Valency: Theoretical,

Descriptive and Cognitive

Issues

Edited by
Thomas Herbst
Katrin Götz-Votteler

Mouton de Gruyter
Valency

≥
Trends in Linguistics
Studies and Monographs 187

Editors
Walter Bisang
(main editor for this volume)
Hans Henrich Hock
Werner Winter

Mouton de Gruyter
Berlin · New York
Valency
Theoretical, Descriptive and Cognitive Issues

edited by
Thomas Herbst
Katrin Götz-Votteler

Mouton de Gruyter
Berlin · New York
Mouton de Gruyter (formerly Mouton, The Hague)
is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.

앪
앝 Printed on acid-free paper which falls within the guidelines
of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication Data

Valency : theoretical, descriptive, and cognitive issues / edited by Tho-

mas Herbst, Katrin Götz-Votteler.
p. cm. ⫺ (Trends in linguistics. Studies and monographs ; 187)
Includes bibliographical references and index.
ISBN 978-3-11-019573-6 (hardcover : alk. paper)
1. Dependency grammar. 2. Cognitive grammar. 3. Contrastive
linguistics. 4. Computational linguistics. 5. Semantics. I. Herbst,
Thomas. II. Götz-Votteler, Katrin, 1975⫺
P162.V345 2007
415⫺dc22
2007031827

ISBN 978-3-11-019573-6
ISSN 1861-4302

Bibliographic information published by the Deutsche Nationalbibliothek

The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie;
detailed bibliographic data are available in the Internet at ⬍http://dnb.d-nb.de⬎.

” Copyright 2007 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin
All rights reserved, including those of translation into foreign languages. No part of this
book may be reproduced or transmitted in any form or by any means, electronic or mechan-
ical, including photocopy, recording or any information storage and retrieval system, with-
out permission in writing from the publisher.
Cover design: Christopher Schneider, Berlin.
Printed in Germany.
Preface: Valency – theoretical, descriptive and
cognitive issues

Thomas Herbst and Katrin Götz-Votteler

As with most other concepts in linguistics, in the discussion of valency one

must distinguish between the linguistic phenomenon of valency on the one
hand and the use of the term valency and the development of theoretical
frameworks associated with it on the other. As far as the former is con-
cerned, it is obvious that valency phenomena have been treated in linguis-
tics under a variety of different labels ranging from government or Rektion
in traditional grammar to subcategorization in generative frameworks or
comparatively neutral labels such as complementation in descriptive gram-
mars such as the Comprehensive Grammar of the English Language. Obvi-
ously, up to a point the use of different terms suggests different ways of
viewing the phenomenon in question.
The notion of valency as such is generally linked with Tesnière’s de-
pendency grammar, although similar concepts had been put forward for
example by Bühler (1934) and de Groot (1949).1 It is probably fair to say
that very significant contributions to the development of a theory of
valency have been made by German linguistics since the 1960s. It is par-
ticularly the work of Gerhard Helbig and the emergence of a number of
German valency dictionaries (Helbig and Schenkel 1969; Engel and
Schumacher 1976; VALBU 2004) that are of importance here. Both lexico-
graphically oriented and theoretical work on valency have resulted in an
extensive discussion of criteria for the distinction between complements
and adjuncts and a distinction between different types of complements with
respect to their various degrees of obligatoriness. In recent years, the term
valency has increasingly been used for the description of English, some-
times with explicit reference to the European tradition of valency theory
and the concepts and criteria developed there,2 sometimes just as a new
term for complementation phenomena.
This volume comprises articles which deal with both the theoretical
notion of valency and the analysis of valency phenomena. The articles in
the first section, theoretical and descriptive aspects of valency, discuss the
valency concept in its theoretical context (Peter Matthews) and the question
of how valency phenomena can be described most appropriately with refer-
vi Thomas Herbst and Katrin Götz-Votteler

ence to certain distinctions such as complement inventories or valency pat-

terns or semantic or syntactic valency (Thomas Herbst, Katrin Götz-
Votteler). Other papers focus on different concepts of grammaticalization
(Lene Schøsler, Dirk Noël) and particular problems of valency in syn-
chronic and diachronic descriptions (Mechthild Habermann, Michael Klotz,
Ilka Mindt). Finally, this section contains an outline of the treatment of
valency phenomena and the underlying theoretical concept in the Berkeley
FrameNet project (Charles Fillmore).
Section II focuses on the important issue of the role of valency phenom-
ena in cognitive linguistics (Gert Rickheit and Lorenz Sichelschmidt, Ru-
dolf Emons), where the acquisition of valency structures is of course a par-
ticularly important aspect (Heike Behrens).
Section III contains a number of papers with a contrastive orientation,
which ranges from descriptive issues comparing different aspects of
valency in English and German (Klaus Fischer, Irene Ickler, Brigitta Mitt-
mann) and English, German and Norwegian (Stig Johansson) to a more
pedagogically oriented account of valency errors in the performance of
German and English learners (Ian Roe).
Finally, Section IV is concerned with computational aspects of valency
analysis, where possible ways of using existing valency descriptions such
as the Valency Dictionary of English (2004) as the basis for programs of
word recognition are demonstrated (Dieter Götz, Ulrich Heid) and other
approaches towards the automatic analysis of valency structures in compu-
tational linguistics are outlined (Roland Hausser, Besim Kabashi, Günther
Görz and Bernd Ludwig).
The volume comprises papers given at a conference entitled Valency:
Valenz − Theoretical, Descriptive and Cognitive Issues held at the Fried-
rich-Alexander-Universität Erlangen-Nürnberg in April 2005, which was
supported by the Deutsche Forschungsgemeinschaft and the Dr.-Alfred-
Vinzl-Stiftung. The editors would like to thank these institutions for the
generous support they gave to the conference, Dr. Anke Beck for attending
the conference and her support of the present volume, David Heath for his
help and advice in all matters linguistic and Susen Schüller for her work on
the index. Above all, our thanks go to all participants of the conference.

Notes

1. Cf. de Groot (1949/1964: 114-115) and Matthews (1981: 117). For the history
of the concept of valency see Ágel (2000); for valency models in German lin-
guistics see Herbst, Heath, and Dederding (1980) and Helbig (1992).
 Valency – theoretical, descriptive and cognitive issues vii

2. See, e.g., Emons (1974), Allerton (1982) and VDE (Herbst et al. 2004).

References

Ágel, Vilmos
2000 Valenztheorie. Tübingen: Gunter Narr Verlag.
Allerton, David J.
1982 Valency and the English Verb. London/New York: Academic Press.
Bühler, Karl
1934 Sprachtheorie. Die Darstellungsfunktion der Sprache. Jena: Fischer
Verlag.
Engel, Ulrich, and Helmut Schumacher
1976 Kleines Valenzlexikon deutscher Verben. Tübingen: Gunter Narr
Verlag.
Emons, Rudolf
1974 Valenzen englischer Prädikatsverben. Tübingen: Max Niemeyer
Verlag.
de Groot, Albert W.
1964 Reprint. Structurele Syntaxis. The Hague: Servire. Original edition,
The Hague: Servire, 1949.
Helbig, Gerhard
1992 Probleme der Valenz- und Kasustheorie. Tübingen: Max Niemeyer
Verlag.
Helbig, Gerhard, and Wolfgang Schenkel
1969 Wörterbuch zur Valenz und Distribution deutscher Verben. Leipzig:
VEB Verlag Enzyklopädie.
Herbst, Thomas, David Heath, and Hans-Martin Dederding
1980 Grimm’s Grandchildren. Current Topics in German Linguistics.
London/New York: Longman.
Herbst, Thomas, David Heath, Ian Roe, and Dieter Götz (eds.)
2004 A Valency Dictionary of English. A Corpus-Based Analysis of the
Complementation Patterns of English Verbs, Nouns and Adjectives.
Berlin/New York: Mouton de Gruyter. [VDE]
Matthews, Peter
1981 Syntax. Cambridge: Cambridge University Press.
Quirk, Randolph, Sydney Greenbaum, Geoffrey Leech, and Jan Svartvik
1985 A Comprehensive Grammar of the English Language. London:
Longman.
Schumacher, Helmut, Jacqueline Kubczak, Renate Schmidt, and Vera de Ruiter
(eds.)
2004 VALBU – Valenzwörterbuch deutscher Verben. Tübingen: Gunter
Narr Verlag [VALBU].
Tesnière, Lucien
1959 Éléments de Syntaxe Structurale. Paris: Klincksieck.
Contents

Preface v
Valency – theoretical, descriptive and cognitive issues
Thomas Herbst and Katrin Götz-Votteler

Section 1: Theoretical and descriptive aspects of valency

The scope of valency in grammar 3

Peter Matthews

Valency complements or valency patterns? 15

Thomas Herbst

Describing semantic valency 37

Katrin Götz-Votteler

The status of valency patterns 51

Lene Schøsler

Verb valency patterns, constructions and grammaticalization 67

Dirk Noël

Aspects of a diachronic valency syntax of German 85

Mechthild Habermann

The valency of experiential and evaluative adjectives 101

Ilka Mindt

Valency rules? The case of verbs with propositional complements 117

Michael Klotz

Valency issues in FrameNet 129

Charles J. Fillmore
x Contents

Section 2: Cognitive issues and valency phenomena

Valency and cognition – a notion in transition 163

Gert Rickheit and Lorenz Sichelschmidt

Valency grammar in mind 183

Rudolf Emons

The acquisition of argument structure 193

Heike Behrens

Section 3: Contrastive aspects of valency

Valency and the errors of learners of English and German 217

Ian Roe

Temporary ambiguity of German and English term complements 229

Klaus Fischer

Sentence patterns and perspective in English and German 253

Irene Ickler

Contrasting valency in English and German 271

Brigitta Mittmann

Valency in a contrastive perspective: Structure and use 287

Stig Johansson

Section 4: Computational aspects of valency analysis

Valency and automatic syntactic and semantic analysis 309

Dieter Götz

Handling valency and coordination in Database Semantics 321

Roland Hausser
 Contents xi

Pronominal clitics and valency in Albanian: A computational 339

linguistics perspective and modelling within the LAG-Framework
Besim Kabashi

The practical use of valencies in the Erlangen speech dialogue 353

system CONALD
Günther Görz and Bernd Ludwig

Valency data for Natural Language Processing: What can the 365
Valency Dictionary of English provide?
Ulrich Heid

Subject index 383

Author index 391
Section 1
Theoretical and descriptive aspects of valency
The scope of valency in grammar
Peter Matthews

1.

Valency or valence is a term originally restricted to the syntax of verbs:

“nombre d’actants”, as Tesnière defined it in the glossary to his Éléments,
“qu’un verbe est susceptible de régir” (1959: 670). It was also linked, by
the same definition, to dependency. Thus, in Alfred parle [‘Alfred speaks’],
the verb as a governor (régissant) “commanded” the actant, Alfred, as a
subordinate term depending on it (Tesnière 1959: ch. 2.1–3). The early
development of valency theory (Valenztheorie) was therefore closely linked
with that of a dependency grammar (Dependenzgrammatik), in Germany
especially, in the 1970’s.
This line of thinking was neatly summarised in English, at the end of
that decade, by Thomas Herbst and his colleagues (1980: ch. 4). It was
obvious, however, that words of other categories could have “semantic
properties”, as I initially put it somewhat nervously, “akin to valency”
(Matthews 1981: 115). Later definitions therefore, following later usage,
have in that respect become more general. In, for example, my own dic-
tionary of linguistics, valency is of “a verb or other lexical unit” (Matthews
1997: 394). For the late Lawrence Trask, whose dictionary of grammar was
familiar to me when I chose this form of wording, the term had both a nar-
rower and a broader definition: “1. The number of arguments for which a
particular verb subcategorizes, ... 2. More generally, the subcategorization
requirements of any lexical item” (Trask 1993: 296; argument defined 20,
lexical item 158). One problem, therefore, is how far the scope of valency
should be extended. This can, if we like, be cast in terms of such a defini-
tion. The questions, that is, are what is a lexical item or a lexical unit, and
what exactly is meant by subcategorisation.
Note too, however, that while Trask’s first definition is in the main close
to Tesnière’s, it says nothing about verbs as governors or their arguments
as depending on them. Neither does the definition I gave, which refers sim-
ply to a “range of syntactic elements”, with no further stipulation of the
relations, whether implicitly of dependence or otherwise, in which they
stand. This may perhaps not quite reflect the way all linguists see things,
outside what one might be tempted to call, however spuriously, an “Anglo-
Saxon” tradition. But dependency and valency are potentially separate. To
4 Peter Matthews

say, for example, that a verb is transitive is one thing; and, if the facts are
agreed, the finding will not be controversial. To say that verbs take objects
as dependents is another, and in some accounts at least no such relation has
been posited. A second problem, therefore, is how far a link between de-
pendency and valency is justified, especially for categories other than
verbs. If X is a lexical item, and its subcategorisation either allows it or
requires it to take other units, is it always the governor, in Tesnière’s sense,
in its relation to them?

2.

The valency of atoms, as defined in chemistry, refers to their capacity to

combine with other atoms, or with groups of atoms, in the formation of
compounds. A verb then, as Tesnière perceived it, could be compared to
“une sorte d’atome crochu” (1959: ch. 97.3), which determined the number
of actants that it too can combine with. To “number” we may add “types”;
and, if units other than verbs can be similar “atoms”, there will be none
which do not in some sense allow some combinations while excluding oth-
ers. Not all definitions in linguistics stress the parallel with chemistry. But
for Crystal, who does, valency refers in general to “the number and type of
bonds which syntactic elements may form with each other” (2003: s.v.,
“syntactic elements” in capitals). If two elements, therefore, of whatever
category can combine in any specific construction, it will be because one or
the other, or perhaps both, has a valency that allows it.
By “syntactic elements” Crystal means, or seems to mean, all units that
form a constituent in a hierarchy (s.v. element). Valency in that sense,
which again is similar to valency in chemistry, would be the foundation not
just of the syntax of verbs, or of verbs and other lexical units, but of syntax
generally. In most accounts, however, its sense is narrower in two ways.
First it is a property of, more precisely, lexemes: of words, that is, as en-
tered in a lexicon or dictionary. Secondly, it has to do, again as Trask de-
fines it, with subcategorisation. Thus, in this view, it is not part of the
valency of clear that it can combine with an intensifying adverb: very clear,
quite clear and so on. This is instead a property of adjectives in general, or
of adjectives in general with specific exceptions. But adjectives in general
cannot be construed with, for example, nominal clauses: thus predicatively
in It was clear that they were coming. This is a property of a particular
subcategory of adjectives, of which clear is a member. Therefore it is part
of the valency of clear, and of every other adjective that in this respect is
like it.
 The scope of valency in grammar 5

This plainly raises problems. There are potential grounds for disagree-
ment as to how we should distinguish lexemes; as to what are categories
and what are subcategories; as to what is a subcategory and what are no
more than “exceptions”. It is now, however, still more obvious that where
the scope of valency might be disputed, issues of dependency in syntax
have no bearing on the argument one way or the other. Most linguists will
agree that in, for example, This is quite clear the intensifier depends on the
adjective. They may use other terminology: quite, for example, is subordi-
nate to clear, or clear is the head of quite clear, or the adjective is again a
governor. Many at least will also see the nominal or that-clause as depend-
ent, or subordinate, in It was clear that they were coming. But suppose that
it were not a dependent, or that clear is not a head or governor in relation to
it. If so, it would still belong to the same major category as all other adjec-
tives. Therefore, once more, that it takes such clauses would be a matter of
subcategorisation. If someone were, despite tradition, to assign it to another
category, it would again not be for such a reason.
Dependency, for its part, was a term that Tesnière did not define in his
glossary. It is simply, in the passage referred to earlier, the equivalent of
being governed. In another account, which is that of, in particular, the re-
cent Cambridge Grammar of the English Language, it is similarly the con-
verse of headship. In This is quite clear, a phrase, quite clear, would be
headed by clear; and, in the same breath, quite would be a dependent com-
bining with the adjective (compare Huddleston and Pullum 2002: 24). “The
term dependent”, as it is then explained, “reflects the fact that in any given
construction what kinds of dependent are permitted depends on the head”.
Thus, for example, quite or very are permitted dependents of head adjec-
tives, but not of nouns. A clause like that they were coming is again a per-
mitted dependent of an adjective like clear, but not of, for example, pink or
pretty.
Valency is thus implicitly, as it could have been for Tesnière, a suffi-
cient criterion for dependency. In, for example, Alfred liked me both the
subject and the object are within the valency of to like. There are some
verbs, that is, which exclude or only optionally take an object; and there are
others, in Tesnière’s term avalent, which in a more sophisticated sense
exclude a subject. Therefore Alfred and me, as “permitted” units, depend
on, in this formulation, a head liked. But the way this evidence is inter-
preted could in principle be quite the opposite. Under what conditions, we
may begin by asking, can a strictly transitive verb, such as to like, enter into
a construction? Part of the answer is, of course, that there must be a subject
and an object with which it can combine. The presence, therefore, of forms
such as liked depends, still in a perfectly natural sense of this term, on the
6 Peter Matthews

presence of units such as Alfred and me, by which these conditions can be
satisfied. More generally, therefore, if X has a valency, the units it takes do
not depend on it. In this view, it instead depends on each of them.
For this concept of dependence compare, for example, the later work of
Zellig Harris (briefly in Harris 1988: 12f. and elsewhere). It is also
matched, for such a sentence, by the much earlier analysis of the Modistae
(survey in, for example, Rosier 1983). Even, however, if this view can be
discounted, the criterion proposed is soon found to conflict with others.
Dependence, in this formulation, is again on heads of phrases: that of an
object, for example, on a verb as head of a verb phrase. But headship is
notoriously problematic, and evidence that a proposed head has a valency
can conflict with other arguments that potentially bear on it.

3.

The problem can be seen most clearly in the case of prepositions. In Eng-
lish especially, different prepositions do take different constructions. In that
way they have properties at least “akin”, once more, to valencies. But it is
far less certain that they are heads, if current definitions of a head are taken
seriously.
For Huddleston and Pullum prepositions include, for example, after in
after I left and, in its wake, most other subordinating conjunctions (2002:
599f.). Some prepositions take accordingly both clauses and noun phrases;
others, such as at, take only phrases; others, like when, only clauses. In
most other accounts the category remains much smaller. But even then, a
preposition such as on has one construction in on the floor and another in
on leaving the building, while, for example, at has only the first. Until, for
example, can combine with adverbs such as recently (Huddleston and Pul-
lum 2002: 599), but during cannot, and so on. In another view, which Hud-
dleston and Pullum also follow, words such as since, in I’ve seen her since
Saturday, are still prepositions, not reclassed as adverbs, in I’ve seen her
since. With some members, therefore, of this category a complement can be
optional while with others it is obligatory. If we talk in this light of the
valency of prepositions, there will obviously be many problems in distin-
guishing in detail the constructions they can take. But in one analysis or
another, different prepositions would have partly different entries in a lexi-
con.
The complement of a preposition might thus be “similar”, as I put it in
the early 1980’s, “to the direct object of a verb, with valencies determining
when it is obligatory, optional and excluded” (1981: 151). “Therefore”, I
 The scope of valency in grammar 7

continued, it was a dependent. But this “therefore”, even at that date, was
rather careless. It is even more so if dependency is defined as the converse
of headship.
In an “informal characterization”, the head of a phrase was “one of its
constituents which in some sense dominates and represents the whole
phrase” (Corbett, Fraser, and McGlashan 1993: 1). In what sense is, of
course, the problem; and, as Zwicky made clear over twenty years ago
(1985), there are several possible candidates. But the formula which many
linguists have since favoured talks of heads as units that “determine” the
external syntax of a whole of which they are part. It is hard to find an illus-
tration that does not raise difficulties. But in, for example, very angry peo-
ple it is easy to establish a relation between very and angry: there is a class
of adverbs, as they are called, by which adjectives can be modified. There
is also a relation between angry and people: one role of adjectives, that is,
is as modifiers of nouns. But it is hard at least to establish any independent
relation between very and people. Substitute for very any other adverb that
forms a similar combination, and it is still the adjective alone that deter-
mines the “distribution”, as a definition on these lines is often formulated,
of the intensifier and the adjective together.
For Huddleston and Pullum the head, “normally obligatory, plays the
primary role in determining the distribution of the phrase, i.e. whereabouts
in sentence structure it can occur” (2002: 24, “distribution” in bold face).
Note, in passing, that the “distribution” of a phrase is relative to “sentence
structure”; also that, though “normally obligatory”, a head can be elliptical.
The main difficulty, however, now lies in the qualification “primary”. The
syntax of a whole can by implication be determined by both a head and a
dependent. But the role of the dependent would be seen as secondary.
It is obvious why the qualification is needed. The distribution of, for
example, angry with me is in general, one might claim, determined by the
adjective; but its position as a post-modifier, in the people angry with me,
reflects in part the presence of with me as its complement. Is it always
clear, though, what is primary and what is secondary? Take, for example, a
phrase such as on leaving the building. Its distribution is not simply that of
on X; of the preposition as such plus whatever can then follow it. Compare,
for example, Put it on the floor, Put it on leaving the building. But does the
preposition even “primarily” determine the constructions in which these
different units can stand? In Put it on the floor, the role of on the floor is as
a locative. In that respect it goes with, for example, here in Put it here, or
where you like in Put it where you like, in neither of which a preposition is
included. In Turn right on leaving the building, the unit introduced by on
belongs instead with clauses such as when you leave the building; and, like
8 Peter Matthews

these, it includes a verbal unit. In another view it is these categories that are
primary, and it is the presence of a preposition, in one kind of locative or in
certain kinds of reduced clause, that would then be secondary.
The headship of prepositions could, of course, be saved by technical
devices. In Put it on the floor, what is locative might in one solution be a
preposition, onloc, which is different from other ons, which merely happen
to be homonyms, in on leaving the building or, for example, on Saturday.
The distribution of a phrase like onloc the floor could accordingly be said to
be determined, absolutely and not merely primarily, by the specific pres-
ence of onloc; that of on leaving the building by a different preposition, oning
that we might establish there, and so on. But this is a solution of a kind not
needed in the same way for verbs, nouns and adjectives. The headship of,
for example, left in left the building, of people in people angry with me, or
again of angry in angry with me, all fit Huddleston and Pullum’s definition
much more easily.

4.

The dependency of complements on prepositions is, in this light, at least

problematic. But let us suppose, for the sake of argument, that prepositions
are not heads. On, for example, will still take a range of possible construc-
tions, distinct from those of other prepositions such as at, or under, or dur-
ing. This could still be valency, if that term applies appropriately to it.
The widest application would again be as implied by Crystal’s diction-
ary. Not only, then, does on as one syntactic element form a “bond” with
the floor or with leaving the building; but, for example, building would
form a bond with the, on leaving the building, in Turn right on leaving the
building, would form a bond with turn right, and so on. These are bonds of
different types, and in Crystal’s definition, which in itself is perfectly co-
herent, they reflect not only the valencies of words like at and building but
also, since a phrase is a syntactic element, of the constituents of which they
form part at all levels. In other accounts, however, valency is again re-
stricted to lexical units. The syntactic elements that they take, moreover,
are traditionally constructions: not constituents individually, but the general
patterns in which any similar constituent will stand.
Take for comparison a straightforward relation of agreement. The con-
struction of, say, die Frau [‘the woman’] is the same, at least as linguists
usually describe it, as that of der Mann [‘the man’]: of, in general, a noun
with an article. It is then a property of certain lexemes, such as Frau, that
they form bonds with articles, such as die, in the feminine. But in this light
 The scope of valency in grammar 9

it is not a property of valency. One justification is that rules for gender are
bound up with those affecting number or case, which are not inherently of
lexical units. Another, however, might be that relations like this are less
obviously asymmetrical. A construction is one thing, and a lexical unit,
which in the traditional term “takes” or “requires” it, is another. But Frau is
a word and die too is a word, and, while grammars have traditionally talked
of articles “agreeing with” nouns, or of nouns as determining the form that
they will take, it would technically be possible to say precisely the oppo-
site. In die Frau, that is, die is inherently feminine; therefore, in an alterna-
tive formulation, it requires nouns, such as Frau, whose properties will
match it. For many linguists, this account fits beautifully with the hypothe-
sis, as they present it, that the construction is of a determiner phrase, with
die as its head determiner.
What do we mean then, more exactly, by constructions? Since the
1980’s this term has taken on a new life, in the work of Fillmore and others
(Fillmore, Kay, and O’Connor 1988; Goldberg 1995). One point, however,
that we need to emphasise is thoroughly traditional: that constructions are
wholes that may not always reduce to a simple hierarchy of parts.
At school, for example, I was taught that certain verbs in Latin, such as
doceo [‘I teach’], took a “double accusative”, or the “double accusative
construction”. The purpose, no doubt, was in practice to discourage me
from putting nouns that should be accusative into the dative, on the model
of, in English, sentences such as I taught it to the children. But this view of
their construction does reflect a fundamental truth, which is brought out
beautifully, from a lexicographer’s viewpoint, by Thomas Herbst in this
volume. A verb, in particular, does not bond independently with individual
syntactic elements, subject only to restrictions that affect each combination
separately. Its valency is a whole of which such elements are parts, and its
relation to each element, as to each of the accusatives in a double accusa-
tive construction, may be bound up with the ones it bears to others, or that
these elements bear among themselves.
Divisions among elements will then be secondary; and in many cases, as
with the constructions of doceo in Latin or to teach in English, they are not
a problem. But take, for a notorious example, the constructions of what
Quirk and his colleagues have called complex transitives. In They made her
their leader, the verb is followed, as they and many other linguists see it,
by two separate elements: first an object, her, and then an object comple-
ment, their leader. In this sense, therefore, made will form bonds with they,
as subject, and with each of these. Other complements of an object, similar
in that way to noun phrases like their leader, include adjective phrases, as
10 Peter Matthews

in, for example, That drove [him] [crazy]; infinitives, as in I felt [it] [to be
falling apart], and so on (Quirk et al. 1985: 1195ff.). But this is, of course,
just one analysis. In another common view, such verbs will take two ele-
ments only: they as subject and a clause of which the so-called “object” is
instead the subject. In two of the examples given, this is of the kind that
followers of Chomsky class as “small”: thus, with brackets again, They
made [her their leader], They drove [him crazy] (compare Fromkin 2000:
133f.). In the third example, it is a clause like others generally, except that
it is not tensed: I felt [it to be falling apart].
Which treatment should we follow? One well-known compromise
would hold that both are right; but at two different levels. In an underlying
structure her remains a subject, in the same role as the subjective pronoun
in He was their leader. But it is superficially “raised”; and, after raising, it
becomes an object. This was an analysis defended at length, thirty years
ago, by Postal (1974). Alternatively, the syntax is of an object and its com-
plement; but the relation between these is semantically like that of predica-
tion. With infinitives in particular, this relation then distinguishes a raising
verb, as many linguists call it, from control verbs, as again a follower of
Chomsky calls them, such as to persuade or to ask. In syntax, that is, both
will take the same constructions. But with verbs of the control kind, as de-
scribed by Huddleston and Pullum, “the syntactic structure matches the
semantics quite straightforwardly” (2002: 1201): compare, for example,
They asked [me] [to leave]. With “raised object verbs”, there is instead a
mismatch. In, for example, They intended [me] [to leave], the syntactic
object is not an argument, at the level of propositional meaning, of to in-
tend; but simply of the subordinate verb to leave (Hudddleston and Pullum
2002: 1201; “propositional meaning” 226). Whatever the solution, how-
ever, there will now be further difficulties. Is there also a “small clause”, if
that is the way we want to describe it, in, for example, They found him ill or
They found him in distress? Or are ill and in distress no more than separate
adjuncts? Which kind of verb is, for example, to expect in I expect you to
leave? Is you, “syntactically” if we so perceive it, no more than the subject
of to leave? What is expected, that is, is an event which involves the ad-
dressee’s departure. Or does the speaker expect it of the addressee, as an
individual or set of individuals, that he, she or they will go away? That
might suggest that you is an object, and the subject of the infinitive, again
in one analysis, a zero “controlled” by it. Or is the sentence structurally
ambiguous, in its syntax or again in no more than its semantics, as we pre-
fer?
Such issues are familiar and it is hard to see how indeterminacy can be
avoided. It seems clear, however, that at least some verbs take networks of
 The scope of valency in grammar 11

relations. In That drove him crazy, there is a link of some kind between him
and crazy. The dispute is simply as to whether it is syntactic or semantic.
Either him, in one account, or him crazy, in another, are in turn related to
drove. But so too, on its own, is crazy. To drive can take a small clause, if
that is how we want to see it, where the predicate is an adjective; or, for
example, an infinitive (That drove him to commit suicide). But, unlike to
make, it does not generally take a noun or noun phrase (compare, for exam-
ple, *That drove him a suicide). There are also limits to the adjectives it
normally goes with (compare, for example That drove him angry or ?That
drove him happy), as with other verbs of this class. Compare, for example,
They painted it green with ?They painted it pretty; or They cut it short with
?
They cut it brief. The construction is a whole in that sense, in that all its
elements are interrelated. But within the class of verbs that take it, whose
valency is at a general level complex transitive, there are again some
where, at a subsidiary level, one relation or the other will be weaker. The
link of verbs to object complements is strongest with what may be called
“group verbs” (Denison 1998: 221ff.) or idioms, such as, we might say, to
cut short. But it is certainly weaker in, for example, The crash left her pen-
niless or They found him ill. The link of verbs to objects is weaker with so-
called “raising verbs”, as in I felt it to be falling apart; and so on. But the
problems this can lead to, in saying what exactly, for example, is the
valency of to expect or to want, are precisely no more than subsidiary.
The network does not, in this case, so obviously include the subject. But
there is also the construction first described by me, I think, as “complex
intransitive” (Matthews 1980). In, for example, She turned green this is
again a whole in which the relation between no two syntactic elements
(subject, verb and subject complement) can be detached from the others.

5.

Valency, to sum up, is in principle independent of dependency, headship or

governorship; it is a property of lexical units in relation to constructions;
and it is specifically of units assigned to subcategories. The remaining
question is, which lexical units? Or, if the answer is all, what is a lexical
unit?
One definition might appeal to a distinction between closed and open
categories. The distinction itself is central for, among others, Quirk and his
colleagues (1985: 67, 71ff.); and in this sense prepositions, in particular, are
not lexical but grammatical. Therefore the bonds they form in varying con-
structions, though “akin” to valency, could belong with those of other
12 Peter Matthews

members of closed categories, such as conjunctions, modal verbs, or arti-

cles. In another view, however, they too form a lexical category. The rea-
sons vary; but one argument might be precisely that each preposition has a
valency. Like verbs, that is, each takes or in more fashionable jargon “li-
censes” a specific range of structures.
The truth, however, is that no single category is quite like the others.
The properties of verbs, in this respect, are clearly lexical. Not only does
each member of the category have a valency; but exactly what it is can vary
between speakers and can change quite easily. Judgments, therefore, are
notoriously difficult. Can to start, for example, be used as a complex transi-
tive: thus The rain is starting the tunnel to collapse? Can to demand take
the construction of They demanded someone to come, or to accord that of
They accorded it with this title? These are modelled, naturally, on examples
I have collected. “Chaque mot”, one might say, “a sa valence”; and al-
though the instinct of many linguists has been to establish ordered series of
subcategories, distinguished by fixed ranges of constructions and semantic
or “cognitive” properties corresponding to them, they are liable to be de-
feated, in the end if not from the outset, by the operations of analogy on the
use of lexemes individually. To say of an intransitive verb that it simply
cannot be used transitively is already imprudent.
If prepositions are grammatical it is, in this light, not just because they
are closed. That statement may in any case need to be qualified. It is be-
cause their meanings and their syntax are fragmented. On, for example,
enters into different contrasts with different sets of opposed units: as a loca-
tive in Put it on the floor; in expressions of time such as on Saturday: in
combination with an ing-form in on leaving the building; in individual
group verbs such as to look on or to run on; and so on. Each use is therefore
subject to its own rules. By, for example, is another preposition that can
take an ing-form: thus by leaving the building. It also enters into locative
constructions, as in I was walking by the river; and, in that use, it can have
a meaning partly similar to that of along, in I was walking along the river.
But there is no basis here for analogical extensions like those that we find
with verbs: by leaving the building, that is; therefore along leaving the
building.
In this respect most adjectives and nouns are also lexical. But nouns
especially raise other problems, which in turn are well-known. Not all, of
course, take even optional complements: the news of their success, not the
cat of their success; her letter to the council, not her cat to the council; and
so on. Is cat to be described in this light as a noun which has a zero
valency, on the lines of verbs such as to rain? Or do such nouns simply
have no valency at all? With nouns like news or letter, complements are
 The scope of valency in grammar 13

then rarely obligatory. Many such nouns are derived, moreover, from verbs:
announcement from to announce; speech, although irregularly, from to
speak, and so on. Their valencies, if that is how we should again describe
them, are in many cases also temptingly derivative: He announced his res-
ignation or She spoke to parliament; hence, as many will argue, the an-
nouncement of his resignation or her speech to parliament. Now the mean-
ing of speech in this example is narrower than that of spoke. But how far,
despite that, are their valencies that of a common stem and not of nominal
and verbal lexemes separately?
The way we answer questions like these may, however, not be that im-
portant. That valencies are above all properties of verbs has been acknow-
ledged from the outset, and most linguists, whether or not they use the term
themselves, see individual argument structures, or what Quirk and his col-
leagues call their complementation, as fundamental to their meanings. The
same is arguably true of adjectives such as clear in It was clear that they
were coming, or sure in I was sure that they were coming, where, in predi-
cative position, they may take complements optional only under ellipsis.
Here too, moreover, usage can be fluid. But many other adjectives, like
many nouns, take modifiers only or have valencies that are temptingly,
again, derivative. If prepositions did have meanings like verbs, their status
as atomes crochus could again be seen as similar: both primitive and fun-
damental to the whole class. If Tesnière did not describe them in that way it
was because, in his analysis, they were grammatical markers and not gov-
ernors. But even if governorship is irrelevant, or the definition of a head
can somehow be made to cover them, there would still be problems that
might lead us to explain their syntax differently.

References

Corbett, Greville G., Norman M. Fraser, and Scott McGlashan (eds.)

1993 Heads in Grammatical Theory. Cambridge: Cambridge University
Press.
Crystal, David
2003 A Dictionary of Phonetics and Linguistics. 5th ed. Oxford: Black-
well.
Denison, David
1998 Syntax. In The Cambridge History of the English Language 1776-
1997, Vol. 4. Richard Hogg, and Suzanne Romaine (eds.), 92–329.
Cambridge: Cambridge University Press.
14 Peter Matthews

Fillmore, Charles, Paul Kay, and Mary Catherine O’Connor

1988 Regularity and idiomaticity in grammatical constructions: The case of
let alone. Language, 501–538.
Fromkin, Victoria (ed.)
2000 Linguistics: An Introduction to Linguistic Theory. Oxford: Black-
well.
Goldberg, Adele E.
1995 Constructions: A Construction Grammar Approach to Argument
Structure. Chicago: University of Chicago Press.
Harris, Zellig S.
1988 Language and Information. New York: Columbia University Press.
Herbst, Thomas, David Heath, and Hans-Martin Dederding
1980 Grimm’s Grandchildren: Current Topics in German Linguistics.
London: Longman.
Herbst, Thomas
2007 Valency complements or valency patterns? This volume.
Matthews, Peter H.
1980 Complex intransitive constructions. In Studies in English Linguistics
for Randolph Quirk, Sidney Greenbaum, Geoffrey Leech, and Jan
Svartvik (eds.), 41–60. London: Longman.
1981 Syntax. Cambridge: Cambridge University Press.
1997 The Concise Oxford Dictionary of Linguistics. Oxford: Oxford Uni-
versity Press.
Postal, Paul M.
1974 On Raising. Cambridge, Mass.: MIT Press.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik
1985 A Comprehensive Grammar of the English Language. London:
Longman.
Rosier, Irène
1983 La Grammaire Spéculative des Modistes. Lille: Presses Universi-
taires de Lille.
Tesnière, Lucien
1959 Éléments de Syntaxe Structurale. Paris: Klincksieck.
Trask, R. Lawrence
1993 A Dictionary of Grammatical Terms in Linguistics. London:
Routledge.
Zwicky, Arnold
1985 Heads. Journal of Linguistics 21: 1–29.
Valency complements or valency patterns?

