . . .
Here the time lapse in (f) also coincides with the end of the conversation. This
surely could be classified as a "change of action" if anything can. The remain­
ing two of the nine cases from Table 2 do not seem to be as amenable to expla­
nation. One of these is given below:
(12) a. kà Hyena said
b. "(Although) I'm so important, you jump up and sit on that
small box."
c. ma grabbed the cock and put it on a large box
d. ma took that one and went home.
e. When he had arrived
f. ma called his compound people and put them in the house.
g. A cripple was there,
h. ma closed the house on him
i. (saying) that he would not be put in clothes with them,
j . ma said again
k. that Cripple should lean sticks against the house (to barricade
the door)
—►1. kà Hyena opened the box
 NARRATIVE CONNECTIVES IN SUPYIRE 9

m, má found its whole interior

n. full of big snakes . . .
In this example, ma is used repeatedly to indicate SS after a variety of non-
ML clauses, including direct and indirect RS. But in (1) kà is used. This may
be due to the importance of the action in (1): the rest of the story has been
building up to this point, and exists in some sense for the sake of this action,
which consequently may be characterized as the peak of the narrative.
Four of the six cases of kà following a SS ML clause and direct RS, from
Table 3, are parallel to the cases following indirect RS: two coincide with the
end of a conversation and a change in place, and two are of the intent-execu­
tion type noted above. The other two cases are more difficult to characterize.
One of them is in (6) above : (6i) is SS as (6a). As noted above, (6i) is not really
ML in the sense that (6a) is, in that it is mostly a repetition of (6a). This use
of kà for repetition occurs in other contexts. In each case, as in the example
(6), a new item of information or clarification is added. Whether the clause
functions mainly to give this added information, or if the added information
is put in to avoid the tediousness of exact repetition, I don't know. What is
clear is that in these cases ma would have implied sequence, and must there­
fore be avoided,
To sum up, 11 out of the total of 15 cases of kà following SS ML and RS
may be plausibly be said to coincide with a thematic break using the criteria
used in this study. There are 7 instances of the intent-execution break (change
in action), four cases co-occur with a change in time and/or place, and three
of these coincide with the end of a conversation.
Turning now to occurrences following adverbial time clauses, we find a
more complicated situation. 7 Table 4 summarizes the findings:

previous adverbial
ML clause clause kà
DS DS 25
DS SS 2
SS DS 12
SS SS 4
TABLE 4
Occurrences following adverbial clauses

From Table 4 it can be seen that if either the time clause or the previous
10 ROBERT CARLSON

ML clause has a different subject, then kà will be used. However, the number
of examples following configuration (b) is perhaps too low to permit generali­
zation. The split between kà and ma following configuration (d) is more
interesting. In all four cases of kà following (d), the time clause ends in the
verb kw) "finish":
(13) a. That day, kà you went (to the graveyard)
b. má gathered them (= firewood)
c. má found the dead there.
d. When you finished gathering the firewood,
—e. kà you said to them
f. that they should help you put it on your head
g. kà they came up to her . . .
The adverbial clause makes clear that a previous event has come to a conclu­
sion. The narrative moves on to a new episode. This thematic break is signal­
led by the use of kà rather than ma even though both preceding clauses have
the same subject. By contrast, the thematic disruption in the cases where ma
is used seems to be less. Following is an example:
(14) a. kà the man got up
b. ma prepared and went.
c. When he arrived,
— d. ma said
e. "Good evening, sister-in-law."
f. kà she glanced up at him and said . . .
Here the preparing and going are for the purpose of conducting the conversa­
tion which follows (14d). The sequence get up — prepare — go — arrive —
say is seen as one complex event, the most important part of which (from the
point of view of the plot of the story) is the last event, say. The other cases of
ma after a time clause are similar. The speaker has chosen to background one
event, the time clause, in a series which coheres together as one thematic
unit, with the ma clause merely continuing the narration within the thematic
unit.8
The situation after high tone complement clauses is rather different. A
common type of complement in Sùpyìré is one in which the subject is a high
toned pronoun. This pronoun must be coreferential with the direct object of
the preceding main clause. Typical verbs taking this kind of complement are
verbs of causation and perception, such as the verb "see" in example (15)
 NARRATIVE CONNECTIVES IN SÙPYIRÉ 11

below. Table 5 gives the number of occurrences of kà and ma clauses follow­

ing this type of complement clause.

previous complement
ML clause clause kà maá mà

a. DS DS 17
b. DS SS 6 6 10
c. SS DS 11 9 1
d. SS SS

TABLE 5
Occurrences following complement clauses

All 16 cases of ma (maá + má) following configuration (b) (that is, when
the preceding complement clause is SS), are in reality continuations of the
complement. In these cases, the complement has two clauses, the second of
which is the ma clause. After a SS complement, kà must be used to return to
ML: it signals discontinuity with the complement. If the preceding comple­
ment is DS while the previous ML is SS (configuration (c)), there is an almost
equal split between kà and ma. All of these ma clauses are ML — they do not
continue the complement. In three of the kà clauses there is a change of place,
whereas there is none in the ma clauses. There is a hint of the more continu­
ous nature of ma in three of the 10 cases, where a kind of parallelism makes
the ma clause in some sense expected. The following example contains two
high tone complements of the verb "see", (15c) and (e). The use of má in
(15d) is evidently due to the parallelism between (15d) and (15b):
(15) a. When you. arrived in the bush
b. kà theyj saw some bush cows
c. they had cut them, off in front
—► d. má saw some
e. they had cut them, off behind
f. má put them, in the middle now
g. kà the bush cows sang their song
h. kà the father climbed into a tree . . .
Note also in this example the use of a ma clause in (f) to continue the SS com­
plement, and the use of kà in (g) to return to ML following a SS complement.
12 ROBERT CARLSON

It should also be noted that the absence of examples following configuration

(d) in Table 5 is systematic: the high tone complement must always be DS
from the preceding main clause.
The situation following negative clauses is somewhat more clear, though
fewer examples are available. Table 6 gives the relevant numbers.

previous negative
ML clause clause kà

DS DS 11
DS SS
ss DS 5
ss SS 2

TABLE 6
Occurrences after negative clauses

ma following a negative returns to ML, and does not continue the negative-
ness of the preceding clause. Table 6 shows that when the negative clause is
DS kà must be used. When both the negative and the previous ML are SS,
there is again a split between kà and ma. It is significant that in one of the two
cases of kà there is a lapse of time, signalled by a time phrase at the beginning
of the kà clause. With the ma clauses, there is no such change in time. The fol­
lowing example has both a kà and a ma clause following negatives:
(16) a. One day, a hunter came
b. ma greeted him (= Monkey).
c. kà he bobbed.his.head up and down.
d. kà the hunter passed.
e. He didn't speak again.
—►f. The next day, kà he again found Monkey there
g. praying
h. ma greeted him.
i. He didn't return the greeting,
—► j . ma bobbed his head up and down.
k. kà the hunter said . . .
Note the time phrase in (f), the kà clause, and the lack of any time lapse in (j),
the ma clause. Here (j) is not strictly comparable to (f) in that the previous
 NARRATIVE CONNECTIVES IN SUPYIRE 13

ML clause is DS — that is, it is following an instance of configuration (b), An

example of a ma clause with both the negative and previous ML as SS is given
below:
(17) a. ká it (— the lion) afterwards came
b. má again passed through the same place.
c. It didn't see another (cow)
—► d. mà passed under that little baobab [gesture]
e. mà again went down to the stream.
f. At that time, the stream was really over-grown . . .
Note here, too, the absence of time lapse. Although there is a change in
place, this is continuous with the change of place in the previous ML clause.
In fact, the lion's passing through the village and out again may be said to be
the theme of the entire paragraph.
The use of kà and ma after anterior clauses provides further evidence
that ma signals more continuity, although, again, the numbers of e x a p e s are
very small.
previous anterior
ML clause clause kà maá mà

a. DS DS 10
b. DS SS 2 2 2
c. SS DS 6 2
d. SS SS

TABLE 7
Occurrences following anterior clauses

Just as with complement clauses, when the anterior clause is SS (configura­

tion (b)), the ma clause continues the anteriority of the previous clause, and
does not return to ML. Example (18) below shows the use of ma (18h) to con­
tinue the anteriority of the previous clause. kà is used in (18i) to return to ML,
even though the previous clause has the same subject. Use of ma in (18i)
would presumably have implied the continuity of the anteriority of the previ­
ous passage. (For another example of kà after a SS anterior, see (5e) above.)
(18) a. kà she said "OK."
b. ma went to the stream
c. ma was washing the pants
14 ROBERT CARLSON

d. ma throwing water on the people.

e. A certain old woman was in the village.
f. She had heard the words of Njible and her mother in the bush.
g. That day, she had hidden
—► h. ma watched them.
—►i. kà she also came to the stream to wash clothes
j . má found Njible
k. she was throwing water on the people . . .
Again just as with complement clauses, when the anterior clause is DS
(and the previous ML is SS) both kà and ma clauses are ML. Two of the six
kà clauses are repetitions of the sort mentioned in the discussion of kà follow­
ing RS above. Two others co-occur with a change of place. The rest are not
explicable at this stage. The following example shows the use of ma in a ML
clause even after a lengthy RS and an anterior. The continuity of action is
obvious:
(19) a. kà she (= Monkey, disguised as a woman) went out quickly far
into the bush
b. mà took her money skin and put it on
c. mà ran round the termite mound
d. saying
e. "Termite mound, termite mound, you said
f. that they would know me,
g. that they would know me,
h. (but) no-one knew me."
i. The man had bought a comb for her, and a scarf, and a cloth,
—► j . ma took them off and put them down
k. ma ran around the termite mound with her monkey skin
1. ma sang the song . . .
It is obvious from the context that (19i) was inserted as a detail necessary for
the proper understanding of the second verse of the song following (191) (the
first verse is (19e-h)). Its intrusion does not greatly disturb the thematic
coherence of the passage as a whole.
The use of ma to continue the habitual aspect has been mentioned above
(see example 2). A kà clause following a habitual clause will return to a punc­
tual/perfective aspect. There are only five such kà clauses in the data, and
only one of these follows a SS habitual clause:
 NARRATIVE CONNECTIVES IN SUPYIRE 15

(20) a. Shei would go out

b. and go to the graveyard
c. and gather firewood.
d. One day, kà you went . . .
Here the habitual clauses (a-c) as a section are arguably ML: that is, they
encode events which are subsequent to the previous ML events (not given
here) and which preceed the event in (20d). A kà clause is used to discontinue
the perfective aspect, and to continue the narration with the more usual punc­
tual encoding of ML events.

Discussion
That part of the function of kà is to mark DS is clear from the evidence
presented above. In this connection, note that in Tables 4-7, only kà appears
after configuration (a), that is, when both the preceding clause and the previ­
ous ML clause are DS. Given the limits of the methodology noted above, a
reasonable case can also be made for hypothesis (3) above: in a significant
number of cases, where kà and ma are used with similar preceding environ­
ments, the kà clause coincides with a change in time, place, or action, ma, on
the other hand, is used to continue aspects such as habitual and anterior, and
the subordination of complement clauses. Some of the unexplained examples
could perhaps be resolved if better syntax-independent techniques for deter­
mining thematic boundaries were available.
Assuming, meanwhile, that the hypothesis holds, it is interesting to note
that the discontinuity signalled by kà is relative to the immediately preceding
context only. Since kà marks a ML clause, it marks continuity of the narra­
tion. The set of ML clauses is in some sense what the narrative is "about".
Each ML proposition continues the narrative, and kà therefore has an impor­
tant cohesive function. In a sense, it returns to, or better, begins a new section
of the larger or global theme. Thus from the point of view of a macro, text-
wide theme, kà may be said to fill a function of continuity, or at least develop­
ment. On a micro, local level, however, as is clear from the evidence pre­
sented in this paper, kà marks discontinuity with what precedes. Frequently,
what precedes is non-ML, i.e. background material. From the text-wide
point of view, this material is non-thematic. ma may often be used inside such
non-thematic material to continue its non-ML status. kà is used to "reinstate"
the (thematic) narrative.
It is well known that in switch reference systems there are almost always
16 ROBERT CARLSON

"exceptions" to the SS/DS marking which pose problems for a simple

analysis. For example, Longacre (1972) points out that in many New Guinea
languages, the switch reference system skips over temporal clauses as if they
weren't there. In at least one New Guinea language, Timbe, the DS marker
is used to introduce a non-ML, background passage, and the SS marker is
used to reinstate the ML. 9
I have found two other cases of the use of a DS marker to mark thematic
discontinuity even when there is not a change in subject. The first of the two
languages (brought to my attention by Robert Longacre) is Paez, a Macro-
Chibchan language of Colombia. Gerdel and Slocum (1976) state that "on the
paragraph and discourse levels . . . a change of situation — even without
change of subject — can result in the selection of atsa' "and-DS"." The sole
example given is from a process text, where the sentence introduced by atsa
gives an alternative to a particular procedure.
More information is available about the other case, the two closely-
related Uto-Aztecan languages Pima and Papago. Scancarelli (1983) states
that uku- is not a different-subject marker [as has been asserted by various
authors], but rather is a marker of discontinuity, and . . . change of subject is
just one kind of discontinuity1'. Looking at the specific places where ku-
marks discontinuity, there are striking parallels with Sùpyiré kà. According
to Scancarelli, "ku- is used to introduce a main clause when that clause repre­
sents a shift (in focus of attention, point of view, or time frame) — a discon­
tinuity of sorts — from the previous clause. A change in subject is one kind
of discontinuity; others include: . . . a shift from direct or indirect speech or
thought to action (i.e., to the sequence of main events in the text); a shift from
description or elaboration to action; or a change in time". In addition "ku- .
. . is used to introduce a minor flashback, or a recapitulation of events already
described." Note the similarities here to the use of kà in Sùpyìré to move to
action after a statement of intention in RS, and as a repetition, or recapitula­
tion of a previously mentioned event.
So many and striking similarities cannot of course be accidental. The evi­
dence strongly supports the claim made here (and in Scancarelli, 1983), that
change or continuity of subject is just one kind of thematic change or con­
tinuity, and this makes the extension of the use of a DS marker to mark other
kinds of discontinuity an effective strategy of communication. This corrobo­
rates, from a different perspective, the claim made in Givón (1983a) that
questions of thematicity are relevant to any understanding of the narrow
notion of switch reference.
 N A R R A T I V E C O N N E C T I V E S IN S Ù P Y Î R É 17

NOTES
9) Sùpyiré is a Senufo language of the Gur or Voltaic family of Niger-Congo. It is spoken by
about 200,000 people in southern Mali. The texts on which this.study is based were collected by
myself in the village of Farakala, Mali, in 1980-83. The dialect they represent is known as Kam-
polondugu. I am not sure how far the findings discussed here are true for other dialects of Sùpyiré.
I would like to thank Ely Sanogo of Farkala for helping me in the painstaking task of collecting and
transcribing texts.
2) I use main line in the same sense as Hopper (1979), that is, to describe narrative clauses which
encode events in their natural chronological order. These clauses constitute the backbone of the
narrative, what the narrative is "about". With the exception of adverbial time clauses (see note 6)
and some complement clauses, virtually all ML clauses are marked with a sequential conjunction,
and most are marked with the sequential aspect marker.
3) Compare the situation in Swahili as reported in Hopper (1979), where the initial verb in a
narrative is marked with the preterite prefix li-, while subsequent main line verbs are marked with
the prefix ka-.
4) The sequential aspect marker may co-occur with the progressive aspect marker to encode a
durative event. Occasionally such an event is simultaneous, or overlaps with the event encoded in
the preceding or following clause.
5) Herber (n.d., vol. 2, p. 12) calls ma a "conjunctive substitute particle and says "it functions
as a conjunction in connecting one verb to another and . . . as a substitute particle in two ways. It
substitutes for the previously introduced subject, and it substitutes for the previously introduced
aspectual particle". I am not aware of any possible diachronic pronominal source of ma, though
given the negligible amount of information available on the history of Senufo languages, such a
possibility cannot be dismissed outright.
6) When it occurs alone, ma has a low tone. Ï am uncertain as to the underlying tone of ma, as
there are no further examples in Sùpyiré of either low tone becoming mid before high, or mid tone
becoming low in isolation, ma may occur in subjectless clauses with a very limited number of verbs.
These clauses, which function as adverbial time clauses for a following ML clause, do not have a
zero anaphor subject: they are truly subjectless. Following is an example:
a. má ù yà? a kil-ni i
PN/Gls leave path-DEF/G3s in
While he was still on the path,
(lit. and leave him in the path)
b. kà it ú pi jiyá. ..
PN/Gls SEQ PN/Glp see
he saw them . . .
There are 17 cases of subjectless ma clauses in the corpus.
7) Strictly speaking, most adverbial time clauses are ML, given the (operational) definition of
ML use here, since they encode events in their natural chronological order. Syntactically, how­
ever, they are subordinate. They always occur to the left of the main clause, they do not take kà
or ma, and they have the perfect aspect rather than the sequential. Finally, they are marked with
a clause-final subordinating particle, ké.
8) There are 6 cases of a ma clause within an adverbial, i.e. the adverbial is two clauses long, and
the ké subordinating particle comes at the end of the ma clause. When a ma clause follows the ké
18 ROBERT CARLSON

marker, it is always a ML clause.

9) This information is from Eileen Gassoway. personal communication.

ABBREVIATIONS

DEF definite
DS different subject
G gender. Sùpyìré has five genders, G l , G2, etc. Genders 1-3 have
both singular and plural forms.
HAB habitual aspect marker
IND indefinite
ML main line
P plural
PN pronoun
RS reported speech
s singular
SEQ sequential aspect marker
SS same subject

REFERENCES

Gerdel, Florence L., and Marianna C. Slocum. 1976. "Paez discourse, para­
graph and sentence structure.11 In Robert E. Longacre. ed. Discourse
Grammar: Studies in Indigenous Languages of Colombia, Panama, and
Ecuador. Dallas: Summer Institute of Linguistics.
Givón, T. 1983b. "Topic continuity in discourse: the functional domain of
switch reference.11 In J. Haiman, ed. Switch Reference. Typological
Studies in Language, vol. 2. Amsterdam: John Benjamins.
——. 1983b. "Topic continuity and word order pragmatics in Ute." In T.
Givón, ed. Topic Continuity in Discourse. Amsterdam: John Benjamins.
Herber, Ralph. n.d. Suppire Unraveled. ms.
Hopper, Paul J. 1979. "Aspect and foregrounding in discourse.11 In T. Givón,
ed. Discourse and Syntax, Syntax and Semantics, vol. 12, New York:
Academic Press.
Longacre, Robert E. 1972. Hierarchy and Universality of Discourse Con­
stituents in New Guinea Languages: Discussion. Washington: Georgetown
 NARRATIVE CONNECTIVES IN SÙPYÏRÉ 19

University Press.
Scancarelli, Janine. 1983. Switch-reference in Pima and Papago. ms.
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW

WALLACE CHAFE
University of California, Santa Barbara

Some years ago (Chafe: 1973, 1974, 1976) I tried to find some order
among the linguistic devices which Î thought of as manifestations of linguistic
"packaging", devices having less to do with the content of an utterance than
with the way that content is wrapped up and presented to a hearer. What fol­
lows is an update and, I would like to think, an improvement on that earlier
attempt. It can be viewed as part of a continuing effort to become clearer on
what is happening in people's minds when they manipulate "given and new
information", "topics and comments", "subjects and predicates", as well as
"intonation units", "clauses", "sentences", and "paragraphs". I assume that
the linguistic phenomena which have been given names like these are man­
ifestations of basic cognitive processes, and that we can never understand
them fully until we understand the psychological phenomena underlying
them. 1
Whatever new things I may say here are based partly on interactions with
other discourse scholars in the intervening years and partly on exposure to,
and reflection upon much new data, notably the "pear stories" (Chafe: 1980)
and materials collected to investigate differences between spoken and written
language (Chafe: 1982, 1985; Chafe and Danielewicz: 1986).2 I will restrict
the discussion here entirely to spontaneous spoken language. In a longer
work there would be much to say about parallel phenomena in more planned
varieties of spoken language, as well as in writing of various kinds.
It seems especially difficult to establish a satisfying terminology in this
area. Terms like "old" or "given" information, "new" information, "topic",
"comment", and so on have been especially prone to misinterpretation, and
have been used in a confusing variety of different ways. I have wondered
whether to keep using them here, and thus invite continued misunderstand­
ing, or whether to sweep the decks clean and introduce other, perhaps more
appropriate terms and invite the disadvantages of terminological prolifera-
22 WALLACE CHAFE

tion. I have finally decided on a mixture of the two strategies. I will retain the
terms "given" and "new", while defining them in quite specific cognitive ways
for which the labels "already active" and "previously inactive" will be seen as
appropriate alternatives. I will introduce the notion of "accessible" or "previ­
ously semi-active" information as a third type which is in a sense intermediate
between given and new. With respect to a different opposition, I have settled
on the term "starting point" for what has sometimes been called a "topic",
along with "added information" for what has sometimes gone been called a
"comment". I believe that my intentions with these various labels will
become clear as we proceed.

Intonation Units
I assume to begin with that our minds contain very large amounts of
knowledge or information, and that only a very small amount of this informa­
tion can be focused on, or be "active" at any one time. It is intuitively obvious
that our capacity to activate information is severely limited. The active por­
tion of our knowledge has sometimes been said to reside in short-term mem­
ory, and from that point of view there is relevance in the idea that short-term
memory can hold only seven plus or minus two items (Miller: 1956). Whether
or not that figure is exactly right, and exactly what an "item" might be, need
not occupy us just now. The important thing is that short-term memory does
not hold very much. In the long run it is better not to think about a kind of
memory, but about a certain limited amount of information in our minds
being temporarily lit up. From that point of view the term "activated" seems
appropriate.
When a speaker is speaking, her or she verbalizes one piece of temporar­
ily active information after another. Each such piece is expressed in what I
will call an "intonation unit" (cf. the "tone unit" of Crystal: 1975 and
elsewhere). An intonation unit is a sequence of words combined under a
single, coherent intonation contour, usually preceded by a pause. An intona­
tion unit in English typically contains about five or six words, and new intona­
tion units typically begin about two seconds apart. Evidently active informa­
tion is replaced by other, partially different information at approximately two
second intervals.
To show the nature of intonation units, and also to provide an ongoing
example for the discussion to follow, I am going to quote a small part of a con­
versation that was tape-recorded during an informal dinner party. At one
point in the evening one of the participants remarked that it was important for
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 23

university teachers to have personal contact with their students. The husband
of the person who had made that remark then contributed the following anec­
dote.
I have segmented the transcription into intonation units, giving each a
number. Two dots indicate a brief break in timing, three dots a full-fledged
pause. Two hyphens indicate lengthening of the preceding segment. The
acute accent shows intonation peaks. Contributions by another participant
are given in parentheses.
1. ... It's fúnny though,
2. ... I dó think that makes a difference .. but,
3. ... I can recall ... uh-- ... a big undergraduate class that I had,
4. ... where .. éverybody loved the instrúctor,
5. ... a—nd .. he was a ... real .. uh .. old world ... Swiss- ... guy,
6. .. this was uh .. a biólogy course,
7. ... a—nd h e - ... left all of the— sort of u h - ... real cóntact with stu­
dents .. up to .. his assistants.
(8. ...Mhm,)
9. ... A—nd .. he would come into class,
10. ... a—t.. uh—you know three or f
11. .. precisely one minute after the hour,
12. or something like that,
13. ... a—nd h e - .. wou-ld .. immédiately open his ... notes up,
14. ... in the front of the róom,
15. .. and he st
16. and évery ... évery lécture,
17. ... after the first,
18. .. stárted the same way.
19. This was .. u - m at Wésleyan,
20. when Wesleyan was still ... a mén's school.
21. ... So every lecture after the first would begin,
22. ... Gentlemen,
23. .. ze last time,
24. ve vere talking about,
25. .. and then he would
(26. (laughter))
27. ... But then .. you know he would just .. give a lécture,
28. .. a-nd .. there was no .. real interáction with t h e - .. the students,
29. .. and then .. a t . . at the end,
24 WALLACE CHAFE

30. .. he would close his notes,

31. .. and walk out of the room.
(32. .. Hm.)
33. ... And he was the
34. . . I I guess that is the .. old world style,
(35. Yéah.)
36. ... of lßcturing.
37. But he was the .. the most extreme exámple I had . . I ever had as
student.
(38. .. But he was interesting.)
39. And he was véry good,
40. .. yéah.
Looking at the general format of language production in this way, we get
a picture of alternating pauses and vocalizations that is schematically repre-
sentable as follows:
... xxxxx, ... xxxxx, ... xxxxx, etc.
where, of course, the number of words in an intonation unit as well as the
number and position of intonation peaks is variable. I will call the pauses in
this picture "initial pauses" to differentiate them from any pauses that may
occur during the periods of vocalization (cf. the "planning phases" of Butter-
worth: 1975).
The pauses as well as the vocalizations are of interest. Although we may
hope that more evidence on this score will eventually be available, it is fruitful
to speculate that what happens during these pauses has something to do with
changes in the activation states of information in the speaker's mind. Such
changes probably require a certain amount of time-consuming cognitive
effort, which may be a chief reason for the periods of silence. Not only is cer­
tain information becoming newly activated in the speaker's mind, but also, if
the total amount of information that can be active at any one time is severely
limited, there must be other information that is passing out of the active
status, being-replaced by other, newly activated material. By the end of the
initial pause, at least under ideal conditions, all the information to be expres­
sed in the upcoming period of vocalization will have become active in the
speaker's mind. This is not to say that changes in activation states never occur
while a speaker is speaking, but that when such changes do occur they repre­
sent a kind of disfluency, or deviation from the ideal.
It is also interesting to consider what may be happening in the mind of
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 25

the hearer. Presumably the periods of vocalization function to activate cer­

tain information for the hearer, while other information leaves the active
state. In fact, the speaker's purpose in speaking is precisely to bring about
changes in the activation states of information in the hearer's mind. There is,
then, a speaker-hearer alternation associated with the pauses and vocaliza­
tions, with changes in activation states taking place in the speaker's mind dur­
ing the pauses, and in the hearer's mind during the periods of vocalization:
Speaker: change change change
xxxxx, ... xxxxx, ... xxxxx,
Hearer: change change change

The Three Activation States

We cannot suppose that each intonation unit expresses a single, indivis­
ible chunk of information. Much of our discussion from now on will have to
do with pieces of information within those chunks, and the kinds of activation
to which those pieces are subject. We will be looking more closely at the way
the pieces are activated and deactivated in the speaker's and hearer's minds
during the pauses and periods of vocalization, and we will be led to identify
more than one state of activation.
Some of what is expressed in intonation units involves people's ideas of
objects, events, and properties. These are the ideas that are typically expres­
sed in noun phrases, verb phrases, and adjective phrases. (Also expressed in
intonation units are hedges, epistemic attitudes, tenses, aspects, and the like,
but they will not be the focus of our attention here.) I will refer to the ideas
of objects, events, and properties as "concepts". The important point is that
when we talk about states of activation, these are not states that apply to the
larger chunks of information expressed in intonation units. Rather, they
apply to the separate concepts that reside within those chunks.
There is evidence that a particular concept, at a particular time, may be
in any one of three different activation states, which I will call "active",
"semi-active", and "inactive". An active concept is one that is currently lit up,
a concept in a person's focus of consciousness. A semi-active concept is one
that is in a person's peripheral consciousness, a concept of which a person has
a background awareness, but which is not being directly focused on. An inac­
tive concept is one that is currently in a person's long-term memory, neither
focally nor peripherally active.
26 WALLACE CHAFE

Active Concepts ("Given Information")

Let us imagine a point in time at the end of the initial pause, when the
speaker is about to utter a new intonation unit, as indicated by the caret:
... xxxxx,
A

At this point, under ideal conditions, all the concepts about to be verbalized
in the upcoming intonation unit are, for the speaker, in the active state. But
at this point the speaker will also have assessed which of these concepts are
already active in the mind of the hearer as well. Those concepts which are
already active for the speaker, and which the speaker judges to be active for
the hearer as well, are verbalized in a special way, having properties which
have often been discussed in terms of "old" or "given" information.
The general thing to say is that given concepts are spoken with an
attenuated pronuncation. The attenuation involves, at the very least, weak
stress. Typically, though not always, it also involves either pronominalization
or omission from verbalization altogether (the maximum degree of attenua­
tion). A few examples are in order. In each example I italicize the words
which express given concepts. In 2:
2. ... / dó think that makes a difference .. but,
the speaker referred to himself with the weakly stressed pronoun 'T" because
he assumed that the concept of him was already active in the consciousness of
the hearer(s), and he made that assumption because he was a participant in
the conversation. First and second person referents acquire the given status
naturally from the conversational context itself (Chafe 1974: 123-124). The
word "that" is a weakly stressed pronoun because it refers to the idea that
good teachers interact with their students, an idea that had been expressed by
another speaker just before this narrative began, and that could thus be
assumed to be still in the active consciousness of all the conversational partic­
ipants.
Once the concept of the instructor had been introduced into everyone's
consciousness in 4:
4. ... where .. éverybody loved the instrúctor,
this concept came to be treated as given for the remainder of the narrative.
Thus, it was referred to with the weakly stressed pronoun "he" in 5:
5. ... a—nd .. he was a ... real .. uh .. old world ... Swiss- ... guy,
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 27

as well as in 7, 9, 13, 15, 25, 27, 30, 33, 37, 38, and 39.
Another straightforward case of given information was the referent of
"this" in 6:
6. .. this was uh .. a biólogy course,
a referent that had been activated through its introduction shortly before in 3:
3. ... I can recall ... uh— ... a big undergraduate class that I had,
The word "that" in 34 provides a similar example:
34. ..I I guess that ìs the .. old world style,
The referent of "this" in 19 is the sequence of events set forth immediately
before:
19. This was .. u—m at Wésleyan,
"Wesleyan" itself then came to be treated as given in the immediately follow­
ing intonation unit 20, where it was weakly stressed:
20. when Wesleyan was still ... a mén's school.
This referent was not pronominalized in 20 because it represented a change
in "starting point" from the preceding intonation unit, as will be discussed
below.
There are a few less straightforward instances of given concepts. For
example, in 1:
1. ... It' s fúnny though,
the "it" referred to a given concept whose nature is difficult to specify. This
was not, apparently, the dummy "it" of "it's raining", since someone might
conceivably have asked, "What's funny?" ("What's raining?" would be
peculiar.) But its referent can perhaps be characterized no more precisely
than "what I am about to say" — not a concept, strictly speaking, that was
already active in the hearer's mind, but one that was formulaically pretended
to be so.
In intonation unit 3:
3. ... / can recall ... u h - ... a big undergraduate class that / had,
there are two occurrences of "I", both of which show strong, not weak stress.
The first person referent must have been active in both the speaker's and the
hearer's minds, but it was also contrastive, and for that reason stressed. The
speaker was contrasting his own understanding with that which had just been
28 WALLACE CHAFE

verbalized by the preceding speaker (Chafe 1974: 117-119, 1976: 33-38).

The referent expressed by the entire object noun phrase in 3:
a big undergraduate class that I had,
was also given. In Chafe (1974: 125-127) I noted that when a particular
instance of a category has been activated, all other instances of the category
are simultaneously activated too. Here the speaker assumed, because of the
preceding conversation, that everyone was already thinking about big under­
graduate classes. It would not have been sufficient simply to pronominalize
this referent:
I can recall one that I had,
because the referent of "one" would not have been clear. The example is use­
ful in showing that a referent may be given even when it needs to be spelled
out in a full noun phrase.

Semi-Active Concepts ("Accessible Information")

What follows will appear in clearer perspective if we keep in mind a
hypothetical frame of reference in which, for the speaker, the initial pause is
occupied by changes in the activation states of one or more concepts, each
such change belonging to one of three types:
(1) a change of a previously inactive concept to an active one.
(2) a change of a previously semi-active concept to an active one.
(3) a change of a previously active concept to a semi-active one.
We can speak of types (1) and (2) as the "activation" of an inactive or semi-
active concept respectively. There is really no difference between (1) and the
psychological notion of "recall". To say that a speaker is recalling something
from long-term memory is equivalent to saying that he is activating something
that was previously inactive. We can speak of (3) as the "deactivation" of an
active concept. Because of the limited capacity of focal consciousness, deacti­
vation takes place as frequently as activation.
It will be instructive now to look at our narrative for some examples of
concepts that were recalled by the speaker from the semi-active into the active
state (type 2). I will call them "accessible" concepts. We can at the same time
consider how these concepts came to be semi-active. I hope in this section to
clarify what it means to be accessible information, but the specifically linguis­
tic consequences of accessibility will not become clear until we have taken up
some additional aspects of verbalization.
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 29

There seem to be two ways in which concepts become semi-active. One

way is through deactivation from an earlier active state, typically through
having been active at an earlier point in the discourse. A concept does not
remain in the active state very long unless its activation is refreshed. But as
concepts become deactivated they do not immediately become fully inactive,
but may remain in the semi-active state for some time. The properties of
semi-active consciousness have never been systematically studied, but it
clearly is not subject to the same capacity and time limitations as active con­
sciousness. Later I will suggest that "paragraph" boundaries are evidence of
major changes in peripheral activation.
In our narrative, intonation unit 5 introduced the property of being "old
world":
5. ... a—nd .. he was a ... real .. uh .. old world ... Swiss- ... guy,
This "old world" property was not mentioned again until 34, by which time it
must have passed out of activation, but not out of semi-activation:
34. . . I I guess that is the .. old world style,
During the initial pause of 34, then, the concept of "old worldness" must have
been recalled from the semi-active state.
The other way in which concepts may become semi-active is amply illus­
trated in this narrative. These are concepts which belong to the set of expec­
tations associated with a schema (Bartlett: 1932, Mandler and Johnson: 1977,
Schank and Abelson: 1977, Tannen: 1979, etc.). A schema is usefully
regarded as a cluster of interrelated expectations. When a schema has
been evoked in a narrative, some if not all of the expectations of which it
is constituted presumably enter the semi-active state. From that point on,
they are more accessible to recall than they would have been as inactive con­
cepts.
The narrative at hand began by evoking the schema of an under­
graduate class. If we disregarded the conversational context of the narrative,
we could suppose that this schema was evoked by mention of the class in into­
nation unit 3:
3. ... I can recall... u h - ... a big undergraduate class that I had,
Evidently, however, the referent of "a big undergraduate class" was already
being treated as an active concept because of prior talk about such classes. It
seems, therefore, that the class schema had already been evoked by the ear­
lier conversation, and that at the beginning of this narrative various concepts
30 WALLACE CHAFE

associated with it were already semi-active or accessible in the minds of those

present.
Schemas can be characterized in part by listing the participants and
events included in them. In the class schema these include at least the follow­
ing:
students
an instructor
teaching assistants
the instructor's notes
a classroom
a lecture
Concepts of this sort must then have been in the semi-active state throughout
the narrative, except during the periods when they were fully active. We can
look at the intonation units in which these semi-active concepts were expres­
sed. In the following examples the words that verbalize such accessible con­
cepts are given in boldface italics.
In 4 there were two accessible concepts: the students (verbalized as
"everybody"), and the instructor. Thus:
4. ... where .. everybody lóved the instrúctor,
In 7 the students and the teaching assistants were accessible:
7. ... a-nd h e - ... left all of t h e - sort of u h - ... real co0ntact with stu­
dents .. up to .. his assistants.
In 9 it was the class itself:
9. ... A - n d .. he would come into class,
In 13, the instructor's notes:
13. ... a-nd h e - .. wou—ld .. immédiately open his ... nótes up,
In 14, the room:
14. ... in the front of the room,
In 16, the lecture:
16. and every ... every lecture,
In 27, the event of giving a lecture:
27. ... But then .. you know he would just .. give a lecture,
In 30, the instructor's notes again:
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 31

30. .. he would close his notes,

In 31, the room again:
31. and walk out of the room.
and in 36, the lecturing again:
36. ... of lécturing.
Before we can appreciate the consequences of accessibility, however, we
need first to look at the third state that concepts may be in.

Inactive Concepts ("New Information")

Another type of change that takes place during the initial pause is the
change of a concept from the inactive to the active state. Such concepts will
not have been active or semi-active because of anything that was said earlier
in the discourse, nor will they have become semi-active through the evocation
of a schema. In the following selected examples, those words that expressed
previously inactive, or "new" concepts are given in boldface:
1. ... It's funny though,
2. ...I dó think that makes a difference .. but,
5. ... a~nd .. he was a ... real .. uh .. old world ... Swiss- ... guy,
6. .. this was uh .. a biólogy course,
This kind of change, which amounts to the recall of a concept from long-
term memory, evidently exacts a greater cost in terms of cognitive effort than
any other kind. Deactivation of a concept from the active to the semi-active
state probably exacts no cost at all; it is an automatic consequence of the
limited capacity of focal consciousness. Activating a semi-active concept
probably exacts only a minimal cost. But the change we are now considering
shows evidence of being cognitively more difficult.
One piece of evidence is the initial pause itself. A principal reason for
this pause is the very process we are considering. To activate something that
was previously inactive evidently takes time. Another piece of evidence is not
as obvious, but it is crucial to our present concern. Simply stated, it is the find­
ing that only one concept can be changed from the inactive to the active state
during any one initial pause. If we ignore the possibility that activation may
occasionally take place during vocalization itself, this amounts to saying that
a single intonation unit can convey no more than one previously inactive, or
new concept. By analogy to the not unrelated "one clause at a time con-
32 WALLACE CHAFE

straint" of Pawley and Syder (1983: 564-565), I will call this the "one new con­
cept at a time constraint", (Something quite similar was suggested by Givón
(1975: 202-204) in terms of "one [new] unit per proposition".) This constraint
results naturally from what I take to be the cognitive basis of an intonation
unit: the expression of a single focus of consciousness. Such a focus can evi­
dently contain no more than one previously inactive concept.
To see the effect of this constraint, it will be useful to have a list of how
all the concepts in this narrative were verbalized, with columns separating
them into (1) concepts that were already active for the speaker before he
uttered the next intonation unit, and that were assumed to be already active
in the mind of the hearer as well ("given" concepts), (2) concepts that the
speaker, during the initial pause, transferred from the semi-active to the
active state in his own mind ("accessible" concepts), and (3) concepts that the
speaker, during the initial pause, transferred from the inactive to the active
state ("new" concepts). As in the examples above, T will cite expressions of
these three types in italics, boldface italics, and boldface respectively. Certain
of the expressions in the lefthand column are given in parentheses, for a
reason that I will come to shortly.

Given Accessible New

Concepts Concepts Concepts
1. (it) is funny
2 (that) makes a difference
3. (I) can recall
a big under­
graduate class
that I had
4. (everybody) loved
the instrúctor
5. (he) was a real old
guy world Swiss
6. (this) was a biólogy
course
7. (he) stúdents left áll of the
his assistants real contact
with—
up t o -
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 33

9. (he) class would come into

10. at three or f
11. precisely one minute
after the hour,
13. (he) his notes would immediately
open -« up
14. the room in the front of
15. (he) st
16. every lecture
17. after the first
18. started the same
way
19. (this) was at Wésleyan,
20. (Wesley an) was still a mén's
school
21. (évery lecture after would begin
thefìrst)
22. Gentlemen,
23. ze last time
24. (ve) vere talking about
25. (he) would
27. (he) would just give
a lecture
28. the students was no real inter­
action with—
29. at the end
30. (he) his notes would close
31. the room walk out of
33. (he) was the
34. (that) is the old world style
36. oflécturing.
37. (he) as a stúdent was the most extreme
exámple I ever
had
38. (he) was interesting
39. (he) was very good
34 WALLACE CHAFE

The one new concept at a time constraint dictates that each of the expres­
sions in the righthand column must express a unitary concept. It would be dis­
confirmed by expressions in that column which could be shown to express
more than one concept. I am not, of course, talking about the number of
words contained in these expressions, but rather about whether the concepts
expressed by these words are unitary or not. It is important to realize that
there is no consistent relation between the status of being a unitary concept
and the length or syntax of the verbalization of the concept. Thus, "Mildred",
"opened the door", or "the house I used to live in" may equally well express
unitary concepts. Given the evident importance of this notion, we can hope
that useful tests for "unitariness" can be devised. For the time being I rely on
the reader's intuitive appreciation of this property.
A single word like "loved" in 4 is the simplest case:
4. ... where .. everybody loved the instructor,
Even though this intonation unit contained three stressed words, two of them
expressed accessible concepts and only one, "loved", expressed a new con­
cept.
One clear type of word sequence that may express a single new concept
is the construction that consists of a copula followed by either an adjective, a
propositional phrase, or a noun phrase. Examples from this narrative include
"is funny" (1), "was interesting" (38), "was at Wesleyan" (19), or "was still a
men's school" (20). In many such cases one can imagine the concept in ques­
tion being expressed by a single word in some other language, where it has
become institutionalized. Similar in nature is a copular phrase which contains
a stressed noun or adjective modifying a concept that is given, as in "was a
biology (course)" (6), or "was a real old world Swiss (guy)" (5).
The last example raises the question of whether, since "old world" and
"Swiss" both carried intonation peaks, they did not express two separate new
concepts. A reasonable reply is that they actually expressed one and the same
property, not two separate ones. Having said "old world", the speaker hesi­
tated and then said "Swiss", evidently in an attempt to clarify the same con­
cept through two different verbalizations.
When an entire intonation unit functions as an adverbial modifier of a
preceding or following clause, it too can be assumed to convey a unitary con­
cept: for example, "ze last time" (23), or "at the end" (29). A similar exam­
ple, "in the front of the room" (14), seems in fact to have conveyed new infor­
mation only with the prepositional sequence "in the front of", since the con-
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 35

cept of "the room'1 was accessible.

There were several cases in this narrative where a unitary concept like
"one minute after the hour" (11) or "was good" (39) was accompanied by a
stressed degree adverb like "precisely" or "very". Evidently such an adverb
does not convey a separate concept, but only expresses the degree of the con­
cept to which it is attached. Intonation unit 13, "he would immediately open
his notes up", also contained a stressed adverb of this sort in the word "im­
mediately". The only new concept in this intonation unit was, then, the verb-
particle combination "open — up". The object of this verb was the concept
expressed by "his notes", which was accessible.
We need to give more thought to those cases where a single new concept
was expressed with a verb-object combination. Of course, those cases where
the object expressed accessible information, as in "would close his notes"
(30) or "walk out of the room" (31) are no problem for the one new concept
at a time hypothesis. In such cases it was only the verb which carried new
information. But what about cases like "makes a difference" (2) or "give a
lecture" (27), where both the verb and the object appear to have expressed
new information? To maintain the hypothesis, we need to suppose that in
those cases the verb-object combination as a whole expressed a unitary con­
cept. Often, in fact, the combination is clearly a lexicalized one. "Make a dif­
ference" is certainly a lexicalized expression, as is "give a lecture". It may
even be generally true that the verb-object construction is a way of presenting
an event as if it were conceptually unitary.
There were several cases here where a construction that expressed new
information contained a noun followed by one or more prepositions, and
where the object(s) of the preposition(s) expressed given or accessible con­
cepts. For example, in "there was no real interaction with the students" (28),
the unitary concept "interaction with" expressed new information, but "the
students" were given. A more complex case of this sort was "he left all of the
sort of real contact with students up to his assistants" (7). If there were one
intonation unit in this narrative which might be thought to violate the one new
concept at a time constraint, this would probably be the one. However, we
can easily interpret the concept expressed by the construction "leave contact
with — up to —" as unitary. It consisted of a verb-object "leave contact",
where "leave" was supplemented by the preposition sequence "up to", and
"contact" was supplemented by the preposition "with". If this construction
was conceptually unitary, there is no problem with the rest, since "students"
and "his assistants" both expressed accessible information.
36 WALLACE CHAFE

Another example of new information expressed in a long sequence of

words was "he was the most extreme example I ever had as a student" (37).
The conceptual essence of this intonation unit, however, was conveyed by
"was the — example". The expression "most extreme — I ever had"
functioned as a degree adverb modifying this concept, and was thus in the
same category as "precisely", "very", and "immediately", as discussed
above.
There seems, then, to be nothing in this narrative which contradicts the
hypothesized constraint: that no more than one concept can be changed from
the inactive to the active state during an initial pause, and that intonation
units, therefore, do not ordinarily express more than one "new" concept of
this kind. 3

Starting Points and Added Information

There are various cognitive factors that interact to determine the content
and shape of an intonation unit. So far we have considered how distributions
of, and changes in activation states can affect pronominalization and the
placement of intonation peaks, and how they can affect the amount of infor­
mation an intonation unit can contain. We come now to another factor, dif­
ferent from but related to the distribution of activation states. It has to do
with the manner in which newly recalled information is linked to other infor­
mation.
A speaker does not simply thrust concepts forward out of nowhere. The
usual technique for presenting information is to choose some concept, typi­
cally some referent, as a starting point and then to add information about it.
As a speaker proceeds to verbalize one focus of consciousness after another,
each added piece of information is attached to some other piece that is in
some sense already present. The linguistic manifestation of this formatting
strategy is the familiar subject-predicate structure. Many information units
conform to this structure, employing a subject to express the starting point
and proceeding with a predicate which adds information about that starting
point.
There are 23 starting points in this narrative, including such varied refe­
rents as what was just talked about (expressed by the "that" in 2), the speaker
himself (the "I" in 3), and "Wesleyan" in 20. Most of these referents enjoyed
the starting point status only once, although "every lecture", which first
appeared as a starting point in 16, was reselected as a starting point in 21 after
the digression concerning Wesleyan. The referent "the instructor", however,
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 37

appeared as a starting point 12 times, thus constituting about half the starting
points in the narrative. On that basis it is natural to regard the instructor as the
referent which this narrative was "mostly about".
How do starting points and added information relate to the three states
of activation discussed earlier? Certain things can be hypothesized about
both the starting point and the added information in this regard. Taking the
starting point first, we might call the following the "light starting point con­
straint":
(1) A starting point is usually a given referent.
(2) Occasionally a starting point is an accessible referent.
(3) A starting point is rarely a new referent, and then only at the begin­
ning of a major section of a discourse (a possibility not illustrated
in the excerpt we have been examining).
We can look back at the three-column division of the narrative into the
three activation states to see how this constraint is supported. In that list I put
those words or expressions which expressed starting points in parentheses. In
almost all cases, expressing a starting point coincided with being the subject
of a clause. The one exception was in 28, where the word "there", if it was
indeed syntactically a subject, did not convey a starting point:
28. .. a-nd .. there was no .. real interáction with t h e - .. students,
The function of the "there" construction is precisely to introduce a new refe­
rent, often one that will be used subsequently as a starting point. But an into­
nation unit containing this construction evidently does not itself contain any
starting point as such.
Of the 23 starting points in the narrative, 20 were given, three (4, 16, and
21) were accessible, and there were none that were new. It may be typical for
about one out of eight starting points to be accessible, rather than given.
But what about the added information? There seems to be a "heavy
added information constraint" which is the converse of the light starting point
constraint just described:
Added information typically contains one new concept, though it may
also contain some accessible concepts, or even some given concepts.
In the three-column list above it can be seen that every instance of added
information (the nonparenthesized material), with the exception of intona­
tion units 3 and 34, did contain one new concept. About half of the examples
of added information also contained accessible cqncepts, and in three cases
38 WALLACE CHAFE

(5, 6, and 28) there was some given information.

Intonation units 3 and 34 thus appear to be exceptions to the heavy
added information constraint, in that their added information contained no
information that was new. What is special about these two intonation units is
the fact that both were contrastive. In 3 the speaker was contrasting his own
recall with that of the previous speaker:
3. ... I can recall ... u h - ... a big undergraduate class that I had,
In 34 he was contrasting his statement about the "old world style1' with the
possibility of its negation:
34. . . I I guess that is the .. old world style,
Thus, we might add a qualification to the heavy added information constraint
to the effect that "added information need not contain a new concept if the
intonation unit expresses a contrast". We can suspect that the reason for this
qualification is that the expression of a contrast is itself cognitively costly, so
that a speaker cannot introduce both a contrast and a new concept simulta­
neously.

What Intonation Units Do

It will be interesting now to look at intonation units with the thought in
mind that they may perform varied functions in furthering the flow of infor­
mation in a discourse. The single most common type of intonation unit is the
kind that takes a starting point and adds information about it, in other words
an intonation unit that exhibits the classic subject-predicate structure. Eigh­
teen of the 40 intonation units in our example were of that kind; specifically
1-7, 9, 13, 19, 20, 27, 28, 30, 34, and 37-39. These intonation units were, that
is, "clauses". We will now see that most of the other intonation units were in
some way parasitic on, or ancillary to clauses. The clause appears to be the
prototypical intonation unit type, from which most other types are derived,
or are deviations.
There were some intonation units that were pieces of clauses. For exam­
ple, an intonation unit may express nothing but a starting point; that is, it may
consist of nothing more than a subject, lacking a predicate. Intonation unit 16
was of that sort:
16. and évery ... évery lécture,
This is a case where the starting point conveyed accessible information. In
such cases, where somewhat more cognitive effort is called for than would be
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 39

necessary just to maintain a given referent as the starting point, a speaker

may produce an accessible "starting point only" intonation unit like this one.
Unless the speaker has a change of plan, such an intonation unit will
sooner or later be followed by another which adds information about that
starting point; that is, which consists of nothing but a predicate. In the present
case this function was performed by 18:
18. .. started the same way.
A different kind of clause fragmentation apears when there is a verb of
saying whose object is a quotation, so that the speaker must focus separately,
first on the clause introducing the quotation, and then on the quotation itself:
21. ... So every lecture after the first would begin,
22. ... Gentlemen,
Because of the one recall at a time constraint, spoken language does not
easily support two or more conjoined items within a single intonation unit if
both items express concepts that are new. For that reason, the second part of
such a conjunction will appear in an intonation unit of its own, as in:
30. .. he would close his notes,
31. and walk out of the room.
I believe that this is a plausible cognitive explanation for the frequent occur­
rence of structures like this one, preferable to the usual explanation in terms
of a deleted subject and auxiliary in the second clause. Speakers are unlikely
to conjoin two verb-object combinations in a single intonation unit if both are
cognitively costly to produce.
Another common function of an intonation unit is to provide what can be
called an ''orientation" for a preceding or following clause. The orientation
may be temporal, spatial, or epistemic. That is, the information expressed in
an intonation unit may provide a time, a location, or an epistemic background
for the information in an adjacent clause. Such an intonation unit may take
the form of a prepositional phrase or of some other adverbial construction.
Examples of temporal orientations in our narrative included:
10. ... a—t.. uh— you know three or f
11. .. precisely one minute after the hour,
17. ... after the first,
23. .. ze last time,
29. .. and then .. a t . . at the end,
An example of a spatial orientation was:
40 WALLACE CHAFE

14. ... in the front of the room,

And of an epistemic orientation:
12. or something like that,
Other non-clausal intonation units fall into the category of "disfluen-
cies". Some are false starts: beginnings of clause structures that were then
reformulated, interrupted, postponed, or abandoned:
15. .. and he st (reformulated with a different starting point)
25. .. and then he would (interrupted by laughter)
33. ... And he was the (postponed to 37)
Others are afterthoughts: material added after a premature closure of some
kind:
34. ..I I guess that is the .. old world style,
(35. yéah.)
36. ... of lécturing.
One last kind of intonation unit which appeared in this narrative was an
expression of agreement of understanding, expressed either by the inter­
locutor:
8. ...Mhrm,
32. .. Hm.
35. Yéah.
or by the main speaker:
40. .. yéah.

Extended Clauses
We have seen that the intonation units in this narrative were of the following
types:
Clauses
Pieces of clauses (subjects alone, predicates alone)
Orientations for clauses (temporal, spatial, epistemic)
Clausal disfluencies (false starts, afterthoughts)
Expressions of agreement or understanding
Except for the last, all of these types are centered around the clause. When
an intonation unit is not itself a clause, it is likely to be either a piece of one,
an orientation for one, one that was falsely started, or a piece tacked onto a
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 41

prematurely finished one.

It is instructive to take the original version of the narrative, formatted as
a list of intonation units, and to reformat it by attaching those intonation units
which were not clauses to the clauses with which they were associated. I will
call the new units thus arrived at "extended clauses". They are as follows (re­
taining the original numbering):
1. ... It's funny though,
2. ... I dó think that makes a difference .. but,
3. ... I can recall... uh— ... a big undergraduate class that I had,
4. ... where .. éverybody lóved the instrúctor,
5. ... a-nd .. he was a ... real .. uh .. old world ... Swiss- ... guy,
6. .. this was uh .. a biology course,
7. ... a—nd h e - ... left all of the— sort of uh— ... real contact with stu­
dents .. up to .. his assistants.
8. (... Mhm,)
9-12. ... A - n d .. he would come into class,... a - t . . u h - you know three
or f.. precisely one minute after the hour, or something like that,
13-14. ... a—nd he— wou—ld .. immediately open his ... notes u p , . . . in the
front of the room,
15-18. .. and he st and every ... évery lécture, ... after the first, .. stárted
the same way.
19. This was .. u—m at Wésleyan,
20. when Wesleyan was still ... a mén's school.
21. ... So évery lecture after the first would begin,
22-24. ... Géntlemen, .. ze last time, ve vere tálking about,
25. .. and then he would
26. (laughter)
27. ... But then .. you know he would just.. give a lécture,
28. .. a-nd .. there was no .. real interáction with t h e - .. the students,
29-31. .. and then .. at.. at the énd, .. he would close his notes, and walk
out of the room.
32. (..Hm.)
33-36. ... And he was the . . I I guess that is the .. old world style, (Yéah.)
... of lécturing.
37. But he was the .. the most extreme exámple I had .. I ever had as
a stúdent.
38. (.. But he was interesting.)
39-40. And he was véry good, .. yéah.
42 WALLACE CHAFE

One thing that is interesting about this reformulation is the number of

extended clauses that began with a connective of some kind. By far the most
common connective was "and", which appeared at the beginning of 5,7,9-12,
13-14, 15-18, 25, 28, 29-31, 33-36, and 39-40; that is, in 10 out of the total of
22 extended clauses (including the false start). This proportion of about 50
percent appears to be typical for clause-initial "and1' in spoken English. Two
of the extended clauses (27 and 37) began with "but", as did the interlocutor's
contribution in 38. Clause 2 ended with "but", a pattern more common in
some other languages, but by no means absent in English. One clause (21)
began with "so". "And", "but", and "so" appear in general to be the three
most common connectives in spoken English. In the same category of parti­
cles which link an extended clause to what precedes, we can include subor­
dinating conjunctions like the "where" in 4 and the "when" in 20.
It is these extended clauses, then, which are the units speakers explicitly
fasten together with overt markers of connection. That in itself makes them
appear to be important building blocks of language. They are, as it were, the
subassemblies to which the overt fasteners of language are applied. Since
every extended clause is built around a subject-predicate construction, we are
led to suppose that that construction, or the starting point plus added infor­
mation format on which it is based, is crucial to language design. Clauses may
have their satellite orientations and disfluencies, but such other intonation
units remain just that: satellite to the basic subject-predicate construction.
Language thus consists in a basic sense of a series of predications. That is
perhaps a trivial and obvious thing to say, but its relevance to spoken lan­
guage can become clearer through studies of this kind.

Paragraphs
There is another level of segmentation which falls out of our narrative
quite naturally, and which it is natural to associate with a division into "para­
graphs". I have written elsewhere (Chafe 1980: 40-47) about the spoken
analogs of written paragraphs, and we have good examples before us here.
Within most narratives one finds certain places where the speaker pauses
longer than normally, where there is likely to be an increase in fumbling and
disfluency, and where an interlocutor is especially likely to contribute some
encouraging noise or remark. When one looks at the content of the narrative
at such places, one usually discovers a significant change in scene, time,
character configuration, event structure, and the like. I have suggested that
these are places in a discourse where the speaker has to make a major change
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 43

in his or her peripheral consciousness. If such changes are costly in terms of

cognitive effort, that explains the unusual amount of pausing and disfluency.
Within the framework of this paper, we can see these points as the places
in a narrative where the speaker introduces a major shift in the set of concepts
which are semi-active. Whereas configurations of active concepts change
rapidly from intonation unit to intonation unit, configurations of semi-active
concepts probably undergo major changes only at these more widely spaced
points. Just as the pauses between intonation units are the change points for
the active state of concepts, these greater disfluencies between paragraphs
appear to be the change points for the semi-active state.
There were just two places within the present narrative where the
speaker paused longer than otherwise, and where one of the interlocutors
injected little noises of encouragement ("mhm" and "hm"). Segmenting the
narrative on these grounds we are led to identify the following three major
segments:
1-7. ... It's funny though, ... I dó think that makes a difference ..
but, ... I can recall ... u h - ... a big undergraduate class that I had,
... where .. everybody loved the instrúctor, ... a-nd .. he was a ...
real .. uh .. old world ... Swiss- ... guy, ... this was uh .. a biology
course, ... a—nd h e - ... left all of t h e - sort of uh— ... real cóntact
with stúdents .. up to .. his assistants.
8. (...Mhm,)
9-31. ... A—nd .. he would come into class, ... a—t.. u h - you know
three or f .. precisely one minute after the hóur, or something like
that, ... a—nd he— .. wou--Id .. immediately open his ... notes up,
... in the front of the room, .. and he st and évery ... évery lécture,
... after the first, .. stárted the same way. This was ..u—mat Wés-
leyan, when Wesleyan was still... a mén's school. ... So évery lec­
ture after the first would begin, ... Géntlemen, .. ze last time, ve
vere tálking about, .. and then he would (laughter) ... But then ..
you know he would just.. give a lécture, .. a—nd .. there was no ..
real interáction with t h e - .. the students, .. and then .. a t . . at the
énd, .. he would close his notes, and walk out of the room.
32. (..Hm.)
33-40. ... And he was the . . I I guess that is the .. old world style,
(Yéah.) ... of lécturing. But he was the .. the most extreme exam­
ple I had .. I ever had as a stúdent. (.. But he was interesting.) And -
he was véry good, .. yéah.
44 WALLACE CHAFE

It is rewarding to look at the content of each of these paragraph-like

segments. We can notice that the first of them provided both a background
and a summary of the entire narrative. It supplied a background orientation
of the unexpectedness of what was to come, it introduced the biology class
and the instructor, and it told that the instructor was loved, at the same time
that he had no contact with his students. That seeming contradiction was the
story's point.
During the pause and the interlocutor's noise of encouragement that fol­
lowed, the speaker shifted his orientation from the generalities of the first
paragraph to a specific instantiation of the instructor's classroom behavior. A
shift from a general to a specific orientation necessitates a shift in the set of
concepts which are semi-activated, and that is evidently a process which takes
time.
The third paragraph wrapped things up with a final evaluation of the
instructor's old world style, paradoxically combined with his positive evalua­
tion. It required a return from the specific to the general stance, and again
that reorientation was time-consuming.
There is one other observation worth making with regard to this tripar­
tite division. We have here, it seems, the major elements of a certain narra­
tive schema:
(1) summary
(2) instantiation
(3) wrap-up
It is my suspicion that such patterns result, not from a top-down application
of a branching tree structure, but rather from a speaker's awareness of what
it is appropriate to verbalize as he or she proceeds through a story. Initially
this speaker knew he needed to orient his listeners by providing the kinds of
information included in his first, summary paragraph. Having done that, dur­
ing the following pause he settled on describing a typical class as a concrete
example. Having done that, during the next pause he decided it would be
appropriate to wrap things up as he did. In other words, a manifestation of a
story schema develops naturally out of the ongoing creation of a narrative in
real time, more than as the stuffing of recalled experience into a preformed
mold (cf. Chafe: 1979).
Spoken paragraphs, then, result from major shifts in a speaker's semi-
active consciousness, including shifts from a general to a specific orientation
and back again, as here. These shifts may be dictated by a schema which
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 45

guides the speaker in knowing what is appropriate to think of and verbalize

next. It is interesting that hearers recognize such shifts, as shown here by the
interlocutor's expressions of understanding. Undoubtedly there are clues for
the hearer in both the speech signal (Lehiste: 1979), and in the recognition of
major breaks in content.

Sentences
But what has happened to sentences in all of this? In jumping from
extended clauses to paragraphs we seem to have ignored a discourse unit of
a familiar and supposedly important kind. There is, to be sure, another way
of segmenting this narrative by dividing it at those points where the speaker
introduced a sentence-final falling pitch. We arrive in that way at the version
which follows:
1-7. ... It's funny though, ... I dó think that makes a difference ..
but, I can recall ... u h - ... a big undergraduate class that I had, ...
where .. éverybody loved the instrúctor,... a—nd .. he was a ... real
.. uh .. old world ... Swiss-- ... guy, .. this was uh .. a biòlogy
course, ... a-nd h e - ... left all of t h e - sort of u h - ... real còntact
with stúdents .. up to .. his assistants.
8. (...Mhm,)
9-18. ... A - n d .. he would come into class, ... a - t . . u h - you know
three or f .. precisely one minute after the hour, or something like
that, ... a—nd he— .. wou—ld .. immediately open his ... notes up,
... in the front of the room, .. and he st and évery ... évery lécturé,
... after the first, .. stárted the same way.
19-20. This was .. u-m at Wésleyan, when Wesleyan was still ... a
mén's school.
21-25. ... So évery lecture after the first would begin,... Géntlemen,
.. ze last time, ve vere talking about, .. and then he would
26. (laughter)
27-31. ... But then .. you know he would just.. give a lécture,.. a-nd
.. there was no .. real interáction with t h e - .. the students, .. and
then .. a t . . at the énd, .. he would close his notes, and walk out of
the room.
32. (..Hm.)
33-36. ... And he was the . . I I guess that is the .. old world style,
(Yéah.) ... of lécturing.
46 WALLACE CHAFE

37. But he was the .. the most extreme exámple I had .. I ever had
as a stúdent.
38. (.. But he was interesting.)
39-40. And he was véry good, .. yéah.
The function of sentences in spoken language is intriguing and prob­
lematic (cf. Chafe 1980: 20-29). There is a useful distinction to be made
between those linguistic units which are determined by basic cognitive
phenomena such as memory and consciousness, and those which result from
passing decisions regarding coherence and rhetorical effect. In the former,
cognitively determined category belong intonation units, extended clauses,
and paragraphs. Sentences, on the other hand, seem to belong to the category
of phenomena which are under more rhetorical control, and are more inde­
pendent of cognitive constraints.
One obvious property of spoken sentences, as defined by the falling
pitch intonation, is their variability in length. This property, clearly evident
in this narrative (compare the lengths of sentences in the list above), is one bit
of evidence that sentences are independent of cognitive limitations. There is
nothing that holds them to a particular size, as there is for intonation units.
It is interesting that the first sentence in our sample coincided exactly
with the first paragraph. If all sentences were like this one, there would be no
need to distinguish between sentences and paragraphs. However, the next
four sentences together constituted only a single paragraph. If we look at the
structure of what was being conveyed, it is clear how these four sentences
came to be separated. They expressed, in effect, the major constituents of the
second paragraph. First there was a sentence that reported the instructor
doing three things: entering the classroom, opening his notes, and beginning
the lecture. Then came a digression to make it clear that all this had happened
at Wesleyan. Then came the climax of the paragraph: the speaker's imitation
of what the instructor said at the beginning of each class. The paragraph
ended with a sentence describing how the instructor performed the very same
actions listed in the first sentence in reverse order: finishing the lecture, clos­
ing his notes, and leaving the classroom. The schema for the paragraph was
thus:
(1) the instructor's arrival
(2) a digression locating thé event at Wesleyan
(3) quoting the instructor
(4) the instructor's departure
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 47

Each of these four divisions ended with a sentence-final intonation. Thus, the
segmentation of this paragraph into sentences was neatly determined by the
paragraph's own internal schema.
Sentence 33-36 gave the speaker's encapsulated evaluation. It was
quickly followed by another idea: an attempt to put this instructor in perspec­
tive as an extreme case. Presumably the speaker would have stopped here,
except that the interlocutor wanted to reiterate the initial point that, in spite
of the instructor's style, he was a successful teacher. The speaker then added
a final sentence confirming this point.
It appears, then, that the sentences in this narrative were determined by
the speaker's rhetorical decisions as to how to present the content of the nar­
rative as it unfolded. Beginning with a long sentence that coincided with the
first paragraph, he then gave impact to the symmetric patterning of the sec­
ond, climactic paragraph by devoting one sentence to each of its major con­
stituents. He then finished off with a series of brief encapsulations of the
story's moral, aided in this by the interlocutor. This division of the narrative
into sentences, as signalled by the sentence-final intonation, had nothing to
do with the activation states of information, or with starting points and added
information, but was determined rather by the speaker's (no doubt intuitive)
decisions as to how most effectively to present what he was saying.

The Narrative
A few words ought to be said here at the end regarding the narrative as
a whole, as a coherent unit of discourse in itself. There are several kinds of
evidence that the excerpt we have been examining was a self-contained seg­
ment of the conversation. Most obviously, it was preceded and followed by
contributions from other speakers; it thus constituted a single "turn". It
ended with a significant pause, after which the wife of this speaker — the per­
son who had made the original point to which this narrative was a kind of
counterexample — added:
... I think you have to be an awfully good lécturer to make up for that
(laugh) ányhow.
thereby providing a closure to the frame within which this segment of the con­
versation had unfolded.
Underlying these external indications is the likehood that the main
speaker verbalized here what it is intriguing to regard as an island of memory.
It seems intuitively right to suppose that what is retrievable from our minds
48 WALLACE CHAFE

consists of a set of islands rather than a continuous record of past experience.

We remember only "a collection of moments" (Salaman: 1970), with vast
blank spaces in between.
It is of some interest that this speaker was telling, not of a singlé event,
but rather of an event type. Many old memories appear to be of this nature.
We remember kinds of things we used to do or used to have happen to us, not
particular instantiations of them. In looking through some materials of the
Berkeley Oral History Project a few years ago, I was struck with how few cases
there were in which one of the interviewees told about a single particular hap­
pening. The characteristic incident was of the type "my father used to ...," not
of the type "I remember the time my father ..." The narrative we have been
considering belongs to the same genre.

Summary
I have tried to present a picture of what was happening in the mind of this
speaker as he produced this narrative, and its effect on the linguistic form
which the narrative took. It is more rewarding, I think, to interpret a piece of
discourse in terms of cognitive processes dynamically unfolding through time
than to analyze it as a static string of words and sentences.
I began by pointing out how a piece of spoken language naturally seg­
ments itself into a series of initial pauses followed by periods of vocalization,
each such "intonation unit" expressing a single focus of the speaker's con­
sciousness. I then noted that the information conveyed by an intonation unit
is not indivisible, but contains within it ideas of objects, events, and proper­
ties, which I lumped together under the term "concepts". Basic to the picture
I tried to construct is the idea that a concept may be in any one of three states
of activation at a particular time: active, semi-active, or inactive. A speaker
normally makes changes in the activation states of certain concepts during the
initial pause, changes which determine the content and form of the following
intonation unit. The effect of an intonation unit on the hearer is to activate all
the concepts it contains, while others are deactivated.
If a speaker assumes, prior to uttering an intonation unit, that a concept
is already active in the hearer's mind, he will verbalize that concept in an
attenuated manner, giving it weak stress and probably pronominalizing it. I
showed which concepts in the narrative had this "given" property, and why.
I then described and exemplified the ways in which a concept may
become semi-active; specifically, though having been fully active at an earlier
point in the discourse or through having been evoked by a schema. I pointed
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 49

out how the "class'' schema evoked in this narrative had caused certain con­
cepts to be semi-active at certain points. I introduced the term "accessible"
for concepts which are recalled from the semi-active state.
Turning to inactive concepts, which I called "new", I pointed out how a
speaker is able to recall only one such concept per initial pause, and how that
normally means that an intonation unit will express no more than one new
concept. I called this the "one new concept at a time constraint".
I then discussed the organization of concepts within an intonation unit
into a referent that expresses a starting point, together with a concept or con­
cepts that add information about that starting point. I introduced the "light
starting point constraint", to the effect that a starting point is usually a given
referent, occasionally an accessible referent, but never a new referent. Î men­
tioned also a "heavy added information constraint", to the effect that added
information typically includes one new concept. Contrastve environments
provide exceptions to this constraint, since apparently an intonation unit can­
not easily express both a contrast and an item of new information.
Having shown that most intonation units are either clauses or satellite to
clauses, I reformulated the example narrative in terms of extended clauses
(clauses plus their satellites). That reformulation illuminated the fact that it is
these extended clauses that are fastened together with connectives, suggest­
ing that the extended clause is an important building block of language, either
equivalent to or an elaboration of the intonation unit.
I showed also how the narrative neatly divided itself into paragraphs
through pausal evidence as well as though the location of responses from the
interlocutor. These paragraphs represented major shifts in the speaker's
semi-active consciousness, and at the same time they manifested the major
schematic structure of the story.
I then reformulated the narrative in terms of sentences, as defined by fal­
ling pitches. I noted that sentences seem to be independent of the activation
factors so important to the other devices discussed in the paper. Instead, they
appear to have been determined by the speaker's decisions based on rhetori­
cal effect.
Finally, I pointed out that the entire narrative was readily isolable from
the rest of the conversation, and that it could be thought of as expressing an
island of memory. I mentioned that many such islands appear to be generic
rather than particular in nature, consisting of event types rather than indi­
vidual events.
I should point out in conclusion that I intended the particular narrative
50 WALLACE CHAFE

discussed here to be illustrative of some very general principles that apply to

all spontaneous spoken language. Of course it will take a lot of additional
work to demonstrate the universality of these principles, but I have seen
enough samples to feel comfortable in claiming that they are by no means
unique to this one case. On the other hand, they may be considerably mod­
ified in written language (and to a lesser extent in more formal kinds of spo­
ken language) by a greater freedom from ongoing cognitive constraints.
That, however, is another story.

NOTES
1) I will not deal here with "definiteness , \ another topic discussed in Chafe: 1976, since its func­
tion of expressing "identifiability" belongs to a somewhat different domain from the interaction
between consciousness and verbalization discussed here.
2) The pear stories research was sponsored by NIMH Grant MH25592, the work on spoken and
written language by NIE Grant NIE-G-80-0125.
3) One difference between written and spoken language is that writers have the opportunity to
dispense with this constraint, an opportunity which can and does lead to cognitive difficulties for
readers. My impression at the moment, however, is that writers adhere surprisingly closely to the
one new concept at a time constraint, although they verbalize accessible information significanlty
more often than speakers do.

REFERENCES

Bartlett, Frederic C. 1932. Remembering. Cambridge University Press.

Butterworth, Brian. 1975. "Hesitation and semantic planning in speech".
Journal of Psycholinguistic Research 4: 75-87.
Chafe, Wallace. 1973. "Language and memory". Language 49: 261-281.
. 1974. "Language and consciousness". Language 50: 111-133.
. 1976. "Givenness, contrastiveness, definiteness, subjects, topics, and
point of view". In: Li 1976.
. 1979. "The flow of thought and the flow of language". In: Givón 1979.
——. 1980. "The deployment of consciousness in the production of a narra­
tive". In: Chafe 1980.
Chafe, Wallace (Ed.). 1980. The pear stories: cognitive, cultural, and linguis­
tic aspects of narrative production. Norwood, NJ: Ablex.
. 1982. "Integration and involvement in speaking, writing, and oral liter­
ature". In: Tannen 1982.
 COGNITIVE CONSTRAINTS ON INFORMATION FLOW 51

. 1985. "Linguistic differences produced by differences between speaking

and writing". In: Olson, Hildyard, and Torrance 1985.
Chafe, Wallace, and Danielewicz, Jane. 1985. "Properties of spoken and
written discourse". In: Horowitz and Samuels 1985.
Clyne, Paul R., William F. Hanks, and Carol L. Hofbauer. (Eds.). 1979. The
elements: a parasession on linguistic units and levels. Chicago: Chicago
Linguistic Society.
Crystal, David. 1975. The English tone of voice. London: St. Martin.
Freedle, Roy O. (Ed.). 1979. Discourse processing: new directions. Nor­
wood, NJ: Ablex.
Givón, T. 1975. "Focus and the scope of assertion: some Bantu evidence".
Studies in African Linguistics 6: 185-205.
Givón T. (Ed.). 1979. Discourse and syntax. New York: Academic Press.
Horowitz, Rosalind, and S.J. Samuels. (Eds.). 1986. Comprehending oral
and written language. New York: Academic Press.
Lehiste, Ilse. 1979. "Sentence boundaries and paragraph boundaries — per­
ceptual evdidence". In: Clyne, Hanks, and Hofbauer 1979.
Li, Charles N. (Ed.). 1976. Subject and topic. New York: Academic Press.
Mandler, Jean M., and Nancy S. Johnson. 1977. "Remembrance of things
parsed". Cognitive Psychology 9: 111-151.
Miller, George A. 1956. "The magic number seven, plus or minus two".
Psychological Review 63: 81-97.
Olson, David, Hildyard, Angela, and Torrance, Nancy (Eds.), Literacy, lan­
guage, and learning. Cambridge: Cambridge University Press.
Pawley, Andrew, and Frances Hodgetts Syder. 1983. "Natural selection in
syntax: notes on adaptive variation and change in vernacular and literary
grammar". Journal of Pragmatics 7: 551-579.
Salaman, Esther. 1970. A collection of moments: a study of involuntary
memories. London: Longman.
Schank, Roger C , and Abelson, Robert P. 1977. Scripts, plans, goals, and
understanding. Hillsdale, NJ: Lawrence Erlbaum.
Tannen, Deborah. 1979. "What's in a frame? Surface evidence for underly­
ing expectations". In: Freedle 1979.
Tannen, Deborah. 1982. Spoken and written language: exploring orality and
literacy. Norwood, NJ: Ablex.
 TRANSITIVITY IN GRAMMAR AND COGNITION

SCOTT DELANCEY
Department of Linguistics
University of Oregon

1. Introduction
The purpose of this paper is to consider the significance to linguistic and
cognitive theory of the complex of semantic and morphosyntactic parameters
which Hopper and Thompson (1980) and Lakoff (1977) have shown to be
intimately connected with, if not to constitute, the phenomenon of transitiv­
ity. Both Lakoff and H&T, working with very different data and assump­
tions, suggest approximately the same list of transitivity parameters, and
point out that generally speaking a clause which has one of a short list of
semantic features will be morphosyntactically more like a canonical transitive
clause than an otherwise identical clause which lacks that feature. In Hopper
and Thompson's well-known list, the canonical transitive clause has two par­
ticipants, reports a kinetic event, is punctual and perfective, has a definite,
referential, individuated, and wholly affected patient and a volitional agent
which ranks high on the animacy hierarchy, and is affirmative and realis. All
available data support the inclusion of all of the listed parameters, though it
is probable that a few more will need to be added to the list. (I will suggest
below that the person of the agent and the evidential value of the clause are
also relevant).
Lakoff and Hopper and Thompson differ, however, on the correct direc­
tion in which to look for an explanation of the transitivity phenomenon.
Lakoff points out the parallel between this list of non-criterial features taken
as a definition of transitivity and the prototype approach to semantics
suggested by Rosch's work on categorization. Hopper and Thompson, on the
other hand, express some scepticism about the possibility of a semantic expla­
nation for the transitivity phenomenon. They point out, however, that trans­
itivity, as defined by these parameters, shows a high degree of correlation
54 SCOTT DELANCEY

with foregrounded information in narrative and procedural discourse, and

suggest a functional explanation according to which the transitivity complex
exists in order to function as a means of coding foregrounded information in
discourse. My purpose here is to argue for the viability of a semantic interpre­
tation of the transitivity phenomenon along the lines suggested by Lakoff,
and to suggest that it is easier to explain the discourse phenomena identified
by Hopper and Thompson in terms of a prototype semantic model than to
explain the semantic facts in terms of a discourse-functional theory of trans­
itivity.
I will further suggest, however, that it is shortsighted to attempt to sim­
ply invert Hopper and Thompson's argument and claim that some linguistic
semantic level can provide the explanation for the behavior of morphosyntax
in connected discourse. An extreme discourse functional theory of mor­
phosyntax must assert that the interpretation of a single sentence requires
that it be treated as though it were in discourse context. In actual language
use, of course, interpretation of any utterance is in fact based at least in con­
siderable part on real-world context; I will suggest that the use of discourse
context in interpreting sentences in connected discourse amounts to using the
discourse context as a substitute for real-world context. Thus we can argue
that the semantics of both clause- and discourse-level constructions are
rooted in a level of cognitive representation prior to either — that, rather
than being derivative one from another, both semantic and discourse-func­
tional facts are reflections of underlying cognitive schemata, the illumination
of which must be the ultimate goal of both semantics and discourse analysis.

2. Transitivity in discourse
There is no doubt that the facts upon which the discourse-functional
explanation for the transitivity phenomenon are based are correct — i.e. that
cross-linguistic morphosyntactic data demonstrate the universality of the
phenomenon, and that in general in narrative discourse there is a strong cor­
relation between transitive morphosyntax and foregrounded information.
Thus the important issue here is one of directionality of explanation. We must
decide, first, whether there is any alternative to the discourse-functional
explanation, and second, if there is, whether the alternative explanation can
predict the facts of discourse patterning of transitivity. To begin with, we
must note that while the statistical correlation of transitivity and foreground­
ing certainly seem to hold cross-linguistically and at least across those genres
which have been adequately investigated, it is quite clear that there is not a
 TRANSITIVITY IN GRAMMAR AND COGNITION 55

constant coding relation between any one of the transitivity parameters and
foregrounding. In the English data analyzed by Hopper and Thompson
(1980:288) we see the more transitive value for each parameter occurring in
a significant number of background clauses, and for all except mode and affir­
mation we see the more transitive value absent in a significant number of
foregrounded clauses. For example, Hopper and Thomspon report that while
88% of the foregrounded clauses in their sample showed the predicted per­
fective aspect, so did 27% of the background clauses. The average for all of
the transitivity parameters is 78% occurrence in foreground clauses, but a
huge 39% occurrence of high transitivity features in background clauses as
well. As Tomlin (1983ms) has pointed out, if there were a direct causal
relationship between foregrounding and the presence of any transitivity fea­
ture, the figures for that feature should approximate 100% and 0. (Tomlin's
paper further demonstrates the absence of a coding relationship between
aspect and foregrounding in unrehearsed on-line descriptive English narra­
tive).
In their discussion of the individual transitivity parameters, Hopper and
Thompson suggest a less direct relationship to discourse structure, identify­
ing the semantic distinction coded by a particular morphosyntactic distinc­
tion, and then suggesting why clauses characterized by the more transitive
value might be expected to constitute foreground information in a narrative,
They argue, for example, that agentivity and volition are associated with
foregrounding because "story lines are typically advanced by people who per­
form actions" (1980:286). We can elaborate their argument by applying the
same type of explanation to the transitivity complex as a whole, arguing that
the various transitivity parameters cohere in the way that they do because
they code aspects of a coherent semantic prototype, and that transitivity in
morphosyntax is associated with foregrounding in discourse because events
which approximate the transitive prototype are more likely to be of interest,
and thus inherently more likely to constitute foregrounded information.

3. Natural and discourse context

The general argument is based on the assumptions that human beings
categorize events according to various relevant features, and that particular
morphosyntactic constructions code particular event categories in precisely
the same way that lexical items code other cognitive categories. Thus, as
suggested first by Lakoff, prototypical transitive morphosyntax codes events
which are close to a semantic transitive morphosyntax. (For further discus-
56 SCOTT DELANCEY

sion and supporting data see DeLancey 1984a, 1984b, 1985c). If we find that
transitive clauses tend to occur as foreground in discourse, this reflects the
cognitive salience of the event type coded by such clauses. It is therefore not
the case that the morphosyntactic expression of transitivity is informed by its
discourse function, nor that the discourse patterning can somehow be
explained in terms of linguistic aspects of the clause type, but rather both
directly reflect aspects of cognitive categorization.
The nature of the argument can be illustrated with what I think is an
uncontroversial example. In most languages the simplest motion verbs either
are lexically or can be morphologically specified for deictic orientation. In
English, for example, the difference between the motion verbs come and go
is that the former typically describes a motional event involving motion
toward the location of the speech act, or some other location associated with
the speaker or hearer, while go describes motion oriented in any other direc­
tion (cf. DeLancey 1981). Thus the basic difference between come and go,
and their analogues in other languages, has to do with the actual context of
the speech act, i.e. with an actual spatial relationship between the location of
the speech act and the event being described. Now, in many languages (and,
I suspect, in all languages where this deictic distinction is marked) 'come' and
'go' have a discourse function in which they are used to indicate the relative
importance of various locations in a narrative. For example, an English bal­
lad begins:
(1) Earl Richard is a-hunting gone.
The second stanza begins:
(2) He rode till he came to his lady's house.
and it is at the latter location that the rest of the action of the ballad takes
place. The use of go and come here marks locus of action; a location to which
the protagonist comes is one where the audience's attention should remain
for a while, as significant action will transpire there. (Detailed and very infor­
mative discussion of this function of 'come' and 'go' in two Tai languages can
be found in Mundhenk 1967 and Bickner 1978ms). In discussing the seman­
tics of the motion verbs one can posit a "general meaning" of 'come' some­
thing like 'motion toward some contextually recoverable center of attention',
applying by default to situations where the location of speaker and hearer
counts automatically as such a center. Certainly this would be an appropriate
dictionary definition, as it encompasses the various important uses of the
word, It is intuitively clear, however, that the fundamental meaning of 'come'
 TRANSITIVITY IN GRAMMAR AND COGNITION 57

is 'motion toward here, where you and I are right now', and that the use of
deictic specification in discourse where it does not refer to the speech situa­
tion is a secondary elaboration of this primary function. The use of 'come' (or
'here', etc.) in narrative to establish a center of attention should be seen as an
instruction to the hearer to accept the specified location as his own point of
view from which to metaphorically watch the further unfolding of the story.

4. Transitivity as a semantic prototype

Lakoff (1977) argues from English data for the existence of a semantic
prototype for the category of events coded by transitive clauses which is
characterized by a set of parameters substantially paralleling those identified
by Hopper and Thompson. The cross-linguistic evidence assembled by Hop­
per and Thompson for the transitivity complex suggests that this prototype
(or some prototype of which it is a subcategory) is universal in language. In
DeLancey 1984a I have demonstrated the utility of such an analysis in
describing the morphosyntax of Lhasa Tibetan. Briefly, the argument con­
cerns the problem of characterizing the distribution of ergative case in Lhasa.
In perfective clauses, ergative case marks the subject of most two-argument
clauses, and of single-argument clauses with volitional actors, as in:
(3) has dkaryol bcag-pa-yin
I-ERG cup break-PERF/VOLITIONAL
T broke the cup (deliberately).'
(4) na s dkaryol bcag-soh
I-ERG cup break-PERF/NONVOLITIONAL
T broke the cup (inadvertently).'
(5) has dkaryol mthoh-byuh
I-ERG cup see-PERF/GOAL ACTOR
T saw the cup.'
(6) has hus-pa-yin
I-ERG cry-PERF/VOLITIONAL
T cried.'
(7) ha si-byuh
I die-PERF/GO AL ACTOR
I died.'
From just these few examples it is clear that it is impossible to describe the dis­
tribution of ergative case marking in terms of a single syntactic or semantic
feature of the clause. Either a volitional actor or the presence of two argu-
58 SCOTT DELANCEY

ments is sufficient in a perfective clause, but neither is necessary. If we make

the assumption that grammatical morphemes and syntactic constructions
have meaning, then the meaning of Lhasa ergative case must be an example
of Fillmore's (1982) CLIMB-type prototype meaning, in which the best
exemplars of the category have both defining features, but the presence of
either feature alone is sufficient to qualify an instance as a member of the cat­
egory.
The picture is further complicated by the pattern of case marking in non-
perfective clauses. With imperfective and future verb endings, the ergative
marking of the subjects of two-argument clauses is optional, which is consis­
tent with Hopper and Thompson's observation that non-perfective clauses
are less highly transitive than perfective clauses. Single argument clauses can­
not have an ergative argument unless they are perfective; which shows that
while both volitionality and transitivity in the traditional sense are among the
defining features of the category coded by ergative case, the presence of two
arguments is the stronger determinant, since the presence of two arguments
is itself sufficient to qualify a clause as transitive, while volitionality alone is
not sufficient without the support of either perfective or the presence of two
arguments.
It is clear, then, that a semantic characterization of Lhasa ergative mark­
ing is possible only in a prototype framework, in which the category is defined
by a bundle of differentially-weighted features, no single one of which is crite­
rial. The more of the relevant features that occur in particular instance, the
better an instance it is of the category, but an instance can lack many of the
defining features and still count as an exemplar of the category of transitive
event. And, since the features which characterize the category are drawn
from Hopper and Thompson's list of universal transitivity features (in fact
more of Hopper and Thompson's features are relevant in Lhasa than I have
discussed here; see DeLancey 1984a for further details) we have some basis
for inferring the universality of this semantic prototype. The next problem is
to find the explanation for the existence of such a universal category.
The cluster of attributes associated with transitivity define a semantic
construct which approximates the notion of EVENT as opposed to STATE.
The extreme version of the discourse-functional hypothesis I would interpret
as asserting that these are purely discourse-based notions, with no indepen­
dent semantic content; the somewhat less extreme version suggested by Hop­
per and Thompson is that only plausible reason for the grammatical promi­
nence of this distinction is the discourse function of treating foreground mate-
 TRANSITIVITY IN GRAMMAR AND COGNITION 59

rial as a sequence of events, and background material as more stative. I would

argue instead that hypotheses about the cognitive process of category forma­
tion suggested by Rosch (e.g. 1978) and others would predict just such a
semantic distinction. Rosch argues that cognitively optimal categories are
maximally distinct from one another, and suggests that the process of cate­
gory formation may operate in such a way as to maximize such distinctive­
ness:
[T]he more prototypical of a category a member is rated, the more attributes
it has in common with other members of the category and the fewer attri­
butes in common with members of the contrasting categories. ... [F]or natu­
ral language categories... the extent to which items have attributes common
to the category was highly negatively correlated with the extent to which they
have attributes belonging to members of contrasting categories... [S]uch
structure may be a result of the human tendency once a contrast exists to
define attributes for contrasting categories so that the categories will be max­
imally distinctive. (Rosch 1978:37)
Thus, given that human beings do categorize types of interaction of entities
in the outside world, we would expect the categories which develop to cluster
around prototypes which are maximally distinct from one another. We might,
for example, anticipate the development of a category of events the prototyp­
ical exemplar of which is a kinetic event in which a human actor, acting in the
most specifically human possible way, that is, volitionally, causes some per­
ceptible change of state in a perceptible object, which would involve contrast
with another category the prototype of which would involve no actors and no
change in anything — typified perhaps by environmental statements such as
it's cold. And, once such a categorial distinction was established, Rosch's
hypothesis would predict that any further attributes which might be
associated with either category would be those which would further maximize
their distinctiveness:
[T]he same laws of cognitive economy leading to the push toward basic-level
categories and prototypes might also lead to the definition of attributes of
categories such that the categories once given would appear maximally dis­
tinctive from one another and such that the more prototypical items would
appear even more representative of their own and less representative of con-
trastive categories. Actually, in the evolution of the meaning of terms in lan­
guages, probably both the constraint of real-world factors and the construc­
tion and reconstruction of attributes are continually present. Thus, given a
particular category system, attributes are defined such as to make the system
appear as logical and economical as possible. (1978:42)
60 SCOTT DELANCEY

Thus we might expect, for example, the association of perfectivity with the
transitivity prototype, since ongoing processes are more like states than are
completed events; similarly, transitive events with individuated patients are
more time-bounded and thus more event-like than events involving multiple,
seriated, or mass objects.

5. The experiential basis of the transitivity prototype

In the actual investigation of natural (as opposed to experimentally
induced) categories carried out by Rosch and her co-workers, it turns out that
the maximization of distinctiveness is not simply a result of the nature of the
cognitive process of category formation, but rather seems to reflect the struc­
ture of the real world:
The second principle of categorization asserts that unlike the sets of stimuli
used in traditional laboratory-concept attainment tasks, the perceived world
is not an unstructured total set of equiprobable co-occurring attributes.
Rather, the material objects of the world are perceived to possess high corre­
lational structure...combinations of what we perceive as the attirbutes of
real objects do not occur uniformly. Some pairs, triples, etc., are quite prob­
able, appearing in combination sometimes with one, sometimes another
attribute ; others are rare ; others logically cannot or empirically do not occur.
(1978:29)
In other words, in categorizing natural objects it often is not necessary to
select out those attributes which will form nice maximally distinct clusters, for
the available — or at least the most salient — attributes cluster naturally. The
fact that dogs and pine trees have relatively few attributes in common is a fact
of biological nature, and not merely something imposed by human categori­
zation; thus we can predict that most languages will have a category which
includes dogs and cats but not pine trees, while few if any will have a category
which includes dogs and pine trees but not cats.
Of course, the categories labelled by morphosyntactic constructions of
the sort we are discussing here are not categories of material objects, but of
event schemata; nevertheless the universality of the transitive prototype
shows that it is somehow an extremely natural category, and not simply a cog-
nitively economical one. I think it can be shown that the natural basis for the
transitivity prototype is the universal human understanding of the physical
fact that events have causes, i.e. that the basis of the transitivity prototype is
a simple CAUSE > EFFECT schema which owes its universality to its
universal utility in dealing with the real world. (Cf. Lakoff and Johnson 1980,
 TRANSITIVITY IN GRAMMAR AND COGNITION 61

Lakoff 1982 for other discussions of the prototype nature of the linguistic and
conceptual schema for causation, and DeLancey 1984b, 1984c, 1985a, 1985b,
1985c, for further discussion of its structure).

6. Deviations from the prototype

The event schema represented by the prototypical transitive clause can
be analyzed as a sequence of two events: a volitional act on the part of the
agent, and a subsequent and consequent change of state on the part of the
patient. This causation schema, which has been widely discussed in the lin­
guistic and philosophical literature, underlies a great many semantic/mor-
phosyntactic categories. One class of deviation from this prototype is that in
which these two events are less than optimally distinct. Thus, for example,
reflexives are less transitive than ordinary events, as is reflected in the partial
or complete syncretism of the expression of reflexivization and detransitiviza-
tion in many Indo-European and many other languages (cf. Hopper and
Thompson 1980:277-8). Single-argument clauses with volitional actors, such
as She jumped, are even less transitive because in such events the CAUSE
and EFFECT are not perceptually distinct.
More complicated classes of deviation from the prototype are those in
which either the CAUSE or the EFFECT event is not fully accessible to an
observer. This is at least part of the reason why imperfective clauses count as
deviations from prototypical transitivity (the case of partitive objects can
probably be considered here as well); for the event which is described in an
imperfective clause is one in which the change of state of the patient has yet
to be accomplished. Here we also find the reason for the inclusion of such
characteristics of the agent as volition and animacy. To take an extreme
example, we may consider the fact, documented and discussed in detail in
DeLancey 1984b, that in some (and probably many) languages, events
involving disease tend to be coded as deviations from prototypical transitiv­
ity, so that in English, for example, (8) is a much more natural sentence than
(9) or (10), and the construction exemplified in (8) occurs only in connection
with a small class of events involving drastic effects of physical or emotional
states:
(8) Somebody/my brother died of pneumonia
(9) ?Somebody/my brother was killed by pneumonia.
(10) ?Pneumonia killed somebody/my brother.
The nature of the deviation is the imperceptibihty of the CAUSE event; only
62 SCOTT DELANCEY

the effect on the patient is directly accessible to an observer. I would suggest

that such event schemata are difficult to categorize (and hence are coded with
unusual morphosyntax) because their cause is not directly perceptible (De-
Lancey 1984b).
I have argued elsewhere (1981, 1984c, 1985a,b,c; cf. also Givón
1979:334-44) that it is here that volitionality enters into the prototype, in that
volitional acts can be traced to an ultimate cause — the "act" of volition —
while non-volitional acts cannot. Thus non-volitional acts, like events involv­
ing disease and similar causes, represent defective instances of the CAUSE -
—> EFFECT schema, and hence deviations from the cognitively more cohe­
rent prototype. Following out the connections between volitionality and evi­
dentially as semantic categories noted in DeLancey 1981, 1985a,b, we might
expect to find evidence that the category of evidentiality is also connected to
the universal transitive prototype; this suggestion of course awaits empirical
verification (See Watanabe 1984 for some suggestive Japanese data).

7. Transitivity, volitionality, and person

This interpretation of volitionality is also closely interwined with another
transitivity parameter not discussed by either Hopper and Thompson or
Lakoff — that of the person of the agent. Hopper and Thompson point out
the support given their analysis of aspect and its place in the transitivity com­
plex by the well-known split ergative case marking pattern in which perfective
clauses have ergative case marking, and imperfective clauses accusative. In
languages exhibiting this pattern we will obviously want to interpret ergative
case as an index of high transitivity; from our earlier discussion of Lhasa Tibe­
tan ergative marking we can see that at least in that language this interpreta­
tion is quite consistent with other facts about the distribution of ergative
marking.
There is, however, another well-known split ergative pattern in which
first and second person agents (hereafter SAP agents, for Speech Act Partic­
ipant) are in nominative/absolutive case, while third person agents are
marked as ergative. Analogizing from our earlier argument and the sugges­
tions of Hopper and Thompson, we would tend to interpret this as evidence
that clauses with SAP agents are less transitive than those with non-SAP
agents. This does not seem particularly plausible on the face of it, and there
is little or no other evidence for such a conclusion, and considerable evidence
for the opposite claim. In at least some languages with direct/inverse marking
systems or analogues thereof (DeLancey 1981 ), inverse forms — i.e. verbs in
 TRANSITIVITY IN GRAMMAR AND COGNITION 63

transitive clauses with SAP patient and non-SAP agent — are morphologi­
cally explicitly marked as lower in transitivity than direct forms (i.e. with SAP
agent and non-SAP patient). A clear example is the Coast Salish language
Lummi, discussed by Jelinek and Demers (1983; similar patterns occur in
some other Salish languages). In Lummi, direct clauses must be in what
Jelinek and Demers analyze as active voice, while inverse clauses are
obligatorily in a form which they label passive. That is, there is no "passive"
counterpart to (11) or (12), and no "active" counterpart to (13) or (14):

know-TRANS-lst the man

T know the man.'

know-TRANS-2nd the man

'You know the man;

know-TRANS-INTER-lst by the man

T am known by the man. 1

know-TRANS-INTER-2nd by the man

'You are known by the man.'
What is particularly relevant to our present concerns is the fact that the n mor­
pheme which characterizes the passive or inverse forms is explicitly identifi­
able as a detransitivizing morpheme, so that (13) and (14) are morphologi­
cally marked as lower in transitivity than (11) and (12). Obviously, then, the
agent marking preposition must be considered an index of lowered transitiv­
ity, like the oblique agent marking in other voice alternations such as that of
.English. Analogizing from this pattern to the SAP/3rd split ergative pattern
(arguments for this analogy are given in Silverstein 1976, Heath 1976, Dixon
1979, and DeLancey 1981), we can conclude that there too the presence of
ergative case is an index of lower rather than of higher transitivity.
We now require an explanation for the apparent fact that a non-SAP
agent represents a deviation from the transitive prototype. At least a partial
answer is suggested by the pattern of volitionality marking in Lhasa Tibetan,
where the morphosyntactic indication of volitionality is restricted to clauses
with first person participants (Jin 1979; DeLancey 1985a,b) similar patterns
occur in some other Tibetan dialects (Schöttelndreyer 1980) and in the
Tibeto-Burman language Newari (Hale 1980, Genetti 1985ms)):
64 SCOTT DELANCEY

(15) nas deb der bzag-pa-yin

I-ERG book there put-PERF/VOLITIONAL
'I put the book there.'
(16) has deb brlagssoh
I-ERG book lose-PERF
'I lost the book.'
(17) khos deb der bzagsoh
he-ERG book there put-PERF
'He put the book there.'
(18) *khos deb der bzag-pa-yin
he-ERG book there put-PERF/VOLITIONAL
I have demonstrated elsewhere (DeLancey 1985, 1986) that this distinction is
in fact an integral part of the Lhasa evidentiality system, and the reason for its
restriction to clauses with first person actors is that one can have direct know­
ledge of the volitionality or non-volitionality only of one's own acts; volition-
ality is a mental state not observable by others. (I suspect that more cross-lin­
guistic evidence of this sort can be found than I have in hand; cf. the active-
type Caucasian language Bats, in which the active case marking pattern —
i.e. the formal indication of control — occurs only with SAP subjects. The rel­
evant data are discussed in DeLancey 1981). Thus the complete transitive
prototype can only be clearly perceived in the subjective observation of one's
own actions; the attribution of volitionality, a crucial parameter of the pro­
totypical transitive event, can only be inferred elsewhere. (The ambiguous
position of second person in this hypothesis, and in the available data, merits
further consideration; some preliminary discussion can be found in DeLan­
cey 1981. It is interesting to note that in Tibetan and Newari volitionality is
distinguished for second person actors in direct yes/no questions).

8. Syntax and discourse

While the preceding argument certainly leaves large gaps, which I hope
will prove tempting to other investigators, I think it is clear that a cognitively-
based semantic account of the transitivity complex is feasible, and can offer
a plausible explanation for a wide range of cross-linguistic morphosyntactic
tendencies. Our suggested basis for the semantic prototype also should satisfy
Hopper and Thompson's demand for
a satisfactory answer to the obvious ... question: what is so important about
Transitivity, and why does it figure so prominently in the grammars of lan­
guage after language? (1980:280)
 TRANSITIVITY IN GRAMMAR AND COGNITION 65

However, it remains to be shown how this hypothesis can account for the cor­
relation between transitivity and foregrounding in discourse. At this point an
answer to this problem can only be suggestive and programmatic, although
there is increasing reason to believe that the empirical investigation of actual
discourse production will soon make it possible to adduce evidence for or
against (undoubtedly for) the suggestions made here. (See, for example,
Tomlin's paper in this volume for an illustration of techniques for investigat­
ing the psychological basis for event structure in discourse).
We can take the basic instance of narrative discourse (at least logically,
and no doubt also ontogenetically and phylogenetically) to be the eyewitness
account, i.e. a narration of an event or series of events which the speaker has
observed first-hand, and the basic purpose of such a narration to be to
recreate in the hearer's mind something like the speaker's mental representa­
tion of the event. (Cf. Goffman 1974:504ff. for further discussion of this
genre of narrative). A good example is a sportscaster's play-by-play commen­
tary (cf. Tomlin 1983), especially when intended for a radio audience. Any
other type of narrative — fictional, speculative, dishonest, or whatever — is
just a step or two further removed from reality. Any type of narrative dis­
course is (perhaps among other things) an attempt to present a series of
events as though it actually happened, and to induce in the hearer's/reader's
mind a representation which simulates the representation which would be
created by actually perceiving the events in reality.
Now, considering narrative as a recreation of actual experience, we can
hypothesize that those events will be foregrounded which in actual percep­
tion of the event were perceptually or psychologically most salient — which
is to say that, in telling a true story, one concentrates on the most interesting
parts. Thus, in considering such reality-based narrative, we can rephrase a
number of Hopper and Thompson's statements about the discourse function
of various transitivity parameters as psychological rather than purely linguis­
tic claims. For example, in explaining the discourse relevance of the kinesis
feature, Hopper and Thompson point out that "Foregrounded clauses typi­
cally narrate events ... It follows that the verb in a foregrounded clause is nor­
mally one which signals such a change". The probable reason for this correla­
tion is that kinetic events, involving changes of place or condition, are percep­
tually and cognitively highly salient; this ensures their prominence in the per­
ceptual and cognitive analysis and representation of a series of events, and
this in turn guarantees them foreground status in a report of the event. Cod­
ing such events as salient amounts to telling the hearer that if he had seen the
action he too would have found these
the most salient events, and that he
66 SCOTT DELANCEY

should so consider them in building his own mental representation which the
speaker's narrative is intended to help him create.
By this line of argument, arguments such as Hopper and Thompson's for
the discourse prominence of highly transitive clauses are in fact arguments for
the psychological salience of events close to the cognitive prototype which is
coded by transitive morphosyntax. It is this notion of psychological salience
which will explain the behavior of grammatical transitivity in discourse. Our
ultimate goal, in the study of discourse as well as of semantics and mor­
phosyntax, should be to find explanations for our data at an extralinguistic
cognitive level. The reason why it turns out to be difficult to explain the dis­
course patterning of morphosyntax in semantic terms, or to explain mor­
phosyntax in terms of discourse function, is that there is no direct causal
relationship in either direction. Rather, both are informed by cognitive and
perceptual structures. We may hypothesize that, at least to a considerable
extent, the semantics and discourse behavior of a given morphosyntactic con­
struction reflect the same underlying cognitive representation, so that the
results of the study of discourse and of semantics should tend to converge;
comparison of the prototype interpretation of transitivity developed here and
by Lakoff with the discourse-functional argument of Hopper and Thompson
will show an example of such convergence.

REFERENCES

Bickner, Robert. 1978ms. Directional modification in Thai fiction: The use

of "come" and "go" in text building. Ms., University of Wisconsin.
DeLancey, Scott. 1981. "An interpretation of split ergativity and related pat­
terns." Language 57.626-57'.
. 1984a. "Transitivity and ergative case in Lhasa Tibetan." Proc. of the
Tenth Annual Meeting of the Berkeley Linguistics Society, pp. 54-63.
. 1984b. Notes on agentivity and causation." Studies in Language 8. 181-
213.
. 1984c. "Categoriesof non-volitional actorin Lhasa Tibetan. "In A. Zide
et al., ed., Proc. of the Conference on Participant Roles: South Asia and
Adjacent Areas. Bloomington: IULC.
. 1985. "Lhasa Tibetan evidential and the semantics of causation."
Proc. of the Eleventh Annual Meeting of the Berkeley Linguistic Society.
. 1986b. "Evidentiality and volitionality in Tibetan." In W. Chafe and J.
 TRANSITIVITY IN GRAMMAR AND COGNITION 67

Nichols, ed., Evidentiality : The Linguistic Coding of Epistemology. Ablex.

. 1985c. "On active typology and the nature of agentivity." In F. Plank,
ed., Typology of Language Structures: Nominatival, Ergatival, Activai.
Mouton.
Dixon, R.M.W. ed. 1976. Grammatical Categories in Australian Languages.
Canberra: Australian Inst. of Aboriginal Studies.
-----. 1979. "Ergativity." Language 55.59-138.
Fillmore, Charles. 1982. "Towards a descriptive framework for spatial
deixis." In R. Jarvella and E. Klein, eds., Speech, Place, and Action. Lon­
don; Wiley.
Genetti, Carol. 1985ms. Grammatical relations in Newari. Ms, University of
Oregon.
Givón, T. 1979. On Understanding Grammar. NY: Academic Press.
Goffman, Erving. 1974. Frame Analysis. Harper.
Hale, Austin. 1980. "Person markers, finite conjunct and disjunct verb forms
in Newari." In R. Trail, ed., pp. 79-106.
Heath, Jeffrey. 1976. "Substantival hierarchies: Addendum to Silverstein."
In Dixon, ed., pp. 172-90.
Hopper, Paul, and S.A. Thompson. 1980. "Transitivity in grammar and dis­
course." Language 56.251-99.
Jelinek, Eloise, and R.A. Demers. 1983. "The Agent Hierarchy and voice in
some Coast Salish languages." UAL 49.167-85.
Jin, Peng. 1979. "Lun Zang-yu Lasa-ko dungci-de tedian yu yufa jiegou-de
guanxi." ("On the relations between the characteristics of the verb and
syntactic structure in spoken Lhasa Tibetan"). Minzu Yuwen 3.173-81.
Lakoff, George. 1977. "Linguistic gestalts." Proc. of the Thirteenth Annual
Meeting of the Chicago Linguistic Society.
. 1982. "Categories and cognitive models." Berkeley Cognitive Science
Report no. 2.
. and M. Johnson. Metaphors We Live By. Chicago: University of
Chicago Press.
Mundhenk, Norman. 1967. Auxiliary verbs in Myang of Northern Thai­
land. {Hartford Studies in Linguistics no. 22). Hartford: Hartford Semi­
nary Foundation.
Rosch, Eleanor. 1978. "Principles of categorization." In E. Rosch and B.
Lloyd, eds., Cognition and Categorization, pp. 27-48. New York: John
Wiley.
Schöttelndryer, B. 1980. "Person markers in Sherpa." In R. Trail, ed., pp.
125-30.
68 SCOTT DELANCEY

Silverstein, Michael. 1976. "Hierarchy of features and ergativity." In Dixon,

ed., pp. 112-71.
Tomlin, Russell. 1983. "On the interaction of syntactic subject, thematic
information, and agent in English." J. of Pragmatics 7.411-432.
——. 1983ms. The identification of foreground-background information in
on-line descriptive discourse. Ms., University of Oregon.
Trail, Ronald, ed, 1980. Papers in South-East Asian Linguistics no. 7. (Pacific
Linguistics Series A, no. 53). Canberra: Dept. of Linguistics, Australian
National University.
Watanabe, Yasuko. 1984. "Transitivity and evidentially in Japanese."
Studies in Language 8.235-51.
 ACTIONS AND PROCEDURAL DIRECTIONS

PETER DIXON
University of AIberta

1. Introduction
Procedural directions have a unique property among discourse types:
people often act on them to accomplish some task. In keeping with this prop­
erty, I will argue in this paper that the appropriate unit of analysis for direc­
tions is the action, rather than the clause, proposition, or event. This conclu­
sion was reached primarily on the basis of psychological research on the com­
prehension process and psychological theories of mental representation.
However, despite these psychological origins, the approach taken here seems
to have some implications for the structure of the directions themselves, and
thus touches on a number of linguistic concerns. In this paper, I will describe
and test a model of how directions are comprehended, and will then consider
how the model is relevant to the structure of the directions.
The first section will discuss a general framework I have used for study­
ing directions. The central point of that framework is that directions are
understood by constructing a mental representation, or plan, for performing
a task. The next section will describe some current psychological notions
about how such mental plans might be organized. When combined with a
model of the comprehension process, the notions make some clear predic­
tions about how directions will be processed. These predictions are then
tested by looking at the patterns of reading times for different forms of a
direction. Finally, I will conjecture about some of the linguistic properties of
directions based on the reading time results.

2. A Framework for Studying Directions

In this section I will describe why the structure of mental plans seems to
be of critical importance in the study of procedural directions. The domain of
interest here is the use of written directions for performing a given task.
70 PETER DIXON

Directions may be presented in forms other than written text (e.g., with
speech or with pictograms). Although analogous processes may occur with
direction forms other than written (cf. Marcel & Barnard, 1979), I will limit
this discussion to the written form only. It is also possible for directions to be
used for purposes other than performing a task. For instance, they could be
used to help understand how a particular device works, they could be used to
find out a specific fact (e.g., an ingredient in a recipe), or they could be read
simply for general interest. However, I am concerned only with the case
where the reader intends to perform a particular task on the basis of the direc­
tions. This intent should have an effect on the nature of the internal represen­
tation constructed from the directions. Specifically, the reader must be sure
to construct a representation that would be adequate for performing the task.
This is not necessarily the case when the same sentences are read for other
purposes. For instance, Dixon (1984) found that the comprehension process
was quite different when subjects merely had to recall directions instead of
carrying them out.
When directions are used to perform a task, their mental representation
can be thought of as a plan for carrying out that task. Although other rep­
resentations might be constructed as well, a mental plan must be present in
some form if one is to do the task. That is, the fact that one can read a set of
directions, put them down, and then perform a task correctly indicates that a
mental representation of those directions must have existed and that that rep­
resentation was adequate for performing the task. Such a representation will
be referred to as a mental plan for the task.
The relationship between a set of directions and its mental plan may be
somewhat different than the one usually assumed to exist between a text and
its mental representation. It is often assumed that the representation gener­
ally reflects the information present in the text, rather than the way in which
the representation may eventually be used. For instance, the distinction
between foreground and background information in discourse is motivated
by the assumption that this distinction affects the mental representation (e.g.,
Hopper & Thompson, 1980). That is, a piece of information will be rep­
resented as either foreground or background depending on how it is stated in
the text. Similarly, Omanson (1982), Kintsch and van Dijk (1978), and Meyer
(1975) present methods of analyzing a text to infer the mental representation
that should be acquired from reading it. Although these authors do consider
how the mental representation might vary with the readers' intentions and
goals, it is still the case that the structure and content of the representation is
 ACTIONS AND PROCEDURAL DIRECTIONS 71

determined primarily by the structure and content of the text. Thus, the main
criterion for the appropriateness of a mental representation would be the dis­
course itself.
In contrast to other dicourse forms, the appropriateness of a mental plan
is determined primarily by the task, not by the directions from which it was
constructed. That is, a plan is appropriate if it allows one to perform the task
correctly and efficiently. If the task cannot be done on basis of the plan, then
the plan is inappropriate, and the reader cannot really be said to have under­
stood the directions. While the mental representation of other forms of dis­
course may reflect the structure and content of the discourse, mental plans
are constrained first of all by what has to be accomplished. This follows from
the assumption that understanding a set of directions involves constructing a
plan that is adequate to perform the task.
The ultimate measure of understanding would then be whether the read­
ers of the directions can perform the task or not. If they can, then it seems
reasonable to suppose that they had possessed an appropriate mental plan
and had understood the directions. On the other hand, failing to perform the
task does not necessarily mean that they did not understand. It is also possible
that their failure was due to limitations other than those of the mental plan.
For instance, subjects in a study by Dixon (1982) read three sentences
describing how knobs and buttons should be adjusted, and then carried out
the steps. Subjects made a large number of errors in this task. However, no
difficulty was encountered at all when each step was read and performed in
isolation. The logical inference is that performance was poor in the first case
because subjects had difficulty remembering all the steps in the plan, not
because the plan itself was inadequate. In order to assert that someone fails
to understand a direction, it must be clear that they could have performed the
task if they had constructed an appropriate mental plan.
A working assumption in this approach is that there is only a single
appropriate plan for a given task. This assumption seems plausible for the
fairly simple and straightforward tasks I have used in my experiments (e.g.,
turning a knob to adjust a meter, pressing a button to turn on a light, or draw­
ing simple schematic pictures). In each of these cases there would seem to be
only a single reasonable and efficient way to do the task, and there may be lit­
tle variation in how that strategy is represented. The assumption seems less
appropriate for tasks that are complex or that can be accomplished in a
number of different ways. For example, to get to Manhatten from New Jer­
sey, one could take the Holland Tunnel, the Lincoln Tunnel, or the George
72 PETER DIXON

Washington bridge. Since each route has its own advantages and disadvan­
tages, there would be no single best plan for making the trip. However, this
situation may not be typical of many of the tasks people perform from direc­
tions, and I will at least start off by assuming that there is only one correct
plan.
A second assumption is that a reader of a set of directions will generally
be able to construct the appropriate plan regardless of how the directions are
stated. In other words, readers will take information from the directions and
use it to construct the best plan for performing the task. Since the plan is
designed specifically for the task, it may not depend on the form of the direc­
tions themselves. This assumption would seem to be true when the optimal
plan structure is readily apparent. That is, the plan may be independent of the
form of the directions when the task is similar to other known procedures, or
when the reader can infer the best plan representation based on other world
knowledge.
For instance, suppose that someone is using a set of directions for chang­
ing a light bulb. If the steps are presented in reverse temporal order, the form
of the directions would be clearly inappropriate for an optimal plan:
1. A new bulb will have to be screwed into the socket.
2. The old bulb will have to be removed.
3. A ladder will be needed to reach the socket.
It may be awkward to use this form for the mental plan because the informa­
tion needed first would actually be at the end of the representation. The prop­
osal here is that the plan would instead represent these steps in the correct
temporal order. Because they are not presented in this order, the steps would
have to be rearranged mentally in order to construct the plan.
Clearly, there are some situations in which this assumption would not
hold, and the plan representation would be strongly affected by the form of the
directions. For instance, if the task was completely novel to the reader, he or
she would have difficulty in deciding what the best plan would be, and would
have to use whatever clues could be found in the directions. Or if the direc­
tions were especially poor and misleading the reader may have difficulty in
rearranging the information appropriately. (For example, this might occur if
the example above had ten or fifteen steps instead of three; readers would
probably have trouble reordering the steps, and the mental plan would be
inefficient or inaccurate.) However, cases such as these may be exceptions;
most of the variations in form one is likely to encounter may have little effect
 ACTIONS AND PROCEDURAL DIRECTIONS 73

on one's ability to construct an appropriate mental plan.

