THE LIMITS OF GRAMMATICALIZATION

TYPOLOGICAL STUDIES IN LANGUAGE (TSL)

A companion series to the journal "STUDIES IN LANGUAGE"

Honorary Editor: Joseph H. Greenberg

General Editor: Michael Noonan
Assistant Editors: Spike Gildea, Suzanne Kemmer

Editorial Board:
Wallace Chafe (Santa Barbara) Ronald Langacker (San Diego)
Bernard Comrie (Los Angeles) Charles Li (Santa Barbara)
R.M.W. Dixon (Canberra) Andrew Pawley (Canberra)
Matthew Dryer (Buffalo) Doris Payne (Oregon)
John Haiman (St Paul) Frans Plank (Konstanz)
Kenneth Hale (Cambridge, Mass.) Jerrold Sadock (Chicago)
Bernd Heine (Köln) Dan Slobin (Berkeley)
Paul Hopper (Pittsburgh) Sandra Thompson (Santa Barbara)
Andrej Kibrik (Moscow)

Volumes in this series will be functionally and typologically oriented, covering 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 37

Anna Giacalone Ramat and Paul J. Hopper (eds)

The Limits of Grammaticalization

 THE LIMITS OF
GRAMMATICALIZATION

Edited by

ANNA GIACALONE RAMAT

University of Pavia
PAUL J. HOPPER
University of Pittsburgh

JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM/PHILADELPHIA
 The paper used in this publication meets the minimum requirements of
American National Standard for Information Sciences — Permanence of
Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data

The limits of grammaticalization / edited by Anna Giacalone Ramat, Paul J. Hopper,
p. cm. — (Typological studies in language, ISSN 0167-7373; v. 37)
Chiefly papers presented at a symposium held during the 28th annual meeting of the
Societas Linguistica Europaea which was held Aug. 1995, Leiden, Netherlands.
Includes bibliographical references and index.
1. Grammar, Comparative and general—Grammaticalization-Congresses. I. Giacalone
Ramat, Anna. 1937- . II. Hopper, Paul J. III. Series.
P299.G73L56 1998
415~dc21 98-21202
ISBN 90 272 2935 X (hb.) / 90 272 2936 8 (pb.) (European; alk. paper) CIP
ISBN 1 55619 649 0 (hb.) / 1 55619 650 4 (pb.) (U.S.; alk. paper)
© Copyright 1998 - 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 75577 • 1070 AN Amsterdam • The Netherlands
John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA
 Table of contents

Introduction 1
Anna Giacalone Ramat and Paul Hopper
Grammaticalization and Language Contact, Constructions and Positions 13
Walter Bisang
Grammaticalization and clause linkage strategies: a typological
approach with particular reference to Ancient Greek 59
Sonia Cristofaro
Some Remarks on Analogy, Reanalysis and Grammaticalization 89
Livio Gaeta
Testing the Boundaries of Grammaticalization 107
Anna Giacalone Ramat
Discourse and Pragmatic Conditions of Grammaticalization. Spatial
deixis and locative configurations in the personal pronoun system
of some Italian dialectal areas 129
Stefania Giannini
The Paradigm at the End of the Universe 147
Paul Hopper
At the Boundaries of Grammaticalization: What Interrogatives Are
Doing in Concessive Conditionals 159
Torsten Leuschner
The Grammaticalization of the Left Sentence Boundary in Hittite 189
Silvia Luraghi
On the Relationships Between Grammaticalization and Lexicalization 211
Juan C. Moreno Cabrera
Structural Scope Expansion and Grammaticalization 229
Whitney Tabor and Elizabeth Closs Traugott
vi Table of Contents

On the Application of the Notion of Grammaticalization to West

African Pidgin English 273
Barbara Turchetta
Language Index 289
Name Index 291
Subject Index 297
 Introduction

Anna Giacalone Ramat and Paul Hopper

University of Pavia and University of Pittsburgh

The study of grammaticalization appears to be entering a new and significant

stage. The pioneers in the field, starting with Meillet, developed the idea of
grammaticalization as a tool of historical linguistics in order to give an
account of the origins and typical changes in grammatical morphemes that
would complement the rich field of etymology and word history. For Meillet,
grammaticalization was a concept that was needed because analogy alone
was inadequate to explain the sources of grammatical morphemes. A search
for the ultimate origins of grammatical morphemes led to the insight that their
source was in the lexicon, through a process of weakening and generalization
of meaning.
With the renewal of interest in grammaticalization that began in the
1980s, interest was naturally focused on expanding the encyclopedic knowl­
edge of grammaticalization by studying its manifestations in a wide variety of
individual languages and typologically by reference to larger groups of
languages. In the course of this work, more and more examples were un­
earthed of borderline phenomena, which as historical processes seemed to
share much in common with the classical type of grammaticalization and yet
lacked some perceived crucial component. Researchers reacted differently to
these phenomena, sometimes, implicitly or explicitly, broadening the notion
of grammaticalization so as to include them, sometimes closing off inquiry at
the point where the term grammaticalization no longer seemed applicable to
the enterprise.
In the third stage that we perceive, there is a growing reflexive interest in
integrating grammaticalization with theoretical work in descriptive and his­
torical linguistics. This interest has naturally focused in part on the question of
the appropriate use of the term grammaticalization itself. Which phenomena
2 Anna Giacalone Ramat and Paul Hopper

are appropriately dealt with under the rubric of grammaticalization? Which

are to be excluded, and on what grounds? As the amount of research in
grammaticalization accumulated, the question of the limits of grammatical­
ization became increasingly urgent. Reviews of current work increasingly
referred to the problem, and one of the most repetitive criticisms of published
work involved the question of whether this or that feature could properly be
referred to as grammaticalization, or should not instead be called something
else, such as morphologization or lexicalization.
A typical problem, and one that was widely noted, was that of lexicaliza­
tion. The source of grammatical morphemes was in the lexicon. Yet the
lexicon itself was susceptible of explanation along lines very similar to those
of grammatical morphemes. One class of examples concerns the fusion of the
parts of a compound noun into stem+suffix, as in English childhood, kingdom.
Former nouns "hood" and "dom" with independent lexical meanings {hood <
"kind, quality", dom < "condition, state, domain") were compounded with
more specific nouns with combined meanings "state of', "condition of',
"domain of'. Here the change of function from noun to derivational suffix
was accompanied by a generalization of meaning and loss of lexical au­
tonomy that was no different from that undergone during the change from
lexical item to grammatical morpheme. An example of a slightly different
type is discussed by Moreno Cabrera in his paper. Certain Spanish adjectives
in -nte, themselves derived from Latin present participles, come to be used as
nouns, and sometimes the source adjective is no longer used, even though the
verb of the original participle is still current. An example is calmante "seda­
tive", from calmar "to soothe". Since the immediate origin in an adjective is
no longer evident, such forms are completely autonomous nouns. We appear
then to have examples of lexical items which are the outcome of a process
(the formation of a present participle from a verb stem and a grammatical
suffix) which is unambiguously grammatical. Evidently the line between
strictly grammatical and strictly lexical processes is a blurred one.
Another area of uncertainty in the placement of conceptual boundaries to
grammaticalization lies in collocations. In Italian the verbs venire and andaré
have come to be used as auxiliaries with gerundal verbs, as in andar dicendo
"to keep on saying", venir dicendo "idem." In the 14th century and for some
time after that, a larger number of verbs could be collocated with andaré and
venire (Brianti 1992). However, in modern Italian the combination andar el
venire + verb is restricted to a handful of verbs: dire "to say", ripetere "to
Introduction 3

repeat", aumentare "to grow", peggiorare "to get worse", consolidare "to
become solid", an the like (Giacalone Ramat 1995). Changes of this kind
represent an apparent countertendency to the general direction of grammati-
calization, for in grammaticalization one expects rather to find an expansion
and loss of constraint in the environments of a form (Lehmann 1985); for
example, modal auxiliaries typically go from requiring human subjects (the
king will..., etc.) to permitting all kinds of subjects (the weather will...), and
many other examples. In the Italian example, however, it seems that the
choice of contexts for the auxiliaries andare and venire has narrowed to the
point where they form a small, closed class of fixed lexical patterns. How are
such cases of emergent "collocation" to be handled within the general theory
of grammaticalization?
The phonological end of the linguistic continuum is also involved in the
question of boundaries to grammaticalization. Often the visible outcome of
linguistic change is one or more meaningless phonological segments whose
function is no longer a grammatical or semantic one, but operates purely on
the phonological level. Thus Dixon (1977) showed how Olgolo syllable
structure, which had been reduced through the loss of word-initial conso­
nants, was being "repaired" by new word initial consonants supplied by
degrammaticalized word class prefixes. Grammaticalization in this instance
presumably went through the sequence identified by Greenberg (1991) for
demonstratives in general, starting from demonstratives, passing through
prefixed articles, and eventually becoming word class markers. In the final
stage of this process described by Dixon, these word class markers, drained of
their original function, serve primarily to restore the natural phonotactics that
had been eroded by the loss of word initial consonants. A remnant of the
original situation is left only in some tendencies for certain consonants to be
associated with very broad lexico-semantic classes. Many words can be
shown to owe part of their phonological substance to earlier morphemes that
have lost their functions. In the German past participle gegessen "eaten", for
example, the middle -g- is historically the ge- of the past participle (MHG
gessen) whose presence idiosyncratically facilitates the prefixing of a new ge.
An extreme formulation to which all borderline cases might lead is that
ultimately grammaticalization is not separately definable from the concept of
change in general. Such a position has in fact been claimed by Hopper (1991),
who has noted that in cases like English miss, Mrs., mistress the semantic and
phonological changes involved are identical with those characteristic of
4 Anna Giacalone Ramat and Paul Hopper

grammaticalization. Speculation of this kind should warn us that our defini­

tion of grammaticalization can only be as trustworthy as our apprehension of
the notion of grammar. But the study of grammaticalization itself has revealed
no clear boundaries to the concept of grammar. Rather, grammatical phenom­
ena have fuzzy borders and are always changing. Grammar itself is the
product of linguists' reflection on language, and thus represents a consensus
about appropriate objects of inquiry rather a sharply delineated, definable
entity. It is not surprising that there should be disagreement as to where to
draw the lines between various pairs of linguistic notions such as morphologi-
zation and lexicalization, grammaticalization and morphologization, and so
on.
With this general caveat, we may still speculate on whether there are
cases of language change that are not part of the phenomenon of grammatical­
ization, and what form such changes might take. They would seem to fall into
two types. In one of them, change is clearly insulated from anything that, by
common assent, would be called grammar. In the second type, change does
not follow any particular direction, but is instead lateral, a case of conversion
among lexical categories of equal rank signaled by simple derivation. Some
such cases might include the following:
(1) Some phonological changes seem to be isolated from other levels of the
language. Among these are low level phonetic changes and changes that do
not immediately contribute to some kind of higher level restructuring of the
kind Gaeta discusses in his paper in the present volume.
(2) Lexical replacements of various kinds may occur without consequences
for the grammar. The replacement of hound by dog, for example, substitutes a
type of dog for a designation of the species as a whole. Yet such idiosyncratic
changes involving single lexical items merge with more problematic ones
when more than one item is involved. For example, the count nouns cherry
and pea replace earlier mass nouns cerys and poise (cf. pease), a change
which might seem to be without influence on grammar. The difficulty of
designating such changes as insulated from grammar is suggested by the fact
that with cerys and poise a subcategory of lexical items referring to a mass has
shifted to one referring to discrete, individual items (no doubt reflecting
changing culinary practices), with an accompanying reanalysis of the final
phonological -s as a plural suffix. These changes might then be seen as well
within the range of grammaticalization phenomena, but confined to a subclass
of two items.
Introduction 5

The pejorative kind of change involved in Latin domina "mistress of a

household" > donna "woman", MHG vrouwe "high ranking lady" > NHG
Frau "woman", or ameliorative changes like OE cwên "woman" (cf. Greek
gune) > English queen appear unlikely to directly acquire grammatical sig­
nificance. In the case of metonymic changes like Latin coxa "hip" > Italian
coscia, "thigh", on the other hand, although specific instances may not be
grammatically relevant, the type of change as a whole is often invoked as
characteristic of grammaticalization: a lexical and a grammatical form can be
said to be metonymically related, in that they share closely similar contexts.
In such cases, the boundary between the lexical and the grammatical may be
especially difficult to trace. For example in English be going to in the sense of
be going in order to is reinterpreted as a future tense be about to. The
emergent English auxiliary manage is another example of such metonymy,
the meaning change being from "organize, be in control of' > "bring about
successfully" > "succeed in, be able to" plus dependent verb, as in the
following:
(a) Jane managed the department >
(b) Jane managed the transfer of funds without difficulty >
(c) Jane managed to save the firm from bankruptcy.
In this and similar examples, the type of change — metonymy — is the same
as for purely lexical changes such as coxa > coscia, but the consequences are
grammatical as well as lexical.
(3) Simple lexical conversions between the two major categories noun and
verb, in which additional morphology is not involved, appear to be separate
from canonically understood shifts from lexical to grammatical category, or
from a major to a minor category. Such 'lateral' conversions between equally
ranked categories are to be kept distinct from derivational processes involv­
ing morphology and from the grammaticalization of nouns, verbs, and ad­
verbs as prepositions, which can more easily assume a grammatical role. Thus
simple changes of the kind involved in English bottle (n.) vs. (to) bottle (v.) do
not involve a shift from less to more grammatical. Another set of examples is
the metonymic substitution of an adjective for the noun in a noun+adjective
collocation, of which there are numerous examples. In one of them, Late
Latin caseum formaticum "shaped cheese" becomes Old French formage,
whence Italian formaggio, Modern French fromage. In another, Latin poreus
singularis "wild pig" becomes French sanglier "wild boar" (Palmer
6 Anna Giacalone Ramat and Paul Hopper

1972:329). On the other hand, the status of category-to-category derivations,

and of compounds, in other words of lexicalization in general, is more
controversial. For example, the creation of a new suffix from Latin mente
"mind (abl.)" for forming adverbs out of adjectives {clara mente "clearly",
dulce mente "softly", etc.) (see Giacalone Ramat, this volume, and Hopper
and Traugott 1993:131) shows many of the same characteristics as grammati-
calization (such as semantic change and morphological fusion, categorial
reduction from noun to suffix), and at the same time is a word formation
process.
Somewhat more obviously, categorial downshifting to prepositions, con­
junctions, and complementizers is typically regarded as grammaticalization.
Thus English beside (prep.) vs. side (n.), Italian durante "during" (prep.) vs.
durare "to last" (v.) could be said to involve a shift in the direction of a more
grammatical category (for example, prepositions are a closed class, typically
assume case functions, become proclitic to NPs, participate in the formation
of paradigms, and so on). Such 'movement down the cline' between lexical
items and purely grammatical markers does not appear to be playing a role in
simple lexical conversions of the "bottle (n.) > bottle (v.)" kind mentioned. A
more general way of putting this might be to say that in the "bottle" example
nothing new is being done but to provide more opportunities for an already
well established grammatical process (the formation of different tenses and
aspects of the verb) to be applied.
Unidirectionality in changes from lexical to grammatical categories is a
strong hypothesis about processes of language change identified as instances
of grammaticalization. Recently, however, doubts have been cast on uni­
directionality as criterial for grammaticalization (see for instance some of the
papers in Traugott and Heine 1991). Tabor and Traugott (this volume) sys­
tematically investigate grammaticalization at the syntactic level with the aim
of verifying the hypothesis of structural scope reduction as a parameter which
correlates to grammaticalization (Lehmann 1995 [1982]). They point to cases
of structural scope increase and conclude that the claims of structural uni­
directionality that have previously been proposed need careful reformulation.
Also the issue of 'functional renewal' has been addressed as potential
challenge to the hypothesis of unidirectionality. Functional renewal (or
'exaptation': Lass 1990) illustrates continuity in form and discontinuity in
function. It calls into question Meillet's view that grammaticalization is
mostly renewal of grammatical forms by showing that function may change
Introduction 1

while form is more resistant to change. Moreover, cases of functional renewal

do represent a challenge for the unidirectionality hypothesis because develop­
ment goes along unexpected paths, jumping from one category to another,
from gender to number (Giacalone Ramat, this volume).
Pidgins and creoles too represent a challenge for unidirectionality: if we
assume that the process of grammaticalization is a language internal develop­
ment, we are faced with the impact of substrate languages, which in the
history of creoles represent an external factor and a source of changes which
may interfere with normal 'natural' changes. Keesing (1991:335) has pointed
to developments in Solomons Pidgin which show a reversal of the metaphori­
cal path whereby spatial meaning becomes extended to temporal meaning
(fastaem from English first time, developed to a spatial mark "in front o f ) ,
and has suggested that such a development was made possible by the already-
existing similar connection in substrate languages. In the same vein Bruyn
(1996) discusses complex prepositional phrases in Sranan, whose patterns
caique similar constructions of West African languages.l
One might speculate that cases of 'apparent' grammaticalization are
more frequent in the history of languages. When no historical evidence is
available, they can be reconstructed, as suggested by Bisang (this volume).
Bisang maintains that the existence of linguistic areas (Sprachbünde) may be
due to the cross-linguistic spreading of mechanisms of grammaticalization, in
particular the introduction of new types of reanalysis in language contact
situations in which speakers/hearers may transfer mechanisms of reanalysis
existing in his/her own language to other languages.
Even though the (relatively) few examples going in the opposite direc­
tion of change should warn us against making a strong claim on unidirection­
ality, we may still expect that certain narrowly defined types of changes are
unidirectional, as suggested by Harris and Campbell (1995:339ff).
In this introduction it has not been our purpose to suggest solutions to any
of the controversies surrounding the various approaches and concepts of
grammaticalization. Instead we have intended to clarify the terms of the
discussion with a view to encouraging further debate.
Several possibilities for approaching the problem of boundaries appear
to exist. The currently dominant one is perhaps to study different grammatical
phenomena with a view to arriving at definitions that will sharply distinguish
different aspects such as morphologization, lexicalization, phonogenesis, etc.
A second one would be to likewise accept the existence of different
8 Anna Giacalone Ramat and Paul Hopper

aspects, but to have prototypical (exemplary) instances determine the core

phenomena and use these as anchor points for exploring the boundary areas.
A final possibility concerning the limits of grammaticalization would be
to understand grammaticalization as a cover term for a wide range of phe­
nomena having common historical processes whose outcomes are grammati­
cal forms. In stating things this way, we need to be aware of two caveats:
(1) As we have noted above, the definition of grammaticalization will never
be separable from that of 'grammar' itself.
(2) The possibility of a close and possibly inextricable relationship between
processes that result in grammar and those that result in lexical items must
never be excluded. The on-going dialectic between grammar and lexicon
cannot be closed off, and we should not allow terminological constraints to
govern our thinking to the point of excluding some higher synthesis of these
two concepts.
Most of the papers in this volume represent a selection of the papers
presented at a Symposium on Grammaticalization held at the XXVIII Annual
Meeting of the Societas Lingüistica Europaea (Leiden, August 1995). One
paper, that by Tabor and Traugott, has been added to this collection since the
Symposium took place.
The contributors to the Symposium approached the question of defining
the limits of grammaticalization in a variety of ways. Bisang emphasizes the
importance of assessing the entire construction, rather than individual words,
in tracing the development of grammar. He shows strikingly how, when this is
done, constructions are often seen to have a long-term durability that survives
the replacement of their component parts by different forms. Nonetheless,
characteristic changes motivated by natural usage and natural cognitive conti­
guities are identifiable. Bisang suggests that we should pay attention to the
interaction between the 'innate', cognitively motivated kinds of change and
sociolinguistic factors such as pragmatics and language contact, which, he
suggests, may override these cognitively driven changes.
Giannini, too, describes the interaction of spatial deixis and everyday
conversational usage in the emergence of complex pronouns in the Luccan
variety of Tuscan. She suggests that in a close-knit traditional society pro­
nouns might be especially liable to be supplemented or even replaced by
spatial deictic markers that refer to the agreed upon typical location of the
referent.
Introduction 9

Gaeta expands the notion of grammaticalization to include problems of

phonological and morphological change, with particular attention to German.
He finds a fundamental theoretical difference between the development that
started as Umlaut and that which began as so-called "Rückumlaut", and
suggests that one of these developments, but not the other, can be included in
the domain of grammaticalization.
Using data from Ancient Greek, Cristofaro evaluates the synchronic
distribution and the diachronic evolution of complementizers out of relational
items such as pronouns. The process is widely attested crosslinguistically,
however Cristofaro's purpose is to show that such cases are not only clear
instances of grammaticalization in the discourse-based sense, proceeding
from pragmatics to syntax: they also are legitimately assimilated to more
traditional lexical/sentence-based instances of grammaticalization, being
governed by the same principles of development from more concrete to more
abstract, and of bleaching of lexical meaning and corresponding strengthen­
ing of grammatical meaning.
Luraghi surveys that happy hunting ground of Wackernagel's Law,
Hittite, for insights into how clitic particles come into being and develop. The
historical development of the second position clitics, she argues, follows
some well-known principles such as increase in frequency of a small number
of particles, and a semantic change from a local, more concrete meaning to a
more general, abstract, aspectual meaning; but a by-product of the changes
turns out to be a new function altogether: that of defining the left-most
boundary of the clause.
Moreno Cabrera aims to sort out the grid of grammaticalization/lexical-
ization, metaphor/metonymy by studying the shift from present participle to
noun in Spanish, Basque, and Hungarian. He concludes that grammaticaliza­
tion and lexicalization processes are two complementary aspects of language
evolution, the one involving metaphorical abstraction processes, the other
metonymical concretion processes.
Hopper construes the problem of the limits of grammaticalization in
terms of the starting points ('incipient grammaticalization') and vanishing
points ('dissipating grammaticalization') of changes. Taking up a theme of
Benveniste, he notes that grammaticalization may often be seen to have a
fairly specific source in a particular kind of discourse. In its end stages, the
dissipation of changes is less determinate, but one of its outcomes is simply
meaningless phonological segments.
10 Anna Giacalone Ramat and Paul Hopper

The paper by Giacalone Ramat also addresses the general issue of this
volume, namely the range and scope of inquiry of grammaticalization, by
exploring a number of critical cases of renewal of function. Such examples
show reassignment of a new functional value to grammatical morphemes
deprived of their previous function along lines which fall outside the expecta­
tions of canonical grammaticalization processes. Giacalone Ramat's concern
is also with the unidirectionality hypothesis, which — she claims — is a
strong constraint on possible language change. Despite some undeniable
counter-examples, the claim is worth of consideration as one major tenet of
grammaticalization theory (see also Moreno Cabrera for similar conclusions).
Leuschner addresses the issue of clause linkage strategies as proceeding
from pragmatics to syntax and from less to more tightly integrated structures,
along the lines suggested by Givón. He argues that the interrogative form
of many concessive-conditional clauses originates in rhetorical dialogues
through a process of grammaticalization of a question-answer sequence.
Interestingly, Leuschner also points out that some already grammaticalized
uses of concessive conditional expressions like German wie auch immer may
receive additional pragmatic functions and allude at a process of 'pragmati-
calization', again a challenge to unidirectionality.
Tabor and Traugott, too, raise the issue of unidirectionality and most
especially 'scope reduction' in lexical-to-grammatical change. In order to
empirically test the claim, they explore a number of changes and demonstrate
for them an 'increase in structural scope', not decrease, as was suggested in
the literature up to now. In spite of the encouraging results of their analyses,
they conclude that too little systematic exploration has been made to draw
generalizations on the unidirectionality claim.
The role of linguistic and extralinguistic context in enforcing processes
of grammaticalization in language contact also figures prominently in
Turchetta's discussion of western varieties of WAPE (West African Pidgin
English) spoken in Cameroon, Nigeria and Ghana. Turchetta argues that
pragmatic analysis of oral discourse may illustrate the origin of grammatical­
ization processes, which generally do not follow in pidgins the same steps as
in languages with a long history. Mechanisms of change are more fluid and
not unidirectional; they include functional splits, as in the case of the locative
marker de, and semantic extension of lexemes which behave differently from
chains of grammaticalization.
Introduction 11

REFERENCES
Brianti, Giovanna. 1992. Périphrases aspectuelles de l'italien. Le cas de andare, venire et
stare + gérondif. Berne: Peter Lang.
Bruyn, Adrienne. 1996. "On identifying instances of grammaticalization in Creole lan­
guages." In Philip Baker and Anand Syea (eds), Changing Meanings, Changing Func­
tions. Papers relating to grammaticalization in contact languages. London: University
of Westminster Press, 29-46.
Dixon, R.M.W. 1982[1969]. "Olgolo syllable structure and what they are doing about it."
In Where Have all the Adjectives Gone?, 207-210. Berlin: Mouton Publishers. (First
published in Linguistic Inquiry 1.273-276,1969).
Giacalone Ramat, Anna. 1995. "Sulla grammaticalizzazione di verbi di movimento:
andare e venire + gerundio," Archivio Glottologico Italiano LXXX: 168-203.
Greenberg, Joseph. 1991. "The last stages of grammatical elements; contractive and
expansive desemanticization." In Elizabeth C. Traugott and Bernd Heine (eds), I, 301-
314.
Harris, Alice and Lyle Campbell. 1995. Historical syntax in cross-linguistic perspective.
Cambridge: Cambridge University Press (Cambridge Studies in Linguistics 74).
Hopper, Paul. 1991. "On some principles of grammaticalization." In Elizabeth C. Traugott
and Bernd Heine (eds), I, 17-35.
Keesing, Roger M. 1991. "Substrates, calquing and grammaticalization in Melanesian
pidgin." In Elizabeth C. Traugott and Bernd Heine (eds), I, 315-342.
Lass, Roger. 1990. "How to do things with junk: exaptation in language evolution,"
Journal of Linguistics 26: 79-102.
Lehmann, Christian. 1995[1982]. Thoughts on Grammaticalization. Munich: LINCOM
EUROPA (originally published as Thoughts on Grammaticalization: A programmatic
Sketch. Arbeiten des Kölner Universalienprojekts 49).
Palmer, Leonard R.1972. Descriptive and Comparative Linguistics. A Critical Introduc­
tion. London: Faber & Faber.
Traugott, Elizabeth C. and Bernd Heine (eds). 1991. Approaches to grammaticalization.
Amsterdam/Philadelphia: John Benjamins.
Grammaticalization and language contact,
constructions and positions

Walter Bisang
Johannes Gutenberg-Universität Mainz

1. Introduction

The basic idea do be presented in this paper will be outlined in the first two
paragraphs. I would like to point out that what I will present is a first
hypothesis which I hope is stimulating enough to instigate further discussions.
Grammaticalization is a phenomenon of language change. According to
functionalist theories, language change is motivated by pragmatic and socio-
linguistic factors (the third factor of language acquisition will not be dis­
cussed). Constructions are of rather central importance for both factors. In
pragmatics, constructions often provide the basic patterns for processes of
reanalysis and analogy. The sociolinguistic factor of language contact en­
forces processes of grammaticalization because it supports the exchange of
mechanisms such as reanalysis and analogy which both can be determined by
constructions. Thus, language contact may support the exchange of construc­
tions and thereby contribute to the emergence of linguistic areas {Sprach­
bünde). Constructions contain one or more positions. These positions are of
crucial importance for processes of grammaticalization. Paradigmatically,
they form the slot or the slots in which certain lexical items can be grammati-
calized. Syntagmatically, they are determined by the construction to which
they belong (for further discussion cf. my explanations on attractor positions
in §3).
Hopper & Traugott (1993) and Croft (1996) both point out that language
change is divided into two parts. There is change as it actually happens in a
given utterance and there is the propagation of that change. Change as such is
14 Walter Bisang

a stochastic process, whereas its propagation, i.e. successful language

change, is a sociolinguistic process (Croft 1996). Although I do not deny the
primary importance of sociolinguistic processes (cf. below), I would like to
argue that the selection of what changes are going to be propagated also must
depend to some degree on other factors. No matter how stochastic changes as
such may be, they are exposed to hearers who will apply the whole of their
knowledge of language to make sense out of them. Those changes which
somehow fit into that knowledge (be it e.g. by reanalysis or analogy) are
potential candidates for propagation. This knowledge is constituted by the
human cognitive equipment and by constructions, which can partly be deter­
mined by this equipment. The degree to which changes can be identified by
one or more constructions somehow preselects those changes which are to be
propagated. The relevance of the cognitive equipment can be seen from the
considerable number of regularities discovered in research on grammatical-
ization, i.e. in research on successfully propagated changes. I cannot see how
to account for these regularities without reference to cognition.
As was pointed out by Bybee, Perkins & Pagliuca (1994) arguing against
Heine, Claudi & Hünnemeyer (1991), the source of grammatical meaning is a
whole construction, not the lexical meaning of a stem or maybe a word:
"Constructions involving movement verbs, for example, are found to be the
sources of markers not only of future, but also of pasts and progressives.
Were we to limit our attention to the lexical stem, we would be able to offer
nothing beyond an unenlightening list of the possible grams which verbs
such as 'go' and 'come' could evolve into." (Bybee, Perkins & Pagliuca
1994: 11)
"Thus we disagree with Heine, Claudi, and Hünnemeyer 1991...: 338) that
'one source concept can give rise to more than one grammatical category.'
It is the entire construction, and not simply the lexical meaning of the stem,
which is the precursor, and hence the source, of the grammatical meaning."
(Bybee, Perkins & Pagliuca 1994: 11)

In this sense, constructions operate as frameworks or as contexts within which

some of its elements are grammaticalized. On the other hand, constructions
with their independent meaning themselves can also become the result of
processes of grammaticalization.
In my definition of constructions, I shall basically follow that of Con­
struction Grammar (cf. e.g. Goldberg 1995, ch. 2) according to which con­
structions have their own independent meaning. The following quotation
from Michaelis & Lambrecht (1996) illustrates how this works with the
Grammaticalization and language contact 15

construction of Nominal Extraposition (NE):

"While syntax has traditionally been restricted to the role of providing the
instructions by which the lexical contents are assembled, syntax in the CG
(Construction Grammar, W.B.) paradigm is capable of contributing concep­
tual content on its own ... In the case of NE construction in particular, the
external semantics of the construction does not merely represent the projec­
tion of the semantics of the main predicator (be amazing, etc.), but also the
fact that this construction is interpreted as a degree exclamation. The fact
that sentences like It's amazing the people he knows receive a scalar
interpretation is not a fact about the predicator be amazing, but about the
formal configuration at issue." (Michaelis & Lambrecht 1996: 226)

The semantic independence of constructions allows them to be organized in

semantic networks and in hierarchies of inheritance. It also makes them
amenable to grammaticalization.1
In my above outline, I pointed out that there are three generally discussed
motivating factors of language change which may also apply to grammatical­
ization (cf. Hopper & Traugott 1993: 63 - 93):
1. Language acquisition
2. Pragmatics (cognitive strategies which lead to the semantic change of
a linguistic sign; the role of speaker and hearer)
3. Sociolinguistics (Linguistic communities and different types of con­
tact)
The problems with the adaptability of models of language acquisition devel­
oped within the framework of formal linguistics (Andersen 1973, Lightfoot
1989) to the theory of grammaticalization are described by Hopper & Traugott
(1993: 33 - 38). In the present paper, language acquisition will not be discussed.
I shall concentrate my argumentation on pragmatic and sociolinguistic motiva­
tions of language change and their consequences for grammaticalization. The
pragmatic motivation is extensively discussed in grammaticalization theory,
while the other functionally motivated factor of language contact and its impact
on processes of grammaticalization is rather neglected, although grammatical­
ization is impossible if the pragmatic motivations enhancing patterns of
reanalysis and analogy in individuals are not spread to other individuals
speaking the same language or even speaking another language in the case of
language contact.
Pragmatics is generally discussed as the most important motivation of
grammaticalization. It can be summarized as follows (Hopper & Traugott
16 Walter Bisang

1993: 63 - 93). Grammaticalization starts from the semantic change of a sign

and the cognitive strategies by which it is caused, i.e. by metonymic inference
and by metaphoric inference which are both of pragmatic nature. Metonymy
is the product of conversational inferencing and is linked to reanalysis. For
that reason, it operates on the syntagmatic level and is based on abduction.
Metaphor is the product of conventional inferencing and is linked to analogy.
It operates on the paradigmatic level. The mechanism of reanalysis seems to
be more important to grammaticalization than analogy and it occurs first. The
following table presents a summary of the most important terms linked with
pragmatic motivations of grammaticalization:

Table 1. Terms linked with pragmatic motivations of grammaticalization

Metonymy Metaphor
Syntagmatic Level Paradigmatic Level
Reanalysis (Abduction) Analogy
Conversational Implicatures Conventional Implicatures
Operates through Interrelated Syntactic Operates through Conceptual Domains
Constituents

Constructions provide very important patterns of reanalysis. For that reason,

they very often form the basis of metonymic inference. Within a given
construction, certain positions can attract further items into a new function by
the mechanism of analogy (cf. §3 on attractor positions).
Situations of language contact generate interferences (Weinreich 1953/
1968)2. Interference among fluent bilingual speakers is connected to Keller's
maxim 'talk like the others talk' (Keller 1994: 100). This maxim is discussed
in two versions in the sociolinguistic literature. In the version of 'talk like the
people around you' (Keller 1994: 100; also cf. Trudgill 1986, Giles & Smith
1979), the speaker accommodates her/his speech to that of her/his interlocu­
tors for reasons of solidarity. In the version of 'talk in such a way that you are
recognized as a member of the group you wish to identify with' (LePage &
Tabouret-Keller 1985), the speaker can either accommodate her/his speech to
that of her/his interlocutors if s/he wants to be identified with them, or s/he
can follow the conventions of either his/her own group or a group s/he wishes
to be identified with.
Grammaticalization and language contact 17

Interference yields important consequences for grammaticalization as

far as mechanisms such as reanalysis and analogy are concerned. In a situa­
tion of language contact, the speech act participants have particular problems
in 'getting their point accross'. These problems may enforce processes of
grammaticalization by means of increasing the need of applying such mecha­
nisms as reanalysis, metonymy, and metaphor and by means of spreading
patterns triggering these mechanisms, i.e. constructions, into wider linguistic
areas. The hearer may reanalyse a certain sequence of a foreign language
from the same area with the mechanisms of reanalysis existing in her/his own
language. Thus, new types of reanalysis in the sense of abduction can be
introduced from the point of view of the hearer as soon as he speaks the
foreign language of her/his interlocutor. In contact with a hearer who speaks a
foreign language, the speaker may use special mechanisms of metonymy and
metaphor which s/he usually does not have in her/his language (but of which
she knows through regular contact that they exist in the language of the
hearer) in order to make herself/himself understood. As I argued in Bisang
(1996) on East and Southeast Asian languages, the existence of linguistic
areas {Sprachbünde) may be due to the cross-linguistic spreading of the above
mechanisms of grammaticalization.
According to Hopper & Traugott (1993: 36), "a distinction needs to be
made between change and spread of the change". As the two authors further
explain, "[t]he problem is that 'one swallow doth not a summer make,' and
change in the grammar of an individual does not constitute what we think of
as a change in 'a language'" (Hopper & Traugott 1993: 36). Croft (1996)
describes language change as a two-step process consisting of the innovation/
actuation of a change and its propagation. "Successful propagation of an
innovation constitutes a change in the language" (Croft 1996). According to
Croft, the factors which trigger the actuation of a change are of the functional
and the pragmatic type, whereas the factors relevant for propagation are of the
sociolinguistic type.
As pointed out by Croft (1996) in his usage-based theory of actuation,
innovation only takes place occasionally, it is not omnipresent in language
use. Apparently, the dominant maxims in language use are such static maxims
as Humboldt's maxim 'speak as you think that the other would speak if he
were in your position' 3 , Keller's phatic maxim of 'talk like others talk' (Keller
1994: 100) and the maxim of 'talk in such a way that you are not misunder­
stood' (Keller 1994: ibid). In the vast majority of speech situations the speech
18 Walter Bisang

act participants simply follow the norm without any problems. What triggers
innovation is the "complexity of the task" (Croft 1996). Each utterance in a
concrete situation reflects a highly specific experience. The process of inter­
preting a given syntactic expression in a particular situation of speech by
filtering the intended meaning out of the considerable range of potential
meanings of that syntactic expression is indeed a very complex task which
Croft understands as the origin of actuation. The speaker, who tries to reduce
the potential range of meaning by following the above static maxims, may
occasionally produce innovation just because s/he tries to stick to the norms
as closely as possible. To the hearer just the same may happen when s/he
painstakingly tries to interpret the concrete meaning of a given utterance
according to the norms.
As I argued in my above outline, the constructions already existing in a
language and the human cognitive equipment are also involved in the propa­
gation of change. These factors somehow seem to preselect the stochastic
actuations/innovations to be propagated. I shall first look at constructions,
then at the human cognitive equipment.
As was shown above in quoting Croft (1996), language use is dominated
by static maxims. We may argue that a relatively close set of constructions
may guarantee the observance of these static maxims in a very favorable way
and thus also guarantee successful communication, i.e. expressing (speaker)
or interpreting (hearer) a particular experience in a given situation as quickly
as possible and as adequately as necessary. In this sense, innovations first of
all have to be measured against the power of what is already there before they
can be propagated to a given language community as a whole. If constructions
represent an important part of what is already there, their influence on the
acceptability and propagation of changes tends to avoid dramatic changes
in language systems. This may explain why there seem to be no "radical
changes" in a situation of Spanish-English bilingualism such as the one of Los
Angeles Spanish described by Silva-Corvalán (1994) even if "bilinguals
develop strategies aimed at lightening the cognitive load of having to remem­
ber and use two different linguistic systems" (Silva-Corvalán 1994: 6). The
existence of mixed languages in the sense of Bakker & Mous (1994) can be
seen as another instance in which constructions remain more or less stable.4
According to Bakker & Mous (1994), mixed languages are the result of
language intertwining, which is defined as a combination of the grammatical
system of one language with the lexicon of another language. In spite of these
Grammaticalization and language contact 19

examples, it looks as if the existence of constructions does not preclude the

integration of even widely different structures into a language, if the socio-
linguistic situation allows for such changes (cf. Thomason & Kaufman 1988).
For that reason, sociolinguistic factors seem to have priority over construc­
tions. Under the most extreme situations such as those applying to the context
of Pidgin and Creole languages, it appears to be inevitable to accept some
kind of universal preconditions directing the emergence of these languages.
The human cognitive equipment is manifested by pathways of grammati­
calization and scales of metaphorical categories and their cognitive back­
ground as described by Bybee (1985), Heine, Claudi & Hünnemeyer (1991)
or Hopper & Traugott (1993). In this paper, I would like to argue that aspects
of human cognition are also involved with constructions. As I pointed out at
the beginning of this section, constructions do not only provide the frame­
work within which grammaticalization takes place, they are also subject to
grammaticalization by themselves, i.e. they can change their meaning or they
can change their structure. In comparison to changes taking place within the
framework of a given construction, which can be quite abrupt, changes of
constructions as a whole seem to be rather slow and gradual. This comes as no
surprise if their function is to guarantee the observance of the above static
maximes. As I hope to be able to show in §3 and §4, at least the structural
change of constructions is partly governed by cognition. In my example on
verb serialization (§3), one part of the grammaticalized verbs is due to
Bybee's principles of generality and relevance. Another part of the distribu­
tion of these positions seems to be purely arbitrary. In my example of
classifier constructions (§4), I would like to show that there are two different
processes of classifier development. The item oriented process was involved
from its very beginning with counting, while the category oriented process
was not. Nevertheless, the category oriented process is tightly, but not com-
pulsorily linked to counting and the construction provided for counting. The
reason for the convergence of the category oriented process with the con­
struction for counting is based on cognitive closeness (cf. (61) and table 3)
and on language contact, because the item oriented process seems to be based
in the northern part of the East and Southeast Asian linguistic area, whereas
the category oriented process developed in the southern part of this area.
In the rest of this paper, i.e. in §2 to §4,I shall try to illustrate some of the
above statements with concrete examples mainly from languages of East and
Southeast Asia. In §5, I shall present a short outlook on the question of
20 Walter Bisang

innateness of language structures. To conclude this section, I would like to

add some more general information on §2 and §3. The main line of argumen­
tation adopted in §4 on counting is already presented above.
In §2, I would like to present two examples of the enormous diachronic
stability connected with constructions. This stability can either be reflected
with regard to the shape of the constructions and the positional arrangement
of their dependent syntactic units or with regard to a form-meaning link in the
sense that a given meaning is always expressed by the same construction even
if that construction significantly changes its structure. The first case is illus­
trated by the position of the relative clause and the head noun in Chinese
(§2.1), the second case is illustrated by the Chinese equational construction
and its function in focus marking (and even in topic marking in Classical
Chinese) (§2.2).
As I postulated above, constructions provide the framework within
which combinations of syntactic units and semantic components can be
analysed in a new way which may lead to language change if it is propagated
from a linguistic individual to a language community. Syntagmatic processes
of reanalysis can yield paradigmatic consequences by changing the meaning
of a syntactic unit. This oscillation between syntagmaticity and paradigmatic-
ity within a construction is what is at the origin of serial units and attractor
positions as they will be described in detail in §3. The position of serial verbs
expressing TAM, direction (towards or away from the speaker) and semantic
roles (e.g. benefactive, instrumental, location, goal, etc.) relative to the main
verb does not seem to be totally random. There seem to be cognitive-semantic
rules which allow only a limited degree of distributional variation. This
variation is enough to yield some interesting areal patterns in East and
Southeast Asian languages such as Chinese, Hmong, Vietnamese, Thai, and
Khmer (Bisang 1992, 1996). The fact that similar positional restrictions also
hold in other languages which use serial verbs for all the three of the above
functions (i.e. Jabêm and Yoruba) shows that the pattern observed in East and
Southeast Asian languages is not simply motivated by contact. There seem to
be more general cognitive processes involved which also operate in language
change and its propagation.
Grammaticalization and language contact 21

2. Continuity through time: The position of the relative clause in

Chinese and the Chinese equational construction

2.1. On the positional stability of the relative clause within the relative
construction

In his well-known paper on word-order correlations, Dryer (1992) presents

data on the correlation between the positions of V and O and the positions of
the relative clause and its head noun. He presents his data in the following
table:

Table 2. Order of noun and relative clause (Dryer 1992: 86)

Africa Eurasia SE-Asia & Australia & North South Total

Oceania New Guinea America America

OV & RelN 5 11 2 2 3 3 26

OV & NRel 9 5 2 6 12 3 37

VO & RelN 0 0 1 0 0 0 1

VO & NRel 21 8 12 3 11 5 60

If we look at the number of genera in which VO word order cooccurs with

NPs in which the relative clause is placed in front of the head noun we only
find one genus, whereas the inverse word order of NRel is attested in 60
genera. Thus, there is a strong tendency for VO to cooccur with NRel. In fact,
the noun turns out to be a verb patterner, while the relative clause is an object
patterner (Dryer 1992: 86 - 87).
The exceptional language among the VO languages in which the relative
clause is in front of the head noun is Chinese. Although the cooccurrence of
RelN with VO is rather marked, Chinese belongs to this word order type back
to the oracle-bone inscriptions between the 14th and the 11th century B.C. In
Classical Chinese (cf. footnote 6) as well as in Modern Standard Chinese,
relative clauses are prenominal. Speculations about OV word order in pre-
archaic Chinese as suggested by Li & Thompson (1974) seem to be rather
22 Walter Bisang

unlikely if one looks at grammatical descriptions of the oracle-bone inscrip­

tions such as the one by Kryukov (1980: 66 - 71). The hypothesis of word
order change from SVO to SOV in Modern Standard Chinese on the basis of
the ba construction (Li & Thompson 1974, 1975) is rather problematic (cf.
e.g. Wang Mingquan 1987) and is not confirmed by statistics based on texts
(Sun & Givón 1985). Thus, we can simply observe that the typologically
rather unlikely word order correlation of VO and RelN exists with remarkable
stability throughout the history of Chinese. The relative construction as a
whole with its RelN word order seems to be strong enough against any
change. This may be due to the fact that the attributive marker used for
relative clauses (Classical Chinese zhï, Modern Standard Chinese de) is also
used with other modifiers. Moreover, RelN word order is highly consistent
inasmuch as any kind of modifier is prenommai in Chinese, even though this
consistency does not improve the typological compatibility with VO word
order. There seem to be instances in which word order stability of a given
construction predominates typological conformity with V and O word order.

2.2. The equational construction in Chinese: Functional stability vs.

change of surface form5

The equational constructions of Classical Chinese and Modern Standard

Chinese of the type N1 is N2 differ considerably. In Modern Standard Chinese,
the first noun is followed by the copula shî 'to be' which is in turn followed by
the second noun (cf. (1) and (3)). In Classical Chinese, the equational marker
yě occurs at the end of the clause after N1 and N 2 (cf. (2) and (4)):
(1) Modern Standard Chinese
N1 shî N 2 .
(2) Classical Chinese
N1 N2 yě.
(3) Modern Standard Chinese
jïngyü shî bürü dôngwù.
whale be mammal
'Whales are mammals.'
Grammaticalization and language contact 23

(4) Classical Chinese (Meng 4.B.28; 5th to 4th century B.C.)

Shùn rén yë.
Shun man EQ
'Shun is aman.'
In spite of their difference in surface structure, both constructions are used to
mark focus. The construction of Classical Chinese is even additionally em­
ployed in the context of topic and of clause combining. Both constructions
will first be presented with their specific discourse function. After this presen­
tation, I shall briefly discuss some processes of grammaticalization which
lead to the equational construction as we find it in Modern Standard Chinese.
In Modern Standard Chinese, the functional range of the copula shî goes
far beyond the mere identification of one noun through another. This is
pointed out in the following quotation from Chao (1968):
"Predication with a more loose relation between subject and object than that
of equality or subsumption is more common in Chinese than in English."
(Chao 1968: 718)

Thus, we find the copula with expressions of space or of time (for the latter cf.
(5)) or with what is called illogical copula by Hashimoto (1969). In example
(6), the copula is used 'illogically ' — it refers to the context of ordering a meal
in a restaurant:
(5) tarnen huí guó de rîqï döu ding-le,
they return country/nation ATTR date all fix-PF
lăo Zhang shî mingtiän, lăo LÎ shî hòutiân.
old Zhang be tomorrow old Li be day_after_tomorrow
T h e date of their return to the country is [already] fixed — for
Zhang it will be tomorrow, for Li it will be the day after tomorrow.'
(6) wo s hi jïfón.
I be chicken_rice
'For me it is chicken rice.' [I have ordered chicken rice.]
(Hashimoto 1968: 86)
The relatively broad functional range of the copula shî may be one of the
reasons why it is generally used in the context of focus. In this context, the
copula often cooccurs with the attributive marker de. One of the functions of
de is to mark relative clauses. If it occurs without a head noun, it can be
interpreted as some kind of nominalizer which allows the nominalization of
24 Walter Bisang

entire states of affairs. As was pointed out by Ross (1983) and Iljic (1987b),
there are three different types of shî ...de constructions which used to be
treated separately, although there is a clear functional link between them. I
shall first present a list of these three types of constructions before each of
them will briefly be described in some more detail:
1. the equational shL.de construction (Chao 1968: 718: equality and
sub sumption', Ross 1983: equational sentences', Iljic 1987b: équatif)
2. the constituent specific shL.de construction (Chao 1968: 296: de
for specification and 1968: 719: nominalizing specifier; Ross 1983:
cleft sentences; Iljic 1987b: [construction] à élément focalisé)
3. the situational shL.de construction (Chao 1968: 296: situational
de; Ross 1983: cleft sentences; Iljic 1987b: [construction]
situationnelle)
In the equational shî ...de construction, the noun in the position of N 2 is a
state of affairs nominalized by de. Apart from that, it does not differ from a
normal equational sentence. The equational shî ...de construction forms the
basis from which the other two focus constructions are developed. In the
position of N1 we find the item to be identified, in the position of N 2 we find
the identifying item:
(7) wo dà gē shî chàng jïngjù lăo
I big elder_brother be sing Beijing_opera old
shëng de.
actor_in_the_role_of_a_man ATTR
'My brother is the one who plays the role of the old man in the
Beijing opera.' (Iljic 1987b)
The constituent specific shî ...de construction is used to focalize an item
which occurs in front of the verb or the verb itself. In this construction, the
focalized item immediately follows the copula shî. The following example by
Iljic (1987b: 132 - 133) illustrates the different possibilities of focalizing an
item in front of the verb or the verb itself:
(8) a. shî bóbó zuótian lái jiē n1
be elder_brother_of_father yesterday come collect you
de.
ATTR
Tt was [your] uncle who came to collect you yesterday.'
Grammaticalization and language contact 25

b. bóbó shi zuótiān lái jië n1

elder_brother_of_father be yesterday corne collect you
de.
ATTR
'It was yesterday that [your] uncle came to collect you.'
c. bóbó zuótian shĺ lái jiē nĭ
elder_brother_of_father yesterday be come collect you
de.
ATTR
'[Your] uncle came to collect you [not e.g. to greet you]
yesterday.'
If the object, i.e. a postverbal element, is to be focalized, Chinese takes
recourse to another construction which also consists of the copula shì and the
attributive marker de. In this construction (cf. (9)), the whole clause except the
object is nominalized by de, the object still occurs in clause final position after
the copula:0
(9) N 1 V de shì N 2
(10) bóbó zuótiān lái jië de shï
elder_brother_of_father yesterday come collect ATTR be
nĭ.
you
'It was you whom [your] uncle came to collect yesterday.' (Iljic
1987b:134)
The function of the situational shì ...de construction is best described by
quoting a very short and concise passage from Chao (1968):
"Sometimes, the de refers to the whole situation with the meaning of 'such is
the case', 'this is the kind of situation' and no particular noun is understood or
can be supplied rather than some other noun."

The following example of a situational shì ...de construction cannot be used as

an answer to the question 'What happened?'. It is used in a situation in which
the speaker wants to make a joke but is not understood by the hearer. In such a
context, the speaker may say:
26 Walter Bisang

(11) wŏ (shì) gēn nĭ kāiwánxiào de.

I be with you make_a_joke ATTR
'I was just joking with you that-was-what-I-was-doing.'
(Chao 1968: 296; Li & Thompson 1981: 591 - 592)
In Classical Chinese, the function of the equational construction is paral­
lel to the constituent specific construction or to the situational construction of
Modern Standard Chinese. However, the equational construction of Classical
Chinese is not only employed to mark focus, but also to mark topic and clause
combining.
To start my discussion of Classical Chinese, I would like to look at the
insertion of headless relative clauses into the equational construction. As we
can see from the following two examples, headless relative clauses can occur
in the position of N1 as well as of N2.6
(12) cĭ [suŏ wèi yăng kŏu tĭ zhë] ye.
this REL call nourish mouth body NOM EQ
This is what is called 'nourishing the mouth and body'. (Meng
4.A.19)
(13) [wú dì yú tiān-xià zhë] tiān lï yě
not_have enemy in world NOM] heaven minister EQ
'He who has no enemy in the world is the minister of heaven.' /
'To have no enemy in the world is to be the minister of heaven.'
(Meng 2.AÍ5)
The next construction to look at is the constituent specific construction. If
the constituent of particular interest is a subject, this subject is separated from
the rest of the clause by the equational marker yě. In example (14), the NP
marked by yě is Qiu, the topic of the clause. In the next example, the
equational marker yě is used no less then four times. In its first use, it
introduces the topics of the sentence, i.e. Zi-gong and Hui. The second yě
further reduces the topic to Zi-gong. The last two instances of yë are used in
the sense of contrastive focus (cf. Dik 1989).
(14) Qiú yë wéi Jï shî zăi.
Qiu EQ make/be Ji Family chief_officer
'As for Qiu, he acted as the chief officer of the Ji family.' (Meng
4.A.14) (also cf. Gassmann 1980: 42)
Grammaticalization and language contact 27

(15) zĭ wèi Zï-gông yuë: "rü yü Huí ye shú

master say/talk_to Zi-gong QUOT you and Hui EQ who
yu?" - duì yuë: "Cì ye hé
better answer QUOT I(i.e. the_speakers_first_name) EQ how
găn wàng Huí? Huí yě wén yï yí
dare look_at/compare Hui Hui EQ hear one in_order_to
zhï shì, Cì yë wén yï yï zhï èr. "
know ten I(i.e. Ci) EQ hear one in_order_to know two
'The master [i.e. Confucius, W.B.] said to Zi-gong: "You and Hui,
who is better?" — [Zi-gong] replied: "As for me, how dare I to
compare [myself with] Hui. Hui has but to hear one part in ten, in
order to understand the whole ten. I understand no more than two
parts if I hear one part.' (Lunyu 5.8)
If the constituent to be pointed out is the predicate, Classical Chinese
employs a special construction which is also based on the equational marker
yě. The use of this construction seems to be limited to focalization. As we can
see from the following example, a verbless complement clause is put in front
of the verb whereby the zero position of the complement clause refers to that
verb. In this construction, the complement clause ends with the equational
marker yě.

(16) [jün-zĭ zhĭ 0 yú qín shôu yë] jiàn qí

superior_man ATTR PREP biped quadruped EQ see POSS:3
shëng bit rën jiàn qí sí.
live NEG bear see P0ss:3 die
'What the superior man does with the animals [lit.: bipeds and
quadrupeds] is, that having seen [them] alive he [cannot] bear to
see them dead.' (Meng l.A.7,8)
If the constituent to be specially referred to is an object, there are two
constructions, one construction for focalization, the other construction for
topicalization. I shall only present the former which is analogous to the
construction presented in (10) from Modern Standard Chinese. Parallel to the
construction of Modern Standard Chinese, we find a headless relative clause
in the position of N1 and the object NP in N2. The whole construction is
concluded by the equational marker yě:
28 Walter Bisang

(17) [Dí-rén zhï suö yù zhë] wú tŭ dì yë.

Di-barbarian ATTR REL want NOM POSS:1S territory EQ
'What the Di-Barbarians want is my territory.' (Meng 1B.15) (vgl.
Gassmann 1980: 43)
Situational constructions are again quite similar to Modern Standard
Chinese. The idea of situationality in the above sense is conveyed by adding
the equational marker yë as in the following examples:
(18) Köng-zï yuë: "shèng zé wú bù néng, wö xué
Confucius say saint then I NEG can I learn
bú yàn ér jiào bú juàn yě.
NEG relent and teach NEG tired EQ
'Confucius said: A saint is what I cannot [rise to] — I am learning
relentlessly and I am teaching without being tired [and that is what
I am doing].' (Meng 2.A.2)
(19) shä rén yĭ tĭng yŭ rèn yŏu yŏ
kill man/other take/with stick and sword there_is take/make
yî hü ? — yuë: "wú yĭ y y ."
difference QUEST say there_is_not take difference EQ
[Mencius asked the king:] "Is there any difference between killing
a man with a stick and with a sword?' [The king] replied: There is
no difference. [The situation simply is that there is no difference.]'
(Meng 1.A.4)
Wang Li's (1958: 353) hypothesis that the copula shî of Modern Stan­
dard Chinese is derived from shî in its meaning of a demonstrative pronoun in
Classical Chinese has become more or less the standard explanation since it
was taken up by Li & Thompson (1977). I shall first present this hypothesis
and then develop an alternative hypothesis based on Yen (1986). This alterna­
tive fits much better with the data found in the centuries immediately follow­
ing the classical period from the 2nd century B.C. to the 5th century A.D. and
it allows us to connect the situation of the classical period with the situation of
the pre-classical period maybe back to the period of the oracle-bone inscrip­
tions (14th - 11th century B.C.).
As we can see from the following two examples, shî can indeed take the
function of a demonstrative pronoun in Classical Chinese:
Grammaticalization and language contact 29

(20) zï yú shï rì kü, zé bù gē.

Master PREP this day weep/mourn then NEG sing
T h e master did not sing on that day in which he had been weeping
(i.e. in his time of mourning)'. (Lunyu 7.9)
(21) fü yü guì shï vén zhì suö yù yě.
riches and honours this man/people ATTR REL want/desire EQ
'Riches and honours, this is what people desire.' (Lunyu 4.5)
Of crucial importance for Wang Li's (1958) and Li & Thompson's (1977)
hypothesis is the fact that shï can be used anaphorically in the sense that it
takes up a preceding topic within the comment. Such topic-comment con­
structions as illustrated by the above example (21) "set the stage for reanaly-
sis: the topic-comment construction without a copula became a subject
predicate construction with the anaphoric demonstrative pronoun shï being
reanalyzed as a copula" (Li & Thompson 1977: 424).
Although this hypothesis sounds rather plausible at first sight there is a
whole series of facts which contradict it. Yen (1986) presents three such facts,
I shall present two more in the course of this section.
1. shï does not only take up topics in Classical Chinese, it also refers to
entire state of affairs:
(22) qän lï ér jiàn wáng, shï yü suö yù yë.
thousand mile and see king/you this I REL want/desire EQ
'[To go] a thousand miles to see you [i.e. the king], this is what I
desire.' (Meng 2.B.12)
2. There is another demonstrative pronoun, i.e. ci, which can occur in
exactly the same anaphoric function as shï. The question therefore is why
only the latter was reanalysed as a copula.
3. The anaphoric use of shï was preserved up to the 5th century A.D.,
whereas the copular function of shï is attested since the first century A.D
(e.g. in Wang Chong's Lunheng). As we can see from Qiu (1979), who
presents examples from the discoveries of Mawangdui which have to be
situated in the second century B.C., shï can occur twice in a sequence. In
the following example referring to a picture of a comet, the first shï
displays demonstrative function, the second is used in copular function:
30 Walter Bisang

(23) shi shî zhú huì

this be bamboo comet
'This is the bamboo comet.' (Qiu 1979: 440, also cf. Yen 1986:
240)
Examples such as (23) lead Peyraube (1988: 130, against Zürcher 1977) to the
conclusion that the origin of the copular construction with shî must go back to
the 2nd century B.C.
The findings presented above clearly show that the demonstrative and
the copular function of shi coexisted for some 500 or 600 years. In such a
context, the reanalysis hypothesis as postulated by Li & Thompson (1977)
does not seem to be very plausible.
For Yen (1971, 1986), the copular function of shî is connected to the
negative copula fëi 'not to be' which occurs between Nj and N 2 followed
optionally be the equational marker yë.
(24) N1 fëi N2(yë).
The following sentence illustrating the use of fëi is very famous in Chinese
logics, although it is not quite clear how to deal with it:
(25) bái ma fëi ma.
white horse not_be horse
'A white horse is not a horse.' (cf. e.g. Graham 1959: 88 - 90)
The use of fai is not limited to the copular function, ƒ / also occurs in the same
pragmatic contexts as its positive counterpart yë. In the following two ex­
amples, fēi is employed to mark contrast in a constituent specific construction
(26) and in a situational construction (27):
(26) fëi shèn bài lîng-yïn, lîng-yïn qí bù
not_be gods defeat chancellor chancellor POSS:3 NEG
qín min shî zî bài yë.
treat_diligently people this himself defeat EQ
'It is not the gods who defeated the chancellor, it is the chancellor
who defeated himself by not treating diligently [his] people.' (Zuo,
Xi 28,6; from Yen 1986: 235, and Yen 1971: 410f.)
(27) fëi bù yuè zē zhï dào,
not_be NEG be-pleased master/you ATTR way/doctrines
li bù zù yě.
force/strength NEG enough EQ
Grammaticalization and language contact 31

'It is not that I don't like your doctrines, but my strength is

insufficient.' (Lunyu 6.10)
The negative copula fëi provided the pattern for reanalysing shî as a
copula. In the following example from the Lunheng by Wang Chong (27 to 97
A.D.), quoted by Yen (1986), we find fëi in the function of the negative
copula, followed by shî in the function of an affirmative copula. The Lunheng
is one of the earliest written documents, in which shl is attested in its copular
function (cf. point 3 above).
(28) rú yĭ guĭ fëi sï rén, zé qì xîn
if take spirits not_be dead man/people then POSS:3 believe
Dù Bó fëi yě. rú yĭ guĭ shĭ sĭ rén,
Du Bo wrong EQ if take spirits be dead man/people
zé qí bó zàng fëi yě.
then POSS:3 simple bury wrong EQ
'If they [i.e. the Mohist; W.B.] do not take the spirits for dead men,
their believing [the story of] Du Bo is wrong, if they take the spirits
for dead men, then their burying [the dead men] in a simple way is
wrong.' (Lunheng 23,7b,10-ll)
[This passage is against the Mohists who opt for simple burials
because they believe that the deceased are changed into spirits. Du
Bo is an example of a dead man who became a spirit.]
Since shî and fëi form a pair of semantic opposition in Classical Chinese
with shi meaning 'right, correct' and fëi meaning 'wrong', it is quite natural to
use shî in contrast to fëi not only in the context of stative verbs but as well in
contrast to fēi in its function of a negative copula as in the above example (28).
Thus, it was not the demonstrative meaning of shì that lead to its reanalysis as
a copula, it was its function of a stative verb with the meaning 'right, correct'
which linked it to fëi which, in turn, already had both functions, that of a
(negative) copula and that of a stative verb. Since the opposition pair shîlfëi is
very common in Classical Chinese, where it is attested in innumerable discus­
sions about behaving correctly, the association of shi with fëi even within a
new context such as that of the equational construction does not come as too
much of a surprise. The next step in the history of the equational construction
with shi is the functional change of yě from the equational marker to a focus
particle with the meaning of 'too, also'. Still later, the negative copula fëi is
replaced by the regular negation bù followed by the copula shì.
32 Walter Bisang

In my view, the fact that the negative copula fëi is commonly used to
mark contrast in the constituent specific construction and in the situational
construction (cf. examples (26) and (27)) is a fourth argument in favour of
Yen's (1986) hypothesis that the copular function of shì is derived from its
meaning as a stative verb. Moreover, the concept of an affirmative counterpart
to fëi is not new in the history of Chinese. If we go back to the pre-classical
period of the Shujing (Book of Documents, 9th to 6th century B.C.), we find
the particle wéi which is known since Uhle (1881) quoted by Gabelentz
(1881: 314) to correspond to Classical Chinese yě (for further discussion cf.
Pulleyblank 1959).8 At that period of time, wéi seems to display the same
function as fëi (apart from the difference in polarity). Thus, we get wéilfēëias a
pair of discourse pragmatic opposition which can well be compared to shîlfëi.
In the following example, wéi and fēi are used together to mark contrast:
(29) fëi yü zî huāng zï dé, wéi rü shě dé.
not_be I self neglect this virtue it_is you conceal virtue
'It is not me myself who neglects this virtue, it is you who conceals
[my] virtue.' (Shujing, Pan'geng, Part I, 8)
As pointed out by Pulleyblank (1959), the words wéi and fëi are not only
functionally and grammatically close, they also show phonetic similarities.fei
has been reconstructed as piwәr, wéi as diwdr. This phonetic similarity leads
Pulleyblank (1959: 183) to analyse ƒē/ as bù wéi, i.e. as a combination of the b-
negation (Modern Standard Chinese bù) with the affirmative copula wéi.
When the symmetry of wéilfëi began to be replaced by Nyě vs. fēi N, it was the
historically more complex form offēi which was preserved. The fact that there
was an affirmative counterpart to fëi in pre-classical times may be a fifth
argument in favour of Yen (1986), although it is almost impossible that wéi
took immediate influence on the development of the copula shî.
If we look at the expression of focus in Chinese, we see a remarkable
continuity of far more than two thousand years in which focus (and some­
times even topic) is expressed by the equational construction. The fact that
this construction showed considerable differences with regard to the internal
distribution of its components did not seriously trouble this continuity. Within
the history of the Chinese equational construction as a whole and as far as it
can be projected back to the past, the characteristic pattern as we find it in
Classical Chinese (iV1 N2 ye) seems to be a rather transitory phenomenon.
Before and after that period, the copula occurred between N{ and N 2 .
Grammaticalization and language contact 33

3. Serial units: Verb serialization and attractor positions9

The construction which I call serial unit (cf. Bisang 1992, 1996) represents
the most grammaticalized part of verb serialization. Its main function seems
to be to keep together individual state of affairs, even if they have to be
expressed by two or more verbs. A serial unit consists of the main verb plus
positions for verbs marking TAM, coverbs, and directional verbs. TAM
markers are used to express tense, aspect or modality. Directional verbs (Vd)
indicate the direction of the action with regard to the speaker or the center of
interest in a particular context. Coverbs (COV) are verbs in the function of
adpositions. In all the languages to be presented in this section (i.e. East and
Southeast Asian languages represented by Chinese and Khmer, Yoruba
[Kwa, West Africa] and Jabêm [Austronesian, Papua New Guinea]) we find a
serial unit which consists in its maximum form of a main verb plus TAM
markers, directional verbs and coverbs.
If verb serialization is defined as "the unmarked juxtaposition of two or
more verbs or verbal phrases (with or without subject and/or object), each of
which would also be able to form a sentence on its own" (Bisang 1992: 9 - 10)
the construction of the serial unit can be used to analyse strings of verbs of the
following type:
(30) (NP) V (NP) V (NP) V (NP) ...
The positional distribution of verbs expressing TAM, coverbs and direc­
tional verbs differs up to a certain degree which I shall further discuss below.
The structure of the serial unit in Chinese, Jabêm, Yoruba and Khmer can be
described by the following patterns:
Chinese: (31) TAM COV V-TAM COV Vd TAM
Jabêm: (32) V COV Vd TAM
Yoruba: (33) COV V Vd COV TAM
Khmer: (34) TAM V Vd COV TAM
Of course, I am aware that verb serialization does not follow the same
principles in Chinese, Jabêm, Yoruba and Khmer. Some of the differences
will be pointed out here. However, none of these differences affects the
positional distribution of verbal TAM markers, directional verbs and coverbs
relative to the main verb as presented in (31) to (34).
Verb serialization in Yoruba is subject to many particular constraints
34 Walter Bisang

which are excellently presented by Déchaîne (1993a, 1993b). With regard to

the serial unit as postulated in this paper I would like to point out only two
characteristics, i.e. the use of one subject for the whole construction of verb
serialization (subject identity) and the use or non-use of the High Tone
Syllable. Except in the imperative each verb in Yoruba must occur with at
least an obligatory subject pronoun. In a serial construction, however, only the
first takes a subject:

(35) ó tà á 0 fún mi.

he sell OBJ;3S cov:give OBJ:1S
'He sold it to me.'
The High Tone Syllable (HTS) is commonly described as a floating tone
which is realized at the last syllable of the subject NP. As is shown by Sonaiya
& Bisang (forth.), it is impossible to describe this marker in the context of the
subject. Maybe the most straightforward case against such a description is the
fact that this tone cannot occur if the clause is negated by (k)à 'not'. Neverthe­
less, the High Tone Syllable is a very good indicator of verb serialization in
the contexts where it has to occur. In such a context the High Tone Syllable
can only mark the subject of the first verb in a construction of verb serializa­
tion (36), whereas in embedded clauses it can also mark embedded subjects
(37). This is illustrated by the following examples from Lord (1974) who
presented this distinction:

(36) Bólá mû Fémi wá síbí.

Bola take Fe0mi come here
'Bola brought Femi here.'
(37) Bólá mú kí Fémi .'. wá síbí.
Bola take COMPL Femi HTS come here
'Bola made Femi come here.'
Unlike in Yoruba, there is no obligatory expression of any arguments in
the languages of East and mainland Southeast Asia (cf. indeterminateness in
Bisang 1992). The constraint on subject identity seems to be rigid in construc­
tions like the serial unit, but I doubt whether this is a general characteristic of
verb serialization in these languages.
The verb in Jabêm is different from the verb in all the other languages
described in the present paper inasmuch as it has a set of two obligatory
prefixes combining the marking of TAM, i.e. realis (R) and imaginativus (I)
Grammaticalization and language contact 35

(Dempwolff 1939), with seven forms to mark person, i.e. 1st, 2nd, 3rd person
singular and p0lural with 1st person plural inclusive vs. exclusive. The connec­
tion between the stem and these TAM prefixes is so tight that the verb stem
cannot occur independently (for further information cf. Dempwolff 1939: 12-
19; Bradshaw 1979). For that reason, the forms which we find in the positions
of the Jabêm serial unit do not only convey information concerning the action
or event expressed by the verb, they further encompass obligatory information
concerning person and TAM. From these forms we can clearly see that
subject identity is of no relevance for the serial unit in Jabêm. While the prefix
ê- (third person imaginativus) occurs with all the three verbs of example (38)
and thus shows subject identity throughout the whole serial unit, the prefixes
differ in example (39). The first verb takes the prefix ka- (first person singular
realis) while the second and the third verbs take the prefix of the third person
realis which is kê- or gê- according to the morphological class of the verb.
Example (39) is thus an illustration of subject change:
(38) Gwamsec e-lac ê-pi Bukawac ê-na
Gwamsec 3sl-sail cov:3sI-move_upward Bukawac Vd:3sI-go
mè?
QUEST
'Will Gwamsec sail up to Bukawac?' (Zahn 1940: 50)
(39) aê ka-kêk bu kê-sêp ku gê-ja.
I lsR-pour water cov:3sR-move_downward pot Vd:3sR-go
T poured water down into the pot.' (Zahn 1940: 49)
The serial units presented in (31) to (34) are maximum patterns in the
sense that none of its verbal elements apart from the main verb itself is
compulsory (cf. indeterminateness of the verb in Bisang 1992, 1996). To
illustrate each of the above maximum patterns and the way it grants the
interpretation of each verb, I shall introduce each language with an example
containing at least two grammaticalized verbs:
(40) Chinese
tā jiù yào bă 'kŏu'-zî xië
s/he just TAM:FUT cov:take 'mouth'-character write
zài hêibăn-shang le.
cov:be_at blackboard-on PF
'S/he is just writing the character 'mouth' at the blackboard.'
36 Walter Bisang

(41) Jabêm
agêc sê-sôm bin gê-dêrn
the_two_of_them 3pR-say speech 3sR-move_towards
tau-n sê-kô è balôm-o gê-lic namalac
self-POSS:3p 3pR-stand CONJ:while ghost-FEM 3sR-see man
gêdô ôlic nadani ma ...
lips thick and
'While the two of them were speaking to each other the ghost
woman became aware that the man had thick lips and ...' (Zahn
1940: 334)
(42) Yoruba
mo ti bá àbùrò mi mú ìwé wá
I PF cov:on__behalf younger_sibling my take book Vd:
fùn o rì.
cov:give OBJ:2S TAM:EXPER
'I once brought you a book for my younger brother.'
(43) Khmer
kbәt ba:n yc:k ?yyvan coh cën
he PST:get V:take luggage Vd:go_down Vd:go_out
md:k ?aoy khjiom.
Vd:come cov:give I
'He took the luggage down and out for me.'
The positions in which we find verbs marking TAM, directional verbs
and coverbs are called attractor positions. They can be described from the
paradigmatic perspective and from the syntagmatic perspective. From the
paradigmatic perspective, they represent slots which attract linguistic items in
order to grammaticalize them. In this sense, they operate as a kind of melting
pot or as a kind of catalyst for linguistic items to be grammaticalized into
different types of grammatical functions. If, e.g., an element falls into the
domain of the attractor position for TAM it will be grammaticalized into a
TAM marker. Each of the three grammatical categories has its own attractor
position which occurs within a certain position relative to the other positions
within the serial unit. The serial unit represents the framework within which
attractor positions can operate syntagmatically. From the syntagmatic per­
spective, the construction of the serial unit turns out to be a generative
mechanism from the point of view of the speaker and a parsing mechanism
Grammaticalization and language contact 37

from the point of view of the hearer to produce or to analyse such verbal
strings as represented in (30). The paradigmatic and the syntagmatic perspec­
tives are both of crucial importance for the initial steps of grammaticalization
as described by Hopper & Traugott (1993). In their paradigmatic function,
attractor positions promote metaphoric processes, in their syntagmatic func­
tion they actuate metonymic processes. Thus, attractor positions and their
linear ordering within the serial unit are excellent catalysts for metonymic and
metaphoric inference (for further explanations cf. Bisang 1996). Further­
more, their existence allows a linguistic item to be rather quickly integrated
into a new function within the grammatical system of a language. Conse­
quently, serial units with their characteristic pattern of attractor positions may
be a good explanation for abrupt processes of grammaticalization. Grammati­
calization is not necessarily a slow and gradual process of reanalysis as
postulated e.g. by Bybee et al. (1994), if it refers to position.
Before concluding my definition of attractor positions, I would like to
point out that an attractor position does not necessarily allow only one item of
the same category to occur in a given construction. This is illustrated by the
three directional verbs occurring in (43) on Khmer. Similarly, we can find two
or even more coverbs at one side of the main verb in one and the same serial
unit if several nouns need to be marked for semantic role. In such a situation,
the whole sequence of coverbs occurs within the positional constraints gov­
erned by the respective attractor position. Thus, an attractor position provides
one or more slots for the expression of one type of grammatical categories.
As we can see from the structures of the serial units as presented in (31)
to (34), the TAM position is always further to the periphery than the COV
position be it to the left or to the right of the main verb. This fact can be
explained in terms of the semantic principles of relevance and generality10 as
presented by Bybee (1985) if we take distance from the main verb as another
indicator of the degree of fusion apart from morphology as suggested by
Bybee. Thus, the relative distance of attractor positions for TAM and COV
from the main verb seems to be basically the same as the relative distance of
its morphologically expressed grammatical analogue from its verb stem. This
can be confirmed by the statistical data presented by Bybee if we subsume
coverbs under the category of valency. As we can see from Bybee (1985: 30-
31), only 6% of the languages of her sample express valency by inflection,
whereas 84% of these languages express valency by derivation. The figures
for aspect (52% inflection, 22% derivation), tense (48% inflection, 2% deriva-
38 Walter Bisang

tion), and modality (68% inflection, 0% derivation) clearly show that valency
is more tightly linked to the stem than TAM. In the same way COV positions
are closer to the main verb than TAM positions.
The TAM position immediately following the main verb in Chinese can
partly be explained as an instance where morphological principles became
more important than distance and partly by its origin from the resultative
construction in which the main verb was immediately followed by its resulta­
tive verb. The following example illustrates the post-verbal TAM markers
-guo (experiential) and -le (perfective):
(44) Chinese (Iljic 1987a: 5)
a. tā duàn-le tuĭ.
s/he break-PFV leg
'S/He has broken his/her leg.' [It is still broken.]
b. tā duàn-guo tuï.
s/he break-EXPER leg
'S/He has broken his/her leg.' [It is cured now.]
My presentation so far shows semantic motivation for the TAM position
and the COV position relative to each other. The next question is what we can
say about directional verbs. According to the principle of relevance they
affect or modify the meaning of the main verb in too strong a way to be
applicable to all the verbs as postulated by the principle of generality. For that
reason, there seems to be more fusion between directional verbs and the main
verb than between TAM markers and directional verbs. This is also reflected
in the above serial units, since there is no language in which Vd occurs after
TAM. Thus, there is a semantic motivation for the position of TAM markers
relative to COV and Vd.
As for coverbs and directional verbs, the situation seems to be different.
Their degree of fusion to the main verb seems to be more or less equal. This is
reflected in the structure of the serial units as presented in (31) to (34). In
Chinese and Jabêm we find the sequence COV - Vd, in Yoruba and Khmer we
find Vd - COV. Thus, the positional distribution of coverbs and directional
verbs seems to be arbitrary. Accordingly, both solutions should be equally
likely to cccur in a language with serial units.
Attractor positions (within their serial units) are not only important
promotors of grammaticalization, but as well of linguistic areas. In Bisang
(1996), I have tried to describe the similarities of grammaticalization in East
Grammaticalization and language contact 39

and mainland Southeast Asia in the framework of language contact. Con­

structions such as serial units play an important role in the interaction of
speaker and hearer in the way I have described above in § 1 which leads into
the development of linguistic areas. In this sense the connection between
constructions (e.g. serial units), grammaticalization and language contact can
turn out to be pretty strong at least in some areas.