Thomas Herbst

1. Valency and the idiom principle

1.1. Valency as a property specific to lexical units

The valency approach as it was developed in German Germanistik can

probably claim to be one of the most systematic attempts to describe com-
plementation structures of verbs, adjectives and nouns. One of its most
important assets is that it has always devoted considerable attention to the
distinction between such elements whose occurrence is dependent on the
presence of a particular valency carrier, i.e. the complements (Ergänzun-
gen), and such elements whose occurrence in a clause is structurally inde-
pendent of the presence of particular other words, i.e. the adjuncts or pe-
ripheral elements (Angaben). Although the distinction between comple-
ments (or, in more refined versions of the approach, different types of com-
plement) and adjuncts takes the form of a gradient rather than that of two
clearly distinct categories, it can be said that within valency frameworks
what is to be considered a complement of a valency carrier is not left to
intuition but based on a number of test criteria.
The phenomenon of valency is one part of the unpredictable, unsystem-
atic aspects of language. It is thus probably more than a historical coinci-
dence that pioneering work within the valency framework has been done
within a general context of foreign language learning and foreign language
teaching, which is equally true, for example, of Gerhard Helbig’s contribu-
tions to the development of valency theory in the GDR and of early valency
work in West Germany (e.g. Engel and Schumacher 1976).1 Equally, it can
hardly be considered a coincidence that valency research should have re-
sulted in valency dictionaries since valency structures represent idiosyn-
cratic, word-specific types of information (e.g. Helbig and Schenkel 21973;
Engel and Schumacher 1976; VALBU 2004 for German or VDE 2004 for
English). Although valency is also an important concept within many syn-
tactic theories, especially those with a dependency component (Matthews
1981; Heringer 1970 or 1996; Herbst and Schüller forthcoming), it is pri-
16 Thomas Herbst

marily to be seen as a property of lexical items or, to be more precise, as a

property of lexical units. This is, of course, no contradiction since, in fact,
the determining influence of individual lexical units on the structure of
sentences has received increasing attention in many theoretical frame-
works.2

1.2. Tidy and messy aspects of language

Unpredictability in language is not restricted to valency, however. A further

case in point is presented by combinations such as guilty conscience or lay
the table, which can be called institutionalized collocations and which can
be characterized as “typical, specific and characteristic relations between
two words” (Hausmann 1985: 118). Again, the idiosyncratic nature of such
combinations is revealed by the comparison with other languages. Neither
*schuldiges Gewissen nor *den Tisch legen would be acceptable transla-
tions in German, for instance. In general, one could argue that the fact that
the well-formedness of sentences or texts cannot easily be described as the
result of applying syntactic rules of some kind is probably particularly ap-
parent in the context of a type of linguistics that takes into account aspects
of foreign language teaching and of translation theory. At a very basic
level, this kind of insight takes the form of the common experience that
learners’ utterances produced in essays or translations which do not violate
any grammatical rules of the target language are nevertheless often judged
not to “sound right” by native speakers although it is difficult to formulate
this in more concrete terms. It is important to realize that although the un-
systematicity of language, for which such observations provide evidence,
may be particularly noticeable in foreign language contexts, it is a central
feature of the phenomenon of language as such and thus has to play an
appropriate role in any comprehensive theory of language.
On the other hand, it cannot be denied that other aspects of language can
indeed be accounted for in terms of general rules or principles. Just as it is
obvious that institutionalized collocations such as guilty conscience, white
coffee or strong tea cannot be explained in terms of rules, one would defi-
nitely assume that at the other end of the extreme the interpretation of utter-
ances is determined by general pragmatic principles.3 Thus an announce-
ment of the type as is made on trains running from Westerland (Sylt) to
Hamburg
 Valency complements or valency patterns? 17

(1) In Kürze erreichen wir Husum. In Husum steigen Sie bitte in

Fahrtrichtung rechts aus.
‘We’ll shortly be arriving at Husum. At Husum please alight
using the doors on the right-hand side of the train.’

is not interpreted by any passenger to mean that one is obliged to get off at
Husum although this would be a literal interpretation of what is being said.
Up to a point, distinctions such as de Saussure’s (1916) between langue
and parole, Coseriu’s (1973) between System and Norm,4 Sinclair’s (1991)
between the open choice principle and the idiom principle5 or Chomsky’s
(1986) between core and periphery recognize the fact that some aspects of
language can be explained rather well in terms of general rules whereas
others apparently cannot. The question is, however, how much importance
is attributed to these two aspects. Chomsky (1986: 221), already in his
choice of terms but also in the description of the concepts, clearly takes
core grammar to be the central aspect of language:

The core, then, consists of the set of values selected for parameters of the
core system of S0; this is the essential part of what is “learned”, if that is the
correct term for this process of fixing knowledge of a particular language.
The grammar of the language L is the linguist’s theory of L, consisting of a
core grammar and an account of the periphery.

Whether core grammar in the sense described by Chomsky is the “essential

part of what is being ‘learned’” or not depends very much on the number of
linguistic facts that – like valency and collocation – fall under the heading
of the unpredictable, idiosyncratic or idiomatic. Opposing Chomsky, Sin-
clair (1991: 110) argues that the principle of idiom, “has been relegated to
an inferior position in most current linguistics, because it does not fit the
open-choice model.”
It is interesting to see that a lot of the empirical research carried out in
corpus linguistics also underlines the importance of idiosyncratic features
as far as the co-occurrence of words in texts is concerned. It is in this light
that Sinclair’s (1991: 110) concept of the idiom principle – “that a language
user has available to him or her a large number of semi-preconstructed
phrases that constitute single choices, even though they might appear to be
analysable into segments” – appears so remarkable. It seems that a consid-
erable amount of the evidence provided by corpus research and the experi-
ence gained in contexts such as that of foreign language teaching must lead
to the conclusion that the idiosyncratic or idiomatic aspect of language may
well be much more important than is often assumed even if this means that
language as a whole appears less tidy and perhaps slightly messy.6
18 Thomas Herbst

It is obvious that the relationship between the rule-driven, tidy and the
messy, idiomatic components of language is of particular relevance to cog-
nitive or psycholinguistic issues. Idiomatic or idiosyncratic aspects require
storage; open choice aspects can be accounted for in terms of rules – irre-
spective of whether such rules are learned or acquired as rules or emerge
from data that are acquired in some form or another. While the investiga-
tion of various valency phenomena provides a considerable amount of evi-
dence to assume that the idiosyncratic component is rather important (Götz-
Votteler, this volume; Klotz, this volume; Herbst, forthcoming), this article
will address the problem from a slightly different angle – namely by look-
ing at the question of how valency relations are best described.

2. Valency patterns or valency complements?

2.1. Complement inventories

Valency is most often seen as the property of a word – or, more precisely,
of a lexical unit as “the union of a single sense with a lexical form” (Cruse
1986: 80) – to determine the occurrence of other elements in a clause. Thus
Helbig and Schenkel (21973: 49) define syntactic valency in the following
way:7

... die Fähigkeit des Verbs, bestimmte Leerstellen um sich herum zu eröff-
nen, die durch obligatorische oder fakultative Mitspieler zu besetzen sind.
[... the ability of the verb to open up certain positions in its syntactic envi-
ronment, which can be filled by obligatory or optional complements.]

This view of valency is expressed in a very similar way in Emons’s (1978:

4) definition of valency:

Die Eigenschaft eines Prädikats, eine bestimmte Anzahl von Ergänzungen

zu fordern, nennen wir seine Val e nz.
[The property of a predicate to demand a certain number of complements is
referred to as its valency.]

Very similar descriptions can be found in other frameworks, for instance,

when Haegeman (1991: 41) says the “verb theta-marks its arguments” or
in Chomsky’s (2004: 111) description of the character of lexical entries of
verbs:
 Valency complements or valency patterns? 19

... every lexical item carries along with it a certain set of thematic roles,
theta-roles, which have to be filled. That is its lexical entry ...

What all these conceptions have in common is that a verb can be associated
with a kind of inventory of syntactic elements, which, depending on the
theoretical framework and terminology, can be described in terms of se-
mantic cases, theta-roles or as arguments or complements. This view of
valency is represented in the Valency Dictionary of English (VDE) in the
form of complement inventories, in which the complements of a particular
lexical unit (e.g. convince) are listed.

The alternative to such a view of valency as an inventory of complements is

to regard valency specifications as information about particular patterns in
which a lexical unit can occur. In the VDE such patterns are indicated
(however usually without any specifications regarding subjects) in the pat-
terns and examples section following the complement inventory. Thus, in
the case of convince, one divalent active pattern and three trivalent active
patterns are identified:

Such a pattern-related view of valency is also reflected in the concept of

Satzbaupläne as outlined by Engelen (1975) or Engel (1977).8 Similarly,
one could argue that Fillmore’s (1968: 27) statement that the “insertion of
verbs ... depends on the particular array of cases, the ‘case frame’ provided
by the sentence” can be taken to refer to a pattern oriented view of such
phenomena.
In many respects, both views of valency – in terms of an inventory of
complements or in terms of valency patterns – are compatible with one
20 Thomas Herbst

another. The question to be discussed here is, however, whether there are
linguistic facts which can be described more appropriately in terms of the
one framework rather than the other.
This will be discussed with respect to the following four levels of de-
scription which a comprehensive valency description as attempted in VDE
should comprise, namely statements about:
(i) the minimum and maximum valency of the lexical unit in question,
(ii) the degree of optionality of the complements as obligatory, contextu-
ally-optional or optional,9
(iii) the formal and functional properties of the complements,10
(iv) the lexical, semantic and collocational description of the comple-
ments.11

2.2. Quantitative valency

The complement inventory account and the pattern-oriented view of verb

valency may already be in conflict when it comes to the relatively simple
question of quantitative valency, which is usually seen as being determined
by the number of obligatory and optional complements a verb requires.
Thus it would be common practice to classify a verb such as meet as diva-
lent on the basis of a sentence such as

(2) This time, she met Jamie at Rital’s wine bar at lunchtime.BNC12

since neither the [N]A-complement she nor the [N]P-complement Jamie can
be deleted without making the sentence ungrammatical. However, monova-
lent uses of meet can be found in sentences such as

(3) These days they meet at conferences ...BNC

(4) A Cabinet committee meets tomorrow to agree to slash public
spending by billions.BNC

The problem with verbs such as meet or kiss is that there is a difference
between syntactic and semantic valency. One could argue that at the se-
mantic level such verbs have an obligatory valency of 2 in that they require
two arguments whose semantic roles could be described as that of a
‘MEETER’ (i.e. a person who meets someone) and that of a ‘MEETEE’
(i.e. a person who is met by someone). At the syntactic level, however, both
these arguments can be expressed by one complement, which can consist of
a coordinated noun phrase or a plural noun phrase as in (3) or of a noun
 Valency complements or valency patterns? 21

phrase containing a singular group noun. In VDE this is represented in a

form which combines the two arguments I + II and provides a list of possi-
ble complements:

Although such an account describes the syntactic possibilities accurately, it

is not entirely unproblematical from a theoretical point of view. First of all,
the monovalent uses of such verbs raise the question of whether the [N]A-
complement (the subject) actually represents the two semantic roles of
‘MEETER’ and ‘MEETEE’ or whether such role assignment is misguided
in such cases. One could indeed argue that since activities such as ‘meet-
ing’ involve more than one person, the unacceptability of a sentence such
as

(2) a. *This time, she met at Rital’s wine bar at lunchtime.

is due to a general semantic or even pragmatic rule. More importantly in

the present context, however, the complement inventory presented in VDE
already contains aspects of a pattern-oriented view of valency since it dem-
onstrates the interrelationship between different kinds of possible realiza-
tions of the arguments identified. In any case, one can conclude that state-
ments about the minimum or maximum valency of verbs are difficult to
make without taking into account the precise form a complement such as
[N]A takes in a particular realization.13

2.3. Optionality

It is common in valency theory to make a distinction between different

kinds of complement with regard to their degree of optionality. Thus it
would be generally accepted to classify [from N] in

(5) We’d love to hear from you about it.VDE

as an optional complement since the verb hear can be used in the same
meaning without that complement, as in

(6) I want to hear about it.BNC

22 Thomas Herbst

At the same time there are uses such as

(7) We’ll hear from an economics writer on why the economy is ex-
panding faster than unemployment is decreasing.VDE

where the [from N] complement cannot be deleted:

(7) a. *We’ll hear on why the economy is expanding faster than

unemployment is decreasing.

This situation is difficult to describe in terms of a complement inventory

since [from N] is optional in patterns with [about N] but obligatory in pat-
terns with [on N].14 An accurate description of the optionality of comple-
ments is complicated further by the fact that with some verbs it may be
affected by special types of text and the grammatical construction in which
the verb is used. A typical example of this is presented by instructions typi-
cal of cookery books:

(8) Fix and wash carefully.VDE

(9) Boil the lime flowers and nettles together in the water, cover and
leave to simmer for ten minutes.VDE

2.4. Alternative realizations

A further problem for an inventory-oriented approach towards the descrip-

tion of valency is presented by alternative patterns which can be considered
to be more or less synonymous. Thus in cases such as

(10) I hurried to pack my thingsII.BNC

(11) I rushed to pack my suitcaseIII.BNC
(12) II pack themII into big bagsIII.VDE
(13) The old fellow has no idea how to pack a shopping basketIII with
goodsII.VDE

it makes sense to consider the underlined elements as representing one type

of complement and the elements marked by dotted underlining as repre-
senting another type of complement. The first one could be seen as repre-
senting an argument that could be described as ‘CONTAINER’, the second
one as an argument labelled ‘ÆFFECTED’ in VDE. The VDE complement
inventory of pack contains the following information:
 Valency complements or valency patterns? 23

What is remarkable about this is that a mere list of complements without

any reference to the patterns in which they can occur would not be suffi-
ciently specific because it would not rule out unacceptable sentences of the
type

(14) a. *I rushed to pack into my suitcaseIII with my thingsII.

b. *I rushed to pack with my thingsII into my suitcaseIII.

Only the fact that [with N] is specified as occurring only in pattern T3 (of
which [13] is an example) and [into N] as occurring only in pattern T2 (as
in [12]) rules out (14a) and (14b). Combinations of two noun phrases of the
type

(14) c. *I rushed to pack my thingsII my suitcaseIII.

d. *I rushed to pack my suitcaseIII my thingsII.

are also excluded by the description of the complements as [N]P-2, which

means that they can occur only as the second noun phrase in a pattern T1 as
in

(15) ShelaghI packed themIV a lunch boxII.VDE

One could argue, of course, that specifying a complement with respect to

its place in a pattern (as indicated by the indices 1 and 2) already provides
information about patterns in the description of the complements. Although
these examples show that referring to the valency patterns of a verb is an
essential component of the description of its valency, this does not mean,
however, that identifying complements as such and establishing a comple-
ment inventory is a pointless or redundant exercise. One should not forget
that it is this kind of complement inventory that provides information about
the semantic roles of the complements in various patterns.
24 Thomas Herbst

2.5. Semantic and lexical properties

A similar kind of problem is presented by rivalling patterns if one takes the

lexical level into account. Klotz (2000) points out that the two trivalent
patterns of the verb cause do not allow the same lexical elements. Thus
although the complements marked by single and double underlining in

(16) ... there are people whose drinking causes them medical or social
harm.BNC

and

(17) It can now be stated that passive smoking causes lung cancer in
non-smokers and serious respiratory illness in babies.BNC

express the same or very similar semantic roles, a sentence such as

(17) a. *passive smoking causes babies serious respiratory illness.

is not acceptable. A similar example is presented by the fact that in the case
of so-called ergative verbs such as

(18) TheyI closed the doorII behind them.VDE

(19) The heavy wooden doorII closed with a thump.VDE

ergativity (in the sense that the [N]P-complement can also occur as an [N]A)
is restricted to certain lexical items:

(20) He closed his book and gazed into the flames.VDE

(21) *His book closed.
(22) ... this book closes with the end of the 1988 season.BNC

where (22) obviously does not correspond semantically to (20). A similar

example is presented by open, where one finds

(23) Suddenly the kitchen door opened ...VDE

(24) He opened the kitchen door and came in and shut it before he
turned to face them.BNC
(25) He opened a bottle of champagne.BNC

but not:
 Valency complements or valency patterns? 25

(25) a. *A bottle of champagne opened.

Observations like these stress the argument that speakers seem to have
available to them information about possible realizations of particular com-
plements in particular patterns and not just information about the comple-
ments of a particular verb.

3. Semantic and lexical information about complements

If valency information comprises not only information about possible com-

plements of a verb (or other valency carriers) but specific knowledge about
the possible combinations in certain valency patterns including lexical in-
formation of the kind which lexical items or sets of lexical items can realize
a complement in a particular pattern, then this certainly increases the
amount of idiosyncratic or idiomatic knowledge that has to be acquired and
stored by the speakers of a language.
This applies particularly to the question of whether it is possible to pro-
vide a description of the semantics of complements that would actually
account for all the lexical items that can realize this complement and ex-
clude others. Empirical work in this field has shown that a finite set of se-
mantic cases as originally proposed by Fillmore (1968) poses a great num-
ber of descriptive problems and is probably not refined enough to provide a
comprehensive description. Helbig and Schenkel (21973) make use of se-
mantic components to characterize semantic properties of complements;
Helbig (1992: 154−155) suggests integrating both semantic components
(Stufe II) and case roles (Stufe III). Although VDE adopts a very flexible
policy and includes semantic descriptions that correspond to stages II and
III of Helbig’s model, it is interesting to see that VDE, VALBU and Frame-
Net independently of each other generally provide descriptions of the se-
mantics of complements which are rather specific to the particular verb.
Thus VALBU characterizes the nominative complement (NomE) and accu-
sative complement (AkkE) of a lexical unit such as gründen (sense 1) as
follows:

NomE: derjenige, der etwas ins Leben ruft: Person/Institution

AkkE: dasjenige, das ins Leben gerufen wird: Institution/Gremium [Kom-
mission, Bürgerinitiative, Selbsthilfegruppe o.Ä.]

This is rather similar to the descriptions provided in VDE for a verb such as
deny:
26 Thomas Herbst

A person or something written or said by a personI can deny

(i) something they are accused of or that has been said about themII
(ii) that something is the case or existsII.

Similarly, FrameNet, which establishes categories that cover more than a

single lexical unit, uses categories that are much more specific than those of
traditional case grammar. Thus the closure frame, to which the verb open
belongs, operates with categories such as ‘Agent’, ‘Fastener’, ‘Containing
object’, ‘Enclosed region’, ‘Container portal’ or ‘Manipulator’. Again, a
parallel can be found to the description provided in VDE:

The description of sense B in VDE finds an interesting parallel in

VALBU’s definition 6 of öffnen:

jemand [Person [als Funktionsträger]/Institution] veranlasst, dass etwas [In-

stitution: Geschäft, Praxis, Behörde o.Ä./[indirekt Räumlichkeit]] irgend-
wann für den Kunden-, Publikumsverkehr zugänglich ist; aufmachen.

What is interesting about the lexicographical treatment of the non-formal

side of the characterization of complements in VDE or VALBU is that both
dictionaries make use of general categories such as someone or derjenige
(which can be seen as equivalent to Helbig’s semantic feature + HUM) but
nevertheless find it necessary to give relatively specific lists of lexical items
such as door, window, etc. or Kommission, Bürgerinitiative. Very often this
is because no suitable label can be found as in the case of the note for the
verb set in VDE

A personI can set someoneIII something such as a deadline, a target, a

task, a test, an examination, etcII .

where it seemed impossible to subsume all possible realizations of com-

plement II under a general heading.
All this provides strong evidence for the messy side of the scale.
 Valency complements or valency patterns? 27

4. Conclusions and questions

4.1. Conclusions

With regard to the question of whether valency phenomena are to be de-

scribed more appropriately in terms of a complement inventory or in terms
of valency patterns, it seems that both views will have to be considered.
Valency patterns can be seen as the basis for generalizations in terms of a
complement inventory or also in terms of argument structure constructions
of the kind discussed, for instance, by Goldberg (2006). The identification
of separate complements in a complement inventory allows certain gene-
ralizations to be made, especially with regard to the semantic contribution
of the complement (in terms of semantic roles or whatever) that must be
part of a valency description. However, the above discussion has revealed
that very important facts about the valency structures of a lexical unit can-
not be covered by a complement inventory: these range from the possible
combinations of complements and their position to the question of the pos-
sible lexical realizations of the complements in different valency patterns.15
That valency dictionaries should provide Satzbaupläne as in VALBU or
valency patterns as in VDE thus is not merely due to considerations of lexi-
cographical or didactic presentation but reflects the nature of valency phe-
nomena as such. This insight is also of psycholinguistic relevance in that it
shows the need to specify valency patterns in the design of a mental lexi-
con.
The discussion has also shown that valency is definitely one of the more
messy aspects of language. Although nobody will deny that certain general
tendencies are also at work – for instance generalizations of the type that
subjects of English active declarative clauses tend to be the most agent-like
entity in the clause –, the discussion has provided ample evidence to illus-
trate that the amount of idiosyncratic word specific knowledge that is in-
volved is considerable.

4.2. Questions

If one considers valency phenomena not only from the point of view of
descriptive linguistics or lexicography but with respect to psycholinguistic
or cognitive questions, then the conclusions outlined above raise a number
of questions. What is obvious is that the messy character of valency phe-
nomena as such and the role attributed to valency patterns – which are nei-
ther general patterns like the patterns of early American structuralism nor
28 Thomas Herbst

identical with the constructions of construction grammar16 – increases the

amount of information that has to be stored in the mental lexicon.
However, if storage is such an important factor in this area, one might
take the issue further and question the idea of valency as a property of lexi-
cal units altogether. Does it make sense to assume that we store different
senses of a lexeme, which then have certain properties such as a particular
valency structure? To what extent are we justified in assuming that

(26) Her excitement shone in her eyes as she showed him her
sketches.VDE

represents a different meaning of show from that exemplified by

(27) Nicholson seized every opportunity to show his work in the mixed
exhibitions now being arranged.AC
(28) Patrick Heron’s work was shown by the Waddington Galleries.AC

simply because (27) and (28) represent divalent uses and semantically are
instances of public showing? Does

(29) Children in this phase show no special anxiety at being separated

from their parent; and no fear of strangers.VDE

have to be treated as a separate sense of show because the showing is non-

intentional? It is obvious that these are questions any descriptive semanti-
cist or lexicographer is faced with every day, and it is equally obvious that
polysemy is not necessarily a property lexical items possess but a property
that is imposed on them by analysts, but nevertheless the question remains
about how such facts are dealt with in language acquisition and how they
are processed and stored in the brain and what role general rules play.
Perhaps it is useful to draw an analogy with regular and irregular phe-
nomena in morphology. Contrary to general opinion, Bybee (1995: 428)
argues that it is not only so-called irregular past tense forms such as stuck
or struck that are stored in the brain but also the regular forms of high fre-
quency verbs such as covered and that past tense forms of a low frequency
words such as hovered can be produced on the basis of the stored informa-
tion:

The basic proposal is that morphological properties of words, paradigms

and morphological patterns once described as rules emerge from associa-
tions made among related words in lexical representation.
 Valency complements or valency patterns? 29

The question to be asked in the valency context is whether storage of un-

analysed information could not equally serve as an explanation for (a) the
apparently idiosyncratic character of many valency phenomena and (b)
those generalizations about the use of certain complements or valency pat-
terns that are actually possible.
Take a simple example such as the verb meet. Presumably one can
safely assume that in the language acquisition process a child will first en-
counter sentences of the type

(2) This time, she met Jamie at Rital’s wine bar at lunchtime.BNC
(30) This morning he’ll meet President Vaclav Havel ...BNC
(3) These days they meet at conferences ...BNC
(4) A Cabinet committee meets tomorrow to agree to slash public
spending by billions.BNC
(31) Heron had met Delia almost immediately on arrival in Welwyn in
1929, when they attended the same school.AC
(32) We had never met before.AC

All of these sentences represent the concept of a coming together of two or

possibly more people, which can be seen as a very simple representation of
the meaning of the verb and a very basic concept of what one might call its
argument structure. If children “are indeed learning utterance-level con-
structions as linguistic gestalts”, as Tomasello (2003: 169) supposes, then
these sentences can serve as the basis for abstractions concerning the se-
mantic features (‘+ human’ or ‘PERSON’) and semantic roles of the com-
plements. However, that the character of the “meeting” described in these
utterances differs can be concluded from one’s world knowledge rather
than from the semantics of the verb. The fact that dictionaries distinguish
between different senses on the grounds of such features as ‘by arrange-
ment’, ‘by chance’ or ‘for the first time’ is an attempt to describe the scope
of situations in which the verb meet can be used rather than a semantic
description; this is made clear by the ambiguity of some of the sentences
above (Herbst and Klotz 2003: 40−41). What one has to bear in mind, how-
ever, is that while for lexicographical purposes it may be necessary to dis-
tinguish between different senses of a verb such as meet, psychologically
this need not be so – at least not for perception purposes. In order to under-
stand sentences such as the ones above, a very general understanding of
what meet ‘means’ together with knowledge of certain facts of the world or
what one could call pragmatic rules is sufficient.
30 Thomas Herbst

On that basis, it is also possible to understand sentences such as

(33) Each service is met by a bus at Dunwich.BNC

(34) Where the land meets the sea, there are hundreds of beaches and
coves ...BNC

in which the arguments are not ‘+ human’ any more. Equally, a sentence
such as

(35) We are moving slowly and carefully to find the best ways to meet
the individual needs of those gifted in math or gifted in any area
...BNC

can be interpreted without any difficulty. The basic suggestion is that on the
basis of storage of highly frequent and thus prototypical uses of a verb such
as meet speakers gain knowledge about the kinds of situations to which this
verb can be applied together with some notion of its meaning. Encountering
less frequent uses does not present any problems of comprehension and
results in further storage of possible uses of that word. Certainly one would
assume that no foreign learner of English who is familiar with the proto-
typical uses of meet referring to the encounter of two or more people will
find it impossible to understand what sentences such as (33−35) mean. On
the other hand it is unlikely that any foreign learner would use such sen-
tences before having heard or read a verb being used in such a context. In
fact, one might argue that this is precisely the reason why many users of
monolingual dictionaries find examples more useful than definitions. What
kind of sentence would a learner be able to produce on the basis of the
definition “to experience a problem, attitude, or situation” without the ex-
ample Wherever she went she met hostility and prejudice (LDOCE4)?
Such an account of verb valency emphasizes storage in combination
with the application of general pragmatic rules. It leaves open the question
of the precise form or amount of the information that is being stored and
the status of generalizations in terms of rules or general principles that cer-
tainly also have to be accommodated, although some of the ideas that have
been proposed under the heading of emergentism seem to promise new
explanatory potential in this respect.17 There seems to be very convincing
evidence that repetition and storage play a major role in language acquisi-
tion. Tomasello (2003: 112), for instance, points out “that many, indeed the
majority of the utterances children hear are grounded in highly-repetitive
item-based frames that they experience dozens, in some cases hundreds, of
times every day”.
 Valency complements or valency patterns? 31

What would make such a model attractive to descriptive linguistics is

that it would account for the impossibility of making a clear distinction
between various senses of a word and it would account for the fact why
much valency information seems messy rather than systematic.
The fact that the descriptive analysis of valency phenomena sometimes
results in rather clear and straightforward generalizations but sometimes
only leads to seemingly unsatisfactory or messy accounts of a phraseologi-
cal or quasi-collocational nature seems highly compatible with Goldberg’s
(2006: 62) view that “both item-specific knowledge and generalizations
coexist” and with the following conclusion reached by Tomasello (2003:
98):

The level of abstraction at which the speaker is working in particular cases

may or may not correspond to the most abstract level the linguist can find; it
is in all cases an empirical question that most often needs psychological ex-
perimentation.

What is interesting in the case of valency is that linguists or lexicographers

do not always seem to be able to find abstract representations – in any case,
empirical psychological evidence would be extremely helpful also for the
design, and the assessment of the plausibility, of various models of valency
phenomena within descriptive linguistics.

Notes

1. For an account of early German valency see Herbst, Heath, and Dederding
(1980).
2. See Sinclair (2004b: 164-165) and Herbst (forthcoming).
3. For the difference between rules and principles see Leech (1983).
4. For a discussion of valency phenomena in terms of Norm see Herbst (1983).
5. See also Sinclair (2004a).
6. Compare also Fillmore, Kay, and O’Connor (1988) for the importance of idio-
maticity.
7. Compare also Helbig (1992: 3–18) and Ágel (2000).
8. Engel (1977: 180) defines Satzmuster as follows: “Die Struktur des Satzes
wird zwar entscheidend durch die Struktur des Verbs bestimmt; dabei spielt
aber die Ar t der jeweiligen Ergänzung eine wenigstens ebenso große Rolle
wie ihre Anzahl. Ein Überblick über die Kombinationsmöglichkeiten von Er-
gänzungen hat also deren Zahl u nd Ar t zu berücksichtigen. Solchermaßen
festgelegte Kombinationsmöglichkeiten werden Sa tz mus t er genannt.” Engel
(1977: 181) distinguishes Satzmuster from Satzbaupläne: “In den Satzmustern
werden zwar die kombinierbaren E. zusammengestellt, es wird aber zwischen
32 Thomas Herbst

obligatorischen und fakultativen E. nicht unterschieden. Dieser wichtige Un-

terschied gehört ebenfalls zur Valenz, ist also wie die Zahl und Art der E. vom
Verb gesteuert. Man berücksichtigt ihn in der Kodierung am besten, indem
man alle fakultativen E. einklammert. Auf diese Art entstehen aus Satzmustern
Sat zb a up lä ne.” Such valency descriptions in terms of Satzbaupläne take the
form of essen<0(1)> or sagen<01(3)> (Engel 1977: 182).
9. For the treatment of these cases in FrameNet see Fillmore (this volume).
10. In VDE complements are characterised in terms of their morphological form –
[N] for noun phrase or [V-ing] for an ing-clause – and in terms of possible
functions in a clause – the index A indicating the ability of a complement to
occur as the subject in an active clause; P indicating the ability to function as a
subject in a passive clause.
11. The first three levels are covered by the information given in the complement
inventory and taken up by the patterns, which are given in the main part of the
entry together with the examples. The fourth level of the description is indi-
cated by separate notes at the end of a dictionary entry.
12. Examples marked BNC are taken from the British National Corpus, those
marked VDE are taken from the Valency Dictionary of English, which is
based on the Bank of English. AC refers to general authentic language mate-
rial.
13. For a more detailed discussion of such problems see Herbst and Klotz (2002).
14. Within the VDE framework, a complement of a verb is classified as obligatory
if it has to be realized when the verb is used. This is different from approaches
in which the distinction between obligatory and optional complements is based
on the use of a verb in active declarative clauses. Cf. Herbst et al. (2004: xxx–
xxxiii).
15. Valency patterns are to be seen as verb specific patterns not as general patterns
in the sense that the concept of patterns was used in early structuralism. In this
respect, the concept of valency patterns also differs from similar ideas in con-
struction grammar. In particular, no claims will be made as to the meaning of
certain valency patterns, which is a position also taken by Engel (1977: 182):
“Insbesondere scheint es mir nicht möglich, jedem Satzbauplan, wie Weisger-
ber offenbar wollte, eine spezifische B ed e ut u n g zuzuschreiben.” Valency
patterns can be modified in their syntactic realization by other syntactic factors
such as theme/rheme considerations, types of clauses etc. For possible varia-
tions on Satzbaupläne see Engel (1977: 183).
16. See, for instance, Croft and Cruse (2004).
17. Compare also MacWhinney (1998 or 2001) for accounts of emergentist ap-
proaches.
 Valency complements or valency patterns? 33

References

Ágel, Vilmos
2000 Valenztheorie. Tübingen: Gunter Narr Verlag.
Bybee, Joan
1995 Regular morphology and the lexicon. Language and cognitive pro-
cesses 10 (5): 425–455.
Chomsky, Noam
1986 Knowledge of Language: Its Nature, Origin and Use. New
York/Westport, Connecticut/London: Praeger Publishers.
2004 The Generative Enterprise Revisited. Berlin/New York: Mouton de
Gruyter.
Coseriu, Eugenio
1973 Probleme der strukturellen Semantik. Tübingen: Gunter Narr Verlag.
Croft, William and D. Alan Cruse
2004 Cognitive Linguistics. Cambridge: Cambridge University Press.
Cruse, David A.
1986 Lexical Semantics. Cambridge: Cambridge University Press.
de Saussure, Ferdinand
1916 Cours de Linguistique Générale (ed. by Charles Bally and Albert
Séchehaye). Berlin: Mouton de Gruyter.
Emons, Rudolf
1974 Valenzen englischer Prädikatsverben. Tübingen: Max Niemeyer
Verlag.
1978 Valenzgrammatik für das Englische. Tübingen: Max Niemeyer Ver-
lag.
Engel, Ulrich
1977 Syntax der deutschen Gegenwartssprache. Berlin: Erich Schmidt
Verlag.
Engel, Ulrich, and Helmut Schumacher
1976 Kleines Valenzlexikon deutscher Verben. Tübingen: Gunter Narr
Verlag.
Engelen, Bernhard
1975 Untersuchungen zu Satzbauplan und Wortfeld in der geschriebenen
deutschen Sprache der Gegenwart. München: Max Hueber Verlag.
Fillmore, Charles
1968 The case for case. In Universals in Linguistic Theory, Emmon Bach,
and Robert T. Harms (eds.), 1–88. New York: Holt, Rinehart and
Winston.
2007 Valency issues in FrameNet. This volume.
Fillmore, Charles, Paul Kay, and Mary Catherine O’Connor
1988 Regularity and idiomaticity in grammatical constructions: The case of
let alone. Language, 501–538.
Götz-Votteler, Katrin
2007 Describing semantic valency. This volume.
34 Thomas Herbst

Goldberg, Adele
2006 Constructions at Work. The Nature of Generalizations in Language.
Oxford/New York: Oxford University Press.
Haegeman, Liliane
1991 Introduction to Government & Binding Theory. Oxford/Cambridge:
Blackwell.
Hausmann, Franz Josef
1985 Kollokationen im deutschen Wörterbuch. Ein Beitrag zur Theorie des
lexikographischen Beispiels. Lexikographie und Grammatik, Henning
Bergenholtz, and Joachim Mugdan (eds.), 118–129. Tübingen: Nie-
meyer.
Helbig, Gerhard
1992 Probleme der Valenz- und Kasustheorie. Tübingen: Max Niemeyer
Verlag.
Helbig, Gerhard, and Wolfgang Schenkel
1973 Wörterbuch zur Valenz und Distribution deutscher Verben. 2d ed.
Leipzig: VEB Verlag Enzyklopädie.
Herbst, Thomas
1983 Untersuchungen zur Valenz englischer Adjektive und ihrer Nominali-
sierungen. Tübingen: Gunter Narr Verlag.
forthc. “Valency – item-specificity and idiom principle.” Exploring the
Grammar-Lexis Interface. Ute Römer, and Rainer Schulze (eds.).
Amsterdam/Philadelphia: John Benjamins.
Herbst, Thomas, and Michael Klotz
2002 Meeting and Kissing as valency problems – some remarks on the
treatment of reciprocity and reflexivity in a valency description of
English. In Reflexives and Intensifiers: The Use of self-Forms in Eng-
lish, Ekkehard König, and Volker Gast (eds.), 239–249. (Zeitschrift
für Anglistik und Amerikanistik 2002.3.) Tübingen: Stauffenburg Ver-
lag.
2003 Lexikografie. Paderborn: Schöningh (UTB).
Herbst, Thomas, David Heath, and Hans-Martin Dederding
1980 Grimm’s Grandchildren. Current Topics in German Linguistics.
London/New York: London.
Herbst, Thomas, David Heath, Ian Roe, and Dieter Götz
2004 A Valency Dictionary of English. A Corpus-Based Analysis of the
Complementation Patterns of English Verbs, Nouns and Adjectives.
Berlin/New York: Mouton de Gruyter. [VDE]
Herbst, Thomas, and Susen Schüller
forthc. Introduction to Syntactic Analysis. A Valency Approach. Tübingen:
Narr.
Heringer, Hans Jürgen
1970 Theorie der deutschen Syntax. München: Max Hueber Verlag.
1996 Deutsche Syntax Dependentiell. Tübingen: Stauffenburg.
 Valency complements or valency patterns? 35

Klotz, Michael
2000 Grammatik und Lexik: Studien zur Syntagmatik englischer Verben.
Tübingen: Stauffenburg.
2007 Valency rules? The case of verbs with propositional complements.
This volume.
Leech, Geoffrey
1983 Principles of Pragmatics. London/New York: Longman.
Matthews, Peter
1981 Syntax. Cambridge: Cambridge University Press.
MacWhinney, Brian
1998 Models of the emergence of language. Annual Review of Psychology
49: 199–227.
2001 Emergentist approaches to language. In Frequency and the Emer-
gence of Linguistic Structrure, Joan Bybee, and Paul Hopper (eds.),
449–470. Amsterdam/Philadelphia: Benjamins.
Schumacher, Helmut, Jacqueline Kubczak, Renate Schmidt, and Vera de Ruiter
2004 VALBU – Valenzwörterbuch deutscher Verben. Tübingen: Gunter
Narr Verlag [VALBU].
Sinclair, John
1991 Corpus, Concordance, Collocation. Oxford: Oxford University Press.
2004a The search for units of meaning. In Trust the Text. Language, corpus
and discourse, John Sinclair (ed., with Ronald Carter), 24–48. Lon-
don/New York: Routledge.
2004b Lexical grammar. In Trust the Text. Language, corpus and discourse,
John Sinclair (ed., with Ronald Carter), 164–193. London/New York:
Routledge.
Tomasello, Michael
2003 Constructing a Language. A Usage-Based Theory of Language Ac-
quisition. Cambridge, Mass./London: Harvard University Press.

Framenet: http://framenet.icsi.berkeley.edu/
Describing semantic valency
Katrin Götz-Votteler

1. Semantic valency

Valency theory as outlined by Tesnière in the 1950s was a primarily syn-

tactic theory. But Tesnière already briefly characterized the actants of a
verb according to their semantic function or, to use his metaphor, according
to their roles as actors in a petit drame (Tesnière 1965: 102). The prime
actant is said to carry out the action (“fait l’action”), the second actant in
an active sentence to support the action (“supporte l’action”), and the tiers
actant to be the one that profits or suffers from the action (“l’action se fait à
son profit ou à son détriment”; Tesnière 1965: 111).
Since then, valency has not only been specified as a theory, but also
been applied as an approach for linguistic description, as in the syntactical
description of such languages as German, English or Danish, for the cogni-
tive exploration of grammar or in computational linguistics.1 As a conse-
quence, the theory has increasingly opened up to semantic, communicative
or cognitive aspects.
In more recent models of valency this has led to a distinction between
different levels of valency: Allerton, for example, establishes three levels of
analysis in valency patterns: the first is concerned with semantic roles and
processes, the second deals with valency structures, i.e. with the comple-
ments required by a verb, and the third examines surface structures like
dummy subjects or transformations of active into passive clauses (Allerton
1982: 40–48).
Helbig also assumes three levels of valency, which differ however from
those established by Allerton: on his first level, logical valency, mental
relations between logical predicates and related elements are analysed. The
logical structure of swim, for instance, comprises one element, called ar-
gument (“Argument”), as swimming is usually done by only one person,
the logical structure of visit, on the other hand, consists of two elements, as
visiting requires a visitor as well as a visitee (Helbig 1992: 7).
At the second level, semantic valency, the semantic properties of the ar-
guments are specified by using semantic properties as well as semantic
cases (Helbig 1992: 8). The third level, syntactic valency, then deals with
the syntactic properties of arguments, which are at this level called actants
or complements (“Aktanten” / “Ergänzungen”) i.e. it concerns their syntac-
38 Katrin Götz-Votteler

tic realisation or their status in relation to optionality or obligatoriness

(Helbig 1992: 9).
What these two models have in common is that they establish a syntac-
tic side of valency on the one hand, a semantic side on the other and some
kind of relationship between the two. The nature of this relationship, how-
ever, is contested: syntactic valency can be seen as an indirect reflection of
semantic relations, it can be regarded as being directly determined by the
verb meaning, or valency can be considered an exclusively semantic phe-
nomenon, i.e. as part of the meaning itself (Helbig 1992: 16).2
While no agreement concerning the exact nature of the relationship be-
tween syntactic and semantic valency has yet been reached, common
ground is shared in the acceptance of the utility of some kind of semantic
description in a valency model. As summarised by the Valency Dictionary
of English (VDE 2004: xxix): “The semantic analysis of valency comple-
ments addresses two questions: firstly, the meanings of the complements,
especially the difference or parallels in meaning between various comple-
ments of the same word; secondly, which lexical items can (or cannot) oc-
cur as a particular complement.”