In summary, directions are understood by constructing a mental plan for
performing the task, and the appropriateness of that plan can be judged by
looking at how well the task is performed. In my approach I have made two
additional assumptions about the nature of these plans: For most tasks there
is only a single appropriate plan representation, and readers of directions can
generally construct that representation regardless of the form of the direc­
tions. Together, these assumptions mean that the process of understanding
directions is tied closely to the structure of mental plans. In the next section
I will describe one view of mental plans, and will discuss how the comprehen­
sion process is affected by them.

3. The Hierarchical Plan Model

According to the previous analysis, the key to understanding how direc­
tions are processed lies in understanding the nature of mental plans and how
they are constructed. This section will present an approach to plans and plan
construction that I have referred to previously as the hierarchical plan model
(Dixon, in press). The balance of the paper will then consider the model's
implications for written directions.
The first assumption of the model is that plans are organized hierarchi­
cally. This idea is common in cognitive psychology and artifical intelligence
(e.g., Miller, Galanter, & Pribraum, 1960; Sacerdoti, 1977). Consider for
example the artificial itelligence system described by Sacerdoti (1977). In this
system, a plan is represented as a hierarchy of actions. The action at the top
of the hierarchy is a general description of what has to be accomplished. The
level below that describes each of the major component steps involved.
Levels below that break up each component into more specific actions, and
so on. For instance, the top level of a plan might be the action, "Go home."
At the second level might be its component steps: "Leave office," "Go to
car," and "Drive home." Furthermore, "Leave office" might be broken down
into "Put on coat," "Turn off light," and "Lock door." Thus, each level of the
plan hierarchy becomes more detailed and specific than the previous one.
The present approach will also assume that each element in the plan hier­
archy is an action, and will represent each action as an action schema (cf.
Schmidt, 1975). In this context, a schema can be though of as a prototypical
description of an action, but containing variables or "slots." By filling in dif­
ferent values in these slots, an action schema can be adapted to particular sit­
uations. For instance, one might have a "measure ingredient" schema for use
74 PETER DIXON

in following recipes. The schema might say something like, "Pour [ingre­
dient] into measuring cup until it reaches [amount]." Here, "[ingredient]"
and "[amount]" are variables that can take on different values. After filling
in the appropriate value, the schema could be used to measure a half cup of
milk, three quarters cup of boiling water, or any other combination of ingre­
dient and quantity.
The use of schemas here is similar to that of Norman (1981) in his discus­
sion of action errors. He assumed that schemas for performing actions are
stored in long-term memory, and are activated under the appropriate cir­
cumstances to accomplish a given task. After the schemas are activated, val­
ues for each of the variables in the schema have to be selected. Some of the
errors Norman discussed may in fact occur because the incorrect value is fil­
led in. For example, an error of putting the lid to the sugar container on the
coffee cup instead of back on the sugar container may be caused by using the
wrong value in a "replace lid" schema.
Two principles will be borrowed from Norman's (1981) theory of action
schemas. The first is that the action schemas are already present in long-term
memory before an action is planned or carried out. The process of creating a
plan on the basis of a set of directions consists of selecting the appropriate
schemas from those already known, rather than construcing each schema
afresh. In order to accomplish a given task, a collection of schemas would
have to be selected and organized to perform the component steps in
sequence. The second principle is that the preparation of a schema proceeds
in two stages. First it must be selected from those in memory, and then values
for each of its variables must be chosen. In other words, each component step
in the task is performed by reference to prior knowledge of similar steps,
which is then adapted to the task at hand.
Plans with this type of organization may have to be constructed from the
top down. That is, the top levels of a plan hierarchy may have to be activated
and filled in first, followed by the lower, more detailed levels. This corres­
ponds to how a lot of explicit and deliberate planning is done. For instance,
when writing a paper, one may first write an outline which describes the paper
at a high level, and only afterwards go on to work on the details. Or when
planning a trip, one may first work out the general itinerary, and then later
decide the details such as where one might eat dinner. The present assump­
tion is that this general-to-specific sequence also applies to understanding
written directions.
A final issue in plan construction concerns how information about differ-
 ACTIONS AND PROCEDURAL DIRECTIONS 75

ent plan components is obtained from a set of directions. My assumptions

here are largely borrowed from current models of reading (Rumelhart, 1977;
Just & Carpenter, 1980). To begin with, I will assume that the analysis of writ­
ten text takes place at a number of levels simultaneously. These levels include
the perception of individual letters and words, the retrieval of appropriate
word meanings, and the analysis of the syntactic structure of the sentence.
Information obtained at each level would be available to all the other levels.
For instance, if the syntactic analysis expected the next word to be a noun,
that information could be used at the lexical level to help select a word mean­
ing.
The process of constructing a mental plan would operate simultaneously
in this same environment. Whenever a piece of information relevant to the
evolving plan is found in the sentence, it would be used by the planning pro­
cess to add another plan component. Thus, by the time the end of the direc­
tions is reached, most of the plan would already be finished. In this view, plan
construction generally overlaps with the perceptual, lexical, and syntactic
processes used in reading. As will be seen below, longer reading times can
result when this overlap cannot occur.
In summary then, constructing a mental plan consists of first finding
information in the directions about the major component steps in the task. In
parallel with other reading processes, this information is used to activate
schemas in memory, and adapt them to the task at hand by putting appropri­
ate values in their variable slots. This procedure then continues with the
lower levels of the plan hierarchy. The completed plan can finally be used to
perform the task. This model already has some support from a series of exper­
iments on how long it took to read directions of different forms (Dixon,
in press). The next section will describe and test another prediction of this
model.

4. A Test of the Hierarchical Plan Model

The model as described so far makes a simple prediction about the time
it takes to understand a direction: The direction should be easiest to under­
stand when the high-level information precedes the low-level information.
As reading proceeds from left to right, the high-level information will be
encountered first. The information can be used immediately to begin con­
structing the mental plan. Subsequently, the low-level information will be
found and that can be put into the plan as well. But when the high-level infor­
mation comes after the low-level information, the direction will be more dif-
76 PETER DIXON

ficult to understand. This is because the low-level information cannot be

incorporated into the plan as soon as it is encountered. Rather, it must be
held in working memory until the upper parts of the hierarchy are completed,
and only then put in the plan.
This may lead to comprehension difficulties in two ways. First of all, the
additional information may tax working memory beyond its capacity limits.
And when information is lost from working memory there may be gaps or
inaccuracies in the mental plan. A second possibility is that longer com­
prehension times may result because the plan can be completed only after the
end of the direction is reached. That is, there would be less overlap between
the plan construction process and other reading processes, and it would take
longer overall to read the directions and construct the plan.
An example may make this prediction more concrete. The direction,
"To make a house draw a square and put a triangle on top" contains two kinds
of information. The first is a high-level description of the task to be per­
formed: "Make a house." The second is the component steps involved in
making a house: "Draw a square" and "Put a triangle on top." To construct
a plan from this direction readers would first search memory for a schema for
drawing a house. The schema would presumably contain information about
the component parts of a house (e.g., walls and roof), and the relationships
among them. Then, at a lower level, schemas for drawing squares and draw­
ing triangles would be activated. These schemas would contain variables
determining their size and orientation. Information from the top-level house
drawing schema could be used to fill in these size and orientation variables.
When the high-level description comes first in the sentence, each one of
these processes could occur as soon as the relevant information is encoun­
tered in the sentence. But when faced with a direction such as, "Draw a
square and put a triangle on top to make a house," the reader would have to
save the component step information until he or she reaches the end of the
sentence and finds out what is being drawn. That is, schemas for the lower-
level components cannot be activated until the top level of the hierarchy is
completed. Since plan construction will not overlap with other reading pro­
cesses in this case, comprehension time should be longer. And, since working
memory is being taxed, more errors should result.
An experiment was performed to test this prediction. Sixty directions
were constructed describing how simple schematic drawings of common
objects could be made. Subjects were then timed while they read and carried
out the directions. The hypothesis would be supported if directions stating
 ACTIONS AND PROCEDURAL DIRECTIONS 77

the high-level information first led to shorter reading times and fewer errors
than those that began with the component steps.
Method
Sixty sentences were constructed using the following model: "To make
a draw a " The first blank contained a common object that subjects
should be familiar with, such as house, bowling ball, or umbrella. The second
blank contained the component steps for making a simple schematic drawing
of the object (e.g., "To make a house draw a square and put a triangle on
top"). There were always at least two component steps, and they were
described in a way that was unrelated to the object being drawn. For exam­
ple, a direction such as, "To make a house draw a square and put a roof on
top" was not allowed because "roof" is related to "house."
Each direction was written in two forms. In object-embedded sentences
the object being drawn was embedded in a subordinate clause (e.g., "To
make a house draw a square and put a triangle on top"). In component-
embedded sentences the component steps were embedded in a subordinate
clause. This form followed the model, "You can make a ... by..." (e.g., "You
can make a house by drawing a square and putting a triangle on top"). In
addition, an object-first and a component-first version was constructed from
each sentence by simply switching the order of the two clauses (see Table 1).
The directions were presented to subjects one at a time on a video
monitor under the control of an Apple II microcomputer. The procedure for
each sentence was as follows. When subjects were ready to read a direction,
they pressed and held down a button labeled "sentence." After a short delay,
one of the directions appeared on the screen. It stayed there as long as the
subjects continued to hold down the button. When they had read and under­
­tood the direction, they released the button and the sentence disappeared.
At this point subjects drew the object in a numbered box in an answer book­
let.
Reading time was defined as the length of time the sentence was on the
screen; drawing time was the time from when the button was released to the
time the subjects pressed the button again to read the next sentence. Subjects
were asked to press the button with their preferred hand to insure that read­
ing and drawing did not overlap. A drawing was judged correct if it was a
reasonable depiction of the object and was consistent with the component
steps in the direction. Accuracy was scored without any knowledge of the
form of the direction seen by the subject.
78 PETER DIXON

Each subject read 30 embedded-object sentences and 30 embedded-

component sentences. Half of each were presented with the object first, and
half were presented with the components first. Which of the sixty objects was
described with each form was randomly determined for each subject. The
data analysis was performed on the median reading and drawing times on cor­
rect trials for each subject and sentence form.
Results
The median reading times averaged across subjects are shown in Figure
1. As can be seen, the directions were read much more quickly when they
began with the object to be drawn, rather than the component steps
(F(l,22)=44.37, p<.001). The order of the subordinate and main clause had
no reliable effect on reading time. There were also no statistically significant
effects on drawing time or percent correct (see Table 1).
Discussion
These results clearly support the assumptions of the hierarchical plan
model. The model assumes that plans are hierarchical and constructed from
the top down, and that plan construction generally occurs as soon as the rele­
vant information is available. Together these assumptions predict that direc­
tions should be read faster when they present information in the order in
which it is used. Thus, object-first sentences should have shorter reading
times because the object is used to activate the top-level schema in these
directions.
This experiment could have conceivably turned out quite differently.
For instance, one possibility is that reading time would have been determined
by the syntactic structure of the sentence, not by the nature of the information
in the two clauses. One might have expected simply that sentences would be
read faster when they began with the main clause. Another possibility is that
sentences would have been read faster when the information was in the cor­
rect temporal order (Spoehr, Morris, & Smith, 1983). In this case, the object
being drawn might be viewed as a consequence of the component steps. For
instance, a picture of a house is the consequence of drawing a square and put­
ting a triangle on top. Thus, it might be argued that the directions should be
read faster when the information is in the correct temporal order with the
object being drawn at the end of the sentence. The fact that neither of these
arguably plausible predictions occurred is support for the hierarchical plan
model.
Contrary to the predictions made above, there was no effect on error
 ACTIONS AND PROCEDURAL DIRECTIONS 79

TABLE 1
Mean Drawing Times and Percent Correct
Sentence Form Drawing Percent
Time (sec) Correct

Object First, Object Embedded 10.45 85%

"To make a house draw a square
and put a triangle on top."
Object First, Component Embedded 9.84 83%
"You can make a house by drawing
a square and putting a triangle
on top."
Component First, Object Embedded 10.29 84%
"Draw a square and put a triangle
on top to make a house."
Component First, Component Embedded 10.04 81%
"By drawing a square and putting
a triangle on top you can make
ahouse. , ,

rate. The hypothesis was that holding on to the component steps while read­
ing the rest of the sentence would tax working memory and lead to an incor­
rect or inadequate plan. The fact that an effect on error rate did not occur
suggests that working memory was not a limiting factor in this task. Although
there were a substantial number of errors, they may have been due to other
causes, such as misinterpreting words or phrases. It is possible that the error
rate effect would be obtained if the sentences were made more demanding,
for instance by adding more syntactic complexity.
With the present materials it is clear why hierarchical plans would be
constructed from the top down. The high-level action of drawing a particular
object exerts a strong influence on the features and relationships of the com­
ponent steps. For instance, the fact that a house is being drawn affects the size
and shape of the triangle used for the roof, and how it should be related to the
square. In other words, how a particular component step fits into the plan is
not clear until after the high-level information about the object being drawn
80 PETER DIXON
 ACTIONS AND PROCEDURAL DIRECTIONS 81

organizes the overall plan structure. This is consistent with the proposal made
by Norman (1981) that lower-level schemas receive many of their variable
values from the schemas on the upper levels. In part, plans are constructed in
a top-down manner in the present model because the top levels impose con­
straints and conditions on the lower levels.
This experiment has provided clear evidence for the hypothesis that
plans are organized hierarchically. Moreover, the results indicate that direc­
tions are understood more quickly when they reflect that hierarchical struc­
ture. That is, reading time is shorter when information at the top of the plan
hierarchy comes first in the sentence. In the following section I will propose
another way in which the form of directions may indicate a plan's structure.

5. The Role of Action Statements in Directions

The hypothesis I will suggest in this section is that explicit action state­
ments in a set of directions often signal a piece of information that is near the
top of the plan hierarchy. From a logical point of view, many pieces of infor­
mation in a plan can be described either as an action, or in terms of the states
or conditions that surround that action. For instance, one can say either,
"Start the motor before proceeding further," or equivalently, "The motor
should be running before proceeding further." In one case the action is stated
explicitly, while in the other it must be inferred from the statement of the
desired state. Thus, the writer of a set of directions generally has a choice of
how to express a component of a plan hierarchy. The hypothesis is that when
the component is at a high level in the hierarchy it will be stated as an explicit
action; when it is lower it will be stated implicitly as a state or condition. The
writer uses this device to signal hierarchical height to the reader.
This hypothesis is reminiscent of the distinction between foreground and
background information in discourse (e.g., Hopper, 1977). Foreground
information describes the sequence of events that define a narrative, while
background information often describes states or conditions that are present
while the events are taking place. Given the nature of the foreground infor­
mation, it is more likely to be expressed with an action clause than is the
background information. For instance, Hopper and Thompson (1980) found
that 88% of the foreground clauses in their corpus were active, while only
41% of the background clauses were. Presumably, this correlation between
action and foregrounding is caused by the nature of the information: the nar­
rative events that make up the foreground are most easily described in terms
82 PETER DIXON

of actions.
High-level information in a plan hierarchy is similar to foreground infor­
mation in narrative in that both consist of the most central or important ele­
ments in the discourse. Consequently, the present hypothesis simply restates
the Hopper and Thompson (1980) findings in terms of written directions.
However, the reasons action statements indicate high-level information may
be quite different for directions than for other discourse types. The hierarchi­
cal plan model assumes that information is represented by action schémas at
all levels of the hierarchy. In the experiment described above, component
steps such as, "Draw a square and put a triangle on top'' are just as much
actions as the object information, "Make a house," even though the former
is lower in the plan hierarchy. Moreover, the component steps would proba­
bly still be represented as actions even if they were not stated as such. For
instance, I would expect the same mental representation for sentence in
which the component actions were not stated at all, as in, "Make a house with
a square and a triangle on top." Thus, the use of actions for high-level infor­
mation may be purely conventional, and not related at all to the nature of the
information being conveyed.
The hypothesis that high-level information in directions is signaled by
explicit actions helps make sense of a previous series of reading time experi­
ments (Dixon, 1982, in press). The sentences used in those experiments were
divided into action information and condition information. For instance, in
the direction, "Press button A while light Z is on," the first clause states an
action (pressing the button), while the second clause states a condition that
should be true before the action is performed. A consistent finding in those
experiments was that sentences were read faster when they began with the
action clause. Table 2 shows some representative sentences, and the reading
times for the action-first and action-second order.
Each of these results fits with the hierarchical plan notion developed
here. For example, consider the direction, "Turn the left knob to set the
alpha meter to 20." The main, central action is to turn the knob. However,
turning the knob actually has a number of more specific components: one has
to move one's hand to the knob, grasp it, twist it until the alpha meter reads
20, and then release it. The action of checking the alpha meter would then
occupy a subordinate position in the action hierarchy because it is simply one
component of the more general action of turning the knob. Thus, the action-
first version is read faster than the action-second version because the action
information is higher in the plan than the condition information.
 ACTIONS AND PROCEDURAL DIRECTIONS 83

The present hypothesis explains why it is always the action information

that is at the top of the hierarchy in these experiments. The hypothesis states
that actions will most naturally be used for high-level information. Since in
each case the direction was intended to be a natural and straightforward
description of the task, the action may have ended up describing information
higher in the plan hierarchy than the conditions did. For example, when the
task involves checking a meter during the course of turning a knob, the most
natural way to express it would be with an explicit action for turning the knob,
and an implicit action for reading the meter, as in, 'Turn the left knob to set
the alpha meter to 20. " Since checking the meter is only a subordinate compo­
nent of turning the knob, it would be odd to state it as an explicit action. Thus,
it is possible that by trying to state the directions naturally, these experiments
unintentionally confounded the action/condition distinction with height in
the hierarchical plan representation.
It would be interesting to consider how planning would proceed if these
two factors were unconfounded. According to the present view, sentences
that begin with the high-level information should be read more quickly
regardless of whether the high-level information is stated as an action or not.
Since the same task is being performed, the same plan would have to be con­
structed, and the same information would be needed at the top level of the
plan. For instance, consider a sentence such as, "You can draw a house with
a square and a triangle on top." Here the high-level information is stated as
an explicit action and the low-level component are only implicit. However,
this situation could be turned around as in, "After drawing a square and put­
ting a triangle on top you will have a house." In this version, the component
actions are explicit and the high-level information is stated as a condition.
Thus, the low-level information is marked incorrectly as high in the hierar­
chy.
I would predict that a person actually trying to carry out the direction
would realize that the information about a house should go at the top of the
hierarchy, regardless of how the sentence is phrased. Sentences with an inap­
propriate structure may take longer to understand, but they would still be
read faster when they begin with the high-level information. However, this
prediction may be limited to cases in which there is a single hierarchical
organization for the plan that is obvious to the reader. When the task can be
done in a number of different ways or with different plan organizations, stat­
ing an action explicitly may cause it to be put near the top of the plan hierar­
chy. In other words, the way in which directions are phrased may affect the
84 PETER DIXON

TABLE 2
Effects of Information Order on Reading Time (sec)
Sentence Form Action-First Action-Second
Order Order

1. The left knob should be turned 3.58 4.18

in order to set the alpha meter
to 20. a
2. As a result of turning the left 3.80 3.96
knob the alpha meter should read
20. a
3. Hold the left button down to get 4.56 5.18
the gamma meter above 20. b
4. Hold the left button down if the 4.80 5.03
gamma meter is above 20. b
5. Before pressing button Alight 2.50 2.60
Z should be on. c
6. Press button A while light Z 2.25 2.35
ison. c
a. Sentence 1, Immediate Performance Condition in Experiment 1,
Dixon (1982)
b. Sentence 3, Immediate performance Condition in Experiment 2,
Dixon (1982)
c. Antecedent conditions in Experiment 1, Dixon (1984)

plan representation if there are few semantic and pragmatic constraints on its
organization.
Several previous results seem to support the hypothesis that the relation­
ship between explicit actions and hierarchical height is only conventional
rather than causal. For instance, in Experiment 2 of Dixon (1984) sentences
like (5) and (6) in Table 2 were used, but the experimental task was arranged
so that the condition clause contained a piece of information needed at the
top of the plan hierarchy. It was argued in that study that turning a light on
and turning a light off were accomplished with two different action schemas.
So a direction such as "Press button A to turn light Z on" would use a "turning
on" schema, while "Press button A to turn light Z off" would use a "turning
 ACTIONS AND PROCEDURAL DIRECTIONS 85

off" schema. Thus the word "on" or "off" in the sentence was needed to
decide which top-level schema should be activated. Since the word occurred
in the condition part of the sentence, the condition-first sentences were read
faster. In this case, the action statement did not indicate the most important
or central piece of information.
Another experiment in which action statements may have failed to indi­
cate the most important information was reported by Spoehr, et al. (1983).
They had subjects read and carry out directions such as, "If the sigma indi­
cator light is on turn the right knob so that the gamma meter reads 20." The
first part of this example ("If the sigma indicator light is on") is an antecedent
condition that should be true before starting the action. The second part is the
main action to be performed ("turn the right knob"). Finally a condition is
stated that should be the consequence of the action ("the gamma meter reads
20"). Spoehr, et al. found that the directions were read fastest when they
stated the antecedent condition first, then the action, and then the con­
sequent condition. Action-first sentences were read more slowly (e.g., "Turn
the right knob if the sigma indicator light is on so that the gamma meter reads
20").
This result can be explained by considering how the antecedent, action,
and consequence might fit into a plan for performing the task. At the top level
of the hierarchy would be a general action to be accomplished, such as "Ad­
just meter." Below that would be two component steps, "Check sigma indi­
cator light," and "Turn right knob." The action of checking the consequent
condition would be a subcomponent of turning the right knob. I have
assumed here that the antecedent and the action would be represented at the
same level in the plan hierarchy. This may be appropriate in the Spoehr, et al.
(1983) task because the antecedent and the action were not causally or logi­
cally related. Thus, it is reasonable to think of them as two independent
actions. Since the action clause would not be higher in the hierarchy than both
of the other clauses, there would be no reason to predict that action-first sen­
tences would be read faster.
This analysis leads to what seems to be a more natural way of stating the
the Spoehr, et al. (1983) directions. By hypothesis, high-level information
should be stated as actions. Since there are two major component steps
("check indicator" and "turn knob"), each should be stated as an explicit
action. Only the subordinate consequence should be stated as a condition.
The direction thus becomes, "Check that the sigma indicator is on and then
turn the right knob to set the gamma meter to 20." This seems less awkward
86 PETER DIXON

than the Spoehr, et al. (1983) version, despite being a little longer.
In summary, the hypothesis that explicit actions signal information near
the top of a plan hierarchy accounts for a number of previous reading time
results. Since directions are read more quickly when they begin with high-
level information, it could account for the finding that action-first sentences
are read more quickly in Dixon (1982) and Dixon (in press). Moreover, since
action statements are only correlated with high-level information by conven­
tion, it is possible to construct examples where lower-level information is
stated as an action, and the high-level information is stated as a condition. In
these cases the action-first order is read more slowly, as predicted by the
hierarchical plan model.

6. Conclusions
In this paper I have argued that understanding directions involves con­
structing a mental plan for performing a given task. According to the prop­
osed model of comprehension, a set of directions will be easy or difficult to
understand depending on how well the directions reflect the structure of the
mental plan being constructed. I have focused here on one aspect of a plan's
structure, the hierarchical arrangements of actions. Since plans are con­
structed from the top down, sentences will be easier to understand (i.e., read fas­
ter) when they begin with high-level information and continue with the lower-
level information. This prediction was confirmed experimentally. It was also
conjectured that high-level information may often be stated as explicit
actions, while less important information would be conveyed more implicitly.
Although an experimental test of this conjecture has not yet been performed,
it does seem to make sense of a number of previous reading time results.
Even though most of the results and examples described so far have been
limited to manipulations within sentences, I would anticipate that the same
conclusions would also apply to larger sets of directions. For instance, in an
experiment by Dixon and LeFevre (1984), an effect of high-level information
was obtained with extended sets of directions for drawing pictures. Subjects
were asked to follow a series of ten steps such as:
1. Draw a horizontal line across the page.
2. Draw a tall thin rectangle just above it on the left-hand side of the
page.
When the steps were followed correctly, a picture such as a city scene or a
landscape would be drawn. In one condition subjects were told the nature of
 ACTIONS A N D P R O C E D U R A L D I R E C T I O N S 87

the scene before they started (e.g., 'This is a landscape"). This information
may have activated a schema which could be used to organize the subsequent
steps. Thus, it may be similar to the high-level information used in Section 2.
Compared to a second condition in which this information was not given, sub­
jects read and carried out the directions faster and with fewer errors when
they knew what was being drawn. The results suggest that at least under some
circumstances the effects described in the present paper may also occur with
directions larger than single sentences.
The model proposed here is related to the linguistic hypothesis that dis­
course contains multiple levels of information. To begin with, the distinction
made here between high-level and lower-level information is similar to the
distinction between backbone and background information made by Lon-
gacre (1979), and the distinction between foreground and background made
by Hopper (1979). The present analysis leaves open the possibility that there
could in principle be more than just two levels (cf. Jones & Jones, 1979). The
number of levels in procedural directions may be constrained only by the
complexity of the mental plan representation.
Secondly, the notion that high-level information is conveyed by explicit
actions is consistent with the finding that foreground clauses are more active
than background clauses (e.g., Hopper & Thompson, 1980). However, in the
present view of directions, the structure of a plan is constrained in part by the
task being performed. Thus, the level of a piece of information in a mental
plan may also be a function of how it is used in performing the task, not only
how it is stated in the directions. Consequently, it would still be possible for
an item of low importance to be stated as an action.
In summary, the present paper provides some evidence that mental
plans are organized hierarchically and that directions are read more quickly
when they take that organization into account. Although I have not attemp­
ted to analyze the structure of directions generally, the results do suggest that
directions that are clear and easy to understand will often begin with high-
level information in the form of an explicit action.

NOTE
This research was supported by Grant A8263 from the Natural Sciences and Engineering Research
Council of Canada. I would like to thank Renée Elio, Jo-Anne LeFevre, and Robert Longacre for
comments on an earlier draft of this manuscript.
88 PETER DIXON

REFERENCES

Dixon, P. in press. The structure of mental plans for following directions.

Journal of Experimental Psychology: Learning, Memory, and Cognition.
. 1982. Plans and written directions for complex tasks. Journal of Verbal
Learning and Verbal Behavior. 21: 70-84.
Dixon, P. and J. LeFevre. 1984. Goal information in procedural directions.
In P. Wright (Chair) Information Design: The contribution of Psychology.
Symposium conducted at the XXIII International Congress of Psychol­
ogy, Acapulco.
Hopper, P. and S. Thompson. 1980. Transitivity in grammar and discourse.
Language. 56: 251-299.
Hopper, P. 1979. Aspect and foregrounding in discourse. In T. Givon (Ed.),
Discourse and Syntax: Syntax and Semantics. 12: 213-242. New York:
Academic Press.
Jones, L. and L. Jones. 1979. Multiple levels of information in discourse. In
L. Jones (Ed.), Discourse Studies in Mesoamerican Languages: Discus­
sion. 3-28. Arlington, Texas: Summer Institute of Linguistics and the Uni­
versity of Texas at Arlington.
Just, M.A. and P. Carpenter. 1980. A theory of reading: From eye fixations
to comprehension. Psychological Review. 87: 329-354.
Kintsch, W. and T. A. van Dijk. 1978. Toward a model of text comprehension
and production. Psychological Review. 85: 363-394.
Longacre, R. 1979. The discourse structure of the flood narrative. In G. Mac-
Rae (Ed.), Society of Biblical Literature: 1976 Seminar Papers. 235-252.
Massoula, Montana: Scholars Press.
Marcel, T. and P. Barnard. 1979. Paragraphs of pictograms: The use of non­
verbal instructions for equipment. In P.A. Kolers, M.E. Wrolstad, & H.
Bouma (Eds.), Processing of Visible Language. New York: Plenum Press.
Meyer, B. 1975. The organization of prose and its effect upon memory.
Amsterdam: North Holland.
Miller, G.A., E. Galanter. and K.H. Pribraum. 1960. Plans and the structure
of behavior. New York: Henry Holt and Company.
Norman, D.A. 1981. Categorization of action slips. Psychological Review.
88:1-15.
Omanson, R.C. 1982. An analysis of narratives: Identifying central, suppor­
tive, and distracting content. Discourse Processes. 5: 195-224.
Rumelhart, E.E. 1977. Toward an interactive model of reading. In S. Dornic
 ACTIONS AND PROCEDURAL DIRECTIONS 89

(Ed.), Attention and Performance VI. Hillsdale, N.J.: Lawrence Erlbaum

Associates.
Sacerdoti, E.D. 1977. A Structure for plans and behavior. New York:
Elsevier North-Holland.
Schmidt, R.A. 1975. A schema theory of discrete motor skill learning.
Psychological Review. 82: 225-260.
Spoehr, K.T., M.E. Morris. and E.E. Smith. 1983. Comprehension of
instructions for operating devices. Paper presented at the meeting of the
Psychonomic Society, San Diego.
 DISCOURSE FUNCTIONS OF WORD ORDER
IN SESOTHO ACQUISITION1

KATHERINE DEMUTH
Boston University

1. Introduction
Word order has been a topic of great interest since the onset of systema­
tic language acquisition studies (Slobin 1966, Bloom 1970, Bowerman 1973,
Brown 1973). Such study, it was thought, might provide evidence for compet­
ing theories of what is innate and what is learned in the process of language
acquisition. Additional research has examined later points in discourse
development where children use three or more constituents in an utterance,
with potential for the use of several different word orders. When data con­
cerning children's word order is compared in different languages, the story
becomes more complex. It appears that most children tend to use a fairly
fixed word order in initial word combinations (English - Maratsos & Chalkley
1980, Turkish - Slobin 1982, Mandarin - Erbaugh 1982), even in languages
that have case marking and allow great flexibility in word order (Russian -
Slobin 1966) and even in cases were adult input contains varied word order.
In Finnish (Bowerman 1973), however, one child followed adult variable
word order at a very early stage. Further investigation has revealed a ten­
dency to delete subject and/or order constituents in a new-old information
sequence (Italian - Fava & Tirondola 1977, Dutch - Snow 1978). Such data
suggests a 'functional' approach to language (Bates & McWhinney 1979),
where the child assumes the hearer is privy to the same background informa­
tion with which he or she is operating. It is only later, with increased discourse
sensitivity and awareness, that children begin to provide the hearer with the
background information necessary for the effective development of a dis­
course topic.
It appears that several variables are involved in what word order a child
may use in a given language. Some languages have more fixed word order
92 KATHERINE DEMUTH

(English, Chinese), while others are extremely flexible (Cayuga, Walpiri).

Some languages rely mainly on word order to indicate grammatical relations
and shifts of focus (English). Others rely primarily on inflection (Russian),
while still others rely on both (Serbo-Croatian). It appears that mixed sys­
tems (i.e. Serbo-Croatian), where both word order and inflections are
involved in constructing discourse, are harder to master than systems which
use only either word order or inflectional strategies. There may, of course, be
individual variation between children which would make the pattern of one
system more accessible than another.
This chapter examines acquisition data from Sesotho, 2 an SVO Bantu
language with inflected subject-verb and noun-complement agreement,
where alternative word orders are used for different discourse functions.
Sesotho speaking children initially use SVO word order. Then, at about 2½
years, they begin to use some different word orders. In this chapter we will
discuss the functions of these word orders in spontaneous child speech, with
special attention given to the linguistic and discourse environments where
children make maximum use of the word order possibilities in this language.
We conclude with some hypotheses concerning the nature of cognitive pro­
cesses involved in learning to manipulate word order.

2. Discourse Context and Word Order Strategies

Many linguists have noted the large number of possible word orders used
in Bantu languages (Aghem-Watters 1979, Makua-Stucky 1985), Chichewa -
Bresnan and Mchombo 1986). The discourse functions of some of these
constructions have been described for the related Bantu language Haya
(Tenenbaum 1977, Byarushayo & Tenenbaum 1978). These constructions
and their uses correspond closely to those of Sesotho. 3 As in Haya we find
word order shifts used mainly to clarify a misunderstanding or a misreading
of what was intended. 4 Adult-child interaction is frequently beset with 'com­
municative breakdowns' and is therefore precisely the kind of context in
which we would expect maximum manipulation of word order to occur. We
predictably find that Basotho adults use a great number of word order shifts
when speaking to young children (aged 2-3). Basotho adults and caregivers
talk animatedly with children even before youngsters can verbalize (Demuth
1983). Once they begin to talk, children are prompted to answer questions
and assume an active role in conversation. In trying to make themselves
understood, Basotho caregivers use the wide repertoire of Sesotho grammat­
ical forms available for discourse strategies.
 WORD ORDER IN SESOTHO ACQUISITION 93

Children, on the other hand, have much of the burden of communication

lifted from them when they are communicating with such accommodating
adults. 5 In fact, few of the word order shifts we find used by children occur
during interactions with adults. Instead, we find that the more challenging
communicative contexts for the child are those involving interactions with
sibling or peers, where the hearer is less linguistically accomodating than the
adult. It is in daily play situations that we find most of the young child's (and
even 5-6 year olds') use of word order shifts, and it is these context which we
examine in this chapter.
In the following sections we discuss the most relevant word orders in
Sesotho, and then proceed to consider how adults use these constructions
when speaking to children. Finally, we examine children's developmental
changes in word order manipulation as a discourse strategy, from about 2
until 6 years of age.

3. Word Order in Sesotho

Basic Sesotho word order is SVO, with object pronoun cliticized before
the verb, as seen in examples (1) and (2) below (sc = subject concord, foc
= focus marker, 6 obj = object pronoun). Each subject concord and object
clitic agree in noun class gender with their co-referent noun. The g is the sub­
ject marker for humans, ę is the object clitic for nouns such as 'dog', etc.
(1) (Thabo) g batla ntjá (2) {Thabo) o-a ę-bâtla
7
sc V o sc-foc obj-V
(Thabo) he want dog (Thabo) he it want
(Thabo) he wants a/the dog. (Thabo) he wants it.
Lexical objects are occasionally preposed, especially in case of a rein­
troduced or contrastive discourse topic.
(3) ntjâ g-a ę-hátla
O sc-foc obj-V
dog he it want
The dog, he (Thabo) wants it.
Preposed subjects are also used, again primarily as reintroduced contras­
tive topics of conversation.
(4) 'nâ kę batla ntjâ
pn sc V O
me I want dog
As for me, I want the/a dog.
94 KATHERINE DEMUTH

The use of postposed lexical object plus preverbal object clitic has been
characterized in other languages as an afterthought. It is said to be used to
clarify the object clitic when the speaker realizes that the referent is not in the
'active memory' (Chafe, this volume) of the hearer. It is also supposed to be
used as an expression of surprise, doubt, emphasis or contrast, or to de-
emphasise the assertion and emphasise the postposed element (Tenenbaum
1977). In Sesotho, however, postponed lexical objects are frequently used
even in cases where the lexical object has just been specified (verbally or
otherwise) and is apparent to both speaker and hearer. Postposed object con­
structions (realized nominally or by demonstrative pronouns) function in
Sesotho primarily to emphasize or contrast the object or additionally to
emphasize the verbal argument, the only 'new' information in the sentence.
(5)
sc-foc obj-V O
he it want dog
He (Thabo) wants it, the dog.
As with postposed objects, postposed subjects are also used to focus
attention on the verbal assertion rather than on the already specified, empha­
tic or contrastive subject. These constructions are found in contexts where
the subject referent has already been identified. Thus, in the following exam­
ple, various aspects of ntjâ 'dog' have already been mentioned.
(6) e-a tsamâea ntjâ
sc-foc V S
it leave dog
It's leaving, the dog.
In each of these cases, the shifted lexical referent has been previously
identified in the discourse. These constructions are not used to introduce a
new referent, but may contrast a referent with others which are also active in
the discourse context.
In addition to extraposition, clefts and passives also play a role in the
pragmatics of Sesotho discourse and reference. Below we examine the use of
these constructions in question formation.
The unmarked form of questions in Sesotho places the question word at
the end of the sentence, as seen below:
(7)
thing this cop what person that cop who
This thing is what? That person is who?
 WORD ORDER IN SESOTHO ACQUISITION 95

Thábo
T. he where
Thabo is where?
Questions are phrased in this manner when an already salient referent is
being brought into the discourse for the first time, or when one referent is
being contrasted with another. In addition to this unmarked question form,
there is also a cleft question or 'preposed' form which is used only in cases
where the referent has already been specified:
(8)
cop what thing this cop who person this
It's what, this thing? It's who, this person?

he where T.
He's where, Thabo?
These are the only two ways to form questions with copular constructions;
there is no separate pseudo-cleft form.
It is frequently the case that unmarked and cleft questions will be used in
sequence: the unmarked form is used first to establish the referent; after this
both forms can be used interchangeably. Such sequences occur not only in
repeated turns by the same speaker, but also in consecutive turns by different
speakers, as seen in the following interaction between child and grand­
mother. Here the child uses the unmarked question form to verbally intro­
duce the referent, while the grandmother uses the cleft question form once
the discourse topic has been established.
(9) Hlobohang (36 months) and grandmother MM are identifying
items on a food chart.
H
N-dem cop wh-
thing-this is what
This thine is what?
MM
cop wh- pn N dem
yes, it's what that-one thing that
Yes, it's what, that thing?
Unmarked and cleft questions also occur with full verbal predicates,
where they then use a relative construction. Here again, the unmarked form
96 KATHERINE DEMUTH

is used with first mentions or reitterations of the referent, while the cleft form
is used only once the referent has been specified.
(10)
rel obj-v-prf-rel O cop wh-
who you gave dog is who
The one that gave you the dog is who?
(11)
cop wh- rel obj-V-prf-rel O
it's who who you gave dog
It's who that gave you the dog?
Examples (7)-(ll) illustrate the active forms of questions. Passive forms are
used when the subject is questioned.
(12)
sc V-prf-pas O cop wh-
you given dog by who
You were given the dog by who?
In passive question forms, the direct object is frequently preposed, resulting
in questions like that in (13) below - similar to the preposed object construc­
tion in (3) above.
(13)
O sc obj-V-prf-pas cop wh-
dog you it were-given by who
The dog, you were given it by who?
While permutations of word order between cleft and passive options 8 are not
frequently employed within the same speech event by children, they are often
used by adults in their attempts to communicate effectively with young inter­
locutors. We now examine how adults manipulate these constructions to
achieve effective communication with children.

4. Adult Uses of Word Order in Talking with Children

As might be expected, adults and caregivers possess a much larger reper­
toire of variations in word order than do children. Further, adults frequently
use several different pragmatically motivated word orders within a single
 WORD ORDER IN SESOTHO ACQUISITION 97

communicative episode. Thus, we find complex word reordering such as that

in (14) where there are shifts from passive to cleft constructions (a & b, c &
d), a minor tense/aspect shift (a & b) a postposed subject (d), and final resort­
ing to an unmarked question form (e). The great aunt here is persistant in try­
ing to extract an answer from her grandniece:
(14) Great aunt J is engaging grandniece 'Neuoe (30 months) in conver­
sation, pressing her for information concerning which child (affec­
tionately called 'grandmother' here) hit 'Neuoe.
a.
se pst V-pas cop wh-
You were lashed by who?
b.
cop wh- rel pst-cont sc V
is who who were you lash
It's who that was lashing you?
(Another child tries to prompt N in between J's repetitions)
(N tries to respond)
c.
sc pst-cont V-pas cop N wh-
you were lash by grandmother who?
You were being lashed bv which old ladv?
d.
cop wh- rel pst-cont obj-V N
is who who you lash grandmother
It's who that was lashing you, the old lady?
(N tries to respond)
e.
wh S rel pst-cont obj-V cop wh- wh
uh grandmother who you lashis who uh
Uh? The old lady that was lashing you is who? Uh?
(N finally answers)
Notice here again the ordering of 'unmarked' passive before the more em­
phatic cleft. The introduction and postposing of nkhono 'grandmother' lends
further emphasis to the agent of the action. This shifting back and forth from
one construction to another reflects an apparent attempt on the part of the
adult to help the child understand the question and produce the desired
response.
98 KATHERINE DEMUTH

during interaction with young children. A possible interpretation for the use
of these constructions is that they represent efforts to help direct the child's
attention to what the speaker considers to be salient in the discourse.

5 Children's Use of Word Order

The data on rural Basotho children's speech is selected from 93 hours of
spontaneous verbal interaction between 4 children, their peers, siblings and
older family members. Each of the children was audio recorded every 5-8
weeks over a period of 12-14 months. The youngest child was 25 months at the
initation of the study, while the older siblings were 5-6 years old. As noted
above, Basotho adults initiate verbal interaction with young children with
great enthusiasm. They introduce the majority of discourse topics and are
largely responsible for topic continuity and topic shift. It is only after the age
of 2½ or so, when young children start to spend more time with their peers,
that they begin to initiate topics of conversation, introduce contrastive topics
and maintain discourse topics which have already been initiated. It is pre­
cisely at this time that Basotho children begin to manipulate word order.
Here we examine the conversational strategies used by children of two years
of age and trace that development until the age of 6.
Young children do not control all the word order possibilities discussed
above. Instead, two year old children use the alternative strategies of repeti­
tion, gesture and prosodic intensity to make themselves understood. Exam­
ple (15) below illustrates a case where a young child is interacting with a less
than competent speaker'. He uses the same construction repeatedly, resort­
ing to the use of prosodic intensity when communication fails.
(15) Hlobohang (27.3 months) has just pointed to the tape recorder,
asking researcher K whose it is. He then tries unsuccessfully to ask
who bought the tape recorder for her.
H a rekélâ é
(ç ę - rek-éts-o-ę kę
you it-buy- cop who
ben/prf/pas
You were bought it by who?
(K doesn't understand)
K
Yes.
 WORD ORDER IN SESOTHO ACQUISITION 99

you it buy- cop who

ben/prf/pas
You were bought it by who?
K Ké éa-ka.
cop my
It's mine.
(H is indignant - his question is not being answered
H

you it-buy- cop who

ben/prf/pas
You were bought it by who?!
Note here that the construction used is the passive, and that there is no
attempt, when communication fails, to switch to the (grammatically more
complex?) cleft form kę ea o rekétséng 'It's who that bought it for
you?' Such a shift might have helped the hearer clarify what the child
intended. At this early age Hlobohang apparently does not possess the lin­
guistic knowledge or discourse skills necessary to switch from passive to cleft
and vice versa. The use of cleft constructions becomes quite noticeable only
a few months later, about 2½ years of age.
Manipulation of unmarked and cleft questions, however, is already pro­
ductive by the age of 25 months, as seen in the following example:
(16) Hlobohang (25 months) and Mololo (4.5 yrs) are playing with tin
can 'cars'.

sc wh- S
It's where, another one?
b.
S sc wh-
Another on is where?
c.
sc wh- S dem
It's where, that other one?
In this interaction Hlobohang is trying to elicit a response on the part of the
100 KATHERINE DEMUTH

hearer. The first question (a) takes the form of a cleft. In adult speech we
would expect an unmarked question to be used upon first mention, so that the
referent could be specified. Here the child apparently assumes that the refe­
rent is known to the hearer and he uses the cleft form. This is consistent with
observations that Dutch and Italian children (Snow 1972, Fava & Tirondola
1977) initially tend to order new information first, assuming the hearer is
privy to the background upon which the utterance is based. In (b) the child
switches to the unmarked question form. After no response from the hearer
again, he shifts back to the cleft form in (c), providing further specification of
referent with the addition of a demonstrative pronoun. Although the child
makes an unwarranted assumption here about the recoverability of the refe­
rent, he is able to manipulate cleft and unmarked question forms from a very
early age.
While children do use occasional cleft constructions with full predicates
by 2 years of age, there is a dramatic increase in the use of this construction
at 30 months (Demuth 1983). The function of clefts at this time was to clarify
misunderstandings, but also to specify who or what was responsible for a par­
ticular action or state of affairs. Concurrently we find that postposed subject
constructions begin to be produced with extreme frequency. As discussed
above, these postposed constructions are not the result of an afterthought or
clarification (as proposed for adult speech in other Bantu languages), or even
for emphasis or contrast (as in adult Sesotho). Rather they are used to focus
on the assertion or activity of the referent which has already been specified.
Postposed subjects are very commonly used at 2½ years of age in contexts
such as the following:
(17) Mololo (5 yrs) and Hlobohang (30 months) are playing cars.
a.
hort se V O
Let me see the car.
b. H tsamâea koloi éa:-ka.
sc-foc V S pos
it go car my
It's going, my car.
c.
sc V-prf N-loc S pos
it turned-over aloe-at car my
It turned over at the aloe, my car.
 WORD ORDER IN SESOTHO ACQUISITION 101

In (a) the first speaker specifies the referent. In (b) and (c) the second speaker
elaborates on the action in which the referent is involved. Postposed subject
constructions of this kind present new verbal information first, again preser­
ving new-given order of presentation frequently found with young children.
Postposed subjects are commonly used at 2½ years. At this time, post­
posed objects start to be produced. Between 3 and 4 years postposed objects
are frequently used, then taper off to more adult-like usage in particular
contexts. Once again, postposed objects have been already introduced into
the discourse and are used either for emphasis, or to illuminate the action to
which the object is subjected. 9
(18) Mololo (5-6 years) tells grandmother that Hlobohang (36 months)
has his (Mololo's) book. Grandmother says she will give Mololo
'her' book, which is actually Hlobolang's. Hlobolang then comp­
lains whiningly to Grandmother in turns (a) + (b) and tells Mololo
to take back his book in turn (c)
a. Tlia buka ena éaka.
V O dem poss
bring book that my
Bring (me) that/my book.
b. 'Na ke-a e-batla buka éa:ka
pn sc-foc obj-V O poss
me I it-want book my
As for me, I want it, my book.
c. É-nke, ha ké é-tsotélle buka émpe.
obj-V neg sc obj-V O adj
it-take neg I it-care book ugly
Take it, I don't care about it, the ugly book.
Turn (a) introduces the book in object position. In utterances (b) + (c) the
lexical form of book is postposed, while in the main clause it takes the form
of an object clitic. As with postposed subjects, we find postposed objects
in contexts where the lexical form has already been introduced in immedi­
ately previous discourse and is postposed in subsequent utterances to give
it emphasis. Such constructions became increasingly frequent about 3 years
of age.
Notice here also the preposing of the independent subject pronoun 'na
in (b), used for contrastive emphasis (Kunene 1975). Preposing of this inde­
pendent pronoun is extremely common and occurs frequently as the sole
102 KATHERINE DEMUTH

subject 'marker', with no subject concord marker (usually before the age of
three, but occasionally even with 5 and 6 year olds). A complete analysis of
the functions of 'na and other preposed subjects in children's speech is the
topic for another paper (see Demuth, in preparation).
Preposed objects, however, were not generally used by very young chil­
dren. They become productive around the age of 3, but only frequently used
by 5 or 6 year olds. This construction, variously referred to as fronted, topi-
calized or left dislocated, is also acquired relatively late in other languages
(English (Gruber 1967), Mandarin Chinese (Erbaugh 1982)). Preposed
objects emphasize a reintroduced or contrastive topic, as seen in example
(19) below. Here the older child uses a preposed object (a), no lexical object
in (b) and a postposed object in (c):
(19) Mololo (4.6 yrs) has been talking to himself while playing with a
flashlight (torch), turning it off and on while Hlobohang (26.2
months) is roasting potatoes in the ashes from the fire. Mololo tells
Hlobohang:
a. :ch ea-hao kę ę timme,
0 pos sc obj V-prf
flashlight your I it turned-off
Your flashlight, I'm turning it off.
b. Kę ę timme hóre,
sc obj V-prf interj
1 it turned-off like-this
I turned it off like this.
c. Kę ę timme :ch
sc obj V-prf O
I it turned-off flashlight
I turned it off, the flashlight.
Note again the particular progression in use of constructions. The referent is
initially preposed in an effort to draw the hearer's attention to the 'rein­
troduced' discourse topic, in part, perhaps, to contrast the torch with the
potatoes which Hlobohang has been playing with. In so doing, it is the dis­
course referent which is ordered first. Once the referent has been established,
it is repeated in the form of an object clitic. When it is lexicalized again in (c),
it is postposed with the propositional emphasis on the verb. This sequence of
word orders seems once again to support the new-given ordering pattern
across turns of a discourse unit. By the age of 3, children begin to use post-
 WORD ORDER IN SESOTHO ACQUISITION 103

posed object constructions with increased frequency. Preposed objects, how­

ever, become more frequently used only by 5-6 year olds and and are used
quite often by adults.

6. Discussion
In sum, there is a specific developmental trend in the learning of differ­
ent word orders in Sesotho. Initially, SVO word order is used. Before two
years of age gesture, repetition and prosodic intensity serve discourse func­
tions later accomplished by the manipulation of word order. Around 2 years
copular questions in unmarked and cleft form are used in sequence, with the
cleft often preceding the unmarked adult-introductory form. By 2½ years
postposed subjects are very commonly used once the discourse referent has
already been specified. Clefts become productive at this age, emphasizing or
contrasting the referent. By 3 years the use of postposed subjects has
diminished, while postposed objects become much more frequent. Preposed
objects are rare at 3 years while 5 and 6 year old children and adults in speech
to children use them frequently. After an experimental stage with each of
these constructions children seem to store them away for future use when dis­
course situations require them. Such use, analysis and 'retention' of gram­
matical form, has been contrasted (Keenan & Schieffelin 1976a, b) with
other language learning that children experience, where overgeneralized
forms, as with irregular plurals or verbal paradigms, are eventually dis­
carded. In the case of Sesotho word order the child retains each new construc­
tion, gradually compiling a set of grammatical devices to be used when
required by the discourse situation.
Data from this study indicate a predisposition for presentation of new
propositions before old information during the period between 2-3V2 years.
Thus, subjects and objects are postposed once they have been introduced in
lexical form. Verbal information is then presented first. This is consistent
with findings from Dutch and Italian youngsters (Snow 1972, Fava & Tiron-
dola 1977). It also appears that certain forms, such as postposed objects or
subjects, do not function primarily as afterthoughts or as clarification devices
for children, but rather as grammaticized ways of encoding discourse infor­
mation. Adults rarely make use of these word orders, except in cases of con­
trast or clarification. One of the productive environments for the use of these
constructions is adult-child interaction. Thus, we see a much larger propor­
tion of word order shifts in adult-child discourse than adult-adult conversa­
tions. Children likewise use alternative word orders more often when speak-
104 KATHERINE DEMUTH

ing to other children and talking to themselves than they do in talking with
adults.
The following table graphically illustrates the incremental progression
by which Sesotho speaking children acquire a working facility with distinct
word orders. Notice how the onset of usage of a given construction is gradual,
as indicated by the dotted line. There is a marked increase in the use of the
construction at the beginning of the solid line. The solid line indicates con­
tinued frequent usage of the form, while a return to the dotted line indicates
a relaxation in usage to those discourse contexts in which adults would also be
expected to use the construction.

Constructions

Preposed Objects

Postposed Objects

Postposed sub­
jects and Clefts

Questions-
U n m a r k e d -I- Cleft

G e s t u r e , Repeti­
tion and Prosody

2 2V2 3 4 5 Adult

Age

TABLE 1

It is important to note that Basotho children, along with children from

other language studies, never order words in an ungrammatical way, even at
initial stages of development. Why this should be, in a language which has the
 W O R D O R D E R IN S E S O T H O A C Q U I S I T I O N 105

potential for several different word orders, and in which adults use these
orders freely in speech to children, is cause for speculation. The fact that chil­
dren appear to 'experiment' with different constructions at various stages in
their development may provide additional evidence for positing some kind of
canonical shape (Slobin 1982, Slobin & Bever 1983) or template construct
(Erbaugh 1982, Demuth, Faraclas & Marchese 1986) which helps the child
organize language. There is a tendency for children across languages to use
some sort of basic word order initially. Once that shape has been firmly estab­
lished, a child then experiments with another. Only after that option has been
thoroughly explored does a child incorporate yet another canonical/template
form, and so forth. Language acquisition evidence from an extremely free
word order language such as Cayuga (Mithun, this volume), or Walpiri
(Bavin 1985), might provide us with evidence of variation in strategies chil­
dren employ in learning such languages.

NOTES
1) Data collection for this paper was supported by Fulbright-Hays and Social Science Research
Council doctoral dissertation grants. This manuscript was prepared while supported by NICHD
Training Grant #5732 HD07181 administered by the University of California at Berkeley.
Acknowledgements go to Knud Lambrecht, Mark Johnson and Russel Tomlin for comments,
while I take full responsibility for the material presented here.
2) Sesotho is a southern Bantu language spoken by the Basotho people in the countries of
Lesotho and South Africa.
3) See Demuth & Johnson 1986 for thorough discussion of the discourse and grammatical func­
tions of word order in Sesotho and Setswara.
4) This contrasts with Spoken French (Lambrecht, this volume), where certain postposed con­
structions are actually the unmarked case.
5) This expectation might not be upheld in societies such as Samoa (Ochs 1982), Kaluli (Schief-
felin 1979) or Maya Quiche (Pye 1980) where adults are minimally 'accomodating1 (Ochs & Schief-
felin 1983) in their physical and verbal interactions with children.
6) Many Bantu languages have a 'focus' marker (Givón, 1975) which is generally suffixed to the
subject-verb agreement marker (subject concord). Languages differ in the extent to which this
marker is grammaticized. The function of this particle in Sesotho is completely grammaticized,
providing foucs only a verb (i.e. in verb final constructions such as intransitives or transitives with
preverbal pronominal object clitic. Focusing and topicalizing other grammatical constituents in
Sesotho are primarily achieved through the reordering of words.
7) The focus marker -a in Sesotho is restricted to verb final indicative constructions like that in
example 2. (with some temporal adverbs permitted postnominally) and negative perfective con­
structions: ha kę-a pheha lijc - neg sc-foc cook food T didn't cook (the/any) food'.
8) For more detail on passive constructions see Demuth (1985).
106 KATHERINE DEMUTH

9) As noted in examples (3) and (4), and seen once again here in examples (18) and (19),
extraposed objects require the use of the preverbal object clitic. Preverbal information (subject
concord, focus marker, tense/aspect, object clitic) becomes progressively more phonologically dif­
ferentiated between ages 2 and 3, but it is especially from 2½ years onwards that object clitics
become more distinct. It is possible that this process might be a 'prerequisite' for being able to
extrapose objects - an additional factor to be considered in the explanation for why these two con­
structions became most productive only after extraposition of subjects.

REFERENCES

Bates, E. & B. MacWhinney. 1979. "A functionalist approach to the acquis­

ition of grammar'. In E. Ochs & B.B. Schieffelin (eds.). Developmental
Pragmatics, New York: Academic Press.
Bavin, E. & T. Shopen. 1985. Walpiri children's comprehension of discon­
tinuous word order, ms. La Trobe University.
Bloom, L. 1970. Language Development: Form and Functions In Emerging
Grammars. Cambridge, Mass.: MIT Press.
Brown, R. 1973. A First Language: The Early Stages. Cambridge, Mass.:
Harvard University Press.
Bowerman, M. 1973. Early Syntactic Development: A Cross-Linguistic Study
with Special Reference to Finnish. Cambridge, Mass.: Cambridge Univ.
Press.
Bresnan, J. & S. Mchombo. 1986. Verb agreement in Bantu: a case study in
grammatical theory and discourse function. ms. Stanford Univ.
Byarushayo, E. & S. Tenenbaum. 1978. "Agreement and word order: A case
of pragmatics." In Haya BLS vol. II.
Chafe, W. (this volume). "Cognitive constraints on information flow."
Demuth, K. In preparation. "Independent pronouns as contrastive topics in
Children's Sesotho."
. 1983. Aspects of Sesotho Language Acquisition. IULC, Bloomington.
. 1985. "Subjects, topics & agentivity: The early use of passive in
Sesotho." Paper presented at the Conference on Language Development,
Boston University.
Demuth, K., N. Faraclas & L. Marchese. 1986. "Niger-Congo noun class
and agreement systems in historical and acquisition perspective." In
Noun Classes and Categorization, C. Craig ed. Amsterdam: John Benja­
mins.
Erbaugh, M. 1982. Coming to order: Natural Selection and the Origins of Syn­
tax in the Mandarin Speaking Child. Ph.D. dissertation, Univ. of Califor-
 WORD ORDER IN SESOTHO ACQUISITION 107

nia at Berkeley.
Fava, E. & G. Tirondola. 1977. Syntactic and pragmatic regularities in Italian
child discourse: Grammatical relations and word order. Unpublished ms.,
Institute di Glottologia, Italy.
Givón, T. 1975. "Focus and the scope of assertion: Some Bantu evidence,"
Studies in African Linguistics 6:185-205.
Gruber, J. 1967. "Topicalization in child language," Foundations of Lan­
guage vol. 3, no. 1:37-65.
Kunene, E.L. 1975. "Zulu pronouns and the structure of discourse". Studies
in African Linguistics 6:171-182.
Keenan, E.O. & B.B. Schieffelin. 1976a. "Foregrounding referents: A
reconstruction of left dislocation in discourse." In Proceedings form the
2nd Annual Meeting of BLS.
. 1976b. "Topic as a discourse notion: a study of topic in the conversations
of children and adults." In Ch. Li (ed.) Subject and Topic. New York:
Academic Press.
Lambrecht, K. (this volume). "On the status of canonical SVO sentences in
French discourse."
Maratsos, M. & M.A. Chalkley. 1980. "The internal language of children's
syntax: The ontogenesis and representation of syntactic categories." In K.
Nelson ed., Children's Language vol. 2. New York: Gardner Press.
Mithun, M. (this volume). "Is basic word order universal?"
Ochs, E. 1982. "Talking to Children in Western Samoa." Language in Society
11:77-104.
Ochs, E. & B.B. Schieffelin. 1983. "Language acquisition and socialization:
Three developmental stories and their implications." In R. Shweder & R.
LeVine eds. Culture and its Acquisition.
Pye, C. 1980. The Acquisition of Grammatical Morphemes in Quiche Mayan.
Ph.D. diss., Univ. of Pittsburgh, Pennsylvania.
Schieffelin, B.B. 1979. How Kaluli Children Learn What to Say, What to Do
and How to Feel: An Ethnographic Study of the Development of Commu­
nicative Competence. Ph.D. diss., Colombia University, New York.
Slobin, D.I. 1966. The acquisition of Russian as a native language. In F.
Smith & C.A. Miller eds., The Genesis of Language: A Psycholinguistic
Approach. Cambridge, Mass: MIT Press.
. 1982. "Universal and particular in the acquisition of language." In E.
Wanner & R. Gleitman eds., Language Acquisition: The State of the Art.
Cambridge, Mass.: Cambridge Univ. Press.
108 KATHERINE DEMUTH

Slobin, D.I. & T. Bever. 1983. "Children develop canonical sentence

schemata." Cognition.
Snow, C. 1972. "Mothers' speech to children learning language." Child
Development 43:549-565.
--. 1978. "The role of social interaction in language acquisition." Minnesota
Symposium on Child Development, 12 Hillsdale, New Jersey: Lawrence
Erlbaum.
Stucky, S. 1985. Order in Makua syntax. Garland Press.
Tenenbaum, S. 1977. "Left and right-dislocations." In E.R. Byarushayo, A.
Duranti & L. Hyman (eds.), Haya Grammatical Structure, SCOPIL 6.
Watters, J. 1979. "Formal correlates of focus word order." In L. Hyman ed.,
Aghem Grammatical Structure. SCOPIL 7.
A UNIFORM PAUSE AND ERROR STRATEGY FOR NATIVE AND
NON-NATIVE SPEAKERS

MARY S.ERBAUGH
University of Oregon

1. Overview of the Foregrounding Hierarchy

Both native and non-native speakers of English pause and err at the
same places in their oral descriptions. The relative importance of the event
being described is critical in prompting speakers to pause, err, or correct
themselves, while language background has astonishingly little influence.
Although non-native speakers pause more and make more errors, both
natives and non-natives have the most difficulty at precisely the same points
in their descriptions of an animated cartoon. These pivotal points were
the most important events in the story, which would not make sense without
them. Regardless of language background, all speakers mentioned the same
core set of events, 1
Paradoxically, the pivotal events were also the most difficult for people
to describe fluently. More peripheral, backgrounded events, and editorial
comments which were outside the plot line, were described more fluently.
This indicates that all speakers share a common, cognitively-based strategy
for choosing which actions in an ongoing scene are most worthy of mention.
In this study, thirty native and non-native speakers described a simple car­
toon of a crab chasing a goldfish. All human beings perceive such prototypical
events in a similar way. Naturally, more deliberate or artistic speech is often
available to embellish the basic descriptions.
Framing a description of the pivotal events in a narrative requires the
greatest amount of planning and effort. The difficulty in overall planning
depletes the resevoir of energy available for fluent speech, so descriptions of
pivotal event are relatively error-prone. All speakers a share a common
monitor which rations their linguistic energy; they only add background and
editorial comments if they have an energy surplus. If they do, then their
 GRAPH 1
DYSFLUENCY RATES FOR PIVOT, F O R E G R O U N D , BACKGROUND, AND O
(as a percentage of all dysfluencies, all errors, and all pauses)

ALL DYSFLUENCY ERROR

PIVOT FORE BKGD OTHER PIVOT FORE BKGD OTHER PIVOT

r=.99 r=.99

■ = NON-NATIVE
• = NATIVE
FOREGROUNDING AND BACKGROUNDING 111

GRAPH 2
TYPES OF DYSFLUENCY
(In number of dysfluencies per proposition)
112 MARY S. ERBAUGH

background and editorial comments will generally be fluent, and remarkably

free of pauses and errors.
Regardless of language background, all speakers allocate their scarce
linguistic planning, production, and monitoring energy according to a com­
mon, cognitively-based Foregrounding Hierarchy:
PIVOT>FOREGROUND>BACKGROUND>OTHER
As Graph 1 shows, both pause and error patterns are almost identical for
native and non-native speaekrs, (r = .99). The Foregrounding Hierarchy
overrides language background.
Both natives and non-natives were far more likely to pause for more planning
time than to err, as Graph 2 demonstrates. While non-natives had three times
as many dysfluencies as natives, both groups shared the same 3:2 ratio of
pauses over errors. Natives and non-natives not only share the same prefer­
ence for pausing rather than committing an error, they share the same cogni­
tively-based set-point for determining when to pause longer or more often
rather than commit an error.
Language repair is also rationed to the Foregrounding Hierarchy. Pivotal
information is the most important, the most error-prone, and also the most
likely to be corrected. Foregrounded information is less likely to be repaired,
and background information the most unlikely.
If the Foregrounding Hierarchy had no psychological reality, we would
see error and pause rates scattered uniformly or randomly across all types of
propositions, possibly according to individual speaker style. In fact, the
Foregrounding Hierarchy predicts dysfluency at highly significant levels,
while the patterns for native and non-native speakers are almost perfectly
correlated, (typically r = .99).
This result is surprising since a syntactically-based argument would pre­
dict that foregrounded propositions should be easy and backgrounded ones
hard because: 1. Foregrounded information typically appears in declarative
main clauses. These use the most common, regular, verb paradigms. Word
order is also highly regular and predictable. Production should be highly
automated, well-practiced, and error-free. 2. Backgrounded information, on
the other hand, tends to appear in embedded or relative clauses. Word order
and verb forms are often archaic or exceptional. These clauses must be coor­
dinated with a main clause which must be maintained in short-term memory
for matching. Production should be much less automatic, and therefore more
error-prone.
 FOREGROUNDING AND BACKGROUNDING 113

An extensive series of experiments unsuccessfully attempted to measure

psycholinguistic difficulty by counting the number of grammatical transfor­
mations required to produce sentences which deviate from a simple subject-
predicate core. 2 According to this model, backgrounded information, which
typically appears in relative clauses, should be less fluent than the simpler
main clauses. In fact, the opposite is true for speech production, and likely for
speech comprehension as well.
This is because in a natural setting, people are able to avoid complex syn­
tactic and rhetorical forms by shifting down from a tightly-structured syntac­
tic mode to the more simply structured pragmatic mode described by Givón
(1979). They can simplify still more to a pidgin or telegraphic speech if their
hearer needs extra simplification (as when talking to foreigners or small chil­
dren) , or when the speaker is fatigued, under stress, or speaking a foreign lan­
guage. People often stumble, pause, change their minds, and misuse words as
they struggle with scripting the central characters in the everyday dramas they
describe. But once the main cast has its movements blocked out, scripting the
rest only elaborates a firm, clear, story line. The dramatist only orders up
extra characters, costumes, props, and settings if he or she can afford to do so.
Mature speakers are so prudent in their budget assessments that they rarely
overspend their linguistic capital. They typically remain intelligible to the
end, even when they are speaking a foreign language under highly pressured
conditions.

2. Methods
I analyzed the thirty oral, online descriptions of an animated cartoon
which Russell Tomlin collected from native and non-native speakers of Eng­
lish.3 Thirty students watched a 108 second videotaped segment from an ani­
mated cartoon which had music but no narration. The cartoon shows a
goldfish trying to escape capture by a crab and other fish. The subjects were
encouraged to describe the cartoon in as much detail as they could while they
were watching the TV screen. All the students were enrolled at the University
of Oregon. The fifteen native speakers were all between eighteen and forty
years old. The fifteen non-natives had all scored above 500 points on the Test
of English as a Foreign Language; all had been in the U.S. for at least two
years. Their native languages included Korean, Arabic, Japanese, Mandarin
and Spanish.
Tomlin transcribed the audiotaped narratives, including all false starts,
errors, and repairs, as well as all pauses. Pause times of one second or longer
114 MARY S. ERBAUGH

are measured and included. Tomlin segmented the utterances into proposi­
tions. Propositions are defined as semantic units consisting of a predicate and
arguments about which a truth value can be obtained. These include full
clauses, as well as partial clauses for which missing arguments are readily
recoverable. Infinitive clauses, participial clauses, and nominalizations did
not count as propositions. Defined this way, propositions are a basic unit of
memory in human cognition, discourse production, and comprehension. 4
Chafe (this volume) suggests that intonation units which contain a single
proposition are the fundamental unit for language production, and that sen­
tences are a secondary, rhetorical development.
Tomlin coded all propositions into the following foregrounding categories:
PIVOTAL INFORMATION the most important events in the narrative;
the most general description of a significant
event.
FOREGROUNDED propositions which describe successive
events in narrative.
BACKGROUNDED propositions which elaborate pivotal or
foregrounded events. Includes motives,
and some locations.
OTHER propositions outside the narrative
sequence, including editorial or evaluative
comments.
SAMPLE CODINGS FOR A NATIVE SPEAKER
The goldfish goes into some coral, no sponge, PIVOT
and the shark follows him, FOREGROUND
can't find him/ FOREGROUND
skulking around in the sponge, BACKGROUND
trying to find him BACKGROUND
Lot of action here. OTHER
Can't quite keep up with this. OTHER
Tomlin also elicited delayed oral and written descriptions. I analyzed the
online set because of the greater likelihood of error and pause. 5 However,
both the online and delayed narratives show the same pattern of putting the
pivotal propositions first in the sentence, with background and other infor­
mation following it. Tomlin found no major difference in the narratives if the
subjects watched the whole cartoon; both plot and musical cues seemed neg-
 FOREGROUNDING AND BACKGROUNDING 115

ligible for this film. Tomlin also found no differences if speakers faced an
interviewer rather than a tape recorder as they described the film.6 7
I define dysfluencies to include both pauses and speech errors. I coded
Tomlin's transcripts in order to calculate dysfluencies per proposition for the
following dysfluency types:
PAUSE TYPES: 1. Filled: e.g. "um" "ah". 2. Empty: momentary, or
timed if longer than one second.
ERROR TYPES 1. REPAIR: a. Repeats of a phoneme, syllable, full
word, or phrase, b. Changes of a phoneme, word or
phrase. 2. OMISSION: of an obligatory grammatical
morpheme or word. 3. MISUSE: a. Word Choice e.g.
"golden fish" for "goldfish." b. Grammar includes errors
in agreement, plural, tense, time aspect, determiner,
prepositions, and word order.

3. Pauses
3.1. Overall Pause Pattern
Natives and non-natives maintained the same 3:2 ratio of pauses to
errors, as Graph 2 shows. Clearly both natives and non-natives prefer pause
to error; both pause longer if they are unsure of either their analysis of the sit­
uation or its linguistic packaging. However, the non-natives showed much
more individual variation in both pause rate and error rate than the native
speakers did, as Graph 3 shows. Although all the non-natives were fluent
speakers, their language skills still varied considerably.
Many more pauses occurred at major event boundaries in the story, before
pivot propositions, and before main clauses than elsewhere. 8 Clearly, speak­
ers can control how much they pause, since the speed of speech is determined
by the amount of pause rather than the rapidity with which individual seg­
ments are pronounced. 9 Pause is a reliable measure of the difficulty of analyz­
ing and verbalizing events. The hardest events to desribe are the ones which
are both most important and least expected, the ones that literally "give one
pause." Goldman-Eisler found speakers paused much more when they were
asked to supply an interpretation of a New Yorker cartoon than when they
were simply asked to describe it.10 Pause rates also vary because of the highly
diversified functions which both filled and empty pause can carry perform.
116 MARY S. ERBAUGH

GRAPH 3
INDIVIDUAL VARIATION

3.2. Empty Pauses

The most obvious function for empty pauses is dramatic emphasis.
Empty pauses punctuate several different rhetorical styles, including
reprimands, storytelling, and political speeches, in French, English, and Ger­
man, as well as in other languages.11 While the pressure of online description
crowded out most stylistic flourishes in the cartoon descriptions, a few native
 FOREGROUNDING AND BACKGROUNDING 117

speakers did add fillips of storytelling style such as, "our hero," or "little does
he know that...." which contrasted with their more usual event-by-event
inventories. Although it is possible that non-natives who were skilled in a
native-language storytelling style would be able to transfer this into their
online descriptions in a second language, I did not find any such stylistic var­
iations in the non-native descriptions.
Empty pauses also serve the much more basic purpose of giving the
speaker additional planning time. In this sample empty pauses indicate addi­
tional planning effort rather than stylistic control. Both natives and non-
natives found the cartoon equally easy to understand, since they both
described exactly the same core set of events. 12 Both natives and non-natives
followed the same strategy of pausing silently nearly three times as often as
they used filled pauses, as Graph 4 shows. Both groups were particularly
likely to pause silently during rapid, confusing, transitional segments of the
cartoon, such as during a whirling, circular chase scene.

GRAPH 4
RATIO OF EMPTY TO FILLED PAUSES
118 MARY S. ERBAUGH

However the non-natives still paused three times as often as the natives did,
as Graph 5 shows.

GRAPH 5
PAUSES PER PROPOSITION

Both natives and non-natives paused far more often in pivot and foreg­
rounded propositions, as shown in Graphs 6a and 6b. In pivot and foreground
clauses, non-natives paused silently once during every proposition, while
even natives paused during one out of three pivots. The empty pause rate
drops off precipitously for backgrounded and other propositions. The non-
native pause rates drop much more sharply because they are much less likely
than natives to elaborate if they cannot do so fluently.
Both natives and non-natives followed the initial pause pattern described by
Chafe (this volume). In initial pauses, the speaker pauses for planning
before the beginning of new proposition, particularly if it introduces a new
topic or event. In this study, 27% of the native speakers' pauses were initial
pauses, while 43% of the non-native pauses were initial, which indicates the
non-natives' need for more linguistic planning time. 13 However, the initial
FOREGROUNDING AND BACKGROUNDING 119

GRAPHS 6a and 6b
PAUSES PER PROPOSITION

6a NON-NATIVE
120 MARY S. ERBAUGH

GRAPHS 6a and 6b
PAUSES PER PROPOSITION

6b NATIVE

pauses followed the Foregrounding Hierarchy exactly as the other pauses and
dysfluencies did (r - .99). In addition, the natives and non-natives both fol-
 FOREGROUNDING AND BACKGROUNDING 121

lowed the Foregrounding Hierarchy in precisely the same way (r = .99) Chafe
also found a similar increase in errors at major event boundaries.

3.3. Filled Pauses

Filled pauses function very differently from empty ones. Filled pauses
hold the floor while the speaker searches for the specific word which he has
already ordered up with a typological description, but not has not yet been
delivered to "the tip of his tongue." Different fillers serve different functions:
"urn" and "ah" mark a successful memory search, while "oh" signals selection
of an example. 14 Virtually all the filled pauses in this study contained "urn" or
"ah" because the pressured task and the lack of a conversation partner pre­
cluded a more leisurely search for examples and clarifications. The native
speaker filled initial pauses with "urn" or "ah," where the non-natives would
typically remain silent. Native speakers were also much more likely to start
new propositions with filler conjunctions such as "and then."
In contrast to the large gap between native and non-native pause rates,
filled pause rates were very similar, as Graphs 6a and 6b show. This indicates
that the search for a specific word to express an already-formulated proposi­
tion is not much more difficult for fluent non-natives than it is for natives.
Still, both groups found word-search for pivots harder than for foregrounds,
and so on down the scale. Bringing an already-ordered word out of storage
seems to be a relatively mechanical, universal, neurologically pre-deter-
mined task, once the difficult jobs of framing major case roles and syntactic
structures are completed.

4. Errors
4.1. Error Overview
Unlike the pause rates, where natives and non-natives patterned alike, the
two groups had exactly inverse error patterns, as shown in Graph 7.
For natives, repairs were more frequent than word omissions, and omissions
more frequent than syntactic or semantic misuse. For non-natives, the pat­
tern was reversed. Not surprisingly, the non-natives were most likely to mis­
use forms, less likely to omit them, and proportionately least likely to repair
them, although they still made a greater number of self corrections than the
natives did.15 There is apparently a rather fixed ceiling to the human capacity
for self-monitoring and self-correction.
Even so, the error rates for these fluent non-native speakers were closer
122 MARY S. ERBAUGH

GRAPH 7
ERROR TYPES

to the natives' than might be expected. Both groups were about equally likely
to correct themselves: natives repaired 1 in 10 propositions, and non-natives
1 in 7. There was more difference in omissions: natives omitted only about 1
word for every 14 propositions, while non-natives omitted more than 1 word
every 5. As expected, misuse rates differed the most: natives misused only 1
word every 100 propositions, while natives misused every 3 propositions.
 FOREGROUNDING AND BACKGROUNDING 123

4.2. Repairs
Repairs include both changes and repeats of a word or phrase. As Graph 8
shows, both natives and non-natives were about equally likely to make
repairs. Both divided their repairs evenly between changes and repeats. Both
allocated their repairs according to the Foregrounding Hierarchy.
124 MARY S. ERBAUGH

This indicates that: 1) natives and non-natives have similar monitor and
repair capacity and 2) changes and repetitions are about equally favored,
regardless of language background. This argues for a very similar, neurolog-
ically set, maximum monitoring capacity for allocating repairs. It also
suggests a similarly universal restriction on the number of repair-interrup­
tions a speaker can tolerate or produce. If he or she repairs more than this
maximum, neither the speaker nor the hearer will be able to hold the sen­
tence in working memory. But if he repairs too much less, the sentence
becomes unintelligible.
We see the same familiar Foregrounding Hierarchy for both native and

GRAPH 9
OMISSIONS
 FOREGROUNDING AND BACKGROUNDING 125

non-native repairs. While both allocate most of their repair budgets to the
more informative pivot and foreground propositions, the non-natives use a
more ruthless triage, since they divert their repairs there almost exclusively.
Native speakers, in contrast, were able to extend themselves enough to repair
an occasional background or other proposition.

4.3. Omissions
Both natives and non-natives omitted grammatically obligatory words and
morphemes, especially informationally redundant determiners and auxiliary
verbs. Graph 9 shows that pivotal and foregrounded propositions are the
most difficult, since both natives and non-natives were most likely to omit
redundant morphemes there. Neither group made any omissions in the edito­
rial other propositions.
Not surprisingly, the non-natives omitted more overall, about one word or
morpheme every 5 propositions; the natives only omitted 1 every 14. The gap
between native and non-native omissions is widest for pivots, which shows
that pivots were much harder for the non-natives than for natives. While
natives only omitted one element every 14 pivots, non-natives omitted about
1 in 4. Still, both groups screened omissions according to the same scale of
informativeness. For both natives and non-natives, some morphemes and
some propositions are created more equal than others.

4.4. Misuse Errors

The biggest difference between the native and non-native speakers
showed up in their misuse errors, as Graph 10 indicates.
The fifteen native speakers only made a total of five misuse errors. These
included: three bad word choices, one agreement error, and one pluralization
error. Three misuses came in pivots, and two in background propositions.
These breakdowns developed largely because of the pressure of online
description. The situation is very different for non-native speakers, since
their English is neither perfectly mastered nor fully automatic. Competence
errors appear in their performance, yet they still adhere to Foregrounding Hier­
archy. For non-natives, almost one of every three pivotal propositions had a
misuse error, as did almost 1 in 4 of the foregrounded ones. The error rate
drops tremendously for backgrounded information: only 1 in 12
backgrounded propositions contains an error. Neither the natives nor the
non-natives had any misuse errors in their optional, other propositions.
126 MARY S. ERBAUGH

GRAPH 10
MISUSE ERRORS

Misuse
errors
per
proposition

PIVOT FORE- BACK- OTHER

GROUND GROUND

= Non-native
= Native
r = .30

If we look at the ranking of error types for non-natives, we see that virtu­
ally all the errors are local errors which do not impede communication,
though they do violate the norms for surface syntactic agreement. Non-
idiomatic word choices are also common.
 FOREGROUNDING AND BACKGROUNDING 127

RANKING OF MISUSE ERRORS FOR NON-NATIVES

ERRORTYPE FREQUENCY
Agreement 21
Word Choice 16
Tense 13
Order 6
Determiner 5
Preposition 3
Plural 2
Time 1
Aspect 1

With the possible exception of the two errors in time and aspect marking,
these misuse errors are all performance problems in surface agreement,
rather than reflections of any cognitively different patterning for either sen­
tential relations or world view. The very redundancy of the misused forms
neutralizes the effect of their concentration in the informationally crucial
pivotal propositions.

5. Conclusions
All the evidence points to the following conclusions:

1. The Foregrounding Hierarchy has psychological reality for

describing prototypical events regardless of language background.
2. Pivotal and foregrounding propositions are much more alike than
backgrounded ones, since pivots and foregrounds describe obliga­
tory information. Other propositions serve a variety of very differ­
ent, optional functions.
3. For both natives and non-natives, pivotal utterances are both the
most necessary, and the most prone to pause and error. Neither
natives nor non-natives will describe events lower on the informa-
tiveness hierarchy unless higher-rated events have already been
described with some fluency.
4. Dysfluency is a single broad phenomenon which includes both
pause and error. When either natives or non-natives are under
stress, they become dysfluent in similar ways. Pivotal and fore­
grounded information is consistently the least fluently expressed; it
 MARY S. ERBAUGH

is also the most likely to be repaired. 16 All speakers have a monitor

which is set to produce more pauses than errors at a ratio of 3:2.
5. Pauses allow additional planning time, but filled and empty pauses
function differently. In this study, differences in pause use reflect
different linguistic processing needs, not different patterns in com­
prehension of events. Both natives and non-natives pause most
frequently for pivots. Empty pauses reflected additional linguistic
planning time rather than rhetorical flourishes.
6. Filled pauses indicate the time for lexical access. Natives and non-
natives have very similar filled pause rates. This indicates that the
difficult, non-automatic part of speech production is framing case
roles for the most important events. Lexical access reflects a per­
formance lag which is relatively unaffected by overall fluency or
native speaker background. There is also less individual variation
in filled pause rates than in empty pause rates.
7. A neurologically set monitor limits the maximum number of
repairs a speaker can produce, so that natives and non-natives are
both restricted to the same maximum number of repairs per prop­
osition. If this repair capacity is overloaded, the speaker will first
pause more, and then, after a maximum pause limit, err more.
8. Speakers would rather pause than err, and they would rather omit
a non-crucial word than misuse it. Even under time pressure,
natives only rarely make grammar or word choice errors. While
non-natives are more likely to err, they still make mostly local
agreement errors which do not disrupt the message.
9. All propositions are not created equal; pivots are more equal than
the others. A common, neurologically-developed monitor allo­
cates linguistic energy to express pivotal events at the expense of
all others, regardless of the speaker's language background.
Speakers are highly sensitive to their current balance. If they find
themselves overextended, they will retreat to a pragmatic mode.
They will not even attempt backgrounding or elaboration if their
accounts are overdrawn.
10. The foregrounding monitor may be disordered in certain sorts of
emotional and neurologically-based language disorders. It may
also be re-set for special stylistic effects. However for normal adult
speech, the foregrounding monitor transcends individual style,
cultural background, native language, and degree of fluency.
 FOREGROUNDING AND BACKGROUNDING 129

Further studies will be needed to determine how well the Foregrounding

Hierarchy predicts both delayed oral and written language, and listening
comprehension. But a strategy as powerful and universal as the foreground­
ing monitor is likely to be re-employed in as many language situations as pos­
sible.

NOTES
1) See also Tomlin 1984 and 1985.
2) Overview in Clark and Clark, 1977, pp. 57-92.
3) See Tomlin 1984 and 1985 for fuller description, and other studies.
4) Anderson and Bower, 1973; van Dijk and Kintsch, 1977.
5) See Nooteboom, 1980
6) Tomlin, personal communication.
7) The interviewer was always in the room, but sat behind the speaker as he or she faced the
screen and the tape recorder in the online set.
8) See Tomlin 1984, 1985 for discussion of event boundaries.
9) Goldman-Eisler, 1968, MacKay, 1980.
10) Goldman-Eisler, 1968, p. 59.
11) See Duez, 1982, for discussion of pauses in French and English. Goldman-Eisler finds strik­
ingly similar pause patterns in both spontaneous and reading style in English, French and German,
(1968, pp. 78-9).
12) See discussion in Tomlin 1984, 1985. It is possible that young children might have described
a different set of events.
13) Goldman-Eisler found 31% initial pauses for descriptions of cartoons, and 43% for sum­
maries, (1968, p. 62.)
14) James 1973.
15) See Levelt for discussion of self-correction.
16) This would not necessarily have to be the case. It is possible to imagine, for example, that
pivotal propositions would have the lowest error rate but be best-monitored and most likely to be
repaired.

REFERENCES

Anderson, J. & G. Bower. 1973. Human Associative Memory. New York:

Wiley.
Boomer, D.S. 1965. "Hesitation and grammatical encoding." Language
130 MARY S. ERBAUGH

and Speech. 8:148-158.

Chafe, Wallace, [this volume]. "Cognitive constraints on information
flow".
Clark, Herbert H. & Eve V. Clark. 1977. Psychology and Language. New
York: Harcourt, Brace, Jovanovich.
Dressier, Wolfgang, ed. 1977. Current Trends in Textlinguistics. Berlin: Wal­
ter de Gruyter.
Duez, Danielle. 1982. "Silent and non-silent pauses in three speech styles."
Language and Speech. 25:1.
Fromkin, Victoria A. ed. 1980. Errors in Linguistic Performance: Slips of the
Tongue, Ear, Pen and Hand. New York: Academic Press.
Givón, T. 1979. "From discourse to syntax: grammar as a processing strat­
egy." In: T. Givón, ed. Discourse and Syntax: Syntax and Semantics. Vol­
ume 12, pp. 81-114. New York: Academic Press.
Goldman-Eisler, F. 1968. Psycholinguistics: Experiments in Spontaneous
Speech. London: Academic Press.
James, Deborah. 1972. "Another look at, say, some grammatical constraints
on, oh, interjections and hesitation." In: Papers from the Ninth
Regional Meeting, Chicago Linguistic Society. Pp. 162-172.
Levelt, Willem J.M. 1983. "Monitoring and self-repair in speech." Cogni­
tion. 14:41-104.
MacKay, Donald G. 1980. "Speech errors: retrospect and prospect." In:
V.A. Fromkin, ed. Errors in Linguistic Performance. New York:
Academic Press.
Nooteboom, S.B. 1980. "Phonological and lexical errors in spontaneous
speech." In: V. Fromkin, ed. Errors in Linguistic Performance. New
York: Academic Press.
Tomlin, Russell S. 1984. "The treatment of foreground-background informa­
tion in the on-line descriptive discourse of second language learners."
Studies in Second Language Acquisition. 6.2.
. 1985. "Foreground-background information and the syntax of subordi­
nation: evidence from English discourse." Text 5.
van Dijk, Teun A. & Walter Kintsch. 1977. "Cognitive psychology and dis­
course: recalling and summarizing stories." In Wolfgang Dressier, ed.
Current Trends in Textlinguistics. Berlin: Walter de Gruyter.
 THE GRAMMATICAL MARKING OF THEME
IN ORAL POLISH NARRATIVE*

VANESSA FLASHNER
University of California, Los Angeles

1. Introduction
The study of themehood in narratives shows how reference to a specific
entity is grammaticized within the discourse context of a story structure.
Theme, here, refers to the protagonist in a story; sub-theme refers to other
characters that take part in the story. This paper describes the use of pro­
nouns, nouns, modifying (descriptive) adjectives, and zero pronoun
anaphora in subject position to refer to theme and subtheme in an oral Polish
narrative. The study shows that continuous predictable themes are rep­
resented by anaphoric zero pronoun. When themes are overtly expressed, it
is due to several factors: narrative discontinuities in the text, changes in
speaker's perspective, and the organizing principle of contrast to develop
background paragraphs.
In Russian, a related Slavic language, Nichols (1984) examined thematic
reference for contemporary prose narratives. She found that the preferred
marking for uninterrupted theme was anaphoric zero pronoun. Overt mark­
ing of theme, however, was predictable under certain conditions: narrative
discontinuities and changes in perspective. The narrative discontinuities were
caused by an intervening potential theme (i.e., secondary characters),
change in syntactic relations, a topic shift, a change in temporal reference, or
a break in narrative boundaries (i.e., episode change and direct speech).
Changes in perspective reflected the difference between thematic and non-
thematic viewpoint. Thematic viewpoint was expressed through the use of
names of characters and non-thematic viewpoint or external viewpoint
through the use of descriptions for characters and patronymics.
Due to similarities in the grammars of Polish and Russian, it is predicta­
ble that contemporary Polish narratives should reflect similar mapping of
132 VANESSA FLASHNER

theme onto grammatical form. The preferred marking of continuous theme

in Polish, then, ought to be anaphoric zero pronoun. Polish, like Russian,
inflects verbs for tense, person, and number in the present. Polish, further­
more, is more explicit as to subject reference in the past and analyzed future
verb forms by also showing gender in addition to the person and number
marking of Russian. Because of the explicit subject reference marked on the
verb through verbal agreement, the overt use of subject pronouns is consid­
ered redundant and mainly a mark of emphasis. Yet, although redundant,
pronouns are surprisingly a feature of spoken Polish.1
Comparison between first person oral and written narratives in Polish
shows a remarkable contrast in overt thematic marking. For example, where
the present oral narrative shows 52% (140/267) overt use of theme, a written
narrative (Mayewski 1975) shows only 1% (2/225) overt marking. 2 This
paper, in examining the function of overt themes, helps explain the discre­
pancy between the high frequency of overt reference in the oral data and its
low frequency in the written language. It claims that despite the obvious
thematic marking shown through verbal agreement, redundant pronouns are
used for three reasons: (1) to delineate the boundaries in a narrative, (2) to
structure contrast relationships, and (3) to show the speaker's personal
involvement. More importantly, each of these factors accounts for overt
thematic marking in unique (i.e. non-overlapping) portions of the text.
Three discourse-pragmatic principles are proposed to account for the
grammatical marking of theme in an oral narrative. 3
1. A textual principle of economy states that in texts reference to con­
tinuous themes is marked by zero pronoun anaphora and to inter­
rupted themes by full nouns for subthemes and pronouns for
theme.
2. A textual principle of expressivity states that in texts the overt use
of themes in sequential clauses develops an idea through com­
parison or contrast.
3. An interpersonal principle of cooperation states that a speaker's
use of marked as opposed to unmarked forms when referring to
themes reflects his/her attitude toward the referent.
The following section introduces the data and describes the structure of
the narrative. Sections 3, 4, and 5 relate thematic use to the structure of the
narrative: sections 3 and 4 describe the use of theme and narrative discon­
tinuity; section 5 examines the overt use of theme to develop background
 THE GRAMMATICAL MARKING OF THEME 133

paragraphs of comparison/contrast. Section 6, in contrast, relates thematic

use to the speaker-hearer dimension and examines the use of theme to show
perspective. Finally, section 7 presents a brief conclusion.

2. Data and Structure of the Narrative

Data. A sixty minute tape of one woman's personal oral narrative pro­
vided the data for this study.4 The material was taped via audio cassette and
transcribed according to Chafe (1980). Clauses or phrases ending with a slight
fall in intonation are punctuated with a comma. Clauses ending with a drama­
tic fall in intonation are marked with a period. Short pauses between words
not extending beyond .5 seconds are separated by periods, and longer pauses
are timed and indicated in parentheses in the transcript. Question marks indi­
cate clause final rising intonation and brackets are used for uncertain trans­
criptions.
Structure. The structure of a narrative cannot be accounted for without
considering a speaker's goals in telling the story. Levy (1979) has noted that
what is called the discourse structure of a text is a complex mixture of inter­
personal, textual, and ideational goals. 5
A speaker has a number of ideas he wants to express and a corresponding set
of communicative goals. Some of these goals (called IDEATIONAL goals)
are concerned directly with the communication of these ideas or proposi­
tions: some (called TEXTUAL goals) are concerned with the weaving of
these ideas into a coherent text: and still others (called INTERPERSONAL
goals) deal with presentation of self in relation to the hearer, with matters of
status and attitude. (Levy 1979: 197)
The speaker's goal in relating this story is to tell how she got from one
place to another. Specifically, the narrative describes a young woman's flight
from a town in Eastern Poland to a town in Western Russia. Since the story
takes place in 1941 during World War II at the time of the German invasion
of the Soviet Union, the protagonist's flight to a sanctuary beyond the Ger­
man-Russian front is repeatedly frustrated as she is unable to outrun the Ger­
man advance. Arrival at a town already or about to be besieged leads to her
setting out for a new place. The story, then, consists of a chain of ten episodes
each separated by pauses of two to five seconds in length, and each describing
her adventures as she traveled from one town to the next. Each episode rep­
resents in Levy's framework one ideational goal (see Figure 1).
n —number of episodes
ns — number of scenes in episodei
n — number of paragraphs in scene j of episode i
nc — number of clauses in paragraph k of scene j which belongs to episode i
 THE GRAMMATICAL MARKING OF THEME 135

The speaker realizes her textual goals in the way she structures the
scenes that comprise an episode (see Figure 1). Each episode is a bundle of
loose event (i.e. simple action) clauses and several scenes which focus on
specific events that occur during travel from one town to another. Every
scene has a core event structure. The core event structure described here
is similar to that described by Mandler and Johnson (1977) for generative
story grammars with the exception that their story grammar does not allow
for the inclusion of background digressions, which were considerable
in the present text. 6 Furthermore, digressions which often express the
interpersonal goals of a story (e.g. moods, reactions, feelings, motives,
etc.) are structured and must be accounted for in a discussion of oral
narratives. Each event structure, then, consists of an action done to the pro­
tagonist, the protagonist's mental reaction to the instigating event, and an
action which is the protagonist's physical follow-up action to his/her reaction.
The plot structure looks as follows: CORE EVENT STRUCTURE:
INSTIGATING ACTION — MENTAL REACTION — FOLLOW UP
PHYSICAL ACTION
The following section describes the interaction between thematic refer­
ence and textual organization in terms of the story's episodes, their scenes
and core event structures, and the introduction of quoted speech.

3. Theme and Discourse Continuity

Because discourses usually consist of chains of clauses about one topic,
reference to that topic is considered to be the unmarked case. Topic con­
tinuity or predictability is thus expected in cohesive units of discourse. The
marked case is topic change or discontinuity. Discontinuous or new topics are
surprising and therefore need adequate coding to clearly identify their refe­
rents. Givón (1983: 18) describes the economics of topic identification in the
following way. "The more disruptive, surprising, discontinuous or hard to
process a topic is, the more coding material must be assigned to it." He has
identified the following cross-linguistic scale for coding topic accessibility
(1983:17).
Most Continuous/Accessible Topic
zero anaphora
unstressed/bound pronouns or grammatical agreement
stressed/independent pronouns
R-dislocated DEF-NP's
neutral-ordered DEF-NP's
136 VANESSA FLASHNER

L-dislocated DEF-NP's
Y-moved NP's ('contrastive topicalization')
cleft/focus constructions
referential indefinite NP's
Most Discontinuous/Inaccessible Topic
Oral narratives are continuous and have no visibly discrete units such as
chapters, sections, or paragraphs to mark topic change. Yet in oral narratives
we can define episode boundaries by identifying ideational goals and by not­
ing the coincidence of described goals with long pauses (2-5 seconds in
length). Given these clues to the internal structure of the oral narrative and
the findings of Givón and Nichols, it was predicted that continuous themes
would be represented by zero anaphora and discontinuous themes by full
noun phrases for subthemes and independent pronouns for the theme. 7 The
following will show that indeed narrative continuity/discontinuity predicts
the nature of thematic reference.
From this narrative's data two continua for topic continuity in Polish can
be derived. The following table accounts for thematic continuity in sub-
themes:
Highest Continuity
a. zero anaphoric pronoun/grammatical agreement
b. independent pronoun
c. modifying adjectives (e.g. this one, the old one)
d. full noun phrases
Lowest Continuity
In the theme, which is a first person narrator, continuity ranges from
anaphoric zero pronoun/grammatical agreement showing highest continuity
to the first person pronoun ja T showing lowest continuity, (this scale is dis­
cussed in part 6.)
Example (1) shows the use of anaphoric zero pronoun as continuous
theme marker after an initial overt introduction of theme. 8 In the passage, the
narrator digresses about herself as she waits to find a train that she can sneak
aboard.
(1) I nastacjiznów..stalpociqg And in the station once again
osbowy, there stood a passenger train,
(1.5) /' ewakuowano zony (1.5) and were evacuated wives
of (1.5). oficerów rosyjskich, of office. .Russian officers,
 THE GRAMMATICAL MARKING OF THEME 137

ja bylam strasznie 'zywotna, I was terribly energetic,

straszna taka wiesz, frightening you know,
wszystko ø moglam się dowie- everything / (0) could find out,
dziec,
ø biegalam tak, I (0) would run like that,
ø mowilam dobrze po I (0) spoke Ukranian well,
ukrainsku,
(1.2) i proszeciebie i ø (1.2) and listen and I (ø)
dowiajuje sie odjednego tarn find out from someone there
nastacji, at the station,
ze jest poçiag w tej Zmerynce, that there is a train in Zmerynko,
i on bedzie bedq ladowac and it will they will be
wnet, boarding right away,
For subthemes anaphoric zero also generally represents uninterrupted
theme. Below, the subtheme is a corpse that the narrator finds lying on the
road.
(2) i jak widzisz to mi zostalo and as you see it has remained
w pamięci w tej szosie, in my memory on this road,
jak tylko wysziam, just as I set out,
to bylo od naszego domu mo'ze it was from our house
wiem..stometrów, I guess one hundred yards,
lezal jeden (.5) zz który.. lay someone (.5) who
mial kulę w plecach, had a bullet in his shoulders,
tak 0 le'zalskulony, like so he (0) lay there bent over,
i tak 0 zginql. and so he (0) died.
Usually themes and subthemes are expressed as the syntactic subject of
a clause and therefore in the nominative case. However, themes and sub­
themes may also appear as inverse subjects in the dative case, as objects in the
accusative, genitive, or dative, or as objects of prepositions in other oblique
cases. There are two syntactic causes for the overt coding of theme or sub­
themes: a theme becomes overt when it repeats a theme that is not in the
nominative (subject) case, and a theme is overt after its introduction through
a presentative construction. In Polish, the presentative construction changes
usual SVO word order to (O)VS. The following example shows the use of the
pronoun on 'he' after the introduction of a character through a presentative
construction.
138 VANESSA FLASHNER

(3) ale do mego wagonu (J) and into my car (1) walks
wchodzi oftcer, an officer,
on uh laduje swoja 'zonę, he uh loads his wife (onto the
kilka worków chleba, train), several sacks of bread,
takie worki chleba, such large bags of bread,
i 0 musial byc jakis wielki, and he (0) must have been someone
great,
i z dziewczynq i sluzqcq, and with a daughter and maid,
to 0 musial byc jakis wielki he (0) must have been somebody
czlowiek, great,
(1) i ø zaladowal..ø and he (0) loaded thorn..he (0)
zalado wal ich. loaded them (onto the train).
In the next example, a subtheme, Blima, becomes overt after appearing as a
direct object in the accusative case ja 'her'. Switch reference does not account
for the use of the pronoun ona 'she' because verb agreement in the verb
pracowala 'she worked' codes for person, number, and gender.
(4) Ale smieszne bylo co, And what was funny was this,
ta Blima.. Blima byla na this Blima..Blima was in
matematyce, mathematics,
i 0 mieszkala w domu and she (0) lived in a
akademickim. dormitory.
A pózniej.. Tatus ja spotkal And later..Daddy met her
w Saratowie, in Saratov,
i ona pracowala w stolówce, and she worked in a mess hall,
i ona.. nosila takie majtki and she. .wore such underwear
które na gumie, which (were held up) by rubber band,
i 0 kradla kotlety albo kasze, and she (0) stole cutlets or kasha,
i ø przynosila Tatusiowi and she (0) brought (them) to Daddy
do jedzenia. to eat.
From the above we see that continuous themes are represented by zero
anaphora and that syntactic discontinuity causes the following thematic refe­
rent to be overt. Out of 20 cases of syntactic discontinuity in the data, 80%
(16/20) were followed by overt themes.

4. Narrative Discontinuities
The major causes of discontinuity, however, are due to the structure of
the narrative. Action discontinuity at episode and scene boundaries and in
 THE GRAMMATICAL MARKING OF THEME 139

the core event structure gives rise to overt thematic reference. Person discon­
tinuity at the introduction of direct speech causes overt thematization as well.
The following describes the nature of these discontinuities.
Change of Episode. Episodes represent the narrator's journey from one
destination (town, village, or city) to another. These major text boundaries
are represented by sentence final intonation and by long pauses of 2 to 5 sec­
onds in length. Overt theme is used at episode junctures because a new series
of events, united around the idea of getting to a certain destination, is
initiated at these points. There were ten episodes in the story: 70% (7/10)
showed overt pronoun themes.
In (5) the narrator describes how she sets out on her journey after a
friend comes to warn her to leave. The pronoun ja T is used as the narrator
leaves her home. Zero ( 0 ) is used for the digression on her leave taking and
ja T repeats the sentence indicating the beginning of the episode which takes
the character from her home town of Zloczow to the neighboring town of Tar-
nopol.
(5) Iona przyszla i môwi, And she came and says,
"Charlotta, wychodzimy, "Charlotta, let's leave,
wychodzimy ze Zlozowa, let's leave Zloczow,
wychodzimy musimy wyjsc let's leave we must leave
ze Zloczow a, Zloczow,
bo uh bo Niemcy wchodzq. " because uh because the Germans
are coming."
(2) No bylo nad ranem, (2) Well it was dawn,
ija wyszlam, and I eft,
nie 0 będę ci opowiadac jak I (0) won't tell you how
się 0 'zegnaiam z mamq, / (0) said goodbye to mother,
bo to 0 będę zaraz, because I (0) will start right away to
0 przestanę mówic na ten I (0) won't talk on that topic,
temat,
tojuz nie. not anymore.
(.6) więc uh ja wyszlam, (.6) well uh/left,
(1.5) I0 poszlismy znowu (1.5) And we (0) went once more
piechotq, on foot,
Another episode near the end of the story is shown in (6) where upon
arriving in the town of Krasnodar the narrator discovers that there is a
sanctuary nearby where she may find her university classmates. After a long
140 VANESSA FLASHNER

pause of 2.5 seconds and a few introductory clauses, the pronoun ja T is used
as she sets out toward this new goal.
(6) to mnie powiedzieli 'ze w so they told me
ze w tym, that in this,
ze jakies cos dwadziescia czy that some twenty or thirty
trzydziesci kilometrów od kilometers from Krasnodar
Krasn odaru jest jak is is some kind of solxoz,
sowcholz,
(1) i tam sa studenci ze (1) and there are students from
Lwowa, Lwow,
(2.5) no jak? (2.5) well then what?
to ale i byl wleczórjuz, but by then it was already evening,
i ja chcialam. .pójsc and I wanted to go on foot,
plechotq,
ho dwadziescia dwadziescia because twenty twenty
kilometrów, kilometers,
to dla mnie nie bylo juz wasn't for me
tak strasznie, that terrible,
Change of scene. Episodes are composed of a series of scenes, picture­
like frames of events, which are held together by loose event/action clauses
such as "and we walked on" or "we slept by the roadside." Each scene calls
up a new situation, thereby demanding that referents be reintroduced
through overt expression. Out of 31 scene junctures 81% (25/31) showed
overt pronoun use.
In the following passage the narrator describes a baker from whom she
and her friends hope to receive a night's lodging. The baker is first introduced
as a full noun in a background digression. The digression is followed with
intervening event clauses mentioning the narrator's and her friend's arrival at
the town of Podwoloczysk. A new scene opens up with the girls' arrival at the
baker's house. The reintroduction of the baker as a noun in the genitive case
is followed by the pronoun on 'he' in the nominative (subject) case. The fol­
lowing references to the baker are expressed through zeroes.
(7) A ta Zosia miala jakiegos And this Zosia had there
tam—
chlopca znajomego, a boy friend,
uh przedtym (który) before that one, (who) she knew
wiedziala 'ze
 THE GRAMMATICAL MARKING OF THEME 141

jego ojciec jest piekarzem, that his father is a

baker,
i 0 mieszka pod Woioczysko, and 0 lives near Woloczyskó,
i—mysmy, (.5) and-we, (.5)
nie wiem jak dlugo szlysmy, I don't know how long we walked,
spalypo drodze gdzies na slept somewhere in the
polach, fields,
juz nie pamiętam dokladnie, I don't now remember exactly where,
ale przyszlysmy do but we came to Podwoloczysko,
Podwoloczysk,
przyszlysmy do we came to Podwoloczysko,
Podwoloczysk, i przy szlysmy and we came to that baker.
do tego piekarza.
To naprzód on się bal nas First he feared to let us
spuscic, in,
bo on powiedzial, because he said,
a jego syn byl komunistq, and his son was a communist,
to my tez napewno, so we also for sure (would be),
(xxx) (xxx)
nikt nie zdawal sobie no one realized that
zprawy ze
Niemcy przyjdq ze to. the Germans would come.
No ale w koncu 0 się But in the end he (0)
zgodzil, tarn, agreed,there,
0 polozyl, 0 pozwolil tej he (0) put, he (0) allowed
trójce
polozyc się spac na the trio to sleep on
podlodze, the floor,
0 dal nam wypic herbatę, he 0 gave us tea to drink,
i kazdemu po kostce cukru, and each a sugar cube,
a 0 byl zly jak pies, and he (0) was as angry as a dog,
chleba nam nie 0 dal. even bread he (0) didn't give us.
Action discontinuity in core event structure. Recall that nearly every
scene has a core event structure consisting of an instigating action, the
theme's mental reaction and her follow-up physical action,
In the excerpt below, the narrator and her two companions have been
advised that they must leave town because the Germans are approaching.
The following shows their reaction and follow-up action. The mental reaction
142 VANESSA FLASHNER

is marked with the pronoun my 'we'. The follow-up action is marked with
zero anaphora.
(8) i-uh wszyscy naokolo and everyone around comes up,
przychodzq,
<instigating action> <instigating action>
to-mówiq, (1.3) and (they) say, (1.3)
Rosja zamknęla granicę, Russia closed the
ale noc jest, borders, but it's night,
<reaction> <reaction>
my nic nie mo'zemy, we cannot do anything,
na na zpowrotem nie ma there's no way to turn back,
co isc,
my spróbujemy we will try it,
a tym czasem nad ranem meanwhile at dawn it had barely
ledwie zaswitalo, become light,
<follow-up action> <follow up action>
0 idziemy, we(0)go..
The next example collapses reaction and follow-up action. The pronoun
signals willful activity. In this scene the train carrying the narrator stops, and
starving from hunger she jumps out of the train to grab some food. Her fol­
lowing actions show action continuity and are represented with zero
anaphora.
(9) <instigating action> <instigating action>
zatrzymalo się gdzies They stopped the train somewhere,
pociqg,
<reaction and follow-up <reaction and follow-up
action> action>
ija zeskoczylam, and /jumped out,
0 wyrwalam trochę grochów I (0) pulled out some
po drodze peas by the side of the road
0 wrócilam. I (0) returned.
Verbs that typically appear with the pronouns ja T or my 'we' at this
"reaction" part of the core event structure are chciec 'want to', postanowic
'decide to', wziqc sobie 'take yourself in hand, set out to', spróbowac 'at­
tempt' and pomyslec 'consider'. The use of pronouns in the reaction part of
the event structure illustrates action discontinuity. It is the reaction part of a
scene that begins a new series of actions initiated by the protagonist. Out of
 THE GRAMMATICAL MARKING OF THEME 143

33 core event structures, 88% (29/33) showed overt pronominal theme.

Introduction of direct speech. In addition to the action discontinuity
shown in episode and scene changes and in the reaction part of a core event
structure, participant discontinuity is reflected in the introduction of direct
speech. Out of 39 introductions of direct speech in the data, 90% (35/39) have
overt marking; from this number 11 out of 12 instances, or 92%, represent the
narrator's use of ja T . Below the narrator describes a friendly gentleman who
offers her lodging while she is hiding in a forest during a bombing raid. Notice
that for each character overt pronouns are used to refer to his/her introduc­
tion of speech.
(10) i kolo mniele'zal taki and next to me lay this older
starszy pan gentlemen,
znaczy 0 le'zal 0 I mean he (0) lay
kryl się, he (0) hid himself,
(1.4)to on mówi pyta (1.4) so he says asks me
mi siq
"co jest..skqd ty jestes?"
"what's the matter..
where are you from?"
a ja mówie "ze Lwowa" and I say "from Lwow",
to on mówi "wiesz ja so he says "You know I
jestem
z Kijowa, am from Kiev,
ja wracam do Kijowa". I am returning to Kiev".
To conclude, the overt expression of usual anaphoric zero theme is
caused by:
a. syntactic discontinuity
b. episode change
c. scene change
d. action discontinuity in core event structure
e. introduction of direct speech
In summary, continuous themes are usually expressed by zero anaphora.
Overt themes are caused by several factors. One factor, syntactic discon­
tinuity, occurs when themes are in non-subject case or in a presentative (O)
VS word order. In addition, narrative discontinuities also trigger overt mark­
ing of theme. See Table 1. So far the overt realization of theme has been due
to episode, scene, and plot divisions, and the introduction of speech. All of
these discontinuities are found in foregrounded narrative events. In the fol-
144 VANESSA FLASHNER

lowing section, overt use of theme is discussed as it relates to backgrounded

portions of the text.