4. Classifier constructions: Two processes of development, one

construction

Since Greenberg (1974: 25) the occurrence of classifiers in a language is

generally linked to the lack of a compulsory expression of plurality:
"Numeral classifier languages generally do not have compulsory expression
of nominal plurality, but at most facultative expression."

The lack of any compulsory marking of number, i.e. transnumerality, is just

one aspect of a more general high degree of indeterminateness in languages of
East and Southeast Asia. In these languages, a noun expresses only a mere
concept of an object which can be further specified — if necessary — by
various operations. Thus, an expression like e.g. Chinese xîn can mean 'letter,
letters, a letter, the letter, etc.' according to a given context. Within the high
degree of indeterminateness, transnumerality has been described as crucial for
the existence of classifiers.11 According to Greenberg (1974), a noun in a
transnumeral language cannot occur in immediate combination with a nu­
meral — it first has to be individualized by the classifier in order to be
countable or enumerable. Thus, Chinese xîn 'letter' has to be individualized
with the classifier fēng if it is used in the context of counting: sān fēng xìn
[three-CL-letter] 'three letters' vs. *sän xîn. This situation may be true for the
majority of the transnumeral languages of East and Southeast Asia, but it does
not hold for all of them.
Apart from individualizing a noun, the classifier also classifies it (cf.
Denny 1976, Allan 1977). The first distinctions made in classifier systems are
almost universally between animates and inanimates and humans and non-
humans (cf. Adams & Conklin 1973). A secondary criterion to [+ human] is
status, i.e. the social status of the person denoted by the noun to be classified.
The second consistent primary criterion of classification discovered by Adams
40 Walter Bisang

& Conklin is shape, i.e. the extension of objects in the three spatial dimensions.
Thus, shape distinguishes long (one-dimensional) from flat (two-dimensional)
and round (three-dimensional) objects. Secondary to the criterion of shape are
physical parameters such as rigidity or flexiblity, relative size, empty vs. full
(hollowness), irregularity vs. regularity, part vs. whole, horizontal vs. vertical
(orientation) and edginess. These secondary parameters can never form an
independent basis for individual classes. A last criterion is based on the nature
or the function of a noun. This criterion seems to be secondary to shape as well.
Thus, the criteria of [± human]/[± animate] and [shape] form the basic semantic
grid at work in classifier systems. It can be represented within a hierarchy such
as the following, which slightly deviates from the one presented by Croft (1994:
152):
(45) Animate/Human < Status
Animacys
Inanimate/Nonhuman <Shape < Orientation/Rigidity
< Nature/Function
In Vietnamese, the main function of the classifier is to make a noun
syntactically referential (cf. Löbel 1996, forthc). Vietnamese classifiers are
not compulsory in most contexts of counting (cf. below). In the following
example (46a), the noun phàng 'room' is part of an attribute to nhà 'house'
and it is not marked by the classifier, although it cooccurs with the numeral ba
'three':
(46) Vietnamese (Löbel 1996: 171)
a. nhà [ba phòng]
house NUM:three N:room
'a three room house'
b. nhà [vó'i ba cái phàng]
house with three CL room
'a house with three rooms'
In example (46a), the noun phàng 'room' is not individualized. What matters
is the fact that the house has three rooms, the rooms as such are of no interest.
Thus, nhà ba phàng has to be translated into English as some kind of
compound. In (46b), however, phàng is individualized, the three rooms with
their individual characteristics are relevant. The following example further
corroborates the fact that classifiers are not compulsory in the context of
counting:
Grammaticalization and language contact 41

(47) Vietnamese (Löbel forthc: 27 - 28)

a. Trongnhà hàt kia cò mu'ò'i bay ghê.
in cinema this have ten seven chair
'There are seventeen chairs in this cinema.',
lit.: This cinema is seventeen-chaired.'
b. Trongnhà hát kia có 17 cái ghê làm bang
in cinema this have 17 CL chair make out_of
cây tôt.
wood good
There are seventeen wooden chairs in this cinema.'
lit.: There are seventeen chairs in this cinema which are made
of precious wood.'
In both examples, (46) and (47), the classifier occurs if the noun has to be
syntactically referentialized, i.e. if it has to be individualized for being further
modified. There is only one instance, in which classifiers seem to be obliga­
tory in the context of counting. This is with nouns denoting humans.
The fact that classifiers are not compulsory in the context of counting in
all the languages of East and Southeast Asia may be due to two different
processes of development which I called item oriented process and category
oriented process in Bisang (forthc. b). The item oriented process starts with
using classifiers with items which are of particular cultural importance. The
result of this process is a new construction containing a new and clearly
defined position for the classifier. The construction applied to the relatively
few nouns successively spreads over to a wider range of nouns until all the
count nouns of a given language become part of a classifier system. This item
oriented process of development is predominant in the Northern parts of East
and Southeast Asia, i.e. Chinese (and also in Japanese)12. The category
oriented process is based on a categorial system already existing in the
language. This system, which is part of word formation, is based on taxonomy
and meronomy. In both cases, we have two positions filled by nouns. In a
process of reanalysis, one of these positions, i.e. the first position in the
languages described here, is reanalysed as a classifier. In the course of further
development, the language may even develop an independent classifier posi­
tion. This category oriented process seems to be predominant in the Southern
parts of East and Southeast Asia, i.e. in Vietnamese, Thai and Hmong. In
Vietnamese, the independence of classifiers from the context of counting
seems to be due to their origin from a category oriented process, i.e. from a
42 Walter Bisang

process which is not immediately linked to counting. However, Vietnamese

seems to be the only language I have analysed in which classifiers are not
largely compulsory in the context of counting, even if they originate from a
category oriented process. This observation leads me to the conclusion that
both processes of development are potentially directed towards the same aim,
i.e. the obligatory individualization of items in the context of counting. The
rather high degree of the likelihood that the two processes converge into one
and the same construction used for counting will be explained after a some­
what more detailed description of each of the two processes.
The item oriented process of classifier development will be illustrated by
the example of Chinese. In the history of this language, classifiers first occur
with single objects of particular cultural value. For that reason, many classifi­
ers have a very concrete basic meaning. With the course of time, however,
they extend their sphere of influence from their highly specific context to the
context of systematic classification. Thus, the development of Chinese classi­
fiers starts from a single noun and its concrete semantics rather than from the
notion of class or category membership.
Although the first flowering period of classifiers starts at the end of Han
dynasty (220 A.D.), some potential classifiers can be postulated for the
earliest written sources, i.e. the oracle-bone inscriptions (14th - 11th century
B.C.). In these records, we find the nouns ren 'man', qiang 'name of a tribe'
and quan 'dog' in the function of repeater classifiers (Takashima 1984/1985).
It should be pointed out, however, that these classifiers are far from being
obligatory. We can just see from the texts that they sometimes occur.
(48) Chinese of oracle-bone inscriptions (Takashima 1984/1985: 235)
ren shi you wu ren
N:man ten and/have five CL:man
'fifteen people of the men; the men, fifteen of them'
The classifier status of the word hing occurring with horses is controversial,
because it may imply a team of two horses and thus display the function of a
quantifier. Similarly, the word peng is not used for counting individual items
of cowries, but to measure a certain quantity of cowries. What we can see
from these classifiers is that humans are the first to be counted by classifiers
followed by socially particularly valuable animals such as dogs and horses, if
we take hing as an item that is at least very close to a classifier.
In addition to the classifiers already mentioned above in the context of
Grammaticalization and language contact 43

the oracle-bone inscriptions some more classifiers denoting highly valued and
culturally significant objects came into use up to the Qin dynasty (221 B.C.):
shèng13 and liàng14 for chariots (especially war chariots), méi 'tree trunk' for
wooden objects, gè for arrows and pĭ for horses. The following example
presents gé 'bamboo tree' which forms the historical basis of the general
classifier ge of Modern Standard Chinese:

(49) Classical Chinese (Xunzi, Yibing; 298 - 235 B.C.; also cf. Wang Li
1958: 238)
shĭ wŭ shi gè.
arrow five ten CL
'fifty arrows.'
At the periods of Wei-Jing and Nanbeichao (220 - 581 A.D.) the vast
majority of the more important classifiers in Modern Standard Chinese are
already attested (for a wealth of details cf. the excellent monograph by Liu
1965). Some of these classifiers are listed here: zhĭl (for animals), kou
'mouth' (for animals), tou 'head' (for animals), shù 'tree', zhl2 'twig' (for
one-dimensional objects), tiáo 'branch' (for one-dimensional objects), zhāng
'to spread, stretch' (for two-dimensional objects), jiàn (for garments), suŏljiän
(for places); special classifiers: ben 'root' (for books), fēng 'seal' (for letters),
sou (for ships). The development of some of the most important classifiers of
Modern Standard Chinese will be briefly sketched in the next three para­
graphs.
As I pointed out, classifiers were first used with culturally valued items.
Later, their use extended to other objects with similar properties. At that
period of extension, classifiers developed prototypical features. This kind of
generalization started at some time during the Han dynasty (206 B.C. to 220
A.D.), when classifiers referred to flowers, fruits, birds and animals. In the
process of generalization nouns denoting trees or parts of trees became of
great importance. The general classifier ge of Modern Standard Chinese as
well as its predecessor méi both go back to nouns denoting trees, gè meaning
'bamboo tree', méi meaning first 'tree trunk', then 'tree' in general15.
Although the basic meaning of méi is 'tree trunk', it was very often used
with fine, subtle varieties of stems or stalks. Furthermore, a particular type of
small stalks denoted by méi was used with a particular instrument for counting
called choumă and with the interpretation of oracles.16 If this type of méi-
stalks was generally used in the context of counting it comes as no surprise
44 Walter Bisang

that méi was later used as a classifier. As is proved by the numerous examples
quoted by Liu (1965: 77 - 82), the classifier méi was able to refer to almost
any noun by the time of the periods of Wei-Jing and Nanbeichao (220 - 581
A.D.). Thus, it is well justified to look at méi as a general classifier at that
time, although one should always bear in mind that classifiers did not have the
same status of obligatoriness as in Modern Standard Chinese. The noun gè
was first used to classify bamboo trees and arrows (cf example (49)). At the
period of time analysed by Liu (1965), gè was also used to refer to other
objects and even to humans in alternation with the specific classifiers denot­
ing these items.17 By the Tang dynasty (618 - 907 A.D.) gè started slowly to
get its way against méi. Other widespread classifiers based on trees are tiáo
'branch' and zhï2 'twig' with the former referring to big one-dimensional
items and the latter denoting small, slender one-dimensional items.
As is pointed out by Erbaugh (1986: 430), the classifier zhāng for two-
dimensional items developed later than tiáo for one-dimensional items. This
classifier is remarkable, because it derives from a verb meaning 'stretch' used
e.g. in the context of stretching a bow. For that reason, it seems quite natural
that this classifier was first used with bows and stringed instruments such as
the qin 'Chinese zither' by the Han dynasty. By the Song dynasty (960 - 1279
A.D.) its extension to two-dimensional items in general was well established.
Other classifiers such as those referring to buildings and places (suŏ, jiāri)
came up still later than zhäng, i.e. in the Tang dynasty.
The diachronic development of classifiers started from a specific lexical
use to the formation of prototypes which formed the basis for further generali­
zations. In a very similar way, classifiers are acquired by children. I shall
describe this parallelism by quoting Erbaugh (1986):
"Early use [of classifiers by children, W.B.] was lexical, specific to a single
referent. Later use marked a prototypical member of a noun class. After this,
the child began generalizing from the prototype, often in plausible but incor­
rect ways. Shape, especially vertical extension and small size, were the most
frequently generalized features. The child acquisition pattern for classifier
strikingly parallels: 1) General semantic acquisition patterns, both in Manda­
rin and cross-linguistically. 2) Individual and adult variation in classifier
choice. 3) The historical pattern of broadening classifier scope from single to
class reference." (Erbaugh 1986: 415)

The item oriented development of classifiers seems to be still reflected in

various ways in Modern Standard Chinese. (1) There is a certain flexibility in
assigning individual classifiers to individual nouns. A noun such as yáng
Grammaticalization and language contact 45

'sheep' can take the classifiers zhi, tóu or tiáo without any significant differ­
ence of meaning. (2) The classifier system does not that strictly mirror the
semantic hierarchy based on human/animate vs. inanimate and dimensionality
as we find it in other classifier languages based on the category oriented
process of development. Nouns belonging to the category of [+human] can
only be classified by the general classifier ge. (3) As described by Erbaugh
(1986), speakers of Chinese either use the special classifier or the general
classifier ge with one and the same noun. A very important criterion governing
the use of one or the other classifier is discourse. New items which are first
introduced strongly tend to be marked by the special classifier, whereas items
which are already known occur with the general classifier (for a similar
phenomenon in a particular variety of Malay cf. Hopper 1986).
The category oriented process of classifier development is based on
taxonomy and meronomy. I shall first describe taxonomic classification. In
Bisang (1993) on Hmong, I presented a process of development based on
class nouns (CN)18. Class nouns form the first step on the pathway of gram­
maticalization from noun to classifier:
(50) N > CN > Q > intQ > CL 19
Class nouns generally represent a rather high level of abstraction from which
more concrete subcategories can be derived by further determination. An
English example would be tree from which we can derive apple tree. In
Hmong, there are class nouns such as ntoo 'tree', txiv 'fruit', noog 'bird',
ntses 'fish', tub 'son' (for agentive nouns), kws 'expert' (for people with a
certain profession), zaj (for sayings, speeches), etc. In (51), the class nouns
txiv 'fruit' and ntoo 'tree' are further determined to designate special subcat­
egories. Example (52) presents the class noun zaj:

(51) txiv kab ntxwv 'orange' ntoo kab ntxwv 'orange tree'
txiv ntseej 'chestnut' ntoo ntseej 'chestnut tree'
(52) zaj dab neeg [CN-spirit-human] 'legend'
zaj teev ntuj [CN-pray-heaven] 'prayer'
zaj tshoob [CN-marriage/wedding] 'wedding song'
The class noun zaj can also be used in the function of a special classifier
with some nouns. In the following example, we find zaj in two positions, in the
classifier position and in the class noun position:
46 Walter Bisang

(53) ob zag20 zaj tshoob

two CL CN wedding
'two wedding songs'
Similar to Hmong, we also find a wealth of class nouns in Thai. In the so
called semi-repeater construction, class nouns are also used in the function of
classifiers. The semi-repeater construction applies with nouns whose first
element is a class noun which will be repeated in the classifier position.
(54) khon-khây sâam khon
CN:man-sick three CL:man
'three patients'
(55) d k-kulàap cèt d k
CN:flower-rose seven CL:flower
'seven roses'
In Vietnamese, there is a pattern of two positions, in which the first noun
is reinterpreted as a classifier with the function of individualizing (and classi-
fying) the item of the second position. This kind of reinterpretation can take
place if the item in the second position is in the relation of a subclass to the
item of the first position. In this case, we have the situation referred to in the
above pathway of grammaticalization (50), in which a class noun is reinter-
preted as a classifier. The second situation under which this kind of reinter­
pretation can be realized is if the items occurring in the two positions are in a
part-whole relation.21 This is shown in the following example (Löbel forthc.:
13):

(56) a. một cửa xe ho'i

one CL:door car
'a door of a car'
b. mot kiền xe ho'i
one CL:mirror car
'a mirror of a car'
If there is no part-whole relation (57a) or if the part-whole reading is not
intended (57b) a core classifier, i.e. a more general classifier, has to be used
(Löbel forthc: 13 - 14):
Grammaticalization and language contact 47

(57) a. mot chiêc xe ho'i

one CL car
'a car'
b. mot cái kien
one CL mirror
'a mirror'
As is pointed out by Löbel (forthc: 14), there is no contextual presupposition
with regard to 'mirror' in example (57b). If the core classifier does not occur
in that position, i.e. if kien occurs in the classifier position (56c), this implies
that it is "a part of a contextually given whole":
(56) c. mot kiên
one CL: mirror
'a mirror (of X)'
To make my pathway of grammaticalization more general with regard to
meronomy, I should thus add the notion of 'part of whole' (PW):
(58) N > CN/PW > Q > intQ > CL
Classifiers maximally express the following four functions of which the
first two can be found in all the classifier languages of East and mainland
South East Asia: individualization, classification, referentialization (i.e.
marking of definiteness/specificity) and relationalization (i.e. marking of
possession or even relative clauses as in Cantonese) (for further details cf.
Bisang 1993, forthc. a).
The evidence presented by Löbel clearly shows that the functional range
of Vietnamese classifiers is limited to individualization and classification.
Constructions of the type 'classifier-noun' are systematically ambiguous with
regard to definiteness vs. indefiniteness.
(59) Vietnamese (Bisang 1996: 541, discussed by Löbel 1996: 228 - 229)
toi mua qua cam
I buy CL orange
T buy the orange./ I buy an orange.'
In other languages, classifiers can take other functions. The following typo­
logical grid based on the above four functions of classifiers can be used to
present a survey of the functional range of classifiers in individual languages:
48 Walter Bisang

I. classification & individualization

Japanese (classifiers occur only with numerals)
Chinese (classifiers occur with numerals and demonstratives)
Vietnamese (individualization, but not necessarily in the context of
counting)
77. classification & individualization & referentialization
Thai (secondary function in combination with adjectives [stative
verbs] in the sequence N-CL-ADJ)
III. classification & individualization & relationalization
Cantonese (classifiers can be used in possessive and relative con­
structions)
IV. classification & individualization & referentialization & relationaliza­
tion
Hmong (with referentialization being a secondary function), Weining
Miao.
The question of why classification and individualization are the core
functions of classifiers and the question of why classifiers expand to referen­
tialization and relationalization in some languages can be tentatively answered
by looking at the cognitive background of these functions. I shall start my
explanatory sketch with classification, which is defined as an operation of
subsuming a given entity under a certain class according to its properties.
Classification can be used for two purposes. It can be employed to compare
one particular sensory perception and its properties to the properties of other
sensory perceptions in order to identify that particular perception by subsum­
ing it under a certain concept or it can be employed to establish a sensory
perception as an individuum by actualizing the inherent properties which
constitute its conceptual unity. The former operation is called identification,
the latter individualization. Identification can take place without referring to
individualization. It seems possible to identify a sensory perception as belong­
ing to the concept of say 'apple' without explicitly referring to its inherent
boundaries, i.e., without referring to it as an individual. On the other hand, it
seems difficult to individualize a sensory perception without identifying it.
Departing from classification, which is "a prerequisite to identification"
(Croft 1994: 161), we may tentatively establish the following hierarchy:

(60) Classification > identification > individualization.

Grammaticalization and language contact 49

The functional expansion of classifiers into the marking of reference or

the marking of possession and relative clauses can be seen in the context of
identification. If we define reference as an act of identifying some entity that
the speaker intends to talk about (cf. Searle 1969, Croft 1991: 110) we may
say that the identificational function of the classifier forms the point of
departure for classifiers to take on the function of referentialization. As for
relationalization, one might argue in a similar way that it is linked to the
previous identification of the head noun before it can be modified by a
possessor or a relative clause. The functional expansion of classifiers may
thus be schematically represented as follows:
(61) Classification > Identification > Individualization
↓↓
Referentialization/ Counting
Relationalization
To conclude this section, I would like to look at the operations involved
with counting in order to explain why classifiers tend to become compulsory
in the context of counting not only within systems developed from the item
oriented process but as well within systems developed from the category
oriented process. I shall start my explanations from the point of view of
mathematical psychology, in which counting is usually subsumed under the
term of measuring (Suppes & Zinnes 1963). Measuring is understood in a
very broad sense as the mapping of objects onto numbers. In the context of

Table 3. Schema of numeric quantification

50 Walter Bisang

numeric quantification, the numeral assigned to the items to be quantified

indicates their number. This is illustrated by the figure above, which is
inspired by Wiese's (1995: 186) stimulating paper on semantic and concep­
tual structures of numeral constructions in German.
From the above approach of mathematical psychology, we can distin­
guish an operation of individualization and an operation of mapping objects
onto the set of natural numbers. The second operation will be called operation
of ordering. Furthermore, there is a notional aspect which defines what
elements are to be counted. This notional aspect can be defined more clearly
if we ask the question of how exactly we find out whether an element qualifies
as a member of the set of items to be counted. If we look at the above table 3
again, we have to find the criteria which lead us to distinguish three different
sets to be counted separately in the above perceptional environment, i.e. the
set of the squares (o), the set of the stars (g) and the set of the crosses (:). The
approach of mathematical psychology offers no solution to this problem,
because it neglects the operation of classification and its two related opera­
tions of identification and individualization. Croft (1994), for whom counting
only consists of identification and individualization, presents the following
scenario:
"One must identify multiple units of the same kind in order to count them.
The two cognitive processes involve[d] are individualization of units, and
identification of them as being of the same kind" (Croft 1994: 162).

Without the operation of classification for identifying an element as belong­

ing to the set of items to be counted it would be impossible to decide whether
to count that element or not. An element which is established as being
relevant for counting is identified. Finally, there seems to be general agree­
ment on the relevance of individualization for counting.
In the above discussion, I have presented four operations involved with
counting, i.e. classification, identification, individualization and ordering.
The first three operations correspond to the core functions of classifiers. This
may explain why classifiers are compulsory in the majority of classifier
languages. The convergence of the item oriented process and the category
oriented process into a classifier construction consisting of a numeral, a
classifier and a noun may therefore be quite natural, given the large common
cognitive background. Nevertheless, this convergence is not compulsory, as
we can see from Vietnamese, where classifiers are not obligatory in the
context of counting.
Grammaticalization and language contact 51

5. Conclusion

In the above discussion, I have tried to show that constructions and the human
cognitive equipment are somehow involved in the propagation of linguistic
changes. On the other hand, we have seen in §1 that the factor of socio-
linguistics seems to be able to disregard both of the other factors and is thus
primary. These remarks should briefly be discussed with regard to the more
general question of the status of language universals and the sociolinguistic
aspect of language use. Of course, what I can say here must necessarily be
speculative and may need corrections.
In general, I do not deny the existence of language universals and I do not
deny that they have a cognitive base and that some of this cognitive base is
innate. The problem is to find out what really are the properties of innate
structures and how they interact with language use. This question is far from
trivial, if one keeps in mind that man is the only living being whose brain is
mainly developed outside of the uterus. Some 75% of the 1400 cubic centime­
ters of an adult human brain is developed ex utero (cf. Wills 1993), i.e. in
interaction with the outside world and with language as one part of it.22
In my view, constructions may be seen as a domain of linguistic knowl­
edge which is only partly innate. Constructions may be the result of the human
cognitive equipment and its interaction with the outside world and with
language in particular. That the effect of syntactic constructions cannot be
explained by universal constraints is supported by Zwicky (1994): "an ex­
pression is ungrammatical only because there is no combination of construc­
tions that licenses it, not because there is some cross-constructional filter that
rules it out" (Zwicky 1994: 614, also see Michaelis & Lambrecht 1996: 218).
As for cross-constructional filters representing universal constraints, Zwicky
goes on saying:
"It is hard to see how the effect of syntactic constructions could be achieved
via a set of universal constraints, however ranked. A universal-constraint
approach is certainly plausible in phonology, but the fact that syntactic form
is associated with semantic interpretation ... in decidedly language- and
dialect-particular ways stands in the way of a universal constraint approach to
syntax." (Zwicky 1994: 614)

To find out more about the interaction of language universals, their

innateness and language use implies that different linguistic schools seriously
and open-mindedly cooperate in trying to point out what properties of gram-
52 Walter Bisang

mar are due to innate structures and which properties are the result of
language use including language contact.

NOTES

1. It would be far beyond the scope of this paper to analyse whether there are parallels
between the diachronic rules of grammaticalization and synchronic hierarchies of inher­
itance and semantic networks.
2. As pointed out by Croft (1996), essentially the same mechanisms are also at work in
certain types of language internal changes. To distinguish this type of changes from
contact-induced inferences, Croft introduces the term intraference.
3. In Humboldt's own words: "Es darf also niemand auf andere Weise zum Andren reden,
als dieser, unter gleichen Umständen, zu ihm gesprochen haben würde" (From Keller
1990: 132).
4. Although not all the languages treated in Bakker & Mous (1994) seem to qualify as mixed
languages as defined by language intertwining there seem to be some clear instances (cf.
Stolz 1996).
5. This section briefly summarizes what I have tried to explain in much more detail with
many examples in Bisang (1991).
6. I shall not present a detailed description of relative clause formation in Classical Chinese
in this paper. In this footnote, I shall only present the most important characteristics of
relative clauses as far as they are necessary to understand my explanations on equational
constructions.
As in Modern Standard Chinese, relative clauses occur in front of the head noun
(also cf. §2.1). Relative constructions with subject coreference simply take the attributive
marker zhl between the head noun and the relative clause. With all the other instances of
coreference, an additional marker suŏ 'place' is used. For relative clauses with non-
subject coreference, we thus get the following pattern illustrated by (ii):
(i) (N3) {zhl) suö V (N2) zhl N1
(N3 = subject of the relative clause; N2 = object of V; N1 = head noun)
(ii) Zhông-zï suö jü zhl shì
Zhong-zi REL live_in ATTR house
'the house in which Zhong-zi is living' (Meng 3.B.10)
The subject of a relative clause with non-subject coreference can be marked again by the
attributive marker zhl. If we find a pronoun in that position, it is selected from the set of
possessive pronouns.
If there is no head noun, i.e. in headless relative clauses, the relative clause may be
marked by the nominalizer (NOM) zhě, which occurs at the end of a relative clause and
can be interpreted as [attributive marker zhl + N], where N very often can be replaced by
such general nouns as 'person', 'thing, fact, matter' or 'cause, reason':
Grammaticalization and language contact 53

(iii) [ai rén zhë] rén héng ai zhï.

love man/people NOM people always love OBJ:3
'The one who loves people people constantly love.' (Meng 4.B.28)
7. This sentence is at the very beginning of Confucius' statement. The equational marker yë
is thus clearly used to introduce a new topic.
8. According to Pulleyblank (1959: 180), wéi in the function of an affirmative copula is
attested since the period of the oracle-bone inscriptions (14th to 11th century B.C.).
9. This section is basically a summary of what I present in Bisang (forthc.a). For that reason,
it will be rather short.
10. The two principles are described as follows:
Relevance: "a meaning element is relevant to another meaning element, if the semantic
content of the first directly affects or modifies the semantic content of the
second." (Bybee 1985: 13)
Generality: "By definition, an inflectional category must be applicable to all stems of
the appropriate semantic and syntactic category and must obligatorily occur
in the appropriate syntactic context. In order for a morphological process
to be so general, it must have only minimal semantic content." (Bybee 1985:
16)
11. To define transnumerality more precisely, we may say that there is no pragmatic con­
straint to mark plurality in a language with transnumerality, so there is no Horn scale of
the type <PL, 0> (Horn 1984, 1988). Furthermore, there is no Quantity Implicature from
non-plural to singular.
12. For my areal typology cf. Bisang (1996).
13. Today, the character representing this word is pronounced shèng in its function of a
classifier and chéng in its function of a verb with the meaning 'ride, go on horseback;
travel'. In its function of a numerative, it is first used to individualize war chariots with
four horses. War chariots used to be of great importance to measure the military power of
a country.
14. The classifier liáng is closely linked to the numeral liăng 'two'. The numeral 'two' is
particularly relevant for chariots which typically had two wheels at the time of Classical
Chinese. The classifier liăng is thus an example of classification on the basis of a part-
whole relationship. Paleographically, the phonetic part of the character of liăang can be
compared to the classifier hing in the oracle bone inscriptions.
15. méi has nothing to do with bamboo as claimed by Erbaugh (1986: 428). In the Shuowen,
méi is interpreted as gàn 'tree trunk'. In the Shijing (Mao 10; 9th to 6th B.C.) we find the
sequence tiáo-méi which is translated as 'branches and slender stems' by Legge (1871:
17) without any reference to bamboo (cf. Legge, James. 1871. The Chinese Classics.
Volume IV: The She King). Finally, méi is written with the tree-radical, not with the
bamboo radical.
16. For some examples cf. Zuo, Zhaogong 12 and Zuo, Xianggong 21 (compilated in the 5th
and 4th century B.C.).
54 Walter Bisang

17. Thus, gè was maybe more widespread even before the Tang dynasty (618 - 907 A.D.) than
Wang Li (1958: 238) assumes.
18. The term 'class noun' corresponds to 'class term' suggested by Haas (1942) and
DeLancey (1986).
19. Q = Quantifier, intQ = intrinsic quantifier. Intrinsic quantifiers refer to the shape in which
a mass noun occurs: a coil of wire, a lump of sugar.
20. Since there are no final consonants in Hmong syllables, the syllable final position is used
to mark tones by using some consonants. In this example, zaj changes its tone into zag
because of the tone of ob 'two', which may yield certain phenomena of tone sandhi.
21. The findings on Vietnamese point to the semantic closeness of taxonomy and meronomy
as described by Cruse (1995; also quoted by Löbel forthc. 21):
"Any taxonomy can be thought of in part-whole terms (although the
converse is not true): a class can be looked on as a whole whose parts
are its subclasses" (Cruse 1995: 179).
As I tried to show in Bisang (forthc. b), the data in Vietnamese do not allow for such a
conclusion without any problems.
22. I don't know enough about this field to judge any of Wills' stimulating ideas, but the
simple fact that the majority of the human brain is developed outside of the uterus seems
me to be interesting enough.

REFERENCES

Adams, Karen L. and Nancy F. Conklin. 1973. "Toward a theory of natural classification."
Papers from the 9th Regional Meeting, Chicago Linguistic Society, 1-10.
Allan, Keith. 1977. "Classifiers." Language 53: 285-311.
Andersen, Henning. 1973. "Abductive and deductive change." Language 49: 765-93.
Bakker, Peter and Maarten Mous (eds). 1994. Mixed Languages. 15 Case Studies in
Language Intertwining. (Studies in language and language use, 13). Amsterdam:
Institute for Functional Research into Language and Language Use.
Bisang, Walter. 1986. "Die Verb-Serialisierung im Jabêm." Lingua 70: 131-162.
. 1991. "Die Hervorhebung im Chinesischen: Zur diachronen Kontinuität des
Äquationalsatzmusters im Rahmen der Diskurspragmatik". In Bisang, Walter and Peter
Rinderknecht, eds. Von Europa bis Ozeanien, von der Antonymie zum Relativsatz -
Gedenkschrift für Meinrad Scheller, 39-75. Zürich: University of Zürich, Seminar of
General Linguistics.
. 1992. Das Verb im Chinesischen, Hmong, Vietnamesischen, Thai und Khmer
(Vergleichende Grammatik im Rahmen der Verbserialisierung, der Gramma-
tikalisierung und der Attraktorpositionen). Tübingen: Narr.
. 1993. "Classifiers, quantifiers and class nouns in Hmong." Studies in Language
17:1,1-51.
Grammaticalization and language contact 55

. 1996. "Areal typology and grammaticalization: Processes of grammaticalization

based on nouns and verbs in East and mainland South East Asian languages." Studies in
Language, 20.3, 519-597.
. Forthc. a. "Verb serialization and attractor positions: Constructions and their
potential impact on language change and language contact." In Kulikov, Leonid and
Heinz Vater, (eds). Festschrift for V. Nedjalkov.
. Forthc. b. "Classifiers in East and Southeast Asian languages. Counting and
beyond." In: Jadranka Gvozdanovic (ed.) Change in Numeral Systems. Berlin: Mouton
de Gruyter.
Bradshaw, Joel. 1979. "Obstruent harmony and tonogenesis in Jabêm." Lingua 49,189-205.
Bybee, Joan L. 1985. Morphology: A study of the relation between meaning and form.
Amsterdam/Philadelphia: John Benjamins.
Bybee, Joan, Revere Perkins, and William Pagliuca. 1994. The Evolution of Grammar.
Tense, Aspect and Modality in the Languages of the World. Chicago and London: The
University of Chicago Press.
Chao. Yuan Ren. 1968. A grammar of Spoken Chinese. Berkeley and Los Angeles:
University of California Press.
Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago & Lon­
don: The University of Chicago Press.
. 1994. "Semantic universals in classifier systems." Word 45.2, 145-171.
. 1996. "Bringing chaos into order: Mechanisms for the actuation of language
change." MS. 61 pp.
Déchaine, Rose-Marie. 1993a. Predicates across categories: Towards a category neutral
syntax. Amherst, Mass. GLSA. (Doctoral dissertation, Univ. of Massachusetts.)
. 1993b. "Serial verb constructions." In J. Jacobs, et al. eds. Syntax. An international
handbook of contemporary research, 799-825. Berlin: de Gruyter.
DeLancey, Scott. 1986. "Towards a history of Tai classifier systems". In Colette Craig
(ed.), Noun Classes and Categorization, 437-452. Amsterdam/Philadelphia: John
Benjamins.
Dempwolff, Otto. 1939. Grammatik der Jabêm-Sprache auf Neuguinea. Hamburg:
Friedrichsen, de Gruyter.
Denny, Peter J. 1976. "What are noun classifiers good for?" In Papers from the 12th
International Meeting, Chicago Linguistic Society, 122-132.
Dik, Simon C. 1989. The Theory of Functional Grammar, Part I: The structure of the
clause. Dordrecht: Foris.
Dryer, Matthew S. 1992. "The Greenbergian word order correlations." Language 68, 81-
138.
Erbaugh, Mary S. 1986. "Taking stock: The development of Chinese noun classifiers
historically and in young children." In Colette Craig (ed.) Noun Classes and Categori­
zation, 399-436. Amsterdam/Philadelphia: John Benjamins.
Fillmore, Charles J., Paul Kay, and Mary Catherine O'Connor. 1988. "Regularity and
idiomaticity in grammatical constructions: The case of let alone." Language 64.3, 501-
538.
Gabelentz, Georg von der. 1881. Chinesische Grammatik (Mit Ausschluss des niederen
Stiles und der heutigen Umgangssprache). Reprinted 1960, Halle (Saale): VEB Max
Niemeyer.
56 Walter Bisang

Gassmann, Robert H. 1980. Das grammatische Morphem ye (Eine Untersuchung seiner

syntaktischen Funktion im Menzius). Bern/Frankfurt/Las Vegas: Lang.
Giles, Howard and Philip M. Smith. 1979. "Accommodation theory: Optimal levels of
convergence." In Howard Giles and Robert N. St Clair (eds), Language and Social
Psychology. Oxford: Basil Blackwell.
Goldberg, Adele. 1995. A Construction Grammar approach to argument structure.
Chicago & London: University of Chicago Press.
Graham, A.C. 1959. "'Being' in western philosophy compared with shihlfei and yulwu in
Chinese philosophy," Asia Major 7, 79-112.
Greenberg, Joseph. 1974. "Numeral classifiers and substantival number: Problems in the
genesis of a linguistic type." Proceedings of the 11th International Congress of Linguis­
tics, Bologna - Florence, Aug-Sept 1972, Bologna: 1974: 17-37.
Haas, Mary. 1942. "The use of numeral classifiers in Thai." Language 18, 201-205.
Hashimoto, Anne Yue 1969. "The verb to be in modern Chinese." Foundations of Lan­
guage Supplementary Series 9.4, 72-111. Dordrecht: D. Reidel.
Heine, Bernd, Ulrike Claudi, and Friederike Hünnemeyer. 1991. Grammaticalization: A
conceptual framework. Chicago: University of Chicago Press.
Hopper, Paul J. and Elizabeth Closs Traugott. 1993. Grammaticalization. Cambridge:
Cambridge University Press.
Hopper, Paul J. 1986. "Some discourse function of classifiers in Malay." In Colette
Craig (ed.), Noun classes and categorization, 309-325. Amsterdam/Philadelphia:
John Benjamins.
Horn, L.R. 1984. "Toward a new taxonomy for pragmatic reference: Q-based and R-
based implicature." In D. Schiffrin (ed.), Meaning, form, and use in context: Linguis­
tic applications, 11-42. Washington, DC: Georgetown University Press.
. 1988. "Pragmatic theory." In F.J. Newmeyer (ed.) Linguistics. The Cambridge
survey. Volume 1, 113-145. Cambridge: Cambridge University Press.
Iljic, Robert. 1987a. L'exploitation aspectuelle de la notion de franchissement en chinois
contemporain. Paris: Editions L'Harmattan.
. 1987b. La marque de détermination nominale 'de' en chinois contemporain
(implications temporelles). Paris: L'Harmattan.
Keller, Rudi. 1990/1994. On Language Change: The invisible hand in language. London:
Routledge. (Translation and expansion of Sprachwandel).
Kryukov, M.V. 1980. The language of Yin inscriptions. Moscow.
LePage, R. B. and Andrée Tabouret-Keller. 1985. Acts of identity. Cambridge: Cam­
bridge University Press.
Li, Charles N. and Sandra A. Thompson. 1974a. "An explanation of word order change
SVO - SOV." Foundations of Language 12.2, 201-14.
. 1975. "The semantic function of word order: A case study in Mandarin." In
Charles N. Li (ed.),Word Order and Word Order Change, 163 - 195. Austin and
London: University of Texas Press.
. 1977. "A mechanism for the development of copula morphemes." In Charles N. Li
(ed.), Mechanisms of syntactic change, 419-444. Austin/London: University of Texas
Press.
. 1981. Mandarin Chinese (afunctional reference grammar). Berkeley, Los Ange­
les and London: University of California Press.
Grammaticalization and language contact 57

Lightfoot, David. 1989. "The child's trigger experience." Brain and Behavioral Sciences
12, 321-75.
Liu, Shiru. 1965. Wei-Jin Nanbeichao liangci yanjiu [A study on classifiers in the Wei-Jin
and in the Nanbeichao periods]. Beijing: Zhonghua shuju chuban.
Löbel, Elisabeth. 1996. Klassifikatoren. Eine Fallstudie am Beispiel des Vietnamesischen.
Universität Köln: Habilitationsschrift. Ms.
. Forthc. "Classifier systems vs. noun class systems: A case study in Vietnamese." In
Barbara Unterbeck (ed.), Approaches to gender. Berlin: Mouton de Gruyter.
Lord, Carol. 1974. "Causative constructions in Yoruba." Studies in African Linguistics
Supplement 5, 195-204.
. 1993. Historical change in serial verb constructions. Amsterdam/Philadelphia:
John Benjamins.
Michaelis, Laura A. and Knud Lambrecht. 1996. "Towards a construction-based theory of
language function: The case of nominal extraposition." Language 72.2, 215-247.
Peyraube, Alain. 1988. Syntaxe diachronique du chinois (évolution des constructions
datives du XlVe siècle av. J.-C. au XVIIIe siècle). Paris: Collège de France, Institut des
hautes études chinoises.
Pulleyblank, E.G. 1959, "fêi, wéi and certain related words." In: S. Eger0d and E. Glahn
(eds), Studia Sérica Bernhard Karlgren Dedicata. Kopenhagen.
Qiu, Xi gui. 1979. "Tantán guwenzi ziliao dui guhanyu yanjiu de zhongyaoxing [On the
importance of ancient inscriptions in the study of old Chinese]." Zhongguo yuwen, 437-
442.
Ross, Claudia. 1983. "On the function of Mandarin de." Journal of Chinese Linguistics
11.2, 214-46.
Searle, John. 1969. Speech Acts: An essay in the philosophy of language. Cambridge:
Cambridge University Press.
Silva-Corvalán, Carmen. 1994. Language contact and change: Spanish in Los Angeles.
Oxford: Clarendon Press.
Sonaiya, Rémi and Walter Bisang. Forthc. "The function of the high tone syllable in
Yoruba."
Stolz, Thomas. 1996. Review article: Peter Bakker & Maarten Mous (eds). Mixed lan­
guages. 15 case studies in language intertwining (Studies in language and language use
13). Amsterdam: Institute for Functional Research into Language and Language Use, in:
Sprachtypologie und Universalienforschung 49.4, 383-386.
Sun, Chao-fen and Talmy Givón. 1985. "On the so-called SOV word order in Mandarin
Chinese: A quantified text study and its implications."Language61, 329-351.
Suppes, Patrick and Joseph L. Zinnes. 1963. "Basic measurement theory." In R. Luce et al.
(eds), Handbook of Mathematical Psychology, Vol. 1, 3-76. New York & London:
Wiley.
Takashima, 1984/1985. "Noun phrases in the oracle-bone inscriptions." Monumenta Sérica
36,229-301.
Thomason, Sarah and Terence Kaufman. 1988. Language contact, creolization, and
genetic linguistics. Berkeley: University of California Press.
Trudgill, Peter. 1986. Dialects in contact. Oxford: Basil Blackwell.
Wang, Li. 1958. Hanyu shigao [A draft history of Chinese]. Beijing.
58 Walter Bisang

Wang, Mingquan. 1987. Transitivity and the Ba-Construction in Mandarin. Ph.D. Disserta­
tion. Boston University, Graduate School.
Weinreich, Uriel. 1953/1968. Languages in Contact: Findings and problems (2nd edition).
The Hague: Mouton.
Wiese, Heike. 1995. "Semantische und konzeptuelle Strukturen von Numeral­
konstruktionen (Untersuchung zu Kardinal-, Ordinal- und Nummer-Konstruktionen)."
Zeitschrift für Sprachwissenschaft 14.2, 180-235.
Wills, Christopher. 1993. The runaway brain; The evolution of human uniqueness. New
York: Basic Books.
Yen, Sian L. 1971, "On negation with fei in Classical Chinese." Journal of the American
Oriental Society 91.3, 409-17.
. 1986. "The origin of the copula shî in Chinese." Journal of Chinese Linguistics
14.2, 227-41.
Zahn, Heinrich. 1940. Lehrbuch der Jabêm-Sprache. Beiheft zur Zeitschrift für Ein­
geborenen-Sprachen Nr. 21. Berlin: Dietrich Reimer.
Zürcher, E. 1977. "Late Han vernacular elements in the earliest Buddhist translations."
Journal of the Chinese language Teachers Association 12.3, 177-203.
Zwicky, Arnold. 1994. "Dealing out meaning: Fundamentals of syntactic constructions."
Berkeley Linguistics Society 20, 611-625.
 Grammaticalization
and clause linkage strategies
A typological approach with particular reference
to Ancient Greek

Sonia Cristofaro
University of Pavia

1. Introduction

Current debate has highlighted the existence of two different approaches to

grammaticalization: the lexical, or sentence-based approach, and the dis­
course-based approach. According to the former, which rests upon the classi­
cal definitions proposed by Meillet (1912) and Kurylowicz (1965),
grammaticalization is a very specific phenomenon exclusively pertaining to
the development of grammatical items and constructions out of lexical or
already grammaticalized ones. According to the latter approach, which is
primarily based on the work of Givón (1979), grammaticalization must be
considered as a broad phenomenon pertaining to the development of gram­
matical structures in general: these may take as their source not only indi­
vidual lexical items, but also discourse-based strategies in which the
individual features of the words or morphemes are subordinated to the prag­
matic function of the device as a whole (Herring 1991). Objections to this
type of approach mainly concern the problem of defining the notion of
discourse in a consistent enough way (Heine, Claudi and Hünnemeyer 1991:
238; Herring 1991: 278); adherents to it have however produced a large body
of evidence suggesting that, despite the lack of a methodology as sophisti­
cated as the one used in the field of historical semantics, there is some
theoretical justification in arguing that pragmatic, discourse-based strategies
60 Sonia Cristofaro

play an important role in grammaticalization phenomena (cf. for instance

Givón 1979; Herring 1991; Hopper and Traugott 1993: chap. 7).
The aim of this paper is to examine the synchronic distribution and the
diachronic evolution of the most common Ancient Greek complementizers,
ÖTL and œç, and to show that their various functions, apparently independent
from each other, can be connected as the result of a general process of
semantic transfer. This process partially falls within the domain of grammati­
calization defined in the sentence-based, or lexical sense, and offers on the
other hand a clear instance of grammaticalization in the discourse-based
sense. and ύ are two pronominal forms, anaphoric in function, that
come to be used as complementizers and adverbial subordinators; they subse­
quently undergo a number of phenomena of pragmatic strengthening of infor-
mativeness (in the sense defined in Traugott and König 1991) that lead them
outside their original domain; they eventually lose their semantic, pragmatic
and occasionally syntactic features, and are sometimes replaced by new
forms. This process is not specific to Ancient Greek. The development of
complementizers and subordinators in general out of full lexical items such as
nouns and verbs and relational items such as pronouns, adpositions and case
markers is widely attested cross-linguistically, and has become a standard
topic in the discussion concerning grammaticalization (Genetti 1991; Givón
1979; Herring 1991; Justus 1976 and 1980; Lord 1976; Noonan 1985; Ran­
som 1988; Saxena 1988). On one hand, it clearly exemplifies the grammati­
calization of lexical items or the increased grammaticalization of already
grammaticalized ones; on the other hand, it offers some evidence that gram­
maticalization must be considered as a broad parameter pertaining to the
reanalysis not only of lexical as grammatical material, but also of discourse
patterns as grammatical patterns and of discourse-level functions as sentence-
level, semantic functions.
It will be argued that, in comparison with traditional instances of gram­
maticalization in which lexical elements take on a grammatical function and
become bound morphemes of reduced phonetic substance, the development of
and ύ (as well as that of other complementizers in various languages)
represents an atypical case. At the beginning of the process, it is possible to
identify some of the features that, according to Hopper (1991), characterize
initial stages of grammaticalization phenomena (layering, specialization, per­
sistence and, more generally, a shift in meaning from a more concrete domain
to a more abstract one). At a later stage, although there is some evidence that
Grammaticalization and clause linkage strategies 61

the process has come to an end, none of the features characterizing final
stages of grammaticalization (condensation, coalescence, paradigmaticiza-
tion, complete loss: cf. Lehmann 1982 and 1985) can be detected. Besides,
one of the outcomes of the process, the adverbial expression /
"clearly, obviously", cannot be considered as a case of grammati­
calization, and the processes from which it originates are not specific for
grammaticalization either. What is more, while the shift from open lexical
categories such as nouns and verbs to closed relational categories such as
adpositions, conjunctions, case markers etc. clearly falls within the traditional
definition of grammaticalization, it is somewhat controversial whether a shift
from a closed relational category (pronouns) to another (subordinators) can
also be included within it.
On the other hand, the development of and as well as similar
cases in the languages of the world, represents a clear instance of grammati­
calization in the discourse-based sense: clause linkage strategies operating on
the syntactic level take as their source pragmatic and semantic strategies
operating on the discourse level. It will be argued that it is legitimate to
assimilate such phenomena to more traditional, sentence-based instances of
grammaticalization because both appear to be governed by the same prin­
ciples: the expression of more abstract domains in terms of more concrete
ones and, more generally, the coding of autonomous and fixed grammatical
relations on the basis of context-dependent strategies.
The paper is organized as follows: section 2 contains some instances of
complementizer development in some unrelated languages; in section 3 the
Greek data, selected from a corpus of literary texts ranging from the Homeric
age to the third century1, are presented; in section 4 some basic theoretical
issues concerning grammaticalization are discussed, and a connection be­
tween grammaticalization and the organization of clause linkage strategies is
established.

2. Some instances of complementizer development

2.1. Theoretical premise

Complementizers are subordinating conjunctions that introduce complement

and sometimes adverbial clauses. The distinction between complement and
62 Sonia Cristofaro

adverbial clauses adopted here reflects the one proposed within Dutch func­
tional grammar between arguments and satellites (Dik 1989: 47-9; Hengeveld
1990: 18-9). Complement clauses represent an argument, subject or object, of
a main predicate: that is, they provide a specification required by the semantic
features of the predicate. Adverbial clauses are satellites, that is they provide
some additional information about the event coded by the main predicate:
unlike complement clauses, they are not necessary for the definition of the
event itself.
Complementizers typically derive from nouns, pronouns, verbs, adposi-
tions and case markers (Noonan 1985: 47; Ransom 1988). The development
from verb to complementizer and from pronoun to complementizer will be
examined here on the basis of data from Banda Linda, an SVO North Central
Niger-Congo language, Biblical Hebrew and Hittite.

2.2. From verb to complementizer: Banda Linda

In Banda Linda (Cloarec-Heiss 1986) complement and main clauses may be

juxtaposed with no formal marker signaling their linkage:
(1) èyi.ngérà pä ènje ?î tèrd gàti.
chief say they attach:INJ Tere to the ground
T h e chief orders to bind Tere.'
(Cloarec-Heiss 1986: 498)
Complement clauses can also be introduced by the complementizer dpa:
this is a frozen form consisting of the verb pā, 'say' and the pronominal
element used for third person inanimate subjects or for animate subjects
whenever they are coreferential with the subject of the preceding clause
(Cloarec-Heiss 1986: 72-3). The original meaning of is approximately
'SS says', as can be seen in constructions like
(2) nà né; ko cé gú.
go:iNJ you:ss say:iNJ it to him ss-say I arrive:AC
'Go and tell him that I arrived.'
(Cloarec-Heiss 1986: 499)
This sequence can be analyzed as consisting either of independent juxta­
posed clauses Cgo, say it, tell him, I arrived') or of two independent clauses
and a complement clause ('go and tell him that I arrived').
Grammaticalization and clause linkage strategies 63

can however be used also when the overall sense of the sentence is
incompatible with its original verbal meaning 'say':

(3) ?à amùnju
we know:AC it know:NEG that white:PL be:AC
jèkôci nē.
on the other side NEG
'We didn't know that the white men lived on the other side.'
(Cloarec-Heiss 1986: 500)

It is clear that in this case, owing to the semantic features of the main
predicate 'know', ^ cannot play in the sentence the role of an
autonomous verb form 'say' and must necessarily be interpreted as a comple­
mentizer. This is a still ongoing process of generalization that is leading 6pä to
become a general subordinator, not restricted to the domain of complementa­
tion. For instance, it may sometimes be employed with predicates that do not
require a complement clause:

(4) . . . ê
certain thing that it:ss banish:PERF you that to you
gù nàè ko nid wiwi rid.
happemsucc CENTR DEICT I see:NEG NEG
T can't see anything that might happen to you and force you to
leave.'
(Cloarec-Heiss 1986: 501)

The reanalysis verb → complementizer →( general subordinator) seems to be

triggered, in the languages where it takes place2, by two types of clause
linkage strategies: parataxis (as in (1) and (2)) and verb serialization. Both are
based on the juxtaposition of the verbs coding main and subordinate events.
While in parataxis each verb represents an independent assertion, can be
independently negated and has no tense, aspect or mood restrictions, serial­
ized verbs usually agree in tense, aspect, mood and person, and cannot be
independently negated (Noonan 1985). Clause linkage is signalled by no
formal means, and can only be inferred from the context. The verbs that give
origin to complementizers are initially employed in paratactic or serialized
constructions in order to indicate clause linkage on semantic and pragmatic
grounds; at a later stage, they are reanalyzed as connectors operating on the
syntactic level, and the clause they introduce is downgraded to a non-main
64 Sonia Cristofaro

status. This seems to be a sign of a general tendency to reduce the number of

predicates in main clauses: multiple-verb constructions are transformed into
single-verb clauses, and new syntactic categories (complementizers) must be
introduced in order to code the connections formerly indicated by parataxis or
verb serialization. This process is not restricted to complementizers, but
involves adpositions and case markers: in a number of languages that use
serialization or parataxis, these too may arise from verbs (Givón 1979: 220-1;
Lord 1976).

2.3. From relative pronoun to complementizer: Biblical Hebrew

In Early Biblical Hebrew (Givón 1979: 219 and 1991) complement clauses
are usually introduced by two different complementizers, ki and ve-hine. The
former, which probably derives from a locative preposition, functions as a
time-adverb marker, a conditional and a comparative marker. As a comple­
mentizer, it is used to convey highly predictable or presupposed information:
(5) va-yar' 'elohim ki-tov.
and-see:iMPF:3sG:M God coMP-good
'And God saw that it was good.'
(Givón 1991: 272)
Ve-hine is a compound of the conjunction v-, 'and' and an archaic verb
*hn, 'be'. It is used to convey surprising, non-presupposed information:
(6) va-yar' ve-hine harvu pney ha-'adamah.
and-see:IMPF:3sG:M and-be dry:PERF:3PL:M faces ofthe-earth
'And he saw that the surface of the earth had dried up.'
(Givón 1991: 272)
Verbs such as see, think or know however display another complementa­
tion strategy: the verb has a direct-object nominal complement followed by a
non-restrictive relative clause, as in
(7) lal tirlu-ni she-lani shaxoret.
NEG see:IMPF:3sG:M-me REL-I dark:SG:F
'Don't see me that I am dark-skinned.'
(Givón 1991: 289)
The relative clause is added to the main clause as an afterthought, and has
Grammaticalization and clause linkage strategies 65

a high communicative value: it conveys a specification about the main clause

object, and this specification represents the communicative focus of the
sentence. This is the reason why, at a later stage, the whole relative clause is
reanalyzed as the actual object of the main predicate, triggering the transfor­
mation of the relative pronoun she/asher into a complementizer, and the
elimination of the direct-object nominal complement of the main verb:
(8) yada"ti she-gam zeh hu ra"yon ruah.
know:PERF:lsGCOMP-also this it notion of wind
'I knew that this too was total nonsense.'
(Givón 1991: 297)
By Late Biblical Hebrew, the use of ve-hineh has been consistently
reduced, and ki, which appears numerically predominant, spreads to contexts
where ve-hineh was formerly employed. The introduction in the complement
system of she/asher appears then to affect ki alone (Givón 1991: 275).

2.4. From focus marker to complementizer: Hittite

In Hittite, at quite late stages, complement clauses may be introduced by the

complementizer kuit, which is also used for reason and time clauses (Justus
1976, 1980; Lehmann 1980):
(9) [KINUN?/GIM?]-ya IDI kuit-za KUR
[now? /when?]-PTCL know-lSG COMP-PTCL land
uru
Mizri KUR uruHatti 1ENKUR™ kisari.
Egypt land Haiti one land become-3sG
'Now I know that the land of Egypt and the land of Hatti are
becoming a single land.'
(Justus 1980: 195)
Kuit originates from the particle ku-, which derives from the Indo-
European relative stem KW-. There is evidence to establish that ku- was
originally used as a focus marker in topical clauses3: the element it referred to
was taken up by a series of anaphoric particles in the following clauses. This
construction type, which is quite common in the early stages of many Indo-
European languages and is usually called correlative diptych (Haudry 1973),
can be illustrated by the following Hittite sentence:
66 Sonia Cristofaro

(10) nu kasô-kuit memai n-at zik sakti.

PTCL this one-what speak-3sG PTCL-it you know-2sG
'You will know what this one says.'
[Literally: 'What this one says, you will know it']
(Justus 1980: 196)
The original function of ku- is to signal the pragmatic connection and the
mutual relevance of a series of clauses; subsequently, it is reanalyzed as a
connector signaling clause linkage on the syntactic level, and acquires the
value of a general subordinator.

3. The Ancient Greek case

3.1. Origin of and

Ancient Greek most common complementizers, and derive from the

same Indo-European pronominal stem YO-, whose sense was presumably
relative-deictic: on originates from an accusative form, œç is probably instru­
mental, although the reconstruction is somewhat controversial (Monteil 1963:
243, 327-9; Schwyzer and Debrunner 1945: 645).4 Both forms are used to
introduce complement as well as adverbial clauses.

3.2. Distribution and diachronic development of

',
Oti is first attested in the Homeric poems, and initially occurs in two
construction types: complement clauses with a pronominal antecedent in the
main clause (correlative diptych structures of the type exemplified in (10))
and complement clauses with no pronominal antecedent:

(11)
know:PRES-IND-lSG in fact DEM:ACC:PL:N
on
plead in defence-FUT-INF:MD DEM:ACC:SG:M COMP DEM-DAT:SG:M

friend-NOM:SG:M be:IMPF:3sG and ART:GEN:PL:N DEM-GEN:PL:N

act-GEN:PL take part:IMPF-3sG:ACT

Grammaticalization and clause linkage strategies 67

'In fact, I know this, that he will plead in defense that he was that
man's friend, and took part in the same acts.'
(Lysias, Against Eratosthenes, 62.5)
(12)
NEG see:PERF:IND-3PL COMP Hippias:NOM PTCL

be supreme:iMPF-3sG:ACT
They don't know that it was Hippias... who actually was supreme.'
(Thucydides, 1.20.2.4)
On the basis of the evidence offered by other Indo-European languages
(Hittite, for which cf. (10) above, Latin, Germanic languages; several ex­
amples can be found in Haudry 1973) it appears reasonable to hypothesize
that the correlative diptych in (11) represents the older construction. c/On is
first used as a pronominal correlative, which refers anaphorically to a corefer-
ential topical pronoun in the preceding clause and introduces a focalized
specification about it; the linkage between the two clauses is signaled by the
semantic relation of coreference between the two pronominal elements, as
well as by their mutual pragmatic relevance. Then on is reanalyzed as a
connector operating exclusively on the syntactic level, and comes to be used
with no pronominal antecedent as in (12).
After Homer, on starts to be used to introduce reason clauses:

(13)
much-GEN:PL:N PTCLbecause of NEG kill-FUT:IND:lsG:ACT

he:ACC:SG and because DEM:DAT:SG:M I:DAT relative-NOM:SG:M

eon.
be:PRES:IND-3sG
'I will not kill him for many reasons, and because he is related to
me.'
(Herodotus, 1.109.9)
This function probably originated in complement clauses introduced by
factive predicates. Besides being the object of the main predicate, these
clauses refer to an event which represents the ground for the occurrence of the
event coded by the main predicate, so that both a complement and a causal
reading are possible:
68 Sonia Cris tof aro

(14)
be glad:PRES-PART:NOM:SG:M:ACT Antilochus-DAT coMP/because

he:DAT dear-NOM:SG:M be:IMPF:3sG friend-NOM:SG

'Being glad that/because Antilochus was his dear friend'
(Homer, Iliad 23.556)
On the basis of such contexts, the use of on may have been extended to
unequivocal reason constructions such as the one in (13).
In the fourth century B.C. a new expression becomes increasingly com­
mon: the adverb originating from the merging of on
and the adjectival predicate 'clear, evident, obvious':
(15)

big-NOM:sG:N PTCL evidence-NOM:SG ART:DAT:SG:M

argument-DAT:SG DEM:NOM:SG:M arrive:PRES-IND-3sG

escape:PRES-PART:NOM:SG:M:ACT at great speed ART:NOM:SG:N

old age:NOM:sG swift:NOM:SG:N be:PRES:PART:NOM:SG:N

obviously
'He himself supplies clear evidence of this; for he flies and flees
from old age — a swift thing obviously'
(Plato, Symposium, 195."b".2)
The merging process involves no predicates but ôfj Xov, and cog, which
may also occur with this predicate, is not affected by it. All the functions of on
listed in (11)-(15) coexist at this stage.
3.3. Distribution and diachronic development of (bç
The distribution of cog is much wider than that of on, and only the cases most
relevant for the purpose of the discussion will be examined here.5 Since the
Homeric age, cbg is used as a subordinator introducing purpose ((16)), result
((17)), time ((18)) and manner clauses ((19)), as well as a comparative
morpheme ((20)):
Grammaticalization and clause linkage strategies 69

(16)

but stay(:PRES-IMP:2sG:ACT) till PTCL PTCL

honey-sweet:ACC:SG:M wine:ACC bring:AOR-suBJ: 1SG:ACT

in order for pour libation-AOR:suBJ:2sG Zeus-DAT

father-DAT and other-DATiPLiM immortal-DAT:PL:M
'But stay till I have brought honey-sweet wine in order for you to
pour libation to Zeus, the father, and to the other immortals.'
(Homer, Iliad, 6.259)
(17)

NEG in fact farmstead-DAT:PL stay-PRES:INF:ACT any longer

so young:NOM:SG:M be:PRES:iND-lSG so that PTCL

order:AOR-PART-DAT:SG*.M:MD overseer-DATall-ACc:PL:N

obey:AOR-INF
'For I am no more of an age to remain at the farmstead, so as to
obey in all things the command of an overseer. '
(Homer, Odyssey, 17.21)
(18)
Hector:NOM PTCL as NEG within find:AOR:iND-3sG
äxoLTiv (...)
spouse:ACC say:AOR:IND-3sG:ACT
'Hector, as he did not find his spouse within (...) said (...)'
(Homer, Iliad, 6.374-5)
(19)
DEM:ACC:PL:N PTCL SO PTCL perform:FUT-IND:lsG:ACT

old man:voc as you:NOM order:PRES-IND-2sG:ACT

'I will perform this, old man, exactly as you order.'
(Homer, Odyssey, 4.485)
70 Sonia Cristofaro

(20)

grant:AOR:IMP-2PL PTCL too DEM:ACC:SG:M become:PERF-INF

son-ACc:sG my-ACC:SG:M as too I:NOM PTCL valiant-ACc:sG:M

Trojan-DAT:PL
'And grant also that this son of mine may become as valiant as I am
among the Trojans.'
(Homer, Iliad, 6.477)
All these functions appear to be semantically connected to the original instru­
mental/ablative meaning of ύ 'by means of which, the way in which'. Being
a relative form, ύ basically establishes a parallelism between two processes:
these may be related either because they take place according to similar or in
some way interdependent patterns, or because the participants involved in
them display similar features (Monteil 1963: 330-5). In purpose and result
constructions, the event coded by the subordinate clause provides an explana­
tion for the main event: in fact, the occurrence of the former represents the
means through which the latter can take place. It is therefore not surprising
that the clauses coding the two events are linked by means of an instrumental
form. In time constructions, the relation of temporal adjacency holding be­
tween the main and the subordinate event provides the semantic ground that
makes it possible to establish a parallelism between them: if two events are
contiguous in time, they can be inferred to occur according to the same
patterns, and possibly to be causally related. Finally, the existence of a
semantic parallelism involving the way in which two events take place is
particularly evident in the manner construction in (19) and in the comparative
construction in (20): here œç explicitly signals that the main and the subordi­
nate event take place according exactly to the same pattern ((19)) or that the
participants involved in them display exactly the same features ((20)).
It is now necessary to account for the functional shift that transforms ύ
from an instrumental/ablative form of a pronominal stem into a subordinator.
The means for the reanalysis may have been constructions such as (16), where
the sense of ύ may be referred either to the whole main clause, or to a single
constituent of it, in this case otvov, so that the alternative reading in (21)
becomes possible:
Grammaticalization and clause linkage strategies 71

(21)

but stay(:PRES-IMP:2sG:ACT) till PTCL PTCL

honey-sweet:ACC:SG:M wine:ACC:SG bring: AOR-SUBJ:1SG:ACT

by means of which/with which pour libation:AOR-suBJ:2sG:ACT

Zeus-DAT father:DAT and other-DAT:PL immortal-DAT:PL

'But stay till I have brought to you honey-sweet wine by means of
which/with which you may pour libation to Zeus, the father, and to
the other immortals.'
(Homer, Iliad, 6.259)
According to this interpretation, ύ operates here on the semantic and the
pragmatic as well as on the syntactic level: it is an anaphoric pronoun referring
back to an item in the preceding clause, and the whole construction has a
correlative diptych structure. It appears then probable that the same kind of
process described for on took place: at a certain point in its history, ύ was
reanalyzed as a connector operating exclusively on the syntactic level, and
was used with no antecedent in the main clause. In fact, cases like (21), where
ύ has a twofold reading, rapidly decrease in frequency after Homer.
Ωç can also be used in complement ((22)) and reason ((23)) clauses:
(22)
COMP PTCL injure:IMPF:3PL:ACT clear:NOM:SG:N
'It's clear that they were injuring us.'
(Thucydides, 1.34.2.1)
(23)
After PTCL PTCL:ACC:SG:N as no one:NOM:sG:M they:DAT
(...)
sail:IMPF-3sG:ACT they-NOM sail:IMPF-3PL:ACT against

ART:ACC:PL:M barbarian-ACC:PL:M
'After this, as no one sailed against them, (...) they themselves
advanced their ships against the barbarians.'
(Herodotus, 8.9.6)
These functions are already attested, albeit quite sporadically, in the
72 Sonia Cristofaro

Homeric poems; their frequency gradually increases from Homer up to the

fifth century B.C.
The use of ύ as a complementizer appears rather problematic, for
complement clauses are subjects or objects of main predicates, and can
therefore hardly be related to an oblique instrumental form. This function may
have as its source constructions such as
(24)
sing :IMPF-3SG: ACT how fortress:ACC destroy:AOR-3sG:ACT

son-NOM:PL Achean-GEN:PL
'He sang how the sons of the Acheans destroyed the fortress.'
(Homer, Odyssey, 8.514)
The communicative focus is placed here on how the subordinate event took
place, rather than on the fact that it actually took place: ύ means 'how, in
which way', so that an instrumental/ablative meaning can still be recovered
from the context. If one describes how an event took place, however, one
usually presupposes that it actually took place (in fact, the dependent event in
(24) is already known to the hearer). This may lead to a neutralization of the
semantic distinction between 'how' and 'that', and this is the reason why ύ
may come to be used with the meaning 'that' in complement constructions like
(22).
As for the use of ύ in reason clauses, Monteil (1963: 358-60) suggests a
development pattern similar to the one proposed in section 3.2 for oTi: this
function originates from constructions introduced by factive predicates,
where both a complement and a causal reading are possible, as in
(25)
rejoice:PRES-IND:3sG:ACT PTCL I:DAT hearth:NOM how I:GEN

always PF:remember-IND:2sG valorous-GEN:SG:M

'My hearth rejoices seeing that/because you remember that I used
to be a valorous man.'
(Homer, Iliad, 23.648)
Cross-linguistic evidence shows however that reason constructions are
often derived from time constructions by means of pragmatic inferences: if
two events are mentioned together as being simultaneous or adjacent in time
Grammaticalization and clause linkage strategies 73

they are often inferred to be causally related (Thompson and Longacre 1985:
179; Traugott and König 1991: 194-9). Reason clauses appear related to
purpose clauses too: both provide an explanation for the occurrence of the
main event. Many languages use for the two the same morphology (Thomp­
son and Longacre 1985: 185). Complement clauses, then, are not the only
possible source for the use of ύ in reason constructions: this may have
originated from purpose or time constructions, which represent the original
domain of this form.

3.4. The distribution of and in complement clauses

3.4.1. Semantic and pragmatic constraints

The distribution of on and ύ in complement clauses is characterized, at least
until the fourth century B.C., by very specific semantic and pragmatic fea­
tures. Both forms can occur in factual contexts, where on typically conveys
new, focalized and non-topical information ((26)), while œç introduces al­
ready known, non-focalized and topical information ((27)):
(26)

First PTCL PTCL DEM:ACC:SG:N be necessary:PRES:IND:3sG

learn-AOR:iNF:ACT VOU-PL:ACC COMP kind-NOM:PL

ART:GEN:PL:M prose-GEN:PL be:PRES:IND-3sG NEG

fewer:ACC:PL:M than ART:GEN:PL:N poetry-GEN:PL

'One thing you must learn first, that the various kinds of prose are
not fewer than those of poetry.'
(Isocrates, Antidosis, 40.5)
(27)

About PTCL PTCL ART:GEN:SG:F hegemony-GEN:SG COMP

rightly PTCL be:PREs:oPT:3sG ART:GEN:SG:F town-GEN:SG

easy:NOM:SG:N from ART:GEN:PL:N say:PERF-PART:MD-GEN:PL:N

understand-AOR:INF:ACT
74 Sonia Cristofaro

'Thus, as to the hegemony, it is easy to understand from what we

said that it should by right belong to our town.'
(Isocrates, Antidosis, 60.2)
In non-factual contexts, only cbg is allowed:
(28) oi

ART:NOM:PL:M PTCL Spartan:NOM:PL:M say:PRES-IND-3PL:ACT

COMP DEM:NOM:PL:M PTCL from Samos:NOM:PL:M

say:PRES-IND-3PL:ACT COMP
T h e Spartans say that (...) People from Samos, on the other hand,
say that (...) '
(Herodotus, 1.70.9.11)
(29)

if somebody-GEN:sG:M hear:PERF-IND-2PL:ACT COMP I:NOM

educate:PRES-iNF:ACT devote myself:PRES-IND:1SG:ACT

man:ACC:PL and profit:ACC:PL make:PRES-IND-lsG-MD NEG

DEM:NOM:SG:N true:NOM:sG:N
'And if you heard that I devote myself to educate people and I make
+ factualprofit from that, this is not true either.' -factual
(Plato,-topic:
+focus, +new, Apology
( 19."d".9)
-focus, -new, +topic: (
This situation can be described by the following schema:
Schema 1. Pragmatic and semantic features characterizing the distribution of . and
in complement clauses

In factual contexts, introduces foreground information with high

communicative value, cog background information with low communicative
value.6 This also explains why ¿>g is chosen for non-factual contexts, while
Grammaticalization and clause linkage strategies 75

is not allowed: in this type of contexts, the information conveyed by the

complement clause cannot be asserted as a fact, and has therefore a low
communicative value, just like the information conveyed by ύ in factual
contexts. Similar phenomena can be found in various languages: Biblical
Hebrew (cf. section 2.2 above) uses two different complementizers to convey
predictable and presupposable information and surprising and non-presup­
posed information respectively; many Romance languages use the same verb
forms (subjunctive) to mark factual presupposed information and non-factual
information (Lunn 1989; Lunn and Cravens 1991). Besides this strategy,
Italian displays two different complementizers: come, 'how', which has
approximately the same range of functions as ύ, and che, 'that'. In certain
types of complement clauses, come has come to mean 'that' through a process
similar to the one described for ύ in sextion 3.3; che is however used to
convey foreground information with high communicative value, while come
introduces background information with low communicative value, although
non-factual information is usually introduced by che, which is the unmarked
complementizer.
The same pragmatic features characterizing the distribution of and ύ
in complement clauses can be detected in the original function of the two
forms. is an accusative form, used to convey a non-topical and focalized
specification about a topical item in the preceding clause ((H)); ύ is an
instrumental/ablative form used to focalize how a given event takes place
((24)): this means that the event itself is topical and non-focalized, and
represents background information. It is then clear that the original function of
on and ύ determines some major constraints on their distribution.