2. Possible ways of describing semantic valency

It is not only the relationship between syntactic and semantic valency that
has not yet been agreed upon, but also which method to use in order to
describe the meaning of complements. Basically, four approaches can be
distinguished here: semantic roles, semantic components, semantic catego-
ries and a verb-specific description.
The concept of a more or less fixed set of semantic roles (also called
“deep cases”, “thematic roles” or “theta-roles”) goes back to Charles Fill-
more (1968) and is mainly used in theoretical approaches to valency.3 The
following chart taken from Allerton (1982: 52) shows a semantic descrip-
tion of several zero-, mono-, di- or trivalent verbs applying a semantic role
model:

rain: (0)
sneeze: (1) subject “patient”
blow: (1) subject “agent/force”,
(2) (object “patient/result”)
see: (1) subject “experiencer”,
(2) object “mental focus”
 Describing semantic valency 39

read: (1) subject “agent”,

(2) (object “mental focus”)
tear: (1) subject “agent”,
(2) object “patient”
give: (1) subject “agent”,
(2) object “patient”,
(3) indirect object “recipient”

More applied approaches, such as VerbNet, also use semantic roles as part
of – in this case – the lexicographical description of a verb, as the following
extract from the entry blow (in the meaning “free of obstruction by blowing
air through: ‘blow one’s nose’”) shows:

Agent[+animate]
Cause[]
Patient[+body_part]
Recipient[+animate]
Theme[+communication]
(http://www.cis.upenn.edu/group/verbnet/)

As can be seen from this entry, VerbNet complements the semantic de-
scription by additionally listing semantic components. Such components as
e.g. +/-animate, +/-human or +/-abstract serve to identify semantic proper-
ties of the noun usually used as complement with a certain verb, as the fol-
lowing entry from Helbig and Schenkel’s Wörterbuch zur Valenz und Dis-
tribution deutscher Verben (51980: 176) shows:
40 Katrin Götz-Votteler

In contrast to semantic roles, semantic components do not depend on the

meaning of the verb, but can be regarded as properties of the noun, i.e. they
are noun-inherent.
Related to the description of semantic valency with semantic compo-
nents is a third option, namely the classification of syntactical complements
into semantic categories. This is done for example in the Valenzwörterbuch
deutscher Verben VALBU, which identifies for each complement possible
categories the expression can refer to, like person, animal, object or force
(VALBU 2004: 89).4 In contrast to semantic components, semantic catego-
ries as defined in VALBU do not refer to inherent properties of a noun, but
to the usage of a noun within the semantic range of a certain category:

Diese Bestimmungen sind nicht so zu verstehen, dass den Wörtern, die an

diesen Stellen eingesetzt werden können, diese Kategorien als inhärente
Merkmale zukommen. Es wird damit behauptet, dass ein Sprecher, der die-
se Verbvariante korrekt verwendet, die Belegung der NomE als einen Aus-
druck interpretiert wissen will, der auf eine Person oder ein Gremium oder
eine Institution referiert. (VALBU 2002: 62)
[These descriptions cannot be regarded as inherent properties of the words
used as complements. They signify that a speaker using this verb correctly
in this meaning wants the expression to be interpreted as a person or a group
or an institution.]

VALBU employs semantic categories in combination with a further option

of describing semantic valency, the verb-specific description of partici-
pants. Therefore, the nominative complement of sich verletzen 1, for exam-
ple, is characterized as follows: “dasjenige, das eine Verletzung durch et-
was erleidet: Person/Tier” (VALBU 2004: 790).
The advantages of a verb-specific description lie in its accuracy and
completion of the verb meaning. This is why this method is mostly used in
lexicographical frameworks, as in VALBU or VDE (2004):

These four different methods of specifying semantic valency, semantic

roles, semantic components, semantic categories and verb-specific descrip-
tion, draw on different starting points of description, as already suggested
above: the semantics of complements can be seen as a result of the verb
 Describing semantic valency 41

meaning; this is the case with semantic role models and verb-specific de-
scription. Characterization of complements via semantic components uses
however noun-inherent properties, so that the head of the noun phrase be-
comes the determining factor of description. Finally, semantic categories
are understood as usage-based categories and therefore as a reflection of
verb pragmatics.
As verbs can be regarded as the core of valency theory, the two methods
that take the verb as starting point appear particularly attractive within a
valency framework. The following section therefore concentrates on these
two methods, semantic roles and verb-specific description, evaluating their
application in descriptions of semantic valency.

3. Semantic roles and verb-specific description

The usefulness of descriptive methods is best assessed by their application.

The following examples show usages of the verb fly; in examples (1–12) fly
takes two noun phrase complements and possibly one or more adverbial
complements, examples (13–16) illustrate a divalent use of fly with one
noun phrase complement and one adverbial complement. The labels after
the examples refer to verb-specific classes of participants that occur as
complements with fly.5

(1) Stan flew helicopters in Vietnam.LDOCE

→ PILOT / MACHINE
(2) He had flown a distance of 169 miles in 3 hours 40 minutes and
established a record.VDE
→ PILOT / DISTANCE
(3) Under the Brabazon recommendations, the aircraft would have six
to eight piston or reciprocating engines, possibly replaced by gas
turbine engines if available, and a pressurized cabin, and be ca-
pable of flying 5,000 miles non-stop at 275 miles per hour.BNC
→ MACHINE / DISTANCE
(4) I find it a little horrifying to think that some Commercial pilots
flying large numbers of passengers may have had very little expo-
sure to reduced ‘g’.BNC
→ PILOT / PASSENGER
42 Katrin Götz-Votteler

(5) British Airways flies passengers to over 150 destinations around

the world and has a cashhandling requirement at each of its air-
ports.NET
→ AIRLINE / PASSENGER / DESTINATION
(6) The couple flew British Airways.VDE
→ PASSENGER / AIRLINE
(7) The camps were spotted by pilots flying supplies to Nagorny
Karabakh, which has been sealed off from the rest of Azerbaijan
for several months.BNC
→ PILOT / GOODS / DESTINATION
(8) From the hub of the operation, the Ilopango Air Base in El Salva-
dor, creaking aircraft flew supplies to the contras’ northern front
billetted in Honduras.BNC
→ MACHINE / GOODS / DESTINATION
(9) His company flew him to Rio to attend the conference.LDOCE
→ INSTITUTION / PASSENGER / DESTINATION
(10) The fully refurbished, vintage plane then flies passengers to Bullo
River Station for three nights at the half-million acre, traditional
Outback cattle ranch before continuing to Kakadu National Park
to accent Aboriginal culture.NET
→ MACHINE / PASSENGER / DESTINATION
(11) The airline flew 89 flights to Sweden in 1941.VDE
→ AIRLINE / OPERATION / DESTINATION
(12) And one of Arrow’s pilots, Jacobo Bolivar, was one of the three
principals of Sur International, the company which flew weapons
from the US to Iran for Lieutenant-Colonel Oliver North in
1985.BNC
→ INSTITUTION / GOODS / SOURCE / DESTINATION
(13) Britain cancelled a 1960 no-visa agreement in June and imposed
entry visas for all Turks after more than 1,500 Turks flew to Brit-
ain and applied for political asylum.BNC
→ PASSENGER / DESTINATION
(14) The symptoms are generally worse the further you fly and they are
more marked after eastward flights than those to the west.BNC
→ PASSENGER / DISTANCE
 Describing semantic valency 43

(15) The plane flew up the fjord, which seemed so narrow that the
mountains were on both wing tips at the same time.BNC
→ MACHINE / ROUTE
(16) While supplies are flying in from all over the world, making sure
that everybody over such a large area receives them is another
matter.NET
→ GOODS / DESTINATION / SOURCE

Eleven different classes of participants can be identified here: PILOT, MA-

CHINE, AIRLINE, PASSENGER, GOODS, INSTITUTION, OPERATION, DESTINA-
TION, SOURCE, DISTANCE and ROUTE. The categories PILOT, PASSENGER,
AIRLINE, MACHINE, INSTITUTION and GOODS can be found in subject posi-
tion.
So far, the complement description with the above categories has been
entirely verb-specific. In order to transfer it to a semantic role model with a
fixed set of roles, it would be necessary to assign these labels to specific
roles, e.g. PILOT – AGENT, PASSENGER – PATIENT. However, the syntactic
arrangement of a clause seems to affect the role of a specific participant, or
in other words, seems to reflect the role of the participant in a specific
situation. Consider the examples in which the same type of participant is
realised in different syntactic positions:

(5) British Airways flies passengers to over 150 destinations around

the world and has a cashhandling requirement at each of its air-
ports.NET
→ AIRLINE / PASSENGER / DESTINATION
(6) The couple flew British Airways.VDE
→ PASSENGER / AIRLINE

The two participants PASSENGER and AIRLINE both occur once in subject
and once in object position, and it seems as if the participant in subject
position is interpreted as the more agentive, more performing character: in
example (5) British Airways can be regarded as the entity doing something,
whereas in example (6) the couple seems to play a more agentive role than
the airline. This result goes hand in hand with findings that agentive entities
are usually encoded in a more fronted position than less agentive entities:

Nominals denoting figures of state and sources of actions consistently pre-

cede those denoting grounds and recipients, except in presentative contexts.
They are also chosen as perspectives. (Sridhar 1988: 82)
44 Katrin Götz-Votteler

These considerations can lead to the hypothesis that elements in subject

position are generally interpreted as the performing character within a spe-
cific situation. This hypothesis also includes inanimate entities:

(3) Under the Brabazon recommendations, the aircraft would have six
to eight piston or reciprocating engines, possibly replaced by gas
turbine engines if available, and a pressurized cabin, and be ca-
pable of flying 5,000 miles non-stop at 275 miles per hour.BNC
→ MACHINE / DISTANCE
(15) The plane flew up the fjord, which seemed so narrow that the
mountains were on both wing tips at the same time.BNC
→ MACHINE / ROUTE

These sentences seem to reflect a perspective within which the MACHINE is

perceived as the only entity that fits the description fly or as the only rele-
vant entity in that process. Other entities to which the role AGENT could be
attributed might either not be perceived, be unknown or just irrelevant. It
can therefore be said that the MACHINE is presented as the participant per-
forming the action of flying.
If the subject is generally perceived as the agentive entity, the conclu-
sion must be drawn that semantic roles cannot be separated from syntactic
structure. Such separation would only be possible if different lexical units
for fly in sentences like (5) and (6) were assumed. Then, fly in example (5)
could be paraphrased as ‘operating flights as an airline’ and British Airways
could be identified as acting entity. On the other hand, a sense like ‘travel
somewhere by plane as a passenger’ would assign the agentive role to the
couple. But even then, the participant in subject position would still be
regarded as performing participant, so that the correlation between subjec-
tivity and agentivity would still hold. Furthermore, assuming different lexi-
cal units for each type of possible subject would come very close to using a
verb-specific description, so that the general applicability of semantic roles
would be weakened. It can therefore be concluded that it is not possible to
assign general semantic roles to a specific type of participant in a one-to-
one relationship.
The hypothesis above can also be extended to ergative verbs. The
Collins Cobuild English Grammar describes ergative verbs as follows:
“Some verbs allow you to describe an action from the point of view of the
performer of the action or from the point of view of something which is
affected by the action.” (1990: 155). In other words: the semantic role of
the participant in subject position is that of the PATIENT.
 Describing semantic valency 45

Following the hypothesis outlined above leads, however, to the conclu-

sion that the subject of an ergative verb is not presented in the role of the
PATIENT but from a perspective within which it seems as if the entity in
subject position were the performer of the action. Consider the following
examples:

(17) He opened the door wide, and gestured for me to come in.LDOCE
→ OPENER / DOOR
(18) The new key opened the door but did not work in the ignition.NET
→ KEY / DOOR
(19) The bedroom door opened and she rushed in.BNC
→ DOOR
(20) After a short discussion with the customs officers, the gates
opened and the truck moved off.LDOCE
→ DOOR
(21) That window doesn’t open.LDOCE
→ WINDOW

The entities in subject position differ semantically: in example (17) he is a

human performer; here, the semantic role AGENT can be assigned without
much hesitation. In example (18), it seems as if it was not in the power of a
human entity to open the door but in the “power” of a key. The key can
therefore be regarded to be the entity responsible for the opening action.
The second clause in example (19) makes it clear that the event is de-
scribed from inside the room. From this point of view, what is seen first is
the moving of the door, i.e. the door is the only participant perceived to be
performing an action. Consequently, it is presented in the position of the
performing participant.
Example (20) is very similar: possibly, the gates open automatically, so
that the element causing the movement, be it the mechanism or an officer
pressing a button, is not perceived. (By the way, would the mechanism or
the officer be regarded as AGENT?) In this specific situation it is again only
the gates which can be seen to be moving. Finally, in the last example it is
impossible for any human actor or for any tool to open the window. Thus,
the “action of non-opening” must be assigned to the window.
Therefore we can sum up our conclusions by saying that in ergative
constructions the entity in subject position does not fulfil the semantic role
PATIENT. Such constructions reflect a certain perspective on a situation in
which the entity in subject position is perceived as the only entity doing
46 Katrin Götz-Votteler

something. The real causer is either not perceived, not known or considered
irrelevant. As put by William Croft (1994: 95):

The subject is the “ultimate cause” in that by making a participant a subject,

the speaker has chosen – to the extent allowed by the grammar of his or her
language – to represent the participant as not significantly acting under con-
trol of someone or something else. In the case of human agency, this nor-
mally implies that the subject is controlling the action himself/herself; how-
ever, lesser degrees of control are not incompatible with being assigned to
subject position ….

This does, however, not apply to passive constructions. The form of the
verb in a passive clause assigns a passive role to the entity in subject posi-
tion and leads to the inference that a more agentive entity is present in the
situation.6 I do therefore not agree with the Collins Cobuild English Gram-
mar, which states: “Note that ergative verbs perform a similar function to
the passive because they allow you to avoid mentioning who or what does
the action” (1990: 157). Whereas in the passive construction the verb form
indicates the presence of a further, more agentive participant, in the erga-
tive construction the entity in subject position is presented as the perceived
performer of the verb action.

4. Conclusion

The analysis above can be summed up as follows:

1. The perception of a certain event that is to be encoded linguistically is
selective.
2. The manner in which a specific situation is perceived is reflected in the
syntactic realisation of a clause.
3. The subject position is assumed by that entity that seems to be perform-
ing the action expressed by the verb. If the “real” causer cannot be per-
ceived or does not play a role in the specific situation, the participant
next likely to perform the action or process is realised as subject.
4. It follows therefore that a model that assumes a fixed relationship be-
tween a verb-specific participant (e.g. PASSENGER) and a semantic role
label (e.g. PATIENT) is not suitable to describe semantic valency.
Thus, it can be concluded that a verb-specific description of participants
can be regarded as the linguistically as well as a psychologically more ac-
curate way to describe semantic valency.
The statements above can also explain hierarchies of semantic features
that are used to express the likelihood of an entity occurring in subject posi-
 Describing semantic valency 47

tion. The following chart taken from Givón (1984: 107) sums up the most
important of these features:

a. Humanity: human > animate > inanimate > abstract

b. Causation: direct cause > indirect cause > non-cause
c. Volition: strong intent > weak intent > non-voluntary
d. Control: clear control > weak control > no control
e. Saliency: very obvious / salient > less obvious / salient > unobvious /
nonsalient

If the subject refers to the participant perceived as the most agentive one in
a situation, then it is just a matter of likelihood that this entity can be de-
scribed as salient, human and intentionally acting. If such a participant is
not present, then the next likely entity to be encoded as the performing
character will possibly be a salient, animate and more or less intentionally
acting one, and the next one perhaps a less salient, animate, less intention-
ally acting one and so on:

To wit, a human is closer to the ego, thus more familiar and obvious. Direct
causes tend to be perceptually more obvious, occupying a clear boundary posi-
tion within the chain (as also does the effect, which is categorically coded as
patient). Intermediate points in the chain are less salient. Strong intent creates a
higher probability of success, i.e. visible effect. Ditto for strong control. (Givón
1984: 107)

All that has been said so far applies to sentences that linguistically encode a
non-linguistic event. It must not be forgotten, however, that apart from
semantic suitability, the subject fulfils a second important function in the
English language, namely that of the topic in an information unit. Very
often, these two aspects, the semantic and informational value of the sub-
ject, coincide. But, as quantitative analyses of fictional texts have shown
(Götz-Votteler forthcoming), up to four percent of the subjects can be said
to function primarily as text-organisational devices. It must therefore be
borne in mind that a description of the semantic valency of a verb applies to
the majority of usages, but cannot embrace all cases of syntactic realisation.
48 Katrin Götz-Votteler

Notes

1. Recent projects that have made use of valency theory as a syntactic model
include the Valenzwörterbuch deutscher Verben (VALBU), the Valency Dic-
tionary of English (VDE), the Odense Valency Dictionary of Danish (OVD),
FrameNet, Contragram and the Prague Dependency Treebank, to name only a
few.
2. For the relationship between syntactic complements and meaning see also
Klotz and Schøsler (both this volume).
3. The model of case roles is not only used in valency frameworks; it is, for ex-
ample, also part of syntactic description within Government and Binding The-
ory. Haegeman, for instance, lists as possible theta-roles for argument specifi-
cation AGENT, THEME, EXPERIENCER, BENEFACTIVE/BENEFICIARY, GOAL,
SOURCE and LOCATION (1991: 41–42).
5. For a list of the most important categories see VALBU (2004: 89–90).
6. The abbreviations at the beginning of the example denote the source of the
sentence: Longman Dictionary of Contemporary English (LDOCE), Valency
Dictionary of English (VDE), British National Corpus (BNC), and the internet
(NET).
9. See Rickheit/Sichelschmidt (this volume).

References

Allerton, David J.
1982 Valency and the English Verb. London/New York: Academic Press.
Croft, William
1994 Voice: Beyond control and affectedness. In Voice – Form and Func-
tion, Barbara Fox, and Paul J. Hopper (eds.), 89–117. Amster-
dam/Philadelphia: John Benjamins Publishing Company.
Fillmore, Charles
1968 The case for case. In Universals in Linguistic Theory, Emmon Bach,
and Robert T. Harms (eds.), 1–88. New York: Holt, Rinehart and
Winston.
Givón, Gerhard
1984 Syntax – A Functional-typological Introduction Vol. 1. Amster-
dam/Philadelphia: John Benjamins Publishing Company.
Götz-Votteler, Katrin
forthc. Aspekte der Informationsentwicklung im Erzähltext.
Haegeman, Liliane
2003 Reprint. Introduction to Government & Binding Theory. Ox-
ford/Cambridge, Mass.: Blackwell.
Helbig, Gerhard
1992 Probleme der Valenz- und Kasustheorie. Tübingen: Max Niemeyer
Verlag.
 Describing semantic valency 49

Helbig, Gerhard, and Wolfgang Schenkel

1980 Wörterbuch zur Valenz und Distribution deutscher Verben. 5th edi-
tion. Leipzig: Bibliographisches Institut.
Herbst, Thomas, David Heath, Ian Roe, and Dieter Götz (eds.)
2004 A Valency Dictionary of English. Berlin/New York: Mouton de
Gruyter.
Klotz, Michael
2007 Valency rules? The case of verbs with propositional complements.
This volume.
Rickheit, Gert, and Lorenz Sichelschmidt
2007 Valency and cognition – a notion in transition. This volume.
Sridhar, Shikaripur N.
1988 Cognition and Sentence Production – A Cross-linguistic Study. New
York/Berlin/Heidelberg/London/Paris/Tokyo: Springer-Verlag.
Schøsler, Lene
2007 The status of valency patterns. This volume.
Schumacher, Helmut, Jacqueline Kubczak, Renate Schmidt, and Vera de Ruiter
(eds.)
2004 VALBU – Valenzwörterbuch deutscher Verben, Tübingen: Gunter
Narr Verlag.
Sinclair, John
1990 Collins Cobuild English Grammar. London/Glasgow: Collins.
Summers, Della
2005 Longman Dictionary of Contemporary English. 4th edition. Harlow:
Longman.
Tesnière, Lucien
1965 Éléments de Syntaxe Structurale. 2d edition. Paris: Librairie C.
Klincksieck.

VerbNet: http://www.cis.upenn.edu/group/verbnet/
The status of valency patterns1
Lene Schøsler

Twas brillig, and the slithy toves

Did gyre and gimble in the wabe;
All mimsy were the borogoves,
And the mome raths outgrabe.
Lewis Carroll: Through the Looking Glass, Jabberwocky

The aim of my paper is to discuss the following questions: is valency a

formal category or a content category in languages like French, English,
and Danish? And if so – or if this is so in part – does such a thing as gram-
maticalisation of valency patterns exist? 2
Concerning the first question, the status of valency, this question can be
rephrased in the following two hypotheses:
a. The valency pattern expresses a cognitive structure; there is a link or an
iconic relation between a valency pattern and the semantic content of
the same pattern;
b. The valency pattern does not express a cognitive structure; there is a
purely grammatical structuration without any link or iconic relation be-
tween expression and content.
When Hopper and Thompson (1980), for example, talk about prototypical
relations of transitivity, they mean a human agent carrying out an action
having an impact on a non-human object. This way of seeing things implies
a valency structuration of the world with, at its centre, an active human
being. So this view suggests an interpretation of valency patterns following
hypothesis (a). On the other hand, hypothesis (a) has to be rejected if we
consider e.g. verbs of perception and verbs of emotion, as already pointed
out, e.g. by Krefeld (1998), because these verbs do not have a prototypical
agent as subject. Others, e.g. François and Broschart (1994: 40–41) insist
that valency patterns are not very informative on the meaning conveyed by
the constructions. Hypothesis (a) is easily adapted to a functional approach
and hypothesis (b) to a formal approach. The goal of my paper is to show
that hypothesis (a) is confirmed at least in some cases and that these cases
can be considered to be cases of specialisation or grammaticalisation of the
valency pattern. My discussion of the status of valency is based on syn-
chronic arguments from Danish, French and English and on diachronic
arguments from French and Danish.
52 Lene Schøsler

When we consider quantitative valency patterns, it appears that some

patterns seem to be very common and others fairly rare. Thus, in e.g. the
Odense Valency Dictionary of Danish (OVD)3, there are eight valency tran-
sitive patterns,4 of which two are extremely common (see table 1) and do
not contain verbs having the same or similar content (see examples 1a and
1b), whereas five other patterns, less frequent ones, contain verbs with re-
lated content (see examples 1c-h). This suggests that at least some patterns
specialise to provide a particular content, whereas others do not. Thus, in
many languages meteorological verbs are avalent,5 and with respect to
valency, they constitute a formal class with a specific content. They are
even productive in the sense that other verbs when used with this specific
pattern will provide a meteorological sense. This is the case in (2a), where
the Danish verb pøse (‘pour’), especially with the particle ned (‘down’), is
used to indicate heavy rain, which is not the sense of the verb when used
with a referential subject (2b). A non-referential subject is impossible with
the divalent valency pattern of the verb pøse (2c).

Table 1. Valency patterns of the OVD.6 Patterns provided with the symbol # con-
tain verbs with related content.

Valency patterns with direct object number of verbs

in the OVD
subject – object (1a) 907
subject - object - indirect object # (1b) 33
verbs of donation and transfer
subject - object - prepositional object (1c) 335
subject - object - prepositional object_1 – (1d) 12
prepositional object_2 #
verbs indicating change of state or change
of concrete or abstract location
subject - object - locative object (1e) 37
subject - object - directional object # (1f) 40
causative verbs of movement
subject - object - object of quantity # (1g) 4
verbs of weighing and paying
subject - object - object of manner # (1h) 54
verbs of personal evaluation or of utilisation

1422
 The status of valency patterns 53

(1) a. Hun afskedigede ham

‘she dismissed him’
b. Hun gav ham bogen
‘she gave him the book’
c. Hun minder mig om hendes mor
‘she reminds me of her mother’
d. Hun oversætter bogen fra dansk til fransk
‘she translates the book from Danish into French’
e. Hun lægger bogen på bordet
‘she puts the book on the table’
f. Han sænker dykkerklokken ned mod havets bund
‘he lowers the diving bell towards the bottom of the sea’
g. Hun betaler ham en million euro
‘she pays him a million euros’
h. Han betragter ham som sin ven
‘he considers him his friend’
(2) a. Det pøsede ned i haven
lit. ‘it poured down in the garden, it rained heavily in the
garden’
b. Han pøser suppe ned i skålen
‘he pours soup into the bowl’
c. *Det pøser suppe ned i skålen
lit. ‘it pours soup into the bowl’

The conclusions to be drawn from the above are: 1, that some valency pat-
terns clearly provide information on the content of verbs having this pattern
(1b, 1d, 1f-h) and 2, that change of pattern, e.g. from avalent to valent con-
structions (2a-b), implies change of content. These points are arguments in
favour of a “light” version of hypothesis (a), implying that at least in some
cases valency is a content category.
Now, let us consider more examples in order to see how widespread
changes of patterns implying change of content are. Cases like these could
indeed be arguments in favour of considering valency patterns a sort of
expression-content-paradigm for verbs, which is a condition for interpreting
these patterns as part of the grammar and not exclusively as part of the
lexicon.
Beth Levin has studied verbal alternations as e.g. in (3a-5b) for English;
see table 2. These cases of alternation imply: 1, a change of verbal aspect
(the a-examples being telic, the b-cases imperfective, as illustrated by the
adverbs); 2, a change of entailment (the a-cases implying that the action has
been completed); and 3, a change of focus (the focus being on the direct
54 Lene Schøsler

objects). Changes like these are known in several languages; (6) is a trans-
lation of (4) into French, and (7) into Danish, with corresponding differ-
ences of verbal aspect, entailment and focus, meaning that in the a-
examples the wall is painted, the car is full of boxes and John has learnt
linguistics, whereas in the b-examples the activity has not come to a (suc-
cessful) completion. In Danish, this alternation is very productive and has
recently spread to contexts where it did not exist previously. The pattern
shown in (8-9), with the preposition på [‘on’], is especially frequent. The
construction without the preposition på is telic, it implies that the activity
expressed by the verb has been completed. The construction with the
preposition på is imperfective; it implies that the activity is ongoing and
that it is not (yet) completed. In other words, the direct construction indi-
cates that the president is dead and the book is finished, whereas the indi-
rect construction indicates that the president is still alive and the book was
not – perhaps never – finished (see Durst-Andersen and Herslund 1996).
The productivity of these patterns confirms, I think, my interpretation of
them as valency-paradigms with specific content.

Table 2. Alternations of valency patterns

telic action imperfective activity

(3a) John sprayed the wall with paint (3b) John sprayed paint on the wall
(in an hour) (for an hour)
(4a) Julie loaded the car with boxes (4b) Julie loaded boxes into the car
(in an hour) (for an hour)
(5a) Mary taught John linguistics (in (5b) Mary taught linguistics to John
an hour) (for an hour)
(6a) Julie a chargé la voiture de (6b) Julie a chargé des caisses sur la
caisses (en une heure) voiture (pendant une heure)
(7a) Julie har læsset vognen med (7b) Julie har læsset kasser på
kasser (på en time) vognen (i en time)
(8a) skyde præsidenten [‘shoot the (8b) skyde på præsidenten [lit. ‘shoot
president’] on the president’]
(9a) skrive en bog [‘write a book’] (9b) skrive på en bog [lit. ‘write on a
book’]

Alternating constructions of the type Marie lui casse le bras, Marie casse
son bras [lit. ‘Marie him breaks the arm’ – ‘Mary breaks his arm’], with
alternating patterns combined with inalienable possession, provide other
well-known pairs that I will consider as valency-paradigms with specific
content. The case is especially clear in Danish, where we find two different
 The status of valency patterns 55

alternation patterns depending on the nature of the verb, a pleasant or an

unpleasant action; see examples (10-11).7

(10) a. jeg kysser ham på kinden b. jeg kysser hans kind

lit. ‘I kiss him on the cheek’ ‘I kiss his cheek’
(11) a. jeg bider ham i kinden b. jeg bider i hans kind
lit. ‘I bite him in the cheek’ lit. ‘I bite in his cheek’

In both cases, there is an agent (I) a possessor (he) and a possessum (the
cheek). In all four examples, I am doing something to him. In the a-
constructions, I am kissing or biting an inalienable part of him, i.e. his
cheek. In the b-constructions, the cheek is presented as not forming an in-
alienable part of him, indeed, the cheek may not be a part of him at all – he
could be a student of medicine, having a cheek-sample for experimentation.
The difference in the constructions (10) and (11) lies in the way the same
constituents are expressed in the parallel constructions. In the a-
constructions, the possessor is expressed as a direct object and the posses-
sum is expressed as a locative; in (10b), they are merged together as the
direct object in the alternating construction; in (11b), they are merged to-
gether as a locative in the alternating construction. Pleasant action-verbs
behave like (10), unpleasant like (11).
Having found valency patterns with specific content and even the possi-
bility of alternation patterns with alternating content, I find it legitimate to
conclude that valency patterns are not purely formal categories, they are –
at least sometimes – content categories. Now, the question is whether we
can find cases where these patterns become “more grammatical”, in the
sense that patterns that used to be open to different verb senses specialise in
order to contain only verbs of the same or related sense. If such a case ex-
ists, I would propose that we have here a case of grammaticalisation of a
valency pattern.
The verbs listed in table 4 of Maurice Gross’ verb lexicon (Gross 1975)
with the divalent pattern subject-indirect object are, I believe, such a case.8
In spite of some differences between the verbs that I cannot comment on
here,9 the verbs of Gross’ table 4 share the particular feature that they all
express – more or less clearly – a psychological relation between on the one
hand a human being who is the experiencer that I will label E and who has
pleasant or unpleasant feelings, and on the other hand another human being
or an object being the cause of these feelings that I will label O. The verbs
of Gross’ table 4 have in common the fact that E is expressed as the indirect
object and O as the subject. Interestingly, the valency patterns of these
verbs have changed over time. Let us just consider three verbs belonging to
56 Lene Schøsler

our list, obéir, ressembler and mentir in a diachronic perspective.10 None of

these verbs exhibit identical patterns in Latin: the etymon of obéir (‘obey’)
was followed by the dative, the etymon of mentir (‘lie’) was followed by
the accusative, and the etymon of ressembler (‘ressemble’) had both con-
structions. In Old French the three verbs had both patterns, and it is not
before Modern French that we find one and the same pattern SVIO for all
three verbs; see table 3, where I present the case-indications throughout the
periods: “accusative” for the direct object and “dative” for the indirect ob-
ject.

Table 3. The evolution of the case-marking for the etymons of the French verbs
obéir, ressembler and mentir

Latin Old Middle 16th 17th Modern

French French century century French
obéir --- accus accus accus --- ---
dative dative dative --- dative dative
ressembler accus accus accus accus accus ---
dative dative dative dative dative dative
mentir accus accus accus accus --- ---
--- dative --- --- dative dative

Now, let us consider a few cases more closely. Following Koch (2001), I
will consider that the verbs of the list, such as plaire, déplaire, répugner
express what can be called “the perspective of O”, implying that O is coded
as subject. Other verbs, that do not belong to the list, such as adorer, aimer,
apprécier, détester, express what can be called “the perspective of E”, im-
plying that E – the experiencer – is coded as subject. Table 4 illustrates how
the verbs are opposed with respect to this perspective.

Table 4. Expression of the perspective of O vs. E with different verbs

O = subject; SVIO E = subject; SVO

plaire [‘please’] adorer, aimer, apprécier [‘like’]

déplaire, répugner etc. [‘displease’] détester, ne pas aimer [‘dislike’]
il me semble que cela est bon [lit. ‘it je trouve que cela est bon [lit. ‘I find
seems to me that this is good’] that this is fine’]
 The status of valency patterns 57

On the other hand we sometimes find a quite different case, as in table 5,

where the same verbs choose the pattern SVIO when the O-perspective is
chosen, and SVPP when the E-perspective is chosen.