TABLE 1
Summary Table for Thematic Representation
in Discontinuous Sections of the Oral Narrative

Overt theme: Zero anaphora: Total

independent pronoun verb agreement

1. Syntactic Discontinuity 16 4 = 20
2. Episode Boundary 7 3 =10
3. Scene Boundary 25 6 =31
4. Reaction in Core Event
Structure 29 4 = 33
5. Introduction of Speech 11 1 = 12

Total 88 (pronouns) 18 (zeroes) =106

chi square = 44.92, df = 1, p<.01

5. Overt Theme as Method of Development for Background Motifs

As noted previously, scenes are composed of a series of events, three of
which usually form a core event structure. These main line events (fore­
ground) are interspersed with digressions (background) which serve to elabo­
rate the main-line events of the story. The digressions can be described as
"paragraphs". That is, they are a series of clauses united by some uniform
orientation which is either spatial, temporal, thematic (i.e. ideational) or par­
ticipant (Grimes 1975:102). In this text, chunks of background clauses often
qualify as "paragraphs' 1 . They may describe physical aspects of the story: for
example, its setting or characters; or they may develop motifs such as loneli­
ness, hunger, poverty, fear of capture, and the like.
More to the point of this discussion is the construction of these
background clause units or paragraphs, and, specifically, the role of their
overt inclusion of subject pronouns. It is proposed that the overt use of pro­
nouns and nouns (names) allows for the construction of paragraphs built
 THE GRAMMATICAL MARKING OF THEME 145

around the organizing principle of comparison/contrast. That is, when pro­

nouns or nouns (names) begin a series of clauses, the clauses usually form a
paragraph of contrast. Fries (1980) has discussed the function of clause initial
elements in organizing discourse. He suggests that there is a correlation
between initial clause elements and the method of development of a text.
Thus, clause or sentence level initial elements are not only determined by
what is old and new information in a discourse but are rather a part of a larger
organization pattern determined by the motif of the discourse. Fries claims
that (1980: 53):
if the themes [Fries means the initial clause elements] of most of the sen­
tences of a paragraph refer to one semantic field (say location, parts of some
object, wisdom vs chance, etc.) then that semantic field will be perceived as
the method of development of the paragraph.
In this study Fries' concept of initial clause element is used, but the first
element in the clause refers to the first semantic element, thereby excluding
conjunctions (e.g. and, so, because, etc.) and fillers (e.g. well, then, so). In
the present data the speaker uses lexical fields, adverbs of place or time or
names and pronouns, to develop background motifs.
Before illustrating the use of overt themes as initial element, I will show
the use of adverbials as a method of development for background motifs. The
first example is a paragraph of description developed through clause-initial
words of space and time. The paragraph describes a castle which the pro­
tagonist passes as she sets out on her journey at the beginning of the narra­
tive.
(11) i tarnmysmymieszkalibarcizo and there we lived very near
blisko zarnku, a castle,
to byl zamek Jana Trzeciego it was the castle of Jan
Sobieskiego, Sobieski III,
tam bylo, there it was,
pozniej to bylo wię zienie, later it was a prison,
a to ten zamek jak za czasów and this castle in
dawnych ancient
byl otoczony takq fosq times was surrounded by a moat,
w której wplynęla woda, in which flowed water,
kiedys, at one time,
a przed tym byl taki most and before then there was
zwodzony, a drawbridge.
ale wtedy tej wody nie bylo, but then there was no water,
146 VANESSA FLASHNER

i jak widzisz to mi zostalo and as you see it has remained

w pamięci w tej szosie. in my memory on this path.
Six out of ten initial clause elements begin with time or place adverbials. Thus
the first elements provide a method of development for the main topic of the
paragraph, the castle. The description of the castle provides a setting for the
action taking place in the narrative.
In the next example the lexical field of place provides the method of
development. The motif or topic of the paragraph is to determine the next
place to which to travel.
(12) to wyszlismy znowu na so we went out on such a road,
takq droge,
i co mysmy dqzyly? and where were we trying to go?
Dqzyly smy (.5) wszyscy We were trying (.5) everyone
mówili.'zeArmia Czerwona said that the Red Army
cofa się do starej granicy, retreats to the old border,
a stara granica to byla pod and the old border was
Woioczyskiej, near Woioczyska,
a Woioczyska to juz byla nowa and Woioczyska was not the
granica, new border,
i tam jak tylko trzeba and there if one can just
siq dostac
do do Woloczysk, get to to Woioczyska,
a tarn juz jest Rosja, and there finally is Russia,
i tarn będzie armia przestanie and there the army will
sie cofac, cease to retreat,
i będzie mo'znajakos zyc. and we'll be able to live somehow.
It is evident that paragraphs of description are often organized around
lexical fields as a method of development. The systematic repetition of lexical
fields as initial clause elements seems, therefore, to be a feature of oral dis­
course as well as of planned written discourse (cf. Fries 1980).
When overt themes begin a series of clauses, the clauses usually form a
paragraph of contrast. Out of 28 instances of contrast in the text, 96% (27/28)
showed overt use of pronoun to represent the narrator, whereas in non-con­
trast background clauses only 30% (20/66) showed overt use. In the para­
graph below, the overt use of pronouns develops an illustration of poverty.
The use of pronouns here is to provide a structural framework where a con­
trast relationship can be expressed:
 THE GRAMMATICAL MARKING OF THEME 147

(13) i tylko trojka dziewczqt idzie, and only a trio of girls walks,
ja mialam tylko teczuszkę, and I had a small bag,
w teczuszce 0 mialam tylko the purse I ( 0 ) had my
swoje dokumenty, documents,
a one mialy zsobq trochę but they had with them
rzeczy, some things,
bo one wyszly ze Lwowa, because they had left from Lwow,
a ja wyszlam z domu, and I had come from home,
i nic z domu nie 0 wzięlam, and nothing from home did I ( 0 )
a one wyszly, take, but they had left,
bo one wyszly ze Lwowa, because they had left from Lwow,
i 0 mialy wplecaku trochę and they ( 0 ) had in
rzeczy, their backpack a few things,
A similar use of pronouns in sentence initial position is shown below.
Once again contrast is the organizing principle to express the motif of pov­
erty. Note that the two final clauses in the paragraph no longer develop the
motif of poverty but rather elaborate on one detail — a watch — so these
clauses do not show overt use of theme and subtheme.
(14) ta dziewczyna znów the girl once again went
pojechala gdzies
indziej kogos szukac, somewhere else to find someone,
ale ona miala pieniqdze, but she had money,
aja nie mialam, and I didn't have (any),
aja mialam tylko zegarek, I had only a watch,
i 0 mialam go sprzedac, and I (0) had to sell it,
i 0 nie moglam go wtedy and ƒ (0) couldn't then sell it,
sprzedac,
Below the use of pronouns in initial position functions to describe the dif­
ferent whereabouts of the characters in the story. Again, the organizing prin­
ciple is contrast:
(15) Wyszlismy tak o trzeciej nad We left at three in the
ranem, morning,
bo ona przyszla, because she came,
ona gdzies byla gdzies indziej, she was somewhere else,
aja bylam u rodzicow, and I was at my parents',
boja..moje rodzice because I.. my parents lived
mieszkali
148 VANESSA FLASHNER

we Zioezowie, in Zloczow,
In the above examples anaphoric zero pronouns are predicted because
the characters have already been introduced. However, by using pronouns as
clause initial elements the speaker sets up the expectation that something or
someone will be contrasted with the referent of the pronoun. And indeed in
the following clauses there is a different nominal or pronominal referent. So,
in this case we have pronominalization setting up contrast through syntactic
location in a clause. See Table 2 for a summary of results.

TABLE 2
Summary Table for Thematic Representation
in Background Paragraphs

Overt theme: Zero anaphora: Total

independent pronoun verb agreement

1. Paragraphs developed
through contrast 27 1 = 28

chi square = 22.32, df = 1,P<.01

2. Paragraphs not developed

through contrast 20 46 = -66

chi square = 9.47, df = 1, p<.01

Thus far theme has been discussed in terms of narrative discontinuity

features and in terms of its overt use as a starting point for clauses in para­
graphs developed through contrast. In the following section the use of theme
within quoted speech and in the loose event clauses that link scenes together
will be examined. It is in these parts of the narrative that the speaker's choice
of thematic reference clearly shows the realization of interpersonal goals in
language. The next section demonstrates how independent pronouns are one
member within a range of devices used to illustrate the narrator's attitude
toward a situation.
 THE GRAMMATICAL MARKING OF THEME 149

6. Theme and Perspective

Some parts of the data do not fit into the previous framework. These
overt uses of theme can be accounted for by the narrator's use of pronouns to
show perspective. They appear in event clauses that are used to connect
scenes, typically when the protagonist is travelling from one place to another,
or in direct speech. There is a scale which indicates speaker's status. It ranges
from complete separation/isolation represented through the pronoun ja T to
neutral attitude represented by the anaphoric zero pronoun. The following is
a scale representing degrees of convergence/divergence:
Separateness
a. ja
T — used to show opposition of "I" vs ''others"
b. my
'we' — used to show opposition of "we" vs "others"
Solidarity
c. mysmy
'we' + 3rd person pl verb ending — used to show integration
within group, i.e., action as one team
Neutrality
d. 0 anaphora — used to show neutral attitude toward events
Within the realm of possible thematic reference, the narrator, depend­
ing upon the context, may use first person singular or plural marking. How­
ever, with each first person marking of T' or 'w' the speaker has a three-way
choice. In the unmarked form, subject reference is affixed to the verb
through verb agreement morphemes. In the marked form an independent
pronoun is used in addition to the verb. Intermediate between the unmarked
and marked forms exists a third choice. This occurs when the verbal affix indi­
cating person is preposed and affixed to a sentence initial element. For exam­
ple, the verbal first person affixes m for T and smy for 'we' may be affixed to
sentence initial words such as to 'then' as in to-smy or jak 'as' in jak-esmy or
to pronouns as ja T' as in ja-m or my 'we' as in mysmy.
Depending on the text, marked forms convey various perspectives.
However, in this text, the narrator uses the range of references to indicate her
separateness from other characters in the story. One of the motifs running
through the story is isolation. The narrator leaves home and family in search
of a place beyond the invading army's reach. In her attempt to reach this
haven she must travel through hostile territory. At certain points during the
150 VANESSA FLASHNER

journey she is accompanied by friends; most of the time, however, she is on

her own. The scale of separateness, as mentioned, ranges from the use of the
pronoun ja T to show extreme isolation to zero anaphoric pronoun to show
neutral perspective. The preposed verbal affix smy is used in contexts where
the narrator acts in unison as a group with her companions.
Below a few examples illustrate the use of this scale in the loose event
clauses. The first example describes the narrator as she leaves home with her
two companions. After describing the sight of a corpse along the road, the
narrator uses a mysmy to indicate her group feeling as she walks in the com­
pany of her two friends. Thereafter the independent pronoun my is used to
indicate what the group looked like as though seen from a bird's vantage
point, "a trio". Then again, the narrator refers to the trio's walking in unison
with the preposed verb suffix mysmy. Afterwards, a contrast is created
between the hordes who travel on the same road and the girls with the isolat­
ing my. This image of the girls as a unique entity set against the others is sum­
marized in the last line "and only a trio of girls walks along".
(16) le'zal jeden (.5) który there lay a person (.5) who had
mial kulę
w plecach, a bullet in his shoulders,
tak le'zal skulony, as he lay hunched over,
i tak zginql. that's how he died.
(1) No i mysmy poszly dalej, (1) So we walked on,
ten obraz.. 'ze to bylpoczqtek that image, .that it was the
mojej drogi, beginning of my journey,
i my w trójkę, and we in a group of three,
i mysmy wyszly, and we walked along,
i moc ludzi szlo wojsko and many people walked
and the army,
i ulica znaczy gosciniec and the street, that is the path
zalany, roadway was overflowing
with people),
i my idziemy, and we walk,
i tylko trójka dziewczqt and only a trio of girls walks,
idzie,
In example (17), the narrator describes how she travels by herself
through a forest to reach her friends who are working on a collective farm:
 THE GRAMMATICAL MARKING OF THEME 151

(17) ijasobie idę tym lasem.. and I walk by myself through

sama the woods..alone..
tak piechotq, yes on foot,
i 0 szlam i 0 szlam, and I (0) walked and (0) walked,
powiedzieli mi tak, they told me this,
"prosta drogajest.. "it's a straight road..
nicsięnie bój, don't be afraid,
znaczy nie zablqdzisz, I mean you won't get lost,
od Krasnodaru będziesz szla from Krasnodar you will go
tylko lasem prosto prosto, only through woods straight straight,
i dojdziesz do Timosiowki and you will come to Timosiowka
swinosollxozu, a pig solxoz,
i tarn sa studenci and there are students
ze lwowskiego uniwersitetu", from Lwow University",
dobra, okay,
bo ja bylam lęk przed because I had a fear of
samotnosciq solitude,
ja bylam sama, /was alone,
ja nikogó nie mam, I don't have anyone,
oprócz tego ze nie besides the fact that I (0)
0 mialampieniędzy, didn't have any money,
(.5) No iproszęja ciebie (.5) So and listen
w poludnie 0 przychodzę in the afternoon I (0) arrive
do tej Timosiowkil to this Timosiowka,
(1.3) imówiqc mówiq tak, (1.3) and saying (they) say this,
"dwa kilometrów stád to oni "two kilometers from here they
dzisiaj pracujq tam i tarn, today are working here and there,
w polujestgrupa kilku in the field is a group a few
studentów lwowskich. " students from Lwow."
(2.2) ø poszlami0 (2.2) I (ø) went a n d /
przyszlam, (0) arrived,
In the above passage the pronoun ja 'I' is used at those points where isolation
or individuation is focal: as the narrator sets out alone on her walk through
the forest in loose event clauses and as she describes her fear of solitude in
background non-contrast clauses. In the text zero anaphoric pronoun is used
with verbs of locomotion where there is no emphasis on separateness. Out of
71 loose event clauses, 1% (1/71) show overt theme, 7% (5/71) show pronoun
152 VANESSA FLASHNER

and clitic (mysmy), and 92% (1/71) show zero anaphora.

The separateness scale also functions within quoted speech. Out of 16
references to herself in quotations, the narrator uses pronouns 94% (15/16)
of the time. In the excerpt below the narrator describes an interrogation con­
ducted by a Russian army officer after she is found speaking Polish on a train
with only Russians on board. The pronoun ja 'I' shows the narrator's isolation
and defensive attitude in a hostile environment.
(18) i ja mówiq, and I say,
"bo ja ucieklam, "because /escaped,
ja nie chcialam zeby byc I didn't want to be under
u Hitlera, Hitler,
ja jestem studentka, I am a student,
ija ucieklam", and / escaped,"
imowie mu, and I say,
"ja wyszlam razem z "I left together with
L wowskim
Uniwersitetem ", Lwow University",
có bylo nieprawda, which wasn't true,
ale jamowiq, but I say,
'ja wyszlam ", " I left",
In sum, the narrator makes use of a range of possibilities for first person
reference to depict attitudes of separation and unity. See Table 3.

7. Conclusion
This study has shown that there are several predictions which we can
make about the use of theme in an oral Polish narrative. The first prediction
is that the preferred marking of continuous theme is anaphoric zero subject.
The second is that theme marking will be overt under certain conditions. (a)
Theme marking is overt when narrative discontinuities appear. These are
scene or episode changes, introduction of direct speech, and change in action
continuity in core event structures. (b) Theme marking is also overt when it
functions as a method of development in backgrounded paragraphs of con­
trast. Finally, (c) theme marking is overt where narrator perspective shows
degrees of separation and solidarity in loose event clauses and within quoted
speech.
There are several principles of language in competition here. The textual
principle of economy, to keep language from being overly redundant, is satis­
fied when thematic (participant) continuity is expressed through anaphoric
 TABLE 3
Summary Table for Thematic Representation
in Loose Event Clauses and Within Quoted Speech

overt theme: Pronoun + preposed ze

independent pronouns 1st person clitic ve
ja T my 'we' my-smy 'we+1st p. pl'

1. Loose event
clauses

2. Within quoted
speech 15
154 VANESSA F L A S H N E R

zero pronoun. This expression of thematic continuity though anaphoric zero

pronoun had also been found to exist in Russian (Nichols 1984) and in a writ­
ten Polish narrative (Mayew7ski 1975). But the principle of economy cannot
explain all of the data. In the narrative, the rule that continuous theme is
anaphoric zero pronoun is often violated because of other pragmatic factors.
The second principle competing here is the textual principle of expressiv­
ity which suggests that language makes overt potentially anaphoric reference
in order to emphasize through unexpected patterns. That is, paragraphs of
digression were often structured by a principle of organization using a par­
ticular lexical field as a method of development. As part of a lexical field, the
overt use of pronouns in sentence initial position provides a framework of sys­
tematic repetition within successive clauses whereby a motif or idea can be
illustrated through contrast.
The third principle competing here is the interpersonal principle of coop­
eration. The maxim of quantity is flaunted when marked forms are chosen to
show speaker perspective. In the present narrative the overt use of theme in
quoted speech and in event lines expresses separateness. Indeed, we see a
scale ranging from overt use of theme to express total separation to zero
anaphora where isolation is irrelevant.
To conclude, the three principles mentioned here compete for overt or
anaphoric expression of theme. We have seen, thus, that looking at the gram­
matical expression of theme in discourse provides a way to study principles of
discourse in narrative texts. In addition, examination of theme provides an
interesting contrast between the written and spoken language and between
planned and unplanned speech. This study shows that unplanned oral dis­
course may also be highly organized and that the reasons for its structure may
be understood.

NOTES
*I would like to thank the narrator Charlotta Wurm Flashner for providing me with this rich data,
and Roger Andersen, Henryk Flashner, Sabrina Peck, and Sandra Thompson for reading and pro­
viding valuable comments on an earlier version of this paper.
1) In Zarebina's (1973) study the distribution of parts of speech in the texts of the three varieties
is as follows (in order of frequency) (42):
General Written Polish General Spoken Polish Regional Polish
Noun 30.73% Verb 26.90% Pronoun 23.64%
Verb 17.03% Noun 20.49% Verb 20.04%
Pronoun 12.90% Pronoun 17.45% Noun 15.70%
 THE GRAMMATICAL MARKING OF THEME 155

The remainder of the list includes prepositions, adjectives, conjunctions, adverbs, particles, and
interjections.
2) These two narratives are comparable in many ways. Both narrators are middle class speakers
of Standard Polish and from the same region of eastern Poland. Both narratives are about personal
encounters during the Second World War. The stories thus show similar occasions for competing
referents, contrasts, and personal viewpoints.
3) The notions of pragmatic principles of textual and interpersonal rhetoric are borrowed from
Leech (1983: 16).
4) The linguistic background of the speaker is as follows. She is a speaker of Standard Polish.
She grew up in Zloczow, a town near the city of Lwow, in eastern Poland. After the war she lived
in Central Poland (Warsaw) until 1957. Since that time she has lived in Israel speaking predomin­
antly Polish and using Hebrew in her professional life. The narrator also speaks Yiddish and Rus­
sian.
5) The terms ideational, interpersonal, and textual are originally taken from Halliday (1970)
and are used to describe the functions of language with respect to grammar.
6) The terms foreground and background are used here in a general sense. Foreground refers
to the event clauses that move a story's action forward. Background refers to all other clauses in
the text.
7) More accurately, zero anaphora in this paper refers to a conflation- of the first two points of
(1) zero anaphora and (2) unstressed/bound pronouns or grammatical agreement found in Givón's
topic accessibility hierarchy. For, technically, no finite verb in Polish has a true anaphoric subject
as Korean, Japanese, or Mandarin can. This is because the subject is always identified through
verb agreement clitics. As an illustration, below is one conjugation for present tense:
l.p.sg. -m l.p.pl. -my
2.p.sg. -sz 2.p.pl. -cie
3.p.sg. -0 3.p.pl. -ja
with the verb pytac 'ask' the forms become:
l.p.sg. pytam l.p.pl. pytamy
2.p.sg. pytasz 2. p. pl. pytacie
3.p.sg. pyta 3.p.pl pytaja
Hence, the use of a pronoun is redundant, and its infrequent use in the written language is not sur­
prising.
8) The following conventions will be used in the examples cited in this paper. Anaphoric zero
pronoun themes are marked by zero ( 0 ) in the Polish text. Although some of these passages also
have other anaphoric zeros referring to entities, zeros are only marked where they pertain in each
example. The English translations are equivalent to the Polish with regard to theme marking; how­
ever, where Polish uses an anaphoric zero theme, the English is also marked with an anaphoric
zero theme, e.g., he ( 0 ) . Overt theme or sub-themes are marked where pertinent in the Polish and
English texts.
156 VANESSA FLASHNER

REFERENCES

Chafe, W. ed. 1980. The Pear Stories: Cognitive, Cultural, and Linguistics
Aspects of Narrative Production. Norwood, N.J.: Ablex.
Flier, M. and R. Brecht. eds. 1984. UCLA Slavic Studies 11. Columbus,
Ohio: Slavica.
Fries, P. "On the status of theme in English: arguments from discourse". To
appear in Forum Linguisticum.
Givón, T. ed. 1979. Discourse and Syntax, Syntax and Semantics, vol. 12,
NY: Academic Press.
Givón, T. 1983. "Topic continuity in discourse: an introduction". In Givón
1983:5-41.
Givón, T. ed. 1983. Topic Continuity in Discourse: A Quantitative Cross-lan­
guage Study, vol. 3, Amsterdam: J. Benjamins.
Grimes, J. 1975. The Thread of Discourse. The Hague: Mouton.
Halliday, M.A.K. 1970. "Clause types and structural functions". In: Lyons
1970: 140-165.
Leech, G. 1983. Principles of Pragmatics. London: Longman.
Levy, D. 1979. "Communicative goals and strategies: between discourse and
syntax". In: Givón 1979: 183-210.
Lyons, T. ed. 1970. New Horizons in Linguistics. Harmondsworth: Penguin.
Mandler, J. and N. Johnson. 1977. "Rememberance of things parsed: story
structure and recall". Cognitive Psychology 9: 111-151.
Mayewski, A. 1975. Wojna Ludzie i Medycyna. Lublin: Wydawnictwo
Lubelskie. 1-40.
Nichols, J. 1984. "The grammatical marking of theme in literary Russian". In
Flier and Brecht 1984:170-186.
Zarebina, M. 1973. "A statistical lexical-semantic characterization of the
main varieties of present-day Polish". Lingua Posnaniensis 17, 37-48.
 ANAPHORA IN POPULAR WRITTEN ENGLISH NARRATIVES

BARBARA FOX
University of Colorado, Boulder

1. Introduction
Within the last ten years, work on discourse anaphora has blossomed in
scope and depth. What had been a neglected area of linguistic behavior has
become a source of interest to researchers in linguistics, psychology, and arti­
ficial intelligence. Anaphora in narratives has become an especially impor­
tant area of research; in this study I would like first to review the highlights
of the recent work on anaphora in narratives, explore the strengths and weak­
nesses of these approaches, and finally provide a broader and more satisfac­
tory account of the anaphoric patterning in written narratives than has previ­
ously been presented.

2. Previous Work
Within linguistics, the major proposals concerning discourse anaphora
in narratives have been Givón (1983), Clancy (1980), and Grimes (1978).
Since Givón (1983) represents the more traditional view of anaphora across
sentences, I will discuss it first.
Givón (1983) deals with anaphora in order to understand the linguistic
coding of the concept topic. He proposes the Continuity Hypothesis, in which
it is claimed that
The more disruptive, surprising, discontinuous or hard to process a
topic is, the more coding material must be assigned to it (original emphasis,
P. 17).
According to Givón, the factors which influence disruptive or discon­
tinuous topics include:
1. Distance to last mention, 'If a topic is definite and returns to the register
after a long gap of absence, it is still difficult to process. The shorter is the
158 BARBARA FOX

gap of absence, the easier is topic identification; so that a topic that was
there in the preceding clause is by definition easiest to identify and file cor­
rectly'' (original emphasis, p. 8).
2. Ambiguity from other referents. "If no other topics are present in the
immediately preceding discourse environment...topic identification is
easiest. The more other topics are present in the immediate register, the
more difficult is the task of correct identification and filing of a topic..."
(original emphasis, p. 8)
3. Availability of thematic information. 'Thematic information available
from the preceding discourse could help in topic identification — espe­
cially when other topics in the register may potentially interfer. Such
information establishes specific probabilities...as to the topic identifica­
tion within a particular clause and in a particular role" (original emphasis,
p. 9)
The first two factors, distance and ambiguity, are the major foci of the
studies in Givón (1983). The third factor is merely hinted at in the introduc­
tion and then neglected. Thus although Givón makes explicit mention of
something that sounds slightly structural and hierarchical, the end result is a
model that views anaphora as a function of distance and a rather vague
notion of ambiguity. In an impressive collection of data from several unre­
lated languages, pronouns are shown to be used when the distance to the last
mention of the referent is small (and there are no interfering referents), while
full NP's are shown to be used when that distance is somewhat great (and/or
if there are interfering referents).
What does such a model imply about text structure and the flow of atten­
tion through a text? From my reading of the claims in Givón (1983), the fol­
lowing analysis seems reasonable. If the degree of a referent's continuity with
the preceding discourse is measured in clauses to most recent mention, then
we can assume that continuity derives from the surface nature of the clauses,
rather than their textual function, and that (presumably) all clauses are equal
in their contribution to the measure of continuity (except perhaps relative
clauses, which were excluded from the continuity counts). That is, whether a
clause is an aside about a character, a source of evidence to support a claim,
etc., is irrelevant to the count: all clauses have the value 1 for the purposes of
measuring continuity. The model thus assumes that discourse is made up of
an undifferentiated string of clauses which follow one another in time but do
not form larger units that could perform communicative functions in relation
to one another. 1 Attention must be equally flat, if all that really matters is dis­
tance, since there would be no need to indicate to the reader that something
new is being started, or that something old is being closed off, or that some
 ANAPHORA IN POPULAR WRITTEN ENGLISH NARRATIVES 159

interruption — after which the interrupted unit will be resumed — is about to

occur. Text structure and attention flow must thus be flat and undifferen­
tiated in this model of discourse.
My criticism of the continuity hypothesis thus rests not on the specific
quantitative predictions made — clearly the gross claims of distance are
borne out by the cross-linguistic counts — but on the model they presuppose
and the accompanying roughness of the predictions.
While it must be acknowledged that much of the emphasis on distance
and ambiguity in the continuity hypothesis arises out of a desire to provide
quantitative evidence (and hierarchical text structure is hard to quantify), the
overemphasis it produces on the linear nature of texts (and the encourage­
ment it brings to such a view) needs to be recognized. It seems to me that at
this point in the study of anaphora it may be wiser to proceed cautiously with
a qualitative approach that incorporates a hierarchical structural view of texts
to see if we can understand the basic mechanism of anaphora before we begin
collecting rough quantitative evidence. 2
Clancy (1980) provides a very rich and detailed look at anaphora in short
spoken English and Japanese narratives. Although the notion of discourse
structure is not fully integrated into the account (cognitive factors of distance
and interference are prominent here, as in Givón (1983)), Clancy brings out
the association between discourse units and full noun phrases, and notes that
this is in some sense an unusual and optional use of anaphora:
In English speakers apparently feel that inexplicit reference [i.e., pro­
noun or zero anaphora] is still comprehensible after the passage of two
clauses...or of one sentence...; Japanese speakers agree....In both lan­
guages at least 97 percent of all inexplicit references were made when no
more than one other character had been mentioned. Yet as the exception to
these trends reveal, time and interference cannot account for all referential
choices. Therefore, in the following sections of this paper the content of the
narratives will be examined in order to clarify how discourse structure may
have influenced referential choice... (pp. 143-144)
The main discourse structures Clancy finds influencing referential choice
are episode boundaries, wherein a new line of action starts, and world
shifts, in which the speaker moves from one mode of talking to another (e.g.,
from digression to the plot line, or from film-viewer mode to story-teller
mode). Both of these structure-types tend to be associated with use of full
noun phrases.
Thus although Clancy does not provide a full account of the relationship
between narrative structure and anaphoric patterning, there is a very strong
160 BARBARA FOX

attempt here to move beyond treating narratives as strings of clauses and

hence to move beyond treating anaphora as responsive only to linear notions
of distance and interference.
Another fairly wide-spread approach to anaphora in narratives is
associated with the Summer Institute of Linguistics, represented in, for
example, Grimes (1978) and Levinsohn (1978). In this framework, anaphoric
patterning in narratives is based on the global status of the characters; that is,
anaphoric devices are distributed according to the centrality or peripheral-
ness of the referents. Certain anaphors are associated with central characters
under certain conditions, while other anaphors are associated with minor
characters, where the centrality of the character is defined either within the
entire narrative or within some subpart of that narrative (e.g., an Episode).
There is no mention here of distance, or ambiguity, and only minimal atten­
tion is paid to discourse structure; the relevant factor is the centrality of the
character to the story.
While this approach may be valid for the languages and narratives
examined in the studies mentioned, it does not seem to transfer well to popu­
lar written English narratives. Consider the following example, in which the
villain of the book (Darth Vader) is referred to with both pronouns and a full
noun phrase, as is a peripheral character (notice especially the pronouns refer­
ring to the peripheral character):
Vader did not waste words. He raised his hand, fingers outstretched,
toward the officer's throat. Ineffably, the officer began to choke. His knees
started buckling, his face turned ashen. (Return of the Jedi, p. 90)
In this passage, there is is no particular device that is reserved for the cen­
tral character and no device reserved for the minor character: both characters
are referred to with full noun phrases and pronouns. Clearly, a different set
of strategies is at work in this passage. It is the goal of the remainder of this
study to describe these strategies in some detail.

3. Anaphoric Patterning in four Popular Written Narratives

I chose four fast-paced popular narratives as the sources of references for
this study. These are:
1. Alien. Foster, Alan Dean. 1979.
2. The Girl of the Sea of Cortex. Benchley, Peter. 1982.
3. Coma. Cook, Robin. 1977.
4. Return of the Jedi. Kahn, James. 1983.
 ANAPHORA IN POPULAR WRITTEN ENGLISH NARRATIVES 161

4. Anaphora in Written Narratives

It first came to my attention that the traditional "distance" theory of
anaphora might not be accurate for stories when I started closely examining
long written monologue narratives. In these texts, I found examples like the
following, in which something like 11 clauses (depending on what one counts
as a clause) separate the two mentions of a character, and yet the second men­
tion is done with a pronoun: 3
She took a deep breath and tested the firmness of her grasp on the wood.
When Jobin had first taught her to swim, he had told her always to get in and
out of the water quickly, for it was in the marginal moment — half in half out
of the water — that a person was most vulnerable to shark attack: It was then
that the person looked truly like a wounded fish; most of the body was out
of the water so it appeared smaller, and what remained in the water (lower
legs and feet) kicked erratically and made a commotion like a struggling ani­
mal.
She spun, grabbed the gunwale... (The Girl of the Sea of Cortez, p. 78)
In addition, I found that two mentions of a character could be separated
by a mention of a character of the same gender and yet the second mention
of the first character could still be done with a pronoun, contrary to the pre­
dictions made by Givón et al (1983) about ambiguity. A passage illustrating
this phenomenon is shown below.
He <Obi Wan Kenobi> paused sadly and looked directly into Luke's eyes,
as if he were asking for the boy's forgiveness. "My pride had terrible con­
sequences for the galaxy." Luke was entranced. That Obi-Wan's hubris
could have caused his father's fall was horrible. Horrible because of what his
father had needlessly become, horrible because Obi-Wan wasn't perfect,
wasn't even a perfect Jedi, horrible because the dark side could strike so
close to home, could turn such right so wrong. Darth Vader must yet have a
spark of Anakin Skywalker deep inside. "There is still good in him," he
declared. (Return of the Jedi, pp. 63-64)
In this passage, two mentions of Luke are separated by references to
Obi-Wan and Darth Vader (Anakin Skywalker is Darth Vader), and yet the
last mention of Luke is done with a pronoun. Clearly, then, the anaphor here
is responding not to the linear sequence of references, but to something else.
Moreover, contrary to the predictions made by the Continuity
Hypothesis, I found references done with full NP's when the referential dis­
tance (i.e., distance to most recent mention) was extremely small, as in the
following passage :
162 BARBARA FOX

The four flights up to her floor seemed longer than usual to Susan. She
paused on several occasions, because of a combination of physical fatigue
and mental effort.
Susan tried to remember if Bellows had said succinylcholine was among
the drugs found in the locker... (Coma, p. 236)

In this passage, there is arguably no clause-gap between the underlined

reference to Susan and the mention just preceeding it, yet the last mention is
done with a full NP. This piece of text thus runs counter to the "distance"
theory of anaphora.
In working through these examples, then, it became apparent to me that
distance was not in and of itself a factor in the determination of anaphora in
narratives. Other factors having to do with the functional, hierarchical struc­
ture of narratives seemed to be much more influential. There isn't room for
a full account here, but some of the critical factors are discussed below.

5. The Basic Patterns

As much as possible, I have couched the statements of the anaphoric pat­
terns in terms that are familiar from our vernacular descriptions of stories and
events. Although a particular text model was used to analyze the narratives
(a modified version of Rumelhart's (1975) Story Grammar), because of limi­
tations on space I have chosen to describe most of the patterns in a model-
independent way, so that a full explanation of the model used would not be
necessary. Where the Story Grammar notation is helpful, I have introduced
just those aspects of it that are relevant to the immediate discussion. It should
be kept in mind throughout this chapter that the statements about anaphora
offered here are meant to be as brief and non-technical as possible. A full exp­
loration of anaphora in written narratives still remains to be performed.
The basic pattern of anaphora in the written narratives I examined seems
to be something like the following:
Temporarily putting aside the environment of two referents of the same
gender (hereafter known as same-gender referents), it seems that a referent is
pronominalizable until another character's goals and actions are introduced,
unless those goals and actions are interactive with the first character's, that is,
unless in the immediately projected text there is some confrontation or active
interaction between the two characters. In other words, if another character
begins planning and performing an action, and there is no immediately pro­
jected interaction between the two characters, then a subsequent mention of
the first character will be done with a full NP. 4
 ANAPHORA IN POPULAR WRITTEN ENGLISH NARRATIVES 163

This basic pattern has several implications. First, it means that a long gap
between mentions of a referent does not necessarily trigger the use of a full
NP for the second mention; if the gap does not introduce another character's
plans and actions, but is, rather, concerned with something off the event-line,
for example, like describing the surrounding scenery or a general truth, then
it will not 'induce" the use of a full NP. Examples of this pattern are given
below.5
She took a deep breath and tested the firmness of her grasp on the
wood. When Jobim had first taught her to swim, he had told her always to get
in and out of the water quickly, for it was in the marginal moment — half in,
half out of the water — that a person was most vulnerable to shark attack: it
was then that the person looked truly like a wounded fish; most of the body
was out of the water so it appeared smaller, and what remained in the water
(lower legs and feet) kicked erratically and made a commotion like a struggl­
ing animal.
She spun, grabbed the gunwale... (The girl of the Sea of Cortez, p. 78)
In this passage, the gap between mentions of the referent in question is
not concerned with describing another character's current plans and actions;
rather, it is taken up with an off-event-line aside about how one should
behave under certain conditions. When the aside is completed, the event-line
is resumed with a pronoun.
Another example of this phenomenon follows:
She reached B deck unopposed, her flamethrower held tightly in both
hands. The food locker lay just ahead. There was an outside chance the alien
had left someone behind, being unable to maneuver itself and two bodies
through the narrow ducts. A chance that someone might still be alive.
She peered around the jamb... (Alien, p. 260)
Here again we have an off-event-line description after an action by the
character in question. When the event-line is resumed, the reference is done
with a pronoun.
Another example:
Luke had known the passing of old mentors before. It was helplessly
sad; and inexorably, a part of his own growing. Is this what coming of age
was, then? Watching beloved friends grow old and die? Gaining a new mea­
sure of strength or maturity from their powerful passages?
A great weight of hopelessness settled upon him... (Return of the Jedi,
p. 61)
Once more, the event-line is interrupted, in this case in order to present
some of the character's thoughts. When the event-line is resumed, the refer-
164 BARBARA FOX

ence is done with a pronoun.

This pattern holds even if the off-event-line gap introduces a character of
the same gender as the character in question. This variation of the phenome­
non is illustrated by the following passage, in which a pronoun is used to refer
to the event-line character in spite of references to other, same-gender, refe­
rents in the off-event-line gap:
He <Obi Wan Kenobi> paused sadly and looked directly into Luke's
eyes, as if he were asking for the boy's forgiveness. "My pride had terrible
consequences for the galaxy." Luke was entranced. That Obi-Wan's hubris
could have caused his father's fall was horrible. Horrible because of what his
father had needlessly become, horrible because Obi-Wan wasn't perfect,
wasn't even a perfect Jedi, horrible because the dark side could strike so
close to home, could turn such right so wrong. Darth Vader must yet have a
spark of Anakin Skywalker deep inside. "There is still good in him," he
declared, (Return of the Jedi, pp. 63-64)
The second implication of the basic pattern described above is that even
if the material separating the references is not off-event-line, as long as it does
not introduce the current actions of another character a pronoun can be used
to refer to the first character. Examples of this use of pronoun are given
below.
She felt her way down the ladder until she reached solid footing. Then
she activated her lightbar. She was in a small maintenance chamber. The
light picked out plastic crates, rarely used tools. It also fell on bones with
shreds of flesh still attached, Her skin crawled as the light moved over frag­
ments of clothing, dried blood, a ruined boot. (Alien, p. 261)
Her gaze rose to the rear-facing screen. A small point of light silently
turned into a majestic, expanding fireball sending out tentacles of torn metal
and shredded plastic. It faded, was followed by a much larger fireball as the
refinery went up. Two billion tons of gas and vaporized machinery filled the
cosmos, obscured her vision until it, too, began to fade. (Alien, p. 266)
She reached down with her hand and tried to squeeze the lower calf
muscle, but it was too late. The muscle fibers had already bailed into a knot
the size of an orange. She rolled onto her back and used both hands to
squeeze her leg. Kneading with her fingertips, she softened the knot and felt
it begin to relax. Suddenly the knot dissolved and she thought the cramp was
finished... (The Girl of the Sea of Cortez, pp. 172-173)
The third implication of the pattern formulated above is that the intro­
duction of a second character (excluding same-gender referents for the time
being) does not necessarily cause the use of a full NP to refer to the first
 ANAPHORA IN POPULAR WRITTEN ENGLISH NARRATIVES 165

character. If the two characters are involved in a fast-paced confrontation or

interaction, such as a fight, a chase, or a conversation, the mention of one
does not cause the next mention of the other to be done with a full NP. Here
again, a referential gap does not "induce" the use of a full NP. Examples of
this pattern follow.
He took a step toward her, reached out helpfully. She bolted, ducking just
beneath his clutching fingers. Then she was out in the corridor, sprinting for
the bridge. She was too busy to scream for help, and she needed the wind.
There was no one on the bridge. Somehow she got around him again,
throwing emergency switches as she ran. (Alien, p. 244)
In this passage a chase between two characters occurs, so we know that
if one of the characters is mentioned, the other is not far behind. Thus in this
case pronouns can be used in spite of intervening mentions of another charac­
ter.
Another instance of this pattern is given below.
"Oh, Han!" she cried, and burst into tears once more. She buried her­
self in his embrace.
His anger turned slowly to confusion and dismay', as he found himself
wrapping his arms around her, caressing her shoulders, comforting her. 'Tm
sorry," he whispered into her hair. 'I'm sorry." He didn't understand, not an
iota — didn't understand her, or himself, or his topsy-turvy feelings, or
women, or the universe. All he knew was that he'd just been furious, and
now he was affectionate, protective, tender. Made no sense.
"Please...just hold me," she whispered. She didn't want to talk. She just
wanted to be held.
He just held her. (Return of theJedi, p. 122)
Here a conversation between Han and Leia is carried out with both
characters referred to using pronouns. Remember that the observation about
anaphora in this case is that the introduction of another character and his/her
plans and actions does not constitute a block to pronominalization if that
character and the first character are engaged in a close interaction (so that the
plans and goals of both characters are always relevant).
The fourth outcome of the basic pattern is that when a second character
is introduced that is not interacting directly with the first, pronominalization
seems to be blocked. In the following passage, for example, the second
character is interacting directly with a group of sharks rather than with the
first character (the first character is watching the human-shark interaction).
A mention of the second character's actions thus blocks pronominalization of
the first character (Jobim is Paloma's father):
166 BARBARA FOX

Now the other two sharks were rising. They kept their distance from the
larger one, seeming to defer to it, but they were growing bolder. And though
they were definitely smaller than the other shark, relativity was the only
comfort: Her [i.e., Paloma's] father was six feet tall, and each of these sharks
was at least as long as he was tall.
Jobim held the half-needlefish out to the big shark and wiggled it with
his fingertips. The circling pattern grew tighter. Now the shark was missing
Palorna by only three or four feet as it swept by. (The Girl of the Sea of Cor-
tez, p. 76)
In the following excerpt from Alien, we have a mention of Ripley
("reached her from outside") followed by an introduction of the goals and
actions of the alien. Notice that in this case the alien is not interacting with
Ripley, but with Jones the cat. We would thus expect, from the patterns pre­
sented above, that a next mention of Ripley would be done with a full NP;
indeed, the second paragraph begins with a reference to Ripley, done with a
full NP.
The locker was not airtight. A distinctive moaning reached her from
outside. Distracted, the alien left the port to inspect the source of the strange
noise. It bent, lifted the sealed catbox, causing Jones to howl more loudly.
Ripley knocked on the glass, trying to draw the creature's attention
away from the helpless animal. (Alien, p. 267)
In the next passage, following a mention of Leia an interaction between
two other characters (an Ewok and an enemy soldier) is initiated. After that
confrontation, a mention of Leia is performed with a full NP. The observa­
tion here is that, in spite of the small gap between the two mentions of Leia,
a full NP is warranted for the second mention because the intervening mate­
rial introduces the actions of another character, and these actions are not
interactive with Leia but with another, third, character.
Slowly she [i.e., Leia] swiveled, to find an Imperial scout standing over
her, his weapon leveled at her head. He reached out his hand for the pistol
she held.
"I'll take that," he ordered.
Without warning, a furry hand came out from under the log and jabbed
the scout in the leg with a knife. The man howled in pain, began jumping
about on one foot.
Leia dove for his fallen laser pistol. (Return of the Jedi, p. 95)

6. Demarcation of Narrative Units

The basic pattern established in section 1.5 does not cover all of the uses
of anaphors in the narrative texts, however. In particular, under certain con-
 ANAPHORA IN POPULAR WRITTEN ENGLISH NARRATIVES 167

ditions full NP's are used where pronouns would have been possible. In the
following passage, for example, full NP's are used in a fast-paced confronta­
tion (fight/chase) where we could have expected pronouns:
Susan herself was amazed at the effect and stepped into the amphithea­
ter, watching D'Ambrosio's fall. She stood there for an instant, thinking that
D'Ambrosio must be unconscious. But the man drew his knees up and pulled
himself into a kneeling position. He looked up at Susan and managed a smile
despite the intense pain of his broken rib.
"I like 'em...when they fight back," he grunted between clenched teeth.
Susan picked up the fire extinguisher and threw it as hard as she could
at the kneeling figure. D'ambrosio tried to move... (Coma p. 241)
And in the next passage, full NP's are used to refer to one character,
although no plans or actions of other characters have been introduced (the
scene takes place in a hospital):
Susan found a concrete wall which she guessed housed the elevator
shafts. Then she discovered that the corridor of the OR area had a dropped
ceiling. Beyond the OR corridor, over what was probably part of central sup­
ply, Susan could see that the maze of pipes and ducts running through the
ceiling space converged in what seemed a tangled vortex. Susan guessed that
was the location of the central chase which housed all the piping and ducts
coursing vertically in the building. (Coma, p. 234)
Another instance of the use of full NP instead of pronoun in a fast-paced
confrontation scene is given below.
Spinning in the chair, her [i.e. Ripley's] heart missing a beat, she saw,
not the creature, but a form and face now become equally alien to her.
Ash smiled slightly. There was no humor in that upturning of lips.
"Command seems a bit too much for you to handle. But then, proper leader­
ship is always difficult under these circumstances. I guess you can't be
blamed."
Ripley slowly backed out of her chair, carefully keeping it between
them. Ash's words might be conciliatory, even sympathetic. His actions were
not. (Alien, p. 243)
I would like to argue that the key to this use of full NP in the narrative
texts lies in the structural organization being displayed by the writer. Let us
suppose, along with researchers in the Story Grammar paradigm, that stories
basically concern the reactions, plans, and actions of different characters. In
small stories we are often given what we might call background setting infor­
mation, then an initiating event ( a happening which causes a character to
respond in some way), a reaction to that event on the part of a character, a
plan to do something about the situation brought about by the initiating
168 BARBARA FOX

event, and then a carrying out of that plan, with some final outcome of the
action. It is, in a sense, a basic problem-solution structure, with the solution
divided up into parts (reaction, plan, action). The basic pieces of a story,
according to this view, are thus:
.1. background information (e.g., setting)
2. initiating event
3. reaction
4. plan
5. action
6. outcome
And these cluster into higher-level units, the most common of which is
the development structure: reaction, plan and action form a development.
There are obviously other units which can occur in extended narratives, but
for our purposes here, let us suppose that the major higher-level unit is the
development structure.
Below I have given a passage which illustrates the development struc­
ture. The first paragraph describes the initiating event (Jones' yowling at
something). The second paragraph presents Ripley's reaction to that initiat­
ing event and thus begins the development. The third paragraph gives two
plans (the first of which is discarded) and starts the action piece.
She [Ripley] did not see the massive hand reaching out for her from the
concealment of deep shadow. But Jones did. He yowled.
Ripley spun, found herself facing the creature. It had been in the shuttle
all the time.
Her first thought was for the flamethrower. It lay on the deck next to the
crouching alien. She hunted wildly for a placed to retreat to. There was a
small locker nearby. Its door had popped open from the shock of the expand­
ing gas. She started to edge toward it. (Alien, pp. 266-267)
Let me now propose that many full NP's in narratives which occur where
one could have expected pronouns are functioning to signal the hierarchical
structure of the text; in other words, I would argue, full NP's are used to
demarcate new narrative units. That is, full NP's can be shown to correlate
with the beginnings of development structures, where the development struc­
ture is seen as the major recurring unit in narratives. In the passage from
Alien immediately above, for example, a full NP is used to begin the develop­
ment structure ("Ripley spun"). Additional examples of this correlation
between full NP and the beginning of a development structure are given
below.
 ANAPHORA IN POPULAR WRITTEN ENGLISH NARRATIVES 169

But the man drew his knees up and pulled himself into a kneeling posi­
tion. He looked at Susan and managed a smile despite the intense pain of his
broken rib.
"I like 'em...when they fight back," he grunted between clenched teeth.
Susan picked up the fire extinguisher and threw it as hard as she could
at the kneeling figure. (Coma, p. 241)
In this passage, the first slot of a development structure contains a full
NP, even though the referent was mentioned approximately 3 clauses before
(depending on what one counts as a clause). It is therefore clear that simple
distance is not at issue here; in addition, the basic pattern established for two
characters interacting seems to be superseded. The critical pattern to be
noticed here is the correlation between the beginning of a narrative unit (the
development structure) and the use of a full NP.
That did it for the Ewok. He jumped up, grabbed a four-foot-long
spear, and held it defensively in her [i.e. Leia's] direction, Warily he circled,
poking the pointed javelin at her, clearly more fearful than aggressive.
"Hey, cut that out,"Leia brushed the weapon away with annoyance.
(Return of the Jedi, p. 94)
Here, as in the preceding example, the first slot in a development struc­
ture contains a full NP, where, by measures of distance or the basic pattern
demonstrated earlier, we could have expected a pronoun. I would like to
argue that it is the demarcation of a new narrative unit which "produces" the
full NP.
Another example of this association between a new narrative unit and
full NP follows:
She shook her head and looked at the pirogue and at the horizon and at
the softly rolling sea swells. She was in at least ten, maybe twenty, fathoms
of water. Then what was she standing on? For, there was no question that she
was standing on something. She drained water from her mask and put her
face down and saw that the manta had come beneath her and had risen, like
a balloon, until it rested just at her feet.
Did it want something? Was it injured again? Paloma took a breath and
knelt on the manta's... (The Girl of the Sea of Cortez, p. 226)
Another instance:
She [Ripley] did not see the massive hand reaching out for her from the
concealment of deep shadow. But Jones did. He yowled.
Ripley spun, found herself facing the creature. It had been in the shuttle
all the time. (Alien, p. 267)
With this collection of examples, and the many others that could be
170 BARBARA FOX

added to it, it seems clear that on top of the patterns for pronominalization I
established above there tends to be an association between the beginning of
a narrative unit (typically the first slot of a development structure) and the use
of a full NP. I do not mean to suggest with this statement that all development
units are started with full NP's; rather, this is a slot in which full NP's can
occur even though we might have expected pronouns. In comparison within
the boundaries of development units there tend not to be full NP's (assuming
by other patterns described that pronouns are expected). Examples illustrat­
ing this predominance of pronouns within narrative unit boundaries are pre­
sented below (see Tomlin (this volume) for similar findings about anaphora
and episode structure).
She stuck her head out into the engine room. It was still deserted.
Smoke curled up around her, making her cough. She climbed out, kicked the
disc back into place, leaving enough of a gap for air to reach the fire. The she
strode resolutely toward the engine-room control cublicle. (Alien, p. 263)
Her mind evaluated every item and discarded it. And then, as she
looked at the wood fibers, she saw beside them other fibers, closely woven
though not as thick as the wood, and she had the answer: her dress. She could
stuff her dress into the hole, and it would keep the water out. The fabric was
already saturated with salt water, so no more could penetrate it. And packed
tightly in a ball, the cloth fibers would bind and become nearly waterproof.
She peeled the sodden shift up over her head, then ducked under the
pirogue and, from the inside, packed the cloth into the hole. It made a tight
plug — nothing that could survive a pounding in a heavy sea, but secure
enough for an easy paddle on calm water.
She ducked out again, hauled herself up onto the bottom and reached
over and grabbed the far edge. (The Girl of the Sea of Cortez, p. 170)
Susan stared at the valve. She looked at the other gas lines coming up
the chase. There were no similar valves on any of the other lines. With her
finger she examined the valve. (Coma p. 235)
We find then that the initial slots of the relevant discourse units are
associated with full NP's, while the non-initial slots of these units are
associated with pronouns.

7. Same-gender referents
The core pattern for anaphora in the environment of same-gender refe­
rents in narratives can be roughly stated as follows:
If a character has been mentioned as participating in an event/action,
then that person can subsequently be referred to with a pronoun, until
another character of the same gender is mentioned participating in another
 ANAPHORA IN POPULAR WRITTEN ENGLISH NARRATIVES 171

event/action. If two referents of the same gender are involved in the same
action, then the grammatical subject of the clauses describing that action can
be referred to in the next event-line mention with a pronoun. The other non-
subject NP will have its referent indexed on the next event-line mention with
a full NP.
Passages illustrating the use of a pronoun to refer to the last person of the
appropriate gender involved in an action follow. Notice that in the sequence
X actionverb, Y actionverb, PRO
where X and Y are same-gender referents, the pronoun refers to Y, not
X. A full NP must be used to perform a second reference to X.
Before Vader could gather his thoughts much further, though, Luke
attacked again — much more aggressively. He advance in a flurry of
lunges... {Return of the Jedi, p. 155)
He [i.e., Luke] took a step back, lowered his sword, relaxed, and tried
to drive the hatred from his being. In that instant. Vader attacked. He lunged
half up the stairs... (Return on the Jedi, p. 156)
Passages illustrating the use of a full NP to refer to a character other than
the last one involved in an action follow.
When Vader moved to parry, Luke feinted and cut low. Vader counter-
parried... (Return of the Jedi, p. 154)
She placed the transparent mask over Ripley's mouth and nose, opened
the valve. Ripley inhaled. (Alien, p. 238)
We can thus see that if a character is involved alone in the current action,
then in the next action it can be referred to with a pronoun; if, on the other
hand, the next action involved another character of the same gender — with­
out mention of the first character — it will be done with a full NP.
If two characters of the same gender are mentioned in the current action,
however, the referent realized by the subject NP of the clause can be refer­
enced using a pronoun in the next action; the other referent must be refer­
enced using a full NP.
In the following same-gender examples, we have two characters of the
same gender involved in the same action. In these instances, the person refer­
red to by the subject NP is referenced with a pronoun in the next event-line
mention, while the non-subject NP is referenced with a full NP.
Subject NP becomes pronoun :
Vader paced the area like a cat, seeking the boy; but he wouldn't enter
the shadows of the overhang. (Return of the Jedi, p. 157)
172 BARBARA FOX

Trembling, he stood above Vader, the point of his glowing blade at the
Dark Lord's throat. He wanted to destroy this thing of Darkness, this thing
that was once his father... (Return of the Jedi, p., 159)
Palpatine raised his spidery arms toward Luke: blinding white bolts of
energy coruscated from his fingers... {Return of the Jedi, p. 160)
Lambert set one of the oxygen tanks down next to her friend. She placed
the transparent mask over Ripley's mouth and nose... (Alien, p. 238)
In all of these examples, the referent realized by the subject NP in the
first action is pronominalized in the next.
Non-Subject NP becomes full NP
When Luke pushed Vader back to break the clinch, the Dark Lord6 hit
his head on an overhanging beam in the cramped space. (Return of the Jedi,
pp. 158-159)
He bound the boy's blade with his own, but Luke7 disengaged... (Return
of the Jedi, p. 156)
Jabba motioned 3PO to his side, then mumbled an order to the golden
droid. 3PO8 stepped up... (Return of the Jedi, p. 42)
In these passages, the referent realized by the non-subject NP in the first
action is not pronominalized in the next action, being realized by a full NP.

8. Conclusion
In the preceding sections we have seen that a small number of patterns
based on the structuring functions in narrative — such as event-line, develop­
ment structures, plans and actions — describe a very large proportion of the
anaphors in the narrative texts examined, including in the environment of
two referents of the same gender. I have also shown, en route, that
approaches which take a more linear view of narrative texts are less than
effective in accounting for the anaphoric patterning displayed by the popular
written narratives examined in this study. There is a reasonable conclusion to
be drawn from this fact: if we are to understand the use of various linguistic
devices in naturally-produced texts, we must accept as a major factor in such
use the structure of those texts. One of the crucial tasks ahead of us, then, is
the development of models of text structure which can be fruitfully used in the
study of linguistic coding. A very simple model of narrative structure was
utilized in the present study; future work in this area will demonstrate the
value of richer, more complex models of a range of text-types.
 A N A P H O R A IN P O P U L A R W R I T T E N ENGLISH N A R R A T I V E S 173

NOTES
*Many people provided helpful comments on earlier drafts of this paper, including Susanna Gum­
ming, Jack DuBois, Sandy Thompson, Russ Tomlin and Fay Wouk. While working on this topic,
I was partially supported by a National Science Foundation graduate fellowship; time to pursue the
research further was generously supported by the Institute of Cognitive Science, University of Col­
orado, Boulder.
1) Givón does mention units such as paragraph, section, chapter, etc., but makes no attempt
to incorporate them into the body of the study.
2) The importance of hierarchical structure for anaphora has been demonstrated for non-narra­
tive discourse by, among others, Grosz (1977), Reichman (1981), Fox (1984), and Linde (1979).
3) Throughout this chapter, the relevant anaphors in real examples are underlined.
4) The other character in this case does not have to be human. It could be some other sort of ani­
mate creature.
5) The event-line is what Hopper and Thompson (1980) have called the foregrounded portion of
a narrative; that is, it is the temporally sequenced events/actions which occur. It thus does not
include descriptions of places, internal monologues, statements of general truths, etc.
6) Vader is often called the Dark Lord,
7) "Luke" and "the boy" are coreferential.
8) 3PO is "the golden droid".

REFERENCES

Bobrow, D, and A. Collins. 1975. Representation and Understanding. New

York: Academic Press.
Clancy, P. 1980. "Referential choice in English and Japanese narrative dis­
course." In W. Chafe (ed.) The Pear Stories: Cognitive, Cultural, and Lin­
guistic Aspects of Narrative Production. New Jersey: Ablex Publishing
Company.
Chafe, W. (ed.) 1980. The Pear Stories: Cognitive, Cultural, and Linguistic
Aspects of Narrative Production. New Jersey: Ablex Publishing Company.
Fox, B. 1984. "Discourse structure and anaphora in written and conversa­
tional English." Unpublished Ph.D dissertation. University of California,
Los Angeles.
Givón, T. (ed.) 1983. Topic Continuity in Discourse: A Quantitative Cross-
Linguistic Study. Amsterdam: John Benjamins.
. (ed.) 1979. Syntax and Semantics, vol. 12. New York: Academic Press.
Grimes, J. (ed.) 1978. Papers in Discourse. Arlington, Texas: SIL.
Grosz, B. 1977. "The representation and use of focus in dialogue understand-
174 BARBARA FOX

ing." SRI Technical Note, 151.

Hinds, J. 1978. Anaphora in Discourse. Edmonton, Canada: Linguistic
Research, Inc.
Hopper, P. and S. Thompson. 1980. 'Transitivity in discourse and gram­
mar." Language 56.251-299.
Levinsohn, S. 1978. "Participant reference in Inga narrative discourse." In J.
Hinds (ed.) Anaphora in Discourse." Edmonton, Canada: Linguistic
Research, Inc.
Linde, C. 1979. "Focus of attention and the choice of pronouns in discourse."
In T. Givón (ed.) Syntax and Semantics, vol. 12. New York: Academic
Press.
Reichman, R. 1981. "Plain speaking: a theory and grammar of spontaneous
discourse." Bolt, Beranek and Newman, report no. 4681.
Rumelhart, D. 1975. "Notes on a schema for stories." In D. Bobrow and A.
Collins (eds.) Representation and Understanding. New York: Academic
Press.
 BEYOND FOREGROUND AND BACKGROUND

T. GIVÓN
University of Oregon

1. Introduction
There are two separate questions that one may wish to ponder regarding
foreground and background in discourse. The first is a purely functional ques­
tion:
"How valid is the binary foreground/background distinction?"
The second is a question about coding:
"How strong or consistent is the correlation between the presumed fore­
ground/background functional distinction and some binary distinctions
in syntax, say main/subordinate clause or some tense-aspect distinc­
tion?"
In this paper I will discuss primarily the first, functional question. I will raise
as many problems that I see impinging upon the validity of this binary distinc­
tion, without necessarily suggesting that we scrap it altogether. I will then cite
one particular series of studies, all leading to and including Violeta Ramsay's
paper [in this volume], to illustrate where some expansion of the foreground/
background tradition might be useful. The paper will lead to, but not directly
attempt to tackle, the second question about what the coding level may or
may not be doing.

2. Problems with the binary functional distinction foreground/background

2.1. The dynamic nature of discourse grounding
The tradition of looking at text once it has been produced and trans­
cribed, then making objective-looking determinations as to what portions are
presupposed, shared, old information or background, and what portions are
asserted, new information or foreground, is a static tradition in discourse
studies. It construes the text from the outside as an objective entity, a finished
176 T. GIVÓN

product. This is obviously one valid perspective, that of the investigator qua
reader. But the production of the text, and the interaction during which it is
produced, is a dynamic process. A proposition that is asserted ('foreground')
at point n in the discourse becomes — in the absence of challenge from the
hearer — a shared presupposition ('background') at point n plus 1. The deter­
mination of what is foreground and what is background in actual discourse
must, then be relative to a particular point in the discourse, the particular
frame. And the frame in discourse tends to shift, to be reassembled, to be re-
framed.

2.2. The correlation between background, presupposition and oid informa­

tion
The connection between what we intuitively consider "background" and
our only slightly less intuitive notions of "presupposition" and "old informa­
tion" is considerably less than perfect. Let us consider a simple example to
illustrate one aspect of this. As pointed out elsewhere (Givón, 1982a],
restrictive relative clauses may modify both definite and indefinite nouns, as
in:
(1) a. The man wearing a frilled dress came into the room and..
b. A man wearing a frilled dress came into the room and...
Now, while in (la) above the RRC is presupposed by the speaker to be known
to the hearer, no such presupposition could possibly obtain in (lb), where
both 'a man' and the proposition underlying 'wearing a frilled dress' are not
assumed to be known to the hearer. Nevertheless, there is some sense, how­
ever intuitive, in which the proposition in both RRCs in (1) is backgrounded.
But that sense could not correspond to either "presupposition" or "old infor­
mation".
In a similar vein, consider the two adverbial subordinate clauses:
(2) a. Because John left, Mary left too.
b. If John leaves, Mary will leave too.
The one in (2a) is presupposed and thus presumably not asserted. The one in
(2b) is not presupposed. In fact, it has no truth value strictly speaking, being
in the irrealis mode. Thus it is not asserted either. Still, there is some intuitive
sense, expressed by linguists over the years [see eg, Haiman, 1978], in which
both adverbial clauses in (2) are equally "background". But whatever that
sense is, it could not correspond to the notion of "presupposition".
 BEYOND FOREGROUND AND BACKGROUND 177

2.3. The correlation between foreground, sequentiality and main-line or

"gist"
A better correlation, although still intuitive, can be seen between our
notion of "foreground" and the semantic notion "information presented
sequentially" or the pragmatic notion "main line (or 'gist') of the informa­
tion". Such correlations have been proposed by many in the past (eg. Givón,
1977; Hopper, 1979; Hopper and Thompson, 1980; Givón, 1982c; inter alia],
But however intuitively appealing, this correlation remains problematic. To
begin with, the coherence structure of many discourse types is not temporally
determined, so that sequentiality — for text such as academic lectures — is
not a useful tool in determining the "gist". Finally, the notion of "fore­
ground" and "gist" are not independent of each other, leastwise not in the
context of our past discussions. Rather, they seem to define each other in a
rather circular fashion. Our intuitive definitions of foreground —- and thus
also background — seem to again lead us onto a rather rocky terrain.

2.4. Binary vs. scalar pragmatic space

Morpho-syntax has a strong tendency to make all functions coded by it
look discrete, and quite often binary. This is inherent in the nature of the cod­
ing instrument. Structure always discretizes function. As I will suggest later
on, this facet of structure may indeed point out to a process of selectivity, by
which only a few — often only two — functional "peaks" along a continuum
become strongly coded, thus often producing the illusion that the semantic or
pragmatic dimension itself is discrete and binary. In many areas of discourse
pragmatics, however, the evidence stronly suggests that the underlying
dimension is itself non-discrete, and even its associated code is n-ary rather
than binary. A conspicuous coding scale of this kind is that of topic continuity/
predictability [Givón,eéd., 1983], which scales noun-marking devices in a
fashion that is highly predictable cross-linguistically. Thus, consider the fol­
lowing widely attested scale:
(3) MOST PREDICTABLE TOPIC
a. zero anaphora
b. unstressed/clitic pronouns
c. stressed/independent pronouns
d. definite nouns
e. modified definite nouns
LEAST PREDICTABLE TOPIC
178 T. GIVON

In most languages this scale is much richer and has many more points, involv­
ing word-order and other syntactic devices. The functional, psychological
dimension involved here concerns how accessible the topic is to the hearer,
given distance from prior mention in discourse, degree of referential interfer­
ence from other referents, amount of semantic redundancy available in the
proposition and amount of thematic redundancy available in the discourse.
This scale of topic predictability is indeed one of the two major components
of an empirically-defined notion of "topicality", the anaphoric or backward-
scanning component. The topic predictability scale applies to nominal partic­
ipants, but there are reasons to assume that a similar scale may apply to prop­
ositions, in terms of their "thematicity", "foregroundedness" etc. Let us
examine a few simple examples suggesting such scales.
While local presupposition is defined as a discrete, binary distinction,
pragmatic presupposition may be n-ary and scalar, determined perhaps by
the strength of the speaker's belief as to how accessible some proposition is in
the hearer's mind. Along such a scale, construction types may be ranked
along at least three major point (Givón, 1982a):
(4) a. Most presuppositionah WH-questions, cleft/focus
constructions, restrictive
relative clauses, some verb
complements
b. Intermediate negatives, if-clauses yes/no questions
c. Least presuppositionah main-declarative-affirmative
(asserted) clauses
Next, it can be shown that the three major speech-acts, declarative,
imperative and interrogative, allow finer scalar gradations to fill the func­
tional space between them. As a simple example consider the following gra­
dation between declaratives and yes/no questions (for further detail see
Givón, 1982a, 1984a; Tsuchihashi, 1983; Bolinger, 1975):
(5) a. John is here. [prototype declarative]
b. John is here, isn't he?
c. John isn't here, is he?
d. John is here?
e. You think John is here?
f. Is John here? [prototype interrogative]
Such a continuum is probably founded upon non-discrete psychological
dimensions such as (Givón, 1982b, 1984a):
 BEYOND FOREGROUND AND BACKGROUND 179

(6) a. The speaker's subjective certainty

b. Evidentiary support available to the speaker
c. Strength of the speaker's wish to seek confirmation from the
hearer
d. Strength of the speaker's willingness to tolerate challenge
from the hearer
e. Status, power and affect relations between speaker and hearer
Next, consider the scalar nature of inter-clausal conjunctive particles,
where a graded continuum exists with an iconic relation between the size of
the interruption between two clauses and the degree of discourse discon­
tinuity between them. Now, here the clauses involved are sequentially deli­
vered main-declarative-asserted clauses, i.e. our prototypical "foreground".
Are they then scaled as to the degree of their "foregroundness"? This phe­
nomenon is very wide-spread cross linguistically, and also involves the sys­
tem of pauses and written punctuation marks. To illustrate it, consider the
following examples from Ute [Givón, 1980, Ch. 17]:
(7) a. O (...he opened the door, O stepped in, O paused and ..)
b. x-'ura (John entered the room, then he stopped...)
c. '-vway-ax (...He finished; so then he...)
d. 'u-vway-ax-'ura (...He finished. So well, then she...)
e. 'u-vway-ax-'ura-'uru (...He finished. So well, it was then that
she...)
f. 'u-vway-ax-'unuv-'ura-'uru (...So it was then, later on, that
she...
The scale in (7) in Ute, English and other languages, involves gradual
increase in the degree of discontinuity, unpredictability or surprise at
that point in the discourse. And the unpredictability may arise from any of the
factors which together make the tapestry of discourse coherence (or their
combinations):
(8) a. Topics-participants continuity
b. Temporal continuity
c. Spatial continuity
d. Action or theme continuity
In sum, then, it is not yet clear that we are justified in assuming that a dis­
crete, binary distinction of foreground/background will remain useful for
characterizing discourse coherence, once we have become more intimately
acquainted with the facts and their subtleties.
180 T. GIVÓN

2.5. Anaphoric vs. cataphoric grounding

As should have been made clear by now, the notion of "grounding" —
even if we can justify, ultimately, its discreteness and binarity — is really a
composite of two separate psycho-communicative processes. The first one is
essentially anaphoric, involving grounding of a particular point in the dis­
course vis-a-vis the preceding discourse background; or, to be more precise,
grounding vis-a-vis what the speaker can assume about shared knowledge
with the hearer. Here we find our traditional correlation of "background"
with "presupposition", "old information" or "topic". The second is a
cataphoric process, involving clues the speaker gives the hearer at a particular
point in the discourse as to how to ground it vis-a-vis the following discourse,
particularly in terms of thematic/topical importance. Here we find our tradi­
tional correlation of "foreground" with the "gist" of the information.
The differentiation between anaphoric and cataphoric orientation — or
grounding — in discourse is very well known to us from the grammar of nom­
inal topics ('referents'). Thus, for example, definite articles, unstressed
anaphoric pronouns and zero anaphora are all primarily anaphoric devices.
On the other hand, indefinite articles ar primarily cataphoric devices, alerting
the hearer as to what to expect in the subsequent discourse. Many topic-
marking devices, however, perform a specific mix of anaphoric and
cataphoric grounding functions. For example, assigning an NP to the
subject clausal position involves both anaphoric continuity and cataphoric
persistence. On the other hand, devices such as subject L-dislocation or sub­
ject stressed/independent pronouns tend to signal anaphoric dis-continuity
but cataphoric persistence or importance. Finally, in Ute the use of clitic pro­
nouns/agreement, as contrasted with zero anaphors, tends to signal — for
subjects — anaphoric continuity but cataphoric dis-continuity. That is, they
tend to be used at the end of equi-topic thematic paragraphs (for many details
supporting this discussion see Givón, ed., 1983).
So far we have considered only examples from the grammar of referential
coherence or the grounding of nominal topics. But similar examples can be
cited from the grammar of thematic coherence. For example, in Biblical
Hebrew (Givón, 1977) the perfect aspect tends to be used at the opening of
new thematic paragraphs, where not only referents but also themes change.
On the other hand, a special tense-aspect in Ute — the -puay-aga suffixal
combination (Givón, 1980, Ch. 4) — signals the end of a large thematic
episode and tags the clause marked by it as a "recapitulation" or "moral" of
the story. As we shall see further below, other devices in discourse are used
 BEYOND FOREGROUND AND BACKGROUND 181

for such combination of grounding, vis-a-vis both the preceding and following
discourse. Our notion of grounding thus requires further elaboration and
clarification.

3. Grounding and syntactic order

In this section I will discuss the relation between grounding and syntactic
order, reviewing some of the recent literature and leading toward a discussion
of Ramsay's paper [in this volume]. In an ingenuous paper, Haiman (1978)
presented massive grammatical evidence to suggest that conditionals (TF-
clauses') are "topics". His notion of "topic" is closely related to "pragmatic
presupposition" or "background old information", without defining these
directly as criteria. Haiman, further, did not differentiate between preposed
and postposed conditionals, perhaps because the impetus — and much of the
data — for his study came from a clause-chaining New Guinea language
which does not allow the postposing of adverbial clauses.
My own early forray into this subject came in Givón (1982a), where I
suggested, based on constructed examples with minimal context, that adver­
bial clauses are "topical" or "pragmatically presupposed" only when they pre­
cede the main clause. But they are "focus" or "asserted information" when
they follow the main clause. The examples I cited there were of the type:
(9) a. Context: Under what conditions would you do it? [M-clause
topic]
b. Response: I'll do it if he leaves. [Postposed ADV-clause]
c. Context: What will you do if he leaves? [ADV-clause topic]
d. Response: If he leaves, I'll follow him. [Preposed ADV-
clause]
This followed the classical dogma of our times since the Prague school,
namely that topical material preceded and focal/asserted material follows in
natural pragmatically-determined order. And since the presumed results
of this rather cursive investigation conformed with the received wisdom,
the matter rested there briefly while I was casting around for a more rigorous
empirical paradigm to test these 'results'.
Two studies made it clear to me that I was probably on the wrong track.
First, Thompson's [1985] paper on preposed and postposed purpose clauses,
which was read at the WECOL Conference in Eugene in 1983. The facts
emerging from Thompson's study, even in a preliminary fashion, suggested
that our notions of "topic" and "foregrounding/background", at least as
182 T. GIVÓN

applied to adverbial clauses, must be radically revised, if not wholly dis­

carded. The following comprises my own interpretation of Thompson's find­
ings, and is not necessarily stated explicitly in her paper:
(i) Preposed P-clauses depend more, for their interpretation, on the
preceding thematic context — a wide chunk of it. On the other
hand, postposed P-clause depend for their interpretation primarily
on the immediately-preceding main clause;
(ii) The thematic link of preposed P-clauses if characteristically wider,
more abstract and plugs into higher thematic nodes. On the other
hand, the thematic link of postposed P-clauses is more concrete,
most commonly referring to the motivation of the participant(s)
mentioned in the directly preceding main clause;
(iii) Preposed P-clauses are more likely to appear at major thematic
breaks in the discourse. They thus tend to connect or ground the
subsequent thematic unit vis-a-vis the preceding one. In terms of
our earlier discussion, they thus function as a combination of
anaphoric and cataphoric grounding device. On the other hand,
postposed P-clauses are more likely to appear in the middle of
thematic units, and are much less likely to function as a grounding
or "re-orientation" device;
(iv) The anaphoric connection of preposed P-clauses is primarily
thematic. Their cataphoric connection to the main clause itself is
primarily semantic, although this does not exclude a parrallel
thematic grounding function vis-a-vis the subsequent discourse.
These tentative conclusions concerning pre-posed P-clauses may be diagram­
med as follows:
(10)
 BEYOND FOREGROUND AND BACKGROUND 183

It thus seems that word order-has a powerful effect on the way some subordi­
nate clauses are used in grounding — or re-orientation — in discourse. But
even Thompson's preliminary results suggest that this function may not be
construed in the simple, binary terms of foreground/background. And as is
made clear in Ramsay's paper [this volume], the phenomenon applies to
other types of adverbial clauses as well:
Another strand of research pertains to the topic continuity studies, as
related to pragmatically-controlled word-order variation. An overview of
these results, starting from Givón [1977] and culminating in Givón [ed., 1983]
suggests the following general principle, applicable across a great variety of
languages and language families:
(11) "More predictable/continuous topic NPs follow the verb;
less predictable/continuous topic NPs precede the verb".
Virtually the same conclusions are outlined in Mithun [this volume] with
respect to Iroquois word-order flexibility. And it is clear that the factors
affecting the use of word-order flexibility to code relative topic predictability
are the same ones discussed in relation to the quantity scale of topic con­
tinuity/predictability, (see (3), section 2.3., above). All this leads one to sus­
pect that the results reported by Thompson (1985) and Ramsay (this vol­
ume) involve a similar principle of ordering, namely that proposed adverbial
clauses are used in environments of greater thematic dis-continuity, and
post-posed ones in environments of greater thematic continuity in the
discourse.

4. Ramsay's study of IF- and WHEN-clauses

The initial hypothesis underlying Ramsay's study has two parts to it, one
methodological and general, the other empirical and specific:
(a) General: It will be unlikely that human language will use same cod­
ing device — in this case word-order variation — to code
two totally separate functional processes, one for the
relative topicality of NPs, the other for the relative
grounding status of clauses.
(b) Specific Preposed ADV-clauses code discourse contexts of lower
thematic continuity, just like pre-verbal NPs code dis­
course contexts of lower referential continuity.
Our general hypothesis (a) is merely a reiteration of the validity of of iconicity
184 T. GIVÓN

in our language code. Our specific hypothesis (b) is the one for which Ramsey
has attempted to devise operationalized discourse measurements, in order to
test its empirical validity. Ramsay's operationalized hypothesis may be sum­
marized as follows:
(i) Referential continuity
"Preposed ADV-clauses will exhibit lower referential continuity
vis-a-vis their (following) M-clauses than would post-posed ADV-
clauses. Further, preposed ADV-clauses will exhibit higher refer­
ential continuity with the preceding discourse — here excluding
the M-clause — than would postposed ADV-clauses. Finally, pre­
posed ADV-clauses will exhibit higher referential continuity with
the preceding discourse than would their M-clauses".
(ii) Thematic continuity
"Preposed ADV-clauses will exhibit their discourse coherence —
or continuity — vis-a-vis the preceding discourse in higher, more
abstract and wider-scoped thematic terms. Postposed ADV-
clauses, on the other hand, will exhibit their coherence/continuity
vis-a-vis the preceding discourse in lower, more concrete and nar­
rower-scoped semantic or referential terms".
(iii) Discourse juncture
"Preposed ADV-clauses will tend to appear at discourse juncture
which are more major, in terms of disruption/break in
(a) Referential continuity
(b) Temporal continuity
(c) Spatial continuity
(d) Sequential action continuity
(e) Thematic continuity".
The challenge of quantified discourse work is, here as elsewhere, the
development of discourse measurements that would operationalize specific
hypotheses as the one summarized above. Like elsewhere, doing science in
linguistics is learning to operationalize one's intuition, regardless what the
source of that intuition is to begin with.

5. Conclusions
5.1. At the functionallevel
While my musings above were aimed at shaking our faith in the validity
of the discrete, binary distinction foreground/background, they were
 BEYOND FOREGROUND AND BACKGROUND 185

not intended to suggest that the entire approach should be scrapped.

There are, to my thinking, two reasons for remaining cautious and ten­
tative about our ultimate conclusions in this respect. The first is historical in
nature, weighing the utility of theoretical distinctions we propose in terms of
the slow, laborious progression of our science. I will express it as the follow­
ing observation:
(12) "Like all discrete, binary distinctions we have constructed in the
past, the foreground/background distinction is both useful and
dangerous. It is useful in carrying us the first step toward a func­
tion-based definition of an important strand in the thematic
coherence of discourse. It is dangerous if we wed ourselves to it
regidly and do not eventually trade it in for more elaborate, more
specific, less circular and empircally better grounded notions".

5.2. At an intermediate level between function and structure

It is hardly an accident that our structural code discretizes and in general
reduces the number of non-discrete and (potentially) infinite functional
dimensions. While the psycho-cognitive dimensions which underly semantics
and pragmatics may indeed be scalar and non-discrete, the imperatives of
processing within finite time require discretization and reduction along any
functional-cognitive continuum. Such reduction and discretization — up to
binarization — must proceed according to which functions are more impor­
tant or urgent, and which ones are secondary and less urgent. There might
therefore be some cogent reasons why, in spite of all I said in section 2.,
above, there is still some room left for the binary distinction of foreground/
background. But such a distinction must be grounded more firmly than
before in non-circular, empirical studies.

5.3. At the coding level

We have taken for granted for too long now that there must be a strong
correlation between main-finite clause syntax and the foregrounding function
in discourse. In the main, our faith in this correlation hinged primarily on
Indo-European facts of grammar, and relatively little on cross-linguistics
studies of matching grammar with discourse. As a result, the "universals"
we have come up with in our earlier methodology (see for example Givón,
1979, Ch. 2; Hopper, 1979; inter alia) tend the characterize, at best, some
levels, genres or language families, rather than present a general and cohe-
186 T. G I V O N

rent picture of human language. In the area of empirical functional studies,

Tomlin's (1985) recent work suggests that the correlation between subordi­
nate and backgrounding in English, with the latter defined in terms of
(cataphoric) informational importance (rather than anaphoric informational
novelty), is both valid and highly problematic. In the area of grammatical
studies, we have the major phenomenon of clause-chaining discourse, which
seems to go against the grain of our earlier Euro-centric assumptions. It is
only by wedding the two methodologies — cross-language comparison and
function-based empircal studies of text — that we may hope to some day
resolve both ends of our semiotic equation — and thus the equasion itself.

NOTE
The other, Cataphoric component, scales the degrees of topic importance, often measured in terms
of persistence/frequency in discourse. See discussion further below. There are probably many finer
sub-gradation within each of the three. Tsuchihashi (1983) suggests that in Japanese the same con­
tinuum is coded by about 15 discrete verb-final particles.
Some of the relevant studies involve Biblical Hebrew (Givón, 1977, 1983 ed., Fox, 1983), Ute
(Givón, ed., 1983), Spanish (Bentivoglio, 1983), Colloquial English (Givón, 1983 ed.). Pidgin
English (Givón, 1984c), Tagalog (Fox, 1984), Indonesian (Rafferty, 1984 as well as in this volume;
Verhaar, 1984), Athabascan (Dryer, 1982), Coos (Dryer, 1983), Papago (Payne, 1984), and the
list is growing, to judge from Mithun's paper (in this volume).
For a general discussion of iconicity in syntax, see Haiman [ed., 1985]. An indirect empirical cor­
roboration of both our general and specific hypotheses may already be found in Tomlin's paper [in
this volume], where it is shown that devices used for coding referential ('topic-participant') con­
tinuity are also used — with exactly the same directionality — to code thematic continuity. Similar
observations were also made in Givón (1977, 1983 ed.).
As in, for example, Chafe (ed., 1980) or Givón (ed., 1983, 1985).

REFERENCES

Bentivoglio, P. 1983. "Topic continuity in spoken Latin American Spanish/ 1

in T. Givón (ed., 1983).
Bolinger, D. 1975. "Yes-no questions are not alternative questions", in H.
Hiz (ed.) Questions, Dordrecht: Reidell.
Chafe, W. 1980. (ed.) The Pear Stories, Ablex.
Dryer, M. 1982. "Tlingit: An object-initial language?" University of Alberta,
Edmonton (ms).
-----. 1983. "Coos word order", University of Alberta, Edmonton (ms) read
at the 13th WECOL Conference, U. of Oregon, Eugene.
 BEYOND FOREGROUND AND BACKGROUND 187

Fox, A. 1983. "Topic continuity in Biblical Hebrew narrative", in T. Givón

(ed., 1983).
Fox, B. 1985. "Word-order inversion and discourse continuity in Tagalog", in
T. Givón (ed., 1985).
Givón, T. 1977. "The drift from VSO to SVO in Biblical Hebrew: The prag­
matics of tense-aspect", in C. Li (ed.) Mechanisms for Syntactic Change,
Austin: University of Texas Press.
. 1979. On Understanding Grammar, NY: Academic Press.
. 1980. Ute Reference Grammar, Ignacio, CO: Ute Press.
. 1982a. "Logic vs. pragmatics, with human language as the referee:
Toward an empirically viable epistemology", J. of Pragmatics, 6.1.
. 1982b. "Evidentiality and epistemic space", Studies in Language, 6.1.
. 1982c. "Tense-aspect-modality: The Creole prototype and beyond", in
P. Hopper (ed.) Tense and Aspect: Between Semantics and Pragmatics,
TSL vol. 1, Amsterdam: J. Benjamins.
. 1983. (ed.) Topic Continuity in Discourse: Quantitative Cross-Language
Studies, TSL vol. 3, Amsterdam: J. Benjamins.
. 1984a. "The speech act continuum", in W. Chisholm (ed.) Interrogativ-
ity, TSL vol. 4, Amsterdam: J. Benjamins.
. 1984b. "Universals of discourse structure and second language acquisi­
tion", in W. Rutherford (ed.) Language Universals and Second Language
Acquisition, TSL vol. 5, Amsterdam: J. Benjamins.
. 1985. (ed.) Quantified Studies in Discourse, special volume of Text.
Haiman, J. 1978. "Conditionals are topics", Language, 54.3.
. 1985. (ed.) Iconicity in Syntax, TSL vol. 6, Amsterdam: J. Benjamins.
Hopper, P. 1979. "Aspect and foregrounding in discourse", in T. Givón (ed.,
1979) Discourse and Syntax, Syntax and Semantics vol. 12, NY: Academic
Press.
Hopper, P. and S. Thompson. 1980. "Transitivity in grammar and dis­
course", Language 56.4.
Mithun, M. [in this volume] "Is basic word order universal?".
Payne, D. 1984. "Information structuring in Pagago", paper read at the Con­
ference of Friends of Uto-Aztecan, UCSD. La Jolla, June 1984 (ms).
Rafferty, E. 1984. "Word order in intransitive clauses in High and Low Malay
of the Late 19th Century", University of Wisconsin, Madison (ms).
. [in this volume] [same title as above]
Ramsay, V. [in this volume] "The functional distribution of preposed and
postposed IF and WHEN clauses in written discourse".
188 T. GIVON

Thompson, S. 1985. "Grammar and written discourse: Initial and final pur-
pose clauses in English", in T. Givón (ed., 1985).
Tomlin, R. 1985. "Foreground-background information and the syntax of
subordination", in T. Givón ed., 1985).
. [in this volume] "Linguistic reflection of cognitive episodes".
Tsuchihashi, M. 1983. "The speech act continuum: An investigation of
Japanese sentence-final particles", J. of Pragmatics, 7.4.
Verhaar, J. 1984. "Topic continuity and word-order in Indonesian", Gon-
zaga University, Spokane (ms).
 THE USE OF PITCH PHENOMENA IN THE
STRUCTURING OF STORIES

LORRAINE E. KUMPF
University of California, Irvine

1. Introduction
In recent years, the concepts of foregrounding and backgrounding have
been employed by linguists in effort to characterize the use of grammatical
structures in discourse, particularly in narrative. Influential examples of such
work are Hopper (1979) and Givón (1982) on aspect morphology, and Hop­
per and Thompson (1980) on transitivity. In much of this literature, the gram­
matical properties of events are contrasted with the grammatical properties
of non-events. "Sequential clauses", "event clauses", "story line", or "foreg­
rounding" is opposed to the out-of-sequence, the other-than-event
phenomena such as scene-setting and evaluation, or the "backgrounding" of
material in the narrative. Although it is preferable to think of these properties
in relative rather than binary terms (that is, as information in relative relief
rather than as "foregrounded or "backgrounded") these linguists have shown
clearly that languages reflect event/non-event or sequence/non-sequence in
their grammars.
While sequence is inherent in narrative, it is also true that the sequential
aspects of a given narrative can be more or less important to the story: some­
times comprising little of the narrative's content, the actual event sequence
may also be unreflective of the storyteller's purpose. The non-narrative seg­
ments — which create conditions of time and place, explain, build motiva­
tion , evaluate, or have particular introductory or concluding purposes — per­
form functions which do not merely fill in extra information, but give the
story its value, making it worth telling and attending to.
In treating "foreground" and "background" in oral narratives, some
further consideration of these terms may be in order. Information is made
more or less prominent largely through prosody: what constitutes prosodic
190 LORRAINE E. KUMPF

"foregrounding" and "backgrounding" is vital to the structure of the story

and to the storyteller's purpose. On examination of oral narratives in terms
of their prosodic characteristics, it is clear that event, commonly seen as "fore­
ground", does not necessarily coincide with prosodic prominence, and non-
event similarly lacks coincidence with "background".
It is the view presented here that the storyteller's use of prosody plays a
central part in his or her structuring of the narrative. The purpose of this
paper is to look at some details of pitch structure, showing how they integrate
into the event structure of the narrative.
It is understood that complex prosodie interactions take place simulta­
neously in the stream of speech, and that it is somewhat deceptive to isolate
pitch from attendant features, particularly the important features of tempo
and loudness. However, pitch — encompassing tone (pitch direction) and
pitch range — is central to the complex of features called "intonation" (see
Crystal, 1969: ch. 5 and 203-4). Pitch defines the basic auditory unit of
analysis used here. This paper will be limited to the characteristics of pitch
changes in event-reporting versus non-event-reporting segments of narra­
tives, but within this limitation lies the most vital prosodie information.

2. Definitions
The prosodie terminology used here is borrowed from Crystal, 1969. The
system of contrasts is simplified for the present purpose, although not sub­
stantively changed. The 1969 volume must be consulted for detailed defini­
tion, motivation and discussion of the concepts and terms.
The primary parameters of prosody are pitch, loudness and duration —
features having the physical counterparts of fundamental frequency,
amplitude, and time, respectively. Pitch is the auditory attribute in terms of
which sounds may be ordered on a scale from high to low. The relative aspect
of pitch change, the degree and direction of the pitch level in relation to other
significant points in the pitch contour, is the important aspect of pitch. Tone
is the direction of pitch movement within the most prominent syllable of a
tone unit. The tone unit, the basic auditory unit of spontaneous speech, is
comprised of one obligatory peak of prominence known as the nucleus of the
unit, and three other optional parts. The nucleus of the unit usually falls on
the most important lexical item in the unit, and carries a glide, usually up or
down. The presence of the nucleus accounts for the intuition that a tone unit
is complete: if there is no nucleus, the auditory effect is that of being cut off,
of non-completion. The three optional units are: (1) the head of the unit,
 PITCH PHENOMENA IN THE STRUCTURING OF STORIES 191

referring to the stressed syllable of onset, (2) the prehead, referring to any
unstressed syllables preceeding the head, and (3) the tail, an unspecified
number of stressed or unstressed syllables following the nucleus in an unbro­
ken series until the end of the unit. A prominent syllable may move in the
same direction as the preceding nuclear tone; such a phenomenon is called
tonal subordination.
Tone unit structure is thus represented:
(prehead) (head) NUCLEUS (tail)
Tone designations describe the direction of pitch movement, and here
we will use a simplified system to discriminate the direction of the pitch glide:
falling (transcribed with over the vowel of the syllable), rising (//), rising-fal­
ling (???), falling-rising (???), and level tone (-).
Pitch-range describes the distance between pitch levels of adjacent sylla­
bles. Crystal claims that pitch-range analysis necessitates isolating the range
distinctions by hypothesizing a pitch constant for any speaker. This constant
is taken to be the first prominent syllable, the onset syllable, of a tone unit.
For any speaker, the onset syllable is articulated around a stable pitch level
for the majority of this tone units. A speaker will occasionally begin a tone
unit at a distinctly higher or lower point to achieve a particular efffect, such
as when expressing contrast (higher) or parenthetical material (lower). Here
such distinct points will be marked / for low and h for high onset.
Syllables after the onset syllable usually descend slightly before the
nucleus. If they are markedly lower, this is called a drop, and is indicated with
a (1) preceeding the syllable. Those which are heard at the same level are
marked ( —►). Those which are markedly higher (called a rise) are marked
(↑). Both drops and rises may be extreme, and such features are marked
(↓↓) and (↑↑), respectively.
The following examples illustrate this notation. The first is an unmarked
tone unit. (The relative pitch height is shown with dot notation, and three
types of syllable are designated by dot size: no stress ( • ), stress ( • ), and
primary stress ( ● ) .
(1) /i took a TÀXi//

(Note that nuclear movement of pitch is shown in the dot notation with a tail
on the dot in the direction of the glide.)
192 LORRAINE E. KUMPF

In (2) the tone unit is meant to describe an unusual circumstance. The

onset is high and followed by dropped unstressed syllable. There is an extra-
high raised tone before the nuclear syllable.
(3) /hit HURT// /hit HuRT//

In (3) the prenuclear syllable is raised high and the nuclear syllable is a
low, falling one. This is an emphatic prenuclear syllable.
A stretch of speech may be articulated at a lower or higher pitch level
than normal, the extent of the contrast being relative. This step up or down
is held over the stretch, contrasting with the preceeding and following pitch
level. The contrasting stretch of speech is marked with parentheses, and
noted below the transcription as high or low. For example:

The first tone unit is the end of a sequence describing the participant's
habitual activity. The high tone unit begins the temporal sequence of the
story with an unusual circumstance. The pitch is very high over the second
tone unit: it not only sets up the action sequence, ushering in a new scene, but
also describes an unusual circumstance vital to that which follows.

because i din/KNoW
 PITCH PHENOMENA IN THE STRUCTURING OF STORIES 193

[ba- new YORK//

The stretch in (5), marked low, is an aside.

Crystal uses the terms narrow and wide to refer to the range of each pitch
glide. He gives a detailed analysis (1969: ch. 5) of the types of possible glides.
We will employ the terms narrow and wide somewhat more loosely, though
in the same general sense. Width in pitch range is seen here, as in Crystal, in
relation to the immediate adjacent syllables. If a stretch of syllables employs
markers of height, rises and drops adjacently, we will call these stretches wide
as in

If the strength is notably flatter in pitch than normal, it will be designated nar­
row in parentheses:

The terms ascending and descending refer to a gradual ascent or descent

in pitch over a series of tone units. In this situation, the relevant tone units are
put in parentheses and labeled below the transcription:

Within the tone and pitch-range system there may be co-occurrences; a

common one is low and narrow, as in an aside,

2. Tone Units in Discourse

Tone units, as the fundamental auditory units of speech, exist at the base
of a hierarchy of units, one which Chafe (this volume) defines cognitively.
194 LORRAINE E. KUMPF

Chafe also identifies 'sentence' as a rhetorical unit of speech, defined by sen­

tence-falling pitch. Chafe offers no further prosodic specification of this unit;
however, the falling pitch as characteristic of a larger division of discourse,
beyond the tone unit and clause, is usualy clearly identifiable in these data.
Identifying this unit as a rhetorical one, a unit of the speaker's planned pre­
sentation, seem intuivively correct, and important for understanding how the
speaker structures his narrative. Although this analysis will be based on the
tone unit rather than the sentence, it is important to note that tone and pitch-
range contrasts exist at once on the prosodic rhetorical, and other levels. For
example, (8) above, illustrating descending pitch, constitutes one sentence. It
is not only the final fall, but also the pattern of descent, that frames the unit.

3. Research Methodology
Hypothesis
The tone units carrying the non-event portions of the narrative will
exhibit greater tone and pitch-range variation than those reporting events.

Data and Subjects

The data are narratives extracted from informal conversations between
Japanese speaking in English and a non-Japanese-speaking interlocutor/
researcher. The three Japanese (two male and one female) are contact lear­
ners of English: that is, they have not studied English in school, but have
picked it up through their experience of living in the US. All Subjects are mid­
dle-aged, all but one having lived in the US for over 20 years. The setting was
informal. The interlocutors were casually acquainted and the conversations
took place either in the home of the subject or that of the researcher.
The subjects were all willing storytellers, and each produced thousands
of narrative clauses in the two to six hours of tape per subject. However, only
a small portion of these data form the basis for this study, first because only
short, complete narratives were used, and second, because the time
demanded for transcription and analysis was considerable. For this study,
about 250 tone units were analyzed per speaker. These consisted of several
short, complete narratives, examples of which will be seen below.
A Note on the Use of Second Language Data
These data are part of a larger project concerning narratives in second
language. If the aim were to detail conventional patterns of English prosody
in narrative, clearly the data would be inappropriate. However, the aim is to
 PITCH PHENOMENA IN THE STRUCTURING OF STORIES 195

show how the story is built prosodically, and these stories contain all the
necessary elements for this task. For each speaker in this study, all of whom
have evolved somewhat individual learners' languages, prosodic patterning is
consistent. Furthermore, all of the basic prosodic structures found in these
data have been recognized for their universality in Bolinger (1978), though
this area will not be investigated here.
Two of the speakers, Hiro and Tomiko, have developed native-like
American English prosody. Although there are aspects of their pronuncia­
tion that are non-native-like, they have acquired, to two different degrees,
the tune of American English. This is of fundamental importance to their skill
as storytellers. The third speaker (Taro) is a case unique to my experience
with non-native speakers: he manges to communicate in monologue regard­
less of the fact that his prosodie patterning is non-native-like. He provides a
contrast to the others of particular interest in this study and in that of transfer
on the prosodie level.

Transcription
To review the notation used here:
Tone
falling
rising
falling-rising
rising-falling
even —
Pitch-range
onset: high h/hh
low ////
other syllables: drop ↓/↓↓
rise ↑/↑↑
height over syllables : {high )
(low )
range of pitch glide: (wide )
(narrow )
gradual descent or
ascent over syllables: (descending )
(ascending )
196 LORRAINE E. KUMPF

The prosodic transcription employs the following convenetions:

1. Tone units. Before the onset syllable there is a vertical line;
After the nuclear syllable (± tail) there are two vertical lines.
2. The nuclear syllables appear in capital letters.
3. Subordinate tone groups appear in square brackets.
4. Juncture. The normal slight pause between tone groups is not
marked. Others pauses are marked, but not in a rigourous man­
ner: three dots note pauses of one second or less; pauses of over a
second are measured and noted in brackets.
5. Hesitations are indicated with a dash.
6. Examples in the text and Hiro narrative below are given with a dot
display supplementing the transcription, in order to make the pitch
changes more graphic for the reader. The dot display locates the
pitch and designates the syllable as unstressed (•), stressed (•), or
nuclear (●), denoting primary stress. Nuclear syllables also show a
tail, noting the direction of the glide.
The transcriptions were reviewed and amended by Dr. Miciej Pakosz of
Lublin University, Poland, and UCLA. Dr. Polly Szatrowski of Cornell Uni­
versity also review some transcripts. Crystal (1969: 13-16) in his justification
of an auditory rather than an acoustic approach to data, notes the importance
of such verification. In the conclusion to this paper, some suggestions will be
made concerning the use of acoustic measures.
Tone unit identification necessitates finding unit boundaries. After the
peak of prominence, there will be a boundary consisting of a pitch change and
a junctural feature, usually a very slight pause. Identification of tone units
was not problematic in these data, given these criteria. One problem, how­
ever, was the status of a unit as independent or subordinate. Where both
criteria of pitch change and juncture were found, a new tone unit was always
indicated. It became clear, however, that adjacent units may vary in degree
of autonomy. Such factors as structural and semantic binding are often
reflected in tone unit structure, but the criteria of pitch and juncture remain
the defining ones. In the following example, the pitch contours are identical,
but the subordinate units differ in semantic and structural binding to the main
unit, (9) being more tightly-bound than (10). Both examples show no
juncture between main and subordinate, and subordinate nuclear tone glides
in the direction of the main one.
 PITCH PHENOMENA IN THE STRUCTURING OF STORIES 197

(9) duck was stuck to PÀTio [frozen STIFF]

(10) i can't SÈE nothin [SNÖW]

Contour:

Method
About 250 tone units of complete narratives were analyzed from each
speaker: the number is approximate because only complete narratives were
used. The narratives were divided into tone units and marked for tone and
pitch-range features.
The data were then examined for the relationship between pitch-range
and tone variations and status as event or non-event. An "event" was defined
as that part of the narrative material that answered the question "What hap-
ned?"; everything else is non-event. Non-events included descriptions of
scenes or conditions, evaluations, abstracts or introductory statements, and
codas or ending statements. Reported speech was used to a significant degree
only by Tomiko. In the analysis of reported speech, the reporting action (usu­
ally say or tell) was considered an event; usually that which was reported, a
non-event.

Sample Narratives
Before the findings are presented, three short narratives, one from each
speaker, will illustrate both the characteristics of the speakers and the
method. The first story is transcribed with the addition of a dot display, to
more graphically illustrate the pitch contours.

1. Hiro1
The story "Automatic Door" is typical of Hiro's output. Although there
is some phrasing and word stress that is decidedly non-nativelike, the sen­
tence contouring and placement of primary stress fits the patterns of Ameri­
can English.
Hiro created no incomprehensible utterances in the data, and in general,
his ability in English, in structure and word choice as well as in prosody,
reflects his experiental acquisition.
198 LORRAINE E. KUMPF
PITCH PHENOMENA IN THE STRUCTURING OF STORIES 199
200 LORRAINE E. KUMPF
PITCH PHENOMENA IN THE STRUCTURING OF STORIES 201
202 LORRAINE E. KUMPF

Notes on the "Automatic Door"

Hiro uses both inter- and intra-unit pitch changes to make material in the
narrative prominent:
1. Descending pitch over tone units is seen in lines 1,5, and 10. In
lines 1 and 5, the affected units are sentences which put forth a hypothetical
condition. In line 10 the units form a sentence which expresses an "expected
event": i.e., one that can be assumed to be in the mind of the hearer because
of the previously-expressed context. In this case, the hearer has in mind the
hypothetical hailing of a taxi, plus the real reported hailing which Hiro starts
to describe in line 8. The sentence "so I stand onna corner, anda raise a hand,
try to get a taxi... anda one empty taxi stop in front of me" is an expected
sequence in context.
In situations in which conditional are used to set up hypothetical contexts
and in situations in which "expected events" are related, Hiro uses this pat­
tern of descending pitch over tone units.
2. Wide range characterizes irrealis situations, e.g.:
I thought taxis supposed to be like that (line 7)
but I didn't know why (line 17)
but door didn't open (line 12)
or continuous situations, as in
I was waiting (line 12)
or evaluative comments, as in
so lotta things different (line 21)
and in one case, a complement:
he's opening a door with himself (line 19)
Wide range is also used when reporting in line 15:
finally taxi driver told me something
 PITCH PHENOMENA IN THE STRUCTURING OF STORIES 203

The final example is the only one in which wide range is used to report
an event in this story. The non-events in wide range are unexpected, surpris­
ing, contrary-to-fact, or emphasized situations.
Narrow range is used in lines 9-10 and 15-16, in both cases asides, and in
line 14, in describing a scene.
3. High onset occurs in 13 units, and 11 of those are non-event units.
Hiro seems to rely on high onset somewhat more than a native speaker
would. Low onset occurs three times, all in afterthoughts or asides; the one
event unit among them (in line 9) expands on an event that was reported
before.

2. Tomiko2
''Stuck Duck" is a story characteristic of Tomiko's narrating ability.
Tomiko, while retaining some non-native characteristics of grammar and dic­
tion, has thoroughly learned the tune of American English. Isolated from all
Japanese in rural and small-town settings in America, she has achieved a high
level of communicative competence, and her acquisition of prosody contri­
butes greatly. Unlike Hiro, she does not "sound Japanese"; many of the non­
standard aspects of her language are characteristic of non-standard native
dialect. Her style as a narrator is also different from Hiro: Tomiko unleashes
extended monologues of continually-flowing language, in which she suggests
her own topics, taking her monologue where she will, without interlocutor
prompting; Hiro exhibits a good deal of planning, as seen through pauses,
hesitation, very slow tempo, and other features.
"Stuck Duck"
204 LORRAINE E. KUMPF

wasjfrozen// /so my HÙSband told me she say// /get WARM WÀter///pour

onna FOOT///and so i DID//... /my HUSband give me more water// /and
(narrow (ascending
P???Rin//and/pretty soon duck MOVE// (laugh) /he was all RIGHT//[1 sec.]/

but/ he↓come down]↑↑every DÀY///messing up myt↑ tPÀTio// i didn't lika

(wide )
d'ÀLL///ÂFter that//
Notes on "Stuck Duck"
1. Tomiko's tone-unit-level production is not highly marked. What is
noticeable is the use of features across tone level: particularly wide/narrown
pitch-range and patterns of descending or high pitch.
The descending pattern in lines 2-4 extends over eight tone units and
frames a description of a usual scene. The wide pitch range units include a
passage containing a contrast (line 8) and an emphasized description of a
habit, (line 11). The high tone units in lines 5-6 mark the switch from the
introductory description of the duck to the situation of the day that the events
took place. They describe an uncharacteristic situation. The narrow units
(which are also fast in tempo) fill in a line of events that are not primary
events; they are all repetitions of material mentioned. They give a sense of
the passage of time in the process of freeing the duck.
The first and second tone units of line 2 show rising pitch in non-inter­
rogative contexts. Repeated rising pitches turn out to be quite common in
Tomiko's speech; here they serve the purpose of affirming understanding
with the interlocutor.
2. It is striking in listening to this passage that the tone units which
contain the event sequences are presented in normal declarative intonation
patterns (with the exception of the narrow sequence mentioned above),
whereas the non-event passages are characterized by such variation as a
sweeping descent in tone or description in wide pitch-range.

3. Taro3
The two previous speakers produced narratives that showed a sophisti­
cated knowledge of English prosody, a knowledge that helped them greatly
as storytellers. Taro also makes use of prosody, but it is used with a very
limited knowledge of English grammatical structure. It is interesting that, in
spite of these limitations, Taro is capable of producing good narratives. "I am
 PITCH PHENOMENA IN THE STRUCTURING OF STORIES 205

Electric" is an example of Taro's distinctive style. (Occasional Standard Eng­

lish translations occur in parenthese.)
"I am Electric"

jaPÀN you know///that TIMu//uh [3 sec.]/number ONE worku//(-I was work­

ing job number one) /na- number TWO you know//(-no- it was number two)
/ hard WORKu///s- pick UP you know//(-pick up the coal) /trailer BRINGu//

uh...ma- /look like country ROADu you know// /just one man TWO man//
uh[2 sec.] /can WORKu you know//look like mo- THAT way///coalman

know///because...ÓNE TON//bu-/trailer is ah very HÈAvy///SOME time

RÀILroadu and ah// ... / wheel between ROCKu or something you know//

HOLDu//
/then-back HURT// /PUJSHu you know// /then elECtric you know///
oh.. .COME to//... / i cannot go [1.5 sec.] OFF you know//(-I couldn't get off///
206 LORRAINE E. KUMPF

Notes on "I am Electric"

1. One is struck immediately with the non-native character of the
sentence intonation produced by Taro. There is one common tone unit pat­
tern in this story: a short statement intonation, having one stressed syllable,
followed by a low tail, you know. The unit generally does not contain more
than one stressed syllable, and the glide is almost always falling. There is a
pre-head of varying length, although the unit is consistently short — usually
three to six syllables. (Only 11.3% of the sampled units were over six sylla­
bles, and none were over nine.) Taro seems to enforce this brevity by drop­
ping off a unit with "you know" at any point in the structure. There is evi­
dence to show that this sentence intonation bears close resemblance to that of
Japanese, 4 though it is not our purpose to present such an argument here.
2. Taro works very hard at storytelling, using gesture, facial expres­
sion and eye contact, searching his limited lexical repertoire, and checking
with the interlocutor as well as employing prosodic devices. The features of
intonation and pitch-range in use by Taro can be seen as an extension of the
physical gestures which he also relies upon — more easily in him than the
other speakers because of the extent to which he exaggerates them. Taro uses
patterns of high and low, descent and ascent, wide and narrow in order to pro­
vide support for his monologue, and often a pair of opposing features will be
juxtaposed. Typical of this is the high stretch in line 13-14 above, followed
immediately by a low stretch of units. This strategy seems exaggerated but
serves Taro well.
3. It is characteristic of Taro's speech that a clause be broken down
into two or more tone units, as in the following:
/just one man TWO man// /can WORK you know//
and
/i did HĂND// /CÀble you know// /HoLDu//
In this way Taro preserves the intonation patterning described in note 1
above. (See footnote for the Japanese nature of this characteristic.) The pat­
tern is the same regardless of the status of units as event or non-event.
4. In this story, and typically for Taro, there is a good deal of descrip­
tion, scene-setting, and evaluation. Above, it is shown that such passages are
marked with the features descend, wide, and high and also exhibit rise and
high onsets. Because of the nature of Taro's delivery (highly marked prosod-
ically) and the nature of the content (highly "backgrounded") we can
generalize that the non-event portions of the narrative exhibit great tone and
pitch-range variation.
 PITCH PHENOMENA IN THE STRUCTURING OF STORIES 207

4. Results and Discussion

It should first be mentioned that the features under discussion are not
numerous in the data. In Hiro's case, the example narrative is more marked
than most of this other stories.

Table 1. Syllable features - Hiro

event non-event totals
↑/↑↑ 10 34 44
features ↓/↓↓ 5 12 17
hlhh 15 27 42
//// 3 7 10
totals 33 80 113

The features which show relative pitch height contrasts between sylla­
bles are best understood in relation to each other, and to the other features:
although their number in themselves do not have much meaning, they define
the pitch contours and thus the features of narrow/wide, descending/ascend­
ing, and high/low.
However, from the above table we can make a few pertinent generaliza­
tions. First, high onset and rise are very numerous. This fact is clear but not
so interesting: it is expected that prosodically marked segments serve some
kind of function of focus on importance, and that function is usually carried
out by raising pitch. Bolinger (1978: 515-16) calls high pitch "the normal sign
of importance " for any language. He goes on to predict that the opposite,
lowering or downtoning, would appear less frequently or not, in some lan­
guages, at all. It may be that the lowering effects counted above need more
explanation that the rising ones; what is obvious in some cases is that the low­
ering comes before a particular rise, in order to exagerate the contrast in
highly marked contexts.
The prevelence of high onset may be explained by the fact that in
Japanese, a stressed onset syllable must be higher than the nuclear syllable
(Szatrowski, personal communication). However, it is not the case that high
onset was applied over all (in fact, only in 42 of 244 tone units); this charac-
terisitc likewise did not render the prosody un-English-like. The role of trans­
fer from Japanese in this case will need to be further investigated.
Finally, the features of Table 1 were used in non-event tone units about
208 LORRAINE E. KUMPF

two and a half times more often than in event units.

Table 2. Polysyllabic Features - Hiro.

event non-event totals

wide 4 12 16
narrow 2 4 6
descending 5 6 10
ascending 0 2 2
high 0 2 2
low 0 4 4
totals 11 30 41

These features, which span a tone unit or series of tone units, are used
three times more often in non-event units than in event units. Notice that
Hiro favors the features wide and descending. He sometimes uses the wide
features in a way that sounds exaggerated to a native speaker, accompanying
it with a slow tempo and precise articulation — no doubt overcompensating
for what he perceives as a lack of effectiveness of his ''English accent".
The importance of these features for the text lies in their relation to the
rhetorical aspects of the stories. Let us see how the features are used in story
structuring, using "The Automatic Door" as an example.
The introductory, orientational material is presented in a descending
pattern at the opening:
When you go to Japan, and when you try to get a taxi, if they see a hand,
taxi will stop...if taxi's empty.
The precedin constitutes a sentence with an afterthought. There follows a few
explanatory units, and then another descent, which signifies the conclusion of
the orientation:
So whenever they got a customer, press the button, door will open
automatically.
This last, also a sentence, ends with a low fall, completing a paragraph.
The next four units are marked with syllable rises, and are accompanied
by tempo slowing, contributing to the weighty effect of Hiro's explanation.
The fourth of these units is a sentence which starts the story action:
 PITCH PHENOMENA IN THE STRUCTURING OF STORIES 209

I took a taxi from Pennsylvania station over New York.

This unit is delivered with wide range, as well as slow tempo: attention is
placed on the beginning of action.
After an aside explanation is made, Hiro talks about what happened
when he hailed a cab in New York. To do this, a descending sequence of
units, constituting "expected events" of standing on the corner and raising a
hand and so on, is produced. These "events" are actually pre-events, setting
the scene for more important ones. The sequence is a sentence; it ends a
scene.
The next units are highly marked. Wide range and high onsets and rises
deliver the facts the door didn't open, that he waited, and that he had had
expectations counter to the events. This is the crucial situation for the narra­
tive, and its success depends largely on these markings. Note that the gram­
mar of the units are typical "backgrounding", however —- telling what did not
happened, describing the scene, and giving explanation.
Two punctual events, also marked wide, and certainly of import to the
storv. come later:

Yet the reaction.to these events is also highly marked:

The subsequent events are related in a straightforward manner: Hiro repeats

his mistake, observes other hailing cabs, and realizes his mistake. The only
highly marked unit is the complement of the verb 'saw' in
then I saw he's/ opening the door [with↑↑himself]//
This unit's high onset and rise conveys Hiro's surprise at the reported fact.
The low fall at the end of the next unit ends the story proper,
An evaluate comment (so iota things different) and a coda (and that was
the first time I had a problem at over here) both constitute sentences.
In this story we see that it is possible to define sentence-length pitch pat­
terns which have identifiable functions in the story: for example, a descend­
ing pitch pattern presents conditional material and "expected events", low
pitch carries asides and some explanatory material, and wide range conveys
importance of explanation or event, and surprising or unexpected conditions.
The patterns frame rhetorical units of the stories. For our purpose here it is
essential to note that the marked tone units create rhetorical units likely to be
non-events.
210 LORRAINE E. KUMPF

Table 3 shows the syllable-level pitch changes for Tomiko.

Table 3. Syllable Features - Tomiko

event non-event totals
\/\\ 7 45 52
ures ↓/↓↓ 6 9
h/hh 5 25 3
l/ll 1 1 3
totals 16 78 94

Again, the contrasts shown for adjacent syllables are less important in
themselves than they are in defining contours. However, we once more
notice a reliance on the high onset and rise. The proportion of these features
is greater in Tomiko's data than in Hiro's, and that is reflective of Tomiko's
speaking style, which is highly animated, contains a lot of contrastive struc­
tures, and relies more heavily on pitch height than on other features.
In Table 3 we see that about 80% of the marking for syllable features
occur in the non-event units of Tomiko's narratives.