3.4.2. Some diachronic issues

The system described in section 3.4.1 holds approximately until the fourth
century B.C., when signs of an incipient breakdown can be found. It is in this
period that the adverbial expression / ovóxi increases in
frequency: the merging of main predicate and complementizer clearly indi­
cates that the latter is losing the features that characterize it as an autonomous
entity. At the same time, öxi and start to be used together, for instance in
combination with superlative forms of adjectives, where formerly only ύ
was allowed:
76 Sonia Cristofaro

(30) L J
as good as possible:ACC:SG:M I:ACC become:AOR-INF
'That I attain the highest possible excellence.'
(Plato, Symposium, 218."d".2)
In the Hellenistic period,. comes to be significantly reduced in use, and
becomes a general complementizer, used in both factual and non-factual
contexts:

(31)
ART:NOM:SG:F pretext:NOM:SG be:IMPF:3sG COMP king:NOM

DEM:ACC:SG:M call:PRES-IND-3sG:MD
'The pretext was that the king had called him'.
(Chariton, Callirhoe, 4.7.8.3)
In the same period, two causal subordinators, and start to be
used as complementizers instead of on:
(32)
come:IMPF:3sG news:NOM:SG COMP be near:PRES:IND-3sG

HannibalI:NOM
'The news came that Hannibal was near.'
(Polybius, 3.61.8.6)
There is then a process of generalization whereby on tends to become a
universal, unmarked complementizer with no specific semantic or pragmatic
features, and is replaced by new complementizers. This process, however,
does not go to completion: on remains quite widespread, nor does œç com­
pletely disappear.

4. Grammaticalization and clause linkage strategies

4.1. Some theoretical issues about grammaticalization

Grammaticalization is usually presented as a process whereby lexical items

come in certain linguistic contexts to serve a grammatical function and, once
grammaticalized, continue to develop new grammatical functions. This pro-
Grammaticalization and clause linkage strategies 11

cess is seen as unidirectional and irreversible: grammatical morphemes origi­

nate from lexical items, disappear through loss and reappear when new lexical
items become grammatical morphemes; this means that grammaticalization
phenomena always proceed from major open categories such as nouns and
verbs to minor closed categories such as adpositions, auxiliaries, conjunc­
tions, demonstratives and pronouns, and not the other way round (Croft 1990:
230; Hopper and Traugott 1993: 103-6). Traugott (1982; 1988) and Traugott
and König (1991) have claimed that there is a further sense in which gram­
maticalization can be said to be unidirectional, namely increasing subjectifi-
cation: the meaning shift determined by grammaticalization phenomena
always proceeds from an objective and referential realm to one encompassing
the speaker's attitudes and opinions. This process is made possible by contex­
tual inferencing: source items acquire in certain contexts some pragmatic
implicatures that may be conventionalized and ultimately become part of, or
replace completely their original meaning, so as to convey the speaker's
subjective point of view. A standard example of such phenomenon is the
diachronic development of English while from a noun meaning 'time, period'
to a textual connector signaling a time relation to, ultimately, a concessive
subordinator conveying the speaker's attitude towards the proposition
(Traugott 1982: 254).
A common claim about the semantic aspects of grammaticalization is that
grammaticalized items undergo a process of bleaching and loss of meaning.
This is certainly true for the last stages of grammaticalization, when the
affected elements become mere indicators of grammatical relations. Strength­
ening and conventionalization of pragmatic implicatures, however, as well as
metaphoric transfer, which is the other major semantic process involved in
grammaticalization, show that in initial stages of grammaticalization it is more
appropriate to speak of meaning exchange, or meaning transfer. In both cases,
although the original meaning of the affected forms may be lost, new mean­
ings are added: in the case of the conventionalization of pragmatic implica­
tures, which seems to be the dominant process in the development of clausal
connectors, the new meaning is in some way inherent to the original one, that
is, it can be inferred to be part of the latter in some specific contexts; in the
case of metaphoric transfer, which is the process operating for instance in the
development of tense, aspect and case markers, the new meaning cannot in
any way be said to be part of the old one, but replaces it as result of a
contextual association (cf. Traugott and König 1991: 190).
78 Sonia Cristofaro

Pragmatic, semantic, morphosyntactic and phonological changes are

correlated in grammaticalization. Lehmann (1982; 1985) has proposed a
number of criteria to classify grammaticalized items according to their degree
of semantic, morphosyntactic and phonological autonomy: the outcomes of
grammaticalization processes are typically forms with a very low degree of
semantic and morphosyntactic autonomy, integrated within a closed class of
grammatical elements and eroded in phonetic substance. As Hopper (1991)
observes, however, these features (none of which is distinctive for grammati­
calization) usually characterize grammaticalization processes that have al­
ready attained a fairly advanced stage, and are unambiguously recognizable as
such; none of them is necessarily present in initial stages, and other criteria are
needed to detect incipient grammaticalization when the stage of morphologi-
zation has not been reached yet. Grammaticalization basically involves a shift
from full lexical categories to secondary relational categories (decategorial-
ization). Initial stages are usually characterized by the emergence of alterna­
tive strategies to code the same functional domain (layering); these strategies
may initially display quite specific semantic nuances, but, as grammaticaliza­
tion takes place, the range of semantic choices narrows and the selected items
take up a general grammatical meaning (specialization). Source items may
coexist with their grammaticalized outcomes (divergence), and the meaning
of the former may influence the distribution of the latter (persistence). These
principles, like those proposed by Lehmann, are not distinctive for grammati­
calization, but rather characterize aspects of language change in general.
They are proposed as heuristic devices to detect potential instances of gram­
maticalization, but, if grammaticalization is not already a given, they fail to
identify it unambiguously (Hopper 1991: 32).
According to the model described so far, prototypical instances of gram­
maticalization should follow the typology described by Hopper (1991) in
their early stages, and to the one described by Lehmann (1982; 1985) in their
final stages: a prototypical grammaticalized item would be a lexical element
taking on a grammatical function and becoming a bound morpheme of
reduced phonetic substance. This considerably narrows the range of phenom­
ena suitable for inclusion within the domain of grammaticalization, and
makes it possible to establish quite clear-cut boundaries between prototypical
and non-prototypical instances of grammaticalization (e.g., grammatical
items originating from other less or equally grammaticalized items or gram­
maticalization processes that do not reach final stages).
Grammaticalization and clause linkage strategies 79

In a broadened framework, initially proposed by Givón (1979: 219-23),

grammaticalization should be considered as a phenomenon encompassing the
reason for and the development of grammatical structures in general: these
may originate either from individual lexical items or already existing gram­
matical elements, or from the larger discourse/pragmatic context. In Givón's
view, grammaticalization proceeds from a pragmatic, discourse-based mode
of communication, characterized by loosely bound topic-comment structures,
to a syntactic mode of communication based on more tightly integrated
structures and elaborate use of grammatical morphology. Instances of this
type of process are represented by such phenomena as shifts from topic to
subject, from topic sentences to relative clauses, from serial verb constructions
to case markers (Givón 1979: 219-23), from rhetorical question markers to
subordinators (Herring 1991) .
Herring (1991: 255) has suggested that the lexical, or sentence-based
path of development and the pragmatic, or discourse-based one are in fact
independent and free to interact. In what follows, the relevance of each of the
two in the development of complementizers will be explored, and the follow­
ing issues will be addressed:
a. To what extent can the development of complementizers, particularly
the development of ort and be said to fall within the domain of grammati­
calization?
b. What is the role played by pragmatic/discourse-based strategies in the
development of complementizers and clause linkage strategies in general?
c. On which grounds can one assimilate prototypical instances of gram­
maticalization involving individual lexical items to the development of gram­
matical structures originating from pragmatic strategies?

4.2. Grammaticalization patterns in the development of and

The development of and which was sketched in section 3, can be

described by means of the following schema:
80 Sonia Cris tof aro

YO-: -> -> used -» adverb, -> COMP -> (as a COMP)
relative- instrumental/ as subordinator is restricted in use
deictic ablative anaphoric (adverbial clauses) —» „_,___
stem —» —» forms —» COMP —» subordinator — » /
accusative in reason clauses
—» generalization
of (as a
COMP),

Schema 2. Diachronie development of on and ύ

Two case forms of a relative-deictic pronominal stem, originally used to

refer anaphorically to an item in the preceding clause, and contributing to
clause linkage on the pragmatic and the semantic level, are transformed into
subordinators exclusively operating on the syntactic level, and pertaining to
the domain of complement and adverbial clauses respectively. They are
subsequently extended outside their original domain (from complement to
adverbial clauses and from adverbial to complement clauses), and undergo a
process of generalization leading, at least as far as complement clauses are
concerned, to the gradual loss of their original semantic, pragmatic and
sometimes syntactic features: gets consistently restricted in use, and its
functions are taken up by on, even in those cases in which the global sense of
the sentence is incompatible with the original semantic and pragmatic features
of the latter; loses in certain contexts its syntactic autonomy, so that it can
merge with the main predicate , and is sometimes
replaced by new complementizers.
The basic shift underlying the whole phenomenon proceeds from a
domain where the relevant forms are used to refer back to concrete and
specific items to one where they signal logical relations between events. Such
shift certainly falls within the framework of grammaticalization, which is
generally recognized as a process extending from more concrete to more
abstract realms. The various phenomena of semantic transfer affecting on and
ύ also correspond to the ones characterizing grammaticalization phenomena:
the shift that transforms on and ύ from pronominal forms into subordinators
can be considered as a case of metaphoric transfer, for the relation linking the
pronoun to its nominal ((21)) or pronominal antecedent ((H)) is assimilated to
the one existing between two events. The extension of on and ύ to reason
and complement clauses respectively, on the other hand, is a case of pragmatic
strengthening of informativeness through the conventionalization of some
Grammaticalization and clause linkage strategies 81

conversational implicatures: on and ύ may be inferred to have causal and

complement features respectively only in a limited number of contexts, and
these represent the ground for generalizing the use of the two forms to all
contexts displaying similar features.
In section 4.1 the features characterizing early and final stages of gram­
maticalization phenomena were described. Some of the former can be recog­
nized in the development of on and ύ:
a. Within the functional domain of complement and adverbial construc­
tions, alternative strategies coexist: the correlative diptych and the use of
subordinators with no pronominal antecedent on one hand, the use of different
complementizers to code factual clauses with different pragmatic features and
non-factual clauses on the other (layering);
b . T h e original meaning of the forms influences their distribution:
and become complementizers and adverbial subordinators following dif­
ferent patterns of development; their domains are kept separate, at least until a
certain point in time, and their different semantic and pragmatic features can
be directly related to their original meaning (persistence);
c . A t a later stage, the range of pragmatic and semantic nuances associ­
ated to and ύ is to a large extent neutralized: on takes up the functions of
in complement clauses and becomes a general complementizer with no
features of its own any more (specialization).
On the other hand, none of the features typical of final stages of gram­
maticalization can be detected. The incipient breakdown involving the
complementizer system that can be observed since the fourth century B.C.
mainly operates on the functional (=semantic and pragmatic) level, and has no
correspondence on the morphosyntactic and the phonological level: on and
ύ undergo no processes of obligatorification, phonological attrition or loss of
morphosyntactic autonomy. The case of an instance of
compounding that could be seen as the result of a loss of morphosyntactic
autonomy and extreme generalization on the part of on, cannot be included
within the framework of grammaticalization: as Hopper and Traugott (1993:
41; 49) show, in order for a process of compounding to be classified as a
grammaticalization case, it must affect the semantics, the syntax, the mor­
phology and the phonology of the forms it involves, and lead to the formation
of new grammatical affixes. Besides being isolated and non-productive, the
case of has no functional consequences: the adverb has
exactly the same function as the combination of main predicate and comple-
82 Sonia Cristofaro

mentizer, and no differences in meaning can be detected; the development of

an adverb out of an adjectival predicate and a complementizer affects the
lexicon, not the grammar.
What is more, while the case of a lexical item taking on a grammatical
function and possibly developing new grammatical functions clearly falls
within the traditional definition of grammaticalization, the case of subordinat­
ing conjunctions originating from pronominal forms is more problematic.
Both pronouns and conjunctions belong to small, closed classes of relational
items. Pronouns can probably be said to be less grammaticalized, for they refer
to individual entities on a deictic or textual level, while conjunctions only
signify abstract logical relations. Grammaticalization is however here a matter
of degree, not a feature that distinguishes a category with respect to the other.
The development of and < displays then a number of semantic shifts
that are typical of grammaticalization although being not distinctive for it, and
shows on the other hand none of the features that characterize grammaticaliza­
tion on the morphosyntactic and the phonological level. Hopper and Traugott
(1993: 95) show that grammaticalization processes do not have to go to
completion and to move all the way along a cline: a change may be arrested
before it is fully implemented, and the outcome of grammaticalization may be
an incomplete subsystem that is not evidently moving in some identifiable
direction. This seems to be the case with and although a number of
features make it possible to assimilate their development to traditional in­
stances of grammaticalization, one should conclude that, according to the
sentence-based/lexical approach, this development represents a non-proto­
typical case.

4.3. The development of complementizers as an instance of

grammaticalization

The instances of complementizer development examined in section 2 basi­

cally display the same features characterizing the development of and ύ.
The processes taken into account show many of the features characterizing
initial stages of grammaticalization phenomena (layering, specialization, di­
vergence); none of them, on the other hand, seems to go to completion or
display any of the features typical of late grammaticalization. In one case, that
of Banda Linda, there is a shift from a major open category (verbs) to a minor
closed one (subordinating conjunctions); in Biblical Hebrew and Hittite,
Grammaticalization and clause linkage strategies 83

however, the shift is again from pronouns to subordinators, that is from a

minor closed category to another.
On the other hand, all the processes taken into account, including the
Greek ones, represent clear instances of grammaticalization in the discourse-
based sense. Lehmann (1988) has shown that clause linkage strategies can be
ranged along a continuum extending from the lowest to the highest degree of
integration of the connected clauses. On one pole, they are independent, and
show maximal elaboration of lexical and grammatical information. On the
other, one of them is reduced in structure and embedded in the other as a
nominal constituent of it. Between these two extremes, various patterns of
clause linkage can be found. These show a decreasing elaboration of the
structure of one of the two clauses and of the information conveyed by it. The
structures from which complementizers originate, namely the correlative
diptych, parataxis and verb serialization, are to be placed on the pole of
minimal integration: the linked clauses are approximately equivalent in struc­
ture, and their linkage is signaled by their mutual pragmatic and semantic
relevance. When the elements indicating this mutual relevance are reanalyzed
as syntactic connectors, the organization of the sentence changes. Although
the structure of the two clauses may remain the same, the complementizer
transforms one of them into a constituent of the other, so that the whole
sentence shifts towards the pole of highest integration. A construction like
in which the former main clause becomes an adverb
which has the former subordinate clause in its scope, may be seen as an
extreme result of this process.
The development of complementizers and clause linkage strategies in
general proceeds then from pragmatics and semantics to syntax, and from less
to more tightly integrated structures. At the beginning the function of the
relevant forms is entirely context-dependent: they can signal clause linkage
only insomuch as the context allows it ( and can function as connectors
only because they can be referred to an item in the preceding clause, and the
same holds for Hittite and Biblical Hebrew complementizers; Banda Linda
dpa can initially be read as a complementizer only when the sentence contains
another verb of saying, cf. (2)). At a subsequent stage, this same function is
generalized and gets constantly and stably associated to the forms.
It is easy to see that this process corresponds to the shift from the
pragmatic to the syntactic mode of communication described by Givón
(1979). Whether such shift can be included within the framework of gram-
84 Sonia Cristofaro

maticalization is a controversial issue (cf. sections 1 and 4.1 above). On the

basis of the data examined so far, it can however be argued that it is fully
legitimate to assimilate it to instances of lexical/sentence-based grammatical-
ization, for the two processes appear to share some fundamental features:
a. Both turn out to be based on problem-solving strategies that code more
abstract domains in terms of more concrete ones: expressions designating
abstract logical relations and subjective points of view (complementizers are
used to convey the speaker's degree of commitment towards the proposition,
or his or her evaluation towards the pragmatic value of the information
conveyed by the complement clause) take as their source lexical items refer­
ring to elements of the external world or pragmatic strategies originating
from the immediate context of face-to-face interaction (Heine, Claudi and
Hunnemeyer 1991).
b. Both display bleaching of lexical meaning and corresponding strength­
ening of grammatical meaning, and appear to be based on the same processes
of pragmatic strengthening of informativeness and metaphoric transfer. More
generally, both lead to the development of context-independent strategies
coding autonomous and fixed grammatical relations out of context-dependent
ones.

NOTES

1. The following texts have been examined: Homer, Iliad, Odyssey; Herodotus, Histories,
books 1, 5, 9; Thucydides, Histories, books 1, 2, 3; Lysias, On the murder of Era­
tosthenes, Against Eratosthenes, On the Refusal of a Pension; Aristophanes, Knights,
Plutus; Hippocrates, Ancient Medicine; Plato, Apology, Symposium, Gorgias; Isocrates,
Against the Sophists, Antidosis; Xenophon, Anabasis, Memorabilia; Demosthenes,
Philippic 1, On the Crown; Aesop, Fables, 1-50; Polybius, Histories, books 1, 2, 3;
Chariton, Callirhoe, books 1, 2, 3, 4; Gospels (Matthew, Mark, Luke, John); Plutarch,
The Parallel Lives: Nicias and Crassus, Agis and Cleomenes, Tiberius and Caius
Gracchus; Lucianus, A True Story, Dialogues of the Courtesans; Eliodorus, Aethiopica,
books 1,5,6, 10.
2. The development of complementizers out of verbs of saying is quite common cross-
linguistically; cf. for instance the well-known case of Kwa languages described by Lord
(1976) and a number of Tibeto-Burman languages (Saxena 1988).
3. The pragmatic notions introduced here are based on Dik 1989, chap. 13. By topic are
meant those entities about which information is provided or requested in the discourse.
By focus are meant those pieces of information with are the most important or salient with
respect to the purpose of the communication. Focality may characterize both topical and
Grammaticalization and clause linkage strategies 85

non-topical entities, but the elements which turn out to have the highest communicative
value, i.e. to be most important for the development of the commmunication, are those
that are new, non-topical and focalized.
4. It has been suggested that ύ may have originated from an ablative form. This hypothesis
does not affect the reconstruction proposed here, according to which some of the
functions of ύ are instrumental in nature. The domains of ablative and instrumental
forms overlap to a large extent in many languages, as is clear, for instance, in Latin, where
they have merged (cf. Monteil 1963: 329).
5. For the other uses of ύ, which can be easily related to the ones discussed here, cf.
Monteil 1963: 327-63.
6. The contexts where only or ύ are allowed are defined by the opposite values of the
three parameters focus, new and topic (+focus, +new, -topic for -focus, -new, +topic
for ύ). Different combinations of the values of these same parameters (for instance
+focus, -new, +topic; +focus,+-new,+Kopic,etc.) are of course possible, and the contexts
characterized by them allow either or with no apparent semantic or pragmatic
difference. The contexts allowing just one complementizer may be seen as the opposite
poles of a continuum encompassing all the contexts allowing both complementizers.

ABBREVIATIONS

AC accomplished M masculine
ACC accusative MD middle
ACT active N neuter
AOR aorist NEG negation
ART article NOM nominative
CENTR centrifugal 0 object
DAT dative PART participle
DEICT deictic PERF perfect
DEM demonstrative PL plural
F feminine PRES present
FUT future PTCL particle
GEN genitive S, SUBJ subject
IMP imperative SG singular
IMPF imperfect SS same subjeel
IND indicative SUCC successive
INF infinitive V verb
INJ injunctive voc vocative
86 Sonia Cristofaro

REFERENCES

Bybee, Joan, Revere Perkins and William Pagliuca. 1994. The Evolution of Grammar.
Chicago and London: The University of Chicago Press.
Burguière, Paul. 1960. Histoire de /'infinitif en grec. Paris: Klincksieck.
Dik, Simon C. 1989. The Theory of Functional Grammar. Part I: The Structure of the
Clause. Dordrecht: Foris.
Cloarec-Heiss, France. 1986. Le banda-linda de Centrafrique. Paris: SELAF.
Cristofaro, Sonia. 1995. "Lo sviluppo dei complementatori come modello di gramma-
ticalizzazione: il caso del greco antico." Archivio Glottologico Italiano 80: 102-121.
. 1996. Aspetti sintattici e semantici delle frasi completive in greco antico. Firenze:
La Nuova Italia.
Croft, William. 1990. Typology and Universals. Cambridge: Cambridge University Press.
Genetti, Carol. 1991. "From postposition to subordinator in Newari." In Elisabeth C.
Traugott and Bernd Heine (eds.), 227-255.
Givón, T. 1979. On Understanding Grammar. New York: Academic Press.
. 1991. "The evolution of dependent clause morpho-syntax in Biblical Hebrew." In
Elisabeth C. Traugott and Bernd Heine (eds.), 257-307.
Haudry, Jean. 1973. "Parataxe, hypotaxe et corrélation dans la phrase latine". Bulletin de la
Société de Linguistique de Paris 68 (1): 147-186.
Heine, Bernd, Ulrike Claudi and Friederike Hùnnemeyer. 1991. Grammaticalization. A
Conceptual Framework. Chicago and London: The University of Chicago Press.
Hengeveld, Kees. 1990. "The hierarchical structure of utterances." In Jan Nuyts, A.
Machtelt Bolkestein and Co Vet (eds.), Layers and Levels of Representation in
Language Theory. A Functional View, 1-23. Amsterdam and Philadelphia: John
Benjamins.
Herring, Susan C. 1991. "The grammaticalization of rhetorical questions in Tamil." In
Elizabeth C. Traugott and Bernd Heine (eds.), 253-284.
Hopper, Paul J. 1991. "On some principles of grammaticalization." In Elizabeth C. Traugott
and Bernd Heine (eds.), 17-35.
Hopper, Paul J. and Elizabeth C. Traugott. 1993. Grammaticalization. Cambridge: Cam­
bridge University Press.
Humbert, Jean. 1986. Syntaxe Grecque. 3e éd. Paris: Klincksieck.
Justus, Carol. 1976. "Relativization and topicalization in Hittite." In Charles N. Li (ed.),
Subject and Topic, 213-246. New York: Academic Press.
. 1980. "Typological symmetries and asymmetries in Hittite and IE complementa­
tion." In Paolo Ramat, Onofrio Carruba, Anna Giacalone Ramat and Giorgio Graffi
(eds.), 183-203.
Kurylowicz, Jerzy. 1965. The Evolution of Grammatical Categories. Reprinted in
Kurylowicz 1976, 38-54.
. 1976. Esquisses Linguistiques. Vol. 2. München: Fink.
Lehmann, Christian. 1982. Thoughts on Grammaticalization: A Programmatic Sketch.
Köln: Institut für Sprachwisenschaft, Universität zu Köln [AKUP 48].
. 1985. "Grammaticalization: Synchronic variation and diachronic change." Lingua
e stile 20 (3): 303-318.
Grammaticalization and clause linkage strategies 87

Lehmann, Christian. 1988. "Towards a typology of clause linkage." In John Haiman and
Sandra A. Thompson (eds.), Clause Combining in Grammar and Discourse, 181-225.
Amsterdam and Philadelphia: John Benjamins.
Lehmann, Winfred P. 1980. "The reconstruction of non-simple sentences in Proto-Indo-
European." In Paolo Ramat, Onofrio Carruba, Anna Giacalone Ramat and Giorgio
Graffi (eds.), 113-144.
Lord, Carol. 1976. "Evidence for syntactic reanalysis: from verb to complementizer in
Kwa." In Salikoko K. Muwfene, Carol A. Walker and Sanford B. Steever (eds.), Papers
from the 12th Regional Meeting of Chicago Linguistic Society, 179-191. Chicago:
Chicago Linguistic Society.
Lunn, Patricia V. 1989. "The Spanish subjunctive and relevance." In Carl Kirschner and
Janet DeCesaris (eds.), Studies in Romance Linguistics, 484-494. Amsterdam and
Philadelphia: John Benjamins.
Lunn, Patricia V. and Thomas D. Cravens. 1991. "A contextual reconsideration of the
Spanish -ra 'indicative'." In Suzanne Fleischman and Linda R. Waugh (eds.), Dis­
course-Pragmatics and the Verb: the Evidence from Romance, 147-163. London:
Routledge.
Meillet, Antoine. 1912. L'évolution des formes grammaticales. Reprinted in Meillet 1958.
. 1958. Linguistique historique et linguistique générale. Paris: Champion.
Monteil, Pierre. 1963. La phrase relative en grec ancien. Paris: Klincksieck.
Noonan, Michael. 1985. "Complementation." In Timothy Shopen (ed.), 42-140.
Ramat, Paolo, Onofrio Carruba, Anna Giacalone Ramat and Giorgio Graffi (eds.). 1980.
Linguistic Reconstruction and Indo-European Syntax. Proceedings of the colloquium
of the 'Indogermanische Gesellschaft', University of Pavia, 6-7 September 1979.
Amsterdam and Philadelphia: John Benjamins.
Ransom, Evelyn N. 1986. Complementation: Its Meanings and Forms. Amsterdam and
Philadelphia: John Benjamins.
. 1988. "The grammaticalization of complementizers." Berkeley Linguistics Society
14: 364-374.
Saxena, Anju. 1988. "On syntactic convergence: The case of the Verb 'say' in Tibeto-
Burman". Berkeley Linguistics Society 14: 375-388.
Schwyzer, Edward and Albert Debrunner. 1959. Griechische Grammatik. 2.Bd. 2. Auf.
München: Beck.
Shopen, Timothy (ed.). 1985. Language Typology and Syntactic Description. Vol. 2:
Complex Constructions. Cambridge: Cambridge University Press.
Thompson, Sandra A. and Robert Longacre. 1985. "Adverbial clauses." In Timothy Shopen
(ed.), 171-234.
Traugott, Elizabeth C. 1982. "From propositional to textual and expressive meanings:
Some semantic-pragmatic aspects of grammaticalization." In Winfred P. Lehmann and
Yakov Malkiel (eds.), Perspectives on Historical Linguistics, 245-271. Amsterdam
and Philadelphia: John Benjamins.
. 1990. "From less to more situated in language: The unidirectionality of semantic
change." In Sylvia Adamson, Vivien A. Law, Nigel Vincent and Susan Wright (eds.),
Papers from the Fifth International Conference on English Historical Linguistics, 496-
517. Amsterdam and Philadelphia: John Benjamins.
88 Sonia Cristofaro

Traugott, Elizabeth C. and Ekkehard König. 1991. "The semantics-pragmatics of gram-

maticalization revisited." In Elizabeth C. Traugott and Bernd Heine (eds.), 189-218.
Traugott, Elizabeth C. and Bernd Heine (eds.). 1991. Approaches to Grammaticalization.
Amsterdam and Philadelphia: John Benjamins.
 Some remarks on analogy,
reanalysis and grammaticalization*

Livio Gaeta
Terza Università di Roma

1.1 Introduction

As is well known, grammaticalization is, according to Meillet (1912), the

term for those linguistic changes that have as input lexemes (or lexical forms)
which become more or less grammatical formatives. Following his seminal
idea, two "clines" of grammaticalization have been singled out (cf. Hopper &
Traugott 1993: 7):
• the cline of grammaticality, concerning "syntactic" phenomena of
grammaticalization, such as cliticization, affix-formation, etc.;
• the cline of lexicality, concerning more specifically the evolution of
free lexical forms into word formation affixes (through a compounding
stage).
Meillet and his followers, however, have used in my opinion the term gram­
maticalization both over-generously and injudiciously. As has been pointed
out by others (cf. Hopper 1994), grammaticalization, in its broader meaning,
has to do with the whole range of phenomena that give rise to grammatical
formatives, not merely with those originating from lexical forms. This as­
sumption that grammaticalization originates mostly from lexical forms is
probably revealing of the working attitude of those scholars who have devel-

* This paper has been presented at the Workshop on "Diachronic perspectives in grammatical­
ization" during the XXVIII Annual Meeting of the Societas Linguistica Europaea held in Leiden
in September, 1995. I thank Georgi Jetchev, the scholars attending the workshop as well as the
editors for helpful comments and suggestions. Needless to say, errors made and views expressed
are my own.
90 Livio Gaeta

oped Meillet's original idea by considering grammar and grammatical change

as more related to syntax and discourse than to other components of the
language. To reverse this custom, I will consider here the case of the "phono­
logical" way to grammaticalization (also called grammaticalization "from
below", cf. Greenberg 1991: 303), i.e. the case of originally phonological
rules that get involved with more or less advanced processes of morphologi-
zation, to use the term closest to grammaticalization.1

1.2 Grammaticalization and morphologization

It is noteworthy, in this perspective, to take into consideration two different

definitions of morphologization, revealing the different points of view of the
respective authors. The first definition is borrowed from the recent handbook
of Hopper & Traugott (1993: 130), in which morphologization is the "com­
pacting — the fusing of erstwhile independent elements with each other, most
especially the development of clitics into inflections". In other words, "mor­
phologization is that part of grammaticalization that primarily involves the
second and third step of the cline [of lexicality] : lexical item > clitic > affix (p.
132)". That other uses of the term morphologization are not ignored by the
authors is shown by the note they add (p. 224), in which they specify that "the
form 'morphologization' is sometimes also used for the development of
phonological alternations that occur as the result of phonological changes in
specific morphological contexts, in other words, of morphophonemic alterna­
tions, as found in the English past tense form -ed ([t - d - ed])". Actually, the
example they provide shows a partial misunderstanding of the kind of linguis­
tic change that goes under the name of morphologization. The sound alterna­
tion involved in the English past tense form is correctly an instance of
morphophonology, namely of a morphologically conditioned phonological
alternation (cf., among others, Dressier 1985). As an instance of
morpho(pho)nemic alternation, the quoted example appears however to be
quite different from the morphologization of phonological rules. In fact, the
crucial point is that it is not anchored in synchronic morphological alternation
(cf. Klausenburger 1979). The purpose of the rule remains, in a sense,
phonologically oriented (in the quoted example: sonority assimilation to the
preceding phoneme), whereas in case of true morphologizations the purpose
becomes of a morphological kind, i.e. to convey a specific semantic/gram-
Some remarks on analogy, reanalysis and grammaticalization 91

matical function. In this sense, a correct and strict definition of morphologiza-

tion is that provided by Wurzel (1980: 444):
"morphologizing, as we understand it, holds whenever a rule, which so far
has held for an operation of permutation, insertion or deletion in a phono­
logical context P, changes in a way that it holds for the same operation (not
necessarily only this one) or its inversion in a context of grammatical
categories C. A rule, whose original (more or less preserved) function it was
to adapt a set of phonetic sequences to human speech organs, takes over the
basically new function of formally marking grammatical categories in
words. Thus the complete or partial phonetic motivation of the rule is
substituted by a semiotic motivation".

This long quotation will serve as a background for the presentation of data
which follows. To briefly summarize the discussion, it seems useful to me not
to restrict the term grammaticalization to the more specific uses mentioned
above. Otherwise, we would be faced with quite a paradoxical (and mislead­
ing) terminology, in which a more general term, i.e. grammaticalization, is
assumed to designate a rather specific set of phenomena, whereas the more
specific term morphologization turns out to be so wide as to cover, besides
instances of grammaticalization, other kinds of changes that somehow "en­
rich" the grammar of a language.

2.1 The path of grammaticality

In the article quoted, Wurzel showed a range of phenomena that go under the
name of morphologizations. They all involve the more general assimilative
rule of Umlaut. In this respect, one can observe that morphologizations
generally find their starting point in morphonologizations, i.e. in the introduc­
tion of grammatical features into the context of a phonological rule (see the
above example of the English past tense form). However, one can speak of
morphologization, when the last remainder of the sequential phonological
environment is deleted from the context of the rule, that, in this way, "becomes
free for categorial marking independently of phonological context condi­
tions" (Wurzel 1980: 445).
Let us give a look now at the following example of morphologization
Wurzel provides. In Old High German (=OHG) there are the following noun
paradigms respectively for masculine /-stems and n-stems:
92 Livio Gaeta

(1) N. Sg. gast 'guest' PI gast Sg. hano 'cock' PL hanun

G. gastes gestio henin hanôno
D. gaste gestim henin hanôn
A. gast gesti hanun hanun
I. gestiu -
The sound alternation ale in the nominal root is caused by the Umlaut rule,
which functions at this stage phonologically:
- consonant 1
(Ur)a V -> [+ front] / _ C1 - back
|_ + high J
After a rather unclear stage where umlauted forms begin to be ruled out
morphologically,2 we find the following situation in Middle High German
(=MHG):
(2) N. Sg. gast PL geste Sg. hane PL hanen
G. gastes geste hanen hanen
D. gaste gesten hanen hanen
A. gast geste hanen hanen
At this stage, the originally phonological rule of Umlaut has become morpho­
logically conditioned: it occurs namely only in the plural of i-nouns, and it has
been completely ruled out from n-nouns. According to Wurzel, we can
represent now the Umlaut rule in the following way:
| + i - Inflect 1
(Ur)b V -> [+ front] / + Masculine
[_+ Plural J
Clearly, for the reanalysis of the (Ur)a to take place, it was necessary for some
other changes to happen, and particularly the weakening of vowels in un­
stressed final syllables. This phonological change was the crucial factor that
favoured the process of reanalysis: it creates what is a kind of affix, or at least
a "discontinous morpheme" (cf. Salmons 1994: 215). This kind of grammati-
calization of an early phonological rule into an inflectional rule can be
understood as belonging to the path of grammaticality, a kind of reverse path
to that starting with lexical forms:
• phonological rule > (morphophonological alternation) > affix
Some remarks on analogy, reanalysis and grammaticalization 93

Notice that this path presupposes an intermediate step, in which the alternation
is morphonologically governed. This corresponds also to the observation of
Wurzel (1980) that morphologizations (of phonological rules) begin as
morphonologizations. Generally, the proper morphologization takes place as a
consequence of the complete "decay of phonological alternations based on
which the alternation had functioned so far" (cf. Wurzel 1980: 457).

2.2 The cline of lexicality

A cline of lexicality, symmetrical to the path of grammaticalization described

above, can be assumed. Here, an originally phonological rule gets grammati-
calized as a word formation device. The latter case points to the grammatical­
ization of early phonological rules as derivational affixes, which are able to
create new lexical words.3
An example of a new derivational affix produced by grammaticalization
to be found in literature is the so-called Cheshirization, where the original
morpheme disappears leaving only a phonological alternation as its trace, like
Lewis Carroll's Cheshire Cat, who disappeared leaving only its smile. For
example, in Lahu, a Lolo-Burmese language of Northern Thailand, an origi­
nal causative prefix, probably *s-, underlies alternations between voiced and
voiceless initials in such pairs as:
(3) a. [do] 'drink' : [to] 'give to drink'
b. [dè] 'come to rest' : [te] 'put down' 4
A similar alternation also existed in Germanic, due to the presence of the
causative suffix -jan, which served to make causative verbs from verbal roots
(cf. Meillet 19304: 155):
(4) got. drigkan 'drink' : dragkjan 'give to drink'
The vocalism of the derived verbal root is a (< PIE *0?)> and the suffix is
stress-bearing, which gives rise, in the course of further developments, to
other sound alternations (see below beißen-beizen), that are however irrel­
evant for the present purposes.
In OHG, the suffix -jan triggered the rule of Umlaut, giving rise to the
following alternations, that still survive in New High German (= NHG, cf.
Paul 1920: 125):
94 Livio Gaeta

(5) trinken 'to drink' : tränken 'to water'

beißen 'to bit' : beizen 'to annoy'
walzen 'to laminate' : wälzen 'to roll'
What is relevant to our purposes is the observation of Wilmanns (19302: 62)
that "erscheint der Umlaut als ein Mittel der Ableitung und dringt so selbst in
solche Verba ein, die ursprünglich der 2. oder 3. Conjugation folgten". This
verb class fluctuation is quite widespread in OHG and it goes further in
MHG, 5 tied in with a semantic differentiation of the verbal "Aktionsart" (cf.
Krämer 1971:66):
(6) sprangôn 'spring' : sprengen 'make to spring'
hwarbôn 'transform' : hwerben 'turn'
zartôn 'flatter' : zerten 'stroke'
krachôn 'creak' : krecken 'shake'
ga-langon 'reach' : lengen 'lengthen'
lazzôn 'slow down' : lezzen 'hold back'
By "Aktionsart" proper, Krämer means a semantic differentiation in which
the verb derived by means of the original jan suffix has a causative-factitive
meaning. As shown in the reported examples, no traces of the phonetic context
motivating Umlaut are present in these verbs. Rather, the weakening of the
final unstressed vowels has levelled out the deverbal jan suffix. As a conse­
quence of reanalysis, the occurrence of Umlaut has been reinterpreted as
responsible for the process of derivation. In fact, after Umlaut lost its phonetic
motivation by the weakening of unstressed syllables, a number of verbs with
factitive-causative meaning that originally did not show -jan suffix take
Umlaut in MHG period:

(7) OHG corônôn, MHG kronen/krœnen, NHG krönen 'to crown'

OHG offanôn, MHG offenen, NHG öffnen 'to open'
OHG roten, MHG roten, NHG röten 'to redden'
My proposal is that beside morphologization of Umlaut as a kind of affix
within inflectional morphology, morphologization as a derivational device has
also taken place (cf. Robinson 1975). The originally phonological rule illus­
trated above has been reanalyzed and reinterpreted as a morphological rule of
derivation:
Some remarks on analogy, reanalysis and grammaticalization 95

+
(Ur)c V → [+ front] / ^erb .
[ + Causative]

This morphological rule of derivation still survives, at least partially, in NHG,

where the following doublets (cf. Wellmann 1973: 24) are to be found:
(8) lahmen 'to be lame' : lahmen 'to paralyze'
blauen 'to be blue' : bläuen 'to dye blue'
dampfen 'to steam' : dämpfen 'to steam (cook.)'
tafeln 'to have dinner' : täfeln 'to cover with panels'

3.1 Grammaticalization vs. analogy: The case of Rückumlaut

The effects of a reanalysis in which the rule is associated with certain

grammatical features are crucial in distinguishing this phenomenon of gram­
maticalization from the contemporary (and apparently similar) case of
Rückumlaut (cf. Vennemann 1986, Ronneberger-Sibold 1990). The latter is
an instance of rule inversion, in which a generalization of the inverse rule has
taken place on the basis of the occurrence of the sound alternation in a
secondary semantic category (cf. Vennemann 1974: 139). In the OHG period,
as a consequence of the loss of phonetic motivation for the Umlaut rule, the
following verbal paradigms emerged:
(9) brennen brannte gebrannt 'to burn'
senden sandte gesandt 'to send'
This alternation concerned verbs of the -jan class (*brannjan, *sandjari),
which displayed, at least historically, particular phonological properties. Here,
a rule of Syncope deleted the -j- in a light syllable between a primary-stressed
heavy syllable and a secondary-stressed one, but the same has not happened to
the glide of the infinitive (cf. *brannjan vs. *brannda, see Vennemann 1986
for details). The rule of Umlaut must be ordered after the Syncope rule in
order to provide the correct form. Interestingly, Rückumlaut has been ex­
tended analogically in the MHG period to some other verbs:
(10) *andi-ô-n *andi-ô-ta 'to end'
OHG entiôn entiôta
MHG enden endete
ante
96 Livio Gaeta

In the form ante, the same alternation occurs as in senden - sandte.

Vennemann speaks in this instance of rule inversion. Where originally the
phonological rule of Umlaut caused sound alternations within the verbal
paradigm, a restructuring that has inverted the interpretation of the surface
forms has taken place. The real change is assumed to occur in the preterital
form — a secondary semantic category with respect to the present tense —
that now alternates with the present on the basis of the well attested model
senden - sandte.
However, the comparison of the latter case with the instances of gram-
maticalization that we presented above (even though determined by the same
Umlaut rule) will show the significant difference existing between them. In
the former case, we observed an example of grammaticalization, in which the
extension of the umlauted forms was closely connected with the birth of a new
derivational rule within the grammar of MHG. In the latter, it is a surface
analogy that triggers the reinterpretation of the sound alternation, not giving
rise to a new rule within the grammar. In fact, no new combinations of
grammatical features take place to characterize grammatically this particular
(and in MHG opaque) inflectional paradigm. It is my opinion that grammati­
calization can be quite useful to distinguish accurately between the two types
of phenomena. Note that Vennemann also, in his early paper on rule inver­
sion, concludes that "the domain of rule inversion within this class [i.e. among
allophonic, phonemic, morphophonemic and morphological rules, LG] is the
phonemic and morphophonemic rules." (1972: 236). In this way, he restricts
the domain of rule inversion to those cases where no grammaticalization is
observable, since the phonetic motivation of the rule is not replaced by a
semiotic motivation (cf. Wurzel's quotation above). Only in those cases
where no reanalysis with consequent grammaticalization takes place can we
really speak of analogy in its original sense of a restructuring of grammar
aimed at "repairing" it, as in Paul's first formulation.6 In Hopper & Traugott's
(1993: 56) words, "analogy refers to the attraction of extant forms to already
existing constructions ... It is overt", whereas "reanalysis refers to the devel­
opment of new out of old structures ... It is covert." It must be specified,
however, that reanalysis has to give rise to new grammaticalized structures,
i.e. new combination of features. In the case of Rückumlaut such a new
combination does not take place, and it is in fact impossible to state a
morphological rule that could show which verbs must undergo it, in so far as
they are provided with particular properties and fulfill a specific grammatical
Some remarks on analogy, reanalysis and grammaticalization 97

target. Rather, it can be suggested which verbs can undergo Rückumlaut on

account of their closeness (primarily in their phonetic substance) to the
analogical model. Therefore it is in this sense that Vennemann (1986) gives
wide documentation of these analogical extensions, without providing, how­
ever, a full formulation of this supposed grammatical rule.

3.2 Syntactic grammaticalization

Up to now we have been examinating cases of grammaticalization of phono­

logical rules restricted to the word. One can speculate further about other
instances of morphologization of earlier phonological rules that have wider
field of application than the word. It is interesting to look for other cases of
reanalysis in which originally phonological rules become grammaticalized in
syntactic fields like the noun phrase or the sentence. In this perspective, a good
example of affix creation can be taken from Bulgarian. In Old Church
Slavonic there was a dissimilation rule that changed the phonetic value of the
so-called jers (cf. Jetchev 1994 and Lunt 1974) in the context of the phono­
logical word:

where ø stands for phonological word

Thereafter, in Bulgarian a deletion rule cancelled those jers that did not
undergo the dissimilation rule in (11). However, whereas in other lexical
morphemes both outcomes are kept distinct, giving rise to two different
phonemes (as shown in (12)),
(12) a. / U / > / Y / , cf. ['lakYt] < *[lakuti] 'elbow'
b. I1l > Ie/, cf. [den] < *[dinu] 'day'
in the case of the masculine definite article a conflation takes place:
(13) a. *[toveku-tu] > *[tfoveko-tu] > [to'veky] 'the man'
b. *[ogunI-tu] > *[ogojne-tu] > ['ogyjiy, 'ogyny] 'the fire'
In ( 13b) -/y/ is identified as the new form of the masculine definite article (<
*/tu/), in paradigmatic relation to 3e'na-ta] 'the woman' and [po'le-to] 'the
98 Livio Gaeta

field (nt.) and therefore extended also to those cases like the reported
['ogyny], where we would expect phonetically *['ogyne]. 7 Thence, we can
assume in diachrony that (11) was redetermined by inserting morphological
features in its context so as to give rise to a new semiotic motivation of the
original phonological alternation:

Once redetermination took place, the new affix was extended to the cases in
(13b). (14) dropped subsequently out since it did not give rise to synchronic
morphological alternations.
A similar point can also be made in relation to the Umlaut rule illustrated
above. Some scholars (cf. Behaghel 19285: 292) have assumed that the
umlauted forms found in the plural present indicative of the so called
"Präterito-Präsentien" in OHG (and still in NHG) are due to the co-occur­
rence with the clitic pronominal forms containing a palatal vowel triggering
Umlaut. In some cases, in fact, Umlaut is triggered in OHG, and still in the
MHG period, by clitic particles across the word boundaries (cf. Behaghel
19285: 292):
(15) sem mir < sam mir 'with me'
drenk-ich < drank-ich 'I drank'
These phenomena show that Umlaut had, in particular conditions, a larger
context than one used to assume, namely the phonological word (cf. Priebsch
& Collinson 19625: 137). However, the question has been scarcely debated in
the literature and even recent reviews of the entire question barely mention it
(cf. Voyles 1991). In Behaghel's opinion, it has been the co-occurrence of the
clitic pronominal particles that has caused the presence of Umlaut in the plural
present indicative of the "Präterito-Präsentien": MHG wir dürfen, günnen,
künnen, mügen (wiegen), müezen, sülen. These forms can be paralleled with
the following forms taken from High German dialects:
(16) Alem. chömme (< cho + mer) 'we come'
gommer 'we go'
stommer 'we stay'
Bavar. gengemer 'we go'
stendemer 'we stay'
Some remarks on analogy, reanalysis and grammaticalization 99

According to Behaghel's subsequent explanation, the umlauted forms, origi­

nally the result of phonological Umlaut, have been morphologized only in the
plural present indicative (whence they passed to the infinitive) of this verbal
class. This explanation has met with several objections (cf. Fiedler 1928, Scur
1961/62). First, there are only a few cases where syntactic Umlaut is reported
in the manuscripts. This is of course true, but it does not necessarily speak
against Behaghel's theory. It is clear that an allophonic variation like Umlaut
is reflected only sporadically in writing (at least until it becomes phonolo-
gized or grammaticalized). Probably, those cases where umlauted forms
occurred in contexts wider than those of a single word were even more
sporadically reflected. Only at a later stage, when these umlauted forms had
slowly been grammaticalized as signals of this specific conjugational class,
do we find a regular notation of the umlauted vowels. On the contrary, since
Umlaut lost phonetic motivation in the course of the OHG period, those cases
where umlauted vowels emerged such as the examples in (15) disappeared
completely.
Second, this explanation avoids the difficulties shown by the other
theory, the Rückumlaut theory (cf. Fiedler 1928). According to Fiedler's
explanation, the alternation of umlauted forms in the present and non-um-
lauted forms in the past is due to the action of an analogical extension of the
Rückumlaut upon this verbal class.
(17) legen → Iahte : megen ← mahte
However, the theory is in my opinion unconvincing because the analogical
reinterpretation must take place here inversely with respect to its normal
behaviour. As we have seen above, the (semantically) secondary forms of the
past are usually remade on the basis of the inverted rule. If we agree with
Fiedler's explanation, we must assume that the analogical conditioning has
operated in the opposite way as it normally did ad hoc for this verbal class.
In any case, I am convinced that the whole question needs a complete revision
and a separate contribution. My intention has merely been to suggest how
different examples of morphologization can be found within the same original
phonological rule. To quote just another case of a grammaticalized phonologi­
cal rule that may be compared with what has been said in this paragraph, it
may be useful to consider the case of some dialects of central-northern
Calabria (cf. Loporcaro 1995), where specific verbal forms (i.e. the 3 rd plural
person of the present) trigger the lengthening of the initial consonant of the
100 Livio Gaeta

following word in external sandhi (well known as "Raddoppiamento

Fonosintattico"):
(18) [tje 'ßiru 'p:oku] 'they can't see well'
['tsiku 'siempre] 'they always come late'
This lengthening represents the survival, on the segmental level, of an earlier
assimilation rule which in its turn assimilated an original Latin dental ending
consonant to the following one. In this case, the whole suffix -(u)nt is assimi­
lated:
(19) ['ßiru 'p:oku] < *vidunt pocu
That this assimilation rule is now only morphologically governed is shown by
the following example, where the lengthening is the only feature that keeps the
two inflectional forms distinct:
(20) ['ruormu 'p:uru] vs. ['ruormu 'puru]
'they also sleep' T also sleep'
Notice that the phonological rule is fully grammaticalized inasmuch as no
relevant features are superficially available to trigger the rule of lengthening;
only reference to the morphosyntactic information can lead to the correct
application of the rule:

In this sense, semiotic remotivation (in Wurzel's terms) of the originally

phonetic rule has taken place. Sometimes, among the world languages, we
find cases of grammaticalization of phonological rules that give rise to
alternations expressing syntactic relations.8 In this perspective, a very inter­
esting example comes from the Celtic languages, that show a wide range of
phenomena of external sandhi rules well known as Initial Consonant Muta­
tions (cf. Awbery 1986, Willis 1986). In Welsh, for instance, there are several
cases of mutations concerning word-initial consonants. A case particularly
relevant for my purposes is the lenition of initial consonants when preceded
by a specific set of words. Diachronically the lenition is triggered by a
preceding vowel, later disappeared (cf. Lewis & Pedersen 1937: 127). Simi-
Some remarks on analogy, reanalysis and grammaticalization 101

larly to other instances of mutations such as spirantization, nasalization, etc.,

lenition takes place in several contexts either triggered by an immediately
preceding word (so-called "projected mutations") or without an apparent
trigger ("incorporated mutations", cf. Willis 1986: 16). The latter cases are
probably to be interpreted as cases of grammaticalization in the syntactic
domain, since the mutation occurs in words that surface as direct objects,
independently both of the preceding word and of the position of the verb:9
(22) a. clywodd Sion gi (< ci)
heard Sion dog
'Sion heard a dog'
b. clywodd gi (< ci)
'S/he heard a dog'
c. clywodd ci
'A dog heard'
Notice that lenition is crucial in distinguishing (22b) from (22c) in which ci
surfaces as the subject.10 Lenition, however, can be found in other contexts,
occurring both word-internally and within syntactic phrases. These cases can
often be explained as instances of grammaticalizations. For example, femi­
nine nouns are lenited after the singular feminine article, so as to distinguish
the following pair:
(23) y gog (< cog) vs. y cog
the (fem.) cuckoo the (masc.) cook
As can be seen from this rather sketchy representation, the question of the
Celtic mutations is extremely intricate. In my opinion, the notion of grammati­
calization can shed some light on the matter, avoiding for example the prolif­
eration of terms for apparently similar changes that we often find in
literature.11 I think however that the main contribution Celtic mutations bring
forth into this discussion concerns the use of the early phonological rule of
lenition as a mark of purely syntactic relations (cf. 22b above); in this
perspective lenition has been grammaticalized as a kind of case marker. The
last example shows how far the process of grammaticalization of phonologi­
cal rules can move: (the alternation produced by) a phonological rule can be
redetermined to express purely syntactic relations.12 However, much further
research must be done in this field, that is, grammaticalization of phonological
rules in the syntactic domain, before we can consider any further generaliza­
tions possible.
102 Livio Gaeta

4.0 Conclusion

The phonological path to grammaticalization has been shown to represent the

symmetrical and reverse path with respect to both clines singled out in
traditional studies on grammaticalization. Moreover, it turns out that this
notion can be useful in distinguishing between morphologizations involving
reanalysis of early phonological rules and cases of surface analogy, where the
restructuring does not involve grammaticalization in the strict sense. Other
"broader" interpretations of analogy, like those recently provided by Becker
(1990), are in my opinion to be avoided, although interesting, because of the
terminological confusion they lead to. Arguing, as he does, that the formula of
analogy and the classical (generative) morphological rule are in principle the
same, Becker fails to take into consideration the main difference between a
rule and an analogical process (cf. Bauer 1993). This difference is properly
that a rule presupposes the presence of a process of grammaticalization that
could "license" it, whereas analogy, in its strict sense, only concerns surface
relations among already grammatical entities.

NOTES

1. Cf. Joseph & Janda (1988: 196): "The diachronic phenomenon of morphologization is
exemplified by the movement of syntactic phenomena into morphology as well as by the
movement of phonological phenomena into that domain".
2. Cf. Salmons (1994) for criticisms that however do not concern the global interpretation of
German Umlaut as a case of morphologization of a phonological alternation. Unfortu­
nately, generative phonology does not seem to share the same interpretation of umlauting,
since it still provides a phonology-driven description of the above alternations, even
though moderated by a two-level lexicalist framework (cf. Wiese 1996).

3. For a discussion of the relationship between grammaticalization and lexicalization, cf.

Lehmann (1989).
4. Cf. Hopper & Traugott (1993: 148), Matisoff (1991). Hopper & Traugott (1993: 149)
observe, moreover, that "there is a tonal change, generally from a lower to a mid or high
tone, which is phonetically (albeit indirectly) linked to the voicing change. Where the
initial consonant is one that does not show a distinction in voicing, such as the nasal [m]
or the affricate [c] in the next example, the tonal difference is the only remaining trace of
the former prefix:
(19) a. mo 'see' : mo 'show'
b. câ 'eat' : cā 'feed'".
5. Cf. Paul (1920: 125): "Im Mhd. ist diese Bildungsweise noch einigermaßen lebendig. Im
Nhd. sind manche früher vorhandene Bildungen untergegangen".
Some remarks on analogy, reanalysis and grammaticalization 103

6. Cf. Kiparksy (1992: 56): "Analogy is sometimes postulated as a supplementary mecha­

nism of derivative generation accounting for the use of utterances that are not directly
generatable by the grammar". A completely different question, for which an adequate
answer has not yet been found, is why and where, i.e. on the basis of which model,
analogy takes place. Cf. the discussion ibidem.
7. Notice that the palatalization triggered in the stems originally containing /I/ and still
visible in the form [ ] has practically disappeared in Colloquial Standard Bulgarian
(Jetchev's personal communication), removing the last segmental remnant of the original
phoneme.
8. In this sense, they become "syntactified", cf. Zwicky (1987: 225): "Syntactification is the
counterpart above the level of the word of the familiar phenomenon of morphologization
below the level of the word".
9. Cf. Awbery (1986: 417): "The direct object of an inflected verbs undergoes the soft
mutation [= lenition, LG] provided that it is phrase-initial in the noun phrase".
10. For a (traditional) explanation of the modern situation, cf. Morris-Jones (1931: 195):
"Originally, of course, the case of the noun had nothing to do with its initial mutation; for
mutation depends not upon the old ending of the word mutated, but on the general
character in Brythonic of the ending of the preceding word. Thus the mutation of the noun
depended wholly upon the form of the verb. The development of the modern rule is in
outline as follows: The 3rd pers. sing, was differentiated from the other persons by the
fact that it could be followed by subjects and objects, while the others could be followed
by objects only. Now, it appears that the 3rd pers. sing, generally ended in a consonant in
Brythonic as in Latin, except in the imperfect tense; hence in Medieval Welsh both
subject and object after the 3rd sing, have the radical [unlenited consonant, LG] with a
large proportion of the soft [lenited consonant, LG] after imperfect. The other persons
mostly ended in vowels, thus caraf ('I sing') represents *carami; hence the soft came to
be associated with the object. Thus a new basis of mutation was unconsciously evolved;
exceptions were gradually done away with, and the radical became the sign of the subject,
the soft of the object." For another interpretation of the Welsh mutation, cf. King (1993:
22-3), criticized however by Alan R. King on the Linguist List 7.1046.

11. Cf. Awbery (1975), who speaks of lexical, categorial, structural and transformational
mutation.
12. However, my opinion is that there are still many possible cases to be found: for example,
an early phonological rule that has assumed a grammatical content, which makes refer­
ence to the domain of the sentence, e.g. becoming an interrogative marker.

REFERENCES

Awbery, Gwenllian M. 1975. "Welsh mutations: Syntax or phonology?" Archivum

Linguisticum 6: 14-25.
. 1986. "Survey of sandhi types in Welsh." In Henning Andersen (ed.), Sandhi
Phenomena in the Languages of Europe, 415-33. Berlin-New York-Amsterdam: Mou­
ton de Gruyter [Trends in Linguistics. Studies and Monographs 33].
104 Livio Gaeta

Bauer, Laurie. 1993. Review Article of Becker (1990). In Geert Booij and Jaap van Marie
(eds), Yearbook of Morphology 1992, Dordrecht: Kluwer, 264-7.
Becker, Thomas. 1990. Analogie und morphologische Theorie. Munich: Fink.
Behaghel, Otto. 19285. Geschichte der deutschen Sprache. 5th ed. Berlin and Leipzig: de
Gruyter [Grundriss der germanischen Philologie 3].
Bybee, Joan L. 1985. Morphology. A Study of the Relation between Meaning and Form.
Amsterdam/Philadelphia: John Benjamins [Typological Studies in Language 9].
Dressier, Wolfgang U. 1985. Morphonology. The Dynamics of Derivation. Ann Arbor:
Karoma Press.
Fiedler, H.G. 1928. "Two Problems of the German Preterite-Present Verbs." The Modern
Language Review 23: 188-96.
Greenberg, Joseph H. 1991. "The last stages of grammatical elements: Contractive and
expansive desemanticization." In Traugott & Heine (eds), vol. 1, 301-14.
Hopper, Paul J. 1994. "Phonogenesis." In William Pagliuca (ed.), Perspectives on Gram-
maticalization, 29-45. Amsterdam/Philadelphia: John Benjamins [Current Issues in
Linguistic Theory 109].
Hopper, Paul J. and Elizabeth C. Traugott. 1993. Grammaticalization. Cambridge:
Cambridge University Press [Cambridge Textbooks in Linguistics].
Jetchev, Georgi. 1994. "From early Slavic to modern Bulgarian: a survey of changes in the
vowel system and the syllable structure." Quaderni del Laboratorio di Linguistica della
Scuola Normale Superiore di Pisa 8: 108-14.
Joseph, Brian D. and Richard D. Janda. 1988. "The how and why of diachronic morpholo-
gization and demorphologization." In Michael Hammond & Michael Noonan (eds),
Theoretical Morphology, 193-210. San Diego: Academic Press.
King, Gareth. 1993. Modern Welsh: A Comprehensive Grammar. London: Routledge
[Routledge Grammars].
Kiparsky, Paul. 1992. "Analogy." In William Bright (ed.), International Encyclopedia of
Linguistics, vol. 1, 56-61. New York-Oxford: Oxford University Press.
Klausenburger, Jürgen. 1979. Morphologization: Studies in Latin and Romance Morpho-
phonology. Tübingen: Niemeyer [Linguistische Arbeiten 71].
Krämer, Peter. 1971. Die Präsensklassen des germanischen schwachen Verbums. Inns­
bruck: Innsbrucker Beiträge zur Sprachwissenschaft.
Lehmann, Christian. 1989. "Grammatikalisierung und Lexikalisierung." Zeitschrift für
Phonetik, Sprachwissenschaft und Kommunikationsforschung 42/1: 11-9.
Lewis, Henry and Holger Pedersen. 1937. A Concise Comparative Celtic Grammar.
Göttingen: Vandenhoeck & Ruprecht.
Loporcaro, Michele. 1995. "Raddoppiamento fonosintattico dopo III persone plurali del
verbo nei dialetti di Conflenti (CZ) e di San Giovanni in Fiore (CS)." Rendiconti della
Classe di Scienze Morali, Storiche e Filologiche dell 'Accademia Nazionale dei Lincei,
s. 9, v. 6, 543-53.
Lunt, Horace G. 1974. Old Church Slavonic. The Hague: Mouton.
Matisoff, James A. 1991. "Areal and universal dimensions of grammaticization in Lahu." In
Traugott & Heine (eds), vol. 2, 383-453.
Meillet, Antoine. 1912. "L'évolution des formes grammaticales." Reprinted in Antoine
Meillet. 1948. Linguistique historique et linguistique générale, Paris: Champion, 130-
48.
Some remarks on analogy, reanalysis and grammaticalization 105

Meillet, Antoine. 19304. Caractères généraux des langues germaniques. 4th ed. Paris:
Hachette.
Morris-Jones, John 1931. Welsh Syntax. An Unified Draft. Cardiff: The University of Wales
Press Board.
Paul, Hermann. 1920. Deutsche Grammatik. V Bd. Teil TV: Wortbildungslehre. Halle/S.:
Niemeyer.
Priebsch, R. and W.E. Collinson. 19625. The German Language. 5th ed. London: Faber &
Faber.
Robinson, Orrin W. 1975. "Abstract phonology and the history of umlaut." Lingua 37: 1-
29.
Ronneberger-Sibold, Elke. 1990. "Zum Auf- und Abbau des Rückumlauts bei den
schwachen Verben." In Armin Bassarak et al. (eds), Wurzel(n) der Natürlichkeit.
Studien zur Morphologie und Phonologie TV. Berlin: Akademie der Wissenschaften der
DDR [Linguistische Studien. Reihe A. 208].
Salmons, Joseph C. 1994. "Umlaut and plurality in Old High German. Some Problems with
a natural morphology account." Diachronica 11: 213-29.
Scur, Georgij S. 1961. "Über den Umlaut der deutschen Modalverben." Neuphilologische
Mitteilungen 62: 206-19.
Traugott, Elisabeth C. and Bernhard Heine (eds). 1991. Approaches to Grammaticaliza­
tion, Amsterdam/Philadelphia: John Benjamins [Typological Studies in Language 19],
2 vols.
Vennemann, Theo. 1972. "Rule inversion." Lingua 29: 209-42.
. 1974. "Restructuring." Lingua 33: 137-56.
. 1986. "Ruckümlaut." In Dieter Kastovsky and Aleksander Szwedek (eds), Lin­
guistics across Historical and Geographical Boundaries. In Honour of Jacek Fisiak on
the occasion of his 50th birthday, vol. 1, 701-23. Berlin-New York-Amsterdam: Mou­
ton de Gruyter.
Voyles, Joseph. B. 1991. "A History of OHG i-umlaut." Beiträge zur Geschichte der
deutschen Sprache und Literatur 113/2: 159-94.
Wellmann, Hans. 1973. Deutsche Wortbildung. Das Verb. Düsseldorf: Schwann.
Wiese, Richard. 1996. "German Umlaut and Ablaut." Journal of Linguistics 32: 113-35.
Willis, Penny 1986. The Initial Consonant Mutations in Welsh and Breton, Bloomington:
Indiana University Linguistic Club.
Wilmanns, Wilhelm. 19302. Deutsche Grammatik. II Bd.: Wortbildung. 2nd ed. Berlin and
Leipzig: de Gruyter.
Wurzel, Wolfgang U. 1980. "Ways of morphologizing phonological rules." In Jacek Fisiak
(ed.), Historical Morphology. The Hague: Mouton [Trends in Linguistics. Studies and
Monographs 17], 443-62.
Zwicky, Arnold M. 1987. "French prepositions: no peeking." Phonology Yearbook 4. 211-
27.
 Testing the boundaries of
grammaticalization

Anna Giacalone Ramat

University of Pavia

1. How is the term grammaticalization to be used

From the considerable number of studies on the processes of grammaticaliza­

tion over the last years the conviction has emerged that grammaticalization is
not a uniform process. A task for the further study of this field is to investigate
in depth the many routes along which grammatical forms and constructions
arise. In this contribution I am going to examine a few critical cases of
grammaticalization: my purpose is to try and throw light on boundary areas in
which the model "lexical item ⇒ morpheme", which in most cases has
appeared to be a successful research paradigm, seems insufficient to capture
the changes under scrutiny.
In sections 1 and 2 I examine the form/function units which are involved
in the grammaticalization process and I discuss cases of functional renewal
using in a critical manner the notion of exaptation (Lass 1990). In section 3 I
concentrate my argumentation on a few cases of loss of grammatical function.
In sections 4 and 5 I resume the topic of the boundaries of grammaticalization
in its initial and final stages and evaluate the consequences of previous
discussion on the hypothesis of unidirectionality in the processes of grammati­
calization. The issue remains open for now; however I suggest that a differen­
tiation should be kept between unidirectional processes and other
developments that are not in terms of form and meaning change. This conclu­
sion speaks in favor of the specificity of grammaticalization processes.
In his seminal article "L'évolution des formes grammaticales" (1948
[1912]) Meillet noted that grammaticalization can have two different functions:
108 Anna Giacalone Ramat

1) to create new forms that replace old forms in existing grammatical

structures, which remain essentially the same from the point of view of
function,
2) to introduce into grammar new categories, i.e. new units of form/func­
tion.
The second aspect has aroused particular interest, as is attested by the volume
of research on the rise and development of auxiliaries in the Romance lan­
guages, and of clitic pronouns, also in the Romance languages. But also the
first case, in which change of form is combined with the preservation of
function, has deserved consideration as a major theme of grammaticalization
studies. Among the examples discussed by Meillet, the case of negation well
illustrates the way different forms that express the function of negation
repeatedly replace one another, reflecting an essential linguistic function,
which languages tend to renew from the formal point of view to achieve
greater expressiveness (1948[1912]:139ff).
But also a third possibility in addition to the renewal of forms and the
creation of new categories can be envisaged. This is the renewal of function,
while the form remains constant. In addition to several well known cases of
grammaticalization, such as the development of complementizers from verbs
meaning "say", we may consider here what Lass (1990) calls exaptation. With
this term he refers to changes concerning the allocation of new functions to
elements already belonging to the grammar which have become useless or
"idle". The fourth logically possible case, which corresponds to the combina­
tion of old function with old form is of no interest here, since it concerns cases
of preservation of the state of affairs. To sum up, we can draw the following
chart:

Table 1. Form/function relations in linguistic units

function form type of grammaticalization
I old new formal renewal
(e.g. negation)
II new new creation of new categories
(e.g. Romance auxiliaries)
III new old exaptation (Lass 1990) (?)

As noted in Vincent (1995:437f), the terms 'new' and 'old' here refer to the
grammatical system of a language: thus the 'new' elements which form the
Testing the boundaries of grammaticalization 109

class of the auxiliaries are not forms 'invented' out of nothing, but are present
in the lexicon of the language. They are extra-grammatical elements which
come into the grammar when they undergo the process of grammaticalization.
The term 'old', on the other hand, refers to elements which are already part of
the grammar, usually of the morphological system.

2. Functional renewal

The process whereby old grammatical forms may take on new functions has
often been discussed in the literature in recent years and different terms have
been proposed to describe it. In particular, we will touch briefly on the term
'exaptation' which Lass derives from evolutionary biology1. However, the
adaptation of this concept to linguistic (morphological) change is open to
criticism (cf. Vincent 1995:435ff). The notion of 'linguistic junk' with which
Lass refers to morphological material deprived of any function, free to be
employed again for functions not linked with the previous ones, does not seem
adequate to explain what happens in linguistic change. Usually a morpheme is
not suddenly left without a function, but goes through a period of more or less
casual variation. Moreover, the process of functional emptying in most cases
is not total, but only partial, in the sense that some features can disappear, while
others persist. Linguistic change starts causing a 'reduction' in functionality
rather than the creation of 'marginal garbage'. There is certainly some truth in
the assumption of Lass that a feature (or a morpheme) can be re-employed for
something different from its original use, but some of his own illustrations
challenge the hypothesis of exaptation. Making the case of apophony (Ablaut)
in the system of the Germanic so-called strong verbs, Lass starts off with the
observation that Indo-European had an aspectual system with an opposition
between perfect and aorist, which was later lost and replaced by temporal
oppositions. Part of the original morphology remained in the Germanic strong
verbs, specifically the Indo-European opposition between o-grade for the
perfect and zero grade for the aorist, which was re-employed to indicate within
the preterit category the opposition of singular/plural:
(1) Apophony in Germanic strong verbs
INF. PRET.lSG. PRET.lPLUR.
Goth, beit-an bait bit-um
hilp-an halp hulp-um
110 Anna Giacalone Ramat

This is considered to be a case of exaptation. But in fact the opposition

between o-grade in the singular and zero grade in the plural is a matter of
preservation both in form and function, as can be seen in the point by point
correspondences between ancient perfects:
(2) The pattern of the Indo-European perfect
1.SG. l.PLUR.
Skt. véda vid-má
Gk. ╒oĩδα ╒ĩδ-μεν
Goth, wait wit-um
The Germanic forms continue an ancient Indo-European perfect pattern, as
proved also by forms like OHG ziohan "to pull", pret. zōh zugum, OHG
kiosan "to choose", Pret. kōs kurum, which show the effect of Verner's law in
so-called 'grammatical change'. Therefore the 'conceptual novelty' of this
case dissolves. When Lass (87) declares: "what is of prime interest here is
conceptual novelty: ablaut in the Indo-European sense was never used for this
kind of thing before", he ignores the fact that already in Germanic preterite
presents of the type wait witum, the apophonic opposition marks number. The
principle of formation is ancient. Germanic has extended it by constructing
further apophonic classes based on oppositions in root vowels, e.g. Goth. las
lēsum,preterite of lisan "to read" (Ramat 1988:192). Thus, the change was
made possible by the fact that there existed an older pattern that could lend
itself to extension of use.
Other cases of linguistic change can perhaps be more appropriate in order
to discover instances of exaptation. The notion of 'exaptation' could be
invoked to describe the evolution of the Latin 'inchoative' suffix -ësc-/-ïsc- in
Romance languages, particularly in Italian. Allen (1995) sums up the question
by noting that in Proto-Indo-European the suffix -sk had no inchoative value
(cf. Skt. gacchati "goes"), but was a suffix for forming present stems, some­
times with iterative value: Hitt. daskizzi "takes repeatedly" (Meillet
1934:220-22). In Latin the suffix enjoyed a certain amount of good fortune
and developed, perhaps following the crescō model, an inchoative meaning:
cf. senescö " I grow old", pallescō "I go pale". But for our argument the
subsequent history is of more interest: in fact, this element which was part of
the derivational morphology of Latin was incorporated into the inflectional
system of Italian (Greenberg 1991:311) and became a kind of interfix (Allen
1995). It was desemanticized and re-employed in a new function as a person
Testing the boundaries of grammaticalization 111

marker, to mark the singular persons and the third person plural in the present
tense conjugation, in subjunctive and imperative. In all these forms stress is on
the suffix, while in the first and second person plural forms, where stress is on
the ending, the interfix is absent. Consider for illustration the indicative
present of finiré: finísco/finísci/finísce/finíscono, as opposed to finiámo/finíte.
The same pattern as Italian holds for Spanish and Rumanian, however Ruma­
nian has considerably increased the number of "inchoative" verbs (Allen
1995). In French in the present tense the element -iss- marks the plural with
respect to the singular: je finis/nous finissons; it has moreover been extended
to the imperfect: je finissais, etc. Ramat (1992) treated the case of -isc- verbs
as a case of lexicalization, focusing on the loss of the inchoative function. In
fact, the list of Italian verbs characterized by the -isc- form does not reveal any
discernible meaning for it (Serianni 1989:420; see however Zamboni
1983:233). One might even argue that the affix is redundant for 'person'
marking since person is expressed by the endings. But a leveling function can
still be discerned for the affix in that it allows to fix the stress for the whole
paradigm in a position after the verb stem, which remains unstressed. At least
in some languages "this eliminates vocalic alternation between stressed and
unstressed verb stems" (Haiman 1988:359 for Rheto-Romance and Allen
1993:6 for Rumanian). It seems possible, then, to classify the forms deriving
from Latin -ēsc-/-īsc- as a case of exaptation. The affix seems to have devel­
oped some marginal (phonological) function different from its original
inchoative value and to have not yet reached the stage of phonogenesis (see
below), i.e. of phonological segment. This evolution, however, does not
constitute a straightforward case of grammaticalization in Meillet's concep­
tion of innovative force in language.
Reinterpretation of grammatical elements in a new grammatical function
is called by Greenberg (1991) re grammaticalization. He mainly deals with
the stages of definite articles, but mentions -as I have noted- as a case in point
also the outcome of the Latin inchoative suffix in Romance languages.
Brinton and Stein (1995:34) resort to the term functional renewal to
describe "the retention or revival of an existing syntactic form with a new or
renewed function". The examples discussed by the authors concern some
changes which have occurred in the history of English: the so-called 'conclu­
sive perfect' (I have a paper written), which expresses the state resulting from
an action, which the authors describe as 'reversal of historical trends' as it is a
revival of ancient meanings after a period of quiescence. Another case cited by
112 Anna Giacalone Ramat

Brinton and Stein deals with the use and frequency of structures with inver­
sion: in Modern English the possibility of moving a constituent into first
position in the sentence (fronting) is distinctly expanding as the result of the
development of new discourse functions such as topicalization or focalization
strategies.
What, then, is the relation between exaptation and functional renewal?
Brinton and Stein (44f) suggest that it might be a case of two names for one
phenomenon. However, the resemblance is not at all evident: by the authors'
own admission, the semantic change in the conclusive perfect is a revival of
values already present in a more ancient stage of the language: it is not then a
genuine 'novelty' in the sense of Lass. But a prerequisite for functional
renewal as understood in this paper is that the older function is no more
available. This seems to be the case for inversions, which in Old English
served the function of introducing referents, or were used after initial adverbs
like pa "then", nu "now (Brinton and Stein 1995:38f).