Table 5. Expression of the perspective of O vs. E with the same verbs, but
different patterns

O = subject; SVIO E = subject; SVPP

bénéficier à [‘be useful’] bénéficier de [‘profit from’]

manquer à [‘miss’] manquer de [‘lack’]
profiter à [‘be useful’] profiter de [‘profit from’]
réussir à (=lui) [‘succeed in’] réussir à (=y) [‘be successful’]

These cases show, I believe, that the pattern SVIO has specialised as ex-
pression of a relation between E and O in such a way that O is the subject
of the sentence. As none of these verbs passivise, the E-perspective cannot
be expressed by means of a passive; the E-perspective is expressed either
by a different verb as seen in table 4, or by a different pattern of the same
verb, as seen in table 5.
So, in certain cases, like the SVIO-pattern, the evolution of language has
as a result a specialisation of the valency pattern which expresses a specific
cognitive relation. We saw above, in table 3, that verbs belonging to Gross’
table 4 have adopted this pattern recently, after some hesitation. Other
verbs that did not match the content of this pattern have indeed been ejected
from it. Let us consider the case of verbs meaning ‘help’. In Old French
these verbs had very different ways of marking the person to whom help is
given: conforter and rescorre were divalent (SVO), secourir was divalent,
but hesitated between the two patterns SVO and SVIO. Later, they all
adopted the pattern SVO. In Modern French the trivalent aider has a direct
object instead of an IO indicating the person to whom help is given. In Old
French, assister and servir also had an IO. Later, all the verbs meaning
‘help’ still in use express the person helped by means of a direct object.11 I
will claim that these verbs changed their pattern because they did not con-
form to what had specialised as the content of the SVIO-pattern and I will
consider this evolution a case of grammaticalisation of the valency pattern.
Danish provides another case of what I will consider grammaticalisa-
tion, where alternations of valency pattern replace previous lexical alterna-
tions. So this is a change implying transfer from lexicon to grammar in the
way of marking a difference of content. Danish had and still has a series of
verb pairs offering the choice between a verb indicating a situation (12a-
58 Lene Schøsler

14a) and a related causative verb (12b-14b), like the English pair to lie – to
lay. In Danish, the two verbs have fused in some cases, and instead of a
previous lexical alternation visible at least in the past tense, see (15a-b), we
find an alternation between two valency patterns of what appears to be one
and the same verb (16a-b). According to Skafte Jensen (2002) Swedish still
distinguishes lexical alternations (17) in cases where Danish has alternating
valency patterns (18). Valency patterns alternating between an intransitive
and a causative (transitive) pattern are well known in other languages as
well, see (19) and (20). I believe that the change from lexical alternation to
valency alternation in Danish confirms my claim that valency patterns
combine expression and content in a paradigmatic way.

intransitive situation causative action

(12) a. stå [‘be situated’] b. stille noget et sted [‘put
something somewhere’]
(13) a. sidde [‘sit’] b. sætte noget et sted [‘put
something somewhere’]
(14) a. ligge [‘lie’] b. lægge noget et sted [‘lay
something somewhere’]
(15) a. hænge / hang [‘hang / b. hænge noget et sted /
hung’] hængte [‘hang something
somewhere / hanged’]
(16) a. hænge / hængte b. hænge noget et sted /
hængte
(17) a. brinna / brann / brunnet b. bränna noget / brände /
[‘burn / burned, burnt’] bränt [‘burn something /
burned, burnt’]
(18) a. brænde / brændte / b. brænde noget / brændte /
brændt [‘burn / burned, brændt [‘burn, burned,
burnt’] burnt’]
(19) a. the branch broke b. Peter broke the branch
(20) a. la branche a cassé b. Pierre a cassé la branche

If my claim about the grammaticalisation of valency patterns is correct, it

should be possible to test it on new verbs being brought into use. Danish,
like other modern languages, has introduced new verbs borrowed from
English, such as those meaning e.g. ‘e-mail’ and ‘fax’. These have adopted
the models for Danish verbs with related meanings. Thus, the Modern Dan-
ish verbs maile, faxe are trivalent, adopting the Danish model of to write or
to send. But the new Danish verb brainstorme, meaning to meet in order to
 The status of valency patterns 59

solve a problem (in American English: to have a brainstorm),12 does not

have such a clear Danish model, and speakers are clearly testing different
valency patterns for the verb, mostly by means of different prepositions,
especially the preposition på [‘on’] (21a-c). Even a transitive construction
is found (21d), as appears from the following examples taken from Google:

(21) Innovative Danish

a. vi brainstormer på egne og andres ideer
[lit. ‘we b. on our own and others’ ideas’]
b. vi forsøgte at brainstorme over emnet
[lit. ‘we tried to b. upon the topic’]
c. det er muligt at brainstorme frem til en kommerciel idé
[lit. ‘it is possible to b. on to a commercial idea’]
d. Lederen beder deltagerne om at “brainstorme” et emne
[lit. ‘the leader asks the participants to b. a topic’]

We have a comparable situation when a verb cannot be understood, either

because we have not heard the verb before or because the verb is a non-
sense verb. In both cases, the person trying to interpret the unknown verb
will apply different strategies, i.e. deriving the sense of the verb from the
context. This strategy makes us understand e.g. that the verbs to whiffle, to
burble and to galumph indicate sounds and manners of movement in the
following passage:

(22) And as in uffish thought he stood,

The Jabberwock, with eyes of flame,
Came whiffling through the tulgey wood,
And burbled as it came!
One, two! One, two!
And through and through
The vorpal blade went snicker-snack!
He left it dead, and with its head
He went galumphing back.
(Lewis Carroll: Through the Looking Glass, Jabberwocky)

The two verbs found at the start of the poem (the slithy toves did gyre and
gimble in the wabe) are clearly verbs of movement: to gyre is to go round
like a gyroscope, and to gimble is to make holes like a gimlet.13 But the
context does not help us to understand outgrabe quoted at the beginning of
this paper: and the mome raths outgrabe. According to Humpty Dumpty,
outgrabe is the past tense of the intransitive verb outgribe, indicating a
60 Lene Schøsler

special sound. I will claim that if the pattern of this nonsense verb had been
different and more specific, e.g. he outgrabe her on the back, or he out-
grabe it from Latin into English, we would clearly interpret this as a verb of
action like to scratch or a verb of transfer like to translate, as these valency
patterns clearly combine expression and content, as shown for Danish in
table 1. I will come back to the implications of this way of reasoning in the
following, as the contribution to this volume by Dirk Noël14 motivates me
to further clarify some key notions of my paper, especially valency patterns
vs. “constructions” and grammaticalisation. I will also need to define what
is included in grammar according to my analysis which is conform to that
of the Danish Functional School (most clearly defined in Engberg-Pedersen
et al. 1996), as this view is clearly not shared by Dirk Noël.
The notion of valency patterns used here should not be confused with
“constructions” in the sense of Goldberg (1995) or Croft (2001). It is, how-
ever, not easy to define exactly what constitutes a construction: is it an
abstraction derived from the lexical semantics of verbs having a special
valency pattern? Or does a construction correspond to the pattern and its
arguments having a specific, but necessarily abstract, meaning? It appears
from the literature that constructions should probably best be understood in
an onomasiological way, related to the meaning of frames, rather than a
semasiological one, derived from verbs. But still, in the literature, construc-
tions are always defined in terms of prototypical and derived cases. My
approach is different, as my starting point is the valency patterns and the
possibility of interpreting them as expressions of a specific content, as we
have seen above. In order to investigate this possibility, I have applied
Hjelmslev’s commutation principle (Harder 1996: 439) on changes of pat-
terns linked to change of content (examples 2a-b) and on alternations of
patterns with corresponding alternations of content (examples 3-11). Thus,
my answer to the question asked by Dirk Noël in his paper is in the af-
firmative: these are indeed cases of patterns having a grammatical meaning.
My goal has been to defend the point of view that we should consider the
existence of paradigms not only in morphology, but also in syntax, and that
the valency patterns illustrated here are comparable to morphological para-
digms, thus belonging to grammar and not to lexicon. A similar argumenta-
tion concerning a paradigmatic conception of another domain of syntax, i.e.
word order, is put forward by Heltoft (1996: 478ff.). A more general dis-
cussion of “sets of options” constituting a “paradigm” is presented by
Harder (1996: 440), and by Andersen (forthcoming b) who uses the term
paradigm “not in the narrow sense of ‘inflectional paradigm’, but in the
general sense of ‘selectional set’, a usage that has been traditional since
Saussure.”
 The status of valency patterns 61

Now, having defined the valency patterns under discussion as paradig-

matic expressions of specific content, it is legitimate to consider the possi-
bility of a grammaticalisation of these patterns. Dirk Noël rejects this pos-
sibility, firstly, because he has a narrower definition of grammaticalisation
than I have, and secondly, because he considers it a case of increase of
meaning, which goes against one of the principles of grammaticalisation.
Let us consider some of the principles of grammaticalisation.
The traditional grammaticalisation cline confuses different levels of
analyses, see Andersen (forthcoming b): “Somehow many historical lin-
guists who accepted this ‘cline’ did not notice that it confuses content,
morphosyntax, and expression: lexical > grammatical is a change in con-
tent, word > clitic > affix is a development in morphosyntax, and item > Ø
refers to phonological attrition.” Moreover, the “cline” is dependent on the
type of language: in analytical languages we will not expect grammaticali-
sation to result in affixes, so we should not cling too heavily to the “cline”,
but instead, as proposed by Andersen (forthcoming b) accept that lexical
and grammatical categories form paradigms, and that content changes
should be divided into “(i) changes into, (ii) changes within or among, and
(iii) changes out of lexical or grammatical paradigms”. I consider the
grammaticalisation of valency patterns as cases of (ii).
The second point raised by Dirk Noël concerning loss or increase of
meaning is interesting. As I see it, we have loss of lexical meaning and
increase of grammatical meaning. I suggest that the grammaticalisation of
valency patterns is the result of a reanalysis where (part of) the meaning of
the lexical verb has been inferred to the pattern.
Dirk Noël is right in pointing out that grammaticalisation is normally
followed by expansion rather than reduction. However, we find both: in the
case of productive patterns, e.g. the pattern of Danish illustrated by exam-
ple (21a), we find that it is actually rapidly spreading to verbs that did not
present this pattern previously. On the other hand, we find patterns that are
not productive any more, but remain a closed, consolidated class, like the
one presented in table 3.
Other scholars might prefer to interpret the facts that I have put forward
here in a different way. They might propose that valency patterns are re-
flections of our experience of the world, having agents doing something to
persons or things in plurivalent patterns or entities moving by themselves in
monovalent patterns, etc.; to put it differently, that we have what we could
call “natural patterns”. This is probably the way Hopper and Thompson
(1980) see it. But such a view does not account for the fact that although
languages often organise their valency patterns in similar ways, there are
important differences, even between closely related languages such as the
62 Lene Schøsler

ones referred to in this paper. Moreover, it cannot account for systematic

changes such as those presented above, for innovation with adoption of
existing valency patterns for new verbs (the case of the Danish verb brain-
storme) nor for the speaker’s reasonable interpretations of unknown verbs
based on existing patterns (see Jabberwocky, Through the Looking Glass).
If we include valency patterns as part of the grammar and accept the
changes presented here as results of inference from lexicon to grammar and
as cases of paradigmatisation of valency patterns, then it is legitimate to
refer to these changes as “grammaticalisations”,15 or rather, following Hen-
ning Andersen’s terminology, as cases of “regramma(ticalisa)tion”.16

Notes

1. This article is part of the project “Linguistic Theory and Grammatical

Change”, conducted at the Centre for Advanced Study (CAS) in Oslo in
2004/05. I want to thank Roger Wright for help, especially for his many useful
comments on a previous version.
2. Here I use the term valency pattern in the most common sense, I think, i.e. in
the sense of quantitative and qualitative valency, implying the number of
valency bound elements, the grammatical and semantic features of the valency
bound elements, etc.
3. The OVD was a project supported by the Danish research council in the 1990s
with the aim of developing the first valency dictionary for Danish verbs to be
used for human inspection as well as in computational applications. It contains
approximately 4000 verb senses corresponding to 1,900 Danish verbs. The
verbs were selected using the criteria of frequency. The OVD records different
types of information about the verb and its combinatory potential and is devel-
oped as an application of the so-called Pronominal Approach, see Schøsler and
Van Durme (1996) and Schøsler and Kirchmeier-Andersen (1997).
4. I.e. valency patterns with a subject and a direct object.
5. The term avalent (from Tesnière) is used for convenience, referring to verbs
like Latin pluit, whose subject is not referential. Avalent verbs are not in-
cluded in table 1 as they have no direct object.
6. Schøsler and Kirchmeier-Andersen (1998) contains a presentation of these
patterns and a detailed discussion of examples.
7. A thorough description of these, very frequent, alternations in found in
Schøsler and Kirchmeier-Andersen (1998).
8. The verbs are: agréer, aller, appartenir, arriver, bénéficier, chanter, convenir,
coûter, déplaire, échapper, échoir, importer, incomber, manquer, mentir, mes-
soir, nuire, obéir, parvenir, peser, plaire, prendre, profiter, répugner, ressem-
bler, réussir, revenir, seoir, sourire, tarder.
9. See Schøsler (2003) for a detailed study of these verbs.
10. See Goyens (2001).
 The status of valency patterns 63

11. Thus it is too simple to claim, as does Hans Geisler (1988), that Modern
French has a strong tendency towards the SVO-pattern – this claim is in fact
inconsistent with the development of the SVIO-pattern.
12. In (21) the verb is abbreviated as b. in my translations.
13. See the explanation of the nonsense words of the poem by Humpty Dumpty in
chapter 6 of Through the Looking Glass.
14. I want to thank the editors for providing the possibility of commenting on Dirk
Noël’s paper. Unfortunately, space does not allow me to go into details, but I
sincerely hope that this interesting discussion on valency, constructions and
grammaticalisation will continue.
15. See Heltoft, Sørensen, and Schøsler (2005).
16. Regrammaticalisation or, more recently, regrammation, is the term proposed
by Henning Andersen (forthcoming a) to cover Lehmann’s “from grammatical
to more grammatical”.

References

Andersen, Henning
forthc. a Grammation, regrammation, and degrammation − tense loss in Rus-
sian.
forthc. b Grammaticalization in a speaker-oriented theory of change.
Blanche-Benveniste, Claire, José Deulofeu, Karel van der Eynde, and Jean
Stefanini
1987 Pronom et Syntaxe. L’Approche Pronominale et son Application au
Français. 2d ed. Paris: Selaf.
Bybee, Joan L.
1985 Morphology: A Study of the Relation between Meaning and Form.
Amsterdam: Benjamins.
Carroll, Lewis
1986 Through the Looking Glass. In The Complete Illustrated Works of
Lewis Carroll. London: Octopus Publishing Group Ltd.
Croft, William
2001 Radical Construction Grammar. Syntactic Theory in Typological
Perspective. Oxford: Oxford University Press.
Durst-Andersen, Per, and Michael Herslund
1996 The syntax of Danish verbs: Lexical and syntactic transitivity. In
Content, Expression and Structure. Studies in Danish Functional
Grammar, Elisabeth Engberg-Pedersen, Michael Fortescue, Peter
Harder, Lars Heltoft, and Lisbeth Falster Jakobsen (eds.), 65–102.
Amsterdam: Benjamins.
Engberg-Pedersen, Elisabeth, Michael Fortescue, Peter Harder, Lars Heltoft, and
Lisbeth Falster Jakobsen (eds.)
1996 Content, Expression and Structure. Studies in Danish Functional
Grammar. Amsterdam: Benjamins.
64 Lene Schøsler

François, Jacques, and Günter Broschart

1994 La mise en ordre des relations actancielles: Les conditions d’accès
des rôles sémantiques aux fonctions de sujet et d’objet. In Les Rela-
tions Actancielles. Sémantique, Syntaxe, Morphologie, Jacques Fran-
çois, and Gisa Rauh (eds.), 7–44. (Langages 113.) Paris: Larousse.
Gaatone, David
1998 Le Passif en Français. Paris/Bruxelles: Duculot.
Geisler, Hans
1988 Das Verhältnis von semantischer und syntaktischer Transitivität im
Französischen. Romanistisches Jahrbuch 39: 22–35.
Goldberg, Adele
1995 A Construction Grammar Approach to Argument Structure. Chicago:
The University of Chicago Press.
Goyens, Michèle
2001 L’origine des verbes français a construction dative. In La Valence,
Perspectives Romanes et Diachroniques, Lene Schøsler (ed.), 43–58.
(ZFSL Beihefte 30.) Stuttgart: Franz Steiner Verlag.
Gross, Gaston
1989 Les Constructions Converses du Français. Genève: Droz.
Gross, Maurice
1975 Méthodes en Syntaxe. Régime des Constructions Complétives. Paris:
Hermann.
Harder, Peter
1996 Linguistic structure in a functional grammar. In Content, Expression
and Structure. Studies in Danish Functional Grammar, Elisabeth
Engberg-Pedersen, Michael Fortescue, Peter Harder, Lars Heltoft,
and Lisbeth Falster Jakobsen (eds.), 423–452. Amsterdam: Benja-
mins.
Heltoft, Lars
1996 Paradigmatic structure, word order and grammaticalisation. In Con-
tent, Expression and Structure. Studies in Danish Functional Gram-
mar, Elisabeth Engberg-Pedersen, Michael Fortescue, Peter Harder,
Lars Heltoft, and Lisbeth Falster Jakobsen (eds.), 469–494. Amster-
dam: Benjamins.
Heltoft, Lars, Jens Nørgård Sørensen, and Lene Schøsler (eds.)
2005 Grammatikalisering og Struktur [Grammaticalisation and structure].
København: Museum Tusculanum Press.
Herslund, Michael
1997 Syntaktiske alternationer og funktionelle kategorier [Syntactic alter-
nations and functional categories]. In Ny Forskning i Grammatik
[New research on grammar], Lisbeth Falster Jakobsen, and Gunver
Skytte (eds.), 49–70. (Fællespublikation 4.) Odense: Odense Univer-
sitetsforlag.
Hopper, Paul J., and Sandra A. Thompson
1980 Transitivity in grammar and discourse. Language 56: 251–299.
 The status of valency patterns 65

Koch, Peter
2001 As you like it. Les métataxes actantielles entre expérient et
phénomène. In La Valence, Perspectives Romanes et Diachroniques,
Lene Schøsler (ed.), 59–81. (ZFSL Beihefte 30.) Stuttgart: Franz
Steiner Verlag.
Krefeld, Thomas
1998 Transitivität aus rollensemantischer Sicht. Eine Fallstudie am Bei-
spiel französischer und italienischer Wahrnehmungsverben. In Tran-
sitivität und Diathese in romanischen Sprachen, Hans Geisler, and
Daniel Jacob (eds.), 155–173. Tübingen: Max Niemeyer Verlag.
Koch, Peter, and Thomas Krefeld (eds.)
1991 Connexiones Romanicae. Dependenz und Valenz in romanischen
Sprachen. (Linguistische Arbeiten 268.) Tübingen: Max Niemeyer
Verlag.
Lehmann, Christian
1985 Grammaticalization: Synchronic variation and diachronic change.
Lingua e Stile 20: 303–318.
Levin, Beth
2003 Objecthood and object alternations,
http://www-csli.stanford.edu/~beth/pubs.html
Noël, Dirk
2007 Verb valency patterns, constructions and grammaticalization. This
volume.
Oesterreicher, Wulf
1991a Verbvalenz und Informationsstruktur. In Connexiones Romanicae.
Dependenz und Valenz in romanischen Sprachen, Peter Koch, and
Thomas Krefeld (eds.), 349–384. (Linguistische Arbeiten 268.) Tü-
bingen: Max Niemeyer Verlag.
1991b Gemeinromanische Tendenzen: Morphosyntax. In Lexikon der Ro-
manistischen Linguistik Band II, Gunter Holtus, Michael Metzeltin,
and Christian Schmitt (eds.), Tübingen: Max Niemeyer Verlag.
Schøsler, Lene
1999a La valence verbale et l’identification des membres valentiels. In
Autour de Jacques Monfrin. Néologie et Création Verbale, Giuseppe
Di Stefano, and Rose M. Bidler (eds.), 527–554. Montréal: CERES.
1999b Réflexions sur optionalité des compléments d’objet direct, en latin,
en ancien français, en moyen français et en français moderne. Etudes
Romanes 44: 9–28.
2000 Le statut de la forme zéro du complément d’objet direct en français
moderne. Etudes Romanes 47: 105–129.
2001 La valence verbale dans une perspective diachronique: Quelques
problèmes méthodologiques. In La Valence, Perspectives Romanes
et Diachroniques, Lene Schøsler (ed.), 98–112. (ZFSL 30.) Stuttgart:
Franz Steiner Verlag.
66 Lene Schøsler

2003 Le rôle de la valence pour une classification sémantique des verbes.

In La Cognition dans le Temps. Etudes Cognitives dans le Champ
Historique des Langues et des Textes, Peter Blumenthal, and Jean
Tyvaert (eds.), 145–159. Tübingen: Max Niemeyer Verlag.
Schøsler, Lene (ed.)
2001 La Valence, Perspectives Romanes et Diachroniques, Stuttgart:
Steiner. (ZFSL 30.)
Schøsler, Lene, and Karen Van Durme
1996 The Odense Valency Dictionary. (Odense Working Papers in Lan-
guage and Communication 13.) Odense University.
Schøsler, Lene, and Sabine Kirchmeier-Andersen
1997 Studies in Valency II: The Pronominal Approach Applied to Danish.
(RASK Supplement 5), Odense: Odense University Press.
1998 The role of the object in a syntactico-semantic classification of Danish
verbs. Leuvense bijdragen: Tijdschrift voor Germaanse Filologie 86
(4): 391–412.
Selig, Maria
1991 Inhaltskonturen des ‘Dativs’. Zur Ablösung des lateinischen Dativs
durch ad und zur differentiellen Objektmarkierung. In Connexiones
Romanicae. Dependenz und Valenz in romanischen Sprachen, Peter
Koch, and Thomas Krefeld (eds.), 187–211. (Linguistische Arbeiten
268.) Tübingen: Max Niemeyer Verlag.
Skafte Jensen, Eva
2002 Historisk lingvistik [Historical linguistics]. Nydanske Sprogstudier
[New Danish studies in language] 31: 7–32.
Verb valency patterns, constructions and
grammaticalization
Dirk Noël

1. Introduction1

Adele Goldberg’s seminal work on argument structure constructions

(Goldberg 1995) has brought to the fore that verb valency patterns have the
potential at least of being symbolic units: valency patterns might not
merely be formal patterns, but pairings of form and meaning. The trivalent
pattern in which a verb is combined with a subject and two nominal com-
plements, for instance, is argued by Goldberg (1995: 151) “to be associated
with a highly specific semantic structure, that of successful transfer be-
tween a volitional agent and a willing recipient”, with a number of “sys-
tematic metaphors” “licens[ing] extensions from the basic sense” (the basic
sense of the construction is illustrated by the sentences in (1), selected from
chapters 1 and 5 in Goldberg 1995; the sentences in (2), selected from
chapter 6, exemplify some of the extensions).

(1) a. Joe gave the earthquake relief fund $5.

b. I brought Pat a glass of water.
c. She threw him a cannonball.
(2) a. She gave me the flu.
b. She gave Jo her thoughts on the subject.
c. She gave him a wink.

More recently, Goldberg has engaged in corpus and experimental psycho-

linguistic research to tackle the question of how argument structure sche-
mas like this ditransitive construction come to be part of the language
user’s linguistic knowledge (Goldberg, Casenhiser, and Sethuraman 2004).
The frequency in the language infants are confronted with of verbs whose
meaning the meaning of the construction can be reduced to appears to play
a crucial role in this process. Native speakers of English, for instance, first,
and most frequently, encounter the ditransive pattern together with the verb
give. A phylogenetic explanation for how such constructions come to be
part and parcel of the repertory of the means of expression available in a
language may be hypothesized to run along similar lines. The ditransitive
68 Dirk Noël

construction could ultimately owe its place as a schematic construction in

the grammar of English to the early and ongoing omnipresence in English
of the verb give. The diachronic investigation of argument structure con-
structions is still largely a virgin territory, however. So far, there have only
been a few studies of partially substantive constructions (i.e. constructions
whose lexical fillers are partly specified; cf. Croft and Cruse 2004: 248);
Israel (1996) and Verhagen (2002), for instance, are enquiries into the his-
tory of Goldberg’s “way-construction” and its Dutch cognate, and Kemmer
and Hilpert (2005) investigate the development of the English make-
causative. The only work I am aware of that could qualify as a tentative
exploration of the diachrony of fully schematic argument structure con-
structions is Schøsler’s (2003, this volume, and forthcoming) treatment of
valency patterns in French.
If we accept that schematic argument structure constructions form part
of the grammar of a language, the question arises of whether their entry
into the grammar is a case of grammaticalization in the technical sense of
the term, i.e. in the sense in which the term is used in grammaticalization
theory, the “research framework for studying the relationships between
lexical, constructional, and grammatical material in language, diachroni-
cally and synchronically” (Hopper and Traugott 2003: 18) that became
established during the last decade of the previous century.2 Nutshell defini-
tions of the phenomenon like the one offered in Hopper and Traugott’s
textbook appear at first sight not to disallow that the entrenchment of a
construction in a language is characterized in terms of grammaticalization:
grammaticalization is “the change whereby lexical items and constructions
come in certain linguistic contexts to serve grammatical functions and, once
grammaticalized, continue to develop new grammatical functions” (Hopper
and Traugott 2003: 18, my emphasis). Schøsler (2003, this volume, and
forthcoming) is the only researcher to date, however, at least to my knowl-
edge, who has applied the term to a change involving a schematic argument
structure construction.3 The purpose of this contribution is to investigate
whether this is warranted. My conclusion will be that it might not be a fe-
licitous option to do so, one reason being that grammaticalization is most
often taken to be a change that affects fairly substantive constructions,
rather than fully schematic ones. I will start by clarifying this distinction
and by exploring the extent to which, and the sense in which, researchers
working within the confines of the grammaticalization theoretical frame-
work are considering constructions.
 Verb valency patterns, constructions and grammaticalization 69

2. Constructions and grammaticalization

Towards the end of the previous century, more or less simultaneously with,
but nevertheless independently of, the surge in interest in grammaticaliza-
tion and grammaticalization theory, a new theoretical approach to language
emerged of which the already mentioned Goldberg (1995) is a major expo-
nent: construction grammar.4 In principle at least, construction grammar is
an all-embracing perspective on language, whereas grammaticalization
theory “merely” covers a particular kind of language change. Both para-
digms can therefore be said to have a different agenda without there being a
conflict of interests between them.
At the basis of construction grammar is the hypothesis that all linguistic
knowledge is uniformly represented in the speaker’s mind as pairings of
form and meaning (Croft and Cruse 2004: 255) or form and function
(Goldberg 2003), in other words as constructions. Constructions vary along
two dimensions: they can either be atomic (morphemes, words) or complex
(phrases, or constructions in the pre-theoretical sense, idioms, valency pat-
terns, …), and more or less abstract. The latter dimension is where the
terms substantive and schematic come in. As pointed out by Fillmore, Kay
and O’Connor (1988: 505, n. 3) and illustrated in Croft and Cruse (2004:
248), there is a cline from maximally substantive to maximally schematic.
Fully substantive constructions are idioms like It takes one to know one, in
which there are not only no lexically open elements, but all grammatical
inflectional categories are specified as well. An example of a slightly more
schematic construction is the idiom kick the bucket, which is not lexically
open (apart from the subject slot) but which has inflectional flexibility
(Jake kicked the bucket / Jake’s gonna kick the bucket). Somewhat more
lexically open is the idiom give NP the lowdown (‘tell NP the news’),
which has two open argument slots as well as inflectional flexibility, as in I
/ He gave / will give him / Janet the lowdown. In the let alone-construction
all content words are lexically open, the only substantive element being the
let alone-connective, as in She gave me more candy than I could carry, let
alone eat and Only a linguist would buy that book, let alone read it. An
example of a maximally schematic construction, in which all elements are
lexically open is the resultative construction, illustrated in He wanted her to
kiss him unconscious and I had brushed my hair very smooth.
Since constructionists are intent on making the point that the meaning of
utterances cannot be reduced to the meaning of the words they contain, but
that structure adds meaning as well, the centre of attention of construction-
ist research to date has been on fully or partially schematic rather than fully
substantive constructions. If partially substantive constructions are de-
70 Dirk Noël

scribed, the focus is not on the meaning of the substantive elements, but on
the meaning of the construction as a whole. Fillmore, Kay and O’Connor’s
(1988) study of the let alone-construction, for instance, does not discuss the
meaning of the verb let, either separately or in combination with alone. In
other words, the approach is holistic rather than componential, which to a
certain extent is also explained by the fact that constructionist descriptive
work normally does not venture off the synchronic plane.
Since grammaticalization theory and construction grammar are not mu-
tually exclusive frameworks, there is nothing to stop a student of gram-
maticalization from subscribing to a construction grammatical view of lan-
guage. However, though the construction word turns up regularly in work
on grammaticalization, more often than not it is used in a non-technical
way5, usually to refer to collocations that turn into fixed units, like sort of
and kind of, discussed (inter alia) in Tabor (1994),6 and instead of (from in
stede of), indeed (from in dede), anyway (from any way), discussed in
Traugott (2003). This is also how the word is used in the definition of
grammaticalization presented earlier. A notable exception is Croft (2001), a
volume that aims to contribute to both constructionist theorizing and
grammaticalization theory. Similarly, Joan Bybee, a leading grammaticali-
zation theorist, has written in a multidisciplinary publication that “grammar
consists of a large number of rather specific constructions which act as
processing units” (Bybee 1998: 272), thereby implicitly subscribing to one
of the tenets of constructionist approaches. On the whole, however, gram-
maticalization theoretical publications refer to a pre-theoretical construction
concept, first and foremost in order to include multi-word units as possible
sources and outcomes of grammaticalization, but also to drive home the
message that neither atomic nor complex items grammaticalize irrespective
of the contexts in which they are used. Another example of the first of these
two senses in which the term is made use of is the often quoted insight that
“[i]t 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, and Pagliuca 1994: 11). The second sense is
apparent in Traugott’s plea for a “focus on grammaticalization as centrally
concerned with the development of lexemes in context-specific construc-
tions (not merely lexemes and constructions)” (Traugott 2003: 627). It is
also evident in the following quote from Himmelmann (2004: 31):7
“Strictly speaking, it is never just the grammaticizing element that under-
goes grammaticization. Instead, it is the grammaticizing element in its syn-
tagmatic context which is grammaticized. That is, the unit to which gram-
maticization properly applies are constructions, not isolated lexical items.”
 Verb valency patterns, constructions and grammaticalization 71

This use of the construction word is what is referred to with “in certain
linguistic contexts” in Hopper and Traugott’s (2003) definition of gram-
maticalization, and with “highly constrained morphosyntactic contexts” in
Traugott’s (2003: 645) personal definition: grammaticalization is “[t]he
process whereby lexical material in highly constrained pragmatic and mor-
phosyntactic contexts is assigned grammatical function, and once gram-
matical, is assigned increasingly grammatical, operator-like function”.8 If
“lexical material” in the second definition is taken to include phrases as
well as words, both definitions therefore at least imply a double and con-
ceptually different reference to constructions.
Whichever way the construction concept is invoked, however, whether
to include non-atomic material as precursors of grammaticalized items or to
highlight the contexts in which the grammaticalization takes place, most of
the quotes supplied so far reveal that grammaticalization theorists are nor-
mally only considering fairly substantive constructions: they are dealing
with constructions containing lexical material. Moreover, since the focus of
their attention is on the change in meaning of the lexical atoms at the centre
of the construction, their approach can be argued to be componential rather
than holistic, as opposed to the typical construction grammatical view on
things.
Not all grammaticalization theorists would agree with limiting the re-
search subject of grammaticalization theory to non-schematic construc-
tions, though. Haspelmath’s (2004: 26) “current definition” of grammati-
calization also contains the construction word but there is nothing in the
definition to restrict its applicability to at least partially substantive con-
structions: “[a] grammaticalization is a diachronic change by which the
parts of a constructional schema come to have stronger internal dependen-
cies.” An attached footnote in fact makes clear that the definition is in-
tended to include fully schematic constructions: “[t]hus, word-order change
consisting of a change from freer to more fixed word order falls under
grammaticalization as well …, not just changes involving free words be-
coming dependent elements …” (Haspelmath 2004: 38). Yet it is not uni-
versally accepted among grammaticalization theorists that the fixation of
word order indeed constitutes a case of grammaticalization. For Himmel-
mann (2004: 33–34), for instance, changes involving fully schematic con-
structions do not represent examples of the phenomenon:

… grammaticization applies only to the context expansion of constructions

which include at least one grammaticizing element (the article in art-noun
constructions, the preposition in pps, etc.). Context expansion may also oc-
cur with other types of constructions, for example a certain word order pat-
72 Dirk Noël

tern, a compounding pattern or a reduplication pattern. These are not con-

sidered instances of grammaticization here.

The footnote he adds is relevant as well:

We may note in passing that there is a tendency in the literature to use

grammaticization as a cover term for all kinds of grammatical change, in-
cluding simple reanalyses, analogical levelings and contact-induced
changes. In this way, the concept grammaticization looses [sic] all theoreti-
cal significance and becomes simply a synonym for grammatical change.
(Himmelmann 2004: 39)

Christian Lehmann, though not himself adverse to including word order

fixation in grammaticalization (e.g. see Lehmann 2002), has in a similar
vein dissociated himself from definitions of grammaticalization like
“grammaticalization is the genesis of grammar/grammatical structure/gram-
matical items” (Lehmann 2005:155), maintaining that

it is unwise to elevate grammaticalization to the status of ‘creation of

grammar’ per se. This necessarily renders the concept wide and heterogene-
ous, with the consequence that it becomes less apt to generate falsifiable
empirical generalizations and to be integrated into an articulated theory of
language change and language activity. (Lehmann 2005:155)

Hopper and Traugott (2003: 24, 60), for their part, oppose the inclusion of
word order. They do discuss Givón’s (1979) work on clause combining and
clause fusion, which also involves schematic constructions, but at the same
time distance themselves from it by saying it can only be included “[i]f
grammaticalization is defined broadly so as to encompass the motivations
for and development of grammatical structures in general” (Hopper and
Traugott 2003: 176).
The recent plea for a construction-based approach to grammaticalization
to replace the morphology-based approach (see Wiemer 2004; and Wiemer
and Bisang 2004) should therefore not be taken to imply that all grammati-
cal constructions are the result of a narrowly defined grammaticalization,
which crucially involves a bundle of changes happening to the substantive
element(s) of non-schematic, or at least not fully schematic, constructions.9
For this reason the term grammaticalization might not apply well to the
establishment of verb valency patterns as argument structure constructions.
In the next section I will examine whether indeed such an evolution can
justifiably be characterized as a case of grammaticalization.
 Verb valency patterns, constructions and grammaticalization 73

3. Verb valency patterns and grammaticalization

As a spin-off of cognitive linguistics, construction grammatical studies of

schematic constructions have so far only considered their ontogenesis (how
do they come to be part of the language user’s knowledge?), not their phy-
logenesis (how do they enter the language?). To my knowledge, no paid-up
member of the construction grammatical paradigm has so far used the term
grammaticalization in connection with schematic constructions (but see
Croft 2001 on substantive constructions). The trigger for the present contri-
bution, however, was work by Lene Schøsler, who has consistently applied
the term in a series of diachronic studies of valency patterns (Schøsler
2003, this volume, and forthcoming).
Though not a proponent of a fully-fledged constructionist approach her-
self, Schøsler (forthcoming) also refers to constructions, but restricts the
term to “specialized” verb valency patterns, which are patterns that have
become “linked to special content”. In other words, Schøsler distinguishes
between valency patterns she terms “default patterns”, which do not ex-
press content, and valency patterns she calls “constructions”, which do
carry content. The latter are claimed to be the result of grammaticalization.
One of her illustrations is based on data from Goyens (2001) on the origin
of French verbs used in the “dative construction”, i.e. verbs used in a diva-
lent pattern taking an indirect object (in addition to a subject). These verbs
“share the particular feature that they all express – more or less clearly – a
psychological relation between on the one hand a human being who is the
experiencer …, and who has pleasant or unpleasant feelings, and on the
other hand another human being or an object being the cause of these feel-
ings …”, the experiencer being expressed by the indirect object and the
cause by the subject (Schøsler this volume: 55). This situation – illustrated
in table 1 with reference to the verbs obéir, ressembler, mentir – is a fairly
recent development, however, because up until the 17th century the experi-
encer did not need to be expressed by an indirect object but could also be
expressed by a direct object.
74 Dirk Noël

Table 1. The evolution of the valency patterning of the etymons of the French
verbs obéir, ressembler and mentir (adapted from Goyens 2001: 56)10

Latin Anc. frç. Moy. frç. XVIe s. XVIIe s. Frç. mod.

obéir --- COD COD COD --- ---

datif COI COI --- COI COI
ressembler accus. COD COD COD COD ---
datif COI COI COI COI COI
mentir accus COD COD COD --- ---
--- COI --- --- COI COI

In other words, there has been a change in the valency patterning of the
verbs expressing this “psychological relation”, leading to a situation where
they need to be used with an indirect object. This valency pattern is there-
fore said to have become “specialized” – it is exclusively used for the ex-
pression of this relation – and such a specialization is interpreted to amount
to becoming “more grammatical”, i.e. as a case of grammaticalization,
more specifically as a case of “secondary grammaticalization” (“the devel-
opment of an already grammatical form into a yet more grammatical one”,
Traugott 2004: 143). This means, in effect, that Schøsler (forthcoming)
considers the crystallization of a construction (i.e. the establishment of a
connection between a morphosyntactic configuration and a meaning) as
being subsumed under the heading grammaticalization.
Though one could find fault with Schøsler’s (2003, this volume) charac-
terization of this particular construction (is there really an experiencer in
the case of ressembler? Does the level of abstractness needed to accommo-
date all verbs that can enter the pattern not preclude its psychological real-
ity?), this is not my intention here. The question I am interested in is the
more general one of whether the establishment of a symbolic link between
a particular syntactic arrangement and a meaning can indeed be argued to
amount to grammaticalization.
A first sub-question that will need to be answered positively to allow
this is whether the meaning of argument structure constructions can justi-
fiably be said to be a grammatical meaning. In the words of Hopper and
Traugott (2003: 24): “how far we shall be prepared to extend the notion of
‘grammaticalization’ will be determined by the limits of our understanding
of what it means for a construction to be ‘grammatical’ or have a gram-
matical function.” Given that the meaning of Goldberg’s ditransitive con-
struction can be reduced to the meaning of the verb give, and given that
Schøsler’s description of her “experiencer” construction makes reference to
feelings incited in somebody by another person or by something, it seems
 Verb valency patterns, constructions and grammaticalization 75

hardly defensible to talk of grammatical notions here, even if we grant that

there is no clear boundary between what is lexical and what is grammatical.
The notions referred to here are of an unquestionable propositional or idea-
tional nature, whereas grammatical meanings are prototypically non-
propositional or interpersonal.11 Valency patterns are part of grammar to
the extent that they assist in organizing the building blocks of a language
into meaningful strings, but it does not follow that the content they might
convey is of a grammatical nature. For Schøsler, however, the very fact that
structure acquires meaning that is typically associated with the lexicon
appears reason enough to talk of grammaticalization when she concludes:

If we include valency patterns as part of the grammar and accept the

changes presented here as results of inference from lexicon to grammar and
as cases of paradigmatisation of valency patterns, then it is legitimate to re-
fer to these changes as “grammaticalisations”, or rather, following Henning
Andersen’s terminology, as cases of “regramma(ticalisa)tion”. (Schøsler
this volume: 62)

In Schøsler (forthcoming) the author goes further and claims these con-
structions to have an ulterior function: they enable speakers and listeners to
identify arguments. Latin had no constructions, only default patterns, and
arguments “were first and foremost identified by means of the lexicon, i.e.
selectional restrictions on predicates and arguments, and by means of the
nominal morphology”. In modern Romance languages, however, “we find a
large variety of grammaticalized devices used to identify the arguments”,
among which word order, use of prepositions, and specialized valency pat-
terns. Leaving aside that this is a psychological claim in need of psycholin-
guistic corroboration, and assuming that argument structure constructions
actually contribute to argument identification, it still does not follow
though that the naissance of such constructions need be the consequence of
a grammaticalization change. Were we to conclude this, it would put a
whole new teleological interpretation on grammaticalization.
A second question we need to address is the extent to which the coming
into being of argument structure constructions is a change that meets the
criteria for grammaticalization put forward in the grammaticalization theo-
retical literature. Here we run into the problem that, since grammaticaliza-
tion theorists have mainly been interested in non-schematic constructions,
these criteria only work well for constructions containing substantive ele-
ments (cf. Fischer 2005). Arguments in favour of their applicability to
schematic constructions might therefore not yield falsifiable statements.
Heine (2003: 579) provides a conveniently concise list of the “mecha-
nisms” involved in grammaticalization (or the “micro-changes” involved in
76 Dirk Noël

the “macro-change” grammaticalization, in the terminology of Andersen,

forthcoming) about which there is a fairly general consensus:
(i) desemanticization (or “bleaching,” semantic reduction): loss in
meaning content;
(ii) extension (or context generalization): use in new contexts;
(iii) decategorialization: loss in morphosyntactic properties characteris-
tic of the source forms, including the loss of independent word
status (cliticization, affixation);
(iv) erosion (or “phonetic reduction”), that is, loss in phonetic sub-
stance.
The latter two mechanisms, especially, will not work if there is no substan-
tive grammaticalizing element: only words and phrases can change catego-
ries (e.g. from lexical verb to auxiliary, or from main clause verb phrase to
adverbial phrase) and lose substance. The first mechanism, on the other
hand, has been brought to bear on the fixation of word order to extend the
domain of grammaticalization to it ever since the term grammaticalization
first entered the linguistic literature (i.e. in Meillet 1912/1958; see Hopper
and Traugott 2003: 23). When word order is free it is used to convey prag-
matic meaning, which is lost when word order is decided by syntax. Sche-
matic constructions are not immune to meaning loss, therefore. But the
converse has happened if, as Schøsler (forthcoming) suggests, argument
structure constructions (or “specialized verb valency patterns”) succeed
“default patterns”. Instead of an evolution from more to less meaning, or
from “expressive” to grammatical meaning, what we have here is a devel-
opment from absence of meaning to presence of meaning, or possibly from
grammatical meaning to referential meaning. Instead of the loss in referen-
tiality usually associated with grammaticalization we are seeing a gain in
referentiality. Rather than a movement away from the ideational plane there
is a progression towards it.
The second mechanism, extension, interlocks with the first: the decrease
in semantic specificity concurs with context expansion. The dwindling of
the expressivity of a particular word order pattern coincides with its gener-
alization, for instance. In the case of the specialization of valency patterns,
on the other hand, the schematic construction’s expressivity swells rather
than dwindles, so that one might expect context reduction instead of expan-
sion. To better examine whether this could indeed be the case, it may be
useful to consider the three kinds of context expansion that were teased
apart by Himmelmann (2004: 32–33). The first of these is host-class expan-
sion, an expansion of the class of elements a grammaticalizing element is in
construction with. Himmelmann’s example: “when demonstratives are
grammaticized to articles they may start to co-occur regularly with proper
 Verb valency patterns, constructions and grammaticalization 77

names or nouns designating unique entities (such as sun, sky, queen, etc.),
i.e. nouns they typically did not co-occur with before” (Himmelmann 2004:
32). Schøsler’s example of the French S-V-IO construction seems to illus-
trate an evolution in the opposite direction, however, when she states that
“verbs that did not match the content of this pattern have … been ejected
from it” (Schøsler this volume: 57). This amounts to host-class reduction
rather than expansion. One of Schøsler’s arguments in favour of the gram-
maticalization of specialized valency patterns therefore in effect detracts
from it.
Himmelmann’s second kind of context expansion, syntactic context ex-
pansion, a change in the larger syntactic context in which the construction
is used (e.g. articles occurring in adpositional expressions in addition to the
core argument positions they typically occur in first), might not be relevant
to argument structure constructions since the context level beyond the one
that is defined by these constructions falls outside the scope of syntax. The
third kind, semantic-pragmatic context expansion, is illustrated in
Himmelmann’s “article” example by the fact that adnominal demonstra-
tives occur only in expressions which involve deictic, anaphoric or recogni-
tional reference, whereas articles also have “larger situation uses” (the
queen, the pub) and “associative anaphoric uses” (a wedding – the bride, a
house – the front door; Himmelmann 2004: 33). A semantic widening of
some kind could at first glance also appear germane to argument structure
constructions when considering Goldberg’s extensions of a construction’s
central sense. Taking up the example of the ditransitive construction again,
the following sentences (the examples in [12] in chapter 6 of Goldberg
1995) illustrate an extension of the construction’s central sense of a suc-
cessful transfer between a volitional agent and a willing recipient in that
they do not involve a volitional agent.