Table 4. Polysyllabic Features - Tomiko

event non-event totals
wide 3 9 12
narrow 5 0 5
features descending 0 3 3
ascending 4 3 7
high 0 10 10
low 1 1 2
totals 13 26 39

Table 4 shows that tone unit features which run across tone unit boundaries
are used twice as often in non-event sequences as in event sequences. We see
that Tomiko has a preference for the wide feature, as Hiro does, but none for
the use of descending tones. Tomiko also favors the feature high, and uses
 PITCH PHENOMENA IN THE STRUCTURING OF STORIES 211

ascending sequences more readily than Hiro. It also seems that certain fea­
tures (narrow and ascending) are more easily used for event reporting than
others. The preference for marking non-events is clear for the other
categories.
The high feature is often used for reported speech, the reported content
being delivered at a substained high pitch, providing a contrast with the
reporting verb and subject, which are usually unmarked or narrow in deliv­
ery.
Tomiko's stories are full of emphatic descriptions, contrastive struc­
tures, reported speech and other structures which are highly marked. Intona­
tion, pause, duration, stress and other prosidic subtleties are native-like in
her data. Although there are other aspects of her speech which are less than
native-like, Tomiko has been very successful at learning the tunes of English.
One pattern which is important in her output is a rising contour which
lasts for one tone unit and is repeated over a sequence of units. This contour
is actually quite flat and narrow, but it drifts upward. This pattern usually
occurs in uccessive units which frame a scene or a series of events. Some­
times the pattern is repeated over units which include both events and non-
events, as in the following report of a hurricane:
(11) WIND was blowing
house trailer have a AWn ing on it
MÍSter rainer-
MÍSsus rainer
was TRYin to take down
so I HÉLP it
CLIMB up toppa trailer
We have a TÈRrible time
These units go together as a paragraph. The same tone pattern is repeated
one unit at a time until a falling tone denotes the end of the paragraph. There
are other examples of this pattern where the characteristic unit is intersperced
for a shorter length, particularly where scenes or sequences of activities are
being described:
(12) I could not open DOOR
I try KÍCKin
an PÚSHin
an PÚLLin
Èverything I done
212 LORRAINE E. KUMPF

There are 41 occurrences of this pattern in the Tomiko data. Occasion­

ally, the pattern sounds non-native-like, as in (11) above. The pattern is also
used for listing, as in (12), where it sounds perfectly natural in English.
Bolinger (1978) points out such rises, signifying "not finished", are universal.
Often the same pattern seems to be used to ask confirmation of the inter­
locutor, to make sure he is attending. Ikasawa (personal communication) and
other Japanese have indicated that such confirmation is constant and
expected between Japanese interlocutors. Perhaps the pattern's frequency
can be explained in part by Tomiko's expectation of confirmation.
Tomiko's use of inter-tone unit patterns can be related to the rhetorical
structure of her stories, but it is clear that this is done in a manner very differ­
ent from Hiro's. The most striking difference is that Hiro's stories seem
highly planned: evidence of planning is particularly seen in pauses which are
sometimes quite long, drawn out "conjunctions" such as "so" at the begin­
ning of units, self-corrections, and so on. Tomiko's stories do not exhibit
planning. Pauses are rare. Stories usually begin in a rather chaotic way, with
Tomiko continuing to talk as she orients herself to what she wants to say.
(This tendency is illustrated in "Stuck Duck".) Despite this lack of planning,
however, Tomiko's stories show the same close relationship between rhetor­
ical unit (the sentence) and pitch contour. This is the case even though the
particular contours which are characteristic for one storyteller appear to
depend on individual style.

Table 5. Syllable Features - Taro

event non-event totals

Ml 2 19 21
features ↓/↓↓ 3 11 14
h/hh 3 25 28
l/ll 5 6 11
totals 13 61 74

Just over 80% of the total marking in Table 5 is found in non-event units,
and again, high onsets and rises constitute the important markings. In Taro's
case, the pitch contrasts may be exaggerated in order to make clear to the
interlocutor exactly what is important: Taro was well aware of his limited
knowledge of English structure and lexicon, and compensated through pro-
 PITCH PHENOMENA IN THE STRUCTURING OF STORIES 213

sodic means. Much of his success in communicating must be attributed to this

and other paralinguistic strategies.

Table 6. Polysyllabic Features - Taro

event non-event totals
wide 1 13 14
narrow 2 6 8
descending 2 2 4
ascending 1 4 5
high 0 5 5
low 1 5 6
totals 7 35 42

In Table 6 we see that Taro favors the use of wide range, but that he uses
all of the features to some degree. Here, five times the numer of non-event
sequences are marked, over that of event sequences.
Taro also makes use of the flat, slightly rising pattern which is repeated
over tone units. This feature seems to be used to list conditions when describ­
ing a situation or to give details when explicating. Taro does not use the pat­
tern for confirmation with the interlocutor, as Tomiko did. The following
example described the conditions in Japan just after World War II, when coal
mining became essential to rebuilding:
(13) look like WÀVES
now uh inFLÀTion
now reCÉSsion
now inFLAtion
uh house Finish (= houses were ruined)
next wave's COMing
means ah COAL you know (= this means coal is needed)
The pattern occurred 22 times in Taro's data. The prominence of this pattern
for both Tomiko and Taro leads to questions about the use of the pattern in
Japanese, the possibility of prosodic transfer, and the reason why Hiro seems
not to have employed the pattern.
It is quite clear that for these speakers, the non-event portions of the nar­
ratives exhibit greater variation of tone and pitch-range than the portions
214 LORRAINE E. KUMPF

which report events. This is the case even though the marking patterns that
are typical for one storyteller may not be characteristic of another — they
vary with individual storytelling style. However, questions may be raised as
to the relevance of this finding.
First, it is true that in the stories, the total number of non-event units
greatly exceeds that of event units. For Hiro the ratio of events to non-event
units is about 1 to 5 ; for Tomiko, 1 to 4 ; and for Taro, 1 to 11. It is not surpris­
ing that description, evaluation, motivation, and other "background" fea­
tures should outweight the event units in number. However, it does not fol­
low that one or the other type of unit should be marked. If we assume that
high marking signifies "importance" or "focus", and low marking signifies the
opposite, no necessary conclusion can be drawn as to which type of unit
would be marked. And in fact, both events and non-events are made promi­
nent or downtoned, according to the storyteller's purpose. What seems to be
true in these stories is that the more prosodically prominent material is the
non-event material, for each of the speakers.
One tentative conclusion which needs further investigation is the sugges­
tion that events are reported in a normal, declarative intonation pattern, with
no marked features, unless the event is a central event, in which it will be
made prominent, or an aside or incidental event, in which case it will be
downtoned. There are relatively few central events in a story, but many ways
to comment on events, and many motivations for marking. We have seen that
high onsets and rises are characteristic of expressions of surprise, evaluations
and so on. High onsets also serve to start a new scene. Patterns of ascent and
descent are often used with material which gets us from one event to another,
particularly through creating conditions. High sentences show surprise,
emphasis, importance; the content of reported speech is often delivered in
high pitch. Low sequence deliver asides or non-essential explanation. With
usually lends weight, surprise, counter-to-expectations, or contrast. Narrow
sequences are often combined with up-tempo to hurry us through a condition
or event sequence and on to something more important. These features com­
bine with tempo, rhythm, loundness and other prosodi characteristics to
create the texture of the stories.

5. Conclusion
In this very preliminary study, it is shown that prosodically marked struc­
tures are characteristic of the "background" portions of narratives. It has also
been suggested that the prosodic structure of the narrative is closely related
 PITCH P H E N O M E N A IN T H E S T R U C T U R I N G O F STORIES 215

to the rhetorical structure — that in fact, prosodic patterns frame the rhetor­
ical units. This study suggests other, and begs comment on some methodolog­
ical points.
First, the study as conducted was impractical and unwieldy from the
standpoint of data analysis. The transcription process was difficult and far too
lengthy; the transcripts also had to be checked by at least one other trans­
criber. This process could be greatly aided by the use of a pitch recorder,
which would provide a more objective measure of the stream of speech. The
recorder would make it easier to observe a stretch of speech, to determine
norms, and to compare different speakers.
What the recorder cannot solve is the problem of deciding on a point at
which high pitch become high, or a wide interval becomes wide. These judge­
ments remain relative to the context of the speech and to the speaker's norm.
The recorder would ease the process of reaching the judgements, however.
This study shows some ways in which rhetorical and prosodic structures
are interwoven. That complex pitch phenomena and other prosodie elements
are integral to story structure seems too obvious to mention, yet such ele­
ments have rarely been studied as indicators, either cognitive or linguistic, of
the structure of discourse, perhaps because of the difficulties in prosodie mea­
surement. Spontaneous personal narrative is a good medium for the investi­
gation of prosody in discourse because narrative is highly structured in ways
that we already know about, and we can see how prosody defines and sup­
ports the discourse units of narratives. Prosody, infinitely subtle and
individuated, is at the same time collectively used in recognizable ways to
create discourse.

NOTES
1) Hiro is a 38-year-old worker in an electronics firm who was sent to the US by his company.
He graduated from a vocational high school in Japan. He claims no formal English education
except in junior high, a learning experience which he termed 'lost'. His speaker status was reported
as 'absolute beginner' upon his arrival in the US eight years before taping.
2) Tomiko came to the US in 1952 as a war bride, and was in her mid-50s when taped. She spoke
no English on arrival, as her husband was fluent in Japanese, but her gregarious nature and non-
contact with the Japanese in the US became the basis for her acquisition. She is an excellent
storyteller, recounting at length and with relish stories based on everyday experiences.
3) Taro was 53 at the time of taping. He is gardener who had been in the US for 27 years, mostly
in California, when taped. He has very limited use of standard English, but manages to commun-
ciate quite effectively. He has developed a number of strategies, including formulaic speech and
216 L O R R A I N E E. K U M P F

paralinguistic/kinetic communication. His social network includes very little contact with native
English speakers. Taro is very gregarious and every verbal despite his limitations with English.
4) At least seven native Japanese, some of. whom were linguists, unanimously attested to the
Japaneseness of Taro's prosody. Some commented that, when listening to Taro and ignoring the
lexical content, they had the impression of hearing Japanese. Shoichi Iwasaki of UCLA noted the
influence of Japanes syllable structure on Taro's brief tone units; he noted also that Japanese had
intraclausal pauses. From this informaton I gather that the relationship between tone unit and
clause in Japanese may be quite different from that in English.
Polly Szatrowski of Cornell University was helpful in showing how the clause structure of
Japanese allows for intonational structures of finality within even tightly-bound clauses. That is,
partial clauses may have final intonation. Taro exhibits this trait often. Szatrowski also pointed out
the Taro's staccato delivery is characteristic of Japanese, as are the afterthoughts which pepper
Taro's speech. Further, a stressed onset syllable in Japanese must be higher than the nuclear sylla­
ble, a fact which may explain Taro's often-high onset. It would be interesting to consider the
characteristics of Taro's speech which are Japanese and those which tend to be universal and thus
presumably aid Taro's ability to communicate. Are there any uniquely American aspects to his
prosody?
Recognizing Taro's heavy reliance on Japanese, one becomes more aware of the acquisition
achievements of Hiro, who has acquired the stress-timing of English very adequately, yet still is
recognized by Japanese and native American English speakers as "sounding Japanese", and of
Tomiko, who, according to the same groups, does not sound Japanese, but has an unplacable
accent which some thought was native American.

REFERENCES

Bolinger, Dwight. 1978. "Intonation across languages'', in Greenberg, J.H.,

ed. Universals of Human Language: Volume 2 - Phonology. Stanford,
CA: Stanford University Press.
Chafe, Wallace, this volume. "Cognitive constraints on information flow".
. 1980. "The deployment of consciousness in the production of a narra­
tive", in Chafe, W., ed. The Pear Stories: Cognitive, Cultural, and Linguis­
tic Aspects of Narrative Production. Norwood, N.J.: ABLEX.
Crystal, David. 1969. Prosodic Systems and Intonation in English. London:
Cambridge University Press.
Givón, T. 1982. "Tense-aspect-modality: The creole prototype and beyond",
in Hopper, P. Tense-Aspect: Between Semantics and Pragmatics. Amster­
dam: John Benjamins.
Hopper, Paul. 1979. "Aspect and foregrounding in discourse", in Givón T.,
ed. Syntax and Semantics 12: Discourse and Syntax. New York: Academic
Press.
Hopper, Paul and Sandra Thompson. "Transitivity in grammar and dis­
course", Language, 56, 251-299.
 ON THE STATUS OF SVO SENTENCES IN FRENCH DISCOURSE

KNUD LAMBRECHT
University of California, Berkeley

1. Introduction1
According to an old linguistic tradition the basic object of grammatical
and logical analysis is a sentence in which all argument positions are filled
with fully referential lexical NPs. This tradition goes back to Medieval and
Greek grammatical theory. Indeed from Plato on, the canonical sentence
type of grammatical theory has been the type Socrates currit 'Socrates is run­
ning', the so-called oratio perfecta, which expresses a "complete thought" (cf.
Stefanini 1981, and references therein). One reason why the oratio perfecta
came to be the canonical model is that the meaning and the truth conditions
of Socrates currit is no doubt easier to state, on the basis of this sentence
alone, than the meaing and truth value of Currit 'he/she/it is running1, which
has no lexical subject. I will refer to such sentences in which all argument
positions of the verb are filled with lexical NPs as SV(O) sentences.
It is typically such SV(O) sentences that are used in linguistic argumenta­
tion. An archetypal SVO example, illustrating the basic form of the English
sentence, is Sapir's model The farmer kills the duckling, which the author calls
"a typical English sentence" (Sapir 1921:ch.V). It is interesting that a great
linguist should imagine as a basic model a sentence which so blatantly con­
tradicts our pragmatic intuitions that it is often misquoted with the verb in the
past tense (The farmer killed the duckling), in an unconscious attempt on the
part of the quoter to bring grammar and the real world a little closer together.
The difference between Sapir's model sentence and what a person might have
said in a real life context to convey the meaning contained in this sentence is
related, I believe, to the distinction Lyons draws between text-sentences and
system -sentences :
System-sentences never occur as the products of ordinary language
behavior. Representations of system-sentences may of course be used in
218 KNUD LAMBRECHT

metalinguistic discussions of the structure and functions of language; and it

is such representations that are customarily cited in grammatical descrip­
tions of particular languages. (Lyons 1977:30f )
As another example, this time from an actual English text, consider the
following short fragment:
When naturalist John Muir stood on the summit and watched the sunrise in
1896, he exclaimed: 'The view from San Jacinto is the most sublime specta­
cle to be found anywhere on this earth!" (John W. Robinson, San Bernar­
dino Mountain Trails, Wilderness Press, Berkeley, 1975:183)
Two things are remarkable here. First, the fact that the writer quotes John
Muir as saying something he would hardly have said in the reported situation.
(He is much more likely to have said something like 'This is the most sublime
spectacle I've ever seen", or some similar sentence with no lexical subject, if
indeed he said anything at all at this particular moment). The second remark­
able fact is that the writer can get away with this fiction, i.e. that we can read
such sentences in a text without finding anything anomalous about them,
even though we would not normally say or hear them as products of "ordinary
language behavior". It is not my purpose here to discuss our ability to engage
in non-ordinary language behavior by creating and accepting system-sen­
tences, although this would be a very interesting endeavour in and of itself.
Rather the purpose of this paper is to investigate the occurrence of SV(O)
sentences in actual speech, in particular in spontaneous oral discourse.
The basis for my investigation will be data from modern French, a lan­
guage that is typologically similar to English in many respects, and in which
the grammatical tradition that has created the SVO model is at least as strong
as in English. A look at any stretch of spontaneous spoken discourse in
French (as in English, or — as I am tempted to predict — in any other lan­
guage2) is sufficient to show that the SVO clause pattern is not the predomin­
ant pattern at the level of surface structure. The empirical evidence concerning
the occurrence of lexical subjects in spoken French is indeed stunning. One
long corpus of conversations between members of a working class family in
Paris (François 1974, vol. 2) contains a total of 1,550 nouns, only 46 of which
are lexical subjects.3 Similar statistical results are reported in Jeanjean
(1981).
This is indeed a noteworthy phenomenon: since subjects are the only
arguments which are necessarily present in all propositions one might expect
the number of nouns in this argument position to be higher than in any other
position. However, nouns occur much more frequently in object position, i. e.
 SVO SENTENCES IN FRENCH DISCOURSE 219

after the verb (roughly 300 occurrencs in François' corpus). This is in fact not
surprising, given what we know about the normal direction of information
flow in the sentence, post-verbal object position being the position in which
"new information" tends to appear in VO languages. What may be more sur­
prising is that the vast majority of nouns appear neither in object nor in
subject position but in prepositional and adverbial phrases, in extra-clausal
topic phrases, and in phrases that have no syntactic connection with the prop­
osition at all. These nouns make up roughly 1,200 occurrences out of the total
of 1,550. In this paper I will be mainly concerned with a subset of these nouns,
those occurring in extra-clausal topic phrases.
The prevalent syntactic unit in spoken French discourse is not the sen­
tence that expands into NP and VP. Rather the corpuses reveal as the
dominating unit a syntactic structure which I will call the preferred clause of
spoken Frech, and which contains no subject NP. 4 In section 2 of this paper
I will describe this preferred clause and the ways in which French speakers
can substitute it for the SV(O) pattern by using various grammatical construc­
tions. These constructions have the effect of ordering the basic constituents of
the clause in such a way that the preferred clause structure is preserved under
the various pragmatic circumstances of the discourse which require a depar­
ture from the normal syntax. In section 3,1 will then investigate the status of
those rarely occurring but nevertheless fully functional sentences that do con­
tain lexical subject NPs. My main argument concerning the function of such
sentences will be that there is a correlation between the occurrence of lexical
subjects and the discourse status of the clauses in which they occur. Lexical
subjects strongly tend to appear in clauses that represent pragmatically
backgrounded portions in the discourse. This pragmatic feature is then shown
to correlate with the syntactic fact that lexical subjects occur almost exclu­
sively in intransitive clauses, i.e. in clauses of the type SV. As a corollary, the
canonical SVO sentence type, containing both a lexical subject and a lexical
object, will appear to have no pragmatic function in spoken French. 5

2. The preferred clause of spoken French6

The low number of lexical subject NPs in spoken French contrasts with
the extremely high number of pronominal (clitic) subjects or other clause-ini­
tial clitics. Thus the above quoted number of 46 lexical subjects in the Fran­
çois corpus should be seen in contrast with the total number of 1,440 clitic
subjects (this number includes l p and 2p clitics but excludes deleted subject
clitics of the type mentioned under (c) below). The preferred clause unit of
220 KNUD LAMBRECHT

spoken French contains one or several clause-initial clitic pronouns which are
bound to the verb 7 and an optional lexical constituent after the verb. Thus,
given the status of clitic pronouns as verbal prefixes, the preferred clause is a
verb-initial structure. This preferred structure can be schematically rep­
resented as [clitic+Verb (X)]. The constituent filling the optional X position
is pragmatically the focus of the clause. Immediately preceding this sequence
is a position for initial focus constituents, including (optionally fronted) ques­
tion phrases and marked focus NPs (cf. section 2.1, ex (1)), For lack of abet­
ter term I will call this position COMP. The preferred clause structure includ­
ing this initial position is thus [(COMP) ciitic + Verb (X)]. I take this syntactic
structure to be the basic information unit of spoken French. It is the domain
of propositional information proper. Constituents that precede or follow it
are lexical topic phrases, adverbials, or other pragmatically determined ele­
ments that do not directly take part in the syntax and semantics of the clause.
The initial clitic in the preferred clause can be:
(a) a subject clitic of the series je tu, if elle, on, vous, ils, elles (on
replaces standard French clitic nous)',
(b) the demonstrative or impersonal 3p clitic ça or c' (standard French
ce does not occur in the spoken language);
(c) the dummy subject marker if used with so-called impersonal verbs
like il faut 'it is necessary', il paraît 'it seems' etc. and in subject-
verb inversion constructions (cf. section 2.2); this il is frequently
omitted; 8
(d) the adverbial clitic y 'there'.
The subject clitics in (a) - (c) can be followed by one or several non-subject
clitics. In this paper, clitics will often be written in the phonetically shortened
forms in which they are used in spoken French (e.g. il→i', tu→t', vous→vs'
or s' etc.; for details on the shape of spoken French clitics cf. Lambrecht
1981:12ff).
The post-verbal focus constituent can be any type of lexical phrase,
including a subject NP of the presentational type (cf. 2.2 below). It can also
be a pronominal NP containing a pronoun of the independent (non-clitic)
series moi, toi, lui etc. An example of a pronominal focus is given in (5)
below. In this paper I will only consider NP foci, since I am mainly interested
in the way in which discourse referents are syntactically coded in the sen­
tence.
The term focus, which stands for the more explicit relational expression
 SVO SENTENCES IN FRENCH DISCOURSE 221

focus of the assertion or focus of the the (new) information, refers to that ele­
ment in the clause which is singled out as the most salient part of the new
information, or, a little more precisely, to that element by which the assertion
expressed by the proposition differs most saliently from the pragmatic pre­
supposition, which I define as the assumed state of information of the addres­
see at the time of the utterance. Note that the focus expresses ''new informa­
tion" only insofar as it indicates an unpredictable (or non-topical, cf. below)
relationship between a referent and a proposition. The referent of the focus
constituent (if the focus is a referential item) does not have to be new in the
discourse. It is important to draw a distinction between the focus of the new
information, which is found in every clause, and a new discourse referent,
which is not an obligatory part of every clause. A new discourse referent is a
referent that is neither given nor (pragmatically) recoverable in the context of
utterance. A given referent is a referent that is assumed by the speaker to be
present in the consciousness of the addressee at the time of utterance (Chafe
1976). A recoverable referent is a referent that is not yet given, but whose
identity the addressee is assumed to be able to recover from the discourse
context. A referent can be recoverable because it was mentioned before,
because it can be inferred from other previously mentioned items, or because
it is saliently present in or inferrable from the extra-linguistic context of the
utterance. 9 In spoken French, every new discourse referent that the speaker
wishes to mark as such must appear in focus position. But, as noted above,
every focus constituent does not have to contain a new discourse referent (for
an example see (5) below). Whereas the "newness" of the focus is due to the
unpredictable relationship of a referent with a proposition, the newness of a
discourse referent consists in the inherent pragmatic status of non-recovera-
bility the referent has in the discourse. Thus, while 'focus' refers to a pragma­
tic relation expressed in a proposition, 'given', 'recoverable' and 'new' refer
to temporary psychological statuses of referents in the mind of the speaker/
hearer, independently of the role these referents play as elements in a propo­
sition.
The relational concept focus as I define it here contrasts with the rela­
tional concept topic. The relation topic of is the relation of aboutness holding
between a referent and a proposition with respect to a discourse (cf. Reinhart
1982). The topic expression in a sentence indicates the referent about which
the speaker wishes to add information or with respect to which the informa­
tion expressed in the proposition is presented as relevant. The topic referent
is a referent that has to be "taken for granted" at the time of utterance. In spo-
222 KNUD LAMBRECHT

ken French, for a proposition to be 'about' a referent this referent must nor­
mally have the inherent pragmatic status given (cf. Lambrecht 1984a).
To prevent some of the misunderstandings that tend to arise when there
is talk about topic let me try to clarify the concept by narrowing it down for
the sake of this discussion. Since topic involves a pragmatic relation between
a referent and a proposition, the word 'topic' is often used to designate the
discourse referent itself about which new information is being added in the
proposition. This referent will be called here the topic referent or sometimes
simply the referent. This common use of the term 'topic' as referring to a non-
linguistic element (a referent) should be sharply distinguished from the use of
the term to refer to a linguistic expression designating the topic referent in a
sentence. This linguistic expression will be called here the topic expression or
the topic. It is this linguistic concept of topic that I am mainly concerned with
in this study.
There are basically two ways in which a topic expression can designate a
referent in a sentence. The topic expression in a sentence can name the topic
referent, by means of lexical phrase and in the syntactic form of an NP. Such
a lexical topic constituent will be designated with the capital letter T (or AT,
which stands for "Antitopic", cf. 2.2.4 below). But the topic expression can
also be, and usually is, a (complex) morpheme that refers deictically to the
topic referent, or anaphorically to some lexical representation of the topic
referent, in pronominal i.e. non-lexical form. This topic morpheme will be
designated with a small t. The t morpheme can be a clitic pronoun, a zero or
the relative pronoun qui, que. Whereas, in spoken French, T is normally an
extra-clausal constituent with a merely lexical naming function and no case
role in the proposition, t indicates the semantic role the topic plays inside the
proposition and the syntactic relation it bears to the clause (cf. section 2.1.3
below, ex. (17) and discussion). This explains, among other things, why t, but
not T, is morphologically case-marked (cf. Lambrecht 1981:34ff.). Thus,
since the T constituent is not part of the proposition proper, it is strictly
speaking not with T but with t that the aboutness relation is expressed within
the proposition. Under specific circumstances, to be analysed in section 3, T
and t can merge and be expressed by a lexical subject NP. In my analysis, a
subject NP with a topical referent is a pragmatic agglomerate, which com­
bines the referential function of T with the relational function of t.
With these distinctions in mind, the structure of the preferred clause,
together with its lexical topic satellites, can now be schematically rep­
resented, in pragmatic terms, as (T) [(F) t V (F)] (AT), where the first F indi-
 SVO SENTENCES IN FRENCH DISCOURSE 223

cates the position for clause-initial focus constituents in COMP (especially

QU-phrases), where the second F indicates the unmarked post-verbal focus
position, and where T and AT are the extra-clausal topic constituents, which
stand in a coreference relation with the intra-clausal t (cf. 2.1.3 below). 10 The
overwhelming frequency with which the preferred clause structure recurs in
natural discourse is evidence, I believe, that the syntactic sequence clitic-
verb-NP, with its pragmatic mapping topic-verb-focus, must have some cog­
nitive or processing advantage over other possible sequences (such as SVO
structures). At the end of this paper I will suggest an interpretation of the pre­
ferred clause as a language-specific manifestation of a universal pragmatic
principle governing the information flow in the sentence.

2.1 Grammatical constructions and the preferred clause

In order to preserve this cognitively preferred unit of syntax as invariably
as possible, the grammar of French offers its speakers a number of grammat­
ical constructions, whose function, I claim, is to order lexical phrases in accor­
dance with the pragmatic requirements of discourse, and in particular to
allow NPs that would become lexical subjects to appear elsewhere than in
preverbal subject position. The term 'grammatical construction' is used here
in the sense of Fillmore, Kay, O'Connor (1983), Fillmore (1985), Lakoff
(1984), and Lambrecht (1984b). Roughly, a construction is a productively
used non-derived structure whose properties cannot be predicted from the
general syntactic and semantic properties of the grammar and whose form is
often directly linked to its pragmatic function. The constructions of spoken
French to be discussed here all have in common that they respect, in one way
or another, the preferred clause structure [clitic+Verb (X)].
In the remainder of this section I will briefly characterize these different
constructions and define their various pragmatic functions in discourse. The
pragmatic functions expressed are the identificational ox focus marking func­
tion (serving to identify a referent as a focus argument in a proposition, sec­
tion 2.1.1), the presentational function (serving to introduce a new referent
into the discourse, section 2.1.2) and the topic marking function (serving to
mark as a topic a referent that the speaker assumes to be pragmatically recov­
erable in the discourse, section 2.1.3).

2.1.1. The identificational or focus marking construction

Unlike English, spoken French has a quasi-absolute constraint against
the mapping of the pragmatic relation of focus onto the syntactic relation of
224 KNUD LAMBRECHT

subject. With the exception of postverbal subjects (cf. the discussion of inver­
sion in 2.1.2), Ï have found no instances of subject NPs with focus function in
the corpuses I have analysed. Initial focus NPs do occur in pragmatically
marked, strongly contrastive contexts, associated with clause-final, i.e. fal­
ling, intonation on the initial constituent and lack of stress on all post-focal
elements in the clause. However these initial focus constituents hardly ever
seem to be subjects. A focus-initial construction is illustrated in example (1).
Here and in the following examples, the focus is in capital letters:
(1) Tas pas faim? - Faim non. SOIF J'ai. (Stempel 1981)
'Aren't you hungry? - Hungry, no, Thirsty I am.'
I will not deal in this paper with such constructions involving initial focus con­
stituents. (For some discussion, cf. Stempel 1981; cf. also Prince 1981b and
Silva-Corvalán 1983 for discussions of focus-initial constructions in English
and Spanish.)
The constraint agains foci in subject position can be illustrated with the
following made-up dialogue:
(2) A: Où est mon rasoir? 'where is my razor?'
B: a. ??PIERREl'a 'Peter has it'
b. C'est PIERRE qui Va 'It is Peter who has if
The answer to A's question is unacceptable with the focus Pierre in initial
position as in (2a). Instead, the c'est-cleft construction has to used, as in (2b).
The c'est-cltit is formally related to the English //-cleft, but unlike in English,
where the use of this construction seems to be quite restricted in the spoken
language, the use of the c'est-cleft is widespread in spoken French. Its range
of use is furthermore much broader than the range of its English equivalent,
as some of the following examples will show (cf. also Lambrecht 1984a for
further examples and discussion). The tendency to keep subject foci out of
initial position is so strong in spoken French that the c'est-eleft is frequently
used to place even question word subjects after the copula, i.e. in the
unmarked focus position, as shown in example (3b):11
(3) a. QUI a fait ça? 'who did that?'
b. C'est QUI qu'a fait ça? 'it is who that did that?'
In the pragmatically oriented functional approach to syntax adopted
here, the c'est-cleft is interpreted as a syntactic device by which a constituent
whose referent plays the pragmatic role of focus in a propositon but whose
normal position does not mark it as focal is allowed to appear in the preferred
 SVO SENTENCES IN FRENCH DISCOURSE 225

post-verbal position for foci via creation of an additional or "auxiliary" clause

of the preferred type, containing the copula être, in whose free focus position
the constituent can appear. Thus the c'est-cleft is essentially a focus marking
device. As a corollary, the clause which now lacks a focus constituent is
marked as non-focal or pragmatically presupposed, i.e. its propositional con­
tent is assumed by the speaker to be given or recoverable in the hearer's mind
at the time of utterance.
The c'est-cltf construction is in a sense a reversal of the pragmatically
unmarked sequence, in which a topical element is followed by a predicate
containing new information about this element. In the c'est-construction, it is
the predicate that is pragmatically recoverable, and the clefted NP that adds
the new, i.e. unpredictable piece of information. This interpretation of the
c'est-cleft also accounts for the observed discourse tendency for the clefted
constituent to correspond to a subject NP or an adverbial phrase in the non-
clefted version of the sentence, but not to an object NP.12 It is well known that
subjects and scene-setting adverbials, but not objects, tend to be topical con­
stituents. Because of this discourse tendency there is a greater functional
need for focus marking when the focus falls on one of these normally topical
constituents.
While the function of the c'est-construction is to mark a constituent as a
focus, the inherent discourse-pragmatic status of the focus referent as given,
recoverable or new is irrelevant. Although a clefted NP can have a new dis­
course referent, as in (4)
(4) A: si, c'était à... comment, celui qui nous a vendu le champ là,
c'était eh
B: oui Beri-, c'était BERITO qua vendu le champ (François
1974:764)
A: 'yes, it belonged to... what's his name, the guy who sold us the
field, that was uh
B: yes Beri, it was Berito that sold the field'
the c'est-clcit can be used also with highly recoverable and even given refe­
rents in focus function, as shown in (5), where the clefted NP is a lp pronoun
of the tonic series:
(5) tous ceux qu'ya dans le quartier, c'est MOI qui leur a donné des
bouts (François 1974:782)
all those that are in the neighborhood, it's me that gave them cut­
tings'
226 KNUD LAMBRECHT

Example (5) shows that the pragmatic function of the c'est-construction can­
not only be to keep lexical material out of subject position, since the non-
clefted version of (5) would have the clitic pronoun je as its subject, i.e. would
in fact conform to the preferred clause pattern (cf. tous ceux qu'ya dans le
quartier, je leur ai donné des bouts). However this non-clefted version would
not exhibit the intended topic-focus distribution. That the c'est-cleft cannot
be uniquely defined as a "subject-NP-avoiding" construction is clear of
course also from the fact the the clefted constituent does not have to corres­
pond to a subject in the non-clefted version, even though clefted objects are
quite rare. Nevertheless the general structural and pragmatic properties of
the c'est-cleft characterize it as a construction whose primary function is to
maintain the preferred clause structure in discourse.

2.1.2. The presentational construction

More directly relevant to the problem of lexical subjects in spoken
French is the strong constraint in this language against the occurrence of new
discourse referents in initial subject position. The existence of this constraint
can be easily verified for indefinite-specific NPs, i.e. for NPs which are for­
mally marked as having referents that the speaker does not assume to be iden­
tifiable by the hearer at the time of utterance (Chafe 1976) and that therefore
are necessarily new in the discourse. In the corpuses I have analysed I have
not found a single instance of an indefinite NP in initial subject position. This
constitutes clear evidence for a pragmatic constraint against the introduction
of "brand-new" (Prince 1981a) referents in subject position. But the con­
straint holds also for "unused" (Prince) referents, i.e. for definite NPs whose
referents are assumed to be identifiable by the hearer but not yet pragmati­
cally available in the context of the utterance. Virtually all subject NPs in the
corpuses have referents that are pragmatically highly recoverable (the only
apparent exceptions being definite NPs with generic referents; cf. section
3.1.3).
New discourse referents, in particular new referents that are intended by
the speaker to become topics in subsequent discourse, are introduced in spo­
ken French, as in many languages, not in canonical subject position but in
post-verbal focus position, by means of a number of presentational construc­
tions. In these constructions, the NP expressing the new referent always has
a non-agentive case role. At the core of all presentational constructions are
semantically highly intransitive clauses containing such verbs as avoir 'to
have', 13 certain verbs of motion and existence, and verbs of perception. These
 SVO SENTENCES IN FRENCH DISCOURSE 227

clauses are among the lowest in transitivity the language possesses. Their
function is not to predicate something about a referent but to simply present
or "locate" the referent in the universe of discourse. In most cases the new
referent introduced by means of one of these constructions is made pragmat­
ically accessible in the discourse through the intermediary of the referent of
a clause-initial clitic. In all cases, the presentational construction conforms to
the preferred clause pattern.
The most frequently used of the presentational constructions are no
doubt those involving the verb avoir 'to have', which I take to be a semanti-
cally intransitive verb (cf. footnote 13). This verb is peculiar in the system of
French syntax in that although it is semantically intransitive it has two argu­
ment positions, a subj ect and an obj ect position. A voir thus allows for the use
of a clause-initial clitic and a post-verbal NP argument without involving the
usual transitivity relation between a subject and an object. I believe it is this
syntactic-semantic peculiarity that makes avoir the preferred candidate for
the presentational function. The most frequent use of avoir is made in the
"idiomatic" sequence y+a (spelled ya in the following examples), which con­
sists of the distal deictic clitic y 'there' and the 3p present form of avoir (tenses
other than the present rarely occur with this construction, but are attested).
Spoken French ya corresponds to standard French il y a, the dummy subject
morpheme il being as usual dropped in the spoken language (cf. section 2 and
footnote 8).
The sequence ya NP is roughly equivalent to English there is NP in its
non-deictic, existential function and can be used, like the existential there­
construction, to introduce a previously unidentified referent into the dis­
course by metaphorically "locating" it in the speech setting (as in ya un livre
sur la table 'there is a book on the table'). 14 But the specific use of ya I am
interested in here is its occurence in what I call the ya-cleft construction. In
this construction, the preferred clause y+a NP, which introduces the new
referent, is immediately followed by a qu(i)-clause, whose subject must be
coreferential with the NP and in which the status of the newly introduced
referent is changed from new to given. The ya-cleft construction is thus a
sequence of two clauses of the preferred type in which a referent first appears
as a focus (the NP) and then as a topic (with t = qu(i)). An example is given
in (6):
(6) à l'heure actuelle, j'm'plains pas, ya un camarade d'usine qui
m'ramène en voiture jusqu'aux Quatre Routes pour prendre l'au­
tobus (François, 1974:818)
228 KNUD LAMBRECHT

'right now Fm not complaining, (there's) a friend of mine from the

factory (who) drives me back to Quatre Routes to take the bus'
In (6) the new referent un camarade d'usine is introduced via ya-clefting in
focus position in the first clause and immediately becomes a topic in the
appended qui-clause, whose subject (qui) must be coreferential with the NP.
(The discourse following the fragment in (6) continues to be about the topic
un camarade d'usine) To summarize, in the va-cleft construction a new dis­
course referent that the speaker wishes to make available as a topic is intro­
duced as a lexical NP in the focus position of a semantically intransitive
clause, so that by the time it assumes the role of a t it has acquired the given-
ness required for this role. The structure [y+a NP qu(î) VP] is a pragmatically
governed syntactic device whereby a new referent that plays the role of topic
in a proposition is kept from appearing as an initial subject NP. Whether the
newly introduced referent actually becomes a "discourse topic", i.e. whether
subsequent discourse will continue to be about this referent or not, depends
on the intention of the speaker. The relevant feature here, as in the other pre­
sentational constructions discussed below, is the avoidance of a lexical sub­
ject NP and preservation of the preferred clause sequence. 15
The ya-cleft is used not only with indefinite NPs, as in example (6), but
also with definite NPs designating referents that are not yet assumed to be
recoverable in the discourse (i.e. with "unused" NPs). This is illustrated in
the next example:
(7) A, a Frenchman living in California, looking at pouring rain:
A: ça sera l'été un jour?
B : c'est pas du tout normal cette année
A: ya mon frère qui vient dans trois semaines, et j'espère bien
qu'i' va faire plus beau quand i'sera là
A: 'is it going to be summer some day?
B: it's not at all normal this year
A: my brother is coming in three weeks and I really hope the
weather is going to be better when he's here'
The canonical SV equivalent mon frère vient dans trois semaines would be dis­
tinctly odd in this context, as would be a left- or right- dislocated construction,
because the referent of the NP mon frère is not yet pragmatically available for
topic status at the time A utters the NP, even though the existence of the
brother is assumed to be known to the addressee.
A presentational device that is closely related to the ya-cleft is the con-
 SVO SENTENCES IN FRENCH DISCOURSE 229

struction in which a new discourse referent is introduced after the verb avoir
but in which the initial clitic is a personal pronoun. As with ya, the new refe­
rent is metaphorically "located" in the discourse, the reference point being
the initial clitic, whose referent is deictically or anaphorically anchored in the
discourse. Here is an example:
(8) j'ai eu mon beau-frère, moi, qui a fait un, un, euh Paris-Nice, le,
la course de lenteur, il a été pénalisé parce qu'il allait trop vite (Fran­
çois 1974:817)
'(I've had) my brother-in-law (me, who) did Paris-Nice, the slow­
ness race, he was penalised because he went too fast'
Semantically this use of avoir is odd in that the cooccurrence of the sequence
j'ai with the following possessive mon beau-frère seems redundant. The literal
meaning of (8), which is reflected in the English gloss, makes little sense if we
try to understand the sentence following the rules of compositional seman­
tics. But the anomaly disappears if we understand the sequence j'ai NP qui as
the pragmatically governed construction it is in spoken French. The verb
avoir in its presentational function does not express possession but existence
or presence in the discourse. This is particularly clear in these two examples
cited by Blanche-Benveniste (1983):
(9) moi j'ai encore un formulaire que j' ai pas
'(me) I have another form I don't have'
(10) moi j'ai pas mon père qui fait les poubelles
'(me) I don't have my father who does trash cans'
Both sentences would express logical contradictions if j'ai were understood
literally as T have, I possess'. Appropriate English glosses are not the ones
given above but rather, for (9), 'there's another form I don't have' and for
(10) 'my father doesn't do trash cans'. 16
Another strategy French speakers make frequent use of in order to intro­
duce new discourse referents is to present the NP as an object of a verb of per­
ception, in particular of the verb voir 'to see'. 17 Examples are given in (11) and
(12):
(11) puis alors maintenant au printemps alors elles tombent, ya la pousse
qui s'fait derrière et tu vois les vieilles qui tombent (François
1974:783)
'but now in the spring they fall, the young shoot develops behind,
and (you see) the old ones (that) fall off
230 KNUD LAMBRECHT

(12) j'm' rappelle, étant gosse, on voyait les avions passer, on disait ah si
c'est joli ça, on voyait tout le monde sortait dans... les rues (...) pour
regarder un avion passer, maintenant on voit les gosses qu'ont ça au
bout d'une ficelle (François 1974:816)
'I remember, when I was a kid, you would see the planes go by,
people would say oh isn't that pretty, (you'd see) everyone would
run out to the street to watch a plane go by, fifty years ago, now
(you see) the kids (that) have them at the end of a string'
As in the previously discussed cases, the strategy involving the verb voir
allows the NP designating the new referent to appear after the verb, in focus
position, without the clause having the degree of transitivity normally
associated with subject-verb-object sequences. The semantic role of the pre­
sented NP is that of an entirely unaffected participant. The proposition con­
taining it does not predicate anything about the referent but simply locates it
perceptually with respect to a deictically anchored given referent (the clitic).
The semantic non-compositionality of these examples is reminiscent of the
semantic anomaly in the avoir strategy. In all cases, a periphrastic construc­
tion is used to break down into two propositional units a piece of information
that would be pragmatically unacceptable if it appeared as a canonical SV(O)
sequence. For example in (12) it would be grammatically unobjectionable but
pragmatically odd to say les avions passaient 'the planes went by' instead of the
actually used on voyait les avions passer, or to say les gosses ont ça au bout
d'une ficelle 'the kids have them at the end of a string' instead of on voit les
gosses qu'ont ça au bout d'une ficelle.
The last presentational device I would like to mention here is the con­
struction known as subject-verb inversion. Except in embedded clauses intro­
duced by QU-complementizers (que, quand, comment etc.) and in a few main
clauses with adverbs in initial position, the corpuses contain mostly examples
in which the subject slot left open after the inversion process has taken place
is filled with the dummy subject marker il. The inversion construction is illus­
trated in the following example:
(13) un beau soir il descend une de mes voisines (Albert)
'one evening one of my neighbors comes down'
(14) après il meurt son fils, il meurt sa belle-fille (Albert)
'afterwards his son dies, his daughter-in-law dies'
(15) il va être construit deux immeubles (François 1974:830)
'there are going to be built two buildings'
 SVO SENTENCES IN FRENCH DISCOURSE 231

(16) à Genève il s'est créé des cooperatives d'enseignement

'in Geneva teaching cooperatives were created'
As in other languages that have similar constructions (wether they involve
subject-inversion or not), the predicates are restricted in French to certain
verbs of motion (venir 'to come', arriver 'to arrive', descendre 'to come down'
etc.) as well as to a few verbs expressing the presence or arrival at, or the
departure from, the scene of the discourse or a more general scene (including
the "scene" of life, as example (14) shows). Note that the detransitivizing
device of the passive and the reflexive-passive can be used to make a predi­
cate accessible to the inversion construction (exx. 15, 16). Unlike in the pre­
sentational constructions discussed earlier, the NP designating the new refe­
rent in the inversion construction still bears the subject relation to the verb. 18
However the fundamental principle is the same here as elsewhere: the gram­
matically possible canonical SV(O) sequence is avoided in favor of the prefer­
red clause sequence with its initial clitic and post-verbal focus constituent.

2.1.3. Marked topic constructions

The third grammatical device a speaker can use to preserve the preferred
clause structure and thus "avoid" the canonical SV(O) pattern is the "disloca­
tion" of a topic NP to the left or to the right of the clause which contains the
information about the referent of this NP. Continuing an earlier usage, I will
often refer to a left-dislocated NP as a Topic NP (T) and to a right-dislocated
NP as an Antitopic NP (AT). I have presented elsewhere a lengthy discussion
of the syntax and pragmatics of T and AT constructions in spoken French
(Lambrecht 1981) and I will content myself here with a brief summary. In
addition, since T and AT constructions present syntactic and pragmatic alter­
natives to the SV(O) pattern, it will be appropriate to discuss further exam­
ples in connection with the discussion of SV(O) sentences in section 3. The
symbol S will be used from now on to designate (non-dislocated) lexical sub­
ject NPs in preverbal position.
Representative examples of a Topic and of an Antitopic construction are
shown in (17). The two examples will be further discussed, in their full con­
text, in section 3.1 (ex. 19).
(17) a. ce lycée, on m'a dit qu'il était pas terminé
'this lycee, I was told it wasn't finished'
b. où est-ce qu'il est, ce lycée?
'where is it, this lycee?'
232 KNUD LAMBRECHT

In these examples, the NP ce lycée, which precedes the preferred clause, as a

Topic, in (a) and follows it, as an Antitopic, in (b), has the function of lexi­
cally expressing the topic referent. Within the preferred clause structures, the
subject clitic il then functions as the t marker (cf. section 2 above) indicating
the syntactic and semantic role the Topic/Antitopic NP plays in the proposi­
tion (the expression où est-ce-qu' in (17b) functions as a single question word
in COMP). 19
Taken as a single group, Topic and Antitopic constructions are more fre­
quently used in spoken French than SV(O) sentences, although the numeri­
cal difference is not overwhelming. The number of Topic, Antitopic and Sub­
ject NPs in the four corpuses I have analysed is listed in table I. Only those
instances of T and AT are counted whose t anaphor is the subject of its clause.

TABLE I
Number of Topic, Antitopic and Subject NPs
Francois I Francois II Francois III lbert Total
Topic NP 39 36 27 18 120
Antitopic NP 8 10 18 8 44
Total T/AT 47 46 45 26 164
Subject NP 40 13 30 21 104

In table I are counted all occurrences of NPs in the three positions, including
non-lexical items like demonstratives, "indefinite pronouns" and quantified
non-specific NPs (e.g. celui-là 'that one', tout 'everything', rien 'nothing', tout
le monde 'everybody' etc; cf. section 3.1). As the table shows, even though
ATs are always fewer in number than Ts or Ss, the total number of disloca­
tions of both types is systematically higher than the number of Ss, with an
average difference of roughly 50%. The proportion would shift further in
favor of T/AT if indefinite pronouns and repeated identical subject NPs were
not counted (e.g. François I has nine S occurrences of les enfants 'the chil­
dren'). A considerably higher proportion of Ts is reported in a recent paper
by Barnes (1984), in which 159 Ss contrast with 310 Ts (not counting ATs).
From the syntactic point of view, the Topic and Antitopic constructions
differ from the previously described Identificational and Presentational con­
structions in that the lexical NP does not appear in post-verbal focus position
but in a "non-relational" position, i.e. in a position outside of the semantic-
syntactic network of the proposition. This entails that the T and AT NP is
essentially autonomous with respect to the syntax and semantics of the clause
 SVO SENTENCES IN FRENCH DISCOURSE 233

(cf. Lambrecht 1981:53ff). There are important syntactic differences

between T and AT constituents, but the main criterion for relational indepen­
dence from the proposition is the same in both cases: both Ts and ATs can
always be omitted from a sentence without causing syntactic or semantic ill-
formedness.
This independence of Ts and ATs from the propositions with which they
are associated correlates with the pragmatic status of their referents as recov­
erable elements in the discourse. This is of course not to say that their
presence is superfluous or redundant. I take their general function to be the
naming, in its lexical form, of an already recoverable but not yet given refe­
rent so that within the proposition expressing the information about it this
referent can acquire the givenness status it needs for t function in the clause.
As I mentioned earlier, givenness is normally the pragmatic status a topic
referent must have acquired in spoken French before it can be involved in an
aboutness relation with a proposition. This explains the overwhelming pre­
ference for t to be a pronoun.
Although the general pragmatic requirement for dislocated NPs to have
recoverable referents is the same for T and AT constructions, there is usually
a pragmatic difference between the two strategies. This difference has to do
with the different degree of topic continuity the referent has in the discourse,
i.e. with the different degree to which the referent of the topic expression can
count as being pragmatically established in the discourse (for the concept of
topic continuity cf. Givón 1983). ATs are more continuous than Ts, i.e. their
referents are assumed to be more easily recoverable in the context than T
referents (but of course less recoverable than t referents, which need no lex­
ical representation). For the more discontinuous strategies of topic switching
and topic establishment a T construction must be used. The difference
between the two strategies can be illustrated with this example:
(18) Husband to wife, complaining about the food on his plate:
H: a. Ça n'a pas de goût, ce poulet
b. ?Ce poulet, ça n'a pas de goût
'this chicken has no taste'
W: a. Le veau, c'est pire
b. ??C'est pire, le veau
'veal is worse'
Even under the assumption that the husband's remark is discourse-initial, the
AT device is appropriate in the first turn. The chicken on the plate is already
a quasi-established topic referent because of its salient presence in the dis-
234 KNUD LAMBRECHT

course setting, particularly since in the given situation food is an expected

topic of conversation. ATs occur frequently in such ''pointing" contexts. A
topic-shifting T construction would here be an overuse of a more powerful
strategy and therefore less appropriate (though not entirely unacceptable, I
believe). In the wife's apologetic reply, however, the topic-shifting T device
is obligatory and the more continuous device of the AT would be unaccepta­
ble. This is so because the new generic topic le veau 'veal' is not yet estab­
lished in the discourse. But the referent of le veau is nevertheless pragmati­
cally inferrable because it takes part in the general "meat scenario" or frame
that was evoked by the husband's remark about the chicken. Without this
possibility of frame inference neither a T nor an AT would be acceptable and
a new-referent-introducing (presentational) construction would have to be
used instead.
In summary, the T and AT constructions can be interpreted from a cog­
nitive-pragmatic point of view as grammatical devices whereby speakers can
name pragmatically recoverable topic referents outside the proposition in
order to make them available for given t status within the clause. Whereas the
previously discussed Presentational constructions can be defined as strategies
"promoting" a referent from focus to t status, the Marked Topic construc­
tions are strategies promoting an already recoverable (T or AT) referent to t
status. In all cases however, application of the strategy results, in one way or
another, in preservation of the preferred clause structure [clitic+Verb (X)].
One peculiarity of the "referent promotion" that takes place in the AT
construction is that by the time the AT referent is named in its lexical form it
has already been referred to in non-lexical (clitic) form inside the clause. It is
important to realize that this does not entail that the AT is an afterthought in
the proper sense of this word (as has often been claimed). The speaker who
taining the associated t and not after some clause higher in the tree structure.
It also explains why, in those cases in which the t is a genitive or dative
to. But the speaker can expect the hearer to be able to temporarily "hold" the
propositional information until the AT is uttered, precisely because the
occurrence of the t pronoun is a signal (under normal cooperative conditions)
that its referent is going to be named immediately afterwards and that this
referent cannot be new in the discourse (because new referents do not have t
agreement). This explains why — unlike the T, which can be indefinitely far
removed from the proposition about its referent — the AT is subject to the
"Right-Roof Constraint", i.e. must appear immediately after the clause con­
taining the associated t and not after some clause higher in the tree structure.
It also explains why, in those cases in which the t is a 'genitive' or 'dative1
 SVO SENTENCES IN FRENCH DISCOURSE 235

clitic, the AT, but not the T, must receive a case marking corresponding to
the case of the clitic, in the form of the preposition de or a.2()
I will now turn to the analsysis of SV(O) constructions involving lexical
subjects, whose occurrence constitutes an exception to the general tendency
in the language towards preservation of the preferred clause structure. In the
course of the discussion, the SV(O) pattern will be contrasted whenever
necessary with topic marking as well as with presentational constructions.

3. The status of lexical subjects

3.1. High and low topicality of referents
As I observed earlier, all occurrences of S, T and AT NPs in the corpuses
have referents that are pragmatically recoverable at the time the NP is
uttered. Thus, even though there can be subtle differences in referential con­
tinuity between these different strategies, the principal difference is not one
of the inherent pragmatic status of the NP referent in the discourse. Rather
the general claim I will make concerning the difference between SV(O) con­
structions and those constructions that preserve the preferred clause unit is
that the choice of one or the other strategy is to a large extent determined by
the higher or lower degree of topicality the referent of the NP has in the dis­
course. This difference in topicality often correlates with the difference
between foregrounded and backgrounded parts of the discourse, in the sense
of Hopper (1979) and Hopper and Thompson (1980). Lexical subject NPs in
spoken French strongly tend to have referents whose topic status is low, and
they often occur in backgrounded portions of a discourse (discourse being
mostly synonymous with conversation in this study). The pragmatic devices
described in the first part of this paper on the other hand tend to be used as
foregrounding devices by which referents are singled out as more salient par­
ticipants in the ongoing discourse. The following parameters will be shown to
be criterial for the choice of one or the other of these strategies:

Topicality parameters
High Topicality Low Topicality
more salient referent less salient referent
more anaphoric referent less anaphoric referent
more specific referent less specific referent
higher transitivity of clause lower transitivity of clause
little or no syntactic subordination frequent syntactic subordination
236 KNUD LAMBRECHT

These parameters interact with and depend on each other in various ways.
Higher topicality of a referent entails that the referent is a more salient pro­
tagonist in the discourse. And salience in the discourse entails that the refe­
rent will tend to be a topic in more than one clause, i.e. will extend anaphor-
ically over a stretch of discourse. A more salient referent is also likely to be
more specific and individuated, because people tend to talk more about
things whose identity matter than about unspecified things. Furthermore a
topic referent is usually described as being involved in some action or process
rather than in a state. And actions and processes tend to involve more agen-
tive participants. Referents with high topicality will therefore tend to appear
in clauses of higher transitivity. And high transitivity is usually considered to
be a property of main rather than subordinate clauses.

3.1.1. Topicality and salience

I would like to begin with a text example that illustrates the general topi­
cality properties of the three main devices at hand, i.e. the T, AT and S
devices. The example is a fragment from a conversation about a new housing
project in a suburb of Paris. The relevant NPs are in boldface:
(19) A: y a des écoles aussi?
B: ah ya des écoles, ah oui ya des écoles (...)
A: et le lycé d'Argentueil (T i )...
B: ça y est
A: ...ça va être par là?
B: oui
C: non, écoute, moi je n 'sais plus, parce que là ce lycée que tu m'dis
sur le boulevard (T.)
B: Oui sur le boulevard
C: ba alors, ce lycée (T.) on m'a dit qu'il était pas terminé et
qu'i'serait à peine termine pour quand les enfants (S) pren­
draient au lycée, alors je voudrais...
B: alors peut-être
C: ...savoir où est-ce qu'il est, ce lycée (AT.), si c'est celui qu'tu
m'dis qu'ya déjà les enfants dedans (François 1974:769)
A: are there schools there too?
B: sure there are schools, oh yes there are schools
A: and the Argentueil lycée...
B: it's right there
 SVO SENTENCES IN FRENCH DISCOURSE 237

A: .. .is going to be over there?

B : yes
C: no, listen, (me) I don't know anymore, because there this
lycée you're telling me on the boulevard
B: yes on the boulevard
C: well, this lycée, I was told it wasn't finished and it would hardly
be finished for when the children would enter the lycée, well
then I would like to...
B: well maybe
C: ...know where it is, this lycée, if it's the one you're telling me
that has already the children in it
After the new discourse referent des écoles 'schools' is introduced in the first
turn by means of a presentational device, the topic T. can be established
directly because its referent is inferrable as an instance of the set écoles. In the
somewhat confusing discussion following the first mention of this topic by
speaker A, speaker C tries to identify a specific highschool whose construc­
tion is not finished and to distinguish if from another school which speaker B
has claimed earlier to be already used by pupils. The relevant fact here is that
speaker C uses the T device first to establish {ce lycée que tu m'dis sur le
boulevard) and immediately afterwards to confirm (alors, ce lycée) a parallel
topic T., but later on resorts to the AT device (AT.) to continue commenting
on the previously established topic T .
There is a clear pragmatic difference between the underlined T and AT
constituents referring to the lycée and the subject NP les enfants 'the chil­
dren'. Whereas the referents of the T and AT NPs are in some intuitive sense
what the propositions they are associated with are about, it is intuitively much
less obvious, in the context of (19), that the subject les enfants expresses a refe­
rent about which the predicate prendraient au lycée adds some new informa­
tion. The non-specific referent of the phrase les enfants does not seem to have
the necessary salience in the context to be considered a topic at all. Rather the
entire information contained in the quand-clause is in a sense semantically
and pragmatically subordinated to the sentence topic le lycée. This semantic
and pragmatic subordination is then reflected in the syntactic subordination
of the quand-clause. This becomes clear if we apply the aboutness test
(Reinhart 1982) to verify the topic status of the NP. It would make sense to
say "I was told about this school that the students wouldn't be able to use it"
but it would hardly make sense, in the context of this example, to say "I was
told about the students that the school wouldn't be finished". Even though in
238 KNUD LAMB RECHT

the proposition les enfants prendraient au lycée, taken in isolation, we can say
that the predicate is about the subject, we cannot say that this aboutness rela­
tion holds in the particular context of example (19). In context, certain sen­
tences, like the one containing the subject NP les enfants, evoke a single
scene, an event or a state in which the subject referent matters only insofar as
it is a necessary component in the scene, not as an independent discourse par­
ticipant with topic status.21
The association of high topicality with the T/AT device and of low or no
topicality with the S device becomes particularly obvious in contexts where
one and the same referent is coded with one or the other device depending on
its varying salience at different points in the discourse. Such a case is illustrat­
ing in the following example. The topic of the discourse from which (20) is
taken is the youngest daughter in the family and her problems in passing a cer­
tain highschool exam. The daughter is referred to in the first passage with the
clitic elle 'she'. The clitic i' 'they' at the beginning refers to the school
authorities; Mademoiselle G. is the daughter's teacher; and ça i t ' at the end
of the second passage refers to l'âge 'the age' of the child:
(20) A: i'regardent l'âge, alors T. (S.) m'a dit du fait
qu'elle (tj.) est tout de même - oh évidemment faut dire c'qu'il est,
y en a beaucoup qui passent à dix ans, normalement elle (t.) est
de l'âge
(twenty-two turns)
A: ah oui j'sais bien, m'enfin quand même, regarde ma... moi je
nsais pas maintenant, Mademoiselle G. (T.) c(ti)'est comme
tout hein, elle(T), c(ti)'est une aucienne institutrice, hein, et elle
(t.) ma dit ça joue quand même un rôle chez les enfants (Fran­
çois 1974:773)
A: they look at the age, well Miss G. told me since after all she is
- oh you gotta admit there are many who pass at age ten, nor­
mally she's reached the age
( )
A: oh yes I know, but nevertheless, look my... me I don't know
now, Miss G. she's really, uh, (she) she's a former elementary
school teacher, uh, and she told me it plays a role after all with
the children
The school teacher Mademoiselle G. has not been mentioned in previous dis­
course but is pragmatically inferrable from the school frame evoked in con­
nection with the daughter. Otherwise the NP could be neither a S nor a T or
 SVO SENTENCES IN FRENCH DISCOURSE 239

AT. When Mademoiselle G. is mentioned for the first time, as S., the salient
sentence (and discourse) topic is the daughter. The clitic elle unambiguously
refers to this topic, not to the subject NP. The teacher is only casually men­
tioned in relation to the daughter and there is no anaphoric pronoun referring
to here after this mention. Her topic status is low, therefore the NP appears
as an S. In the second passage however, twenty-two turns later in the conver­
sation, the teacher Mademoiselle G. acquires greater salience, even though
the primary topic is still the daughter. Her professional background is now
mentioned and her opinion in school matters is reported. The referent has
acquired topic status, therefore the coding of the NP changes from S to T,
with the anaphoric clitics (c' and elle) now referring to her, not the daughter. 22
The difference between high and low topicality of referents and the
association of S coding with low pragmatic salience is well illustrated also in
the following passage, which is taken from a discussion about the problem of
obesity in the United States (data from Barnes 1983). Speaker M. is a French­
woman married to an American. The passage narrates the first encounter
between the speaker's parents and the parents of her husband:
(21) M: ben alors, moi je vais te dire, quand mes parents (S) sont venus
pour le marriage, alors euh... évidemment, mon père (S) a la
même taille que moi, ma mère (S) est plus petite, euh, mon père
(S) fait euh cent dix pounds cent dix pounds, c'est à dire que cin­
quante deux kilos
E: oui
M: cinquante deux kilos
C: c'est un moustique!
M: mais quand on l'a vu, alors la famille de Bill... évidemment, son
frère (T), il fait deux cents et quelques pounds
E : deux fois ton père !
M: son père (T), qui, qui est vraiment trop gros... la mère (T) bon,
ça va, mais enfin, enfin, une famille typiquement américaine de
c'point d'vue-là. Quand ils ont vu papa, tout petit, tout chétif ya
mon beau-frère (ya-cleft) qui l'a porté comme ça. Il l'a soulevé
comme ça. Il en revenait pas de sa légèreté!
M: well then, (me) I'm going to tell you, when my parents came
for the wedding, then uh... of course, my father is the same
size as me, my mother is smaller, uh, my father weighs, uh, 110
pounds, 110 pounds, that is, only 52 kilos
240 KNUD LAMBRECHT

E: yes
M: 52 kilos
C: he's a mosquito!
M: but when people saw him, well. Bill's family,., of course his
brother he weighs two hundred and some pounds
E: twice as much as your father!
M: his father, who, who is really too fat... the mother, well she's
ok, but anyway, anyway, a typically American family from
that point of view. When they saw papa, so little, so puny, my
brother-in-law carried him like this. He picked him up like
this. He couldn't get over his lightness!
The main event speaker M, wishes to narrate is the encounter of the two
families, highlighted by the striking physical difference between them. In
order to introduce this event, M. first gives a minimum of background infor­
mation, in the form of the preposed adverbial clause quand mes parents sont
venus pour le marriage 'when my parents came for the wedding', in which the
referent of the NP mes parents, being an element of the background, is
characteristically coded in S form. However the speaker then realizes that
this background information is not sufficient because her audience knows
nothing about her parents' physical appearance. She therefore interrupts her­
self at the beginning of the main clause, which is introduced by alors 'then',
in order to provide more information about her parents. She does so in the
form of three short clauses describing her father and her mother, preceded by
the pragmatic particle évidemment (roughly 'of course'), which marks the
information provided in these clauses as the obvious yet necessary
background. The relevant fact here is that even though these clauses clearly
present information about her parents, i.e. even though mon père and ma
mère can be interpreted as the topics of these clauses, the referents are coded
as lexical subject NPs. S coding thus marks these referents as pragmatically
backgrounded elements with respect to the yet to be narrated main portion of
the story.
Having provided the background information about her parents,
speaker M. then proceeds to narrate the main event. The form she chooses
for the narration is the same as in her first attempt: an adverbial clause fol­
lowed by a main clause introduced with the adverb alors. However at this
point the speaker again realizes that in order to make the main event under­
standable she has to provide more background information, this time about
her husband's family. Again she interrupts herself to provide this infor-
 SVO SENTENCES IN FRENCH DISCOURSE 241

mation, and again she uses the particle évidemment, marking the information
both as necessary and as obvious. This time however, in describing the phys­
ical appearance of her husband's family, beginning with her brother-in-law,
the speaker resorts to the T strategy. Note that this change in syntactic coding
occurs despite the fact that the descriptive vocabulary is almost identical in
both passages. I believe that this difference in syntax, in spite of the similarity
in semantic content, is the result of two competing pragmatic factors. On the
one hand the description of the American family represents background
information with respect to the main event, on the other hand this
background itself pertains directly to the issue of obesity, i.e. to the discourse
topic of which the whole narration is intended to be a relevant illustration.
Thus compared to the members of the French family, the Americans
described here are more topical discourse referents. This difference in rela­
tive topicality is then expressed in the formal difference between S and T cod­
ing.
Having now provided all the necessary elements for the audience to
understand the main piece of information she is about to present, the
speaker, for the third time, begins her narrative with an adverbial quand-
clause. And this time the following main clause expresses the main event.
which is the high-light of the narration. Of particular interest here is the use
of the presentational ya-cleft construction as a foregrounding device for the
presentation of salient new information. Even though the referent of the ya-
clefted NP mon beau-frère is clearly already recoverable, i.e. even though the
newness of the referent which is normally required in the ya-construction
does not obtain, the more dynamic presentational construction is used
instead of the dislocation device, which would be less acceptable in this con­
text. Due to the fact that in the presentational ya-construction normally a new
referent is introduced into the discourse and immediately followed by a prop­
osition presenting new information about this referent, this construction has
a pragmatic "all new" character that makes it the ideal coding device for the
pragmatic function of event-reporting or for the marking of a piece of infor­
mation as unexpected (cf. Lambrecht 1984a). This dynamic function can be
exploited here, despite the fact that one of the appropriateness conditions for
the presentational device, the newness of the referent, is not satisfied.
Note that in all three syntactic devices used here, the T, S and ya devices,
the relevant selection criterion is neither a difference in the inherent semantic
features of the nouns (all nouns are kinship terms) nor a difference in the
pragmatic status assigned to the referents in the discourse (all referents are
242 KNUD LAMBRECHT

equally recoverable from the evoked kinship frame), but rather a difference
in the relative salience and topicality of the referents in the context.
In view of examples like (21) it seems necessary to define the pragmatic
relation topic-of as a scalar notion. Even though in the passages describing
the French and the American family the predicates of all clauses can be said
to add information about the different family members, i.e. even though the
various NPs all are in some sense the topics of the clauses with which they are
associated, the different referents are topical to a higher or lower degree in
their respective contexts. Nevertheless on the syntactic level the scalar notion
becomes necessarily a polar one: within the background-foreground contrast
expressed in (21), only the more foregrounded referents, i.e. those with rela­
tively higher topicality, are formally marked as topics.

3.1.2. Anaphoric continuity

As I mentioned earlier, the non-salient status of an S referent in a dis­
course entails that such a referent is normally not anaphorically referred to
beyond the single clause in which it appears. But a referent that is salient
enough to acquire topic status typically extends anaphorically over more than
one clause. This difference in anaphoric continuity between more vs. less
topical referents can be observed in all three fragments discussed so far. A
particularly clear case is the difference in anaphoric "scope" between le lycée
and les enfants in example (19), which was analysed in section 3.1.1. As an
apparent exception to the principle that associates high anaphoricity with
topic NPs and low anaphoricity with subject NPs we might cite the case of the
subject NP mon père in (21). This NP is anaphorically referred to after its first
mention with the clitics c and l'. It may be possible to explain this exception
psychologically, as a result of the unplanned character of speaker M's dis­
course, in which the referent gradually takes on a degree of salience it was
originally not meant to have. In addition, since we analysed the referent of
mon père as a case of relative low topicality, a certain degree of anaphoric con­
tinuity seems naturally compatible with this subject referent.
The relationship between high topicality and referential continuity is
described by Hopper (1979) in a discussion of Old English narrative prose:
Now although new characters can be, and often are, introduced in foreg­
rounded narrative, it is usually with a view to a role of some kind in the nar­
rative, which is then related immediately; the casual presentation of new
personages is characteristic rather of backgrounded material. (...) (We find)
high topicality of the subject, which is almost always either an anaphoric pro-
 SVO SENTENCES IN FRENCH DISCOURSE 243

noun or a definite noun without focus. The characteristic "oldness" of the

subject in foregrounding is, of course, a natural consequence of the tendency
for narratives to be concerned primarily with a small number of participants
and, hence, to have continuity of topic-subject in the main story-line. In
background, on the other hand, a variety of other topics can be introduced
to support and amplify the story-line. (Hopper 1979:224)
Although the rules for literary written prose are of course not in all respects
comparable to the rules of spontaneous conversation (in particular in spoken
French definite subject NPs do not code referents with high topicality), the
common syntactic characteristics are obvious in Old English and in contem­
porary spoken French.

3.1.3. Referential specificity

An important semantic parameter influencing the degree of topicality of
a referent is the parameter of referential specificity. As a general rule, the less
specific a referent, the more likely it is to be coded in S form, although non­
specific referents do sometimes appear as T or AT NPs. This is a natural con­
sequence of the fact that non-specific referents are inherently less topic-
worthy, i.e. less 'interesting" to talk about, than specific, individuated refe­
rents. In spoken French, NPs designating non-specific referents can be
divided into two classes: those which, for pragmatic reasons, tend to be S-
coded but which can grammatically also appear as Ts or ATs, and those which
the grammar only allows in S-coding. The first group is the set of generic NPs,
the second group is the set traditionally referred to as indefinite pronouns.
An already discussed example of a non-specific generic S is the NP les
enfants 'the (school) children' in (19), which contrasted with the highly topi­
cal and specific NP le lycée 'the highsehool'. Two more examples of generic S
referents, taken from the same conversation as (19), are given in (22) and
(23).
(22) pour le moment, les enfants qui vont au lycée vont à Champagne là
haut (François 1974:769)
'for the time being the children that go to (the) lycee go to Cham­
pagne'
(23) tu comprends, les élèves qu'on trouve en sixième ont pour la
plupart...auront leur douze ans revolus dans Vannée (François
1974:773)
'you know, the children that you find in sixth grade have for the
most part... will have completed their twelfth year during the year'
244 KNUD LAMBRECHT

The S phrases in boldface in these two examples do not refer to specific indi­
viduals or groups of individuals, but to all members of a particular class (the
class of all high school students in a geographic area (22), and the class of all
sixth-graders (23)). In (23), the class is then restricted by the following quan­
tifier phrase pour la plupart Tor the most part'. 23 Generic NPs are often inter­
preted as a subset of (universally) quantified NPs, and quantification is to
some extent incompatible with the pragmatic relation of aboutness (cf.
Reinhart 1982, and below). There are many more examples of S-coded
generic referents in the corpuses, e.g. les enfants 'children' (with nine occur­
rences in Francois I), les cités 'housing projects', les gens de la mer 'people
who live near the ocean' etc.
Examples (22) and (23) are particularly interesting in that they involve
complex subject NPs, i.e. NPs of the type sometimes characterized as
"heavy". Heaviness of an NP is sometimes invoked as one of the psychologi­
cal factors favoring dislocation: because it can happen with such NPs that a
clause is separated from its subject noun by an intervening relative clause, the
thus created distance between the subject and the verb is thought to favor the
"repetition" of the subject before the verb, in the form of a "resumptive" pro­
noun. The fact that in spite of the heaviness of the subject phrases dislocation
does not take place in (22) and (23) may be interpreted as another piece of
evidence that dislocation is essentially a pragmatically motivated phenome­
non, involving the status of referents in a discourse, not a purely syntactic
process determined by formal properties of the sentence,
That genericity is not in principle incompatible with topic status is shown
in the following example, in which a semantically highly non-specific NP (les
gens '(the) people') is coded first as an S, then as an AT. The example is taken
from a conversation about the difficult living conditions in Marseille under
the Nazi occupation.
(24) tenez, quand on vous joue des films de cette époque-là, hein, ben, les
gens (S) doivent se dire, mais c'est pas vrai, ils vivaient comme ça,
les gens (A T), fallait toujours se cacher (Albert)
look, when they show you movies from that period, uh, well,
people must say to themselves, I can't believe it, they lived like
that, the people, you always had to hide'
In its first occurrence, the NP les gens refers to the entirely indefinite set of all
people who might watch these movies. It is this sort of highly indefinite refe­
rent that often gets grammaticized in languages into an indefinite 3p pronoun
 SVO SENTENCES IN FRENCH DISCOURSE 245

(cf. e.g. French on from Latin homo 'man', or German man etc.). It is clear
here that the predicate doivent se dire does not add information about the "re­
ferent" of les gens. The NP is not a topic, therefore S-coding is used. But in
the second occurrence, les gens refers to the people living under the occupation
and whose life is described in the movies mentioned in the example. The prop­
osition ils vivaient comme ça adds information about these people, as does the
last clause in the fragment (and further discourse following it in the corpus).
Therefore the NP is coded as an AT. Example (24) contains a revealing min­
imal pair demonstrating that non-specific referents can be topics when they
play a role of some sort in the discourse but must be S-coded when such a role
is not intended.
The case of the indefinite S les gens in (24) with its quasi-pronominal
character is similar to the case of so-called indefinite pronouns, which cannot
normally appear dislocated in T or AT position. Such indefinite NPs are
quelqu'un 'someone' (in its non-specific sense), rien 'nothing', tout 'every­
thing', tout le monde, tous les gens 'everybody'. These quantified NPs have in
common that they either do not have any referent at all or that their referents
are so indefinite that they cannot be talked about, i.e. that no information can
be added about them. Consequently they do not appear as Ts or ATs in the
corpuses. However it sometimes happens that an indefinite quantified NP
globally refers to the members of a specific set, and in such cases topic status
is again possible. One such case is the NP tous 'all of them' in example (25):
(25) A: oui, enfin, lui, Guerini (?i) meurt (...) après il meurt son fils, il
meurt sa belle-fille
B: tous i' sont morts, tous (Albert)
'yes, okay, (him) Guerini (?he) dies (...) afterwards his son
dies, his daughter-in-law dies all of them they died, all of thenv
The NP tous here refers to the set of all members of the specific family being
talked about and as a consequence T-coding is used.

3.1.4. Transitivity
An important parameter associated with the use of subject NPs and their
backgrounding function in discourse is the low transitivity of the clauses in
which they occur. I am using 'transitivity' here as a scalar notion, in the sense
of Hopper and Thompson (1980), i.e. as a global property of a clause deter­
mined by various semantic, syntactic and pragmatic parameters. In spoken
246 KNUD LAMBRECHT

French Ss occur in their vast majority in intransitive clauses, in particular in

clauses containing the verb être 'to be'. A text count of the François corpus
revealed the following clause types. Among the 83 clauses containing subject
NPs, 69 (83%) are syntactically intransitive and have semantically non-agen-
tive subjects. In the few cases where a post-verbal NP cooccurs with these Ss
this NP is semantically a locative or it is a predicate noun (as in les cités longent
la ligne de ceinture 'the housing projects are located along the belt line', la
femme était marchande de légumes 'the wife was a produce seller'). More than
half of the intransitive clauses have être 'to be' as their predicate (38 occur­
rences). The remaining 31 intransitive clauses have stative predicates {longer
'to be located along sth', se reposer 'to rest' etc.), predicates that indicate a
change of state (se développer 'to develop', se former 'to form' etc.) or pred­
icates that express motion (venir 'to come', partir 'to leave' etc.).
In the 14 syntactically transitive clauses containing lexical subject NPs,
i.e. in the clauses with more or less agentive subjects, we find six occurrences
of dire 'to say', two occurrences of raconter 'to tell', and single occurrences of
donner 'to give', remettre 'to hand', mettre 'to put', se regarder 'to look at one­
self/each other' (a reflexive-reciprocal verb, thus a verb of reduced transitiv­
ity), tirer 'to pull' and photographier 'to photograph'. Note that in almost all
of these transitive clauses the S is the only lexical NP element. The object is
typically a clitic pronoun or it is not expressed at all. Non-expression of the
object is observable in particular with the verb dire 'to say' (as in
Mademoiselle G. m'a dit 'Mademoiselle G. told me', ex. 20). Concerning the
verb dire, it is interesting to refer to a recent paper by Munro (1982), in which
'say' verbs are cross-linguistically analysed as a special class of intransitive
verbs. Munro observes that non-expression of the object of 'say' verbs is a
common feature in many languages. If, following Munro, we count the six
occurrences of dire 'to say' and the two occurrences of raconter 'to tell' as
occurrences of intransitive verbs, we eliminate the only recurring exceptions
to the rule which associates Ss with intransitive clauses. Finally it should be
observed that among the few occurrences of agentive S's many are proper
names, whose exceptional status as "pronoun-like" NPs will be discussed in
section 3.1.6.

3.1.5. Syntactic subordination

The last parameter associated with lexical subject NPs which I would like
to discuss here is the relationship between the backgrounding function of SV
clauses and the fact that they frequently occur as subordinate clauses. The
 SVO SENTENCES IN FRENCH DISCOURSE 247

correlation between backgrounding and syntactic subordination is well-

known (cf. e.g. Hopper and Thompson 1980), as is the correlation between
certain so-called main-clause phenomena and pragmatic foregrounding
(Hopper and Thompson 1973, Givón 1979).24 In spoken French, S's generally
appear more freely in embedded clauses than T's or AT's. Among the 83 sub­
ject NPs in the François corpus, 22 (26,5%) occur in embedded clauses. But
only 10 out of the 102 occurring T NPs (9,8%) and 4 out of the 36 AT NPs
(11,1%) are found in subordinate position. As recent research has shown
(e.g. Haiman and Thompson 1984, Lakoff 1984), the difference between
main and subordinate clauses is not as clear-cut as has usually been assumed
and is often a matter of degree rather than of categorial distinctness. In order
for a comparison between embedded Ss and embedded Ts/ATs to be fully
meaningful it would therefore be necessary in each case to determine the
degree to which the clause can be considered subordinate, taking into
account such factors as the preposing vs. postposing of the embedded clause
with respect to the matrix as well as the presuppositional status of the different
preposed and postposed clauses. Not being able to go into sufficient detail
here I will content myself with presenting a few clear cases illustrating the
association of backgrounded S status and syntactic subordination.
The occurrence of subject NPs in temporal quand-clauses was already
mentioned in the discussion of examples (19) and (21) above. Possible
psychological evidence for a quasi-grammaticalized association of low topi­
cality with syntactic subordination can be seen in the following fragment
taken from a discussion of the respective advantages of the ocean and the
mountains as vacation places:
(26) A: c'est plus varié, bah oui mais tu parlera ça à des gens de la
mer...chez nous on a une Bretonne, elle te dira qu'la mer (S)
n'est...
B: oui ah ah
A: .. jamais la même
B: hein? ah ba ah
A: elle te dira la mer (T) c est jamais la même (François 1974:796)
A: 'it (i.e. the mountains) is more varied, okay yes, but you say
that to ocean people...at our place we have a woman from
Brittany, she'll tell you that the ocean is...
B: yes uh uh
A: ... never the same
B: what? oh well uh
248 KNUD LAMBRECHT

A: she'll tell you the ocean it's never the same'

Since the referent of la mer is topical in the context of (26) the speaker may
have felt that S-coding of this referent, motivated by an unconscious associa­
tion of the conjunction que with S, was no the appropriate coding. This may
then have motivated the switch from S to T and from an embedded to a main
clause. I have found no other examples of such apparently automatic associ­
ation between the syntactic status of the clause and the pragmatic status of a
referent, and it may therefore be premature to speak of grammaticalization
here.
A relatively strong correlation between syntactic subordination and S-
coding is found in the constraint against left-dislocation in subordinate
clauses that are parts of syntactic "islands", in particular in relative clauses
(cf. Lambrecht 1981:58ff for some discussion). An example of S-coding in a
relative clause is given in (27):
(27) A: alors, tu as du chèvre-feuille maintenant.,.
B: ah le chèvre-feuille, j'l'ai adopté
C: ah oui du chèvre-feuille
A: ...que maman fa donné (François 1974:782)
A: so you've got honeysuckle now...
B: oh the honeysuckle I've adopted it
C: oh yes honeysuckle
A: .. .that mom gave you
A T construction would seem less acceptable here:
(28) ? le chèvre-feuille que maman elle t'a donné il a pas pris
'the honeysuckle that mom she gave you it didn't catch on'
Clear acceptability judgments are hard to come by in the domain of topic
embedding. 25 Pending further research, the correlation between S-coding
and syntactic embedding has to be characterized as relatively weak, but
nevertheless observable.

3.1.6. A class of exceptions: Proper names

I should mention here one class of exceptions to the general rule which
associates S-coding with low topicality and pragmatic backgrounding. This
exception has to do with the particular referential properties of proper names
and certain common nouns that functions like proper names. In any dis­
course, certain referents occupy particularly prominent positions and are
 SVO SENTENCES IN FRENCH DISCOURSE 249

therefore pragmatically more easily accessible than others. This is always the
case for the speaker and the hearer, who are deictically referred to with lp and
2p pronouns. But there can be other referents in a discourse whose recovera-
bility is almost as high as that of the speaker/hearer, because they are physi­
cally present as prominent participants in the speech situation or because they
stand in a particularly close relationship with the speaker/hearer and are in a
sense "mentally present". Such highly recoverable referents are typically the
members of the family of the speaker and/or the hearer and sometimes close
friends. They are referred to with proper (first) names (like "Mary" or
"John") or with certain name-like expressions (like "mommy" or "daddy").
In a sense, such expressions have a function analogous to the function of deic­
tic or anaphoric pronouns, because in a given context names, like pronouns,
"point to" or identify their referents directly rather than via a category out of
which the hearer has to pick the intended individual.26 Therefore these
expressions can sometimes (but by no means always) be "substituted" for t
clitics. They will then appear as Ss, but their topicality will be much higher
than that of other lexical subjects.
The following example illustrates this peculiar use of proper names
designating familiar referents. The two main referents in (29), Guerini and
Nicolas, are salient topics in the discourse preceding this fragment. Guerini
is a person the speaker had known a long time ago, and Nicolas is the
speaker's husband, who also participates in the conversation as speaker B:
(29) A: oui enfin lui, Guerini (?i) meurt - Guerini ou Martini i's'ap­
pelait, j'm'en souviens plus, ah oui, bon
B: non Guerini, Guerini
A: après, il meurt son fils, il meurt sa belle-fille
B : to us i'sont morts, to us
A: et Nicolas (S) est toujours en vie, lui, parce qu'à l'époque il était
maigre, mon mari hein (Albert)
A: 'yes okay (him) Guerini (?he) dies - Guerini or Martini was his
name, I don't remember, oh yes okay
B: no Guerini, Guerini
A: afterwards his son dies, his daughter-in-law dies
B: all of them they died, all of them
A: and Nicolas is still alive, (him), because at that time he was
skinny, my husband uh'
The syntactic status of Guerini as a T or an S is not entirely clear because the
250 KNUD LAMBRECHT

3p agreement clitic i is phonetically indistinguishable from the final i in

Guerini. However T-coding seems highly likely. But the case of Nicolas is
clear. Even though the topic Nicolas is contrasted with the topic Guerini, i.e.
even though the appropriateness conditions for the topic-shifting use of the T
device clearly obtain, the proper name appears directly in subject position.
Contrastiveness is instead expressed through the AT pronoun lui (cf. foot­
note 22). It is clear that the NP does not designate a backgrounded referent
in this context and that the proposition is about the topic Nicolas. That the NP
Nicolas has here a quasi-pronominal function is confirmed by the fact that an
AT pronoun like lui can normally occur only in coreference with a t clitic or
with another T or AT constituent, but not with an S (cf. e.g. the contrast
between i'va bien, lui 'he's doing fine, him' and ?cet homme va bien, lui 'this
man is doing fine, him'). Striking evidence for the special status of the proper
name Nicolas is also provided by the fact that in the same sentence the com­
mon noun mon mari 'my husband', although coreferential with Nicolas, is not
coded as an S but as an AT.
As another example consider (30), in which the noun maman 'mom'
functions like a proper name:
(30) A; moi j'dis allez, en vacances j'aime mieux aller à Charreau, c'est
pour s'retirer
B: ba oui mais maman (S) se repose pas à Charreau, ya trop de
bruit à l'hôtel (François 1974:806)
A: me I say okay, on vacation I prefer going to Charreau, it's to
get away
B : okay yes but mom doesn't get any rest in Charreau, there's too
much noise in the hotel
Even though the individual referred to as maman does not participate in the
conversation, the high pragmatic prominence of this referent allows the
speaker to code it as a quasi-pronominal subject. And as with Nicolas in (29),
S-coding does not entail backgrounding here. Note that this syntactic coding
of Nicolas and maman is pragmatically different from S-coding in the case of
e.g. mes parents 'my parents' or mon père 'my father' in example (21) above.
Although these nouns also designate a close kinship relation with the
speaker, they are regular common nouns, and the topicality of their referents
is low (cf. also the subject NPs mes enfants and ma soeur in example (31)
below). The case illustrated by (29) and (30) thus constitutes an exception to
the general rule which associates S-coding with low topicality and pragmatic
 SVO SENTENCES IN FRENCH DISCOURSE 251

backgrounding.

3.1.7. A short narrative

I would like to conclude this section on the status of lexical subject NPs
with the discussion of a short narrative in which the pragmatic functions of the
different syntactic constructions described in this paper are well illustrated.
The speaker in this narrative is a working class woman from Marseille who is
talking about the food shortage under the Nazi occupation. The passage
occurs at the beginning of the corpus:
(31) 1 Pournous personnellement voyez, nous personnellement parce
que - je vous dis mes enfants (S) se contentaient de peu - je vais
vous citer un cas, euh, je montais souvent à... ma soeur (S) avait
un bar à Aubagne - et elle - et ce bar (S) était fréquenté beaucoup
5 par des paysans - et avec les paysans ma soeur (S) avait de tout -
ma soeur avait de tout - et je montais une fois par semaine avec
mes enfants - et mes enfants (S) se faisaient un banquet là-haut
parce que ma soeur (S) les privait de rien, hein -j'avais la petite
de ma soeur (present. O) - ma nièce - si vous aviez vu cette petite
10 - elle avait de tout - elle mangeait des frites avec, euh, de la
bonne huile - des bons bifteks - elle était maigre comme un clou
- maigre comme un clou - mes enfants (T) qu'est-ce qu'ils
avaient eu à manger - peu - peu mais tout ce qu'i mangeaient i'l'
mangeaient euh avec appetit - et elle qu'elle avait de tout (T)
15 elle mangeait tout avec délicatesse hein - mes enfants (T) ils
avaient une figure euh on aurait dit des lunes - et elle (T) on
aurait dit qu'elle sortait de Buchenwald.
(Albert)
'for us personally you see, we personally because - I'm telling
you, my children were happy with very little - I'm going to tell
you a case, uh, I would often go to... my sister had a bar in
Aubagne - and she - and this bar was frequented often by far­
mers - and with the farmers my sister had everything - my sister
had everything - and I would go once a week with my children
- and my children would have a banquet up there because my
sister didn't deprive them of anything -I had the little daughter
of my sister - my niece - if you had seen this girl - she had every­
thing - she would eat French fries with uh good oil - good steaks
252 KNUD LAMBRECHT

- she was skinny as a nail - skinny as a nail - my children, what

did they have to eat, little - little - but everything they ate they
would eat with appetite - and she who (she) had everything she
ate everything in a picky manner uh - my children they had uh
faces like moons - and she she looked like she came right out
of Buchenwald.'
The narrative proper, which starts in line 3 with the sentence je vais vous citer
un cas is divided into two parts, the background information (line 2-8) and the
main story (9-19), in which the speaker makes a comparison between the
eating habits of her own children and those of her sister's daughter. Out of the
seven clauses (or eight, if we count the fragment in 3) that make up the
background portion, six contain subject NPs. The referents of all these sub­
jects are restricted to the one clause in which they occur, i.e. they are not
anaphorically referred to in subsequent clauses. Although the referent of mes
enfants (2) is to some extent topical given that the whole narrative is about the
speaker's family and in particular her children, neither mes enfants nor any of
the other S referents is marked as a topic in this backgrounded part of the nar­
rative. The various referents mentioned here are relevant only as necessary
elements leading up to the main story.
The topicality of the referents changes drastically with the beginning of
this main part of the narrative (line 9). The first protagonist, the sister's
daughter, is introduced with a presentational device (j'avais la petite de ma
soeur; cf. section 2.1.2), i.e. in focus position, even though this referent is
pragmatically recoverable from the general family frame, which in the first
part allowed e.g. ma soeur to appear in subject position without prior men­
tion. (Note that the only S referent in the first part of the narrative that is not
inferrable from this frame, ce bar (line 4), is first introduced as an object in
focus position {un bar, line 4).) The presentational device in line (9) marks
la petite de ma soeur as a salient discourse participant, which is to acquire
topic status in subsequent discourse: the next four clauses following the first
mention of this referent are about the daughter. The second set of pro­
tagonists in the main part of the story, the speaker's children, is introduced in
a T-construction {mes enfants, line 12), rather than with a presentational
device, because the children were already mentioned earlier in the story.
These referents are then anaphorically maintained over three clauses. The
end of the narrative consists in successive T-constructions expressing topic
shifts in which are contrasted the two topic NPs elle (cf. fn. 22) and mes
enfants.
 SVO SENTENCES IN FRENCH DISCOURSE 253

4. Summary and conclusion

In this paper I have tried to establish two facts concerning the structure
of the sentence in spoken French. The first is that the canonical SVO sentence
type of linguistic theory is not the type used in actual discourse, and that a pre­
ferred clause type is used instead, whose basic form is V(X), the subject being
a preverbal clitic bound to the verb. I have furthermore tried to show that the
language has systematic ways of preserving this preferred structure by using
grammatical constructions which have the effect of placing NPs in positions
other than preverbal S position. In these constructions, the NP appears either
as a focus constituent after the verb or as a topic constituent outside the
clause, depending on whether the referent of the NP has the pragmatic rela­
tion of focus or of topic with the proposition.
The second fact I have tried to establish is that there exists a class of
rarely occurring but nevertheless fully functional exceptions to this preferred
clause structure, in the form of SV clauses, where S is an NP. In these SV
clauses, the referents of the subject NP (in as much as the NP is referential)
are pragmatically backgrounded discourse participants. This distinguishes
SV clauses from the alternative syntactic strategies involving the preferred
clause structure, in which the referents are pragmatically salient, topical par­
ticipants. The backgrounded character of SV clauses entails that the S refe­
rents have low topic status, i.e. that their pragmatic relation to the proposi­
tion is not or not to the same degree, the aboutness relation found with more
highly topical referents. The backgrounded S referents tend to be non­
specific, non-anaphoric and non-agentive. There are virtually no cooccurring
lexical object NPs in SV clauses.
Thus lexical subjects are not topics; but neither are they foci. They repre­
sent in some sense the pragmatic "default case", They usually occur when the
referent of the NP plays the role of an obligatory part in some conceptual
frame, of an argument in a proposition, not of a topic in a discourse. Some­
times lexical subjects occur as minor topics, when the referent as well as the
proposition about the referent are backgrounded with respect to some more
foregrounded piece of information elsewhere in the discourse. In all cases,
the referent of the subject NP tends to be contained within the narrow limits
of the clause in which it occurs. By its syntactic coding, the S referent is
marked as an element of no or secondary importance for subsequent dis­
course. 27 SV clauses are thus to some extent self-contained pieces of informa­
tion, unlike the preferred clauses, which are marked by their syntax as ele-
254 KNUD LAMBRECHT

ments in an informational sequence. This pragmatically autonomous charac­

ter of SV clauses may well be the main reason why the SV(O) sentence type
has been the preferred model of grammatical description (cf. Introduction),
and why SVO sentences are so easy to elicit in metalinguistic situations, even
though their pragmatic function in discourse is so highly restricted.
From the point of view of discourse, it seems clear that the preferred
clause structure, the structure that contains no S, is the fundamental propos-
itional unit of information. It is fundamental also from the cognitive point of
view, as the optimal processing unit in spoken communication. The cogni-
tively most efficient unit at the clause level seems to be a structure in which
the topic referent(s) can be taken for granted, and in which non-topical refe­
rents, i.e. referents whose identification normally requires lexical coding,
appear in a position reserved for non-agentive arguments, which in a VO lan­
guage like French is the position after the verb. There thus seems to exist a
correlation between constituent order, transitivity and lexical coding such
that lexical NPs are preferred in post-verbal (object) position as the normal
position for less agentive NPs, and avoided in initial (subject) position as the
normal position for agents.
Any lexical NP that is an argument in a proposition codes in fact two
functions, the naming function, which identifies the referent of the NP, and
the relational function, which indicates the syntactic-semantic role this refe­
rent plays in the clause. Although from grammatical tradition we are used to
seeing these two functions combined, they are nevertheless cognitively inde­
pendent from each other. The data suggest that cumulation of the two func­
tions, the naming function and the relational function, is cognitively more dif­
ficult when the position of the NP is one of greater agentivity. From the point
of view of discourse this entails that introducing a referent in a proposition
and talking about that referent as involved in some action are two tasks that
are best carried out independently of each other. This discourse constraint
imposed on the cumulation of the referential and the relational function can
be expressed in the form of a pragmatic maxim: "Do not introduce a referent
and talk about it at the same time."
The asymmetry between subjects and objects (or between agents and
patients) with respect to their lexical naming function has been sporadically
observed in the literature (e.g. Limber 1976, DuBois 1981, Chafe this vol­
ume, Ochs forthcoming). Chafe calls it the "light topic" and the "heavy com­
ment" constraint, with discourse-active elements appearing to the left side
(the subject side), and not yet activated elements to the right side (the object
 SVO SENTENCES IN FRENCH DISCOURSE 255

side) of the verb. A corollary of this subject-object assymetry is that the opti­
mal processing unit on the level of the proposition has the further property of
containing at most one major "piece of new information". This piece of new
information is what I have called the focus, which may be any syntactic con­
stituent except the topic. The focus can be the verb itself or, perhaps more
rarely, an adverbial phrase (two cases I have excluded from consideration in
this paper). Or it can be a lexical NP, i.e. a phrase whose principal pragmatic
function is to name a referent whose discourse role is not yet established.
The idea that the cognitively preferred number of "new elements ,, per
clause is one has also been sporadically discussed in the literature. It appears
for example in Givon's suggestion that "there exists a strategy of information
processing in language such that the amount of new information per a certain
unit of message transaction is restricted in a fashion — say "one unit per prop­
osition" (Givón 1975:202); it is expressed in Chafe's "one-recall-at-a-time
constraint" (Chafe, this volume) and in DuBois' "one noun phrase con­
straint" (DuBois 1984). The form of the preferred clause in spoken French
and its high frequency in discourse is a nice piece of evidence in favor of this
idea.
It should be emphasized that the occurrence of SV clauses in spoken
French, while being an exception to the preferred clause structure, is not an
exception to this processing constraint on the number of lexical elements per
clause. SV and VO/VS structures have in common that they contain no more
than one lexical NP in a primary grammatical relation. Furthermore, even
though the S in SV clauses is in the wrong place, so to speak, it does share the
property of low agentivity with the focus constituent in the preferred clause,
since SV sentences with agentive referents, let alone SVO sentences, are vir­
tually absent from the corpuses. Thus S constituents (whether pre-verbal or
post-verbal) share one important property with O constituents: they strongly
tend to be non-agentive. This entails that in spoken French there is a level at
which the group of Ss and Os consistently contrasts with the group of (agen­
tive) As since As do not occur as lexical NPs but as clitic t markers. It is but
a small step from here to the conclusion that there exists an ergative undercur­
rent in the syntax of spoken French. This ergative undercurrent is caused by
the powerful flow that directs the information in the clause from topic to
focus, from predictable to unpredictable, from u old" to "new".
256 KNUD L A M B R E C H T

NOTES
1) I am grateful to Claudia Brugman, Wallace Chafe, Katherine Demuth, Pamela Downing,
Martin Harris, Charles Fillmore, Tom Larsen, David Solnit, Russ Tomlin, Marie-Paule Woodley
and to members of the UCLA Discourse Group for many helpful comments on earlier versions of
this paper.
2) Cf. Limber (1976), Prince (1981) and Chafe (this volume) for English, DuBois (1984) for
Sacapultec and Ochs (forthcoming) for Samoan.
3) These numbers are quoted by François. My own count (cf. section 2.3 below) differs some­
what from hers because of certain differences of interpretation and because I have counted
repeated instances of the same lexical noun. These differences do not invalidate the general point
I am making here.
4) The concept of the preferred clause is discussed by DuBois in his work on Sacapultec dis­
course (DuBois 1981, 1984), under the name of "preferred argument structure"'. Cf. Also Ochs'
discussion of the "basic utterance type" in oral Samoan discourse (Ochs. forthcoming). Interest­
ingly, the preferred clause in French as discussed below has the same basic form as the preferred
clause in Sacapultec and Samoan, namely [V (NP)].
5) The corpuses used in this analysis are the three corpuses in phonetic transcription in vol. 2
of François (1974), as well as a corpus established by Suzanne Albert (Université de Provence,
Aix-en-Provence), I would like to thank Colette Jeanjean from the University of Aix for making
this corpus available to me. I have also used data from Barnes (1983). Examples with no indication
of the source are from my own data collection. Most data used in this paper are from spontaneous
conversations among more than two speakers.
6) This section is a modified version of Lambrecht (1984a).
7) Cf. Lambrecht (1981) for an analysis of the spoken French pronoun system as a topic-agree­
ment marking system.
8) //-deletion is much more sporadic in the subject-verb inversion construction than with
impersonal verbs. In Lambrecht (1981:271f) Ï argue that the weak status of the dummy subject il
has to do with the fact that this clitic never stands in an agreement relation with a noun. This lack
of grammatical function can then lead to total deletion of the phonetically weak subject mor­
pheme.
9) Cf. Chafe (this volume) for a psychological interpretation of the concepts given, recovera­
ble, new, for which Chafe now suggests the terms (hearer-)active, (previously) semi-active and (pre­
viously) inactive respectively. For a taxonomy of the different types of referential newness or
recoverability cf. Prince (1981a).
10) 1 am leaving open here the important question of the grammatical category to which belongs
the entire structure including T and AT, at whose core is the preferred clause. It would seem nat­
ural to call it sentence, if it is understood that 'sentence' designates here not a syntactic but a dis­
course unit. For some discussion of the problematic status of the notion 'sentence' in spoken dis­
course cf. Chafe (this volume).
11) There is a general tendency in spoken French not to place questioned constituents (QU-
constituents) in COMP position. This tendency too can be interpreted as a result of the general
preferred-structure-preserving force in the language. For a type of question formation similar to
(3b) in Sesotho, as well as for a number of Sesotho constructions that are reminiscent of those
described here for spoken French cf. Demuth (this volume).
 SVO SENTENCES IN FRENCH DISCOURSE 257

12) The rareness of clefted direct objects in English it-clefts is noticed in Prince (1978).
13) For an interpretation of have as an intransitive verb of location or existence cf. Clark 1970
and Van Oosten 1978. The semantic interpretation of the subject argument of have (as well as of
perception verbs) as a locative is also adopted by Foley and Van Valin 1984:48f.
14) As in the English existential there-construction, the original deictic meaning of the adverb
is all but lost in the ya-cleft. The deictic presentational function is expressed in French with the
voilà-construction, as in voilà Pierre 'there is Pierre'. This voilà-construction is etymologically
related to the presentational construction involving the verb voir 'to see', which is discussed below.
15) Ï am mentioning here only the simplest presentational function of the ya-cleft. Another
important function is the event-reporting function (as in ya le téléphone qui sonne! 'the phone's
ringing!') and the marking of unexpected information. Cf. example (21) and discussion.
16) A similar use of the verb 'have' is found in such (attested) English constructions as I have a
friend of mine from Chicago's gonna meet me downstairs or We had a friend of mine from Norway
was staying here, whose pragmatic function seems identical to the function of the French avoir-
construction. The main difference between the English and the French construction, besides the
presence vs absence of a '"relative" pronoun, seems to be that in English the "presented" NP must
be indefinite.
17) Presentational use of verbs of perception in spoken English discourse is noticed by Ochs-
Keenan and Schieffelin (1976:249). The authors call these verbs "locating verbs". Cf. Also foot­
note 13 above.
18) In fact, subject status of the inverted NP is debatable since the NP lacks one crucial subject
property: verb agreement. The NP agrees neither in number (exx. 15, 16) nor in gender (exx. 13,
14) with the verb. In Lambrecht (1984a) I define as one of several formal differences between topic
and focus in spoken French that topic involves verb agreement and focus does not. The formal sim­
ilarity between (what I call) presentational subjects and direct objects has been noticed e.g. by
Burzio 1981 (following Perlmutter 1978). Burzio proposes to generate such subjects in Italian
directly as objects in VP, i.e. to the right of V.
19) For the purposes of this discussion, I am ignoring autonomous topic constructions of the
type la mer, tu vois de lean 'the ocean, you see water', in which the T constituent has no coreferen-
tial t marker inside the clause. Cf. Lambrecht 1981:chapter 3.
20) For a detailed discussion of the syntactic and case-marking properties of T and AT con­
stituents cf. Lambrecht 1981, Chapter 3.
21) The difference between the high topic status of le lycée and the low topic status of les enfants
shows, incidentally, that the inherent semantic property of animacy or humanness, often invoked
as a topicality criterion (cf. e.g. Givón 1976), is not an essential factor in spoken French as far as
3p NP topics are concerned.
22) That Mademoiselle G. is now a topic is confirmed by the occurrence of the autonomous pro­
noun elle (in elle, c'est une ancienne institutrice). This elle is not a t clitic but a T pronoun of the
topic-shifting series {moi, toi, lui, elle etc.), whose members can cooccur with lexical topic NPs in
marked cases of topic-shift. (Cf. the cooccurrence of lui, Guerini and Nicolas, lui in example (29)
below.) The topic-agreement sequence is thus: [Mademoiselle G.(T)[c'(t)est...] elle(T)
[c'(t)est...] et [elle(t) m'a dit...]]. The difference between the two clitic t markers c and elle is due
to a rule in French that obligatorily converts a personal clitic to ç(a) when it is followed by an inde­
finite predicate NP; thus e.g. elle est institutrice contrasts with c'est une institutrice, both meaning
'she's a school teacher.'
258 KNUD LAMBRECHT

23) Martin Harris (p.c.) has pointed out to me that changing the article les in these two examples
into the demonstrative ces 'these/those' would have the effect of individuating the generic refe­
rents, thereby making T-coding quasi-obligatory.
24) In Japanese, the association of topic with main clause status is grammaticalized to the extent
that the topic marker wa may not appear in subordinate clauses (Pamela Downing, p . c ) .
25) Marie-Paule Woodley (p.c.) has observed that the low acceptability of (28) seems to
improve with the NP ma mère 'my mother' replacing maman. This would confirm the analysis of
maman as a "quasi-pronominal" NP presented in the next section.
26) That names and pronouns can form a single pragmatic category is observed also by Limber
(1976), who notices that only one out of thirty subjects in a child language corpus are NPs like the
baby. All other subjects are either pronouns or names. Limber observes that the pragmatic similar­
ity between names and pronouns has a syntactic correlate: the two categories have in common that
they are syntactically non-expandable (e.g. by adjectives or relative clauses). The special pragma­
tic status of proper names has also been observed by Givón (1983:10), who writes that proper
names "often constitute exceptions to the text measurements that reveal the rules which govern the
discourse distribution of topics."
27) In other languages, the presence vs. absence of a numeral associated with the noun can have
a function analogous to the function served by the contrast between T-coding and S-coding in spo­
ken French. In certain languages that have numeral classifiers, the NP that is preceded by the clas­
sifier is marked as topical for subsequent discourse (cf. Dowing (1984:ch.7) for Japanese, and
Hopper (this volume) for Malay.). In other languages, the distinction is expressed by the contrast
between presence or absence of the numeral 'one', as with Latin unus (Wehr 1984:39ff), Turkish
bir (combined with the accusative case suffix on the noun, Comrie 1981:128), and Hebrew exad
(Givón 1983:26 and fn. 14).

REFERENCES

Andreu, Flora-Klein (ed). 1983. Discourse perspectives on syntax. New

York: Academic Press.
Barnes, Betsy. 1983. "Left-dislocation in French: a multi-level model of
topic." Paper presented at the Minnesota Regional Conference on Lan­
guage and Linguistics.
. 1984. "An empirical study of the syntax and pragmatics of Left Disloca­
tions in spoken French." Paper presented at the 14th LSRL meeting, USC
1984.
Blanche-Benveniste, Claire. 1983. "Examen de la notion de subordination."
Recherches sur le Français Parlé, vol. 4, 1982. 71-115. Aix-en-Provence:
Publications Université de Provence.
Burzio, Luigi. 1981. Intransitive verbs and Italian Auxiliaries. Unpublished
MIT dissertation.
Chafe, Wallace. 1976. "Giveness, contrastiveness, definiteness, subjects,
 SVO SENTENCES IN FRENCH DISCOURSE 259

topics and point of view." In: Li 1976.

. (this volume). "Cognitive constraints on information flow."
Clark, Eve. 1970. "Locationals: A study of the relations between 'existen­
tial', 'locative' and 'possessive' constructions." Working Papers in Lan­
guage Universals, Stanford University.
Cole, Peter (ed). 1981. Radical Pragmatics. New York: Academic Press.
Comrie, Bernard. 1981. Language universals and linguistic typology.
Chicago: The University of Chicago Press.
Demuth, Katherine (this volume). "The development of pragmatic word
order in Sesotho acquisition."
Downing, Pamela. 1984. Japanese numeral classifiers: a syntactic, semantic,
and functional profile. Unpublished PhD dissertation, UC Berkeley.
DuBois, John. 1981. "Ergativity and preferred argument structure." Ms. UC
Santa Barbara.
. 1984. "Competing motivations." In: Haiman 1984.
Fillmore, Charles J. 1985. "Syntactic intrusions and the notion of Grammat­
ical Construction." BLS, vol. 11.
Fillmore, Charles, Paul Kay and Catherine O'Connor. 1983. "Let alone."
Ms. University of California, Berkeley.
Foley, William A. and Robert D. Van Valin Jr. 1984. Functional syntax and
universal grammar. Cambridge: Cambridge University Press.
François, Denise. 1974. Français parlé. Analyse des unités phoniques et sig­
nificatives d'un corpus recueilli dans la région parisienne. 2 vols. Paris:
S.E.L.A.F.
Givón, T. 1975. "Focus and the scope of assertion: some Bantu evidence."
Studies in African Linguistics. 6. 185-205.
. 1976. "Topic, pronoun and grammatical agreement." In: Li 1976.
. 1979. On understanding grammar. New York: Academic Press.
. (ed). 1979. Discourse and Syntax. Syntax and Semantics, vol. 12. New
York: Academic Press.
. 1983. "Topic continuity in discourse: An Introduction." In: Givón (ed)
1983.
. (ed). 1983. Topic continuity in discourse. Amsterdam: Benjamins.
Haiman, John (ed). 1984. Iconicity in Syntax. Typological Studies in Lan­
guage, vol. 6. Amsterdam: Benjamins.
Haiman, John and Sandra A. Thompson. 1984. ""Subordination" in Univer­
sal Grammar." BLS, vol. 10.
Hooper, Joan and Sandra A. Thompson. 1973. "On the application of Root
260 KNUD LAMBRECHT

Transformations." Linguistic Inquiry 4, 465-497.

Hopper, Paul. 1979. "Aspect and foregrounding in discourse." In: Givón
(ed) 1979.
. (this volume) "A discourse function of classifiers in Malay."
. and Sandra A. Thompson. 1980. "Transitivity in grammar and dis­
course." Language 56, 2. 251-299.
- — . and Sandra A. Thompson (eds). 1982. Studies in Transitivity. Syntax and
Semantics, vol. 15. New York: Academic Press.
Jeanjean, Colette. 1981. "L'organisation des formes sujets en français de
conversation: étude quantitative et grammaticale de deux corpus."
Recherches sur le Français Parlé, vol. 3, 1981. 99-134.
Lakoff, George. 1984a. "There-constructions: A case study in Grammatical
Construction Theory and Prototype Theory." Berkeley Cognitive Science
Report No. 18, January 1984.
— - . 1984b. "Performative subordinate clauses." BLS, vol. 10.
Lambrecht, Knud. 1981. Topic, antitopic and verb-agreement in non-stan­
dard French. Series Pragmatics and Beyond. Amsterdam: Benjamins.
1984a. "A pragmatic constraint on lexical subjects in spoken French."
CLS, vol. 20.
. 1984b. "Formulaicity, frame semantics and pragmatics in German bino­
mial expressions." Language 60, 4. 753-796.
Li, Charles, (ed). 1976. Subject and Topic, New York: Academic Press.
Limber, John. 1973. "The genesis of complex sentences." In: Moore 1973.
-----. 1976. "Unravelling competence, performance and pragmatics in the
speech of young children." Journal of Child Language, vol 3. 309-318.
Lyons, John. 1977. Semantics. Vol I. Cambridge: Cambridge University
Press.
Moore, Timothy (ed). 1973. Cognitive development and the acquisition of
syntax. New York: Academic Press.
Munro, Pamela. 1982. "On the transitivity of 'say' verbs." In: Hopper and
Thompson (eds) 1982.
Ochs, Elinor. Forthcoming. Culture and Language Acquisition: Acquiring
Communicative Competence in a Samoan Village. Cambridge: Cambridge
University Press.
Ochs Keenan, Elinor and Bambi Schieffelin. 1976. "Foregrounding refe­
rents: a reconsideration of Left-dislocation in discourse." BLS, vol. 2.
Perlmutter, David. 1978. "Impersonal passives and the unaccusative
hypothesis." BLS, vol. 4. 157-189.
 SVO SENTENCES IN FRENCH DISCOURSE 261

Prince, Ellen. 1978. "A comparison of WH-clefts and It-clefts in discourse."

Language, 54. 883-906.
. 1981a. "Toward a taxonomy of given-new information." In: Cole (ed)
1981.
. 1981b. "Topicalization, focus movement and Yiddish movement: a
pragmatic differentiation." BLS, vol. 7.
Reinhart, Tanya. 1982. "Pragmatics and linguistics: An analysis of sentence
topics." Distributed by IULC.
Rohrer, Christian (ed). 1981. Studia linguistica in honorem Eugenio Coseriu,
vol. 4, Grammatik. Berlin: De Gruyter.
Sapir, Edward. 1921. Language. New York: Harcourt, Brace and World,
Inc.
Silva-Corvalán, Carmen. 1983. "On the interaction of word order and intona­
tion: some OV constructions in Spanish. In: Andreu 1983. 117-140.
Stefanini, Jean. 1981. "Sur la notion de phrase et son histoire." Recherches
sur le Français Parlé, vol. 3, 1981. 7-18.
Stempel, W.D. 1981. "L'amour elle appelle ça-L'amour tu ne connais pas."
In Rohrer (ed) 1981.
Van Oosten, Jeanne. 1978. "Expletive-there sentences." Ms. UC Berkeley.
Wandruszka, Ulrich. 1981. "Typen romanischer Subjektinversion." In:
Rohrer (ed) 1981.
Wehr, B. 1984. Diskursstrategien im Romanischen. Tübingen: Günter Narr
Verlag.
 ON THE ROLE OF CONDITIONALS IN GODIE
PROCEDURAL DISCOURSE

LYNELL MARCHESE
United Bible Societies

Most discourse studies assume that there are grammatical units larger
than the sentence, typically called "paragraphs" (Longacre, 1979) or
"episodes" (van Dijk, 1981). Research is currently being carried out to deter­
mine what constitutes such units in various languages and how these units
are linguistically encoded. While most of the evidence for such units comes
from studies of narrative, it appears that other discourse genres have "higher
level" discourse organization as well (Hinds, 1979). In this paper I attempt to
examine procedural texts from Godié, a West African language. 1 First it is
observed that in this particular genre, an unusually high number of condi­
tional clauses occur. It is then claimed that these clauses have an important
text-organizing function, breaking the discourse into significant units.
Finally, it is suggested that the high frequency of conditionals is directly
related to the goal of this particular discourse type.

1. The data
Data in this paper come from eleven procedural texts in Godie. 'Pro­
cedural' is a term applied to a discourse genre which gives directions for car­
rying out a given task (Grimes, 1975). In English, we encounter procedurals
in how-to manuals of all types (cookbooks, car-repair manuals, etc.), as well
as each time we ask someone for directions on how to accomplish some task.
While this discourse genre is perhaps not attested in every language, 2 oral
procedurals are common place in West African society and are a standard
way of transmitting knowledge within the community.
The data here consist of 10 oral and 1 written text collected from five dif­
ferent speakers. They explain how to carry out various tasks relevant in a
Godie context:
264 LYNELL MARCHESE

Speaker3 Topic Text

1 how to plant rice A
1 how to make palm wine B
1 how to catch monkeys C
1 how Godiés stop the rain D
1 how to make kids walk E
1 what to do when someone F
drowns
1 what to do when someone G
dies
2 how to build a house H
3 women's work I
4 (written) how to plant rice J
5 how to find snails K
For this study I have also examined procedural text from three other Kru lan­
guages: Tepo Kroumen, Nyabwa, and Wobé:
6 (Tepo) how to make palm nut stew L
7 (Nyabwa) how to make fufu M
8 (Wobe) how to make attieke N
From the data, a very clear picture emerges as to what constitutes a pro­
cedural text, not only Godié, but across the language family as well.