3. Loss of grammatical function without renewal?

In the light of the above considerations, we suggest to define exaptation as

refunctionalization under conditions of discontinuity in the developmental
continuum. This kind of change may be logically complementary to the first
two types described in Table 1, as Vincent suggests (1995:434). But exaptive
changes are not isolated cases. They deserve to be explored in depth in order to
establish to what extent the characterization of unidirectionality is appropriate
for them.
As a starting point, we might recall that Benveniste (1968:85ff) intro­
duced a distinction in the evolution of languages between "innovating muta­
tions", which show up in the disappearance or emergence of categories, and
"conservative mutations". For the latter Benveniste cites as an exemplary
case the periphrastic forms which arise in the evolution from Latin to Ro­
mance to replace categories expressed by morphological means, while he
describes as innovating mutation the loss of the neuter in the gender category
and of the dual in the number category. The loss of the Latin neuter gender
deserves a few comments: we will try to show that the morphology of neuter
gender has been partly utilized with functions different from those of the
denotation of grammatical gender.
Testing the boundaries of grammaticalization 113

The Latin category of gender is reduced to the masculine/feminine

opposition in all Romance languages.2 However, neuter inflections have been
variously preserved in Romance: e.g. Southern and Central Italian dialects
distinguish a masculine article lu or u < Latin illu and a neuter lo or o < illud,
the latter mostly used for mass nouns: cf. o latte "the milk", but u lope "the
wolf' (Rohlfs 1968[1949]§419). But the largest traces of neuter are to be
found in feminine plurals such as Italian le labbra "the lips", le ciglia " the
eyebrows" or feminines singular la foglia "the leaf"< Latin folia (neuter
plural), la pecora "the sheep (sg.)" < Latin pecora (neuter plural). The main
cause of the decadence of gender systems in languages is the erosion of the
formal marks which distinguish them (Corbett 1991:315). In Latin the neuter
nouns of the second declension in -urn, distinguished from the masculine
nouns only in the nominative and vocative singular and in the nominative/
accusative/vocative plural, were mostly absorbed into the masculine class.3
What is particularly important here is that a part of the neuter plurals was not
dumped, but was reinterpreted as feminine: this difference in outcome sug­
gests that the re-assignment operated on the basis of a purely formal resem­
blance, the neuter plural ending -a being identical with the singular of the
feminine nouns in -a. However, if the formal equivalence justifies the Italian
type foglia < Lat. folia (the feminine singular is already attested in late Latin),
a different explanation based on functional principles is needed for that
substantial group of nouns which has gone over to the feminine gender, but
has remained plural: to cite some examples: Lat. labia > It. le labbra (with
masculine singular il labbro), Lat. brachia > It. le braccia, sing. il braccio.
The model has also been productive, since it has been extended to a few
masculine nouns: Lat. digitus (masc.) "finger" > It. il dito, le dita, Lat. fructus
(masc.) "fruit" > It. il frutto, i frutti, but also la frutta. Despite the many
analogical formations, a semantic core can be singled out for the nouns with
plural in -a. Such a prototypical value can be found in the designation of
double members (already of neuter gender in Latin: brachia, cilia, labia), as
Magni (1995) proposes, or in the designation of "totality" which allows to
include collective items such as Italian le ossa "the bones", le membra "the
limbs", le mura "the walls", le fondamenta "the foundations", etc. In actual
fact the Italian plurals in -a introduce a distinction, a subcategory within the
category of number. This state of affairs lends itself to treatment in the
framework of exaptation: the neutral plural form in -a has been refunctional-
ized as mark of collective plural. It could be questioned whether the collective
114 Anna Giacalone Ramat

value is a real innovation or rather the original collective meaning of the Latin
ending -a has re-emerged;4 but what is relatively new is the development of
an explicit form to express a sort of holistic vision in contrast with masculine
plural endings. To be noted that for all nouns which show a collective
feminine plural in -a, a masculine plural in -i was developed, with an indi­
vidualizing value (Serianni 1989:143ff, Brunei 1978:30ff). Thus the different
"plurals" correlate with divergent meanings.5 The collective feminine plural
in -a is a marked class, less frequent (today not productive), which constitutes
an innovation, a renewal of function for an old form.
As far as the theory of grammaticalization is concerned, it is of special
interest to note that the reassignment of a new functional value to the Latin
neuter forms took place in parallel with the disappearance of the neuter as a
grammatical category. Indeed, the data show variations and fluctuations in Late
Latin leading us to think that the propagation of change took a long time. Finally,
the forms in -a came to be employed for other more basic functions (according
to Greenberg's universals, number is more basic than gender). As an outcome
of such change, the category of number has gained a new distinction. But the
change in itself doesn't belong to the typical cases of grammaticalization along
the lines set for historical morphological change because it shows a kind of jump
from one function to the other. This point may be relevant for the problem of
limits of grammaticalization and will be resumed below.
The last step in the process of functional loss can be "the creation of
phonological segments out of earlier morphemes", a process for which Hop­
per (1994:32) has suggested the term phono genesis. This is a more radical
phenomenon than the ones previously discussed and catalogued under the
name of 'exaptation', because renewal of function is excluded. One example
put forward by Hopper concerns the German prefixes g(e)-, b(e)- found in
forms that can no longer be analyzed by speakers: bleiben "to remain"<be-
līben, genug "enough", genügen "to be enough", vergnügen "to amuse" (the
forms -leib or -nug do not exist and the former prefixes are pure phonological
segments).
Hopper points out that the process of phonogenesis "presupposes a
seamless continuum between phonology and grammar" (42) and defends the
idea that phonogenesis is part of grammaticalization in the later stages "in
which meanings typically become more diffuse"(39). It should be noted,
however, that phonogenesis implies a total loss of function and exit from
grammar. When it becomes impossible to attribute any grammatical function
Testing the boundaries of grammaticalization 115

to one or more phonological segments, such remnants do not appear to take

part in the grammaticalization process any longer, rather they contribute in
some way to the creation of new lexical material.

4. Perspectives on directionality

The claim that grammaticalization is irreversible, that all grammaticalization

involves shifts in specific contexts from lexical item to grammatical item, or
from less grammatical to more grammatical item (Hopper and Traugott
1993:128) is one main tenet of grammaticalization theory. In this section I will
discuss some changes which for different reasons challenge the principle and I
will re-examine the cases of functional renewal presented thus far in order to
evaluate their relevance for the unidirectionality principle.
A consequence of the unidirectionality hypothesis specifies that gram­
matical elements do not turn back in the direction of the lexicon. Let us
consider the formulation given by Bybee, Perkins and Pagliuca (1994:13):
"Once affixation has occurred, grams do not ordinarily detach themselves and
assume a free form again, so that growing dependence on surrounding lexical
material is not usually reversed". Counter-examples are rare, but not as rare as
some scholars seem to assume. The authors cite as "a very rare case of
change" Irish pronouns, where a change from affix to autonomous form took
place for the first person plural ending -mid/-muid, thus proceeding in the
opposite direction to what was expected.
Numerous instances of changes proceeding from grammar to the lexicon
come from those grammatical morphemes that receive a new function as
autonomous lexical items, such as English ade (from lemonade, etc.) "fruit
juice", teens "age between thirteen and nineteen", Italian anta "age from forty
upwards". Ramat (1992) interpreted these cases as examples of de-grammati-
calization intended as a process which goes backwards along the pathway of
grammaticalization, while Hopper and Traugott (1993:127) consider them
lexicalization or "recruitment of linguistic material to enrich the lexicon".
Other counter-examples can be collected from the literature such as preposi­
tional or adverbial elements ending up as verbs, at least in informal speech: cf.
English to up the sale price. According to Hopper and Traugott (1993:126ff),
however, the impact of such problematic changes for the theory is low, since
they are somewhat idiosyncratic cases and infrequent.
116 Anna Giacalone Ramat

Yet there are other problematic cases, in addition to the creation of new
lexical items: the view of unidirectionality that we shall call 'linear' is called
critically into question by all those cases that seem to prove a discontinuity, or
a reversal of historical trends. In this perspective the examples of exaptation
or renewal of function previously discussed deserve further examination. As I
have noted above, the case of the formal mark of gender that goes over to
indicating number shows a discontinuity in unidirectional development. The
morphological marks have not followed the expected course of grammatical-
ization over time, which in this case would result in zero (loss of overt
segment) as a consequence of the loss of the neuter, but have slid towards
adjacent (and more central) areas of morphology. In fact the renewals of
function examined, although their outcome remains the expression of gram­
matical distinctions through affixed elements, do not entail any directionality
of development.
In the following I would like to discuss two further examples of changes
in grammar, which do not conform to the unidirectionality hypothesis. A
critical case for unidirectionality is represented by increase in morphological
complexity, if we take for granted what Lehmann (1985:307, 1995:132)
claims, namely that morphological degeneration is an attribute of grammati-
calization. Morphological degeneration is defined as the loss of the ability to
inflect. It can be shown, however, that sometimes the process of grammatical-
ization brings about rise rather than loss of inflections. This may happen when
personal pronouns grammaticalize to verb inflections involving a richer dis­
play of morphological distinctions than the original verbal paradigms. A case
in point is provided by South Alemannic (Walser) dialects spoken in Italy
(Giacalone Ramat 1990, 1992, Dal Negro 1996). While a series of weak
subject pronouns in postverbal position was developed in all Alemannic
dialects (including Swiss dialects) (Nübling 1992), in Southern Walser dia­
lects of Italy a further development is taking place which is typical of ongoing
grammaticalization. Subject pronouns increasingly tend to be suffixed to the
verb (univerbation), irrespective of the V2 constraint which would require
subject-verb inversion and consequently suffixation in non interrogative sen­
tences only if the first position is taken by a constituent different from the
subject. This change which calls into question one major rule of German
syntax could have taken place under Italian influence in language shift
situation, since in Italian overt subject pronouns are not required (Giacalone
Ramat 1992, Nübling 1992: 259). Be it as it may, the grammaticalization of
Testing the boundaries of grammaticalization 117

subject pronouns results in an increase of morphological complexity since

clitic pronouns are reanalyzed as verbal inflections through boundary weak­
ening. Thus clitics tend to become fixed in their syntactic position, giving rise
to a new configuration of verbal paradigms where new distinctions are
present, including a gender distinction triggered by the affixation of 3rd
person singular pronouns inflected for gender. For example, in the Walser
dialect of Gressoney (Aosta Valley) we find:
(3) hätter gseit "has-he said"
hätsch gseit "has-she said"
häts gseit "has-it said"
Apparently, the change has progressed at different speed in the South-Walser
area, along uniform patterns of development and with the dialect of Rimella
leading the process: in it the first person plural has only the cliticized forms:
gawer "go-we", asswer "eat-we".
My second example concerns the development of clitic pronouns to
affixes, which on the basis of a vast evidence is regarded as a cline of
grammaticality reflecting a unidirectional movement. There are, however,
problematic cases that seem to proceed in the opposite direction. Portuguese
has an inflected infinitive {infinitivo pessoal) with person/number endings as
well as an impersonal infinitive form. The inflected forms probably derive
from the Latin imperfect subjunctive which was reanalyzed as infinitive
(Lausberg 1969, Parkinson 1988, Silva Neto 1988):
(4) depois de partires, o Jorge chegou
after leave-INF-you-suB-CL, George arrived
"after you left, George arrived"
Viewed synchronically, the inflections of the inflected infinitive behave more
like subject clitics than verbal endings, since they can alternate with the simple
infinitive: cf. the following example from Körner (1983:87).
(5) é melhor nos sair cedo, apois de meus irmãos chegar
é melhor nos sair cedo, apois de meus irmäos chegarem
it is better that we get off early, after my brothers arrive-INF-they-CL
The whole process seems to involve a reanalysis of inflections into clitics and
to reverse the usual grammaticalization path from clitic to affix. Evidence for
a similar kind of development is provided by Janda (1995): Regional Spanish
118 Anna Giacalone Ramat

affix -mos "1st person plural" > nos "subject marker".

To resume, I have argued that in dealing with possible counter-examples
to the hypothesis of unidirectionality it seems appropriate to distinguish
different subcategories of unexpected developments. I have suggested to
distinguish non-continuity of historical development whose results are still
within the boundaries of grammar (as in the case of neuter gender), from those
cases in which grammatical elements slide in the direction of the lexicon as in
ade or anta. A third type of problematic changes was seen in the bi-directional
clitic/affix developments. Such evidence is not to be minimized or disre­
garded, however it does not contradict the claim of a genuine directional
pathway regarding the nature of the grammaticalization process. Such pro­
posed counter-examples cannot be recognized as true reversals of unidirec­
tional processes in the sense of return to the original state. On the whole they
represent local changes effected under marked contextual or sociolinguistic
conditions (informal speech, or contact situations).

5. Testing the boundaries of grammaticalization

Since the current paper focuses primarily on the limits of grammaticalization,

let us now briefly turn to some considerations on the category "auxiliary".
Despite the many problems in determining the entities that should belong to
this category, or in dealing with their various behavior, the development of
auxiliaries is considered prototypical of grammaticalization (Heine 1993,
Heine et al. 1993). The category is new in Romance languages, even though
Latin esse "to be" was used for analytic forms in the passive perfectum and in
the perfectum of so-called deponent verbs; moreover, for some Latin verbs
the onset of the auxiliation process may be located in the Late Latin period
(Pinkster 1987). Auxiliaries nicely illustrate all crucial assumptions of the
grammaticalization theory: gradualness, overlapping, noncompletion, loss of
semantic content and categorial properties (Heine 1993, Hopper and Traugott
1993, Bybee, Perkins and Pagliuca 1994).
Besides the two auxiliaries "be" and "have", which appear to be com­
pletely grammaticalized, though with different distribution in Romance lan­
guages and dialects (Vincent 1982, Green 1982, Ramat 1987, Squartini
1995), there is another group of less grammaticalized forms which allow us to
single out a continuum of grammaticalization and to highlight the scalar
Testing the boundaries of grammaticalization 119

nature of the auxiliary category. Among Romance languages, Italian and

Spanish in particular tend to extend the use of periphrases with auxiliaries and
to increase the number of members of the category, which includes verbs of
motion ("come", "go"), verbs of position ("stay", "remain"), leaving aside the
constructions with so-called 'aspectuals': "continue", "finish", etc..
Let us consider as an example of incomplete grammaticalization the
Italian motion verbs andare "go" and venire "come" when used as auxiliaries
for the passive. First of all, in that function they share a specific feature of all
auxiliaries, that is they have no longer an argument structure of their own;
moreover they undergo a series of morphosyntactic and semantic-lexical
restrictions, as shown by Giacalone Ramat (1995b): venire is excluded from
compound tenses, as in (6), while andare is used only with verbs denoting
disappearance, destruction or damage of the Patient (7). In addition to that,
andare has developed a deontic reading for simple tenses, except simple past
(8). Some examples to illustrate the point:
(6) *questa casa è venuta venduta l'anno scorso
this house is come sold last year
"this house was sold last year"
(7) questa casa è andata distrutta durante il terremoto
this house is gone destroyed during the earthquake
"this house was destroyed during the earthquake"
(8) questa facccnda va sbrigata immediatamente
this matter goes dealt with immediately
"this matter has to be dealt with immediately"
The defective paradigm of andare and venire in auxiliary function is an
essential part of their process of grammaticalization, since increase in mor­
phosyntactic restrictions is a characterizing feature, according to Heine
(1993: 58ff).
On the other hand, the incomplete grammaticalization of andare and
venire is to be attributed to the partial preservation of their semantic content
as motion verbs (semantic retention: Bybee, Perkins and Pagliuca 1994). The
lexical restrictions between andare and the verb associated with it seem to
depend on the fact that the prototypical meaning of atelic movement away
from the speaker or from the deictic center still persists. The two auxiliaries
are not equivalent and interchangeable: one of the diversification features
120 Anna Giacalone Ramat

consists of the fact that andare with passive value does not admit expression
of the agent, which on the contrary is possible with venire (Giacalone Ramat,
1995b).
These explorations on the development of auxiliaries from verbs of
motion do not challenge unidirectionality: they rather illustrate the long
lasting ambiguity in the initial stages of grammaticalization (Heine 1993:48f).
In Italian for many centuries up to now the same verbs have been referring to
lexical or to grammatical concepts. The boundary between the lexical and the
grammatical domain is in such cases difficult to trace.
There is considerable disagreement among linguists on where to set the
dividing line between grammaticalization and lexicalization. On the one
hand, lexicalization is sometimes considered as the final stage of grammati­
calization. On the other, in linguistic literature the same phenomena are
sometimes cited as exemplary cases of either linguistic process.
The development of adverbs from inflected nouns is one instance of
lexicalization for Anttila (1989:151) who declares: "when an adverb splits off
from a noun, it has to be learned separately and is thus a new lexical item". On
the other hand, Hopper and Traugott (1993:131f) consider adverbial forma­
tion in Romance languages as "a straightforward instance of grammaticaliza­
tion: a new grammatical formative has come into existence out of a formerly
autonomous word", involving semantic change and decategorization from
noun to affix. Indeed, Lat. mente is an ablative form, an inflected form which
has split off to become an adverb formation device. It is certainly not appropri­
ate to take -mente as a lexical device.
The development of pronouns into conjunctions (e.g. the English
complementizer that from demonstrative pronoun), is again a case of lexical­
ization according to Anttila (151). In my view, such development lends itself
to treatment as increased grammaticalization of already grammatical items
which serve to express the relations between clauses (Hopper and Traugott
1993). The problem with Anttila's suggestion is that he seems to reject the
possibility that the canonical cline:
lexical item > clitic > affix
is not completed, but only parts of it undergo evolution: e.g. pronouns
(autonomous words) may shift to conjunctions (autonomous words as well) in
specific collocations. The meaning shift is accompanied by decategorization,
i.e. loss of properties of the pronominal item along a pathway leading from a
Testing the boundaries of grammaticalization 121

paratactic pattern of juxtaposed clauses to embedding through a complemen­

tizer of the that-typo (see Hopper & Traugott [1993:185ff] on that-comple-
mentation in the history of English). Both the source and the target forms in
this development are autonomous words, no affixation is involved: what
changes are the syntactic and pragmatic contexts in which particular colloca­
tions may arise. The study of grammaticalization does not only concern the
origin of affixes; autonomous words also may reflect grammaticalization. This
discussion shows that apparent instances of lexicalization can be better de­
scribed as grammaticalization processes.
In the classical version of grammaticalization going back to Meillet
(1912:131), all grammatical items originate in lexemes. According to this
view, the lexicon feeds grammar, while, at the opposite end of the cline,
grammatical elements in their later stages may simply disappear or survive as
phonological segments (phonogenesis). Despite the large amount of evidence
in this sense building up a corpus of typical cases of grammaticalization, I
argue that at the end of the cline the limits between grammatical elements and
lexemes may be blurred, and propose to rethink the traditional view according
to which grammaticalization and lexicalization are quite distinct, even oppo­
site processes. Rather, they seem to be complementary or overlapping and
processes of change such as loss of autonomy or univerbation are similar both
in grammar and in the lexicon.
For our purposes it is important to stress that grammatical materials may
become lexical through a number of developments which do not mirror the
ones occurring at the initial stages of grammaticalization of lexemes. What is
striking is that all counter-examples to unidirectionality discussed in the
literature, including cases like Italian anta, etc., refer to idiosyncratic
changes. It is not possible to identify a tendency of language change, as in the
case of unidirectional changes.
In investigating more closely the final stages of grammaticalization vari­
ous instances of lexicalization can be shown:
1) gradual development across time of affixes or prefixes may lead to lex­
emes in which the original affix is no more recognizable: this is the case of the
Latin comparative suffix -ior- in Italian signore, French seigneur, with se­
mantic shift from "older" to "noble, respectable man". The process could lend
itself to an interpretation in terms of phonogenesis, as Lazzeroni (in press)
proposes, if we look at the fate of the comparative suffix. But at the same time
the result of the process is a new lexeme, a new form/meaning unit, not just a
meaningless phonological sequence.
122 Anna Giacalone Ramat

2) compound words tend to become opaque, as time passes: "only specialists

of etymological research know that English lord comes from Anglo-Saxon
hlaf+weard 'bread-guard'" (Hagège 1993:183).
3) gradual evolution of constructions across time may lead to new lexemes:
well known cases are OHG hiu tagu (old Instrumental case) > German heute,
Latin ad ipsum (Accusative) > Italian adesso "now". These examples show
lexicalization emerging from syntagmatic units via univerbation: Hopper and
Traugott (1993:23) question Meillet's suggestion (1912:138f) that German
heute is an instance of grammaticalization showing strong phonetic reduction
typical of such "mots accessoires".
Also larger units or phrases entering highly conventional collocations
may give rise to lexemes. Italian forse "perhaps" derives from a Latin con­
struction fors sit an "be the case that..", French peut-être, Dutch misschien,
Danish måske, kanske "perhaps" mean "it may be" or "it may happen"
(possibly a European pattern: Ramat and Ricca 1994:297f). Here belongs also
the example (already in Meillet 1912:145) of Modern Greek tha used as a
morpheme introducing future tense, which derives from Ancient Greek thelô
hina "I wish that...". Lexicalization of syntactic units is also shown by the
"forget-me-not" type (cf. Moreno Cabrera, this volume).
4) finally, cases of periphrastic constructions undergoing a process of
'idiomatization' might be considered here. A case in point would be the
Italian aspectual periphrasis venire+gerund, a grammaticalized construction
expressing continuous and iterative aspect which has undergone obsoles­
cence and reduction in use in Modern Italian (Giacalone Ramat 1995a). This
is a reversal of the tendency to expand contexts for grammaticalized forms
(Heine et al. 1991) and its final outcome is the creation of almost idiomatic
expressions to be located in the lexicon rather than in grammar.
The discussion in this section has made reference to various ways of
creating lexical items out of morphological or syntactic constructions. Such
cases apparently contradict the predictions made by unidirectionality, accord­
ing to which grammatical morphemes do not become lexical morphemes.
However, the constraint of irreversibility should not apply to cases which are
not point-for-point reversals of any grammaticalization process. Since con­
structions becoming one lexeme involve the interaction of diachronic pro­
cesses pertaining to the lexicon, one possibility would be not to include them
into the class of changes a theory of grammaticalization should explain.
As has been noted in the opening section, multiple paths of development
Testing the boundaries of grammaticalization 123

have emerged from work on grammaticalization. Based on a critical discus­

sion of Lass (1990) and Vincent (1995), this paper has tried to show that the
kind of processes involving a refunctionalization of old grammatical forms,
represents a serious challenge to the unidirectionality hypothesis. For concep­
tual clarity they should be kept apart from cases of grammaticalization as a
process proceeding from lexical forms to grammatical forms along a unidirec­
tional cline. These considerations lead us to conclude that grammaticalization
is a specific form of language change. Another possibility would be to reject
unidirectionality as a necessary condition to decide what changes a grammati­
calization theory should explain. I feel, however, that in this case what we call
"grammaticalization theory" would be left with a too vague definition of its
field, including almost every instance of change. The unidirectionality of
changes from lexical categories to grammatical (functional) categories consti­
tutes a significant constraint on possible language changes. In the light of this
constraint, possible counterexamples can be excluded because they do not
adhere to the sequence of changes entailed in grammaticalization.

NOTES

1. In evolutionary biology exaptation is defined as those useful structures that arose for
other reasons, or for no conventional reasons at all, and were then fortuitously available
for other changes (Gould 1983:171, cited by Lass 1990:80) and exemplified by the
development of feathers by those dinosaurs that are the ancestors of birds. Feathers had
originally the function of preserving body temperature and were later "opportunistically
capitalized on or coopted for flight".

2. Rumanian has a class of nouns which are masculine in the singular and feminine in the
plural ('third class nouns'). Some of them continue the Latin neuter pattern: timp, timpuri
"time". Some linguists (Rosetti 1985) see the third class nouns as a direct inheritance
from Latin neuter class which was reinterpreted as representing the inanimate category.
Others (Mallinson 1986:246) point to influence from Slavic languages for the Rumanian
three gender system.
Schön (1971) gives a general overview of remnants of Latin neuter forms in
Romance languages: some cases of gender distinctions pertain to the lexicon rather than
to grammar, as French le cerveau "the mind" vs. la cervelle "the brain". To be also noted
the Swiss Rheto-Romance la bratscha "both arms" which is a collective singular (Schön
1971:87).
3. The loss of gender "may result from a conspiracy of contributing factors" (Corbett
1991:316): besides phonological change, which remains the main factor, a restructuring
on the basis of parameters such as animate/inanimate can contribute, and also a change in
the view, or the role of derivational morphology.
124 Anna Giacalone Ramat

Inconsistencies in gender assignment can be found in Latin: e.g. aevus/aevum and

also feminine doubles in -a: armentum (n.), but armenias in Ennius and Pacuvius, or
rāpum and rāpa "turnip" (Väänänen 1981:102).
4. Kuryłowicz (1975:53) recalls that the Latin neuter plural ending -a has its origin in an
Indo-European collective of feminine gender, which originally had its own declension
and was successively incorporated in the declension of neuters as nominative-accusative-
vocative plural. The link between feminine and collective is therefore very old: the events
in later Latin and the Romance languages might contain a conservative element rather
than being a totally innovative change.
5. The meaning of the plural in -i in this class of nouns is defined as "analytical" by Serianni:
e.g. "the masculine lenzuoli indicates two or more lenzuoli taken one by one...the
feminine lenzuola indicates the pair of sheets used to make a bed" (1989:144-45). In Old
Italian the feminine plural in -a concerned a much larger number of nouns: le castella, le
intestina, le anella, etc.(Serianni, loc.cit., Magni 1995:158ff). The extension of use of the
morpheme -a might be the signal of the increase in functionality of the morpheme as a
collective mark.
Another plural form in -ora is found in Medieval Latin and Old Italian: corpora,
pratora, luogora, as well as in a few modern dialects (Aebischer 1933, Rohlfs 1968
[1949]§370; Väänänen 1981:105). The formation is transparent: the model is offered by
the type tempora or corpora, in which the ancient Indo-European suffix for the formation
of neuter nouns -es-/-os- has been re-analysed as part of the ending, with the consequent
creation of a new morpheme of the plural -ora. The more salient ending -ora seems to
have been favored in some ancient documents (Aebischer 1933); however its evolution,
which would require to be examined in more detail, is quite independent from -a plurals
and not central to our argument.

REFERENCES

Aebischer, Paul. 1933. "Les pluriels analogiques en -ora dans les chartes latines de l'Italie."
Archivum Latinitatis Medii Aevi 8: 5-76.
Allen, Andrew. 1995. "Regrammaticalization and degrammaticalization of the inchoative
suffix." In Henning Andersen (ed.), 1-8.
Andersen, Henning (ed.). 1995. Historical Linguistics 1993. Amsterdam/Philadelphia:
John Benjamins.
Anttila, Raimo. 1989. Historical and Comparative Linguistics, 2nd edition. Amsterdam/
Philadelphia: John Benjamins.
Benveniste, Emile. 1968. "Mutations of linguistic categories." In Winfried P. Lehmann
and Yakov Malkiel (eds), Directions for Historical Linguistics. A Symposium, 85-94.
Austin & London: University of Texas Press.
Brinton, Laurel and Dieter Stein. 1995. "Functional renewal." In Henning Andersen (ed.),
33-47.
Brunet, Jacqueline. 1978. Grammaire critique de l'italien. vol 1. Vincennes: Université
de Paris VIII.
Bybee, Joan, Revere Perkins, and William Pagliuca. 1994. The Evolution of Grammar.
Testing the boundaries of grammaticalization 125

Tense, Aspect and Modality in the Languages of the World. Chicago: The University
of Chicago Press.
Corbett, Greville. 1991. Gender. Cambridge: Cambridge University Press.
Dal Negro, Silvia. 1996. "Fenomeni di grammaticalizzazione e decadenza linguistica nel
titsch di Formazza." Linguistica e Filología 2, 123-134: Universit degli Studi di
Bergamo.
Giacalone Ramat, Anna. 1990. "Il mutamento linguístico in contesto plurilingue." In
Liliana Spinozzi Monai (ed.), Aspetti metodologici e teorici nello studio del pluri-
linguismo nei territori dell'Alpe Adria, 79-100. Udine: Aviani Editore.
. 1992. "The pairing of structure and function in syntactic development." In Marinel
Gerritsen and Dieter Stein (eds), Internal and external factors in syntactic change, 317-
339. Berlin: Mouton De Gruyter. [Trends in Linguistics: Studies and Monographs 61].
. 1995a. "Sulla grammaticalizzazione di verbi di movimento: andare e venire +
gerundio." Archivio Glottologico Italiano LXXX: 168-203.
. 1995b. "On some grammaticalization patterns for auxiliaries." Paper delivered at
XII ICHL, University of Manchester, August 1995 (to be published in D. Bentley & J.C.
Smith (eds), Proceedings of the XII ICHL. Amsterdam/Philadelphia: John Benjamins).
Green, John N. 1982. "The status of the Romance auxiliaries of voice." In Nigel Vincent
and Martin Harris (eds), 97-138.
Greenberg, Joseph. 1991. "The last stages of grammatical elements: contractive and expan­
sive desemanticization." In Elisabeth C. Traugott and Bernard Heine (eds), vol. I, 301-
314.
Hagège, Claude. 1993. The language builder: An essay on the human signature in
linguistic morphogenesis. Amsterdam/Philadelphia: John Benjamins.
Haiman, John. 1988. "Rhaeto-Romance." In Martin Harris and Nigel Vincent (eds), 351-
'390.
Harris, Martin and Nigel Vincent (eds). 1988. The Romance Languages, London &
Sydney: Croom Helm.
Heine, Bernd. 1993. Auxiliaries. Cognitive forces and grammaticalization. New York
Oxford: Oxford University Press.
Heine, Bernd, Ulrike Claudi and Friederike Hünnemeyer. 1991. Grammaticalization: A
conceptual framework. Chicago: The University of Chicago Press.
Heine, Bernd, Tom Güldeman, Christa Kilian-Hatz, Donald A. Lessau, Heinz Roberg,
Mathias Schladt and Thomas Stolz. 1993. Conceptual Shift. A lexicon of grammatical­
ization processes in African languages. Afrikanistische Arbeitspapiere 34/35.
Universität zu Köln.
Hopper, Paul. 1991. "On some principles of grammaticalization." In Elizabeth C.
Traugott & Bernd Heine, vol. I, 17-35.
. 1994. "Phonogenesis." In William Pagliuca (ed.), Perspectives on Grammatical­
ization, 29-45. Amsterdam/Philadelphia: John Benjamins.
Hopper, Paul and Elisabeth C. Traugott. 1993. Grammaticalization. Cambridge: Cam­
bridge University Press.
Janda, Richard D. 1995. "From agreement affix to subject "clitic" — and bound root:
-mos > -nos vs. (-)nos(-) and nos-otros in New Mexican and other regional Spanish
dialects". Papers from the Chicago Linguistic Society 31:118-139.
126 Anna Giacalone Ramat

Körner, Karl-Hermann. 1983. "Wie originell ist der flektierte Infinitiv des Portugiesi­
schen? Eine Studie zum Subjekt in den romanischen Sprachen." In J. Schmidt-
Radefeld (ed.), Portugiesische Sprachwissenschaft, 77-103. Tübingen: Narr.
Kuryłowicz, Jerzy. 1965=1976. "The evolution of grammatical categories." Reprinted in
Jerzy Kurylowicz, 1976, Esquisses linguistiques, vol. II, 38-54. München: Fink.
Lass, Roger. 1990. "How to do things with junk: exaptation in language evolution."
Journal of Linguistics 26: 79-102.
Lausberg, Heinrich, 1969. Romanische Sprachwissenschaft III: Formenlehre, Berlin: De
Gruyter.
Lazzeroni, Romano. Forthc. "Divagazioni sulla degrammaticalizzazione." In Giuliano
Bernini, Pierluigi Cuzzolin and Piera Molinelli (eds), Ars linguistica. Studi per Paolo
Ramat. Roma: Bulzoni.
Lehmann, Christian. 1985. "Grammaticalization: Synchronic variation and diachronic
change." Lingua e Stile 20: 303-318.
. 1995. Thoughts on Grammaticalization. München-Newcastle: Lincom Europa.
First published as Arbeiten des Kölner Universalien-Projekts 48. Köln 1982.
Magni, Elisabetta. 1995. "II neutro nelle lingue romanze: tra relitti e prototipi." Studi e
Saggi Linguistici 35: 127-178.
Mallinson, Graham. 1986. Rumanian. London and Sydney: Croom Helm.
Meillet, Antoine. 1912=1948. "L'évolution des formes grammaticales." In Linguistique
historique et linguistique générale, Paris: Champion, 130-148.
Meillet, Antoine. 1934. Introduction à l'étude comparative des langues indo-européennes
(7ème éd.). Paris. Librairie Hachette.
Nübling, Damaris. 1992. Klitika im Deutschen: Schriftsprache, Umgangssprache, aleman­
nische Dialekte. Tübingen: Narr.
Parkinson, Stephen. 1988. "Portuguese." In Martin Harris and Nigel Vincent (eds), 131-
169.
Pinkster, Harm. 1987. "The strategy and chronology of the development of future and
perfect tenses auxiliaries in Latin." In Martin Harris and Paolo Ramat (eds), Historical
Development of Auxiliaries, 193-223. Berlin: Mouton De Gruyter.
Ramat, Paolo. 1987. "Introductory paper." In Martin Harris and Paolo Ramat (eds),
Historical development of auxiliaries. Berlin-New York-Amsterdam: De Gruyter.
. 1988. Introduzione alla linguistica germanica2. Bologna: Il Mulino.
. 1992. "Thoughts on degrammaticalization." Linguistics 30: 549-60.
Ramat, Paolo and Davide Ricca. 1994. "Prototypical adverbs: On the scalarity/radiality of
the notion of ADVERB." Rivista di Linguistica 6: 289-326.
Renzi, Lorenzo (a cura di). 1988. Grande grammatica italiana di consultazione. vol. I.
Bologna: II Mulino.
Rohlfs, Gerhard. 1949=1968. Historische Grammatik der italienischen Sprache und ihrer
Mundarten. II Formenlehre und Syntax. Berlin: A. Francke AG. Italian translation:
Grammatica storica della lingua italiana e dei suoi dilaletti. II Morfología. Torino:
Einaudi: 1968.
Rosetti, Alexandru. 1985. "Sur le neutre en roumain." In La linguistique balkanique, 360-
384. Bucure§ti: Editura Univers (reprinted from Etudes linguistiques. Den Haag Paris
1973, 59-71).
Testing the boundaries of grammaticalization 127

Salvi, Giampaolo. 1988. "La frase semplice." In L. Renzi (a cura di). Grande grammatica
italiana di consultazione, 29-113.
Schön, use. 1971. Neutrum und Kollektivum. Innsbruck: Institut für vergleichende
Sprachwissenschaft.
Serianni, Luca. 1989. Grammatica Italiana. Italiano comune e lingua letteraria. Torino:
UTET.
Silva Neto, Serafím da. 1988. Historia da lingua portuguesa. 5a ediçâo. Rio de Janeiro:
Presença.
Squartini, Mario. 1995. On the Grammaticalization Path of Some Romance Verbal
Periphrases. Tesi di perfezionamento. Pisa: Scuola Normale Superiore.
Traugott, Elisabeth C. and Bernd Heine (eds). 1991. Approaches to Grammaticalization.
Amsterdam/Philadelphia: John Benjamins.
Väänänen, Veikko. 1981. (3 éd.). Introduction au latin vulgaire. Paris: Klincksieck.
Vincent, Nigel. 1982. "The development of the Auxiliaries HABERE and ESSE in Ro­
mance." In Nigel Vincent and Martin Harris (eds), 71-96.
. 1995. "Exaptation and grammaticalization." In Henning Andersen (ed.), 433-445.
Vincent, Nigel and Martin Harris (eds). 1982. Studies in the Romance Verb. London:
Croom Helm.
Zamboni, Alberto. 1983. "Note aggiuntive alla questione dei verbi in -isco." Studi di
grammatica italiana 12:230-237.
 Discourse and pragmatic conditions
of grammaticalization
Spatial deixis and locative configurations
in the personal pronoun system of some
Italian dialectal areas*

Stefania Giannini
Università per Stranieri di Perugia

0. Introduction

This paper explores the insertion of locative adverbs (qui = "here", lì/là =
"there") in the personal pronoun system (luqqui/lullì "he here", "he there"
types) of a north-central Italian dialect (the Lucca dialect), both in a
synchronic (how the person-deictic system works) and in a diachronic per­
spective (how the grammaticalization process may have arisen).
On a first level the study focuses on the semantic and pragmatic condi­
tions which determine the grammatical and lexical structure of the person-
deictic pronouns in Lucchese. As previous research has clearly showed (cf.
Hopper & Traugott 1993), shifts from lexicon to grammar can be variously
conditioned by discourse and pragmatic forces.1 In this paper I want to show
that the insertion of spatial and deictic indicators in the third-person pronouns
largely depends on discourse strategies typically selected in face-to-face
interaction. The two questions we are about to answer are the following:

* A wider paper concerning the same topic was published in Archivio Glottologico Italiano
Vol. LXXX - Fasc, I/II (1995:204-238).
130 Stefania Giannini

(a) What linguistic and discourse conditions have allowed deictic adverbs to
gain ground in the cline of grammaticality (from content word to function
word status ?)
(b) To what extent can the process here discussed be considered as a case of
grammaticalization?
On the second level we will try to illustrate the historical and geographi­
cal roots of the process, taking into account some other tendencies well-
attested in that linguistic area since Old and Middle Lucchese. Definiteness,
as a semantic and conceptual category, seems to have played a relevant role in
selecting new linguistic items expressing the specificities of space.2 The
person-deictic pronoun system is supposed to be imported from north dialec­
tal varieties, whose history was closely linked to Lucca's northern posses­
sions.

1. Analysis and interpretation of data

The Lucca dialect belongs to the north-west Tuscan variety and it is spoken in
the municipal land surrounding the town of Lucca. As regards its internal
structuring, Lucchese seems to be a heterogeneous system, both in a diastratic
(dialect spoken by educated speakers vs. dialect spoken by uneducated; this
distinction mostly reflects the opposition between urban and rural dialect) and
in a diatopic perspective (east and southern vs. northern idioms). The phe­
nomena we are discussing in this paper are generally widespread in the
system, although it must be pointed out that they become more frequent in the
northern rural variety.3
The Lucchese personal pronoun system is affected by final vowel dele­
tion (apocope in traditional terms of Italian dialectology, cf. Rohlfs 1968), as
with most central Italian dialects (Tuscany and Umbria). Personal pronouns
lu' "he/him" - le' "she, her" are differentiated with respect to gender (male vs.
female, respectively). Such a system is synchronically productive and re­
places the common standard personal forms (lui, lei); see, for instance:
(1) Lu' è arivato tardi, ma a le' unn'importava
'He was late, but she did not care'.
Besides this system (which we propose to define as Simple Personal
Pronoun (SPRO, from now on) there are also personal pronouns produced by
Discourse and pragmatic conditions of grammaticalization 131

the merging of personal items with locative adverbs qui "here", lì- là "there",
su "up", giù "down". The effect of this process is the following pattern:
(i) pronouns encoding expression of proximity:
proximal luqqui/leqqui vs. non proximal lullì/lullà; lellì/lellà
(ii) pronouns encoding expression of spatial configuration (without
strictly deictic reference):
lussù, lessù vs. lug giù, leggiù
For simplicity of exposition, we shall define this personal pronouns
system we are presenting here as Complex Personal Pronouns (CPRO from
now on). Locative and deictic adverbs cannot associate with other stressed
personal forms such as the second-person *teqqui. This property is surely
related to the fact that the third-person pronoun is neutral with respect to the
participant-roles of the speech act (speaker and listener). As Lyons has
pointed out (1977:638), "it does not correlate with any positive participant
role". 4
CPRO is employed in a large range of different speech events. Depend­
ing on the presence or absence of personal referents the pronoun is referring
to, one or another form will be chosen as in the following:
(2) Luqqui un ista bene
(3) Lullì un ista bene
(4) *Lullà un ista bene
(5) *Lu' un ista bene
'He is not well'.
(6) Me l'ha detto lullà
(7) Me l'ha detto lu'
(8) *Me l'ha detto luqqui
'He told me that'.
(9) Lu' è partito ieri
(10) * Lullà è partito ieri
(11) * Luqqui è partito ieri
(12) *Lullì è partito ieri
'He left yesterday'.
132 Stefania Giannini

(13) N'hai messo a lullì il cappellino ?

(14) *N'hai messo a lu' il cappellino
'Did you put on his hat?'.
(15) Vada un po ' luqqui che fa!
(16) Vada un po ' lullì che fa!
(17) *Vada un po' lu' che fa!
'Look at what he is doing !'.
(18) Io un lo so, chiedini a lullì
(19) * Chiedini a luqqui
(20) * Chiedini a lu'
'I don't know, ask him !'
(21) Leggiù/lessù umm 'ha pagato ancòra la pigione
'She hasn't yet paid the rent'
On the basis of the data quoted above, semantic and pragmatic properties
of CPRO can be gathered. The sentences suggest several different kinds of
situation:
(a) In (2)-(5), CPRO specifies the proximity (+ /- proximal) of referent with
respect to the speaker and listener. In such a function, it cannot be replaced by
SPRO.
(b) Concerning (6)-(8), it should be noticed that the action referred to is at a
distance (in time and space) from the speech event. Consequently, forms
encoding proximity (luqqui type) would make the sentence unacceptable. For
the same reasons, deictic pronouns expressing non proximity (lullà type,
where là indicates non proximity to the zero-point of the deictic context) are
allowed.
(c) Examples (9)-(20) refer to persons who physically participate in the
speech act. This automatically determines the selection of CPRO.
(d) Finally, as regards the situation described in (21), the zero-point is here
the bodily mass of the speaker (as for Bühler 1965=[1983]).5 Deictic specifi­
cations here referred to are up or down with respect to the speakers spatial
position in the speech event. Luggiù or lussù (I choose here the masculine
form as the unmarked member of gender opposition, but the same can be said
for the feminine ones) must be interpreted as "he who is usually situated
above / under the place I am". Such a spatial configuration generally refers to
neighbours or people whose usual location is already known by listeners.
Discourse and pragmatic conditions of grammaticalization 133

A third and less frequent form parallels luggiù/lussù types, being consti­
tuted by sequential association of nonproximal deictic adverb (là) + spatial
specification expressed by su/giù. A four-member system follows from the
combination of two spatial references with the personal base-form, conserv­
ing the same gender opposition as CPRO:
(22) lullassù/lullaggiù
'he up/down there'.
(23) lellassù/lellaggiù
'she up/down there'
This less frequent pronoun system can be employed in cases such as
those described in:
(24) Lullassù s'è scordato di noaltri
'God the Father has forgotten us'
(25) Lellaggiù non s'è piùfatta viva
'she, who lives over there, hasn't shown up any more'
Therefore, spatial indications recoverable from the double locative mark­
ers are always referred to the normal location the referent spatially keeps. In
such a way, God the Father can be addressed as "that who lives up there in
heaven", speaker and listener bearing in mind his actual location, as it is
depicted in the common iconography shared by speakers from Catholic
communities.
On both semantic and pragmatic grounds, the personal and deictic pro­
noun system of the Lucca dialect can be interpreted by means of the following
parameters:
(a) parameter of proximity/non proximity: origo - speaker and lis­
tener spatial position
luqqui proximal to the speaker
lullì proximal to the listener
(b) parameter of presence /absence: origo = speech event
luqqui present at the speech event
lullî absent from the speech event
(c) parameter of the spatial configuration: origo = ego of the speaker
lussù above me (=speaker)
luggiù under me (=speaker)
134 Stefania Giannini

The CPRO system specifies the spatial co-ordinates of personal entities

they are referring to, merging two grammatical categories in the same linguis­
tic unit: person and space. According to Heine, Claudi and Hünnemeyer
(1991:157) grammatical categories can be arranged along a linear scale
whose individual steps represent a gradual path starting from the most con­
crete categories (on left) to the most abstract ones (on right):
(d) Person > Object > Process > Space > Time > Quality
Grammaticalization processes usually reflect the tendency of categories to shift
from left to right (in some languages, for example, there are attested forms
which express the metaphor "time is space"). In the same perspective, we can
assert that the spatial metaphor encoded in CPRO implements the person
category with additional spatial information. In other words, space becomes a
member of the person category and it seems to be essential for referring to
[+human] referents. Therefore, grammaticalization of deictic and locative
adverbs in a personal pronoun system can be interpreted as a problem-solving
strategy which translates a more abstract concept into a more concrete speci­
fication. CPRO syncretically associates information about "who" (traditionally
assigned to the linguistic category of person) to information about "where"
(traditionally expressed by the linguistic category of space). It could be said that
in Lucchese CPRO we can observe a clear example of the localization/
spazialition of person.
Furthermore, CPRO can be said to be produced by the need for identify­
ing and spatially collocating referents, which can be mostly found in 'face-to-
face' speech events. These situations are especially common in linguistic
interaction between speakers belonging to small and narrow communities.
A very similar view has been taken by sociolinguistic and dialectologic
field researchers. Recent work on the problem is reviewed by Sobrero (1993:
442), who concludes that the conceptual representation of space varies ac­
cording to the "cognitive pattern/habit" (stile cognitivo) of each linguistic
community. Thus in very narrow and well known places spatial conceptual­
ization and its linguistic counterpart become, in some way, simpler than in
large and less delimited places.6 This has important consequences for the case
at issue. The renewal of a personal pronoun system through the use of an
original locative adverb rests on a set of pragmatic conditions:
Discourse and pragmatic conditions of grammaticalization 135

(e) Speaker and listener knowledge mostly overlap (deictic function)

(f) Oral structuring of discourse (topic continuity)
(g) Concreteness of discourse subject-matter (identifying function)
Each feature links to specific properties of the face-to-face conversa­
tional pattern which is characteristic of small communities. Broadly speaking,
the first way in which one may identify an object (either a thing or a human) is
by locating it for the addressee (Lyons 1977:648). Common and direct knowl­
edge of people and places (including the properties they have and their home-
base in the village) thus forces the speaker to intensify the use of deictic
expressions (in other languages, demonstrative pronouns mostly operate as
referring and deictic indicators: see this and that in colloquial English) (e). In
doing so, the speaker simultaneously informs the addressee that a particular
individual is being referred to: spatial information contributes to making the
identifying process successful (g). The third pragmatic component in CPRO
is the so-called topic continuity, by which is meant, as is suggested by Givón
(1983), a specific role that certain elements can be said to have for keeping
the discourse topic alive. These coding points are distributed on a hierarchical
scale, whose topic continuity strength decreases from right to left:

(h) definite and indefinite articles > demonstratives > case-markers

For spoken Italian, it has already been pointed out (Duranti 1980:118)
that stressed personal pronouns (lui, lei) and, on the other side, the linear
association of demonstrative + proximity or remoteness deictics (the questo
qui/quello lì types) are the best attested forms in those speech events where
turn-taking is freely decided by speaker-listener alternation. Luqquì/lullì can
play the same role, as may be illustrated by means of utterances like:
(26) S. Ma chi dici?
'Who are you talking about?'
'Who do you mean?'
(27) L. Lullì/lullà, che lavora con te
'Him there, who works with you'
On the evidence of such examples, topic continuity does look relevant.
CPRO really works as a morphosyntactic device referring to lexical items
previously mentioned in the discourse. In the hierarchical scale above-quoted
(h), it is situated in the same position as demonstratives and deictic forms are.
136 Stefania Giannini

The difference between lì and là, equally allowed in utterances which

obey the same conditions as (26)-(27), is rooted in the different attitude that
the speaker takes up towards the referent. Lullà will be selected when the
speaker implicitly underlines a certain remoteness between the person re­
ferred to and himself; it means something like "he there, who I feel very far
from me", also metaphorically speaking without pointing to his spatial posi­
tion. Lullì, as it is opposed to lullà, does not contain such a pragmatic nuance.
We will not investigate here the intriguing matter of the metaphorization of
spatial perception to express psycho-affective attitudes (Lakoff and Johnson
1980:14-21). Suffice it to say that under certain conditions some grammatical
elements, usually devoted to expressing proximity/remoteness as mere physi­
cal conditions, can be used to encode far more complex affective interplay
which links the participants to each other.
In order to focus more clearly upon the nature of CPRO and the subunits
it is composed of, we will briefly examine their morphological and syntactic
properties, in formal and distributional terms. Deictic and locative adverbs
which represent the input of grammaticalization as autonomous lexemes
show a particular behaviour in the process. It can be schematized as in (i-iv)
below:
(i) They (i.e. locative adverbs) morphologically depend on the pro­
nouns' base-form
(ii) They cannot be transposed in word-initial position (from word-
final position) (*qui lu)
(iii) They cannot undergo the insertion of other items (*luX qui)
(iv) They are stressed (and perceptually prominent) [luk'kwi]
According to the properties (i-iii) deictic adverbs would seem to fall
within the realm of bound morphemes; but it also holds true that they maintain
morphological invariability (gender markers are signaled by the personal
base-form). Prosodic relevance can be typical of derivative morphemes (see
It. -oso, famoso "famous" < Lt. -ōsus), whereas it does not usually affect
inflectional; thus the Italian oxytones (as for città "town", caffè "coffee",
tribu "tribe") are mostly ruled as monomorphemic words. Also according to
general prosodic criteria, a distinction can be drawn between deictics consti­
tuting CPRO and clitics, inasmuch as clitics can both precede base-forms (and
are proclitic as in Engl, 's me) and follow them (and are enclitic, as in Engl.
Fm). -qui,-lì as morphological formatives of luqquì, lullì are postponed and
Discourse and pragmatic conditions of grammaticalization 137

stressed. Prosodically, indeed, in CPRO the personal pronoun base-form

tends to be proclitic.
Considering now the well-known cline of grammaticalization proposed
by Hopper and Traugott (1993):
(v) a. content item > b. grammatical word > c. clitic >
d. inflectional word
it can be observed that deictic adverbs, as they became bound forms
inside the emergent CPRO, have reached position c. in terms of lexical
autonomy, position d. in terms of morphological features and position b. in
terms of semantic properties.
The results given here for morphological and syntactic properties of
deictics indicate an affirmative answer to the question whether or not we have
to do with a case of grammaticalization. Nevertheless, the individual func­
tions and values that the formative has acquired during the process seem to
proceed and stop at different stages along the way from word to morpheme.

2. Geographical spread and diachronic roots of CPRO

Searching now for the geographical distribution and diachronic roots of

CPRO, an attempt must be made to answer the following crucial questions:
(a) Where does CPRO come from ?
(b) Can it be thought of as an internal creation of the Lucchese
linguistic system, or is it easier to conceive of its external roots?
The former interpretation can be accounted for by finding philological
evidence of the phenomenon in older linguistic periods. The second hypoth­
esis can be explained by means of historical and geographical reconstruction
of plausible antecedents.
Both paths have been taken, but the second has borne more fruit. Briefly
setting out the different steps of my inquiry, let us first envisage the picture we
find on looking at the internal structure of the Lucca dialect from the medieval
era to more recent linguistic documents (the 18th and 19th centuries). Vulgar
and documentary texts, as published by Marcheschi (1983) and Castellani
(1990), have been examined for ancient times, while theatrical and poetic
poems (edited by Ambrosini 1979) have been selected for exploring the
language of the last two centuries.
138 Stefania Giannini

The first point to be made is that diachronically Lucchese clearly seems

to lack direct and 'premonitory signs' of the subsequent complex pronominal
system. Notwithstanding, documentary research has revealed two facts of
utmost importance, which together account for the emergence of a morpho­
logical component. In fact, Lucchese had a noticeable gap in demonstratives
since ancient times: undoubtedly there were no traces of cotesto/codesto "this
here", a widespread form uniformly found throughout the rest part of
Tuscany. It is not without consequence for the thesis here suggested that such
a gap partly coincides with the semantic and pragmatic functions of CPRO.
Actually cotesto (still alive in spoken and familiar Tuscan) is the indexical
member of the demonstrative pronoun system.
This fact had already been pointed out in pioneering morphological notes
by Pieri (1890:163), who concluded "The dialect of the town and the sur­
rounding land lacks cotesto". Pieri also suggested that the absent form was
replaced by periphrastic forms constituted by a nonproximal demonstrative +
remoteness deictic adverb: quello lì would operate as a substitute for the
missing cotesto at the end of last century. In the subsequent development of
spoken dialect such an expression has quickly disappeared in favour of the
CPRO luqquì/lullì, whose range by now includes utterances implying deictic
reference (as it has been outlined above). It may be added, however, that
embedding a locative adverb in demonstrative pronouns testifies to the se­
mantic erosion of the demonstrative itself, and the linguistic system was ready
to remedy such a deficiency by introducing replacement forms having the
same function. In other words, this morphological gap can be said to represent
a weak point in the pronouns and referring expressions of Lucchese (in
comparison with what is well documented for the other Tuscan varieties) and
is the first reason why a new-coined form could find an easy way to enter and
settle in the system.
On the other hand, the diachronic perspective has brought to light inter­
esting morpho-syntactic phrases, which variously encode the semantic cat­
egory of definiteness. They represent a class of singular definite referring
expressions observable across the centuries at least from the 18th up to the
present-day language, and whose basic function is the indirect allocution. The
pattern works as in the following schema: proper name + preposition di
(which replaces Latin genitive case in most Romance languages) + patro­
nymic:
Discourse and pragmatic conditions of grammaticalization 139

(28) Mauro di Simon dar

PropN 'of' Prep Patronymic 'from' Prep+Art
Trogolo
Locative Specific.
(29) Ceceo di Lampedon Castagna
PropN 'of' Prep Patronymic
(30) Momo di Giulian dar Cesso
PropN 'of'Prep Patronymic ' from' Prep+Art Loc.Specif.
and so on. The locative specification can be omitted.
An example of this allocutionary pattern was discovered by Ambrosini
(1979) in a eighteenth-century poetic text. It is a list of the family-heads
drawn up on the occasion of a funeral. It is reasonable to think of the
determinative and referential phrase "X di Y" as being a very widespread
allocutive form. Such a pattern survives in Modern Lucchese in a tripartite
structure, whose members are the following:
(31) Beppe di Viviana
X (Proper Name) + di + Y(Proper Name)
(32) Dora la Mennucci
X (Proper Name) + Def.Art. + (Family Name)
(33) Mario il cicciaio
X (Proper Name) + Def.Art. + "butcher" (dialect.word) (nomen
agentis)
It will be noted that in (31) Y can be embodied by different determiners:
either close relatives such as parents, brothers (in this case the phrase must be
read as "X, the son/brother of Y"), or husband and wife (thus the meaning will
be "X, the husband/wife of Y"). As far as (32) is concerned, personal deixis,
normally expressed in Modern Standard Italian by means of proper name+
family name (the Mario Rossi type), has been modified by the insertion of the
definite article. The identifying property of proper names turns out to be
increasingly powerful; and the task of the structure is to let the addressee
immediately identify referents and collocate them in their own place in the
social community. Successfulness of indexical specification is guaranteed by
the common knowledge of people and places: the more similar speaker and
listener encyclopedia is, the easier and securer it is for the addressee to identify
140 Stefania Giannini

its referent. Such a condition is perfectly satisfied in small and narrow

communities, where giving information about a [+human] referent automati­
cally implies telling the addressee "what family it belongs to" and "what its job
is". These two pieces of qualitative information fulfill the range of specifying
properties normally encoded by personal pronouns and demonstratives.
We can argue therefore that the referential patronymic and determining
phrases represent the second plausible reason why indexical function might
also be implemented in the pronominal system.
So far we have been dealing with internal evidence. Given the external
roots of CPRO, a further search strategy would appear to be to investigate its
potential source (either geographically or historically). Neighbouring dialects
gave evidence for its northern origin. Two geographical areas especially
exhibit personal pronoun systems variously consisting of locative and deictic
indicators. This is true for some Apennine dialects (Grizzana, Lizzano in
Belvedere, situated in the province of Bologna) and for Lunigiana (a region
bordering on the northwest part of the province of Lucca and included in the
municipal land of the town of Massa). In the last case, personal pronouns very
similar to the pattern of CPRO are well-attested, such as lu-kvì "this one (for a
male)" and le-kvì "this one (for a female)". It is worth noticing that
Lunigianese and Apennine Emilian dialects include a plural form too (lor-là),
which is excluded from the Lucchese CPRO system (with the exception of
some sporadic cases such as lóro là). That Lucchese lacks the plural form can
be understood as a sign of the chronological priority of the Emilian system. It
can reasonably be assumed that CPRO reached the Lucca municipal area
down the course of the river Serchio (see the map). As previous research has
clearly shown (Ambrosini 1978), many other interesting phonological and
morphological features have entered Lucchese dialect along the same route.
CPRO seems to be part of a set of imported linguistic elements. To what
extent West Tuscany dialects owed new structures to northern idioms is a
matter for further investigation. What can be said here is that the undoubted
specificity (in the realm of Tuscany) of such a linguistic variety could be
partly explained and justified on these grounds also.
Putting together the various facts examined in the previous pages, one
may summarize with the following conclusions, as an attempt to answer the
main questions we started from:
(a) Grammaticalization of locative adverbs in the personal pronoun
system of Lucchese has primarily been achieved by pragmatic and discourse
forces. By means of the syncretic association of two grammatical categories
Discourse and pragmatic conditions of grammaticalization 141
142 Stefania Giannini

(person and spatial deixis), the new system expresses new identifying and
referential properties.
(b) The pragmatic strategies that CPRO encodes reflect the general
representation of space as it is assignable to narrow linguistic communities.
Space becomes an essential feature of the person category. It has been
discovered that the semantics of space (by means of locative expressions) and
the pragmatics of speech participation (across the person category) interpen­
etrate with one another in such a way that what (i.e. the qualitative informa­
tion about animate referents) is synergically associated with where (the
locative specification of the place where referents are).
(c) Diachronically CPRO seems to originate from northern neighbouring
dialects which variously possessed similar constructs. A morphological gap
(Lucchese lacking the widespread indexical pronoun cotesto) and a highly
developed referring nominal system created fit conditions for importing a new
complex form. It is an appealing idea that not only certain developments of
grammatical structures (as for the classic and well known movement from
lexicon to grammar, which represents the core of grammaticalization), but also
the stabilization and success of their results can profoundly rest on the way
speakers live and represent their social space.
And agreeing with the assumption that "Pragmatics is the study of those
relations between language and context that are grammaticalized, or encoded
in the structure of a language" (Levinson 1983:9), then the study of how and
why the structure of a language modifies over time can not leave out of
consideration the social and cultural context in which such a language is
spoken.
In other words, if it is commonly accepted by linguists that "today's
morphology is yesterday's syntax" (to quote a famous slogan by Givón),
could it be equally true that "today's grammar is everyday pragmatics"?
Further investigations, I hope, will enable us to answer such an intriguing
question. The luqquì/lullì case can simply testify to the rightness of assuming
that socio-cultural conditions of speakers (that is to say: kinds of family and
community they belong to) deeply interplay with the grammatical representa­
tion of space.
Discourse and pragmatic conditions of grammaticalization 143

NOTES

1. This topic has not yet been investigated: namely, it must be clarified to what extent
pragmatics can be discovered as the set of non-linguistic causes which determine how
speakers codify the external world. We implicitly accept the definition of pragmatics
proposed by Levinson (1983:9): "Pragmatics is the study of those relations between
language and context that are grammaticalized, or encoded in the structure of a lan­
guage".
2. We are referring here to definiteness as a conceptual category, which is often grammati­
calized in natural languages. In this sense it has been defined by Givón (1978:296): "The
notions 'definite' and 'indefinite', so far as referential nominals are concerned, are used
here strictly in their discourse-pragmatic sense, i.e. 'assumed by the speaker to be
uniquely identifiable to the hearer' vs. 'not so assumed', respectively".
3. There is no up-to-date linguistic analysis and description of Lucchese. Texts we can
usually refer to for diachronic evidence are Nieri (1901) and Pieri (1890).
4. Lyons (1977:638) also argued that: "Only the speaker and addressee are actually partici­
pating in the drama. The term 'third person' is negatively defined with respect to 'first
person' and 'second person' [...] The so-called third-person pronouns are quite different
in this respect from the first-person and second-person pronouns."
5. Egocentricity is commonly identified as the canonical situation-of-utterance; see Lyons
(1977:636 sgg.): "The canonical situation-of-utterance is egocentric in the sense that the
speaker, by virtue of being the speaker, casts himself in the role of ego and relates
everything to his viewpoint. He is at the zero-point of the spatiotemporal co-ordinates of
what we will refer to as the deictic context".
6. Sobrero (1993:421) underlines the importance of individual perception of space in
determining the linguistic specification of locations: "La deissi spaziale è anche legata
alla percezione individuale dello spazio, la quale - oltre che a fattori psicologici - a sua
volta è legata alla rappresentazione mentale dello spazio che 1'individuo acquisisce nelle
prime fasi dell'apprendimento: dunque, la scelta deittica dipende, in ultima analisi, anche
dalla storia e dalla cultura (in senso antropologico) della famiglia e delia comunità a cui
appartiene il parlante".

REFERENCES

Ambrosini, Riccardo. 1978. "Di alcuni toponimi della valle della Lima e della posizione
del dialetto lucchese." Vitalia Dialettale XLI [Nuova serie XVIII]: 8-28.
. 1979. "Testimonianze lucchesi della seconda metà del '700." L'Italia Dialettale
XLII [Nuova serie XIX]: 7-23.
Ambrosini, Riccardo. 1980. "Appunti lucchesi." Vitalia Dialettale XLIII [Nuova serie,
XX]: 1-35.
Blasco Ferrer, Eduardo. 1992. "Io e te." Studi Linguistici Italiani XVIII/I: 45-71.
144 Stefania Giannini

Bottiglioni, Giovanni. 1911. "Note morfologiche sui dialetti di Sarzana, San Lazzaro,
Castelnuovo Magra, Serravalle, Nicola, Casano, Ortonovo." Revue de dialectologie
romane III: 339-401.
Bühler, Karl. 1965. Sprachtheorie. Die Darstellungsfunktionen der Sprache, Gustav
Fischer Verlag [trad. it. 1983. Teoria del linguaggio. Lafunzione rappresentativa del
linguaggio, Roma: Armando].
Castellani, Arrigo. 1990. "Capitoli d'un'introduzione alla grammatica storica italiana. V:
Le varietà toscane nel Medioevo." Studi Linguistici Italiani XVI [IX della nuova serie]
fasc. II: 155-222.
Cherubini, F. 1827. Vocabolario mantovano-italiano, Milano: Gio. Batista Bianchi.
Clark, Eve. 1978. "Locationals: existentials, locative, and possessive constructions." In
John Greenberg (ed.), Universals of Human Language, vol. 4, Syntax, 85-126. Stanford:
Stanford University Press.
Coronedi Berti, Carolina. 1869-1874. Vocabolario bolognese-italiano. Bologna: Stab.
tipografico di G. Monti.
Danziger, Eve. 1994. "Out of sight, out of mind: person, perception, and function in
Mopan Maya spatial deixis." Linguistics 32: 885-907.
Duranti, Alessandro. 1980. "Sull'uso dei pronomi tonici nelle conversazioni." In
Pierangiolo Berrettoni (a cura di), Problemi di analisi linguistica, 103-123. Roma.
Fillmore, Charles, J. 1971. Toward a Theory of Deixis, The PCCLLU Papers (Department
of Linguistics), University of Hawai.
Giannarelli, D. 1913. "Studi sui dialetti Lunigianesi compresi fra la Magra e l'Appennino
Reggiano." Revue de dialectologie romane V: 261-311.
Giannini Giovanni and Idelfonso Nieri. 1917. Lucchesismi. Manualetto per lo studio del
vernacolo in relatione con la lingua, Livorno: Raffaello Giusti editore.
Givón, T. 1976. "Topic, pronoun and grammatical agreement." In Charles Li (ed.), Subject
and Topic, 149-188. New York Academic Press Incorporated.
. 1978. "Definiteness and referentiality." In J.H. Greenberg (ed.), Universals of
Human Language, vol. 4, Syntax, 291-330. Stanford: Stanford University Press.
. 1983. Topic Continuity in Discourse: Quantitative Cross-Language Studies.
Amsterdam/Philadelphia: John Benjamins.
Heine, Bernd, Ulrike Claudi, and Friederike Hünnemeyer. 1991. "From cognition to
grammar. Evidence from African languages." In Elizabeth Closs Traugott and Bernd
Heine (eds), 149-187. Approaches to Grammaticalization. Amsterdam/Philadelphia:
John Benjamins.
Hopper, Paul and Elizabeth Closs Traugott. 1993. Grammaticalization. Cambridge: Cam­
bridge University Press.
Lakoff, George and Mark Johnson. 1980. Metaphors we live by. Chicago: The University
of Chicago Press.
Levinson, Stephen C. 1983. Pragmatics. Cambridge: Cambridge University Press.
Levinson, Stephen C. and John B. Haviland. 1994. "Introduction: Spatial conceptualiza­
tion in Mayan languages." Linguistics 31: 613-622.
Lyons, John. 1977. Semantics, I-II, Cambridge: Cambridge University Press.
Maccarrone, N. 1914. "Appunti sulla lingua di G.A. Faye, speziale lunigianese del sec.
XV". Archivio Glottologico Italiano XVIII: 475-532.
Discourse and pragmatic conditions of grammaticalization 145

Malagoli, Giovanni 1940. "Appunti di Morfología e Sintassi del Dialetto di Lizzano in

Belvedere." Vitalia Dialettale XVI: 191-211.
Marcheschi, Daniela. 1983. Ingiurie, improperi, contumelie etc. Saggio di lingua parlata del
Trecento cavato dai libri criminali di Lucca per opera di Salvatore Bongi. Nuova
edizione rivista e corretta con introduzione, lessico e indici onomastici a cura di Daniela
Marcheschi, Lucca: Maria Pacini Fazzi.
Marello, Carla. 1994. "Deissi." In Gian Luigi Beccaria (a cura di), Dizionario di
Linguistica e di filología, metrica e retorica, 203-204. Torino: Einaudi.
Nieri, Idelfonso. 1901. Vocabolario lucchese. Lucca: Tipografía Giusti.
Pieri, Silvio. 1890. "Appunti morfologici, concernenti il dialetto lucchese e il pisano."
Archivio Glottologico Italiano 12: 161-180.
Rohlfs, Gerhard. 1968. Grammatica storica della lingua italiana e dei suoi dialetti.
Morfología, II. Torino: Einaudi.
Salvioni, Carlo. 1905. "Appunti sull'antico e moderno lucchese." Archivio Glottologico
Italiano 16/3: 395-477.
Sobrero, Alberto A. 1993. "Pragmatica". In Alberto A. Sobrero (a cura di). Introduzione
all' italiano contemporaneó. Le strutture. I. Roma-Bari: Laterza, 403-450.
Tiraboschi, A. 1873. Vocabolario dei dialetti bergamaschi antichi e moderni. 2ed.,
Bergamo: Tipografía Editrice Fratelli Bolis.
Traugott, Elizabeth Closs and Bernd Heine (eds). 1991. Approaches to Grammaticaliza­
tion. Amsterdam/Philadelphia: John Benjamins.
 The paradigm at the end of the universe*

Paul Hopper
Carnegie Mellon University

The ninth century German grammarian Smaragdus noted, in elegiac couplets,

a startling fact about the category Adverb:
Pars omnis proprium tandem si perderit actum,
Istius in vasto gurgite mersa iacet,
Adgregat ut pelagus partes sorbendo marinum,
Et violenter eas ad sua iura trahit.

"If in the end the parts of speech lost their own proper functions, they would
lie drowned in the vast whirlpool of adverbs. The adverb draws to itself
parts of speech like the all-absorbing sea, and violently forces them to obey
its laws." (Michael, 1970, p. 73.)

It is perhaps the first statement of a principle of inescapable weakening

and generalization among categories, the first recognition of "naturalness"—
underlined by the metaphor of the invincible forces of nature—in linguistic
change, and the first statement of category shift, that is, of decategorializa-
tion, from major categories to more general 'graveyard' categories such as
adverb.
In recent years, the story of the weakening, bleaching, and pragmatic
strengthening of lexical forms has been a preoccupation of students of gram-
maticalization. The Carolingian landscape in which Smaragdus' grammar is
set invites an obvious allegory: grammaticalization can be thought of as a
salvation narrative. It is the tragedy of lexical items young and pure and
strong in heart but carrying within them the fatal flaw of original sin; their

* Thanks are due to Chris Werry for editorial assistance on this paper, including a number of valuable
suggestions.
148 Paul Hopper

inexorable weakening as they encounter the corrupt world of Discourse; their

fall into the Slough of Grammar; and their eventual redemption in the cleans­
ing waters of Pragmatics.
The story of grammaticalization can be seen as a trope, a conventional if
somewhat repetitive plot. As with any good yarn, we are inclined to suspend
our curiosity about the early childhood of the hero and what happened to him
after the last chapter. Indeed, not to do so would incur the danger of changing
the genre from salvation narrative to Bildungsroman, or perhaps from a
Christian narrative of redemption to a Buddhist one of a cycle of birth, death,
and rebirth. The paradigm at the end of the universe that I talk about in this
paper is as much a story of beginnings as of finishings, and it is with these
'ends' that I will be concerned with, what I will call incipient grammaticaliza­
tion and dissipating grammaticalization. I will refer to grammaticalization as
it is presented in studies that conform to the received narrative, and which
focus on grammatical forms as these are conventionally understood, as 'ca­
nonical grammaticalization.'
The exclusive attention given to canonical grammaticalization no doubt
has a historical reason. Since Meillet, linguists have seen grammaticalization
as an answer to the question of the origins of grammatical forms (cf. Hopper
and Traugott 1993:1-31). These forms have furthermore been seen as lexical
in origin, a view that as it is presently understood also goes back to Meillet.
The first chapter of the grammaticalization narrative has therefore been
framed as a set of lexical items, and the last chapter as a set of corresponding
grammatical items, and the salvation story as what goes on in between.
In fact, for Meillet, the story was not confined to grammatical forms,
since he includes the shift from Old High German hiu tagu "on this day" to
New High German heute "today" in his discussion of grammaticalization, a
change which some would see as lexicalization (Meillet 1912). It is worth
emphasizing that because there is no accepted definition of grammar, there
can be no distinct process of grammaticalization. A wider view of grammati­
calization demands a modification of our perspective on grammar, one which
sees structure in language as intrinsically unfixed and unstable, in other words
as emergent (Hopper 1987, 1988). From this perspective, there would be no
neat distinction between lexicalization and grammaticalization, or any other
aspect of change. It seems reasonable to suggest that the term 'grammatical­
ization' itself is too narrow to accommodate these perspectives. We need
perhaps to think in wider terms, and to adopt the sociologists' term 'structura-
The paradigm at the end of the universe 149

tion' (Giddens 1984), or John Haiman's 'routinization' (Haiman, 1994).