(3) a. The medicine brought him relief.

b. The rain bought us some time.
c. She got me a ticket while distracting me while I was driving.
d. She gave me the flu.
e. The music lent the party a festive air.
f. The missed ball handed him the victory on a silver platter.

Such metaphorical extensions do not move the construction off the idea-
tional plane, however, and cannot therefore be argued to be constitutive of
a grammaticalization change.
The mechanisms that are generally taken to define grammaticalization
do not, therefore, seem to be at work in the “specialization” of verb valency
78 Dirk Noël

patterns, or the creation of argument structure constructions. Some of these

mechanisms can only affect substantive constructions, and those that sche-
matic constructions could be subject to do not seem to work in the direction
typical of grammaticalization.

4. Conclusion

Schøsler (2003, this volume, and forthcoming) has pointed the way to a
whole new area of research, the diachronic study of schematic argument
structure constructions, i.e. the study of how and when verb valency pat-
terns crystallize into pairings of form and meaning. If, as Wiemer and
Bisang (2004: 4) have advocated, grammaticalization is “extended to all the
processes involved in the diachronic change and in the emergence of
[grammatical] systems” (systems “of more or less stable, regular and pro-
ductive form-function mappings”), i.e. if it is taken as “a general perspec-
tive from which to analyse changes in the expression formats of grammati-
cal structure or [in] the distribution of certain morphological or syntactic
units in the languages of the world”, studies in this area will undoubtedly
find their place within grammaticalization theory. The nature of their re-
search subject will make them fall outside the bounds of a more narrowly
defined grammaticalization, however. Though constructions have, at the
least, a double relevance for grammaticalization theory – as grammaticaliz-
ing units and as the structural contexts in which grammaticalization takes
place – the core business of the field to date has been (at least partially)
substantive constructions.12 It is on their basis that the principles involved
in grammaticalization have been defined. Being schematic constructions,
argument structure constructions are less susceptible to them. But argument
structure constructions also differ from those schematic constructions that
have so far been considered by grammaticalization theorists, in that their
meaning is propositional rather than grammatical. The semantic change
attending their development is a movement towards greater referentiality,
rather than the converse. If the diachronic study of argument structure con-
structions does find a place in grammaticalization theory, such differences
will have to be integrated in future taxonomies of the different natured
changes subsumed under grammaticalization.
 Verb valency patterns, constructions and grammaticalization 79

Notes

1. This paper was written during a sponsorship by the Research Fund of the Uni-
versity of Leuven. I am grateful to the Functional Linguistics Leuven research
unit, and especially Kristin Davidse, for their hospitality. I must also thank
Lene Schøsler for passing on two of her manuscripts (Schøsler this volume
and forthcoming). The organizers of the “Valency – Valenz: Theoretical, De-
scriptive and Cognitive Issues” symposium must be thanked for their kind in-
vitation to contribute a paper, and for prodding me into writing up my contri-
bution. Lieselotte Brems and Timothy Colleman are owed words of gratitude
for their comments on an earlier version.
2. Hopper and Traugott (2003) do not themselves refer to grammaticalization
theory but use the term grammaticalization to refer to the phenomena it covers
as well as the study of these phenomena, analogous to such linguistic terms as
syntax, morphology and semantics. I will employ grammaticalization theory
for stylistic reasons, not least because it allows reference to grammaticaliza-
tion theorists, in the absence of a coinage like grammaticalizationists. Gram-
maticists is already in use but has a much wider reference.
3. In a series of (at the time of writing) unpublished or yet to be published con-
ference papers Suzanne Kemmer has talked about constructional grammati-
calization, but the term seems so far only to have been applied to constructions
containing a substantive element (e.g., see Kemmer and Hilpert 2005).
4. Grammaticalization and constructionism alike are not uniform frameworks. I
am using construction grammar as a cover term for all constructionist ap-
proaches. Croft and Cruse (2004: 257) distinguish between four variants: Con-
struction Grammar (in capital letters; e.g. Kay and Fillmore 1999), construc-
tion grammar (without capitals; e.g. Lakoff 1987 and Goldberg 1995), Cogni-
tive Grammar (Langacker 1987, 1991) and Radical Construction Grammar
(Croft 2001).
5. Leaving aside certain notational conventions of individual constructionist
approaches, very little is “technical” in construction grammar, not least the
definition of what constitutes a construction, but I qualify uses of the construc-
tion word as “pre-theoretical” or “non-technical” when those who use it do not
pledge their adherence to a constructionist stance.
6. Denison (2002) tries out a construction grammatical analysis of these con-
structions, and thus constitutes an exception to the generalization formulated
here.
7. Some grammaticalization theorists prefer the term grammaticization to gram-
maticalization.
8. The reference to a process rather than a change shows that this definition
actually predates the one offered in Hopper and Traugott (2003: xv). Elizabeth
Traugott has confirmed (personal communication) that the “Constructions in
grammaticalization” article was written in 1995 and revised in 1998, to finally
come out in 2003.
80 Dirk Noël

9. In Fischer’s (2005) terminology, these changes require tokens as well as types,

whereas the “grammaticalization” of clause types and the fixation of word or-
der only involve types. (This use of the type/token distinction roughly com-
pares with the way the two concepts are distinguished in “usage-based” cogni-
tive linguistics, tokens being more specific instantiations of more general
types (cf. Bybee 1985).) In Andersen’s (forthcoming) classification of types of
“macro-changes” (which involve a chain of changes) on the basis of “the ob-
server’s wider or narrower focus” grammaticalization is categorized as a
change that is observed as a result of a “single element view” (as opposed to a
“whole-language view” and a “subsystem view”) focusing on expressions
(rather than content).
10. Goyens’ original terminology was replaced by Schøsler’s (2003). COD and
COI stand for direct object and indirect object, respectively. For another adap-
tation, see Schøsler (this volume).
11. Ideational and interpersonal are Hallidayan terms (e.g. see Halliday 1970).
12. A third, paradigmatic, sense in which constructions are relevant for grammati-
calization does involve schematic constructions, when these act as triggers for
grammaticalization (cf. Bisang 1998a, b; Hoffmann 2004; Fischer 2005; Noël
2005).

References

Andersen, Henning
forthc. Grammaticalization in a speaker-oriented theory of change. In Gram-
matical Change and Linguistic Theory: The Rosendal Papers, Þórhal-
lur Eyþórsson (ed.). Amsterdam: Benjamins.
Bisang, Walter
1998a Grammaticalization and language contact, constructions and posi-
tions. In The Limits of Grammaticalization, Anna Giacalone Ramat,
and Paul J. Hopper (eds.), 13–58. Amsterdam: Benjamins.
1998b Verb serialization and attractor positions: Constructions and their
potential impact on language change and language contact. In Typol-
ogy of Verbal Categories, Leonid Kulikov, and Heinz Vater (eds.),
254–271. Tübingen: Niemeyer.
Bisang, Walter, Nikolaus P. Himmelmann, and Björn Wiemer (eds.)
2004 What Makes Grammaticalization? A Look from its Fringes and its
Components. Berlin: Mouton de Gruyter.
Bybee, Joan
1985 Morphology: A Study of the Relation between Meaning and Form.
Amsterdam: Benjamins.
1998 A functionalist approach to grammar and its evolution. Evolution of
Communication 2 (2): 249–278.
 Verb valency patterns, constructions and grammaticalization 81

Bybee, Joan, Revere Perkins, and William Pagliuca

1994 The Evolution of Grammar: Tense, Aspect, and Modality in the Lan-
guages of the World. Chicago: The University of Chicago Press.
Croft, William
2001 Radical Construction Grammar: Syntactic Theory in Typological
Perspective. Oxford: Oxford University Press.
Croft, William, and D. Alan Cruse
2004 Cognitive Linguistics. Cambridge: Cambridge University Press.
Denison, David
2002 History of the sort of construction family. Paper presented at the 2nd
International Conference on Construction Grammar, Helsinki.
Eyþórsson, Þórhallur (ed.)
forthc. Grammatical Change and Linguistic Theory: The Rosendal Papers.
Amsterdam: Benjamins.
Fillmore, Charles John
2006 Construction Grammar. Chicago: University of Chicago Press.
Fillmore, Charles John, Paul Kay, and Mary Kay O’Connor
1988 Regularity and idiomaticity in grammatical constructions: The case of
let alone. Language 64: 501–538.
Fischer, Olga
2005 Coming to terms with grammaticalization. Plenary paper read at the
international conference From ideational to interpersonal: Perspec-
tives from grammaticalization, Leuven, 10-12 February 2005.
Givón, Talmy
1979 On Understanding Grammar. New York: Academic Press.
Goldberg, Adele E.
1995 Constructions: A Construction Grammar Approach to Argument
Structure. Chicago: University of Chicago Press.
2003 Constructions: A new theoretical approach to language. Trends in
Cognitive Sciences 7 (5): 219–224.
Goldberg, Adele E., Devin M. Casenhiser, and Nitya Sethuraman
2004 Learning argument structure generalizations. Cognitive Linguistics 15
(3): 289–316.
Goyens, Michèle
2001 L’origine des verbes français a construction dative. In La Valence,
Perspectives Romanes et Diachroniques, Lene Schøsler (ed.), 43–58.
(= ZFSL Beihefte 30.) Stuttgart: Steiner.
Halliday, Michael Alexander Kirkwood
1970 Language structure and language function. In New Horizons in Lin-
guistics, John Lyons (ed.), 140–165. Harmondsworth: Penguin.
Haspelmath, Martin
2004 On directionality in language change with particular reference to
grammaticalization. In Up and Down the Cline: The Nature of Gram-
maticalization, Olga Fischer, Muriel Norde, and Harry Perridon
(eds.), 17–44. Amsterdam: Benjamins.
82 Dirk Noël

Heine, Bernd
2003 Grammaticalization. In The Handbook of Historical Linguistics,
Brian D. Joseph, and Richard D. Janda (eds.), 575–601. Oxford:
Blackwell.
Himmelmann, Nikolaus P.
2004 Lexicalization and grammaticalization: Opposite or orthogonal? In
What makes Grammaticalization? A Look from its Fringes and its
Components, Walter Bisang, Nikolaus P. Himmelmann, and Björn
Wiemer (eds.), 21–42. Berlin: Mouton de Gruyter.
Hoffmann, Sebastian
2004 Are low-frequency complex prepositions grammaticalized? On the
limits of corpus data – and the importance of intuition. In Corpus Ap-
proaches to Grammaticalization in English, Hans Lindquist, and
Christian Mair (eds.), 171–210. Amsterdam: Benjamins.
Hopper, Paul J., and Elizabeth Closs Traugott
2003 Grammaticalization. 2d ed. Cambridge: Cambridge University Press.
Israel, Michael
1996 The way constructions grow. In Conceptual Structure, Discourse and
Language, Adele Goldberg (ed.), 217–230. Stanford: CSLI.
Kay, Paul, and Charles John Fillmore
1999 Grammatical constructions and linguistic generalizations: The What’s
X doing Y? construction. Language 75: 1–33.
Kemmer, Suzanne, and Martin Hilpert
2005 Constructional grammaticalization in the make-causative. Paper pre-
sented at the Workshop on Constructions and Language Change, held
at the XVIIth International Conference on Historical Linguistics,
Madison, Wisconsin, 31 July-5 August 2005.
Lakoff, George
1987 Women, Fire and Dangerous Things: What Categories Reveal about
the Mind. Chicago: University of Chicago Press.
Langacker, Ronald W.
1987 Foundations of Cognitive Grammar. Vol. 1: Theoretical Prerequi-
sites. Stanford: Stanford University Press.
1991 Foundations of Cognitive Grammar. Vol. 2: Descriptive Application.
Stanford: Stanford University Press.
Lehmann, Christian
2002 Thoughts on Grammaticalization. 2d, revised ed. Erfurt: Seminar für
Sprachwissenschaft der Universität.
2005 Theory and method in grammaticalization. In Grammatikalisierung,
Gabriele Diewald (ed.), 152–187. Berlin: de Gruyter [Zeitschrift für
Germanistische Linguistik, Themenheft].
Meillet, Antoine
1958 Reprint. L’évolution des formes grammaticales. In Linguistique his-
torique et linguistique générale, 130–48. Paris: Champion. Original
edition, Scientia 12, (26, 6), 1912.
 Verb valency patterns, constructions and grammaticalization 83

Noël, Dirk
2005 The productivity of a “source of information” construction: Or, where
grammaticalization theory and construction grammar meet. Paper
read at the international conference From ideational to interpersonal:
Perspectives from grammaticalization, Leuven, 10-12 February 2005.
Schøsler, Lene
2003 Le role de la valence pour une classification sémantique des verbes.
In La Cognition dans le Temps: Études Cognitives dans le Champ
Historique des Langues et des Texts, Peter Blumenthal, and Jean-
Emmanuel Tyvaert (eds.), 145–159. Tübingen: Niemeyer.
2007 The status of valency patterns. This volume.
forthc. Argument marking from Latin to modern Romance languages: An
illustration of “combined grammaticalization processes”. In Gram-
matical Change and Linguistic Theory: The Rosendal Papers, Þórhal-
lur Eyþórsson (ed.). Amsterdam: Benjamins.
Tabor, Whitney
1994 The gradual development of degree modifier sort of and kind of: A
corpus proximity model. In Papers from the 29th Regional Meeting
of the Chicago Linguistics Society, Katherine Beals, Gina Cooke,
David Kathman, Sotaro Kita, Karl-Erik McCullough, and David
Testen (eds.), 451–465. Chicago: Chicago Linguistic Society.
Traugott, Elizabeth Closs
1995 Subjectification in grammaticalisation. In Subjectivity and Subjectivi-
sation, Dieter Stein, and Susan Wright (eds.), 31–54. Cambridge:
Cambridge University Press.
2003 Constructions in grammaticalization. In The Handbook of Historical
Linguistics, Brian D. Joseph, and Richard D. Janda (eds.), 624–647.
Oxford: Blackwell.
2004 Exaptation and grammaticalization. In Linguistic Studies Based on
Corpora, Minoji Akimoto (ed.), 133–156. Tokyo: Hituzi Syobo Pub-
lishing Co.
Verhagen, Arie
2002 From parts to wholes and back again. Cognitive Linguistics 13: 403–
439.
Wiemer, Björn
2004 The evolution of passives as grammatical constructions in Northern
Slavic and Baltic languages. In What makes Grammaticalization? A
Look from its Fringes and its Components, Walter Bisang, Nikolaus
P. Himmelmann, and Björn Wiemer (eds.), 271–331. Berlin: Mouton
de Gruyter.
Wiemer, Björn, and Walter Bisang
2004 What makes grammaticalization? An appraisal of its components and
its fringes. In What makes Grammaticalization? A Look from its
Fringes and its Components, Walter Bisang, Nikolaus P. Himmel-
mann, and Björn Wiemer (eds.), 3–20. Berlin: Mouton de Gruyter.
Aspects of a diachronic valency syntax of German

Mechthild Habermann

1. Introduction

According to Ágel (2000: 269), research into the historical valency of Ger-
man has become a neglected area of linguistics due to a condescending
attitude towards language dynamics, or rather shifts. Ágel states that what
is lacking is a theory of valency dynamics and shifts.
There are only very few studies devoted to the historical valency of Ger-
man, and those that we have are strictly synchronic. To quote names:
− Albrecht Greule started in 1973 with an article on Valenz und his-
torische Grammatik [valency and historical grammar], in the first issue
of the Zeitschrift für Germanistische Linguistik.
− In 1978, Jarmo Korhonen worked on clause patterns and valency based
on texts written by Martin Luther.
− In 1982, Hugh Maxwell wrote a valency grammar for Middle High Ger-
man verbs based on the Nibelungenlied.
− In his Habilitationsschrift, Albrecht Greule (1982) worked on valency in
Old High German based on the Gospel by Otfrid von Weißenburg.
− In 1988, Vilmos Ágel wrote a verb valency dictionary based on the
Early New German text Denkwürdigkeiten der Helene Kottannerin
(1439–1440).
Only rarely are diachronic studies carried out, that is to say, studies which
investigate the valency shifts, or the problems of polyvalency in historical
language periods, as did, for example, Korhonen (1995).
Today, it is generally agreed that the field of historical valency in the
scope of a history of German syntax needs to be investigated in more depth.
This is the heritage left by the so-called Neogrammarians of the second half
of the 19th century, who directed their attention mainly to phonetics and
morphology when dealing with the historical condition and development of
language.
Most of the few studies on historical valency are – rightfully – descrip-
tive. They are orientated towards the description of the signifier in terms of
the formal means, i.e. the morphological cases, which are then cautiously
classified alongside (syntactic-)semantic functions of cases. Analysing and
86 Mechthild Habermann

describing the historical texts is the only possible way to improve our
knowledge of historical languages and to be able to build up a so-called
Ersatzkompetenz compensating for the competence of the native speaker.
This is because our sentence structures, our conventional valency sche-
mata and patterns are not of a formal kind, i.e. determined by morphologi-
cal cases. In addition to this, prototypical case roles are usually assigned to
certain formal cases, as, for instance, “agent” for the subject and “patient”
for the accusative object. Both, formal structures and the assignment of
case roles, are not necessarily the same throughout the different historical
language periods.
Werner Abraham (2005: 211−218) sees the difference in the assignment
of cases as follows: in his opinion, Old and Middle High German only have
an inherent lexical use of cases, whereas New High German developed a
structurally, i.e. syntactically governed use.
I would like to extend Werner Abraham’s theory: the valency of Old
and even Middle High German depends not only on lexical factors but also
on text linguistics. The prototypical clause patterns of New High German
are different from those of other historical language periods.
My critique of historical syntax valency is centred on a fundamental
point: possible solutions to problems arising from the analysis of Modern
German are all too often and easily transposed to earlier language periods.
As in the present-day language, the main task is to determine complements
and adjuncts.1 Borderline cases, especially adverbial phrases, are catego-
rized quantitatively, i.e. if an adverbial phrase accompanies a verb or mean-
ing of a verb relatively constantly, it is categorized as a complement. Oth-
erwise, it is considered to be an adjunct.2 However, the question of whether
quantitative valency alone can determine whether a phrase is a complement
or an adjunct is sometimes left open (in the case of optional complements
for instance).

2. Analysing historical valency of German: Difficulties and problems

When examining historical valency, some preconceptions must be revised,

or even abandoned since they obstruct the description of historical syntax.
These are:
a) Uncertainty as to the limits of a clause. Punctuation is often missing
as a clause is punctuated according to pauses in speech, hence there are no
criteria for identifying the beginning and the end of sentences. Conse-
quently, a given phrase cannot easily be classified as a complement of a
 Aspects of a diachronic valency syntax of German 87

certain verb. The classification is not always unequivocal, on the contrary,

it often remains ambiguous or vague.
b) Uncertainty as to the status of the clause. Subordinate clauses are, as
such, not unequivocally identified in every case, since the end position of
the finite verb first appears as a rule in New High German. In addition,
many conjunctions can just as well be read as hypotactic subordinators or
coordinating elements. The relative pronouns der, die, das [‘who’, ‘which’]
are also demonstrative pronouns. It thus often remains vague whether there
is a relationship of dependence or not. To sum up: the difference between
parataxis und hypotaxis is nowhere as clear and unequivocal as in Modern
German.
c) Uncertainty as to the verbal stem. For a long time, noun compounds
and, what is more important for verb valency, verb compounds were not
usually written as one word. With regard to the stem and its valency, it is
essential to determine whether Middle High German adverbs such as an,
auf, durch, or heran, hinauf, herum have the status of phrases or not and
whether, as a consequence, they could be complements or adjuncts; or
whether we are dealing with verb particles, and thus with verbs which take
a particle, as is the case with ankommen [‘arrive’], aufsteigen [‘rise, go
up’], or durchfahren [‘pass’].
d) Uncertainty as to the morphological identification of cases. Because
of early syncretism of form, especially since Middle High German, certain
cases are no longer identifiable. This means that there is a coincidence be-
tween the forms of the genitive and the dative singular of feminine nouns.
Thus, der zît [‘of the time / to the time’] can just as well be genitive sin-
gular or dative singular; der vrouwen [‘of the woman / to the woman, of the
women’] can just as well be genitive singular, or dative singular, or even
genitive plural feminine.3
It is very risqué to transpose conventional valency schemata of New
High German to historical language. In Modern German, the use of the
dative is more plausible than assuming a genitive complement, whereas for
Middle High German, this assumption would be no more than a shallow
prejudice.
e) The polyvalency of verbs. In contrast to New High German verbs,
Old and even Middle High German verbs do not have a stable, or should I
say prototypical valency.4
In New High German the meaning of the verb introduces a valency
framework which, although it is slightly modifiable, as for instance in the
case of optional complements, is quite stable for this particular meaning of
polysemic lexemes.
88 Mechthild Habermann

A wider range of structures is often recognisable in historical periods of

language, so that prototypes cannot easily be defined. Thus here, and this is
the import of the following statement, historical valency is greatly influ-
enced by co-textual and contextual factors.
f) The effect of the Indo-European meanings of the case. It seems that
the meanings of the case in Old and Middle High German are still strongly
influenced by their ancient Indo-European meanings, and more so in the
cases of genitive and dative than for the accusative as the direct object case.
In the Germanic dative, the Indo-European dative merges with the instru-
mental, locative and ablative. Until Old High German, there still are rare
occurrences of instrumental and locative, whereas during the later language
periods, the non-dative functions are generally expressed by prepositional
phrases.5
Basically, the three morphological cases genitive, dative and accusative
can appear as adverbial phrases. The disparity and diversity of meanings of
genitive and dative render the assignment of semantic roles difficult.

3. Reasons for valency shifts

Before I look more closely at the idiosyncrasy of valency in historical lan-

guage periods, I would like to put forward several arguments that help ex-
plain the phenomenon of valency shifts. The question is, which factors are
responsible for prototypical clause patterns of the valency system in Mod-
ern German? Which factors condition these shifts? Valency shift is particu-
larly affected:
a) by phonetic shifts, as a consequence of which a syncretism of form
leads to re-analysis: the merging of different s-sounds in Late Middle High
German influenced verb valency. Indeed, the fact that the differentiation
between the inherited (MHG <s>) and the unvoiced s from the Germanic t
(MHG <a> or <aa>) in Late Middle High German was dropped (NHG <s>
or <ss>) has made it impossible since Early Modern to distinguish between
the neutral personal pronouns. In Middle High German, there were distinc-
tive forms for the genitive on the one hand, that is to say es (today: seiner
‘his’) and the nominative and accusative on the other hand, that is to say ea.
In Late Middle High German there is only one form, namely: es, for all
three cases.6
 Aspects of a diachronic valency syntax of German 89

(1) a. MHG es verdriuzet mich

compl (gen) V compl (acc)
MHG ea (nominative and accusative singular)
NHG es verdrießt mich7
compl (nom) V compl (acc)
‘it irritates me’

The tendency to introduce a formal subject is illustrated in this example. In

New High German impersonal structures without a subject, which are
mostly ergative structures such as mich friert [‘I am cold’] or mich hungert
[‘I am hungry’], are adapted to prototypical clause patterns with subject
(experiencer) and object.8
The s-merging also affects the so-called strong declension of the adjec-
tive, which has a number of forms identical with the declension of the pro-
noun:

(1) b. LU Matt. 9,4

ENHG Warumb denckt jr so arges in ewren
hertzen?
compl (gen) or
compl (acc)9
‘Why do you think such a bad thing in your
hearts?’

With the occurrence of the syncretism of form, a reinterpretation and thus

re-analysis by the hearer/reader is possible, meaning that an earlier genitive
as in the Lutheran example (so arges) can also be understood as an accusa-
tive form. The valency shift happens here through the re-organisation of the
morphological case of a genitive into an accusative complement.
b) Another reason for valency shift is the decline of the morphological
case for the benefit of analytically formed prepositional cases.

(2) Sie rühmte sich ihrer Taten. Sie rühmte sich wegen ihrer Taten.
‘She prided herself on her actions.’
Sie würdigte ihn keines Blickes. Sie würdigte ihn mit keinem Blick.
‘She did not deign to look at him.’

In sentences like these, the genitive phrase is becoming less common and is
gradually being replaced by the prepositional phrase. It is common knowl-
edge that verbs with a genitive complement have become rare. For Old
High German, 198 verbs followed by the genitive were counted in Otfrid’s
90 Mechthild Habermann

Evangelienbuch and at least 260 such verbs are recorded for Middle High
German.10 With around 40 verbs followed by the genitive, Modern German
only retains 15% of the original number.11
c) Valency shift can be caused by a decrease in the variety of possible
constructions linked with a gradual development of prototypical clause
patterns:
There are many verbs with multiple valency patterns in Old and Middle
High German which have simultaneously two possible valency patterns. In
addition to the subject (or nominative complement) there is a genitive or
accusative complement, occasionally a dative complement, apparently arbi-
trarily.
Subsequently, I would like to examine verbal patterns and thus show the
idiosyncrasy of historical valency syntax in order to lay the basis for a dia-
chronic valency. It is assumed that the distinction between genitive and
accusative complements is not arbitrary but indeed deliberate and inten-
tional.

4. Historical valency of German: Diachronic perspectives

As in Modern German, the meaning of the verb always has a great influ-
ence on valency. The distinction between genitive and accusative comple-
ments remains in Old High German particularly in verbs of certain groups,
i.e.:12

action and effort e.g. biginnan [‘begin’], geban [‘give’]

striving and desire e.g. gerōn [‘desire’], āhten [‘respect’]
mental activity e.g. thenken [‘think’], gilouben [‘believe’]
speech and communi- e.g. giwahan [‘mention’], manōn [‘urge’],
cation [‘remind of’]
question and request e.g. frāgēn [‘ask’], bitten [‘ask for’]
mental processes e.g. frewen [‘be glad’], sorgēn [‘worry’]
separation e.g. tharbēn [‘not participate’], mīdan
[‘avoid’]

Alongside the meaning of the verb the Old Indo-European meaning of the
case plays an essential part in the genitive. Genitive actually means “case of
origin”, which includes the functions of the ablative, such as separative.
According to this definition, both the partitive meaning (‘part of some-
thing’) which was common in Old and Middle High German, and its rela-
 Aspects of a diachronic valency syntax of German 91

tional meaning, which I could paraphrase as a relationship building func-

tion, are thus classified.
The next question is whether the difference in meaning for one and the
same verb lies in alternating structural possibilities. I will thus now discuss
the difference in meaning between the structure of the genitive complement
on the one hand, and the accusative complement on the other.
Jacob Grimm (1837/1989: 646) had already noticed a difference and
stated that genitive was the case of “geringere objectivisierung” [‘case of
the lesser objectification’]. Its primary meaning is that of participation. The
primary meaning of the accusative is that of involvement.13
Verbs of thought and perception such as gedenken [‘think of’] offer the
best opportunity to describe alternative structures.14 Compare the example
of gedenken, with a genitive complement:

(3) a. NL 1757,1 [1695,1]

Er gedâhte langer mære, diu wâren ê geschehen
compl V compl (gen)
(nom)
‘Er erinnert(e) sich an lange Geschichten, die einst geschehen
waren’
‘He remembered / he remembers long tales which once oc-
curred’

The genitive complement has a looser relationship to the predication than

the accusative. To some extent, the meaning expressed by the use of geni-
tive externally influences that of the nominative (subject). It is not so much
the agent as the experiencer, which is embodied in the subject.
In the example of the Nibelungenlied (3a), the existing meaning of the
genitive appears independently of the range of influence of the verb. That is
to say, the long tales already exist, before the action of gedenken is imple-
mented. The meaning of the verb and the meaning of the case operate to-
gether. In this context gedenken [‘think of’] includes the meaning ‘think
out’, so that one knows or is reminded of.
The meaning ‘remember’ in gedenken is not coincidental. Indeed, the
act of memorising uses a process of thought, that is, recalling. Therefore,
the very frequent link between past tense and genitive complement is –
probably – not a coincidence either. It is similarly the case with verbs such
as sich erschrecken [‘get a fright’], vergessen [‘forget’] and sich befreien
[‘free so.’].
92 Mechthild Habermann

A very different meaning of the structure is introduced in the case of

gedenken followed by an accusative complement, the case of the direct
object:

(3) b. Herb. 13450

so gedenke ich wol die list
adjunct V compl adjunct compl
(caus) (nom) (mod) (acc)
‘Deshalb denke ich mir gutüberlegt die / meine Vorgehens-
weise aus’
‘Therefore I think out well my action’

In this example, the meaning of the accusative complement is directly af-

fected by the action of the verb. It is the result of the “thought process”.
Here, the meaning of über etwas nachdenken [‘think about something’]
is linked with etwas ausdenken [‘think something up’]. The action of the
verb governs the object, in a rather abstract way that emerges gradually or
is created during the action. The meaning of the accusative is the content or
the result of the action carried out. This is also the case with verbs such as
schreiben [‘write’] and bauen [‘build’].
In these examples, the occurrence of valency arises through both the
meaning of the verb and the meaning of the case. The genitive complement
refers to an external, already present object (which is brought to light
through the action of the verb), the accusative object refers to an internal
emerging object, to an object of result.
A further distinction, connected with the dichotomy of external and in-
ternal object defines the choice of the case in Old and Middle High Ger-
man. The semantic class of the object may also be influential. The accusa-
tive complement is used for an abstract noun, the genitive complement for a
concrete noun. This is not so clear in the case of gedenken, but is easily
recognisable in the case of the Old High German niazan [‘enjoy, turn to
profit’]:15

(4) a. O 5,22,5
Thie selbun gotes liuti th$r niazent liohto z0ti
compl (nom) mod (loc) V compl (acc)
‘Das Gottesvolk dort genießt die lichten Zeiten’
‘God’s people there are enjoying the bright times’
 Aspects of a diachronic valency syntax of German 93

b. O 1,11,8
thaz se erdr0ches niez)n
conj compl compl V
(nom) (gen)
‘... sofern sie das Erdreich (das Land) genießen’
‘... while they enjoy the land’

There is undoubtedly a relic of the old partitive meaning in the genitive

complement. Distinctions were later blurred, particularly in the eighteenth
century, when abstract objects are expressed by the genitive complement.
This brings me to another essential point: when dealing with a verb that
can take either a genitive or an accusative complement, textlinguistics play
a central role alongside the external object and result object in choosing
which case is appropriate. Especially Richard Schrodt (1992: 385; 2004:
82–83, § S 78) has referred to this point.
Depending on whether the object was previously mentioned or not, the
genitive or the accusative, respectively, are used. The text-deictic and the
text-phoric function are closely interrelated here: in the case of the genitive
complement, the external object correlates with the factor of being previ-
ously mentioned or given, and in the case of the accusative complement the
object of result correlates with the factor of being new. In the case of some
verbs, verb valency depends essentially on the dichotomy of given and
new, or on the structure of theme and rheme.
The verb h@ren [‘hear’] in Old High German (di- or trivalent pattern)
might illustrate this: the person “who is being listened to or obeyed” is in
the dative (beside the nominative), the command, which the person (sub-
ject) obeys, can be in accusative, dative or genitive or occur as a thaz-
clause:

(5) a. with dative G ‘auf jmdn. hören, ihm gehorchen’

E ‘obey sb.’
with accusative G ‘etw. hören, etw. erhören, auf etwas
hören’
E ‘hear sth.; listen to sth.’
with genitive G ‘auf etwas (schon Vorhandenes,
schon Gesagtes) hören, es beachten’
E ‘listen to sth. (which has been said /
mentioned before)’.16
94 Mechthild Habermann

b. T 52,7
sie h@rent imo?
compl (nom) V compl (dat)
‘Gehorchen sie ihm?’
‘Do they obey him?’
c. O 1,17,53
Thaz imbot sie gih@rtun
compl (acc) compl V
(nom)
‘Sie hörten den Befehl’
‘They heard the order’
d. O 2,9,55−7
quad ... thaz er got forahta tho er sul0h werk worahta
‘er sagte, ... dass er Gott fürchte, weil er eine solche Tat
vollbrachte’
‘he said ... that he feared God because he accomplished such
an act’
Ioh s0n)ro worto er h@rta filu harto
conj compl compl V adjunct
(gen) (nom) (mod)
‘Hinsichtlich seiner Worte (seinen Worten) gehorchte er
nämlich sehr’
‘As to his words he obeyed well’

In example (5d), the anaphoric reference is doubled, coded by the case and
anaphors (personal and possessive pronouns er, s0n)ro).
A further example: in the case of a question, the circumstances which
are being questioned are previously mentioned or known.17 Therefore, this
verb is followed by the genitive. When the circumstances are unknown, it is
followed by the accusative (or prepositional case):

(6) a. O 3,12,5
thes ich nu fr$g)n iuih
compl compl adjunct V compl
(gen) (nom) (temp) (dat)
‘... wonach ich euch jetzt frage’
‘... what I am going to ask you about’
 Aspects of a diachronic valency syntax of German 95

b. T 170,30f.
[tho quuat her imo]
uuaz mih fr$g)s fon guote
compl (acc) compl (acc) V mod (prep) /
compl (prep)
‘[da sprach er zu ihm:]
Was fragst du mich nach den guten Dingen?’
‘[Then he said to him:]
What are you asking me about the good thing?’

Very frequently, the use of the genitive complement in a textlinguistic

function also involves the use of a pronoun as the part of speech which
additionally intensifies the cataphoric function. This includes personal,
possessive, demonstrative and relative pronouns.
The results presented here have so far been verified only with limited
groups of verbs, namely the verbs of thought, perception and interrogation,
of seeing and hearing, of eating and drinking and the like. Whether these
characteristics are also found with other groups of verbs is at this time open
to debate. As yet, we have no studies based on large corpora and they will
not be undertaken until we have digital historical language corpora for the
German language. The individual occurrences of verbs can then be ana-
lysed in context, including textlinguistic aspects, and statements on the
frequency of certain types of occurrences can be made.
Another question is if polyvalency, especially the choice of genitive
versus accusative complement for text linguistic reasons, is actually a phe-
nomenon of verb valency or if it is merely a phenomenon of language use,
such as omitting optional complements. If the latter were the case, it would
not touch the level of the language system.
The distinction proves to be extremely difficult. In my opinion, there is
only one, in practice vague criterion, which can hardly be employed for
historical language periods: valency variation is rooted in the language
system if the meaning of the verb changes depending on the choice of the
case.

5. Conclusion

The verb valency of Old and Middle High German is determined by many
different factors. Different possible structures are applied to increase the
polysemy of verbs, much more so than in Modern German. The different
partial significances are activated according to the structural framework
96 Mechthild Habermann

which covers the range of meaning of gedenken from ‘denken’ [‘think’] to

‘erinnern’ [‘remember’]; of h@ren from ‘etwas hören’ [‘hear something’] to
‘etwas hören, gehorchen’ [‘obey’].
This diversity of meaning(s) is often still effective in New High German
when connected to prepositional phrases. There are, however, particular
lexemes which lexically codify the equivalent partial significance of Old
High German polysemic verbs. This is essentially different from historical
language periods.
The lexical meaning of Old and Middle High German is not clearly de-
fined. It is formed within a context, and according to the combination of the
meaning of the verb and the meaning of the case. Nevertheless, it remains
variable and imprecise in lexicographical terms. It is difficult to compile
polyvalency in verb valency dictionaries, and until Early New Modern
German it is in my view almost impossible.
The differentiation of meanings according to context is conditioned by
the scantiness of verbs. So far we have no verbs that take particles; at most
they are polysemic simplex verbs. Until well into the eighteenth century,
the vernacular language lacks something crucial: it lacks a copia verborum,
an adequately varied vocabulary, which is an achievement of modern times.
The specific meaning of the case plays a decisive role in activating the
meaning of the verb. It is strongly determined by its Indo-European pri-
mary meaning, at least in the cases of genitive and dative.
Using this argument, I have tried to establish the differences in the use
of polyvalent verbs, working from the primary meaning of the genitive as a
linking case. One and the same verb (from a specific semantic group) is
constructed with a genitive complement if the genitive meaning in the (ex-
tralinguistic) reality is already available – it carries the semantic mark of
the concrete or has already been mentioned in the text.
However, the accusative complement is then used if the object is not yet
available but appears through the use of the verb, which has a semantic
mark of abstraction or has not previously been mentioned in the text.
The function of the case develops from this past relationship between
meanings, i.e. the linguistic function of the text in the sentence, but also of
the sense of coding of the given and new, of theme and rheme. This func-
tion is achieved through textual elements of New High German. Old and
Middle High German valency is not only part of the grammar of the sen-
tence, but also of the grammar of the text.
 Aspects of a diachronic valency syntax of German 97

Notes

1. In the following English complement will correspond to German Ergänzung

and adjunct to German Angabe.
2. Cf. in detail Greule (1982: 192–219, 1983: 85, 1995: 358–362).
3. On the declension of nouns cf. Paul (1881/1998: 184–207, § 174–195).
4. The variety of constructions found with Old High German verbs is docu-
mented impressively in Greule’s valency dictionary (1999). The best-known
phenomenon is that in the case of negation in a sentence an accusative com-
plement can be replaced with a genitive phrase; cf. Ebert (1986: 38–39).
5. Cf. at length Delbrück (1893: 172–199, § 55–80) and Braune (1886/2004:
182–183, § 192e) with further reading.
6. On the declension of pronouns and on the merging of the s-sounds cf. Paul
(1881/1998: 162–163, § 151–154 and 220–221, § 214) and Ebert et al. (1993:
110–113, § L 52, and 214–215, § M 63 Anm. 3).
7. Cf. Ebert (1986: 60).
8. Cf. Ágel (2000: 270–273); for the terminology cf. Fillmore (2003: 464).
9. It is also possible to interpret arges as a modifier (genitivus partitivus) to so.
10. Cf. Schrodt (1996: 78–82) and Rausch (1897: 54).
11. Cf. Kolvenbach (1973: 123). Lenz (1998: 3) still identifies 56 verbs followed
by genitive, many of which however can be shown to be outdated or stylisti-
cally marked; cf. Dürscheid (1999: 34–37).
12. Cf. Schrodt (2004: 80, § S 75), Behaghel (1923/1989: 562–574, § 407–409),
Ebert (1986: 40–43) and Paul (1881/1998: 341, § 361).
13. Cf. Delbrück (1893: 308, § 148, and 360–361, § 176).
14. This group of verbs has already been investigated by Milligan (1960). More
attention has been directed to them through the works of Schrodt (1992: 372–
377, 387–391, and 2004: 81–82, § S 76); cf. Donhauser (1998: 73–76).
15. Cf. Greule (1999: 184–185, cf. niozan) and Schrodt (2004: 83–84, § S 79).
16. Cf. Greule (1999: 130–133, cf. hôren).
17. Cf. Greule (1999: 82f., cf. frâgên) and Schrodt (2004: 83, § S 78).