2. Features of the Godié procedural

Like other discourse types (such as narratives, histories, folktales, etc.),
the procedural begins with an introductory statement. Eight out of eleven
texts make reference to 'our country' in the introduction as the setting
for the procedural. Speaker (3) begins her discourse on women's work in the
following way:4
(1)

our country Godié on place on NF

day if break NF you if sleep:loc leave NF

you sweep:imp your ground on and you entenimp

nàa jàlèé
your kitchem:loc
'In our Godie land, if it's morning, if you wake up, you sweep out-
 CONDITIONALS IN GODIE PROCEDURAL DISCOURSE 265

side and then you go in the kitchen.'

Sometimes a more elaborate introduction is given stating who requested the
information and who the speaker is.5 If the introduction is lengthy, greetings
(ayayoka) are inserted before beginning the real procedural. A similar strat­
egy is used in narratives.
After this introduction, the procedural begins to diverge from other dis­
course types. The most obvious ways in which procedurals differ from the
other genres are (i) the theme and (ii) overt reference to the addressee. As
noted above, the theme of a procedural centers around activities to be per­
formed in accomplishing some task. The addressee in a procedural is nor­
mally second person singular and this reference is made explicit in procedural
texts, as can be seen in example (1). In contrast, narratives, folktales, riddles,
etc., constantly refer to third person and the addressee is only rarely referred
to overtly. In the sample of eleven texts, seven made direct and constant ref­
erence to  'you (sg.)'. 6
Another feature of Godíe procedurals is the frequent use of the focus
construction. Focused elements occur in sentence-initial position where they
are optionally followed by a focus marker mu. There is a gap in the clause
where the focussed item would normally appear. As the speaker goes through
the various procedures which accompany some task, s/he may use the focus
construction to draw attention to certain noun phrases which are crucial parts
of the procedure. The following example comes from the text on house-build-
ing. (H):
(2) a.
day if break,

your machete:D FOC you FUT sharpen

c. naa
your machete :D you if-it sharpen NF day if break
d.
forest you go now PART
e.
you if forest go NF
f.
trees FOC you FUT cut
'If it's morning, you sharpen your machete. Your machete, if
you've sharpened it, if it's morning, you go to the forest. If
you've gone to the forest, you cut trees.'
266 LYNELL MARCHESE

In this example, every main clause contains a focussed item (lines b., d., and
f. In line c., dvànvv 'the machete' is topicalized; not the resumptive pronoun
v i t ' occurring in the clause. While focussing does not occur in every clause
in procedurals (compare example (1)), it is typical of this discourse genre.
Another major feature of Godíe procedurals is their irrealis viewpoint.
Procedurals contain a very high number of futures, connected so closely to
realis modes — is not a feature of procedurals. In (1) above, two conditionals
are followed by two imperfective clauses (indicating habituality). More com­
monly, future modals occur, as seen in (2) above and in the following exam­
ple:

(3) from text B (how to make palm wine)

you if -it side clean NF

you FUT now adze up take

you FUT now palnr.D dig until

and palm:d FUT fall

'If you've cleaned the side of it, then you pick up your adze and you
dig around the palm a long time and then it falls down.'
In Godié, there are two future modals, a volitive kA (or related l???k)
and a potential yi:
(4)
I VOL down lie
'I want to/will lie down'
(5)
he POT come
'He will/can/may come'
Interestingly, in procedurals, it is the volitive future which occurs over­
whelmingly in this context. In the texts studied here, 45% of all main clauses
contain while only 3.4% contain yi. One possible explanation
fork???is derived from the 'have'.
In Kru (and, indeed, in languages throughout the world), there is evi­
dence that the verb 'have' may pass through various stages of semantic shift
(Marchese, 1979):
have > must/should (have to) > will > want
 CONDITIONALS IN GODIE PROCEDURAL DISCOURSE 267

The frequent use of in procedurals could be a reflection of the older

'obligatory/should' meaning.
Another feature of procedurals is the special use of the perfect auxiliary
In narratives, may be used to slow down the action (and thus build
suspense) right before the climax of the story (Marchese, forthcoming). It is
also used in both narratives and procedurals to signal events which are out of
the natural temporal sequence (Marchese, 1978):7
(6) from text 1 (women's work)

and you put:IMP-it fire side and you cook: there

IMP

cut-up you PERF-it wash you PERF-it pot:LOC put

'...and you put it (rice) next to the fire and you cook your sauce.
The meat, you have already cut it. You have washed it. You have
put it in the pot...'
A special use of the perfect in procedural discourse is its appearance at the
end of the discourse, where it signals that the task has been completed. For
example, in text F (what to do when someone drowns), the desired result of
the procedure — retrieving the body — appears at the end of the procedure
in the perfect:
(7)
your man:D PERF up come-out
'Your man has appeared'
Similarly, in the text on women's work, the end of the procedure is signalled
by the perfect aspect:

(8) from text I

you if wash NF you PERF down be

your work PERF now finish

Tf you wash, you have sat down. Your work has finished.'
A similar shift to the perfect has been observed in Swahili (Wald, MS:22) and
in at least one speaker of Liberian English (Singler, 1984).
268 LYNELL MARCHESE

But of all the features associated with procedurals, certainly the most
striking is the use of the conditional clause. In Godié, the condition-consequ­
ence relationship is expressed by two clauses, the first of which contains the
conditional auxiliary ku and a non-final marker HA (which is sometimes
elided, leaving behind a low tone). Conditionals are typically used in Godié
to predict future events or to describe unreal or imaginative situations:
(9)
he if want NF he POT come
'If he wants, he'll come'
(10)
he if PAST money see NF he POT PAST car buy
Tf he had some money, he would buy a car.'
They also occur frequently in proverbs where they present general truths:
(11)
blind man he if blind man carry NF
kofd wa
hole they fall
Tf a blind man leads a blind man, they both fall in the ditch' (kagbo
dialect).
Conditional clauses in Godié are also used to express the occurrence of an
expected or inevitable event:
(12)
you if go NF tell him hello
Tf/when you go, tell him hello.'
Despite this semantic overlap, conditionals contrast with temporal (when)
clauses, which never contain the conditional ku and always have a 'real world'
reading:
(13)
I see-him:PFTV NF and he SEQ:me up take..
'When I found him, he took me.... '
It is the conditional with ku (and not the temporal subordinate clause)
which occurs with high frequency in procedural texts. In this discourse sample
from Godié, conditionals make up 79.6% of all initial subordinate clauses
and 25.6% of the total number of clauses:
 It is the conditional with ku (and not the temporal subordinate clause) which oc
in procedural texts. In this discourse sample from Godié, conditionals make up 79
clauses and 25.6% of the total number of clauses:
% of conditional clauses per text
speaker Text total# #main #initial #conditionals % CO
clauses clauses sub clauses (of sub c
A 71 50 21 16 76.
B 123 93 40 33 82.
C 61 35 26 20 76.
D 18 1.1 7 4 57.
E 19 12 7 5 71.
F 53 37 16 11 68.
G 32 16 6 5 83.
2 H 166 117 49 46 93.
3 I 124 76 48 46 95.
4 J 29 19 10 7 70.
5 K 14 11 3 3 100.
AVERAGE 79.

Though clearly a smaller sample, the data from the other Kru languages show
higher) scores:
6 L(Tepo) 26 14 12 12 100.
7 M(Nyabwa) 44 323 12 12 100.
8 N (Wobe) 24 19 5 4 80
270 LYNELL MARCHESE

From the tables above, it can be seen that the percentage of conditionals
in Godié procedural texts (A-K) range from 15.6% to 37%. In other Kru lan­
guages (L-N), they make up from 16.6% to 46.1% of all clauses. To show that
a high frequency of conditionals is associated with this particular discourse
genre, a count was made of conditionals in several discourse types in Godié:

Average % of conditionals per text according to discourse genre

procedurals 25.6% (10 texts)
proverbs 15.8% (200 proverbs)
riddles 3.3% (5 texts)
songs 3.3% (10 songs)
animal folktales 2.3% (8 texts)
narratives 2.0% (5 texts)
prayers 1.2% (2 texts, one lengthy)

These figures show that in long stretches of speech, conditionals are most fre­
quent in procedurals. With the exception of proverbs, conditionals account
for less than 4% of clauses in non-procedural texts. The question then arises:
what motivates the appearance of high numbers of conditionals in this genre
and what is their discourse role?

3. Role of conditional clauses

In an earlier study (Marchese, 1977), I reported that clause initial subor­
dinate clauses in Godié function as topics. To summarize the findings of that
study, it was discovered that conditionals as well as temporal adverbials and
preposed relatives share the following characteristics with simple noun topics
(Chafe, 1976; Haiman, 1978):
(i) they are only loosely connected to the main clause
(ii) they do not function as arguments of the main predicate
(iii) they are highly repetitive, containing old or assumed information
(definite)
(iv) they serve to frame the event in the following clause
These features can be seen in the examples below, where each initial subordi­
nate clause repeats information of the previous clause and 'frames' the fol­
lowing comment:
 CONDITIONALS IN GODIE PROCEDURAL DISCOURSE 271

(14) a. initial relative clause

man one certain he marry:PFTV women three

I say women:D three he marry PART NF

they gave-birth:PFTV children three

'A certain man married three women. I say, the three women
he married, they gave birth to three children/
b. initial temporal clause

and I SEQ him see

'and I found him. When I had found him, he took me...'

c. initial conditional clause (from 2 d-f)

forest you go now PART

You go to the forest. If you've gone to the forest, you cut trees.
Note that the clause 'if you go to the forest' contains known
information that sets the scene for the next procedure to be performed. Thus,
on a local level (i.e. from sentence to sentence), these clauses appear to be
linking devices (Grimes, 1975). However, eventually we will see that sen­
tence-initial conditionals play a more crucial role in discourse organization.
Taking the analysis one step further, we could try to apply the notion of
backgrounding and foregrounding to this data. Hopper (1979: 214-6) charac­
terizes backgrounding and foregrounding in narratives in the following way:
foreground background
main line events comment on narration
iconic order not in sequence
REALIS IRREALIS, more modals
main route thru lower degree of assertiveness
narrative
new information in new info can appear in subject
predicate
272 LYNELL MARCHESE

Thus foregrounded actions typically trace out the skeletion of main events in
a narrative and are usually in the perfective aspect, while backgrounded
events provide supportive material and are typically in imperfective or dura-
tive-type aspects. Hopper applies this distinction only to the narrative dis­
course type, but one is tempted to see if it can characterize the procedural dis­
course type as well. Indeed, there are some parallels. The consequence
clauses of conditionals do appear to correspond to narrative foregrounding as
described by Hopper. They provide the steps of the procedure, are typically
iconic, and if irrealis (future, imperfective, imperative), are at least more
'real' than conditional clauses containing ku. They supply new information,
and, as can be seen in subsequent examples (15e, f) may contain focussed
(sentence-initial) elements. In contrast, conditionals (i.e. antecedents con­
taining ku) appear to be backgrounded in that they do not report main line
events, they are more irrealis than any other clauses, and they have a lower
degree of assertiveness. While topics may appear in backgrounded clauses
(L5, gwΛsuu 'that medicine' in (15k), focussed elements do not occur. How­
ever, conditionals do occur in strict sequence and thus are iconic. And, as we
will see later, they in some sense "trace a main route" through the procedural
text itself. Thus the background/foreground distinction as it aplies to narra­
tive does not exactly fit procedural discourse genre.
Probably the most helpful clue to the function of conditionals in pro­
cedurals is the observation made by both Hopper (1979) and Givón (MS) that
subordinate clauses are discontinuous. Let us examine a more complete text
to see what this means. The following extract is from a procedural telling what
steps must be taken to ensure that a Godié child learns to walk:
(15) (from text E)
a. Ayayoka
greetings
b.
now really our country woman if give-birth NF
c. yí??? kv dàaboó
child:D is cloth:LOC

medicine they FUT him after do

FOC
 CONDITIONALS IN GODIE PROCEDURAL DISCOURSE 273

and medicine:D it you FUT there heart rub

FOC

and he FUT spider hit (= turn over)

and they FUT again another med. up take

and he FUT down be (sit)

there med:D he if now down-be-NOM finish NF

lí
and they FUT now there again med. there kind

up take

he FUT walk

his crawling if now also there finish...

'Greeting! Now really, if a woman from our country gives
birth, the child is placed in a (special) cloth. If he leaves the
cloth, they will put medicine on him. This medicine you will
rub on his chest and he will (learn to) turn over. If he (knows
how to) turn over, they will take another kind of medicine and
he will (learn to) sit up. If he has finished sitting up (by using
that medicine), they will get a different kind of medicine and
he will crawl. If his crawling has finished...'
From this text, it can clearly be seen that conditionals have a text-
organizing function. Notice that conditionals occur every two or three clauses
(marked by arrows). Thus, the text is broken into a set of steps outlining the
procedure. Each conditional along with its consequent seems to represent a
sequence of events which the speaker sees as closely connected. It seems
likely that these "chunks" represent some kind of cognitive unit. In fact, tak­
ing the content of the conditional clauses alone, one can construct an outline
of the basic steps in the procedure:
274 LYNELL MARCHESE

1. birth
2. leaving the daaboo (special cloth)
3. hitting the spider (turning over)
4. sitting up
5. crawling
These breaks occur where English might use 'first', 'second', 'third' for indi­
cating significant steps in a procedure. Conditionals in Godié seem to be the
oral equivalent of such breaking devices.
Further evidence for the breaking effect of conditionals is the presence
of the marker mà (lines (b) and (d)) and the discontinuity of subject NP's. Mà
has been recognized as serving as a paragraph marker (Gratrix, 1978: 313).
While it does not occur with every conditional in the data base, it often co-
occurs with them. Perhaps more convincing evidence comes from the data
concerning subject discontinuity. Well over half (57.5%) of all conditional
clauses examined in this study precede a change in subject — a confirmation
that conditionals are signalled some kind of natural break. Notice, for exam­
ple, that the subject of the conditional in (b) is rjwlc 'woman', while the sub­
ject of the main clause switches to yd 'child' (line (c)). 'Child' is taken up
again as the subject of the next conditional (line (d)), but again the subject of
the main clause switches to wa 'they' in (e). These switches have been rep­
resented in the text by diagonal arrows Thus, as noted earlier, conditional
clauses are entirely given. This may be a point of difference between narra­
tives and procedures, since, according to Hopper, backgrounded clauses in
narratives tend to tolerate new information in pre-verbal position.
The actual size of these units varies and may be dependent on several fac­
tors. Counting ten of eleven texts, 1 x the longest unit is 8 clauses long (exclud­
ing the conditional), though such a long string is rare. More often, there is
only one clause following the conditional, and this clause is immediately fol­
lowed by another unit (see example (2)). The average number of clauses in a
unit is 2.53. As noted above, each clause following the conditional seems to
represent an event. Turning back to example (3):

(3) (from how to make palm wine)

you if-it side clean NF

you FUT now adze up take

 CONDITIONALS IN GODIE PROCEDURAL DISCOURSE 275

you FUT now palm:D dig until

and palm:D FUT fall

If you've cleaned the side of it, then you pick up your adze and you
dig around the palm a long time and then it falls down.
We see that there are three steps in the wine-making process after the side of
the palm is cleaned:
(i) you pick up the adze
(ii) you dig around the palm
(iii) the palm falls down
It is difficult to predict the number of clauses which will occur in each unit. It
seems likely that the length of the unit is dependent on how difficult the pro­
cedure is to process or how much the speaker gauges the hearer to already
know. As Hinds (1979: 148) notes in his study of Japanese procedurals:
The decision for how a procedural discourse should be organized is largely
an individual matter. There are often a number of different ways a process
could be completed successfully, so there may be several extremely different
"correct" organizational patterns. It is there impossible to predict with any
degree of certainty how an individual...will segment a recipe; but, once the
individual has constructed the discourse, it is possible to determine what he
considers to be the discrete steps, [emphasis, L.M.]
similarly in Godié, the discourse reveals two things:
(i) the events the speaker sees as crucial to carrying out the task (en­
coded in each clause)
(ii) the grouping of events which the speaker chooses to impose on an
otherwise long list of procedural steps (encoded in each "chunk"
following a conditional break)
Many questions could be raised as to the factors contributing to (i) and (ii).
For example, one could ask why the speaker in (3) spells out 'you take up an
adze', rather than simply saying 'you dig around the tree with an adze'. From
this text, it would appear that the speaker is restricted the amount of new
information per clause to a verb and one 'new' object complement {take adze,
dig palm fall). While an interesting hypothesis, more study needs to be done
to determine the actual limits of new information per clause in this discourse
context.
276 LYNELL MARCHESE

4. Frequency of conditional clauses

Now that it has been determined that conditionals play a role in text-
organization in procedural text, it is important to understand why these con­
ditionals are so frequent in this discourse type. It is claimed here that the fre­
quency of conditionals can be directly attributed to the discourse goal of
teaching someone a procedure. It stands to reason that the smaller the
chunks, the easier it will be for the hearer to identify and remember the pro­
cesses involved. Thus, conditionals are more frequent in procedural dis­
course because we are dealing with a process which must be remembered and
carried out. It must be noted that in narrative texts in Godié, sentence-initial
temporal clauses perform a similar role in discourse — dividing the text into
"episodes" of some sort. In narrative text, however, there do not need to be
as many divisions and the units do not have to be so small, since the goal is
merely to process the information coming to the hearer. A likely hypothesis
is that cross-linguistically units in procedurals will be smaller than in narra­
tive-type discourse. This seems to be true in English, where procedurals have
short, quick steps:
(12) (from a Datsun repair manual)
When changing tires, carefully take the following steps:
1. Park on a level surface and set parking brake firmly. Set manual
transmission in reverse (automatic transmission in "P")
2. If parked on or near road, activate hazard warning flasher.
3. Remove the spare tire and tools from the storage compartment
4. Place wheel chock at both the front and back of the wheel diagon­
ally opposite the jack position
5. Place the jack under the jack-up point illustrated....
In attempting to transmit a procedure orally, Godié speakers lay out the steps
similarly, but they do so through the use of conditional clauses.
Along with this, conditional clauses are an excellent pedagogical device.
They are repetitive, a standard method of ensuring that the hearer will
remember the message. Furthermore, they offer an opportunity for a rather
unique speaker-hearer interaction. As noted in Marchese, 1977, conditionals
clauses, like other subordinate clauses, are followed by a non-final marker
HA, Here the speaker pauses and quite often, it is at this point that the hearer
acknowledges that he is following the message. He does so by making a verbal
response:
 C O N D I T I O N A L S IN G O D I E P R O C E D U R A L D I S C O U R S E 277

(13) SPEAKER:
you if bush go NF
HEARER:
This verbal response can occur in other discourse types — for example, in
narratives, but the frequency of conditionals in procedurals means that this
interchange is more frequent in this context. Thus, the speaker-hearer
relationship is much tighter in this discourse genre than in any other discourse
type. The "teacher" is continually checking up to see if his "student" is follow­
ing him (see Wald, MS: 11, 12 for a discussion of a similar phenomenon in
Swahili).

5. Conclusion
The role of conditional clauses in Godié procedural texts is clear: they
break the text into manageable units and facilitate the learning process. Judg­
ing from the procedural text from three other Kru languages, conditionals are
used in exactly the same way throughout the Kru family. In fact, this may be
a general African strategy, since others report hearing similar phenomena in
other parts of West and East Africa (C. Lord, P. Bennett, M. Lewis, p.c.).
And if it is indeed true that languages generally use subordinate clauses as
markers of discontinuity, many more languages may use conditionals in a way
similar to Godié. 12 What is interesting in the Godié case is, not only do condi­
tionals signal a breaking point in the discourse, but they also seem to have
been "singled out", and in a sense "grammaticized, emerging as the most dis­
tinctive feature of a specific discourse genre, the procedural text.

NOTES
*I would like to thank Sandy Thompson and Tucker Childs for their helpful comments on an ear­
lier version of this paper, as well as Doris Payne and members of my advanced syntax class for dis­
cussing various points with me.
1) Godié is a Kru language spoken in southwest Ivory Coast by approximately 20,000 people.
Kru is part of the Niger-Kordofanian language family which covers most of sub-Saharan Africa.
2) Both W. Chafe and D. Payne have noted (p.c.) that the procedural genre does not seem to
exist in some Amer-Indian languages.
3) Speaker 1 is Zadi Sassi Michel, 29 at the time the texts were recorded. Speaker 2 is "Grego-
rire", about the same age. Speaker 3 is "Marguerite", a young woman about 24 years old at the
time of the recording. Speaker 4 (the writer) is Dago Jean-Claude, from an eastern Godié dialect.
278 LYNELL MARCHESE

Speaker 5 is Nyapi Djawli Matthieu, approximately 12 years old at the time of the recording. With
the exception of the text by Dago Jean-Claude, all the above texts were collected by Carol Gratrix-
Brinneman and myself during 1972-1975 in the village of Dakpadou, Ivory Coast. The remaining
texts in Tep, Wobe, and Nyabwa were supplied to be by P. Thalmann, I. Enger, and J. Bentinck.
Unfortunately names of these speakers are not available.
4) In this and subsequent example, the following abbreviations are used:
D definite NOM nominalizer
FOC focus PART particle
FUT future PERF perfect
IMP imperfective PFTV perfective
LOC locative POT potential future
NF non-final SEQ sequential
VOL volitive future
Godié has three tones : high ( ' ), mid (unmarked), and low (???). While Godié does not have contras-
tive vowels, two vowels are written when a vowel carries two tones. The symbol 6 designates an
implosive bilabial stop, g, a voiced velar fricative.
5) There appears to be a strong tendency in Godié discourse to justify "holding the floor". Pro­
cedurals and stories of all types are very often prefaced with lengthy justification which may
include (i) a summary of the request for information (ii) the identity of the person making the
request (iii) the identity of the speaker (iv) reason why the speaker is qualified to answer the
request (cf. Walk, MS: 16).
6) The use of A 'you' seems to depend on whether the speaker can see himself in the agentive
role in the procedural or not. For example, speaker 1 uses 'you' when describing male-oriented
tasks, whether his audience is male or female. He switches to third person when he would not be
actively involved as in text extract (15). In the sample of a written procedural, the author uses only
third person. This may be due to the fact that the author was in the capital, Abidjan, describing
how villagers plant rice, and was thus divorced from the situation. We will have to wait for Godié
written literature to develop before we can tell if a switch to third person will be a feature of written
style.
7) Note that what is called 'perfect' here was labelled 'perfective' in Marchese (1978). Since
that time I have changed labels to come more in line with traditional terminology, with perfective
designating an event viewed in its entirety and perfect referring to a past event with current rele­
vance (Marchese, 1979).
8) Conditions follow consequence clauses only rarely when the relationship is entirely logical
(i.e. non-temporal) (Marchese, 1976).
9) Conversational material was not considered in this study. I have restricted my data to more
structured 'monologue'-type material.
10) The case of proverbs is quite interesting. Usually proverbs are 3-4 clauses long. Often times
they begin with a conditional. It may be significant that both procedural texts and proverbs can be
considered teaching tools.
11 ) One text was not analyzed because there were several complex complements and it was not
immediately clear to me how these should be counted.
12) English sometimes uses conditionals in a similar way. For example, in a five page essay by
J. Wright, "TV commercials that move the merchandise", a conditional clause "breaks" the essay
into two essential parts. Wright argues that there are two basic commercial types: music-related
 C O N D I T I O N A L S IN G O D I E P R O C E D U R A L D I S C O U R S E 279

and celebrity-related. As he shifts in mid-essay from one subject to the other, we find paragraph
initially:
If good music can actually get consumers to sing about an advertiser's products, the
use of a star or celebrity helps to catch the viewer's attention.... (italics, L.M.)
In this essay, the conditional presents old information, serving as an introduction to a new topic. It
thus parallels the Godié examples, except that the units it creates are much larger. Of course, this
discourse use of conditionals is much rarer in English than it is in Godié.

REFERENCES

Chafe, W. 1976. "Givenness, contrastiveness, definiteness, subjects and

topics" in Subject and Topic, edited by C. Li. New York: Academic Press.
Givón, T. ed. 1979. Discourse and Syntax, Syntax and Semantics, Vol. 12.
New York: Academic Press.
. MS. 'Topic continuity in discourse: the functional domain of switch ref­
erence".
Gratrix, Carol. 1978. "Godié narrative" in Papers on Discourse, J.E.
Grimes, ed., 311-323.
. 1978. Papers on Discourse. SIL/University of Texas at Arlington.
Hinds, J. 1979. Organizational Patterns in Discourse. In T. Givón, ed. Syntax
and Semantics, vol. 12.
Hopper, P. 1979. "Aspect and Foregrounding in Discourse". In T. Givón,
ed. Syntax and Semantics, vol. 12.
Li and Thompson. 1976. "Subject and topics: a new typology of language."
In C. Li, ed. Subject and Topic (457-489).
Longacre, R. 1979. "The paragraph as a grammatical unit". In T. Givón, ed.
Syntax and Semantics, vol. 12.
Marchese, L. 1976. Subordination in Godié, UCLA MA thesis, published in
French (1978) Subordination en Godié ILA/SIL serie conjointe, no. 4.
. 1978. "Time reference in Godié", in Papers on Discourse edited by J.
Grimes. SIL. Dallas. (63-75).
1977. "Subordinate clauses as topics in Godié", Studies in African Lin­
guistics. Supplement 7.
. 1979. Tense/aspect and the development of auxiliaries in the Kru lan­
guage family. UCLA PhD dissertation.
. forthcoming. "Apercus sur un conte folklorique Godié" to appear in
Traditions Orales, ILA, Abidjan, Simone Ehivet, ed.
Tannen, D. (ed.) 1981. Analyzing Discourse: Text and Talk. Georgetown
280 LYNELL MARCHESE

University Round Table on Language and Linguistics.

Thompson, S. MS. "Grammar and Written Discourse: Initial vs. Final Pur­
pose Clauses", to appear in Text.
Singler, J. 1984. Variation in Tense-Aspect-Modality in Liberian English.
UCLA PhD. dissertation.
van Dijk, 1981. "Episodes as unit of discourse analysis". In D. Tannen, ed.
Analyzing Discourse: Text and Talk.
Wald, B. MS. "On the Discourse Unit", to appear in L. Polanyi, ed. Structure
of Discourse.
Wright, John W. 1979. "TV Commercials that move the merchandise," The
Commercial Connection: Advertising and the American Mass Media. Dell
Publishing Co.
 IS BASIC WORD ORDER UNIVERSAL?

MARIANNE MITHUN
State University of New York Albany

A fundamental assumption underlying much current work in syntactic

typology is that all languages have some basic, syntactically defined, con­
stituent order. It is generally recognized that this order may be altered some­
what for pragmatic purposes, but the basic order is considered a primary
characteristic, from which other features of the language can be predicted. It
is questionable, however, whether all languages actually have such a basic
order. Discourse data from a number of genetically and areally distinct lan­
guages indicate that syntactic ordering and pragmatic reordering processes
may not in fact be universal. In what follows, it will be shown that forcing such
languages into the mold of any basic word order at all is at best descriptively
unnecessary, and at worst an obstacle to the discovery of interesting univer­
sals.
For many languages, the basic constituent order seems so obvious that
criteria for determining it need not be specified. In languages where one
order predominates statistically, and any rarer alternative orders are highly
marked pragmatically, several reasonable criteria converge to indicate the
same choice. For other languages, however, these criteria lead to conflicting
results. Sometimes the choice is the statistically most frequent order (Haw­
kins 1983, Dryer 1983). Sometimes it is whatever order permits the simplest
overally syntactic description (McCawley 1970). Sometimes, it is the order
accompanied by the least morphological marking (Hawkins 1983). Finally, it
may be the least pragmatically marked, or neutral order, although identifying
pragmatically neutral sentences is itself problematic. In some of these cases,
discourse-intial sentences have been considered the most neutral because
they presuppose no preceding context (Pullum 1977: 266). In others, the pre­
ferred order for potentially ambiguous clauses has been judged the most neu­
tral (Chomsky 1965: 127). In still others, 'simple, declarative, active clauses
282 MARIANNE MITHUN

with no complex verb or noun phrases' are assumed to exhibit neutral order
(Chomsky 1957:107, Greenberg 1966: 74, Pullum 1981), There have been a
few observations on the interrelationships of these criteria. Givón (1979: 87-
8) hypothesizes that the statistically most frequent order is actually that found
in main, declarative, affirmative, active clauses, and, furthermore, that this
is also the least presuppositional. Hawkins, in his ambitious study of word
order universals, states that of his three criteria (textual frequency, frequency
within the grammatical pattern, and grammatical markedness) simple textual
frequency usually provides a sufficiently sensitive basis for the identification
of basic order (1983: 13-14). For many languages, however, grammarians
note that nearly all logically possible constituent orders appear with sufficient
regularity in main clauses to render identification of even a 'preferred order,
difficult.
The isolation of pragmatic effects on word order in various languages has
also been somewhat problematic, due in part to the well known diversity of
terminology and its usage in the analysis of discourse. Linguists associated
with the Prague School have traditionally described pragmatic ordering in
terms of the concepts 'theme' and 'rheme'. Mathesius (1939) defined the
theme as 'that which is known or at least obvious in the given situation, and
from which the speaker proceeds' in his discourse (cited in Firbas 1964: 268).
Firbas (1964, 1972) sought to refine the notion of theme in terms of 'commu­
nicative dynamism1. 'By the degree of communicative dynamism carried by a
linguistic element, I understand the extent to which the element contributes
to the development of the communication, to which, as it were, it "pushes the
communication forward"' (1972: 78). That element carrying the lowest
degree of communicative dynamism is called the 'theme', that carrying the
highest, the 'rheme'. The Prague School linguists and others (see, for exam­
ple, Firbas 1964: 270, Greenberg 1966: 100, Lyons 1977: 508, Givón 1979:
296) have remarked that the normal pragmatic ordering of constituents seems
to be theme-rheme, or topic-comment. In Firbas' terms, the basic distribu­
tion of communicative dynamism is a gradual increase in degree from the
beginning of the sentence to the end. Lyons notes that 'Not surprisingly there
is a very high correlation, not only in English, but in all languages, between
occupying initial position in the utterance and being thematic, rather than
rhematic ... To many scholars it has seemed natural that the cognitive point
of departure and the communicative point of departure should coincide.'
(1977: 507-8)
A number of facts indicate that the interaction between basic word order
 IS BASIC WORD ORDER UNIVERSAL? 283

and pragmatic theme-rheme reordering principles is not constant from one

language to the next. Following work initiated by Mathesius (1928),
Thompson (1978) pointed out that languages can vary in the relative effects
of syntactic, semantic, and pragmatic considerations on surface word order.
In languages like English, the syntactic roles of constituents are the primary
determinants of word order, while in languages like Russian and Czech, prag­
matic considerations may have a stronger effect. When pragmatic factors do
play a role, furthermore, it is not clear that all languages follow the 'natural'
progression from theme to rheme. Tomlin and Rhodes (1979: 307) point out
that 'in Ojibwa, a VOS Algonquian language, the unmarked distribution of
thematic information is reversed from the language-general tendency for
thematic information to come earlier in a sentence or clause.' Givón (1938:
145) reports that Ute, 'a mature SOV language with a high degree of prag­
matically-controlled word-order flexibility', shows comment-topic order.
Similarly, Biblical Hebrew 'is rigidly VO but shows a pragmatically-control­
led VS/SV variation' (Givón 1983: 28), also with the order comment-topic.
In the sections below, ordering principles will be examined in some of
the languages that have posed problems for grammarians and syntactic typ-
logists. It will be shown that for languages of this type, the assumption of any
syntactically defined word order is unmotivated and misleading, as is the
assumption of theme, rheme pragmatic reordering. It will also be shown that
the existence of such languages has important consequences both for the
reliability of standard strategies used in detecting basic word order, and for
word order typologies.

1. Word Order in Some Perplexing Cases

In this section, the ordering of constituents will be examined in Cayuga,
Ngandi, and Coos. Although unrelated geneticaly and areally, the three lan­
guages show surprising parallelisms in their surface constituent orders.

1.1. The General Character of the Languages

All three of these languages are highly polysynthetic with obligatory pro­
nominal affixes within all verbs. Cayuga is an Iroquoian language spoken in
Ontario. Each verb contains obligatory pronominal prefixes referring essen­
tially to its agent and/or patient. Thus a verb like -atekany- 'bite' contains an
agent pronoun, while a verb like -nqhqktani 'be sick' contains a patient pro­
noun. A verb like -kqhek- 'hit' contains a transitive pronoun, referring to a
284 MARIANNE MITHUN

particular combination of agent and patient. Since no more than two argu­
ments can be referenced in a verb, agent pronouns are used not just for pro­
totypical agents, but for the most agentlike of the primary arguments. Thus
in verbs like -ke- 'see' or -nohwe'- 'like', which do not involve agents, the
agentive pronouns refer to the seer and the liker.
Because of their pronominal prefixes, single verbs can stand alone as
predications in themselves, and often do.*
(1)
he/her-like-HABITUAL
'He likes her/
One or both arguments of the verb may be further identified by a separate
nominal. The pronominal prefixes remain unchanged.
(2)
John he/her-like-HABITUAL
'John likes her.1
(3)
Mary he/her-like-HABITUAL
He likes Mary.'
Interestingly, although speakers agree that it is perfectly grammatical to
include both a separate agent/subject noun phrase and a patient/object noun
phrase within a single sentence, such full sentences occur relatively rarely in
spontaneous discourse. It is easy enough to elicit them from bilingual speak­
ers, however. When asked, speakers agree that all logically possible con­
stituent orders are grammatical: SOV, SVO, VSO, VOS, OSV, and VSO.
Ngandi, an Australian aboriginal language of eastern Arnhem Land, is
also polysynthetic with obligatory pronominal prefixes within the verb. As in
Cayuga, the prefixes remain in the verb whether separate noun phrases
further identify the subject or object or not. All Ngandi data here come from
texts recorded by Heath (1978).
Ngandi (Heath 1978: 192)
(4) Barma-ma-ŋi.
3Pl/MA-get-PAST CONTINUOUS
They used to get it.'
(5)
MA-water lily fruit-ABS 3Pl/MA-get-PAST CONTINUOUS
They used to get water lily fruit.'
 IS BASIC WORD ORDER UNIVERSAL? 285

(6) Barma-ma-rŋ
3Pl/MA-get-PAST CONTINUOUS Pl-woman-ERG ATIVE-
ABS
T h e women used to get it.'
Ngandi nouns are classified into noun classes. The noun referring to the water
lily fruit above is of the MA class, and accordingly carries a prefix -ma-. The
noun referring to the women, of the BA class, carries a prefix -ba-.
Case relations are marked on nouns by suffixes, so that the noun
'women', referring to the subject of the transitive verb 'get', carries an erga-
tive suffix (Objects of transitives and subjects of intransitives carry zero
suffixes. The suffixes glossed as ABS by Heath are not case markers.) While
nominal case marking is ergative, pronominal reference within verbs is
accusative.
As in Cayuga, clauses containing both a separate subject and a separate
object are possible, but rare. When they do occur, any order is possible.
Coos, an Oregon language, is also polysynthetic with pronominal
affixes. Subjects of intransitive verbs are referenced by 'loosely prefixed' pro­
nominal clitics on the verb. Subjects of transitive verbs are referenced by
combinations of the proclitics and transitive verbal suffixes. All data here
come from the Hanis Coos texts recorded by Frachtenberg in 1909 and pub­
lished in 1913 and 1922.
Coos (Frachtenberg (1922:351, 425)

you-take-I
T take you along'

look at-TRANSITIVE the maggot

'he looked at the maggots'

DISCRIMINATIVE-maggot eat-TRANSITIVE
'maggots ate him'
(Third person pronominal affixes appear as zero.) Subjects of transitive
verbs, like the maggots above, carry the prefix x-, termed 'discriminative' by
Frachtenberg. Thus case marking within both nouns and verbs is essentially
ergative.
As in Cayuga and Ngandi, clauses containing both a separate subject and
286 MARIANNE MITHUN

a separate object are relatively rare, and word order is variable. As Frachten-
berg noted, T h e syntactic structure of the Coos sentence is very simple, and
is characterized by the facility with which the different parts of speech may
shift their position without changing in the least the meaning of the sentence.'
(1922:319)

1.2. Strategies for Determining Basic Word Order

Since all word orders occur in all of these languages, the identification of
a basic order is not as straightforward as in many other languages. Several
strategies are possible.

1.2.1. Statistical Frequency

As noted above, a common diagnostic of basic order is statistical fre­
quency (Dryer 1983). Whichever order appears the most often might be con­
sidered basic. The crucial constructions, however, clauses containing both
separate subject and separate object noun phrases, are relatively rare in
spontaneous discourse in all of the languages. Sample counts of clauses in
texts indicate that in Cayuga, perhaps 1% - 2% of all clauses contain three
major constituents, in Ngandi, approximately 2%, and in Coos, 2% - 3 % .
Yet even among these small sets of clauses, all orders are represented. Since
such constructions appear so rarely, strong statistical evidence for any order
is simply lacking.

1.2.2. The Ambiguity Test

Another frequently cited diagnostic is the word order preferred in poten­
tially ambiguous sentences (Chomsky 1965). A Cayuga speaker was pre­
sented with the sentences below, constructed from shorter sentences he had
produced in a narrative.
(10) (SOV/OSV)
Buffalo Six Nations they beat them
(VSO/VOS)
they beat them Buffalo Six Nations
(SVO/OSV)
Buffalo they beat them Six Nations
He remarked that they were all good Cayuga sentences, but that in all cases,
it was unclear who beat whom. He could not choose a preferred reading.
Other similar sentences produced the same reaction. The overall rarity of
 IS BASIC WORD ORDER UNIVERSAL? 287

such full sentences suggests that their very appearance is a marked phenome­
non , so that none of them should be considered reliable models of unmarked
constituent structure. (In the following discussions, all subjects will be label­
led S, whether they are agents or patients, ergatives or absolutives, since all
follow the same principles, and this facilitates comparison among languages.
Direct objects will be labelled O, and predicates V.)

1.2.3. Relative Order within Pairs

Predicates do appear frequently with a single nominal constituent. This
fact suggests that in order to discover the basic constituent order of Cayuga,
one could determine the orders found between pairs of constituents, then
combine the results. (If SV and VO, then SVO.) This is of course analagous
to the usual strategy for determining the relative order of morphemes in a lan­
guage when all of them could not possibly cooccur in a single word.
Separate subject nominals are somewhat rare in Cayuga discourse,
appearing in approximately 12% of the clauses. (All examples cited below
were taken from spontaneous narratives or conversation unless otherwise
specified.) Sometimes subjects appear before their predicates. The sentence
below is from a discussion of how severe the winter is likely to be. The Whites
have predicted a hard winter, but the speaker is not convinced.
SV S V

it is just most fishworm still contr dwell

There are still plenty of fish worms around.
S

birds also still not contr

The birds haven't left yet either/
V

this they have not gone

Subjects can also appear after their predicates. The sentence below comes
from a description of how to hunt rabbits.
VS V S

not they don't know today young men

T h e young men of today do not seem to know how to do it.'
288 MARIANNE MITHUN

Direct object nominals often appear before their predicates.

O V

it is the this the rabbit used to we hunted

when I was a child

'This is how we used to hunt rabbits when I was a kid.'
Objects also appear after their predicates,
V

it is usually the long ago they sell usually

'A long time ago, they used to sell rabbits.'
O

the rabbit
Time adverbials can occur on either side of the predicate.
T V

night this you know I say they stir themselves

'As I said, at night, they walk around.'
V

then usually contr there body hangs

Then usually the body will be hanging there
T

very early that

in the morning.'
Locative constituents can appear on either side of the predicate.
L V
(17)
in the bush there you will go
'You will go out in the bush.'
 IS BASIC WORD ORDER UNIVERSAL? 289

V L

they are stirring just it is always in the water

They are always moving around in the water.'
Manner adverbs can also precede or follow verbs.
Adv V

slow as so you will be able you wall walk

'Walk as slowly as you can.'
V Adv

you will come around slow

'You should come around carefully.'
Nearly any word order is possible. As can be seen from the examples
above, even the predicates have no constant position.
Ngandi shows the same variation. Separate subjects occur in only about
10% of Ngandi clauses, but they can precede or follow their predicates,
whether these are transitive or not.
(Heath 1978:206) S V

then-ABS GU-water-ABS GU-now-go through

'Then water passes through'
(Heath 1978:206) V S
gu-jark-yuŋ
then-ABS GU-now-throw-AUG-PR GU-water-ABS

this way
'Then the water rushes through.'
Objects also appear either before or after their predicates.
(Heath 1978: 199)
O V

NI-honey-ABS lPlEx:NI-now-go out-CAUS-PR properly

'We take out the honey entirely.
(We then put it entirely into the what's-it, the cooliman.)
290 MARIANNE MITHUN

V O

lPlEx/NI-put in-PAST CONT NI-honey-ABS

We used to put the honey in.'
Oblique nominals appear anywhere as well. Instrumentais appear either
early or late in the clause.
(Heath 1978:197) I V

MA-string-INST-ABS lPlEx/A-now-attach
'We attach it (the spearhead) with string.'
(Heath 1978:241) V I

3Pl/3Pl-now-poison-AUG-PCON maybe MA-poison-INST

'Maybe someone poisoned them, with magical poison.'
Datives appear both early and late.
(Heath 1978: 212)
Dat V

NI-bee-DAT lPlEx/NI-BEN-look
'We look for bees.'
(Heath 1978:213) V Dat

longtime IPlEx-now-go-PRES GU-egg-DAT

'We go for a long time looking for eggs.'
Allative noun phrases, indicating locative goal, can appear early or late.
(Heath 1978:197) Allative V

A-shovelspear-ABS MA-spearshaft-ALL lPlEx/A-put on-PR

'We put the shovel spearhead onto the spear shaft.'
V Allative

also A-wirespear-ABS lPlEx/A-put on-PR MA-shaft-ALL

We also put the wire spear prongs onto their spear shaft.'
As in Cayuga, all constituent orders appear. Even the predicate can occupy
almost any position within the clause.
 IS BASIC WORD ORDER UNIVERSAL? 291

Constituent ordering in Coos is as variable as it is in Cayuga and Ngandi.

Subjects can appear either early or late in their clauses, whether transitive or
intransitive.
(Frachtenberg 1922: 426)
S L V

snake his at thighs it wraps around him

T h e snake coiled around his thigh.
L V S

his to waist that it arrived the snake

It crawled up to his waist.'
Objects can also appear either early or late.
(Frachtenberg 1913: 7)
O V
(31) TE kā'wil hanL îs yō'qat.
that basket shall we two split it in two
'Let us split this basket in two!'
(Frachtenberg 1913: 6)
V O

they two saw it that land

They saw the land.1
The same mobility is typical of oblique nominals. Compare the position of the
locative nominals in the sentence below.
L V

now ocean beach on that they two put it down.

They put it on the sand beach.'
(Frachtenberg 1913: 18)

he saw it shore close to

'He saw (different kinds of food lying) along the beach.'
No basic word order emerges from the examination of pairs of con­
stituents. All alternatives occur. The obvious question raised by sets of sen-
292 MARIANNE MITHUN

tences like these is just what function the alternatives serve. Surely speakers
would not randomly vary a feature as salient as word order. Since order does
not signal the syntactic or semantic roles of constituents, perhaps it has a
pragmatic function.

1.3. Definiteness
As noted earlier, pragmatic ordering has most often been described as a
tendency for thematic information, or topics, to appear before rhematic
information, or comments. Unfortunately, themes, or topics, have been
defined in a variety of ways. One characteristic often associated with themes
is givenness, or predictability, in contrast to the newness or unpredictability
of rhemes. Recall that Mathesius defined the theme as 'that which is known
or at least obvious in te given situation, and from which the speaker proceeds'
(1939: 234, cited in Firbas 1964: 268). Since definite nominals refer to entities
that the speaker assumes the hearer can identify, either from general know­
ledge or specific context, a comparison of the positions of definite and indefi­
nite noun phrases might provide a key to the identification of pragmatic fac­
tors in word order. Li and Thompson pointed out, for example, that the ten­
dency in Mandarin and Russian 'to place indefinite nouns after the verb and
definite nouns before the verb seems to be a manifestation of a general and
widespread tendency among languages to put known information near the
beginning of the sentence and new information near the end of the sentence'
(1976: 172).
None of the three languages described here marks definiteness
obligatorily, but definiteness can be specified by means of certain nominaliz-
ing particles and demonstratives. Cayuga has a particle ne' that can optionally
precede definite nominals, including proper and possessed nouns. When this
definite particle is inserted into manufactured sentences containing both sep­
arate subject and object nominals, a relationship between definiteness and
word order can be detected. The particle cannot grammatically precede a
nominal early in the sentence below.
(*ne'S-V-0)
the John he likes her Mary
(*ne'0-V-S)
the Mary he likes her John.
(*S-ne'O-V)
John the Mary he likes her
 IS BASIC WORD ORDER UNIVERSAL? 293

d. *Mary ne John shakónçhwe's. (*0-ne'S-V)

Mary the John he likes her
It can, however, precede a nominal near the end of a sentence, and is often
added when full sentences are repeated by speakers. All of the (elicited) sen­
tences below mean 'John loves Mary'.
(36) a. John Shako hwés ne' Mary. S-V-ne'O)
John he likes her the Mary
b. Mary shako hwes ne John. (O-V-ne'S)
Mary he likes her the John
c. Shako hwes John ne Mary (V-S-ne'O)
he likes her John the Mary
d. Shako hwes Mary ne John. (V-O-ne'S)
he likes her Mary the John
e. Shako hwes ne' Mary né John. (V-ne'O-ne'S)
he likes her the Mary the John
This ordering of indefinite before definite is the reverse of that found in
Chinese, Russian, Czech, and the more familiar Indo-European languages
for which pragmatic ordering has been described. It should be emphasized
that the sentences cited above were elicited, not culled from narratives or
conversation. A survey of spontaneous discourse does indicate that indefinite
nominals tend to appear near the beginning of their clauses, while definite
nominals tend to appear near the end.
Indef V
(37) Katsihwa kihsa:s.
hammer I seek
'I am looking for a hammer/ (said in a hardware store)
V Def
To: ti' nika:no:' nę:kyę katsihwa'?
how then so it costs this hammer
'How much does this hammer cost?'
The indefinite-definite order appears to be characteristic of Ngandi dis­
course as well. Note the translation of (22) above. When the water is first
mentioned, it appears early in the clause and is translated with an indefinite
noun. At the second mention, in (23), it appears late, and is translated with
a definite noun. In the passage below, the narrator is describing how
Aborigines used to get yams and roast and peel them. When the yams are
294 MARIANNE MITHUN

introduced, they are indefinite, and appear early. When the skin first appears
in the second line, it is identifiable from the preceding context, so it appears
late. In the third line, it is identifiable from previous mention, so it appears
late again.
(Heath 1978: 210-211)
Indef V
(38) buluki? ma-jalma barma-ma-ni, ...
also MA-yamsp 3Pl/MA-get-PRESENT
'They get round yams (and roast them)
V Def
ma-ja-bolk- u-ni, ma-gula? -nu ayi-yun,
MA-now-appear-AUG-PRES MA-skin-its-ABS
Their skin comes off.
V Def
barma-geyk, barma-geyk ma-gula?-yuŋ guniň,
3Pl/MA-throw 3Pl/MA-throw MA-skin-ABS that's all
They throw the skin away, and that is that.'
The indefinite order also appears characteristic of Coos. The identity of
the child in the second line below is inferrable from the preceding line; as a
definite nominal, it appears late. The person mentioned in the last line is new
on the scene, indefinite, and clause-initial.
(Frachtenberg 1913: 11)
(39) Má lau mîtsiilti'ye.
nevertheless that (he) pregnant became
'Nevertheless he became pregnant.
V Def
Ta lau qano'tca l'nuwît lE ā'la.
and that outside to (he) pulls the child
The child was all the time trying to come out,
V Def
Hats īn qantc lau L!ēitc la ā'la.
just not way that go out his child
but could not do it.
Indef V
Tsō ma lilt.
now person (he) sent (it)
So they sent someone (to the north).'
 IS BASIC WORD ORDER UNIVERSAL? 295

1.4. Old versus New Information

Overt definite marking accounts for a significant proportion of con­
stituent ordering in all three languages, yet it does not account for all of it. A
principle must be found that correlates highly with definiteness, but explains
the remaining cases as well. Consider the following Cayuga passage. A dinner
guest was asked whether he liked baked potatoes. He replied that yes, he
thought he probably did.
(40) Ne:1 ky thréhs i:n kyę:'ę 'k:k.
it is I guess too far I guess I will eat
'It's just that I eat them so seldom.
Def NEW V
Ne:' tsh : ne' oa' wista the' ní:' ťe:ke:s.
it is only the peelings not I do I eat
I just don't eat the skins.
V Def Old
Kwiskwís kyeę:' hne:' tsh : ka:ti:s ne' oá'wista'.
pig just contr just they eat the peelings
Only pigs eat the skins.'
In the second line, the skins appear before the verb, although they are defi­
nite, inferrable and thus identifiable from the previous mention of the
potatoes. The skins represent newer information than the verb 'eat', which
had just appeared in the preceding sentence, however. In the third line, the
skins, now old information, appear late, while the pigs, completely new infor­
mation, appear first.
The same is true in Ngandi. In the passage below, the wood wulčum
spears are generic. When first introduced, they appear early in the clause.
After that, they appear late.
(Heath 1978: 187)
New V
(41) gu-wulčum balaka ňaru-ga-?-yaw-^ u-ni,
GU-wood spear before lPlEx/3MaSg-Sub-DUR-spear-AUG-P
V Old
a-jeň-uŋ bara-ga-yaw- u-ni, gu-wulčum- .
A-fish-ABS 3Pl/A-Sub-spear-AUG-P GU-wood spear-INST
They used to spear fish with wulčum spears.'
The same is true of Coos. New information tends to precede old informa-
296 MARIANNE MITHUN

tion, as can be seen in the following passage. Both noun phrases referring to
the mat are definite, but the first time the mat occurs, it appears at the begin­
ning of its clause. The second time, it is at the end.
(Frachtenberg 1913: 7)
New V
(43) TE tc!î 'cil yuL îs yo 'qat...
that mattingwe two split it
'"Let us split this mat."
(They did so, and went down to examine the earth. The earth was
still not solid, even ...)
V Old
î lau tci  hî'outs hE tc!î'cîl.
when that there they two put it down the matting
after they had put down the mat.'
The principle of new information before old predicts the order of con­
stituents in a large majority of clauses. It is not surprising that this should cor­
relate so often with the indefinite/definite distinction, since new enities are
most often definite. There are still some constructions that cannot be
explained purely in terms of a preference for new before old, however.

1.5. Importance
In some cases, both constituents are equally given or equally new. A
Cayuga speaker telephoned his friend to announce that he had lost his wallet.
New V New 
(44) Ni: ke: thóne:' okahtó:' ne' akétkw'ęta'.
just here there Ilostit the my wallet
'Mind you, I lost my wallet.'
Both constituents are completely new, neither present nor alluded to in pre­
vious discourse (there was none) nor in extralinguistic context (also absent,
since this was a telephone conversation). In this case, the new verb precedes
the new object. Yet consider the utterance below. A man has just told his
friend that he cut his foot with an axe. His friend, horrified, asks if he is badly
hurt. He replies no, not really,
New  New V
(45) thréhs kyę:'ó to:kęhs wahtahkwatę:s tewakę:so:.
because just-suppose really thick shoes I wear
'I guess because I had really thick shoes on.'
 IS BASIC WORD ORDER UNIVERSAL? 297

Again, both the verb 'wear' and the object 'thick shoes' are completely new,
neither mentioned in previous discourse, nor referred to subsequently. This
time, however, the new object precedes the new verb.
A father, trying to make his daughter hurry in the morning, said,
New V New S
(46) 0:nę kokhwáihse: sanó:ha'.
now she has finished the food your mother
'Your mother has already finished cooking breakfast.'
Both elements are equally new, neither mentioned or alluded to in previous
discourse, neither within view of the speaker or hearer, but equally identifi­
able. Neither is referred to again in subsequent discourse. The new predicate
precedes the new subject.
The speaker cited below was describing his misadventures in the woods.
No saplings or bushes had been mentioned previously, nor of course any
grabbing, or even the coat. The saplings are not totally unexpected, since the
setting is the woods, but the grabbing is also not totally unexpected, since the
speaker's clothing is torn. Neither is mentioned subsequently. The order is
new subject before new predicate in this case.
New S
(47) Shę nyó:' n'ato:ta:ké:' thó hne:' oho:ta
so far on the way there contr sapling
'Along the way, bushes
New V
taka:ye:nâ:' akaky'ataw'ithrâ'keh.
it grabbed me on my coat.
caught on my coat.'
In each of the sentences above, the constituent conveying the principal
information of the utterance appears first. The most important part of (44)
the loss, of (45) the thickness of the shoes, of (46) the finished state of the
breakfast, and (47), the bushes. Since new information is usually more impor­
tant than old information, the principle of new before old usually accounts for
constituent order.
The importance principle works equally well in Ngandi. In the passage
below, the narrator is describing how Aborigines used to fish with a hook and
line, in addition to the spears. (The verb root -woyk- is translated as 'angling,
fishing with line and hook', but contains neither the noun root -jen- 'fish' nor
the verb root -ma- 'get'.)
298 MARIANNE MITHUN

(Heath 1978: 198)

(48) buluki? -yuŋ ñar-ga-woyk,
also-ABS lPlEx-SUB-angle
'We also go angling.
ñar-ja-woyk- u-ni a-jara- u,
lPlEx-now-angle-AUG-PRESENT A-what's it?-INST
We go angling with what's it?
o-monarŋa-ku-yiñurŋ ñar-ja-bak-woyk
A-White-GEN-RELlPlEx-now-BEN-angle
the thing belonging to Whites, we go angling then.
Old V Old 
ñar-ga-woyk- u-ni ñara-ga-ma-ni a-jeñ-urŋ,
lPlEx-SUB-angle-AUG-PR lPlEx/A-SUB-get-PR A-fish-ABS
We go angling and catch fish/
In the last line, both the verb 'geť and the object 'fish' are inferrable old infor­
mation, since the entire discussion has been about fishing. (Neither has been
mentioned lexically.) Note that here, the old predicate precedes the old
object. This is because the point of the discussion is the getting, the fact that
when they use a hook and line, they still do catch fish.
The passage below is from a description of plum gathering.
(Heath 1978: 195)
(49) ma-mala-galič-uŋ narma-ņu-ni,
MA-group-other-ABS lPlEx/Ma-eat-PAST CONT
'Some (of the plums) we ate (as they were),
New V
ma-mala-galič-uŋ ñarma-gul?- u-ŋi
MA-group-other-ABS lPlEx/MA-pound-AUG-PC
New Loc
gu-jundu-gi
GU-stone-LOC
others we pounded on a stone (so that they became soft).'
Both the predicate and the locative nominal are completely new. The locative
follows the predicate. Compare the passage below.
(Heath 1978: 211)
New Loc New V
(50) a- andiya?-gi barma-ja-yo-ŋana,
A-mat-LOC 3Pl/MA-now-put in-PRESENT
'We put (that food) on mats.'
 IS BASIC WORD ORDER UNIVERSAL? 299

Here again, both the locative nominal and the predicate are completely new.
No previous mention had been made of mats or putting. In this case, how­
ever, the locative precedes the verb. The reason is clear. In (49), the pound­
ing is more important than the rock while in (50), the mats convey the most
important information of the clause.
The same principle can be seen in the Coos texts. The sentence below is
near the beginning of a narrative. There has been no previous mention of any
body of water nor of dryness of wetness.
(Frachtenberg 1913: 14)
New V New S
(51) In tclle 'xEm tE la'nik:.
not dry that there river
There was no low tide.'
Here, the new predicate precedes the new subject. Compare the sentence
below, however. Again, both the subject and predicate are new. This time,
however, the new subject precedes the new predicate.
(Frachtenberg 1913:9)
New S New V
(52) Haqa 'ti laā 'ya łtce isītc le ' nhü 'wis L!tā.
tracks go to it beach on their ready land
('Suddenly they saw tracks on the ocean beach.')
The main point of the clause in (51) is the lack of dryness (because Crow had
no chance to get food), while the most important constituent in (52) is the
tracks. Sentences like this one suggests a particular way in which constituents
may be considered important in discourse.

1.6. Topic Shift

The Cayuga passage below comes from the cosmology legend. A woman
has fallen from the heavens through a hole in the sky, and as she falls, she
wonders what will become of her. Suddenly she notices something.
(53) Ne' nę:ę ne' nę eyatokyé'
the this the then her body is flying along
'And then, as she was falling,
nękwa a' hna:tó:k tho:kyeh ne:' n ne:'
and then she noticed it there it is you know
she noticed
300 MARIANNE MITHUN

ne:' tsi't' ęsh : ' o h , ha'tekatiy'atá:ke:

it is birds many bodies
birds, all different kinds of them,
tho katikyęno:kyé's teyakotík hne:'
there they are flying around they are looking at her
flying around there, looking at her,
shęnh : nę:kyę eyé'tokye'.
where this her body is flying
as she fell.
S
Neť ne' o:nę nę:kyę tsi't'ęsh
it is the now this birds
Now at this time, these birds
V
teyotiya' towéhto atkatiya'to:wéht ne' n ne:'
they are thinking they thought the you know
were thinking that
nę:ky a:kenaťeny :te ...
this they should try
they should really try (to lessen her misfortune).'
Note the early position of the underlined tsťťęshoťo 'birds' in the second
sentence. This noun is neither new information, since the birds were just
mentioned in the previous clause, nor necessary for disambiguation, since the
zoic plural pronoun in the verb 'they are thinking' clearly refers to the birds
and not to the woman. It is prominent for another reason. It represents a new
topic, a new point of view. The text continues to describe the birds' decisions
and their resulting attempts to save the poor woman. A shift in topic can thus
be considered sufficiently important to appear early in a sentence.
Not all new subjects appear clause-initially. Not all new subjects intro­
duce new topics. A man and his wife had left a tape recorder running for a
long time as they conversed about a wide range of different things, including
what was scheduled at the longhouse that evening, who might be putting on
a supper and how it would be done, a neighbor who was to get her fortune
told, and when the husband was planning to return from his weekend trip. At
that point a car was heard outside. The wife said:
V S
(54) Kwé: sakáeyo', thó:kyeh.
well they (F) arrived again that
 IS BASIC WORD ORDER UNIVERSAL? 301

'Well, they're back.'

The husband answered:
V S
(55) 0:n ki key:' sakáey ', kashehawáhksh '.
now just then they returned your daughters
'Yes, your daughters have returned.'
The daughters had not been mentioned at all up to this point. The significant
part of the message here was the return, rather than the daughters, however,
because it meant that the conversation was over. The daughters were not
introduced as new topics.
The same importance of new topics can be seen in the Ngandi texts. The
passage below is part of the narrator's reminiscences about his experiences as
a police tracker. He and two others had captured a criminal and taken him to
the government office, where they sat waiting. Finally,
(Heath 1978: 250)
New Topic S V
(56) ni-Ted Ervin-uŋ ni-yimi-ñ-? "..."
MaSg-Ted Ervin-ABS 3MaSg-say-PPUNC-0
'Ted Ervin (a high-ranking government official) said, "...'"
Ted Ervin, a new topic, appears at the beginning of the clause.
The function of topics within narratives can often be seen by comparing
the contexts of different constructions. The first sentence below is part of a
description of the various kinds of foods Aborigines used to collect. The nar­
rator has just mentioned that they used to roast and eat euros and antelopine
kangaroos.
(Heath 1978: 192)
New Topic  V S
(57) ma- atam-yuŋ, barma-ga-ma-ŋi ba- iŋ?-yuŋ,
MA-water lily fruit-ABS 3Pl/MA-SUB-get-PC Pl-woman-ABS
'The women used to get water lily fruits (seed pods).'
Both the water lily fruit and the women are new information here, neither
previously mentioned, nor inferrable. This discussion is about types of food,
however, so the water lily fruit appears first. The narrator continues discus­
sing the food: 'We ate that food, we ate vegetable food'.
In a different narrative, the same speaker mentioned the same custom.
This time, he had said, 'Then we get up and leave. We are going now for veg-
302 MARIANNE MITHUN

etable food instead of meat, for water lily root eorms, fruits, and stems — we
eat that.'
(Heath 1978: 210)
New Topic S V
(58) ba- in? yanaci ba-ja-wulup, ba-ja- iŋ?-gu
Pl-woman all along 3Pl-now-bathe 3Pl-now-woman-DAT
T h e women go into the water, (that work) is for women.
manga? ma-guyk barma-ma-ni, ...
maybe MA-water lily species 3Pl/Ma-get-PRESENT
Maybe they get guyk, ...'
Both the women and going into the water are completely new here, as before,
but this time the women appear first in the clause instead of last. This passage
is about women and women's work. The women remain the subject of the
next several clauses.
The same prominence of new topics can be seen in the Coos texts. The
narrative cited below begins with a description of Crow and his habits. Then
a new character is introduced.
(Frachtenberg 1913: 15)
New Topic S V
(59) Xyî'xei dä'mil lau ha'lqait.
one man (to) that one (he) came to him
'Once a man came to Crow,
Lau xwändj īilt
that one that way (he) told it to him
and said, "..."'
The narrative continues with the man's suggestions.

1.7. Contrast
Important contrasts are not limited to new topics. Any constituent repre­
senting a focus of contrast is generally considered sufficiently important to
occur early in the clause, whether it is indefinite or definite, new or old, a
topic or not. These constituents represent a focus of contrast.
The following (elicited) Cayuga sentences illustrate the positions of con­
trasting constituents.
 IS BASIC WORD ORDER UNIVERSAL? 303

V S
(60) The' ťa:ke:kas ohya, kehswahéhs ní:'.
not do I like it fruit I hate it I
T don't like fruit, I hate it.'
S V
(61) Thę' ní:' ťa:ke:kas ohya , Péte hne:' hó:kďs.
not I do I like it fruit Pete contr he likes it
'I don't like fruit, Pete does.'
The use of the separate pronoun ní:' T, myself in the last sentence above
is interesting. Languages with pronominal affixes always contain separate
pronouns as well, although they appear much less frequently than in lan­
guages without the affixes. Grammarians of such languages often note that
the separate prefixes seem to appear near the beginning of clauses unusually
often. This is no accident. In polysynthetic languages, separate pronouns
have a special function; they generally indicate special emphasis or contrast.
The contrastive force of the independent pronouns can be seen in the Cayuga
sentence below. The speaker was provided with a context and asked to trans­
late the English sentence 'I'm the one who broke it.'
(62) I:' atkriht.
I I broke it
T m the one who broke it.'
Pronominal contrasts are not automatically the most important elements of
their clauses, as can be seen in (60) above. They may also be used to set up a
double contrast. Offered a platter of chicken, the dinner guest said:
Cayuga
(63) Ohsi:ná ki ni:' ę:ke:k.
leg just I I will eat it
'I'll have a leg, myself.'
This reply contrasts two entities: the drumstick as opposed to other pieces,
and the speaker as opposed to other diners. Here, the drumstick contrast was
more important than the diner.
The other languages exhibit similar ordering or contrastive information.
In each Ngandi clause below, the initial constituent represents the focus of
some contrast.
304 MARIANNE MITHUN

Ngandi (Heath 1978: 201)

(64) gu- awal-?ñirayi-gi-yurj ma-gami-bugi? ñar-ga-jal- u-ŋi,
GU-country-our-LOC-ABS MA-spear-only lPlEx-SUB-hunt kg
'In our country we used to hunt kangaroos with spears only.
gu-ni-?-yuŋ gu~ awal-yuŋ ba-wan-gu,
GU-this-0-ABS GU-country-ABS P1-PRO-GEN
This country belongs to someone else.
ñer-yuŋ gu--? ŋuri,
we-ABS GU-that-O north
As for us, (we were) there to the north.'
Before the Coos sentence below, the narrator described a bargain
suggested by Crow. Crow wants to exchange his lightning for the evening low
tide. The bargain is accepted, and Crow obtains the tide.
Coos (Frachtenberg 1913: 18)
(65) Halt! ä'ka h lo'waku L!āa.
now he the lightning has as booty
'while the other man came into possession of the lightning'
The contrastive pronoun xa 'ka 'he' occurs at the beginning of the clause. The
second focus of contrast, the lightning, follows. Any kind of constituent, pro­
nominal, nominal, or verbal, can be the focus of a contrast. In all cases the
ordering is the same. If the contrast is the main point of the predication, and
thus the most important, the focus of the contrast will appear initially.

1.8. The Determination of Word Order in Cayuga, Ngandi, and Coos

1.8.1. The Newsworthiness Principle
Word order in these languages is thus based on pragmatic considera­
tions, on the relative newsworthiness of the constituents to the discourse. An
element may be newsworthy because it represents significant new informa­
tion, because it introduces a new topic, or because it points out a significant
contrast.
A test for the 'most newsworthy first' principle is provided by questions
and answers. Presumably in normal conversation, the most important con­
stituent of an answer is that which corresponds to the interrogative word of
the question. In Cayuga, this word appears initially, whether it be a subject,
object, time, location, or anything else. (All of the questions and answers
below were spontaneous, not elicited.)
 IS BASIC WORD ORDER UNIVERSAL? 305

SV
(66) Q. S : ęsne:'?
who you two will go
'Who are you going with?'
A. Sámęya:khne:'.Samwe two will go
T m going with Sam.'

OV
(67) Q. Tę h 'te' a:yę:' ihse: a:shni:n '?
what it seems you think you would buy
'What do you think you'd like to buy?'
A. O:, ahya'tawi'thrá ki a:y '  hsa:s a:khní:n '.
Oh dress just seems I seek I would buy
'Well, I am just looking for a dress.'
TV
(68) Q. To: ti' n'aonishe tho hekae's?
how then so it lasted there they two were there
'So how long were they there?'
A. Tekhní: akya tatoehthé' konáhteky :.
two weeks they two were away
'They were away for two weeks.'
LV
(69) Q. Kqe ti' hęswe:'?
where then you all will go there
'Where are you all going then?'
A. Ó:, othow'eké hęya:kwe:\
Oh at the cold we will go there
'Oh, we'll go up north.'
QV
(70) Q. To: ti' nika:n :'?
How then so it costs
'So how much does it cost then?'
A. Kéi n'ate'wenya'.w'é sikwa:ti:há nikam :'.
four so hundreds a bit beyond so it costs
'It costs a little more than four hundred dollars.'
Answers to alternative questions yield the same evidence. The constituent
306 MARIANNE MITHUN

that provides the most important information, the one whose information
answers the question, appears first.
SV
(71) Q. Atis htháę' k h, John, Mary k'ishęh?
you two talked ? . John Mary or
'Did you talk to John or Mary?'
A. Máry ki akyakh htháę'.
Mary just we two talked
'I talked to Mary.'
OV
(72) Q. Oti:,  h, kha:fí nik ęhsnék ha?
tea ? coffee either you will drink
'Will you have tea or coffee?'
A. Otí: ękhnék ha'.
tea I will drink
' I ' l l have tea.'
Answers to yes-no questions provide the same evidence again.
LV
(73) Q.  h wak  h ha'ká tahk?
in boat ? they went there
'Did they go by boat?'
A. Th Tekat hné hakáęhtahk.
no in it flies they went there
'No. They went by airplane.'
PossV
(74) Q. I:s kęh satshe:n thó:kyę só:wa:s?
you ? your pet that dog
'Is that your dog?'
A. Eh í:' ake:tshé:nę'.
yes I my pet
'Yes, it's mine.'
The same ordering characterizes Ngandi answers. The constituent which
answers the question appears first.
 IS BASIC WORD ORDER UNIVERSAL? 307

(Heath 1978: 250)

SV
(75) Q. ni-nja rain? ni-ni-?-yuŋ?
MaSg-who Q MaSg-this-O-ABS
'Who is this?'
A. Ni-wacinbuy ni-na-ri-yuŋ
MaSg-Wacinbuy MaSg-that-IMM-ABS
That is Wacinbuy.'
Coos questions and answers show the same pattern.
(Frachtenberg 1913: 10)
SV
(76) Q. Extcī'tcü mä?
thou what sort person (are)
'Who are you?'
A. Nîloxqai'nis ma il.
I medicine person (am) surely
'I am a medicine man.'
This newsworthy-first principle appears to be the same as that first
described by Firbas in terms of communicative dynamism ('the degree to
which a sentence element contributes to the development of the communica­
tion'), but in reverse. Do Cayuga, Ngandi, and Coos exhibit essentially the
same type of ordering as a language like Czech, but backwards? If they do,
and if it is assumed that a progression from theme to rheme is inherently more
natural cognitively, are Cayuga, Ngandi, and Coos somehow less natural or
logical?

1.8.2. The Naturalness Issue

Themes, or topics, as noted earlier, have been variously defined as the
elements carrying the lowest degree of communicative dynamism or oldest
information, as the starting point of an utterance (Mathesius 1939), and as the
focus of the speaker's empathy (Kuno 1976). Themes establish an orientation
and a perspective, so they typically apear first in a sentence (Halliday 1967).
If themes do indeed provide such a point of departure, how can Cayuga,
Ngandi, and Coos speakers leave them until the end?
An examination of discourse shows that these speakers do not save
orienting material until the end any more than English speakers do. Narra-
308 MARIANNE  HUN

tives typically open with an establishment fo the general topic of discussion.

This is usually sufficiently significant to fill an entire sentence or intonation
unit. (See Chafe 1980.) Other orienting devices, such as time and perhaps
location, are set early as well. The following passages open narratives.
Cayuga
(77) Ta: ahi:' ne.' a:kathro:wí'
now I thought this I would tell
'Now I thought I would talk about
shę niyoht hné:' ne' swé;'keh.
how it used to be the long ago
how things used to be a long time ago.'
Ngandi (Heath 1978: 229)
(78) walkundu-yun naki?
Walkundu-ABS there
'There at Walkundu (a place south of the Roper River),
walkundu baru-ga-maka-na,
Walkundu 3Pl/Bu-SUB-call-PRESENT
they call that place Walkundu,
i- u u-ŋi i-yul-yu:::ŋ
3MaSg-go-PASTCONT MaSg-man-ABS
a man was going along there/
Coos (Frachtenberg 1922: 419)
(79) Ûx sla 'tcinī.
they two cousins (were) mutually
'Once upon a time there were two cousins.'
If new themes appear early, what of the most common themes, those
already established and present in the mind of the speaker? Speakers typi­
cally establish a topic and stay with it for a certain length of time. In the
absence of counterindications, hearers normally expect the topic to remain
constant. Since it is expected, a continuing topic need not occupy a prominent
position in the clause. Reference to it within the pronominal prefixes on the
verb confirms its continuation without unduly distracting the hearer. The
hearer is not actually waiting in suspense until the verb appears with its pro-
nonimal markers, since a topic shift would be signalled early in the clause.
Word order in Cayuga, Ngandi, and Coos, is thus not simply a mirror
image of that in a language like Czech. The overall principle is somewhat
similar, in that items are arranged according to their newsworthiness, but the
 IS BASIC WORD ORDER UNIVERSAL? 309

mirror image model is inappropriate in two ways. For one, elements that
establish a significant orientation for the first time, whether it be the point of
view of the topic, the time, the location, or the reliability of the statement,
occur early, just as they do in Czech. For another, items that signal the con­
tinuation of such orientation, such as an unchanged topic, time, or location,
often do not appear as separate constituents at all, but rather as bound
affixes.

1.8.3. The Markedness of Pragmatic Ordering

These are not the only differences between languages like Cayuga,
Ngandi, and Coos, and those like Czech. As mentioned above, members of
the Prague School and, more recently, Thompson (1978), have noted that
languages can vary considerably in the extent to which surface word order is
controlled by syntactic or pragmatic considerations. In languages like Eng­
lish, order is determined primarily by the syntactic functions of constituents.
In languages like Czech, their pragmatic functions may play a greater role.
Cayuga, Ngandi, and Coos, appear closer to Czech in this respect.
They are not the same, however. Recall that when presented with sen­
tences containing alternative word orders, Cayuga speakers will not even
choose a preferred order out of context. Marta Roth informs me that Czech
speakers, on the other hand, are very conscious of 'normal' word order.
Alternate orders apparently do occur more frequently in Czech than in Eng­
lish, especially in written language. Yet when presented with these orders out
of context, speakers are strongly aware of their marked status.
Asked to translate transitive sentences like 'Daniel quickly drank the
milk' out of context, this Czech speaker consistently supplied SVO versions
like that below.
S V O
(80) Daniel rychle vypil mléko.
Daniel quickly drank milk
'Daniel quickly drank the milk.'
(Vanessa Flashner informs me that among her spoken Polish texts, approxi­
mately 87% of the clauses show SVO or VO order.)
Asked for an appropriate full answer to subject questions like 'Who
drank the milk?', the Czech speaker simply added heavy stress to the subject,
rather than altering word order. (As in English, one-word answers like
'Daniel' are perfectly appropriate. Transitive sentences with full subject and
310 MARIANNE MITHUN

object noun phrases are probably no more frequent in natural Czech dis­
course than in English.) Verb-initial orders were interpreted as questions,
and object-initial orders as fragments of relative clauses. When presented
with the alternative order SOV, the speaker agreed that this was grammati­
cal, but would require some obvious reason for the added emphasis on the
verb and adverb, such as the added clause below.
S O V V
(81) Daniel mléko rychle vypil a odesel.
Daniel milk quickly drank and left
'Daniel quickly drank up the milk and left.'
These responses do not contradict the work of the Prague School lin­
guists. Firbas states, for example, 'Even in Czech, of course, the possibility of
freely changing the order of words is limited' (1964:278 note 17). Pragmatic
considerations can apparently enter into surface word ordering in Czech
more freely than in English, probably because of the case suffixes on nouns.
It is still a very different process from that operating in languages like Cayuga,
Ngandi, and Coos, however. Pragmatic reordering in Czech results in rela­
tively marked structures sometimes described as 'archaic' or 'overly literary',
whereas Cayuga speakers seem less inclined to find any order more marked
than the others.
Does this mean that languages like Cayuga, Ngandi, and Coos have no
mechanisms for highlighting unusual pragmatic situations? Not surprisingly,
they all do have constructions exactly for this purpose, and these construc­
tions are used somewhat more frequently than such devices as clefing and
pseudo clefting or topicalization in English.
In Ngandi, such constructions involve the prefix -ga-, which sometimes
functions as a weak subordinator, although it occurs freely on main verbs.
Heath (1978:122-3) describes the construction as follows. 'The usual way to
focus a constituent is to put it at the beginning of the clause, followed by a sub­
ordinated verb [with] -ga-. There appear to be no significant restrictions on
the type of constituent which may be focussed in this way, and examples are
attested of NP's in virtually all surface cases (except perhaps the Genitive)
and of various kinds of adverbs occurring in focused position. Ngandi focus
constructions may be literally translated with English topicalized or cleft sen­
tences, but it should be emphasized that the Ngandi constructions are much
more common than these English types.' He then provides examples of
focused constituents of all types. Here are a few.
 IS BASIC WORD ORDER UNIVERSAL? 311

(82) ņi-Conklin, yaya, ņi-jambularŋa, ñar-ga- i -i.

MaSg-Conklin I MaSg-Wallace lPlEx-SUB-go-PPUN
'Conklin, I, and Wallace were the ones who went.1
(83) jipa? gunukuwic ñar-ga-ñawk- u-ŋ.
later tomorrow lPlEx-SUB-speak-AUG-FUT
Tomorrow is when we will talk.'
Coos shows similar constructions.
(Frachtenberg 1913:17)
(84) Tsäyä'nautc wîx:i'lîs lElau qaL!axex:ī'we.
small (pl) in food that is it begins to flop
the manner of the one back and forth.
'All kinds of food (fishes) began to flop around.'
Cayuga also makes frequent use of such devices.
(85) Ha'te:y ki' ase:'sh. h,
all kinds just new ones
'All kinds of vegetables,
né:' thó:kyę teyéhsnye'.
it is that one cares for it
that is what they're growing.1
Cayuga, Ngandi, and Coos, have devices for accomplishing the same syntac­
tic and pragmatic functions as languages like English and Czech, but these
devices are distributed differently over various areas of the grammar. One
result of this is a radical difference in the degree of markedness of alternative
word orders.

2. Standard Strategies for Detecting Basic Order and Pragmatically Based

Languages
As noted earlier, the usual criteria for establishing the basic word order
of a language include statistical frequency, descriptive simplicity, and prag­
matic neutrality. What do these criteria indicate when there is no arbitrarily
defined basic order, that is, when all ordering is the result of pragmatic con­
siderations?

2.1. Statistical Frequency

Does the fact that word order is pragmatically based mean that all possi­
ble orders appear with equal frequency? In fact it does not. On the relatively
312 MARIANNE MITHUN

rare occasions when a single clause contains both a separate subject nominal
and an object nominal, the order OVS appears slightly more often than the
other logical possibilities. Does this mean that OVS should be considered the
basic order after all? Establishing a fundamental order on the basis of a slight
statistical advantage in a comparatively rare construction seems unnecessary,
unless it can provide some significant descriptive or typological advantage. In
fact, the assumption of an arbitrary basis could cut off fruitful exploration
prematurely. The inequalities in occurring orders in languages like Cayuga,
Ngandi, and Coos, reflect interesting facts about the actual workings of lan­
guage. The fact that subjects appear near the end of clauses more often than
at the beginning in a pragmatically based system indicates that subjects are
typically the least newsworthy. This finding is not unrelated to Givón's statis­
tical studies of definiteness. As he notes, 'in human language in context, the
subject is overwhelmingly definite' (1979: 51). The fact that separate objects
more often appear near the beginning of clauses indicates that objects are
more often used to convey newsworthy information. Although Givón found
that in general, direct objects are roughly 50% indefinite and 50% definite in
English texts, 'the 50% indefinites are the bulk of the indefinite nouns in the
text ... The accusative or direct object position is thus the major avenue for
introducing new referential arguments into discourse, at least in English.'
(1979:52).

2.2. Descriptive Simplicity

A major justification for assuming the existence of an arbitrary, syntacti­
cally determined constituent order in the description of a language would be
its power as a descriptive device. For languages with relatively rigid, syntacti­
cally defined surface word order, the establishment of this order at the outset
has obvious utility. The description of rarer, morphologically and pragmati­
cally marked alternative orders as the result of the movement of constituents
out of their normal position is mechanically simple. For languages like
Cayuga, Ngandi, and Coos, however, it is not at all clear that arbitrarily
selecting one order as basic, then scrambling this order most of the time, is
simple, revealing, or realistic. Since alternative orders are unaccompanied by
additional morphological material, there is also no formal motivation for one
choice over another.
 IS BASIC WORD ORDER UNIVERSAL? 313

2.3. Pragmatic Neutrality

A third criterion for determining the basic constituent order of a lan­
guage is the selection of the least pragmatically marked order, that order
which presupposes the least. As noted above, it has been suggested that the
most pragmatically neutral sentences of all must be discourse-initial, since
there is no preceding linguistic context to establish information. Pullum
(1977:266) remarks, 'where a discourse environment could not be present,
i.e. discourse-initially,,.. the basic order would be expected.' In fact, as men­
tioned above, the beginnings of narratives generally represent a highly
marked situation. Topics must be established before anything else can be
attended to. Confining a study of word order to initial utterances would limit
the investigation to a highly specialized corpus of utterances with a relatively
unusual function, that of establishing referents and point of view.
Most discourse does not even open with 'main, declarative, affirmative,
active clauses'. Conversations usually begin with questions. In fact, of 30
recorded Cayuga conversations, 21 began with questions. (Greetings were
not counted as initial sentences.) Another two opened with commands.
The relatively rare declarative conversation openers provide little indi­
cation of basic constituent order, since they consist almost solely of verbs and
particles.
(86) Ák :kę' s :té' syąthwáhs h.
I saw you last night you were planting
T saw you planting last night.'
(87) Kwé:, tętwa'ęnáę' ak ęyó:hę'.
Well, we're playing snowsnake it seems tomorrow
'Well, it seems we're playing snowsnake tomorrow.'
(88) A:y s'atreht hétk'ęse:'.
it seems your car got bad on you
'You seem to be having car trouble.'
Narrative openers present the same problem. The two sentences below
open long narratives, the first highly formal, the second more informal.
(89) Eke:ka:t :' shę niky hwętsahky
I will tell it how so the earth originated
ne:' tshihwa' hé tsha' hwętsá:tęh.
it is when first when the earth began
314 MARIANNE MITHUN

(90) Né:' ki' kyę: thó:kyę ne' a:sanits' ta:tó:wa:t

it is just this that the you would hunt fish
'This is how you hunt fish.'
All of these opening sentences are characterized by their small number
of major constituents. This is due in part to the high productivity of noun
incorporation in Cayuga. (See Mithun 1984). Most of the verbs in the exam­
ples above contain an incorporated noun stem ('pole', ' c a r ' ,'earth', "fish").
These nouns are incorporated for a reason. Cayuga speakers normally intro­
duce one new concept at a time into discourse, not unlike other speakers. The
single verb represents a single, complex but unified concept.
As can be seen in (78) and (79) above, narrative openings in Ngandi and
Coos generally illustrate the same characteristic of few major constituents,
because their function is to introduce one new idea at a time.
The one-idea-at-a-time tendency is also reflected in the so-called 'after­
thought constructions' which appear frequently throughout texts in all of
these languages. Often when a full nominal appears, it is not an integral part
of the intonation unit containing its predicate. During the description of how
to hunt rabbits, for example, the Cayuga speaker said:
(91) Omítate:no' s'ekyę:', ne' kw'ay '.
they have roads you know the rabbit
'They have roads, you know, rabbits/
The intonation break between s'ekyę:' 'you know' and ne' kwa'y ' 'the rab­
bit(s)' is more a change in pitch than a pause. Final nominals like the rabbits
above are pronounced with significantly lower pitch and often somewhat sof­
ter volume than the preceding constituents. Such nominals are not literally
'afterthoughts', in the sense that the speaker simply forgot to mention them
earlier. They are provided as insurance that the hearer will be able to keep
reference straight. Often they repeat a referent which has not been men­
tioned for a while, like the rabbits above. In other cases, they are used to
clarify the identify of a referent, as below. The verb 'to fish' below does not
contain any overt mention of 'fish', so when the speaker said 'that is what we
will live on', he felt it necessary to clarify what 'that' referred to.
(92) Ęy kwatahny :k. Né: ' ki' tsh : kwáhs
we will put hook in water it is just only all
'We will fish. That is mainly what
 IS BASIC WORD ORDER UNIVERSAL? 315

ęyakyonhéhk h :k, ots 'ta kanyo'sh

we will live on it fish wild animals
we will live on, fish, game, too, I guess,
ne kyę:' hni t ho'te ' ęyakwatsh :i'.
the I guess too what we will find
whatever we find.'
Such constructions are not exactly equivalent to regular subject or object
noun phrases in languages like English, in that they are not as tightly bound
to their predicates intonationally. Because of the pronominal prefixes, they
are never necessary for grammaticality. Their lower and softer pronuncation
mirrors their function as backgrounded appositives.
Ngandi and Coos also exhibit such constructions frequently. Heath
(1978: 53) mentions 'the 'afterthought' construction so common to languages
in this area, where a core nuclear clause pronounced and then one or more
constituents giving more precise specification of arguments in the clause are
added after a pause.' During his description of life in the old days, a Ngandi
speaker was discussing various tools. He said:
(Heath 1978: 190)
(93) a-jeler bara-ma-ŋi, ba-jawu?-jawulpa- u-yuŋ,
A-stone axe 3Pl/A-get-PCONT Pl-RDP-old man-ERG-ABS
T h e old people used to get stone axes.'
The final noun phrase serves simply to remind us of the continuing topic.
The Coos narrator cited below had just mentioned that all kinds of food
began to flop around. Hearing the noise, Crow decided to open his eyes, but
someone yelled at him. After a while, Crow was finally permitted to open his
eyes.
(Frachtenberg 1913: 18)
(94) Tsö k:ilo 'wît h E wix:ī'lis,
now (he) saw it the food
'He saw the different kinds of food.'
The final noun phrase serves to ensure that we remember what was there to
be seen.
Such 'afterthought' constructions also provide a device for keeping
heavy information from blocking the flow of the discourse. When a con­
stituent is so heavy that its early appearance would interfere with the presen-
316 MARIANNE MITHUN

tation of information, it may be represented early in the clause by a deictic

particle, then filled in later by the 'afterthought/
(Frachtenberg 1913: 16)
(95) MīL halt! eene xle itc eeL!äts
please now thou with it with thou speak
te, xL!e ' 'yis.
this my with language
'Now try my language.'
The usual devices for discovering basic constituent order thus provide lit­
tle clear evidence for any underlying order in these languages. The statis­
tically more frequent orders are no more pragmatically neutral than any
others. Discourse-initial sentences do not provide good models of neutral
order, since their pragmatic function is to establish the initial theme of the dis­
course, they are usually non-declarative, and they rarely contain enough
nominal constituents to show any order at all.

3. Word Order Typology and Pragmatically Based Ordering

A justification for assuming the existence of an arbitrary, syntactically
defined constituent order underlying every language would be its utility as a
basis from which to predict other structural features. Discovering correla­
tions among different constituent orders has been a major goal of word order
typologists since the pioneering work of Greenberg (1966). Greenberg and
many others, including Lehmann (1971, 1973, 1974), Venneman (1974,
1975), and Hawkins (1979, 1980, 1983), have uncovered strong patterns
among the types of word orders that occur within languages. At the same
time, their work has revealed the complexity of ordering relations. Many
statements must include provisions like 'with overwhelmingly greater than
chance frequency' or 'usually', etc. The formulation of such principles as ten­
dencies rather than absolute universals does not render them invalid, but,
rather points up the number of factors involved. Hawkins has carefully
excluded from his work those languages in which the determination of under­
lying order is problematic, so that the strength of his conclusions is not com­
promised by inaccurate starting points. This exclusion has been of course
both appropriate and necessary in initial investigations, it would now seem,
however, that an understanding of the applicability of word order universals
to languages exhibiting little evidence of basic word order should be a useful
part of the refinement of such universals.
 IS BASIC WORD ORDER UNIVERSAL? 317

Greenberg's first universal involves the relative order of subjects and

objects.
1. In declarative sentences with nominal subject and object, the dominant
order is almost always one in which the subject precedes the object, (1966:
110)
If the dominant order is assumed to be the most frequent, then Cayuga,
Ngandi, and Coos are clearly exceptions to this first principle, since the order
OVS shows a small statistical advantage. Greenberg included the word 'al­
most' for a reason explained in a note:
5. Siuslaw and Coos, which are Penutian languages of Oregon, and Coeur
d'Alene, a Salishan language, are exceptions. (1966: 105)
(The sources of Greenberg's Coos information were Frachtenberg 1913 and
1922, those used here.)
Other universals posited by Greenberg involve ordering relations
between verbs and their auxiliaries, adpositions and their objects, and
between nouns and modifiers such as adjectives, relative clauses, number,
and demonstratives. As noted earlier, a salient characteristic of languages
like Cayuga, Ngandi, and Coos, is their high polysynthesis. What other lan­
guages express in several words, these languages often express in one. This
fact has a significant effect on questions of ordering.
In some cases, there is no basic order simply because a concept is expres­
sed in a single word. The relative positions of verbs and auxiliaries in Cayuga
and Ngandi are such a case. Like a number of the languages in Greenberg's
original survey, these languages have no separate auxiliaries. Tense, aspect,
and mode are expressed by combinations of verbal prefixes and suffixes.
Coos has several separate particles with temporal meanings such as 'about
to', 'shall/will', 'intend', and 'usually/frequently/habitually'. Each of these
can either precede or follow the verb (Frachtenberg 1922: 383-4))
In Cayuga, Ngandi and Coos, functions performed by prepositions or
postpositions in other languages are accomplished by suffixes on nouns.
Cayuga
(96) a. Kanyatar'ké ha'he.
lake-LOCATIVE there he went
'He went to the lake.'
318 MARIANNE MITHUN

Ngandi (Heath 1978: 189)

(97) ba-ga-  u-ŋi gu- awal-gic-uŋ,
3Pl-SUB-go-PCONT GU-country-ALLA TWE-ABS
'They went to (their) country/
Coos (Frachtenberg 1922: 323)
(98) x-kwile 'L-eitc n-djī
from-sweathouse-in I-came
'I came from the sweat-house.'
Mixing discussions of the relative orders of roots and affixes with those of the
orders of words would seriously interfere with an understanding of universals
syntactic principles.
Some locations are expressed by separate words in these languages,
however. A Cayuga speaker was aksed to translate the English sentences
given here as glosses. The sentences were thus not spontaneous responses to
actual situations, so, although technically grammatical, they may be some­
what unnatural pragmatically. In the first sentence, ohnake: 'behind'
appears to function like a preposition.'
(99) a. Ohnaké: she kanqhsó:t haétakse:\
behind where house stands there she ran
'She ran behind the house. 1
A few moments later, however, the sentence below was given. This time,
ohnake: appears to function like a postposition.
b. Shę kan hsó:t ohnaké: haétakse:\
where house stands behind there she ran
'She ran behind the house.'
According to the speaker, both are equally grammatical. In fact, an
examination of the use of ohnake: in spontaneous discourse indicates that it
is not an adposition at all, but, rather, a deictic particle, used appositionally
in the sentences above. Hawkins (1983:16) has suggested that prepositions
and postpositions are better and more general type indicators than con­
stituent orders VSO and SOV. Languages like Cayuga, Ngandi, and Coos, do
not provide counterexamples to this hypothesis, but the hypothesis does not
provide motivation for chosing an arbitrary constituent order, either.
Other ordering relations first investigated by Greenberg involve nouns
and their modifiers, such as adjectives, relative clauses, and genitives. In lan­
guages like Cayuga, Ngandi, and Coos, adjectival words tend to be predica-
 IS BASIC WORD ORDER UNIVERSAL? 319

tions. Heath notes: 'Noun-phrases which have more than one constituent are
typically formed by apposition ... By using the term 'apposition' I am trying
to indicate that the various constituents are often formally independent of
each other; they often each have a complete set of affixes and may be sepa­
rated from each other by pauses and even by other constituents such as a
verb.' (1978: 52) This is of course reflective of the one-idea-at-a-time ten­
dency. It is difficult to find single noun phrases containing both an adjectival
constituent and a separate noun in spontaneous discourse. Instead, the mod­
ifier is normally a separate predication, as in the Cayuga, Ngandi, and Coos
passages below.
Cayuga
(100) Ne:' ki' he' hne:' wakyes'aké ne' a:sató:wa:t.
it is justalso contr it is easy the you would hunt
'Also, it's an easy way to hunt, as well.'
Ngandi (Heath 1978: 268)
(101) ni-wolo ni-yul-yun, ni-warjak,
MaSg-that MaSg-man-Abs MaSg-bad
'that man is a bad man.'
Coos (Frachtenberg 1922: 424)
(102) L!a'nex yeeneu kwä'sîs.
new is thy ball
'Your ball is new.'
Since these are predicate adjectives rather than attributive adjectives, the
prediction does not apply.
In Cayuga and Ngandi, adjectival verbs may incorporate the nouns they
modify to form a single constituent. The resulting complex verb can then
function either as a predication or as a nominal.
Cayuga
(103) akya'tawithriiyo:
it-dress-nice
'a/the dress is nice or a/the nice dress'
Ngandi (Heath 1978: 262)
(104) ŋi-yuŋ buluki? gu-dawal-wiripu-gi ŋa-ga-n-i:,
I-ABS also Gu-country-other-LOC 1Sg-SUB-sit-PC
T was staying in a different country,'
320 MARIANNE MITHUN

Again, since such expressions consist of single words, the relative orders of
their constituent morphemes cannot be compared to the relative orders of
nouns and adjectives in languages like English or French.
Because words corresponding to adjectives in languages like English are
full predications in Cayuga, Ngandi, and Coos, the distinction between adjec­
tives and relative clauses is not a sharp one. As in the case of adjectives, mate­
rial translated into English relative clauses may simply be a separate sen­
tence, as in the Coos example below.
Coos (Frachtenberg 1913: 6)
(105) Ûx kwîna'eiwat HE hemkwî'tîs.
they two look at it frequently the heavy waves
They looked frequently at the waves,
Hats yî'qa xwändj weL!L!a'ni lE xaap.
just conti- in this goes over the water
nually manner back and forth
that rolled back and forth continually.'
Alternatively, material corresponding to English relative clauses and their
heads may be incorporated into single words, as in the Cayuga example
below.
(106) A'awehthé' ne' aket'ithro:ni'.
it got strong the I made tea
T h e tea I made really got strong.'
On the extremely rare occasions when separate constituents appear
comparable to adjectives or relative clauses in other languages, the modifiers
and heads can appear in either order. (107) was elicited as a translation.
Cayuga Mod N
(107) Kę:kę: nikay'ato'tę:' aketshe:nę taku:s oká:nyas.
it is such is its my pet cat it is
white body lousy
'My white cat has fleas.'
(108) Thó ti ni:yóht nq:kyé ne:'
there so so it is this it is
That's the way it is with this
N Mod
akç:kwé kowiyaętatre'.
she person she is getting a child
woman who is expecting a child.'
 IS BASIC WORD ORDER UNIVERSAL? 321

Ngandi (Heath 1978:230) Mod N

(109) a-ja-bolk-d-i a-darpal a-jar a a-ñalk,
A-now-appear-AUG-PP A-big A-what'sit A-rain
1
'A big rain appeared then.
N Mod
(110) ba-vmn?-du-ni gu-jark gu-wanar
3PL-look-AUG-PR GU-water GU-huge
They see a huge body of water/
Coos (Frachtenberg 1913:14) Mod N
(111) Hekwa'în lEái xkwî'nautc le'u hä'wis L!tā.
very good appearance their ready land
T h e appearance of the world which they had created was very
good.'
Coos N/Mod (Frachtenberg 1913:13) N Mod
(112) Tso a'yu ûx kwîna'eiwaî le'ûx mî'laq sīL'nēi.
now surely they two look at their arrows joined
them frequently two together
They shook the joined arrows/
Nominal-genitive order show the same variation. A noun identifying a
possessor may appear on either side of the possessed noun. Both of the geni­
tive constructions below were taken from the Cayuga cosmology legend.
Cayuga
(113) ne' tho nę:kyę aethwe '-kyę k'anów'ake
the there this we would put it this on its back
'we would put this (dirt) there on his back
Gen N
shęnho: niyakota:tęo, kanyahtęko:wá k'anów'ake.
where so she remains great turtle on its back
where she is sitting, on the big turtle's back/
Cayuga N Gen
(114) akoti'skhwáę' nę:kyę k'anów'ake nę:kyę kanyahtękó:wa
they set her this on its back this great turtle
'they set her on the back of this great turtle'
Compare the two Coos genitive constructions below.
322 MARIANNE MITHUN

(Frachtenberg 1913:8) Gen N

(116) ... ûxlemi'yat IE mexa'ye û kwä'xu.
they two to stand the eagle his feathers
up caused it
Then they stuck into the ground the feathers of an eagle'
(Frachtenberg 1913: 15) N Gen
(117) Halt! xä lä û L!ē'yîs hE tsņ'na.
now he his his language the thunder
'the other man received Thunder's language'
Finally, several constructions do exist in which the order of constituents
is constant. Numbers tend to precede the nominals they quantify, but this is
of course consistent with the newsworthiness principle.
The relative order of determiners and nominals is generally invariable.
Definiteness is not obligatorily marked in Cayuga, and no indefinite article is
used. As noted earlier, an optional particle ne may precede definite nomi­
nals, including proper and possessed nouns. When it appears, it precedes the
nominal it modifies. A closer look at the function of the particle reveals why
its order does not change. In Cayuga, morphological verbs, like any clauses,
can function as syntactic nominals. They need carry no overt markers of
nominalization. The result is that normal discourse can consist largely of
verbs. The particle ne is most often used to indicate that what follows is func­
tioning syntactically as a nominal.
(118) akaqnihnató:k ' nóne:' nę:kyę
they noticed the you know this
'they noticed
ne' kowiyáętatre'.
the she is getting a baby
that she was expecting.'
The Coos article lE/hE is similar. Frachtenberg notes, T h e article has a
general nominalizing function, and when prefixed to adverbs, adjectives,
etc., gives them the force of nouns.' (1922: 320)
Coos (Frachtenberg 1913: 50)
(119) 'ne ītE lE eedowayExta'is qau'wa
I (am) emphatic the you wanted night
T am the one (whom) you wanted last night.'
The invariability of the positions of these articles is not arbitrary. Their prim-
 IS BASIC WORD ORDER UNIVERSAL? 323

ary function as a signal of constituent identity would be lost if they floated.

The fact that they precede rather than follow the constituents they modify is
also functional. Due to the potential decoding difficulties posed by long series
of verbs in normal discourse, speakers need an early cue to the syntactic func­
tion of nominalizations.
Except for Greenberg's first syntactic universal, Cayuga, Ngandi, and
Coos provide strong evidence neither for nor against the most discussed word
order universals. Most of the universals simply do not apply, because they are
defined over rigid word orders. Positing a basic, syntactically defined con­
stituent order for such language provides little predictive power. The recogni­
tion of pragmatically based languages is crucial to serious work on syntactic
topology, however, not only because they represent a significant proportion
of the world's languages, but also because of the obvious danger of misclas-
sifying them. As many of the elicited examples cited earlier demonstrate, it
is only too easy to force a language into an inappropriate syntactic model on
the basis of data elicited or analyzed out of context.

4. The Pragmatically Based Type

Against a backdrop of Indo-European languages, the Cayuga, Ngandi,
and Coos strict pragmatic ordering seems unusual. The phenomenon is not at
all rare, however. An obvious question is whether purely pragmatically
ordered languages share any other characteristics which might combine to
define a type.
Their most obvious shared feature is their high polysynthesis. Clauses,
or intonation units, typically consist of relatively few major constituents, but
these may be morphologically very complex. Verbs contain not only pronom­
inal and adverbial affixes, they can also contain nouns.
In one sense, Cayuga, Ngandi, and Coos, actually have as much arbit­
rary, rigid ordering as English, if not more, since so much information is
packed into the morphological structure of words, a domain where order is
the most rigid of all. The ordering of morphemes does differ in a significant
way from that of words, however. Speakers normally know where word
boundaries fall, whether they are literate in their languages or not. They can
easily pronounce sentences one word at a time, but few can pronounce words
one morpheme at a time, particularly when the words are morphologically
complex. In some cases, where morphemes correspond to syllables, there is
little fusion, and the meanings of morphemes are transparent, some speakers
may be aware of the identity of some morphemes. Some Lakhota speakers, for
324 MARIANNE MITHUN

example, can identify a morpheme like the first person agentive pronominal
prefix wa-, but not the allomorph bl- that fuses with y-initial verb stems. They
are likely to be aware of the plural suffix -pi, because its meaning is relatively
concrete, but not the nominalizer of the same form. In the Iroquoian lan­
guages, by contrast, morphemes only very rarely correspond to complete syl­
lables, and speakers without rigorous linguistic training generally have no
idea whatsoever which segment of sound corresponds to which element of
meaning. They could never recognize a verb root in isolation. These facts
reflect a fundamental aspect of the word. It comprises a unitary symbol for a
unitary concept. It is no accident that the order of morphemes within words
is not normally altered for semantic or pragmatic effect, as is the order of
words.
Of course, pragmatic ordering is not restricted to highly polysynthetic
languages. It can be found both in Indo-European languages like Sanskrit,
Latin, and Russian, and Czech, and in non-Indo-European languages like
Cayuga, Ngandi, and Coos. All of these languages differ from languages with
more rigidly syntactic order, like English, in having special morphology for
encoding the syntactic relations of constituents. Sanskrit, Latin, Russian, and
Czech use case affixes on nouns, Cayuga uses pronominal affixes on verbs,
and Ngandi and Coos use both. The pragmatic ordering is not the same in all
cases, however, as discussed earlier. Those using only nominal case markers
show syntactically based order with some theme-rheme reordering. Those
with more complex verbal morphology, in particular pronominal affixes or
clitics, show pragmatically based order in the reverse direction of decreasing
newsworthiness.
Does this mean that all languages with pronominal affixes exhibit purely
pragmatically determined constituent order? In fact, it does not. Lakhota, for
example, has pronominal prefixes marking agent and/or patient. Word order
in Lakhota is relatively rigid, however, generally SOV. Parengi (Gorum), a
South Munda language of India, also has obligatory pronominal affixes
within the verb referring to subjects and objects. In this language, word order
is basically verb-final, and the surface order generally appears to progress
from theme to rheme (Aze and Aze 1973).
A crucial feature of purely pragmatically ordering languages may be the
nature of the relationship between the verb and associated constituents. In
highly polysynthetic languages like Cayuga, Ngandi, and Coos, with obliga­
tory pronominal marking of arguments, it is the pronouns which bear the
primary case relations of arguments to the predication, not external noun
 IS BASIC WORD ORDER UNIVERSAL? 325

phrases. The associated noun phrases serve as appositives to the pronominal

affixes, rather than directly as subject and direct objects. Although Lakhota
verbs carry pronominal prefixes marking first and second person arguments,
third persons are not marked on the verb. Speakers will not accept an
unmarked verb alone in isolation as a complete statement, as they will in
Cayuga, Ngandi, and Coos.

5. Conclusion
An assumption upon which much current descriptive and typological
theory is based, namely, that all languages have some basic, syntactically
defined, word order, is thus not universally valid. In a number of languages,
the order of constituents does not reflect their syntactic functions at all, but
rather their pragmatic functions: their relative newsworthiness within the dis­
course at hand. Constituents may be newsworthy because they introduce per­
tinent, new information, present new topics, or indicate a contrast.
These pragmatically based languages differ in several important ways
from some of the more familiar, syntactically based languages which exhibit
'pragmatic' reordering' such as right and left dislocation. First, in syntacti­
cally based languages, pragmatic reordering is highly marked. Deviation
from the basic, syntactically defined word order indicates an unusual situa­
tion. In pragmatically based languages, on the other hand, all ordering
reflects pragmatic considerations. Unusual situations are marked by other
means. Second, in syntactically based languages, pragmatic reordering is usu­
ally assumed to result in a theme-rheme order, with elements of lower com­
municative dynamism at the beginning of clauses, followed by increasingly
more important or newsworthy elements. In the pragmatically based lan­
guages examined here, the order is nearly the reverse. Constituents appear in
descending order of newsworthiness. This does not result in a simple rheme-
theme order, however. New themes, newsworthy in their own right, appear
early, as do other orienting elements like time and location. Continuing
themes, however, as well as continuing times and locations, usually do not
appear as separate constituents at all. Pragmatically based languages are typ­
ically highly polysynthetic, and such information is simply referenced mor­
phologically within the verb.
All in all, pragmtically based languages do not provide strong evidence
against most word order typologies. Most of the implicational universals they
suggest are simply inapplicable. It is only too easy, however, to misclassify
such languages on the basis of the criteria usually employed to determine
326 MARIANNE MITHUN

basic order, and such misclassification can seriously obscur important

typological generalizations.

NOTE
*I am grateful to Cayuga speakers Marge Henry, Reg Henry, and Jim Skye, who generously con­
tributed their expertise. I especially appreciate the many long hours of patience contributed by Mr.
Henry, as well as his insight and keen sensitivity to the intricacies of his language. I am also grateful
to Marta Roth for sharing her expertise on Czech, to Sandra Thompson for many helpful sugges­
tions, and to Wallace Chafe for fruitful discussions about discourse in general.

REFERENCES

Anderson, J.M. and C. Jones, ed. 1974. Historical linguistics 1. Amsterdam:

North Holland.
Aze, F. Richard and Trish Aze. 1973. "Parengi texts". In: Trail 1973. 213-362.
Boas, Franz, ed, 1922. Handbook of American Indian languages. Bureau of
American Ethnology Bulletin 40.2. Washington: Government Printing
Office.
Chafe, Wallace. 1980. 'The deployment of consciousness in the production
of a narrative". In: Chafe 1980. 9-50.
. ed. 1980. The pear stories: cognitive, cultural, and linguistic aspects of
narrative production. Norwood, N.J.: Albex Publishing Corp.
Chomsky, Noam. 1957. Syntactic structures. The Hague: Mouton.
. 1965. Aspects of the theory of syntax. Cambridge, Mass.: MIT Press.
Clyne, P. et al, eds. 1979. Papers from the fifteenth regional meeting of the
Chicago linguistics society. Chicago: University of Chicago.
Cole, Peter and Jerrold Sadock, eds. 1977. Grammatical relations. Syntax
and semantics Vol 8. New York: Academic Press.
Dryer, Matthew. 1983. "Coos word order". Paper presented at the Western
Conference on Linguistics, University of Oregon, Eugene.
Firbas, Jan. 1964. "On defining the theme in functional sentence analysis".
Travaux linguistiques de Prague 1: 267-280.
. 1972. "On the interplay of prosodic and non-prosodic means of func­
tional sentence perspective". In: Fried 1972.
Frachtenberg, Leo J. 1913. Coos texts. Columbia University Contributions to
Anthropology Vol 1. New York: Columbia University Press.
— . 1922. "Coos." In: Boas 1922. 297-430.
 IS BASIC WORD ORDER UNIVERSAL? 327

Fried, V., ed. 1972. The Prague school of linguistics and language teaching.
London: Oxford University Press. 77-94.
Givón Talmy. 1979. On understanding grammar. New York: Academic
Press.
. 1983. "Topic continuity in discourse: an introduction". In: Givón 1983.
1-43.
. 1983. Topic continuity in discourse. Amsterdam: John Benjamins.
Greenberg, J.H. 1966. "Some universals of grammar with particular refer­
ence to the order of meaningful elements". In: Greenberg 1966. 73-113.
. ed. 1966. Universals of language, 2nd ed. Cambridge, Mass.: MIT Press.
Halliday, M.A.K. 1967. "Notes on transitivity and theme in English". Jour­
nal of Linguistics 3: 37-81, 199-244.
Hawkins, John A. 1979. "Implicational universals as predictors of word order
change". Language 55: 618-648.
. 1980. "On implicational and distributional universals of word order".
Journal of Linguistics 16.2: 193-235.
. 1983. Word order universals. New York: Academic Press.
Heath, Jeffrey. 1978. Ngandi grammar, texts, and dictionary. Canberra:
Australian Institute of Aboriginal Studies.
Kuno, Susumo. 1976. "Subject, theme and speaker's empathy: a re-examina­
tion of relativization phenomena". In: Li 1976. 417-444.
Lehmann, W.P. 1973. "A structural principle of language and its implica­
tions". Language 49: 47-66.
. 1974. Proto-Indo-European syntax. Austin: University of Texas Press.
Li, Charles N., ed. 1975. Word order and word order change. Austin: Univer­
sity of Texas Press.
. ed. 1976. Subject and topic. New York: Academic Press.
. and Sandra Thompson. 1976. "On the issue of word order in a syn­
chronic grammar: a case against 'movement transformations'". Lingua 39:
169-181.
Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press.
Mathesius, Vilem. 1928. "On the linguistic characterology of modern Eng­
lish". Actes du premier congres international de linguistes a la Haye.
. 1939. "O tak zvanem aktualnim cleneni vetnem". (On the socalled func­
tional sentence perspective) SaS 5: 171-174. Reprinted in Cestina a obecny
jazykozpyt (The Czech language and general linguistics). Prague 1947. 234-
242.
McCawley, James D. 1970. "English as a VSO language". Language 46.286-
328 MARIANNE MITHUN

299.
Mithun, Marianne. 1984. " T h e evolution of noun incorporation 1 '. Language
60: 847-94.
Nichols, Johanna. 1984 ms. "Head-marking and dependent-marking gram­
mar". University of California, Berkeley.
Pullum, Geoffrey K. 1977. "Word order universals and grammatical rela­
tions". In: Cole and Sadock. 1977. 249-77.
— - . 1981. "Object-initial Languages". International Journal of American
Linguistics 47: 192-214.
Thompson, Sandra A. 1978. "Modern English from a typological point of
view: same implications of the function of word order". Linguïstische
Berichte 54: 19-35.
Tomlin, Russ and Richard Rhodes. 1979. "An introduction to information
distribution in Ojibwa". In: Clyne et al. 1979. 302-320.
Trail, R. ed. 1973. Patterns in clause, sentence, and discourse in selected lan­
guages of India and Nepal. Part HI, Texts. Norman, Oklahoma: Summer
Institute of Linguistics.
Vennemann, Theo. 1974. "Topics, subjects and word order: from SXV to
SVX via TVX". In: Anderson and Jones 1974.
— . 1975. "An Explanation of Drift". In: Li 1975.
 ENCODING EVENTS IN KALAM AND ENGLISH:
DIFFERENT LOGICS FOR REPORTING EXPERIENCE

ANDREW PAWLEY
University of Auckland

1. Introduction
In Kalam, a language spoken in the Highlands of Papua New Guinea,
conventions for talking about events and event sequences differ in striking
ways from those followed by English speakers. l The flavour of the contrast is
suggested by pairs (l)-(3), each of which consists of a fragment of Kalam nar­
rative with an idiomatic translation: 2
(1) Mnek am mon pk d ap ay-a-k.
next: go wood hit get come he-placed
morning
'Next morning he gathered firewood.'
(2) Mnmon Lae nb am md-y, toytk ow-n-k,
place Lae place go having- yesterday I-came
stayed-SS
'I came back from Lae yesterday'
(3) Tmwk ag-e-k nŋ-b-yn.
thunder it-having- I-perceived
sounded-DS
'I heard thunder'
and by (4)-(5), each of which consists of a fragment of English narrative with
its idiomatic Kalam translation:
(4) 'A stone broke the glass.'
Kab anan ap yap pk-e-k, pag-p ok.
stone glass come fall it-having- it-has- that
hit-DS broken
330 ANDREW PAWLEY

(5) T h e thief was sent to prison.'