A full account of grammaticalization in this wider sense demands an
account not only of canonical grammaticalization, but also of the incipient
and dissipating ends. This means paying more attention to groups of words,
rather than individual lexical items, especially in the earlier stages. In the later
stages of grammaticalization, it means that we must study the renovation of
the lexicon through the sedimentation of disused morphemes (cf. Hopper
1992), and the range of phenomena involved in 'dying' grammaticalization.
In his treatment of grammatical innovations in German, Lehmann has
drawn attention to a crucially important fact about incipient grammaticaliza­
tion. He notes: "An examination of my data reveals a feature that so far seems
to have gone unnoticed in research on grammaticalization: The complex
prepositions arise in a narrowly circumscribed set of contexts" (Lehmann,
1991:503). Some of the examples he gives are:
(1) im Wege "by (way/means of)"
(2) im Vorfeld "on the eve (of), before"
(3) im Gefolge "as a sequel (of), after" (Lehmann 1991:501)
Lehmann notes that the prepositional complexes are not contextually
neutral, but begin in specific types of text. For example, im Wege "by means
of' always appears in the context of some new technology, such as data
storage; im Vorfeld "on the eve of' always refers to a social or diplomatic
activity such as a meeting; im Gefolge "as a consequence of' in the context of
a trendsetting activity, preferably scientific. A purely structural account of
grammaticalization is likely to overlook the very precise cultural sources of
the collocations that stand at the head of the drift toward grammaticalization.
A second example of this is Mendoza-Denton's historical study of the
preposition concerning (Mendoza-Denton, 1995). Görlach (1991 [1978]:108-
110) has pointed to the emergence of a set of prepositions derived from
participles from the 14th century onward, and attributes it generally to the
growing demand for precision in the written language. He does not, however,
trace the step-by-step linguistic movement from participle to preposition in
texts. Mendoza-Denton, in her lecture at the Linguistics Institute in Albuquer­
que in 1995, focused on the textual environment that fostered the grammati­
calization of "concerning." Originally the present participle of a verb
meaning "see, discern," concerning comes to be detached from the predicate
of the sentence in very specific contexts, those of legal statements, where it is
150 Paul Hopper

used to announce a title for an act, as in:

(4) 1438 Petition to the council concerning the ship Maryknight.
(5) 1500-1570 An acte concerning phesicions & surgeons.
Forasmoche as the science and connyng of Physyke and Surgerie...
(etc.)
This titular use sets the stage for the topic-announcing, domain specifying use
of concerning, and for its decategorialization to the status of preposition. For
most of the 16th century, concerning retained many of the syntactic properties
of a participle. Mendoza-Denton's careful sifting of the examples follows the
course of change. In the 1517 example (6), concerning is still a verbal
participle, as can be seen in its behavior in gapping:
(6) 1517 ...the one concerning the office of Clerk of the market...&
the other my officers...
We are permitted to say that concerning is here not a preposition because
prepositions do not 'gap' leaving their noun phrase complements intact, e.g., a
sentence like "*Their son is at Harvard and their daughter Smith" would be
ungrammatical. By the turn of the century (ca. 1600) it is coordinated with
other prepositions, as in (7): for and concerning:
(7) 1588-1604 ...for and concerninge all Offences to be committed
after the same Feaste.
A second example is Susan Herring's study of Tamil rhetorical questions
(Herring, 1991). In Tamil, the grammaticalization of complex sentences
begins with a particular pragmatic turn whereby new information is com­
monly introduced through the medium of rhetorical questions. Such questions
were a common device among story tellers to gain the involvement of their
audiences. They passed into general Tamil culture at first as a way of
distributing interesting information over several adjacent utterances. As they
became more frequent, question-and-answer pairs became grammaticalized
as main and subordinate clauses (in 9, the slash / is glossed as Rising
Intonation):

(8) âka inta poNNu enna âyirutu; karppamâ âkutu inta poNNu
thus this girl what become:PF pregnant become this girl
"And so what happens to this girl? She gets pregnant, this girl."
> "And what happens next is, the girl becomes pregnant" (261)
The paradigm at the end of the universe 151

(9) avan ên / inkê Mai nnâ, ovan ûrukku pônân

he why RIS. INT. here NEG SUBD he town:DAT go:PAST
"Why he isn't here is, he went to the village"
(10) avan inkê Mai ênnâ, avan ûrukku pônân
he here NEG CONJ (!) he town:DAT go:PAST
"He isn't here because he went to the village" (273)
Apparently the difference between a rhetorical question and its answer (9),
and a main and subordinate clause (10), is very slight, and rests largely on
intonation. Herring's work is important not only for its application to gram-
maticalization, but for its insistence on a close link between specific cultural
practices and linguistic changes. The connection with Haiman's idea of
'routinization' (Haiman 1994) is especially relevant.
Two further examples show grammaticalization owing its origins to
discourse roles. One of these is the English Present Perfect, as discussed in
papers by Slobin 1993, and Carey 1993. Using examples from child language,
Slobin identifies two discourse functions characteristic of the perfect. One of
them is to mark a right boundary with a view to focusing on the result of an
action, as in (11) and (12):
(11) Child (1.1): Read that one, Mum.
Mother: I've read it.
Child: Read it again, Mum.
(12) Mother: Goodness sake, you've come out in your tights. Af­
ter I've just dressed you - taken ages to get you
ready. (Slobin, 1993:123)
The other is to identify a left boundary as a starting point for negotiation or
new topic, as in (13):
(13) Mother: You draw a letter for me.
Jonathan [draws letter]: I've drawed a letter for you.
(Slobin, 1993, p. 124)
Examples from Old English (Carey, 1993) suggest that such a discourse
origin has a precise counterpart in the diachronic emergence of the have
perfect, which marks transitions, setting things up for the start of a new topic:
152 Paul Hopper

(14) Nu hœbbe we ymb Affrica land gemœro gesœd

"Now we have talked about the African land" (Alfred, Orosius)
(Carey, 1993:115)
Grammaticalization starts as collocations and contextually bound forms
become habitual and hence 'routinized,' released from their restricted con­
text. Sometimes it is possible to be fairly precise about the origins of a
collocation that goes on to have a career as a grammatical construction.
Timberlake and Nichols (1991), in their analysis of the Russian predicative
instrumental (the rule that the NP that is the subject complement in a copular
sentence is in the instrumental case), suggest that it had its origin in a very
specific phrase "went as commander":
(15) V' to zhe lêto ide [AOR] Nezdlylo Pextinych' voevodoju [INST] na
Luky
"In that year Nezdlylo Pextinych' went [AOR] to Luky as com­
mander [INST]" (p. 138)
In this phrase the word "commander" is in apposition to the subject, and
therefore appropriately in the instrumental, and the verb is the aorist "went".
The combination gives rise to a general principle that the instrumental and the
aorist form a textual unit.
This principle is transferred by blending to the copula, restricted to the
predication of a role. That is, "He was the commander" (perfect and nomina­
tive) is confused with "He went as commander" (perfect and instrumental), so
that the instrumental comes to be used whenever the sense of the verb is
perfect or "bounded" in some way, even when the verb is "to be" rather than
"to go". To begin with, the role is a temporary one, and is found in bounded,
i.e., perfect, contexts, as in:
(16) A byl' [PERF] arxiepiskokom' [INST] Feofil' do vzlatia Novagrada
VI lêt
"And Feofil was [PERF] archbishop [INST] six years until Novgorod
was seized" (133)
In unbounded, i.e., imperfect, contexts the nominative is still used:
(17) V lêto 6864 prestivisja Ivan' episkop' Rostov'skii, chto byl' [IMPF]
perezhe arximandrit' [NOM] U svjatogo Spasa na Moskvê
The paradigm at the end of the universe 153

"In the year 6864 Bishop Ivan of Rostov died, he who was [IMPF]
formerly the archimandarite [NOM] at the Church of the Savior in
Moscow" (132)
But it spreads to all copular contexts, including unbounded ones, provided a
role is being predicated of the subject:
(18) A David' be [IMPF] vladykoju [INST] 17 lêt prestavisya fevralja 5, i
polozhisha i v ' pritvorê svjatyja Sofêi podlê Klimenta
"David was [IMPF] the leader [INST] for 17 years and he died on
Feb. 5, and they laid him in the portico of St Sofia alongside
Clement." (133)
The instrumental construction might have spread to all predicate nomi­
nais, and at various times in the history of Russian has threatened to do just
this; but in the contemporary language it has not only remained constrained to
roles, but has even receded in favor of the more numerous nominatives. If
Nichols and Timberlake are correct, the source of the predicate instrumental
in Russian is a very narrowly constricted one, traceable to one specific idiom,
"he went as commander".
These examples all suggest the importance of seeking the sources of a
grammatical construction in a specific discourse collocation, or at least a
context, wherever possible. Often, of course, this source is already lost by the
time of its first textual attestation. We can surmise that the ultimate origins of
the French negator pas was in a verb of motion "didn't walk a step", etc., but
we can never know this for certain. Nonetheless, the principle is important.
We should not lose sight of the fact that context is all important, nor succumb
too readily to the temptation to assume cognitive prototypes and cognitive
leaps in grammaticalization chains.
At the other end of the universe, grammaticalization moves into phonol­
ogy. The stage of 'dissipating' grammaticalization involves what has vari­
ously been called fossilization and 'mummification' (by Gabelentz,
1891:242.) Again, this terminal stage has not been examined sufficiently for
its own interest. For example, in several German words old unstressed pre­
fixes have been absorbed into the stem and it is possible, contrary to the usual
rule of German morphology, to add a second prefix. Thus to the root {nug,
nüg} was added the prefix ge- {genug "enough", genügen "suffice"), to which
now a further unstressed prefix can be added ([sich] vergnügen "enjoy
oneself). Similarly, bleiben (Middle High German belîben) displays a frozen
154 Paul Hopper

prefix be-, which has a second unstressed prefix in verbleiben "remain". That
the prefixes g(e)- and b(e)- in these forms (viz. begnügen, verbleiben) are
now purely phonological can be inferred from the facts that (1) there are no
longer any corresponding unaffixed forms *leib and *nug, and (2) normally a
stem may not take two unstressed prefixes (there is no *verenthalten "?", for
example, although both (sich) verhalten "relate to" and enthalten "contain"
exist. I have argued (Hopper 1991) that in such cases (and there are very
many of them), new phonological bulk is being created by the sedimentation
of moribund morphology, a process I have referred to as 'phonogenesis'.
In our book on grammaticalization (Hopper and Traugott, 1993), Eliza­
beth Traugott and I have stressed the importance of seeing grammaticaliza­
tion as always involving a stage of alternation: A > A ~ B > B . While this
process has been generally accepted, the focus has always been on the
innovating 'B' element, often a periphrastic form, and less attention has been
paid to the recessive A form. It is clear that the decline of the A form involves
a loss of environments, the opposite of the analogical spread that character­
izes the up-and-coming B form. It is less clear what general statements are
possible about the loss of the A form.
How complex and apparently random this process can be is seen in the
case of the Danish (and general Scandinavian) s-passive. Its origins were in a
pre-Norse amalgamation of the reflexive pronoun sik with a verb, from where
it went to the paradigmatization (Lehmann, 1985) of sik at the expense of the
other pronouns, to its grammaticalization as a morphological medio-passive,
and finally its replacement by a periphrastic passive with blive (="become",
werden). One of these is in the infinitive after a modal, as in:
(19) Verden vil bedrages "The world wants to be deceived"
(20) Det kan g0res "That can be done."
(21) Brevet må shrives nu i aften "The letter must be written this very
evening."
Combinations of modal + infinitive in Danish have the peculiarity that
the modal when it precedes an s--passive assumes its "full sense" (Spore
1965:183-187, from which the forms cited in the following discussion are
taken.) Thus vil means "want to" rather than "will [future]" when the depen­
dent infinitive is a morphological passive, and kan means "be able to" rather
than "might" or "could." Compare, with a dependent periphrastic passive
using the auxiliary blive:
The paradigm at the end of the universe 155

(19)' K0bmanden vil ikke blive bedraget "the merchant won't be de­
ceived"
(20)' Det kan blive gjort en anden gang "That can be done another
time."
(21)' Brevet må blive skrevet senere "The letter may be written later."
The collocation of the older, sometimes deontic (cf. må in [21] above), sense
of the modal with the older morphological passive suggests an isolation of the
s-passive as it recedes into fixed lexical patterns, leaving the field free for the
newer periphrastic formation. The retreat of the s-passive leaves behind
several 'patches', or relics, including:
(a) Fixed expressions with der: der behøves "it is necessary, 'il faut"',
der findes "there is/are", der trœnges "it is necessary, 'il faut'"
(b) Reciprocals: m0des "meet", slås "fight", and a few reciprocals
tantum: enes "agree", kappes "rival", kives "quarrel", kævles "quarrel",
mundhugges "exchange insults", omgås "keep company with".
(c) Deponents, that is, verbs that are passive in form but active in
meaning. These include some impersonal expressions: det dages "dawn
breaks", det afines "night is falling", det lykkes mig "I succeed", det mislykkes
mig "I fail", det rygtes "rumor has it", det times mig "it befalls me". There are
also a few personal forms, such as blues "be ashamed", gr0nnes "become
green", lœnges "long for", synes "seem".
Probably the only remaining context for the s-passive with any vitality is
the present 'mediopassive' used generically, as in:
(22) Der spises meget smør i Danmark, "Much butter is eaten in Den­
mark."
Its sense is almost always generic, iterative, or stative; the s-passive is
virtually impossible for single actions. This fact suggests, of course, an
imperfective meaning for the ¿-passive that in turn points to a loss of transitiv­
ity for the suffix. Such a progressive decline in transitivity may be the general
fate of morphology that starts out indexing the arguments of the verb.
In the past tense the ¿-passive has an affinity for weak verbs, where it is
quite productive in forms like de elskedes "they loved one another", but in
strong verbs it has become fossilized in a very small number of set expres­
sions ([der] fandtes "there was/were (found), 'es gab'", [der] gaves "there
was/were, 'es gab'", sås "was seen", sloges "fought". Since the only consis-
156 Paul Hopper

tent difference between strong and weak verbs is a morphological, rather than
a semantic or grammatical, one, this is on the face of it a rather puzzling
restriction. Yet here at the finishing end the particularity of change is seen in
the same way as it is in the origins. Danish has lost its person-number
distinctions in the past tense, and there is thus a tendency to understand the
final -edes as a single, unanalyzed suffix. This suffix is, of course, only
available to weak verbs.
In a general sense what has happened here surely represents a move from
agglutination to synthesis, in that -edes is now a portmanteau form of
past+passive, rather than the passive of the preterite. Yet such a characteriza­
tion of the process risks misrepresenting the nature of the events. The Danish
s-passive survives in several environments, with different degrees of lexical-
ity and freedom of occurrence, none of which has much in common with the
others. But one of these environments, the weak preterite, is paradigmatically
defined. The partial picture derived by focusing attention on the weak para­
digm tends to make it all appear much more predictable and systematic. By
attending exclusively to canonical grammaticalization we naturally identify
the most systematic and regular parts, when a more global picture might
present a more fragmentary and messier situation.
Both incipient and dissipating grammaticalization occur in particular
restricted discourse contexts. Heine et al., 1991:20-21 and 238-243, have
identified two directions in the study of grammaticalization, a lexical-cogni­
tive one and a discourse-textual one. Certainly the lexical-cognitive direction
has proven extremely fruitful. Yet evidently if we are to understand at least
the initial and the terminal phases of grammaticalization, we cannot dispense
with the study of usage from texts. This is hardly a new observation, of
course; it was a cornerstone of Meillet's philological method, and was reiter­
ated by Benveniste, for example in his "Problèmes sémantiques de la
réconstruction" (Benveniste 1966 [1954]: 289-307.) The important achieve­
ments of the typology of semantic change and the recent decade's work in
grammaticalization should not be allowed to overshadow the constant neces­
sity to investigate texts for insights into the emergence of grammatical forms.
Expanding the picture of the linguistic field to include, crucially, contextual
information brings into focus the complex ensemble of processes and rela­
tions involved in language change. It makes available a processual, emergent
and dynamic representation of language change. Such an understanding of
the 'whole' of linguistic activity works not by assuming an integrated 'cogni-
The paradigm at the end of the universe 157

tive' totality in advance, but by piecing together bits of textual evidence from
here and there to build a more integrated picture from below.

REFERENCES

Benveniste, Emile. 1966. Problèmes de linguistique générale. Paris: Editions Gallimard.

Carey, K. 1993. "The grammaticalization of the perfect in Old English: An account based
on pragmatics and metaphor." In W. Pagliuca (ed.), Perspectives on Grammaticaliza­
tion, 103-119. Amsterdam/Philadelphia: John Benjamins.
Gabelentz, George von der. 1891. Die Sprachwissenschaft: ihre Aufgaben, Methoden,
und bisherigen Ergebnisse. Leipzig: Weigel.
Giddens, Anthony. 1984. The Constitution of Society: Outline of a Theory of Structura­
tion. Cambridge: Polity Press.
Görlach, Manfred. 1991. Introduction to Early Modern English. Cambridge: Cambridge
University Press.
Haiman, J. 1994. "Ritualization and the development of language." In W. Pagliuca (ed.),
Perspectives on Grammaticalization, 3-29. Amsterdam/Philadelphia: John Benjamins.
Heine, B., U. Claudi, F. Hiinnemeyer. 1991a. Grammaticalization: A Conceptual Frame­
work. Chicago: University of Chicago Press.
Herring, S. 1991. "The grammaticalization of rhetorical questions in Tamil." In E. Traugott,
and B. Heine (eds), Approaches to Grammaticalization, volume I: 253-285.
Amsterdam/Philadelphia: John Benjamins.
Hopper, P. J. and E. Traugott. 1993. Grammaticalization. Cambridge University Press.
Hopper, P.J. 1987. "Emergent grammar and the a priori grammar postulate." In D. Tannen
(ed.), Linguistics in Context: Connecting Observation and Understanding. Lectures
from the 1985 LSA/TESOL, and NEH Institutes, 117-134. Norwood, NJ: Ablex.
. 1992. "Phonogenesis." In W. Pagliuca (ed.), Perspectives on Grammaticalization,
119-135. Amsterdam/Philadelphia: John Benjamins.
. 1987. "Emergent grammar." In Jon Aske, Natasha Beery, Laura Michaels and
Hana Filip (eds), Papers of the Annual Meeting, Berkeley Linguistics Society 13:139-
157.
Lehmann, Chr. 1985. "Grammaticalization: Synchronic variation and diachronic change."
Lingua e Stile, 20:3, 303-318.
. 1991. "Grammaticalization and related changes in contemporary German." In E.
Traugott and B. Heine (eds), Approaches to Grammaticalization, volume II: 493-535.
Amsterdam/Philadelphia: John Benjamins.
Meillet A. 1912. "L'évolution des formes grammaticales." Scienza (Rivista di Scienza)
12 (1912), No. 26,6; reprinted in Linguistique historique et linguistique générale, 1958,
130-48. Paris: Champion.
Mendoza-Denton, Norma. 1995. 'The grammaticalization of 'concerning'." Lecture at the
Linguistics Institute, Albuquerque, June 1995.
Michael, Ian. 1970. English Grammatical Categories. Cambridge: Cambridge University
158 Paul Hopper

Press.
Slobin, D. 1994. "Talking perfectly: Discourse origins of the present perfect." In W.
Pagliuca (ed.), Perspectives on Grammaticalization, 119-135. Amsterdam/Philadel­
phia: John Benjamins.
Spore, P. 1965. La langue danoise: Phonétique et grammaire contemporaines.
Copenhagen: Akad.
Timberlake, A. and J. Nichols. 1991 "Grammaticalization as recontextualization." In E.
Traugott and B. Heine (eds), Approaches to Grammaticalization, volume I: 129-146.
Amsterdam/Philadelphia: John Benjamins.
 At the boundaries of grammaticalization
What interrogatives are doing
in concessive conditionals*

Torsten Leuschner
Freie Universität Berlin

0. Introduction: Discourse as a boundary of subordination

Among the literature which has in recent years explored the relationship of
subordinate clauses to discourse, studies on conditionals are particularly
numerous. Especially the grammaticalization of many conditionals from such
interactive structures as interrogatives is widely accepted in the literature,
given the great similiarities in the semantics of these two construction types
(Traugott 1985: 294). Nonetheless the diachronic roots in discourse of spe­
cific subordinating constructions, including conditionals, have rarely been
investigated since Haiman (1978); a notable exception is found in Herring
(1991, on the grammaticalization of rhetorical questions as clause-linkers in
Tamil). The present paper attempts to help fill that gap by investigating the
syntacticization of a specific set of subordinating structures: concessive-
conditional adverbial clauses.
In Givón's well-known grammaticalization cycle (Givón 1979: 83, see
also Lehmann 1995 [1982]: 13, and Heine et al. 1991: 239),
discourse > syntax > morphology > morphophonemics > zero,

* After the Grammaticalization Workshop during the 1995 meeting of the Societas Linguistica
Europaea in Leiden, an earlier revised version of this paper was also read as part of the
Linguistik am Montag lecture series at the Freie Universität Berlin, May 1996. I am much
indebted to both audiences and Ekkehard König for helpful discussion.
160 Torsten Leuschner

'syntacticization' occurs near the beginning: it is the diachronic process

leading from loosely joined paratactic structures (Givón's 'pragmatic mode')
to tightly joined subordinating structures (Givón's 'syntactic mode'). Echoing
Matthiessen and Thompson's claim (1988: 301, 286) that there is a "funda­
mental analogy" between subordinate clause structure and text structure,
Hopper and Traugott (1993: 169, 171) place the evolution of clause-combin­
ing structures firmly within the broad range of grammaticalization phenom­
ena. In their words, the rise of complex sentence grammar represents "the
fixing of structure through (...) discourse strategies" (1993: 189): "two sepa­
rate and autonomous nuclei that are mutually relevant" are condensed into a
subordinating structure in which one of the nuclei is de-ranked as satellite to
the other, thus bringing about 'enhancing hypotaxis' (1993:169).
We start, therefore, from the assumption that concessive conditionals
exemplify particularly clearly the syntacticization of a nucleus-satellite rela­
tionship in discourse (König 1992: 432). More precisely, I would like to
explore the hypothesis that the interrogative form of many concessive condi­
tionals is due to their origin in rhetorical dialogues broadly similar to those
described by Herring (1991) for Tamil, in other words that they originate
from a question-answer sequence in which the 'question' becomes grammati-
calized as satellite to the 'answer' or nucleus (see similarly Haiman 1978 on
conditionals). Concessive conditionals are a particularly suitable object for
such an approach because of the transparent nature of their forms and their
great formal diversity, which suggest that they are only partially syntacticized
and that their rate of syntacticization is exceptionally slow (König and van der
Auwera 1988: 128f.). Thus they lend themselves readily to an investigation
not only of the output of the grammaticalization process in the syntactic mode
but of the entire range of transitional variants closer to the pragmatic mode.
After a brief survey of their semantic and formal characteristics (ch.1),
which suggest a basis in interrogative structures, we will therefore look for
traces of dialogical interaction in concessive conditionals from mainly En­
glish (E.) and German (G.), and occasionally from Dutch (D.) and Swedish
(S.). Working from authentic, contextualized text samples1, we will investi­
gate the origin of one subtype of concessive conditionals in Rhetorical Dia­
logues (Popovici 1981, Herring 1991; chs. 2&3), subsequently expanding our
perspective to include one more, formally different subtype (ch.4). The cross-
linguistic perspective not only provides a broader basis for synchronic com­
parison but, following König and van der Auwera (1988), will eventually
Interrogatives in concessive conditionals 161

allow us to make additional diachronic observations in a sketch of the syntac-

ticization of concessive conditionals in the Germanic languages (ch.5).

1. A survey of concessive conditionals

There are three basic types of concessive conditionals (constructed examples,

adapted from Quirk et al. 1985: 1099ff.):
(1) a. Whatever/no matter what I say to them, I can't keep them quiet.
b. (No matter) whether he finds a job or not, he's getting married.
c. Even ify ou dislike ancient monuments, Warwick Castle is worth
a visit.
The fact that the first two subtypes contain different kinds of interrogatives,
whereas the third type is transparently derived from a conditional (if), has
until recently prevented most grammarians from recognizing them as a single
adverbial clause category. In fact, all three examples in (1) represent condi­
tionals whose protases state or imply a series of sufficient conditions, exhaust­
ing the spectrum of possibilities within a given parameter, for the truth of the
apodosis (König 1986: 231f.):
(1) a.' If I say a to them, I can't keep them quiet; if I say b to them, I
can't keep them quiet; if I say c to them, I can't keep them quiet;

b.' If he finds a job, he is getting married; if he doesn't find a job,

he is getting married.
c.' If you dislike ancient monuments, Warwick Castle is worth a
visit; if you are indifferent to ancient monuments, Warwick
Castle is worth a visit; if you like ancient monuments, Warwick
Castle is worth a visit.
The three types differ as to how the set of conditions is specified in the
protasis: by a free-choice expression in (l)a, involving a WH-pronoun and
generalizing particle or expression of irrelevance ('universal' concessive
conditionals); by a disjunction 'p or not-p' or 'p or q' as used in embedded
interrogatives in (l)b, sometimes also involving an expression of irrelevance
('alternative' concessive conditionals)2; by if and a focus particle, asserting a
conditional relationship for an extreme value on a scale of possible values, in
162 Torsten Leuschner

(l)c ('scalar' concessive conditionals). What is in fact asserted in (l)c. is thus

a conditional relationship which holds true contrary to expectation ('if p then
normally not-q'); this expectation could be spelled out such that a person not
interested in monuments is not normally likely to find fascination in Warwick
Castle. The unfavourable condition is implied by the generalizing particle in
the 'universal' variety and exemplified by one of the disjuncts in the 'alterna­
tive' one (Haspelmath and König: 2ff.). On the other hand, since concessive
conditionals relate a whole series of protases to an apodosis, they are distin­
guished from normal conditionals by the lack of conditional perfection ('if p, q
> if not-p, not-q'), in other words they deny the inference that any condition
sufficient for the truth of the apodosis should be a necessary condition as well
(König 1986: 236).
As far as the conditional semantics of the sentences in (1) have been
recognized in the literature, they have led to a wide variety of labels.
Zaefferer (1991: 211), for instance, described them as 'unconditionals' and
the function of the protasis with respect to the apodosis as
'deconditionalizing'. These terms are reminiscent of similar terminology
such as 'irrelevance conditionals' (König and Eisenberg 1984), a term based
on Visser's name 'clause of indifference' for the concessive-conditional
protasis (Visser 1966). All these labels refer to the fact that the meaning of
some universal and alternative concessive conditionals comes about composi-
tionally from the interrogative pronoun and a more or less reduced
superordinate expression stating the irrelevance of the actual protasis value to
the truth of the apodosis, such as es ist ganz gleich WH/ob in German, it
doesn't matter WH/whether or no matter WH/whether in English. In other
varieties of universal concessive conditionals, the irrelevance effect is
achieved in the same compositional manner by generalizing (E. -ever, G.
immer), additive (G. auch, D. ook) or other particles focussing the WH-
pronoun (Haspelmath and König: 7ff.).
None of these labels, however, can fully capture the semantics of the
sentences in (1) — not, in particular, the strong concessive element in the
meaning of ( l ) c , which is in fact shared by the types (l)a and b, suggesting
that 'concessive conditionals' (e.g. König 1986) is the most appropriate label.
All the sentences in (1) are clearly distinguished from normal conditionals by
the fact that, as in concessives, the apodosis is asserted; this implies a similar
scalar ordering among the set of conditions in all three types as specified in the
protasis as in (l)c, such that even the condition least likely to lead to the truth
Interrogatives in concessive conditionals 163

of the apodosis does in fact make it true. This is particularly clear in (l)b,
where being unemployed could be understood as a good reason for postpon­
ing marriage. The set of possible values specified in the protases of conces­
sive conditionals is therefore not completely random but can be said to be
structured along a semantic parameter which is generalized over in universal,
exemplified by two opposite values in alternative, and implied by naming just
one extreme case in scalar concessive conditionals, and indeed the three types
of quantification are often combined in various ways to spell out the precise
parameter referred to, or explicate the scalar ordering (Haspelmath and König:
8f.).
Of particular interest to our inquiry is the connection with interrogatives.
As Haspelmath and König (6f.) point out, many concessive conditionals do
not express a link between events or situations as illustrated by our examples
so far. In line with other types of adverbial clauses, they can encode a link
between pieces of evidence or knowledge, or between speech acts
('epistemic-level' and 'illocutionary-level linkage') as well as 'content-level
linkage' (see Sweetser 1990: 113-144 for conditionals). On the illocutionary
level, pragmatic conditions for uttering the main clause are pushed aside
(whether you like it or not, ...); since we are not going to focus on this kind of
concessive conditionals, however, we will not exemplify them any further
(see Haspelmath and König: 7). As for epistemic linkage, the following
examples (2), taken from Haspelmath and König, illustrate the link between a
conclusion and potential evidence for it; the evidence, expressed in the
protasis, is rejected as irrelevant:
(2) a. Whatever his motive was, it was certainly not altruistic.
b. Whether he actually was at his office or not, he certainly did
not pick up his mail.
c. Even if it had not been his intention, he certainly managed to
alienate most of his colleagues.
The meaning of these sentences is made particularly clear in alternative and
universal concessive conditionals, on which we will focus exclusively, by
superordinate expressions of irrelevance like It doesn't matter... or No mat­
ter..., but also expressions of ignorance (/ don't know...), doubt (G. Ich
bezweifle...) and, of course, of questioning (I wonder...): professing ignorance
or uncertainty as to potential evidence and asserting the conclusion nonethe­
less is after all equivalent to saying that the evidence is irrelevant for the
164 Torsten Leuschner

conclusion. Another way of describing epistemic linkage in concessive con­

ditionals is therefore to say that a question is raised in the protasis which is left
undecided in order to assert the apodosis with more emphasis, and this link
between a "dialogue-type of discourse" (König 1992: 432) and the form and
meaning of concessive conditionals provides extra motivation for an inquiry
into the discourse basis of this clause type.

2. "A dialogue type of discourse": Question and answer in alternative

concessive conditionals

According to the typology of interrogative situations set out by Wunderlich

(1976: 181), one can respond to questions in at least six ways; expressing
indifference as to the answer is one of them. The least syntacticized kind of
concessive conditionals consists of precisely this type of question-answer
sequence. In the following example involving the German conjunctional
adverb jedenfalls ('anyway'), the interior monologue is quoted of a girl about
to cross a road and be hit by a car:

(3) Mit schnellen Schritten ging sie weiter und riß dann vor Angst die
Augen weit auf. Hatte der Fahrer sie nicht gesehen? Jedenfalls
verminderte er sein Tempo nicht. Der rechte Kotflügel erfaßte
Barbaras Bein und schleuderte sie mit voller Wucht in den Rinn­
stein.
('She kept walking at a fast pace and then opened her eyes wide
with fear. Hadn't the driver seen her? In any case he did not reduce
his speed. The right fender caught Barbara's leg and threw her into
the curb with full force.')
(MK2/TSL: S. Stephan, Ihre Liebe gab ihr Leben, light fiction.
Bergisch-Gladbach, n.d., p. 50)
Rather than by any means indicating grammatical subordination, the conces­
sive-conditional relationship between the question and the subsequent assertion
is expressed solely by means of the conjunctional adverb. Since concessive-
conditional conjunctional adverbs are a particularly frequent feature of conces­
sive conditionals in actual discourse3, it is helpful to recall the analysis by
Schelling (1982) of de toute façon, a close French equivalent of such conjunc-
Interrogatives in concessive conditionals 165

tional adverbs as in jedem Fall and in any case. The function of de toute façon
is described by Schelling as a resolution of conflicting 'argumentative orienta­
tions' ('visées argumentatives') implicit in the preceding discourse: by using de
toute façon, a speaker indicates that no choice between the arguments is to be
made (Schelling 1982: 68, 96ff.). In the structure of an argument, de toute façon
therefore operates a kind of 'closure' ('clôture', Schelling 1982: 66): rather
than align herself with one or all of the arguments, the speaker declares all their
conflicting argumentative orientations invalid with respect to the point she
proceeds to make. In (3), the 'arguments' are the answers implicit in the
question: if the driver hasn't seen the girl, we are less surprised that she is hit
than if he has seen her and still does not avoid her. Y'et jedenfalls 'closes' this
question undecided so that the precise answer, including the most plausible one,
appears as irrelevant for the subsequent assertion.
Exactly the same structure is found when the question is phrased as an
embedded interrogative. In this example, the 'arguments' are explicitly raised
in an ob - oder ('whether - or') clause:
(4) "Ziehen Sie das an, Hoheit, sonst klappern Sie mit den Zähnen. "
Er fragte sich, ob sie ein bißchen verrückt war oder nur ein
Abenteuer suchte. In jedem Fall war es merkwürdig, daß sie
ausgerechnet hierher auf diese Klippe gekommen war, denn das
war sein Refugium, das war streng abgegrenzt, hier hatte niemand
Zutritt.
("Put this on, Your Highness, or your teeth will start chattering."
He wondered whether she was a little crazy or simply looking for
some adventure. In any case it was strange for her to have come out
on this cliff, because this was his refuge, strictly demarcated, no-
one had access here.')
(MK2/TLP: V. Larsen, Die heimlichen Wege der schönen Prinzes­
sin, light fiction. Bergisch-Gladbach, n.d., p. 9)
The embedded question clearly poses two alternatives, but the precise expla­
nation for the girl's coming is left open. On the background of the available
alternatives, the subsequent sentence is instead asserted using another con­
junctional adverb, in jedem Fall ('in any (lit. every) case'); we therefore have
a topic-comment structure precisely as in normal conditionals (Haiman
1978). In other words, a subordinating or satellite-nucleus relationship is
166 Torsten Leuschner

construed on the purely textual level (Schelling 1982: 102) while the indi­
vidual clauses retain full grammatical independence.
The potential grammatical effect of textual subordination is easily seen
in some so-called 'non-canonical' concessive conditionals, which show in­
version of the verb rather than an overt subordinator as in straightforward
questions (cf. Haiman 1978 on conditionals, with reference to Jespersen
1940). Therefore the following example of a non-canonical alternative con­
cessive conditional, taken from a summary of European history, consists
simply of a question as protasis and an assertion, again with conjunctional
adverb, as apodosis:
(5) War dieser Schluß nun optimistisch oder pessimistisch, auf jeden
Fall hatte er, verbunden mit einem fast mystischen Glauben an
Rußlands spezifisch slawische Art, weitreichende Konsequenzen,
die alle Slawen betrafen.
('Whether this conclusion was optimistic or pessimistic, it had —
being associated with an almost mystical belief in Russia's specifi­
cally Slavic character — far-reaching consequences which affected
all Slavs.')
(Oskar Halecki: Europa. Grenzen und Gliederung seiner Ge­
schichte. Darmstadt: Wiss. Buchgesellschaft 1957, p.79)
Question and answer have been united in a single complex sentence by the
omission of the question-mark, which has been replaced by a comma indicat­
ing a more unified intonation contour (Givón 1979: 98). Precisely the same
loss of independence is found in the embedded interrogative in the next
example, where protasis and apodosis are now separated by a comma:
(6) No other manufacturer makes such a wide variety of shirts as the
CWS. Society shirts are made to please all types of wearer, from the
artisan to the executive, and for all occasions. It doesn't matter
either whether you are a giant or a dwarf, your Co-operative
society can fit you out with a CWS shirt.
(LOB/H30: Home Magazine, October, 1961)
From here, loss of grammatical independence proceeds through erosion of the
expression of irrelevance. (It is) no matter, e.g., may be reduced to the
adverbial phrase, no matter, modifying the interrogative pronoun:
Interrogatives in concessive conditionals 167

(7) Occasionally I notice my patient has a quiver round the mouth or

lip. This indicates nervous strain. No matter how you try to control
yourself, the muscles round the mouth are first to give way.
(LOB/F33: People's Friend magazine, August 26, 1961)
Note the increase in positional variability; like any normal conditional, the
new concessive conditional may follow the main clause:
(8) (M)any of the cases removed from the water whilst still alive are
doomed to die within a few minutes from the devastating changes
which have already taken place, no matter whether the water was
fresh or salt. This knowledge explains the very high mortality rate
in drowning. There are, however, a small number of cases which
are rescued from water before large amounts of water have appar­
ently been inhaled,
(LOB/J16: from Medico-Legal Journal vol. 29, 1961)
Finally, in alternative questions the superordinate expression can disappear
altogether; the result is an alternative concessive-conditional clause with
'whether - or' as conjunction:
(9) Ich habe mich selbst darüber gewundert, welche Kraft der Verstel­
lung ein Mädchen überkommt, wenn es — ja, soll ich wirklich
sagen: wenn es liebt? Nun, ob man meinen Zustand nun Liebe
nennen oder ihn mit irgendeinem neumodischeren Ausdruck be­
zeichnen will, so mit irgendetwas aus der Psychologie, in jedem
Falle hatte Jerome sich alle meine Gedanken Untertan gemacht.
('Even I marvelled at the powers of make-believe which sieze a girl
when she — ought I really to say: when she loves? Nun, whether
my condition be called love or labelled something more fashion­
able, something or other from psychology, in any case Jerome had
subjected all my thoughts to him.')
(MK1/LBT: Werner Bergengruen, Das Tempelchen, novella.
Zürich: Arche 1950, p. 27)
The question-answer structure is particularly evident in this example due to
the discourse particle nun, which in English could well be rendered as anyway
(see König 1986: 242, on anyway).
168 Torsten Leuschner

The effect that losing the superordinate expression of irrelevance has on

the meaning of the former interrogative clause is particularly well illustrated in
the following example. Observe what happens when ob - oder ('whether - or')
is replaced with wenn - oder ('if - or'):
(10) Es ist wie eine große, nie zu fixierende Gemeinschaft der Funken,
selten in gleichbleibender Höhe. (...) Von jedem wird verlangt, daß
er reflektierend in sich selbst die Schuld sucht in beiden Fällen:
ob/wenn er ein Sinken zu sehen meint oder selbst in dieses Sinken
gerät Es gibt keinen Richter.
('It [i.e. the total of human individuals] is a great, never-to-be-fixed
community of sparks, rarely flying at a uniform height. (...) Each is
expected to search within himself for the fault in either case:
whether/if he believes to see another spark sink or finds himself
sinking. There is no judge.')

(adapted from MK1/WJA: Karl Jaspers, Die Atombombe und die

Zukunft des Menschen, non-fiction. München: Piper 1958, p. 306)
Clearly, ob - oder has acquired a similar conditional meaning to wenn - oder,
yet only ob - oder is concessive-conditional. Although the conjunctional
adverb in beiden Fällen ('in both cases') makes clear that no more than two
alternative conditions will be referred to, only ob - oder implies that the
alternative conditions named exhaust the semantic parameter implied. Once it
has fully shed any superordinate expression of irrelevance, 'whether - or' most
truly functions as an adverbial-clause conjunction in its own right.

3. Rhetorical dialogues: Concessive conditionals as polyphonic

discourse

There is a striking parallel between question-answer sequences suggesting

concessive-conditional sentence relationships in written, monologic dis­
course, and the use and grammaticalization of rhetorical questions. In Tamil
narrative discourse (Herring 1991), rhetorical questions are used to engage
the audience without jeopardizing the narrator's hold of the floor. By posing
rhetorical questions which he immediately answers for himself, the traditional
Tamil narrator interacts with an ideal hypothetical listener in order to move the
storyline forward, as in 'If a schoolboy fails an exam, will he come straight
Interrogatives in concessive conditionals 169

home to tell about it? He won't.' or 'And so what happens to the girl? She
becomes pregnant, the girl.' (Herring 1991: 261ff.). Indeed there are real-
dialogue types of narrative in Tamil in which a secondary narrator is present
who interacts with the primary narrator to pose or answer questions for him
(Herring 1991).
Like rhetorical questions in Tamil, concessive conditionals represent
what Popovici (1981) called 'Rhetorical Dialogues', i.e. typical examples of
what in the tradition influenced by Bakhtin and Oswald Ducrot is known as
'polyphonic' discourse, in which a speaker integrates another's contribution
into her own (see Roulet et al. 1985: 9ff.). This may imply a pretense of
cooperative behaviour: "not only are Rhetorical Dialogues feigned dialogues,
they are a manner of cutting off dialogue altogether" (Popovici 1981: 14,
Herring's translation). By setting up an apparent dialogue around an
interlocutor's potential intervention, a speaker may foresee and avoid an
actual exchange, indeed hide behind another's 'voice' in order to make all the
more effectively her own rhetorical move (Popovici 1981, also Herring 1991:
260). A disguised hierarchy is thus set up within the discourse: "the apparent
cooperation is in reality a subordination [of the hearer's 'voice' to the
speaker's] on the pragmatic level by verbal means" (Popovici 1981: 14, my
translation). Concessive conditionals can therefore be explained as the syn-
tacticization of subordinating relationships implicit in Rhetorical Dialogues,
which in turn are modelled on a type of actual interrogative dialogue.
Strong evidence for the 'Rhetorical Dialogue' analysis comes from
irrelevance expressions like it doesn't matter and I don't know, in which less
grammaticalized varieties of concessive conditionals are embedded as inter­
rogatives with or without an answer. Here is a classic example of a parenthesis
in which a question is raised, embedded in an expression of ignorance:
(11) Construed as an internal impression which is thought to function
as a cause that issues in some item of so-called overt behaviour
(whether this be some bodily movement or an action is of no
matter for our present purposes), the impression must be describ-
able without reference to any event or object distinct from it.
(LOB/J54: A.I. Melden, Free Action, non-fiction. London:
Routledge 1961)
It is not necessarily obvious why new, seemingly uncalled-for infomation
should be inserted into the discourse only to be immediately declared irrel-
170 Torsten Leuschner

evant: wouldn't the maxim of relevance prevent the speaker from mentioning
it in the first place? Worse: why would a speaker want to risk a severe loss of
face by raising a question, just to admit that she does not know the answer?
The 'feigned dialogue' hypothesis can help solve these riddles. Irrel­
evance expressions, and concessive conditionals in particular, are a means of
acknowledging and at the same time disarming potential hearer objections
which the speaker foresees (or believes herself able to foresee) due to a shared
knowledge of the world, such as typical relationships between states of affairs
and conclusions typically drawn from evidence. The opportunity to make a
point safe against potential objections and disruptions is rated higher by
speakers and writers than the loss of face incurred when ignorance is con­
ceded. In other words: losing the point one is making or, in conversation,
losing hold of the floor due to a foreseeable objection, is considered a greater
risk than a frank admission of ignorance. An interlocutor pursuing an objec­
tion would be more compromised than a speaker who has already conceded
ignorance.
Further evidence that concessive conditionals originate from a polyphonic,
'feigned dialogue' type of discourse comes from typology: concessive-condi­
tional expressions across many languages clearly originated interactively in
addresses to the hearer. In their typological study of concessive conditionals in
the European languages, Haspelmath and König (32) distinguish between
speaker-oriented and hearer-oriented strategies of expressing free choice and
arbitrariness: whereas the former are based on a statement of irrelevance such
as (it is) no matter, es ist gleichgültig (lit. 'it is equally valid' ), I don't care and
so on, hearer-oriented strategies express irrelevance by leaving the choice of
protasis value to the hearer. Spanish is one of many languages in which an
address to the hearer like '(you) want' is the source of a fairly grammaticalized
concessive-conditional marker equivalent to '-ever' (Haspelmath and König:
40):

(12) Donde-quiera que vaya, nunca la dejará.

('Wherever she goes (lit. where you want her to go), he will never
leave her.')
This kind of strategy is found in the Germanic languages, too, where similar
expressions complement the large number of speaker-based ones. A particu­
larly suitable strategy to mark hearer-involvement is imperatives (constructed
examples from Haspelmath and König 16):
Interrogatives in concessive conditionals 111

(13) a. Say what you will, we cannot afford it.

b. Let him be ever so bad, he has some good points.
One such imperative is English you name it, whose Dutch equivalent noem
maar op (~ 'make your list!') appears in the following example from a water-
sports magazine's special report on the leisure fair, HISWA:
(14) Zo'n HISWA-nummer als dit brengt als altijd een grote hoeveel­
heid nieuws. Nieuwe boten, nieuwe apparatuur, nieuwe motoren,
noem maar op.
('As always, a special HISWA issue like this one contains a lot of
news. New boats, new equipment, new engines, you name it.')
(INL: Waterkampioen magazine, No.3, 1993)
A similar challenge to the hearer appears to be at the origin of Swedish WH
som helst ('-ever', lit. 'as favourite'):
(15) En sak hade hon i alla fall begripit och det var att han hade kängor
av en kvalitet hon aldrig sett tidigare sä hon sa: Du kan gå hur
längt som helst.
('One thing at any rate she had understood and that was that he had
boots of a quality she had never seen before, when he said: you can
walk however far you like/no matter how far/any distance.')
(SUC: kk37)
Such expressions request a challenge from the hearer to the truth of the
speaker's assertion, thus giving the speaker a particularly strong claim to truth.
An intriguing example of a hearer-oriented free-choice strategy occurred
in older English, which until the Early Modern English period had indefinite
pronouns based on a WH-pronoun embedded in the imperative look! (ex­
amples from Visser 1966: 911f.):
(16) a. Loca hwa ut gange, licge he of slagen. (Ælfric)
b. Looke who that is moost pacient in love/ He is at his advantage
al above. (Chaucer)
c. The loue of God is honourable wisdom; loke unto whom it
appeareth, they loue it. (Coverdale)
172 Torsten Leuschner

According to Horn (1923: 64f.), this structure, also known in Old High
German (sih-wer 'anyone' < sehan 'look'), originated in question-answer
sequences such as 'Look! Who comes? He is welcome!'. Along with choose-
WH pronouns found in some English dialects (Horn 1923), look-WH is a
prime if rare example of a concessive-conditional expression gone all the way
from dialogue to subordinating conjunction similar to rhetorical questions in
Tamil (Herring 1991). It also reminds us that syntacticization is by no means
a new discovery of late 20th-century linguistics.

4. Universal concessive conditionals: Introducing the common

integrator

After reconstructing the likely discourse origin of alternative concessive

conditionals and other individual structures within this paradigm, we can now
expand our perspective towards universal concessive conditionals: after all,
the hypothesis might seem likely that expressions involving generalizing
particles like E. -ever, G. auch/immer/auch immer, D. (dan) ook, S. som helst,
are also somehow related to embedded interrogatives. To begin with, the
striking similarities between the three types of concessive conditionals and
parametric ('constituent'), alternative and polar ('yes-no') embedded ques­
tions can be summarized as follows, using the respective structures of English
(adapted from Haspelmath and König 48):

Table 1. Formal Correspondances between Concessive Conditionals and Embedded

Questions

parametric: alternative: polar:

concessive wherever she goes whether she comes even if she comes
conditionals: or goes
interrogatives: where(*ever) she whether she comes if/whether she comes
goes or goes

The formal similarity is closest, amounting to complete identity, in the 'alter­

native' varieties: indeed, as we have argued, alternative concessive condition­
als are simply alternative embedded questions with the superordinate clause
left out. The relationship is considerably more obscure than that in the polar
varieties: scalar concessive conditionals in English are obviously based di-
Interrogatives in concessive conditionals 173

rectly on conditionals, which often provide markers for embedded polar

questions. In German, on the other hand, whether no longer has a cognate and
is replaced with the former conditional marker, ob, in polar and alternative
embedded questions and alternative concessive conditionals, whereas scalar
concessive conditionals are based on the modern (temporal-)conditional
marker, wenn.
The most interesting disturbance in the parallelism between concessive
conditionals and questions is due to the particle or particle combinations
(corresponding to -ever) in universal concessive conditionals. Some of these
particles, including E. -(so)ever, are very old, going back to the common West
Germanic pattern swa WH swa or so WH so to which they were added at
different points in history (Poutsma 1928/9: II.1214f. on English; Behaghel
1923: 369 on German; Stoett 1977: 33 on Dutch). However, some (as yet
limited) historical data suggest that at least some universal concessive condi­
tionals in Germanic languages other than English remained particle-less until
quite recently, looking just like former embedded interrogatives. The follow­
ing example comes from a treatise, first published in 1872, in which the
philosopher Nietzsche celebrates Ancient Greek life as reflected in the images
of the gods:
(17) Und so mag der Beschauer recht betroffen vor diesem phantasti­
schen Überschwang des Lebens stehn, um sich zu fragen, mit
welchem Zaubertrank im Leibe diese übermütigen Menschen das
Dasein genossen haben mögen, daß, wohin sie sehen, Helena, das
'in süßer Sinnlichkeit schwebende' Idealbild ihrer eigenen Exi­
stenz, ihnen entgegenlacht.
('And thus the beholder may stand quite perplexed before this
phantastic superabundance of life and wonder what magic potion
must have helped these high-spirited people to so relish their exist­
ence that, wherever they look, Helena smiles towards them, the
ideal image of their own being, 'floating in sweet sensuousnessV)
(Friedrich Nietzsche, Die Geburt der Tragödie oder Griechentum
und Pessimismus. In: Werke in drei Bänden, Vol. 1. München and
Wien: Hanser 1973, p.29)
In present-day German, wohin sie sehen would have to be rendered as wohin
sie auch (immer) sehen or wo immer sie (auch) hinsehen, thus disrupting the
174 Torsten Leuschner

precise formal identity between parametric interrogatives and some universal

concessive conditionals that still obtained in Nietzsche's German.
On this hypothesis, at least some universal concessive conditionals
would have come from embedded parametric interrogatives via the reduction
and eventual omission (as in alternative concessive conditionals) of
superordinate clauses of indifference, gradually acquiring generalizing par­
ticles. In this case, of course, given the slow evolution of concessive condi­
tionals, we would expect to find plenty of intermediate forms in the data
combining a reduced clause of indifference and a WH-'ever' pronoun. Yet
according to intuition, expressions like *no matter wherever she goes are
impossible, and this impression is confirmed empirically: in a database of six
and a half thousand entries from four Germanic languages, there is just one
instance (in MK2/TPN) of any expression of irrelevance combining with a
form of universal-concessive-conditional conjunction: G. egal wer Sie auch
sind, 'no matter who you are', involving auch 'also' rather than *egal wer
immer ('whoever') Sie sind. Therefore, even if former particle-less universal
concessive conditionals can, on the model of alternative concessive condi­
tionals, be explained as embedded interrogatives with the expression of
irrelevance left out, generalizing particles are in fact more likely to have come
from another source.
The solution proposed by Haspelmath and König (49f.) is that WH-'ever'
pronouns may have come, not from embedded interrogatives, but from non­
specific free relative clauses like whoever comes will be welcome, which after
all are easily obtained from concessive conditionals by inserting a suitable
pronoun in the main clause {whoever comes, he will be welcome). This
proposal is strongly confirmed by Lötscher (1993): relative clauses were often
used as conditionals in Middle High German, where syntactic rules were more
readily adapted to pragmatic processes such as topicalization, leading to
extraposition in precisely the manner described (Lötscher 1993: 177). It seems
that when this practice ceased with normal conditionals in Early Modern High
German, non-specific free relative clauses continued to serve as universal
concessive conditionals in the absence of other readily available markers
(such as wenn for normal conditionals). In older stages of English and Dutch,
too, universal concessive conditionals are often difficult to tell apart from
relative clauses for the same reason (see examples in Visser 1966: 912 and
Poutsma 1928/9: II.1214f. for English, and Stoett 1977: 40 for Dutch).
Interrogatives in concessive conditionals 175

The possibility of expressing concessive conditionality by means of a

relative clause is due not only to the fact that both clause types are semanti-
cally non-specific. In present-day usage, too, their discourse functions are
sometimes identical, and this must have encouraged the spread of generaliz­
ing particles to particle-less structures as in (17). An even more recent
instance of the particle-less pattern comes from the Dutch historian Johan
Huizinga. In a work first published in 1919, Huizinga discusses possible
cross-influences between various representations of the Danse Macabre in
fifteenth-century art, closing the issue as follows:

(18) Hoe het zij, de Dodendans van het kerkhof der Innocents (...) is de
meest populaire verbeelding van de dood geweest, die de Middel­
eeuwen hebben gekend.
('Be this as it may, the Danse Macabre in the churchyard des
Innocents [in Paris] (...) was the most popular representation of
Death in the entire Middle Ages.')
(Johan Huizinga, Herfstij der Middeleeuwen. Groningen: Wolters-
Noordh. 1986, p. 142)
In examples like this, the concessive conditional resembles the conversational
'uptake' by some speakers of the interlocutor's previous utterance (de Castro
Campos 1985: 108f. on conditionals in children's dialogues): rather than the
question to which the nucleus provides the answer, the concessive conditional
is part of the answer itself, 'arresting' (see Poutsma 1928/9: I.593ff.) an
undecided issue and dispatching it as it were out of the way of the subsequent
assertion. Here is an present-day English example with -ever, illustrating
precisely the same function:
(19) The expert will get splendid results from a cheap box camera;
others will get poor results from an expensive model. The greater
the amount paid for a camera, and the more gadgets it has is no
sure way of guaranteeing good results. But whatever model you
have, study it carefully and know thoroughly how to work it and
what its capabilities and limitations are. Once you are master of
your camera, you have gone a long way to good pictures.
(LOB/E24: Educational Development magazine, February, 1961)
That a concessive conditional should thus provide a smooth transition be-
176 Torsten Leuschner

tween two sequences of discourse is hardly surprising; indeed example (9),

adapted below as (20), reveals that given a certain degree of syntacticization,
this is a potential of all concessive conditionals:
(20) [1:] Ich habe mich selbst darüber gewundert, welche Kraft der
Verstellung ein Mädchen überkommt, wenn es —ja, soll ich wirk­
lich sagen: wenn es liebt? [2:] Nun, ob man meinen Zustand nun
Liebe nennen oder ihn mit irgendeinem neumodischeren Aus­
druck bezeichnen will, so mit irgendetwas aus der Psychologie,
[3:] in jedem Falle hatte Jerome sich alle meine Gedanken Untertan
gemacht.
('Even I marvelled at the powers of make-believe which sieze a girl
when she — ought I really to say: when she loves? Nun, whether
my condition be called love or labelled something more fashion­
able, something or other from psychology, in any case Jerome had
subjected all my thoughts to him.')
(MK1/LBT: Werner Bergengruen, Das Tempelchen, novella.
Zürich: Arche 1950, p. 27)
The uncertainty what to call her state of mind leads on to an alternative
concessive conditional explicating two possible answers to the question which
in turn introduces an assertion not dependent on the precise label. This
structure can be spelled out in three steps: old topic [1]; closing the old topic
with '-ever' [2]; new topic [3]. Yet the 'arresting' function [2] common to ob
- oder in (20) and whatever in (19) is fulfilled in entirely different ways by
these two structures: ob - oder makes explicit two possible answers to the
preceding question, whereas ever avoids precisely such explicitness. Instead,
whatever in (19) summarizes the preceding discussion under the heading
'models (of cameras)': the previous two sentences are to be understood as
instances and elaborations of this parameter, over which -ever generalizes. At
the same time, but makes clear that we are at a turning-point in the text: the
discussion of different kinds of cameras is to be put aside in favour of a return
to the original subject of the text ('how to take good pictures') to which the
precise model of camera is in fact irrelevant.
It seems therefore that the particular contribution of WH-'ever' pronouns,
leading to their intrusion into universal concessive conditionals as in (17) and
(18), is due to what Lang (1984[1977]) describes as the Common Integrator
Interrogatives in concessive conditionals 177

(CI). As a classic example of a CI, Lang (276) quotes the second sentence of
the Communist Manifesto:
(21) All the powers of old Europe have entered into a holy alliance to
exorcise this spectre: Pope and Czar, Metternich and Guizot,
French radicals and German police spies.
CIs are present at two levels in this example. At the higher level, the CI is
explicated as 'all the powers of old Europe which have entered into the holy
alliance against the spectre of communism'. Rather than simply enumerated,
the powers are then exemplified in appropriate groupings, each of which is
subject to an implicit CI according to particular aspects of power: Pope and
Czar stand for 'ecclesiastical and secular feudal power structures', Metternich
and Guizot for 'aristocratic and bourgeois governments', French radicals and
German police-spies for 'representatives of these forces' (ibd.: 277). While
each pair of conjuncts, and each conjunct within each pair, is carefully chosen
so as to contrast with its neighbour, the implicit evaluation of all these powers
as 'reactionary' is common to all the conjuncts in this effectively balanced
phrase. The CI is thus the result of an interpretative act finding in the conjuncts
a shared meaning parameter (Common Share) of which they are to be under­
stood as exemplifications.
The following example, in which an instructor talks about the use of
various 'aids' in controlling a pedigree dog during an exhibition, shows how a
WH-'ever' pronoun can be used to set up a CI:

(22) "Your aids are your attitude, which comes through your voice,
your hands and legs — voice to encourage, discourage or whatever
the need may be; hands to guide or restrain; legs to produce
motion and rate of speed. Without right attitude the other aids just
do not work right".
(BROWN/E05: Dog World magazine, April, 1961)
The contribution of whatever the need may be can be described on two levels:
on the one hand, the phrase states the relevant parameter, 'verbal behaviour
necessary for controlling a dog', making clear that the teacher's intention in
starting the list was to exemplify this parameter; on the other hand, the specific
contribution of whatever is to cut short the list by generalizing over it. Since
the instructor has named two extremes ('en-courage' and 'dis-courage'), the
generalizing force of whatever can be said to cover at least everything in
178 Torsten Leuschner

between, i.e. as filling up the list without going into any more specifics.
The extent to which CIs can be subject to the dynamics of the text is
borne out by the following example of a universal concessive conditional.
Stemming from the 1960's, it reports on the 360 affiliated radio stations then
serving the African-American communities in the U.S., "whose signals reach
more than half the total U.S. Negro population":
(23) [1:] One question which inevitably crops up is whether such sta­
tions have a future in a nation where the Negro is moving into a
fully integrated status. [2:] Whatever the long-range impact of
integration, [3:] the owners of Negro-appeal radio stations these
days know they have an audience and that it is loyal. [3':] Advertis­
ers have discovered the tendency of Negroes to shop for brand
names, (...).

(BROWN/C12: Christian Science Monitor, October 17, 1961)

At first, the explicit question [1] is raised whether radio-stations aiming at an
all-Black audience have a future in a society putatively approaching racial
equality. A CI under which the question could be subsumed ('the long-range
impact of integration') is offered in the concessive conditional [2], appropri­
ately chosen to contrast with the nucleus [3], which turns from the future to the
present. Full thematic transition is finally achieved in the follow-up sentence
[3'], which picks up on the topic of loyalty but drops the discussion of radio
and the future.
We are now in a position to compare universal concessive conditionals
with relative clauses. The following example suggests potential contexts of
transition:
(24) From the town surveyor, Hans learned drawing and mathematics
and, from a university student, some academic subjects. The mayor
of the town taught them English and French. Whatever Hans or
Anders learned separately they passed on to each other; they read
every book that they could borrow in the village. At 12, Hans was
sufficiently mature to help his father in the apothecary shop, which
helped stimulate his interest in medicine.
(BROWN/E26: Electrical Engineering magazine, June, 1961)
Another diachronic consequence of the topic-shifting use of concessive con-
Interrogatives in concessive conditionals 179

ditionals is the emergence of concessive clause-linkers. Similar to however,

with which it is synonymous, the concessive conjunctional adverb howbeit,
now archaic, is clearly derived from a universal concessive conditional.
Poutsma (1928/9: I.594, §7) mentions the following biblical example of a
discourse function which he calls "arrestive adversative coordination":
(25) He prayed him that he might be with him. Howbeit Jesus suffered
him not. (St. Mark)
Both the original how be it and the more recent Dutch hoe het zij in (18)
strikingly resemble G. wie dem auch sei, another universal concessive condi­
tional with a subjunctive form of 'be'. Along with such concessive-condi­
tional pronouns as G. wie auch immer and D. hoe dan ook (often spelled
hoedanook), wie dem auch sei has become fixed as a topic-shifting expres­
sion:
(26) "In der heutigen Zeit ist es zumindest ebenso wahrscheinlich, daß
derartige Dinge sich als kostenlose Reklame erweisen. " — "Na?"
sagte der Arzt zweifelnd, "wir haben Kurgäste aus aller Welt und
recht seriöse —. Wie dem auch sei, ich muß mich jetzt der selbst­
verordneten Bewegungstherapie in Form eines einstündigen Spa­
zierganges unterziehen und lasse euch zwei solange allein. " —
"Ja, tu das, Paps! " fügte Katja mit Schalk in der Stimme zu.

('"It's at least as likely nowadays that such things turn into a free-
of-charge advertising campaign." — "NaT said the doctor doubt­
fully, "we have patients from around the world and quite
respectable ones, too. — Be this as it may, I must now take my self-
prescribed kinetotherapy and go on a sixty-minute walk. In the
meantime I'll leave the two of you alone." — "Oh yes Daddy,
please do!" replied Katja in a roguish tone of voice.')
(MK1/TPM: PINKWART, Mord ist schlecht für hohen Blutdruck,
detective novel. München: Goldmann 1963, p. 180)
This dialogue, in which a health-spa doctor uses wie dem auch sei to avert a
potential disagreement with his daughter's boyfriend, reveals an interesting
pragmatic property of concessive conditionals: thanks to the non-factual
nature of the protasis, they can be used to lay by a dispute peacefully without
either of the parties having to agree on what is a fact. This is clearly seen in a
180 Torsten Leuschner

comparison with concessives, a sentence type where both clauses are factual.
In both concessive conditionals and concessives, the validity of a general rule
('if p, then normally q') is at stake. From an interactive point of view,
concessives can be regarded as Rhetorical Dialogues in which the speaker
acknowledges the truth of the interlocutor's potential objection; the impact of
her own assertion is thus softened by appearing as an exception to the rule,
whose general validity is not disputed (Klein 1980: 160f.). In concessive
conditionals, on the other hand, the speaker does not concede the truth of the
hearer's potential objection, which is instead declared irrelevant. Thus an
important element of cooperative interaction in concessives, the agreement
element, is missing from concessive conditionals. The situations where con­
cessive conditionals can become particularly useful are therefore those where
there is no basis for agreement, as in (26): the doctor is aware of the potential
for conflict induced by his disagreement and, leaving open the outcome,
changes the subject.
The origin of concessive adverbs like howbeit lies in 'arrestive' uses like
(26). All that was required for a concessive interpretation was a contextual
understanding in some situations that the question which it 'arrested' was
already answered; from there the phrase could be applied to explicit facts,
yielding a typical concessive construction (see König 1986: 239ff.). This
development can of course be anticipated for German, too; in no Germanic
language, however, have such expressions as yet acquired a conventional
concessive meaning or reached remotely the same degree of grammaticaliza-
tion as in English.

5. From more to less pragmatic: Concessives conditionals and

syntacticization

Comparing the pragmatic and syntactic modes, Givón sets up a list of features
for each mode which with regard to concessive conditionals can be repre­
sented as in Table 2 below (adapted from Givón 1979: 223, cf. Herring 1991:
277).
The development of embedded interrogatives into concessive conditionals
turns out to be a typical process of syntacticization on the view of Givón (1979).
In Givón's terms, it shows the "balance of gain and loss" with respect to
'communicative fidelity' and 'economy'. On the one hand, the original se-
Interrogatives in concessive conditionals 181

quence of independent sentences, spells out precisely the irrelevance relation­

ship between the satellite and its future nucleus, thus ensuring a "one-to-one
correlation between code and message". On the other hand, in the syntacticized
concessive-conditional, this high degree of "resolution" in the message has
decreased in favour of a considerable gain in processing speed (Givón 1979:
108).

Table 2. Comparative Features of the Pragmatic and Syntactic Modes

Pragmatic Mode: Syntactic Mode:

loose parataxis tight subordination
satellite-nucleus structure subordinate clause - main clause structure
no use of grammatical morphology elaborate use of grammatical morphology
slow rate of delivery under several fast rate of delivery under a single
intonation contours intonation contour

Leaving aside intonation4, cross-linguistic comparison reveals three argu­

ments that a process of syntacticization is indeed under way in concessive
conditionals among the Germanic language family:
(a) embedded interrogatives are (in a loose sense, see Herring 1991: 279)
reanalyzed as concessive-conditional adverbial clauses, leading from the
paratactic conjunction of sentences to grammatical subordination within
the concessive-conditional clause complex;
(b) conjunctions evolve, providing sentence-type specific subordinating
morphology which takes its place in the larger paradigm of adverbial
clause markers;
(c) the concessive-conditional satellite gradually comes to fill the forefield of
the verb in the nucleus, i.e. V/2 is increasingly observed for the clause
complex as a whole rather than the nucleus alone (to be discussed be­
low).
Argument (a) summarizes our discussion in section 2 as to why the West
Germanic languages at any rate use the same conjunctions in alternative
concessive-conditional clauses as in embedded interrogatives (whether, ob,
of): a whether-or interrogative embedded in a grammatically independent
expression of irrelevance is reanalysed as an alternative concessive-condi­
tional satellite clause to the independent nucleus following it. This involves a
182 Torsten Leuschner

simultaneous reanalysis of the irrelevance expression, so far a nucleus in its

own right, as an adverbial phrase within the satellite. Whereas the structure
was [nucleus-satellite]-[nucleus], it is now [satellite-nucleus]. As a result of
the syntacticization process, 'whether - or' comes to mark a connection
between protasis and apodosis rather than between the interrogative and the
superordinate clause in which the latter is embedded (König 1992: 430).
Since it is accompanied by the acquisition of conditional and irrelevance
meaning by an interrogative, this process clearly represents yet another case
of semantic strengthening rather than bleaching in the course of grammatical-
ization (Traugott and König 1991).
Clearly, a nucleus-satellite clause complex being absorbed into depen­
dency on another nucleus is an intriguing development, particularly since, as
argument (c) below shows, the new satellite's degree of grammatical integration
into the resulting adverbial clause complex is rather low in most Germanic
languages. It provides clear evidence of the evolutionary chain postulated by
Herring (1991: 278), i.e. for the claim that pragmatic-based meanings, rather
than late concomitants of grammaticalization (Traugott 1982), "may consitute
the very roots of grammar":
pragmatic (interactive/expressive) > textual (organizational) > grammatical
We encountered the text-organizational use of concessive conditionals in
chapter 4: what started life as a Rhetorical Dialogue modelled on typical
question-answer sequences can, given a certain degree of syntacticization and
"pragmatic bleaching" (Herring 1991: 279), take on an 'arrestive' text-
organizational function that, in the case of universal concessive conditionals,
may be described as explicating a Common Integrator. Examples of the
further evolution into the grammatical domain are less easily found in conces­
sive-conditional expressions in Germanic, given the slow diachronic evolu­
tion of this clause type, although the reduction of concessive conditionals via
verbless 'abbreviated clauses' (Backhand 1984) to such adverbs as willy-nilly,
derived from the non-canonical alternative concessive conditional clause will
he, nill he (König 1992: 432f.), does in fact demonstrate such a development.
Nothing perhaps illustrates the evolution of concessive-conditional ex­
pressions into grammar better than look-who in example (16) above, where
this imperative has become a subordinating conjunction. It is therefore in this
context that argument (b) must be seen, which is largely based on the evolu­
tion of WH-'ever'-compounds. Whereas in English ever is very closely
Interrogatives in concessive conditionals 183

amalgamated with the interrogative pronoun, thus pointing to a high degree of

grammaticalization, the corresponding expressions in German are much less
developed: immer, for instance, can co-occur in various positions with auch
and is never univerbated with the WH-pronoun. Older English, too, shows
considerable variation in this area (see Visser 1966: 913ff. for examples), but
the only alternatives available nowadays are formed by inserting -so-, an
option already largely (though not completely) archaic. Other languages
employ slightly different strategies for the formation of free-choice pronouns:
Dutch (WH dan ook, WH ook maar) has completely abandoned earlier mark­
ing with immer, and Swedish has än 'still' and som helst. A process of
differentiation appears to be under way splitting up languages whose strate­
gies of free-choice pronoun formation started out extremely similar a thousand
years ago, even though, as Haspelmath and König's typological survey re­
veals, any radically new strategies have yet to develop in the process
(Haspelmath and König forthcoming).
Argument (c), fully adopted here from König and van der Auwera
(1988), finally refers to all subtypes of concessive conditionals. It starts from
the observation that full grammatical integration ('hierarchical downgrad­
ing', Lehmann 1988: 183) of a subordinate clause into the clause complex is
indicated in the Germanic languages (other than English) by the preservation
of overall V/2 consitutent order, i.e. when the subordinate clause fills the
forefield of the verb. Although this is fully the case in (non-speech-act)
canonical conditionals, it is true of concessive conditionals only in the Scan­
dinavian languages. Examples like the following (König and van der Auwera
1988: 109; König 1992: 432) from Swedish are impossible in German and
Dutch:

(27) a. Hur mycket du än vet, (så) behöver du veta mera.

('However much you know, you ought to know more'; cf. G.
'Wieviel du auch weißt, du solltest noch mehr wissen/*solltest
du noch mehr wissen.')
b. Antingen du går eller inte, (sä) blir resultatet bra.
('Whether you go or not, the result will be good' ; cf. G. ,Ob du
gehst oder nicht, das Ergebnis wird gut/*wird das Ergebnis
gut')
In the course of centuries, V/2 has become almost obligatory in Swedish
concessive conditionals but not yet in German and Dutch, where concessive
184 Torsten Leuschner

conditionals continue to remain at the margin of the main clause (Lehmann

1988: 192). Despite minor differences in the rates at which German and
Dutch develop clause-integrating patterns in concessive conditionals, non-
integration of the protasis on the whole remains the unmarked option in these
two languages, with German slightly ahead of Dutch, whereas integration is
the unmarked option in Scandinavian. König and van der Auwera (1988: 128,
124) therefore suggest that the relevant languages have been affected to
varying degrees by a diachronic process of clause integration leading to tighter
subordination, which extends to conditionals and concessives as well as
concessive conditionals and whose extreme slowness in the latter seems to be
influenced by the 'separate assertibility' of the main clause.
Our study thus confirms the expectation that rhetorical question strate­
gies are likely to correlate with grammatical subordination in many lan­
guages, and that "the evolution of the former into the latter is well-motivated
in terms of basic communicative functions which all speakers of all languages
share" (Herring 1991: 279f.). Yet in the light of chapter 4, parts of the
evolution from discourse of concessive conditionals could also be interpreted
as a challenge to the unidirectionality hypothesis: barely grammaticalized,
concessive conditional expressions like G. wie auch immer, wie dem auch sei,
D. hoedanook and S. hur som helst (as well as their English equivalent,
anyway) are already being fed back into the pragmatic mode for uses based on
the 'arresting' function, including cooperative topic-switching in conversa­
tions, and as a source for concessives. On this view, concessive conditionals
are an interesting example of the phenomenon now receiving increasing
attention in the literature under the label of 'pragmaticalization' (Erman and
Kotsinas 1993), whose precise relationship with grammaticalization is as yet
an open question.

NOTES

1. The examples are part of a database of concessive conditionals in English, German,

Dutch and Swedish which scholarships from the DAAD and the Senat von Berlin (Nafög)
enabled me to assemble in 1994/5. The data came from the following corpora:
BROWN: Brown University corpus of written American English
LOB: Lancaster-Oslo/Bergen corpus of written British English
MK: Mannheimer Korpus of written German at the Institut für deutsche Sprache
INL: 5-Million-Word corpus of written Dutch at the Institute for Dutch Lexicology
SUC: Stockholm-Umeå corpus of written Swedish
Interrogatives in concessive conditionals 185

Source information (unavailable for SUC) is given at the end of each example; all
translations of examples into English are by myself. The search programmes originally
used were either supplied with the corpora (MK, INL) or home-made. Many thanks are
due to the Institut für deutsche Sprache in Mannheim and the Instituut voor Nederlandse
Lexicologie at Leiden University for kindly providing on-line access to their corpora and
programmes, and I am above all indebted to Gert Durieux at the University of Antwerp
and Gunnel Källgren and Gunnar Eriksson at Stockholm University for supplying the
remaining corpora and helping with the search programmes.

2. While this holds true for the West Germanic languages, Swedish has antingen - eller
('either - or', cf. Norwegian and Danish enten - eller), a construction not mentioned by
Haspelmath and König (forthcoming) which never combines with superordinate expres­
sions of irrelevance. I shall therefore focus exclusively on the 'whether - or' pattern.
3. Concessive-conditional conjunctional adverbs reflect precisely the three varieties of
concessive-conditional subordinating conjunctions seen in (1). Typical examples from
English include anyway, anyhow, at any rate, in any case, in any event, at all events;
either way, one way or the other; even then, even so. Concessive conditionality can also
be expressed by means of the prepositions irrespective of and regardless of (Haspelmath
and König forthcoming: 11).
4. For the role of intonation in clause-combining, see Couper-Kuhlen (1996).