References

1. Sources

Herb. = Herbort’s von Fritslâr liet von Troye

1966 Reprint. Georg Karl Frommann (ed.), (Bibliothek der gesammten
deutschen National-Literatur 5.), Amsterdam: Rodopi. Original editi-
on, Quedlinburg/Leipzig, 1837.
LU = Martin Luther. Biblia. Das ist die gantze Heilige Schrifft Deudsch auffs new
zugericht
1974 Reprint. Vol.3 Hans Volz (ed.), (dtv Text-Bibliothek 6033) München:
Deutscher Taschenbuch Verlag. Original edition, Wittenberg, 1545.
98 Mechthild Habermann

NL = Der Nibelunge Nôt. Mit den Abweichungen von der Nibelunge liet, den Les-
arten sämtlicher Handschriften und einem Wörterbuche 1. Theil:
Text; 2. Theil 1. Hälfte: Lesarten; 2. Theil 2. Hälfte: Wörterbuch
1966 Reprint. Karl Bartsch (ed.), Hildesheim: Olms. Original edition:
Leipzig, 1870; 1876; 1880.
O = Otfrid’s Evangelienbuch
1982 Reprint. Paul Piper (ed.), Hildesheim: Olms. Original edition, Tübin-
gen, 1882. 2d ed. Freiburg i. Br.
T = Tatian. Lateinisch und deutsch mit ausführlichem Glossar
1966 Reprint. Eduard Sievers (ed.), 293−515 (Bibliothek der ältesten deut-
schen Litteratur-Denkmäler 5.), Paderborn: Schöningh. 2d revised
edition, Paderborn, 1892.

2. Secondary literature

Abraham, Werner
2005 Deutsche Syntax im Sprachenvergleich: Grundlegung einer typologi-
schen Syntax des Deutschen. 2d improved and enlarged ed. (Studien
zur deutschen Grammatik 41.) Tübingen: Stauffenburg.
Ágel, Vilmos
1988 Überlegungen zur Theorie und Methode der historisch-synchronen
Valenzsyntax und Valenzlexikographie: Mit einem Verbvalenzlexi-
kon zu den „Denkwürdigkeiten der Helene Kottannerin (1439–
1440)”. (Lexicographica Series Maior 25): Tübingen: Niemeyer.
2000 Valenztheorie. Tübingen: Narr.
Behaghel, Otto
1989 Reprint. Deutsche Syntax. Eine geschichtliche Darstellung. Vol. 1:
Die Wortklassen und Wortformen. A. Nomen. Pronomen. Heidelberg:
Winter. Original ed., Heidelberg, 1923.
Braune, Wilhelm
2004 Reprint. Althochdeutsche Grammatik I. Laut- und Formenlehre. 15th
revised ed. Ingo Reiffenstein (ed.), (Sammlung kurzer Grammatiken
germanischer Dialekte A, 5/1.) Tübingen: Niemeyer. Original edition,
1886.
Delbrück, Berthold
1893 Vergleichende Syntax der indogermanischen Sprachen. Erster Theil.
(Grundriß der Vergleichenden Grammatik der Indogermanischen
Sprachen 3). Straßburg: Trübner.
Donhauser, Karin
1998 Das Genitivproblem und (k)ein Ende? Anmerkungen zur aktuellen
Diskussion um die Ursachen des Genitivschwundes im Deutschen. In
Historische germanische und deutsche Syntax. Akten des internatio-
nalen Symposiums anläßlich des 100. Geburtstages von Ingerid Dal,
Oslo, 27.9.-1.10.1995, John Ole Askedal (ed.), 69–86 (Osloer Beiträ-
 Aspects of a diachronic valency syntax of German 99

ge zur Germanistik 21.) Frankfurt M./Berlin/Bern/New York/Paris/

Wien: Lang.
Dürscheid, Christa
1999 Die verbalen Kasus im Deutschen. Untersuchungen zur Syntax, Se-
mantik und Perspektive. (Studia linguistica Germanica 53.) Ber-
lin/New York: de Gruyter.
Ebert, Robert Peter
1986 Historische Syntax des Deutschen II: 1300–1750. (Germanistische
Lehrbuchsammlung 6.) Bern/Frankfurt M./New York: Lang.
Ebert, Robert Peter, Oskar Reichmann, Hans Joachim Solms, and Klaus-Peter
Wegera
1993 Frühneuhochdeutsche Grammatik. (Sammlung kurzer Grammatiken
germanischer Dialekte A, 12.) Tübingen: Niemeyer.
Fillmore, Charles
2003 Valency and semantic roles: The concept of deep structure case. In
Dependenz und Valenz. Ein internationales Handbuch der zeitgenös-
sischen Forschung, Vilmos Ágel, Ludwig M. Eichinger, Hans-
Werner Eroms, Peter Hellwig, Hans Jürgen Heringer, and Henning
Lobin (eds.), 457–475. (Handbücher zur Sprach- und Kommunika-
tionswissenschaft 25.1.) Berlin/New York: de Gruyter.
Greule, Albrecht
1973 Valenz und historische Grammatik. Zeitschrift für Germanistische
Linguistik 1: 284–294.
1982 Valenz, Satz und Text. Syntaktische Untersuchungen zum Evange-
lienbuch Otfrids von Weißenburg auf der Grundlage des Codex Vin-
dobonensis. München: Fink.
1983 Zum Aufbau einer dependenziellen althochdeutschen Syntax. Sprach-
wissenschaft 8: 81–92.
1995 Valenz im historischen Korpus. In Dependenz und Valenz, Ludwig
M. Eichinger, and Hans-Werner Eroms (eds.), 357–363. (Beiträge zur
germanistischen Sprachwissenschaft 10.) Hamburg: Buske.
1999 Syntaktisches Verbwörterbuch zu den althochdeutschen Texten des 9.
Jahrhunderts. Altalemannische Psalmenfragmente, Benediktinerre-
gel, Hildebrandslied, Monseer Fragmente, Murbacher Hymnen,
Otfrid, Tatian und kleinere Sprachdenkmäler. (Regensburger Beiträ-
ge zur deutschen Sprach- und Literaturwissenschaft B 73.) Frankfurt
M./Berlin/Bern/New York/Paris/Wien: Lang.
Grimm, Jacob
1989 2d reprint. Deutsche Grammatik. Vol. 4, 2. Gustav Roethe, and Ed-
ward Schröder (eds.). Hildesheim: Olms. Original edition, 1837.
Kolvenbach, Monika
1973 Das Genitivobjekt im Deutschen. Seine Interrelation zu den Präposi-
tionalphrasen und dem Akkusativ. In Linguistische Studien IV. Fest-
gabe für P. Grebe zum 65. Geburtstag. Teil 2, 123–134. (Sprache der
Gegenwart 24.) Düsseldorf: Schwann.
100 Mechthild Habermann

Korhonen, Jarmo
1978 Studien zu Dependenz, Valenz und Satzmodell. Vol. 2: Untersuchung
anhand eines Luther-Textes. (Europäische Hochschulschriften II;
271.) Bern/Frankfurt M./Las Vegas: Lang.
1995 Zum Wesen der Polyvalenz in der deutschen Sprachgeschichte. In
Dependenz und Valenz, Ludwig M. Eichinger, and Hans-Werner
Eroms (eds.), 365−382. (Beiträge zur germanistischen Sprachwissen-
schaft 10.) Hamburg: Buske.
Lenz, Barbara
1998 Objektvariation bei Genitiv-Verben. Papiere zur Linguistik 58: 3–34.
Maxwell, Hugh
1982 Valenzgrammatik mittelhochdeutscher Verben. (Europäische Hoch-
schulschriften 1.504.) Frankfurt M./Bern: Lang.
Milligan, Thomas R.
1960 The German verb-genitive locution from Old High German to the
present: A study in structure of content. Ph.D. diss., New York Uni-
versity.
Paul, Hermann
1998 Reprint. Mittelhochdeutsche Grammatik. 24th revised ed. Peter
Wiehl, and Siegfried Grosse (eds.), (Sammlung kurzer Grammatiken
germanischer Dialekte A, 2.) Tübingen: Niemeyer. Original edition,
1881.
Rausch, Georg
1897 Zur Geschichte des deutschen Genitivs seit der mittelhochdeutschen
Zeit. Ph.D. diss., University of Gießen.
Schrodt, Richard
1992 Die Opposition von Objektsgenitiv und Objektsakkusativ in der deut-
schen Sprachgeschichte: Syntax oder Semantik oder beides? Beiträge
zur Geschichte der deutschen Sprache 114: 361–394.
1996 Aspekt, Aktionsart und Objektsgenitiv im Deutschen: Wie weit kann
eine systematische Erklärungsmöglichkeit für den Schwund des Ge-
nitivobjekts gehen? In Language Change and Generative Grammar,
Ellen Brandner, and Gisella Ferraresi (eds.), 71–94. (Linguistische
Berichte, Sonderheft 7/1995–96.) Opladen: Westdeutscher Verlag.
2004 Althochdeutsche Grammatik II. Syntax. (Sammlung kurzer Gramma-
tiken germanischer Dialekte A, 5/2.) Tübingen: Niemeyer.
The valency of experiential and evaluative
adjectives
Ilka Mindt

1. Introduction and aim

This paper focuses on adjectives which are followed by that-clauses. An

example is

(1) It is obvious that sovereignty does not mean dictatorship.

(AMK 481)

The valency pattern for the adjective obvious in this example is given in A
Valency Dictionary of English (VDE) as “[it] + (that)-CL” (Herbst et al.
2004: 559). The square brackets around the pronoun it indicate that the
adjective obvious has to be preceded by impersonal it. The adjective obvi-
ous is followed by a that-clause. The round brackets around the conjunction
that specify the conjunction that as an optional element. The conjunction
that can occur but it need not occur as can be seen in (2) where a zero reali-
sation is found.

(2) It was obvious a storm was coming in. (H9C 338)

The research reported here is part of a research project (for more details see
Mindt forthcoming) which focuses on the 51 most frequent adjectives in
the pattern “adjective + conjunction that”. These 51 adjectives account for
75% of all adjectives in the pattern “adjective + conjunction that”. For the
research reported here about 44,000 cases, all taken from the British Na-
tional Corpus (BNC), have been considered.
The aim of this paper is to offer a new description of adjectives in the
pattern “adjective + conjunction that”, which will then be compared with
the valency patterns given in the VDE. Section 2 focuses on the description
of the pattern “adjective + conjunction that” in three reference grammars of
English. In section 3 the empirical approach is presented, which leads to a
new classification of adjectives followed by that-clauses. A comparison
between the description found in the reference grammars with the new clas-
102 Ilka Mindt

sification is attempted in section 4 before the adjective classification is

compared with the valency patterns presented in the VDE in section 5.

2. Description of the pattern “adjective + conjunction that” in three

reference grammars

I will use the description of adjective complementation by that-clauses of

three present-day reference grammars of English as a basis for comparison.
Other studies which group adjectives followed by that-clauses are House-
holder, Alexander, and Matthews (1964) and Francis, Hunston, and Man-
ning (1996). Their grouping of adjectives is similar to the one given in the
three reference grammars. There are some studies that concentrate on one
particular group of adjectives followed by that-clauses: namely adjectives
in the construction “it + verb + adjective + that-clause”. Examples are Erd-
mann (1987), Hunston and Sinclair (2000) and Kaltenböck (2004). These
studies are not considered here in greater detail. The three reference gram-
mars are:
a) A Comprehensive Grammar of the English Language (Quirk et al.
1985),
b) Longman Grammar of Spoken and Written English (Biber et al. 1999),
and
c) The Cambridge Grammar of the English Language (Huddleston and
Pullum 2002).
A Comprehensive Grammar of the English Language distinguishes between
two sets of adjectives: (i) “Adjectives with experiencer as subject” and (ii)
“Adjectives with anticipatory it as subject” (Quirk et al. 1985: 1223−1225).
A typical example of an adjective with an experiencer as subject from the
BNC is aware in (3), whereas the adjective apparent in (4) occurs with
anticipatory it as subject.

(3) I am also very aware that all this is relative. (A0F 2162)
(4) It was apparent that a genius had been born. (EX1 463)

The type of grammatical structure found in (4) is referred to as subject ex-

traposition1 in the CGEL (Quirk et al. 1985: 1224). Quirk et al. (1985:
1391−1392) describe extraposition in terms of a postponement of nominal
clauses. “The subject is moved to the end of the sentence, and the normal
subject position is filled by the anticipatory pronoun it.” (Quirk et al. 1985:
1391). Quirk et al. state that postponement is “more usual” (1985: 1392) for
clausal subjects − that-clauses in my research − “than the canonical posi-
 The valency of experiential and evaluative adjectives 103

tion before the verb” (1985: 1392). An example taken from the BNC of a
that-clause occurring before the verb in subject position is given in (5).

(5) That this was clearly a tactical decision quickly became apparent.
(AHK 85)

Extraposition is a special device in order to structure the information. It is

employed when end-weight or focus should be given to the postponed ele-
ment.
The Longman Grammar of Spoken and Written English (Biber et al.
1999: 671−674) differentiates between (i) “Adjectival predicates taking
post-predicate that-clauses” as in (3) and (ii) “Adjectival predicates taking
extraposed that-clauses” as in (4) (1999: 672).
The Cambridge Grammar of the English Language (Huddleston and
Pullum 2002: 957−964) distinguishes between (i) “Adjectives in predica-
tive function” (2002: 964) which may take declarative content clauses as
exemplified in (3) and (ii) adjectives with extraposed subjects as in (4)
where the that-clause can also occur as the subject (see [5]).
All three grammars basically distinguish two groups of adjectives. One
group comprises adjectives which are followed by a that-clause. These
adjectives are described differently in the three reference grammars. Quirk
et al. state that they occur with an experiencer in subject position. Biber et
al. and Huddleston and Pullum do not comment on the subject in the matrix
clause. They just describe this group of adjectives as being complemented
by a that-clause as opposed to the second group of adjectives. The second
group consists of adjectives which can occur with extraposed subjects.
They are referred to as “Adjectives with anticipatory it as subject” by Quirk
et al. (1985: 1224), as “Adjectival predicates taking extraposed that-
clauses” by Biber et al. (1999: 672) and as “adjectives taking a clause as
(extraposed) subject” (2002: 964) by Huddleston and Pullum.
Table 1 lists examples of the two groups of adjectives as given by Quirk
et al. (1985: 1223−1225). Only those adjectives are listed in table 1 which
are found in the CGEL and are also considered in the empirical research
outlined in section 3.
104 Ilka Mindt

Table 1. The grouping of adjectives by Quirk et al.

adjectives with experiencer as subject adjectives with anticipatory it as

subject
afraid, angry, anxious, aware, cer- apparent, appropriate, arguable,
tain, confident, disappointed, glad, certain, clear, essential, evident, im-
grateful, happy, hopeful, pleased, portant, inconceivable, inevitable,
sad, sorry, sure, surprised likely, obvious, odd, possible, prob-
able, sad, strange, surprising, true,
unfortunate, unlikely, vital

In the following, I will not use the term anticipatory it but instead refer to
the pronoun it in so-called extraposition as impersonal it. This terminology
is in accordance with the VDE (Herbst et al. 2004: xx). Impersonal it must
be distinguished from another use of the pronoun it as in (6).

(6) The commission says it’s adamant that the public will have the
final say. (K1R 897)

The pronoun it in (6) will be termed referring it, because the pronoun it
refers to the noun phrase the commission.

3. An empirical approach to adjectives followed by that-clauses

The aim of the empirical approach described here is to arrive at a system-

atic description of adjectives followed by that-clauses, which accounts for
all cases without any exception. The approach is not based on any linguistic
theory or framework. The findings rest exclusively on the language sam-
ples. The analysis of the language samples leads to a new classification of
adjectives and is based on the subject types of the matrix clause. It will be
explained in the following (for more details see Mindt forthcoming).
The adjectives are first distinguished according to whether they co-
occur with a subject or without a subject.

(7) Keith kept up a brisk pace, glad that it was a full moon.
(HUA 1319)
(8) But he was glad that there was no mirror in this room. (ADA 824)
 The valency of experiential and evaluative adjectives 105

In (7) no subject precedes the adjective glad whereas in (8) the subject he
co-occurs with glad. Cases with a subject are then further distinguished
according to the following subject types:
a) pronominal subject vs. non-pronominal subject and
b) intentional subject vs. non-intentional subject.
The subjects we and it in (9) and (11) are examples of pronominal subjects,
environmentalists in (10) and the Bible in (12) are non-pronominal sub-
jects. The subjects we (9) and environmentalists (10) are intentional sub-
jects, it (11) and the Bible (12) are non-intentional subjects.

(9) We are hopeful that it will be a true festival of football. (A9N 135)
(10) Environmentalists are worried that the fumes from the fire are
hazardous. (K23 1445)
(11) It’s likely that trade will rise, but it doesn’t automatically follow.
(HYN 63)
(12) The Bible is quite clear that these evil spirits (and the things that
they do) are dangerous. (C8N 916)

The four subject types can be cross-classified as shown in table 2.

Table 2. Cross-classification of subjects

intentional subject non-intentional subject

pronominal subject we it
non-pronominal subject environmentalists the Bible

The pronouns I, you2, he, she, it (both impersonal it and referring it), we
and they are pronominal subjects, all other subjects are non-pronominal. An
intentional subject can act intentionally or may be able to act intentionally.
This distinguishes it from a non-intentional subject. The subject we in (9) is
an intentional subject which does not perform an intentional action because
the adjective hopeful refers to a feeling. Non-intentional subjects as imper-
sonal it in (11) or the Bible in (12) cannot act intentionally. Impersonal it as
a non-intentional subject is used in (11) as an option in the English lan-
guage to make a statement about the likelihood of something without ex-
plicitly naming the source who/that considered this as likely. The subject
the Bible in (12) is also non-intentional because no intentional action can or
may be performed by it. For more details on intentional and non-intentional
subjects see Mindt (forthcoming).
106 Ilka Mindt

The subject types of all 44,000 cases have been analysed. The analysis
of the subject types into intentional and non-intentional subjects cannot be
made on the basis of formal criteria but has to be carried out by the re-
searcher for each individual case. The decision on whether the subject is
intentional or non-intentional has to be made on the basis of the context.
The next step in the research was to discover if co-occurrences exist
between the different subject types and the 51 adjectives. Because of the
large number of cases it would have been prohibitive to analyse them
manually. But statistical procedures provide powerful tools to explore rela-
tionships between the subject types and the adjectives. I have employed
hierarchical cluster analysis. The purpose of cluster analysis “is to group
objects on the basis of the characteristics they possess” (Hair et al. 1998:
473). This means that the adjectives are grouped together into clusters on
the basis of their subject types.
Hierarchical cluster analysis yields two clusters of adjectives (see Mindt
forthcoming for further details on the application of cluster analysis). The
two classes of adjectives are presented in table 3.

Table 3. Classification of adjectives

Experiential adjectives Evaluative adjectives

adamant, afraid, angry, anxious, apparent, appropriate, arguable,

aware, certain, concerned, confident, certain, clear, essential, evident,
conscious, convinced, delighted, disap- good, great, important, inconceiv-
pointed, glad, grateful, happy, hopeful, able, inevitable, interesting, ironic,
pleased, sad, satisfied, sorry, sure, likely, natural, obvious, odd, possi-
surprised, unaware, worried ble, probable, sad, significant,
strange, strong, surprising, true,
unfortunate, unlikely, vital

The adjectives aware, adamant, glad, hopeful, and worried (examples [3],
[6-10]) belong to the first class. The adjectives obvious, apparent, likely,
and clear in examples (1), (2), (4), (11), and (12) are part of the second
class. The adjectives of each class have one semantic characteristic in
common. The adjectives of the first class are all experiential. Experiential
adjectives either convey a feeling or express certainty and confidence.
Those of the second class are evaluative because they convey a judgement
or an assessment. The adjectives certain and sad can occur as members of
both classes. Examples of the adjective certain are given in (13) and (14).
 The valency of experiential and evaluative adjectives 107

(13) An informed insider told me last night: “It is absolutely certain

that Nigel has done the deal.” (CH3 3179)
(14) He is almost certain that they went up Charterhouse Street.
(ANL 1315)

In (13) the adjective certain co-occurs with the non-intentional subject

impersonal it. The adjective certain conveys an evaluation. On the other
hand, the subject he in (14) is intentional and co-occurs with the experien-
tial adjective certain which expresses a meaning similar to that of sure (see
Mindt forthcoming for more details).
The adjectives have been classified with regard to their subject types in
the matrix clause and thus reveal typical co-occurrence patterns. Experien-
tial adjectives co-occur with intentional subjects, evaluative adjectives with
non-intentional subjects. But it is also an inherent semantic property of
experiential adjectives to favour intentional subjects because an intentional
subject is the carrier of the emotion or certainty/confidence expressed by
the adjective. Evaluative adjectives – on the other hand – co-occur with
non-intentional subjects. In contrast to an intentional subject, the non-
intentional subject cannot act intentionally. The evaluative adjective ex-
presses the judgement or assessment which is assigned to the non-
intentional subject. By using impersonal it as a non-intentional subject, a
speaker or writer has the possibility to express a judgement or an assess-
ment without explicitly naming the source of the evaluation. Impersonal it
as a non-intentional subject expresses the evaluation from a non-involved,
neutral and indeterminate point of view. This view is also shared by
Collins, who states that a speaker, when using impersonal it as the subject
in the matrix clause, ascribes “to an unspecified source the responsibility
for an assertion” (1994: 19). The empirical research outlined above shows
that the non-intentional subject impersonal it refers to such an unspecified
source.
The terms “experiential” and “evaluative” illustrate the semantic charac-
teristics of the adjectives. They should be considered as umbrella terms that
cover the specific meanings of these adjectives.

4. Differences in the adjective classifications

The grouping of the adjectives in the three reference grammars (Quirk et al.
1985; Biber et al. 1999; Huddleston and Pullum 2002) mainly rests on a
syntactic criterion which has been described as extraposition. Adjectives
108 Ilka Mindt

occurring with extraposed that-clauses and thus with impersonal it in sub-

ject position are differentiated from adjectives which are not followed by an
extraposed that-clause. Quirk et al. also introduce a semantic criterion for
adjectives followed by a that-clause: they have an experiencer as subject.
The adjectives comprising each group can then be further subdivided ac-
cording to their semantics (see for example Quirk et al. 1985: 1223−1225
or Biber et al. 1999: 671−674).
The new classification of adjectives can be explained in terms of lexico-
semantic features. One group of adjectives expresses either a feeling or
certainty/confidence and thus has been described as experiential. The other
group of adjectives conveys a judgement or an assessment and has been
termed evaluative. The two classes of adjectives emerged on the basis of
the co-occurrence patterns with their subject types. Experiential adjectives
typically co-occur with intentional subjects. These intentional subjects are
the carriers of the emotion or certainty conveyed by the adjective. Inten-
tional subjects are expressed by different forms of subjects, the most fre-
quent being a personal pronoun. The personal pronouns I, you, he, she,
(referring) it, we, and they account for 78% of all intentional subjects.
Evaluative adjectives co-occur with non-intentional subjects of which im-
personal it is the most frequent, accounting for 96.9% of all evaluative ad-
jectives with non-intentional subjects (for more information on frequency
distributions see Mindt forthcoming). By using impersonal it as the subject,
a speaker or writer can make a judgement or an assessment without the
need to explicitly refer to the source of the evaluation. Another example of
a non-intentional subject co-occuring with an evaluative adjective is the
Bible in (12).
The empirical analysis of more than 44,000 cases of adjectives followed
by that-clauses shows that there is no group of adjectives which exclusively
occurs with impersonal it as their subject. The class of evaluative adjectives
co-occurs with non-intentional subjects, of which impersonal it is one pos-
sible form. A subclassification of evaluative adjectives into those which co-
occur with impersonal it only and those which do not occur with imper-
sonal it results in two classes which list the same adjectives. Such a sub-
classification does not reveal any contrast and is therefore useless.
The grouping of adjectives given in the reference grammars as well as
the classification of the adjectives based on an empirical analysis revealed
two groups or classes of adjectives. The adjectives which comprise these
two groups or classes are largely identical (compare tables 1 and 3). The
group of adjectives which have been described in the reference grammars
as complemented by a that-clause largely match those which have been
termed experiential adjectives. The adjectives which are followed by an
 The valency of experiential and evaluative adjectives 109

extraposed that-clause are mostly part of the class of evaluative adjectives.

This means that the classification of adjectives into two classes as outlined
by the empirical research resulted in the same division of the individual
adjectives which has already been described by the reference grammars.
The main difference between the two accounts lies in the fact that the
distinction in the reference grammars is based mainly on syntactic criteria
whereas the new classification reflects systematic co-occurrences of the
adjectives with their subject types and can additionally be explained in
terms of lexico-semantic features inherent in the two adjective classes.
Cases such as (12) above or (15) below are not accounted for explicitly
by the reference grammars. The adjective clear is listed in all three refer-
ence grammars as being followed by an extraposed that-clause. This is not
the case in (12) and (15).

(15) The Equal Opportunities Commission says the Sex Discrimination

Act is clear that restricting taxi cab jobs to one sex is potentially
unlawful. (K26 1633)

Within the new adjective classification the adjective clear is considered as

an evaluative adjective co-occurring with non-intentional subjects of which
both The Bible and the Sex Discrimination Act are examples.
Huddleston and Pullum (2002: 964) mention cases similar to (16) (“I’m
quite clear that …”).

(16) Our audiences are clear that we are the most trusted source of
information and news in Britain. (J1L 62)

The subject our audiences in (16) as well as I in the example given by Hud-
dleston and Pullum is analysed in the empirical research as an intentional
subject. The meaning of the adjective clear in both examples is similar to
the meaning of certain or sure and conveys the semantic characteristics of
an experiential adjective. In a small number of cases which account for less
than 1% of all 44,000 cases an experiential adjective may acquire an
evaluative meaning or an evaluative adjective (as clear) may be analysed as
an experiential adjective. This shift in meaning is also reflected by the co-
occurrence of experiential clear with an intentional subject, whereas the
evaluative adjective clear co-occurs with a non-intentional subject.
Impersonal it is the most frequent form a non-intentional subject takes.
It can only be assumed that the high frequency of impersonal it is the rea-
son why cases with this subject have been considered separately in previous
accounts of adjective classifications. The high frequency might also be the
110 Ilka Mindt

reason why cases with impersonal it have received so much attention and
are the topic of many studies within different grammatical frameworks and
theories.

5. Valency patterns

In section 5.1. I will briefly refer to two accounts that have analysed imper-
sonal it in terms of its valency. Section 5.2. discusses the valency patterns
of experiential and evaluative adjectives as given in the VDE.

5.1. Analysing impersonal it

Herbst (1983: 33−38) argues that adjectives should be described in terms of

patterns. He states that impersonal it fulfils merely a syntactic function, but
goes on that impersonal it is necessary in terms of the structural require-
ments of a sentence and thus should be regarded as a complement in a
valency description. This implies that impersonal it is assigned a valency.
Herbst analyses adjectives that occur with impersonal it in subject position
as divalent (1983: 37): “Zwar läßt sich argumentieren, daß it, auch wenn es
nicht gegen andere Elemente austauschbar ist, eine Valenzstelle besetzt,
also die Funktion von E 1 wahrnimmt; dennoch erscheint es sinnvoll, bei
einer Valenzbeschreibung anzuführen, ob E 1 nur von it oder auch von
anderen Elementen besetzt werden kann” (1983: 33). In this quote he ex-
presses his view on the pedagogic description of cases with impersonal it in
subject position: when impersonal it occurs as a complement which cannot
be substituted by other elements then the valency pattern should make it
clear that impersonal it is the only subject of this particular adjective.
Seppänen (2002: 452−459) employs valency to analyse the status of
impersonal it in constructions as It is raining or It is cold. He states that
“[a]ll the verbs and multi-word predicates quoted above are – in the rele-
vant weather sense – zero valent, rather than monovalent” (2002: 452). He
does not give an analysis of the valency in constructions such as (1). Sep-
pänen cites examples identical in structure to (17) where impersonal it is in
direct object position and analysed as a placeholder for the that-clause fol-
lowing the adjective. This leads to the conclusion that the verb make is
trivalent, taking a subject, a direct object and an object complement or PC,
as Seppänen calls it. Herbst et al. (2004) also consider the verb make to be
trivalent. Herbst et al. give the following valency pattern for make which
 The valency of experiential and evaluative adjectives 111

describes (17) and (18): “+ NP/V-ingP + ADJ / it + ADJ-pattern” (2004:

514).

(17) In another extract Morton makes it clear that the Queen has sup-
ported Diana. (CEK 2749)
(18) Recent work makes us much less confident that any such clear
correlation is possible. (A6S 1278)

(17) consists of impersonal it followed by an adjective (clear) and a that-

clause which is dependent on the valency pattern of the adjective. This is
depicted in the valency pattern “+ it + ADJ-pattern”. (18) consists of the
object us and the adjective confident followed by a that-clause. The valency
pattern “+ NP + ADJ” describes this construction. It is clear from this out-
line that the that-clause is not part of the valency pattern of make but is a
feature of the adjective.

5.2. Adjectives in the VDE

The distinction of adjectives into two classes (evaluative and experiential)

is reflected in the valency patterns found in the VDE. Not all adjectives
which have been considered in the empirical approach are listed in the
VDE. Those that have been included in the VDE are presented with the
following pattern description for each of the two classes of adjectives:
a) experiential adjectives: “+ (that)-CL” or “+ that-CL”, and
b) evaluative adjectives: “[it] + (that)-CL” or “[it] + that-CL”
The valency pattern of experiential adjectives takes the form of a that-
clause with or without the conjunction that. The adjective surprised is an
experiential adjective and the valency pattern is reflected in the VDE in
examples (19) without the conjunction that and in (20) with the conjunction
that.

(19) “The family were surprised she’d found someone else so quickly.”
(Herbst et al. 2004: 834)
(20) “I was surprised that the police hadn’t followed me to New York.”
(Herbst et al. 2004: 834)

Evaluative adjectives are also described in the VDE as being followed by

that-clauses with or without the conjunction that. Additionally, all adjec-
tives classified as evaluative in the empirical research are specified in the
112 Ilka Mindt

VDE according to their subject: impersonal it. Examples for the evaluative
adjective possible taken from the VDE are found in (21) and (22).

(21) “It was possible they had been at the gatehouse for only a few
days.” (Herbst et al. 2004: 600)
(22) “It is possible that he thought that by alliance with her his career
would progress.” (Herbst et al. 2004: 600)

The valency patterns in the VDE serve a descriptive and a pedagogic pur-
pose. Its descriptive purpose is to provide the dictionary user with the most
frequent valency patterns. This is reflected in the valency pattern for expe-
riential and evaluative adjectives. Whereas experiential adjectives co-occur
with a wide range of different intentional subjects, it has been outlined
above that the most frequent form a non-intentional subject takes when co-
occurring with an evaluative adjective is impersonal it. Almost 97% of all
occurrences of evaluative adjectives are found with impersonal it in subject
position. The pedagogic purpose of the valency patterns in the VDE is to
give the dictionary user a clear outline of the valency for each adjective.
The pedagogic purpose is fulfilled in that evaluative adjectives are all as-
signed the valency pattern “[it] + (that)-CL” or “[it] + that-CL” and experi-
ential adjectives the valency pattern “+ (that)-CL” or “+ that-CL”.
The adjectives certain and sad have been classified in the empirical
research as belonging to both classes of adjectives. This is also reflected in
the VDE. The adjective certain as an experiential adjective is assigned the
valency pattern “+ (that)-CL” (Herbst et al. 2004: 120). An example from
the VDE is (23). Certain co-occurs with an intentional subject and conveys
an experiential meaning.

(23) “He was certain that no one he knew had seen him.”
(Herbst et al. 2004: 120)

As an evaluative adjective certain is assigned the valency pattern “[it] +

(that)-CL” (Herbst et al. 2004: 120), an example from the VDE is shown in
(24).

(24) “Now it seems certain that elections will go ahead.”

(Herbst et al. 2004: 120)

In (24) the evaluative adjective certain – similar in meaning to true – co-

occurs with the non-intentional subject impersonal it.
 The valency of experiential and evaluative adjectives 113

The adjective sad has also been assigned two different valency patterns
in the VDE: as an experiential adjective it is followed by a that-clause,
reflected in the valency pattern “+ that-CL” (Herbst et al. 2004: 716), as an
evaluative adjective it is described with the pattern “[it] + that-CL” (Herbst
et al. 2004: 716). An example from the VDE for the experiential adjective
sad is given in (25), for the evaluative adjective sad in (26).

(25) “I’m sad that we are leaving.” (Herbst et al. 2004: 716)
(26) “He had a truly original talent, and it is sad that in the end there
was so little to show for it.” (Herbst et al. 2004: 716)

In (25) the experiential adjective sad co-occurs with the intentional subject
I, the non-intentional subject impersonal it is found in (26) together with
the evaluative adjective sad.
The adjective clear has been classified as an evaluative adjective. As
has been outlined above it may also be found as an experiential adjective,
co-occurring with an intentional subject. Of all cases of the pattern “clear +
that-clause” in the empirical research only 2.9% occur with an intentional
subject. In all of these cases, the adjective clear has an experiential mean-
ing. This difference is also taken into account in the VDE. The adjective
clear can be complemented by a that-clause, which is reflected in the
valency pattern “+ that-CL” (Herbst et al. 2004: 137). Clear is also de-
scribed as an evaluative adjective, corresponding to the valency pattern “[it]
+ (that)-CL” (Herbst et al. 2004: 137). Again, the examples given in the
VDE clearly show that experiential clear co-occurs with intentional sub-
jects (“I”, “The security forces”; Herbst et al. 2004: 137), whereas evalua-
tive clear co-occurs with impersonal it.3 The frequency distribution based
on the empirical research is also corroborated by the information found in
the VDE. In the empirical research, the experiential adjective clear is only
found in 2.9% of all cases, whereas the evaluative adjective clear accounts
for 97.1% of all cases of the adjective clear in the pattern “adjective + that-
clause”. In contrast to the valency pattern “+ that-CL” the valency pattern
“[it] + (that)-CL” is labelled “frequent” in the VDE (Herbst et al. 2004:
137).

6. Conclusion

The aim of the research reported here is twofold: first, a new classification
of adjectives followed by that-clauses has been presented. Second, it has
114 Ilka Mindt

been demonstrated that this classification is also reflected in the valency

patterns of adjectives given in the VDE.
The classification of adjectives into evaluative adjectives and experien-
tial adjectives is based on more than 44,000 cases occurring in the pattern
“adjective + conjunction that”. By employing an empirical approach which
neither rested on previous descriptions of adjectives nor attempted a pre-
selection of cases it was possible to present a novel description of the pat-
tern “adjective + conjunction that”. This description not only leads to a new
classification of adjectives, but also accounts for all cases in the pattern
“adjective + conjunction that”.
The results from the empirical approach are reflected in the valency
patterns given in A Valency Dictionary of English (VDE). The VDE ac-
counts for both classes of adjectives. Experiential adjectives co-occur with
intentional subjects which take a wide range of forms and are therefore not
accounted for in the VDE. Evaluative adjectives occur most frequently with
only one subject: impersonal it. This is also reflected in the valency patterns
of the VDE.

Notes

1. Cases of so-called object extraposition are not considered in this paper. They
are discussed in Mindt forthcoming.
2. The pronoun you refers both to the singular and plural usage.
3. Two of the four examples in the VDE have the structure “make it clear that”
as in “He has made it clear that he does not want the job.” (Herbst et al., 2004:
137). These are considered as examples of object extraposition and are not
part of the research reported in this paper. For more information on the struc-
ture make it clear that see Mindt forthcoming.