B tap sy d-p, d am, kot g-y,
man stuff illegal he- get go court having-done-SS
got
kalabws ay-p-ay.
jail they-have-put
My first concern in this paper is to explore some of the conventions which
lead English and Kalam speakers customarily to report similar objective
events in different ways. My starting point is that of a translator who wants to
have, for both languages, a grammar of usages for encoding events.
What exactly is entailed in such a grammar? One can hardly consider this
matter without asking some fundamental questions about the relation
between language and the world, and about the tasks of linguistics. For exam­
ple, when someone reports an observed event, what is being reported? Seem­
ingly without effort, a person will describe an incident that he witnessed, and
his audience will hear the report as a description of a piece of reality, an actual
happening. But, as Grace (1982: 4) says, "it is quite misleading to speak...as
though linguistic utterances were about reality directly." What is reported as
an event or happening is the outcome of a complex interaction between many
variables: between certain physical phenomena and the speaker's sensory
impressions of these phenomena; habits and expectations, and limitations of
attention span, biassing the speaker's interpretation of these impressions;
limitations of long term memory influencing his recall; and, among other
things, structural patterns provided by the language which shape the form of
his report and possibly his initial perception and memory of what happened.
The question arises, then, as to what role language plays in the structur­
ing of experience, in people's conceptions of reality. With a few notable
exceptions, such as B.L. Whorf, modern linguists — at least in the public
arena — have generally steered clear of this question, or placed it outside the
main agenda of linguistic research. Such systematic investigations as have
been made have been mainly in the domain of lexicon.
In fact, however, much of the discussion of grammatical universals over
the past 20 years has rested on assumption about the relation between the
grammatical or semantic structure of particular languages, on the one hand,
and the world as known by human experience, on the other. Some propo­
nents of universal grammar have adopted a strong version of the hypothesis
that languages are essentially culture-free systems for expressing thoughts:
deep structure is universal. Thus, a description of an event in language A will
 ENCODING EVENTS IN KALAM AND ENGLISH 331

be equivalent, at a deep structure level, to its proper translation in language

B. The logic behind such an assumption is, presumably, that Nature has inhe­
rent structure, and that Mankind's common biological makeup will cause
people in all language communities to perceive that structure and to reflect it
— identically, at some level — in their linguistic representations.
In recent years a number of linguists have returned to Whorf's view that
the relation between language, thought and reality is a subject too important
for linguistics to ignore. Two scholars who have been much concerned with
this matter, and whose work I will draw on here, are Wallace Chafe (e.g.
1973, 1974, 1977, 1979, 1980) and George Grace (1981, 1981-83).
In several recent essays, Grace has focused on the functions of human
language in relation to the problem of translation. A central function of lan­
guage, he argues, is the construction of conceptual worlds or conceptual
realities. A conceptual world is "a relatively permanent system of resources
available for the characterization of situations", 3 such as is provided by the
syntactic devices and lexicon of a language.
If languages have markedly different resources for the characterization
of situations, i.e. if their grammars require them to report the same bits of
observed reality in very different ways, it may be that translation between
such reports is impossible — at least, accurate translation. Grace concludes
that what he calls 'isomorphic' or 'quasi-isomorphic , translation is rarely pos­
sible between languages that are genetically unrelated or associated with rad­
ically different cultures. In an isomorphic translation the source text and its
translation specify the same conceptual situation (a construct to be discussed
in some detail below). Usually, the best one can hope to achieve is that kind
of matching which Grace refers to as 'paraphrastic' translation — in which the
speaker's general communicative intent is more or less accurately captured.
What is being translated in such cases is not the linguistic meaning or the con­
ceptual situation specified by the source text; rather, it is the translator's
reconstruction of the speaker's pragmatic meaning or communicative pur­
pose. In fact, paraphrastic translation does not specify meaning so much as
indicate it.
Grace (1982:10-11) comments as follows on the prelinguistic activity that
goes on in encoding — on the selective perceptual processes that precede
'saying something about the world' in a particular language:
(1) We cannot deal with the whole of reality at once — some more limited segment
of it must be singled out....
(2) Even in such a narrowly confined segment....one decides what is of interest
332 ANDREW PAWLEY

and...leaves out of the account much of the actual array of sensory impression.
(3) One perceives relations in what has been singled out to be reported, i.e. one
makes sense of it.
Wallace Chafe and his associates have carried out several experiments
dealing with how people interpret and report large chunks of past experience.
These experiments suggest to Chafe that "there is little that happens to us that
we do not interpret in terms of patterns already existent in our minds (1977:
222). Following Bartlett (1932) he uses the term 'schema' for habitually-used
patterns which are imposed when interpreted the external world:
A schema...is a stereotyped pattern by which experience is organized, and
more specifically a pattern that dictates the way in which a particular larger
chunk will be broken down into smaller chunks. (Chafe 1977: 222).
Chafe and Grace each go on to say something about linguistic schemas
which are used to interpret experience, and in particular, about schemas used
to encode those chunks of experience which they call events and situations. I
will take up their specific proposals in the next section. At this point it is worth
mentioning that Grace wishes to locate, in this business of putting a linguistic
construction on our experiences and thoughts, a central question of linguis­
tics: what is it to say something? Specifically, what is the nature of the 'things'
that are said in in human discourse? And when can two sayable things be con­
sidered the same? These are questions which in one way or another deeply
concern lawyers, philosophers, psychologists, translators, language learners,
sociologists, poets — everyone, in fact, who has an interest in real-world lan­
guage use. But it is linguists who are most competent to tackle them. Yet we
largely bypass these questions and do not understand exactly what they
involve.
Grace suggests that saying something analytically — a very different
matter from indicating or hinting — is an extraordinary innovation that has
perhaps arisen only once in the history of the universe. It is the crux of the dif­
ference between human language and the communication systems of other
species.4
In the course of outlining what he thinks is entailed in an act of saying
something, Grace adumbrates a theory about the relation between bits of
external reality, on the one hand, and linguistic constructions of reality, on
the other. Linguistic constructions themselves entail a further relation
between conceptual schemas and syntactic schemas. A particular formula­
tion of this theory is of special interest to us here.
 ENCODING EVENTS IN KALAM AND ENGLISH 333

2. The hypothesis of a natural relation between clause, conceptual event

and objective event
Let us propose, for the sake of argument, that there is in all languages a
natural relationship between, 'conceptual' events as expressed by clauses,
and 'natural' or 'actual' events. To make this hypothesis we must first suppose
that there are in nature certain phenomena that human beings are biologi­
cally constructed to (learn to) perceive as single, bounded events. Such
events, then, are a natural class, exhibiting objective characteristics that
human brains are adapted to notice. We must suppose, further, that there is
a linguistic schema specifically designed to encode our conceptions of the
structure of single, bounded events, namely, the clause.
A strong and a weak version of this hypothesis can be distinguished. The
strong version will maintain that clause structure is similar in all languages,
and that the same (or nearly the same) range of objective phenomena are
treated as single conceptual events, encoded by single clauses, in all lan­
guages. A weak version will maintain only that there is some degree of agree­
ment across languages in these matters. The qualifications may be of various
kinds, e.g. that only a small core of conceptual events is universal, and that
other conceptual events are particular to individual languages; that only some
characteristics of typical, objective events are universally picked out for
encoding in clause structure, while others are given variable treatment, being
mentioned in some languages and not mentioned in others, or being spread
across more than one clause in some languages; or that the basic constituents
of clause structure vary somewhat between languages.
I am not sure that anyone has explicitly proposed any version of this gen­
eral hypothesis before. 5 I believe, however, that certain components of the
hypothesis have been spelt out before, and that something like the strong ver­
sion of the hypothesis can be detected as an assumption in many writings
about languages.
In outlining the syntactic mechanisms needed in order for a speaker to
say something about the world, Grace (1983b: 7-8) says:
The syntactic function which more than any other seems to hold the key to
human language is that which permits the specification of what Ï call the
"conceptual situation"....
The conceptual situation is a model of a clause-sized chunk of reality or
imagined reality. The syntactic mechanisms involved are primariiy those
which mark the case relations of the verb.
Discussing the 'propositionalizing' stage in the recall and verbalizing of
334 ANDREW PAWLEY

experience, Chafe (1977: 222) makes a similar connection between the ele­
ments of situations and their expression in clause structure:
"...in propositionalizing a chunk [of experience or thought] is replaced by a
different kind of entity: a structure consisting of an event (or situation) plus,
as separate element elements, the objects the speaker has chosen to ver­
balize as participants in the event... [E]ach of the objects plays its own
specific role in the event or situation. The major contribution of...case gram­
mar has been to recognize the existence and importance of these roles. Thus,
in Then I ate a sandwich' there has been not only a factoring out from the
holistic event of the ideas of me and the sandwich but also a decision to treat
me as the 'agent' of the event and the sandwich as 'patient'.
As far as I can tell, Grace's 'conceptual situation' closely matches
Chafes 'proposition'. Both seem to entail a commitment to report a chunk of
thought by a clause, a factoring out of objects or participants in the situation
from the action or state, a classification of the action or state as being of a cer­
tain type, having a particular set of case relations, and an assignment of case
roles to the objects or participants. But there is no necessary commitment to
particular words. 6
According to Chafe, the final stage of encoding a particular thought is
reached, when each participant, etc. in the situation is categorized — given a
name or characterization in words. Grace remarks that in addition to naming
the action or process and indicating its relevance by naming, e.g. time and
place, or participants, a speaker who wishes to 'say something' needs also to
specify the modality or condition or instantiation; that is, he must indicate
whether the particular proposition is being presented as, an assertion, a ques­
tion, a hypothetical situation, etc.
Both writers seem to use 'event' in two senses. In one sense an event is
a component of a conceptual situation, specifically that component which is
named by the verb. In its second sense, an event is a whole conceptual situa­
tion, one involving an action or process rather than a state.
If I understand Chafe correctly, he suggests that events in the second
sense are a natural class, in as much as prototypical events have various objec­
tive characteristics that distinguish them from prototypical objects and states.
The difference between (the idea of) an event and (the idea of) an object can
be specified in terms of space, time and particularity. Unlike objects and
states. typical events occupy "a single, unique, limited segment of time", and
indeed, occupy a segment of time that is no greater than an observer's span
of focused consciousness — a few seconds. Such is not true of all events, but
it is true of those events that are typically encoded as a single conceptual situ-
 ENCODING EVENTS IN KALAM AND ENGLISH 335

ation, expressed by a single clause. In a narrative, clauses tend to express

basic units of memory, chunks of experience which the narrator as observer
or participant was able to devote his central attention to, store as a raw per­
ception or memory, and recall in essentially the same form on each occasion
of recall.
It is an attractive idea, Chafe's idea that a clause typically gives just that
amount of information that a narrator can recall and verbalize in a single span
of focused attention. When transcribing large amounts of spontaneous
speech a decade ago Frances Syder and I were led to a similar conclusion
(Pawley and Syder 1975, 1976). However, our work dealt exclusively with
English speech. While Chafe's experiments included narrations in a number
of different languages, the evidence he cites is also mainly from English, and
related languages. It remains to be seen how similar or diverse human lan­
guages are in respect of what they include in reports of comparable events. It
remains to be seen, too, how consistently different languages package such
reports in the form of single clauses.

3. Comparing English and Kalam: Definitions

We turn now to the comparison of English and Kalam. Such a compari­
son is of interest, I believe, in that it may indicate roughly the outer limits of
variation among languages in resources and conventions encoding event-like
phenomena. In these respects Kalam may be as different from English as any
language on earth.
It is not easy to characterize the differences in a precise, technical way —
largely because linguistics lacks a well-developed metalanguage for talking
about conventions for encoding events. English speakers who are learning
Kalam, or translating Kalam discourse into English, find many Kalam,
accounts of happenings extraordinarily explicit and long-winded. When
describing an activity Kalam speakers will routinely single out for mention
certain aspects or components of the activity that English speakers normally
leave out or fuse together with other components. Kalam speakers often
found my descriptions of events to be cryptically or ambiguously telegraphic,
relying heavily on assumed knowledge and inference.
It will help to have a few more technical terms. While the following labels
and definitions are not altogether satisfactory, they provide a rough working
platform.
Event classifier, a verb stem denoting a kind of action, conceptually simple
336 ANDREW PAWLEY

or complex. (Let 'action' stand for action or process.)

Conceptual event, the meaning of a clause containing a single event clas­
sifier, and denoting a bounded activity (usually placed in a particular
time and place, but see generic event).
Event sequence, two or more conceptual events, each of which is expres­
sed by a separate clause.
Simple event, a conceptual event which comprises a single action, unre­
pealed. The typical case is an event that begins and ends within the space
of a few seconds or less, e.g. Bill released the rope; John winked; Mary
gave me a slap in the face.
Episode or Episodic event, a conceptual event which denotes a sequence
of more or less separate acts, e.g. Mary painted a landscape: Bill built his
own house; Lisa bludgeoned her father to death with forty whacks of an
axe; then she gave her mother forty-one.
Complex event, any event that is not a simple event but is not necessarily
episodic. Between simple and episodic events falls a large body of con­
ceptual events occupying various points on a scale of complexity, e.g.
Bill hit a ball through the window; Fido is fetching his stick; Mary's words
made me angry.
Specific event, a conceptual event consisting of a specific action, taking
place at a specific time and place.
Generic event, a conceptual event in which the action, time and place are
not specific, e.g. Mary used to drink coffee last year; Mary always drinks
coffee at breakfast.
Objective event, the reality which event expressions refer to, or which lan­
guage users imagine they refer to.
When I use 'event' without a modifier, it should be read (unless other­
wise obvious) as meaning 'conceptual event'. I will sometimes speak also of
episodic verbs and episodic events, simple event verbs, etc.
As we shall see, one important difference between English and Kalam is
that Kalam has few event expressions that are very high on the scale of con­
ceptual complexity. Kalam clauses (if they are clearly single clauses) usually
denote simple events. English speakers, on the other hand, freely use single
clauses to represent episodes and other complex conceptual events.
These restrictions in Kalam are related to the small number of verbs.
Verb stems are a closed set, comprising about 90 members. They are sharply
distinguished from ail other parts of speech by numerous morphological and
syntactic criteria. All verb stems (and no other class) may carry suffixes mark-
 ENCODING EVENTS IN KALAM AND ENGLISH 337

ing (1) absolute tense, (2) relative tense — indicating whether a non-final verb
refers to an event before, simultaneously or after the time of the final verb in
the sentence, (3) aspect, (4) mood, (5) absolute reference of subject by person
and number, (6) relative reference of subject — indicating whether the sub­
ject of a non-final verb is the same as that of the final verb in the sentence, or
different. In examples given here, 'same subject' suffixes are glossed 'SS' and
different subject suffixes 'DS'.
Verb stems also occur in their bare form (without suffixes) as non-final
members of serial verb constructions, containing up to five or six bare verbs
plus one inflected verb. Serial verb constructions are best regarded as strings
of separate words. Kalam speakers are able to say each constituent verb stem
in isolation and to give it a separate gloss.
A clause consists minimally of an inflected verb. The verb must be
inflected for subject reference (either (5) or (6)) and for one more of the other
grammatical categories (1) - (4) mentioned above.
Fewer than 30 verbs stems, all having very broad or abstract meanings,
account for more than 90 percent of verb stem occurrences in Kalam texts.
Each of these has a very general meaning, but translates many English verbs
having more specific meaning. Some of these recurrent verbs are:
d-, control, constrain, get, hold, touch, handle, grasp, take, catch,
finish, stop (an activity), attain (a position, etc.), capture, possess,
have, etc.
g-, do, act, make, work, occur, happen, function, operate, create,
cause.
md-, exist, be live, dwell, stay, remain, persist, be alive, be located at.
nŋ-, perceive, sense, be aware; thus: see, hear, feel, smell, know,
think, be conscious, awake, intelligent, remember, etc.
n-, transfer control, change the placement of s.t. ; thus; give, transfer,
transmit, fit (a lid), close (a door), put on or apply (e.g. makeup),
connect, reposition, set (a table), etc.
ay-, stabilize, thus: put, place, set, form, take shape, become, turn into
(a new state), put in order, be in a stable condition, have a visible
condition (scar, boils, warts, baldness, etc.), etc.
pag-, be destabilized, be in a disturbed condition; thus: break, shatter,
smash, bend, collapse, fold, chip, dent, buckle, ripple (as water
surface), etc. (all intransitive).
If a Kalam speaker wishes to be more specific about an action that any
338 ANDREW PAWLEY

single verb allows, he may use a combination of words or clauses. While such
conventional combinations provide specific equivalents of many English
verbs, most of the Kalam collocations are not idioms in the strict sense. For
example:
wdn nŋ- 'see' tmwd nŋ- 'hear'
eye perceive ear perceive
wsn nŋ- 'dream' gos nŋ- 'think'
sleep perceive mind perceive
d nŋ- 'feel nb nŋ- 'taste'
touch perceive (by touching)' eat perceive
am d ap- 'fetch ' nŋ tep gy ag- 'recall'
go get come perceive well having say
done-SS
English and Kalam differ in the first place, then, in respect of what ideas
can be expressed by a single verb. But there are other important differences
in encoding conventions, beyond those that are strictly lexical. When Kalam
narrators report an event sequence, the component events that they single
out for mention are rarely the same as those that an English speaking narrator
would single out. The next section will illustrate this point.

4. Some Familiar Event Sequences and Episodes

The data which follow concern familiar kinds of event sequences and
episodes, activities that are part of everyday life in Kalam or English speaking
societies. Accounts of such activities are a good data source for several
reasons. Some of the activities will be similar in both societies. Accounts will
be frequent. And ways of reporting familiar events are likely to be highly
codified. Among such reports we may hope to discover conventional patterns
or schemas for encoding complex experiences. 8

5. Hunting Game and Related Activities

(6) '...When that land came into existence,1 [people hunted game
mammals (and cooked and ate them)]
1
mneb ak lgl mdek [kmn ak pak dad
land that having come about it-existed-DS game that kill carry
apl, ty ty gl, adl nbek]...
having-come what what having- having- he-ate
done cooked
 ENCODING EVENTS IN KALAM AND ENGLISH 339

Extract (6) and the five Kalam text fragments that follow in 4.1. are
taken from the first chapter of Animals Our Ancestors Hunted, a book in pre­
paration by Ian Saem Majnep, a native speaker of Kalam, and Ralph Buhner,
a social anthropologist. In his introduction, Saem Majnep describes the
beliefs and practices of his people concerning the hunting of kmn or game
mammals.
In extract (6) five parts of the hunting sequence are mentioned by the
Kalam narrator: pak 'kill' (lit. 'strike', referring to the manner of killing), dad
'carry', apl 'having come (= ap- 'come' plus -1 'prior action by same subject
as final verb') — this refers to the bringing of the game back to the living site
(camp, or house) of the hunter, adl 'having baked', and nbek 'he ate'. The
intervening ty ty gl 'having done how many' refers to the performing of vari­
ous rituals and magic before cooking an animal.
Hunting game is an important, traditional activity both among the
Kalam and among rural English-speaking communities. Like many other
activities with utilitarian origins, it has been turned by the men who practise
it into a prestigious, rule- and ritual-governed enterprise. Any hunt is a com­
plex sequence of actions, generally beginning with a search for game, or the
flushing of a targeted animal from its lair, or the capture of game in traps or
snares, followed by the killing of the animal (in the European tradition usu­
ally by shooting, but among the Kalam usually by a blow after hand-capture),
bringing the carcass home, skinning or singeing it, cleaning it, and disposing of
the meat, offal and skin. The skin may be cured, the offal given to the dogs
and the meat eaten or given away, but practices vary according to the nature
of the game as well as between cultures. Among the Kalam, game is some­
times smoked for future use but most often is baked and eaten soon after the
animal is killed. In the latter circumstances cooking and eating game are vie­
wed as actions bound closely to the rest of the hunting sequence. In the Eng­
lish tradition, cooking and eating the catch is not so closely integrated into the
hunting sequence — some game animals, e.g. foxes, are not eaten at all, and
some are hung, or smoked or cooked and salted, to be eaten at a later date.
English has an episodic verb hunt, and also compound verbs like go hunt­
ing, go N hunting, etc, (where N stands for the kind of animal hunted) which
embraces all those activities considered to be part of the hunting sequence. In
communities where game are shot, the preferred episodic verb may be shoot.
The point is that one part of the hunting sequence is taken as standing for the
whole. It is not necessary for a narrator talking about a particular hunting
episode, or various hunting episodes, to specify the constituent events —
340 ANDREW PAWLEY

unless he wishes to highlight these events: He may simply say 'We hunted
every weekend'; or 'Bill went pig-hunting yesterday and got two'.
Kalam usage is different. Saem's references to hunting always specify a
sequence of acts, never fewer than three, usually five or six, sometimes more.
There is one standard sequence for frequently-caught arboreal animals,
another for burrowing animals (bandicoots), another for birds, and so on.
The narratives quoted here use variants of a sequence which refers to the
hunting of kmn 'game mammals' generally. Most kmn are arboreal marsu­
pials but some, such as the wallaby and bandicoot, are ground dwellers.
Often the verbs representing each act in a sequence are adjacent, either form­
ing a serial verb construction — a kind of extended single clause — or a
sequence of clause comprised entirely of verbs. Sometimes the verbs fall into
two (occasionally three) discontinuous sub-sequences.
In (6) the killing and transporting of the game fall into a separate sub­
sequence from the cooking and eating of it, although both subsequences
belong to a single multi-clause construction. To an English-speaking trans­
lator, mention of bringing, cooking and eating the game seems unnecessary
in this context.
(7) '... [certain ancestors] living there 1 [used to cook (and eat) in places
around there the game they killed].'
...[basd yes ogok] mdl kty l [am kmn pak dad apl,
ancestor certain having- they go game kill carry having-
distant stayed-SS come-SS
nb okok ad nbelgpal].
there around bake they-used-to-eat.
The larger discourse context indicates that here the narrator is focusing
on the fact that people cooked and ate game mammals in cordyline enclosures
(which individuals had planted for their own use) throughout the forest. An
English narrator recounting the same circumstances might be content to say
'the ancestors used to cook their game in these various cordyline enclosures
around there (the forest)'. Saem describes the circumstances as a sequence of
five or six events:
GO / KILL GAME / CARRY AND COME / BAKE / EAT (GAME)
The initial verb in this description, am 'go', refers to the fact that hunters
went out from their homes to hunt in the forest. The rest of the sequence cor­
responds in essential structure and detail to (6), with kmn pak dad apl form­
ing one clause, and nb okok ad nbelgpal a separate clause.
 ENCODING EVENTS IN KALAM AND ENGLISH 341

(8) ...And if you should wonder about the cordyline plants you see
when you go in the forest, what these will show you is the locations
where the earliest people had their house sites, where l[they
cooked and ate the animals they hunted].
...sblam nb ogok mey am nŋl mey ag nŋ gabm,
cordyline like certain there go having- so speak perceive you-do
ones (=ask)
that seen-SS
byn-b ak ned mdl, sblam mey ognl,
woman-man that first having- cordyline those hereabouts
lived-
SS
1
[kmn pak dad apl ad nbl], katp seŋ those
game kill carry having- bake having- house old-site those
come eaten-
SS
ognl agl mey tap nb ak nbep ned yomnggab.
hereabouts having those thing like- that you first it-will-show
asked that
Extract (8) spells out the focus on the cooking and eating of game in the
cordyline enclosures which was indicated less explicitly in (7). Once again the
elements kill game carry and come / bake / eat are specified, in that order, and
in this case without any intervening material.
(9) This was the basis on which1 [our forefathers hunted,... and had
their camp-sites in the same cordyline enclosures].'
Mey l[basd skop yes ogok am... tagl, sblam mgan nb ak
that ancestor distant some go walk- cord- enclosure that
(previous group about yline which-
phrase)
knl, kmn pak d ap ad nbl apelgpal].
having- game kill get come bake having- they-used-to-come.
slept-SS eaten-
SS
In (9) the hunting sequence is reported in full detail, using nine verbs,
grouped into four subsequences. The first is am, referring to going forth to
hunt in the forest; the second is tagl 'having walked about', referring to the
hunter's movements while hunting; the third is knl 'having slept', referring to
342 ANDREW PAWLEY

camping in the forest. The fourth and final subsequence contains six verbs in
a row — the five we have already encountered in (1-3) above plus apelgpal
'they used to come'; the latter refers to the hunters returning home from the
forest.
The single English verb 'hunted' is an idiomatic translation of the whole
sequence. It is noteworthy that in contrast to their extraordinary explicitness
in specifying the component actions in the hunting sequence, Kalam nar­
rators are very sparing in their mention of other elements — places, instru­
ments, objects affected, etc. The audience is left to infer these from his know­
ledge of customary usages. It is not simply a function of the sequence of verbs
being lexicalised, or as denoting a familiar series of events. As we shall see
later, Kalam speakers follow the same conventions when talking about
unfamiliar events — detailing the component actions (according to certain
conventions) while omitting reference to many of the other elements that an
English speaker feels it necessary to mention.
(10) 'The cordylines are of two kinds...both originally planted at Waym
and Mobn, the first places where, 1[(after hunting game mammals)
they made ovens with heated stones and ate the game]. 'Sskanay'
is the cross-cousin (close relative) of the real game mammal cor-
dyline, the one used in ovens when game mammals 2[after being
killed and brought home) are cooked (and eaten)].'
Sblam nb ak almŋal,...Waym yp Mobn yp ned mdl
cordy like- that two Waym and Mobn and first having-
line that lived
1
[kmn pak dad apl, kab g nbl], sblam tk
game kill carry having stone having- cordyline cut
eaten-SS
come do
ym getek nb ogok mey: nmwd ney sskanay
plant they-did the- some those cross- it sskanay
-DS which cousin
apal, nmwd ney, mey kmn sblam yb ak, kmn sblam
they- cross- it that game real that game cordy-
say cousin cordyline line
yb ak mey, 2 [kmn ak pak dad apl ad nbal] ak...
true that game kill carry having bake they-ate that
there that come-SS
 ENCODING EVENTS IN KALAM AND ENGLISH 343

(10) contains two references to hunting and cooking game. In the first,
the narrator seems to be principally concerned with the hunters' use of heated
stone to cook their catch. Nevertheless, convention requires that he specify
the main parts of the whole hunting sequence: kill/carry and come/cook/eat.
In the second, the focus is on the use of one of the cordyline kinds in cooking
game. But again, the narrator is constrained by custom to mention the events
that came before and after the cooking component in the hunting sequence,
and to mention all components in their temporal order of occurrence.
One final extract, to convince the reader that we are dealing here with a
highly conventionalised pattern of usage, a way of chunking and talking
about perceived reality that is standard procedure:
(11) 'And today, when some of us come across the ancient cordyline
groves of our ancestors, 1[we bring the game mammals we catch to
cook and eat there]... '
...'[cn] mny,..ognap am nŋl apwn, basd skop
we now some go having- we- ancestor group
seen- say
SS
1
sblam mgan pet agl, [kmn pak dad apl
cordyline always having- game kill carry having-
thought-
enclosure SS come-SS
nb okok ad nbl, ognap opwn]...
around bake having- some we-come
that eaten-SS
Once again the narrator's main concern is with one part of the sequence
but he does not extract that part from the whole nor does he give it particular
emphasis, syntactically. His main concern is to explain that hunters still use
the old cordyline enclosures for cooking game. An English-speaking narrator
in a similar situation would either omit reference to the events he wished to
background or diminish their syntactic prominence, e.g. by putting them in a
relative clause (hunters cook the game they have killed, hunters bring the
game they have killed to be cooked, etc.)
Kalam conventions for reporting the hunting sequence can be rep­
resented schematically as follows. Constituents or functions in parentheses
are optional, or contextually conditioned; the remaining elements are more
or less obligatory.
344 ANDREW PAWLEY

1 2 3 4 5 6
(GO KILL BRING IT COOK IT EAT IT (RETURN
FORTH) GAME TO CAMP HOME)
OR HOME
The question arises as to how typical such conventions are of Kalam
usage in general.

6. Gathering Nuts
Before leaving Saem's narrative, we may look at some extracts where he
refers to another traditional forest activity, having economic and ritual
importance: gathering and eating the nuts of the mountain pandanus (alrjaw),
which grows at high altitudes (above 2000 m.) in the mountain forest.
(12) '...and 1[when we gather mountain pandanus nuts we bring these
to the ancestral cordyline enclosures]1 2[to cook (and eat)].'
..1[alrjaw kab tk dad apl], shlam mgan pet
mt. nuts cut carry having- cordyline always
pandanus come-SS enclosure
nb ak 2[ad nbwn sek].
like- that bake we-eat together
that
(13) Thus the many hollows in the ground which can be found around
the cordyline shrubbery are old oven pits where 1 [pandanus fruit
was cooked on the heated stones].'
Kaw-twb ogok, konay mdengab, sblam mgan okok,...
hollow certain many they- cordyline around
ones will
exist enclosure
l
[tap alrjaw kab ak tk d apl adelgpal].
food mt. nuts that get having-
pandanus cut come- they-used-to-bake
SS
(14) ...1[and when men went...eating mountain pandanus nuts], 2[they
used to (cut them and bring them and) cook them in the cordyline
enclosures].'
..1[b ak...am...alnaw tk nb] 2[tagl, alrjaw kab tk dad
men that go mt. cut eat having- pandanus nuts cut carry
pandanus walked-about
 ENCODING EVENTS IN KALAM AND ENGLISH 345

apl sblam mgan nb okok adl gelgpal].

having- cordyline like-that having- they-used-to-do
come-SS enclosure around baked
Harvesting mountain pandanus nuts is broadly comparable to harvesting
and fruit-picking activities which English speakers refer to by such phrases as
digging potatoes, picking apples, picking raspberries for jam, etc. English
speakers refer to a single phrase in what is understood to be a standard
sequence of actions, and let this single phrase stand for the whole. The other
components in the sequence may be mentioned, if the narrator wishes to
highlight them (" All day I've been digging potatoes, picking up potatoes, put­
ting potatoes in bags,..."). In Kalam it is customary to mention several com­
ponents of the pandanus harvesting sequence:
1 2 3 4 5 6
(GO CUT BRING COOK EAT (RETURN
FORTH) NUTS THEM THEM THEM HOME)
TO CAMP
OR HOME
Extract (12) exhibits components 2-5, with 2-3 falling into one clause,
and 4-5 into another. (13) shows 2-4, standing in a single serial verb construc­
tion. (14) has two references to harvesting pandanus, though these are
interdepent. The first consists of components 1-2 and 5, and the second of 1-3,
while both share the same coda, 4: adl gelgpal 'having baked they used to do'
(i.e. they used to bake the nuts in an earth oven and do all those things previ­
ously mentioned). Component 4 is placed in a separate clause. Component 6
does not show up in these extracts, but has been observed elsewhere as part
of the pandanus harvesting sequence.

7. Other Event Sequences

Looking further afield, we find that Kalam has many other common
usages for designating everyday events that are broadly comparable in struc­
ture to those described above. Furthermore, the broad pattern is not con­
fined to collocations that are lexicalised or in common use. It is equally
characteristic of Kalam descriptions of unfamiliar events or of event
sequences having unfamiliar components.
I am speaking here of patterns or schémas for reporting event sequences
which have a deliberate action as one of the central components. There are
other sorts of event-sequences, e.g. those in which a person experiences a
346 ANDREW PAWLEY

sensation, or an involuntary bodily condition, which are encoded with differ­

ent schemas and which will not be described here.
For reporting a deliberate action the event-sequence seems to be roughly
as follows.
1 2 3 4
MOVEMENT ACTION MOVEMENT ACTION(S)
TO SCENE FROM SCENE AT PRESENT
OF FIRST OF 2 TO OR FINAL
ACTION PRESENT OR SCENE
FINAL SCENE
Each of these components may be internally complex, and components
2 and 4, in particular, may contain two or more verbs or contain an event-
sequence nested within them. In the case of lexicalised or semi-lexicalised
encodings, however, the usual realization is a serial verb sequence without
nestings. For example, gathering firewood, a daily routine, is usually
reported (as in (6) above) with the sequence:
1 2 3 4
am i mon pk / d ap / ay-
go wood strike get come put
where pk- 'strike' refers to the breaking of wood into convenient lengths
(traditionally done with hands and feet or by crashing pieces of wood against
something, rather than by chopping with an axe), and where ay- 'puť refers
to the bundle of wood being put away for use after being brought home.
Some lexical items which often occur in action schemas:
1 2 3
am 'go' mon pk 'break dam 'take'
ap 'come' firewood' get go
tag 'travel, kmn pk 'kill
walk about' game' dap 'bring'
kby 'leave' ñg mal 'fill with get come
water'
ap yap 'fall' dad ap 'bring'
come descend md- 'stay' carry come
4
ay 'place' (= stack, put away, keep, etc.)
ñb 'eat'
 ENCODING EVENTS IN KALAM AND ENGLISH 347

ad 'bake'
ad nb 'bake and eat'
The conventions of Kalam preclude the kinds of ambiguity that the
short-hand style of English often presents. Thus, an English utterance, I have
come from Lae, is ambiguous between the readings 'Lae was my home and
now I have come here', T have returned after a visit to Lae' and Lae was the
last place I stopped at on my journey'. These three situations will normally be
distinguished by Kalam narrators, as e.g. in (15), (16) and (17).
(15) I've just been to Lae/I've just come from Lae.
1 2 3
Mnmon Lae nb am/ mdy / opyn
place Lae at go having- I have-come
stayed-SS
Here the central event is the speaker's being in Lae, but it is placed in
relation to certain other circumstances: (1) Lae is not his reference location.
He had to travel to Lae in order to be there. (This perspective is captured in
English by using be + in/at Place.) (2) He is not in Lae now7. The action is seen
as completed. This perspective is captured in English by using the perfective
have..en. Kalam also uses a perfective (the medial verb ending -y, after md-
'stay') but must also specify where the actor went after he was in Lae: he came
to his present location. We can capture this perspective in English by using
come from, but I've come from Lae is indeterminate on the question of
whether I was based in Lae or had to travel to Lae to be there. Kalam does
not allow such indeterminacy. Thus (15) contrasts with (16):
(16) I've (just) come from Lae.
1 2 3
0 / Mnmon Lae nb kby / opyn
place Lae at having- I-have-come
left
In (16) the speaker is telling us that he has left Lae to come here. There
is perhaps no single central event here, but the focus is on the source — the
speaker has come from Lae. He was in Lae before he came here. The fact that
the first position in the schema is empty tells us that the speaker was residing
in Lae before leaving it; he did not travel there just beforehand. The selection
of the verb kby 'leave' is consistent with this interpretation — compare 15,
where am mdy opyn 'go having-stayed I-have-come' is the verb sequence —
but it is possible to say am mdy kby opyn 'go having-stayed having-left I-have-
348 ANDREW PAWLEY

come', if the speaker wishes to emphasize that Lae was not his home but he
did stay there for some time before leaving it.
If Lae was not his starting point, the speaker must tell us how he came to
be in Lae, as in (17).
(17) 'I have come from Lae, after being in Port Moresby' (or 'I came
from Port Moresby via Lae/)
yad balws dy Mosby nb, am mnmon Lae pwŋy
I plane having- Moresby at go place Lae having-
taken-SS landed-
SS
mdy opyn.
having I-have come
stayed-SS
(18) shows an episode which English treats as a relation between two
constituent events and Kalam as a relation between three events.
(18) 'I'll do it when I get back.'
1 3 4
amy apy wog gng gpyn.
having- having- work intending-
gone come doing I-do
The Kalam speaker of (18) was reassuring me that he would do a certain job
at the place where we lived. However, the speaker intended to go away for a
while before returning to do the job. In English only the returning needs to
be reported; departing is presupposed in the use of a verb such as get back,
come back, return. In Kalam the departing as well as the returning must be
overtly stated, each event reported by a separate clause.
(19) 'The garden was where I shot the pig' (or T h e pig was in the gar­
den when I shot it.')
1 2
Kaj wog day ap mdek / mey nagnk.
pig garden enclos- come it- therefore I-shot
sure having-
stayed-
DS
In (19) the central event is my shooting the pig. However, a Kalam
speaker will not report straight out that he shot a pig in the garden. Pigs do
not live in garden-enclosures; they have to go there first. (A report will also
 ENCODING EVENTS IN KALAM AND ENGLISH 349

commonly state how the pig entered the enclosure, e.g. by breaking through
the fence.) And one does not ordinarily shoot another man's pig at the
moment of its arrival in a garden. The speaker states that the pig had already
been in the garden for some time when he shot it. A fuller report might state
additional circumstances, but (19) gives the minimum required by the Kalam
discourse conventions.
The speaker of (20) has met a friend who asks what he is carrying:
(20) T have bought a gift from the store for Sawan.'
1 2 3 4
Stoa apy / tapskoy tawy / d am/ Sawan-nwp nng gpyn.
store having- small having- hold go Sawan-him intending-I-do
come- thing bought- giving-
SS SS SS
Here the speaker tells us that he had first to move to the scene of buying
(selection of ap 'come' indicates that the conversation takes place at the
scene), and that the object he has bought is to be given to Sawan. In order to
give it to Sawan it must first be taken to him. Components 3 and 4 thus form
a subsequence, itself consisting of a 1 + 2 , within the larger sequence.
(21) 'We have put the thief in jail.'
1 2 3
B tap sy dp d am / kot gy / kalabws aypwn.
man stuff he-got
illegally hold go court having- jail we-have-put
done-SS
T h e thief is expressed here by a relative construction 'man who stole
stuff. Before putting someone in jail it is necessary to taken him to court, and
in Kalam this was stated literally: the man was first taken into custody,
brought to the court and then tried, before he was jailed. The English expres­
sions 'send s.o. to jail', or 'have s.o. put in jail' loosely cover the sequence of
events that Kalam makes explicit.
(22) a. I forgot the bow.' b. T forgot to bring the bow.'
2 3
Cm saky gy / ownk.
bow out- having-done- I came
SS
of-mind
350 ANDREW PAWLEY

English allows an alternation between (a) and (b), with (a) having (b) as
a possible reading. (cf. also I came without the bow, which may imply forget­
ting to bring it.) In Kalam the nearest equivalent expression breaks the
episode into two parts — forgetting about the bow, and then coming —
expressed by two clauses.

8. Two Analyses of the English Metonymic Strategy

When referring to institutionalized event sequences, English narrators
prefer what might be called a metonymic strategy, leting the name of one part
of the sequence stand for the whole.
(23) I went to the movies (I went to the theatre, bought a ticket, went
into the theatre, watched the movie, etc.)
(24) I went to the supermarket (and went into it, selected food, carried
it to the cashier, paid for it, brought it home)
(25) I went to Chicago last week (and stayed there for a while and came
home again)
In (24-25) it is the first event in the sequence, movement to the scene of
main action, that stands for the whole. While this is a favourite strategy, Eng­
lish speakers often select other components:
(26) I saw a good movie yesterday.
(27) I found some bargains at the supermarket today.
(28) I was in Chicago last week and met a friend of yours.
My impression is that these alternatives are used when the speaker
wishes to focus on a particular component or aspect of the whole, in contrast
to a general, unfocused reporting of the whole. Thus, in (27) it is the discov­
ery of the bargains that is noteworthy. In (28) it is the fact that an unexpected
meeting took place while I was in Chicago.
In the following, however, it is the final stage of the event sequence that
stands for the whole:
(29) John loaded the truck with boxes (John picked up boxes, carried
them and put them on the tray of the truck, and kept doing this
until the tray was loaded)
(30) John took/delivered a load of onions to the supermarket (The
onions were loaded onto John's vehicle, he drove it to the super­
market, where the onions were unloaded and a receipt for delivery
was signed by a supermarket official).
 ENCODING EVENTS IN KALAM AND ENGLISH 351

Two superficially different analyses of these examples are available. Let

us call the first analysis — the one implicit in our remarks above —
backgrounding-by-omission. This analysis assumes that the speaker men­
tions a single component in an event-sequence, and omits the rest. The hearer
has a strategy available which allows him to work out whether or not the
single component stands for a whole series of events. If the activity referred
to is part of an institutionalised sequence, and if the other parts of the
sequence are not mentioned, the hearer is to infer that the other parts did
take place, as a matter of course, but that nothing of note was associated with
them. Not any component in the sequence can stand for the whole in this way,
however. Certain components or stages in a sequence are customarily used to
stand for the whole sequence, while others are used rarely or not at all.
The second analysis might be called lexical incorporation. In the case of
load or deliver in (29) and (30), for instance, one can argue that the whole
event-sequence is part of the lexical meaning of the verb in one of its senses,
just as the verb incorporates a conceptual event-sequence go, get and come or
go and bring. In similar vein, one might suggest that one of the senses of go
is 'go and carry out all the activities customarily associated with the following
noun', this sense being associated with constructions of the form go to
PLACE or a go N-ing.
I do not think it matters very much whether one opts for the lexical incor­
poration analysis or the backgrounding-by-omission analysis. The facts are
that when no other activities are mentioned by a narrator, the hearer under­
stands the single verb plus its arguments to stand for the whole customary
sequence. With institutionalised activities, the lexical incorporation and
backgrounding-by-omission analyses are probably notational variants.

9. Conclusion
We may conclude from the foregoing that there is no universal set of
episodic conceptual events. Indeed, it seems that languages may vary enorm­
ously in the kinds of resources they have for the characterization of episodes
and other complex events.

10. Clause and Case Relations as Putative Universals

It is an old idea that there is a natural, and perhaps universal connection
between the idea of a situation and its expression by a clause. Modern formu­
lations of case grammar (e.g. Fillmore 1968, Starosta 1973, 1978), have
352 ANDREW PAWLEY

refined this idea and made it the basis of a model of generative syntax. Even
though languages vary a great deal in the precise syntactic means (verb, case
forms, etc.) which express the elements of conceptual situations (action, case
roles, etc.), there is an impressive amount of agreement between languages
as to which elements are distinguished.
If I understand case grammarians correctly, the agreements go beyond
this. Languages also agree in that they typically encode the action or process
by a verb — a universally distinctive syntactic category — and certain other
elements by noun phrases; furthermore, the noun phrases are linked to the
verb by virtue of standing in the same clause (another universal syntactic cat­
egory) with it, as surface subject, direct object, etc.
Case grammarians have not, however, assumed that all languages agree
as to what counts as a conceptual situation. Presumably, the case model
allows for some variation between languages in respect of what actual situa­
tions are characteristically conceived of as a single conceptual situation, to be
mapped onto clause structure. But for the model to have general application,
there must be a large core of agreements between languages: a large set of
prototypical, universal conceptual situations.
The strongest version of the general theory outlined in section 2 is that
isomorphic translation, or at least clause-by-clause translation of event
descriptions is always possible between languages. This is, essentially, the
hypothesis that there is a universal deep or semantic structure. In the domain
of events, for example, any conceptual event in English is translatable into a
conceptual event (has a one clause translation exhibiting the same basic pac­
kaging of action and case relations) in Kalam or any other language.
We saw in section 4 that the hypothesis does not hold for episodic events.
The possibility remains that a weaker hypothesis is true, in which a certain set
of simpler conceptual events is universal. The question is, which events? We
might say that is as it may turn out — the definition of a simple conceptual
event will be known when it is discovered which conceptual events (if any) are
universal (have isomorphic encodings).

11. A Core of Conceptual Events Common to English and Kalam

English and Kalam speakers do share a core of conceptual events,
namely, those that are expressible in each language by a single clause. Some
examples:
 ENCODING EVENTS IN KALAM AND ENGLISH 353

(31) His house collapsed.

Kotp-nwk pagp.
house-his it-destabilized
(32) The man took hold of an axe.
B tw dp.
man axe he-took
(3) The man cut wood.
B mon thp.
man wood he-cut
(34) The wood split.
mon lakp.
wood it-split
(35) I am going home.
kotp amjpyn.
house I-am-going
(36) I feel hungry.
yp ywan gp.
me hunger it-acts
(37) I gave you money yesterday.
Yad toytk np mony nnk
I yesterday you money I-gave
The following conceptual elements are (sometimes) factored out and
expressed as verb and arguments of the same clause in both languages: action/
process (as verb), actor/agent, patient (which I take to include place of con­
tact verbs and goal of simple verbs of motion, as well as experiencer and
affected object, etc.), time and beneficiary (only of the verb n- 'give, etc.' in
Kalam).

12. Differences in the Treatment of Case Relations

The syntactic resources of English allow some quite complex action
sequences to be treated as a single conceptual event. For example:
(38) The man threw a stick over the fence into the garden.
In Kalam such an 'event' must be encoded as an episode, a sequence of
four conceptual events: (1) the man takes hold of the stick, (2) the stick is
354 ANDREW PAWLEY

thrown, (3) it flies over the fence, (4) it falls into the garden. These may be
compacted into three surface clauses, as in:
(39) B mon-day d yokek, waty at amb,
man stick hold he-displaced-DS fence above it-went
wog-mgan yowp.
garden-inside it-fell
English speakers have the choice of encoding an 'instrumental' action by
a prepositional phrase or by a full clause. That is to say, the instrumental
action may be conceptualized as an element of the same event as the resulting
action, as in (40), or as a separate event, as in (41):
(40) The man split the wood with an axe
(41) The man used an axe to split the wood
In Kalam, an instrumental action, such as using a tool in order to bring
about a resulting action or state, counts as a separate event from the resulting
action. Thus both (40) and (41) are translated by:
(42) b tw dy, mon tb lak-p
man axe having-taken wood cut he-split
SS
English speakers usually encode the beneficiary of a verb of object crea­
tion or control, like make, keep, collect, send, and steal, as an argument of the
verb, the beneficiary relation being marked by the preposition for. In Kalam
it takes at least two clauses to express such a relation. The first describes the
creation or control of the object. The second describes the transfer of the
object to a receiver, using the verb n- 'give'.
(43) kotp gy, np ring gspyn
house having- you intending I-am-doing
built-SS to-give-SS
T am building a house for you'
There are three clauses in (43): having built a house, to give you, I am doing.
A beneficiary relationship cannot hold directly between kotp g- 'build a
house' and np 'you', but the purpose of the building clause may be shown by
marking it as occurring prior to the giving clause, and marking the giving as
an intended or purposive act. The final verb places the speakers' action in an
absolute tense, as against the relative tense of the preceding two verbs.
English treats the direction of movement of an affected object as a case
relation of the verb denoting the causal act. The movement may be expressed
 ENCODING EVENTS IN KALAM AND ENGLISH 355

by a preposition (such as through, over, into, around) or as an adverbial ele­

ment (out, off, over, etc.). In Kalam the movement of an affected object
counts as a separate event (see example (39)), and its clause must be tensed
to show that it occurs after the causal action.
English allows the idea of cause to be incorporated in a verb which also
expresses an action or process, as e.g. in A stone broke the glass ox I smashed
the window with a bat. But in Kalam one cannot say X broke Y. One can only
say Y broke. What led to Y breaking is another matter. In order to attribute
a change of state event to an agent or cause, a Kalam speaker must say what
happened prior to the change of state. In the case of breaking glass, this
account will normally include mention of something coming into contact with
the glass, and of a movement or other action that preceded the event of con­
tact. Thus:
(44) Kab anan ap yap pkek, pagak ok.
stone glass come fall it-having-struck-DS it-broke that
where kab 'stone' is the subject of pkek 'it having struck' (different subject
from next verb) and anan 'glass' is the subject of pagak 'it broke.' The idioma­
tic English translation of (44), "A stone broke the glass,' is framed as a single
conceptual event. A more isomorphic translation would be 'A stone fell and
struck the glass and it broke,' representing a three-stage episode or event-
sequence.
Finally, in this brief survey, we may note that whereas English clause
structure provides a place for a prepositional phrase expressing the location
or source of a conceptual event, Kalam expresses location or source in a sep­
arate clause from the related event. Thus:
(45) Kiyas is working in his house.
Kiyas kotp-nwk mdyg, wog gsap.
Kiyas house-his staying- work he-is-doing
SS
(46) Where have you come from?
Akay mdaban, opan?
where you-stay you-have-come
(recent past)
vSummarizing, we can say that the differences illustrated by the preceding
examples are broadly of three sorts. First, Kalam speakers single out for
obligatory mention certain aspects of complex episodes that English speakers
356 ANDREW PAWLEY

usually do not refer to.

Second, clauses do rather different jobs in the two languages. English
clause structure is a syntactic Procrustean Bed into which a wide range
of diverse conceptual structures are squeezed. For example, unlike Kalam,
English allows several conceptual situations/events to be fused into a single
clause. This result is largely achieved by reducing certain situations/events
to the status of peripheral or backgrounded elements in the clause, expressed
as arguments of the verb. Use of an instrument, intention to transfer posses­
sion, direction of movement of an affected object, the location of a partici­
pant in an event — such things can be backgrounded in English not just by
expressing them as subordinate clauses (John was working while (he was) in
his house), but by reducing them to the status of noun phrases (John was
working in his house).
Third, verbs do rather different jobs in the two languages. In the case of
English, it is convenient to speak of a division of labor between full verbs,
which do lexical-referential work, and grammatical functors (auxiliaries, pre­
positions, etc.) which do grammatical work. In Kalam verbs do both kinds of
work, but always as full verbs. But Kalam is more restrictive than English in
the amount of information it allows to be compressed into a verb stem, and
in the kinds of case relations which may be associated with a single verb.

13. Conclusions
What general implications can be drawn from the foregoing comparison
of English and Kalam?
I think the evidence indicates, that while there is some connection
between (a) events in nature, (b) conceptual events and (c) clause structure,
it is a loose and indirect connection. Kalam and English do share a body of
more or less isomorphic conceptual events and situations, namely those
which both languages may express by a single clause. This common core pre­
sumably reflects certain characteristics of the external world and human
experience that are salient for people everywhere. But it is a fairly small core,
in relation to the total set of conceptual situations which English can reduce
to a single clause expression.
The comparison suggests, then, that the notions 'conceptual situation'
and 'conceptual event' are very largely language specific. A corollary of this
is that the notion of 'clause' is also to some extent language specific. English
and Kalam each exhibit certain types of construction that we feel comfortable
about labelling as clauses. But as with conceptual events, the clause struc-
 E N C O D I N G E V E N T S IN K A L A M A N D E N G L I S H 357

tures that are more or less isomorphic between Kalam and English are only
a modest part of the total range of English clause structures.
I do not think that this observation invalidates the idea that certain ele­
ments of conceptual situations, and specifically those singled out in the theory
of case grammar, are universal. But it does invalidate the notion that in all
languages there are syntactic resources that allow all these elements to be
expressed in a single clause, by one verb and its arguments.
Finally, space permits only passing mention of a point that deserves ful­
ler treatment. I have said that 'clause' is not the same thing (exactly) in the
grammar of English as it is in the grammar of Kalam. It needs to be said that
'clause' is not a well-defined entity in either language. What we have, in fact,
is more like a scale of constructions ranging from phrases to prototypical
clauses to sequences of prototypical clauses. In between are various construc­
tion types that are something less than or something more than the prototyp­
ical clause structure (whether the English type or the Kalam type). In Kalam,
for example, a number of verb stems (up to six or seven) may occur in succes­
sion, with only the final verb carrying inflections, as e.g.
(47) am d owan! 'Fetch it!'
go get you-come
(48) d am yokan! 'Get rid of it!'
get go you-displace
See also examples (6-14) above. One may well wish to argue that such
sequences show an attempt to crunch several underlying clauses into one sur­
face clause. But one cannot, I believe, argue that the crunching process has
been completed, i.e. serial verb constructions are not prototypical clauses.
They are clause-like, but are something more than a clause.
I alluded above to English constructions which are intermediate between
one and two clauses, namely those in which there is a reduced subordinate
clause. For a fuller discussion of a scale of clause integration or binding, see
Givón's (1980) account of complement typology.

NOTES
1) I am indebted to Ralph Bulmer and Ian Saem Majnep, and to George Grace for allowing me
to quote from unpublished material. Ralph Bulmer provided valuable comments on the draft. My
research among the Kalam people, a total of 12 months fieldwork carried out between 1963 and
1975, was funded by the Wenner-Gren Foundation, the New Zealand University Grants Resarch
Committee and the University of Papua New Guinea.
358 ANDREW PAWLEY

The Kalam number some 15,000 and live around the junction of the Bismarck and Schrader
Ranges, near Simbai, Madang Province, on the northern fringes of the central Highlands of Papua
New Guinea. Most Kalam live at altitudes of between 1500 and 2000 metres, cultivating sweet
potatoes as their main subsistence crop, with taro and yams as more prestigious ceremonial crops.
eaten, together with pork, during the annual cycle of smy or dance festivals which take place from
August to October. Hunting in the forests which cover the upper mountain slopes is an important
recreation as well as a subsidiary food source.
Kalam is usually regarded as belonging to the East New Guinea Highlands Stock of the Trans
New Guinea Phylum (Wurm 1975). Its only close relative is the nearby Kobon (or Kopon) lan­
guage, and possibly Gaj ; it shares less than 20 percent of basic vocabulary cognates (200 word list)
with all other East New Guinea Highlands Stock languages.
Kalam is spoken in a number of dialects and it should be noted that two different dialects are
represented in examples cited here. Examples (6-14) are in the ty mnm dialect spoken by Iam Saem
Majnep, of the Gobnem territorial group in the Upper Kaironk Valley. I worked mainly in the etp
mnm dialect region, among the Kaytog people of the Upper Kaironk. The two dialects show
considerable morphophonemic differences.
2) A phonemic orthography for Kalam was devised in the 1960s, and a modified form of this is
coming into use by the people as they begin to write in Kalam. The phonemic orthography will be
used here. There are four stops /b, d, j , g/, pronounced with homorganic nasal onset; initially and
medially the usual allophones are [mb. nd, nj, ng], and finally [mp. nt, nc, nk]. There are four oral
obstruents /p, t, c, k/, voiceless in initial position as [Φ, t, c, k], voiced and fricativised in medial
position [ß, r, j , y]; final /p/ may be either [b] or [p]; otherwise final allophones are as for initial
position. There are four nasals /m, n, n, ŋ]; and a lateral /1/ which is flapped and retroflexed. /w/
and /y/ pronounced as [w] and [y] before a vowel and in initial position, as [u] and [i] between con­
sonants, and as [uw] and [iy] finally. There are three pure vowels /a, e, o/.
All consonants are articulated with a predictable vocalic release when standing alone in a
word or when followed by another consonant in a word. The vocalic release is usually a short high
central [i], e.g. nŋbyn 'I perceived' is [niŋimbin], kmn 'game mammal' is [kimin], or a mid-central
[a] after a lone consonant, e.g. b 'man' is [mbe], m 'taro' is [ma]. However, the vocalic colouring
of the release show considerable variation according to the adjacent consonants, e.g. b-yad 'my
(kins)man' is [mbiyant], m-wog 'taro garden' is [muwonk] and m-yob "big taro' is [miyomp].
3) Grace (1983b). See also Grace 1982, 1983a, c.
4) Grace 1981b: 8ff. 1 assume that Grace is speaking here of natural language use, and might
wish to exclude the use of human language by apes.
5) See sec. 5 for some discussion of the connection between this hypothesis and models of case
grammar. If I read them right. Chafe and Grace (cited in the text below) would espouse a weak ver­
sion of the hypothesis.
6) Specification of the condition of instantiation is necessary to specify the modality which the
speaker places the conceptual situation he has invoked. Grace also refers to other syntactic devices
which have the functions of putting the utterance into an appropriate context in the discourse, or
playing a role in expository strategy (e.g. devices for foregrounding elements) (1983a: 15).
7) Chafe says (1977:225):
Perhaps most of the events that we deal with are possible to comprehend in their
entirety within what I have elsewhere called 'surface memory' (Chafe 1973), where
the entire segment of time from beginning to end can be held in consciousness with­
out being relegated to deeper levels of memory.
 ENCODING EVENTS IN KALAM A N D ENGLISH 359

....My suggestion is that the typical lifetime of an object is of a different order of

magnitude from that of the typical segment of time occupied by an event, and that
we conceive of the two as being different sorts of things for that reason. An insect
whole lifetime is only a fraction of a second may in immediate conceptual terms be
more like an even than an object.
8) For convenience, a free English translation of each Kalam extract is given first. A word-by­
word gloss appears under the Kalam text. The free translation (which I have adapted from
Bulmer's translation of Saem's Kalam text) makes some concessions to the style and structure of
the original, but generally conforms to the idiom of English narrative. Material which an idiomatic
English translation would omit is sometimes retained here but is enclosed in parentheses. Those
portions of the Kalam text and the English translation that refer to what I call the 'hunting
sequence' or the 'pandanus-gathering sequence' are square-bracketed and numbered for ease of
reference.
9) Starosta (1978:109, fn.) comments:
The 'discovery' ...that the semantic representation and the syntactic deep structure
of a sentence are identical, then, turns out to be simply a consequence of the arbit­
rary decision to have a grammar account for paraphrase; while the 'Universal Base
Hypothesis'...just turns out to be a claim that all humans perceive the same range
of situations in the same way, a proposition which would be rather difficult to sup­
port empirically. There would then be no difference in principle in a 'generative'
semantic framework between saying two sentences in Language L were synonym­
ous, and saying that a sentence of L was a translation of another sentence in L, since
these 'deep structures' are perceptual representations of situations, and not of the
structures of sentences in any particular language.

REFERENCES

Bartlett, F.C. 1932. Remembering. Cambridge, University Press.

Chafe, Wallace L. 1973. "Language and memory." Language 49:261-81.
. 1974. "Language and consciousness. Language 50:111-33.
. 1977. "The recall and verbalization of past experience." In P. Cole et al
(eds.) Current Issues in Linguistic Theory. pp. 215-246.
. 1979. "The flow of thought and the flow of language." In Givón (ed.),
pp. 159-81.
. 1980. "The deployment of consciousness in the production of a narra­
tive." In W.L. Chafe (ed.) The Pear Stories: cognitive, cultural and linguis­
tic aspects of narrative production. Norwood, NJ, Ablex. pp. 9-50.
Fillmore, Charles J. 1968. "The case for case." In E. Bach and R. harms
(eds.) Universals in Linguistic Theory. New York, Holt, Rinehart and
Winston. pp. 1-88.
Givón, T. (ed.) 1979. Syntax and Semantics. Vol. 12. Discourse and Syntax.
New York, Academic Press.
360 ANDREW PAWLEY

- - - . 1980. 'The binding hierarchy and the typology of complements."

Studies in Language 4(3): 333-77.
Grace, George W. 1981a. An Essay on Language. Columbia, S.C.,
Hornbeam Press.
. 1981-83. Ethnolinguistic Notes. Dept. of Linguistics, University of
Hawaii.
. 1981b. "Defining what linguistics is about (and why).' Series 3, No. 2 of
Grace 1981-83.
. 1982. "The question of the nature of language." Series 3, No. 4 of Grace
1981-83.
. 1983a. "More on the nature of language: the intertranslatability post­
ulate and its consequences." Series 3, No. 2 of Grace 1981-83.
——. 1983b. 'The linguistic construction of reality." Series 3, No. 11 of Grace
1981-83.
. 1983c. "A second progress report: on the nature of language." Series 3,
No. 16 of Grace 1981-83.
Pawley, Andrew and Frances Syder. 1975. "On sentence formulation in
spontaneous speech." N.Z. Speech Therapists' Journal 30(2):2-ll.
-—-. 1976. "The one clause at a time hypothesis." Paper read at 1st congress
of N.Z. Linguistic Society, Auckland, August 1976. Typescript. pp. 84.
Starosta, Stanley. 1973. "The faces of case." Language Sciences 25: 1-14.
. 1978. "Generative syntax: a case approach." Typescript. Dept. Linguis­
tics, University of Hawaii.
. 1982. "Case relations, perspective, and patient centrality." University
of Hawaii Working Papers in linguistics, vol. 14, no. 1.
Wurm, Stephen. (ed.) 1975. New Guinea Area Languages and Language
Study. Volume 1: Papuan Languages and the New Guinea Linguistic
Scene. Pacific Linguistics, Series C:38. Canberra: ANU.
 WORD ORDER IN INTRANSITIVE CLAUSES IN HIGH AND
LOW MALAY OF THE LATE NINETEENTH CENTURY

ELLEN RAFFERTY
University of Wisconsin

1. Introduction
Although it is generally accepted that contemporary Malay is an SVO
language (J. Greenberg 1963:107, J.U. Wolff 1982) data from the late
nineteenth and early twentieth centuries show Malay to have a high degree of
predicate-initial clauses in both transitive and intransitive clauses. This fact is
not surprising since many contemporary Austronesian languages such as
Malagasy, Tongan and Tagalog are predicate initial, and Old Javanese
(Kawi) was predicate initial. The discussion below examines the role of pred­
icate clauses in High and Low Malay at the turn of this century.
Although predicate initial clauses occur in both transitive and intransi­
tive clauses, this paper considers only the intransitive clauses because it is
here that the greatest amount of variation occurs. The intransitive clause was
also chosen because two argument clauses are rather rare in texts. In the High
Malay data examined 10% of the total 311 clauses had two nuclear argu­
ments, and in the Low Malay data the percentage rose to 19% of 397 clauses.
In this paper it is argued that the functions of predicate initial-clauses of
intransitive clauses in High and Low Malay texts are quite different and that
these differences are related to a modernizing trend in Malay that involves
not only (1) an increasing percentage of subject-initial clauses, (2) a decreas­
ing use of the focus particles -lah and -pun, and (3) a decreasing use of clause
linking words, such as shahadan, hatta and maka that mark new scenes or
paragraphs in the discourse. This modernizing trend has a number of causes
including the influence of subject-initial languages such as Dutch, English
and modern Javanese, as well as, a change in literary style from poetic form
to prose and from an oral to a written medium of dissemination of literary
works.
362 ELLEN RAFFERTY

One of the functions of predicate-initial intransitive clauses (VS clauses)

of High Malay is to mark transitions from one era or scene to the next. High
Malay literature of the late nineteenth century was emerging from an oral
tradition in which written literature was performed for an audience. This oral
presentation necessitates clear marking of the development of a literary piece
since continual audience attention can not be assumed. The VS clauses are
also more highly associated with major (not minor) characters and with the
use of the predicate marking particle -lah but not with the subject marking
particle -pun. All of the above characteristics of the VS clause conspire to give
greater prominence to this clause than to the SV clause.
Low Malay literature of the late nineteenth century arose from a modern
literary tradition with considerable Dutch influence in the early stages. The
Low Malay discourse has fewer predicate initial-clauses, a decreased use of
the paragraph marking words (shahadan and maka), and decreased use of the
focus particles. Word order in Low Malay reflects the definiteness of the sub­
ject, yielding a dominant SV order with a definite subject followed by verb.
There are small number of clauses with the verb followed by an indefinite
subject. In Low Malay the new information appears in second position, exp­
laining why clausal subject follow the verbs.
Malay literature from the late nineteenth and early twentieth centuries
is generally classified as either High or Low Malay, although High Malay
from this period is more accurately labelled early modern Malay because of
important differences between it and classical High Malay literature of the
sixteenth and seventeenth centuries. The styles of early modern Malay that
developed during the nineteenth century have several characteristics that dis­
tinguish them from classical Malay. Becker has noted that the most signific­
ant developments in the transition from classical to modern Malay are the loss
of the -lah/-pun structure and the closer syntactic resemblance of Malay to
European languages (Becker 1979:249, note 16). One of the important syn­
tactic changes was the shift from verb-initial to verb-medial clauses.
Before proceeding further with the discussion of differences among var­
ieties of Malay, I pause to offer some classifications of Malay that have been
suggested by prominent scholars. Marsden (1812:xv-xvii), a British linguist
who worked in West Sumatra, defined four types of Malay: 1) bahasa dalam
'the language of the court', 2) bahasa bangsawan 'the language of the elite',
3) bahasa dagang 'the language of trade' and 4) bahasa kachukan 'the mixed
language of the markets'. Windstedt (1927:177) followed this same basic clas­
sification adding only the category, literary Malay, which he characterized as
 INTRANSITIVE CLAUSES IN HIGH AND LOW MALAY 363

lacking abbreviations and regional expressions, while having complete verb

morphology and a complex sentence structure.
The above four way division of Malay was collapsed by most British and
Dutch scholars into a High/Low Malay dichotomy where High Malay was the
literary language and Low Malay was any of the numerous spoken dialects of
Malay. (See Alisjahbana 1957:44-47 for further discussion.) Although this
two-fold distinction was generally accepted at the turn of the century, it was
also acknowledged by scholars that there were varieties of both High and
Low Malay. Within the Low Malay category reference was made to Batavian
Malay, Java Malay, officialese or foreigners Malay and Chinese Malay to
name a few. Within the High Malay category variation occurred regionally,
creating varieties such as Banjarese Malay, and Riau Malay.1
For the purposes of this paper I use the dichotomy of High and Low
Malay but note that High Malay here refers only to early modern Malay of the
late nineteenth and early twentieth centuries, not to classical High Malay.
The situation is complex not only because both High and Low Malay cover a
number of varieties of language, but also because there is a continuum
between High and Low Malay on which features of High Malay, such as the
focus particles, -lah and -pun, and the clause combining words, such as shaha-
dan and maka, vary in frequency. The ends of the continuum can be iden­
tified, but all literature is not easily classified as either High or Low Malay
because some authors use 'proper' verb morphology but deviate from High
Malay discourse structure with respect to the use of the focus particles and
clause linking words, blurring the High/Low Malay distinction.
High Malay of the late nineteenth century flourished in the Malay courts
where sultans supported scribes to continue the literary traditions and record
the important events of the kingdom. This literary Malay developed distinc­
tive regional styles; it was not a stable or unitary phenomenon but exhibited
flexibility and adaptability. The two styles of the late nineteenth century High
Malay examined below are texts written in the Riau region and in Malacca.
Low Malay in Indonesia at this time was just emerging as a literary lan­
guage. Traditionally, that is, from the time of the Sriwijaya Empire (6th-12th
century), it was the oral language of inter-ethnic communication used
throughout the archipelago for trade and government. Only after the liberali­
zation of press regulations in the Dutch East Indies in 1856, was a local Malay
press established by Dutch residents of the Indies. In the last few decades of
the nineteenth century the Dutch editors were joined by Eurasians, Chinese
and then by indigenous literati in the early part of the twentieth century. The
364 ELLEN RAFFERTY

Low Malay of the press was, in general, similar to the Malay of Batavia
because the press developed most rapidly on Java. The audience of this press
included Dutch, Eurasians, Chinese and local peoples and for many of these
readers this was the only language in which they were literate.
The initial purpose of most Low Malay newpapers was to facilitate trade,
but soon the goals expanded to include entertainment and education. Many
of the early novels were first serialized in newspapers. To a large extent, it
was through this medium that Low Malay developed into a literary language.
It is from this literature that I have selected samples of Low Malay narratives.
The syntactic features of High and Low Malay of the late nineteenth cen­
tury which are contrasted in this paper are: 1) word order in intransitive
clauses, 2) the use of the empahtic particles -lah and -pun, and 3) the use of
the clause combining words, maka and shahadan. A description of these dif­
ferences as found in several High and Low Malay texts is presented below.

2 The Data
The data examined come from two High Malay and three Low Malay
texts. All texts are written prose, relating a series of events and thus may
broadly be labelled narrative although there are significant stylistic differ­
ences among the texts.
The first High Malay text is a section of the Tuhfat al-Nafis (The Precious
Gift) by Raja Ali Haji ibn Ahmad who was born into a ruling family of the
Riau region of Indonesia. Raj a Ali was well trained in Malay literary arts and
completed this work, which had been begun by his father, in the late 1860s.
The Tuhfat is a history of the Johor-Riau-Lingga area from the late seven­
teenth to the mid-nineteenth century. Parts of the book contain long sections
of genealogies of the royal families, including the major achievements of each
reign. The passage chosen is from one of these sections.
The second High Malay text, Kesah Pelayaran Abdullah (The Voyage of
Abdullah), by Abdullah bin Abdul-Kadir Munshi relates the activities and
impressioris of the author as he sailed up the east coast of Malaya from Singa­
pore carrying letters for merchants. Although Abdullah was born in Malaya,
he was of Arab and Tamil descent and thus learned Tamil and Arabic, as well
as, Malay as a child. As an adult, he became proficient in English and Hindus­
tani. He was a translator for many foreigners and became Raffles' secretary.
He was strongly influenced by British culture and values and this influence is
reflected in the content and style of his writings. It has been claimed that mod­
ern Malay literature began with his writings because he was the first major
 INTRANSITIVE CLAUSES IN HIGH AND LOW MALAY 365

author to embrace the Western literary tradition (Teeuw 1967:1). Western

literature, in which an author expresses his own views, is in contrast with trad­
itional Malay literature, where the cultural values are restated. Another major
difference between these two traditions is the means of transmission of the lit­
erature; Western literature through reading and Malay literature through lis­
tening. Because of Abdullah's proposed significance in the development of
modern Malay literary traditions, it is interesting to note to what extent his
style varies from other early modern Malay writers.
The three Low Malay texts studied here are short stones or novels. One
is an unsigned serialized story in the Chinese newspaper, Tjahaja Timoer,
published in 1917. The other two pieces are: Tjerita Njai Isah (The Story of
Mistress Isah) written by F. Wiggers and printed in 1901, and Boenga Ram­
pai: Berbagai-Bagai Tjerita (Anthology of Various Stories) written by A.F.
Von De Wall and printed in 1890.

3. Word Orders in High and Low Malay

Table one summarizes the differences in word order of the intransitive
clauses of the two High Malay texts and the three Low Malay texts. In the two
High Malay texts there are 311 clauses of which 38% (or 117) are intransitive
with an expressed S. In the three Low Malay texts there is a total of 397
clauses of which 35% (or 137) are intransitive with an expressed S. In the
High Malay data 58% of the intransitive clauses have VS word order and 42%
SV word order, while in the Low Malay data the word order is 33% VS and
67% SV. These data show that Low Malay has moved away from a VS word
order towards a more predominantly SV word order. 2

Table 1. Word Order in Intransitive Clauses.

no. clauses percentages
High Malay
SV 49 42%
VS 68 58%
total 117
Low Malay
SV 92 67%
VS 45 33%
total 137
366 ELLEN RAFFERTY

In both the High and Low Malay sets of texts there is a range of word
order percentages. The High Malay text by Abdullah, a man who knew west­
ern European languages, has a higher percentage of SV clauses than the text
by Raja Ali, a man educated in Malay literary arts. See table 2. It might be
suggested that knowledge of a prestigious SV language promotes a shift to SV
use in one's language. Within the Low Malay texts little is known about the
background and education of the authors, but one can note a difference
between the 1890 work in which 58% of the clauses are SV and in the other
two pieces where the percentages of SV clauses are over 70. This difference
may indicate a drift over time toward a more dominant SV word order, but
other factors, in addition to time, would have to be considered. Other factors
influencing this type of change in language might include exposure to a pres­
tige language with a dominant SV word order and a change in genre or
medium of dissemination of literature.

Table 2. Word Orders in the Five Malay Texts.

HIGH MALAY
no. clauses percentages
Raja Ali SV 24 39%
1860 VS 37 61%
total 61
Abdullah SV 25 45%
1838 VS 31 55%
total 56
LOW MALAY
Von De Wall SV 25 58%
1890 VS 18 42%
total 43
Wiggers SV 38 78%
1901 VS 11 22%
total 49
TjahajaT. SV 27 71%
1917 VS 11 29%
total 38

4. Functions of SV and VS Word Orders in High Malay

In this section a passage of 106 clauses of the Tuhfat text by Raja Ali is
examined in order to see the relationship between the two word orders of
 INTRANSITIVE CLAUSES IN HIGH AND LOW MALAY 367

intransitive clauses and the occurrence of emphatic particles and clause link­
ing words. Becker has noted that the loss of the -lah/-pun structure and closer
syntactic resemblance to European languages are among the important
developments in the transition from classical to modern Malay (Becker
1979:249). These three factors (word order, particles, and clause linking
words) are here treated as a complex through which change occurred to trans­
form classical Malay into modern Malay.
The Tuhfat text by Raj a Ali, rather than the Kesah text by Abdullah, was
selected because of its higher percentage of VS word order and its higher fre­
quency of emphatic particles and clause linking words. See table 3. The more
conservative text, that is the text which shares more features with classical
Malay, is used to present a contrast between VS and SV word orders.

Table 3. Use of Emphatic Particles and Clause Linking Words.

-lah -pun maka shahadan
Kesah N= 24 8 27 1
total no. % = 32% 11% 36% 1%
cls. 74
Tuhfat N= 23 10 12 11
total n. % 46% 20% 24% 22%
cls. 49
(The percentages refer to the number of clauses with a feature compared with
the total number of intransitive clauses in the text.)

Table 3 shows the lower percentages of the emphatic particles in Abdul­

lah's Kesah text as compared with Raja Ali's Tuhfat text. The percentage for
the -lah particle drops from 47% to 32%, while the percentage for the -pun
particle drops from 20% to 11%. With respect to the clause linking words,
shahadan is almost completely absent in Abdullah's text while maka is used
more frequently. The discourse functions of maka are reduced, making its
use similar to that found in Low Malay. (See the section on Low Malay for an
explanation of the discourse function maka.) Although the use of maka
increases in the Kesah text, the general hypothesis drawn from these data is
that VS word order and high frequency use of the particles and clause linking
words are correlated and characterize pre-modern Malay. Before discussing
in detail a section of the Tuhfat text, the particles and clause linking words are
defined.
The linking words, maka and shahadan, are often translated equiva-
368 ELLEN RAFFERTY

lently into English as then, consequently or furthermore; but their functions

in discourse are distinct. Shahadan is characterized by Winstedt (1927:160) as
a punctuation word that is used in written Malay to introduce a new topic or
paragraph, Shahadan can begin a story with the meaning, once upon a time,
or can begin a paragraph with the meaning, next or then. Winstedt states that
shahadan is composed of the Sanskrit word, saha, and the Malay word, dan.
Both of these words mean and, making the word shahadan a compound of a
loan word plus its translation. Although the words have similar English trans­
lations, the functions of shahadan and saha are quite different. In addition to
introducing a new topic, shahadan in older literature was used to connect two
clauses or two adverbial phrases. For example:
(1) Maharaja Ruana karar-lah dengan adilnya
king Ravana establish with justice
shahadan dengan murahan.
and with graciousness
'King Ravana was established with justice and with graciousness.'
(Winstedt 1927:160)
Saha in Sanskrit means with or and in the sense of accompaniment and
neither introduces a new discourse unit nor conjoins two clauses or phrases.
For example:
(2) Raja putr-ena saha grham gacchami.
king son-instr with go home
'The king goes home with his son.' 3
Although shahadan may be derived from the Sanskrit word saha, the dis­
course functions of the two words are very different. Shahadan is a prominent
marker of transitions and begins new units of discourse while saha connects
two nouns.
Maka is an indigenous word4 that links clauses which are conceptually
closely tied to one another. It expresses a range of meanings including tem­
poral, spacial and conceptual sequencing. In contrast to shahadan, maka can
not begin a unit of discourse because it is understood as a non-initial step in
a sequence.
No extensive study of the functions of the -lah and -pun particles has
been made, but Becker has suggested that these emphatic particles mark the
event and topic respectively.5 The -lah particle is most often suffixed to the
verb but other elements of the predicate can receive the -lah suffix. The par­
ticle -pun is typically affixed to a noun but can also occur with an adjective,
adverb or demonstrative. Hopper has suggested that -lah in early modern
 INTRANSITIVE CLAUSES IN HIGH AND LOW MALAY 369

Malay is a focus particle which when attached to a verb is a mechanism for

foregrounding that verb, while the particle -pun is a topic marker (Hopper
1979:227ff.). In the next few paragraphs I examine the occurrence of these
particles and linking words with SV and VS word orders in a High and a Low
Malay text and speculate on the interactions among these features.

5 The Tuhfat al-Nafis

In the 106 clause passage selected from the Tuhfat al-Nafis by Raja Ali,
there are 49 intransitive clauses of which 19 (39%) have SV word order and
30 (61%) have VS word order. All of the 6 presentative verb clauses have VS
word order.
Table 4 summarizes that distribution of the particles (-lah and -pun) and
the linking words (maka and shahadan) in the two word orders.

Table 4 Distribution of Particles and Clause Linking Words in SV and VS

Word Orders.
-lah -pun maka shahadan
SV 6(33%) 10(56%) 5(28%) 1 (6%)
19 cls
VS 17(57%) 0 8(27%) 10(33%)
The percentages in parenthesis indicate the percentage of clauses in that par­
ticular word order that have the feature given in that column.

Two points illustrated in table 4 are 1) the lack of the particle -pun in the
VS clause and 2) the near lack of the linking word, shahadan, in the SV
clause. Also, the VS clauses have a markedly higher percentage of -lah than
the SV clauses. Thus, -pun is characteristic of SV clauses, and shahadan and
the frequent use of -lah are characteristic of VS clauses.
In the Tuhfat section examined, VS clauses are typically introduced by
shahadan and/or maka and the verb is often suffixed with -lah. See sentence
(3).
(3) Shahadan apahila mangkat-lah Sultan Abdul Jalil Shah itu
then when die Sultan Abdul Jalil Shah that
maka Raja Mansor-lah menggantikan kerajaan-nya.
then King Mansor replace throne-the
Then when Sultan Abdul Jalil Shah died King Mansor replaced
him on the throne/
370 ELLEN RAFFERTY

The above shahadan/maka set of clauses should be considered as a unit

because in the passage examined all of the 11 shahadan clauses are followed
by a maka clause. In 7 of the 11 cases the maka clause immediately follows the
shahadan clause, in one the maka is contained in the shahadan clause, and in
the other three instances the maka clause is 2 to 4 clauses away but is still
within the same idea unit. In 9 of the 11 maka clauses following a shahadan
clause the verb is in initial position. In the two clauses where a nuclear argu­
ment precedes the verb that argument is suffixed with -lah or -pun, drawing
attention to this argument. See sentence 4 below.
(4) Shahadan apabila mangkat-lah Sultan Ibrahim itu;
then when die Sultan Ibrahim that
maka baginda itu-pun pindah-lah ka Johor.
then majesty that move to Johor
'Then when Sultan Ibrahim died; his majesty moved to Johor.'
The same generalization applies to the shahadan clauses as applied to the
maka clauses. They both have VS word order or SV word order in which the
particle -pun marks the initial subject. See sentence 5. Thus, both clauses in
the shahadan/maka sequence are typically verb initial.
(5) Shahadan Bendahara Seri Maharaja Tun Pekrama Tun
then prime minister Seri Maharaja Tun Pekrama Tun
Habib-pun mangkat-lah pula.
Habib die also
'Then the prime minister Seri Maharaja Tun Pekrama Tun Habib
died also.'
In the Tuhfat text the shahadan/maka sequence sets a new scene. The
new scenes presented in the passage are the ascent of a new ruler upon the
death of his father. In 8 of the 11 instances of the shahadan/maka set, the
death of one ruler occurs in the shahadan clause and the ascent of his son in
the maka clause. See sentence 2. In two of the remaining 3 sets of clauses the
shahadan clause states that the story continues and the maka clause describes
a new setting. See sentence 6.
(6) Shahadan kata sahibu'l-hika-yat, maka ada-lah
then say story-teller then is
putera baginda itu dengan gundeknya ada tiga orang
son majesty that with mistress is three person
'And so the story continues, his majesty had three sons by his sec­
ondary wives.'
 INTRANSITIVE CLAUSES IN HIGH AND LOW MALAY 371

In the final instance of the shahadan/maka set both linking words occur
in the same clause and report that the reigning king has been given a new title.
The change here is less dramatic and involves only one actor, thus there is
only one clause but two sequencing words. See sentence (7).
(7) Shahadan pada ketika itu
then at then that
maka bergelar-lah ia Sultan Abdul Jalil Shah.
then entitle he Sultan Abdul Jalil Shah
Then at that time he was given the title of Sultan Abdul Jalil
Shah.'
The shahadan/maka complex in this genealogy marks the end of one era
and the beginning of a new one ; in most cases this involves the ascent of a new
king but also includes the setting of a new scene or the change in status of a
major character. These transitions are marked by verb initial clauses and thus
constitute one discourse function of the VS clause.
The S of the VS clause in this text is referential and either definite or
indefinite. The S is never a first mention. 6 Further it is characteristically a
major participant in the discourse. A major participant in this passage is con­
servatively defined as a member of the royal family, that is the king/sultan or
one of his successors. All other nouns are considered minor participants even
if they appear several times in the discourse. Of the 30 Ss in VS clauses 21
(70%) are major participants. This fact gives prominence to the VS clause.
Finally mention should be made of the particle -lah which is found in
57% of the VS clauses. This particle is found as frequently in the shahadan/
maka clauses as in other VS clauses and is found in all lexico-semantic classes
of verbs (presentative, stative and active), indicating that it is not specifically
associated with the sequencing of events. The -lah particle in the VS clause is
associated with the major characters. Of the 17 -lah particles in VS clauses 15
(88%) occur in clauses where the S is a major participant. Of the 21 VS
clauses with a major participant in the S position, 15 (71%) have the -lah par­
ticle. The -lah particle draws audience attention to the VS clause by marking
the relationship between the predicate and the S but does not mark the
sequencing of events.
The most characteristic feature of the SV clause is the presence of the
particle -pun affixed to S. Before discussing the characteristics of -pun, I will
briefly describe the S of the SV clause. It is always referential and in 17 of 19
instances it is definite. Three of the definite Ss are first mentions, indicating
that a new participant may enter the discourse in the S position of the SV
372 ELLEN RAFFERTY

clause. First mentions also occur as patients and agents of transitive verbs, as
objects of prepositions and as predicate nominals. Ten (53%) of the Ss of SV
clauses are major participants (as defined above). This is considerably lower
than the 70% foud in VS clauses. In summary the SV clause may be regarded
as a lighter clause in that it carries less information about the central charac­
ters of the text and is thus more peripheral.
The -pun particle which appears in 10 (56%) of the SV clauses is never
found the VS clause. It is most frequently affixed to the S of the clause but
may also be affixed to the existential, ada. In the three clauses where the -pun
is affixed to the verb, ada, the S is a first mention or of low topicality. An
example of the ada-pun construction is given in sentence 8.
(8) Ada-pun Bendahara-nya Tun Pekrama Habib bergelar
is prime minister Tun Pekrama Habib entitle
Bendahara Seri Maharaja
prime minister Seri Maharaja
'And so the prime minister Tun Pekrama Habib was given the title
to Sri Maharaja.'
In the remaining 7 instances, -pun is suffixed to an NP and in 5 of these
cases the NP is a major participant. In the two cases where the NP is a minor
participant, the noun has just been mentioned in the previous clause and is
thus prominent in that section of the passage. The -pun has a switch reference
function returning an NP to S position or placing a highly topical NP that has
been in non-A or non-S position into S position. The S with the affixed -pun
particle is highly topical and therefore has a low referential distance. 7 The
degree of topicality of S in the SV clauses with -pun ranged from 3 to 12 with
the average being 3.4. The referential distances for the Ss in sentences 13 and
14 below are 1 and 4 respectively.
(9) Shahadan pada satengah chetera,
then at some story
(10) maka kerajaan Raja Abdullah ini-lah,
then rule king Abdullah this
(11) Raja Aceh datang,
king Aceh come
(12) melanggar Johor.
attack Johor
 INTRANSITIVE CLAUSES IN HIGH AND LOW MALAY 373

(13) maka Johor-pun alah-lah,

then Johor defeated
(14) maka baginda itu-pun berundor ka Lingga.
then majesty that retreat to Lingga
'Then according to some stories, during the reign of King Abdul­
lah, the king of Aceh came and attacked Johor. Johor was
defeated and then his majesty (King Abdullah) retreated to
Lingga.'
The -pun in sentence 13 marks the S which in the preceding sentence is
a patient and the -pun in sentence 14 returns the S of sentence 10 to S position.
In these cases, the particle -pun is serving a switch reference function and
occurs most frequently with major participants and/or with participants with
high topicality (low referential distance).
Another less common function of -pun is to note surprise or a notion of
contrary to expectation where there is no switch in subject referent. Sen­
tences 15 and 16 illustrate this function.
(15) Pada masa baginda ini-lah datang langgar
at time majesty this come attack
Peringgi ka Seluyot,
Portuguese to Seluyot
(16) maka Perginni-pun alah.
then Portuguese defeated
'During his majesty's reign the Portuguese came and attacked
Seluyot, and then the Portuguese were defeated.'
Although the Portuguese is the subject of both sentences 15 and 16 and
therefore does not need to be repeated, it is repeated with the particle -pun
in 16, indicating that the statement carries counter to audience expectation.
A few other characteristics of the SV clause should be noted. The maka
in the SV clause serves the same sequencing function as it does in the VS
clause, but it is never used in conjunction with the shahadan clause to mark
major breaks in the discourse. The -lah particle, which draws attention to the
verb or an element in the predicate, never occurred alone in an SV clause in
the data but always co-occurred with the particle -pun. In this way, the pred­
icate of an SV clause is never singled out as the one significant element of the
clause. This fact indicates the less forcefull status of the predicates of SV
clauses.
374 ELLEN RAFFERTY

Several factors conspire to place the SV clause in a less prominent posi­

tion in the Malay discourse of the Tuhfat al-Nafis. This is achieved by
associating the SV clause with minor characters, with new subjects, and with
the use of the emphatic particle -pun.

6. Topicality of SV and VS Clauses

Comparing the SV and VS clauses with respect to the topicality measure
of referential distance produces another means of assessing discourse func­
tions within a text. After excluding the 6 presentative clauses of the VS cate­
gory, all of which have a referential distance of 20, we are left with two groups
of VS clauses: those associated with the turning points in the discoure, that is
the shahadan/maka sets of clauses, and those clauses following these sets. The
clauses of the shahadan/maka sets have long referential distances and thus
low topicality. The shahadan clause has low topicality because it returns to
the reigning king who was introduced at the last major break. The maka
clause introduces the new king, whose name is likely to have been previously
mentioned but still has low topicality at this point in the text. The second
type of VS clause immediately follows the maka clause and elaborates on the
new king, his achievements and his associates. The Ss of these clauses have
low referential distances and high topicality because they build on the newly
introduced king. The section is ended with another shahadan/maka set, relat­
ing the death of the king and the ascent of his son. The rhythm that is devel­
oped is one of VS clauses of low topicality at major breaks and VS clauses of
high topicality immediately following these breaks. The average referential
distance of the 9 VS clauses associated with the shahdan/maka sets is 9.1,
while that of the other 15 VS clauses is 1.53. The VS clauses can not be charac­
terized by the degree of topicality of their subjects but rather by their func­
tions. VS clauses are used for presentatives, for clauses at major breaks in the
narrative; and for the development of events involving major participants. To
group all VS clauses together would mask the wide differences in topicality
and the interaction between degree of topicality and the clause linking words.
In this case, the interaction is an important structural feature of the discourse.
The S of the SV clause, on the other hand, has a more uniformly low
degree of topicality. The first mentioned NPs, of course, are very low in topi­
cality, while the Ss with the -pun affix tend to have high topicality. The dis­
tribution of the level of topicality follows no discourse structure in the text.
The average referential distance of the Ss of the 18 SV clauses is 9.43, sup­
porting other indications that the SV clause is less topical and less prominent
 INTRANSITIVE CLAUSES IN HIGH AND LOW MALAY 375

in the development of the major characters. It is associated with new and/or

minor characters which generally have lower levels of topicality.

7. The Functions of SV and VS Word Orders in Low Malay

Low Malay, like High Malay, has SV/VS word order variation, but in
contrast to High Malay, the dominant word order in Low Malay is SV (not
VS). Low Malay also differs from High Malay in that it shows a very limited
use of the emphatic particles -lah and -pun and the clause linking words,
shahadan and maka.
A passage from F. Wiggers' novel, Njai hah, is examined in order to dis­
cover the different functions of SV and VS word orders and their relations to
the emphatic particles and clause linking words. Little is known about Wig­
gers' early life or education, but we do know that he was a civil servant in the
Dutch East Indies and later became an active journalist, writer and translator
(from Dutch to Malay) (Pramoedya 1928:17-18). The novel, Njai hah, was
written at the turn of this century for the popular entertainment of a multi­
ethnic readership. The passage chosen here relates the events in the life of a
young Dutch woman in the home of a local Indonesian family after she has
left her family home because she has become pregnant.
The NjaiI s a hpassage chosen has 141 clauses of which 49 are intransitive
with an expressed S. Of these 49 clause, 38 (78%) have SV word order and 11
(22%) have VS word order. The distribution of the emphatic particles, -lah
and -pun, and the clause linking words, shahadan and maka, in the two word
orders is given in table 5.

Table 5. Distribution of Particles and Clause Linking Words in SV and VS

clauses.
-lah -pun maka shahadan
SV 0 1(3%) 4(11%) 0
38 clauses
VS 3(27%) 0 1 (9%) 0
11 clauses

The use of the emphatic particles which in High Malay distinguished SV

from VS word orders and the use of the clause linking words that gave hierar­
chical organization to the High Malay texts are insignificant in the Low Malay
data.
376 ELLEN RAFFERTY

In the Njai Isah passage selected the SV clause dominates and the choice
of word order here appears to be based on the definitiness of the subject. See
table 6. Structurally, the clause proceeds from definite S to verb or from verb
to indefinite S. In the few instances where this sequencing of information is
not followed the statement carries information that is counter to general
expectation.

Table 6 Definiteness of S in Low Malay Intransitive Clauses.

definite S indefinite S
SV 39 0
VS 4 7
total 43 7

Seven of the 11 VS clauses have indefinite Ss; the three with definite Ss
ail occur with the verb ber-kata 'say'. This verb is used with VS word order
despite its definite S because verb-initial clauses imply that new information
follows the verb and in the case of the verb ber-kata the important new infor­
mation is indeed contained in the clausal complement which follows the VS
clause. See sentence 17.
(17) maka ber-kata poela njonja Abrams,
then say also Mrs. Abrams
Angkau, kasian masih moeda tida taoe apa artinja,
you pity still young not know what meaning
Then Mrs. Abrams also said, 'You poor dear, you are still young
and do not know the meaning (of this).'
Frequently in written and spoken discourse the subject of (ber-)kata
becomes cliticized yielding, kata-nya. This illustrates the strong tendency for
the post-verbal position of the S with this verb.
Of the remaining 8 VS clauses, 7 have indefinite Ss. Two of these 7 inde­
finite Ss are clausal subjects such as, in sentence 18.
(18) Tetapi baik nona djangan terlaloe bergerak-gerak.
but good miss not too move move
'But it is good if the young woman is not too active. '
Clausal subjects do not occur in SV word order, again demonstrating the
need for the new information of the clausal subject to appear in post-verbal
position. An example of the 5 remaining VS clauses is found in sentence 19.
 INTRANSITIVE CLAUSES IN HIGH AND LOW MALAY 377

(19) Serta timboel-lah djoega pengrasaan.

and rise also feeling
'And a feeling also arose (in her).'
The one other instance of VS word order occurred with a definite sub­
ject. In sentence 20 this clause has counter to expected word order and lexi­
cally marked surprise.
(20) Wah, semangkin bertjahaja moeka non Poppi,
Wow the more radiant face ms. Poppi
'Wow, Ms. Poppi's face is all the more radiant.'
Although there are not enough examples in these data to generalize, it
is reasonable to hypothesize that one function of the VS word order with defi­
nite Ss is to indicate surprise or counter to expected information.
In the SV clauses of this passage, -lah is never used and -pun is used only
once. The -pun is affixed to a continuous NP (an NP with a referential dis­
tance of 1). See sentence 21.
(21) Dalem itoe ampat lima hari non Poppi soedah bisa
in that four five day ms. Poppi already can
sekali bawa adat orang oedik,
very carry custom people village
dia-poen soedah beladjar-beladjar omong Sunda.
she already study study speech Sunda
Tn those four or five days Ms. Poppi was able to pick up the ways
of the villagers, and she even began to study Sundanese.'
The emphatic particle -pun (here spelled poen) is used in the second
clause of 21 with a continuous subject which is ordinarily indicated by zero
anaphora. This repetition gives the statement an element of surprise. This
usage is similar to the second function of -pun in the High Malay texts.
The four instances of maka have the same sequencing function as was
found in the High Malay texts. In the Low Malay texts there is no use of the
linking word, shahadan, or any other marker of paragraph-sized units. The
Low Malay texts thus have a flatter, more linear development to their plots.
The lack of paragraph-sized markers does not imply a paucity of types of
clause linking words. Indeed, the variety of such words is greater in the Low
Malay text than in the High Malay text examined. In the 141 clauses there
were 14 different clause linking words with a variety of meanings more exten­
sive than those found in the High Malay texts. Several of the additional link-
378 ELLEN RAFFERTY

ing words are derived from verbs, such as lantas (to penetrate) meaning next,
sampai (to arrive) meaning until, and lalu (to pass by) meaning then.
In summary the Low Malay discourse is characterized by SV clauses
where the Ss are definite, producing a pattern of information flow of old to
new. A typical intransitive clause in Low Malay resembles the one in sentence
22.
(22) Maka Njonja Abrams pegi ka roemah makan.
then Mrs. Abrams go to house eat
Then Mrs. Abrams went to the restaurant.'
The VS clause in Low Malay is used in clauses with indefinite subjects
including clausal subjects, with definite subjects in clauses with the verb ber-
kata 'to say', and in clauses where an element of surprise is desired. In Low
Malay the particles -lah and -pun are rarely used. Finally the Low Malay dis­
course is flatter because it lacks lexically speciallized paragraph markers.

8. Conclusion
The two contemporaneous types of Malay used in literature at the turn of the
twentieth century, Low and High Malay, show markedly different word
order patterns and different discourse functions of the two word orders. In
addition, the two types of Malay differ with respect to the use of the clause
linking words, shahadan and maka, and the emphatic particles, -lah and -pun.
The choice between VS and SV in High Malay is based upon several fac­
tors all conspiring to draw audience attention to the VS clause. The VS
clause appears in the presentative clause, the shahadan/maka clause com­
plex that marks major breaks, and in clauses following these major breaks
that relate developments concerning the major participants. The SV clause,
in contrast, is used for first mentions, especially for minor characters, and is
characterized by the use of the emphatic particle -pun on the S. Through the
use of the shahadan/maka complex, High Malay establishes a rhythm
wherein the VS clauses are associated with major changes and major
developments, while the SV clauses are associated with minor characters and
minor events or states.
The use of the linking words, shahadan and maka, which create a hierar­
chical structure in High Malay texts, I suggest, is a mechanism more appropri­
ate for literature that is intended to be listened to than for literature that is to
be read. With oral literature one can not assume continuous audience atten­
tion, therefore the language must develop mechanisms for informing the
 INTRANSITIVE CLAUSES IN HIGH AND LOW MALAY 379

audience as to the progress of the plot. 8 One manner of marking progres is by

the use of a formulaic set of words or phrases to mark off time or scenes in the
story. This appears to be the function of the shahadan/maka complex.
Windstedt's definition of shahadan as a punctuation word, emphasizing its
structural importance over its semantic content lends support to such an
analysis of the discourse functions of the clause linking words. The goals of lit­
erature, the genre, and the traditions out of which a literature develops all
contribute to shaping the form of its discourse structure. Each genre and each
style of language must therefore be examined individually in order to dis­
cover the particularities of its discourse. The assumption that a language will
employ similar discourse strategies in diverse genres is untenable.
Low Malay in the late nineteenth century emerged from an oral language
that had no literary traditions. The early Low Malay literature was written by
Dutch, Eurasians and Chinese and often was translations of Chinese, Dutch
or other European works. The over-arching tradition for most Low Malay
works was the western European novel or short story. The literature was
intended to express the opinions of the author (not to state cultural values as
traditional literature did) and was to be read (not listened to). The goals of a
literature and the traditions out of which it emerges contribute in shaping the
form of its discourse. In the case of Low Malay, the linking words such as
shahadan and maka were no longer needed, and a greater variety of linking
words appropriate for written literature developed. The shift from a domin­
ant VS to a dominant SV word order may also be attributed to the change
from an oral to a written tradition where a distinction between major and
minor characters is not as important as in oral literature. The role of outside
languages on the shift from VS to SV must be considered and these outside
languages must include not only the western languages, Dutch and English
but also Javanese which had shifted to SV word order several centuries before
this period.
This study of two closely related styles of Malay at the turn of this century
points out the need to study discourse structures of distinct styles and genres
separately. The assumption that a language will employ similar patterns of
organization even in such fundamental areas as word order can not be
assumed. The structure of a language is too closely tied to its social functions
and context to allow similarities across genre and style to be taken for
granted.
380 ELLEN RAFFERTY

NOTES
1) For definitions of a number of varieties of Malay see Rafferty 1984a:251-252.
2) The reason I say that Low Malay has moved away from VS word order is because in both
High and Low Malay texts the older texts show proportionately greater use of VS word order than
the more recent texts. Also, among colloquial Malay speakers today, the uneducated speakers
(therefore the ones less affected by Indonesian) use a high percentage of VS word order. There­
fore, I conclude that VS is the more conservative word order.
3) I am grateful to Frances Wilson and Manindra Verma for explaining the use of saha to me.
F. Wilson provided me with the example sentence. Saha may be used with the instrumental case
as is shown in the example or with the accusative or genitive cases.
4) Maka may be related to the Javenese word, mangka, meaning but, whereas or now.
5) Becker 1979:248-250. Topic is here used in the Thompson and Li 1976 sense, not the Givón
1983 sense of topicality.
6) Not included in the first mention category are the nouns that are implied by a frame.
7) Î am using the term referential distance as defined by Givón 1983. The longest referential dis­
tance is arbitarily set at 20.
8) See Sweeney 1980 for a discussion of how this phenomenon relates to the structure of classi­
cal Malay literature.

REFERENCES

Secondary Sources
Alisjahbana, T. 1957. Dari Perjuangan dan Pertumbuhan Bahasa Indonesia.
Djakarta: P.T. Pustaka Rakjat.
Becker, A.L. 1979. 'The figure a sentence makes: An interpretation of a
classical Malay sentence." In: T. Givón ed., Discourse and Syntax (Syntax
and Semantics vol. 12) New York: Academic Press.
Givón, T. ed. 1983. Topic Continuity in Discourse: A Quantitative Cross-lan­
guage Study. Amsterdam: J. Benjamins.
Greenberg, J.H. 1963. "Some universals of grammar with particular refer­
ence to the order of meaningful elements." In: J.H. Greenberg, Univer­
sals oƒ Language. Cambridge, Mass.: MIT Press.
Hopper, P. 1979. "Aspect and foregrounding in discourse." In: T. Givón
ed,, Discourse and Syntax (Syntax and Semantics vol. 12) New York:
Academic Press.
Li, C. ed. 1976. Subject and Topic. New York: Academic Press.
Li, C. and Thompson S. ed. 1976. "Subject and topic: A new typology of
language.) In C. Li ed., Subject and Topic. New York: Academic Press.
 INTRANSITIVE CLAUSES IN HIGH AND LOW MALAY 381

Monier Monier-Williams, Sir 1899. A Sanskrit-English Dictionary. Oxford:

Clarendom Press.
Pramoedya, Ananta Toer. 1982. Tempo Doeloe. Jakarta: Hasta Mitra.
Rafferty, E. 1984a. "Languages of the Chinese of Java: An historical
review". Journal of Asian Studies 43(2): 247-272.
--—. 1984b. "The emergence of Indonesian from High and Low Malay of the
late nineteenth century". Paper at the SEAsian Summer Institute, Ann
Arbor, Michigan, August 2-4.
Sweeney. Amin 1980. Authors and Audiences in Traditional Malay Litera­
ture. Berkeley: U. California, SEAsia Center monograph series no. 20.
Teeuw, A. 1967. Modern Indonesian Literature, vol. 1. The Hague: Martinus
Nijhoff.
Winstedt, R.O. 1927. Malay Grammar. second edition, Oxford: Clarendon
Press.
. 1969. A History of Classical Malay Literature. Kuala Lumpur: Oxford
U. Press.
Wolff, J.U. 1980. Formal Indonesian. Ithaca, N.Y.: Cornell U. SEAsia Prog­
ram.

Primary Sources
Ali Haji Ibn Ahmad, Raja. 1965. Tuhfat al-Nafis. Singapore: Malaysia Pub­
lications Ltd.
Abdullah bin Abdul-Kadir Munshi. 1928. Kesah Pelayaran Abdullah. Singa­
pore: Malay Publishing House Ltd.
Tjahaja Timoer. March 2, 1917 issue, Malang, East Java.
Von De Wall, A.F. 1890. Boenga Rampai: Berbagai-Bagai Tjeritera. Batawi:
Pertjitakan Goebernemen.
Wiggers, F. 1901. Tjerita Njailsah. djilid ka-ampat, Betawi: Albrecht & Co.
 THE FUNCTIONAL DISTRIBUTION OF PREPOSED AND
POSTPOSED "IF" AND "WHEN" CLAUSES
IN WRITTEN DISCOURSE

VIOLETA RAMSAY
University of Oregon

1. Introduction
The study of the different types of adverbial clauses in the past has been
done almost exclusively by logicians, who analyzed the relationship between
the adverbial clause and the main clause in terms of truth value, material
implication, presupposition, etc. This merely semantic analysis did not allow
for notions such as organization and content of the discourse, communicative
intent or pragmatic motivations of speaker and hearer (Givón 1984:252-4).
Since these notions did not become the concern of linguists until recently, we
still do not have a clear undertstanding of the behaviour of the different types
of adverbial clauses in discourse.
One of the most recent linguistic analysis of adverbial clauses is
Haiman's (1978). This study analyzes cross linguistic data and suggests a
relationship between topics and conditionals, arguing that in terms of gram­
matical marking and in terms of function, conditionals behave like topics of
their sentences. But although he is dealing with a notion as topic, which can
only be analyzed at the discourse level, Haiman presents sentences in isola­
tion and centers the argumentation on the causal connection between propo­
sitions. There is also the problem that he does not differentiate between
clauses that are preposed and those that are postposed to the main clause, and
as this paper will propose, preposed clauses perform a different job in dis­
course than do postposed ones.
Thompson's (to appear) paper on the distribution of purpose clauses in
English texts is one recent attempt at defining the distribution in discourse of
subordinate clauses. She analyzes the discourse factors that determine
whether a purpose clause will be placed before or after its main clause and
384 VIOLETA RAMSEY

concludes that initial and final clauses are performing different functions in
discourse. Initial clauses are functioning to guide the attention of the reader
by signaling how he/she is expected to associate the material following the
purpose clause with the material preceding it. The final clause does not have
the same function but has a very localized and different one: it only serves to
state the purpose for which the action named in the main clause is underta­
ken.
Marchese (this volume) studies the function of conditional clauses at the
discourse level too, based on data from procedural discourse from Godie.
She determines that conditional clauses have a recapitulative function in that
they summarize what has come before, working in this way as 'topics1 and
sharing characteristics with noun topics.
Chafe (1984) suggests that adverbial clauses vary their functions with
respect to two factors. One has to do with their position with respect to the
main clause: the other has to do with how tightly the adverbial clause is bound
to its main clause.
Another work concerning this type of phenomenon of 'pragmatic posi­
tioning' is Fagerber's (1983) study of the use of cleft and pseudocleft sen­
tences in Pulaar. It shows how the notion of discourse connectedness serves
to explain the choice of one of these two focus patterns in a given context. She
says that an emphasized NP is positioned at the beginning of the sentence
when it is 'connected' to something which has preceded it in the discourse,
whereas the emphasized element is postposed if it is connected to something
which is still to come.
The above mentioned studies offered good insights into the behaviour of
different types of adverbial clauses and most of them suggest, to one degree
or another, that the positioning of some types of adverbial clauses is dictated
by the organization of information in discourse. However, most of these
works are based almost exclusively on intuitive observations or present a
minimum of empirical evidence but as Givón (this volume) proposes, we
need to devise operationalized discourse measurements in order to test the
empirical validity of our hypotheses.

2. Hypotheses
The goal of this study is to perform a contextual and quantitative analysis
of the distribution of 'if clauses (IC) and 'when' clauses (WC) in order to
determine what are the discourse factors that dictate their positioning vis-a­
vis the main clause. I suggest that:
 PREPOSED AND POSTPOSED IF' AND 'WHEN1 CLAUSES 385

In written discourse, preposed IC and WC perform different pragmatic

functions than those of postposed IC and WC:
a) Preposed IC and WC are thematically associated to the preceding dis­
course as well as to the main clause, thus have a broader scope. Post-
posed IC and WC are only related to their main clause, thus have a very
localized scope.
b) Preposed WC form an integral part of the narrative in that they help
to sequence the temporal contour of the main line of the narrative
whereas postposed WC only function as added comments to the main
clause.

3. Method
3.1. Data base
I used a modern murder mystery novel for this analysis of English written
narrative: Death in the High C's by Robert Barnard.

3.2. Data collection

For most of the counts performed for this study I considered separately
the discourse of the narrative (narrator) from the discourse in quoted speech
(participants). I found a total of 109 IC and 95 WC distributed in the following
way:

Table I

IF CLAUSES WHEN CLAUSES

Preposed Postposed Preposed Postposed

Narrator 12 8 42 27

Participants 59 30 17 9

Totals 71 38 59 36
386 VIOLETA RAMSEY

The only clauses considered for this analysis are those that relate to their
main clauses as adverbial clauses. Thus I did not consider relative or comple­
ment clauses such as in the following examples:
(1) a. Little Mr. Pettifer, the repetiteur, was seated at the piano and
told when he could start (p. 12)
b. As usual at moments when no underlining was needed, Owen
felt he had to make things...(p.13)
c. 'Perhaps you should warn her,' suggested Raymond, his
heavy-lidded eyes watching to see if his humour got through.
(p.25)
Neither did I consider the 'as if or 'if only' type clauses, for example:
(2) a. Gaylene looked at him for some moments, as if quite unsure of
what he was trying to say. (p.26)
b. and was beginning to feel in total control of herself, If only the
voice comes good, she said to herself (p.68).
The reason for this exclusion is that this type of clauses generally does
not have a consequent. This is also the case with the 'meta-comment' type of
clauses, which were also excluded. This refers to those comments that partic­
ipants of the story add sometimes to what they just said (a lot of them 'if
clauses). An example follows:
(3) 'Most of the time we have to fight Owen Caulfield to get to do what
we want, but that's by the way. The point is that the three of us —
and Ricci, too — are really trying to do justice to the piece, if that
doesn't sound too pompous. (p. 138)
In order to assess the different functional distribution in discourse of ini­
tial and final IC and WC, the following types of measurements were
employed:
a) Referential distance — for subjects only — for preposed and post-
posed IC and WC. This type of measurement was also performed for
subjects of those main clauses which have their adverbial clause preced­
ing the main clause.
b) Scope in number of clauses to the left, for the four types of adverbial
clauses.
c) Aspect-Modality categorization of verbs used in 'when' clauses.
d) Distribution of commas for all IC and WC.
e) Use of IC and WC in paragraph breaks.
 PREPOSED AND POSTPOSED 'IF' AND WHEN' CLAUSES 387

The results of all these measurements are given in the tables below where
I present them in such a way that the results of measurements performed for
initial clauses appear side by side to those obtained from final clauses. The
motivation for this presentation is to let the reader compare the behaviour of
initial clauses with that of final clauses.

3.2.1. Referential distance

I used the method introduced by Givón (1983d) for assessing topic con­
tinuity, except that I only did it for subject NPs. I counted the number of
clauses (or gaps) to the left between the appearance of the subject NP in the
four types of clauses examined, and the previous appearance of a reference
to that same NP, regardless of what grammatical device marked that previous
appearance.
In counting clauses I followed these restrictions:
a) Non-finite verb clauses and relative clauses were not counted as gaps,
although the appearance of the previous reference inside any of these
clauses was considered.
b) As explained above, the speech of the narrator and that of the partic­
ipants of the story were considered separately. For example, in measur­
ing referential distance for the subject of a clause used by the narrator.
I considered as gaps only clauses in pure narrative, and not those in
quoted speech, although the appearance of a referent therein was
counted. And viceversa, if the clause to be considered was used by one
of the participants, only clauses in quoted speech were counted as gaps.
c) Those clauses whose subject NPs do not have a specific previous ref­
erence in the preceding discourse, are not considered on the tables that
show the results of these measurements. This is the case of names of par­
ticipants that appear only once in the novel: of indefinite NPs such as
'anyone' or 'people': of first and second person singular pronouns: and
of non-referential pronouns such as 'it' or of non-referential NPs such as
in the following example:
(4) When rationality and skepticism regained some foothold in his
mind, it occurred to him that...(p. 186)
I also counted referential distance on those main clauses that have their
adverbial clause preceding them (this includes IC and WC) so as to be able to
compare the referential continuity of both initial and final clauses with
respect to the main clause. That is, since the counts on referential distance are
388 VIOLETA RAMSEY

done to the left, we can not analyze the thematic connection that exists
between preposed clauses and the main clauses when we do referential dis­
tance on the former. Thus performing referential distance on the main clauses
we can have a way of comparing the thematic link that exists between initial
and final clauses vis-a-vis the main clause.

3.2.2. Scope
This type of measurement was also performed in order to assess the con­
nection that exists between the adverbial clause and the previous context. I
counted the number of clauses to the left of each clause, that were needed in
order to understand the whole clause. The number of 20+ clauses was
assigned arbitrarily if the scope of a specific clause exceeded that number. For
this restriction I am also following Givón's (1983d) methodology. I also
excluded from this measurement non-finite verb clauses, as I did in counting
referential distance. For an illustration of what I mean by 'scope' two exam­
ples follow, the first one with a scope of two clause, the second with a scope
of eleven:
(5) a. The early rehearsals were to be in Italian. Mike Turner, the
company's director, who was to conduct, was insistent on this.
If they knew the Italian, this would smooth over the early
rehearsals with Giulia Contini...(p.l2)
b. Everyone went back to first positions, and Mike began the
scene again. It had been perfectly sensible advice but the pati­
ence was exaggerated, and as Barbara stood in the wings she
remembered only that 'well, never mind.' and she mulled over
to herself the variety of wounding phrases it could have been
designed to hide. She was a Lancashire girl, and she hated sar­
casm, as she hated condescension, and she stiffened with
resentment. When her moment came she marched purpose­
fully forward....(p. 162)
Notice that in order to give sense to the underlined IC or WC one needs
to look back at a number of clauses. This looking back' is similar to what we
do in counting referential distance, except that in this case one looks for a pre­
vious reference for an entire clause.