REFERENCES

Bäcklund, Ingegerd. 1984. Abbreviated Conjunction-Headed Clauses in English. Uppsala:

Almqvist & Wiksell.
Behagel, Otto. 1923. Deutsche Syntax. Eine geschichtliche Darstellung. Band 1. Heidel­
berg: Winter.
Couper-Kuhlen, Elizabeth. 1996. 'Intonational and Clause-Combining in Discourse: the
Case of because." Pragmatics 6: 389-426.
De Castro Campos, M. Fausta P. 1985. "On conditionals as dialogue constructs." In
Marcelo Dascal (ed.), Dialogue: an Interdisciplinary Approach, 101-113. Amsterdam/
Philadelphia: John Benjamins.
Erman, Britt and Ulla-Britt Kotsinas. 1993. "Pragmaticalization: the Case of ba' and you
know." Stockholm Studies in Modern Philology N.S. 10: 76-93.
Givón, T. 1979. On Understanding Grammar. New York: Academic Press.
Haiman, John. 1978. "Conditionals are Topics." Language 54: 564-589.
Haiman, John and Sandra A. Thompson (eds). 1988. Clause Combining in Grammar and
Discourse. Amsterdam/Philadelphia: John Benjamins.
Haspelmath, Martin and Ekkehard König. Forthc. "Concessive conditionals in the lan­
guages of Europe." In Johan van der Auwera (ed.), Adverbial Relations in the Lan­
guages of Europe. Berlin and New York: Mouton de Gruyter.
Heine, Bernd, Ulrike Claudi and Friederike Hünnemeyer. 1991. Grammaticalization: a
Conceptual Framework. Chicago and London: Chicago U.P.
186 Torsten Leuschner

Herring, Susan C. 1991. "The grammaticalization of rhetorical questions in Tamil." In

Traugott and Heine (eds), Vol. 1, 253-284.
Hopper, Paul J. and Elizabeth C. Traugott. 1993. Grammaticalization. Cambridge: C.U.P.
Horn, Wilhelm. 1923. Sprachkörper und Sprachfunktion. Second ed. Leipzig: Mayer &
Müller
Jespersen, Otto. 1940. A Modern English Grammar on Historical Principles. Vol. 5:
Syntax. London: Allen & Unwin.
Klein, Josef. 1980. "Die Konzessiv-Relation als argumentationstheoretisches Problem."
Zeitschrift für germanistische Linguistik 8: 154-169.
König, Ekkehard. 1986. "Conditionals, concessive conditionals and concessives: Areas of
contrast, overlap and neutralization." In Traugott et al. (eds), 229-246.
König, Ekkehard. 1992. "From discourse to syntax: the case of concessive conditionals." In
Rosemarie Tracy (ed.), Who Climbs the Grammar Tree, 423-434. Tübingen: Niemeyer.
König, Ekkehard and Peter Eisenberg. 1984. "Zur Pragmatik von Konzessivsätzen." In
Gerhard Stickel (ed.), Pragmatik in der Grammatik, 313-332. Düsseldorf: Schwann.
König, Ekkehard and Johan van der Auwera. 1988. "Clause integration in German and
Dutch conditionals, concessive conditionals, and concessives." In Haiman and Thomp­
son (eds), 101-134.
Lang, Ewald. 1984[1977]. The Semantics of Coordination. Amsterdam/Philadelphia: John
Benjamins [originally published as Semantik der koordinativen Verknüpfung. Berlin:
Akademie-Verlag].
Lehmann, Christian. 1988. "Towards a typology of clause linkage." In Haiman and
Thompson (eds), 181-226.
. 1995[1982]. Thoughts on Grammaticalization. München: Lincom [originally
published as Thoughts on Grammaticalization, Vol.1. Cologne: Institut für
Sprachwissenschaft der Universität Köln].
Lötscher, Andreas. 1993. "Relativsätze mit konditionaler Funktion im Mittelhochdeut­
schen: Reanalyse oder Grammatikalisierung?" In Werner Abraham (ed.), Grammatika-
lisierung und Reanalyse: Konfrontation. Special Issue of Folia Linguistica Historica
13: 167-187.
Matthiessen, Christian and Sandra A. Thompson. 1988. "The structure of discourse and
'subordination'." In Haiman and Thompson (eds), 275-329.
Popovici, Vasile. 1981. "Dialogues rhétoriques." Degrés 9: il-i6.
Poutsma, Hendrik. 1928/9. A Grammar of Late Modern English for the Use of Continental,
Especially Dutch, Students. Second ed., 2 Vols. Groningen: Noordhoff.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik. 1985. A Compre­
hensive Grammar of the English Language. London: Longman.
Roulet, E., A. Auchlin, J. Moeschler, C. Rubattel and M. Schelling. 1985. L'articulation du
discours en français contemporain. Berne: Peter Lang.
Schelling, Marianne. 1982. "Quelques modalités de clotûre, les conclusifs: finalement, en
somme, au fond, de toute façon." Cahiers de linguistique française 4: 63-106.
Stoett, F. A. 1977. Middelnederlands che Spraakkunst: Syntaxis. Fifth ed. The Hague:
Nijhoff.
Sweetser, Eve. 1990. From Etymology to Pragmatics: Metaphorical and Cultural Aspects
of Semantic Structure. Cambridge: C.U.P.
Interrogatives in concessive conditionals 187

Traugott, Elizabeth C. 1982. "From propositional to textual and expressive meanings:

some semantic-pragmatic aspects of grammaticalization." In Winfred P. Lehmann
and Yakov Malkiel (eds), Perspectives on Historical Linguistics, 245-271.
Amsterdam/Philadelphia: John Benjamins.
. 1985. "Conditional Markers." In John Haiman (ed.), Iconicity in Syntax, 289-307.
Amsterdam/Philadelphia: John Benjamins.
Traugott, Elizabeth C. and Bernd Heine (eds). 1991. Approaches to Grammaticalization.
Amsterdam/Philadelphia: John Benjamins.
Traugott, Elizabeth C. and Ekkehard König. 1991. "The semantics-pragmatics of gram­
maticalization revisited." In Traugott and Heine (eds), Vol. 1, 189-218.
Traugott, Elizabeth C , Alice ter Meulen, Judith S. Reilly and Charles Ferguson (eds). 1986.
On Conditionals. Cambridge: C.U.P.
Visser, Fredericus Th. 1966. An Historical Syntax of the English Language, Part 2:
Syntactical Units with One Verb, Continued. Leiden: Brill.
Wunderlich, Dieter. 1976. "Fragesätze und Fragen." In D. Wunderlich, Studien zur Sprech­
akttheorie, 181-250. Frankfurt: Suhrkamp.
Zaefferer, Dietmar. 1991. "Conditionals and unconditionals." In D. Zaefferer (ed.), Univer­
sal Semantics and Semantic Universals, 210-236. Berlin and New York: Foris.
 The grammaticalization of
the left sentence boundary in Hittite

Silvia Luraghi
Third University of Rome

0. Introduction

In the present paper, the term 'grammaticalization' does not refer to the
creation of new grammatical forms out of lexical items; rather, the processes
described are cases of increasing obligatoriness of grammatical forms, i.e.
various types of clitic (connectives, particles, pronouns). Besides, the main
point in the paper is not so much in describing the further grammaticalization
of these items as in showing how their occurrence in virtually all sentences
caused the left sentence boundary to have a peculiar structure and how this
was used for pragmatic purposes.
A well known word order rule of the ancient Indo-European languages
concerns the placing of enclitic sentence particles, unstressed pronouns and
other types of postpositives1 close to the left border of the sentence in which
they occur, i.e. mostly after the first accented word. This tendency is known as
Wackernagel's Law, and it was first observed in Wackernagel (1892).
Given the state of Indo-European linguistics at that time, Wackernagel
based his observations especially on Old Indic, Ancient Greek, and Latin.
When Hittite was finally deciphered in the early decades of the present
century, it became apparent that this language was to bear out Wackernagel's
remarks, given its strict rules of clitic placement, accompanied by an unusual
richness in various types of clitics.
In my paper I will contrast the development of Wackernagel's Law in
Hittite and in Greek. I will show how the use of Wackernagel's clitics,
including particles and pronouns, increased during the attested history of the
190 Silvia Luraghi

Hittite language, and how this development brought about changes in the use
and distribution of prepositive sentence connectives. I will argue that the
extension of most particles was the result of grammaticalization, since par­
ticles and even pronouns have often become obligatory after undergoing
semantic bleaching or loosing pragmatic force. Besides, I will show that some
word clitics, which were not restricted to second position, have been replaced
by second position clitics, thus contributing to the overall increase of the
latter. I will start with some observations about Wackernagel's Law (§ 1) and
its effects in Indo-European (§ 2); in the central section of my paper (§ 3) I
will discuss the Hittite data, both synchronically and diachronically. I will
show how the rigid structure of the left sentence boundary could allow for two
sentence patterns, i.e. one with the enclitics hosted by an otherwise empty
connective, which served the purpose of extracting them from the sentence,
and one with a topicalized left dislocated word, which was separated from the
remainder of the sentence through the intervention of the enclitics. Section 4
contains the conclusions.

1. Wackernagel's Law

As I have shown in Luraghi (1990a) enclitics and postpositives of different

type occur in second position for two different reasons, connected with their
grammatical and discourse status.
a. Sentence particles such as coordinators and connectives, discourse mark­
ers, and modal particles, which have the whole sentence as their scope,
tend to occur as early as possible in the sentence. Note that postpositive
connectives have placement rules similar to those of prepositive, which
occur at the beginning of a sentence (Hittite examples will be discussed
below): unstressed particles and postpositives occur after the first ac­
cented word in the sentence, this being the leftmost accessible position
for items that cannot begin a sentence;
b. enclitic pronouns, which belong in the VP, in their turn, are attracted
close to the left sentence boundary for pragmatic reasons. As Steele
(1977) has shown, enclitic forms often are morphological variants of
stressed forms. Unstressed pronouns have a low communicative dyna­
mism, since they do not convey new information; they rather refer back to
items which have already been introduced in the preceding discourse.
The grammaticalization of the left sentence boundary in Hittite 191

Thus, they also fulfill a textual function, connecting sentences with each
other, and contributing to the building of discourse continuity. When
unstressed variants of pronouns develop into special clitics2 they usually
follow two paths: either they are hosted by verb, as Romance clitics do,
or they follow Wackernagel's Law, as shown in various Indo-European
and non-Indo-European languages (see Renzi, 1989).
The basic difference between clitics and particles in (a) and clitic pro­
nouns in (b) lies in the relation between their structural and their phonological
host. The (a) forms are attached phonologically to the whole sentence (i.e. to
its border), to which they also refer structurally; the (b) forms, on the contrary,
have the VP as their structural host, but they take the sentence border as their
phonological host.
Keeping this important difference in mind, we now turn to the Indo-
European data.

2. Wackernagel's Law in Indo-European

In the Indo-European languages where Wackernagel's Law applies, clitic

sentence connectives always precede unstressed pronouns when several forms
occur together in second position.3 Hittite examples are particularly clear,
given the wide use of enclitics:
mD
(1) piran =ma= at= mu XXXPU-as DUMU mzida
before CONN 3SG-N/A 1SG-OBL A.-NOM child Z.
maniyahhiskit
administrate-3SG-PRET-ITER
'before me Armadatta, the son of Zida, had administrated it', see
ex. (31);
(2) n= as= mu= kan huwais
CONN 3SG-NOM 1SG-OBLPTC escape-3SG-PRET
'he ran away from me', see ex. (24).4
In the other Indo-European languages Wackernagel's Law displays a
variety of patterns which depend on different definitions of second position.
The following examples from Old Indic illustrate various possibilities. In
examples (3) and (4) it is shown that, especially with compound verbs, the
192 Silvia Luraghi

preverb can either host the enclitics, or it can be counted together with the
verb, and consequently skipped:
(3) práti enam osati... ná enam práti osati
PREV 3SG-ACC burn-3SG-PRES NEG 3SG-ACC PREV burn-3SG-PRES
'it burns him ... it does not burn him', TS 1.5.9.7;

(4) apá hnute evá asmai tád

PREV hide-3SG-PRES-M PTC 3SG-DAT it-ACC
'he hides it to him', TS 1.5.9.7.
Example (5) shows that connectives can be skipped altogether (here we
find the additive connective áthah, which is most often skipped):
(5) átho mánas a vái Prajápatis yajñám
CONN mind-INSTR PTC P.-NOM sacrifice-ACC
atanuta
accomplish-3sG-AOR
'and then P. performed the sacrifice with his mind', TS 1.6.8.4.

Furthermore, left dislocated constituents, in particular in wh-clauses, do

not often count as initial for the purpose of Wackernagel's Law, and the
enclitics are placed after the first word which follows the left dislocated
constituent (often the wh-form):

(6) ukté ká u svid ántamah

hymn-LOC who-NOM PTC 2PL-DAT dearest-NOM
'in a hymn, who is the dearest one for you?', RV 8.64.9c.
This tendency creates the possibility of having two positions for enclitics
within the same sentence:
(7) utá va yó no marcáyad ánagasah
even PTC who-NOM 1PL-ACC damage-3sG-suBJ innocent-ACC
'or who would even damage innocent us', RV 2.27.3.

The same possibilities in defining second position are found in Classical

Greek, where enclitic pronouns mostly do not separate the article from the
noun5 and conjunctions are often skipped (this holds for subordinating con­
junctions, as ei, 'if', in (8), too):
The grammaticalization of the left sentence boundary in Hittite 193

(8) ei kaì min Olumpios autòs egeírei

if also 3SG-ACC Olympian-NOM self-NOM push-3sG-PRES
'if even the Olympian himself pushes him', Homer Il. 13.58.

The possibility of skipping left dislocated constituents is extended to all

types of clauses. Often enclitic pronouns end up following the verb:
(9) kai he gunè: eporãi min exiónta
and the woman-NOM see-3sG-PRES 3SG-ACC go-out-PART-ACC
'and the woman saw him going out', Herodotus Histories, 1.10.2.
According to Marshall (1987) the position after the finite verb is the
rightmost one which is accessible to enclitic pronouns.6 With little variants,
this position is the only one found in the New Testament:

(10) ho paté:r sou kagd: oduno:menoi

the father-NOM 2SG-GEN and-I-NOM suffer-PRES-PART-NOM-PL
eze:toûmén se
ask-1 PL-PRES 2SG-ACC
'your father and I ask you in sadness', Lc. 2.48,
and it is the regular position for unstressed pronouns in Modern Greek.
Example (10) also shows another tendency of Greek that goes contrary to
Wackernagel's Law, the placing of enclitic genitive forms of personal pro­
nouns after the NP to which they belong syntactically. This is an early
tendency, as shown in
(11) tò dè lógon au mè: tôi hré:matí mou
the PTC word-ACC PTC NEG the-DAT discourse-DAT 1SG-GEN
dw:ke
f0ll0W-2SG-IMPER
'do not try to follow my demonstration in ...', Plato Theaet. 168d.

Examples where the genitive form of personal pronouns follows

Wackernagel's Law are also attested, but they are not a significant majority.
The pattern in which enclitics occur in two different positions in the
sentence, as in Old Indic (7), occurs very frequently in Classical Greek, and it
does not depend on the sentence type (in Old Indic it is mostly limited to wh-
caluses). Placing clitics in two (or more) different positions offers the possibil­
ity of establishing weak accentual boundaries among constituents, thus adding
to a possible pragmatic use of unaccented particles shown in (7), where both
194 Silvia Luraghi

the adverb utá, 'even', and the pronoun yó, 'who', are isolated by the enclitic
no from the rest of the sentence, and are left dislocated for emphasis. In (12)
the same tendency is shown in Classical Greek:
(12) taûta men dé: isa pros ísa
that-N/A-PL-N PTC PTC same-N/A-PL-N to same-N/A-PL-N
sphi genésthai metà dè taûta Héllenas
3PL-DAT become-INF after PTC that-N/A-PL-N Greek-ACC-PL
aitíous... genésthai
guilty-ACC-PL become-INF
'at that moment their condition were equal: after that, the Greeks
became guilty ...', Herodotus Histories, 1.2.1.
In (12) the constituent taûta, that indicates the temporal setting of the first
sentence, is left dislocated and contrasted with the constituent metà taûta in
the second sentence. Then the constituent ísa pròs ísa is isolated from the verb
by the intervention of the enclitic pronoun sphi and is thus given emphasis.
This sentence is a sort of summary of the preceding passage and at the same
time it introduces the next passage, where a change in conditions is described
('the Greeks became guilty').
Scattering of clitics, especially pronouns, in an internal position often
had the effect of bringing them close to their syntactic host (i.e. the verb), as
Marshall noted and as I have shown in example (11). According to Dover
(1960: 18), the disruption of Wackernagel's Law in the case of clitics might
be an effect of their 'natural' tendency toward their structural host. Whereas
this is a possible explanation, mainly supported by the data on the placement
of the modal particle an and of genitive forms of pronouns, it must be
remarked that the tendency for unaccented pronouns to follow the verb is
clearly attested only from the New Testament onwards. In Classical Greek, as
Dover himself writes, it is doubtful "whether the distribution of q [i.e.
postpositives] over the constituent word-groups of a clause is motivated to
any significant degree by the desire to bring together words which 'belong
together'. ... the many clauses in which distribution has the effect which
seems 'natural' to speakers of modern English are matched by an equally
large number in which it has the opposite effect" (1960: 18-19).
A relation between Wackernagel's Law and topicalization has been
noted in Steele (1977). On the base of Uto-Aztecan languages, Steele argues
that Wackernagel's Law was created by the occurrence of left dislocated,
The grammaticalization of the left sentence boundary in Hittite 195

topicalized constituents. The Indo-European data point rather towards an­

other direction: second position clitics had the effect of highlighting left
dislocation, by setting a boundary within the sentence.

3. Hittite

As I have mentioned above Hittite has a wide variety of enclitics, some of

which have no correspondents in the other Indo-European languages. Second
position clitics occur in fixed order:
a. sentence connectives and conjunctions: -(y)a-, coordinator; -ma-, -a-,
adversative particles, -man- modal particle (which also has an accented
variant);7
b. -wa(r)-, direct speech particle, not found elsewhere in Indo-European;
c. nominative or accusative of third person pronoun singular or plural.
While unstressed accusative forms of third person pronouns are attested
in the other Indo-European languages as well, Hittite is alone in having
developed an unaccented nominative form. Since the Ancient Indo-
European languages were Null Subject languages, third person nomina­
tive pronouns usually had contrastive value and were stressed. Hittite has
special enclitic subject forms although it allowed null subjects, too, as we
will see below;
d. oblique forms of first and second person singular and plural or dative of
third person singular or plural.8 Note that, whereas third person nomina­
tive and accusative clitic pronouns cannot co-occur with each other, they
can co-occur with any dative form, including the third person;
e. -z(a)-, reflexive particle;
f. -kan, -(a)sta, -san, -an, -(a)pa, so-called local particles. This type of
particle does not occur in any of the other Indo-European languages.
All the above mentioned forms have a slot that they can fill (here
symbolized by letters (a) through (f)); each slot can be filled by one only of the
possible clitics, which means that clitics in each of the above groups are
mutually exclusive. Clitics in slot (a) can appear only if none of the prepositive
connectives (nu, ta, su) occurs in the initial position. The occurrence of certain
particles after clitic pronouns in Hittite does not go contrary to the tendency
for clitics with a wider scope to precede those whose scope is more restricted.
In Old Hittite, local particles had a local NP or adverb as their scope; the
196 Silvia Luraghi

particle -z(a)- most likely had the subject as its scope. Only later both types of
particle acquired a different scope, which in any case, remained restricted to a
sentence constituent only (the VP, see § 3.2.2. and 3.2.3).
There obviously are some exceptions to these rules, some of which will
be discussed below, but on the whole they apply throughout the history of the
Hittite language consistently. Major changes between Old Hittite9 and later
language stages concern the use of prepositives connectives and the fre­
quency of second position clitics, rather than their placement.

3.1. Old Hittite

All regularities described in the preceding section held in Old Hittite already.
However, many Old Hittite sentences did not contain enclitics and/or preposi­
tive connectives, as shown below, in example (28).
Beside the enclitics described in § 3, Old Hittite also had a set of enclitic
possessives. The latter were inflected forms that functioned as nominal modi­
fiers; they did not follow Wackernagel's Law, but were hosted by their head
noun, as shown in
(13) nu- us appa ishi= ssi pennai
CONN they-ACC back master-D/L 3SG-POS-D/L drive-3sG-PRES
'(s)he takes them (= the oxen) back to their (sg.) owner', HG § 79
(=Friedrich, 1959).
(14) takku LÚ.ULULU-as ELLAM-as KAxKAK=set
if man-GEN free-GEN nose 3SG-POSS-N/A
kuiski waki
someone-NOM-SG bite-3sG-PRES
'if someone bites the nose of a free man', HG § 13 (A i 24)
(=Friedrich, 1959).
The disappearance of enclitic possessives after the Old Hittite period has
a bearing on the increase in the use of second position clitics, as I will show in
§ 3.2.5.
The grammaticalization of the left sentence boundary in Hittite 197

3.2. Development after Old Hittite

3.2.1. Direct speech particle

Complement clauses are a late development in Hittite, and their use never
reached a great frequency; in particular, they were never used with verbs of
saying. In the place of complement clauses, paratactic clauses are normally
found, either without any particles, or with the clitic -wa(r)-, that introduces
direct speech. In Old Hittite the particle is mostly found where no verbs of
saying occur, as in
(15) s= as sara URU-ya pait uk- wa
CONN 3SG-NOM up city-DIR go-3SG-PRET lSG-NOM PTC
LUGAL-us=s mis kisha
king-NOM your become-1SG-PRES
'he went up against the city, saying: 'I shall become your king!",
StBoT 17 Rev. 14-15 (=Otten, 1973).
In the next example, from New Hittite, the particle co-occurs with the verb
mema-, 'to speak' :
(16) nu= mu memir paiueni- war- an-
CONN 1SG-OBL speak-3PL-PRET gO-1PL-PRES PTC 3SG-ACC
kan kuennumeni nu- wa= tta SAG.DU-an
PTC kill-lPL-PRES CONN PTC 2SG-OBL head-ACC
utumeni
bring-1PL-PRES
'they told me: 'we shall (go) kill him and shall bring his head to
you', StBoT 24 ii 24-26 (= Otten, 1981).
In general, one notices an extension in the occurrence of the particle after
the Old Hittite period. Note further that the particle -wa(r)- was never obliga­
tory in Old Hittite, not even in cases where no verbs of saying occurred. Its
use was pragmatically oriented, and emphasized the introduction of direct
speech in a reported narrative. Since the particle has its etymology in a verbal
root (Hittite weriya- means 'to call' and it is cognate to Latin verbum and
English 'word'), it had already undergone partial grammaticalization, involv­
ing semantic bleaching. Later on as the process of grammaticalization contin­
ued, -wa(r)- lost its pragmatic force as well, and it became virtually obligatory
in sentences such as (16).10
198 Silvia Luraghi

3.2.2. Reflexive particle

The particle -z(a) is generally referred to as 'reflexive particle'. In Old Hittite,
it expressed a major involvement of the subject in a certain event, and it
appeared to be in complementary distribution with the middle voice, since it
never occurred in sentences with middle verbs. An example of -z(a) in Old
Hittite is
(17) nu= zza DUMU.NITA MEŠ karti= smi piran
CONN PTC boy-PL heart-D/L 3PL-POSS-D/L before
memir
speak-3PL-PRET
'the boys said to themselves', StBoT 17, Obv. 3-14 (=Otten, 1973).
Later on, both the particle -z(a) and middle voice underwent semantic
change. In particular, the particle took over reflexive and reciprocal functions.
As an example, one can quote the verb es-, 'to sit down', which never occurs
with the particle in Old Hittite, but regularly does in later texts. Besides, it is
consistently found with certain verbs, especially when they can convey differ­
ent meaning with or without the particle. An example of this tendency is the
verb kis-, which can mean either 'to become', or 'to happen'. Both meanings
are conveyed without the aid of the particle in Old Hittite, whereas in the post
OH period the particle regularly occurs when the verb means 'to become'.
Compare the occurrence of kisha, 'I have become', without particle in (15),
with kisat, 'he became', and kishahat, 'I became' with -z(a) in (18) (example
(18) also contains an occurrence of es-, 'to sit down' with the particle):
(18) mahhan= ma= za ABU-YA mmursilis DINGIRLIM
when CONN PTC father my M.-NOM god
kisat SES- YA=ma= za= kan m NIR.GÁL ANA
become-3sG-PRET brother my CONN PTC PTC M. on
GlŠ
U . Z A ABI- ŠU esat ammuk- ma- za
throne father his sit-3sG-PRET 1SG-NOM CONN PTC
ANA PANI ŠEŠ- YA EN.KARAS kishahat
in-front-of brother my army-commander become-1SG-PRET
'when my father Mursili became a god (i.e. died), my brother
Muwatalli sat on his father's throne while I became army com­
mander in front of my brother' (see example (31)).
Again, as in the case of -wa(r)-, the occurrence of the particle increased
The grammaticalization of the left sentence boundary in Hittite 199

through grammaticalization, since -z(a) became obligatory in contexts where,

at an earlier stage in the language, it sometimes occurred and sometimes did
not.

3.2.3. Local particles

In Old Hittite, the so-called 'local particles' are a group of five particles, -kan,
-asta, -san, -(a)pa, -an, which occur in sentences that contain some kind of
space expressions, such as a NP in the d/1, or in the directive, or a local adverb,
as in the following examples:
NINDA
(19) sarrui= ma= ssan ERÍNMEŠ-az eszi
s.bread-D/L CONN PTC troop-NOM lie-3sG-PRES
'a (clay figurine of a) troop lies on the s.bread', StBoT 8, i 30 (=
Otten & Soucek, 1969);
(20) ug= an namma anda [p]aimi
1SG-NOM PTC besides into go-1sG-PRES
'I go inside again', StBoT 8, ii 45 (= Otten & Soucek, 1969);
Local particles are the set of enclitics that display the biggest irregularities
with respect to placement rules. As Neu (1993) has shown, sporadically two
particles co-occur in the same sentence. In this case, one takes Wackernagel's
position, whereas the other one is hosted by a Location or Direction expression,
as in
GIS
(21) n= asta MA istappesnas PA5 -as istappesnaz
CONN PTC ship basin-GEN channel-NOM basin-ABL
para ÌD=kan anda pedai
towards river-PTC into lead-3sG-PRES
'the channel of the basin leads the ship outside toward the river
from the basin', KUB 29.7+ rev. 51-52.
In a small number of other passages, only one local particle occurs in a
sentence, and it is placed in internal position, again hosted by some local
constituents. The sentence in (22) is an example of two coordinated constitu­
ents at the beginning of a sentence, where the particle -san is placed after the
second constituent, thus avoiding an interruption of the unit formed by coordi­
nation:
200 Silvia Luraghi

(22) halmasuitti hassi-ya=ssan tiyanzi

throne-D/L hearth and PTC put-3PL-PRES
'and they (sc. the king and the queen) put (them, sc. the bowls) on
the throne (and) on the hearth', StBoT 8, ii 49 (=Otten & Soucek,
1969).
After the Old Hittite period, the use of sentence particles underwent two
major changes, both of which are typical of grammaticalization:
a. an increase in the overall frequency of some particles (in particular -kan
and, to a lesser extent, -asta);
b. a semantic development, that changed their concrete local meaning into a
more abstract aspectual one.
Especially in New Hittite, the particles appear to be related with space
expressions only occasionally, whereas their occurrence is often connected to
the meaning of the verb. Very often, telic verbs take the particle -kan (less
frequently -asta): an example of this tendency is given by the verb kuen-, 'to
kill', that never occurs without a particle after the Old Hittite period. Later on,
kuen- always takes the particle-ten, which clearly has no spatial implications:10
(23) [man=t]a= kkan kuenta
PTC 2SG-OBL PTC kill-3SG-PRET
'he would have killed you', Madd. Ro. 3-4 (= Goetze, 1927).
Furthermore, the loss of concrete meaning also brings about a reduction
in the number of particles, so that only -kan actually survives and remains
productive: -asta is lexically constrained and mostly limited to archaizing
texts in New Hittite, -an, already very infrequent in Old Hittite period,
disappears immediately thereafter, and -san and -(a)pa survive through the
Middle Hittite period and then disappear as well.

3.2.4. Obligatory third person subjects

Hittite is a peculiar type of null subject language, where zero subjects are
regularly found with all kinds of verbs for first and second person singular and
plural only. With third person sg. and pl. on the contrary, NS and nominative
forms of the third person enclitic pronoun are in complementary distribution,
depending on verbal transitivity: all transitive verbs that can be found with a
direct object can take third person zero subjects, while intransitive verbs
cannot have null subjects and consequently their subject must be either a NP
or accented pronoun, or it has to be the nominative form of enclitic personal
The grammaticalization of the left sentence boundary in Hittite 201

pronouns. This means, among other things, that enclitic third person nomina­
tive and enclitic third person accusative, which are built from the same root
(-a-), also occur in complementary distribution: transitive verbs can take a
clitic third person object, but not a subject, whereas intransitive verbs can take
a clitic third person subject, but not an object. In the following example the
verb pai-, 'to go', occurs with a first person zero subjects in (24a,b), and with
a third person clitic in (24e):
(24) a) nu- kan INA KUR URUarzauwa paranda paun
CONN PTC in country A. upwards go-1SG-PRET
URU
b) nu INA apasa ANA U R U LIM
SA muhha-LÚ andan paun
CONN in A. to city of U. into go-lSG-PRET
m
c) nu- mu uhha-LÚ-is UL mazzasta
CONN 1SG-OBL U.-NOM not resist-3SG-PRET-M/P
d) n- as= mu- kan huwais
CONN 3SG-NOM lSG-OBL PTC eSCape-3SG-PRET
e) n- as- kan aruni paranda gursawanza
CONN 3SG-NOM PTC Sea-D/L t o w a r d s sail-PART-NOM-SG-C
pait
go-3SG-PRET
f) n- as= kan apiya anda esta
CONN 3SG-NOM PTC there in be-3sG-PRET
'I went up to the country of Arzawa. In the city of Apasa I went
into Uhhaziti's quarters and Uhhaziti did not make any resistance.
He escaped me and went sailing on the sea and stayed there',
(Goetze,1933 50).
In the second part of the above passage the subject remains the same over
several connected sentences: from a discourse point of view, the degree of
continuity should be high enough to allow omission of the subject; however,
the clitic -as (third person singular nominative) is repeated in each sentence,
since all verbs are intransitive.
In Old Hittite texts there is a small number of occurrences12 where third
person null subjects are allowed with intransitive verbs as well, apparently
based on discourse conditions, as with the verb ar-, 'to arrive', in:
202 Silvia Luraghi

(25) DUMU.NITA MEŠ appa URU nesa ianzi nu ANŠE-in

son-PL back N.-DIR go-3PL-PRES CONN donkey-ACC
nannianzi man URUtamarra arir nu
drive-3PL-PRES when T.-DIR reach-3PL-PRET CONN
tarsikanzi
Say-3PL-PRES-ITER
'the sons go back to Nesa, and drive a/the donkey. On arrival in
Tamarra, they say:...', StBoT 17 obv. 7-8 (=Otten, 1973).
It must be stressed in any case that example (25) is rather exceptional even for
Old Hittite, where the use of third person clitic subjects was already well
established with intransitive verbs.

3.2.5. The loss of enclitic possessives

The use of enclitic possessive adjectives has been shown above, in example
(13). In Old Hittite, enclitic possessives could modify nominal heads as well
as local adverbs, as in (25):13
(26) ser=samet= a GÍR ZABAR kitta
over 3PL-A-POSS PTC dagger bronze lie-3sG-PRES
'a bronze dagger lies over them', StBoT 8 i 31' (=Otten & Soucek,
1969).
After the Old Hittite period possessive adjectives disappear. However,
the accented forms of personal pronouns do not appear to be used more
frequently: rather, possession is increasingly expressed through second posi­
tion enclitic personal pronouns in the oblique (dative for third person); i.e.
non-Wackernagel enclitics are replaced by Wackernagel ones:
LÚ
(27) KÚR MEŠ= mu= kan LÚ.MEŠ
arsanatallus D
IŠTAR
enemy-PL 1SG-OBL PTC envious-ACC-PL I.
GASAN-YA SU-i dais
Lady my hand-D/L put-3sG-PRET
'Istar my Lady put in my hand my enemies and those who were
envious of me', StBoT 24 (=Otten: 1981: 8).
The grammaticalization of the left sentence boundary in Hittite 203

3.2.6. Sentence connectives

3.2.6.1. Prepositive connectives

Additive connectives in Old Hittite are nu, ta, and su. They are used in order
to 'push forward' a narration, and express progression in the course of the
events. Additive connectives are sensitive to modality: they can appear in
assertions or in orders only, whereas they are incompatible with potential and
contrary-to-fact modality (particle -man-), and with prohibitive (negation le).
An example where nu is consistently found at the beginning of each sentence
in a context of narrative continuity is given above, in (24).
In Old Hittite asyndeton was often found in the same type of context,
provided that there were no clitics. When clitics occurred, they were hosted
by a connective, as in:
m
(28) a) zidis LÚZABAR.DIB esta
Z. cup-bearer be-3sG-PRET.
b) ABI LUGAL DUGharharan GESTIN-it ANA fhistaiyara
father king measure-N/A-Nwine-INSTR to H
m
maratti=ya maniyahhis
M. and assign-3sG-PRET
c) LUGAL-i SIG5-antan GESTIN-an hinkatta
king-D/L good-ACC wine-ACC pour-3sG-PRET
d) apedass- a tamain GESTIN-an pier
that-D/L-PL and other-ACC wine-ACC give-3PL-PRET
e) apass- a uit LUGAL-i tet
that-NOM and come-3sG-PRET king-D/L tell-3sG-PRET
f) natta apun GESTIN-an pier LUGAL-us
not that-ACC wine-ACC give-3PL-PRET king-NOM
kuin austa
which-ACC see-3sG-PRET
g) apass- a uit QATAMMA IQBI
that-NOM and come-3sG-PRET the-same (s)he-said
h) s= an- asta arha pehuter
CONN 3SG-ACC PTC out bring-3PL-PRET
i) s= an essikir
CONN 3SG-ACC beat-3PL-PRET
1) s= as BA.UG6
CONN 3SG-NOM die
204 Silvia Luraghi

'Zidi was cup-bearer. The king's father had assigned a measure of

wine to Histaraya and Maratti. He (sc. Zidi) gave the king good
wine and to them he gave another wine. One of them came and told
the king: 'I was not given the wine that the king has seen!' The
other one came and said the same. Zidi was taken away, beaten up
and killed' (from Luraghi, 1990b, 170 = Kbo 4 iii 36, l l ' - 1 7 ' ) .
After the Old Hittite period, not only asyndeton, but the connectives su
and ta, too, disappeared. Note that the distribution of nu and ta was very
similar in Old Hittite, except for a higher overall frequency of ta, and a higher
frequency of nu between preposed subordinate and main clauses, where the
use of a connective appeared to be an innovation, connected with the occur­
rence of clitics and with the need to mark sentence boundaries. The connec­
tive su was practically only found with pronominal clitics (as in (29)).

3.2.6.1. Adversative connectives

Adversative connectives in Old Hittite are enclitic -ma- and -a-. The connec­
tive -ma- denotes that something referred to in a certain clause is contrary to
expectations. Thus, it indicates discontinuity, either on the text level or in the
course of the events. Initial verbs, which also may be used in order to denote
that something is contrary to expectations, are frequently found with -ma-:
(29) anda- kan halinas tesummius tarlipit suwamus
inside PTC clay-GEN vessel-ACC-PL T.-INSTR full-ACC-PL
II-ki petumini tarueni- ma- at eshar
twice bring-1PL-PRES say-1PL-PRES CONN it blood-N/A-N
DUMU.E.GAL-is Dhantasepan LUGAL -i kissari
servant-NOM H.-ACC king-D/L hand-D/L
dai tesumminn- a pai
put-3sG-PRES vessel-ACC-SG and give-3sG-PRES
Twice we bring inside the clay vessels full of blood (we call it t.);
the Palace servant puts a H. divinity in the hand of the king and
gives (him) a vessel', StBoT 8, i 26-29 (=Otten & Soucek, 1969).
The connective -a- has the function of indicating topic shift. As an
example of the difference between -a- and additive connectives, see the
passage in the following example:
The grammaticalization of the left sentence boundary in Hittite 205

(30) a) SAL.LUGAL URUkanis XXX DUMUMEŠ IEN MU-anti

queen K. 30 boy-PL one year-D/L
hasta...
generate- 3SG-PRET
b) tuppus sakanda sunnas
basket-ACC-PL oil-INSTR fill-3SG-PRET
c) nu D U M U M E Š - ŠU anda ziket
CONN boys her inside put-3sG-PRET
d) s= us ÍD-a tarnas
CONN 3PL-ACC river-DIR leave-3sG-PRET
e) ÍD-s= a ANA A.AB.BA KUR URUzalpuwa
river-NOM CONN to sea country Z
pedas
carry-3sG-PRET god-NOM-PL
f) DINGIRDIDLI-s= a D U M U M E Š - U S A.AB.BA= az sara
god-N-PL CONN boy-ACC-PL sea-ABL outside
dair
take-3PL-PRET
g) s= us sallanuskir
CONN 3PL-ACC bring-Up-3PL-PRET
'The queen of Kanis generated thirty boys during the same
year. ... She filled some baskets with oil, laid her children
inside and abandoned them to the river. The river took the
children to the sea, in the country of Zalpuwa. The gods took
the children out of the sea and bought them up', StBoT 17
(=Otten, 1981), obv. 1-5.
In later texts the particle -a- disappeared. This is shown in Middle Hittite
texts by increasing confusion in the spelling of final consonants before -a-,
which was by then being interpreted as -(y)a- by scribes copying Old Hittite
texts. The function of -a- is consistently taken over by -ma- in Late Hittite.
Therefore, the function of -ma- in Late Hittite is twofold, as shown in example
(31) below:
a) introducing background information, typically in temporal clauses, as in
(31a), but possibly also elsewhere, as in (3If), where some background
information is provided for events that will be narrated further on in the
text in which the man called Armadatta will play a major role;
b) indicating topic switch, as in (31b) and (c).
206 Silvia Luraghi

In the following passage, the difference between coordination, marked

by -(y)a-, and addition, with nu, is neatly borne out in sentences (31d-e):
(31) a) mahhan=ma= za ABU-YA mmursilis DINGIRLIM
when CONN PTC father my M.-NOM god
kisat
become-3sG-PRET
b) SES- YA= ma= za= kan m NIR.GÁL
brother my CONN PTC PTC M.
ANA GIS GU.ZA ABI-ŠU esat
on throne father his sit-3sG-PRET
c) ammuk=ma= za ANA PANI SES-YA
1SG-NOM CONN PTC in front of brother my
EN.KARAS kishahat
army-commander become-3sG-PRET
d) KUR UGU=ya=mu maniyahhanni pesta
country upper and 1SG-OBL administrate-iNF give-3sG-PRET
e) nu KUR UGU 77 taparha
CONN country upper rule-lSG-PRET
mD
f) piran= ma= at= mu XXX. D U-as DUMU
before CONN 3SG-N/A ISG-OBL A.-NOM child
m
zida maniyahhiskit
Z administrate-3sG-PRET-ITER
'when my father Mursili became a god (i.e. died), my brother
Muwatalli sat on his father's throne while I became army
commander in front of my brother, and he gave me the Upper
Country to administrate. So I ruled the country. (Before me
Zida's son, Armadatta, had administrated it for a long time)',
StBoT 24 i 22-28 (= Otten, 1981).

3.3. The left sentence boundary in New Hittite

It has often been observed that second position clitics build an 'information
chain' (the term is from Rosenkranz, 1979) that conveyed relevant informa­
tion relative to the linking of the sentence with the preceding discourse as well
as to its inner grammatical structure. Some important semantic properties of
the clause and of the verb, as modality, degrees of transitivity, and aspect, are
The grammaticalization of the left sentence boundary in Hittite 207

anticipated to the left sentence boundary, which possibly was a way of

making up for the late occurrence of the verb at the right sentence boundary.14
However the rigid sentence structure of Hittite still allowed for word
order to be used for pragmatic purposes. The alternative between nu and -ma-
ox -(y)a- (or accented and unaccented man in non-assertive clauses) resulted in
two distinct patterns:
a. sentences with no topicalized or contrasted constituent started with nu
followed by the enclitics;
b. the other sentences had some accented constituent in the initial position,
which was separated from the remaining part of the sentence by the
enclitics.
The extension of nu as a sentence introducer was probably brought about
by the need to extract all clitics from the sentence, in order to allow for a
sentence pattern where no constituents were separated from the others.
Wackernagel's clitics marked the left sentence boundary in such a way that any
word or constituent that preceded them was extraposed, thus receiving particu­
lar emphasis. Wackernagel's clitics were no longer real second position clitics;
rather, they were placed at the beginning of the sentence, and the sentence
introducer nu occurred for prosodic reasons only.

4. Conclusions

As we have seen above, the rigid structure of the left sentence boundary in
Hittite was the result of a number of converging processes of grammaticaliza­
tion, mostly concerning items that occurred to the left of nominal constituents.
Grammaticalization of discourse particles (various types of connective, in­
cluding -wa(r)-), unstressed pronouns (in particular, third person nominative
forms), and other types of particles (reflexive and local particles) brought
about a consistent sentence pattern, in which the left sentence boundary was
always marked by the occurrence of unstressed forms. The connective nu
became obligatory as a host for clitics in sentences where no constituents
needed to be left dislocated for emphasis or contrast, whereas sentences with
left dislocated constituents mostly took the particle -ma- and were clearly set
aside of the sentence by the clitic chain.
Note that the loss of enclitic possessives, described in § 3.2.2, also had
the effect of extracting all clitics from internal position: the only word clitics
208 Silvia Luraghi

which were allowed in an internal position in New Hittite were the coordinat­
ing conjunction -(y)a-, and the particle -pat, a focalizer. The loss of sentence
internal clitics goes contrary to the development shown in Greek, where the
genitive of unstressed pronouns displayed an early tendency toward being
placed close to its nominal head. The development in possessive placing is a
hint to the overall difference in the evolution of placement rule for second
position clitics in Greek and Hittite. In spite of diverging developments, both
languages were able to exploit the occurrence of second position clitics for
pragmatic purposes. Only in Hittite, given the increasing frequency of such
clitics, the difference between narrative non-emphatic sentences and adversa­
tive sentences, which conveyed unexpected information, became strictly de­
fined by the position of the clitic chain.15

NOTES

1. The word 'postpositive' is used , especially in Greek Linguistics (cf. Dover, 1960), for
items that can never occur in sentence initial position. Already Wackernagel (1892)
observed that some of the particles regularly placed in second position in Greek and Old
Indic carried a graphic accent, and supposed that they were weakly stressed. The contrary
of 'postpositive' is 'prepositive', used here for particles that always occur in sentence
initial position. Again, the word 'prepositive' has no implications on the prosodic nature
of particular particles, i.e. it does not imply that they are proclitic, although many
prepositives in fact are.

2. According to Zwicky (1977) special clitics are unaccented forms that do not share the
same distribution as corresponding accented forms (i.e. Romance clitics are special
clitics; English unstressed personal pronouns are simply unaccented variants of accented
pronouns, but they are not special clitics).
3. The most significant exception to this regularity is constituted by the Greek modal
particle án, which displays a tendency to stand closer to the verb; see Marshall (1987).
4. The clitic -kan is a local particle; its meaning and placement rules will be discussed
below, § 3.2.3.
5. In Homer, where the demonstrative ho, he, tó had not developed into a determinative
article yet, placing of all enclitics, including pronouns, after initial demonstratives is
frequent; such pattern also occurs in Herodotus, where the article was well established as
such, but it is avoided in Attic for enclitic pronouns, while a number of other post-
positives, such as the discourse particles gár, mén, dé, etc., are regularly placed between
the article and the noun (cf. Dover, 1960: 16).
6. Marshall (1987) lists a few exceptional cases in which the enclitic is hosted by a word or
constituent on the right of the finite verb, as in lége td psé:phisma moi, 'tell (lége) me
The grammaticalization of the left sentence boundary in Hittite 209

{moi) what has been deliberated', Dem. 18.180.

7. The particle man can sometimes co-occur with the connective -ma-.
8. In the plural dative enclitic pronouns normally precede possible nominative or accusative
enclitics.
9. The Hittite language is usually divided into three periods, Old, Middle, and New Hittite. It
must be remarked that the most important language change occurred at the end of the Old
Hittite period, so that the difference between Old and Middle Hittite is much bigger than
the difference between Middle and New Hittite.
10. On the grammaticalization of -wa(r)-, see Luraghi (1996); on its etymology and historical
development, see Pecora (1984), and on its use with verbs of saying see Cotticelli (1994).
11. See Josephson (1995).
12. Garret (1990: 130-133) gives a full list of passages where intransitive verbs occur with
NS. Beside the Old Hittite examples, that come from all text types, he also gives some
Middle Hittite examples, all coming from the same text (a protocol for the royal guard),
and some Late Hittite examples from copies of Old Hittite ritual texts.
13. Local adverbs originated from grammaticalized nominal forms, which explains the
possibility for them to take enclitic possessives, as well as genitive modifiers. Neither
enclitic possessives nor nouns in the genitive occur with local adverbs in New Hittite.
14. Hittite was a rather strict SOV language, see Luraghi (1990b) and (1997).
15. Obviously, it needs to be remarked that this comparison brings together two languages
whose attested history covered quite different time spans: about three millennia for
Greek, and less than five hundred years for Hittite. No one could say what would have
happened to Hittite, if its history had had a length comparable to Greek.

REFERENCES

Cotticelli-Kurras, Paola. 1995. "Hethitische Konstruktionen mit verba dicendi und

sentiendV In Onofrio Carruba, Mauro Giorgieri e Clelia Mora (a cura di). 1995, Atti
del ii Congresso Internazionale di Hittitologia, 87-100. Pavia: Iuculano.
Dover, Kenneth J. 1960. Greek Word Order. Cambridge: Cambridge UP.
Friedrich, Johannes. 1959. Die hethitischen Gesetze. Leiden: Brill.
Garrett, Andrew. 1990. The Syntax of Anetolian Pronominal Clitics. Ph.D. Dissertation,
Harvard University.
Goetze, Albrecht. 1927. Madduwatta. Leipzig: Hinrichs.
. 1933. Die Annalen des Mursilis. Leipzig: Hinrichs.
Josephson, Folke. 1995. "Directionality in Hittite." Analecta Indoeuropaea Cracoviensia
2: 165-176.
Luraghi, Silvia. 1990a. "Osservazioni sulla Legge di Wackernagel e la posizione del
verbo nelle lingue indoeuropee." In Maria-Elisabeth Conte, Anna Giacalone Ramat e
Paolo Ramat (a cura di). Dimensioni delia Linguistica, 31-60. Milano: Franco Angeli.
210 Silvia Luraghi

. 1990b. Old Hittite Sentence Structure. London and New York: Routledge.
. 1996a. "Processi di grammaticalizzazione in ittita".Archivio Glottologico Italiano
81/1:45-75.
. 1997. Hittite. München, LINCOM Europa.
Marshall, M.H.B. 1987. Verbs, Nouns and Postpositives in Attic Prose. Edinburgh: Scottish
Academic Press.
Neu, Erich. 1993. "Zu den hethitischen Ortspartikeln." Linguistica 33: 137-152.
Otten, Heinrich. 1973. Eine althethitische Erzählung um die Stadt Zalpa. Wiesbaden:
Harrassowitz.
. 1981. Die Apologie Hattusilis III. Wiesbaden: Harrassowitz.
Otten, Heinrich and Vladimir Soucek. 1969. Ein althethitisches Ritual für das Königspaar.
Wiesbaden: Harrassowitz.
Pecora, Laura. 1984. "La particella -wa(r)- e il discorso diretto in antico eteo." Indo­
germanische Forschungen 89: 104-124.
Renzi, Lorenzo. 1989. "Two types of clitics in natural languages." Rivista di Linguistica 1/
1: 355-372.
Rosenkranz, Bernhard. 1979. "Archaismen im Hethitischen." In Erich Neu und Wolfgang
Meid (eds), Hethitisch und Indogermanisch, 219-229. Innsbruck: Institut für Sprach­
wissenschaft der Universität Innsbruck.
Steele, Suzanne. 1977. "Clisis and diachrony." In Charles Li (ed.), Mechanisms of Syntactic
Change, 539-579. Austin: Texas UP.
Wackernagel, Jacob. 1892. "Über ein Gesetz der indogermanischen Wortstellung." Indo­
germanische Forschungen 1: 333-435.
Zwicky, Arnold. 1977. On Clitics. Bloomington: Indiana UP.
 On the relationships between
grammaticalization and lexicalization*

Juan C. Moreno Cabrera

Autonomous University of Madrid

1. Introduction

In this paper I will investigate the interrelations between grammaticalization

and lexicalization. It will be argued that the two processes are complementary
and can be semantically characterized by using one single conceptual hierar­
chy. Morever it will be shown that, from a semantic point of view, grammati­
calization is a metaphorical process and lexicalization is a metonymical
process.
In section 2 I will briefly characterize grammaticalization and review
some of the proposals concerning its study. In section 3 I will sketch an
analysis of lexicalization based on the idea that this process is complementary
with respect to grammaticalization. Section 4 will be devoted to showing how
grammaticalization feeds lexicalization. Three concrete examples will be
briefly presented: they will concern the Spanish nouns in -nte, the Hungarian
nouns in -ó and the Basque nouns ending in ko. In all the three cases
grammaticalized syntactic units are shown to undergo a lexicalization pro­
cess. In section 4, I will consider the consequences of the points made in the
preceding sections for the understanding of the unidirectionality hypothesis

* This is a revised, expanded and improved version of the paper I delivered at the Innaugural
Conference of the ALT (Vitoria-Gasteiz, Spain, September 7-10, 1995). Thanks are due to A.
King, W. Bisang and E. A. Moravcsik for useful comments on that paper. I wish also to thank the
editors of this volume and P. Ramat for their valuable and constructive criticism.
The following abbreviations will be used: ACC(usative), ADL(ative), ART(icle), FEM(inine),
GEN(itive), MASC(uline), PART(iciple), l(rst person),SG (singular), PL (plural).
212 Juan C. Moreno Cabrera

as well as of the interactions between metaphor and metonymy in language

evolution.

2. On the nature of grammaticalization

Current research on grammaticalization focuses mainly on the process

whereby lexical items develop grammatical meanings. Many important
works in the field are entirely devoted to describing how such a process takes
place and determining its general laws and principles1. In this process, we
obtain grammatical elements out of lexical items. Adpositions and mor­
phemes are typical cases of these grammatical elements.
In many languages of the world certain lexical items tend to become
adpositions conveying a particular grammatical function. Let me illustrate
this point with the following examples. A 'reach'-verb is used in several
African languages as an adposition conveying the grammatical functions
PLACE a., MALEFACTIVE b. or as a modal auxiliary c. (Heine and Reh 1984:
180-181).
(1) Possible grammaticalizations of a reach-verb in three African
languages
a. Zande
Mi nekandu da diyo
I go reach the river
'I am going as far as the river'
b. Ewe
Wo- no aha dé -m
3PL drink palmwine reach 1SG.OBJ.
They drank palmwine at my expense'
c. Kono
à éé cé à má-á
3sG. NEG reach 3SG. do-AUX
'He is not able to do it'
As Greenberg suggested in a series of important papers (Greenberg
1978, 1981 and 1991) grammaticalization process can proceed still further
producing noun-class, gender or classifier systems. For example, Greenberg
reconstructs a root *-kwa for Proto-Chibchan-Paezan denoting some round
On the relationships between grammaticalization and lexicalization 213

object (cfr. Terraba gwa 'egg'). This root becomes a numeral classifier in
languages such as Cuna; this classifier is originally used with nouns denoting
round objects, but as Greenberg notes, it is spreading to other noun types
(such as those denoting people2).
As Greenberg himself observed, this type of development gives rise to
complex agreement phenomena such as those of the Bantu languages, whose
agreement morphemes come from demonstratives by means of a grammati­
calization process (see Greenberg 1978: 75-78).
From the preceding considerations, it follows that grammaticalization
processes depart from the lexicon and proceed towards the syntax. This
means that they enrich the syntax by enlarging the syntactic contexts in which
a particular word can appear and by creating new syntactic constituents
(adpositional phrases, serial verb constructions, auxiliaries) and rules (con­
cord rules, for instance). For this reason I characterize them as syntactotelic or
syntax-creating processes.
From the semantic point of view, grammaticalization processes have
been claimed to be processes of metaphorical abstraction. They are said to be
constrained by the following hierarchy (see Heine, Claudi and Hünnemeyer
1991b:157).
(2) Metaphorical Abstraction Hierarchy (Heine, Claudi, Hünnemeyer)
PERSON > OBJECT > PROCESS > SPACE > TIME > QUALITY

I interpret this hierarchy in the sense that lexical items having their denotation
in the conceptual domain PERSON can develop by metaphorical abstraction a
new meaning in one of the domains at its right. The same applies to the rest of
the conceptual domains included in the hierarchy. An impressive example is
provided by Heine, Claudi and Hünnemeyer (1991b: 161). In the African
language Ewe the word megbé denoting a body part (the back), can be used to
denote SPACE (behind), TIME (after) and QUALITY (backward). This type of
metaphorical abstraction is also common in European languages; think for
example of English back that, in a similar way, can be used to convey PLACE
(back of), TIME (some months back) and QUALITY (backwardness). In Spanish,
pie 'foot' has developed a PLACE meaning as in al pie de la colina 'at the foot
of the hill', and a QUALITY meaning as in entrar con buen pie 'to get off to a
good start' or nacer de pie 'be born lucky'. In Basque begi 'eye' develops a
PROCESS meaning as in begira 'looking' and a QUALITY meaning as in begiko
'pleasant' or begiragarri 'worth seeing'.
214 Juan C. Moreno Cabrera

To sum up, let me list the main properties of grammaticalization.

(3) Properties of grammaticalization
a. It is a syntactotelic process (it goes from the lexicon to the
syntax)
b. It affects lexical items (it is a lexicogenetic process)
c. It abides by the Metaphorical Abstraction Hierarchy
d. It feeds the syntax and bleeds the lexicon

3. On the nature of lexicalization

By lexicalization I mean the process creating lexical items out of syntactic

units3.
Idioms constitute the best known cases of this process. In essence, idioms
are syntactic constructions that lose their compositionality and acquire a new
idiosyncratic content. They become lexicalized and can be listed in the
lexicon as unanalyzable wholes. In this case, we obtain lexical items out of
syntactic constructions; that is, we are before a lexicotelic process. It is clear,
therefore, that lexicalization goes in exactly the opposite direction in regard to
grammaticalization. If this is so, lexicalization processes should abide by the
very same grammaticalization hierarchy but read in exactly the opposite
direction. Let me explore now this proposal.
In general, lexicalization obtains when a phrase or a syntactically-deter­
mined lexical item4 becomes a full-fledged lexical item in itself. Consider for
example the English word reading. In principle this word is an inflected form
coming from the verb to read; this inflected form is originally used as a noun
modifier but it has been lexicalized and appears in the dictionaries as a noun.
Among other readings it means 'matter read or for reading'. That is, it denotes
a concrete object: the thing or things being read or to be read. It also develops
an abstract-object reading, reading in the sense of one of the possible inter­
pretations of a word, for example. If we say that originally reading had a
PROCESS reading, we can see that in this case this word has developed an
OBJECT meaning. If we look at the hierarchy in (2) we will realize that a
semantic shift towards the left-end of the hierarchy has taken place. It is,
therefore, a lexically-oriented or lexicotelic process.
The same phenomenon is common in other languages. Consider the
following examples from Spanish.
On the relationships between grammaticalization and lexicalization 215

(4) Three Examples of Lexicalization in Spanish

a. Corneta 'bugle'
i) La corneta
ART-FEM bugle
T h e bugle'
ii) El corneta
ART-MASC bugle
'The bugler'
b. Trompeta 'trumpet'
i) La trompeta
ART-FEM trumpet
'The trumpet'
ii) El trompeta
ART-MASC trumpet
'The trumpeter'
c. Espada 'sword'
i) La espada
ART-FEM SWOrd
'The sword'
ii) El espada
ART-MASC sword
'The bullfighter'
All three words are originally feminine nouns denoting a concrete object,
but by a lexicalization process they can denote a person. In the first and
second case it is the person who blows the instrument and in the third case it is
the person who carries a sword; in bullfighting the only one allowed to use a
sword is the bullfighter (and only for killing the bull, of course!). From a
syntactic point of view, the source for this new use of the original feminine
words are elliptical phrases such as el de la corneta, literally 'the (one) of the
bugle', that is 'the one having or blowing the bugle'. Elliptical phrases such as
el bigotes 'the one (man) wearing a moustache' (cfr. los bigotes 'the mous­
tache'; bigotes is plural in Spanish) or la rizos 'the one (female) having curly
(hair)'(cfr. los rizos 'the curls', rizos is plural masculine), are widely used in
colloquial Spanish. A schematization of the elliptical process originating
these phrases follows.
216 Juan C. Moreno Cabrera

(5) Ellipsis in Spanish noun phrases

a. Step one:
ART+NOUN+PREPOSITION+ART+NOUN
el+hombre+de+los+bigotes
the+man+of+the+moustache
b. Step two: elision of the main noun
ART+PREPOSITION+ART+NOUN
el+de+los+bigotes
the+of+the+moustache
c. Step three: elision of the dependent noun article
ART+PREPOSITION+NOUN
*el+de+bigotes
the+of+moustache
d. Step four: elision of the preposition
ART+NOUN
el+bigotes
ART-MASC-Sg+NOUN+MASC+PL
the+moustache
'the one wearing a moustache'
The source construction of step one consists of a main noun phrase {el hombre
'the man') and a dependent noun phrase {los bigotes 'the moustache'). It is
the dependent noun phrase that bears the semantic burden of the whole phrase
since it denotes the most salient feature of the person referred to. In step two,
the main noun is deleted, since it has no identificational information, but the
element conveying reference, gender and number (the article) is left. In step
three, the determiner of the dependent noun is deleted since bigotes 'mous­
tache' does not function as an identifying noun; it is instead a characterizing
noun. In step four, the preposition is deleted since it is no longer needed.
In all these cases, a semantic shift from OBJECT to PERSON has been
carried out. Again, this is a leftward movement in the grammaticalization
hierarchy as given in (2). So I propose that lexicalization as defined here
abides by the following hierarchy.
(6) Lexicalization hierarchy
QUALITY > TIME > SPACE > PROCESS > OBJECT > PERSON

If the grammaticalization hierarchy tries to account for metaphorical

abstraction processes, the lexicalization hierarchy reflects metonymical con-
On the relationships between grammaticalization and lexicalization 217

cretion processes. Consider as an example the lexicalized interpretation of

reading: from denoting an abstract entity (a process) it moves to a concrete
object denotation. This is not a metaphorical process, but a metonymical one.
In every act of reading there is something contiguous: the thing being read.
Contiguity is the leading principle of metonymy, as similarity is that of
metaphor. In the Spanish lexicalization examples, the semantic shift from an
object to the person who manipulates it is also an example of metonymical
semantic development.
We can therefore conclude that the hierarchy in (6) is a metonymie
concretion hierarchy regulating lexicalization processes.
Lexicalization proceeds from syntax towards the lexicon. The source
units for lexicalization are not lexical items but syntactically-determined
words or phrases.
It is very common that phrases and even whole sentences lexicalize in the
world's languages. As an example, consider the following names for a flower
species in different European languages.
(7) Some European Denominations for myosotis palustris
a. English
forget-me-not
b. Spanish
nomeolvides (not-me-forget)
c. German
Vergissmeinnicht (forget-me-not)
d. Dutch
vergeet-mij-nietje (forget-me-not)
e. Russian
nezabudka (not-forget)
f. Hungarian
nefelejcs (not-forget)
In all these cases a syntactic constructions becomes lexicalized as an
independent word denoting an object, so we also obtain here a concretion
process. It is easy to see that this process cannot be thought of as a metaphor;
it is indeed a metonymy what is involved here. As it is known, flowers are
involved in human relations as tokens of love or devotion and for that reason
they can symbolize the wishes and desires they are intended to convey.
As a summary, let me list the more salient properties of lexicalization
processes.
218 Juan C. Moreno Cabrera

(8) Properties of lexicalization

a. It is a lexicotelic process (it goes from syntax to the lexicon).
b. It affects syntactically-determined words and phrases or
sentences (it is a syntactogenetic process)
c. It abides by the Metonymical Concretion Hierarchy
d. It feeds the lexicon and bleeds the syntax

4. From grammaticalization to lexicalization: three examples

In this section I want to show how grammaticalization and lexicalization

interact with each other. It will be argued that grammaticalization processes
can feed many lexicalization developments. In order to demonstrate this point
I will briefly discuss three lexicalization phenomena. First I will examine the
Romance present participles in -ent(e), then I will investigate the Hungarian
affix -ó and finally I will analyse the Basque suffix -ko.
In Latin, a present participle in -ens was widely used in an adjectival
function. Let me list a couple of examples (taken from Woodcock
1959:72,74):
(9) The use of the Latin present participle
a. Plato scribens est mortuus
write-PART is dead
'Plato died while writing'
b. Video puerum currentern
See-1SG child-ACC run-PART-ACC
'I see the boy running'
In both cases, the participle is modifying a noun {Plato and puerum, respec­
tively). Both participles come from an action verb and both of them denote
that action as a property or as a quality. From my point of view, it is clear that
a semantic shift in the metaphorical abstraction hierarchy has taken place. We
obtain a QUALITY meaning out of a PROCESS meaning. This means that the
morphological process turning a verb into an adjectival modifier is a
grammaticalization process. This process reaches the right-most end of the
Metaphorical Abstraction hierarchy. But this does not mean that the evolution
of present participles stops here. On the contrary, in Latin, the tendency exists
to use present participles as nouns denoting the agents of the actions formerly
viewed as qualities. For example, in the following Latin sentence from Caesar
On the relationships between grammaticalization and lexicalization 219

(Woodcock 1959:80) timentes confirmat 'He encourages the fearful', timen­

tes is the present participle of timeo 'to fear' and here it does not denote the
quality of being fearful but the persons who are fearful. Therefore, a semantic
shift from QUALITY to PERSON has taken place. Recall that I maintained in the
preceding section that this type of semantic shift concerns a lexicalization
process. What we would expect is that a strong tendency towards obtaining
nouns out of participles is to be observed in the evolution of Latin. And this is
exactly what happened in the evolution from Latin to the modern Romance
languages. In Spanish for example, the affix -nte no longer signals a parti­
ciple. It is used to form adjectives and nouns. Indeed there are many nouns
ending in -nte that are listed in the dictionaries as full-fledged nouns. Some of
them are not commonly used as adjectives. Others cannot be used at all in that
function. I give a list of some of them.
(10) Some Spanish -nte nouns
a. Calmante (> calmar 'to soothe')
'sedative'
b. Dibujante (> dibujar 'to draw')
'draughtsman'
c. Cantante (> cantar 'to sing')
'singer'
d. Intermitente (> Lat. intermittens)
'flashing light'
e. Viajante (> viajar 'to travel')
'salesman'
f. Conferenciante (> conferenciar 'to confer')
'lecturer'
g. Navegante (> navegar 'to sail')
'navigator'
h. Presidente (> presidir 'to preside over')
'president', 'chairman'
i. Tripulante (> tripular 'man', 'drive')
'crewman'
j . Amante (> amar 'to love')
'lover'
From this list, only intermitente, calmante and amante can be used freely as
adjectives as well. The rest of them can only be used as nouns. In the case of
calmante and intermitente we have a QUALITY > OBJECT semantic shift in
220 Juan C. Moreno Cabrera

accordance with the Metonymic Concretion Hierarchy. The rest of the cases
exemplify a QUALITY > PERSON move.
Let me turn now to the case of the Hungarian suffix -ó. This suffix, that
can show up as -S or -ó depending on vowel harmony, is described as a
present participle suffix5. Two examples are included in (11) (Tompa
1972:146).
(11) Hungarian -ó present participles
a. A bejelent-ó' személy
ART announce-PART person
'The announced person'
b. A hiányz-ó iv
ART lack-PART arch
T h e lacking arch'
As in the Latin case, the participles have an adjectival function. This means
that the -ó suffix is a grammaticalizing morpheme. Many Hungarian -ó
participles become full-fledged nouns denoting the person involved in the
action implied in the participial form exactly in the same way as in Spanish.
Some illustrative examples are given in (12) (Bencédy, Fabián, Rácz and
Velcsov 1985:38).
(12) Some Hungarian -ó nouns
a. Igazgató (> igazgat 'to direct')
'director'
b. Költö (> költ 'to compose')
'poet'
c. Hallgató (> hallgat 'to listen to')
'student'
d. Elárusító (> elárusít 'to sell')
'employee', 'clerk'
e. Ásó (> ás 'to dig')
'spade'
f. Vasoló (> vasal 'to iron')
'iron'
g. Seprö (> seper 'to sweep')
'broom'
h. Ternetö (> temet 'to bury')
'cemetery'
On the relationships between grammaticalization and lexicalization 221

i. Ebédl'ó (> ebédel 'to lunch)

'dining room'
j. Haló (> hal 'to spend the night')
'bedroom'
In (12a-d) there is a semantic shift from QUALITY to PERSON; in (12e-g) a move
from QUALITY to OBJECT is observed and, finally, in (12h-j) a shift from
QUALITY to PLACE is evident. These shifts are conceived of here as
lexicalizations and they not only abide by the Metonymic Concretion Hierar­
chy but also yield full-fledged lexical items listed as such in the dictionary.
That is the final result of a lexicalization process. As indicated in (12) all these
nouns come from a grammaticalized form (the participial form) of a verb.
The Spanish and Hungarian data clearly show that grammaticalization
feeds lexicalization. It could be argued that the similarity between Spanish
and Hungarian in the evolution of present participles is due to the fact that the
same grammatical category is involved in both languages. Similar grammati­
cal categories are expected to develop in a similar way and therefore the
conjunction of the Spanish and Hungarian data does not appear to strongly
support the claim that grammaticalization feeds lexicalization.
In order to give more evidence for my claim, I will now present some
Basque data showing exactly the same type of interaction between the two
processes. These data have nothing to do with participles and verbs; in spite of
that, they show exactly the same developments observed in the Spanish and
Hungarian cases.
In Basque there is a suffix -ko with several uses. It can function as a
locative genitive as in:
(13) Gernika-ko arbol-a
Gernika-GEN tree-ART
'The tree of Gernika'
But it has another important function. It is in fact an index of grammaticaliza­
tion. It can be added to a number of different constructions (words and
phrases) in order to obtain a word or phrase denoting QUALITY. It is a general
adjectivalizer. Some examples follow:
(14) The Basque Adjectivalizing -ko suffix6
a. aldizkari-entza-ko diru-kopurua
j ournal-for-of money-quantity
'the amount of money for the journals'
222 Juan C. Moreno Cabrera

b. ordurarte-ko kompromisuak
now-until-of commitments
'the commitments until now'
c. ikerketa- ra- ko bide berriak
investigation-ADL-of way news
'New ways of investigation'
d. Komunikabideeta-ko askatasun eta aniztasuna-ren
means of communication-ko freedom and pluralism-GEN
alde-ko manifestu-a
in favor-ko manifesto-ART
'Manifesto in favor of the liberty and pluralism in the means of
communication'
The suffix -ko modifies aldizkarientzat 'for the journals', ordurarte 'until
now', ikerketara 'towards investigation' and kumikabideetako askatasun eta
aniztasunaren alde 'in favor of the liberty and pluralism in the means of
communication', respectively. The semantic shifts the suffix conveys in the
above examples are: DESTINATION > QUALITY (case 14a), TIME > QUALITY
(14b), PLACE > QUALITY (14c) and BENEFAcnvE > QUALITY (14d). All these
changes are cases of metaphorical abstraction and can therefore be conceived
of as grammaticalization processes.
But in Basque a strong tendency toward the lexicalization of ko words
can be observed by simply browsing a Basque dictionary. The following is a
minimal list that could be easily expanded.
(15) Some Basque -ko nouns
a. QUALITY > PERSON
Aurretiko 'guide' (lit. 'the one going ahead')
Etxekoak 'family' (lit. 'those of home')
Gerokoak 'descendants' (lit. 'those after')
Bitarteko 'mediator' (lit. 'the one of the space between two')
Mendeko 'employee' (lit. 'that subjected')
Oinezko 'walker' (lit. 'that of with the feet')
b. QUALITY > OBJECT
Buruko 'pillow' (lit. 'that of the head')
Gerriko 'belt' (lit. 'that of the waist')
Gerripeko 'loincloth' (lit. 'that of under the waist')
Soineko 'dress' (lit. 'that of the body')
On the relationships between grammaticalization and lexicalization 223

Belarriko 'headphones' (lit. 'that of the ear')

Bularretako 'brassière' (lit. 'that of the breasts')
C. QUALITY > PROCESS
Hileroko 'menstruation' (lit. 'that of every month')
d. QUALITY > TIME
Geroko 'future, result' (lit. 'that afterwards')
All of these words are listed in Basque dictionaries as such but clearly
come from other original lexical items inflected for case and provided with
the -ko suffix. For example oinezko 'walker', comes from oinez; that is the
instrumental case form for oin 'foot', plus the adjectivizing suffix -ko. This is
not the only new noun oin gives rise to. In Basque the word for shoe is
oinetako, also coming from oin, in this case inflected for plural number.
The following diagram illustrates how grammaticalization feeds lexical­
ization in this example.
(16) An example of grammaticalization feeding lexicalization in Basque7

We see again that a strong tendency exists for grammaticalization processes to

feed lexicalization processes, and that there is a close interaction between both
procedures. This interaction is crucial for having a better understanding of the
evolutionary dynamics of lexical and grammatical elements.

5. Conclusion: Unidirectionality, Reversibility, Metaphor and

Metonymy in the evolution of grammatical and lexical elements

In this paper I have tried to demonstrate that grammaticalization and lexical­

ization processes are two complementary aspects of essentially one single
type of evolutionary dynamics for grammatical and lexical items. From this it
224 Juan C. Moreno Cabrera

follows that exactly the same hierarchy that some scholars have proposed for
explaining how grammaticalization takes place can be succesfully applied to
the opposed and in some sense contradictory process of lexicalization.
This conclusion is highly relevant for the so-called unidirectionality
property of grammaticalization. In fact, it is claimed that the unidirectionality
hypothesis is one of the major axioms of grammaticalization theory8. It is
claimed that when a lexical item grammaticalizes as a morpheme it is not in
general possible for this morpheme to de-grammaticalize into a lexical item.
Only a few exceptions to this tendency have been pointed out in the literature
(Ramat 1992, Hopper and Traugott 1993: 126-128). I think that it would be
much better to characterize this process as irreversible. In general it can be
said that the grammaticalization of lexical elements is not reversible. The
directionality issue should not be confined to grammaticalization only; it
should be judged relevant to the evolution of grammar in general. If that
evolution were unidirectional and we considered only grammaticalization we
would expect languages to become more and more grammaticalized. But that
is not confirmed by the facts. Language evolution is, on the contrary, bi­
directional and comprises both grammaticalization and lexicalization. In
language change there is a constant movement from the lexicon to the syntax
and the other way around. We do not observe languages gradually losing their
lexicon and enriching their morphology and syntax. Nor do we observe
languages gradually increasing their lexicon and losing their morphology and
syntax. This means that language evolution is not exclusively a process of
grammaticalization or lexicalization. Only the interaction of the two pro­
cesses can produce the balanced results we observe in language evolution9.
One of the suggestions of this paper is that grammaticalization processes
are predominantly metaphorical and that lexicalization processes are mostly
metonymical. Metaphor has also been much discussed in relation to grammati­
calization (Heine, Claudi and Hunnemeyer 1991:45-65 and 70-98). Heine,
Claudi and Hiinnemeyer (1991:70-78) try to demonstrate that both metonymy
and metaphor are involved in grammaticalization. Indeed they say that gram­
maticalization itself has a metonymical component. The metonymic part of
metaphorical change, they argue, comes from the linguistic context inviting a
particular conceptual interpretation (Heine, Claudi and Hiinnemeyer 1991:
72). But I think that we should distinguish the metaphorical change in itself from
the conceptual mechanisms whereby a metaphor can be created; some of those
mechanisms could be metonymical, but what we obtain is a metaphor, not a
On the relationships between grammaticalization and lexicalization 225

metonymy. Grammaticalization should be therefore characterized as a meta­

phorical process.
I have claimed in this paper that metonymy is involved in lexicalization.
Since grammaticalization and lexicalization are two complementary prin­
ciples of language evolution, they are guided by two complementary and
equally important cognitive strategies: metaphor and metonymy.
One question arising inmediately is why grammaticalization and lexical­
ization are metaphorical and metonymical processes respectively. Let my try
to give a tentative answer, to conclude the paper.
It is clear that the path from the lexicon to the grammar, which character­
izes grammaticalization from my point of view, is highly abstractive. Gram­
matical categories are very abstract entities, since they can be used to convey
a wide range of different entities or situations. For this to be possible, one or
more common conceptual features must be assumed to characterize all those
entities or situations. For example, in English John is a man, John is tall and
John is in the classroom denote three different situations but they are con­
veyed by the same type of construction (a copulative sentence): the common
thing between the three situations is that they are states. In order to create a
grammatical category state, a bundle of features common to these and many
more different situations must be postulated. This is an abstraction process in
which the similarities between a great amount of different things must be
discovered, discarding those specific differences distinguishing all these
things from each other.
Abstraction and similarity are precisely the two basic operations giving
rise to metaphorical thinking. In metaphors, two or more different objects are
seen as identical: this is possible because we discard the features distinguish­
ing them and focus on those features which they share. For example, the foot
of a man and the foot of a hill are different entities. Nevertheless, we can use
the same word for denoting both things. For this to be possible we must first
discard all those properties distinguishing both entitities and focus only on
those features which they share. As a consequence, it is no surprise that
grammaticalization should be a metaphorical process.
I have maintained in section 3 that the lexicalization path proceeds from
the grammar to the lexicon. The input units for lexicalization are phrases or
syntactically-determined words. These elements will be highly context-de­
pendent and therefore concretion and contiguity, being two of the main
characteristics of context-dependent elements, will play the starring role in
226 Juan C. Moreno Cabrera

lexicalization processes. It is well known that both factors characterize

metonymic reasoning, and as a consequence the lexicalization process will be
essentially metonymic.
I think that an understanding of the dialectical relationships between
grammaticalization and lexicalization can shed new light on the interactions
of metaphorical and metonymical processes in language evolution.

NOTES

1. See Heine and Reh (1984), Heine, Claudi and Hünnemeyer (1991a,b), Lehmann (1982)
and Hopper and Traugott (1993).
2. Greenberg (1991:310) cites the following example from Cuna:
tule pakke-kwa
person four-CLASS
four people
3. In the literature, lexicalization processes have been studied as degrammaticalization
processes; cfr. Ramat 1992
4. By syntactically-determined lexical item I mean an inflected lexical item or a lexical item
affected by an adposition or grammatical particle. That is, a lexical item playing a
particular syntactic role
5. Cfr. Tompa 1972:146, who uses Präsenspartizip for naming this suffix and Bencédy,
Fabián, Rácz and Velcsov 1985:36 who use the expression melléknévi igenév (adjectival
verbal noun) for referring to verbal forms presenting this -Ó/-Ö ending
6. The examples appear in a widely known Basque cultural journal (Jakin 1994, 83:13,14,30
and 68).
7. A similar diagram for various Indo-European languages can be found in Ramat 1992: 555
8. Hopper and Traugott devote an entire chapter to the hypothesis of unidirectionality in
their book on grammaticalization (1993: 94-129).
9. This was already noted by the eminent Polish linguist J. Kuryłowicz (1965), who
maintained that an interaction exists between grammaticalization and lexicalization in the
evolution of grammatical categories

REFERENCES

Abraham, Werner. (ed.) 1992. "Grammatikalisierung und Reanalyse: Konfrontation"

Folia Linguistica Historica, XIII no. 1-2,
Bencédy, József, Pál Fábián, Endre Rácz and Mártonné Velcsov. 1985. A Mai Magyar
Nyelv, Budapest: Tankönyvkiadó.
On the relationships between grammaticalization and lexicalization 227

Greenberg, Joseph. 1978. "How does a language acquire gender markers?" In J.