References

Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad, and Edward
Finegan
1999 Longman Grammar of Spoken and Written English. London: Long-
man.
Collins, Peter
1994 Extraposition in English. Functions of Language 1: 7−24.
Erdmann, Peter
1987 It-Sätze im Englischen. Heidelberg: Carl Winter.
 The valency of experiential and evaluative adjectives 115

Francis, Gill, Susan Hunston, and Elisabeth Manning

1996 Collins COBUILD Grammar Patterns 2: Nouns and Adjectives.
London: HarperCollins.
Hair, Joseph F., Rolph E. Anderson, Ronald L. Tatham, and William C. Black
1998 Multivariate Data Analysis. New Jersey: Prentice Hall.
Herbst, Thomas
1983 Untersuchungen zur Valenz englischer Adjektive und ihrer Nomina-
lisierungen. Tübingen: Narr.
Herbst, Thomas, David Heath, Ian F. Roe, and Dieter Götz
2004 A Valency Dictionary of English. A Corpus-Based Analysis of the
Complementation Patterns of English Verbs, Nouns and Adjectives.
Berlin: Mouton de Gruyter.
Householder, Fred. W. Jr., Dee Alexander, and Peter H. Matthews
1964 Adjectives before That-Clauses in English. Indiana University Lin-
guistics Club, Indiana University. Indiana: Bloomington.
Huddleston, Rodney, and Geoffrey K. Pullum
2002 The Cambridge Grammar of the English Language. Cambridge:
Cambridge University Press.
Hunston, Susan, and John Sinclair
2000 A local grammar of evaluation. In Evaluation in text. Authorial
Stance and the Construction of Discourse, Susan Hunston, and Geof-
frey Thompson (eds.), 74−101. Oxford: Oxford University Press.
Kaltenböck, Gunther
2004 It-Extraposition and Non-Extraposition in English. A Study of Syntax
in Spoken and Written Texts. Wien: Braumüller.
Mindt, Ilka
forthc. Adjective complementation by that-clauses: An empirical study.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik
1985 A Comprehensive Grammar of the English Language. London:
Longman.
Seppänen, Aimo
2002 On analysing the pronoun it. English Studies 5: 442−462.
Valency rules? The case of verbs with propositional
complements
Michael Klotz

1. Introduction

Almost 20 years ago Noam Chomsky (1986) suggested that syntactic

valency properties of verbs (or c-selection in his terminology) may be seen
as an automatic consequence of the semantic properties of the verb and its
complements (or s-selection): “Let us assume that if a verb (or other head)
s-selects a semantic category C, then it c-selects a syntactic category that is
the ‘canonical structural realisation of C’ (CSR(C))” (Chomsky 1986: 87).
Investigating the verb persuade he concludes that

… the lexical entry for persuade need only indicate that it s-selects two
complements, one a goal, the other a proposition. All other features of the
VP headed by persuade are determined by general properties of the UG. A
child learning English must, of course, learn the meaning of the word per-
suade including its properties of s-selection … . Nothing more must be
learned … . In particular, no properties of c-selection and no rules of phrase
structure are required in this case. (Chomsky 1986: 88)

That these were more than just casual remarks can be seen from the fact
that Chomsky reiterated that view nine years later in 1995, when he noted
“… that subcategorization follows almost entirely from θ-role specifica-
tion” (Chomsky 1995: 31).
Of course, most of the grammatical machinery that Chomsky proposed
in the 1980s has been thrown overboard since, but the questions underlying
his remarks remain: rephrased in valency terminology we might ask to
which extent the syntactic valency of verbs can be reduced to semantic
facts. Are there (semantically motivated) valency rules or must we consider
valency as a purely lexical property of individual words? As Herbst sug-
gests elsewhere in this volume, the two views at stake here are those of rule
vs. storage.
If we accept the rule view for the moment, there seem to be at least three
kinds of semantic basis from which the form type (cf. Huddleston and Pul-
lum 2002: 1173) of complements could be deduced:
118 Michael Klotz

Firstly, the form type of the complement may depend on the meaning of
the verb itself. In this vein, the COBUILD Grammar Patterns 1: Verbs (Fran-
cis, Hunston, and Manning 1996) correlates syntactic patterns with verb
meanings. As John Sinclair puts it in the foreword (1996: iv): “… verbs can
be subdivided according to pattern, and patterns can be seen to correlate
with meaning – that is to say, verbs with similar patterns have similar
meanings”.
Secondly, the form may depend on the semantic (or theta) role relation
between the valency carrier and its complement. This is what Chomsky
seems to have in mind when he talks about subcategorisation following
almost entirely from θ-role specification (see above).
Thirdly, the form type itself may be associated with meaning. Thus, in
connection with non-finite clausal complements of verbs Quirk et al. (1985:
1191) suggest that “the infinitive gives a sense of mere ‘potentiality’ for
action … while the participle gives a sense of the actual ‘performance’ of
the action itself”.
Other authors, however, doubt whether the syntactic valency behaviour
of verbs can be predicted from such semantic facts. Thus Gazdar et al.
(1985: 32) conclude “that there are restrictions on contexts of occurrence
for lexical items which the grammar must specify, and which cannot be
reduced to facts about meaning”. Helbig (1992: 9) essentially takes the
same line when he asserts, „[d]ass diese verschiedenen Valenzebenen nicht
einfach isomorph aufeinander abbildbar sind“ [that these different levels of
valency cannot be mapped isomorphically onto each other]. Noël (2003:
369) reviews several authors who believe that syntactic constructions are
essentially semantically driven, but remains “convinced that in the absence
of convincing evidence to the contrary the clausal complements of the kind
considered here, do not have a different semantics”. Huddleston and Pul-
lum (2002: 1241) take an in-between position, asserting that “we cannot
assign distinct meanings to the form-types and treat the selection as seman-
tically determined. On the other hand, the selection is not random: verbs
with similar meanings tend to select the same form-types …”.
The present study is a contribution to this ongoing discussion. In par-
ticular, it will attempt to empirically test the assumption of a correlation
between the lexical meaning of and syntactic patterning around a valency
carrier by investigating the data contained in the Valency Dictionary of
English (Herbst et al. 2004). Taking its lead from Chomsky’s persuade
example, the study focuses on verbs which take propositions as arguments.
Although on the syntactic level these propositional arguments can be real-
ised in a considerable variety of ways, we will limit ourselves to cases
 Valency rules? The case of verbs with propositional complements 119

where the entire valency pattern to the right of the verb is made up alterna-
tively by the following three basic form types:

that-clause
VERB + N to-INF
N V-ing

The latter two non-finite constructions will be considered as the most im-
portant constructional equivalents to the that-clause, because like the that-
clause they also overtly express subject and predicate.
The following discussion will provide statistical data which shed some
light on the following two hypotheses:
– Hypothesis 1:
That-clause, N to-INF and N V-ing are constructional synonyms of each
other, i.e. they can replace each other in the same context (i.e. after a
given verb).
– Hypothesis 2:
Differences in complementation behaviour correspond to differences in
verb meaning.

2. That-clause, N to-INF and N V-ing as constructional synonyms

In order to test hypothesis 1, all occurrences of the above mentioned three

form types as complements to verbs were extracted from the electronic text
of the Valency Dictionary of English. Here it has to be pointed out that both
non-finite form types can be further subdivided so that we are actually not
looking at three single form types but rather three families of form types.
For example, with respect to the N to-INF as well as N V-ing type the
Valency Dictionary of English differentiates between cases like expect and
leave where the noun phrase can become subject of the passivised matrix
clause and cases such as demand and mind where it cannot. Compare the
following examples from the Valency Dictionary of English (VDE) and the
British National Corpus (BNC):

(1) a. This year I expect market conditions … to im- Np to-INF

prove somewhat … VDE
b. The meeting is expected to last two days. VDE
(2) a. I left the engine running and the lights on. VDE Np V-ing
120 Michael Klotz

b. Thomas Seton was left dangling in the air,

swinging and twitching grotesquely … BNC
(3) a. They demanded more planes to be made N to-INF
available. VDE
b. *More planes were demanded to be made
available.
(4) a. Where’re you from, if you don’t mind me N V-ing
asking? VDE
b. * … if I am not minded asking.

Both types were included in the count. Also included was the N INF type as
well as the for N to-INF type; the latter because the introductory for acts as
a construction marker (Matthews 1981: 59−61; cf. also Allerton 1982:
16−17) rather than a lexical preposition.

(5) Instantly, unmistakably, he felt her recoil. VDE N INF

(6) Originally the scheme allowed for pensions to be for N to-INF
calculated on the best twenty years of earnings. VDE

In contrast, constructions with prepositions followed by N V-ing as in ac-

count for them being here were excluded from the statistics, because here
the prepositions are much more variable and lexical in character. They of-
ten form a semantic unit with the verb, i.e. a prepositional verb in the sense
of Quirk et al. (1985: 1150−1167). In the following we will revert to simply
talking about form types, but it should be born in mind that this includes
their respective variants.
Combining the three form types in all logically possible ways results in
seven complementation classes:

Table 1. Seven complementation classes

that-CL N to-INF N V-ing

class 1 2 2 2
class 2 2 2
class 3 2 2
class 4 2 2
class 5 2
class 6 2
class 7 2
 Valency rules? The case of verbs with propositional complements 121

The Valency Dictionary of English contains 511 verbs. Of these, 211 take
one or several of the form types under study as their second complement.
The following statistics show the number of verbs in each group.

70
60
50
40
30
Σ Verbs
20
10
0
class 1 class 2 class 3 class 4 class 5 class 6 class 7

Figure 1. Number of verbs in each complement class

Several points are immediately apparent:

− Firstly, class 1 verbs which alternatively take all three form types as
their second complement are relatively rare.
− Secondly, none of the seven groups is empty; this means that there is no
form type which actually occurs with all verbs that take propositional
complementation. Thus none of the three form types could be called a
canonical realisation of propositional arguments in the sense that it
could be applied invariably across the board whenever a propositional
argument needs to be encoded in a grammatical form.
− Thirdly, the distribution of verbs across the classes is very uneven. Thus
the that-clause and the N to-INF construction in class 2 are more likely
alternants of each other than the that-clause and N V-ing in class 3.
− Fourthly, it is common for verbs to allow only one form type as realisa-
tion of the propositional argument. In fact, class 5 verbs which allow a
that-clause but none of the non-finite constructions constitute the largest
group.
The picture that emerges is complex but it clearly contradicts Chomsky’s
suggestion that the formal realisation is largely an automatic consequence
of the propositional character of the argument. However, some important
qualifications to the statistics presented above are in order.
Firstly, the statistics give the number of verbal lexemes in each class.
However, it is generally accepted that valency is not a property of lexemes
but rather of lexical units in Cruse’s (1986: 80) sense, i.e. understood as
“the union of a single sense with a lexical form”. Thus, the co-occurrence
122 Michael Klotz

of two form types in the valency of one verb in the statistics above does not
mean that the two form types actually are constructional synonyms; they
may just as well belong to different senses. However, using lexical units
instead of lexemes in statistics of the above kind would have increased the
complexity of the count considerably, all the more since the lexical unit is
only well defined in theory.
For that reason it seemed safer to count lexemes and just speculate about
how counting lexical units might have changed the statistics. Since count-
ing lexical units would have resulted in a much greater degree of diversifi-
cation, it seems clear that counts in classes 1 to 4, where form types co-
occur as alternatives, would have come out considerably lower. The view
emerging from my statistics – that the three form types are alternatives only
to a very limited extent – would have been strengthened further.
The second qualification concerns those verbs which allow N to-INF or
N V-ing but no that-clause. For a number of verbs there are good syntactic
reasons which prevent the that-clause, as appears from the examples below,
and these verbs should be counted separately.

(7) But in your particular case I prefer the precautions N to-INF

to be extreme. VDE
(8) … their parents will have to attend a meeting of the N + to-INF
school governors to try to convince them not to go
ahead with the expulsion. VDE

The N to-INF type after prefer and convince as they can be seen in these
example sentences are only superficially similar. Following Huddleston
and Pullum (2002: 1201), we can draw a distinction between a raised object
after verbs like prefer and an ordinary object after verbs like convince. Es-
sentially, prefer must be seen as a verb taking two arguments whereas con-
vince takes three. The Valency Dictionary of English makes the same dis-
tinction by distinguishing N to-INF from N + to-INF patterns.

prefer convince

I the precautions to be extreme (parents) them not to go ahead…

Since the that-clause always realises only one argument, it cannot replace
N + to-INF, i.e. those cases, where N and to-INF have to be seen as realisa-
 Valency rules? The case of verbs with propositional complements 123

tions of separate arguments. The same is true for the N V-ing type re-
spectively and the statistics have to take this distinction into account.

80
60 that-CL excluded on
40 syntactic grounds
20 other
0
class 1 class 2 class 3 class 4 class 5 class 6 class 7

Figure 2. Number of verbs in each complement class – amended

The amended graph therefore shows those cases where the that-clause is
excluded on syntactic grounds in dark grey. As can be seen, the number of
cases where the that-clause does not function as an alternative to the N to-
INF or N V-ing type as realisation of one argument is much more limited
now. Thus, the that-clause is certainly the one form type which comes
closest to being the canonical realisation of a propositional argument, how-
ever without being applicable in all cases either.

3. The correspondence of verb meaning and complement form type

We will now turn to hypothesis 2 and restate it here for matters of conven-
ience:
– Hypothesis 2: differences in complementation behaviour correspond to
differences in verb meaning.
To test the hypothesis, verbs were semantically classified for comple-
mentation classes 5 and 2, i.e. those verbs which only allow for a that-
clause and those which alternatively admit the N to-INF type. With respect
to the latter only those verbs were counted where the that-clause and N to-
INF represented real alternatives. As has already been pointed out, two
requirements have to be met for this:
Firstly, the N to-INF type has to represent a single argument. This ex-
cludes verbs like promise, where N and to-INF actually represent separate
arguments.

(9) SheA1 promised BerylA2 to keep an eye on himA3. VDE

124 Michael Klotz

Secondly, both form types have to be complements to the same lexical unit.
This excludes verbs like allow, which has clearly different meanings in
combination with the that-clause and N to-INF.

(10) I certainly would allow that things had taken + that-CL

an unfortunate turn. VDE ‘admit’
(11) … my wife has never allowed me to see them. VDE + N + to-INF
‘permit’

In the analysis 112 verbal lexical units were classified into the following
semantic groups:
− communication verbs: add, admit, beg, claim, command, demand, deny,
direct, explain, indicate, insist, joke, lie, maintain, object, pronounce,
question, request, respond, rule, say, shout, signal, (let) slip, state,
swear, threaten, whisper, write ...;
− opinion verbs: accept, agree, assume, believe, bet, conceive, consider,
doubt, fear, gather, guess, know, question, suppose, suspect, think, trust;
− fact finding verbs: calculate, conclude, decide, establish, estimate,
judge, learn, read, realize/realise, reason, recognise/recognize, reflect;
− fact demonstrating verbs: confirm, indicate, prove, reveal;
− fact manipulating verbs: conceal, hide, ignore;
− fact establishing verbs: arrange, check, ensure, guarantee, intend, plan,
pretend, provide;
− emotion verbs: pray, desire, hate, hope;
− imagination verbs: dream, suppose;
− unclassified: respect, vote, wonder.
The groups can be characterised in the following way: communication
verbs constitute the largest class by far. As the name suggests, they all have
in common that the proposition is communicated to somebody. Frequently,
though not always, these verbs allow complementation by a direct quote.
This latter fact distinguishes them from the opinion verbs; these verbs es-
sentially express the AGENT’s stance towards the truth value of the proposi-
tion on a scale from certainty (know, believe) to disbelief (doubt). Fact
finding verbs are similar to opinion verbs, but in contrast to those they have
an inchoative element to their meaning. Thus realise could be paraphrased
as ‘come to believe’. Fact demonstrating verbs also lead to opinions about
the truth of a proposition, but their AGENTS are not those individuals who
come to have these opinions; in fact, the AGENTS need not be animate at all.
Fact manipulating verbs are those which signify some reaction by the
AGENT to a true proposition. Fact establishing verbs like check, guarantee
and provide put the onus on the AGENT to make the proposition a true one.
 Valency rules? The case of verbs with propositional complements 125

Emotion verbs signify the emotional stance of the AGENT towards the
proposition. Nothing about the truth value of the proposition is implied.
The group includes verbs like pray, hope and wish. Finally, there are the
imagination verbs dream and suppose. Like emotion verbs they leave the
truth value of the proposition entirely open, but they do not express an
emotional stance either.
A few remarks about this semantic classification are in order: firstly, no
claim is being made as to the completeness of these groups. There may be
more semantic groups and the groups established may contain further
verbs. Secondly, it is clear that polysemic verbs can occur in more than one
semantic class. Thus suppose is an opinion verb in (12) and an imagination
verb in (13).

(12) She supposed that there was a copy of the book in ‘believe’
the library. VDE
(13) Suppose somebody found an unlimited supply of ‘imagine’
energy …

Thirdly, there are a few verbs which could not be subsumed in any one of
the semantic groups, but their number is surprisingly small. Most verbs
could be included in one of the groups with some degree of confidence
although it is clear that the groups should be seen as prototype categories
which allow for some gradience. Such gradience is exemplified by the verb
fear in

(14) I fear that John will be late again.

Like hope and wish it expresses an emotional stance towards the propo-
sition. But in contrast to these emotional verbs it also expresses the
AGENT’S stance towards the truth value of the proposition: a useful para-
phrase might be ‘I believe that John will be late again and I dislike this
possibility’. For that reason, fear was classified as an opinion verb in the
analysis.
The following graph shows the correlation between the semantic groups
outlined above and complementation classes 2 and 5.
126 Michael Klotz

70

60

50

40

30

20
class 2:
that-CL and N to-INF
10
class 5: that-CL only

0
co op fa fa fa fa em im un
m in ct ct ct ct ot ag cla
m io fin de m es i o i n ss
un n di m an tab n at ifi
ica 'be ng on ip l ' a io ed
tio lie s u l is h r r n
ve ' jud trati atio m a ng 's
n en e up
're ' ge on n t 'a ' p
po ' 'pr 'co os
rt' nc rra e'
ov e n
e' al' ge
'

Figure 3. Correlation between semantic classes and complementation classes 2 and 5

It is immediately apparent that both complementation classes co-occur with

all semantic groups. The only exception to this is the small group of the
three fact manipulating verbs hide, conceal and ignore. However, given the
small size of the group it is not clear whether this is actually a correlation
between semantic and valency properties or simply a coincidence. Apart
from this there is very little to suggest that the realisational possibilities of a
propositional argument could be predicted from the meaning of the verb to
which it belongs. The χ2 test lends support to this conclusion: at df = 8 the
resulting χ2 value of 7.7 is only about half of what would be necessary for a
significant result at the 95% level of confidence.
Slightly more significant results can be obtained, if we just compare the
two largest groups, verbs of communication and opinion:
 Valency rules? The case of verbs with propositional complements 127

Table 2. Correlation between two semantic groups and complement classes 2 and 5

communication verbs opinion verbs

class 2 20 10
class 5 40 7
χ2 (df = 1, N = 67) = 3,61
χ2 (necessary for 95% confidence level with df = 1) = 3,84

Here we might tentatively conclude that communication verbs prefer the

that-clause only, whereas opinion verbs show a slight preference for having
that-clause and N to-INF as alternatives. However, even this distribution
does not result in a χ2-value which would be significant at the 95% level of
confidence, although it is close to it.
Assuming that this distribution is not just coincidence, we are still no-
where near anything that might be called a regularity. If anything, it is
rather a tendency, which begs the question whether such tendencies have a
role to play from a psycholinguistic point of view. Although they obviously
do not allow to make any predictions, one might suggest that the psycho-
linguistic correlate to such statistical tendencies may be certain expecta-
tions in the speaker which allow the speaker to retrieve the actual valency
patterns more easily from memory.
In sum it seems fair to say that the statistical analysis of complementa-
tion data from the Valency Dictionary of English does not lend support to
the view that the valency of a verb can be deduced from its meaning. The
storage view of valency which sees it as an irregular lexical rather than
semantically rule-based phenomenon is strengthened further.

References

Allerton David J.
1982 Valency and the English Verb. London/New York: Academic Press.
Chomsky, Noam
1986 Knowledge of Language: Its Nature, Origin and Use. New York:
Praeger Publishers.
1995 The Minimalist Program. Cambridge, Mass.: Massachusetts Institute
of Technology Press.
Cruse, David A.
1986 Lexical Semantics. Cambridge: Cambridge University Press.
Francis, Gill, Susan Hunston, and Elizabeth Manning
1996 Collins Cobuild Grammar Patterns 1: Verbs. London/Glasgow:
HarperCollins.
128 Michael Klotz

Gazdar, Gerald, Ewan Klein, Geoffrey K. Pullum, and Ivan A. Sag

1985 Generalized Phrase Structure Grammar. Cambridge, Mass.: Harvard
University Press.
Helbig, Gerhard
1992 Probleme der Valenz- und Kausustheorie. Tübingen: Max Niemeyer
Verlag
Herbst, Thomas
2007 Valency complements or valency patterns? This volume.
Herbst, Thomas, David Heath, Ian Roe, and Dieter Götz (eds.)
2004 A Valency Dictionary of English. Berlin/New York: Mouton de-
Gruyter.
Huddleston, Rodney, and Geoffrey K. Pullum
2002 The Cambridge Grammar of the English Language. Cambridge:
Cambridge University Press.
Matthews, Peter H.
1981 Syntax. Cambridge: Cambridge University Press.
Noël, Dirk
2003 Is there semantics in all syntax? The case of accusative and infinitive
constructions vs. that-clauses. In Determinants of Grammatical
Variation in English, Günter Rohdenburg, and Britta Mondorf (eds.),
347−377. Berlin/New York: Mouton de Gruyter.
Quirk, Randolph, Sydney Greenbaum, Geoffrey Leech, and Jan Svartvik
1985 A Comprehensive Grammar of the English Language. London:
Longman.
Valency issues in FrameNet1

Charles J. Fillmore

1. Introduction

This chapter describes the assumptions and practices of the Berkeley Frame-
Net project and shows how these have led to a particular treatment of the
concept of valency.2 The FrameNet project is dedicated to producing
valency descriptions of frame-bearing lexical units (LUs)3, in both semantic
and syntactic terms, and it bases this work on attestations of word usage
taken from a very large digital corpus. The semantic descriptors of each
valency pattern are taken from frame-specific semantic role names (called
frame elements), and the syntactic terms are taken from a restricted set of
grammatical function names and a detailed set of phrase types. Sentences
extracted from the British National Corpus provide both the empirical evi-
dence for the analysis and an example database for both human and ma-
chine users. Sentences in the example database, chosen to illustrate each of
the lexical units we analyze, are annotated according to the LU’s semantic
and syntactic combinatory properties, and the valency patterns are auto-
matically derived from the annotations.
The treatment of valency in the FrameNet database differs from certain
other electronic lexical resources in several ways, by: (1) relying on corpus
evidence; (2) basing the semantic layer of valency on an understanding of
the cognitive frames that motivate and underlie the meanings of each lexi-
cal unit; (3) recognizing various kinds of discrepancy between units on the
semantic/functional level and patterns of syntactic form; and (4) providing
the means of assigning partial interpretations to valents that are conceptu-
ally present, but syntactically unexpressed.
After a brief introduction to the work of the FrameNet project and a
summary of the kinds of information it produces, the discussion will pro-
ceed to the special nature of building a frame-based dictionary and the
ways in which such a commitment leads the analyst to a “splitting” rather
than “lumping” approach to polysemy, ending with a survey of the discrep-
ancies between the frame structures evoked by particular LUs and the syn-
tactic structures which realize them.
130 Charles J. Fillmore

2. The project

FrameNet is a lexicon-building project, one of whose missions is to con-

struct valency descriptions for frame-bearing words in English – verbs,
nouns, and adjectives, as well as some adverbs and prepositions. The mean-
ing of frame and the scope of valency have determined the unique features
of the descriptions produced by this research.
The project is administered at the International Computer Science Insti-
tute in Berkeley, California, and is in its eighth year. The original aim was
purely lexicographic, but in recent years the project has taken on a number
of frame-based full-text semantic annotation assignments.4 The National
Science Foundation provided two three-year grants for the lexicographic
work5 and two sub-contracts for text-analysis;6 further support has been
provided by the DARPA7 and ARDA8 agencies of the U. S. government.
Two important methodological and theoretical commitments assumed
for FrameNet are the corpus9 and the frame (Fillmore and Atkins 1992,
1994; Fontenelle 2003). We use corpus evidence to derive information
about the combinatory possibilities of English lexical units, and to charac-
terize the manner in which phrases that are grammatically dependent on the
LU fill in details about the semantic frames which underlie each LU.
The process starts with extracting sample sentences from the corpus
containing the words being examined, determining which of these contain
instances of the LU under analysis, selecting sentences that show the varie-
ties of that LU’s combinatory properties in perspicuous ways, and annotat-
ing these with respect to their syntactic and frame-semantic properties.
The word frame10 in this context is used to refer to a schematic represen-
tation of speakers’ knowledge of the situations or states of affair that under-
lie the meanings of lexical items. The named components of a frame, called
frame elements (FEs), stand for the participants, props, phases, and parts of
the kinds of situations named by the frame.11 For very schematic frames,
such as those involving simple movement, the main FEs can be quite ab-
stract: Theme (an object seen as moving), Source (the starting point of a
movement), Goal (the endpoint or destination of a movement), and Path
(information relevant to the itinerary of the movement). In sentence (1)
below, the subject (the horse) expresses the Theme and the prepositional
phrase (out of the barn) the Source. For narrowly defined frames involving
complex scenarios, they can be quite specific. For example, in the Revenge
frame, we recognize Avenger (the individual who carries out an act of re-
venge), Offender (the individual whose prior act is to be punished through
an act on the part of the avenger’s), Injured_party (the individual who is
harmed or offended by the offender, who of course might be identical to the
 Valency issues in FrameNet 131

Avenger), Injury (the act or insult perpetrated on the injured party), and
Punishment (the act carried out on the offender by the avenger). In sentence
(2) the subject (I) expresses the Avenger, the prepositional phrase (at him)
is taken as indicating the Offender, and the prepositional gerund (for insult-
ing my sister) stands for the Injury and indirectly indicates the In-
jured_party.

(1) [The horse] bolted [out of the barn].12

(2) [I] got back [at him] [for insulting my sister].

The segmentation seen in these examples show an important feature of

FrameNet annotation. Since the purpose is to link the syntactic valents of
the governing LU with its semantic valents, the syntactic and morphologi-
cal markers of the relevant phrases are included in the sentence segmenta-
tions: thus, in the case of sentences (1−2), the words out of, at, and for are
included within the marked-off phrases.
With respect to the naming of frame elements, we learned early that for
many of the complex frames there is no non-arbitrary way of fitting them
into the “standard” sets of case roles or thematic roles in recent literature,13
except perhaps for the association of the familiar “Agent” with the primary
active participant in a scene, e.g., the Avenger in the case of the Revenge
frame.14
Using FrameNet terminology, a (frame-bearing15) lexical unit evokes a
frame and a valency description of a given lexical unit presents the set of
ways in which the syntactic accompaniments of the lexical unit introduce
information about the meaning elements of the evoked frame – or, stated
the other way around, the ways in which the semantic valents are expressed
in the sentence or phrase built around the frame-bearing unit. Since words
with different context requirements can evoke the same frame, the align-
ment of syntactic and semantic valents needs to be specified LU by LU.16
For example, the grammatical markings of the Offender for various expres-
sions in the Revenge vocabulary include simple direct object position in the
case of pay (someone) back, and a variety of prepositional markings, as
with retaliation against, get even with, wreak vengeance on, get back at.
Since there are kinds of text annotation that seek to connect all of the
information found in a sentence or text as tightly as possible, and since
FrameNet examples are not limited to simple finite sentences, the limits of
FrameNet annotation practice need to be made clear. In particular, for each
targeted LU FrameNet annotates, just those realizations of the frame’s FEs
which are in grammatical construction with the LU itself. That is, for lexi-
cographic purposes FrameNet records the amount of information about a
132 Charles J. Fillmore

frame that is provided in grammatically relevant positions within phrases

headed by the LU that evokes the frame.17
Suppose we were to select sentence (3) as illustrative of the use of the
verb send out:

(3) My attorney requested the files early last week, and [they] were
sent out the following morning.

FrameNet annotation would record for the token of send out in this sen-
tence that the pronoun they expresses one of its frame elements, as subject
of the passive form of the verb, and that the other two FEs (the Sender and
the Destination) are not locally identified within the grammatical structure
headed by that verb.18 This is distinct from the kind of annotation, built on
principles of discourse coherence, that would probabilistically recognize
that in the world of the text, my attorney stands for the intended recipient of
this sending act and the files are the things that got sent. Since the phrases
my attorney and the files hold no grammatical relation to the verb sent in
this sentence, they are not part of the FrameNet annotation for this token of
the verb in this sentence.19 In other words, given FrameNet’s lexicographic
purposes, FrameNet sentence annotations alone cannot be interpreted as
marking all of the participants in the situations evoked by the lexical units
analyzed; sometimes frame-relevant information is outside of the valency
range of the relevant lexical units.

3. The product

The database produced from FrameNet activities includes:20

1. a collection of informally characterized frame descriptions, including
the assignment of names and definitions to the frame elements;
2. the set of annotations for each lexical unit, where each sentence is anno-
tated with respect to one token of the LU whose function is illustrated in
it;21
3. lexical entries, which identify, for each LU, the frame itself, and the
variety of ways in which individual FEs are syntactically realized in the
corpus (including zero realization), together with the full patterns of FE
realizations found in individual sentences;22 and
4. a network of frame-to-frame relations, showing how some frames are
elaborations of others, how some frames are components of others, and
so on.23
 Valency issues in FrameNet 133

3.1. Frames

There are at present more than 700 frames recognized in the FN database.
Frame descriptions are formulated informally, using the frame element
names in definitional contexts as ways of indicating the frame elements.
Among the collection of frames is one called Arranging, the description of
which reads as follows: an Agent puts a complex Theme into a Configura-
tion, which can be a proper order, a correct or suitable sequence, or spatial
position.
The frame description mentions the core frame elements, Agent, Theme
and Configuration. Non-core, or peripheral frame elements are separately
described, and these include Manner, Location, Means, and several others.
The core vs. periphery distinction is analogous to, but not identical with,
the distinction in Tesnière (1959) between actants and circonstants. Core
elements are those which are necessary to the central meaning of the frame,
and peripheral elements provide aspects of the setting which can modify
any frame of the relevant type, i.e., act, state, happening, or the like. The
category core is not limited to obligatory elements, since we distinguish
two main functions of missing FEs, and these apply only to core FEs. The
distinction does not separate nuclear from “oblique” syntactic constituents,
since many core FEs are expressed adverbially or in prepositional phrases.
Core prepositional phrases can have their prepositions explicitly selected by
the head LU; peripheral prepositional phrases have forms determined by
their meaning, independently of the frame to which they are attached.
In order to have GF labels on all the phrases around a lexical governor,
FrameNet annotations provide a third kind of FE, called extrathematic, a
word or phrase which can be thought of as introducing a new frame, rather
than filling out the details of the frame evoked by the head. Comparing
sentences (4) and (5), we can see that each of them provides two layers of
information about the same event. The Letter_writing frame gives the con-
tent of the event, and the Revenge frame gives an evaluation or interpreta-
tion of the event in respect to a larger scheme. The brackets decorating
sentence (4) show core FEs in square brackets, peripheral in parentheses,
and the extrathematic in retaliation in wavy brackets. A sentence bearing
essentially the same information is sentence (5).

(4) [She] wrote [the letter] (yesterday) {in retaliation}.

(5) Yesterday she retaliated by writing a letter to my boss.

The fact that we cannot depend on a syntactic analysis that places such
“adjuncts” outside the scope of the verbal predicator is suggested by the
134 Charles J. Fillmore

VP-internal presence of a Beneficiary in certain sentences, e.g., in an ap-

parent direct object position in a double-NP construction. Comparing sen-
tences (6) and (7), note that the person indicated as my sister is a core par-
ticipant in the activity designated by the verb sell in (6), but is understood
as an intended participant in a secondary act in (7).

(6) [I] sold [my sister] [a harmonica].

(7) [I] bought {my sister} [a harmonica].

Recognizing my sister as the Beneficiary in (7) is acknowledging that this

sentence evokes a complex scenario involving two phases or sub-events:
the purchase of the harmonica and its subsequent presentation to the
speaker’s sister.

3.2. Annotations

As of this writing the FrameNet database has approximately 150,000 anno-

tation sets,24 each representing one valency possibility of a single target LU
as exhibited in the sentence. Though there are large differences in the rep-
resentativeness of attestations between highly frequent and infrequently
occurring members of a given frame, the general goal is to include a small
number of examples of each observed syntactic context. FrameNet does not
present relative frequency information for frames, lexical units, or valency
patterns, assuming that such information should be derived by procedures
sensitive to genre, register, regional variation, and the like.
For the verb arrange in the Arranging frame we find such examples as
those seen in figure 1.

[AgentHe] [targetARRANGED] [Themethe jharo] [Configurationin two pyramids]

[Locationat the edge of the roof, between the flowers].

[targetARRANGE] [Themerice] [Configurationin a ring] and place chicken in the centre.

[AgentCNI]

[ThemeThe waxworks] were [targetARRANGED] [Configurationin groups] [Locationbeyond

a rope, which was supposed to separate them from their admirers]. [AgentCNI]

Place a slice of avocado over each slice of grapefruit and [targetARRANGE]

[Manner attractively] [Configurationin lines down a serving plate]. [AgentCNI]
[ThemeCNI]

Figure 1. Arranging.arrange.v: tagging of phrases by their frame element name

 Valency issues in FrameNet 135

The annotations displayed in these samples identify only the constituent

boundaries and the frame element name. In the corresponding XML repre-
sentation the information characterizing the form of each syntactic valent is
expressed, in two separate layers, in terms of grammatical functions and
phrase types.
The principal GF names are External, Object, and Dependent: a distinc-
tion corresponding to that between arguments and adjuncts is not shown at
the level of grammatical function, but in terms of the core vs. non-core
distinction discussed above. Additional GFs are Appositive, Modifier,
Head, Genitive, and Quantifier, important especially for nouns.
External and Dependent need special explanations. The External GF
corresponds not only to the subject of a finite sentence, but also to the
phrases that stand for the subject function of non-finite verbs, e.g., the con-
trollers of subject roles in Raising and Equi constructions and subordinated
participial constructions, and to the primary arguments of frame-bearing
nouns and predicatively used adjectives. We do not use “Subject” because
in many cases the relevant constituent is not, in its own location, a subject
of anything, and because it does not seem natural to use the term as a GF of
a noun. Thus, with annotations centered on the verb release, as in sentences
(8) and (9), we categorize the general’s (the Genitive modifier of decision)
and the general (the Object of persuade) as the External argument of re-
lease in those contexts, and as being in grammatical construction with the
target verb.

(8) The [general’s] decision to release the prisoners was surprising.

(9) We persuaded [the general] to release the prisoners.

The decision to annotate these elements at all comes from the history of
FrameNet annotation practice: we do not work with parsed sentences,
within which controllers could in principle be recovered, and we have
wanted the annotations to yield collocational information about “subjects”
(here between general and the VP release the prisoners). Even a very large
corpus cannot always be relied on to present examples for each verb in
which a lexical subject (as opposed to a pronoun) is adjacent to a finite
verb.
The function name Dependent is a cover term for all other dependents of
an ordinary verbal predicate. Thus the “second object” in V+NP+NP pat-
terns and their transformations is referred to as Dependent, i.e., as an
“oblique nominal”25, one of the areas in which the FrameNet annotation
does not preserve theory-neutrality.
136 Charles J. Fillmore

The phrase types recognized in FrameNet are those that we believe are
relevant for the description of the lexically specified canonical syntactic
contexts of English frame-bearing words, corresponding more or less to
traditional subcategorization features. A near-exhaustive classification of
such contexts can be found in Atkins, Fillmore, and Johnson (2003:
277−279).

3.3. Lexical entries

The lexical entry database includes for each LU a reference to its frame, a
simple definition,26 indications of how each FE is grammatically repre-
sented (the complement inventories, in Thomas Herbst’s presentation), and
patterns of FE realization (the valency patterns) found in corpus-attested
sentences.27
Current valency pattern displays – automatically derived from the anno-
tations – are organized in a brute force way, with the FEs in alphabetical
order and the realizations given as pairs of grammatical function (GF) and
phrase type. Figure 2 is a fragment of the valency description provided for
the verb accuse, which is in the Judgment_communication frame, the frame
for a situation in which a Communicator (the person uttering a judgment)
speaks negatively of an Evaluee (the person whose behavior is being
judged) and offers (or is understood to have) a Reason for this judgment.
The active use is shown by the examples in which the Communicator has
the GF “External” (Subject) and the Evaluee has the GF Object; in the pas-
sive use the Evaluee is the subject and the Communicator is realized either
as CNI (unexpressed for constructional reasons) or as PP (by), i.e., with the
preposition by. The Reason is given with the preposition of followed by
either a NP or a gerund, or it is missing and given the DNI interpretation.

Total (54) Communicator Evaluee Reason

(2) CNI NP PP(of)
--- Ext Dep
(9) CNI NP PPing(of)
--- Ext Dep
(4) NP NP DNI
Ext Obj ---
(8) NP NP PP(of)
Ext Obj Dep
(19) NP NP PPing(of)
Ext Obj Dep
(1) PP(by) NP DNI
Dep Ext ---

Figure 2. Fragment of the valency description for the verb accuse

 Valency issues in FrameNet 137

The numbers in the left-most column indicate the number of sentences re-
ceiving the coding found on that line. The relevant part of a sentence for the
first line is Anyone could be arrested and accused of communist sympa-
thies. An example of the pattern shown in the fourth line is [In your edito-
rial] you accuse the Australian government of hypocrisy. An example of
the most common pattern, with of+gerund, is The Defence Minister Moshe
Arens accused the Fatah-affiliated Black Panther group of carrying out the
attack. The Reason is shown as DNI in the third and last lines, meaning that
it is understood as contextually “given”; an example from the annotations is
Now she was accusing me in front of a stranger.
In the case of accuse the alphabetical order of the FE names coincides
accidentally with the familiar ranking of GFs, but in many cases this is not
the case. Eventually the valencies will be shown in a normalized abstract
form from which all of the actual realizations can be predicted, and this will
require the identification of the nuclear GFs (subject in the case of verbs
and adjectives, subject and object in the case of transitive verbs, parallels of
these in the case of nouns derived from verbs or adjectives) together with
information about lexically licensed valent omissibility.
With frame-evoking nouns of a particular type we occasionally find that
a lexical unit that evokes a frame is itself (or is the lexical head of) the con-
stituent in the sentence which represents one of the frame’s FEs. This is
true, for example, of the ‘product’ interpretation of the noun replacement in
a sentence like We need a replacement for this part. Here the Replacement
frame is evoked by the noun, and the NP a replacement counts as one of
the FEs of that same frame (the other being for this part).