3.2.3. Aspect-Modality categorization of 'when' clauses

In order to assess the different discourse roles of initial WC, as compared
 PREPOSED AND POSTPOSED 'IF' AND 'WHEN' CLAUSES 389

with final WC. I categorized them in two groups, in terms of the aspect and
modality of their verbs. The two distinctions are: Perfective aspect/realis
modality — that includes only simple past tense — and Imperfective aspect/
Irrealis modality — that includes present and future tenses, progressive,
habitual and perfect aspects and the conditional. For this categorization I am
following Givón's (1984:287-9) suggestions for discourse foreground/
background correlations of tense-aspect-modality.
I had two reasons for not considering tense. One is that the distinctions
of time are different for the narrator and for the participants, that is, the rela­
tive time for the participants is that of the story, logically; but that of the nar­
rator is a later one from which he looks back at the story and describes it to
the reader. Derived from this fact we get different grammatical tenses
describing the same semantic actions. Some examples will clarify this point:
(6) a. 'Not so much of this Eyetalian,' said Gaylene French, stomp­
ing over to the far end of the hall and dumping down on a stray
chair the bulging plastic bag that contained here sustenance for
the day. 'We're doing the thing in English, guest star not­
withstanding.' (p.11)
b. If they knew the Italian, this would smooth over the early
rehearlsals with Giulia Contini when she arrived, and in any
case a thorough knowledge of the text helped singers to pro­
ject the words of the translations. They would change to the
English version when la Contini had been to a few rehearsals.
(p.12)
In the first example it is one of the participants who says the performance
of the opera will be in English and the verb is in the present-continuous. In the
second example it is the narrator who says the same thing and the verbs used
in the sentence are in the conditional and past perfect. In both cases it is the
same action that has not yet (in the story) taken place (imperfective/irrealis).
The other reason is that one same grammatical tense can describe differ­
ent aspectual distinctions. The following example includes two sentences in
which the narrator uses the past tense in all verbs but the first sentence
describes an action tha thas not yet occurred (imperfective) and the second
describes two actions that happened as the story unfolded (perfective):
(7) If they knew the Italian, this would smooth over the early rehear­
sals with Giulia Contini When she arrived and in any case a
thorough knowledge of the text helped singers to project the words
of the translation. They would change to the English version when
390 VIOLETA RAMSEY

la contini had been to a few rehearsals. So when Simon Mulley, the

veteran Rigoletto, began the apparently common place dialogue
with Gilda that begins Act III, he said:...(p. 12)

3.2.4. Distribution of commas

I counted the number of clauses in which a comma separates the adver­
bial clause from the main clause, in each category. The motivation was to find
out whether preposed and postposed clauses pattern differently in this
respect. If preposed clauses show a higher thematic continuity with the pre­
ceding discourse than with the main clause, then they are more likely to have
a comma. If postposed clauses are thematically connected to the main clause
then they are likely to go without a comma.

3.2.5. Distribution of IC and WC in paragraph breaks

I counted the number of clauses, of each category, that appeared in the
first sentence of a paragraph: the motivation being that since it is at paragraph
breaks where thematic discontinuities are likely to appear then, of all
categories, preposed WC will be used more frequently in that position than
any other clause type.

4. Results of measurements
4.1. Referential distance
The comparison of the results of measurements of referential distance
show the following: The majority of subjects of final clauses find their previous
reference in the clause immediately preceding, that is, the main clause, while
initial clauses find their previous reference in one of the various clauses pre­
ceding them. The results are shown in the tables below.
- Table II shows the results of referential distance for initial and final 'if
clauses used in narrative. One can see that while 100% of subjects of the final
clauses have a value of 1, that is they find their previous referent in the main
clause, that is not the case for initial clauses. 60% of initial clauses have a
value of 1 and the other 40% have varied values from 2 to 13 clauses.
- Table III shows initial and final 'if clauses used by participants. Here we find
that more than 90% of final clauses have a value of 1 while in the case of initial
clauses only 41.9% of the clauses have a value of 1 : the rest have varied values
from 2 to 13 clauses.
- Table IV shows the measurements of referential distance for initial and final
 PREPOSED AND POSTPOSED 'IF' AND 'WHEN' CLAUSES 391

'when' clauses used in narrative. Here again the results show roughly the same
pattern. Almost 95% of final clauses have a value of 1 whereas in the case of
initial clauses we see a much broader distribution. 43.2% have a value of 1
and the other 56.8% have varied values of 2 to 20+ clauses.
- Table V shows initial and final 'when' clauses used by participants. We find
that 66.6% of final clauses have a value of 1 whereas for initial clauses only
44.5% have that value. The rest of initial clauses vary from 2 to 20+ clauses.
- Table VI and Table VII show the results of referential distance for subjects
of those main clauses that have a preposed IC or WC. These measurements
were performed, as explained above, in order to assess the thematic link that
exists between the main clause and the preposed clause. One should notice that
although the percentage of main clauses whose subjects NP's find their previ­
ous reference in the preposed adverbial clause appears to be high, there is a
large number of subjects that are not considered in those percentages. For
example, for IC used in narrative, 5 subjects were not considered; for IC used
by participants as many as 36 were not considered, and so on. This is due to
the fact that those NP's are of the type that usually do not find a previous ref­
erence in the previous discourse (proper names, indefinite NP's. 1st, and 2nd
person singular pronouns, etc.). On the other hand, if one looks at the other
tables on referential distance one can see that they do not have as many of
those type of subjects.
What these last results show is that on the one hand the thematic link of
preposed clauses is much broader than of postposed clauses, and on the other
that preposed clauses show lower referential continuity vis-a-vis the main
clause than do postposed clauses.
392 VIOLETA RAMSEY

Table II
Initial and final 'If clauses used in narrative
Referential distance for subjects
INITIAL CLAUSES Distance FINAL CLAUSES
N % to total N % to total N N
of clauses of clauses
6 60 1 100 6
1 10 2
1 10 3
1 10 4
/ / 5
/ / 6
/ / 7
/ / 8
1 10 9-13
14-20+
10 100 100 6
clauses whose subjects don't
have a previous referent
2 (not counted in percentages) 6
12 8

Table III
Initial and final 'If' clauses used by participants
Referential distance for subjects
INITIAL CLAUSES Distance FINAL CLAUSES
N % to total N % to total N N
of clauses of clauses
18 41.9 1 91.3 21
11 25.6 2 8.7 2
5 11.6 3
2 4.7 4
/ / 5
1 2.3 6
3 7 7
1 2.3 8
1 2.3 9-13
1 2.3 14-20
43 100 100 23
clauses whose subjects don't
have a previous referent
16 (not counted in percentages) 7
59 total N of clauses 30
 PREPOSED AND POSTPOSED ' I F AND 'WHEN' CLAUSES 393

Table IV
Initial and final 'when' clauses used in narrative
Referential distance for subjects
INITIAL CLAUSES Distance FINALCLAUSES
N % to total N % to total N N
of clauses of clauses
16 43.2 1 94.7 18
7 18.9 2 / /
5 13.6 3 5.3 1
1 2.7 4
3 8.1 5
1 2.7 6
1 2.7 7
1 2.7 8-10
1 2.7 11-14
/ / 15-19
1 2.7 20+
37 100 100 19
clauses whose subjects don't
have a previous referent
5 (not counted in percentages) 8
42 total N of clauses 27

Table V
Initial and final 'when' clauses used by particpants
Referential distance for subjects
INITIAL CLAUSES Distance FINALCLAUSES
N % to total N % to total N N
of clauses of clauses
4 44.5 1 66.6 4
2 22.2 2 16.7 1
1 11.1 3 16.7 1
/ / 4
/ / 5
1 11.1 6
/ / 7
/ / 8-10
/ / 11-14
/ / 15-19
1 11.1 20+
9 100 100 6
clauses whose subjects don't
have a previous referent
8 (not counted in percentages) 3
17 total N of clauses 9
394 VIOLETA RAMSEY

Table VI
Referential distance for subjects of main clauses
('IF' CLAUSES)
NARRATIVE Distance PARTICIPANTS
N % to total N % to total N N
of clauses of clauses
5 83.3 1 73.9 17
1 16.7 2 17.4 4
3 / /
4 4.3 1
5 4.4 1
6-20+
6 100 100 23
clauses whose subjects don't
have a previous referent
5 (not counted in percentages) 36
12 total N of clauses 59

Table VII
Referential distance for subjects of main clauses
('WHEN' CLAUSES)
NARRATOR Distance PARTICIPANTS
N % to total N % to total N N
of clauses of clauses
21 91.3 1 87.5 7
2 8.7 2 12.5 1
/ / 3
/ / 4
5-13
14-20+
23 100 100 8
clauses whose subjects don't
have a previous referent
19 (not counted in percentages) 09
42 total N of clauses 17
 PREPOSED AND POSTPOSED 'IF' AND 'WHEN' CLAUSES 395

4.2. Scope
As explained above, this measurement is similar to referential distance
but instead of looking for a previous referent for the subject NP, here the
entire clause is measured within the context in which it appears. The tables
below show that the scope of final IC and WC is much more restricted than
that of initial clauses. The results are the following:
Table VIII shows the results of scope in previous discourse for initial and
final IC used in narrative. We find here that 87.5% of final clauses have a
scope of 1 to 3 clauses. On the other hand, initial clauses show a scope of 1 to
3 clauses in 66.7% of the cases and the rest show values of 6 to 20+. The aver­
age scope for final clauses is 2.3 clauses, whereas the average scope for initial
clauses is 6.3.
Table IX shows the results of scope for initial and final IC used by partici­
pants. Here we can see that 83.4% of final clauses have a scope of 1 to 3
clauses, whereas initial IC show a much broader distribution: Only 6.7%
have a value of 1 to 3 clauses while all the rest get varied values of 4 to 20+,
in a fairly even distribution. The average scope for final clauses is 2.1 clauses,
while that for initial clauses is 10.5.
Table X shows the results of scope for initial and final WC used by the
narrator. We also find here that postposed clauses have a much restricted
scope than preposed clauses. 77.8% of postposed clauses have a scope of 1 to
3 clauses while in the case of preposed clauses only 13.8% get values of 1 to 3
clauses; the other 66.2% get varied values of 4 to 20+ clauses. The average
scope for final clauses is 3.2 clauses while that for initial clauses is 7.8.
Table XI has the results of scope for initial and final WC used by partici­
pants. As expected, here we also find that for postposed clauses 66.7% of the
total number of clauses have a scope of 1 to 3 clauses whereas preposed
clauses show a value of 1 to 3 clauses in only 6.2% of the cases; all the rest get
varied values of 4 to 20+ clauses. The average scope of final clauses is 3
clauses, while that for initial clauses 10.5.
396 VIOLETA RAMSEY

Table VIII
Initial and final 'if clauses used in narrative
Scope in previous discourse
N of clauses
INITIAL CLAUSES to the left FINAL CLAUSES
N % to total N % to total N N
of clauses of clauses
1 8.4 1 25 2
3 25 2 37.5 3
4 33.3 3 25 2
/ / 4 / /
/ / 5 12.5 1
1 8.3 6
1 8.3 7
/ / 8-10
/ / 11-13
/ / 14-17
22 16.7 19-20+
12 100 100 8
Average scope per clause
Initial clauses: 6.3 clauses
final clauses: 2.3 clauses
 PREPOSED AND POSTPOSED 'IF' AND 'WHEN' CLAUSES 397

Table IX
Initial and final 'if clauses used by participants
Scope in previous discourse
N of clauses
INITIAL CLAUSES to the left FINAL CLAUSES
N % to total N % to total N N
of clauses of clauses
1 1.7 1 53.4 16
/ / 2 23.3 7
3 5 3 6.7 2
5 8.5 4 3.3 2
4 6.8 5 6.7 2
3 5 6 / /
3 5 7 3.3 1
4 6.8 8 3.3 1
9 15.3 8
5 8.5 10
6 10.2 11-13
4 6.7 14-17
12 20.5 19-20+
59 100 100 30
Average scope per clause
Initial clauses: 10.5 clauses
final clauses: 2.3 clauses
398 VIOLETA RAMSEY

Table X
Initial and final 'when' clauses used in narrative
Scope in previous discourse
N of clauses
INITIAL CLAUSES to the left FINAL CLAUSES
N % to total N % to total N N
of clauses of clauses
2 4.6 1 55.6 15
2 4.6 2 14.8 4
2 4.6 3 7.4 2
3 7 4 11.1 3
6 14 5 / /
4 9.4 6 / /
4 9.4 7 / /
3 7 8 / /
5 11.6 8 / /
3 7 10 / /
2 4.6 11 / /
2 4.6 12 3.7 1
2 4.6 13 / /
/ / 14-16 3.7 1
/ / 17-19 / /
3 7 20+ 3.7 1
43 100 100 27
Average scope per clause
Initial clauses: 7.8 clauses
final clauses: 3.2 clauses
 PREPOSED AND POSTPOSED 'IF' AND 'WHEN' CLAUSES 399

Table XI
Initial and final 'when' clauses used by participants
Scope in previous discourse
N of clauses
INITIAL CLAUSES to the left FINAL CLAUSES
N % to total N % to total N N
of clauses of clauses
1 6.2 1 44.5 4
/ / 2 11.1 1
/ / 3 11.1 1
/ / 4 11.1 1
1 6.2 5 11.1 1
2 12.5 6 / /
2 12.5 7 / /
3 18.9 8 / /
/ / 9 11.1 1
/ / 10
2 12.5 11
/ / 12
/ / 13
2 12.5 14-16
/ / 17-19
3 18.9 20+
16 100 100 9
Average scope per clause
Initial clauses: 10.5 clauses
final clauses: 3 clauses
400 VIOLETA RAMSEY

4.3. Aspect/Modality categorization of 'when' clauses

The results of this quantification show that 'when' clauses used in the
narrative pattern differently from those used in quoted speech.
For those clauses used in the narrative (Table XII) we find that a great
majority of the ones preposed to the main clause, more precisely 90.2%, are
perfctive/realis and only 9.8% are imperfective/irrealis. Conversely for post-
posed clauses, only 26% are perfective/realis while 74% are imperfective/
irrealis.
For those clauses used by participants (Table XIII) the pattern is
reversed: With respect to preposed clauses, 38.9% are perfective/realis and
61.1% are imperfective/irrealis. With respect to postposed clauses we find
that 66.7% are perfective/realis and 33.3% are imperfective/irrealis.
Notice that most 'when' clauses are used in the narrative. Out of a total
of 95 WC (this includes initial and final clauses) 68 — or 72.6% — are used
by the narrator and only 27 — or 27.4% — are used by the participants. This
is to expect since WC serve to sequence the temporal contour of the narra­
tive. (Regarding 'if clauses it is the reverse that is found; out of a total of 109
clauses only 20 — or 18.3% — are used in the narrative and 89 — or 81.7%
— are used by the participants).

Table XII
Aspect/Modality categorization of 'when' clauses
used by the narrator
Preposed clauses % Postposed clauses %
Perfective/realis 37 90.2 7 26
Imperfective/irrealis 4 9.81 20 74
41 100 27 100

Table XIII
Aspect/Modality categorization of 'when' clauses
used by the participants
Preposed clauses % ļ Postposed clauses %
Perfective/realis 7 38.9 7 66.7
Imperfective/irrealis 11 61.1 3 33.3
18 100 9 100
 PREPOSED AND POSTPOSED 'IF' AND 'WHEN' CLAUSES 401

4.4. Distribution of commas

The results of these counts show that commas are used much more fre­
quently with preposed clauses than with postposed clauses. In Table XIV one
can see that in the case of I (this includes narrative and quoted speech)
73.2% of the initial clauses have a comma separating the adverbial clause
from the main clause, while for final clauses 94.7% are not separated from the
main clauses by a comma.
With regard to WC (Table XV) we find the same pattern: 62.7% of pre­
posed clauses have a comma whereas as much as 94.5% of postposed clauses
do not.

Table XIV
Distribution of commas in i f clauses
INITIAL CLAUSES FINAL CLAUSES
N % % N
52 73.2 with a comma 5.3 2
19 26.8 without a comma 94.7 36

71 100 100 38

Table XV
Distribution of commas in 'when' clauses
INITIAL CLAUSES FINAL CLAUSES
N % % N
37 62.7 with a comma 5.5 2
22 37.3 without a comma 94.5 34

59 100 100 36

4.5. Distribution of WC and I at paragraph breaks

The results of this measurement show that very few I and WC are used
in paragraph initial position (only 13 in the entire novel). As one can expect,
all of these clauses are used in the narrative; of those, 12 are WC (7 initial and
5 final). There is only one 'if clause in this position and it is an initial clause.
These results appear in Table XVI.
402 VIOLETA RAMSEY

Table XVI
Paragraph break clauses
Perfective Imperfective
WHEN CLAUSES
Preposed clauses 7 /
Postposed clauses 3 2
IF CLAUSES
Preposed clauses / 1
Postposed clauses / /

5. Discussion
The quantifications performed for this study might not, each by itself,
'prove' that an initial clause is performing a different discourse function than
that performed by a final clause. But taken all together they come to show
that there is a big difference between the role of the former and that of the lat­
ter.
The quantifications on referential distance, for example, show that final
clauses exhibit a very high referential continuity with the main clause. This
only means that the subject of the main clause tends to be the same as the one
on the postposed clause. They also show that initial clauses exhibit a lower
referential continuity vis-a-vis the main clause, that is, the subject of an initial
clause tends to be different from the subject of a main clause. Moreover, the
subject of a preposed clause is sometimes the same as the one of the clause
immediately preceding, but in most cases the previous reference for the sub­
ject of an initial clause is in one of the various clauses appearing in the preced­
ing discourse. Thus preposed IC and WC appear to be thematically linked to
the main clause as well as, and even more frequently, to the preceding dis­
course. Postposed clauses on the other hand, appear to have a much higher
referential continuity with the main clause.
As for the quantifications on scope, they come to support the results of
referential distance by showing that the thematic link of a preposed clause is
much wider than that of a postposed clause. As we have seen, the average
scope of a postposed clause is 2 or 3 clauses whereas that of a preposed clause
is from 7 to 10 clauses.
With regard to the measurements on the aspect/modality of verbs of
'when' clauses, they come as a very strong support of the hypotheses pre­
sented in this paper. We already said that most 'when' clauses are used in the
 PREPOSED AND POSTPOSED 'IF' AND 'WHEN' CLAUSES 403

narrative and that a majority of them are used in a preposed position, some­
thing that we can expect since this type of clauses are most frequently used for
sequencing the temporal events of the narrative. There is also the fact that a
great majority of those 'when' clauses used in initial position show verbs in
the perfective/realis while most of those used in final position have verbs in
the imperfective/irrealis (most specifically, the majority of them are in the
habitual aspect). Thus preposed WC seem to have all the characteristics of
what Hopper (1979:214-6) calls 'foregrounded' clauses: They carry main line
information, reflect an iconic order, are in the realis modality and show new
information in their predicate. There is the fact that these characteristics are
not commonly found in adverbial clauses, but as Hopper says, they tend to
tolerate new information when they are in clause initial position. In any case,
it seems that the notion of foreground/background information is not a
theoretically stable one and that, as Givón (this volume) suggests, the
grounding function of preposed clauses may not be construed in simple
binary terms of foreground/background.
By looking at some examples one can illustrate some of the differences
in the function of the two types of clauses:
(8) a. Some had removed their make-up already, some were still
crudely overcoloured for their parts, and Jim McKaid was
already costumed and made up for the part of Don Alfonso in
Cosi, and looked the elderly cynic to the life.
When they saw that Hurtle was in the theatre, one or two went
over to him to express sympathy. Nichols watched in amuse­
ment...(p.71)
b. Giulia Contini was floating some delicious high notes as she
prepared to go and join her mother in heaven. She was at last
giving some indication of how she had made her reputation.
Nichols suspected that she was one of these Italians who
thrived on a good blow-up row. 'That's a real nice sound,' said
Hurtle, and staved on to listen.
When it was all over, and the stage-hands prepared to disman­
tle the mantuan inn.... Nichols slipped down and had a word
with Mike...(p.170)
One can see in these examples the scope of the two preposed WC, the
switch of time and action they establish, and how they provide a frame for the
material that follows advancing in this way the main line of the narrative.
404 VIOLETA RAMSEY

Now to illustrate how final WC do not perform any of these functions, I quote
here two examples:
(9) a. At present it was her recent Carmen with the Welsh National
that was fresh in everyone's mind. Brazen, blatant, torrid and
vulgar, it had had some critics reaching for their superlatives,
and other simply reaching. She even claimed that one non­
conformist minister had preached a sermon against her some­
where in the valleys, but nobody believed her. It was as well
not to believe Gaylne when she said things of this sort. In fact,
it was rumored...(p. 10)
b. For the duets Bridget, after consulting with Simon and Mr.
Pettifer, was determined not to stint of voice. This Gilda was
to be an emotional adult, a woman who knew her own mind.
Simon followed suit, opening up with the splendid, full line
that was his when he didn't let his approach get too complicated
by introspection. (p.28)
In these examples it is easy to see how the information conveyed by the
final 'when' clause does not advance the main line of the narrative (it is not
temporally sequenced either), it only completes the information given in the
main clause. One can say that it looks like a parenthetical comment on the
narration. Notice also the verbs are in the habitual aspect.
Exactly the same difference in function is seen in 'if' clauses. Consider
the following examples:
(9) a. It was a sweet tenor voice, and the hall, which in its time had
made hollow, hungry voices sound strong and full, did its best
for Calvin's...'splendid little hall,' he said.
'Not so much of this Eyetalian!' said Gaylene French, stomp­
ing over to the far end of the hall...'We're doing the thing in
English, guest star notwithstanding.'
She undid the top buttom of her blouse, threw apart her arms
and bellowed. 'Land of Ho-ope and Gllory!' Then she put her
hands on her hips, looked at the pair by the door with infinite
self-satisfaction. and said: 'If the halVll stand that, it'll stand
anything.' (p.9)
b. He turned to a passing constable, and said: 'We'll want the
best shorthand writer in the force for this case. It's going to be
one hell of a complicated one, if I'm not mistaken. If we cant
 PREPOSED AND POSTPOSED 'IF' AND 'WHEN' CLAUSES 405

have Chappel. I want nothing less than McLintock, and make

that clear to them at Headquarters.' (p.85)
c. The music was racy and seductive, and Gaylene, hands on
hips, delivered it with a bouncy vulgarity, too loud. There was
something else wrong too, for the voice didn't quite suit the
part; there was a suggestion of British Contralto about it, a
touch of the Kathleen Ferriers. The listener felt that she might
be better employed telling good tidings to Zion, though
Gaylene did not in any other respect suggest a messenger of
the Lord, and would not for a moment have accepted her
unsuitability for the role of Maddalena, if anyone had had the
temerity to suggest it. (p. 14)
d. 'She went on elsewhere, then. Who to, do you know?
'Owen Caulfield, in the first place, I think. And then
Raymond Ricci, and then-God knows who.'
'Why did you split up, sir?'
'Split up? I don't know that we split up, not as I understand the
term. You can only split up if you've been going together, can't
you?
We just went to bed a couple of times...(p.89)
One can see in the first two examples how the initial clauses associate
with a portion of the discourse that precedes them, how their scope goes
many clauses back and how they help the dynamics of the sequencing of infor­
mation. The final clauses, on the contrary, carry information that is not
exactly part of the main line of the narrative, which makes them look more
like a parenthetical comment for the main clause.
If we look now at the results on the measurements on the distribution of
commas, we find more support to our findings. We said that preposed clauses
are most of the times separated from the main clause by a comma, whereas
postposed clauses are almost never separated by a comma. One could try to
attribute this fact to stylistic choices or to the fact that since the adverbial
clause is in a way an independent structural unit, it calls for a pause between
the two structures; but none of these reasons could explain why postposed
clauses are almost never separated from the main clause with a comma. One
has to say instead that the initial clause is 'recalling' part of the preceding dis­
course and in so doing it is connecting or grounding that preceding discourse
to the material that follows. In this way it calls for a pause before the intro­
duction of the information conveyed by the main clause. On the other hand
406 VIOLETA RAMSEY

the postposed clause only seems to be extending the semantic information

given by the main clause, thus its role is not at the thematic level.
Chafe (1984) also presents similar results. He examined all types of
adverbial clauses and looked for the differences between the behaviour of ini­
tial and final clauses in spoken and written discourse. With respect to the
analysis of written discourse he also found that a great majority of the pre-
posed clauses are not 'bound' to the main clause, that is, there is a comma
between the two clauses (or it is not in the same intonation group in the case
of spoken discourse). As for postposed clauses, he says that a majority of
them do not have a comma. Another part of his study consists of an analysis
of the discourse role that initial and final clauses perform and concludes that
a preposed adverbial clause represents a limitation of focus, 'signalling a path
or orientation in terms of which the following information is to be under­
stood'. A postposed clause, on the contrary, only adds something to the
assertion made by the main clause or modifies part of what was stated there.
The analysis presented by Marchese (1977) also comes to support our
hypotheses. She determined that clause initial subordinate clauses in Godie
function as topics in that they serve as 'frames' for the text that follows. As for
her analysis of conditionals on procedural discourse (this volume), it also
shows that conditionals (which are all in a preposed position) have a text-
organizing function since they serve to break the discourse into significant
units and to frame the event in the following clause.
Fagerber's findings (1983) also show the same kind of phenomenon. The
choice for clefting or pseudoclefting an element in Pulaar has to do with the
needs of the discourse and the communicative choices of the speaker.
Yet another work that supports our findings is Thompson's (to appear)
work on the behaviour of initial and final purpose clauses. Her argument cen­
ters on the semantic characteristics of purpose clauses which, explains the
'pragmatic positioning' of only that specific type of adverbial clause.
Nevertheless, she shows that it is discourse factors that determine whether
purpose clauses will be preposed or postposed to their main clauses.
And of course further research on the behaviour of different types of
adverbial clauses might require a combined analysis of their functions both at
the discourse and at the semantic level. We have seen that different types
exhibit different characteristics. 'When' clauses, for examples, are used in the
narrative and in initial position most of the times. 'If' clauses are used more
frequently in quoted speech and in initial position and, because of their
semantic character, they are always irrealis. Purpose clauses, on the other
 PREPOSED AND POSTPOSED ' I F ' AND 'WHEN' CLAUSES 407

hand, are used most frequently in final position (Thompson to appear) and
usually do not have a subject of their own but share the one of the main
clause. In any case, we can say now that the positioning of the adverbial
clause, before or after the main clause, is determined by the organization of
discourse.
One should mention the fact that this phenomenon of positioning of
adverbial clauses in written discourse is of course related also to factors other
than connecting or sequencing information. The writer is also trying to give
a certain style to his writing, to create certain moods in the narrative, to sur­
prise or keep the reader in tension in certain ways, to create certain climaxes,
etc., besides the obvious job of sequencing the action/events of the story.
And in so doing he handles the information with all of these factors in mind
and adjusts his writting accordingly. Thus if our numbers do not show higher
figures is because this phenomenon is a very complex one that will require
much more work before we can understand it entirely.
We might also deepen our understanding on the workings of this
phenomena as we get more data from other languages and from other types
of discourse. We also have to wait until more research is done on the organi­
zation and development of written discourse. And of course we need more
empirical studies that will give validity to the intuitive observations we are
creating about these notions.

REFERENCES

Chafe, W. 1984. "How people use adverbial clauses". In Claudia Brugman

and Monica Macaulay, eds., Proceedings of the Tenth Annual Meeting of
the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society.
Fagerber, Sonja. 1983. "Discourse strategies in Pulaar: The uses of focus"
Studies in African Linguistics Vol. 14-2:141-157.
Givón, T. 1984. Syntax: A Functional-Typological Introduction, vol. 1
Amsterdam: J. Benjamins.
1983d. "Topic continuity and word-order pragmatics in lite" in Topic
Continuity in Discourse: A Quantitative Cross-Language Study, ed. by T.
Givón, Typological Studies in Language, vol. 3. Amsterdam: J. Benja­
mins.
Haiman, John. 1978. "Conditionals are topics" Language vol. 54-3:564-589.
Hopper, P. 1979. "Aspect and foregrounding in discourse" in Syntax and
408 VIOLETA RAMSEY

Semantics vol. 12. ed. by T. Givón.

Marchese, L. 1977. "Subordinate clauses as topics." in Papers in African
Linguistics. Supplement 7.
Thompson, Sandra A. (to appear) "Grammar and written discourse: Initial
vs final purpose clauses in English." Text.
 FASTNESS" AND "NARRATIVE EVENTS" IN JAPANESE
CONVERSATIONAL NARRATIVES

POLLY E. SZATROWSKI
University of Michigan

1. Introduction
This paper is a quantitative analysis of the intersubjective agreement
among Japanese native speakers regarding the use of the tense-aspect forms,
RU and TA, in Japanese conversational narratives. 1 Previous analyses of
Japanese tense-aspect forms have been concerned with how to fit the RU and
TA forms into the tense and aspect paradigms which have been developed for
western languages. The RU/TA distinction has been analyzed on the sen­
tence level as tense (non-past/past) and aspect (non-complete/complete,
uncompleted/completed, incompletive/completive) (Teramura 1971, Szat-
rowski 1981, Kusanagi 1981, 1982; Soga 1981, 1983, 1984). However, these
arguments for and against tense and aspect are difficult to evaluate because
the evidence is often based on one linguist's intuitions against another's.
More recent work has shown that there is a high frequency of RU forms in
Japanese written narratives (Soga 1981, 1983, 1984; Makino 1981, 1983) and
conversational narratives (Szatrowski 1985), that is, in environments where
it is difficult to interpret the meaning as non-past or non-complete, uncom­
pleted, incompletive. The question of whether RU and TA are tense or
aspect seems of limited value in comparison with the more global question of
how these forms function in actual use in discourse.
Soga (1981, 1983, 1984) has pointed out that foreground events tend to
be given in the TA form and background events in the RU form. The present
study is a refinement of Soga's analysis and is concerned with the variability
in native speakers' associations of "pastness" and "narrative events" with the
use of RU and TA forms in conversational narratives. The question addres­
sed is an ecological one: "How do mechanisms like tense and aspect function
in Japanese and why?" By "ecological" I mean the function of forms and how
410 POLLY E. SZATROWSKI

they interact with other components in their environment, i.e., the discourse.
Recent linguistic research has stressed the importance of viewing forms
in context on the discourse level and suggests that speakers use different strat­
egies in creating their discourse. Polanyi (1979) claims that the point of a nar­
rative is not static but changes during the course of a conversation as it is
negotiated by the speakers and hearers involved in the speech situation. Hop­
per and Thompson (1980) present a view of transitivity as a continuum and
suggest that speakers can negotiate the relative foregrounding of events
through the categories they choose in the presentation of their experience.
Schegloff (1981) points out the following weakness in research on oral narra­
tive.
The common discourse-analytic standpoint treats the lecture, or ser­
mon, or story told in an elicitation interview, campfire setting, or around
the table, as the product of a single speaker and a single mind; the con­
versation-analytic angle of inquiry does not let go of the fact that speech-
exchange systems are involved, in which more than one participant is
present and relevant to the talk, even when only one does the talking.
(Schegloff 1981: 71-2)
In this study, language is viewed as a negotiation between speaker and
hearer(s), that is, a product of human agents participating in a social activity.
The view that language forms have functions as opposed to meanings necessi­
tates the study of language in discourse because functions have no role out­
side the context in which they are used.
The methodology used in this study differs radically from that used in the
previous research on Japanese tense and aspect cited above. Previous studies
relied primarily on the linguist's intuitions or the judgements of several native
speakers on sentences, often isolated from context. Questions to the infor­
mants were typically posed in the form, "What does the TA form mean in this
sentence?" The methodological assumption seemed to be that native speak­
ers are consciously aware of their use of language and that this awareness reli­
ably indicates what guides their actual usage of language. However, like many
others, I have found that it is often difficult to ascertain the nature of these
articulated associations of meanings or functions with form. My research on
the Japanese tense-aspect forms chosen in negative responses to questions
about the past showed a discrepancy between speakers' intuitions and their
actual usage of the forms, which could well be caused by their educational
background in Japanese and English grammar (Szatrowski 1983). The
methodology used here investigates speakers' intuitions indirectly in an
attempt to bypass some of these problems.
 "VIVIDNESS" AND "NARRATIVE EVENTS" 411

Results of the present study indicate that tense-aspect forms are

implemented variably. In particular, the use of TA or RU served to enhance
either pastness or narrative events but not both at a particular clause. This
complementary of the functions of pastness and narrative events is shown to
relate to communicative strategies in the discourse.

2. The Study
The procedure used in this study was as follows. First, the data were col­
lected using observational as opposed to elicitation techniques. Conversa­
tional narratives about past experiences were recorded from live television
talk shows and natural conversations. Next, native speakers, when possible
the original speaker and hearer(s) in the conversation, were questioned
directly and indirectly about the verbal forms used in the narrative, for exam­
ple, how they interpreted the use of a given form and what the effect of chang­
ing a RU form to a TA form or vice versa might be. Based on these native
speakers' judgements, a list of criteria potentially relevant for the use of the
RU and TA forms in narratives was established. The feature pastness was
selected as the focus of this paper since it occurred most often in speakers'
characterizations of the TA form. The feature narrative events will also be
discussed because recent work has shown that event relations play a major
role in narrative. The notion of "narrative events" proved important in
understanding the use of Japanese tense-aspect forms.
This study was designed to investigate the function of pastness and narra­
tive events in narratives, in particular in clauses which allow the RU/TA sub­
stitution. A transcript of the narrative chosen for this study is given in the
appendix. A set of three tapes was made based on this transcript, making sys­
tematic changes of RU to TA and vice versa.2 The tapes were recorded by the
same two Japanese native speakers. Tape 1 was made with the RU clauses,
marked with a star in the transcript, changed to TA. Tape 2 was a tape of the
original narrative. Finally, tape 3 was a recording of the original narrative
with the TA clauses, marked with a circle in the transcript, changed to RU.
Thus, tape 1 was characterized as having a higher number of TA forms than
the original narrative, and will be referred to as the "high TA" tape. Tape 2
will be referred to as the control. Tape 3, which was characterized as having
a higher number of RU forms than the original narrative, will be called the
"high RU" tape.
Three undergraduate classes at the University of Tsukuba, Japan were
chosen as the subjects for this study. Each student was given a packet of trans­
cripts, similar to that in the appendix, only in Japanese. The first class was
412 POLLY E. SZATROWSKI

given transcripts of tape 1, second class, tape 2 and third class, tape 3. The
number of students in each class was 29, 49, and 41, respectively.
The investigation was conducted as follows. First, I played the tape for
the students to familiarize them with its content. Next, I defined a feature, for
example, narrative events, and asked students to indicate the clauses in the
transcript where they felt the feature narrative events applied, while listening
to a repetition of the tape. A similar procedure was followed with the feature
of pastness. Thus, on separate repetitions of the tape, students were asked to
mark the clauses in the transcripts where they felt each feature applied.
It is important to note here that the students' attention was not drawn to
the particular verbal forms used in the clauses. They were not directed to
associate the features of narrative events and pastness with the tense-aspect
forms used. Rather they were asked to consider the narrative as a whole and
judge in terms of the clausal units as given in the transcript.
A comparison of students' responses across the three surveys made it
possible to measure the heightening effect of TA over RU for the features
investigated in groups of clauses as well as in individual clauses. The approach
used in this study allowed for a more direct inference of cause than has been
allowed in previous studies. A change in the surveyees' intersubjective agree­
ment on the relevance of a feature to a given clause across the three surveys
could be associated more directly with the tense-aspect form used, since only
changes of RU to TA and TA to RU were made among the three tapes.

3. Results for the Pastness Feature

For the purpose of this study, pastness was defined as follows.3
(1) Kakosee. Kakosee to iu no wa, hanasite iru koto ga kako ni okotta
to kanzirareru tokoro o sasimasu.
Pastness. The thing called "pastness" indicates the places which
are felt that the things which (one) is talking (about) occurred in
the past.
Pastness. "Pastness" refers to the parts where it is felt that what is being
talked about occurred in the past.
The average pastness ratings obtained in this study are given in Table 1.4
The headings in Table 1 (and Table 2 which is discussed later in the text)
stand for the groups of clauses which were analyzed in this study. "All
Clauses" refers to all of the clauses changed in the transcripts. The remaining
headings are subsets of the clauses in this group. The group "Verbal-1
 "VIVIDNESS" AND "NARRATIVE EVENTS" 413

Table 1: Heightening Effects of Pastness, Comparing RU and TA

Classes All Verbal-1 Verbal-2 N +Copula Htg
Clauses Clauses Clauses Clauses Effect
of TA
1+2 TA TA TA TA
29.2 14.8 12.6 38.7
(.0001) (.0006) (.014) (.0001) N>V
2+3 TA TA TA TA (.002)
16.7 15.8 14.0 24.4
(.03) (.04) (.03) (.0001)
P-values are given in parentheses.
Capital letters indicate significant differences of p<.05.

Clauses" consisted of all changed affirmative verbal clauses. These were final
and non-final verbal clauses including clauses in which the verbals preceded
a nominalizer such as no(ñ), wake or mon, and verbal clauses which modified
other nominals. "Verbal-2 Clauses11 are a subset of "Verbal-1 Clauses". This
group consisted of affirmative final and non-final verbal clauses including
clauses in which the verbals preceded nominalizers such as no(ñ), wake and
mon, but not verbal clauses which modified nominals. The "Nominal +
Copula Clauses" group consisted of affirmative final and non-final nominal +
copula clauses, including clauses in which the nominal + copula preceded a
nominalizer such as no(ñ), wake and mon, but not clauses which modified
nominals. Finally, "Htg Effect of TA" stands for the heightening effect of
TA over RU. "N>V" in this column indicates that the heightening effect of
TA over RU in nominal + copula clauses was greater than in verbal clauses.
Table 1 shows that TA heightened the pastness rating over RU in all of
the 8 comparisons of TA and RU studied and this was significant in all cases.
Previous analyses would end here. The fact that the overall average effects of
TA over RU are consistent across the board would be sufficient for designat­
ing the meaning of TA as pastness in most studies.
However a closer look at Table 1 reveals that the heightening effect of
TA was significantly different for verbal and nominal + copula clauses. That
is, although the heightening effect of TA over RU was significant overall, this
tendency was much stronger for nominal + copula clauses than for verbal
clauses. The magnitude of the heightening effects in Table 1 is higher for
nominal + copula clauses than for verbal clauses. Specifically, the heighten­
ing effects for nominal + copula clauses were 38.7 and 24.4, as compared to
 414 Graph 1: "Pastness" ResultsPOLLY
for Class 1E. SZATROWSKI
(high TA) and Class 2 (control)
Class 1 (high TA)
Class 2 (control)
Precentage of Students

Graph 2: "Pastness" Results for Class 2 (control) and Class 3 (high RU)
Class 2 (control)
Class 3 (high RU)
 "VIVIDNESS" AND "NARRATIVE EVENTS" 415

14.8 and 15.8 for the Verbal-1 group (which included clauses which modified
nominate) and 12.6 and 14.0 for the Verbal-2 group.
Results for the pastness feature are summarized in Graphs 1 and 2, which
give the percentage of students who rated each particular clause past. In
Graph 1, the changed clauses, that is, the clauses where RU in the original
transcript were changed to TA in Survey 1, are marked with stars under the
x-axis. In Graph 2, the clauses where TA in the original were changed to RU
in Survey 3 are marked with circles under the x-axis.
The most striking thing to notice is that the peaks and valleys in the dot­
ted and solid lines of each of these graphs occur at the same clauses. Thus, the
distinction between the RU and TA forms is not a black and white distinction
of pastness vs. non-pastness. If this were the case. we would expect the solid
line to have peaks where the dotted line has valleys, or vice versa. Graphs 1
and 2 give evidence to support the claim that it is not possible to dichotomize
the meanings of TA and RU as past and non-past, respectively.
Now we turn to a discussion of the variable implementation of the TA
and RU forms. Graph 3 is a graph of the variation in the heightening effect
of pastness of TA over RU. The x-axis gives the mean pastness rating of the
changed clauses and ranges from 0 to 1. This mean pastness rating was calcu­
lated by taking the average of the pastness rating when RU was used and the
pastness rating when TA was used, for each clause. Thus, clauses which were
considered past regardless of whether TA or RU was used, are on the right
side of this graph. Clauses to which the feature of pastness applied less
strongly on average are on the left side of this graph. The y-axis gives the
heightening effect of TA over RU, that is, the difference in pastness rating
when TA was used minus the pastness rating when RU was used. Each point
on the graph is labeled N, V or A and represents a single clause. N stands for
a nominal + copula clause, V for a verbal clause and A for an Adjectival
clause.
It is important to note that the heightening effect of TA over RU varies
with each clause. Therefore, like Graphs 1 and 2, Graph 3 also offers evi­
dence against an analysis of TA as past and RU as non-past. In particular,
there are a number of points near the horizontal zero line. The pastness rating
for these clauses was the same regardless of whether the clause contained a
RU or a TA form. In other words, the use of RU or TA did not make much
difference in the pastness rating of these clauses.
The following summarizes three points which account for the variation in
Graph 3. First, the heightening effect of TA over RU is very small for clauses
416 POLLY E. SZATROWSKI

Graph 3: Variation in the Heightening Effect of "Pastness"

Mean "Pastness" Rating of the Clause

with high mean pastness ratings. That is, if a clause has a high mean pastness
rating, it is judged past regardless of whether TA or RU is used. Further­
more, as the mean pastness rating decreases, the heightening effect of TA
over RU increases. That is, clauses which were less strongly associated with
pastness regardless of whether TA or RU was used, nonetheless had their
degree of pastness significantly enhanced when TA was used over RU.
 "VIVIDNESS" AND "NARRATIVE EVENTS" 417

Therefore, one could draw a iine from the upper left to the lower righthand
corner of Graph 3 and all the clauses would fall near that line.
Second, there are more V's in the lower right and more N's in the upper
left portion of Graph 3. This indicates that there was a tendency for verbal
clauses to be associated with pastness regardless of whether TA or RU was
used, while clauses containing nominal + copula constructions were
associated with pastness only if TA was used.
Third, if one looks at the individual clauses in this graph, clauses in the
lower right tend to be active and high in Transitivity. As clause points go up
towards the left they become more stative and lower in Transitivity. The
notion of Transitivity used here follows that of Hopper and Thompson
(1980).5 For example, the following clauses at the bottom right are active and
high in Transitivity.6
(2) V77 totTa ñ desu yo.
take N COP SP
'It's that (I) took (his picture), you know/
V103 baa to kaetTA ñ desu yo.
whoosh QUOT go home N COP SP
'It's that (I) went straight home, you know.'
V81 Koo, kamera no hoo mukli ñ desu yo.
like this camera POSS direction face N COP SP
'Like this, it's that (he) faces the camera, you know.'
Clauses in the upper left are more stative and low in Transitivity.
(3) N79 De, kirai NA no ni ne,
and hate COP although SP
'And even though (he) hates it,'
N119 Sore mo zeñzeñ piñboke zyanal ñ desu yo.
that also at all out of focus COP-(neg) N COP SP
'also it's that (they) are not out of focus at all, you know.'
N15 Koo yaruno ga suki NA ñ desu yo.
like this do SUB J like COP N COP SP
'Also it's that (I) like doing this, you know.'
N123 De, husigi desU nee.
and amazing COP SP
'And, (it)'s amazing, isn't it.'
Verbal clauses in the middle range tend to be negatives or questions, that is,
lower in Transitivity than the affirmative declarative clauses which tend to
418 POLLY E. SZATROWSKI

cluster in the lower right.

(4) V57 anoo, 'syokuzi-dekimasEÑ kara ne.
um meal-do(pot)(neg) so SP
'um, (he) can't eat so,'
V97 Kami kañzi de motte rasiTÁ ñ desyoo?
light feeling with take go N COP(tent)
i t ' s that (you) took (it) with a light feeling, right?'
In summary, on the average, TA heightened the pastness of a clause
more than RU. However, since this heightening effect was variable, an
analysis of the meanings of TA and RU as past vs. non-past is inadequate. The
variability in the heightening effect of TA over RU can be accounted for in
terms of Hopper and Thompson's (1980) Transitivity components. Specifi­
cally, for clauses that rate high on the Transitivity scale, the use of TA vs. RU
does not change the pastness rating of the clause. For clauses with low Trans­
itivity ratings, the heightening effect of pastness of TA over RU is very
strong. Results also correlate with Hopper and Thompson's (1984) catego-
riality hypothesis. Nouns as a category do not typically function to narrate
events. Therefore, one would not expect them to be located in a time frame.
On the other hand, it is easier to infer the time location of verbal clauses
because they are often associated with events in the discourse. One might
speculate that when clauses are high in mean pastness rating (i.e., considered
past regardless of whether RU or TA is used), such as is the case with the ver­
bal clauses in the lower right of Graph 3, the speaker has the option of using
TA for a different function without altering the pastness of the clause. As is
demonstrated in the next section, TA functioned to heighten the narrative
event rating of these clauses.

4. Results for the Narrative Event Feature

The feature narrative events was defined after Labov and Waletzky
(1967), Labov (1972) and Schiffrin (1981) as follows for the purpose of this
study.7
(5) Orno na Dekigoto.   dekigoto to iu no wa, toki no nagare no
naka de zyuñbañ ni okovu dekigoto desu. Kono zyuñbañ ģa kuruu
to sutoorii wa kawatte simaimasu. Tatoeba, "(a) Samui hi da. (b)
Heya ni haitta. (c) Kotatu ni haitta. (d) Gohān o tabeta. " de wa, (a)
no zyuñbañ o kaete  sutoorii wa kawarimaseñ ga, (b), (c), (d) 
zyuñban ga tiģau to sutoorii wa kawarimasu. (b), (c), (d) wa 
na dekigoto de ari, (a) wa tiģau koto ni narimasu.
 "VIVIDNESS" AND "NARRATIVE EVENTS" 419

'Main Event(s). The thing called "main event" is an event

which occurs in order in the stream of time. If this order goes
wrong the story ends up changing. For example, in "(a) (It)'s a cold
day. (b) (I) entered the room. (c) (I) got in the kotatu. (d) (I) ate
a meal." if (one) changes the order of (a), the story does not
change but if the order of (b), (c), (d) is different, the story
changes. (It) becomes the fact that (b), (c), (d) are main events and
(a) is different.'
'Main Events. "Main Events" are the events which occur in
order in the flow of time. If this order is mixed up the story
changes. For example, in "(a) It's a cold day. (b) (I) entered the
room. (c) (I) sat down at the kotatu (low table with heating ele­
ment attached). (d) (I) ate dinner.", if one changes the position of
(a), the story does not change but if the order of (b), (c), (d) is dif­
ferent, the story changes. (B), (c), (d) are main events and (a) is
not.'
Recent analysis of the historical present in English and Spanish conver­
sational narratives (Labov 1972, Schiffrin 1981, Silva-Corvalán 1983) relies
Narratives are oral versions of experience in which events are relayed in the
order in which they presumably occurred. Their defining characteristic is a
relationship of TEMPORAL JUNCTURE between at least two clauses: if a
change in the order of the two clauses results in a change in the interpretation
of what actually happened, then those two clauses are NARRATIVE
CLAUSES and the events reported are NARRATIVE EVENTS. (Labov
and Waletzky 1967, Labov 1972) (Schiffrin 1981: 47)
These definitions suggest that the speaker has an experience in mind made up
of a temporal sequence of events which language can accommodate with a
one-to-one correspondence between narrative event and narrative clause.
The literature cited above accepts Labov and Waletzky and Labov's defini­
tions as given and suggest that narrative events are easy to identify in English
and Spanish narratives. There is no mention of the subjective nature of these
definitions.
However, Graph 4, which is frequency histogram showing the number of
clauses with particular narrative event percentage ratings, indicates that
Labov and Waletzky and Labov's criteria are not objective in the case of
Japanese conversational narratives. If the criteria were clearcut and objec­
tive, we would expect a graph with two peaks, one at the lower end of the nar­
rative event rating scale indicating agreement on what does not constitute a
 420Graph 4: Frequency HistogramPOLLY
of "Narrative Event" Ratings (Survey 2)
E. SZATROWSKI
Number of clauses

"Narrative event"
percentage ratings
 'VIVIDNESS" AND "NARRATIVE EVENTS" 421

narrative event, and the other on the higher end of the scale indicating agree­
ment on what does constitute a narrative event. Instead, we obtained a
smooth curve which strongly suggests that the criteria for narrative event-
hood are relative. This supports the claim made by Hopper and Thompson
(1980) that narrative events form a continuum.
It should be stressed that narratives are not static discourses, prepac­
kaged units of experience composed of discrete events. Experiments by
Newtson (1976) indicate that people describe the events of a movie differ­
ently depending on the stimuli they are given beforehand. This suggests that
when speakers recount their experience they do not simply encode predeter­
mined set of events in temporal sequence but rather are free to present their
experience in terms of a variety of event-types, some more or less prominent
than others. Similarly, hearers' interpretations of a speaker's presentation of
events may vary. This view of speakers and hearers having options to
negotiate the status of events in the discourse is relevant to the discussion of
the results obtained for the narrative event feature which follows.
Results for the narrative event feature are given in Graphs 5 and 6. If we
look at the peaks and valleys for the lines in these two graphs, we see that they
occur at the same clauses. Therefore, we cannot conclude that the TA clauses
mark narrative events while the RU clauses do not, or vice versa. In addition,
we see that a number of other clauses, which do not allow the RU/TA sub­
stitution are perceived as narrative events. Preliminary results indicate that
clauses ending in the V-te gerund form are perceived as being very prominent
narrative events. The narrative event ratings obtained for all of the clauses in
this survey suggest that clauses which contain an affirmative non -te iru verbal
form were perceived as being the most prominent "narrative events." These
include clauses ending the V-te gerund form, V-(r)u form and V-ta form.
Next, clauses ending in the V-te iru and V-te ita forms are perceived as less
prominent narrative events. Finally, clauses containing adjectivals, nominal
plus copula predicates, and negatives received the lowest narrative events
ratings. Again these results correlate with Hopper and Thompson's (1984)
categoriality hypothesis which predicts that nominal 4- copula clauses are less
likely to be associated with narrative events than verbal clauses.
 422 POLLY E. SZATROWSKI
Percentage of Students
 "VIVIDNESS" AND "NARRATIVE EVENTS" 423

Now, we turn to the question this experiment was designed to answer: In

the same environments, does the use of a TA form highlight the perception
of the clause as a narrative event more that the RU form? The narrative event
ratings for clauses which allow the RU/TA substitution are summarized in
Table 2.

Table 2: Heightening Effects of Narrative Events, Comparing RU and TA

Classes All Verbal-1 Verbal-2 N+Copula Htg
Clauses Clauses Clauses Clauses Effect
of TA
1+2 ta ta TA ru
.5 2.4 9.5 2.3
(.05) V>N
2+3 TA TA TA TA
6.7 6.5 7.6 12.9
(.02) (.05) (.04) (.05)
P-values are given in parentheses for p<.05.
Capital letters indicate significant differences of p<.05.
Lowercase letters indicates possible tendences at p < . l .

Table 2 indicates that on the average TA heightens narrative event rat­

ings. Specifically, TA heightened the effect of narrative events over RU in 7
of the 8 RU/TA comparisons studied in this investigation and 5 of these were
significant. However, more interestingly, the analysis showed that there was
a negative correlation between the heightening effect of pastness and the
heightening effect of narrative events. In particular, it was found that clauses
in the lower right of Graph 3, that is, clauses where the heightening effect of
TA over RU for pastness was near zero, had higher than average narrative
event ratings. Clauses in the upper left, where TA strongly enhanced the
heightening effect of pastness over RU, had lower than average narrative
event ratings.
To summarize, the function of TA served to enhance either the pastness
or narrative event rating, but not both, at a particular clause. In other words,
the functions of pastness and narrative events complemented one another in
clauses which allowed the RU/TA substitution.
424 POLLY E. SZATROWSKI

5. Conclusion
The above study gives empirical evidence to support the claim that an
explanation of the use of Japanese tense-aspect forms involves more than
morphological distinctions like non-past/past and noncomplete/complete,
uncompleted/completed, incompletive/completive. The TA and RU forms
are shown to be implemented variably in the discourse and this has important
implications for research methodology. Specifically, it indicates that the func­
tion of tense-aspect forms cannot be derived using an analysis which concen­
trates only on the sentence level.
On the average TA heightens the perception of pastness and narrative
events over RU, but responses are not consistent for all clauses. However, it
is possible to give an ecological explanation for this variability in view of the
interaction between tense-aspect forms and other components in the dis­
course. When the mean pastness rating of a clause is very high, i.e., the clause
is perceived as past regardless of whether TA or RU is used, there is less of a
need to negotiate pastness, presumably because the time location of the
clause can be inferred from other forms in the clause or from the position of
the clause in the discourse. In particular, this is the case for affirmative verbal
clauses which are high in Transitivity. One can speculate that in these clauses
RU and TA are free to be used for other functions besides pastness. In fact,
TA is shown to enhance narrative event ratings over RU in these clauses,
which suggests that the function of TA in these contexts is to foreground the
clause.
On the other hand, when the mean pastness rating of a clause is in the
medial-low range, there is a tendency for TA to strongly enhance the pastness
of the clause over RU and the effect of TA and RU on the narrative event rat­
ing is less prominent. This is the case for nominal + copula clauses, which pre­
sumably require the most negotiation with respect to pastness because it is
difficult to infer their time location from the discourse. The following exam­
ple taken from the conversation used in this study illustrates the kind of
negotiation which can occur when a RU form is used in nominal + copula
contexts.
(6) A- Soo iu koo, iiisai no o ne hikinobasite,
like that say like this little one OBJSP enlarge(ger)
Those, like this, (I) enlarge the little ones, and/
(7) koo yaruno ga suki NA n desu yo.
like this do N SUBJ like COP N COP SP
'it's that (I) like doing this, you know/
 "VIVIDNESS" AND "NARRATIVE EVENTS" 425

(8) B- Zya, ima demo odeki ni naru no?

well now even do(pot) (hon) N
'Well, is it that you can do (it) even now?'
(9) A- I maa, titi ga ikite RU toki wa ne,
now father SUBJ live when TOP SP
'
Now, when my father is living,'
(10) anoo, are NA n desu yo.
urn that COP N COP SP
'um, it's that, you know.'
(11) tyotto naku narimasita mon de,
just pass away N COP(ger)
'just, it's that (he) died, and'
Speaker A's use of RU (the NA form of the copula) in (7) leads B to think that
she enlarges pictures even now. A and B negotiate this point through B's
question in (8) and A's response in (9) through (11). Further inquiry into
interactions of this sort will shed new light on the types of situation which can
be negotiated by tense-aspect forms.
Finally, it was found that notions like pastness and narrative events are
also relevant to clauses which do not allow the RU/TA substitution. This
suggests a need for further investigation into clause combining and the rela­
tions between preceding and subsequent clauses in the discourse.
There are undoubtedly many other functions associated with the RU and
TA forms in Japanese conversational narratives. However, the more impor­
tant question to ask is what strategies do people use to create and interpret
their discourse. Just as people recall their experience of a movie differently
depending on the stimuli given beforehand, one would expect a narrative to
be presented differently depending on the impetus in the pre-narrative dis­
course and the interaction between speaker and hearer as the narrative
develops. The task for the future is to describe the negotiations of tense-
aspect forms in the interactional context.

ACKNOWLEDGEMENTS
I would like to thank Paul Hopper and Sandra Thompson for their wil­
lingness to discuss the ideas in this paper and for their valuable suggestions
and encouragement. I would also like to thank Ted Szatrowski, David
Andow and Ron Breiger for their help on the statistical analysis and Sally
McConnell-Ginet, Linda Waugh, and Eleanor Jorden for their helpful com-
426 P O L L Y E. S Z A T R O W S K I

ments on drafts of this paper. Finally, I am grateful to Professors Hideo

Tcramura and Shiro Hayashi and the students of the University of Tsukuba,
Japan who participated in the surveys and would like to express my warm
appreciation to Kazuyo Kusakari and Yuko Nakagawa for their invaluable
assistance.

NOTES
1) RU is used in this study to refer to the three major predicates in Japanese, V-(r)u (including
V-te iru), A-i and N + da (V= verbal stem, A= adjectival stem, N = nominal) and all their varia­
tions in politeness and formality, affirmative or negative, whether they occur in: sentence final
position; sentence medial or sentence final position before a particle, before a nominalizer such
as no (n). wake, mono, (mon) before daroo/desyoo; or in the sentence modifying constructions of
other nominals. TA refers to the three major predicates in Japanese, V-ta (including V-te ita). A-
katta and N + datta in the same environments as specified for RU above. The RU and TA varia­
tions are capitalized in the examples and transcript. The Japanese romanization used throughout
this paper follows that of Jorden ( 1963).
2) The decision as to which clauses to change was made in consultation with native speakers
using two methods. First, some native speakers were shown the transcript and asked if it were pos­
sible to make changes of RU to TA or vice versa. Second, copies of transcripts with these changes
made were shown to other native speakers who were asked to comment on whether anything in the
transcript seemed unnatural. That speakers took this task quite seriously is evidenced by the fact
that they pointed out a number of places in the transcripts besides tense-aspect forms which they
felt were strange. Finally, these speakers were asked specifically to comment on the acceptability
of tense-aspect forms in the transcript. Only clauses which speakers said would allow the RU/TA
substitution were changed in the study presented here.
3) Two English translations are given for the definitions of the features studied in this paper.
The first is a word-for-word literal translation of the Japanese and the second, a more smooth
translation.
4) The results given in Tables 1 and 2 were calculated using the following statistical methods.
The example presented here shows how the heightening effect for the pastness feature was calcu­
lated for Vcrbal-2 clauses in the comparison of results for Classes 1 and 2 (Table 1).
Example'. Heightening Effect of Pastness in Verbal-2 Clauses (Classes 1 and 2)
Class 1 (High TA) Class 2 (Control)
TA (unchanged clauses) 63.72 53.84
RU --> TA (changed clauses) TA used 90.51 y
RU --> TA (changed clauses) RU used x 68.00
Adjusted pastness effect of TA= 12.63 p= .014
Ideally we would like to estimate 90.51 - x , or y - 68.00 because this would estimate the heighten­
ing effect of TA over RU at the same clause for the same people in the same survey. However, we
do not have direct estimates of x and y because it is impossible for a single clause to be both RU
and TA in the survey for the same class. Assuming an additive model holds, we can use (90.51 —
63.72) - (68.00 - 53.84) = 26.79 - 14.16 = 12.63% to estimate the amount that the use of TA
 "VIVIDNESS" A N D " N A R R A T I V E E V E N T S " 427

heightens pastness over RU. In other words, differences in average pastness ratings between the
two classes can be corrected by subtracting off the pastness ratings for unchanged clauses, i.e.,
63.72 and 53.84, respectively. Thus, we estimate that there is a 13% increase in the perception of
pastness of Verbal-z Clauses if TA is used over RU. A two independent population t-test on popu­
lation means was used to determine whether or not this effect was signficantly different from zero
(Ott 1977: 112).
5) Hopper and Thompson (1980) asses the relative Transitivity of a clause according to the fol­
lowing 10 components of Transitivity.
TRANSITIVITY HIGH LOW
A- PARTICIPANTS 2 or more participants 1 participant
B- KINESIS action non-action
C- ASPECT telic non-telic
D- PUNCTUALITY punctual non-punctual
E- VOLITIONALITY volitional non-volitional
F- AFFIRMATION affirmative negative
G- MODE realis irrealis
H- AGENCY agent high in potency agent low in potency
I- AFFECTEDNESS of  object totally affected object not affected
J- INDIVIDUATION of  object highly individuated object non-individuated
(Hopper and Thompson 1980: 252)
6) The following abbreviations are used in the Japanese examples. N= nominalizer (no, ñ,
wake, mono, mon) The nominalizers no and mono are distinguished from nominals no and mono,
respectively, because they do not allow ga/no substitution in the predicate which precedes them.
(Eleanor Jorden, p..)
COP= copula, SP= sentence particle, QUOT= quotative particle, POSS= possessive, SUBJ =
subject, OBJ= object, pot= potential, neg= negative, tent = tentative, hon= honorific, ger=
gerund
7) The Japanese definition for this feature follows that of Labov and Waletzky (1967), Labov
(1972) and Schiffrin (1981). However, it was necessary to translate the feature itself, "narrative
events", as  na dekigoto, 'main event', because native speakers felt that more literal equiva­
lents were unnatural.

REFERENCES

Chafe, Wallace L. (ed.). 1980. The Pear Stories. Norwood: Ablex Publishing
Corporation.
Hopper, Paul J. and Sandra Thompson. 1980. "Transitivity in grammar and
discourse". Language 56: 251-99.
. 1984. "The discourse basis for lexical categories in universal gram­
mar". Language 60:4:703-52.
Jorden, Eleanor Harz. 1963. Beginning Japanese. New Haven: Yale Univer­
sity Press.
428 POLLY E. SZATROWSKI

Kusanagi, Hiroshi. 1981. "Tense and aspect in subordinate and relative

clauses". In Mabuti Kazuo Hakase Taikan Kinen Kokugogaku Ronsyuu.
Tokyo: Taisyuukan Publishers. 201-24.
. 1982. "Nihongo no tensu asupekuto kaiseki arugorizumu", "An
algorithm for the interpretation of Japanese tense and aspect". Bungei
Gengo Kenkyuu Gengohen 8: 69-83.
Labov, William. 1972. Language in the Inner City. Philadelphia: University
of Pennsylvania Press.
--- and Joshua Waletzky. 1967. "Narrative analysis: Oral versions of per­
sonal experience". In June Helm (ed.), Essays on the Verbal and Visual
Arts. Seattle: Washington University Press. 12-44.
Makino, Seiichi. 1981. "Tense switching in Japanese written discourse". In
Seiichi Makino (ed.), Papers from the Middlebury Symposium on
Japanese Discourse Analysis. University of Illinois. 125-250.
—--. 1983. "Monogatari no bunsyoo ni okeru zisei no tenkan", "Tense
switching in written narratives". Gengo: 109-117.
Newtson, Darren. 1976. "Foundations of attribution: The perception of
ongoing behavior". In John H. Harvey, William John Ickes, Robert F.
Kidd (eds.). New Directions in Attribution Reasearch: I. Hillsdale N.J.:
Lawrence Erlbaum. 223-47.
Ott, Lyman. 1977. An Introduction to Statistical Methods and Data Analysis.
North Scituate: Duxbury Press. 112.
Polanyi-Bowditch, Livia. 1976. "Why the whats are when: 'Mutually con-
textualizing realms of narrative'". BLS 2: 59-77.
— - . 1979. "So what's the point?" Semiotica 25: 3/4: 209-41.
Schegloff, Emanuel A. 1981. "Discourse as an interactional achievement:
Some uses of 'uh huh' and other things that come between sentences".
In Deborah Tannen (ed.), Georgetown University Round Table on Lan­
guages and Linguistics 1981. Washington, D.C.: Georgetown University
Press. 71-93.
Schriffrin, Deborah. 1981. "Tense variation in narrative". Language: 57:1:
45-62.
Silva-Corvalân, Carmen. 1983. 'Tense and aspect in oral Spanish narra­
tive: Context and meaning". Language 59:4: 760-70.
Soga, Matsuo. 1981. "Tense and aspect in conversations and narratives". In
Seiichi Makino (ed.), Papers from the Middlebury Symposium on
Japanese Discourse Analysis. University of Illinois. 276-94.
. 1983. Tense and Aspect in Modern Colloquial Japanese. Vancouver:
 "VIVIDNESS" A N D ' N A R R A T I V E EVENTS" 429

University of British Columbia Press.

. 1984. "Nihongo no danwa ni okeru zisei to soo ni tuite", "Tense and
aspect in Japanese conversation". Gengo 13:4; 120-27.
Szatrowski, Polly E. 1981. A Linguistic Analysis and Pedagogical Ordering of
the Contextual Variants of the -te iru Form in Japanese. Cornell University
M.A. Thesis.
. 1983. "Puragumatikkusu kara mita Nihongo no doosi no asupekuto —
toko ni hiteikei no baai ni oite", "A pragmatic analysis of Japanese nega­
tive verbal aspect forms". Tsukuba Working Papers in Linguistics 2:
48-64.
. 1985. The Function of Tense-Aspect Forms in Japanese Conversations:
Empirical and Methodological Considerations. Cornell University Ph.D.
Thesis.
Tannen, Deborah. 1983. "'I take out the rock — dok!': How Greek women
tell about being molested (and create involvement)". Anthropological
Linguistics 23:3: 359-74.
Teramura, Hideo. 1971. "'Ta' no imi to kinoo — asupekuto tensu muudo
no koobunteki itituke—", " T h e meaning and function of 'ta' — structural
position of aspect, tense and mood —". In Z. Toki, T. Sugimoto, et al.,
Gengogaku to Nihongo Mondai, Linguistics and Problems in the Japanese
Language. Tokyo: Kurosio Publishers. 288-256.
Wolfson, Nessa. 1979. "The conversational historical present alternation' 1 .
Language 51'. 168-82.

APPENDIX
Narrative transcript taken from the Japanese interview program "Tetuko no Heya",
'Tetsuko's Room'; A= actress Kumi Mizuno, B= interviewer Tetsuko Kuroyanagi
Class 1 Class 2 Class 3
TA ←- RU RU
   → RU
1. A- Sore kara toru hoo wa daisuki ni 1. A- Then it got to the point where (I)
narimasite, loved to take (pictures) too, and
2. iroiro titi kara osowatte, 2. (I) learned various things from my
father, and
3. B- A. Syasiño? 3. B- Oh. (Taking) pictures?
4. A- . Toru hoo mo daisuki ni 4. A- Yes. It got to the point where (I) loved
narimasite, to take (pictures) too, and
5. B- A. Soodesu(ka). 5.B- Oh. Is that so.
430 POLLY E. SZATROWSKI

6. - . Sore kara ansitu tte aru n desu 6. A- Yes. Then, there's a dark room, you
yo no? know?
7 . _ B - . 7.B- Yes.
8. A- De,soko e haitte, 8. A- And, (I) go in there, and
9. ano,hikinobasi tte aru ñ desu yo. 9. um, there's an enlarger, you know.
10. - . 10. B- Yes.
11. A- Sore o, ano, tetudatte, 11. A- That, um, (1) help (with it), and
12. anoo, me ga totemo yokatta moñ desu 12. um, my eyes were very good
kara, so,
13._B- Araa. 13. B- Oh.
14. A- Sooiu,koo,tiisai no o ne 14. A- Those, like this, the little ones
hikinobasite, (I) enlarge (them), and
15. * koo yaru no ga suki NA n desu yo. 15. (I) like doing this, you know.
16. - Zya,ima d e mo odeki ninaruno? 16. B- Well, can (you) do (it) even now?
17. * A- Imaa,titi ga ikite RU toki wa ne, 17. A- Now, when my father is living,
18.* anoo, are NA ñ desu yo. 18. um, it's that, you know.
19. tyotto naku narimasita mon de. 19. just, it's that (he) died, and
20 B- A.Soo desu(ka). 20. B- Oh. Is that so.
21 A- Hai. 21. A- Yes.
22. B- Wari t o saikin desu ka. 22. B- Is (it) relatively recent?
23. * A-Kyoneñno si-gatu nizyuu ku n i t i desU.23. A- (It) is April 29 of last year.
2 4 . B - Araa. 24. B- Oh.
25. A- Anoo,tyotto,ee,kootoogañ de, 25. A- Um, just, yes, from throat cancer,
26. B- Soo desu ka. 26. B- Is that so.
2 7 A - . 27. A- Yes.
28. - Oi, otoosama oikutu datta no? 28. B- How old was your father?
29. * A- Anoo ne, nanazyuu-sañ desU. 29. A- Um, seventy-three.
30. * B- Nañka,sono toki desU ka. 30. B- Something, is (it) that time?
31. Otoosama no toko, otoosama wa 31. Your father, your father
daitai syasin utusareru no wa for the most part was a person who
okirai na kata dat... didn't like having (his) picture
taken...
32. o Moo, daikirai datTA ñ desu you nee. 32. A- Oh, (he) hated (it), you know.
3 3 . B - Ñ. 33. B- Hm.
34. A-.De,soree,ñ , soo iu no tte, nañka, 34. A- And, that, um, that kind of thing,
sore koso, tyotto reikañ tte iu no something, that, just (I) wonder if
ka sira. (one) calls (it) a sixth sense.
35. Nani ka, ano, yoku hito ga naku naru 35. Something, um, often when people
toki tte, are going to die,
36. koo, nañka, koo, yobiyoseru tte iu 36. like, there's something, like, that
no arimasu desyoo? (one) says comes to call, isn't there?
3 7 . B - . 37. B- Yes.
38. A- Maa,sore ka doo ka wakañnai ñ desu 38. A- Well, (I) don't know whether (it) is
kedo ne, that or not, but
39. tyotto husigi na koto ga atta ñ 39. just, an amazing thing happened, but
desu kedo,
40. yo,yorosii desu ka. 40. is (it) alright (to talk about it)?
41. B- Ii desu yo 41. B- (It)'s alright.
 "VIVIDNESS" A N D " N A R R A T I V E EVENTS" 431

42. o A- Anoo, zenzen ne, ano maa, omimai wa 42. A- Um, not at all, um well, (I/we) had
tokidoki itte masiTA kedo ne, been going to see (him) sometimes
but,
43._B- Ee.Tookyoode. 43. B- Yes. In Tokyo.
44._* - . . Tookyoo des U kedo, 44. A- Yes. Yes. (It)'s Tokyo but,
45. ano, nañka, kazoku de, zeñiñ de ne 45. um, something, with the family,
nañka, tonikaku asita zettai something, anyway (I) got the feeling
ikenakya ikenai tte iu ki ni natte, that (we) should all go tomorrow, and
46.  anoo, sore de, sono maa, miñna de 46. um, then, that well, (we) all went, you
itTA wake desu yo. know.
47.  - . n ni sore hodo sepatumat TA 47. B- Yes. (It was)n't especially, that
zyootai de wa na(ku)... pressing circumstances...
48. * A- Zya nal ñ desu. 48. A- (It) isn't.
49.__B- Aa soo. 49. B- Oh (I) see.
50. * A- Tada, anoo, byooin ni nyuuiñ-site RU 50. A- Just, um, (his) being in the hospital,
tte iu koto de, and
5 1 _ B - . 51. - .
52._A- Sore de ittara, 52. A- Then, when (I/we) went,
53._ anoo, mo, kanari ne, yasehosotte, 53. um, oh, (he) had grown very thin, and
54._ moo, kokkara kuda o toosite, 54. oh, (they) pass a tube from here, and
55. _ hidol ñ desu yo. 55. (it)'s awful', you know.
56._* kootoogañ desU kedo, 56. (It)'s throat cancer but,
57._ anoo, syokuzi-dekimas EÑ kara ne, 57. um, (he) can't eat so,
58._ de. moo, yasehosotte masiTA kara, 58. and, oh, (he) had grown thin so,
59.  aa, dame ka naa nañte omotte 59. oh, (I) was thinking this is it but,
masiTA kedo,
6 0 . _ B - . 60. B- Yes.
61._* A- geñki yoku syabeRU ñ desu yo ne. 61. A- (he) talks in good spirits, you know.
6 2 . _ B - Aa. 62. B- Oh.
63. A- Koñna ni yasetyatte na, sore de, 63. A- (He) says, "(I)'ve gotten so thin,
nañka yami zyoozu no sinibeta da haven't (I)" then something like (he)'s
to ka itte, good at getting sick and bad at dying,
and
64._ syabette RU ñ desu yo ne. 64. (he)'s talking, you know.
65._B- . 65. B- Yes.
66. * A- De, sono toki, itu mo kamera o 66. A- And, that time, (I) don't always carry
motanal ñ desu yo. carry a camera, you know.
67._B- . 67. B- Yes.
68. A- De,yoñ-mai, 68. A- And, four pictures,
69. - Anata ga... 69. B- You...
7 0 . _  - . Yon-mai nokotte RU kara, 70. A- Yes. (There) are four pictures left so,
71. motte ikoo to omotte, 71. (I) think about taking (it), and
72._o motte tTa ñ desu yo. 72. (I) took (it), you know.
73._B- . 73. B- Yes.
74. A- Sore zyaa... 74. A- Well...
75. - Kazoku-zyuu de irasite, 75. B- (You) go with the whole family, and
76. A- Totte ageyoo ka. . Totte ageyoo 76. A- "Shall (I) take (your picture)?" Yes.
ka tte tte ne, And saying, "Shall (I) take (your
picture)?"
432 POLLY E. SZATROWSKI

77  totTa ñ desu yo. 77. (Î) took (his picture), you know.
78. - Otoosama... 78. B- Your father...
79._A- De,kirai NA no ni ne, 79. A- And, even though (he) hates it,
80.__B- Ñ. 80._B- m.
81 . _ A - K, kamera no hoo mukU ñ desu yo. 81. A- Like this, (he) faces the camera, you
know.
82._B- . 82. B- Yes.
83._ A- Sore de, anoo, kodomo mo ne, turete 83. A- Then, um, (I) brought the children too
ikimasita kara, so,
84. zyaa, mago to issyo ni doo tte ne, 84. saying, "Well, how about with the
grand-children",
85._B- Ñ. 85. B- m.
86._A- Ano, issyo ni pati pati to totte, 86. A- Um, (1) took (them) together snap
snap, and
87._o sono yoñ-mai totTA wake desu yo ne. 87. (I) took those four pictures, you
know.
88.  - . Ano, nokotte  no ne, sono 88. B- Yes. Um, the left over ones, (you)
firumu ga nokotte TA no o omoti ni took the one which had the left over
natTA no ne? film?
89. A- Soo na ñ desu. 89. A- That's right.
90. Sore ga doo site yon na no ka 90. (I) don't know why it's four but,
91._B- Aa. 91. B- Oh.
92. A- Yoñ te, añmari yoku nai rasii desu 92. A- Apparently, four is not very lucky but,
kedo ne,
93. B- Ñ. Sore wa, sosite, atarasii no wa 93. B- Hm. That, then, (you) put new one
irete, (film) (in), and
94. A- Irete ikimasu yo ne. 94. A- (I) put (it in) and go, you know.
95. - De mo, maa, 95. B- But, well,
96. A- De, watasi tte, soo iu toko 96. A- And, I, have that kind of thing so,
arimasu kara,
97.  - Karui kañzi de motte rasiTA ñ 97. B- (You) took (it) with a light feeling,
desyoo? didn't you?
98._A- Soo na ñ desu. 98. A- That's right.
98._ B- N. 99. B- Hm.
100. A- Sore de, anoo, maa, minna 100. A- Then, um, well, everyone talked, and
syabette ne,
101. zyaa, ano, watasi, nani ka attara 101. well, um, (I) said call me if there's
yonde ne to ka itte, something, and
102. ano, zya, kaeru wa yoo, nañte 102. um, saying well (I)'m going home
103._o_ baa to kaet TA ñ desu yo. 103. (I) went straight home, you know.
104._B- . 104. B- Yes.
105. A- De, uti e tuite, 105. A- And, (I) arrived home, and
106. go-huñ go ni ne, deñwa ga 106. five minutes later, there was a phone
kimasite, call, and
107. sakki naku narimasita nañte 107. saying (that he) just died, um there
ano, oneesañ kara deñwa arimasite, was a phone call from my older sister,
and
108._B- Aa. 108. B- Oh.
 "VIVIDNESS" A N D "NARRATIVE EVENTS" 433

109.  A- Bikkuri-sityat TA n desunyo. 109. A- (I) was totally dumbfounded, you

know.
110.o_ Da kara, sore ga saigo dat TA ñ 110. So, that was the last time.
desu.
111. - Donna syasin ga dekita ñ desu ka, 111. B- What kind of pictures came out?
112. A- Sore ga nee, tyañío ne, kamera no 112. A- That, properly, (he) faced towards the
hoo muite ne, camera, and
113. - . 113. B- Yes.
114. A- Koo, nañ te iu nok sira. 114. A- Like, (I) wonder how to say (it).
115 * Hoñto ni, koo. kamera ni sugarU 115. Really, like, like (he) is appealing to
mitai no ne, hyoozyoo o site, the camera, (he) makes an expression,
and
116._B- . . 116.B- Yes. Yes.
117. A- Sore de, mago to wa tyañto te o ne, 117. A- Then, (he) held hands with the grand-
niģitte ne, children, and
118. - A soo. 118. B- Oh. Is that right.
119. :: - . Soo i u syasin ga . . . ,soremo 119. A- Yes. Those kind of pictures..., also
zeñzeñ piñboke zya nal ñ desu yo. they are not out of focus at all, you
know.
12()._B- Asoo. 120. B- Oh. Is that right.
121. A- De,haha ni s o r e o agemasita 121. A- And, (I) gave those to my mother but,
kedo nee.
122.__B- Ñ. 122. B- Hm.
123. * A- De, husigi desU nee. 123. A- And, (it)'s amazing, isn't it.
1 2 4 . _ B - Ñ. 124. B- Hm.
125. A- Ima, koo kero kero syabette masu 126. A- Now, (I)'mtalking away like this but,
kedo nee,
126. maa, sono toki wa, tyotto ne, da 126. well, then, a bit, so, (it) seems like
kara, soo iu tokoro tte aru mitai that kind of thing happens, you know.
desu yo.
127. - Soo na no ne-kitto 127. B- That's right, for sure.
128. Ma, soo iu mono ga, yahari 128. Well, that sort of thing may be, afterall
niñgeñ no ikite ru tte koto na no what life is all about...
ka mo sirenai si ne...
 SUBORDINATION" AND NARRATIVE EVENT STRUCTURE

SANDRA A. THOMPSON
University of California, Santa Barbara

1. Introduction
There has been a fair amount of dicussion in the literature from the past
several years on the possibility of a distinction in narrative discourse between
two types of linguistic material: material whose primary function is to report
the narrative EVENTS of the story, those which are temporally ordered with
respect to each other, and material whose primary function is something
other than this. One of the earliest such proposals and one of the most precise
is that of Labov and Waletzky (1967), in which narrative sequence is
described in terms of "displacement sets" and "temporal juncture":
Displacement set of clause c-: the set consisting of the clauses before
which c can be placed without affecting the temporal sequence of the seman­
tic interpretation, c itself, and the clauses after which it can be placed without
changing the temporal sequence of the original semantic interpretation. p.
22
Temporal Juncture: the boundary between two clauses which are tem­
porally ordered with respect to each other, i.e., whose displacement sets do
not include each other. p. 25
Any clauses whose range does not cross a temporal juncture between
two other clauses is defined as a narrative clause.
Labov (1972) repeats this distinction between "narrative" and "non-nar­
rative" clauses: a narrative text is a "sequence of two clauses which are tem­
porally ordered: that is, a change in their order will result in a change in the
temporal sequence of the original semantic interpretation" (p. 360).
A number of linguists have taken this definition as a starting point in
attempting to relate various grammatical phenomena to this distinction
between narrative and non-narrative portions of text (see especially Du Bois
(1980), Grimes (1975, chaps. 4, 5, 6), Hopper (1979a), (1979b), Hopper and
Thompson (1980), Longacre (1976, 1983), Sheffler (1978), Silva-Corvalan
436 SANDRA A. THOMPSON

(1983), and Wald (1973)). The work of Hopper and of Hopper and
Thompson introduced the terms "foregrounding" and "backgrounding" for
this distinction; since then there has been much discussion as to how the dis­
tinction should be defined, what the criteria might be for determining, in a
given text, which portions are "foregrounded", and whether the distinction
exists at all.
While the terms "foregrounding" and "backgrounding" were intended
to reflect the temporal ordering criterion of Labov, in both Hopper's papers
and in Hopper and Thompson there are suggestions that the foregrounded
portion of a narrative discourse also carries the most "important" material, or
the "backbone" of the story. Since then, it has become clear that the temporal
ordering criterion and the "important event" criterion need to be sharply dis­
tinguished. At least two studies (Kalmar (1982) for Czech and McCleary
(1982) for Brazilian Portuguese), for example, have shown that while sequen-
tiality might be relevant to aspect marking, "importance" or "backbone" may
not be. 2
In a very interesting recent paper, Reinhart (to appear) discusses the
relationship between subordinate clauses and event structure in the context
of the relationship between the temporal organization of narratives and the
principals of gestalt perception. Reinhart defines "foreground" as the
sequence of narrative clauses, as defined by Labov, and suggests that "a pow­
erful means" for marking background is "the use of syntactic embedding"
(ms., p. 22). She further claims that "material presented in subordinate
clauses cannot normally be foregrounded" (ms., p. 22), but that writers can
sometimes "play" with the foreground-background relations such that a nar­
rative clause can "function as background" if it is marked syntactically as sub­
ordinate (ms., p. 26-7), just as long as the events are still ordered "on the
same time axis" as in the represented world (ms., p. 27).
In this paper I would like to discuss the correlation between "subordi­
nate clauses" in English written narrative and the notion of temporal
sequencing. The question, then, is: in written English narrative discourse, is
"subordination" inversely correlated with "foregrounding" in the sense of
"sequentially ordered"? If not, why not?

2 Data Base
For this study, I took two narrative passages from the personal narrative
by Herbert Terrace called Mm, the story of Terraces's experience attempting
to teach the chimpanzee Nim Chimpsky to use American Sign Language. In
 "SUBORDINATION" AND NARRATIVE EVENT STRUCTURE 437

order to investigate the relationship between "subordinate" clauses and tem­

poral sequencing in a narrative, I attempted to identify the predicates which
were in temporal sequence, or "on the time line", as well as those which
occurred in "subordinate" clauses. Both these notions will be elaborated in
the context of the passages, which are given just below; the predicates
evaluated as being in temporal sequence are in capital letters, and those in
"subordinate clauses" are italicized:3
Passage A
The most memorable reunion I observed occurred between Nim and
Stephanie and her family, It took place at Delafield two days before Nim was
returned to Oklahoma. Knowing how excited Nim gets when he sees a former
caretaker, even from a distance, I took pains to insure that Nim had as little
an inkling as possible of the presence of his visitors. I ASKED Stephanie and
her family to sit quietly, around the corner from the main entrance to the
house, so that Nim, who was playing in front of the house, would not see
them. I then ASKED Dick Sanders, who was with Nim at the time, to walk
with him toward the site I had chosen for the rendezvous between Nim and
his first family. Nim was midly excited when he SAW me and SIGNED
"PLAY". That reaction was quite familiar. At that time I was visting the
house on a daily basis. Just about every time Nim saw me, he reacted the
same way: a momentary round of playfulness, a quick hug, and no great dis­
appointment when I put him down.
Stephanie had visited Nim at Delafield on two prior occasions. I had
witnessed the most recent of these reunions. From what I saw and from what
Stephanie told me about her earlier reunion, it was apparent that Nim was
especially delighted to see her. But none of the reunions I had witnessed or
heard about between Nim and Laura, Stephanie, Marika, or Walter had pre­
pared me for the intensity of Nim's reaction when he suddenly CAME
UPON Stephanie and her family sitting in a semicircle near one of his favo­
rite trees.
Nim SMILED a smile the size of which I had never seen and
SHRIEKED in a way Yd never heard. At first he seemed too excited even to
hug Stephanie. His smiling and shrieking continued for what seemed to be at
least three minutes. During that time he SAT DOWN across from
Stephanie. While looking back and forth at Stephanie, WER, Joshua, and
Jennie, he CONTINUED TO SHRIEK, SMILE, and POUND the ground
with joy. Only after he STOPPED SMILING and SHRIEKING did he GO
to Stephanie and HUG her. That hug was also interrupted by additional
shrieks. Quite a lot of noise from a normally silent chimpanzee!
After spending about fifteen minutes with Stephanie, Nim WENT over
to WER, Josh, and Jennie, and HUGGED each of them in turn. He then
RETURNED to WER and BEGAN TO GROOM him. A few minutes later
he MOVED over to Jennie, GROOMED her for a while, and then DID the
same with Josh. Nim's total involvement in hugging and grooming and in
438 SANDRA A. THOMPSON

playing with the LaFarges was all the more impressive in that he seemed
oblivious to the presence of a group of familiar and well-liked teachers who
were watehing from afar (Dick Sanders, Bill Tynan, Joyce Butler, Mary
Wamback, and me). To Nim, it seemed as if nothing mattered but being reun­
ited with his original family.
When he FINISHED GROOMING Josh, Nim T U R N E D to Stephanie
and her family and repeatedly S I G N E D "PLAY". In turn, each member of
the family responded. Josh and Nim C L I M B E D a tree together. Later, Jen­
nie and Nim C H A S E D each other around the base of the tree. Stephanie
and W E R also got into the act by running with Nim. Nim P L A C E D himself
between Stephanie and W E R , G R A B B E D their hands, and P U L L E D them
to and fro around the grounds of Delafield. All of this seemed to be an
incredible treat for Nim. Even after spending more than an hour with
Stephanie and her family, Nim was still smiling. I had seldom seen even brief
smiles of the kind that almost seemed pasted on Nim's face while he was with
the LaFarge family. The only other times I saw NIM smile that way were dur­
ing the first few seconds of playing with one of his favorite cats and during the
first few minutes of a reunion with a returning caretaker. I do not recall any
situation in which Nim's smile persisted for so long or any time when he dis­
played such unrestrained joy. 135-42.

Passage B

Even though I exerted considerable dominance over Nim and was wise
to most of his tricks, there were times he found new ways to test me. One
memorable incident occurred during February 1976, shortly after I had
begun a session with him in his Columbia classroom. That session was my
first opportunity to spend more than a few minutes with him in more than a
week. I could not be sure whether it was my recent absenteeism from the
classroom or a bad session with his previous teacher or both that caused the
cool reception I received from Nim. Even though he was well behaved, he
was unresponsive to my attempts to engage him in various activities.
One of the first things I did was TO SEE whether Nim wanted to use the
potty. Since he could not yet be relied on to sign " D I R T Y " every time he
needed to use the potty, his teachers made a practice of pointing to the potty
and asking " D I R T Y " ? . In response to my question Nim W A L K E D over to
the potty, P U L L E D D O W N his pants, and SAT on the potty. I
C R O U C H E D D O W N next to him and P A T T E D his back in praise of his
good toilet habits. Having been away from the classroom for a week, I paid
less attention than I should have.
In a flash Nim R A N out of the classroom and O P E N E D the door lead­
ing to the outside corridor. When I had entered the classroom complex, I had
carelessly left that door unlocked. This was not the first time that Nim had
escaped from the classroom. Because of his curiosity about the hallways and
offices of Schermerhorn Hall, I instituted a security system that called for all
teachers to lock themselves into the classroom complex. Not only should I
"SUBORDINATION" AND NARRATIVE EVENT STRUCTURE 439

have followed my own directive to lock the doors to the classroom and the
hallway, I should have also heeded the advice I had given to innumerable
new teachers: Never trust Nim. Particularly when a teacher was content with
his behavior, Nim would sense that the teacher had lowered his or her guard.
Unless he was reminded that the teacher was watching him very carefully he
would usually find a way to exploit any lapse of vigilance.

Catching Nim was no easy task, particularly since no one else from the
project was around. Even though I knew it would be difficult to catch Nim by
myself, I D E C I D E D A G A I N S T enlisting help from students and sec­
retaries. Nim had previously caused enough of a disturbance among the
department's secretaries, many of whom wanted to have nothing to do with
him.
It took me more than ten minutes TO CORNER Nim at the end of a cor­
ridor. During that time he led me on a merry chase through the second- and
third-floor corridors and stairwells of Schermerhorn Hall. Even when I had
cornered him he was able to keep out of my reach by hiding underneath a
table at the end of the corridor. Each time I reached under the table, Nim
either squirmed out of my reach or tried to bite my hand. Finally, I SEC­
U R E D a good grip around his wrist and with just enough of a twist, I was
able to persuade him TO COME OUT. In the process I incurred my first and
only bite from Nim.
I can still recall my anger as I MARCHED him back to the classroom.
Nim watched me very carefully as we walked down the hallways, looking for
another opening for escape. Unfortunately for him I had my hand wrapped
tightly around his wrist, and I was not about to relax that grip until we got
back to the classroom. After LOCKING both the inner and outer doors, I
SAT D O W N with Nim to see what he would do. I hoped that he would sign
" S O R R Y " and that he would try to convince me that he intended to behave.
Instead he M A X I M I Z E D the distance between us and T R I E D unsuccess­
fully to open the door.
I knew that the bite was a defensive response, but I still felt angry. This
was the first time Nim had bitten me and the first time he had tried to run
away from me. I felt that unless Nim was made to understand the strength of
my anger he might be encouraged to repeat his behavior. After Nim's second
attempt to get out of the room, I P I C K E D him UP and T H R E W him away
from me. I was quite surprised by what followed. I had thrown him so hard
that he ENDED UP HITTING the cinder block wall and not the carpeted
floor, as I had intended. I quickly D I S C O V E R E D that there was no reason
to feel concerned that I might have hurt Nim. GETTING UP, he H A L F
SMILED and S I G N E D " P L A Y " , which I INTERPRETED as a request to
throw him against the wall again. I O B L I G E D a second time before I
REALIZED that, far from punishing Nim, I was engaging him in a game of
roughhousing, which he loved.
At that point I was tempted to slap Nim across the face. On other occa­
sions I used that form of punishment after Nim delberately bit somebody or
440 SANDRA A. THOMPSON

made a bad mess in the kitchen at Delafield. Nim's reaction to my slap was
always one of instant terror. He screamed loudly and spun himself around,
often throwing himself into a tantrum. I DECIDED to save hitting Nim for
graver infractions. Instead, I tried a new form of punishment. Again, I
THREW Nim against the wall but only to provide myself with an opportunity
TO MAKE, from Nim's point of view, an unexpected exit from the class­
room. Nim's response was instant panic. From the adjacent observation
room I could see him banging on the locked door of the classroom. DIS­
COVERING that he could not open the door, he BEGAN TO SCREAM an
ear-piercing scream. That was followed by a full-fledged temper tantrum. I
watched and listened to Nim's temper tantrum as long as I could bear it. As
much as it hurt me to see Nim so upset, I knew that the longer I delayed my
return to the classroom, the more certain I could feel that Nim understood
that I was angry at him.
When I REAPPEARED in the classroom, Nim TRIED to jump into my
arms. To underscore his request to be reassured, he SIGNED "SORRY" and
"HUG" many times and continued to whimper, I REPLIED by showing him
the bite at the base of my thumb and signing "HURT". Nim STUDIED my
wound and SIGNED "SORRY" repeatedly. Then I WALKED to the outer
door and SIGNED "NO OPEN" repeatedly. Again Nim SIGNED
"SORRY" and this time ADDED "HUG". At that point I PICKED him UP
and ALLOWED him to hug me. 145-48.

3. "Subordination" and Temporal Sequentiality

Several researchers have commented on the relationship between "sub­
ordination" and sequential ordering. In Labov and Waletzky (1967) and in
Labov (1972), it is definitional that "subordinate clauses do not serve as nar­
rative clauses" (Labov (1972:362). This follows from their definition of narra­
tive given above: narrative clauses are those whose order cannot be reversed
without disturbing "the temporal sequence of the original semantic interpre­
tation", but with most clauses considered "subordinate", their order relative
to the main clause often CAN be reversed without disturbing the temporal
sequence. As Labov puts it, "once a clause is subordinated to another, it is
not possible to disturb semantic interpretation by reversing it". Thus, "it is
only independent clauses which can function as narrative clauses" (1972:
362).
However, dispensing with the "reversability" criterion, other linguists
have been inclined to consider certain "subordinate" clauses as part of the
temporal sequence. One example is Polanyi-Bowditch (1976). In the narra­
tive she discusses, the following two clauses are included as being part of the
"temporal structure":
 "
SUBORDINATION" AND NARRATIVE EVENT STRUCTURE 441

(1) a. When she began to arrange the flowers in a bowl

b. a small fly flew out.
It is clear that because the two events are presented in iconic order, it is
reasonable to consider thern to be part of the "temporal structure". However,
as McCleary (1982:8) points out, there is some difference between presenting
them this way and, say, as a pair of co-ordinate clauses, we will return to this
point below.
Reinhart (to appear), as mentioned above, has suggested that "syntactic
subordination" is a means for distinguishing foreground and background.
However, she also proposes to distinguish between "content" criteria for
identifying foreground and background, primarily iconic temporal order, and
"linguistic" criteria, that is, whether the clause is main or subordinate. Nor­
mally, she says, the two types of criteria coincide, but there may be cases
where "the linguistic criteria can impose a selection of foreground and
background which does not follow directly from the content criteria" (ms., p.
25). To illustrate, she takes the following portion from an oral narrative
reported by Labov and Waletzky (1967) (with italicization reflecting the
judgements of Hopper and Thompson (1980) as to what was foreground,
which Reinhart accepts):
This was just a few days after my father had died, and we were sitting shive.
And the reason the fight started ...He sort of ran out in the yard — this was
way out on Coney Island — and he started to talk about it. And my mother
had just sat down to have a cup of coffee. And I told him to cut it out.
Course kids, you know — he don't hafta listen to me. So that's when /
grabbed him by the arm, and twisted it up behind him. When I let go his arm,
there was a knife on the table, he just picked it up and he let me have it. And
I started to bleed like a pig...
Here it is worth quoting Reinhart:
The foreground units all consist of narrative clauses, i.e., temporally
ordered sequences. The point is, however, that at least one of the
background clauses also meets the definition of a narrative clause. This is the
clause, When I let go his arm, in the second paragraph. This clause is ordered
temporally after the event depicted in the preceding clause, and before the
event depicted in the next foreground clause. So given only content and
ordering criteria this should have been considered a foreground clause. But
the reason it functions as background here is that it is marked syntactically
as subordinate (ms., p. 26-7).4
From this quote, it is clear that Reinhart is taking Labov and Waletzky's
definition seriously: "foreground" can only include independent (main)
442 SANDRA A. THOMPSON

clauses which are temporally sequenced. It follows that if a clause is marked

as "subordinate'', then it must be part of the background. But this freedom to
"play" with the foreground-background relations is only possible, she says, if
the iconicity of the sequencing is left intact. That is, if clause A reports an
event belonging to the temporal foreground sequence and clause B reports a
temporally background event, then what cannot happen is this: A cannot be
marked as syntactically subordinate to B. So while (2)a. may be quite accept­
able, (2)b. is not:
(2) a. The host was telling another joke. Having already heard this
joke many times before, Rosa started to yawn.
b. The host was telling another joke. Starting to yawn / having
started to yawn, Rosa has already heard this joke many time
before. 5
Reinhart's paper gives us much to think about on the question of "subor­
dination" and "foregrounding". She takes "foreground" to exclude "subordi­
nation" on definitional grounds, but she points out that the choice an author
has between representing a temporally sequenced event as part of the "fore­
ground" (i.e., as a "main" clause) or as part of the "background" (i.e., as a
"subordinate" clause) is part of the artistry that makes a written text worth
reading. 6
According to Labov and Waletzky and Reinhart, then, foreground (nar­
ratives) clauses must not be syntactically dependent. But this exclusion by
definition precludes the possibility of determining whether there is any inde­
pendent relationship between "foreground" and "subordination". It is this
question that the present section is intended to address.
To identify the temporally sequenced clauses, I simply tried to deter­
mine which predicates named a punctual event that followed the previous
sequenced event and preceded the following sequenced event. But what
counts as "temporally sequenced" or "on the time line" is not always a
straightforward matter. Let me disucss one recurring type of problem which
arises in determining temporal sequence in narrative discourse.
The major type of difficulty in applying the criterion just given is that
there are predicates which appear superficially to be part of the temporal line
of the narrative, but which in fact serve to summarize either the previous few
temporally sequenced predicates or the following few temporally sequenced
predicates. In Passage A we can find an example of each situation. Consider
the following extract:
 "SUBORDINATION" AND NARRATIVE EVENT STRUCTURE 443

(3) Only after he STOPPED SMILING and SHRIEKING did he GO

to Stephanie and HUG her. That hug was also interrupted by addi­
tional shrieks. Quite a lot of noise from a normally silent chimpan­
zee!
After spending about fifteen minutes with Stephanie, Nim WENT
over to WER, Josh, and Jennie, and HUGGED each of them in
turn.
The predicate in question is spending, which might appear at first glance
to be part of the temporal squence, since it seems to follow the preceding
event of hugging Stephanie and precede the following event of going over to
WER, Josh, and Jennie. However, closer inspection reveals that this spend­
ing is actually recapitulative, and summarizes the preceding events of going,
hugging, and shrieking.
The other type of situation, where a predicate serves to provide an
advance summary, can be seen in the following excerpt form the third para­
graph, where Terrace summarizes what Nim did with each member of the
family:
(4) In turn, each member of the family responded. Josh and Nim
CLIMBED a tree together. Later, Jennie and Nim CHASED each
other around the base of the tree. Stephanie and WER also got
into the act by running with Nim. Nim PLACED himself between
Stephanie and WER, GRABBED their hands, and PULLED
them to and fro around the grounds of Delafield.
The verb responded in the first sentence of (4) is not part of the temporal
sequence because it anticipates the next several temporal events.
For the purposes of this discussion, then, predicates which summarize
preceding or following events will not be considered as part of the temporal
sequence.
Because one of the criteria for "temporally sequenced" is punctuality,
durative predicates also do not count as part of the temporal sequence. Thus
durative forms such as watched, walked, and looking for in the following
excerpt from Passage B are not considered as temporally sequenced:
(5) I can still recall my anger as I MARCHED him back to the class­
room. Nom watched me very carefully as we walked down the hall­
ways, looking for another opening for escape.
While the notion of "temporal sequence" is admittedly somewhat of an
444 SANDRA A. THOMPSON

idealization, then, it does seem possible to retain it as a framework for analyz­

ing narrative discourse by taking it to be rather strictly the set of predicates
denoting punctual events such that each one of these punctual events follows
the preceding one and precedes the following one. As mentioned above, the
predicates which are temporally sequenced by this definition are those which
appear in capital letters in Passages A and B.
Before we can move to a discussion of the relationship between temporal
sequencing and "subordination", however, I must first make precise just
what I mean by "subordination", and reveal why I am using it in quotes.
Characterizing the category "subordinate clause" in a given language is
not a straightforward matter. In view of the highly composite nature of the
items generally included in this category, of the lack of agreement among
grammarians as to what should be included, and of the difficulty in achieving
any cross-linguistic comparability with the term 7 , it seems appropriate to
define "subordinate clause" carefully before using it. My definition will be
extensional. I will be discussing four types of grammatically dependent
clauses by name; hereafter, when I use the term "subordinate clause", it will
refer to the heterogeneous set which includes all these types.
In the passages given above, each of the underlined predicates occurs in
a clause type which is dependent on another clause in one of these four ways:
1. it is a complement to another predicate, as in:
(6) Nim would sense that the teacher had lowered his or her guard.
2. it is non-finite, as in:
(7) Not only should I have followed my own directive to lock the
doors to the classroom and the hallway, ...
(8) ...he suddenly CAME UPON Stephanie and her family sitting
in a semicircle
3. it begins with a conjunction that marks it as an adverbial clause, as
in:
(9) Particularly when a teacher was content with his behavior,...
4. it is in a relative clause (including non-restrictives and "free rela­
tives"), as in:
(10) I do not recall any situation in which Nim's smile persisted for
so long
Any clause which exhibits one of these four types of dependency, then,
will be called a "subordinate" clause for the purposes of the remainder of this
discussion.
Having shown what I mean by "temporal sequencing" and "subordina-
 "SUBORDINATION" AND NARRATIVE EVENT STRUCTURE 445

tion", I will proceed to the hypothesis and the discussion of the results.
Unlike Labov and Waletzky, I do not define away the possibility of sub­
ordinate clauses' predicates being part of the set of temporally sequenced
predicates, and unlike Reinhart, I do not call the set of temporally sequenced
clauses the "foreground". The hypothesis, then, is a two-part one:
a. The vast majority of subordinate clause predicates will not be on
the time line.
b. Those which are on the time line are doing other discourse work in
addition to naming a temporally sequenced event.
Les us consider each of these points in turn.
Point a. is intuitively correct, and is easy to verify. Table 1 presents the
relevant numbers:

Table 1
Subordinate predicates on and off the time line
ON TIME LINE OFF TIME LINE TOTAL
PASSAGE A 5(=11%) 43 (=89%) 48(=100%)
PASSAGER 12 ( = 11%) 101 (=89%) 113 (= 100%)
Table 1 shows that part a. of the hypothesis is strongly and consistently
confirmed for these two passages: 89% of the subordinate predicates, as here
defined, occcur in non-temporally sequenced clauses.
Part b. of the hypothesis deals with those subordinate predicates that can
be counted as part of the temporal sequence: in the passages given above,
these are the predicates which both appear in capital letters (to show that they
are temporally sequenced) and are underlined (to show that they occur in
subordinate clauses). Part b. of the hypothesis predicts that it will always be
possible to demonstrate that these are performing some other discourse func­
tion in addition to indicating a successive event in the temporal sequence. To
show this, we must look at each case in turn.
Let's consider Passage A first. In this passage, it happens that all of the
subordinate temporally sequenced predicates are in adverbial clauses. Recall
that Reinhart (to appear) suggested that writers can code "foreground"
events with "background" (i.e., subordinate) clauses (her example being the
When I let go his arm from the Labov and Waletzky story) as long as the
iconicity in order is maintained. She goes on to say that there is much free­
dom on the part of the writer as to which events to represent as "part of the
446 SANDRA A. THOMPSON

foreground" and which not to; the choice, she says, "is subject to various
(widely studied) aesthetic and functional, or perceptual considerations and it
is part of what enables narration to be a work of art" (p. 29).
What I would like to do here is to pursue this implicit line of reasoning:
without making the assumption-laden statement that writers may choose to
represent foreground events as "part of the background", we can neutrally
ask WHY in fact a writer does choose to represent a temporally sequenced
event with an adverbial clause, thereby making it dependent on another
clause?
Part b. of our hypothesis predicts that the author is always trying to do
something else in addition to conveying temporal order. That is, there are
other discourse-organizing principles involved in constructing texts besides
those of temporal ordering. Thus, in the third paragraph, we have the follow­
ing sentence:
(11) Only after he STOPPED SMILING and SHRIEKING did he GO
to Stephanie and HUG her.
In terms of temporal sequencing, Nim's ceasing smiling and shrieking
preceded his going to Stephanie and hugging her. But Terrace also wants to
convey the intensity of Nim's reaction to Stephanie and her family, and one
way he accomplishes this is to frame these sequential events such that the hug­
ging (Nim's normal first reaction) is reported to have been possible only after
he'd calmed down enough to stop smiling and shrieking. Thus a dependency
of one event on the other has been created which cuts across their temporal
relationship.
Similar in principle though different in detail are the SAW in the first
paragraph and the CAME UPON at the end of paragraph two; the sentences
in which these occur are give as (12) and (13):
(12) Nim was mildy excited when he SAW me and SIGNED "PLAY".
(13) But none of the reunions I had witnessed or heard about between
Nim and Laura, Stephanie, Marika, or Walter had prepared me
for the intensity of Nim's reaction when he suddenly CAME
UPON Stephanie and her family sitting in a semicircle near one of
his favorite trees.
In the case of SAW in (12), Nim's seeing Terrace and signing "PLAY"
took place consecutively after Terrace asked Dick Sanders to walk with Nim
towards the reunion site. But Terrace is relating Nim's seeing him as being
simultaneous to Nim's being excited rather than as being consecutive to the
 "SUBORDINATION" AND NARRATIVE EVENT STRUCTURE 447

asking of Dick Sanders, and this is reflected in the coding of the punctual and
temporally sequenced SAW with a when clause.
Exactly the same explanation seems appropriate for CAME UPON in
(13): though CAME UPON occurs temporally after SAW and SIGNED, the
two previous temporally sequenced events, Terrace is here describing the
simultaneity of the intensity of Nim's reaction and his coming upon Stephanie
and her family. Once again, this simultaneity is what is being coded rather
than the temporal relationship between the sequenced event and the previous
sequenced events.
In these instances of temporally sequenced events being reported in
adverbial clauses, then, we can make a case for these clauses having other dis­
course connections to make besides the temporal linking one. Support for
this position can be found in the fact that in both (12) and (13), there is some
distance between the event verb in question and the previous sequenced
event verb, five verbs in the case of SAW, but sixteen verbs in the case of
CAME UPON.
This point about distance between two temporally sequenced event
verbs seems to be involved in the explanation of the last instance of a tempor­
ally sequenced event being represented by an adverbial clause in Passage A.
This is the first verb in the last paragraph, FINISHED GROOMING.
(14) When he FINISHED GROOMING Josh, Nim TURNED to
Stephanie and her family and repeatedly SIGNED "PLAY".
In this case, there has been some intervening descriptive material, unre­
lated to the temporal line, between the predicate in question and the preced­
ing temporally sequenced event predicate, a discussion of how impressive
Nim's reaction to the LaFarges was in view of the other favorite teachers on
the scene. Here what the adverbial clause seems to be doing that couldn't be
done by an independent clause is to relate the clause following it back to the
ongoing temporal line.
The ability of adverbial clauses to perform this relating-back function
has been described by Longacre in Longacre and Thompson (1985) as the
"cohesive" function of front-placed adverbial clauses. Chafe (1984), has
described the same phenomenon in terms of front-placed adverbial clauses
serving as "guideposts" to information flow, "providing a temporal, condi­
tional, causal, or other such orientation for the information in the upcoming
main clause" (p. 444). Following Fries (1983), I have talked about the way in
which such "orientation" is provided in terms of such a preposed adverbial
448 SANDRA A. THOMPSON

clause's role in a chain of expectations in Thompson (1985), and a similar sort

of explanation seems to hold here: in a narrative, the unmarked expectation
is that the temporal line will be returned to after a digression of any kind. One
function of adverbial clauses on the temporal line in a position preceding the
next temporal event, then, would be to respond to this expectation by explic­
itly relating that next event to the already-established temporal line, and that
is just what is happening here. In example (14), which opens the final para­
graph of the passage, our attention is brought back to the time line by the
adverbial clause When he FINISHED GROOMING Josh. The preceding
temporal event was Nim's grooming Josh, seven verbs back in the text. Then
Terrace tells us about the other teachers observing the reunion. The when
clause in (14), by directly mentioning the grooming of Josh again, orients us
back to the time line in a way that a simple independent clause could not have
done.
This "orienting" function, then, is perhaps the most prominent function
of initial adverbial clauses. The reason why they are able to function this way,
I suggest, is precisely because a temporally sequenced event is being coded in
a marked form, that is, in a form which makes it grammatically dependent on
another clause.
So far, then, we have seen that the five cases of subordinate clause pred­
icates occurring on the time line in Passage A provide strong support for Part
b. of our hypothesis, that such predicates are always demonstrably perform­
ing additional discourse work besides simply reporting the next event in the
temporal sequence. Let us now consider Passage B.
Passage B is somewhat more complex, in that the types of subordinate
clause predicates form a more diverse set. It will be useful to consider each
type separately.
Two of the twelve subordinate time-line predicates in Passage B are in
initial adverbial clauses of the type we discussed just above for Passage A.
The first is the verb REAPPEARED at the beginning of the last paragraph.
The sentence in which it occurs, after a description of Nim's panic at being left
alone in his classroom, is this:
(15) When I REAPPEARED in the classroom Nim TRIED to jump
into my arms.
That this initial when-clause orients the reader back to the unexpected
exit Terrace had previously made from the classroom is obvious. Its orienting
function is precisely parallel to that of the initial adverbial clauses discussed
for Passage A.
 "SUBORDINATION" AND NARRATIVE EVENT STRUCTURE 449

The second case of a time-line predicate in an initial adverbial clause is

the one found in the middle of the story, where Terrace has marched Nim
back to the classroom:
(16) After LOCKING both the inner and outer doors, I SAT DOWN
with Nim to see what he would do,
Once again, the orienting-back function here is clear: it carries the
reader back to the lengthy discussion Terrace has provided about the need for
all teachers to lock themselves into the classroom to guard against just such
escapes at the one he is describing.
Note, incidentally, that the three "initial" adverbial clauses we have just
considered in (14), (15), and (16), that is, those which precede the main
clause with which they are associated, in addition to this cohesive function as
predicted by Fries, Longacre, Chafe, and Thompson, all also contain lexical
material which is recapitulative, which provides "cohesion", in the sense of
Halliday and Hasan (1976) and Longacre and Thompson (1985), with previ­
ous material (FINISHED GROOMING in (14), REAPPEARED in (15),
and LOCKING in (16)). This is clearly not the case with ordinary main clause
temporally sequenced predicates, which do not "look back" at all, but simply
report the next event in the sequence.
Functioning very similarly to these initial adverbial clauses are the initial
participial clauses in the following examples:
(17) GETTING UP, he HALF SMILED and SIGNED "PLAY",...
(18) DISCOVERING that he could not open the door,...
Just as we saw for (14)-(16) above, these participials both revert back to
earlier events in the text: GETTING UP in (17) clearly refers back to having
been thrown against the wall, while the DISCOVERING clause in (18) refers
back to the locked door.
Two other adverbial clause temporally sequenced predicates in the data
are of the type illustrated by when he suddenely CAME UPON in Passage A
(example (13)). That is, some other temporal relation is being described
besides the sequence of the predicate in question with the previous and fol­
lowing events. Here are the examples from Passage B:
(19) I can still recall my anger as I MARCHED him back to the class­
room.
(20) I OBLIGED a second time before I REALIZED that, far from
punishing Nim, I was engaging him in a game of roughhousing,
which he loved.
450 SANDRA A. THOMPSON

In both these cases, Terrace is indicating another temporal relation: in

(19), it is simultaneity with the anger, and in (20), it is the fact that it took a
second throwing before he realized how Nim was manipulating him.
Another group of subordinate predicates on the time line consists of
infinitives:
(21) One of the first things I did was TO SEE whether Nim wanted to
use the potty.
(22) It took me more than ten minutes TO CORNER Nim at the end of
a corridor.
(23) Finally, I SECURED a good grip around his wrist and with just
enough of a twist, I was able to persuade him TO COME OUT.
(24) Again, I THREW Nim against the wall but only to provide myself
with an opportunity TO MAKE, from Nim's point of view, an
unexpected exit from the classroom.
Each of these four cases illustrates a grammatical consequence (in-
finitivization) of having chosen a means of expression which reports more
than just that an event happened. In the case of (21), Terrace is not simply
saying that he determined whether Nim wanted to use the potty, but that it
was one of the first things he did with Nim. In the case of (22), once again,
Terrace is clearly not simply narrating that he cornered Nim; that could have
been done with an independent clause. He is including in one grammatical
construction the amount of time it took along with the event being reported,
which results in infinitivization. And similarly in (23), Terrace is not just
announcing that Nim came out, but that it took some persuasive tactics on his
part; in these cases, the grammatical consequence of choosing to express
more than just that the event happened results in a construction in which the
time-line predicate appears in the form of an infinitive. Finally, in (24), we
see a substantial amount of grammatical complexity: the predicate TO
MAKE an unexpected exit is embedded as an adjunct to the noun opportun­
ity, which is itself part of an infinitival purpose clause. In this way, the author
can relate his leaving the classroom causally to the third occurrence of throw­
ing Nim against the wall.
Similar sorts of explanations are available for the last two subordinate
predicates which are temporally sequenced. The first is found in the non-
restrictive relative clause in this sentence:
(25) ... he half smiled and SIGNED "PLAY", which I INTER­
PRETED as a request to throw him against the wall again.
 "SUBORDINATION" AND NARRATIVE EVENT STRUCTURE 451

The second is in the complement to the degree adverb so hard in this sen­
tence:
(26) I had thrown him so hard that he ENDED UP HITTING the cin­
der block wall ...
Once again, both of these examples are straighforward cases, it seems to
me, of the principle that subordinate predicates on the time line are doing
other discourse work than just reporting an event. (25) illustrates the dual
role of the predicate INTERPRETED as both the next temporal event and as
part of the non-restrictive relative clause characterizing Nim's behavior. (26),
similarly, shows the dual role of ENDED UP HITTING as the next temporal
event and as part of the clause indicating how hard Nim had been thrown.