Greenberg (ed.), Universals of Human Language. Vol 3. Word Structure, 47-82.
Stanford: Stanford University Press.
. 1981. "Nilo-Saharian movable -k as a stage III article (with a Penutian parallel)".
Journal of African Languages and Linguistics 3:105-112.
. 1991. "The last stages of grammatical elements: contractive and expansive
desemanticization." In E. Traugott and B. Heine (eds), vol 1: 301-314.
Heine, B. and M. Reh. 1984. Grammaticalization and Reanalysis in African Languages.
Hamburg: Helmut Buske.
Heine, Bernd, Ulrike Claudi and Friederike Hünnemeyer. 1991a. Grammaticalization. A
Conceptual Framework. Chicago: The University of Chicago Press.
. 1991b. "From cognition to grammar - evidence from African languages." In E.C.
Traugott and B. Heine (eds) 1991, vol 1: 149-188.
Heine, Bernd et al. (eds). 1993. Conceptual Shift. A lexicon of grammaticalization in
African languages. Köln: Afrikanistische Arbeitspapiere.
Hopper, Paul J. and Elizabeth C. Traugott. 1991. Grammaticalization. Cambridge:
Cambridge University Press.
Lehmann, Christian. 1992. Thoughts on Grammaticalization. LINCOM Studies in Theo­
retical Linguistics. Munich: LINCOM EUROPA.
Kurylowicz, Jerzy. 1965. "The Evolution of Grammatical Categories." In Jerzy
Kuryłowicz, Esquisses Linguistiques II, 38-45. Munich: Fink.
Ramat, Paolo. 1992. "Thoughts on degrammaticalization." Linguistics 30:549-560.
Tompa, József. 1972. Kleine Ungarische Grammatik. Leipzig:VEB
Traugott, Elizabeth C. and Bernd Heine (eds). 1991. Approaches to Grammaticalization,
2 vols. Amsterdam/Philadelphia: John Benjamins.
Woodcock, E.C. 1959. A New Latin Syntax. Bristol: Bristol Classical Press.
 Structural scope expansion and
grammaticalization

Whitney Tabor and Elizabeth Closs Traugott

Cornell University and Stanford University

1. Introduction

Unidirectionality in grammatical change is a tantalizing idea that has both

substantial empirical plausibility and appealing theoretical simplicity, but it
lacks tooth in its current formulation because of a number of hard-to-dismiss
counterexamples.1 The idea comes in several forms which correspond roughly
to different aspects of linguistic analysis: it is claimed that:
1. Grammatical change persistently involves a shift from more referential to
less referential meanings (semantics/pragmatics) (e.g., Traugott 1982;
Traugott and König 1991; Dasher 1995),
2. The meanings of grammatically metamorphosing elements tend to
become more, rather than less, abstract (semantics) (e.g., Sweetser 1990;
Heine, Claudi and Hünnemeyer 1991; Bybee, Perkins and Pagliuca
1994; Pagliuca 1994),
3. Their phonetic substance tends to be reduced (phonetic form) (e.g.,
Heine and Reh 1984; Bybee, Perkins and Pagliuca 1994),
4. Their frequencies tend to increase (statistical form) (e.g., Bybee 1985;
Givón 1991),
5. Grammatical change proceeds across a cline of structural types (phrasal/
morphemic form), which Givón formulated as:
(1) discourse > syntax > morphology > morphophonemics > zero
(Givón 1979: 209)
230 Whitney Tabor and Elizabeth Closs Traugott

and Lehmann (1995[1982]) has described in the following terms:

[W]e assume that grammaticalization starts from a free collocation of
potentially uninflected lexical words in discourse. This is converted into a
syntactic construction by syntacticization, whereby some of the lexemes
assume grammatical functions so that the construction may be called
analytic. Morphologization, which here means the same as agglutination,
reduces the analytic construction to a synthetic one ... In the next phase, the
unity of the word is tightened, as the morphological technique changes from
agglutinative to flexional. (Lehmann 1995[1982]: 13-14)

Lehmann identifies six "parameters" that typically correlate in processes of

language change identified as instances of grammaticalization. Three are
paradigmatic and will not concern us here, three syntagmatic: including
bondedness, syntagmatic variability, and structural scope. Bondedness is
morphophonological and involves "coalescence" primarily at the level of the
word; it is the subject matter of many major studies, including Bybee (1985),
and Bybee, Perkins and Pagliuca (1994). Syntagmatic variability is character­
ized by Lehmann as a change via "fixation" from a language state in which an
item can be shifted around freely to one in which the item occupies a fixed
slot. Structural scope, which is the major focus of our paper, is characterized
as a change via "condensation" from a language state in which an item relates
to constituents of arbitrary complexity to one in which the item modifies a
word or stem (ibid. 145-178).
Givón's and Lehmann's formulations attempt to make explicit
Langacker's appealingly vivid metaphor: languages "in their diachronic as­
pect" may be regarded as "gigantic expression-compacting machines"
(1977:106). But the question is: Is this is the right metaphor for all aspects of
language? Has a metaphor that applies well at the morphophonological level
been overgeneralized to syntax and thereby created a conceptual dilemma for
the field of grammaticalization studies?
Many of the troublesome challenges to Givón's and Lehmann's claims
that have been reported on so far are somewhat idiosyncratic and primarily
morphological in character, e.g., the English possessive (Janda 1980; Tabor
1992), various other clitics (Jeffers and Zwicky 1980; Joseph and Janda
1994), regional Spanish agreement affix -mos '1st pers. pl.' > nos- 'subject­
marker' (Janda 1995, 1996), Estonian suffix ep > clitic 'affirmative adverb'
(Campbell 1991), "regrammaticalization" of erstwhile articles in Salishan
and Nilo-Saharan languages (Greenberg 1991). Others (not necessarily cited
Structural scope expansion and grammaticalization 231

as counter-examples), however, are more systematic, especially at the syntac­

tic level. These include cases of shifts "from reduced (VP) to nonreduced (IP)
clauses" in the history of Spanish and Portuguese (Davies 1994:48 on
biclausal → monoclausal → biclausal causatives), a change that would be
unexpected given a theory of structural scope reduction (see also Lehmann
1988 on correlated "parameters" of clause linking, among them grammatical­
ization and "desentialization" of the subordinate clauses, e.g., reduction to a
nominalization, that is, to a constituent of the main clause). Another change
that is problematic for scope reduction involves reanalysis of subordinate
clauses as independent (a change known as "insubordination"), which has
been reported in Australian languages (Evans 1988). Recently, a new class of
counterexamples has been explored in some detail: the development of dis­
course markers which are clause-external and often disjunct, like indeed and
in fact out of VP-internal adverbials in English (e.g., Traugott 1995), clause-
external demo 'but, marker of turn taking' out of VP-internal, clause-final -te
mo 'even if in Japanese (Onodera 1995). All these episodes seem to involve
structural scope increase. Nevertheless, they show many other signs of being
episodes of the type of change called "grammaticalization".
One possibility, at this stage, is to reject structural unidirectionality as
criterial for grammaticalization, an approach taken for various reasons by
e.g., Nichols and Timberlake (1991); Hagège (1993); Traugott (1995). An­
other is to conceptualize it as irrelevant because it is not predicted by the
theory within which the author is working — the position of linguists working
in parametric theory, sometimes explicitly, sometimes implicitly (e.g.,
Lightfoot 1979, 1991; Roberts 1993; Warner 1990). In this paper we take a
third position: that the claim of structural unidirectionality, and most espe­
cially scope reduction, deserves to be empirically tested with an explicit
definition of "scope change". We will explore the hypothesis that episodes of
so-called "grammaticalization" persistently involve increase of structural
scope (defined as C-command Scope Increase), not decrease in structural
scope. We will show that this hypothesis is surprisingly robust, given that it
seems to be intuitively opposite to the scope claims that have been prominent
in the literature up to now. Consequently, we argue that the jury is still out on
the status of structural unidirectionality as a criterion for deciding what
change episodes come within the purview of "grammaticalization studies". It
is certainly not appropriate to take scope decrease (without further articula­
tion and demonstration of its validity) to be a criterion. Nor is it appropriate to
232 Whitney Tabor and Elizabeth Closs Traugott

conclude, yet, that there is no structural unidirectionality principle, for rela­

tively little systematic exploration of the formal possibilities has so far been
done.
The paper is organized as follows. Section 2 provides motivations and
definitions of C-command and the method of "diachronic string comparison"
by which scope increase or decrease can be tested. Section 3 considers several
change episodes, showing that the C-command Scope Increase Hypothesis
applies in a surprising number of cases, including some that have been taken
to exemplify scope decrease. Finally, Section 4 reviews the results, and
assesses the implications for the limits of grammaticalization studies. Ex­
amples are primarily from English.
Like earlier hypotheses concerning scope decrease, our hypothesis is
unidirectional. Some serious conceptual questions about the ultimate viability
of unidirectionality claims have been raised by Lass (p.c.). One strong version
of unidirectionality holds that all grammatical formatives originate in lexical
formatives. Lass points out that there is no evidence in Indo-European and
Uralic of a lexical origin for many grammatical items; worse, in apparent
violation of the uniformitarian principle, we would have to postulate proto-
languages of a non-attested sort—ones with lexical items and no function
elements or morphology. Givón (1979) and Bickerton (1984) have hypoth­
esized that just such languages can be found in early creoles, but such creoles
are only postulated, not empirically attested. Our claim, in this paper, how­
ever, does not depend on the existence of such languages. It says only that
when change episodes occur in which an item or collocation changes its
structural status gradually via reanalysis, then the change may involve C-
command Scope Increase.

2. Scope change

To make what we mean by scope increase/decrease very clear, we define

scope contrast in terms of C-command and propose an explicit method
("diachronic string comparison") for comparing scopes across different states
of a language.
Def. 1. Node A C-commands node B if A does not dominate B and every
branching node which dominates A also dominates B (Haegeman
1994[1991], based on Reinhart 1981).
Structural scope expansion and grammaticalization 233

Def. 2. Node A asymmetrically C-commands node B if A C-commands B

but B does not C-command A.
Def. 3. Node X dominates Y if Y is a daughter of X or there is a daughter of
X which dominates Y. If the latter is the case, we say that X remotely
dominates Y.
Def. 4. Node R is a daughter of node Q if there is a phrase-structure rule in
which Q is to the left of the arrow and R is to the right.
It is standard to equate C-command with scope and say that if node A
asymmetrically C-commands node B, then node A has greater scope than
node B. But we immediately run into a problem with vagueness when we try
to extend this interpretation to the diachronic arena. Consider, for example,
the claim that the Present Day English (PDE) modal will has higher scope
than its Old English (OE) ancestor, which meant something like 'want.' One
might try to support such a claim by noting that there is a natural syntactic and
semantic correspondence between OE willan 'to want' and PDE want, and
that we have good reasons to believe that there is a PDE structure along the
lines of (2), in which will has C-command scope over want.
(2) [IP Those people [I' will [VP want many things]]]
But this argument runs aground on examples like
(3) Some people [VP want to believe [CP that [IP nothing [I' will [VP
change]]]]]
in which want has C-command scope over will. Consequently, we propose the
following more constrained notion of what it means to say that the scope of
morpheme X has increased/decreased diachronically. We limit our attention
to cases in which we can replace each word in an older construction with a
descendent from some specific later stage to produce a legitimate sentence
from the later stage. Then, if possible, we align phrase-brackets across the two
diachronically related constructions and examine the changes in any words or
subconstructions which have changed their syntactic class.
As an illustration, consider the following example:
(4) Min Drihten, ic wille gangan to Rome.
My Lord, I want go:INF to Rome
'My lord, I will/want to go to Rome.'
(c. 970 Blickl. Horn. [DOE])
234 Whitney Tabor and Elizabeth Closs Traugott

If we replace each word in this construction, in order, with its modern or early
modern English descendant, we obtain the following diachronic comparison:
(5) a. Min Drihten, ic wille gangan to Rome.
b. My Lord, I will go to Rome.
There have been many analyses of the history of the English modal system
and there is still a question as to exactly which analysis is right. But one
contrast seems to be agreed upon by all historical researchers: the modern
modals occupy an auxiliary position which dominates the verb phrase in
which main verbs are generated, while the OE ancestors of modal verbs were
like main verbs and were presumably generated in the verb phrase. Thus, we
can reasonably make a Diachronic String Comparison (DSC) along the lines
of (6):
(6) Diachronic String Comparison. Main verb to modal verb.
a. [CP Min Drihten [IP ic [I'0 [VP wille [VP gangan [PP to [NP Rome]]] ]]]]
b. [CP My Lord [IP I [I'will [VP go [PP to [NP Rome]] ]]]]

Here, there is a natural correspondence between certain levels of embedding or

shells of matched brackets across these two examples. Given this alignment of
brackets, there is a clear sense in which the position ofwill in (6b) has scope over
the position of wille in (6a).2 Thus, under the most plausible structural analyses
we know of, this case involves increase, not decrease, in the scope of will. On
this basis, we claim that this episode of grammatical change exhibits C-
command Scope Increase. The validity of this claim depends crucially on the
assumption that it is the element wille/will that we should be paying attention to
when we compare scopes — for example, the scopal status of ic T in this
example has not changed under the analyses given. Therefore, to be explicit, we
assume that scope change claims must always be evaluated with respect to an
item that changes its syntactic class status. We also assume that it is not
necessary for the shells corresponding to the changing item in each example to
line up (since it is changing) but we assume that all other shells must line up. We
use the term "diachronic string comparison" to refer to this method of focusing
on contrasting analyses of identically ordered, diachronically related, construc­
tions from different time periods, lining up bracket shells across the analyses,
and comparing the scopes of a reclassified item.
Now we turn to our central point. We test the following hypothesis:
Structural scope expansion and grammaticalization 235

(7) The C-command Scope-Increase Hypothesis: When an item un­

dergoes gradual syntactic reclassification, resulting in a state in
which diachronic string comparison can be applied, then its C-
command Scope increases.
This hypothesis turns out to be surprisingly robust, even though it is probably
not a universal generalization. Thus, we believe it merits further looking into.
We hope that eventually it will become possible to decide, on independent
grounds, which change episodes will exhibit C-command Scope Increase,
which will exhibit decrease, and which will show no change in scope relation­
ships. We see the formulation of a clear testable hypothesis like our C-
command Scope-Increase Hypothesis as an important step in this direction.
Our thesis presupposes a particular view of the "limits of grammaticaliza­
tion". We take the volume title to be posing questions about what methods we
should use to decide which historical change events our theories should
explain, that is: What are the limits on the subject matter of the field of study
which goes by the name of "grammaticalization"? It is clear that we cannot
take structural unidirectionality as a definition of the subject matter, for
structural unidirectionality is a hypothesis which needs to be formalized and
tested before it can be presupposed. Instead, we suggest using the following
correlated hallmarks to identify change episodes of relevance:
1. Morphosyntactic change
2. Pragmatic/Semantic change
3. Gradualness in the sense that some subtypes of a new construction
become possible before others.
These hallmarks are fairly easy to use.3 Even when, as we advocate, these
features are treated as necessary conditions for deciding what episodes our
theories of change are responsible for, these criteria include a wide range of
changes — almost everything, in fact, that researchers have called "grammati­
calization" in the past. We feel it is better to use uncontroversial features such
as these to identify the realm of study. Controversial features, like unidirec­
tionality, should be given explicit formulation and then tested. If someone
succeeds someday in formulating a version of unidirectionality which is
highly correlated with the above, uncontroversial hallmarks, then we would
support including it in the list of field-defining features. But, given our current
results, and the lack of another explicit formulation, we feel unidirectionality
is best treated as a hypothesis, not a defining characteristic.
236 Whitney Tabor and Elizabeth Closs Traugott

Our list of hallmarks includes much, but it also rules out some important
kinds of change which we do not aspire to make claims about. First of all,
coinages, and lexical borrowings, and other "lexicalization" processes are
excluded for they are essentially instantaneous. Thus regular "lexicalization"
processes are also excluded that involve major category shifts such as use of
up as a verb, upper as a noun. If, following the common English pattern, a
speaker innovates the usage to window a process, meaning to open a computer
window for a program to display its output, then the change, here N → V, is
not within the purview of grammaticalization because a productive rule of the
language has been used to shift an element from one major category into
another.
On the other hand, if this new verb window subsequently undergoes a
protracted development (involving the acquisition of special semantic and
morphosyntactic properties and almost a continuum of intermediate phases)
eventually changing its class from verb to auxiliary verb, then the changes
will be in the purview of grammaticalization. We predict that there will be no
such gradual development which changes an auxiliary verb into a main verb.4
In sum, we take the limits of the field of inquiry to be gradual morphosyntac­
tic and semantic change which results in grammatical reanalysis and we
hypothesize that such change brings about an increase, not a decrease, in C-
command scope.

3. Four cases

In this section we discuss four change episodes in the history of English:5 the
development of the -s possessive, the VP-gerund, adverbial and conjunctive
instead (of), and the discourse marker anyway.

3.1 The English possessive

Our first example concerns a morphological change which we will argue is

best explained in terms of a shift in status from case inflection to clitic: the
history of the English possessive -s (see in PDE Celine's oak, the postal
worker's bicycle). As Janda (1980) already pointed out, it is a potential
challenge to Givón's famous slogan "Yesterday's syntax is today's morphol­
ogy" (Givón 1971: 413).
Structural scope expansion and grammaticalization 237

The OE ancestor of the -s was a genitive case marker that occurred on

masculine and neuter a- and /-stems. Several pieces of evidence support the
view that this marker was generated in the morphology of OE: it alternated
with other markers of genitive case (e.g., -a, zero) on a lexical basis (8), all
lexical daughters of NP exhibited concordant marking (9), phrasal modifiers
were always extraposed and thus never intervened between the -s and its host
lexical item (10):
(8) NOM fot vs. bot
GEN fot-es vs. bot-e
'foot' 'remedy'
(9) pœs arwurd-an wer-es gebedrœdden-e
the:GEN honorable:GEN man:GEN prayer.DAT
'to the honorable man's prayer'
(c.1000 Gregory's Dialogs H: 39.12 [DOE])
(10) a. mid Rodbeardes eorl:es fultume of Flandran
with Robert:GEN earl:GEN support of Flanders
'with Robert, Earl of Flanders' help'
(1085 ChronE (Plummer) [DOE])
b. dœs cyning:es sweoster Ecgfrid:es
the:GEN king:GEN sister:NOM Ecgfrid'.GEN
'the sister of Ecgfrid the king'
(c. 1000 Aelfric Horn 11, 10 87, 215 [DOE])
By contrast, in PDE, corresponding evidence supports the view that the
possessive -s is a kind of clitic6: there is no variation across possessive phrases
that is sensitive to the lexical identity of the head noun; no daughters of NP
exhibit specific possessive marking, as the translation of (9) shows; the -s can
appear at the end of the possessor noun phrase even when there are post-head
phrasal modifiers as in (11):
(11) a. the King of Thailand's response
b. a room in the attic's quietude
c. the woman we met yesterday's umbrella
d. * the king's response of Thailand
e. * a room's quietude in the attic
f. the woman's umbrella *(? that) we met yesterday
238 Whitney Tabor and Elizabeth Closs Traugott

These data suggest that, structurally, the -s made the following transition:

The development of the possessive shows all three of the "hallmarks of

grammaticalization" that we identified in the introduction. It clearly shows
syntactic change, as we have argued above. It also shows semantic change,
for the OE genitive (including the -s genitive) was used for many noun-noun
relations that are not possible with the modern possessive (see Mitchell,
1985). Furthermore it shows signs of gradualness in the sense that some
grammatical changes happened before others. Possible evidence consists in
the fact that we see -s marked genitive nouns with uninflected definite articles
in the Peterborough Chronicle c. 1155 — so new forms compatible with the
clitic analysis had already begun to appear at that time, but it is not until two
centuries later, in Chaucer's time, that we find the first unequivocal evidence
for a clitic analysis in the form of a modifier separating the noun from its head:
Structural scope expansion and grammaticalization 239

(13) the god of slepes heyr

'the god of sleep's heir'
(c. 1368 Chaucer, Book of Duchess 168)
There are three such instances in Chaucer's works. Of course, it is possible
that this delay is due to a low overall likelihood of observing post-head
modifiers on genitive noun phrases. We also have evidence for incremental
change in the fact that in Chaucer's works there are still some lexical items
which systematically take a special form — zero marking — in the genitive
singular (e.g., chirche 'church's', lady 'lady's', sonne 'son's', widwe
'widow's'). Such systematic exceptions suggest that the clitic analysis had
not yet become completely general by Chaucer's time.
The structural analyses given above suggest the DSC (14):
(14) Diachronic String Comparison. Lexical genitive to possessive
clitic (XD = "X has dative case"; X G = "X has genitive case").
a. [NPD [NPG [DetG pœs] [NG Nstem wer] es]] [ND [Nstem gebed] e]]
b. [NP [Det[NP [Det the] [N man ]] s] [N prayer ]]

Here, we have aligned brackets corresponding to major class labels, and not
required alignment for the brackets corresponding to the changing item, or to
case-categories. The former move is consistent with our assumptions as laid
out in the beginning. The latter seems reasonable since English has lost case
marking except on pronouns. Given this comparison, we can say that the OE
genitive marker that appears on nouns has narrower scope than the modern
possessive marker. Since this is the only item that has significantly changed its
morphosyntactic class (from being a marker of a subclass of nouns to being a
marker of full noun phrases), we conclude that this episode exhibits C-
command Scope Increase.
Janda (1980) suggests a different analysis of the history of the possessive
clitic. If his analysis is correct, this example may not be problematic for
Lehmann's and Givón's scope-decrease claims after all. Janda notes that
there is evidence for what has been called the "his-genitive" during this
crucial transitional period. The [h] of the pronoun his had lost its aspiration in
unstressed positions (Wyld 1953: 314)) and sometimes forms like the follow­
ing are attested in texts:
(15) a. Modred is hafd
'Modred's head'
(1225 Lay. Brut [MED])
240 Whitney Tabor and Elizabeth Closs Traugott

b. Seynt Dunstone his lore

'Saint Dunstan's lore'
(1420 [Janda (1980: 248])
Janda hypothesizes that the immediate source of the clitic was not the inflec­
tional genitive but the resumptive pronoun, itself a clitic, although the sur­
vival of the possessive marker -s may have been helped along by its
coincidental homophony with the pronoun. However, Tabor (1992) notes that
the development of a similar possessive clitic in Norwegian and Danish
without a homophonous his-genitive makes this hypothesis less convincing
and that the paucity of gender agreement early in the his-genitive period
makes it unlikely that the to-genitive was around early enough to play a role
in the differential erosion of case on genitive noun phrases. In sum, we
believe the English possessive is a problem for the Scope Decrease Hypoth­
esis, but is consistent with the C-command Scope Increase Hypothesis.

3.2. The English gerund

In discussing the grammaticalization scale of verbal nouns (gerunds), Lehmann

cites:
(16) a. John's constantly reading magazines
b. John's constant reading of magazines
c. *the (constantly) reading magazines
d. the constant reading of magazines
(Lehmann 1995[1982]: 62)
and comments "we have two stages of our grammaticalization scale embod­
ied in the English POSS-ing construction. At the latter stage, the nominalized
verb has assumed all the relevant features of a noun; -ing-nominalizations are
even pluralizable" (Lehmann 1995[1982]: 64). While it is not clear whether
Lehmann is making a synchronic or diachronic claim here,7 the diachronic
version of his claim is incorrect: types (16b) and (16d) are historically earlier
than type (16a).
A number of synchronic analysts have noted the pervasive differences
between gerunds (that is -ing words) like those in (16b) and (16d) on the one
hand, which we will call "NP gerunds", and that in (16a), which we will call a
"VP gerund" (see e.g., Lees 1960; Chomsky 1970; Wasow and Roeper 1972;
Jackendoff 1977; Abney 1987). The NP gerunds accommodate only adjectival
Structural scope expansion and grammaticalization 241

modification, they take any kind of nominal quantifier, and, if a direct object is
present, it occurs in a prepositional phrase headed by of. The VP gerunds, on
the other hand, accommodate only adverbial modification, they take only
possessive quantifiers, and if a direct object is present, it occurs immediately
after the verb. The auxiliary verbs have, be[copula], and be[passive] occur in
VP gerund constructions, as does the main verb be[predicate nominal] and the
verbs which take adjectival complements (be[predicate adjective],feel, seem,
look, etc.) None of these special kinds of verbs can occur in the NP gerund
construction. Jackendoff (1977) suggests the following contrasting phrase
structure analyses of these constructions:

He assumes that the -ing marking is base-generated in a position to the left of

the phrase it marks and then jumps over intervening words to be united with
the verb in a process related to Chomsky (1965)'s "Affix Hopping."
Jackendoff notes that we can describe the relationship between the two types
as involving two different instantiations of his Deverbalizing Rule Schema:
Xn → Af Vn, with X = N and n = 0 for the NP gerund and n = 2 for the VP
gerund.
There has been much debate about the manner in which the modern
contrast in gerunds arose (see e.g., Curme 1912; Einenkel 1913; Poutsma
242 Whitney Tabor and Elizabeth Closs Traugott

1923; Visser 1966; Emonds 1971; Tajima 1985; Houston 1989; and a sum­
mary in Fischer 1992), but the following points are agreed on by most: the OE
ancestor of the gerund -ing was an ending -ung (or occasionally -ing) used
mainly in forming feminine abstract nouns from members of the second class
of weak verbs:
(18) abidung 'waiting'
ascung 'interrogation'
blacnung 'pallor'
brocung ' affliction'
handlung 'handling'
niþerung 'humiliation'
wiccung 'witchcraft'
wending 'turning'
(Wright and Wright 1925[1908]: 319)
These nouns behaved like other nouns in all relevant regards; in particular
they showed none of the distinctive properties associated with the VP gerund.
In late OE and eME, the -ung ending spread to all verbs while its spelling
changed to -ing} From earliest eME, the -ing nouns could occur with direct
objects in a prepositional phrase headed by of (19). Then, according to a
tabulation based on a sample of approximately 22,000 pages of eME and ME
texts examined by Tajima (1985), the development proceeded as follows.
Around 1200, the deverbal nouns in -ing began taking adverbial modifiers
(both before and after the noun) (20); such constructions are not attested in
OE. Around 1300 the first instances of immediately following direct objects
appeared (21). Around 1450, the first adjectival complements appeared (22).
A century later, around 1550, the first gerundive forms of be [passive] (23) and
have[perfect] (24) appeared. In this way, the current VP gerund came to exist
along side the NP gerund.
(19) a. pe lichames festing is widtiging of estmetes and
the body's fasting is resisting of delicacies and
over-etes.
excessive eating.
(c. 1200 Trin. Horn. 63/21-22 [Tajima 1985: 62])
b. wipouten doyng of any harme
without doing of any harm
'without doing any harm'
(C.1300 K Alex. 558 [ibid.])
Structural scope expansion and grammaticalization 243

(20) pe teares pe man weped for longenge to heuene ben cleped rein
water oder deu water.
'the tears that man weeps in longing for heaven are called rain
water or dew water.'
(c. 1200 Trin. Horn. 151/17-18 [ibid.: 107])
(21) a. bisi In ordaining of priestes, and clerkes, And in casting kirc
werkes
'busy ordaining priests and clerics, and in planning church
works'
(c.1300 [MS. 1400] NHom 112/2-4 [ibid.: 76])
b. he shewed obedyens Yn fulfyllyng hys faders comoundemens.
'he showed obedience in fulfilling his father's commands.'
(c. 1325 Med. Supper 173-4 [ibid.])
(22) y haue more repented me of spekinge than y haue do of beinge still.
'I have regretted speaking more than being quiet.'
(1450 Scrope DSP 236/27-28 [ibid.: 92])
(23) shoulde take more honour in being coupled to Englande
(1545 Ascham Tox. [ibid.: 115])
(24) after having failed to take him into the fisher boate
(c. 1580 Sidney, Arcadia I. 36 [ibid: 112])
Abney (1987) and Tabor (1993) suggest that this development was
incremental, passing from the V0 stage through a V1 stage to the V2 stage.
Indeed, Tajima's analysis makes it appear that there was a V0 —> V1 transi­
tion around 1250 and a V1 → V2 transition around 1550. However, it is not
clear that these two transitions were distinct, for the modern, presumably
asymptotic, rate of use of passive and perfect (VP) gerunds, so it may be that
the V2 types appear later simply because they are much less likely to occur in
texts (see Kroch 1989a, 1989b), or because of independent developments in
the auxiliary system. Nevertheless, there is little doubt that the main transi­
tion, the advent of the VP gerund, occurred over this period of time.
Like the possessive, the gerund development also shows the three "hall­
marks of grammaticalization". As we have argued above, the VP-gerund is a
new syntactic form which did not exist in OE. The semantic function of the
ending has changed as well, for the original OE -ung ending was used mainly
in forming abstract nouns from the second class of weak verbs (Wright and
244 Whitney Tabor and Elizabeth Closs Traugott

Wright 1908: 299) while its modern descendant can be applied to essentially
any main verb. The outline of the development of the gerund based on
Tajima's tabulations, which we summarized above, suggests an incremental
development, but this evidence is not conclusive because the later emergence
of the predicate adjective, passive, and perfect gerunds may reflect the low
odds of seeing them used at all. However, the earlier possibility of gerunds
with non-possessive quantifiers (25), now ungrammatical, suggests that the
constraints on the modern construction did indeed emerge incrementally.
(25) The wythholdyng you fro it can doo yow no good.
(1481 Caxton, History of Reynard Fox [Tajima, p. 81])
Moreover, it is clear that at least by 1550, certain strings had become ambigu­
ous between the NP- and VP-gerund analyses. Thus we can make the following
comparison for God's blessing'.
(26) Diachronic String Comparison. NP gerund to VP gerund.
a. (N3 Godes [N2 [N1 [N° [Vo bliss] -ynge] ]] ]
b. (N God's [N -ing [V2 [V1
3 2
[Vo bless] ]]] ]

Under this analysis, the later -ing (26b) has greater scope than the earlier -ing
(26a). In this case, there are missing shells opposite both the old and the new
shells specified by the -ing marker. All the other shells line up on the
assumption that it is fair to make the correspondences, N2 ↔ V2 and Nl ↔
V1. But these correspondences are well-motivated in Jackendoff's schema: he
shows that modifiers on the same level share many properties across different
types of projections. Thus, this case also arguably involves C-command
Scope Increase. It certainly does not illustrate condensing of constructions
relating to a "constituent of arbitrary complexity" into one where the item (-
ing) modifies a word or stem as Lehmann (1995[1982]: 64, 164) seems to
suggest.

3.3. Instead of and instead

The development in English of the possessive and gerund are challenges to

the Scope Decrease Hypothesis, but they seem to have happened just once in
the few languages where they have happened at all. The case for C-command
Scope Increase will be strengthened if we can show that it characterizes a
common kind of grammaticalization episode. A clear example of this sort is
Structural scope expansion and grammaticalization 245

provided by a number of adverbial phrases in English which started out as

adjuncts of the verb and evolved into sentence adverbs, and in some cases,
discourse markers. In this section we will investigate instead (of), and in the
next anyway.
PDE has at least three syntactically distinct adverbial formatives which
derive from the OE noun stede meaning 'place.' Instead of functions as a
preposition heading an adverbial phrase and as a conjunction. Instead alone
functions as an adverbial phrase, with a distribution similar to that of adverbial
instead of We justify these assertions and then outline the developments that
led to the modern situation, showing how several strands involve Scope
Increase.
In his groundbreaking work on adverbials, Jackendoff (1972) distin­
guished three adverbial positions based on distributional evidence: sentence-
initial, within the range of the auxiliary, and in sentence-final position. He
suggested three levels of configurational scope, sometimes described in the
following terms: those adverbs that are sisters to S (e.g., probably), those that
are sisters to V-bar (e.g., intentionally), and those that are sisters to V (e.g.,
completely). Subsequent work has refined the analysis, with particular atten­
tion to the fact that most adverbs can occur in several different slots, sometimes
with different meanings, and to the constraints on the distributional character­
istics of particular classes of adverbs with respect to the auxiliary (e.g.,
McConnell-Ginet 1982; Ernst 1984; McCawley 1988; Radford 1988). No
current linguistic theory has a full account of these facts, but most agree on the
distinctions that are important for the points we make here. McCawley (1988)
gives a summary of distribution on which we expand to account for instead (of):9

(27) Distribution of PDE adverbial instead, instead of NP, and in Al's

place in comparison to various adverbs (based on the table given
by McCawley 1988 (p. 632)).
ɸ The carpenter ɸ will ɸ have ɸ constructed a chimney ɸ, ɸ.
perhaps OK OK OK ?? * OK
intentionally * * OK OK OK *
completely * * ? OK OK *
instead OK ?? OK ?? OK ??
instead OK OK ?? ?? OK ??
... of installing a window
in Al's place OK * * * OK OK
246 Whitney Tabor and Elizabeth Closs Traugott

Comparing the distributions of instead and instead of NP against this chart

suggests sentence adverb status for both of these expressions, although the
correspondence is not quite perfect.
Further evidence that the two adverbial instead constructions are not V-
bar or V-adverbs comes from a semantic scope contrast: like S-adverbs
(SAdvs), the instead adverbs strongly prefer to be scoped outside of the
quantification on an embedded NP, while the V adverbs (VAdvs) resist wide
scope readings, and the V-bar adverbs accommodate both scopings (cf.
McCawley 1988: 634):
(28) a. Marvin completely ate all three bagels. (prefer narrow)
b. Marvin intentionally ate all three bagels. (narrow or wide)
c. Marvin probably ate all three bagels. (prefer wide)
d. Marvin instead (of shaving) ate all three bagels. (prefer wide)
While the phrase instead of NP functions as an adverbial, the substring
instead of seems best analyzed as a complex preposition. It can do most of the
sorts of things prepositions can do:
(29) a. Nominal embedding: A boy with a marshmallow instead of/
on a stick
b. Preposition stranding: What did you eat watercress instead of
/with?
c. Subordination: The judge dismissed the case instead
of / without filing charges.
However, unlike any other English prepositions, instead of can also
behave as a coordinating conjunction, combining a wide range of categories.
In this regard, it behaves almost identically with the canonical conjunctions
(e.g., and, or, but not, etc.). In some cases there is structural ambiguity with
the prepositional use, as in (30a), but usually the constructions are totally
distinct. Examples of conjunctive instead of include:
(30) a. We're now getting hostility instead of / and / but not hospitality.
(4 July 1990 UPI10 (NP conjunction)
b. Gonzalez agreed to be sentenced in New York instead o f / and
/ but not Miami.
(27 Sept. 1990 UPI (PP conjunction).
c. Dealing with abortion ..., instead of / and / but not making it a
crime, is the most effective way to reduce abortions.
(14 July 1990 UPI (VP-gerund conjunction)
Structural scope expansion and grammaticalization 247

d. You can sit instead of / and / but not stand.

(finite VP conjunction)
e. to put things on instead of /and /but not to strip things off
(non-finite VP conjunction)
However, instead of does not enter into conjunction of inflected clauses
(31a); nor is it very felicitous as a conjoiner of nonmaximal projections (31b-
d):
(31) a. He runs to her and / * instead of she runs to him.
(finite S conjunction)
b. *The king and / * instead of queen of England wrote the note.
(N conjunction)
c. A rather contrite and / ? instead of miserable face appeared in
the doorway.
(Adj conjunction)
d. We build and / ? instead of buy picture frames. (V conjunction)
All three of the instead constructions treat in and stead as inseparable (in
*Jasmine's / *the / *good stead). Moreover, the conjunction instead of does
not tolerate separation of the of:
(32) a. Instead, by the way, of clapping, perhaps you could just stomp
softly. (adverbial instead of)
b. You can sit instead (*by the way) of stand. (conjunction in­
stead of)
These data suggest the following analyses of the various instead (of)
constructions:
(32) a. PDE Adverbial instead (of NP).
248 Whitney Tabor and Elizabeth Closs Traugott

Tree Diagram (32a) captures the fact that the adverbial instead's can
appear in the outermost shell of the matrix clause as sentence adverbs. It does
not provide a mechanism for generating sentential adverbs in the many
intermediate positions in which they occur, but we know of no fleshed-out
syntactic theory that does at the moment. Our arguments for C-command
Scope Increase hinge only on the claim that instead(of) can occupy this
outermost position in IP. Tree Diagram (32b) captures the fact that conjunc­
tive instead of can conjoin maximal projections of any type except IP.
With these analyses in mind, we now review the historical developments
(based on Schwenter and Traugott 1995).
The modern preposition instead of stems indirectly from the concrete
sense of OE stede as physical place. In OE we find only stede 'place' and a
metaphorically extended use expressing substitution (originally of one person
for another in a position or role). We will call this the "deputative" type:
(33) & God silf hine bebirigde & gesette Iosue on
& God himself him buried & appointed Joshua in
Moyses stede pam mannum to heretogan.
Moses' place those men:DAT as leader
'and God himself buried him (Moses) and appointed Joshua in
Moses' place, as leader of the people.'
(c. 1000 Aelfric, Letter Sigeweard, p. 31 [HC])
In ME we find this type with the of-genitive:
(34) I schall haue a-nothyre mann jn the stede of Pekoke.
T shall have another man in the place of Peacock.'
(c. 1450 John Paston II, letter 231])
This early use clearly involves stede as a noun, syntactically distinct from
the preceding preposition: while we cannot place any words between in and
stead in the modern adverbial sense, OE speakers could do so. Thus we posit
the following analysis:
Structural scope expansion and grammaticalization 249

By the later ME period the substitutive construction begins to be attested

as the head not only of animate nouns but also of concrete objects and abstract
(nominalized) actions, including NP gerunds, in constructions that suggest it
had acquired its subordinator use. These generalized uses of the construction
are most frequently found in topicalized position, after the complementizer, if
there is one:
(36) Therfore in stede of wepynge and preyeres
Men moote yeve silver to the povre freres.
Therefore instead of weeping and prayers, people should give
silver to the poor friars.'
(c. 1388 Chaucer, Prol. Cant. Tales, p. 27 [HC])
By EMdE the contexts have been expanded to include VP gerunds:
(37) a. Grim-visaged war hath smoothed his wrinkled front,
And now instead of mounting barbed steeds ...
He capers nimbly in a lady's chamber.
(1597 Shakespeare, Richard III I.i.5-8)
b. studies ... which will corrupt and hurt in stead of doing good.
(1627 Brinsley, Ludus Literarius p. 45 [HC])
c. The senate there, instead of being elected for a term of six
years, and of being unconfined to particular families or for­
tunes, is an hereditary assembly of the opulent nobles.
(1787 Federalist papers p. 430)
250 Whitney Tabor and Elizabeth Closs Traugott

It seems reasonable to conjecture that at this point instead of is function­

ing as a complex preposition and instead of NP has joined the class of SAdvs in
the language:11 it frequently effects contrast between event-denoting clauses
rather than person-denoting noun phrases and it can occur clause-initially
(37a), between the subject and main verb (37c), and also at the end of VP
(37b). We can surmise that it had, by the 17th century and probably signifi­
cantly earlier, a different analysis from the earlier locative PP, given the fact
that modern locative PPs have a distribution more similar to that of VAdvs
than SAdvs. Also, locative PPs can participate in nominal embedding (38) and
locative inversion (39) while SAdvs cannot:
(38) a. Which umbrella are you looking for?
b. The umbrella in Loretta's chair
c. * The umbrella probably
d. * The umbrella instead (of a cane)
(39) In Sonny's place stood a statue of Venus.
* Probably stood a statue of Venus.
?? Instead (of his rifle) stood a statute of Venus.
Further evidence that instead of had developed adverbial status by the
17th century is provided by the appearance, around this time, of the similarly
distributed form instead, without an overt PP (presumably via ellipsis of the
anaphoric instead of this):
(40) a. and in derision sets
Upon their tongues a various spirit, to rase
Quite out thir Native Language, and instead
To sow a jangling noise of words unknown.
[instead of their native language]
(1667 Milton, Paradise Lost xii. 51 [OED])
b. Elizabeth felt herself completely taken in. She had fully pro­
posed being engaged by Wickham for those very dances: —
and to have Mr. Collins instead! her liveliness had been never
worse timed.
(1813 Austin, Pride & Prejudice, p. 87)
c. No doubt the hope of praise and liking and services from one's
fellowmen, and the fear of forfeiting these and incurring in­
stead aversion, ...
(1874 Sidgwick, Methods of Ethics, p. 166)
Structural scope expansion and grammaticalization 251

In Austen's novels adverbial instead is used only in direct speech, indirect

speech (40b), or free indirect style and not in narrative, which suggests it was
regarded at the beginning of the 19th century as a feature of spoken language.
However, it appears in a philosophical work at the end of the century (40c).
While the history of the adverbial instead's sketched above suggests that
instead of NP underwent C-command Scope Increase during the 17th century,
instead does not appear in our data base preceding a full clause until the 20th
century (41), suggesting that the emergence of the full-fledged SAdv instead
was gradual:
(41) Michaelis tried to find out what had happened, but Wilson
wouldn't say a word — instead he began to throw curious, suspi­
cious glances at his visitors
(1925 Fitzgerald, Great Gatsby p. 164)
Meanwhile, starting in the 17th century, we find instead of spread to
contexts preceding a participle (42a) and PP (42b, c), thus beginning to take on
the behaviors of a conjunction:
(42) a. Would he not think himself mocked, instead of taught, with
such an account as his?
(1690 Locke, Essay Human Understanding, np 175, dp 230)
b. It was very strange that he should come to Longbourn instead
of to Lucas Lodge.
(1813 Austen, Pride & Prejudice, p. 128)
More recent types to appear involve finite and non-finite VP coordination: all
of the examples in our database involve ellipsis of modals:
(43) a. You are in the devil of a hurry, sir. Are you clear (= "sure") that
this haste will not mar, instead of make an understanding? I
think I am, sir.
(1794 Godwin, Caleb Williams, IV, p. 28)
b. It must be so frightful to have to put things on in order to look
better, instead of to strip things off.
(1962 Drabble, A Summer Bird Cage [Quirk et al. 1985: 661])
c. "Mr. Rose," US. District Judge Arthur Spiegel politely asked
the man called "Pete" by most people, "are you in pain because
of your leg? If so, you can sit instead of stand."
(July 19 1990 UPI)
252 Whitney Tabor and Elizabeth Closs Traugott

The data outlined here suggest that the conjunction instead of, like the
adverbial instead (of) spread gradually, step by step to more and more con­
structions. It is likely that conjunctive instead of arose directly out of the
deputative instead of type in (33), once inanimate nouns had become available
in this construction, rather than out of adverbial instead. The deputative
construction typified by God appointed Joshua in Moses ' stead had persisted
alongside the adverbial construction (it is now largely replaced by in (the)
place of Moses). There is a semantically plausible reanalysis for the deputative
—» conjunctive transition, motivated simply by loss of the constraint that the
entity X existed in the position before Y was substituted.
Regarding "hallmarks of grammaticalization", we have presented evi­
dence for several syntactic changes. Each of these syntactic changes was
clearly associated with a semantic change (see Schwenter and Traugott 1995
for treatment of developments within deputative instead of and the emergence
of substitutive instead of). Moreover, Schwenter and Traugott show that
modern day in place of has a distribution very similar to that of an earlier,
intermediate, stage of instead of when the construction had locative senses,
social role replacement senses, and inanimate role replacement senses, but
not gerund or tensed verb phrase uses. Thus, we also have evidence for a
grammatically incremental development here.
In a brief passage, Ramat (1992) characterizes the development of
adverbial instead as a case of "degrammaticalization" or return to the lexicon.
However, this ignores the properties just mentioned, and the fact that instead
has no characteristics of major class lexical items. The history of all three
instead constructions is fully consistent with processes of grammaticalization,
except that adverbial instead (of) violates the postulated scope reduction.
Based on the analyses we presented above, and using a modified version
of example (36a), we can make the following DSC for the adverbial instead of
NP:

(44) Diachronic String Comparison. Deputative to adverbial instead of

a. [IP [IP Men • [VP yeve [NP silver] [PP in stede of wepynge ]]] ]
b. [IP [IP Men [VP give [NP silver] 11 [PP instead of weeping ]]

In (44a), men are being asked to "give silver" and not "give weeping"
(consistent with Chaucer's example (36a)), while in (44b), men are being
asked to "give silver" rather than "engage in the act of weeping". Here we
have scope increase of the prepositional phrase headed by in stede/instead.
Structural scope expansion and grammaticalization 253

Thus this case supports the C-command Scope Increase hypothesis.

As indicated above, we assume that conjunctive instead of arose from
deputative instead of rather than from adverbial instead (of NP). This suggests
the following DSC:
(45) Diachronic String Comparison. Deputative to Conjunctive instead of
a. [IP I [VP have [NP a-nothyre ] [PP jn in the stede of Pekoke ]]]
b. [IP I [VP have [NP [NP another] instead-of [NP Pekoke]] ]]

Unlike the adverbial case, this does not exhibit C-command Scope Increase.
In fact, one might argue that it involves C-command Scope Decrease since the
PP jn the stede of Pekoke in (45 a) C-commands the NP which instead-of
Pekoke becomes a subpart of in (45b). However, the analyses may be mislead­
ing. Larson (1981) and Pesetsky (1996) present evidence from negative
polarity environments, pronoun binding, and reflexive binding that the second
constituent following a double complement verb (like have NP PP) is C-
commanded by the first argument. Under such an analysis, there is a clear
sense in which in (the) stede/instead-of Pekoke has "moved up" in the struc­
ture, although some elaboration of the notion of C-command may be required
to make this case consistent with the other cases we have examined. We will
not pursue this elaboration here. We conclude, thus, that the development of
conjunctive instead of is a possible exception to C-command Scope Increase,
while the development of adverbial instead (of NP) is clearly consistent with it.

3.4. Anyway

We turn now to one example of a frequently attested syntactic development of

a clause-internal adverb to a sentence adverb and sometimes ultimately a
discourse marker. Hanson (1987) discussed the development of manner ad­
verbs like probably and possibly into epistemic sentence adverbs; Powell
(1992) showed how a whole class of "stance" adverbs (Biber and Finegan
1988) including actually, generally, really likewise arose from clause-internal
adverbial constructions; and Bullen (1995) investigated the development of
clause-internal adverbs such as once, still, besides into clause connectives.
Traugott (1995, Forthc.) discusses the development of clause-internal indeed,
in fact, besides, anyway etc. into epistemic sentence adverbs that express
attitude to the truth-fullness of the proposition, and then into discourse mark­
ers; Brinton (1996) discusses a significantly wider set of similar changes. The
254 Whitney Tabor and Elizabeth Closs Traugott

focus of many of these studies has been the semantics and pragmatics of the
changes, and their significance for grammaticalization. However, there are
clearly important syntactic structural changes as well.
In Traugott (1995, Forthc.) the same kinds of syntactic criteria are
outlined as were given in Section 3.3. to distinguish between VAdvs and
SAdvs. VAdvs typically occur after the verb, SAdvs either clause-initially
(where they follow the Complementizer, if there is one) or within the auxil­
iary or (in some cases) clause-finally. An additional category, Discourse
Markers (DMs), is discussed. DMs are items that "bracket" units of discourse
(Schiffrin 1987). In a more restrictive definition of discourse markers build­
ing on Schiffrin's subclass of "discourse deictics", Fraser has defined dis­
course markers as the class of pragmatic markers that "signal a comment
specifying the type of sequential discourse relationship that holds between the
current utterance — the utterance of which the discourse marker is a part —
and the prior discourse" (Fraser 1988: 21-22), and that is the definition used
here. Whereas SAdvs have content-meaning, and are subject to truth condi­
tions, DMs are primarily pragmatic in nature and serve to signal the speaker's
attitude about the discourse relationship between what preceded and what
follows.12 More importantly for our immediate purposes here, DMs have
structural properties. Typically they are disjunctive, requiring comma intona­
tion, and they occur externally to SAdvs, as in:
(46) a. So, probably you don't approve.
b. I like it. Indeed. I love it.
Historically, it can be shown that many DMs in English derive from VAdvs,
often via an SAdv stage. That they arise last is totally unexpected given the
traditional view of grammaticalization as a condensing and fixing machine,
since they not only have the widest syntactic scope but they also can occur in
multiple positions in the clause, and furthermore, they require comma intona­
tion. These characteristics, and the fact that they have primarily pragmatic
rather than content meaning, have led some researchers to question whether
adverbs with DM function can be considered as cases of grammaticalization.
For example, in synchronic studies of adverbials like y'know that function as
discourse markers in the broad sense first used by Schiffrin (1987), not the
narrower one used here, Erman and Kotsinas (1993) label them cases of
"pragmaticalization", and Vincent, Votre, and LaForest (1993) label them
(synchronically) as cases of "post-grammaticalization". However, they are
Structural scope expansion and grammaticalization 255

syntactically and intornationally constrained, and even if not fully integrated

into the clause, must be considered "part of the grammar of a language"
(Fraser 1988:32). Furthermore, they show all the hallmarks of grammatical­
ization other than structural scope and variability reduction.
Here we discuss anyway in some detail from a syntactic perspective.
Syntactically, anyway has at least two distinct uses in contemporary English,
which we will label according to their senses without meaning to imply any
particular semantic/pragmatic analysis thereby (for a detailed study of uses in
Texas English, see Ferrara 1997). There is a concessive anyway (C-anyway)
which means something similar to 'nonetheless', and appears to be restricted
to the right periphery of a VP:
(47) a. In one instance a resterilizing service returned a pacemaker to
Walton labeled "Not For Human Implant," but he sold it any­
way and it was implanted in a patient.
(23 June 1992 UPI)
b. They cooked the fish anyway, probably. (enclitic anyway)
c. They cooked the fish, probably/*anyway. (enclitic anyway)
There is also a topic-resuming anyway (TR-anyway) which is used to
signal a return to a previous topic and serves as a DM (48):
(48) a. So uhm — anyway just think about this offer.
(London Lund Corpus, line 4,200)
b. Anyway, probably the enemy is surviving on birds and squirrels.
TR-anyway can occur in many syntactic positions, including before SAdvs, but
it must be surrounded by pauses wherever it occurs. We again extend
McCawley (1988)'s table of adverbs, to compare its distribution with other
adverbials:
(49) Distribution of PDE TR-anyway, C-anyway, and M-any way in
comparison to various adverbs. (based on the table given by
McCawley 1988 (p. 632)).
ɸ The sailors ɸ will not ɸ have ɸ scanned the horizon □, □ .
perhaps OK OK OK ?? * OK
intentionally * * OK OK OK *
completely * * ? OK OK *
TR anyway OK OK OK OK * OK
C anyway * * * * OK *
M in any way * * ?? ?? OK OK
256 Whitney Tabor and Elizabeth Closs Traugott

In addition, there is also a phrase any way meaning 'in any manner, by
any means', which contrasts with these anyway's in permitting material to
intervene between any and way, and functions like a manner adverb (we refer
to it as M-anyway):
(50) She did it (in) any (old) way she pleased.
We hypothesize that M-anyway can be characterized as in (51a), and C-
anyway as in (51b). We are not aware of any complete generative analysis of
DM's, but note they have some syntactic properties in common with the left­
most "E-node" posited for expressive adjuncts by Banfield (1973) and
Emonds (1985), and recently argued for external topics in Mayan by Aissen
(1992).13 We therefore propose the tentative analysis for TR-anyway in
(51c).14
Structural scope expansion and grammaticalization 257

We now turn to an examination of the historical development of TR-

anyway, C-anyway, and M-anyway (elaborating on Traugott Forthc.). As may
be surmised from their form, the adverbial anyway's ultimately derive from a
directional construction:
(52) hi da æfter gereorde on heora weg ferdon.
they then after meal on their way went.
(c. 1000 Aelfric Horn 11, 10, 84.113 [DOE])
As an adverbial phrase, directional any way (without other modifiers) is often
semantically ambiguous between path, manner, and extent. Guessing from its
semantics and the lack of any evidence to the contrary, ME M-any way was
probably a VP internal adjunct like OE directional anyway and PDE M-
anyway:
(53) Sche wolde 3et excusyn hir yf sche myth in any wey.
She wanted still excuse:INF her if she might in any way
'She still wanted to excuse herself if she could in any way.'
(c. 1438 Kempe p. 1,227 [HC])
This adverbial phrase appears in the early 17th century unambiguously in the
meaning of extent. The contexts are always negatives as in (54), quantified
indefinites or other polarity items such as conditionals:
(54) that they bee not any way overloaded or discouraged.
(1627 Brinsley, Ludus Literarius, p. 12 [HC])
The textual data show possible uses of C-anyway 'nonetheless' appearing first
in the mid 17th century. In this meaning it is also at first used only in negative,
conditional or indefinite contexts.
258 Whitney Tabor and Elizabeth Closs Traugott

(55) a. For the word heir does not itself imply the children, or nearest
kin of a man; but whomsoever a man shall any way declare, he
would have to succeed him in his estate.
(1651 Hobbes, Leviathan LVI p. 182)
Note that this non-final use of C-anyway is ungrammatical in current English.
It is probably not fully developed as a concessive at this stage. The clause-
final concessive appears first in the mid 19th century:
(55) b. "I don't know whether the story about Lady Laura is true."
"He was always there... The mischief he has done is incalcu­
lable. There's a Conservative sitting in poor Kennedy's seat
for Dunross shire."
'That might have been the case anyway."
"Nothing could have turned Kennedy out."
(c. 1848 Trollope, Phineas Redux)
We make the assumption that these 18th and 19th century C-anyway's, like
20th century C-anyway, were VP adverbs.
The first clear examples of TR-anyway in our database are shown in (56).
Each involves a return after a digression.
(56) a. It's queer; very queer; and he's queer too; aye, take him fore
and aft, he's about the queerest old man Stubb ever sailed with.
How he flashed at me! — his eyes like powder-pans! is he
mad? Anyway there's something on his mind, as sure as there
must be something on a deck when it cracks.
(1851 Melville, Moby Dick, p. 125 )
b. He has ... possibly the feeling that he has 'balked' me — or that
I am a detective from the enemies' camp. Anyway, he did not
encourage conversation.
(1891 Beatrice Webb, Letters)15
Similar histories are outlined in the OED for related forms like anyways and
anyhow.
The anyway episode also shows the three "hallmarks of grammaticaliza-
tion". We have argued for a syntactic contrast between the original direc­
tional/manner (in) any way, the subsequent C-anyway, and the most modern
TR-anyway. Traugott (Forthc.) identifies semantic and pragmatic differences
between these three types and also notes that the loss of the preposition in
Structural scope expansion and grammaticalization 259

preceded the first evidence for C-anyway by at least one hundred years (thus
exhibiting structural gradualness). Furthermore, as is typical of most recently
grammaticalized elements (see Kroch 1989b; Bybee, Perkins and Pagliuca
1994; Tabor 1994a), TR-anyway is by far the most frequent type of anyway
(Ferrara 1997).
We assume that Directional anyway, Manner anyway, and Extent any­
way have been V-adverbs since their inception. There may have been subtle
syntactic changes accompanying the semantic development from one to the
next, but we do not have enough data on this part of the episode to assess it
effectively. The developments from M-anyway to C-anyway, and from C-
anyway to TR-anyway , on the other hand, clearly involve C-command Scope
Increase. Using a plausible simplification of example (53) above, and follow­
ing our analyses of M-anyway and C-anyway, we can make the following
DSCs: 16
(57) Diachronic String Comparison.
(a) M-anyway, (b) C-anyway, and (c) TR-anyway.
a. [IP Sche [VP excusyth [NP hir] [PP in any wey] ] ]
b. [IP She [VP [VP excuses [NP herself] ] anyway] ]
c. [E [IP She [VP excuses [NPherslef] ] 1 anyway]

Clearly, the transition from M-anyway to C-anyway involves scope increase,

as does the shift from C-anyway to TR-anyway.17
In sum, the examples of instead and anyway provide evidence that
although discourse use is without question an issue in the development of the
different syntactic and semantic functions of the constructions in question,
they do not exemplify the kinds of changes from "loose" to "tight" syntax that
Givón (1979) and others seem to have had in mind when they characterize
grammaticalization as an instance of discourse > syntax. In each case a
construction originated in a syntactically structured context. Through use in
discourse, most particularly relatively fixed formulae, the constructions ac­
quired new meaning (in stede 'in the location of > 'in role of; any way 'any
path' > 'any extent'). Once these semantic changes had occurred, the con­
structions became eligible for use in topicalizations and other movement rules,
and eventually for base-generation as SAdvs (instead) or DMs (anyway), etc.
Full NP status and topicalization are and have been available for the full lexical
nouns from which these forms developed throughout the history of English,
e.g.:
260 Whitney Tabor and Elizabeth Closs Traugott

(58) As for his deeds, they did not match his intentions.
but the forms and uses discussed entered the language at different times and in
an order which appears to be entirely generalizable:
(59) VAdv > SAdv (> DM)
The later constructions are disjunct in the case of DMs, and therefore do not
illustrate tighter syntax; each stage involves C-command Scope Increase.

4. Conclusions

4.1. Summary

We started by noting that structural unidirectionality, as usually interpreted,

seems troubled by some hard-to-dismiss counterexamples. We went on to
observe that there is a need for careful formalization of claims about unidirec­
tionality. We proposed one system for doing this: diachronic string comparison,
combined with a synchronic definition of scope-change based on C-command.
Using our formulation to give one explicit interpretation to Lehmann's scope
decrease claim, we showed that the latter claim is clearly violated in a number
of cases (consistent with the suggestions of Joseph and Janda 1994; Janda 1995;
Campbell 1991, and others cited in Section 1). We noted that instead of Scope
Decrease, these cases exhibited C-command Scope Increase.

4.2. The plausibility of Scope Increase as a generalization.

Most of the evidence we have presented here appears to support a generaliza­

tion about structural unidirectionality that is opposite to the generalization that
most previous research on the topic has considered. We think that the number
of cases consistent with this hypothesis makes it well worth investigating
further. Some others of a different type that are particularly relevant include
the changes from monoclausal —» biclausal structures, and subordinate clause
→ independent clause cited in Section 1. However, we acknowledge that the
C-command Scope Increase Hypothesis is not likely to prove universal—the
example of the development of conjunctive instead of has shown this already.
It remains to test the hypothesis against changes that are often considered
prototypical of grammaticalization, for example, the development of auxilia-
Structural scope expansion and grammaticalization 261

ries. Although there is no question that morphophonological bonding is char­

acteristic of many processes of auxiliation (see Bybee 1985; Bybee, Perkins
and Pagliuca 1994), recent syntactic analyses of the syntax of auxiliation raise
questions about such claims as: "An auxiliary of the 'have' or 'be' type starts
as a main verb which takes a nominalized VP as a complement; that is, it starts
at the clause level. When it has become an auxiliary, it functions at the VP
level" (Lehmann 1995[1982]:144). It is unquestionably true that from a
morphological perspective inflectional (e.g., tense) affixes on verbs descend
from syntactically independent periphrastic forms (e.g. dicere habeo 'say:INF
have:lPers' > Fr. dirai 'say-FUT:lPers), arguably via a clitic stage. Syntacti­
cally, however, this set of changes is more complex. Roberts (1993) analyses
the early stage of this development in Sardinian (in which aio corresponds to
habeo) as:
(60) Diachronic String Comparison.
Latin perfect to early Romance future (based on Roberts 1993: 235)
a. [IP [VP dicer ti. ] habeo. ]
b. [IP [VP/TP dicere ] aio ]

This analysis demonstrates C-command Scope Increase, provided we evalu­

ate the syntactic changes at the level of D-structure. While it is difficult to
align the kinds of shells we have been using with the various levels into which
the once unitary INFL node have been split (AgrP, TP, etc.), it is worth noting
that Roberts' primary proposed mechanism for grammaticalization is "move­
ment chain elimination", which involves eliminating the D-structure to S-
structure movement in favor of base-generation in the S-structure position.
Since most movement is upward in parse-tree structure, this hypothesis looks
somewhat consistent with a scope-increase viewpoint, under the assumption
of comparison at D-structure. Clearly, though, more work needs to be done on
these cases before a conclusion can be reached.
Even if the syntactic changes characteristic of auxiliation were to evi­
dence something like C-command Scope Increase on a systematic basis cross-
linguistically, it must still be acknowledged that not all changes can be
accounted for in terms of morphological bonding and syntactic Scope In­
crease. For example, the expressions sort of and kind of (sorta, kinda) have
developed degree adverbial uses that contrast with historically prior (and still
co-existent Noun-Prep uses (Tabor 1994a, 1994b). Starting in the early 19th
century, the following comparison becomes possible:
262 Whitney Tabor and Elizabeth Closs Traugott

(61) Diachronic String Comparison.

Noun-Preposition to Degree Modifier sort of.
a. [NP a [N sort [PP of [NP [AdjP [Adj breezy]] [N day]]]] ]
b. [NP a [NP [AdjP sort-of [Adj breezy]] [N day]] ]

Tabor presents quantitative evidence suggesting that constructions of this

form played a central role in the development of the new usage. Here,
however, the two items that have changed their syntactic status, sort and of,
have decreased their C-command scope. We know of no way to distinguish
this case from the others we have considered on independent grounds, so it
remains a challenge to the Scope Increase Hypothesis. There are other rea­
sons as well to be cautious about the result at this point. For one, our claim is
somewhat limited in its purview, because we must be able to perform diachro-
nic string comparison in order to evaluate it. This means that there may be
some changes that show the "hallmarks of grammaticalization" that we identi­
fied in Section 2, but for which we cannot say that constructional scope either
increased or decreased.
We do not, however, recommend abandoning the investigation of struc­
tural unidirectionality in the study of grammaticalization. The exact nature of
individual changes needs to be understood, and the question remains to be
answered whether there are generalizations to be made about particular
classes of change.
Furthermore, there is a diachronic-theoretical reason to prefer the C-
command Scope Increase Hypothesis over scope-decrease hypotheses: it is
more consistent with independently motivated evidence for unidirectionality
in pragmatics/semantics and statistical distributional change. In particular, the
change from being a marker of something in the world (referential) to being a
marker of something in the language (non-referential) is expected to increase
the scope of a formative since functional (language-based) markers generally
mark whole classes of lexical (world-based) markers. The change from being
less abstract to more abstract is also naturally correlated with grammatical
scope because more abstract meanings apply to a wider variety of circum­
stances. The change from being a low frequency word to a high frequency
word is aligned with Scope Increase in our sense because increasing the set of
cooccurrence possibilities of an item tends to increase the number of occurring
situations in which it is used. There is also a reason to expect a correlation
between greater C-command scope and less phonetic substance, for most
functional morphemes are short. This last, somewhat noisy generalization,
Structural scope expansion and grammaticalization 263

may be due to the fact that an efficient code (that is, one that minimizes the
average message length) should use less bandwidth for more frequent mes­
sages (e.g., Hamming 1980).
Given these overarching reasons to be interested in Scope Increase
claims, and the large number of cases consistent with it, we think the hypoth­
esis is worth looking into further.

4.3. On the putative isomorphism between synchronic and diachronic

"grammaticalization clines."

One of the methodological assumptions of much synchronic work from the

perspective of grammaticalization has been that historical development and
synchronic position on a cline of grammaticality will in general coincide, in
other words, that there is isomorphism between historical development and
synchronic relations among polysemous items.
This assumption may have its more distant origins in the hypothesis that
ontogeny recapitulates phylogeny (for a summary of views since the 19th
century, see Baron 1977; Harris and Campbell 1995:29-30 trace the idea back
to the 17th century). Evidence for the hypothesis that ontogeny recapitulates
phylogeny in language acquisition has become less and less convincing,
leading one of its major original proponents to currently find the parallels
illusory (see Slobin 1994, reassessing Slobin 1977). The assumption of a
synchronic-diachronic isomorphism in grammaticalization has its more im­
mediate origins in Givón's "Today's morphology is yesterday's syntax", and
his hypothesis cited in (1) in Section 1. Givón's proposals were widely
explored, not only in historical syntax but also in semantics (e.g., Bybee 1985;
Heine, Claudi and Hunnemeyer 1991; several papers in Traugott and Heine
1991). It promoted many studies of internal reconstruction (e.g., Traugott
1988) and cross-linguistic lexical and typological work (e.g., Bybee, Perkins
and Pagliuca 1994; Heine et al. 1993). However, it was early shown that
putative syntactic matches (e.g., OV > VO, as proposed in Givón 1971) can
be falsifiable because of the phenomenon of second position and other
properties of cliticization (Steele 1977). Also, putative semantic matches can
be falsified if the relationship between core or salient and derived meanings
changes over time (Michaelis 1993).
A related kind of isomorphism is proposed by Lehmann (1995[1982]),
and also Heine, Claudi and Hiinnemeyer (1991): that between diachronic
264 Whitney Tabor and Elizabeth Closs Traugott

change and typological panchrony. Lehmann's panchronic approach permits

typological work to be done locating typologically but not necessarily histori­
cally related elements on a "grammaticalization scale". Of this scale
Lehmann says it is "a theoretical construct along which functionally similar
signs types [sic] are ordered according to their degree of grammaticality"
(Lehmann 1995[1982]: 25), as measured by six structural paradigmatic and
syntagmatic parameters. He also says that the diachronic approach permits
historically related items to be measured on a "grammaticalization channel",
defined as "a frequently recurring route which signs with a given function
may take when they are grammaticalized in language change" (ibid). This
distinction between grammaticalization scale and grammaticalization chan­
nel is extremely useful. However, by Lehmann's own admission, he some­
times uses them "interchangeably" (ibid.). Indeed it is the confusion over the
two that leads, in part, to Lehmann's claim cited in section 3.2 about the
gerund in English. His conflation of the two terms seems to presuppose that
his synchronic cline is always lined up with the diachronic order-of-emer-
gence facts. But, as we noted in the case of the gerund, this is false.
The example of the gerund highlights the importance of testing any
hypothesis regarding matches. Historical evidence is the main testing ground.
When no historical evidence is available, reconstruction based on an assump­
tion of unidirectional isomorphism especially in the domain of Scope De­
crease must be considered hypothetical, not factual, and should be framed as
a testable hypothesis. Bybee, Pagliuca, and Perkins, for example, argue that
form-meaning correlations based on a principle of semantic relevance of one
element to another (which predicts closeness to the stem of a verbal affix)
intersect form-meaning correlations based on degree of grammaticalization
(which predicts degree of bonding), such that: "(w)e would then expect grams
that are older — i.e., that have undergone more development — to be closer to
the stem, more fused and shorter or more reduced in segmental material than
younger grams of equal relevance" (Bybee, Pagliuca, and Perkins 1991:33).
This is a properly testable hypothesis about matches.

4.4. The limits of grammaticalization

Lehmann and others have implied that unidirectionality involving structural

condensing is an appropriate criterion for identifying a class of language
change phenomena that form a natural topic of study (from not grammatical to
Structural scope expansion and grammaticalization 265

grammatical, or from less grammatical to more grammatical, to paraphrase

Kuryłowicz 1965). We have argued, however, that the notion of structural
compacting, including scope-reduction, has been too vaguely articulated up
to now. Because of this, we propose using other criteria — semantic and
syntactic reclassification, increase in abstraction and non-referentiality,
gradual step-by-step change — to identify the phenomenon of grammatical­
ization. We also propose taking structural unidirectionality as a hypothesis to
be tested. We have argued for the viability of this approach by showing how it
is possible to define an explicit unidirectionality hypothesis and test it on a
wide range of cases. Indeed, our C-command Scope Increase Hypothesis,
which has come against relatively few exceptions so far, makes a claim that is
opposite to that of the structural unidirectionality hypotheses and assumptions
that have previously been proposed. We take this as evidence that structural
unidirectionality (in any formulation) is not an appropriate presupposition at
this stage in the development of the field. We do not mean to imply that no
formulation will ever become a reasonable background assumption, but we
believe that it is crucial — indeed it is one of the most fascinating challenges
in the field — to establish the plausibility of an explicit theory before making
an assumption along these lines. We offer the present paper as a step in that
direction.

NOTES

1. Elizabeth Traugott thanks Roger Lass for inspiration derived from lengthy correspon­
dence about various aspects of unidirectionality.
2. If one adopts the analysis of Kemenade (1987); Kroch (1989b); Pintzuk (1991), and
others for OE, then wille in (6a) moves up to the INFL position (the lexical daughter of I')
in the surface structure. On this view, our claim about C-command Scope Increase is still
tenable, but the comparison of scopes must be applied to the D-structure representations
of strings (i.e., before movement), and we must assume that the modern modals are base-
generated in the INFL position.
3. The results of Kroch (1989a, 1989b) indicate that one must use some caution in assessing
the gradualness of a change which can only be observed through the medium of historical
texts: after all, it may be that some uses of a new construction fail to appear in the texts at
an early date simply because they are rarely used. However, we can use several methods
to avoid this pitfall: (1) if the texts show an incremental progression for construction X,
and we find that there is a current construction Y with similar semantics that is still, based
on negative felicity and/or grammaticality judgments, in one of the earlier stages of X,
then we can be fairly confident in the chronology of the historical record; (2) as Kroch
266 Whitney Tabor and Elizabeth Closs Traugott

(1989b) notes, if we find that the texts show an ultimately low-frequency usage appearing
earlier than a high-frequency usage, then it is likely that the luck of the draw reflects the
real chronology; (3) if a construction or formative starts out unattested, then becomes
significantly attested during the transition period, and finally becomes ungrammatical
again, we can be fairly confident of an intermediate grammatical stage.
4. We do not exclude the possibility, however, that a word may gradually reduce its range of
cooccurrence possibilities. The reduction of English brethren from meaning brothers in
general to members of certain male religious groups was probably a gradual change but it
did not result in a shift in grammatical status, only in lexical subclass.
5. The main periods of English are: Old English (OE) 700-1150, Middle English (ME) 1150-
1500, Early Modern English (EMdE) 1500-1750, Modern English (MdE) 1750-.
Our main data bases are:
a) The Dictionary of Old English Corpus in Electronic Form (Cameron et al. 1981)
(DOE).
a) The Helsinki Corpus of the English Language (see e.g., Rissanen et al. 1993) (HC)
b) The on-line Oxford English Dictionary (OED)
c) The Middle English Dictionary (MED)
Stanford Academic Text Services made access to these and a variety of other computer­
ized corpora possible.
6. Zwicky (1987) argues that the modern -s is still lexically combined with its host because
of examples like The boy who sold only a few apples/* apples's hat in which the plural
marking on the embedded noun blocks the appearance of the possessive marker on the
subject noun phrase.
7. The reason for this unclarity is briefly mentioned in 4.3.: interchangeable use of "scale"
(synchronic term) and "channel" (diachronic term).
8. At the same time, the spelling of the present participle ending -ende/-inde also changed to
-ing.
9. The boxes designate potential positions for adverbials; the final position is reserved for
"parenthetical, after-thought" uses of adverbials, as in She drew the sculpture well,
probably / *intentionally / *completely. In the afterthought uses, there is high stress on
the preceding phrase while the adverb has low stress.
10. In this and the following examples, the instead of version is from the United Press
International Top Stories on the date given; alternatives have been added.
11. SAdvs have been attested in English since its inception (see Swan 1988).
12. Blakemore (1987) introduced the useful term "procedural" for such functions.
13. This suggestion was originally made in Mendoza-Denton's (1994) study of the develop­
ment of topic-specifying concerning-NP constructions.
14. A fuller analysis would specify additional structural analysis for the positions which
anyway can enter.
15. Thanks to David Denison for making available A Corpus of Late Modern English Prose
(Department of English, University of Manchester, compiled by David Denison with
Graeme Trousdale and Linda van Bergen, 1994), from which this example comes.
Structural scope expansion and grammaticalization 267

16. (57) abstracts away from changes in the syntactic position of anyway.
17. The hypothesis that TR-anyway developed from C~anyway rather than M-anyway is
pragmatically motivated in terms of a shift from contentful counter-expectation to
pragmatic metatextual (DM) counter-expectation (see also the developments of indeed,
in fact, after all, etc.).