4. Conclusion
What we have seen, then, in this paper is that the temporally sequenced
events in a written narrative need not form a homogeneous class in terms of
their grammatical representation. While the majority are coded by indepen­
dent clauses, a significant subset are "subordinate". In each of these cases I
have shown how the use of a subordinate clause allows the writer to
accomplish a text-creation goal in addition to the obvious one of maintaining
the temporal line. As Reinhart says, this is certainly part of what makes a text
a work of art; I would suggest that it is also part of what makes a text readable.
A strictly linearly organized written narrative text would be not only boring,
but hard to attend to, for the well-known reasons discussed in the gestalt per­
ception literature. In written narrative, then, in a language with a well-devel­
oped written tradition, some use of subordinate clauses on the time line
seems to correlate with the multiple purposes to which a writer puts temporal
sequenced predicates.

NOTES

1) I wish to thank the following people for their help in making suggestions for improving an
earlier draft of this paper: Marianne Celce-Murcia, Hyo Sang Lee, Robert Longacre, Carol Lord,
Jim Martin, Susan Mordechay, Mickey Noonan, R. McM.Thompson, Russell Tomlin, Benji
Wald, and members of my UCLA seminar on clause combining, Spring, 1984. This paper has
benefitted greatly from their input, even though some of them may not recognize it.
2) Of some interest to our attempts to understand "importance"1 in discourse is Reinhart's posi­
ton that "foregrounding" (for her, the material on the time line) is not related to "importance" in
a discourse: "there is no reason to expect that the "narrative" temporal sequence should be more
452 S A N D R A A. T H O M P S O N

important...than the non-narrative units" (ms., p. 11). Her arguments for their independence rest
on the perceptual neutrality of the gestalt figure-ground distinction, where the figure need not be
more "important" than the ground. I find these arguments persuasive, and I think they strongly
support a careful distinction between what linguists have meant by "foreground" (or temporally
sequenced clauses) and "importance". Chvany (to appear) takes a similar position.
3) Because I am using capitalization to indicate temporally sequenced predicates, I have
enclosed the capitalized signed words in the original text in quotes for ease of reading.
4) A reasonable analysis of the sentence
(i) So that's when I grabbed him by the arm, and twisted it up behind him.
in this passage might well also posit its when clause as a subordinate, and thus, for Reinhart,
background, clause, but she does not consider this possibility.
5) Note, incidentally, that the participial clause type with which Reinhart chooses to illustrate
her point about "subordination" is precisely one about which there will be little agreement as to
whether it is "subordinate" or not. Here the term "dependent" might have been a better choice.
6) There is much else in this paper of great interest, especially with respect to the issue of gestalt
perception and the organization of discourse.
7) For discussion see Haiman and Thompson (1984) and Mattheissen and Thompson (to
appear) and references cited there.

REFERENCES

Chafe, Waliace. 1984. "How people use adverbial clauses". Proceedings of

the Tenth Annual Meeting of the Berkeley Linguistics Society. Berkeley:
Berkeley Linguistics Society.
Chvany, Catherine V. To appear. "Foregrounding, 'transitivity', saliency (in
sequential and non-sequential prose) 1 '. Essays in Poetics,
Du Bois, John. 1980. "Beyond definiteness: The trace of identity in dis­
course". In Chafe, Wallace, ed., The Pear Stories, Norwood, N.J.: Ablex.
Fries, Peter. 1983. "On the status of theme in English: Arguments from dis­
course". In Petofi, J.S. and E. Sozer, eds., Micro and Macro Connexity of
Texts. Hamburg: Helmut Buske Verlag.
Grimes, Joseph. 1972. The Thread of Discourse. The Hague: Mouton.
Haiman, John and Sandra A. Thompson. 1984. ""Subordination" in univer­
sal grammar". Proceedings of the Tenth Annual Meeting of the Berkeley
Linguistics Society. Berkeley: Berkeley Linguistics Society.
Halliday, M.A.K, and Ruqiaya Hasan. 1976. Cohesion in English. London:
Longman's.
Hopper, Paul. 1979a. "Some observations on the typology of focus and
aspect in narrative language", Studies in Language 3.1: 37-64.
 "SUBORDINATION" AND NARRATIVE EVENT STRUCTURE 453

—--. 1979b. "Aspect and foregrounding in discourse". In T. Givón, ed.,

Discourse and Syntax. (Syntax and Semantics Vol. 12.) New York:
Academic Press.
. and Sandra A. Thompson. 1980. "Transitivity in grammar and dis­
course". Language 56.2: 251-299.
Kalmar, Ivan. "Transitivity in a Czech folk tale". In Paul J. Hopper and
Sandra A. Thompson, eds., Studies in Transitivity. (Syntax and Semantics
Vol. 15.) New York: Academic Press.
Labov, William. 1972. 'The transformation of experience in narrative syn­
tax". In Language in the Inner City. Philadelphia: University of Pennsyl­
vania Press.
. and Joshua Waletzky. 1967. "Narrative analysis: oral versions of per­
sonal experience". In Helm, June, ed., Essays on the Verbal and Visual
Arts, Proceedings of the 1966 Annual Spring Meeting of the American
Ethnological Society. Seattle: University of Washington Press.
Longacre, Robert E. 1976. "The discourse structure of the flood narrative".
In MacRae, George, ed., Society of Biblical Literature, 1976 Seminar
Papers. Missoula, Montana: Scholars Press.
. 1983. "Vertical threads of cohesion in discourse". In Neubauer, F.,
ed., Coherence in Natural Language Texts. Hamburg: Helmut Buske Ver­
lag.
Longacre, Robert and Sandra A. Thompson. 1985. "Adverbial clauses". In
Shopen, Timothy, ed., Linguistic Typology and Syntactic Description.
Cambridge: Cambridge University Press.
Matthiessen, Christian and Sandra A. Thompson. To appear. "The struc­
tures of discourse and 'subordination'". In Haiman, John and Sandra A.
Thompson, eds., Clause Combining in Grammar and Discourse. Amster­
dam: Benjamins.
McCleary, Lelan. 1982. Transitivity and grounding in a Portuguese narrative
Text". UCLA M.A. thesis.
Polanyi-Bowditch. 1976. "Why the whats are when: Mututally contextualiz-
ing realms of narrative". Proceedings of the Second Annual Meeting of the
Berkeley Linguistics Society, Berkeley: Berkeley Linguistics Society.
Reinhart, Tanya. To appear. "Principles of gestalt perception in the temporal
organization of narrative texts". Linguistics.
Sheffler, Margaret. 1978. "Mundurku discourse". In Grimes Joseph, ed.,
Papers on Discourse. Dallas: Summer Institute of Linguistics.
Silva-Corvalan, Carmen. 1983. "Tense and aspect in oral Spanish literature".
454 SANDRA A. THOMPSON

Language 59.4:760-780.
Thompson, Sandra A. 1985, "Grammar and written discourse: Initial vs.
final purpose clauses in English". Text 5: 55-84.
Wald, Benji. 1973. "Variation in the system of tense markers of Mombasa
Swahili". Columbia University Ph.D. dissertation.
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS

RUSSELL S.TOMLIN
University of Oregon

1. Introduction
This paper examines the relationship between the thematic organization
of narrative discourse production, as reflected in its episodic organization,
and the syntax of reference, the differential use of nominal and pronominal
form in discourse production. It is argued, using data drawn from two inde­
pendent data sources, that the syntax of reference is directly a function of
episodic or thematic boundaries at a relatively local level. The episodic boun­
daries in turn can be seen as a speaker-based re-orientation of attention dur­
ing the on-line process of discourse production.
Background. Investigation of the syntax of reference and its functions,
particularly in discourse comprehension, has been focused in a variety of
areas: syntactic constraints on reference, semantic constraints on reference,
formal text and discourse constraints, cognitive constraints, and so on. Van
Dijk and Kintsch (1983:161-182) provide a detailed survey of psycholinguistic
research in this area, while Clancy (1980) offers a careful review of linguistic
research in this area.
There are two major, and interrelated, theoretical orientations taken in
the study of the function of the syntax of reference. One orientation, which
can be called the recency or distance approach, considers the alternation
between noun and pronoun to be primarily a function of time (recency of ref­
erence), which is manifested in the text artifact by the number of clauses or
propositions intervening between a given referent and its antecedent (dis­
tance). Relying primarily on quantitative measurement of referential dis­
tance (the number of clauses between subsequent mentions of a given refe­
rent), Givón and his associates (Givón 1983; Givón, Kellogg, Posner. and
Yee 1984, for instance) have shown convincingly an iconic relationship hold­
ing between referential distance and the amount of coding material used to
456 RUSSELL S. TOMLIN

maintain reference. As referential distance increases, so does the amount of

coding material required to maintain reference.
Clancy (1980) offers a similar conclusion in her analysis of the pear
stories data (Chafe 1980) for both English and Japanese speakers. Within the
psychological literature, Clark and Sengul (1979) and others have conducted
studies of discourse comprehension in which the time of retrieval of referents
in pronominal form increases as the distance between referent and anteced­
ent increases.
The recency/distance approach for referential syntax is attractive in its
relative simplicity, and does capture well some important generalizations
about languages and groups of speakers. But there are two classes of coun­
terexamples this model fails to address. First, it is clear that full nouns can be
used to refer to an antecedent even only a single clause away, and such excep­
tion can occur even in cases in which ambiguity resolution is not at issue.
Second, it also appears that in some cases pronominal reference can be sus­
tained for more than a clause or two. Thus, while a recency/distance
approach, does address well general characteristics of the syntax of refer­
ence, it does not account for the full range of use exhibited by individuals
engaged in discourse production and comprehension.
The second orientation, what can be called the episode/paragraph
model, considers the alternation between noun and pronoun to be a function
of the limited capacity of working memory, which is manifested in the text
artifact primarily through its paragraph, or episodic, organization. Hinds
(1977, 1979) argues this for Japanese discourse; van Dijk (1982) also consid­
ers the importance of episodes to the syntax of reference. Kintsch and van
Dijk (1978, 1983) offer a detailed treatment of subject and topic interaction
with the syntax of reference in which pronominal form is linked with the
maintenance of reference across an adjoining clause but mediated by a higher
level thematic macroproposition. Givón (1983) has also discussed the role of
thematic organization on the syntax of reference. More recently. Fox (1984)
provides convincing empirical evidence from written narrative to support an
episode/paragraph approach to the syntax of reference.
One important complication of the episodic/paragraph approach consid­
ers the alternation of noun and pronoun to be due to differential "focusing",
or foregrounding, of a given referent. There are a number of important
studies in the psychological, linguistics, and artificial intelligence literatures
that pursue this complication: Reichman (1981), Marslen-Wilson, Levy, and
Tyler (1982), Tyler and Marslen-Wilson (1982), Chafe (1974, 1976, 1984),
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 457

Grosz (1977), among others. These studies argue that pronouns are used to
make reference when a particular referent is in a state of high focus, or
"foregrounded", in a given linguistic context (presumably something like a
conceptual paragraph). Full NPs are used for referents in lower focus. Chafe's
notion of 'activation' (Chafe 1974, 1976, 1984) represents a very similar point
of view within linguistics.
The episode/paragraph approach, with its accompanying focus or fore­
grounding augmentation, is attractive in two major respects. First, it shows a
greater sensitivity to subject or text specific variation in distance between ref­
erence and antecedent. Second, it articulates a natural connection between a
particularly important linguistic unit, the paragraph, and its psychological
correlate, the limited capacity working memory.
The weakness of this approach also lies in two areas. First, the critical
theoretical linguistic notions, — paragraph, episode, and focus, — are
weakly defined and generally resistant to empirical analysis. Second, and not
independent of the first weakness, the practical identification of such notions
in genuine text data depends on relativistic thematic notions of relevance and
salience, which are similarly vague at the present moment (van Dijk and
Kintsch 1983:171; but see also Tomlin 1984, 1985).
This paper argues for the episode/focus approach of the functional syn­
tax of reference. It is demonstrated that during the on-line task of discourse
production subjects will use a full noun to reinstate reference across an
episode boundary, and they will use a pronoun to maintain reference within
a particular episode. The recency/distance approach is specifically rejected
for it provides an inadequate treatment of the present experimental data.
There is one additional parameter affecting the syntax of reference
which cannot be ignored: ambiguity resolution. Full nouns are clearly used to
maintain reference when there exists a possible ambiguity of referential
interpretation by the listener. Ambiguity resolution will not be considered at
all in this study, except to filter from the data some rather straightforward
instances, switch-subject references.
Finally, it should be pointed out that almost all studies of the syntax of
reference focus on strategies of the listener, on discourse comprehension. The
historical reason for this is clearly methodologically grounded: comprehen­
sion studies lend themselves better to needed experimental control. Still, it is
important to understand the strategies used by the speaker in producing dis­
course, for it is by no means necessary that comprehension and production
strategies be the same. This study examines the syntax of reference from the
458 RUSSELL S. TOMLIN

point of view of the speaker.

Hypothesis. The general hypothesis be considered here can be found in
(1):
( 1 ) General hypothesis: The syntax of reference in discourse production
is tied directly to psychological processes of attention as reflected in the
episodic organization of natural discourse data.
The basic claim of this hypothesis is that the alternative use of a noun or
pronoun in discourse production is a function of attention allocation by the
speaker. During the on-line process of discourse production, the speaker uses
a pronoun to maintain reference as long as attention is sustained on that refe­
rent. Whenever attention focus is disrupted, the speaker reinstates reference
with a full noun, no matter how few clauses intervene between subsequent
references.
The linkage of the syntax of reference to processes of attention is, in
turn, linked to the episodic organization of discourse production. Episodes,
or following Lackstrom, Selinker and Trimble (1973) what might be called
conceptual paragraphs, represent the next highest structural unit of discourse
organization after the sentence. The episode is governed by a paragraph level
theme or macroproposition (van Dijk and Kintsch 1983, inter alia), which can
be sustained and elaborated upon as long as attention is directed toward it.
Episodes in prototypical narrative discourse probably average some five to
eight propositions in length, but such an average, at best representing only a
structural definition of episode, does not define the episode, it merely charac­
terizes it in its typical realization. Episodes are defined ultimately by the sus­
taining of attention on a particular paragraph level theme, a pragmatic instan­
tiation of a rhetorical act.
Two different sorts of data are used to explore the general hypotheses in
(1). There is experimental data, in which subjects created narratives in
response to the presentation of visual slide sequences. And, there is elicita-
tion data, in which subjects described on-line a short animated video-taped
cartoon. (Representative samples of each type of discourse can be found in
Appendices A and B.)
In each data set, it is shown that the use of noun and pronoun form is very
much a function of episodic boundaries, where episodic boundaries are
defined and identified explicitly and independently of linguistic information.
Further, because the episodic units in each discourse type are of significantly
different length, it is shown that an episodic/focus approach to the syntax of
reference provides a more complete accounting for the use of nouns and pro-
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 459

nouns in the collected data than does the recency/distance approach.

2. Theoretical Model of Discourse Production

Figure 1 presents a simplified model of the discourse production process.
The discussion here is intended to ground rather generally the specific data
analyses which follow. This outline in general follows the more complete
descriptions of van Dijk and Kintsch (1983); interested readers are referred
to that work for a more detailed discussion.

Figure 1. A simplified cognitive model for discourse production.

Speaker

Mental Text Functional Text

model represen­ syntax artifact
of → tation → →
X of
X

In this process, the speaker has in mind a particular mental representa­

tion of some event or other subject matter, which he intends the hearer to
create during the complementary process of discourse comprehension. To
assist the listener create this mental representation, the speaker engages in
discourse with the listener, creating in on-line processing a text representa­
tion which reflects the speaker's estimates of the listerner's familiarity with
the subject matter (information organization) and the speaker's attempts to
help the listener appreciate some particular point of view toward that subject
matter (thematic organization).
The text representation is operated on by the syntax of the particular lan­
guage used to produce the specific text artifact. Many local syntactic alterna­
tions, — constituent ordering, voice, the syntax of reference, — are con­
ditioned by local pragmatic factors (Givón 1983, Chafe 1980, and so on),
which are, in turn, a function of higher level pragmatic and rhetorical factors
(van Dijk 1981, 1982; van Dijk and Kintsch 1978, 1983; Rumelhart 1975,
Mandler and Johnson 1977, and so on).
Within this sort of general model, local syntactic alternations, like the
460 RUSSELL S. TOMLIN

pronoun-noun alternation considered here, may turn out to be linguistic

codes which facilitate the cognitive processes required of the listener to build
the intended mental representation. Some syntactic alternations facilitate
retrieval of information from memory; other syntactic alternations facilitate
selective focusing and emphasis on information during the model building
process.
Empirical investigation of these local syntactic alternations in discourse
production entails a number of basic requirements. First, empirical study of
discourse production requires establishing a database of reasonably compara­
ble text artifacts produced by an appropriate number of subjects. Second, it
requires that the basic analytical concepts utilized receive independent defin­
ition, both conceptual and operational. Third, it requires sensitivity to, and
some means for filtering out, irrelevant data, or 'noise'. And, fourth, it pre­
fers results that account for the productions of individuals within groups to
those that present only group characteristics.
These four requirements are met in the present study. Data were drawn
from two diverse sources: (1) on-line oral narrative production created by
three different groups of ten subjects describing a set of slide pictures, and (2)
on-line oral descriptive discourse production produced by ten subjects view­
ing an animated videotape. The basic analytical units, — proposition and
episode —, are defined below. One source of 'noise1, ambiguity resolution,
is systematically and explicitly filtered from the data. And, finally, the
hypothesis proposed here accounts well not only for the overall groups con­
sidered but for the performance of each specific individual.

3. Analytical Categories and Discourse Production Measurements

Analytical categories
Episode. An episode is defined conceptually as a semantic unit in dis­
course organization consisting of a set of related propositions governed by a
macroproposition or paragraph level theme. It represents sustained atten-
tional effort devoted to the macroproposition and endures until attention is
diverted; that is, it is sustained until an episode boundary is reached.
Episode boundaries represent major breaks, or attention shifts, in the
flow of information in discourse. In typical narrative discourse, major changes
in time, place, or characters correspond to episode boundaries (see van Dijk
and Kintsch 1983:204), though they do not themselves represent such bound­
aries. Similarly, in procedural or expository discourse major changes in
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 461

action or argument correspond to episode boundaries.

For experimental work on discourse production, such text-oriented
symptoms of attention shifting cannot be utilized in identifying episodes in
discourse data without risking circularity in functional argumentation (Tom-
lin 1983, 1984, 1985). For the present study, episode boundaries are iden­
tified following the general insights on event perception by Newtson and his
colleagues (Newtson 1973, Newtson and Engquist 1976, Newtson et al.
1977). Episode boundaries are identified here as major disruptions in the
flow of the non-linguistic visual material perceived by experimental subjects.
For the experimental data, episode boundaries in discourse production are
triggered by attention shifts caused by the perceptual disruption of the slide
projector shutter release cycle. For the elicitation data, episode boundaries in
discourse production are triggered by major thematic breaks, operationally
defined for this study as video cuts accompanied by major scenery changes
(Monaco 1981,Tomlin 1984).
Proposition. Discourse production is measured here in part in terms of
propositions. A proposition is defined here as a semantic unit composed of a
predicate plus its arguments for which a truth value can be determined. The
proposition, as the term is used here, also represents a basic unit of mem­
ory in human cognition (Anderson and Bower 1973, Anderson 1980) and in
discourse comprehension and production (Kintsch and van Dijk 1978,
Kintsch 1973, van Dijk and Kintsch 1983).
The identification of propositions in the collected discourse data is simi­
larly straightforward. An utterance in the text data counts as a proposition if
and only if it realized by a full clause or by a partial clause for which missing
arguments are readily recoverable. Embedded complement clauses are not
counted as separate propositions but as arguments of the matrix clause they
are embedded in. Infinitive and participial clauses in adjunct relations in sen­
tences are counted as realizing independent propositions. Nominalizations
and other complex phrasal constructions are not unpacked into sets of prop­
ositions.
Ambiguity resolutions. In discourse production data involving multiple
characters undistinguished by gender, numerous opportunities arise for pos­
sible ambiguous reference. In such cases, full nouns are used to provide suffi­
cient information to the listener to avoid actual ambiguity. Such cases of
ambiguity resolution, if simply ignored, would result in an unfairly conserva­
tive test of the specific hypothesis examined here, for this additional parame­
ter would represent a third and intervening variable for data analysis. There-
462 RUSSELL S. TOMLIN

fore, clear cases of ambiguity resolution are not counted in the analyses
below. Clear cases are explicitly defined as instances of switch subject refer­
ence only (Haiman and Munro 1983), even though other cases no doubt exist.
The example in (2) illustrates a filtered case:
(2) Even condition: 009
Episode Proposition Text
8 21 The ape looks down at the dog
→8 22 and the dog starts to run away/
9 23 And then the dog is lying dead
10 24 and the ape looks like he's
going to eat him
11 25 and along come this dinosaur/

Discourse production measurements

Absolute production. Raw measurements of discourse production were
collected simply by counting the number of propositions produced in each
protocol and the number of episodes described.
Discourse density. Refined density measurements are calculated by com­
puting the proportion of propositions to episodes.
These two sets of measurements permit one to determine the variability
in production performance of individual subjects. Subjects within each data
type produce comparable amounts of discourse, with little individual varia­
tion in these measurements. Subjects compared across data types produce
significantly different amounts of discourse.
Referential distance (Givón 1983). Referential distance is calculated by
counting the number of clauses which intervene between a given referent and
its last antecedent. Though clause and proposition are distinct linguistic
entities, there is no practical significance for this study in counting clauses
rather than propositions.
Referential distance measurements permit a good test of the recency/dis­
tance approach by allowing detailed comparative figures on the general use
of nouns and pronouns within each data type and, more crucially, between
the data types.
Episode boundary results: hits and misses. Data were collected on the
proportion of instances of referential syntax consistent with (hits) or inconsis­
tent with (misses) the general hypothesis. For a given referent, the syntactic
form represents a hit if and only if it is a full noun and the first mention of the
referent after an episode boundary, or if it is a pronoun and not the first men-
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 463

tion after an episode boundary. The syntactic form represents a miss if and
only if it is a full noun and not the first mention after an episode boundary
(excluding filtered out ambiguity resolutions), or if it is a pronoun and is the
first mention of the referent after an episode boundary.

4. The Experimental Data

The basic idea for this experiment was to stimulate artificially episode
boundaries for subjects engaged in narrative production for a slide picture
sequence. It was assumed that the shutter release cycle of the slide projector
represented a sufficiently strong perceptual disruption for the subject that the
subject would be forced to re-orient attention in order to continue with the
narrative production task. Nouns would be used to reinstate reference on first
mention after the forced boundary, independent of the amount of preceding
discourse and independent of where in the structural organization of the slide
sequence the episode boundary was placed. Pronouns would be used to main­
tain reference on subsequent mentions until the next episode boundary is
encountered. This is exactly what happened.

Method
A set of twenty-one slide pictures was presented to three different
groups of ten subjects. A representation of these slides can be found in
Figure 2. Each subject was asked to produce a story based on the slides pre­
sented as they appeared. Subjects self-paced through the task.
One group, the singles condition, saw each slide individually, one at a
time. The second group, the odd condition, saw slide one alone and then the
remaining slides in pairs. The third group, the even condition, saw slides one
and two together and then the remaining slides in pairs, except for slide 21
which was presented singly.

Results
Discourse density measurements. In absolute terms, subject in the singles
condition produced about twice as much discourse as did subjects in the other
two groups. Subjects in the singles condition averaged 54.36 propositions for
the task, or 2.59 propositions per slide. Subjects in the other two conditions
averaged only 29.92 propositions for the task, or 1.43 propositions per slide.
More interesting, however, is to see that each group produced the same
density of production when propositions are averaged per episode. The graph
in Table 1 displays the discourse density measurements for each of the three
464 RUSSELL S. TOMLIN

Odd condition Etc.

Even condition

Etc.

Figure 2. A representation of the stimulus slides (I= flying insect, B= bird,

C= cat, D = dog, G = gorilla, X= dinosaur/dragon). In singles condition,
each slide seen separately. In odd condition, slide pair denoted above by dot­
ted line. In even condition, slide pair denoted above by solid line.
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 465

groups. There is no statistically significant difference among or between the

three conditions, as measured by the Kruskal-Wallis test for non-parametric
analysis of variance (p<.05).
These results are important in that they show that subjects in each of the
three conditions are responding to the discourse production task in compara­
ble ways. They provide good grounding for the comparisons which are
described below.

Singles Odd Even Crab Video

Condition Condition Condition Data

Table 1. Mean discourse density measurements (propositions per episode)

for three experimental conditions and crab video data. Significant
difference among and between crab video data and the three experi­
mental groups. No significant difference among or between the
three experimental groups.

Episode boundary results. The crucial data are found in Table 2. If the
syntax of reference is a function of episode boundaries, then one would
expect the same performance behavior in each of the three conditions. One
should observe reinstatement of reference through nouns on first mention
after a boundary followed by pronominal reference until the next episode
boundary. All such instances represent hits for the hypothesis.
466 RUSSELL S. TOMLIN

One expects not to observe pronominals across episode boundaries or

nominals within an episode, except for ambiguity resolutions, which have
been filtered out.
Table 2 shows that no matter where one puts the boundaries, the propor­
tion of hits remains the same, about 84%. If the structure of the slide
sequence or simple distance controlled the syntax of reference, one would
expect differences across the conditions. Instead, one observes identical dis­
course production performances in precise harmony with the experimentally
placed episode boundaries.

Singles Odd Even Crab Video

Condition Condition Condition Data

Table 2. Proportion of predicted hits for three experimental groups and crab
video data. No significant difference among the four data sets or
between any pair.

5. The Video Data

The video data provides strong corroborative evidence for the experi­
mental results. Here 10 subjects produced on-line oral descriptions of a 108
second videotaped cartoon (Tomlin 1984, 1985).
The video cartoon was independently analyzed as composed of a set of
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 467

eight major episodes. Episode boundaries are in this case, like the experi­
mental conditions described above, taken to be perceptually salient break­
points or disruptions in the flow of visual material. These disruptions were
operationalized as video cuts (Monaco 1981) at major scenery changes. A
representation of the video organization is found in Figure 3.

SECOND0 14 3
1 A P P E A R A N C E OF 2
EPISODE FISH AND CRAB SNA1LSCENE (SNAIL = GOLDFISH)
AND FIRST ATTACK
EVENT NO. 1 2 3
4 5 6 7 8 9 lu­ 11 12
fish crab crab snail crab snail fish crab snail shed snail crab
appears appears alone crawl­ taken crawl­ peers looks con­ lost con­ climbs
and ing aback ing back tinues tinues down
attacks
fish
DURATION 4 7 3 3 2 5 1 2 5 1 2 4
SIG + + - + - - - - - + - -

SECOND 46 52 58 74 7
3 4 5 6
SECOND CRAB-FISH EYES T H I R D CRAB-FISH
EPISODE CHASE
INTERACTION SCENE INTERACTION, PIKE

EVENT NO. 13 14 15 16 17 18 19 20 21 22 23 24
fish crab crab fish fish eyes crab fish fish crab pike pike
re­ con­ attacks es­ meets come attacks es- es­ attacks ehases chases
lieved fronts fish capes eyes out fish eapes capes fish fish fish
fish pike

DURATION 2 5 51/2-6 1/2 2 4 5 1 2 1 2 5

S1G + + + - + - + - + + - +

SECOND 85 93 108
6 7 8
EPISODE SUNKEN SHIPS C E N E SPONGE SCENE BUBBLES SCENE

EVENTNO. 25 26 27 28 29 30 31 32 33 34 35 36
two both pike pike fish pike pike bubbles pike pike pike pike
enter gone gets es­ into into looks rise rises circles opens and
hatch stuck capes sponges sponges for crab
fish attack
fish |
DURATION 2 2 6 1 2 3 3 2 4 31/2 1/2 5
SIG + - + - + - - + - - - +

Figure 3. A representation of the episodic and event structure of the 108 second animated videotape.

If one examines Table 1 again, one can see that subjects in this elicitation
produced more than twice as many propositions per episode as did the experi­
mental groups.
468 RUSSELL S. TOMLIN

These data permit a better test of the recency/distance approach. If the

syntax of reference were primarily a function of linear distance, one would
expect a significantly different hit rate for this group than the other two.
Further, one would expect that the mean referential distance (Givón 1983)
for nominal and pronominal forms should be the same in the experimental
data and the video data.
But another look at Table 2 reveals that this is not the case. The propor­
tion of hits as predicted by the general episode boundary hypothesis is identi­
cal to that of the experimental groups. Further, the data in Table 3 show that
there is a statistically significant difference between the experimental condi­
tions and the video data in the referential distance measurements for nouns
and pronouns.

N P 0 N P 0 N P 0 N P 0
Singles Odd Even Crab Video
Condition Condition Condition Data

Table 3. Mean referential distance measurements (Givón 1983) for three

experimental groups and crab video data. Significant difference in N
and P measurements between crab video data and experimental
groups. Otherwise, no significant differences.

What this means is that subjects for this elicitation procedure perform
similarly to those in the experimental situation described above. They do so
despite producing significantly greater amounts of discourse per episode.
Further, they do so even though it results in statistically significant differ-
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 469

ences in the average distances occurring between referent and antecedent.

The video data and the experimental data seem to converge to show that
episode boundaries do control the syntax of reference, for one obtains consis­
tent and harmonious results in two very different discourse production tasks.

6. Exceptions and Counterexamples

While the general hypothesis in(1), as tested in the two distinct data sets,
accounts for some 84% of the discourse production of the 40 subjects, there
remain about 16% of cases that run counter to expectations, the misses.
These fall into two major classes: intra-episode nominals and inter-episode
pronominals. There are several non ad hoc explanations that account for
most of these exceptions, though a small class of genuine counterexamples
remains.

Intra-episode nominals
As illustrated in (3), intra-episode nominal exceptions occur whenever a
full NP is used to maintain reference after a noun has been used once within
a given episode:
(3) Video data: 011
Episode Proposition Text
3 23 That crab just tipped the top of
its shell,
3 24 Id o n ' tbelieve that,
3 25 do you?
→3 26 And that crab's going away,...
The use of the full NP, that crab, is exceptional in that it occurs within a given
episode, episode 3, but after this referent has been reinstated already by a full
nominal, that crab, in proposition 23.
Intra-episode exceptions are the more common kind observed in the col­
lected data. The represent some 90% of all the observed exceptions. But
there are at least three non ad hoc explanations that can reasonably account
for most of these exceptions.
First, there are numerous cases in which non-narrative evaluative dis­
course intervenes between propositions devoted to narration. The speaker in
such cases appears to switch his rhetorical activity from narrating events to
evaluating them. Such a switch seems to represent another case of attention
shifting of just the kind we have been considering here.
470 RUSSELL S. TOMLIN

Examples (4) and (5) illustrate this.

(4) Singles condition: 013
Episode Proposition Text
3 10 ...the bird standing on the ground
3 11 and he's looking at the bug
3 12 he might be (1 ) uh about to eat it
3 13 I don't know/
3 14 it seems kind of strange
→3 15 that the butterfly would
so close to this (.) to the
little bird
(5) Video data: 003
Episode Proposition Text
2 11 and the creature is watching
snail
2 12 Going along/
2 13 Leaving bubbles behind him/
→2 14 The snail is sort of
Disney-like creature/
2 15 Looks very cute/
In both cases, propositions appear in which the speaker stops narrating
events or actions occurring and begins to evaluate something. Though the
speaker technically does not leave the current episode, it seems reasonably
clear that his attention has shifted as the rhetorical activity shifted, requiring
reinstatement of reference by the full nominal form.
Second, some subjects overtly treated the dual slide presentation condi­
tion as though the slides were presented singly. This can be seen in (6):
(6) Even condition: 008
Episode Proposition Text
2 2 ...(.) I think the bird is
looking at the fly
2 3 = I think maybe he ate it
→2 4 the second slide the bird is
chasing the fly/
In such a case, the subject is clearly performing the task differently than other
subjects, keeping each slide individually and distinctly in mind as he pro­
ceeded through the task. Only one subject performed so differently, but he
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 471

produced a fair number of exceptions.

Third, there appear to be cases of ambiguity resolution which were not
filtered out. As described above, the only cases of ambiguity resolution which
were filtered out were cases of switch subject. Such cases were identified
operationally and do not require semantic interpretation. Thus, filtering
them out presents no problem for the present analysis. Cases like the one in
(7) below, probably also represent ambiguity resolution, but since the
ambiguity cannot be explicitly identified, such cases were counted as excep­
tions instead.
(7) Odd condition: 005
Episode Proposition Text
7 17 And then the dog chases the cat
→ 7 18 (.) and it looks like he
catches the cat/
These three explanations account for most of the observed exceptions. The
remaining cases, totaling only some 5% of the total numbers of cases, or
about 1.5 per subject, must be considered at this point to be genuine coun­
terexamples.

Inter-event pronominals
Inter-episode pronominal exceptions occur when the speaker uses a pro­
nominal form to make reference despite the presence of an episode bound­
ary. This is illustrated in example (8). There are altogether very few excep­
tions of this sort. One non ad hoc explanation accounts for many of them.
After completing the task, one subject reported that she regularly read
and told stories to children and said she performed the task as though she
were reading a storybook to a child. Most of the inter-episode pronominal
exceptions were produced by this one subject. As illustrated in (8), this sub­
ject used the relative pronoun, who, across the episode boundary between
episode 1 and 2. The general hypothesis predicts that reference would be
reinstated across that boundary with the full NP, the butterfly.
(8) Odd condition: 004
Episode Proposition Text
1 1 Once upon a time there was a
butterfly
→2 2 who saw a little bird
2 3 and greeted it/
472 RUSSELL S. TOMLIN

The remaining cases were all counted as counterexamples. These represent

about 1% of all the cases observed, or .3 exception per subject.

7. Discussion
Summary conclusions. The analysis of two independent classes of data,
the experimental slide data and the video data, seem to converge on the same
demonstration. The syntax of reference is clearly a function of episodic boun­
daries in narrative production, which are related to shifts in attention during
the on-line process of discourse production.
In the experimental data, subjects, both as groups and as individuals,
consistently used nouns and pronouns as predicted by the episodic/paragraph
approach, when episode boundaries are experimentally controlled and non-
linguistically determined. In the video data, similar results are obtained, but
in this case episode boundaries are more traditionally identified through
change in place. During discourse production, subject do use nouns to
reinstate reference after an episode boundary, and they use pronouns to
maintain reference within an episode. Episode boundaries, in turn, seem to
be a function of attention orientation, weakly demonstrated in the video data
by video cuts and scene changes, and strongly demonstrated by independent
control in the experimental data.
Episodic organization and attention. While the results here argue
strongly for the episode/paragraph approach to the syntax of reference over
a recency/distance approach, the argument for an episode-attention connec­
tion has been less clearly made. The crucial argument required is to show that
attention allocation, a more general cognitive process, stimulates the episode
bounaries, and that they are not due simply to something special about the
organization of the elicitation stimuli. A review of the experimental data per­
mits one such argument to be made.
To begin, the most plausible alternative to an attention-driven episode
boundary hypothesis is to argue that episodic organization merely reflects a
particular instantiation of a given story schema. In this view, the episodic
organization represents the consequences of higher level decisions about the
structuring of the story.
The slide sequence used to collect the experimental data does have a
very clear schematic structure. It is composed of five iterations of a basic pre­
dation sequence: (1) introduce new character, (2) two characters meet, (3)
larger character chases smaller character, and (4) larger character eats smal­
ler character. Its transparent constituent structure is presented in Figure (4) :
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 473

Figure 4. The schematic structure of the slide sequence.

Predation sequence

Bug-Bird Bird-Cat Cat-Dog Dog-Gorilla Gorilla-Dinosaur

The natural episodic units in this sequence should be some consistent set
of nodes, clearly either the terminal nodes individually or the next highest
level in the tree. In addition, the two bracketed pairs, A and B, clearly do not
represent episodic units in this story schema.
This schematic representation of the slide sequence appears to have
been recognized by nearly every subject in the experiment, irrespective of
which group they were assigned to. Subjects appear to understand the basic
story organization by no later than the beginning of the third iteration (Cat-
Dog). This is reflected empirically in two ways: (1) overt meta-comment
during the experimental task, the subject observing the iterative nature of
sequence, and (2) solicited acknowledgement during debriefing of having
recognized the iterative organization of the slide sequence.
If episodic organization were simply schema-based, then one would
expect that at least some subjects would reveal differences in their discourse
production strategies after learning the slide sequence structure. Having
learned the basic predation sequence, the subject only needs to be introduced
to the next new character to be able to predict precisely what will happen.
Given such a state of affairs, one might expect to see differences in the
distribution of hits and misses between the first half (schema unknown) and
the second half (schema learned) of subjects' production protocols for the
two paired slide conditions, odd and even. Episodic organization might be
attention-driven as long as no concrete schema were available, but as soon as
one were might shift to an available schema. However, comparison of the
474 RUSSELL S. TOMLIN

first and second halves of the odd and even protocols reveals no statistically
significant difference in the proportions of hits and misses. That is, subjects
perform identically across the two halves, even though during the second half
subjects appear to have access to some general schematic representation of
the story structure.
Subject performances are clearly dictated by the placement of the shut­
ter release cycle, after either even or odd slides. Neither the odd condition,
represented by A in figure 4, nor the even condition, represented by B in fig­
ure 4, break the slide sequence into any sort of natural or expected schema-
grounded episodes. On the contrary, the two conditions were designed spec­
ifically to run counter to the expected episode analysis of the slide sequence.
If the shutter release cycle does represent, as assumed here, an adequate
disruption of attention focus for the experimental task, then it becomes
increasingly likely that it must be attention driving episodic organization and
not the other way around. Research on discourse comprehension does not
significantly address this issue, for in such studies attention focus and episode
boundaries are generally confounded. Thus, while the argument for atten­
tion-driven episodic units may remain at the present moment incomplete, it
should prove easier to sustain than a schema-based model of episodic organi­
zation, for such a model will be hard pressed to account for on-line discourse
production data of the type collected in this study.
Recency versus episodes and the syntax of reference. The present study
argues strongly for an attention-driven episodic/paragraph model of the func­
tion syntax of reference, and it argues strongly against a recency/distance
approach.
The attraction of the recency/distance approach lies in its apparent
simplicity and generality. Increasing the time or distance between subsequent
references does increase the likelihood that full nominal reference will occur
for the second reference. Further, measurement of average distances holding
between a given referent and between a subsequent noun and a given referent
and a subsequent pronoun reveal that the observed average is always higher
for the full noun. While these observations represent important general
observations about the use of nouns and pronouns, they do not represent
adequate evidence for arguments about the specific circumstances that trig­
ger the use of a noun or a pronoun at any specific moment in the discourse
production of individual speakers.
In the end, the pertinent question to ask is this: exactly what are the con-
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 475

ditions that trigger the use of a noun or pronoun in making references during
the time course of discourse production?
The fundamental limitation of the recency/distance approach is that it
admits a great many potential counterexamples but offers no systematic
explanation for them. The question must be raised just how much time or dis­
tance is required to trigger full noun reference during discourse production
and how little is needed to guarantee that a pronoun will occur.
While the recency/distance approach cannot answer this question, the
episode/paragraph model does exactly this. It makes specific predictions
about the performance of individuals as well as groups in the discourse pro­
duction task. Individuals will use full nouns on first mention after an episode
boundary; individuals will use pronouns to sustain reference during an
episode.
The primary drawback to the episode/paragraph approach lies in the dif­
ficulty of providing explicit and structure-independent means of identifying
episodes and episode boundaries. In this study episodes are argued to be a
function of attention allocation, and episode boundaries are identified and
manipulated independently of text structure and without dependence on
introspection. It may even suggest that in the end discourse units, like the
paragraph, are more likely to be the artifacts of linguistic analysis than they
are cognitive units utilized by speakers in discourse production.

REFERENCES

Anderson, J. 1980. Cognitive Psychology and Its Implications. San Francisco:

W.H. Freeman.
Anderson, J. and G. Bower. 1973. Human Associative Memory. New York:
Wiley.
Chafe, W. 1974. "Language and consciousness". Language 50: 11-133.
. 1976. "Givenness, contrastiveness, definiteness, subjects, topics, and
points of view." In C.N. Li, ed. Subject and Topic. 25-56. New York:
Academic Press.
—--. [this volume]. "Cognitive constraints on information flow".
Clancy, P. 1980. "Referential choice in English and Japanese narrative dis­
course." in W. Chafe, ed. The Pear Stories: Cognitive, Cultural, and Lin­
guistic Aspects of Narrative Production. 127-201. Norwood, N.J.: Ablex.
476 RUSSELL S. TOMLIN

Clark, H. and C. Sengul. 1979. "In search of referents for nouns and pro­
nouns". Memory and Cognition 7: 35-41.
van Dijk, T. 1981. Macrostructures. Hillsdale, N.J.: Lawrence Erlbaum.
. 1982. "Episodes as units of discourse analysis." In D. Tannen, ed.
Analyzing discourse: Text and Talk. Washington D.C.: Georgetown Uni­
versity Press.
van Dijk, T. and W. Kintsch. 1983. Strategies of Discourse Comprehension.
New York: Academic Press.
Fox, B. 1984. Discourse Structure and Anaphora in Written and Conversa­
tional English. Unpublished PhD. dissertation. Department of Linguis­
tics, UCLA.
Givón, T. (ed.) 1983. Topic Continuity in Discourse: A Quantitative Cross-
Language Study. Amsterdam: John Benjamins.
Givón, T., W. Kellogg., M. Posner. and P. Yee. 1984. "Time dynamics of ref­
erential processes in discourse." Paper presented at the 6th Annual Meet­
ing of the Cognitive Science Society. Boulder, Colorado. June 1984.
Grosz, B. 1977. "The representation and use of focus in dialogue understand­
ing." SRI Technical Notes 151.
Haiman, J. and P. Munro. (eds.) 1983. Switch Reference and Universal
Grammar. Amsterdam: John Benjamins.
Hinds, John. 1977. "Paragraph structure and pronominalization." Papers in
Linguistics 10: 77-99.
Hinds, J. 1979. "Organizational patterns in discourse." In T. Givón ed. Dis­
course and Syntax. Syntax and Semantics 12. 135-158. New York:
Academic Press.
Kintsch, W. 1974. The Representation of Meaning in Memory. Hillsdale,
N.J.: Lawrence Erlbaum.
Kintsch, W. and T. van Dijk. 1978. "Toward a model of text comprehension
and production." Psychological Review 85: 363-394.
Lackstrom, J., L. Selinker. and L. Trimble. 1973. "Technical rhetorical prin­
ciples and grammatical choice." TESOL Quarterly 7: 127-136.
Mandler, J. and N. Johnson. 1977. "Remembrance of things parsed: Story
structure and recall." Cognitive Psychology 9: 111-191.
Marslen-Wilson, W., E. Levy. and L. Tyler. 1982. "Producing interpretable
discourse: The establishment and maintenance of reference." In R.J. Jar-
vella & W. Klein, eds. Speech, Place, and Action. Chichester: Wiley.
Monaco, J. 1981. How to Read a Film. Oxford: Oxford University Press.
Newtson, D. 1973. "Attribution and the unit of perception of ongoing
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 477

behavior. " Journal of Personality and Social Psychology 35:28-38.

Newtson, D. and G. Engquist. 1976. "The perceptual organization of ongo­
ing behavior." Journal of Experimental Psychology 12: 436-450.
Newtson, D., G. Engquist. and J. Bois. 1977. "The objective basis of
behavior units." Journal of Personality and Social Psychology 35: 847-862.
Reichman, R. 1981. Plain Speaking: A Theory and Grammar of Spontaneous
Discourse. Technical Report. Bolt, Beranek and Newman, Inc.
Rumelhart, D. 1975. "Notes on a schema for stories." In D. Bobrow & A.
Collins, eds. Representation and Understanding. 211-236. New York:
Academic Press.
Tomlin, R. 1983. "On the interaction of syntactic subject, thematic informa­
tion, and agent in English." Journal of Pragmatics 7: 411-432.
. 1984. "The identification of foreground-background information in on­
line descriptive discourse." (ms.) Department of Linguistics, University of
Oregon.
. 1985. "Foreground-background information and the syntax of subordi­
nation." Text 5.
Tyler, L. and W. Marslen-Wilson. 1982. "The resolution of discourse
anaphors: some outline studies." Text 2: 263-291.

APPENDIX A

A representative protocol elicited in the experiment (odd condition)

Odd Condition: 003
Episode Proposition Text
1 1 There's a butterfly
1 2 flying/
2 3 The bird's talking to the butterfly/
3 4 The bird's chasing the butterfly
3 5 and he's going to eat it/
4 6 (2) The cat sees the bird/
5 7 The cat chases the bird
5 8 and kills it/
6 9 Um the dog is walking along
6 10 and sees the cat/
7 11 The dog chases the cat
7 12 and eats it/
478 RUSSELL S. TOMLIN

8 13 What is that
8 14 (. ) the gorilla sees the dog
9 15 The gorilla grabs the dog
9 16 and kills it/
10 17 The dragon sees the gorilla
10 18 and talks to it or something/
11 19 The gorilla tries to get away
11 20 and he eats it/

APPENDIX 

A representative on-line protocol elicited with the cartoon videotape

On-Line Description: 010
Episode Event PropText
1 1 1 Okay, there's a fish in the ocean/
1 1 2 It's a cartoon/
1 2 3 And up comes a crab
1 2 4 and tries to get it with its pinchers
1 2 5 and it seems to be avoiding it
1 3 6 and now it disappears/
2 4 7 Okay, the crab's looking around
2 4 8 and he sees a snail or something
2 4 9 walking on the ocean floor/
2 5 10 He spots it/
2 5 11 He blinks his eyes/
2 5 12 He looks at it
2 6 13 crawling along kind of in a weird fashion, okay/
2 8 14 Okay, he's been spotted now, the fish that's
being chased/
2 9 15 Okay, he keeps walking
2 10 16 Okay, he goes out of the shell
2 10 17 so he's uh lost his uh shell
2 12 18 so the here comes the crab/
2 12 19 Crawls down from the rock/
3 14 20 Okay, he lifts his hat up or his shell up
to the fish/
3 15 21 Okay, tries to get him with his pinchers/
 LINGUISTIC REFLECTIONS OF COGNITIVE EVENTS 479

3 16 22 Okay, unsuccessful/
4 17 23 All right, there are a new set of eyes, a
new kind of fish
4 18 24 that seems to be all eyes/
5 19 25 Okay, the crab's still at it,
5 19 26 trying to catch the fish
5 21 27 and, and then he almost goes in, the fish
almost goes in the mouth of a very big whale
or shark or something, some some kind of
fish/
6 24 28 Now they're swimming around uh(.) an old
wreck of a ship, okay/
6 24 29 Obviously been there some time/
6 27 30 Urn the big fish stretches a little/
6 27 31 That was kind of cute/
6 28 32 Um, they're still in a chase,
7 29 33 the little fish has gone into a bunch of
sponges it looks like,
7 30 34 and the big fish has gotten lost
7 30 35 and crashed into it and/
8 32 36 Okay, now he's hidden in a bubble/
8 33 37 Okay, so he's, the big fish is going up
8 33 38 to try to (.) get in the bubble
8 33 39 and catch him
8 36 40 and out come the crab out of the fish's
mouth/
8 36 41 Okay, they popped the bubble/
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE
IN NARRATIVE AND BEYOND

BENJI WALD
National Center for Bilingual Research

1. Introduction
My purpose in this paper is to discuss the relation of narrative to other
types of discourse as a nexus for the study of the relation of syntax to dis­
course. Particularly in the light of those views of discourse which have arisen
as an outgrowth of syntactic research, a guiding question underlying the fol­
lowing discussion is: what, if anything, is special about narrative that has war­
ranted the attention it has received in syntactic studies?
As knowledge of the syntactic characteristics of discourse genres other
than narrative progresses, there is a corresponding recognition that narrative
has syntactic properties of problematic generalizability to discourse in gen­
eral. It will be important for the rest of this paper not to overstate the case.
However, for introductory purposes, consider the following observations.
Linde (1974) noted that spoken spatial tours told by New York City English
speakers are rich in inversion, exemplified by the rotation of subject and loca­
tive phrase around the verb of a clause. In contrast, narrative in current spo­
ken English is poor in the use of inversion (of any type). l This makes spatial
tours a more appropriate site than narrative for the study of inversion. Mov­
ing from clause-internal ordering or constituents to the ordering of clauses
themselves, a similar issue follows from Thompson's (1984) study of the posi­
tion of reduced purpose clauses relative to the associated main clauses.
Thompson found that preposing of the purpose clause appears to be rela­
tively rare in written narrative, but common in the procedural discourse of an
auto tune-up manual. As in Linde's case, a similar ordering of constituents
does not appear to be equally characteristic of all discourse genres. One con­
sequence of observations like these is to call into question the use of narrative
alone to make generalizations about the use of syntax in discourse.
482 BENJI WALD

This in turn raises the issue of the interpretation of "markedness" in

terms of statistical frequency in narrative. To the extent that a theory of uni-
versais in terms of markedness is based on statistical counts from narrative
data, what justifies the emphasis on narrative, if not an assumption that nar­
rative data is representative of the grammar in general? This issue is implicit
in Mithun's (this volume) discussion of the problem of defining "basic word
order" in the unrelated languages Cayuga, Ngandi and Coos. Mithun notes that
in these languages, subject and object relations do not seem to play a role in
ordering with respect to the verb. Instead, she proposes the concept of news-
worthiness, through which either relation can be preposed depending on how
the discourse context affects topicality. In view of the inequitable distribution
of some syntactic devices across discourse genres, it is likely that the more
newsworthy relations to the verb change with discourse genre. Should narra­
tive have a privileged position in determining the markedness of word order
in languages like these?
Underlying these issues of the relation of narrative syntax to that of other
discourse genres is the basic linguistic issue of the "productivity" or "creativ­
ity" of grammar, the principle that grammar is a set of systems for organizing
information for a potentially infinite number of purposes. 2 If grammar is used
differently in different discourse genres in a single language, generalizations
about the grammar of languages cannot be made on the basis of narrative (or
any other discourse genre) alone without further justification.
To this end the following discussion will be aimed at putting the defining
characteristics of narrative syntax into a larger discourse perspective, using
syntactic and discourse data from several languages which appear to gram­
maticality encode temporal sequence in ways relatively rarely found among
languages.
Discussion wall proceed from basic concepts such as "narrative", "tem­
poral sequence", "subordination", "grounding" (section 2), to a general dis­
cussion of the relationships among expression of these concepts in specific
languages such as English, Hausa, Bemba and Swahili (section 3), then to a
more extended and detailed discussion of the verb marking of clause relations
in Swahili (section 4), and, finally, to conclusions to be drawn from the pre­
ceding discussion and directions for further research on the relation of narra­
tive syntax to the syntax of discourse in general.
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 483

2. Basic Concepts
2.1. This section is divided into two parts. The first establishes a particular
restricted meaning for the term narrative (necessary since different writers
vary between wider and narrower applications of the term "narrative" to dis­
course genres which are conceptually distinct), and focuses attention on the
concept of temporal sequence as one of several properties of narrative (and
many other types of discourse unit). The second part discusses subordination
and grounding.

2.2. Narrative and temporal sequence

A minimal syntactic characterization of narrative is given by Labov &
Waletzky (1967) in their seminal work on the analysis of spoken narrative. By
definition they require a narrative to consist of at least two main clauses in a
linear order which reflects the inferred temporal order of events. The
requirement that the narrative clauses be "main" is of special significance to
the development of a syntactic theory of discourse (coherent multi-sentence)
units, because it establishes a relationship between clauses which goes
beyond sentence grammar.
The crucial criterion for distinguishing narrative clauses from other
multi-clause sequences is the effect of clause reordering on the inferred tem­
poral order of events. For example, consider the following episode from a
written text, based on a filmed cartoon serving as a controlled elicitation
stimulus in a study of clausal processing of texts across differences in memory
condition and oral/written channel by Tomlin (1984:34).
(1) 19 The fish swam around fast
20 and almost ran in to (sic) something with very big eyes.
21 It got scared
22 and turned the other way.
All of the clauses in (1) are narrative clauses by virtue of mutual irreversibility
without changing inferred temporal order. It must be noted that the test of
reversibility is concerned with temporal reference only, and ignores ungram­
maticality caused by changes in nominal reference, i.e., it allows test interpre­
tation to adjust nominal reference so that the test of reversal of, for example,
(1)21-22 is:
(1) 22 It turned the other way
21 and O got scared,
484 BENJI WALD

rather than the mechanically anti-grammatical

(1) 22 And O turned the other way.
21 It got scared
Note that the lexical substitution of be for get in (2) below changes (1) 21
from a punctual to a stative clause. This change destroys narrative sequence,
because clause order does not affect inferred temporal order, i.e., "it was
already scared BEFORE it turned away".
(2) 21 It was scared
22 It turned away
or
22 It turned away
21 It was scared
In justifying the distinction between a sequence of narrative clauses and
a subordinate + associated main clause, L&W use the syntactic test of reor­
dering to show that subordinate clauses share a property with non-narrative
main clauses of not reversing the inferred temporal order, e.g.
(3) 21 AFTER it got scared
22 it turned the other way
or
22 it turned the other way
21 AFTER it got scared
Labov (1972:361) insists that narrative clauses must refer to single past
events. He explicitly excludes clause pairs oriented toward another modality,
e.g. past habituais or general "presents", arguing on logical grounds that they
are reversible, e.g.
(4) a. ya get up there
b. n you feel like (you're gonna die)
The argument is that, since (4)a and b are events which are inferred to recur,
either order is "true". However, this is a different argument from the one
which tests the effect of linear reordering on temporal interpretation. One
can see that on any single occasion there is an internal temporal order of
events, i.e. GET - FEEL. Most importantly, if the clauses of (4) are reor­
dered, the inferred internal temporal order is reversed, just as across narra­
tive clauses. Silva (1983:20) provides a dramatic example of the importance
of temporal order in procedural discourse. In a study of the development of
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 485

the English conjunctions among middle-class Bay Area children, she notes a
gag which was used in one episode of the TV program MASH, which plays on
deviations from temporal ordering in procedural discourse. While following
the written instructions for dismantling a live bomb, the characters come
across:
(5) a. Carefully cut the wires leading to the clockwork...
b. BUT FIRST remove the fuse... (Silva: 1983,20)
Though (5) demonstrates that English has the semantic resources to present
two main clauses in reverse order, the gag shows that using these devices is not
the best strategy for certain types of discourse.
While the L&W restriction of narrative clause to single past events does
serve to define narrative as referring to single past experience, in contradis­
tinction to other kinds of discourse, it does not recognize that narrative shares
an internal temporal order with many other types of discourse. That is,
iconicity of temporal order is a property of narrative independent of single­
ness and pastness of events.
For purposes of further discussion I will reverse the term narrative
sequence to refer to a set of two or more narrative clauses, in accordance with
L&W's definition. However, I will use the term temporal sequence to refer to
the more general class of clause relations by which two events are referred to
in incomc order, regardless of external time or modal orientation. Temporal
order, then, is the inferred referential order of events. Temporal sequence is
the iconic match of temporal order and clause order.
At this point the special syntax of narrative has been resolved into four
more general components.
1. clause status (main or subordinate)
2. temporal sequence
3. singularity of events in temporal sequence
4. pastness of events in temporal sequence
With the glaring exception of the clause status, these notions are clear and
independent of grammatical encoding in particular languages. 3

2.3. Subordination and grounding

Some recent literature in syntax has questioned the basic notion of sub­
ordination as a uniform characteristic of ail languages, in view of its differen­
tial syntactic and semantic properties in different languages (e.g. van Valin,
1984; Haiman & Thompson, 1984; Mithun, 1984a). In syntactic terms, the
486 BENJI WALD

coherence of subordination as a language-independent concept has been

questioned on the basis of the language-dependent nature of distinctions
between the marking and/or internal word order of main and subordinate
clauses.
Among discourse-oriented studies, the function of "subordination" is
linked to "background" in discourse. In keeping with the focus on narrative,
the distinction between "foreground" in discourse. In keeping with the focus
on narrative, the distinction between "foreground" and "background" has
been discussed virtually exclusively in the context of narrative. The semantic
distinction between "foreground" and "background" has proven difficult to
explicate in precise terms. Generally, terms like "main", "highlighted", "im­
portant", "significant", "central" are used to characterize foregrounding
(e.g. Labov, 1972; Hopper, 1979; Jones & Jones, 1979; Hopper and
Thompson, 1980; Ohmanson, 1982; Tomlin, 1984). A basic question addres­
sed to the fundamental notion underlying these terms must be: is the notion
language-independent? If so, why is it difficult to decide between translations
using English when or then clauses, or main and relative clauses in some lan­
guages (cf. Mithun, 1984a)?4
This difficulty is avoided by Jones & Jones (1967) in their proposal for
variable degrees of grounding across languages. However, their solution is
not without a cost to clarity in the description of grammatical paradigms.
According to Jones & Jones' notion of grounding, languages differ in the rela­
tive levels of "importance" which they grammatically encode on clauses in
discourse, e.g., English has at least three in narrative: (lowest) past subordi­
nate clause — past main clause — historical present (highest). In some lan­
guages, e.g. Apache, these levels belong to a single paradigm. However, J&J
recognize a problem in identifying the highest level (called "peak") in terms
of a specific grammatical paradigm among many languages (op cit:23ff).
Thus, in English, subordination and tense-marking belong to separate gram­
matical paradigms. Why are areas within a single semantic continuum
encoded in such grammatically different ways, and are there any limits across
languages on how such semantic continua are grammatically encoded?
Regardless of terminology or conceptual framework, scholars agree on
the status of the syntactic representation of single past events in narrative
sequence as constituting a special conceptual ground in narrative. Still, recog­
nition of narrative sequence depends crucially on distinguishing main from
subordinate clauses. Furthermore, this characterization of "foregrounding"
only applies to narrative, whereas both the main:subordinate distinction and
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 487

temporal sequence are conceptually independent of time orientation (i.e.

past/non-past, etc.), as discussed in 2.1. above.
In considering the maimsubordinate distinction, and its relation to
grounding, the effect of clause reordering on temporal sequence will be a use­
ful analytic tool. Temporal sequences which may be linearly reordered with­
out changing the inferred temporal order can be distinguished from temporal
sequences for which reordering does affect interpretation of temporal order.
With regard to the status of narrative syntax within the grammar of any par­
ticular language, distinct types of languages can be recognized. One type of
distinction is between languages which regularly encode temporal sequence
and languages which do not. A second type of distinction, which turns out to
crosscut the first, is between languages which regularly encode time orienta­
tion to the moment of speaking (henceforth speech time) at the clause level,
and languages which do not. The following sections will briefly discuss the
grammatical independence of temporal sequence and orientation to speech
time, and then examine in more detail the syntactic construction of narrative
and other types of discourse in Swahili, a language which grammatically
marks temporal sequence in some discourse genres but not in others.

3. The explicit marking of temporal sequence

It seems to be the case that all languages have devices to make explicit
the relative timing of events represented by adjacent clauses, but how regu­
larly these devices are actually used varies across languages. Perhaps tem­
poral sequence is most commonly represented iconically without further
marking, and it is left to pragmatic inference to determine whether or not the
linear ordering of clauses indeed represents a temporal sequence. Example
(2) above indicates that in the English get/be distinction lexical encoding can
play a role in establishing temporal sequence; copula clauses represent states
rather than events and may break a temporal sequence. Clauses (6)b-c below
give further evidence of the pragmatic basis of temporal sequence in English.
(6) a. you gotta get it (= the mop:BW)
b. n squeeze it yk (= y'know)
c. take all the water out
d. then mop (MT16m,ELA)
Clauses (6)a and d are strictly temporally ordered with respect to each other,
and with respect to (6)b and c, but the temporal ordering between (6)b and c
is problematic. One view might be that squeeze initiates a single event which
488 BENJI WALD

ends in take (ALL,..). Another view might say that both clauses are unor­
dered but in a manner-purpose relation to each other. Still a third view might
take (6)b-c as a temporal sequence, analyzing (6)b as an inceptive event and
(6)c as a completive event. The problem is that the strategy of temporal
sequence and the lexical features of the verbs of (6)b-c alone are not sufficient
to decide the analysis. As a general rule, where pragmatics operates without
grammatical support, indeterminate cases will occur.
Languages which more regularly mark temporal sequence seem to be
rare. In addition, they vary in type amongst themselves. An example is
Hausa, like most Afro-Asiatic languages, an aspect-only language (i.e., does
not grammatically encode past/nonpast). Hausa obligatorily marks aspect
(through a morpheme suffixed to the subject marker, preceding the verb).
The two major aspects are usually called "completive" and "continuous"
(e.g. Kraft & Kirk-Greene, 1973). The continuous, like the Slavic imperfec-
tive, is used in progressive and habitual contexts (cf. Comrie:1976, 25ff).
There is a further special habitual "aspect" which is never used as a progres­
sive. In a habitual context (e.g. procedural or past routine), clauses in tem­
poral sequence are marked with the subjunctive, e.g.
(7) a. ya KÀN tàfi barga:
he HAB go stable
b. yà zàunà
he+SJN sit
c. mà:su sàn kwabàn-sà su zo:
those want money-his they+SJN come
d. sú ke:wàye:shi... (Abraham 1959:69-70)
they + SJN surround him...
a. "he used to go to the stable
b. and stay
c. and then those who wanted his money would come
d. and surround him..."
In the context of the story, the translation calls for a past habitual, but since
Hausa does not mark tense, in other contexts the time orientation could be
"general present" (i.e., he usually goes...stays...).
Clauses representing a temporal sequence of single events use a special
form of the completive, (obligatorily used in relative clauses), e.g.
(8) a. ka:tsi kùwa: shi: nè: tàdo: na tukunya-r ba:ba wàn-dà
katsi tp it is mud from pot-of indigo which
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 489

b. a-KÀ di:bà
one-CR gather
c. a-KÀ dunkùle: cu:ri eu:rí
one-CR roll.into balls one
d. a-KA ko:nè
one-CR burn
a. "katsi is mud from an indigo pot which
b. someone gathers
c. and rolls into bails
d. and bakes." (Abraham 1959:29-30)
In another context the temporal sequence in (8)b-d might refer to a narrative
sequence (somebody GATHERED...ROLLED...BAKED /the mud/). 5
Although Kraft & Kirk-Green (1973:109) note that the simple comple­
tive can also be used in past temporal sequences (except in relative clauses,
where the relative completive must be used in any case), the relative comple­
tive seems to be more common. In contrast, the simple completive is more
regularly used as an anterior marker in past contexts, as in (9)c.
(9) a. dà su-KÀ shiryà:
RM they-CR ready
b. su-KÀ tàfi
they-CR go
c. màka:ni:kî: ya-:gya:rà: ma-sù mo:tà-r-sù.
mechanic he-CM fix for-them car-of-them
a. "when they were (i.e. got) ready
b. they went
c. the mechanic HAD (already) fixed their car." (K&K-G
1973:171)
Thus, in view of its use in both temporal and reverse sequences, the
generalization can be offered that the simple completive does not inherently
encode anteriority, but due to the common use of the relative to mark tem­
poral sequence, the simple completive gravitates toward use in the context of
anteriority.
In contrast to Hausa, Bemba marks both absolute tense and temporal
sequence. Givón (1972:137ff) notes that verbal prefixes mark temporal
orientation to speech time, e.g. À 'relatively distant past', but that there is
suffixal contrast between temporal sequence and anteriority represented by
-i/ele:Ø, respectively. This language, then, has both overt marking of tern-
490 BENJI WALD

poral sequence and tense marking in certain discourse contexts. Bemba will
be discussed further below in relation to Swahili.
In languages like Hausa and Bemba, which regularly grammatically
encode temporal sequence, clause reordering necessarily changes inferred
temporal order. Hausa and Bemba contrast with English in the regular mark­
ing of temporal sequence. However, they differ from each other in whether
or not they mark tense as well. Hausa has no specific tense marking, while
Bemba has an elaborate system of tense distinctions in the past, typical of
Bantu languages (cf. Givón, 1972). In marking tense relative to speech time
Bemba is similar to English, but distinct from Hausa.
In the next section I will consider the verb marking system of Swahili,
particularly as it is used in the construction of narrative and other types of dis­
course containing temporal sequences. It will be seen that in some ways the
Swahili system is intermediate between its distant Bantu relative Bemba, and
an aspect-only language like Hausa. In still other ways, it is unlike either of
these languages.

4. Verb marking in Swahili discourse

The three most prominent verb markers in Swahili narrative are KA, LI
and KI. Some of the basic characteristics of these markers are discussed by
Hopper (1979) and Hopper & Thompson (1980:281ff) in the context of narra­
tive. Essentially H&T characterize KA as marking clauses of the "main line"
of the narrative (thus, a temporal sequence), KI as marking deviations from
that line, and LI as the usual "narrative past tense". It is, in fact, the only past
tense in Swahili. In the sense of encoding reference to speech time, it is the
only Swahili "tense". KI and KA, on the other hand, are neutral to the exter­
nal time orientation of discourse.
The following discussion will examine KA, LI and KI in that order, in
spoken Swahili discourse. It will examine the interplay between Swahili
grammar, lexicon and pragmatics in the construction of narrative and other
types of discourse.

4.1. KA
The Swahili verbal prefix KA exemplifies the regular grammatical
encoding of temporal sequence in certain discourse types, including narrative
and general procedural discourse. KA explicitly marks a temporal sequence
without need to refer to wider context for recognition. The indifference of
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 491

this marker to speech time (i.e. between past/nonpast) is indicative of the

more general characteristics of the Swahili system of marking tense, aspect
and mood, with qualifications to be discussed subsequently with respect to
the markers LI and KI.
Consider now the following discourse segment featuring KA. (In (10)
and further examples, upper case to the left of each cited clause isolates the
verb marker prefixed to the italicized verb).
(10) a. NA....halafu tunachota kwenye maji
b. KA tukatia kwenye ndoo
c. KI 'kikoma
d. KA tukabeba
e. KA tukaja zetu mpaka nyumbani
f. KA tukamimina kwenye mtungi...
a. "...and then we draw up water
b. and put it in a pail
c. when we finish (i.e. and then)
d. we take it
e. and come home
f. and pour it into jugs. (ML38f, Lamu)
ML is explaining how water is brought to houses in her hometown. The dis­
course does not refer to a single past event, and therefore is not a narrative.
It is a general routine, or, from the speaker's perspective, a past routine, since
she now lives in a house in a town with plumbing. However, the external time
orientation is not evident from the context given in (10) alone. As Hopper
(1979,214) notes, the Swahili system does not express orientation to the time
of speaking in its marking of temporal sequence. Orientation is usually pro­
vided by a marker in a clause preceding the temporal sequence. Independent
of external time orientation, the temporal sequence is marked from (10)b to
(10)f. All verbs marked by KA are in temporal sequence. No pair of KA
clauses can be reordered without reversing the temporal sequence. On this
basis KA appears to be a fully encoded consecutive marker.
There is, however, more to be said about how KA is interacting with the
lexical verbs in (10) — and more generally, about how lexical verbs are
semantically structured in Swahili. First, reconsider the translation given for
(10)d-f. The translation reflects the three-clause temporal sequence of the
Swahili original "TAKE...COME...POUR". A more idiomatic English
translation might collapse the serial-like (10)d-e into a single clause "we
492 BENJI WALD

CARRY it home'1. In point of fact, the standard Swahili-English dictionaries

(e.g. Johnson, 1975) define the verb beba of (10)d in terms of the English verb
"carry (on shoulder)". As discussed immediately below, the lexical meaning
of beba would more accurately be glossed as a clause sequence 'TAKE...and
CARRY on upper body (head or shoulders rather than hand)". This decom­
poses the lexical meaning of beba into a punctual event TAKE followed by a
non-punctual event CARRY.
The interaction between lexicon and grammatical marking of Swahili
verbs indicates that Swahili differs from English by not lexically encoding
punctuality in verbs. There is widespread agreement among linguists that
punctuality refers to the encoding of an event as "instantaneous", i.e., it
begins and ends so fast that no other event could occur within its temporal
boundaries. 6 Hopper & Thompson (1980:252ff) exemplify lexical encoding
of punctuality with the difference between English "kick" (punctual) and
"carry" (durative). In contrast, one cannot find Swahili verbs which differ
with respect to punctuality irrespective of wider discourse context. Instead,
one finds evidence that where English makes a lexical distinction, Swahili
uses grammatical marking alone.
One of the most prominent distinctions occurs in the interpretation of
verbs as state, marked by ME in most dialects, or process (single event or
habitual), marked by A (or NA in some dialects), e.g. a-ME-choka/simama/
etc. 'he s tired/(in the state of) standing up/etc.' vs. yu-A-choka/simama/etc.
'he's getting tired/(in the process of) standing up...' (as single events) or 'he
gets tired/stands up...' (either general habitual events or "historical" present
single events, as in the on-line narration of the English present exemplified in
Tomlin, 1984). Indeed, the use of grammatical rather than lexical marking for
the state:process distinction is common to a great number of languages in
Africa and elsewhere, (cf. Welmers, 1973:347ff; Comrie, 1976:57). However,
how extensive this neutrality to punctuality is among lexical verbs in general
within these languages has not been investigated cross-linguistically. Thus, in
a discussion of this phenomenon among West African languages, Welmers
(op cit) considers the case of a Kpelle verb used both as equivalent to 'see' and
'catch sight of (cf. 'notice') to be the extreme, to his knowledge. Of course,
some instances of the phenomenon are familiar to students of various Indo-
European Languages, e.g. the Old Greek "perfect with present meaning", as
in the case of verbs such as (w)oid- 'know' (of root meaning 'find out', cog­
nate with Latin vid- 'see' and English wit-ness), or the set of Spanish verbs
with alternative English durative or punctual equivalents depending on
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 493

aspect, as in the case of conocer 'know/be familiar with' (present/imperfect =

durative), "meet (for the first time)' (preterite = punctual). In Swahili all
verbs may be distinguished according to the contrast given above for ME and
A. In addition, all verbs may be marked by KA, in which case they will select
a punctual verb in the English translation, where this is possible, e.g. a-KA-
choka/simama 'he got or gets tired/he got or gets up (on his feet)../ Hence,
the translation (10)d 'take' instead of 'carry' is appropriate in temporal
sequence before the English durative verb of motion 'come'.
Beside grammatical marking of the difference between state and change
of state, rather than lexical encoding, the Swahili analogues to English lexical
verbs show a more general indifference to punctuality. A prime example is
the verb ruka, which glosses lexically in English as either 'fly' (durative) or
jump' (punctual). Thus, in English she kept flying has a continuous reading,
while she kept jumping can only be iterative. A more accurate gloss of Swahili
ruka is 'go up in the air' (cf. the causative form rusha 'throw', and see (13)d
below). Only pragmatic inference will associate ruka with 'fly' or 'jump 1 in a
given context.
Next, the relation between punctuality and perfectivity needs to be dis­
cussed with respect to KA marking. The notions labelled "punctuality" and
"perfectivity" are subject to some variation in the recent literature. Of par­
ticular relevance to the discussion of Swahili is Givón's (1982:136ft) discus­
sion of Bemba aspectual distinctions. Givón uses the term nonpunctual to
refer to 'progressive/continuous', 'habitual' or 'iterative' aspect.7 He uses the
term "perfective" to refer to events in temporal sequence. On the other hand,
Comrie (1976:16ff) uses the terms "perfective" and "punctual" in a different
way, e.g., he argues that in Russian, an obligatory-aspect language, a verb like
'wait' (po/pro-stojit') is necessarily perfective, but is lexically durative rather
than punctual, since one may "wait" for a prolonged period of time. In view
of my analysis of KA as a punctualizer and the neutrality of the Swahili verb
lexicon to punctuality, Givón's characterization of Bemba aspect as perva­
sively concerned with marking punctuality also applies to Swahili. Perfectiv­
ity in Comrie's sense is of no consequence to Swahili grammar apart from
punctuality. Verbs in temporal sequence are necessarily "punctual".
Givón goes on to contrast the marker À-...-I/ELE as "perfective" (i.e.,
in temporal sequence — in past contexts) with À-...-0 as "anterior", i.e.,
necessarily representing a reverse sequence. Within anteriority, he makes a
further conceptual distinction between whether or not an anterior event "lin­
gers" (i.e., is 'perfect') on the basis of the lexical verb, e.g. anterior marking
494 BENJI WALD

of ya 'leave' is pluperfect in a past context (i.e. 'had left'), but shipa 'be(come)
brave' lingers to the time reference point (i.e. 'was brave').
The Swahili system differs from the Bemba system with respect to
anteriority. Swahili has a perfect marker distinct from an anterior marker.
The perfect marker, as discussed above, is ME (in most dialects). It specifi­
cally marks 'linger' as an event relevant to whatever time reference point has
been independently established. For example, ni-LI/A/TA-m-kuta a-ME-
choka T found/find/will or was going to find him tired', uses kuta 'find' as the
time reference point for choka 'get/be tired', but in isolation a-ME-choka
simply uses speech time as a reference point, resulting in the translation 'he's
tired' (i.e. he got tired and still is). While ME necessarily denotes an event
which began before the time referent point, and marks anteriority in this
sense, it is distinct from another marker commonly used to signal anteriority
in the past, LI. The following discussion indicates that LI, like KA, is punctual.

4.2. LI
LI, like KA, is associated directly with punctual events. However, there
is a special construction which allows LI to establish a time reference point for
states. Example (11) below shows how LI functions to mark anteriority
regardless of the punctuality of the clause.
(11) a. LI nilikwenda
b. LI kwa kuwa mimi nlitoka hapa
c. LI ... roho yangu ilikuwa
d. A yapenda kwenda kusoma katika Saudi Arabia
e. LI maana mi nlisoma lugha ya kiarabu kidogo ...
a. "I went
b. inasmuch as I left here
c. ... I had
d. wanted to go study in Saudi Arabia
e. because I had studied a little Arabic..." (AH53m,Msa)
The time reference of successive Li-marked clauses move progressively back­
wards. The temporal sequence would be "STUDY...WANT...LEAVE...GO
(to)". The Li-marked events in (ll)a,b and e may be read as punctual.
"Clauses" (ll)c-d orient toward a state by means of the "compound tense"
sequence
c. LI-kuwa + d. A,
marking penda 'like'. Marked by LI, ku-wa 'be(come)' acts as a "dummy"
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 495

punctual event anterior to the established time reference point (of (9)b in this
case). Unlike ME, LI does not necessarily imply that events so marked are
relevant to the time reference point. 8
There is more to be said about LI. First, it is strictly bound to anteriority
before speech time, unlike ME. In this respect it is unique among the Swahili
verb markers. No other Swahili verb marker encodes reference to speech
time. Second, there are specific discourse contexts in which it is not anterior,
but rather in temporal sequence with a preceding clause, as in (12) below:
(12) a. LI alipolishwa
b. LI alishika ndia
a. "As soon as he was fed
b. he hit the road (i.e. he ate and ran) (SW19m,Msa)
As the English gloss for (12)a "as soon as" suggests, the consecutive use of LI
in (12)b implies a minimal interval between two punctual events. It is impor­
tant to note that the temporal sequence is not a narrative sequence in either
the Swahili original or the English translation. The clauses can be reversed
without affecting inferred temporal order. However, examples like (12),
although relatively rare in narrative because of their specific emphasis on the
temporal closeness of two successive events, suggest that LI is indeed 'past',
rather than truly anterior. It may either reverse or create a temporal
sequence, but the event it marks must always be prior to speech time. In both
respects it differs from KA.
It will soon be made clear that the special status of LI derives from a stage
at which absolute tense marking played a greater role in Swahili, as it does in
Bemba and some much closer relatives of Swahili in Northeast Bantu. For the
moment, however, it is important to note that LI marking a verb in temporal
sequence is restricted to contexts in which it follows a clause either marked by
LI-PO (i.e. LI + the temporal relative marker -PO 'when') or KI. The
replacement of LI-PO by KI in this context does not affect interpretation.
Thus compare (12) with (13) below:
(13) a. KI ikija
b. LI ilipiga tu
c. KI ikipiga
d. LI iliruka hivi
e. KA ikapinduka
f. KI 'kisha
g. KAikarudi tena hivi
496 BENJI WALD

a. "as soon as it (= the car:BW) came

b. it hit (= the other car:BW)
c. as soon as it hit
d. it went up (in the air)
e. and turned over
f. when it finished (i.e. and then, cf. (10)c above)
g. it turned back (upright)," (FA23f, Msa)
Again LI marked verbs in (13)b and d denote events in temporal sequence
with the events denoted by the immediately preceeding clauses. The "nor­
mal" marker of temporal sequence, KA, begins in (13)e, where the shortness
of the time interval between the temporally ordered events is not
emphasized.
Despite the neutralization of KI and LI-PO in punctual past temporal
sequences, KI has a much more general range of discourse uses than LI-PO.
In many respects, KI presents a contrast with KA most revealing of the nature
of the Swahili verb marking system.

4.3. KI
In contrast to KA, KI clauses are normally reorderable in principle,
without changing the inferred temporal order, e.g.
(14) a. KI akipita adui
b. TA mkubwa atasema "mkanyageni!"
a. "if/when an enemy passd (below)
b. the leader would say "step on him!" (AD16m, Msa)
(15) a. ng sijui
b. TA ntafanya nini
c. KI nkiolewa.
a. "I didn't know
b. what I would do
c. if/when I got married." (ML42f,Lamu)
KI is seen above associated with clauses marked by TA irrealis' (i.e. 'future'
or 'conditional'). When postposed, as in (15)c, KI is interpreted as a reverse
sequence, i.e. MARRY...DO is the temporal order. In this respect the KI
clause is identical to the adverbial clause of English and many other lan­
guages.
KI has additional semantic properties which are not associated with sub-
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 497

ordination in English. Thus, in the "compound tense" construction following

the "dummy'1 auxiliary ku-wa, it behaves like the "imperfective" of many lan­
guages with obligatory aspect marking, having a habitual or progressive read­
ing, e.g. ni-LI-kuwa ni-KI-tembea T used to walk/was walking'.
The reason I said above that a KI clause is "normally" reorderable is
because of the case past habitual discourse, as in (16) below, in which reversi­
bility of the KI marked clauses is problematic.
(16) a. LI Kuweit walikuwa
b. KI wakija hapa na majahazi yao...
c. KI majahazi yao yakifika hapa
d. KI wakipiga ngoma siku zote.
a. "From Kuweit they used to
b. come here with their dhows
c. and (when) their dhows arrived
d. they used to party every day." (Abd23m,Msa)
"Clauses" (16)a-b feature the compound tense construction. In effect, clause
(16)a is simply orientation to the past. Clauses (16) b-d form a past habitual
temporal sequence. The choice of KI rather than KA in this context makes it
explicit that the sequence is past habitual. While KA functions in a general
context, as in (10) above, as well as in narrative, KI is the rule in past habitual
contexts.
Note that the gloss for (16)c equivocates on whether or not the clause is
subordinate. Our test for subordination, the effect on interpretation of reor­
dering, works in principle. Each clause is permutable with the following
clause as long as it is interpreted as an "adverbial" clause. According to the
strategies actually observed in Swahili discourse, however, a series of past
habitual events are invariably presented in temporal sequence.
It is interesting to note in this context that in the relatively rare cases in
which a sequence of KI-marked clauses follows an associated main clause, it
is a reverse sequence, as in the following line of verse by Muyaka, the early
nineteenth century Mombasan poet.
(17) a. NGE nyote mngelia
b. KI mkiniona
c. KI hipita (= n-ki-pita)
a. "you would (hypothetical) cry (tears of jealousy)
b. if you saw me
c. passingby." (Hichens, 1940:52)
498 BENJI WALD

In order to recognize the sequence as reverse, interpret each clause as punc­

tual and note the temporal order is PASS...SEE...CRY.
The temporal sequencing of KI clauses extends beyond past habitual dis­
course to "subordinate" narrative clauses (amidst complex preposed adver­
bial clauses containing verbs in temporal sequence, in general). Thus,
(18) a. KI sasa akitoka yule bwana
b. KI akikumbuka lile koti lake
c. KA akarudi mbio mpaka nyumbani
d. KI akirudi nyumbani
e. KI sasa akija
f. KI akishika koti
g. A asikia
h. A vitu vyalia ndani.
a. "so when the guy left
b. and remembered his overcoat
c. he ran back to the house
d. when he ran back (i.e. and then)
e. and came
f. and took the coat
g. he heard
h. some things making a noise inside." (CH19m,Msa)
The temporal sequence is divided into two discrete sections (in this case "nar­
rative episodes", i.e. "little narratives" on their own). The first is (18)a-c; the
second is (18)d-g. The division between the sections is marked by repetition
in (18)d of rudi 'return'. This repetition suspends the temporal sequence.
Repetition of the type in (18)d is a common device in Swahili narrative (as in
many languages, cf. Grimes & Gloebe, 1970 for Saramaccan narrative and
Marchese, this volume for Godié procedural discourse). The repeated clause is
analogous to either a repeated clause marked by 'when' in English, as in the
(infelicitous but literal) translation of (18)d, or more usually pronominaliza-
tion of the entire clause by '(and) then'.
Following the suspension of (18)d, the temporal sequence is resumed in
(18)e-g. In principle, KI can be replaced by KA without changing the tem­
poral sequence. However, it is not unusual in short episodes for KI marking
to continue across clauses with a shift in marking as the final clause of the tem­
poral sequence is approached. Thus, (18) exemplifies this twice. The first
case is (18)c followed by the repetition (18)d. In the second case (18)g is the
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 499

last clause in the sequence. Then (19)h breaks the temporal sequence with a
clause temporally unordered with respect to (18)g. Conceived of as punctual
events, the temporal order of (18)g-h is reverse: MAKE NOISE - HEAR.
Temporal sequences of the type in (18) are reminiscent of those found in
languages which may specially mark the last verb of a temporal sequence.
Givón (1972:232) describes this situation for Bemba. In a Bemba past or fu­
ture temporal sequence, tense marking may be suspended for the last verb,
e.g.
(19) a-À-ish-ile a-À-ike-ele a-À-li-ile a-Â-ya
he-P-come-sf he-P-sit-sf he-P-eat-sf he-SH-leave
"he came and sat and ate and left".
Here P stands for the relatively distant past tense morpheme and sf for the
suffix marking verbs in past temporal sequence. The last verb shifts to a
marker (labelled SH above) which does not show tense.
Swahili: KI...KI...SHIFT
Bemba TENSE.. .TENSE.. .SHIFT
Both the Swahili and the Bemba systems fit a general pattern of shift from a
sequence of identically marked verbs in temporal sequence to another
marker (simply labelled "shift" in the above scheme) without breaking tem­
poral sequence. This pattern fits the continuum of levels of grounding discus­
sed in section 2.3. above with reference to the Jones & Jones (1976) proposal.
In the Bemba example the special marker of shift (which is neutral to tense)
is equivalent to "peak", the highest "ground" in the discourse. From prior
discussion, it should be clear that Swahili does not have a special "peak"
marker. Instead, peak is marked in a local section of discourse by down-shift­
ing all but the last clause of a temporal sequence. This downshifting is
accomplished by KI marking. Thus, KI marking is not necessarily an indi­
cator of "background" as opposed to "foreground" (though it may be used in
that way), but rather a marker of relative shift downward.
In narrative and many other types of discourse (e.g. procedural),
downshifting through KI-marking is commonly used to break the main line
into two local episodes. Two conventions are commonly used in this case:
1. KI marking a repetition of the last verb, as in (13)c and (18)d
above.
2. KI marking a verb meaning 'finish', usually isha, as in (13)f, but
also koma in some dialects, as in (10)c.
500 BENJl WALD

Technically there is no difference between these two options. Both are fre­
quently used in narrative. Lexically, 'finish' has a specific meaning distinct
from other verbs. However, since KA explicitly marks a verb as punctual,
when 'finish' refers to a preceeding verb marked by KA in a discourse con­
text, the event denoted in the KA clause and the event referred to in the fol­
lowing KI-SHA "clause" are identical. This is why KI-SHA is simply translat­
able into English as "and then", "afterwards11, or "next11. Both strategies of
sectioning discourse may be found together in the same clause; thus, note the
following segment of general routine.
(19') a. KA... tukatoka kule nyumbani
b. KI Sasa tu-ki-sha kutoka ...
c. NAtunakwenda mpaka ile dau .,.
a. "and then we leave home
b. when we finish leaving ... (i.e. AFTER we leave)
c. we go to the dhow...11 (AB21m,Msa)
In view of the lexical power of isha, it is perhaps not surprising to find
dialects in which is ha suspends a narrative sequence in connection with KA,
as in the following passage:
(19") a. KAruka/tf
b. KArukQsha (= ru-ka-isha)
c. KArukenda (= ru-ka-enda) madukani
a. "we ate
b. (and then/when) we finished (i.e. eating)
c. we went to the shops.11 (YH 16m, Vanga)
Despite the loss of the downshift marker KI in these dialects, it is relatively
easy in most discourse contexts to recognize by means of the attached lexical
verb if KI would be used in the urban dialects of Swahili. However, the
"peak" constructions of (18) are lost in the rural dialects.
I have not observed the rhetorical arrangement represented in (19") in
any current urban dialect of Swahili, where KI would be used instead. It is,
however, commonly observed in the central rural dialects of Swahili, where
KI has totally merged into KA.
Having now associated the function of KI with downshifting, let us con­
sider how the use of KI in past habitual discourse (e.g. past routines) fits into
the general function of KI. The question is: by what perspective can the past
habitual be considered a downshift? Downshift from what? This is a particu-
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 501

larly pressing question in view of the widespread use among languages of the
same marker for progressive and habitual reference. Progressives occur in
narrative to mark the part of "background" concerned with ongoing actions
coincident with the temporal sequence of the main line. As such they are not
in temporal sequence with the "foreground", and can be reordered with
respect to perfective or punctual clauses without change in inferrred temporal
order. Thus, a progressive behaves on a discourse level like an adverbial
clause does with respect to its associates main clause on a more local level in
discourse (namely, the local level identified with the "sentence" in written
texts).
Habitual or general discourse contrasts with narrative in that it refers to
an iterative temporal sequence rather than a single temporal sequence. There­
fore, it is, of course, true that there is some narrative implied by the iterative
temporal sequence to which the iterative temporal sequence may be pre-
posed or postposed (as noted by Labov (1972) and critically discussed in sec­
tion 2.1. above). For example, there is some particular occasion, expressible
as a narrative for (16), on which they partied both before and after two par­
ticular occasions on which they arrived in Mombasa. That is, they partied on
some occasion before they left and then returned to Mombasa, and on some
occasion after they returned to Mombasa. The shift from narrative to routine
or procedure is from a specific time to a set of times with which the specific
time of the implied narrative is unordered. The answer to downshift from
what, then, is simply downshift from narrative.
The discourse "ordering" of habitual and narrative discourse depends on
whether or not the rhetorical conventions of the occasion and/or culture favor
deductive (habitual first then narrative) or inductive (narrative first) organi­
zation of information. It seem that deductive rhetoric is more usual in every­
day discourse, favoring habitual-first discourse, at least in English and
Swahili. The rhetoric of scientific discourse tends to follow the same pattern:
first the hypothesis (habitual, i.e. "general"), then the experiment (narra­
tive). Of course, in many situations the narrative implication of habitual/gen­
eral discourse remains implicit.
It should be clear that narrative implication follows simply from the
nature of habitual or general discourse, whether past or continuing into
speech time. Habitual discourse is a set of narratives collapsed (and con­
densed from the specifics of any particular occasion) into a single discourse.
Narrative implication may well be a universal feature of discourse whether or
not it is specifically encoded into the grammar of a language. Under various
502 BENJI WALD

social circumstances, habitual discourse is explicitly converted to narrative.

A simple example from English is found in (20), where 11-year-old YL con­
verts the habitual discourse of BP's question into a narrative answer.
(20) BP: ... how do you make your money - your spending money?
YL: Like when I go to the store, like YESTERDAY I went to the
store and my - when I came back my dad gave me a dollar for
going...
(Upper case in this and the next example orients to narrative)
Sometimes the generalization represented by habitual discourse has a specific
prototypical narrative behind it. Thus, in (21) two 18-year-olds' discourse
about the personality of a mutual friend NC, one particular narrative expe-
rientially shared by LD and EY underlies the opening general statement.
(21) LD: And it - it bothers NC to see when I don't wear make-up,
huh? She yells it to the world.
EY: (to a third person) You know what she DID? Oh my God! (to
LD) Tell her! Tell her!
LD: I was walking to go get my lunch n she goes... I just looked at
hern I go...
(The italicized verbs are in narrative sequence. Note in passing
the progressive orientation of 'walk' to the narrative sequence.)
The preceding discussion on the relationship between habitual and nar­
rative discourse may be represented diagramatically as follows:

a a a
/ \
b b b

Diagram 1. Habitual Discourse.

The external boundaries of the diagram represent the envelope of the

habitual discourse. The vertical dimension represents a single occasion on
which events a and b are temporally ordered. The horizontal dimension rep­
resents recurrences of events a and b on separate occasions. Reading from
left (earlier occasion) to right (later occasion), the arrows arbitrarily show
that some event b occurs both before and after some other event a, but on dif­
ferent occasions. If the difference between occasions is ignored, as it is in the
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 503

external frame for habitual discourse, then, naturally events a and b are tem­
porally unordered.

a a a

b b b

Diagram 2. The narrative implication of habitual discourse.

Diagram 2 retains the same meaning as Diagram 1 above. However, in Diag­

ram 2 the narrative implication of the habitual discourse is singled out by an
internal set of boundaries representing a single occasion. On this occasion
events a and b are strictly temporally ordered internally, as they are, of
course, on each occasion. Furthermore, the occasion represented by the nar­
rative implication is temporally ordered with respect to some other prior
occasion and some other subsequent occasion.

Diagram 3. deductive Diagram 4. inductive

rhetoric rhetoric

Diagrams 3 and 4 represent the two different kinds of rhetoric, where the nar­
rative implication of habitual discourse is made explicit. The deductive
rhetoric postposes the narrative to the habitual discourse, as exemplified in
(20) and (21) above. The inductive rhetoric preposes the narrative to the gen­
eral discourse. This is typical of just-so stories or moral tales. It might be used,
for example, in telling the Greek myth of the Trojan horse, and then ending
with "So (that's why they say) beware of Greeks bearing gifts." Of course,
one may also both prépose and postpose the habitual discourse to the narra­
tive. We might call this didactic rhetoric. In any case, the narrative tends to
be much richer in event structure (number of events explicitly expressed)
than the habitual discourse abstracted from a series of narratives of similar
occasions. Similarly, the number of events downshifted (e.g. by subordina­
tion) in narrative tends to be relatively small in number compared to the
504 BENJI WALD

number of events featured in "main" clauses, with the possible exception of

languages, like Swahili, which may use downshift in order to create a peak
construction, as in (18) above. In point of fact, however, with regard to
Swahili, a simple eye inspection of any extended Swahili narrative reveals
that KA is so much more frequent than KI marking that a count of each
marker is superfluous. That is, peak is too sparingly used in Swahili narrative
to affect the numerical supriority of KA over KI. On the other hand, there are
other discourse genres in which the peak construction is much more common;
for example, in the hypothetical discourse of rhetorical questions such as (22)
below, where FT justifies his distaste for the violence of cowboy and gangster
movies:
(22) a. KI sasa mimi nkija
b. KI nkikupiga wee mwili wako
c. KI nkikukata
d. TA utasikia raha?
a. "So if I came
b. and struck your body
c. and cut you
d. would you enjoy it?" (FT 19m, Msa)
In (23) below the overall downshift feature of KI in Swahili can be seen
most clearly.
(23) a. KI zamani sote sisi tukicheza kirikiri
b. KA skmngia yule shetwani ' soul.
a. "Before, we used to dance kirikiri
b. (and/but) then that demon (called) soul came in (lit. entered).'''
(KB17m,Msa)
Inference from context leads to a past habitual interpretation of KI. In other
contexts, the sequence might be more appropriately translated "while we
were dancing kirikiri, soul (another musical style) came in" or "when we
finished dancing kirikiri, soul came in". KI simply encodes downshift. Other
features of interpretation depend on further information in the discourse con­
text.
A final observation about the evolution of KI in Swahili is of interest
toward an understanding of the special role that narrative has played in the
evolution of grammatical encoding in Swahili discourse. This can be seen
from a comparison of Swahili with Giriama (among other languages particu­
larly closely related to Swahili in Northeast coastal Bantu).
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 505

As stated above, Bemba is more typical of Bantu than Swahili in encod­

ing speech time on the clause-level in the main line of narrative discourse. It
is also more typical of Bantu in encoding different degrees of pastness from
speech time. This situation remains true in most of Swahili's closest relatives
as well. For example, unlike Swahili, Giriama distinguishes "today past"
(DZA) from "before past" (A). In addition, like Bemba, it does not distin­
guish perfect from (today) past (cf. Wald, 1976). Furthermore, like Bemba,
it encodes both speech time and temporal sequence on the main line of narra­
tive. However, unlike Bemba, Giriama encodes temporal sequence and
speech time into a single marker, KA ("today" consecutive): KI ("before
today" consecutive). The following parallel translations from a passage in the
New Testament exemplify.
(24) SWAHILI GIRIAMA
a. LI alipokuwa katika RI hata arihokala yunadza
kumwendea
b. KA pepo akambwaga chini KI pepo akimugwaga
c. KA akamtia kifofa KI akimutarura-tarura
a. "and as he was yet a-coming
b. the de vel threw him down
c. and tare him (i.e. gave him an epileptic fit)" (Luke 9:42)
This is the only way in which Giriama and Swahili differ in the use of KI.
The Giriama system is indicative of an earlier stage of the Swahili sys­
tem. As Swahili collapsed degrees of pastness into LI, KA became the con­
secutive marker not only for "today" (past or general, i.e. time reference
includes "today"), as in Giriama, but also for the "before past". The collapse
did not, however, extend to the past habitual. Thus, Giriama and Swahili
maintain identical uses of KI in past habitual contexts, e.g.
(25) SWAHILI GIRIAMA
a. KI ... kilamchanaalikuwa KI Bai kila mutsana were
akxfundisha hekaluni... akifundya kahi za m..
b. KI watu wote walikuwa KI atu osi makirauka...
wakiamka mapema...
a. "... every day he used to teach in the temple...
b. (and) everyone used to get up early..." (Luke 21:37)
By virtue of the extension of KA to narrative, the distinction between main
line and downshift has become more general in Swahili than in Giriama. Nar­
rative has become formally more distinct from past habitual discourse. Evi-
506 BENJI WALD

dentally, Swahili reanalyzed the "before" (or "remote") past feature of KI,
common to its single and habitual past uses in Giriama, as habitual. In group­
ing it together with the "subordinate" use of KI, the common feature became
simply "downshift" both within and across discourse types. By retaining LI as
a marker of orientation to the past, Swahili has become "aspect-prominent"
but not "aspect-only".

5. Conclusions and directions for research

With regard to the focus on narrative in linguistic (and psycholinguistic
studies), the question asked at the outset of this paper was: why study narra­
tive as opposed to some other discourse genre? We have seen that on the con­
ceptual level narrative consists of four discrete properties: 1. distinction in
"ground", 2. temporal sequence, 3. reference to the past, 4. reference to
single events. Altering any one of these properties results in a different genre
of discourse. Narrative differs from past routines in referring to temporal
sequences of single events. It differs from current routines and procedural
discourse in referring to past events. It differs from subsections of any larger
discourse type by having a "main" ground (reflected in a "main" temporal
sequence).
Tannen (1982:5) speaks for the experience of sociolinguists in general in
suggesting that narrative has been a common topic of study because it is a
relatively common genre,is easy to elicit, and has (relatively) easily recogniz­
able boundaries. Certainly much remains to be said about why this is so, e.g.,
the social and psychological underpinnings of the narrative genre. However,
it remains to further empirical study to establish if narrative is indeed more
common, easier to elicit (among all age groups in all languages), and has
more easily recognizable boundaries than, say, past or current routines and
procedural discourse. In any case, these consideration bespeak a concern
with convenience in data collection and analysis, and perhaps stem from the
fact that from an interpersonal point of view, narrative is inherently more
interesting than other genres, since it is concerned at best with the transmis­
sion of personal experience. On the other hand, these considerations do not
speak to a concern with grammar. Is it the case, for example, that narrative
is grammatically richer than other discourse genres? 9
The discussion of Swahili narrative and other discourse genres confirms
the observation discussed in the beginning of this paper (re: Linde, 1974;
Thompson, 1984), with respect to the particular grammatical bias of looking
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 507

at narrative alone without considering the relation of the use of grammar in

narrative to its use in other discourse genres. The discussion of the downshift­
ing function of KI within and across discourse genres illustrates the need to
look at genres beside narrative in understanding the operation of grammati­
cal systems in constructing discourse. It is evident from the discussion of
Hausa, Bemba, Swahili and English, that languages differ in what aspects of
discourse they grammatically encode. Hausa obligatorily encodes aspect,
optionally encodes temporal sequence, and has no tense encoding markers;
English optionally encodes aspect and temporal sequence, but obligatorily
encodes tense; Bemba obligatorily encodes tense, aspect and temporal
sequence; Swahili is intermediate between Bemba and Hausa in the encoding
of tense. In all cases, these general characterizations of the encoding proces­
ses are subject to qualifications on the basis of "grounding" and discourse
type. For example, most dialects of English suspend coding for modality but
not tense in adverbial clauses, e.g. whenever they asked her, she would (=
modal 4- past) tell them "/or"whenever they would ask....
Despite its evolution toward an aspect-only system, Swahili shows resis­
tance to the loss of the resource or explicitly marking the past. Thus, LI
remains as an unambiguously past punctual marker, where it serves to orient
discourse toward the past, or, if pastness is already established, toward
anteriority in the past. The KA:KI distinction remains in past contexts to dis­
tinguish narrative from past habitual discourse respectively. More work
needs to be done across languages to see to what extent the evolutionary pro­
cesses affecting their grammars in discourse reveal a resistance of the past, or
other prominent features of narrative, e.g. temporal sequence, singleness of
events and grounding, to the collapse of distinctions.
Considering further the larger discourse restrictions on the encoding of
temporal sequence, the following hypothesis may be offered.
(HI) No language encodes temporal sequence unambiguously on the
clause-level for all discourse types.
This is clear for English, where temporal sequence is rarely encoded at
all. In Swahili temporal sequence is encoded by KA on the clause-level, but
in past habitual discourse KI is used. KI, however, is a marker of downshift,
not temporal sequence, so that a KI-marked clause out of further context is
not recognizable for temporal sequence. Similarly, in Hausa, the use of the
subjunctive (in habitual/general discourse) and the "relative completive" (in
narrative) to express temporal sequence is only recognizable in a context
larger than the clause. A subjunctive or relative completive clause alone is
508 BENJI WALD

not recognizable as a clause in temporal sequence. With regard to the relative

completive, the use of the same marker for temporal sequence and relative
clauses suggests an original ground shifting function for the marker. The
Bemba marking of temporal sequence in narrative bears further investiga­
tion. Givón (1982) proposes that historically the Bemba marker Aiele, which
marks temporal sequence, was an anterior marker, not a consecutive marker.
That is, it was earlier a marker of deviation from temporal sequence. Thus,
it appears to have originated as a marker of downshift. The route by which it
came to mark temporal sequence needs further explication.
At the present state of knowledge about the encoding of temporal
sequence, two further hypotheses may be offered.
(H2) If a language encodes temporal sequence at all on the clause-level,
it will do so in the context of narrative discourse, but not necessarily
elsewhere.
(H3) The encoding of temporal sequence has its origin in some other dis­
course function.
The most likely origins for regular temporal sequence marking is the
marking of punctuality (as in the case of Swahili KA) and/or groundshifting
(as in the cases of the markers sensitive to temporal sequence in Hausa and
Bemba.
Finally, more attention needs to be paid to the relation between lexical
and grammatical encoding in the construction of discourse. Most recent work
in discourse appears to assume that either languages do not have a lexical
structure, or the lexical structure is similar or the same in most languages, e.g.,
distinguishes punctual and durative verbs. As discussed above, the Swahili
verb lexicon shows indifference to this dimension. Consequently, the verb
marking system is fully productive, without any lexical restrictions. A very
important exception, in my view, to the general neglect of the study of lexical-
grammatical interaction is Pawley's (1984) study of Kalam as a language with
a small set of verbs of fairly abstract meaning. Equivalents to the much larger
set of verbs of English, and most other languages, are accomplished by verb
sequences, i.e. little discourse units in themselves, e.g. "hunt (for food) and
then eat" is expressed by the temporal sequence '*(go)-kill-carry-come-bake-
eat" (in that order). Here the "narrative implication" of lexical items is seen
even more clearly and regularly than in the comparison of the English and
Swahili verb lexicon, e.g., carry vs. beba "take and transport on upper part of
body". It follows from this that languages not only differ in the encoding of
semantic features (like punctual/durative) into lexical structure, but also in
 CROSS-CLAUSE R E L A T I O N S A N D T E M P O R A L S E Q U E N C E 509

how they map events onto verbs and then clauses. It is alien to English to con­
ceive of verbs of transportation as punctual. One may punctually "take" or
"grasp" something but one can only duratively "carry" or "hold" it. In
Swahili, only larger discourse context will tell if a "taken" object is trans­
ported. A common way of expressing transportation is simply a following
clause with a directional verb (like come in (10)e). Hence two clauses for
Swahili where English commonly manages with one.

NOTES
1 ) As Hopper (1979) notes, inversion was much more characteristic of narrative during the Old
English period. See the end of fn 4 below for one common use of inversion extending well into the
Middle English period.
2) The characterization of productivity/creativity in terms of an infinite number of purposes
obviously derives from Chomsky's model for characterizing the linguistic creativity of sentence. It
should be noted that Chomsky's innovation lay in his proposal for how to model linguistic produc­
tivity at the sentence level. The general issue of grammatical productivity is traditionally of central
importance to linguistic description.
3) This is not to say, for example, that the choice between lexical and grammatical encoding of
events is language-independent, but that all languages recognize a set of events which are encoded
lexically into verbs, forming the nucleus of clauses, and leading to the identification of narrative
sequences. The language-dependent aspect of lexical encoding is disucced further later on in the
text.
4) For example, Mithun notes that aqa in the following example from Kathlamet is glossed
'then' by Boas, but translated as 'wfhen':
aqa ió.maqt yaXi iqcxé:Lau Ictó.pa
THEN it.died that monster they.two.went.out
"WHEN the monster was dead, they went out".
Similar phenomena are pointed out in the discussion for other unrelated languages, e.g. Mohawk
and Gunwinggu. In each case the clauses are in temporal sequence, but whether or not the
sequence is reversible is not discussed. It is evident that forms like aqa are temporal pronouns
meaning 'at that time'. It is worth pointing out in passing that the Germanic systems of subordina­
tion evolved from a similar system. As late as the mid 15th century examples of THEN introducing
a subordinate time clause are still found in English, e.g. THEN (=when) hys houndys began to
baye, that harde (=heard) the jean (=Genoan) there (=where) he laye. (1440, OED then 6.)
In the case of Germanic, however, clause-internal word order signals the difference between main
and subordinate clause. Thus, subordination was marked in English with SV order, while a follow­
ing associated main clause had VS. Hence THEN SV = "when SV" and THEN VS = "then SV".
5) CR is used in examples (8) and (9) to refer to the "relative" form of the completive. This form
is kà or 0 depending on the person of the subject marker. In (9), RM refers to the 'relative marker'
dà, cf. wan-da (demonstrative + RM) in (8); and CM refers to the non-relative completive aspect,
formed by lengthening the vowel of subject marker or -n, corresponding to the use of Ø/kà respec­
tively in CR. With regard to (9), the use of the RM alone is quite common in West African lan-
510 BENJI W A L D

guages, regardless of genetic affiliation, as a marker of temporal clauses, as if condensed from

'(time) which'.
6) It is important to realize that durability and punctuality are not "objective" features of
events, but rather of how events are lexically encoded. Thus, for example, the English verb burn
is a (durative) process verb. It implies the initial punctual event "catch on fire/ignite (intrans.)" and
is implied by the terminal punctual event "burn up" or "burn down". In contrast, the English verb
fall has duration in time, but is not graced with idiomatic lexical expressions for punctual events
which may be associated with falling, such as "get destabilized" ("slip"?) or "hit the ground (i.e.
finish falling)". That is, it burned does not imply "it burned up (to completion)", typical of process
verbs, while it fell does imply "it hit the ground". Where languages, like English, do make a lexical
distinction for punctuality between types of verbs, it seems that the distinction is based on "real
world" considerations of the relative length of time it takes for a process to reach completion.
Thus, "burning up" usually takes longer than "falling".
7) In Swahili, these aspects and more are marked by KI in certain contexts, as discussed in more
detail later in the text.
8) As a matter of fact, in context (11) c initiates a more extensive reorientation from (11 )b than
is apparent from the citation. This is because kwenda 'go' in (ll)a refers to Uganda, not Saudi
Arabia. As the narrative progresses it becomes clear that (11)c-e is relevant to explaining how AH
became a teacher of Arabic and Islam in Uganda.
9) In certain respects, involved in the development of syntactic studies of discourse, narrative
is indeed "richer" than most other genres, namely, in the complexity of cross-clause nominal rela­
tions. The number of references to nominal entities and their persistence in narrative discourse has
favored this type of discourse for analysis of topicalization, passivization, pronominalization, etc.

REFERENCES

Abraham, R.C. 1959. Hausa Literature and the Hausa Sound System. Lon­
don: University of London Press.
Ashton, E.O. 1944. Swahili Grammar (including intonation). London:
Longmans.
BLS 10. 1984. Proceedings of the 10th Annual meeting of the Berkeley Lin­
guistics Society. Berkeley Linguistics Department of the University of
California at Berkeley.
Comrie, B. 1976. Aspect. Cambridge: Cambridge University Press.
Givón, T. 1972. "Studies in Chi Bemba and Bantu grammar." Studies in Afri­
can Linguistics. Supplement 3.
. 1982. "Tense-aspect-modality: The creole proto-type and beyond". In
P.J. Hopper, ed. Tense-Aspect: Between Semantics and Pragmatics. 115-
166. Amsterdam, J. Benjamins.
Grimes, J.E. and N. Gloebe. 1970. "A Saramaccan narrative pattern." Lan­
guage. 46.2,408-425.
 CROSS-CLAUSE RELATIONS AND TEMPORAL SEQUENCE 511

Haiman, J. and S.A. Thompson 1984. "Subordination in universal gram­

mar". BLS 10.510-523
Hichens, W. 1940. Diwaniya Muyaka bin Haji al-Ghassaniy. Johannesburg:
Witwatersrand University Press.
Hopper, P.J. and S.A. Thompson. 1980. "Transitivity in grammar and dis­
course." Language 59:2.251-299.
Hopper, P.J. 1979. "Aspect and foregrounding in discourse." In T. Givón,
ed. Discourse and syntax. 213-242. New York: Academic Press.
Johnson, F. 1975. A Standard Swahili-English Dictionary. Nairobi: Oxford
University Press.
Jones, L.B. and L.K. 1979. "Multiple levels of information in discourse." In
L.K. Jones & R. Longacre, eds. Discourse studies in Mesoamerican lan­
guages. 3-27. SIL Publication 58. Vol. 1. Arlington: University of Texas at
Arlington.
Kraft, G.H. and A.H.M. Kirk-Greene. 1973. Hausa. (Teach Yourself
Series). London: Hodder & Stoughton.
Labov, W. and J. Waltezky. 1967. "Narrative analysis." In June Helm, ed.
Essays in the verbal and visual arts. 12-44. Seattle: University of
Washington Press.
. 1972. "The transformation of experience in narrative syntax." In Lan­
guage in inner city: Studies in the Black English vernacular. 354-396.
Philadelphia: University of Pennsylvania Press.
Linde, C. 1974. The linguistic encoding of spatial information. Unpublished
PhD. dissertation. Columbia University.
Marchese, L. this volume. "Conditionals in Godié procedural discourse."
Mithun, M. this volume. "Is basic word order universal?"
----. 1984a. "How to avoid subordination." BLS 10.493-509.
Ohmanson, R.C. 1982. "An analysis of narratives: Identifying central, sup­
portive, and distractive content." Discourse Processes. 5.195-224.
Pawley, A. this volume. "Encoding events in Kalam and English: Different
logics for reporting experience."
Silva, M.N. 1983. The development of conjunction in later childhood.
Unpublished PhD. dissertation. University of California at Berkeley.
Tannen, D. 1982. "Oral and literate strategies in spoken and written narra­
tives." Language. 58:1, 1-21.
Thompson, S.A. 1984. "Subordination in formal and informal discourse."
Paper given at the Georgetown Round Table, 1984.
Tomlin, R. 1985. "Foreground-background information and the syntax of
512 BENJI WALD

subordination Text 5.
Van Valin, R.D. Jr. 1984. "Subordination in universal grammar." BLS
10.542-558.
Wald, B. 1973. "Variation in the tense-markers of Mombasa Swahili."
Unpublished PhD dissertation. Columbia University.
. 1976. "Comparative notes on past tenses in Kenyan Northeast Bantu
languages." Studies in African Linguistics. Supplement 6.
Welmers, W.E. 1973. African Language Structures. Berkeley: University of
California Press.