REFERENCES

Abney, Stephen P. 1987. The English Noun Phrase in its Sentential Aspect. Unpublished
Ph.D. dissertation, MIT.
Aissen, Judith. 1992. "Topic and focus in Mayan." Language 68: 43-80.
Banfield, Ann. 1973. "Narrative style and the grammar of direct and indirect speech."
Foundations of Language 10: 1-39.
Baron, Naomi S. 1977. Language Acquisition and Historical Change. New York: North-
Holland.
Biber, Douglas and Edward Finegan. 1988. "Adverbial stance types in English." Discourse
Processes 11: 1-34.
Bickerton, Derek. 1984. "The language bioprogram hypothesis." The Behavioral and Brain
Sciences 7:173-221.
Blakemore, Diane. 1987. Semantic Constraints on Relevance. New York: Blackwell.
Brinton, Laurel J. 1990. "The development of discourse markers in English." In Jacek
Fisiak (ed.), Historical Linguistics and Philology, 45-71. Berlin: De Gruyter,
Brinton, Laurel J. 1996. Pragmatic Markers in English: Grammaticalization and Discourse
Functions. Berlin: Mouton de Gruyter.
Bullen, James. 1995. "The histories of English connectives: A study of the paths and
processes by which words have acquired connective uses in Middle and Modern
English." Honors subthesis, Australian National University.
Bybee, Joan L. 1985. Morphology: A Study of the Relation between Meaning and Form.
Amsterdam/Philadelphia: John Benjamins.
Bybee, Joan L. and William Pagliuca. 1985. "Cross-linguistic comparison and the develop­
ment of grammatical meaning." In Jacek Fisiak (ed.), Historical Semantics and Histori­
cal Word Formation, 59-83. Berlin: de Gruyter.
Bybee, Joan L., William Pagliuca, Revere D. Perkins. 1991. "Back to the future." In
Traugott and Heine (eds), Vol. II: 17-58.
Bybee, Joan, Revere Perkins, and William Pagliuca. 1994. The Evolution of Grammar:
Tense, Aspect, and Modality in the Languages of the World. Chicago: University of
Chicago Press.
Cameron, Angus, Ashley Crandell Amos, Sharon Butler, Antonette diPaolo Healey. 1981.
The Dictionary of Old English Corpus in Electronic Form. University of Toronto:
Dictionary of Old English Project.
Campbell, Lyle. 1991. "Some grammaticalization changes in Estonian and their implica­
tions." In Traugott and Heine (eds), Vol. I: 285-299.
Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
. 1970. "Remarks on nominalization." In Rodney A. Jacobs and Peter S.
268 Whitney Tabor and Elizabeth Closs Traugott

Rosenbaum (eds), Readings in English Transformational Grammar, 184-221.

Waltham, MA: Ginn.
Curme, George O. 1912. "The history of the English gerund." Englische Studien 45: 249-
280.
Dasher, Richard. 1995. Grammaticalization in the System of Japanese Predicate Honorif-
ics. Unpublished Ph.D. Dissertation, Stanford University.
Davies, Mark. 1994. "Parameters, passives, and parsing: Motivating diachronic and
synchronic variation in Spanish and Portuguese." Parasession on Variation and Lin­
guistic Theory, Chicago Linguistics Society 30: 46-60.
Einenkel, Eugen. 1913. "Zur Geschichte des englischen Gerundiums."Anglia 37: 382- 392.
Emonds, Joseph. 1971. "The derived nominals, gerunds and participles in Chaucer's
English." In Braj B. Kachru, Robert B. Lees, Yakov Malkiel, Angelina Pietrangeli, and
Sol Saporta (eds), Issues in Linguistics: Papers in Honor of René Kahane, 185-198.
Urbana: University of Illinois Press.
Emonds, Joseph. 1985. A Unified Theory of Syntactic Categories. Dordrecht: Foris.
Erman, Britt and Ulla-Britt Kotsinas. 1993. "Pragmaticalization: The Case of ba' and you
know." Studier i Modern Sprakvetenskap. Acta Universitatis Stockholmiensis, New
Series 10, 76-93.
Ernst, Thomas Boyden. 1984. Toward an Integrated Theory of Adverb Position in English.
Indiana University Linguistics Club.
Evans, Nick. 1988. "Odd topic marking in Kayardild." In Peter Austin (ed.), Complex
Sentences in Australian languages, 219-266. Amsterdam/Philadelphia: John
Benjamins.
Ferrara, Kathleen. 1997. "Form and function of the discourse marker anyway: Implica­
tions for discourse analysis." Linguistics 35: 343-378.
Fischer, Olga. 1992. "Syntax." In Norman Blake (ed.), The Cambridge History of the
English Language: Vol. II 1066-1476, 207-408. Cambridge, UK: Cambridge Univer­
sity Press.
Fraser, Bruce. 1988. "Types of English discourse markers." Acta Linguistica Hungarica
38:19-33.
Givón, T. 1971. "Historical syntax and historical morphology: An archeologist's field
trip." Papers from the Chicago Linguistic Society 7: 394-415.
. 1979. On Understanding Grammar. New York: Academic Press.
. 1991. "Serial verbs and the mental reality of 'Event': Grammatical vs. cognitive
packaging." In Traugott and Heine (eds), Vol. I: 81-127.
Greenberg, Joseph H. 1991. "The last stages of grammatical elements: Contrastive and
expansive desemanticization." In Traugott and Heine (eds), Vol. I: 301-314.
Haegeman, Liliane. 1994[1991]. Introduction to Government and Binding Theory. Oxford:
Blackwell.
Hagège, Claude. 1993. The Language Builder: An Essay on the Human Signature in
Linguistic Morphogenesis. Amsterdam/Philadelphia: John Benjamins.
Hamming, Richard W. 1980. Coding and Information Theory. Englewood Cliffs, NJ:
Prentice-Hall.
Hanson, Kristin. 1987. "On subjectivity and the history of epistemic expressions in
English." Papers from the Chicago Linguistic Society 15: 23-52.
Structural scope expansion and grammaticalization 269

Harris, Alice C. and Lyle Campbell. 1995. Historical Syntax in Cross-Linguistic Perspec­
tive. Cambridge, UK: Cambridge University Press.
Heine, Bernd, and Mechthild Reh. 1984. Grammaticalization and Reanalysis in African
Languages. Hamburg: Helmut Buske.
Heine, Bernd, Ulrike Claudi, and Friederike Hünnemeyer. 1991. Grammaticalization: A
Conceptual Framework. Chicago: University of Chicago Press.
Heine, Bernd, Tom Güldemann, Christa Kilian-Hatz, Donald A. Lessau, Heinz Roberg,
Mathias Schladt, and Thomas Stolz. 1993. Conceptual Shift: A Lexicon of Grammati­
calization Processes in African Language. Afrikanistische Arbeitspapiere 34/35. Uni­
versity of Cologne: Institut für Afrikanistik.
Hopper, Paul J. and Elizabeth Closs Traugott. 1993. Grammaticalization. Cambridge, UK:
Cambridge University Press.
Houston, Ann. 1989. "The English gerund: Syntactic change and discourse function." In
Ralph W. Fasold and Deborah Schiffrin (eds), Language Change and Variation, 173-
196. Amsterdam/Philadelphia: John Benjamins.
Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, MA:
MIT Press.
. 1977. X-bar Syntax. A Study of Phrase Structure. Cambridge, MA: MIT Press.
Janda, Richard D. 1980. "On the decline of declensional systems: The overall loss of OE
nominal case inflections and the ME reanalysis of -es as his." In Elizabeth Closs
Traugott, Rebecca Labrum and Susan Shepherd (eds), Papers from the 4th International
Conference on Historical Linguistics, 243-252.
. 1995. "From agreement affix to subject 'Clitic' — and Bound Root: -mos > -nos
vs. (-)nos(-) and nos-otros in New Mexican and other Regional Spanish Dialects."
Papers from the Chicago Linguistic Society 31: 118-139.
. 1996. "The alledged unidirectionality of grammaticalization vs. the discontinuity
of language transmission." Paper delivered at the LSA Annual meeting, San Diego.
Jeffers, Robert J. and Arnold M. Zwicky. 1980. "The evolution of clitics." In Elizabeth
Closs Traugott, Rebecca Labrum and Susan Shepherd (eds), Papers from the 4th
International Conference on Historical Linguistics, 221-231. Amsterdam/Philadelphia:
John Benjamins.
Joseph, Brian D. and Richard D. Janda. 1994. "The how and why of diachronic morpholo-
gization and demorphologization." In Michael Hammond and Michael Noonan (eds),
Theoretical Morphology: Approaches in Modern Linguistics, 193-210. San Diego:
Academic Press.
Jucker, Andreas (ed.). 1995. Historical Pragmatics. Amsterdam: John Benjamins.
Kemenade, Ans van. 1987. Syntactic Case and Morphological Case in the History of
English. Dordrecht: Foris.
Kroch, Anthony S. 1989a. "Function and grammar in the history of English; Periphrastic
DO." In Ralph W. Fasold and Deborah Schiffrin (eds). Language Change and Varia­
tion, 134-172. Amsterdam: John Benjamins.
. 1989b. "Reflexes of grammar in patterns of language change." Language Varia­
tion and Change 1:199-244.
Langacker, Ronald W. 1990. "Subjectification." Cognitive Linguistics 1: 5-38.
270 Whitney Tabor and Elizabeth Closs Traugott

Lees, Robert B. 1960. The Grammar of English Nominalizations. The Hague: Mouton.
(Also supplement to International Journal of American Linguistics 12).
Lehmann, Christian. 1988. "Toward a typology of clause linkage." In John Haiman and
Sandra Thompson (eds), Clause Combining in Grammar and Discourse, 181-225.
Amsterdam/Philadelphia: John Benjamins.
Lehmann, Christian. 1995[1982]. Thoughts on Grammaticalization. Munich: LINCOM
EUROPA (originally published as Thoughts on Grammaticalization: A Programmatic
Sketch, Vol 1. University of Cologne: Arbeiten des Kölner Universalienprojekts 49).
Langacker, Ronald. W. 1977. "Syntactic reanalysis." In Charles N. Li (ed.), Mechanisms of
Syntactic Change, 57-139. Austin: University of Texas Press.
Lightfoot, David. 1979. Principles of Diachronic Syntax. Cambridge, UK: Cambridge
University Press.
. 1991. How to Set Parameters: Arguments from Language Change. Cambridge,
MA: MIT Press.
Kurylowicz, Jerzy. 1965. "The evolution of grammatical categories." Repr. in Jerzy
Kuryłowicz, 1976, Esquisses linguistiques, Vol. II: 38-54. Munich: Fink.
McCawley, James D. 1988. The Syntactic Phenomena of English. Chicago: University of
Chicago Press, Vol. II.
McConnell-Ginet, Sally. 1982. "Adverbs and logical form: A linguistically realistic
theory." Language 58: 144-187.
Mendoza-Denton, Norma. 1994."Concerning-NPconstructions." MS, Stanford University.
Michaelis, Laura. 1993. "'Continuity' within three scalar models: The polysemy of adver­
bial Still" Journal of Semantics 10: 193-237.
Mitchell, Bruce. 1985. Old English Syntax. Oxford: Clarendon, II Vols.
Nichols, Johanna and Alan Timberlake. 1991. "Grammaticalization as retextualization." In
Traugott and Heine (eds), Vol. I: 129-146.
Onodera, Noriko Okada. 1995. "Diachronic analysis of Japanese discourse markers." In
Jucker (ed.), 393-437.
Pagliuca, William. 1994. "Introduction." In Pagliuca (ed.), ix-xx.
Pagliuca, William (ed.). 1994. Perspectives on Grammaticalization. Amsterdam/Philadel­
phia: John Benjamins.
Pintzuk, Susan. 1991. Phrase Structures in Competition: Variation and Change in Old
English Word Order. Unpublished Ph.D. dissertation, University of Pennsylvania.
Poutsma, Hendrick. 1923. The Infinitive, the Gerund and the Participles of the English
Verb. Groningen: Noordhoff.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik. A Comprehensive
Grammar of the English Language. London: Longman.
Radford, Andrew. 1988. Transformational Grammar: A First Course. Cambridge, UK:
Cambridge University Press.
Ramat, Paolo. 1992. "Thoughts on degrammaticalization." Linguistics 30:549-560.
Reinhart, Tanya. 1981. "Definite NP anaphora and C-command domains." Linguistic In­
quiry 12: 605-635.
Rissanen, Matti, Merja Kytö and Minna Palander-Collin (eds). 1993. Early English in the
Computer Age: Explorations through the Helsinki Corpus. Berlin: Mouton de Gruyter.
Roberts, Ian G. 1993. "A formal account of grammaticalization in the history of Romance
futures." Folia Linguistica Historica 13: 219-258.
Structural scope expansion and grammaticalization 271

Schiffrin, Deborah. 1987. Discourse Markers. Cambridge, UK: Cambridge University

Press.
Schwenter, Scott A. and Elizabeth Closs Traugott. 1995. "The semantic and pragmatic
development of substitutive complex prepositions in English." In Jucker (ed.), 243-273.
Slobin, Dan I. 1977. "Language change in childhood and in history." In John MacNamara
(ed.), Language Learning and Thought, 185-214. New York: Academic Press.
. 1994. "Talking perfectly: Discourse origins of the present perfect." In Pagliuca
(ed.), 119-133.
Steele, Susan. 1977. "Clisis and diachrony." In Charles N. Li (ed.), Mechanisms of
Syntactic Change, 539-579. Austin: University of Texas Press.
Swan, Toril. 1988. Sentence Adverbials in English: A Synchronic and Diachronic Investi­
gation. Troms0-Studier i Sprâkvitenskap X. Oslo: Novus.
Sweetser, Eve E. 1990. From Etymology to Pragmatics: Metaphorical and Cultural
Aspects of Semantic Structure. Cambridge, UK: Cambridge University Press.
Tabor, Whitney. 1992. "Reanalysis is early." Paper presented at the Annual Meeting of
the Linguistic Society of America, Philadelphia.
. 1993. "Rule frequency and syntactic innovation." Paper presented at the Annual
Meeting of the Linguistic Society of America, Los Angeles.
. 1994a. Syntactic Innovation: A Connectionist Model. Unpublished Ph.D. disserta­
tion, Stanford University.
. 1994b. "The gradual development of degree modifier Sort of and Kind of. A corpus
proximity model." Papers from the Chicago Linguistic Society 29: 451-65.
Tajima, Matsuji. 1985. The Development of the Gerund in Middle English. Tokyo:
Nan'un-do.
Traugott, Elizabeth Closs. 1982. "From propositional to textual and expressive meanings;
Some semantic-pragmatic aspects of grammaticalization." In Winfred P. Lehmann and
Yakov Malkiel (eds), Perspectives on Historical Linguistics, 245-271. Amsterdam/
Philadelphia: John Benjamins.
. 1988. "Is internal semantic-pragmatic reconstruction possible?" In Caroline
Duncan-Rose and Theo Vennemann (eds), Rhetorica, Phonologica, Syntactica: A
Festschrift for Robert P. Stockwell from his Friends and Colleagues, 128-144. New
York: Routledge.
. 1995. "The role of the development of discourse markers in a theory of grammati­
calization." Paper given at the International Conference on Historical Linguistics 12,
Manchester.
. Forthc. "Constructions in grammaticalization." In Richard D. Janda and Brian D.
Joseph (eds), A Handbook of Historical Linguistics. Oxford: Blackwell.
Traugott, Elizabeth Closs and Bernd Heine (eds). 1991. Approaches to Grammaticaliza­
tion. Amsterdam/Philadelphia: John Benjamins, II Vols.
Traugott, Elizabeth Closs and Ekkehard König. 1991. "The semantics-pragmatics of gram­
maticalization revisited." In Traugott and Heine (eds), Vol. I: 189-218.
Vincent, Diane, Sebastião Votre, Marty LaForest. 1993. "Grammaticalisation et post-
grammaticalisation". Langues et Linguistique 19: 71-103.
Visser, F. Th. 1966. An Historical Syntax of the English Language, Part II. Leiden: E.J.
Brill.
272 Whitney Tabor and Elizabeth Closs Traugott

Warner, Anthony. 1990. English Auxiliaries: Structure and History. Cambridge, UK:
Cambridge University Press.
Wasow, Thomas and Thomas Roeper. 1972. "On the subject of gerunds." Foundations of
Language 8: 44-61.
Wright, Joseph and Elizabeth M. Wright. 1928[1908]. Old English Grammar. London:
Oxford University Press.
Wyld, Henry C. K. 1953. A History of Modern Colloquial English. Blackwell: Oxford:
Blackwell (repr. with additions; first ed. 1920).
Zwicky, Arnold M. 1987. "Suppressing the Zs." Journal of Linguistics 23: 122-148.
 On the application of the notion
of grammaticalization to
West African Pidgin English

Barbara Turchetta
University of Viterbo

1. The relevance of pidgins in the study of grammaticalization

West African Pidgin English (WAPE) is a group of different varieties of a

pidgin spoken in a wide area in West Africa, within the boundaries of three
countries: Cameroon, Nigeria and Ghana.
According to the development scale given by Mühlhäusler (1986), east­
ern varieties of the language have been shown to be in the expansion stage,
while western varieties spoken in Ghana can be considered to be in the
previous stage of stabilization (Turchetta 1996).
The instability of western varieties of this pidgin constitutes an interest­
ing field of research in the origin of the grammaticalization phenomena. In
fact, grammaticalization processes are highly productive in all contact lan­
guages, where language change seems to be faster in time than in other
natural languages.
It is well known that both syntactic and semantic roles of elements
undergoing grammaticalization processes are not clearly defined, even when
different stages of a linguistic chain synchronically coexist in a linguistic
system. Because of this, a consideration of the pragmatic context is often
needed in the interpretation of such roles, especially when the analysis of
linguistic data is done through a corpus of oral language.
Non-expanded and unstable pidgins are characterized by the absence of
systematic codification, especially in the areas of the Tense/Aspect/Modality
system, morphosyntactic markers, complementation and subordination. The
274 Barbara Turchetta

lack or instability of these features makes linguistic and extralinguistic con­

text highly relevant in the interpretation of utterances, especially when the
observed language is subject to the conditions identified by Givón (1989:
248) as belonging to the pre-grammatical mode, characterized by a higher
context dependence and a pragmatic use of word order.
Grammatical properties of the pre-grammatical mode can still be found
in western varieties of WAPE, which are characterized by a near absence of
grammatical morphology, simple/conjoined syntactic constructions, prag­
matic use of word order and a high degree of context dependence.
It is possible that the need for functional morphemes in a pidgin for the
expression of grammatical concepts is the main source of grammaticalization
processes in these kinds of languages. Grammaticalization processes in pidgins
probably originated from a basically pragmatic need to reach a satisfactory
level of adequacy in language, by speakers who need to express themselves in
a given context, according to their communication needs. In regard to this, we
could state that grammaticalization is the result of speakers' needs, rather than
being simply an effect of language change due to internal factors of the
language system itself, as can happen in other natural languages. We could
consider this phenomenon as a way to fill in the gaps in the emerging grammar.
This would explain better the several linguistic chains which are still observ­
able in WAPE, where different functions for the same word still coexist on a
synchronic level, especially in the Tense/Aspect/Modality system of the verb.
One of the characteristics of grammaticalization chains is the violation of
the principle of 'one form, one meaning' (Mühlhäusler 1986: 151; Hopper
and Traugott 1993: 71, 99), which has been shown to be systematically
violated in pidgins, where a single linguistic item can fulfill different gram­
matical functions. In that sense, linguistic analysis can point up some of the
pragmatic principles, such as those of iconic order and discursive preference
in narrative sequence (Givón 1989: 104; Hopper and Traugott 1993: 51),
which have been shown to be at work in linguistic execution and might
contribute to generate grammaticalization chains.
On the basis of data from oral conversations in West African Pidgin
English, I will investigate some of the phenomena linked to the origin and
development of grammaticalization processes, with a view to answering the
following questions:
1. To what extent can we define a grammatical morpheme as being the result
of a grammaticalization process when it seems to have originated in a func-
Grammaticalization in West African Pidgin English 275

tional split rather than in a grammaticalization chain, either synchronically or

diachronically identifiable, and in what terms can we speak of change in
linguistic categories when a grammatical chain is synchronically present in
the system?
2. In what sense can the pragmatic analysis of oral discourse contribute to an
explanation of the origin of grammaticalization processes?
Grammaticalization processes are constantly active in all natural lan­
guages and seem to be highly productive in contact languages, where devel­
opments proceed more rapidly. It is because of this fact that there is a need to
consider sudden developments in pidgin languages as not following the same
steps as slower and more gradual grammaticalization processes in languages
with a long history. In fact, together with intrinsic factors we have to consider
external influences from native languages where a pidgin is spoken as vehicu­
lar language. Interference and borrowing phenomena can certainly contribute
to some grammatical developments, perhaps more significantly so than the
lexifier language. A typical example in this sense is the development of serial
constructions in most Atlantic pidgins and creoles (Alleyne 1980), whereas
lexifier European languages do not show the same high frequency of this
verbal construction. It is important to remember that in this case there might
be influences from many languages of the western branches of the Niger
Congo language family, where serial constructions are abundant (Bamgbose
1973, 1974; Hyman 1971; Sebba 1987; Weimers 1973).
A serious difficulty in the interpretation of the origin of grammaticaliza­
tion processes in a pidgin language is given by the evaluation of the linguistic
item that is undergoing the process, but was originally present in the lexifier
language system with a different grammatical function. We can give some
examples of the use of two verbs which retain their full semantic content
while also fulfilling grammatical functions: go and kam, an irrealis marker
and a tense marker respectively. For both of them we face a problem of
interpretation for the development of the grammaticalization chains they
seem to have generated (see § 2.1.).
276 Barbara Turchetta

2. The quest for grammar

Categorization of linguistic items is the most important aim of descriptive

grammar. When describing a language it is always very important to catego­
rize appropriately the linguistic data we observe, with the intention of reach­
ing a reliable level of description of the language we are working on.
Naturally, categorization is considerably easier in the case of a language with
a written tradition or with a long tradition in the standardization of at least one
of its varieties. Because of the long tradition of research in Indo-European
languages, it is always difficult to work on languages which are typologically
close to them but do not precisely match all their linguistic categories.
Linguistic categories drawn from the Indo-European tradition are widely
used in descriptions of all the languages of the world. The category Adjective
in Niger-Congo languages is notoriously controversial (Pulleyblank 1987:
966). It is also not yet clear if the existence of auxiliaries is a unique category
in the different languages of the world (Heine 1993: 9-10).
Things become even more complicated when talking about contact lan­
guages, such as pidgins, where the lexifier language happens to be a European
language of colonial heritage. Grammatical categories of pidgins easily be­
come a speculum of grammatical categories of European languages, and
anything that does not conform to the model can be easily considered as
inadequate. In fact, it is true that most pidgins show a reduced number of
grammatical markers, compared to the markers found in the lexifier lan­
guages, but it is also true that this lack is mostly due to the lack of grammatical
bound morphemes in lexifier languages of European heritage. If we consider
the Tense/Aspect/Modality system of English, for example, we find many
grammatical morphemes each with a different semantic content, whereas
comparing that system to the one we have in WAPE we find the latter to have
a decidedly smaller number of grammatical markers. Many of the items we
have in WAPE for the Tense/Aspect/Modality system of the verb are actually
the product of a grammaticalization process and are still undergoing similar
processes, though, as we will see later, it is not possible to consider each of
them as the effect of a single grammaticalization chain, consecutively devel­
oped.
On the other hand, typical examples of diachronically visible grammati­
calization chains are given by the past tense marker bin and the aspect
completive marker don, the latter mostly present in eastern varieties of the
Grammaticalization in West African Pidgin English 277

language. Both are the result of a grammaticalization process, one that

probably started in the 19th century when the language was still a nautical
pidgin and was still spoken in restricted linguistic domains. Evidence of this
process is found in trade documents of that time, as can be seen from
examples such as the following, taken from the diary of an African slave
trader in West Africa:
"Wee 3 go on bord for ask him and his answer be & say he will not stay
for us River soon (...)" (Forde 1956: 81; 12.3.1785) "We three go aboard his
ship to ask him and his answer was that he was not going to stay in our river."
"ƒ have send pound with Esin to give Combesboch for get 8 slav to pay
captain Osatam so I done pay Osatam for all I owe." (Forde 1956: 86;
21.6.1785) "I sent a pound with Esin to give Combesboch to get eight slaves
to pay Captain Osatam. So I paid Osatam for all I owe him."

2.1. Grammatical categories and grammaticalization

Many of the items undergoing grammaticalization processes cannot be attrib­

uted to a specific linguistic category, especially when they are isolated from
their linguistic context. In fact lack of grammatical morphology, a condition
which is prominent in pidgins, prevents one from identifying the semantic
content of a word like tok in WAPE, and indicating whether it is a verb or a
noun.
(1) I bin tok
3s COMP N/V
'there has been a speech/quarrel' / 'he/she has talked' (Personal
data)
Polysemy is found in a high percentage of the basic vocabulary of
pidgins. Many items undergo processes of semantic extension, as in the case
of WAPE wata, which means both 'water' and 'river' or 'lake'. The idea that
all senses of a polysemous item must share a core meaning is a consequence
of the classical definition of "category", implying a set of necessary and
sufficient conditions for membership within a category. In this view, two
words in any language, such as talk and talk in English, being a noun or the
present tense of a verb, cannot be considered as to be a single polysemous
item because they belong to two different linguistic categories. The condi­
tions for membership from a formal point of view might be stated as follows
(Taylor 1989):
278 Barbara Turchetta

1. The item must undergo the same morphological rules as the other
members of the category;
2. The item must be paradigmatically interchangeable with the other
members of the same category in a given syntactic context;
3. The item must be syntagmatically dependent on other items occurring
in the syntactic sequence to the same degree as any other item belonging to the
same grammatical category.
Many WAPE words may be either verbs or nouns, and may have full
semantic content, but are undergoing grammaticalization processes and do
not fulfill all of these conditions.
Because of the lack of morphology in WAPE it is not possible to test the
first statement. However, a very interesting counterexample to statements
number 2 and 3 is provided by two WAPE locatives, hia and de. Both of them
belong to the same category, both being spatial deictics, but they are not
paradigmatically interchangeable. In fact, de is undergoing grammaticali-
tazion and can occasionally have a different syntactic behaviour even when
functioning as a locative adverb:

(2) I kam fo hia

3s come PREP LOC
'He comes here.'
(3) I go tok de
3s FUT speak LOC
'He will speak there.'
Locative de can be used either with or without the preposition, while
locative hia can only be used in a prepositional phrase. These two examples
show at least two important things: the two locatives are not interchangeable
in the same syntactic context; they have different syntactic constituents, the
first being necessarily in a prepositional phrase and the latter either in a verbal
phrase or in a prepositional phrase. Moreover this locative de has gained new
grammatical functions, which suggest an interesting field of investigation for
the origin and boundaries of grammaticalization processes. If the two loca­
tives are still sharing the same function while showing a different syntactic
behaviour, we can state that the first condition for a grammaticalization
process to start is the loss of constituents eventually predictable for the
category.
Grammaticalization in West African Pidgin English 279

When considering grammaticalization as a process working through loss

of meaning and gain of grammatical functions it is not easy to shape and
determine the boundary between a grammatical shift, implying a loss of
semantic content, and a functional split not implying it.
As has already been shown for many languages, metaphorical conceptu­
alization plays a central role in the rise of a new grammatical morpheme and
may have an important role in functional splits too, as we will try to show for
our language.
One of the less clear metaphorical processes in grammaticalization is the
one linked to the semantic base of the metaphor given by Claudi and Heine
(1986: 308) as "PROCESS IS SPACE". The two authors give evidence from
Ewe for this metaphor, where actions are conceived as locations, for example:
(4) é -le dɔ gbɔ
3s-be at work at
'He/she is at work.'
It has been noticed that in a group of English related pidgins and creoles,
including WAPE, there is a progressive marker de derived from the English
'there'. In our case, the progressive marker is certainly a grammaticalization
of the locative we have seen in example (3). In fact, Holm (1988) has pointed
out the existence of a relation between progressive aspect markers and
locatives in many Creole languages. The WAPE locative de has three syntac­
tic positions: after the first NP (which is always to be considered in WAPE as
the subject, unless there is a NP object focalization construction), after the
preposition fo (which can also be a complementizer in different syntactic
contexts), and after a verb as a locative but without the preposition. When it is
after the first NP it can be either an existential marker (EM) or a progressive/
durative aspect marker (AM); if it is in the first position, or the last one in a
sentence, it has to be considered as a tense adverb (TADV). All possible
constructions are given in the examples below (personal data):

(5) i wok fo de
3s work PREP LOC
'He works there'
(6) i de fo de
3 s EM Prep LOC
'He is from there.'
280 Barbara Turchetta

(7) De taim yu kom yu se fud no de yu no go chop

ART time 2s corne 2s say food NEG EM 2S NEG go eat
'If you just got there and there was no food, you would not eat'.
trobul de, yu go adasaid fo bush, yu go wid de ol
trouble EM 2S go somewhere PREP forest 2s FM weed TADV ADJ
'It would be a problem, so you would go somewhere into the forest
to gather food'.
ledi i se wi de wid fo her.
lady 3s say 1PL AM weed prep 3s
There you would find a lady who would say that we were gathering
for her.'
All these functions are given for the locative 'there' in Heine et al. (1993:
219-222), together with other functions for the same item, except for the
conditions given in (7), line 3. The aspect marker function actually seems to
be part of the functions given in the same lexicon (Heine et al. 1993: 62),
under the cluster 'copula locative', as a durative/progressive aspect marker.
This copula locative is given in conjunction with a locative preposition as in
the example given for Diola:
(8) burɔk n -εn di bɔ.
work 1s-be in it
T am working.'
The word de shows the development of at least two different functional
splits. Another way to observe this phenomenon is by investigating the
metaphor to see if it applies to a functional split as much as it does to a process
of grammaticalization. If it is the metaphor that is the moving force for the first
phases of a grammaticalization chain, where the lexical semantic content is
still visible, it is plausible to think that it is the metaphor again that is the
moving force in functional splits, either when they occur in the last phases of
a grammaticalization chain or when they start from a linguistic item which
does not belong to the lexicon of a language.
Let us consider again example (7). The only way we have to explain de in
this context is through a functional split from a locative to a new function of
copula, motivated by a metaphor such as: "TO BE SOMEWHERE IS TO
EXIST". But this statement is in contrast with the one given previously:
"PROCESS IS SPACE". In fact, if something or someone is statically some-
Grammaticalization in West African Pidgin English 281

where, it or he cannot be at the same time moving in a process. This would

explain from a semantic point of view the difference of meaning between
example (6) and line 3 of example (7), showing clearly that there is no
connection between the functional split of the locative as a progressive marker
and the one of the same locative as a stative verb or copula.
The locative, then, is going through two processes of functional change.
As we have already seen, this morpheme has at least three syntactic distribu­
tions and three grammatical functions. Moreover, it does not fulfill the three
conditions defining central members of a category and its transformation is
mainly due to a metaphorical conceptualization.
According to the general theory given by Heine et al. (1991), grammati­
calization processes originate from metaphorical processes and cause seman­
tic and syntactic changes in the item involved. As we have seen here,
functional splits, though not originally motivated by a semantic loss of a
lexical item, undergo a similar process which is caused by the same forces.
We will now consider two grammaticalization processes in their syntactic
and semantic evolution and will then consider them together with the func­
tional split we have already seen, to observe if there are analogies in their
developmental pattern.
The lexical item go, a derivation of the English verb, shows grammatical
processes both in the lexifier language and in all the Atlantic pidgins having
English as a lexical base. It is for this reason that it is difficult to know
whether or not the process observable in WAPE had its origin in the original
system rather than in a developmental stage of the pidgin itself.
As a verb of motion, go is still widely used in western varieties of WAPE,
either isolated or in serial constructions (SV). Grammaticalization of the same
item has produced three different verbal markers — a prospective marker
indicating couterfactuality in the past (CM), a purpose modality marker
(MM), a future marker (FM). The following are examples:

(9) Fufu go kam.

fufu sv come
There will be food available.'
(10) Bifo i chop i se i go brin da famil.
before 3s eat 3s say 3s CM bring ART family
'He said he would have brought the family before eating.'
282 Barbara Turchetta

(11) I go fain som wata.

3 s MM look for some water
'It is his intention to look for some water.'
(12) I fren i go tok.
3 s friend 3 s FM speak
'His friend will speak (will give a speech).'
Another verb of motion, kam, is undergoing a grammaticalization pro­
cess as a tense marker (TM). Recent data collected by the author at the
beginning of 1996 show the same linguistic item with a new grammatical
function as a causative marker (CsM) and with a new lexical semantic
content. The following are examples:
(13) I kam stop fo hia.
3 s TM stop PREP here
'He stopped here.'
(14) I kam yu wont fud
3s CsM 2s want food
'You happen to be hungry.'
(15) Drom go kam big fud.
drum CsM bring big food
'The drum brings you a lot of food.'
Both the lexical shift and the two grammaticalization processes of kam
seem to be motivated by a metaphor.
As was noticed by Claudi and Heine (1986: 301) and Heine, Claudi and
Hunnemeyer (1991: 157), prototypical units of basic categories of conceptu­
alization show the following developmental process:
person → object → process → space → time → quality
Physical units standing for person refer to animate entities, while units
standing for object stand for inanimate entities. The sequence of concepts
implies that metaphorical changes occur from left to right. Concepts standing
for person are therefore the metaphorical vehicle to express object, space and
so on. It is interesting to notice how in our case both the lexical semantic shift
of kam as it is given in example (15) and the grammatical changes shown in
examples (13) and (14) have rendered the content of kam metaphorically
Grammaticalization in West African Pidgin English 283

applicable to an inanimate entity, moving from left to right according to the

scale of the metaphorical structure of grammaticalization given above,
though corresponding to different metaphorical conceptualizations.
The same conceptual shift is shown by the grammaticalization process of
go, though in that case no other semantic extension of the lexical content has
been noticed.
The pattern of a grammaticalization chain is not always relevant for the
explanation of expansion in a pidgin grammar. It might be more appropriate
to think in terms of nets of grammatical expansion, starting from a single item
either from the lexicon or from the already existing grammatical morphemes.
In this pidgin, functional splits of grammatical morphemes, grammaticaliza­
tion processes and semantic extensions of lexemes seem to be driven by
communicational needs and produced by way of metaphorical conceptualiza­
tions. Moreover, they do not show a single and uniform developmental
pattern, as would be found in a grammaticalization chain. It is more often the
case that they turn to new functions not necessarily linked among themselves.
The syntactic and pragmatic contexts give us the key to the interpretation
of the items undergoing grammaticalization processes. The interpretation of
oral discourse is probably the best field to understand what kind of pragmatic
needs cause grammaticalization in contact languages.

3. Grammaticalization and oral discourse

The interpretation of sentence meaning in pidgins is strictly dependent on the

linguistic context of the whole utterance. Speakers of a pidgin constantly use
pragmatic devices to disambiguate their execution and to interpret interlocu­
tors' speech. The need for pragmatic informations, such as that given by
deixis and intonation, becomes more relevant than in other natural languages
where grammatical morphology is abundant. It is for this reason that any
linguistic analysis of a pidgin will give good results only when both linguistic
and extralinguistic contexts in oral discourse are taken into consideration,
including all linguistic features of the speaker's performance.
Oral discourse represents a very interesting field for the investigation of
grammaticalization. Different styles of discourse in a language might reveal
interesting grammaticalization phenomena at the first stage. In Kwa lan­
guages, for example, special lexical items are used instead of grammatical
284 Barbara Turchetta

morphemes in ritual discourse. Personal data on some languages of this

cluster show a high degree of iconicity in the use of lexical items instead of
grammatical morphemes for the expression of tense and aspect in verbs. The
following example is from a Gonja ritual song at a public funeral. Tense
markers are normally placed before verbs and after the pronoun. In the
following example, a past tense marker has been replaced by a noun of verbal
origin (lùwé), following both the verb and the pronoun of reference:
(16) búdè be wú búmó bùmɔ'-pd 'n-lùwé
this PROG die 3PL(O) 3PL(s)-kill PL-the finished one
keni ái!
imagine that!
T h e death was everywhere and somebody killed them, imagine
that!'
The lack of grammatical morphology in pidgins, especially when they are
in the first phases of development, makes pragmatic factors extremely impor­
tant in the interpretation of meaning. Linguistic and extralinguistic contexts
play a fundamental role in interpretation of utterances, together with other
elements such as intonation, style of execution and shared knowledge.This
last factor makes the interpretation of meaning difficult when trying to analyse
pidgins out of the cultural context where they are spoken. This is even more
evident in the jargon phase of a pidgin, where reliance on the extralinguistic
context is quite strong.
Pragmatic strategies are often used in pidgins to supply lack of informa­
tion in the Tense/Aspect/Modality verbal system. Tense adverbs are often
used to locate the action. A portion of the Tense/Aspect conceptualization of
the lexifier language may enter the new language to supply lack of time and
aspect deixis, through the use of new markers which were originally part of
the lexicon. A very interesting example of this linguistic device is given in
WAPE by the occasionally occurring completive marker (CP) don, which
originated from the past participle of the English verb to do:
(17) Dem don tok so
3P CP speak like-this
They spoke in this way.'
In pidgin languages it is important to disambiguate in utterances many of
the plurifunctional grammatical items. Let us take again the example of
Grammaticalization in West African Pidgin English 285

WAPE de, which, as we saw in § 2.1, works with different grammatical

functions.
The following utterance is a section of the story collected by the author in
1996. The general meaning of this utterance is not clear in its written form,
unless we make at least some references to both the intonation and pauses of
the speaker, which are fundamental to understanding sequences of homopho-
nous words. When notations of different intonational pitches, together with
voice pauses are supplied, it is easy to single out different functions of the
morpheme de. The existential marker function of de is preceded by a rising
intonation and high pitch on the noun preceding it, while the following
locative is characterised by a falling intonation. A pause of nearly two
seconds always separates the tense adverb from the rest of the utterance. The
pause disambiguates the tense adverb and the homophonous article preceding
the following noun in line 4 (+ is equivalent to a pause of nearly one second,
--/ marks rising intonation, and —\ marks falling intonation).
------------------/-------------------------/ ----------------------------------------/-----------\-------------------/----------------------\--------------------------------------------------/

(18) Kokanensi go put da tin daun + i mekam kingrin kingrin de++

Kokanansi SV put ART thing LOC 3 S make-TRANS (onomatopoeia) TADV
'K. went to put the thing down, he played the drum and then
----------------------------------/--------------------/-------------------------------------/-----/-------\------------/
bifo i go si snek big big haus kam + hau pipo de ron gra gra ha! +
TADV 3s sv see snake big big house come how people 3P run (onomatopoeia)
before he could see the snake a very big house appeared and everyone rushed to see
it.
-----------//--------------------\-------------------//------------------\--------------------/
De snek de de kokanensi de de + don bait mi ol
ART snake EM LOC K. EM LOC NEG bite Is (onomatopoeia)
The snake was there and K. was there. Don't bite alright?
----------------------------/------------/---------------------/-------------------------/--------
de ++ de king kacham we + i kot mi hed + bat bifo i go kot mi +
TADV ART king catch-TRANS TADV 3s cut 1s head but TADV 3s FM cut 1s
Then the king took him while he was cutting my head, but before he could cut me
----------------------/----------------\
si kokanensi i bi sensibol man.
see K. 3s be clever man
he saw K. who is very clever.'

In section 2.1. we saw different functions of go in isolated sentences. If we

consider again example (7), we can see how disambiguation in meaning is
given both by diversity in syntactic distribution and deictic devices.
286 Barbara Turchetta

In observing the utterance again within a wider linguistic context (ex­

ample 19 below) we realize that the English translation we had previously
given, which reads as if it were a narration about counterfactual events in the
present, is actually a report of past events. The deictic information is given in
line three of example (19) with a tense periphrasis and in line six with the
combination of two tense adverbs. The two linguistic devices make the
utterance a report of real events, switching from past to present. The speaker is
referring to two different characters while narrating: a father on the one side
and a son on the other.
(19) (Father:) We yu pipo go a fain yu pipo i de fo de ol kondre
LOC 2P people go 1s find 2P people 3s EM PREP ART old country
'Wherever you go I can find you. He comes from an old place
a no si di pipo we yu go. (Son:) yes yes fada.
1 s NEG see ART people LOC 2P go yes yes father
I do not know the people where you go to. Yes father
De taim yu kom yu se fud no de yu no go chop
ART time 2s come 2s say food NEG EM 2S NEG go eat
When you got here you realized that there was no food and you would not have eaten.
trobul de, yu go adasaid fo bush, yu go wid de ol
trouble EM 2s go somewhere PREP forest 2s FM wit TADV ADJ
It was a problem so you went somewhere into the forest to gather food. There an old
ledi i se wi de wit fo her faia.
lady 3s say 1PL AM wit PREP 3s fire
lady said that we had gathered for her.
I kol as fo de kitchen giv as fud + wi kuk + wi chop + + den nau
3s call 1p PREP ART kitchen give 1P food 1P cook 1P eat TADV TADV
She called us into the kitchen and gave us some food. We cooked and ate. But now that
wi de kam haus+giv as de drom se if i kam yu wont fud hit de drom.
1P AM come home give 1P ART drum that if 3 S CSM 2S want food play ART drum
we are coming back home, give us the drum so that if you happen to be hungry you
will play it.'

Probably most of the Tense/Aspect/Modality markers of the verbal system in

WAPE have their origin in similar pragmatic strategies. Plurifunctionality of
most of the morphemes at work in the space and time deixis in this pidgin is a
potential basis for new developments of grammaticalization processes. Dis­
ambiguation in function and meaning is still done by speakers through prag­
matic devices. The end of the stabilization process is likely to lead to a more
stabilized developmental pattern of grammatical functions, as is observable in
Grammaticalization in West African Pidgin English 287

eastern varieties of this language, which are evidently less dependent on

context for the disambiguation of meaning.

4. Conclusion

In considering the high percentage of grammaticalization processes in WAPE

on the one hand, and the many grammaticalization processes already com­
pleted in the Tense/Aspect/Modality system of this contact language on the
other, it is plausible that speakers' communication needs cause the process to
start when other extralinguistic factors, such as intonation in oral speech, are
not sufficient to disambiguate the utterance.
Functional splits of grammatical elements originally present in the
lexifier language show processes of development that are quite similar in their
metaphorical conceptualizations to the ones observable in grammaticalization
processes. A single development chain is not always predictable in functional
splits, as has been shown here for locative de, undergoing two developmental
processes. An analogy can be made with the grammaticalization process of a
lexical item {kam) showing a similar developmental pattern in the absence of
unidirectionality.
It is reasonable to conclude that the pattern of a grammaticalization chain
in this language would force the explanation of a phenomenon that would be
better explained in terms of nets of grammatical expansion rather than unidi­
rectional movements toward the target of a full grammatical stage.
The existence of more fluid mechanisms of grammatical functions like
the ones we have seen here in comparison with a more stable grammar in
expanded pidgins and creoles probably marks a difference in developmental
patterns of grammaticalization, which are definitely more systematic and
predictable in their evolution in highly structured languages.

REFERENCES

Alleyne, Mark C. 1980. Comparative Afro-American. Ann Arbor: Karoma.

Bambgbose, Ayo. 1973. "The modifying serial construction: A reply." Studies in African
Linguistics 4, 2: 207-217'.
Claudi, Ulrike and Bernd Heine. 1986. "On the metaphorical base of grammar." Studies in
288 Barbara Turchetta

Language 10, 2: 297-335.

Forde, Daryll. 1956. Efik Traders of Old Calabar. Containing the diary of Antera Duke,
an Efik slave-trading chief of the eighteenth century. London: Oxford University
Press.
Givón, T. 1989. Mind, Code, and Context. Essays in Pragmatics. Hillsdale N.J.: Lawrence
Erlbaum.
Heine, Bernd. 1993. Auxiliaries. Cognitive forces and grammaticalization. Oxford:
Oxford University Press.
Heine, Bernd, Tom Güldemann, Christa Kilian-Hatz, Donald A. Lessau, Heinz Roberg,
Mathias Schladt and Thomas Stolz. 1993. A Lexicon of Grammaticalization Processes
in African Languages. Köln: Institut für Afrikanistik [Afrikanistische Arbeitspapiere
34/35].
Heine, Bernd, Ulrike Claudi and Friederike Hünnemeyer. 1991. "From cognition to
grammar - Evidence from African languages." In Elizabeth Closs Traugott and Bernd
Heine (eds), Approaches to Grammaticalization, 149-187. Vol. 1. Amsterdam/Phila­
delphia: John Benjamins. [Typological Studies in Language 19]
Holm, John. 1988. Pidgins and Creoles. Theory and Structure. Vol. 1. Cambridge: Cam­
bridge University Press.
Hopper, Paul J. and Elizabeth Closs Traugott. 1993. Grammaticalization. Cambridge:
Cambridge University Press.
Hyman, Larry. 1971. "Consecutivization in Fe? Fe?" Journal of West African Languages
10, 2: 29-42.
Mühlhäusler, Peter. 1986. Pidgin and Creole Linguistics. London: Basil Blackwell.
Pulleyblank, Douglas. 1987. "Niger-Kordofanian languages". In B. Comrie (ed.) The
World's Major Languages, 959-1014. London and Sydney: Croom Helm.
Sebba, Mark. 1987. The Syntax of Serial Verbs: An Investigation into Serialization in
Sranan and Other Languages. Amsterdam/Philadelphia: John Benjamins. (Creole Lan­
guage Library, 2).
Taylor, John R. 1989. Linguistic Categorization. Prototypes in Linguistic Theory. Oxford:
Clarendon Press.
Turchetta, Barbara. 1996. Lingua e diversita. Multilinguismo e lingue veicolari in Africa
occidentale. Milan: Franco Angeli. (Materiali Linguistici dell'Universita di Pavia, 19).
Weimers, William. 1973. African Language Structures. Berkeley: University of California
Press.
 Language Index

A 254, 255, 258, 259, 264, 266, 276, 279,

Apennine Emilian dialects, 140 281, 284
Atlantic creoles, 275 Modern, 171
Atlantic pidgins, 275, 281 Old (OE), 5, 151, 233, 237, 238, 242,
243, 248, 249, 257
B Present Day English (PDE) 233, 236,
Banda Linda, 62, 82, 83 237
Bantu, 213 Estonian, 230
Basque, 9, 211, 213, 218, 221, 222, 223, Ewe, 212, 213, 279
226
Biblical Hebrew, 62, 64, 65, 75, 82, 83 F
Brythonic, 103 French, 111, 121, 122, 153, 164
Bulgarian, 97, 103 Modern, 5
Old, 5
C
Celtic, 100 G
Chinese, 20, 21, 22, 23, 25, 32, 33, 35, 38, German, 3, 9, 10, 50, 114, 122, 149, 153,
41,42,45,48 160, 162, 164, 173, 180, 183, 184, 217
Cantonese, 47, 48 High, 91, 92, 93, 95, 98, 148, 153, 172,
Classical, 20, 22, 23, 26, 27, 28, 29, 31, 174
32, 43, 52, 53 Germanic, 67, 93, 109, 110, 170, 173, 174,
Modern Standard, 21, 22, 23, 26, 27, 28, 180, 181, 182, 183, 185
43, 44, 52 Gonja, 284
Cuna, 213, 226 Greek, 5, 83, 189, 193, 208, 209
Ancient, 9, 60, 66, 122, 189
D Classical, 192, 193, 194
Danish, 122, 154, 156, 185, 240 Modern, 122, 193
Diola, 280
Dutch, 122, 160, 171, 173, 174, 179, 183, H
184, 217 Hittite, 9, 62, 65, 67, 82, 83, 189, 190, 191,
195, 196, 197, 198, 199, 200, 201, 202,
E 203, 204, 205, 206, 207, 208, 209
English, 2, 3, 5, 6, 7, 23, 40, 45, 77, 90, 91, Hmong, 20, 42, 45, 46, 48, 54
111, 115, 135, 151, 160, 162, 167, 171, Hungarian, 9, 211, 217, 218, 220, 221
172, 173, 174, 175, 180, 182, 183, 184,
185, 197, 208, 213, 214, 217, 225, 230, I
231, 232, 233, 234, 236, 240, 244, 245, Indo-European, 65, 66, 110, 190, 191, 195
290 Language Index

Irish, 115 R
Italian, 2, 3, 6, 75, 113, 115, 119, 121, 122, Regional Spanish, 117
124, 135, 136, 139 Romance, 75, 108, 111, 112, 113, 118,
119, 120, 123, 138,208
J Rumanian, 111, 123
Jabêm, 20, 33, 34, 35, 36, 38 Russian, 152, 153, 217
Japanese, 41, 48, 231 S
Sardinian, 261
K Scandinavian, 154, 183, 184
Khmer, 20, 33, 36, 37, 38 Solomons Pidgin, 7
Kono, 212 Spanish, 2, 9, 111, 119, 170, 211, 213,
Kwa, 84, 283 214, 215, 216, 217, 219, 221, 230, 231
Sranan, 7
L Swedish, 160, 171, 183, 184, 185
Lahu, 93 Swiss dialects, 116
Latin, 2, 5, 6, 67, 85, 100, 103, 110, 112,
113, 114, 117, 118, 121, 122, 123, 124, T
138, 189, 197, 218, 219, 220, 261 Tabor, 262
Liu, 43 Takashima, 43
Lucchese, 8, 129, 130, 137, 138, 139, 140, Tamil, 150, 159, 160, 168, 169, 172
143 Terraba, 213
Lunigianese, 140 Thai, 20, 42, 46, 48
Tibeto-Burman, 84
M
Malay, 45 V
Mayan, 256 Vietnamese, 20, 40, 41, 42, 46, 47, 48, 50,
MHG, 5, 95 54

N W
NHG, 5, 95 Walser dialects, 116, 117
Niger Congo languages, 275, 276 South Alemannic, 116
Norse, 154 Weining Miao, 48
Norwegian, 185, 240 Welsh, 100
Medieval, 103
O West African Pidgin English (WAPE), 7,
Old Church Slavonic, 97 10, 273, 276, 278, 279, 281, 284, 285,
Old Indic, 189, 191, 193,208 286, 287
Olgolo, 3
Y
P Yoruba, 20, 33, 34, 36, 38
Portuguese, 117, 231
Z
Zande, 212
 Name Index

A
Abney, 240, 243 C
Adams, 39 Campbell, 7, 230, 260, 263
Aebischer, 124 Carey, 151, 152
Aissen, 256 Castellani, 137
Allan, 39 Chao, 23, 24, 25, 26
Allen, 110, 111 Chomsky, 240, 241
Alleyne, 275 Cinque, 245
Ambrosini, 137, 139, 140 Claudi, 14, 19, 59, 84, 134, 213, 224, 226,
Andersen, 15 229, 263, 279, 282
Anttila, 120 Cloarec-Heiss, 62, 63
Awbery, 100, 103 Collinson, 98
Conklin, 40
B Corbett, 113, 123
Backhand, 182 Cotticelli, 209
Bakhtin, 169 Cravens, 75
Bakker, 18, 52 Cristofaro, 9
Bamgbose, 275 Croft, 13, 14, 17, 18, 40, 48, 49, 50, 52, 77
Banfield, 256 Cruse, 54
Baron, 263 Curme, 241
Bauer, 102
Becker, 102 D
Behaghel, 98, 99, 173 Dal Negro, 116
Bencédy, 220, 226 Dasher, 229
Benveniste, 112, 156 Davies, 231
Bickerton, 232 de Castro Campos, 175
Bisang, 7, 8, 17, 20, 33, 34, 35, 37, 38, 41, Debrunner, 66
45, 47, 52, 53, 54 Déchaîne, 34
Blakemore, 266 DeLancey, 54
Bradshaw, 35 Dempwolff, 35
Brianti, 2 Denison, 266
Brinton, 111, 112, 253 Denny, 39
Brunei, 114 Dik, 26, 62, 84
Bruyn, 7 Dixon, 3
Bühler, 132 Dover, 194, 208
Bullen, 253 Dressier, 90
Bybee, 14, 19, 37, 53, 115, 118, 119, 229, Dryer, 21
230, 259, 261, 263, 264 Ducrot, 169
292 Name Index

Duranti, 135 Hamming, 263

Hanson, 253
E Harris, 7, 263
Einenkel, 241 Hashimoto, 23
Eisenberg, 162 Haspelmath, 162, 163, 170, 172, 174, 183,
Emonds, 242, 256 185
Erbaugh, 44, 45, 53 Haudry, 65, 67
Erman, 184, 254 Heine, 6, 14, 19, 59, 84, 118, 119, 120,
Ernst, 245 122, 134, 156, 159, 212, 213, 224, 226,
Evans, 231 229, 263, 276, 279, 280, 281, 282
Hengeveld, 62
F Herring, 59, 60, 79, 150, 151, 159, 160,
Fábián, 220, 226 168, 169, 172, 180, 181, 182, 184
Ferrara, 255, 259 Holm, 279
Fiedler, 99 Hoppe, 19
Fraser, 254, 255 Hopper, 3, 6, 9, 13, 15, 17, 37, 45, 60, 77,
Friedrich, 196 78, 81, 82, 89, 90, 96, 102, 114, 115,
118, 120, 121, 122, 129, 137, 148, 149,
G 154, 160, 224, 226, 274
Gabelentz, 32, 153 Horn, 53, 172
Gaeta, 4, 9 Houston, 242
Garret, 209 Hünnemeyer, 14, 19, 59, 84, 134, 213,
Gassmann, 26, 28 224, 226, 229, 263, 282
Genetti, 60 Hyman, 275
Giacalone Ramat, 3, 6, 7, 10, 116, 119,
120, 122 I
Giannini, 8 Iljic, 24, 25, 38
Giddens, 149
Giles, 16 J
Givón, 10, 22, 59, 60, 64, 65, 79, 83, 135, Jackendoff, 240, 241, 245
142, 143, 159, 160, 166, 180, 181, 229, James, 53
230, 232, 236, 239, 259, 263, 274 Janda, 102, 117, 230, 236, 239, 240, 260
Goetze, 200, 201 Jeffers, 230
Goldberg, 14 Jespersen, 166
Görlach, 149 Jetchev, 97, 103
Graham, 30 Johnson, 136
Green, 118 Joseph, 102, 230, 260
Greenberg, 3, 39, 90, 110, 111, 212, 213, Josephson, 209
226, 230 Justus, 60, 65, 66

H K
Haas, 54 Kaufman, 19
Haegeman, 232 Keesing, 7
Hagège, 122, 231 Keller, 16, 17, 52
Haiman, 111, 149, 151, 159, 160, 165, 166 Kemenade, 265
Name Index 293

King, 103
Kiparksy, 103 M
Klausenburger, 90 Magni, 113, 124
Klein, 180 Mallinson, 123
König, 60, 73, 77, 159, 160, 161, 162, 163, Marcheschi, 137
164, 167, 170, 172, 174, 180, 182, 183, Marshall, 193, 194, 208
184, 185, 229 Matisoff, 102
Körner, 117 Matthiessen, 160
Kotsinas, 184, 254 McCawley, 245, 246, 255
Krämer, 94 McConnell-Ginet, 245
Kroch, 243, 259, 265 Meillet, 1, 6, 59, 89, 90, 107, 108, 110,
Kryukov, 22 121, 122, 148
Kuryłowicz, 59, 124, 226, 265 Mendoza-Denton, 149, 150, 266
Michaelis, 14, 15, 51, 263
L Mitchell, 238
LaForest, 254 Monteil, 66, 70, 72, 85
Lakoff, 136 Moreno Cabrera, 2, 9, 10
Lambrecht, 14, 15, 51 Morris-Jones, 103
Lang, 176, 177 Mous, 18, 52
Langacker, 230 Mühlhäusler, 273, 274
Larson, 253
Lass, 6, 107, 108, 109, 110, 123, 232, 265 N
Lausberg, 117 Neu, 199
Lazzeroni, 121 Nichols, 152, 153, 231
Lees, 240 Nieri, 143
Legge 53 Noonan, 60, 62, 63
Lehmann, 3, 6, 61, 65, 78, 83, 102, 116, Nübling, 116
149, 154, 159, 183, 184, 226, 230, 231,
239, 240, 244, 260, 261, 263, 264 O
LePage, 16 Onodera, 231
Leuschner, 10 Otten, 197, 198, 199, 200, 202, 204, 205,
Levinson, 142, 143 206
Lewis, 100
Li, 21, 22, 26, 28, 29, 30 P
Lightfoot, 15, 231 Pagliuca, 14, 115, 118, 119, 229, 230, 259,
Liu, 43, 44 261, 263, 264
Löbel, 40, 41, 46, 47, 54 Palmer, 5
Longacre, 73 Parkinson, 117
Loporcaro, 99 Paul, 93, 96, 102
Lord, 34, 60, 64, 84 Pecora, 209
Lötscher, 174 Pedersen, 100
Lunn, 75 Perkins, 14, 115, 118, 119, 229, 230, 259,
Lunt, 97 261, 263, 264
Luraghi, 9, 190, 204, 209 Pesetsky, 253
Lyons, 131, 135, 143 Peyraube, 30
294 Name Index

Pieri, 138, 143 Silva-Corvalán, 18

Pinkster, 118 Slobin, 151, 263
Pintzuk, 265 Smith, 16
Popovici, 160, 169 Sobrero, 134, 143
Poutsma, 173, 174, 175, 179, 241 Sonaiya, 34
Priebsch, 98 Soucek, 199, 200, 202, 204
Pulleyblank, 32, 53, 276 Spore, 154
Squartini, 118
Q Steele, 190, 194, 263
Qiu, 29, 30 Stein, 111, 112
Quirk, 161 Stoett, 173, 174
Stolz, 52
R Sun, 22
Rácz, 220, 226 Suppes, 49
Radford, 245 Sweetser, 163, 229
Ramat, 110, 111, 115, 118, 122, 224, 226,
252 T
Ransom, 60, 62 Tabor, 6, 230, 240, 243, 259, 262
Reh, 212, 226, 229 Tabouret-Keller, 16
Reinhart, 232 Tajima, 242, 243, 244
Renzi, 191 Takashima, 42, 43
Ricca, 122 Taylor, 277
Roberts, 231, 261 Thomason, 19
Robinson, 94 Thompson, 21, 22, 26, 28, 29, 30, 73, 160
Roeper, 240 Timberlake, 152, 153, 231
Rohlfs, 113, 124, 130 Tompa, 220, 226
Ronneberger-Sibold, 95 Traugott, 6, 10, 13, 15, 17, 19, 37, 60, 73,
Rosenkranz, 206 77, 81, 82, 89, 90, 96, 102, 115, 118,
Rosetti, 123 120, 121, 122, 129, 137, 148, 154, 159,
Ross, 24 160, 182, 224, 226, 229, 231, 248, 252,
Roulet, 169 253, 254, 258, 263, 265, 274
Trudgill, 16
S Turchetta, 10, 273
Salmons, 92, 102
Saxena, 60, 84 U
Schelling, 164, 165, 166 Uhle, 32
Schiffrin, 254
Schön, 123 V
Schwenter, 248, 252 Väänänen, 124
Schwyzer, 66 van der Auwera, 160, 183, 184
Scur, 99 Velcsov, 220, 226
Searle, 49 Vennemann, 95, 96, 97
Sebba, 275 Vincent, 108, 109, 112, 118, 123, 254
Serianni, 111, 114, 124 Visser, 162, 171, 174, 183, 242
Silva Neto, 117 Votre, 254
Name Index 295

Voyles, 98 Wright, 243, 244

Wunderlich, 164
W Wurzel, 91, 92, 93, 96, 100
Wackernagel, 189, 208 Wyld, 239
Wang Li, 28, 29, 43, 54
Wang Mingquan, 22 Y
Warner, 231 Yen, 28, 29, 30, 31,32
Wasow, 240
Weinreich, 16 Z
Wellmann, 95 Zaefferer, 162
Weimers, 275 Zahn, 35, 36
Wiese, 50, 102 Zamboni, 111
Willis, 100, 101 Zinnes, 49
Wills, 51, 54 Zürcher, 30
Wilmanns, 94 Zwicky, 51, 103, 208, 230, 266
Woodcock, 218, 219
 Subject Index

A C-command Scope Increase, 232, 234,

abduction, 16, 17 235, 240, 244, 248, 251, 253, 259, 260,
abstraction, 225 261, 262, 265
actuation, 17, 18 canonical grammaticalization, 148, 149
adposition, 212 categories, 134, 147
adverb, 147 categorization, 276
adverbial clauses, 61, 62, 163 category of number, 113
adverbial phrase, 245, 257 category oriented process, 19, 41, 42, 45,
adverbs, 136 49,50
adversitive connectives, 202 category-to-category derivations, 6
affixes, 117 causative verbs, 93
agglutination, 156 Cheshirization, 93
aktionsart, 94 class markers, 3
alternative concessive conditionals, 161, class nouns, 45, 54
163, 172, 173, 174, 181, 182 classification, 47, 48, 49, 50
analogy, 13, 15, 16, 17, 95, 96, 102, 103 classifiers, 19, 39, 40, 41, 42, 43, 44, 45,
apocope, 130 46, 47, 48, 49, 50
apophony (Ablaut), 109 clause combining, 23, 26
apparent grammaticalization, 7 cline
approaches to grammaticalization, 59 cline of grammaticality, 89
arguments, 62 cline of lexicality, 89, 90, 93
arrestive adversative coordination, 179 clitics, 9, 189, 190, 191, 194, 195, 196,
aspect marker, 280 201, 203, 204, 207, 208, 236, 237, 239
attractor positions, 20, 33, 36, 37, 38 clitic/affix developments, 118
auxiliaries, 2, 5, 118 particles, 9
auxiliary verbs, 236, 241 pronouns, 117
close-knit traditional society, 8
B closure, 165
bilingualism, 18 coding points, 135
bleaching, 147 cognition, 14, 19
borderline phenomena, 1 cognitive pattern/habit, 134
boundaries of grammaticalization, 7, 107, collocation, 2, 3, 152
118,278 complement clauses, 61, 62
complementizers, 9, 60, 61, 64, 121
C completive marker, 284
C-command, 231, 232, 233, 236, 253, Complex Personal Pronouns, 131
260, 262 complex prepositions, 149
298 Subject Index

conceptual shift, 283 demonstratives, 3, 28, 29, 30, 31

concessive conditionals, 10, 159, 160, deponents, 155
161, 162, 163, 164, 166, 169, 170, 172, desentialization, 231
173, 174, 176, 178, 179, 180, 181, 182, diachronic perspective, 138
183, 184 diachronic roots, 137
concessives, 180, 184 diachronic string comparison, 232, 234,
conclusive perfect, 111 239, 244, 252, 253, 260, 261, 262
concretion, 225 direct speech particle, 197
condensation, 230 direction of grammaticalization, 3
conditionals, 159, 160, 161, 162, 165, directional verbs, 33, 36, 37, 38
166, 173, 174, 184 directionality, 115, 224
conflicting argumentative orientations, disambiguation of meaning, 287
165 discontinuity, 116
conjunction, 120, 246, 247, 251 discourse collocation, 153
conjunctional adverbs, 164, 165, 185 discourse functions, 151
conservative mutations, 112 discourse markers, 231, 253, 254
Construction Grammar, 14, 15 dissipating grammaticalization, 9, 148,
constructions, 8, 13, 14, 15, 16, 17, 18, 153, 156
19, 20, 39, 51 divergence, 78, 82
contact languages, 273 dual, 112
contextual inferencing, 77
contiguity, 217, 225 E
continuum, 118 egocentricity, 143
implicatures ellipsis, 216
conventional, 16 embedded interrogatives, 161, 165, 166,
conversational, 16 172, 173, 174, 180, 181
conversions, 5, 6 enclitics, 189, 190, 191, 192, 193, 194,
copula, 23, 28, 29, 32, 280 195, 196, 199, 200, 201, 202, 204, 207,
correlative diptych, 66, 67, 71 208, 209
counting, 40, 41, 42, 44, 48, 49, 50 equational constructions, 20, 22, 23, 24,
coverbs, 33, 36, 37, 38 26,31
creoles, 7 exaptation, 6, 108, 109, 112, 113, 123
cultural context, 142 expansion stage, 273
expression of irrelevance, 161
D extraposition, 15
decadence of gender systems, 113
decategorialization, 78 F
decategorization, 120 factive predicates, 67, 72
deconditionalizing, 162 fluid mechanisms of grammatical
decrease in structural scope, 231 functions, 287
definiteness, 47, 130, 143 focus, 20, 23, 26, 32, 84, 85
degrammaticalization, 115, 226, 252 form/function relations, 108
degree of fusion, 37, 38 form/function units, 107
deictic function, 135 free relative clauses, 174
deictic indicators, 129 free-choice strategy, 171
Subject Index 299

free-choice pronouns, 183 identifying function, 135

functional renewal, 6, 7, 109, 111, 112 idiomatization, 122
functional splits, 279, 280, 281, 287 idioms, 214
future marker, 281 idiosyncratic changes, 121
'inchoative' suffix, 110
G increase in structural scope, 10
gender, 112, 113, 116, 117, 124, 132, 133 increase of morphological complexity,
generality, 37, 53 117
generalizing particles, 172, 174, 175 increase of structural scope, 231, 232
geographical distribution, 137 indexical function, 140
gerund, 240, 241, 242, 243, 244, 247, indexical specification, 139
249, 264 indifference, 162, 164, 174
grammar, 4, 8 individualization, 47, 48, 49, 50
grammatical categories, 277 inflected infinitive (infinitivo pessoal),
grammatical change, 229 117
grammatical expansion, 283 Initial Consonant Mutations, 100
grammatical forms, 156 innateness, 20, 51
grammatical shift, 279 innovating mutations, 112
grammaticalization, 4, 6, 8, 9, 89, 90, 91, innovation, 17, 18
92, 93, 95, 96, 97, 100, 101, 102, 118, instrumental construction, 153
119, 120, 121, 123, 147, 153, 221, 223, insubordination, 231
230, 274, 277, 282, 283 interference, 16, 17
incipient grammaticalization, 9, 148, interrogatives, 159, 161, 163, 169, 174
149, 156 intonation, 285
incomplete grammaticalization, 199 inversion, 112
interest in grammaticalization, 1 irrelevance, 161, 162, 163, 166, 168, 169,
limits of grammaticalization, 2, 8, 9, 170, 174, 181, 182
114,202,235 -isc- verbs, 111
grammaticalization chain, 110, 274, 287 isomorphism, 263
grammaticalization clines, 263 item oriented process, 19, 41, 42, 49, 50
grammaticalization feeds
lexicalization, 221 K
grammaticalization processes, 275, 278, Keller's maxim, 16, 17
281
grammaticalization scale, 264 L
language acquisition, 13, 15
H language change, 3, 4, 5, 10, 13, 123, 230
have perfect, 151 gradualness of change, 235, 238, 265,
hierarchical downgrading, 183 266
High Tone Syllable, 34 propagation of change, 13, 14, 17, 18,
historical linguistics, 1 20,51
Humboldt's maxim, 17 language contact, 7, 8, 13, 15, 16, 17, 19,
39,52
I language intertwining, 18, 52
identification, 48, 49, 50 'lateral' conversions, 5
300 Subject Index

layering, 78, 81, 82 nominalizer, 23

left sentence boundary, 206 norms, 18
lenition, 100, 101 nucleus-satellite, 160, 182
lexical autonomy, 137 number category, 112
lexical replacements, 4 numeral, 39, 50
lexical-cognitive direction, 156 numeral classifier, 213
lexicalization, 2, 6, 9, 115, 120, 121, 122,
211, 214, 215, 217, 218, 219, 221, 222, O
223, 224, 225, 226 object patterner, 21
lexicalization hierarchy, 216 one form, one meaning, 274
lexifier language, 275, 276, 284 oracle-bone inscriptions, 21, 22, 28, 42,
limits of grammaticalization, 2, 8, 9, 114, 43
202, 235 oral discourse, 10, 275, 283
linguistic and extralinguistic contexts, ordering, 50
284 oriented process of classifier develop­
linguistic areas, 7, 13, 17, 38, 39 ment, 42
local particles, 199 origin of grammaticalization processes,
locatives, 131, 133, 136, 278, 280, 281 275, 278
origo, 133
M
mathematical psychology, 49, 50 P
measuring, 49 panchrony, 264
meronomy, 41, 45, 54 paradigmatic, 13, 16, 20, 36, 37
metaphor, 9, 16, 17, 37, 77, 212, 217, parataxis, 63
223, 224, 225, 279, 280, 281, 282 parsing, 36
Metaphorical Abstraction Hierarchy, 213, participles, 219
214, 218 particles, 162, 189
Metonymical Concretion Hierarchy, 218, passive, 154, 155
220, 221 past tense marker, 276
metonymy, 5, 9, 16, 17, 212, 217, 223, pauses, 285
224, 225 pejorative, 5
mixed languages, 18, 52 perfect, 151
modal auxiliaries, 3 persistence, 78, 81
morphological change, 236 person, 134
morphological degeneration, 116 person and spatial deixis, 142
morphologization, 90, 91, 93, 94, 97, 99, person-deictic pronoun system, 130
102, 103 personal pronouns, 135
motion verbs, 119, 281, 282 phonogenesis, 114, 121, 154
mummification, 153 phonological and morphological change,
4,9
N phonological end, 3
narrow linguistic communities, 142 phonological rule, 92
negative copula, 30, 31, 32 phonological segments, 3, 114
neuter, 112, 123, 124 phonology, 153
nominalization, 23 pidgins, 7, 10 273, 274, 275, 276, 279,
Subject Index 301

281, 283, 284, 287 reduction in functionality, 109

'polyphonic' discourse, 169 referentialization, 49
polysemy, 277 reflexive particle, 198
possessive, 237, 238, 239, 240, 243, 244 reflexive pronoun, 154
postpositive, 189, 190, 208 regrammaticalization, 111
pragmatic and discourse forces, 140 relationalization, 47, 48, 49
pragmatic factors, 284 relative clauses, 20, 21, 23, 26, 27, 49,
pragmatic informations, 283 52, 174, 178
pragmatic mode, 160, 181, 184 relevance, 53
pragmatic strategies, 142, 284 renewal of function, 10
pragmatic strengthening, 147 result constructions, 70
pragmaticalization, 184 reversibility, 223
pragmatics, 10, 15, 142 Rhetorical Dialogue, 10, 160, 169, 180,
predicative instrumental, 152 182
prefixed articles, 3 rhetorical questions, 150, 151, 184
prefixes, 153 Rückumlaut, 9, 95, 96, 97, 99
preposition, 149, 150 rule inversion, 95, 96
prepositional phrase, 252
prepositive connectives, 190, 196, 203, S
208 s-passive, 155, 156
present participle, 2, 218, 220, 221 satellites, 62, 181
present perfect, 151 'scalar' concessive conditionals, 162,
principle of generality, 19, 38 163, 172
principle of relevance, 19, 37, 38 scope increase, 232, 245, 262, 263
progressive marker, 279 scope reduction, 10, 265
pronoun, 8, 120, 174, 189 semantic change, 156, 235, 236, 238, 252
prosodic criteria, 136 semantic content, 119
prospective marker, 281 semantic erosion, 138
prototypes, 44 semantic retention, 119
prototypical value, 113 semantic shift, 216
proximity, 131, 132 semantic strengthening, 182
purpose clauses, 70, 73 semiotic motivation, 91, 96
purpose modality marker, 281 semiotic remotivation, 100
sentence connectives, 203
Q serial constructions, 275
quantification, 50 serial unit, 33, 34, 35, 36, 37, 38, 39
quantifier, 54 serial verbs, 20
question-answer sequence, 10, 160, 164 shi ...de construction, 24, 25
similarity, 217, 225
R simple personal pronoun, 130
Raddoppiamento Fonosintattico, 100 small and narrow communities, 140
reanalysis, 7, 16, 17, 20, 29, 30, 31, 37, social status, 39
41,92,95,96,97, 102, 117 sociolinguistics, 15, 51
reason clauses, 73 sources of a grammatical construction,
reciprocals, 155 153
302 Subject Index

space, 134, 142, 143 topic, 20, 23, 26, 29, 32, 84, 85
spatial configuration, 131 topic continuity, 135
spatial deictics, 8, 278 topic-comment constructions, 29
spatial meaning, 7 topic-comment structure, 165
specialization, 78, 81, 82 topic-switching, 184
Sprachbünde, 7 topicalization, 194
stabilization, 273, 286 transnumerality, 39, 53
stage of alternation, 154
stile cognitivo, 134 U
strong verbs, 109, 155 Umlaut, 9, 91, 92, 93, 94, 95, 96, 98, 99,
structural scope increase, 6 102
structural scope reduction, 6 unconditionals, 162
structural unidirectionality, 231, 232, unidirectional processes, 107
235, 260, 262, 265 unidirectionality, 6, 7, 10, 115, 116, 118,
subcategory, 113 123, 184, 211, 223, 224, 226, 235, 229,
subject identity, 34, 35 260, 262, 264, 265, 287
subject pronouns, 116 univerbation, 116
subjectification, 77 universal concessive conditionals, 163,
subordinating conjunction, 182 172, 173, 174, 176, 178
subordination, 246 universals, 51
suffix -ko, 221
syntactic grammaticalization, 97 V
syntactic mode, 160, 180, 181 V/2, 181, 183, 243, 244
syntacticization, 159, 160, 161, 169, 176, verb patterner, 21
180, 181, 182 verb serialization, 19, 33, 34, 64
syntactotelic, 213 verbal markers, 281
syntagmatic, 16, 20, 36, 37 vowel deletion, 130
synthesis, 156
W
T Wackernagel's clitics, 207
taxonomy, 41, 45, 54 Wackernagel's Law, 9, 190, 191, 192,
tense markers, 282, 284 193, 194, 196
Tense/Aspect/Modality, 273, 276, 284, weak verbs, 155
286, 287 weakening, 147
third-person pronouns, 129 WH-pronouns, 162, 171, 172
time, 70, 73 word-order, 21