3.4. Frame-to-Frame relations

A system of frame-to-frame relations links semantically related frames to

each other in any of several ways. One frame can be seen as a sub-type of
another (inheritance relation), or as a component of another (subframe rela-
tion), or frames can be related as cause-effect (causative relation), or event-
state (inchoative relation). There is also what we call a perspectivizing rela-
tion, by which one frame is seen as taking a point of view on a more ab-
stract frame.
As an example (that isn’t as perfect as it should be, but will do for illus-
trating the point), the verbs buy, sell, pay and charge have uses in which
they give, or imply, information about the main individual transactions in a
commercial event. They differ, however, in how they highlight the human
participants’ relation to the goods or the money. From one point of view,
138 Charles J. Fillmore

then, they all are capable of indexing all of the participants in a well-
defined commercial event – and therefore can support inference operations
about which party ended up owning the money and which ended up owning
the goods – but buy and charge are instances of Taking, and sell and pay
are instances of Giving, from the perspective of one human participant, and
at the same time buy and sell are instances of Goods_transfer while pay and
charge are instances of Money_transfer.
Certain kinds of semantic inferences can be read off of frame descrip-
tions by way of frame-to-frame relations. For certain purposes, such infor-
mal descriptions appear to be satisfactory, since the descriptions can be the
same – or can be paraphrases of each other – for a great many frames
across languages. Sentences that report instances of the same frame, with
FE instances being translations of each other, can at least partially be ac-
cepted as translation of each other. For more technical purposes, e.g., infer-
encing, the mapping of frame descriptions to semantic simulations or to
logical expressions may be possible, but this is not included in the principal
activity of the FrameNet researchers.28
A fragment of a display of such links, centered in Placing, is seen in
figure 3.

Figure 3. A fragment of a display showing frame-to-frame relations among a set

of frames centered in Placing (i.e., putting something in a place). The
various relations include inheritance and using.

The interpretation of the different relations indicated by the arrows (colored

in various ways on the website) can be found on the FrameNet public web-
site.
 Valency issues in FrameNet 139

4. Frame-based vs. word-based progression

By working one frame at a time, rather than one word at a time from an
alphabetical word list, frame-based lexicography necessarily pays attention
to paraphrase relations and postpones thorough treatment of polysemy
structures. That is, when we look for all the expressions that can be called
on to talk about a particular well-defined situation type, we cannot simulta-
neously look into all of the other meanings of each of the words in our list.
Word-by-word lexicography, the traditional way to build dictionaries,
tends to include all the senses29 of each of the words that have been in-
cluded so far. When an ordinary dictionary project is mid-way in its work
there will be many words that have not yet been touched (possibly the pro-
ject hasn’t reached the letter “K” yet); a mid-stream frame-by-frame project
is likely to have covered paraphrase-relations between expressions in each
of the frames that have been included so far, but will not yet have all the
meanings of each of the words in its current word list.
While working through an analysis of a single frame we try to accumu-
late all of the words that evoke that frame. For example, in the case of Re-
venge, mentioned earlier, the list will include nouns like revenge, ven-
geance, retribution, reprisal and retaliation; verbs like avenge, revenge,
retaliate (against), get back (at), get even (with), and pay back; adjectives
like vengeful, retaliatory, and vindictive, and a large number of support
constructions – V+N examples like take revenge, exact retribution, wreak
vengeance, and P+N examples like in retaliation and in revenge. To say
that these LUs (these words in the intended meanings) are all members of
the Revenge frame is to say that, in spite of their grammatical and organiza-
tional differences, they all evoke, and require an understanding of, the full
Revenge scenario as described earlier. That is, independently of whether an
instance of the frame evoked by a particular LU is asserted, presupposed,
denied, or merely imagined, in that evoked scene an Avenger is meting out
Punishment to an Offender in return for an Injury to an Injured_party.
There are at least two reasons for examining all of the words in a frame
together, rather than describing one word at a time. First of all, frame-
sharing words often need to be delimited from each other in ways we dis-
cover by doing corpus analyses of all of them, in parallel; sometimes we
notice that what seemed like one frame at the beginning might in the end
need to be separated into two or more. Secondly, since semantically related
words, of the same part of speech, often have similar syntactic behavior –
in spite of the variety just noticed for marking the Offender in the Revenge
frame – the search parameters for finding example sentences in the corpus
are largely reusable across a frame.
140 Charles J. Fillmore

When the project is complete, the FrameNet lexicon will be of use to

researchers working on broad-domain automatic sense selection, but at
intermediate stages we are not in position to document all senses of each
word. It is easy to see that having both goals simultaneously – exploring all
the words in a frame and all the senses of each word – would have explo-
sive ramifications. Starting with a word, you determine the different frames
it belongs to – its different senses; taking each of these frames, you look for
the different words that serve it; taking each of these words, you determine
their frames; and so on.
To support this point with a concrete example, let’s assume that our
current target of analysis is the verb depend and that we have begun by
describing its meaning in a sentence like (10). Calling this the Reliance
sense, we examine samples of the word in context and notice that not all
instances of the word in the corpus fit this frame. We then see the need to
identify a Contingency sense as seen in example (11), where a connection
is predicted between the weather and the success of a festival.

(10) You can depend on me to get the job done.

(11) The success of the festival will depend on the weather.

The Reliance frame involves humans trusting other humans to do what is

expected of them. The Contingency frame poses a causing or enabling rela-
tion between occurrences. In order to pin down the details of the frame
structures needed for each of these frames, we have to find out what depend
has in common with the other words in each of these frames. We quickly
realize that in the Reliance frame we also have the verbs rely and count,
with essentially the same syntax and meaning, as seen in sentences (12) and
(13).

(12) You can rely on me to finish it in time.

(13) I hope we can count on the boys to get home before dark.

Neither of these new additions to the Reliance frame is usable in the Con-
tingency frame, as shown in the contrasts between acceptable and unac-
ceptable versions of (14) and (15). (In [14] the thing depended on is unex-
pressed, i.e., is left implicit.)

(14) That depends. (*That counts.30 *That relies.)

(15) The salary level will depend partly on the age of the applicant.
(*count, *rely)
 Valency issues in FrameNet 141

We have made some progress towards including different senses of depend,

but if we felt, in order to include count, we had to cover the rest of its
meanings, we soon find what we might call a Validity or Importance
sense31 (your vote doesn’t count), a Categorization sense (I count you as a
friend), a Counting_out sense (my grandson can already count to twenty),
and seven or eight others among the verbs, in addition to what we find with
nouns sharing that form, including Nobility_rank (the count and countess),
and a number of others in sports and weaving. In exploring each of these,
we would need to deal with other words in those frames: the Validity frame
will have the adjective valid; the Nobility_rank frame will have duke,
prince, earl; and so on. The Categorization frame also contains regard,
consider, deem, categorize, and many others, plus their nominal deriva-
tives.
We at FrameNet want to explore all the words in a frame and to build up
paraphrase relations. While this may be disappointing to those of our col-
leagues who expect to find all senses of each of the words that we touch, no
rational plan of research is compatible with such conflicting goals, short of
bringing the work to completion.

5. Polysemy and FrameNet

As noted, while traditional lexicographic practice leads immediately to an

exploration of polysemy and provides data allowing the discovery of pat-
terned polysemy structures, FrameNet’s methods lead naturally to an explo-
ration of paraphrase (and near-paraphrase) relations, showing how identical
or similar semantic structures can be syntactically expressed in a variety of
ways.
This does not mean that FrameNet researchers can ignore polysemy.
There are both empirical and theoretical reasons for a concern with multiple
meanings. The empirical issue is that we obviously have to choose the par-
ticular sense of a lemma that is going to define a particular LU, and this
requires an initial survey of the most important of the word’s other senses.
Although it might not be possible to determine full descriptions of other
meanings of a word from the start, we obviously need to delimit one par-
ticular LU from others that use the same form. The theoretical issue has to
do with competing possible explanations of the same phenomena.
In the matter of deciding how many separate senses a word has, Frame-
Net is more closely aligned to the splitters than the lumpers. Monosemists32
are semanticists who wish to cover word senses as generally or abstractly
as possible. While a concern with maximal generalization may be attractive
142 Charles J. Fillmore

to the theoretician, it does not always satisfy the criteria according to which
we need to decide which meanings belong to which frames. In FrameNet,
we seek to maximize the regularity of LU-to-LU relations (antonymy, syn-
onymy, etc.),33 LU to valency patterns, LU to various grammatical patterns
(tense, aspect, etc., in the case of verbs, determination and countability
possibilities in the case of nouns, gradability in the case of adjectives and
adverbs, etc.), inference possibilities associated with one sense rather than
another, ambiguities not associated with complementation types, and mor-
phological relations between words of different parts of speech. (To illus-
trate the last point, the nominalizations of two verbal LUs might have the
same form: deduce > deduction vs. deduct > deduction; two verbal LUs
with the same form might have different nominalizations: observe > obser-
vance vs. observe > observation. Arguments for the monosemy of deduc-
tion or observe will not be convincing.)
When asked for a synonym of depend, we need to know which frame it
is in: rely fits in one frame but not in the other. When asked for a nominali-
zation of observe, we need to ask which frame our inquirer has in mind:
observation works in one frame, observance in the other. When asked
about the valency of remember, it will depend on which sense of the verb is
intended.
The case of remember is instructive. Monosemists are likely to find one
meaning to this word, and to assign it a fairly large set of valency patterns.
Remember has many different complement types, and it might be tempting
to treat the differences in interpretation as explainable merely in terms of
semantic accommodations to its various complement types. This verb can
occur without an object, with a NP object, with a finite clause complement,
with an infinitive complement, or with a verbal or clausal gerundial com-
plement, and the verb can function as either a stative verb or an active verb.
In studying the language appropriate for describing episodic memory,
we can create a frame called Remembering_experience. In this frame the
verb remember has recall and recollect as its partners, as well as phrasal
expressions like have memories of. Complements that participate in this
meaning include VP gerunds or V-ing complements (16), sentential ger-
unds or NP V-ing complements (17), and simple NP (18).

(16) I remember as a child [falling down the steps in my grandma’s

house].
(17) I remember [something touching me in the dark].
(18) I can still vividly remember [the accident].
 Valency issues in FrameNet 143

Another use of this verb has to do with Retaining_information, a sense in

which it can be linked, through Frame-to-Frame relations, to forget (losing
information) and know (having information). The complements of remem-
ber in this frame can be that-clauses (19) interrogative clauses (20), infini-
tival interrogatives (21), and NPs (22). As in the remember of Remem-
ber_experience, this LU can be used statively, representing an ability.

(19) Everybody else remembered [that the meeting was scheduled for
today].
(20) Do you remember [what she said]?
(21) They don’t remember [what to do next].
(22) Do you remember [my phone number]?

Yet another frame is needed for Remembering_to_do_something. This

occurs with infinitive VPs (23) and simple NPs (24).

(23) Did you remember [to feed the cat]?

(24) Did you remember [your umbrella]?

This use of remember is paired with one sense of forget having the same
presupposition about intending or resolving to do something, but has no
semantic relation with know.
One syntactic complement type found in all three frames is the simple
direct object. If all three putative senses can take NP objects, can we still
find reasons for separating them in this context? The combination of differ-
ences in the kinds of NPs and aspect differences in the verb’s main uses
sorts these out. For episodic memory, a sentence like (25); for Remember-
ing_to_do, something like (26); e.g., in the meaning ‘did you remember to
bring your umbrella’; and for retaining information a concealed question
NP as in (27).

(25) Do you remember [the umbrella]? (The one we gave you when
you were seven, the one with the Mickey Mouse design?)
(26) Did you remember [the umbrella]? (Or did you leave it at home?)
(27) Do you remember [my name]? (Do you remember what my name
is?)

The separation of remember into different frames, then, permits association

with synonyms: in one sense it goes with recall and recollect, with a vari-
ety of antonyms (senses 2 and 3 go with forget), associations with infer-
ences, with know, with aspectual possibilities of the verb (senses 1 and 2
144 Charles J. Fillmore

can be stative), etc., and with different ways of interpreting NPs. Impor-
tantly for present purposes, the three senses already identified – there are
several more – have only partially overlapping valencies.
The theoretical issue of polysemy for FrameNet has to do with the ex-
tent to which frame-to-frame relations can make it possible to recognize
both commonalities and differences between closely related frames. In the
course of the project’s treatment of what Fillmore and Atkins (1992) pro-
posed as the Risk frame, what was regarded as a single frame at the begin-
ning eventually split into three frames, all of them using a super-frame that
recognized “risky” situations in the abstract. The polysemy issue here in-
volves the verb risk. In one instance of risk, the verb has what can be called
an Asset as its direct object: one risks one’s life, one’s fortune, one’s
health, and the like. Another has some unfortunate consequence (Mishap)
as its complement: either a nominal object, as in one risks failure, infection,
etc., or a verbal gerund as in one risks losing one’s job, falling off the cliff,
or a sentential gerund, as in one risks everyone getting angry. The third
sense has an Action as its nominal or gerundial complement: I wouldn’t
risk a trip into the jungle at this time, I wouldn’t risk swimming in the dark.
The original analysis simply had Asset, Mishap and Action as different
frame elements in the same frame, allowing any of them to be the primary
verbal complement. The resulting three frames – called Jeopardizing, Incur-
ring and Daring – allow a more systematic treatment of the data, a tighter
recognition of synonymy relations and more consistent valency descrip-
tions.

6. Discrepant links between semantic roles and syntactic constituents

In the simplest case, a linking of syntactic and semantic valents will have a
governor, the valency-bearing word, and a set of dependents; each depend-
ent in a given sentence will be assigned a syntactic status and a semantic
role. A single lexical unit in the same frame might have more than one syn-
tactic valency pattern, but for each one, in this starting idealization, a sim-
ple matching holds. Using everyone’s first example of valency, we can
assign Giver, Gift and Receiver to the three arguments of the verb give,
mapping the Giver to the subject, the Gift to the direct object, and the Re-
ceiver to an oblique, marking it with the preposition to. Because of the so-
called Dative Alternation and Passivization, of course, we can describe
three other patterns involving different syntactic assignments to each of the
semantic roles.
 Valency issues in FrameNet 145

Limiting ourselves to the versions without passive ellipsis, a schema for

the one-to-one linking of units of grammatical form (G) with the compo-
nents of meaning (M) can look like this: {{Ga:Ma}, {Gb:Mb},
{Gc:Mc}}.34 The combined syntactic/semantic valency is a set of linked
relations pairing units of grammatical form with units of meaning (the
frame elements), the linking suggested by the lowercase letters: the first
valent of give in the primary valency mentioned above is Subject:Giver,
where Subject is Ga and Giver is Ma.
There are numerous ways in which observed valency patterns differ
from this simplest case:
1. There can be syntactic valents for which there is no corresponding se-
mantic valent, schematically {{Ga:Mzero}, {Gb:Mb}, …}.
2. There can be semantic valents for which there is no corresponding syn-
tactic valent – i.e., there can be semantic roles that are understood but
not expressed, schematically {{Gzero:Ma], {Gb:Mb}, …}.
3. There can be single syntactic valents that incorporate more than one
semantic valent {{Ga:Ma&Mb}, {Gc:Mc}, …}.
4. There can be single semantic valents whose realizations are distributed
over more than one syntactic constituents, schematically
{{Ga&Gb:Mab}, {Gc:Mc}, …}.
5. While the preceding examples show variations in the pairing of syntac-
tic and semantic valents, other cases involve the frame-bearing words
themselves. There can be simple-word valents of a verb (typically the
particles of English particle-verbs, for which it makes most sense to
claim that the verb and its particle jointly bear the verb’s meaning).
6. And there are examples for which a frame-bearing noun participates as
the syntactic dependent of a verb or preposition, where the verb contrib-
utes a subordinate semantic content, as with the support constructions.
The linkings examined in this section fall into a class of phenomena re-
ferred to by Francis and Michaelis (2003: 1−27) as mismatches. In our case
the discrepancies are departures from the pattern that syntactic valents are
expressors of semantic valents and that individual semantic valents are
expressed by individual syntactic valents.

6.1. Syntactic valents with no assigned semantic role

{{Ga:Mzero}, {Gb:Mb}, …}

The semantically empty word there can occur in sentence formats express-
ing existence, and occurrence, as in sentences (28−30):
146 Charles J. Fillmore

(28) There suddenly appeared a great host of angels. (Presentative)

(29) There is some beer in the fridge. (Locative/Existential)
(30) There was a loud explosion. (Support construction for event
nouns)

The epenthetic pronoun it can serve as syntactic place-holder in the clause

types known as it-extraposition, as seen in sentences (31−35):

(31) I regard it as obvious that ...

(32) It is well-known that …
(33) It is impossible to read this small print.
(34) See to it that the kids are home before dark.35
(35) You can depend on it that we’ll be home before dark.

In the extraposition case, the two constituents (it and the re-located clause)
occupy well-defined syntactic positions relative to the governor.

6.2. Semantic valents with no syntactic expressors

{{Gzero:Ma], {Gb:Mb}, …}

Ignoring grammatical constructions of ellipsis, gapping and various kinds

of coordination reduction, all of which do not target the dependents of spe-
cific governing words, we can describe two main kinds of lexically licensed
ways of omitting valents in English. FrameNet annotations identify three
kinds of valent omissions, but one of these is not treated as a part of spe-
cifically lexical descriptions. These are indicated with the symbols CNI,
INI and DNI, paired with the name of the frame element that is not realized
in the sentence.36 The labels stand for: constructional null instantiation,
indefinite null instantiation, and definite null instantiation.37 Each of these
will be treated in turn in the next three subsections.

6.2.1. CNI: Constructionally licensed valent omission

In the CNI cases the valent omissibility is not relevant to the description of
individual lexical items, beyond whatever limits there may be in the condi-
tions that limit lexical items to participating in the relevant constructions.38
Examples are the missing subjects of imperative sentences (38), the missing
“agents” of reduced-passive sentences (39), and the generic or “free” sub-
ject of locally interpreted infinitives or gerunds (40−41). CNI marking is
 Valency issues in FrameNet 147

included in the annotated sentences, if only to show how all valents are
accounted for in the annotations. The omission of arguments in various
sub-genres (recipes, instruction manuals, newspaper headlines, diaries,
telegraphese, etc.) is also generally not determinative of lexically interest-
ing principles (42).

(38) Please leave the room. (Imperative)

(39) The work was finished years ago. (Passive)
(40) To expect more would have been foolish. (“Free” a.k.a. PRO)
(41) Sleeping in the park overnight is against the law.
(“Free” a.k.a. PRO)
(42) Mix well and cook until done. (Instructional imperative)39

All such absences are registered in FrameNet sentence annotations, to make

it possible to see how all of the valents are accounted for, but are not in-
cluded in information about the lexical units. In (38) the ‘second person’
indicator is missing; in (39) the passive’s agent is unexpressed; in (40) and
(41) a generically interpreted subject is unexpressed; and in (42) the im-
plicit object of the verbs mix and cook is understood from a previous in-
struction.40
The other two Null Instantiation types are necessary components of the
descriptions of individual lexical units.

6.2.2. INI: Indefinite null instantiation

Indefinite null instantiation has something in common with canonical In-

definite NPs, in that they both represent the introduction of something into
the discourse and invite the understanding that the speaker is (or ought to
be) prepared to say more about that. Thus, the speaker of (43) can be ex-
pected to say what his or her question is, perhaps in the next utterance.
Similarly, with unspoken indefinite arguments, as with (44) or (45), the
speaker can be expected to say something, on inquiry, about what has been
being baked or what something depends on.

(43) I have a question. (indefinite NP)

(44) I’ve been baking all day. (missing object)
(45) That depends. (missing PP complement)41

FrameNet records only those missing elements that belong to the core type,
since peripheral elements are always optional. One can assume that infor-
148 Charles J. Fillmore

mation about such circumstantial elements as Place, Time and Manner are
open for inquiry at any time and so specific annotations indicating their
availability for further elaboration in the ongoing discourse does not need
to be included in sentence annotations.

6.2.3. DNI: Definite null instantiation, lexically licensed zero anaphora

Definite null instantiation has something in common with Definite NPs, in

that it points back to something available in the interlocutors’ context of
discourse. Thus, in (46), the NP the problem refers to something already
under discussion. DNI omissions, with similar contextual resolutions, can
be found with adjectives (47), adverbs (48), nouns (49) and verbs (50).

(46) I have a solution to the problem. (definite NP)

(47) My answer was similar. (similar to something given in the con-
text)
(48) That happened a year later. (a year later than something just dis-
cussed)
(49) Were there any witnesses? (witnesses to an incident just identi-
fied)
(50) When are they likely to arrive? (arrive at the place we all have in
mind)

Note that the link from DNI-marked valents to entities mentioned in previ-
ous discourse is similar to what is sometimes referred to as indirect or
bridging anaphora. In the DNI case, it is specific lexically-identified infor-
mation about unrealized arguments that invites the search for antecedents,
and not real-world assumptions about what is likely to accompany particu-
lar referents in a text world.42 Most discussions of bridging anaphora are
concerned only with NPs, and in fact only with definite NPs. As the above
examples show, the DNI phenomenon is not limited to such conditions; the
unexpressed argument is what is construed as definite.

6.3. Syntactic valents incorporating more than one semantic valent

{{Ga:Ma&Mb}, {Gc:Mc}, …}

We sometimes find situations in which a valency alternation allows refer-

ence to two entities to be expressed as a single syntactic valent in one case
and as two syntactic valents in the other case.
 Valency issues in FrameNet 149

With verbs of medical attention like cure, heal and treat, the situation
calls for a Patient and a Disorder, and these can be represented jointly, as in
(51), or disjointly, as in (52). The Disorder is implicit, as DNI, in (53). The
generically interpreted sentences of (54) can be thought of as having the
Patient as INI, the Disorder as incorporated, or the Disorder as Modifier.

(51) The medicine cured [my asthma]. (Patient as “possessor” of Dis-

order)
(52) The medicine cured [me] [of asthma].43 (Patient and Disorder are
separated)
(53) The medicine cured [me]. (Disorder is DNI-omitted.)
(54) The medicine cures [asthma].
(54’) The medicine cures [asthmatics].
(54’’) The medicine cures [asthma patients].

In the case of reciprocal or symmetric predicates, plural NPs can appear in

subject (or object) position, representing the entities involved as a single
syntactic constituent, or the two roles can be distributed over a subject (or
object) and an oblique constituent. A number of examples are given below.

(55) [John and his brother] are quite different.

(55’) [John] is quite different [from his brother].
(56) [Figures A and B] are similar.
(56’) [Figure A] is similar [to figure B].
(57) I find it difficult to distinguish [your sons] [from each other].
(57’) I find it difficult to distinguish [John] [from his brother].

Certain kinds of relational nouns, like price, population, etc., allow the
relatum to be expressed in the NP headed by the relational noun, or allow
the attribute to be identified in a separate oblique PP. This is clear in Frame-
Net’s Change_position_on_a_scale frame, as seen in (58−59).

(58) [The price of oil] is rising.

(58’) [Oil] is rising [in price].
(59) [The population of such communities] is increasing.
(59’) [Such communities] are increasing [in population].

Frames dealing with body contacts allow such separate treatments between
body-part names and their possessors, as in (60−61).
150 Charles J. Fillmore

(60) She pinched [my nose].

(60’) She pinched [me] [in the nose].
(61) He slapped [my face].
(61’) He slapped [me] [in the face].

Another involves the relation between a role and the occupant of a role, as
in (62).

(62) We elected [Harry] [as president]. (occupant and role as separate

constituents)
(62’) We elected [a new president]. (individual and role simultaneously
represented)

6.4. Semantic valents distributed over more than one syntactic valent
{Ga&Gb:Mab}, {Gc:Mc}, …}

This description covers two situations, one of which does, and one of which
does not, involve syntactically describable lexically headed constructions.
There are instances of discontinuities which could be thought of as in-
terruptions of phrasal or clausal complements of verbs of speaking or think-
ing. FrameNet annotates these as discontinuous complements of the
speech/thought verb, but there is nothing semantically or syntactically de-
terminate about the pieces of the discontinuous structure. The point of sepa-
ration is determined more by the structure of the complement clause than
by any properties of the governing verb (see sentences [63−64]). FrameNet
annotation uses the pseudo-phrase-type “Quotation” to represent the con-
stituent as a whole, and assigns that label to each of piece of the discontinu-
ity, whether or not the two parts can be given determinate phrase-type
names.44 The lexical heads tend to be – or to be construed as – verbs of
speaking or thinking (say, ask, think, suppose, sigh, huff, etc.).

(63) “But why,” she asked, “did you do that?”

(64) The main issues are, I think, beyond resolution at this point.

The phenomena of preposition-stranding and prepositional passives are

further instances of contexts in which complements can be discontinuous. It
is not an interesting property of either put or shout, beyond the reality that
these verbs can take prepositional complements, that accounts for the dis-
continuities in (65) or (66).
 Valency issues in FrameNet 151

(65) [Which box] did you put it [in]?

(66) [I] don’t like being shouted [at]?

In other words, the ability to participate in such structures would not be

separately described as significant parts of the valency of these verbs.
Secondary predication and raising are among the more lexically specific
grammatical processes that create situations of distributed syntactic realiza-
tion of single semantic role notions. Thus, all of the examples (67−70) have
as their semantic complement Message, Proposition, or Content, depending
on the frame, in which ‘being my friend’ is predicated of ‘Harry’, and the
FE label is associated with both pieces.

(67) [Harry] appears [to be my friend].

(68) I consider [Harry] [my friend].
(69) I regard [Harry] [as my friend].
(70) I want [Harry] [to be my friend].

The complement of appear, consider, regard and want is expressed as the

stretch ‘Harry … my friend’ (give or take the “marking”) in all four cases.
The passivizability of the NP Harry in both cases suggests the correctness
of assigning it a syntactic role on its own.45

6.5. Syntactic valents that are parts of multiwords

Syntactically viewed, the particles in English particle verbs are their de-
pendents, but it is often useful to regard the combination of the verb and the
particle as defining a single lexical unit. Alternatively it is possible to re-
gard the verb itself as the frame-bearing element and specify that the parti-
cle is a semantically empty obligatory syntactic valent. These two views
could be represented in the same way.
This will be the case for put off meaning ‘postpone’, carry on meaning
‘continue’, check out meaning ‘examine’, call off meaning ‘cancel’, and
many hundreds of others. FrameNet does not give the same treatment to
subcategorized prepositions, since these can be taken as markers of the
relevant frame element. Thus for put up [with] meaning ‘tolerate’ the lexi-
cal unit itself is expressed as put up, and with is the selected marker of its
complement.
152 Charles J. Fillmore

6.6. Syntactic dependents that are semantic heads

While both nouns and verbs have valencies, in some cases, a noun that is a
syntactic valent of a verb is itself the principal or only frame evoker, and
the verb which syntactically governs it makes no (or little) contribution to
the semantics of the clause, serving mainly to provide, within its own syn-
tactic valence, frame elements in the noun’s frame, while adding tense and
aspect information: these are the support verb constructions. Many but by
no means all of these are V+N paraphrases of morphologically related
verbs, in the way that have a fight pairs with fight, make a choice with
choose, say a prayer with pray, give advice with advise, and take a bath
with bathe. The concept is not limited to deverbal nouns but also includes
cases like wreak havoc and wage war.
Support verbs have many functions in addition to allowing a noun-
evoked frame to be expressed in a verb-headed context. In some cases dif-
ferent support verbs for the same noun introduce different perspectives on a
single event type: perform vs. undergo an operation, inflict vs. sustain an
injury, pay attention vs. draw attention, etc. In some cases a manner or
“setting” component varies with different support verbs for the same noun:
make vs. lodge, file or register a complaint, for example.
There are more complex lexical functions, in the sense of Mel’cuk
(1998), whose verbs share arguments with the noun’s frame but identify
subevents in a larger frame. Giving and getting advice express participating
in an advising event; taking someone’s advice presupposes participating in
such an event but adds the concept of uptake; making a promise is a prom-
ising event, but keeping and breaking a promise are separate acts, on the
part of the promiser, having to do with acting on the promise; giving and
taking a test are perspectives on an examining event, but passing or failing
a test are separate events, affecting the examinee, related to the “licensing”
function of the test.
There are also prepositional supports, i.e., cases in which a preposition
governing a frame-bearing noun creates a structure which functions as a
predicate adjective (in danger, at risk, on fire, etc.) or as a verb-modifying
adverb (under pressure, in retaliation). In such cases the primary valent of
the frame is controlled by the surrounding syntax (subject of be as in be
under pressure, object of find as found him in trouble, subject of an inde-
pendent verb as in acted under pressure).
 Valency issues in FrameNet 153

7. Summary and moving forward

In a frame-based lexicon words whose meaning descriptions require an

appeal to a common underlying conceptual structure are described together.
Somewhat analogous to word-grouping by frames is the practice in most
dictionary projects of defining a standard way of describing words that
differ from each other along clearly statable parameters. This includes color
words, compass points, kinship terms, weekday names, weights and mea-
surements, and perhaps a few dozen others. In the case of color words, for
example, the style manual will likely dictate that the definitions are to be
made with reference to (a) the color spectrum, in terms of neighbors within
the spectrum, (b) the reader’s knowledge of familiar colored objects, such
as grass, blood, the cloudless daytime sky, etc., or (c) wavelength range, or
some line-up of more than one of these kinds of information.
The difference is that for a frame-based lexicon, the grouping practice
extends more or less to all words. Belonging to a semantic frame and hav-
ing a valency are not identical. Color words have modifiers (tending, in
English, to be adjectives – light blue, dark green, pale yellow, etc.) but they
lack the kinds of FEs associated with relations, events, complex states of
affairs, etc. Kinship terms have of course the terminals of the kinship rela-
tions (and in some cases an intermediary – paternal uncle), compass points
used for indicating directions necessarily imply a starting point, and
weights and measurements are the weights and measurements of some-
thing, and they are expressed in terms of types of units.
For words that don’t really have valents in the usual sense, FrameNet
chooses to treat categories of modification, including the qualia in the work
of Pustejovsky (1995), whereby we can label the modifiers of simple
nouns, for example, in terms of substance, dimensions, orientation, func-
tion, and a whole host of others. This has not been satisfying, in the way
that modification in general provides some problems for semantic analysis.
For a phrase like angry child, the adjective identifies a disposition of the
child, but at the same time the noun stands as a semantic valent of the ad-
jective.
It was pointed out above that semantic analysis in the FrameNet style
begins by characterizing the kind of schema or event that motivates our
understanding of the word’s meaning. Once one has one such frame in
mind, the question quickly turns around, and the analyst begins to ask what
are the various means provided in this language for talking about the rela-
tions, processes and individuals within such a frame, and the kind of in-
quiry becomes onomasiological, or of the encoding type, rather than sema-
siological, or of the decoding type. This is part of what predisposes Frame-
154 Charles J. Fillmore

Net researchers to assume a sense-splitting strategy, since a common word

that has a special use within a particular frame is likely to seem like a sepa-
rate sense of that word.
Much work in the semantic analysis of English modals has treated the
monosemy/polysemy issue; there are persuasive arguments that the im-
agined differences of meaning of individual modals have more to do with
the interaction of their uniform senses with differing contexts or domains of
application (Sweetser 1990: 49−75). It’s not likely that FrameNet research-
ers would even find a point in their research in which that question comes
up. If the frame being worked on has to do with probability estimation, then
the modal may of He may retire early is assigned to this frame (in a context
where the intended meaning is clear); and if another frame is one of per-
mission-granting, the may of You may leave the room now is assigned to
this frame. Here the intellectual process by which a monosemist would seek
to construct a single overarching sense to cover each of these uses has been
foreshortened. The valence of the probability sense would undoubtedly be
the (disjoint) predication that included both the subject and the VP com-
plement of may; the valence of the permission sense would likely separate
the subject (as the person receiving permission) from the VP complement
(as the permitted action). I should point out that FrameNet has not yet tack-
led the modals.
One property of a valency description which FrameNet has not managed
to provide directly is an account of the typical semantic types of the phrases
that serve as frame elements. It is hoped that later research based on further
corpus evidence can spot the semantic types found for particular FEs of
particular LUs and incorporate such results in the valency descriptions –
beyond such limited high-level indications as animate, concrete, and ab-
stract. Various efforts to find nodes in WordNet to cover such categories as
criminal acts (murder, arson, treason) for predicting the content of expres-
sions occurring in contexts like charged with ___, accused of ___, guilty of
___, or commit ___, have been made, but they have not been impressively
successful (Mohit and Narayanan 2003). Furthermore, there are many cases
where particular FEs do not themselves have distinct semantic types, but
pairs of FEs might be expected to have similar semantic types. Thus, for
reciprocal structures, involving LUs like similar, replace, combine, etc., it
is the pairing of similar semantic types that is typical. In “A can replace B”,
the A and B can both be holiday destinations, recipe ingredients, football
players, words, etc., but there is simply no semantic type unique to either
one of those positions.
 Valency issues in FrameNet 155

Notes

1. The author is grateful to Miriam Petruck and Josef Ruppenhofer for their ears
and their red pencils.
2. In all other writings on FrameNet or by FrameNet members, the word is given
its American form valence. Since I want to use the noun valent to cover both
semantic role notions (semantic valent) and syntactic dependents (syntactic
valent), I resort to the form valency to avoid homophony between valence and
valents.
3. Following Cruse (1986: 77) we use lexical unit to refer to a pairing of a word
with a sense: in our case, it is the pairing of a word with the semantic frame to
which it belongs. Thus, the ties in (a) tie a knot, (b) research has tied cancer
to smoking and (c) she tied the pony to a hitching post are three different lexi-
cal units.
4. In standard FrameNet annotations, each sentence is annotated with respect to
the single LU whose use is illustrated in it; in full-text annotation, the sentence
is annotated – in multiple layers – with respect to each frame-bearing word
found in it.
5. IRI-9618838, March 1997 - February 2000, “Tools for lexicon-building”; then
under grant ITR/HCI-0086132, September 2000 - August 2003, entitled “Frame-
Net++: An On-Line Lexical Semantic Resource and its Application to Speech
and Language Technology”; with a smaller but much appreciated supplement
in 2004.
6. A subcontract from grant IIS-0325646 (Dan Jurafsky, PI) entitled “Domain-
Independent Semantic Interpretation”.
7. Defense Advanced Research Projects Agency, http://www.darpa.mil/.
DARPA, FA8750-04-2-0026, “Steps Toward the Alignment of Complemen-
tary Lexical Resources and Knowledge Databases”.
8. Advanced Research and Development Activity, http://www.ic-arda.org/. Since
Sept, 2004, we have been a subcontractor on an ARDA contract for work on
question answering through the University of Texas at Dallas, as part of the
AQUAINT project; we are annotating AQUAINT texts, questions, and an-
swers with frames and FEs, testing whether this will improve the QA results.
Both the PropBank and the AQUAINT annotated texts are browsable on the
FN public website.
9. The corpus used for most of the time of the project has been the British Na-
tional Corpus (BNC). In recent years the BNC has been supplemented by
samples of US newswire text provided through the Linguistic Data Consor-
tium of the University of Pennsylvania (http://www.ldc.upenn.edu/), selected
texts provided by funders, and some material from the WorldWideWeb.
10. In the author’s personal history the concept “frame” evolved from an earlier
use as “case frame”, as readers who are over sixty may remember. The con-
cept has obvious similarities to structures covered by terms like frame,
schema, scenario, script, etc., common in educational, artificial intelligence
and cognitive psychology research in past decades, but in our case it is limited
156 Charles J. Fillmore

to such structures as they are keyed to – and “evoked by” – specific linguistic
objects, i.e., words and grammatical patterns. See discussion in Fillmore
(1985).
11. The FEs are to be taken as role names, not names of entities. As with the Re-
venge frame described below, the Avenger in a Revenge situation might be the
same as or different from the victim of the original Injury, and may be a group
of people rather than a single individual.
12. Although the FrameNet database is faithful to its corpus commitment, in that
all annotated examples are attested in the corpus, the examples offered in this
paper are simplified or invented.
13. Such lists are likely to contain agent, instrument, patient, theme, experiencer,
source, goal and path among others. For examples see Fillmore (2003).
14. For various technical reasons we require that the FEs be seen as distinct for
each frame, but rather than invent ever newer FE names for each new frame,
we can satisfy this requirement by using dotted names that combine the frame
name with the FE name: thus Placement.Theme is distinct from Arriv-
ing.Theme.
15. The qualifier “frame-bearing” is to distinguish the words that are directly
targeted for FrameNet analysis from those that have mainly functions con-
trolled by the grammatical system (tense, aspect, support verbs, etc.) or are de-
termined by the subcategorization requirements of governing words, such as
highly selected prepositions and particles in the case of phrasal verbs. Such
words – together with vast numbers of names of artifacts, species, chemical
compounds, as well as persons, places and institutions – do not receive treat-
ment in the FrameNet database.
16. The groupings by which semantic and syntactic structures show parallel be-
havior, as in the important work of B. Levin (1985), do not always match
frame-relationships of the kind developed in FrameNet.
17. This isn’t quite true, but the places where we reach beyond the LU-headed
phrase are precisely those places where familiar syntactic theories provide
construals for “empty categories”: the antecedents of gaps for WH-extracted
elements, missing subjects in non-initial conjoined VPs, and the controllers of
the subjects of non-finite VPs.
18. As will become clear below, the Sender is in fact recognized in the annotation,
as unrealized in a way licensed by the passive construction, and the Recipient
is indirectly expressed by the particle out.
19. It is for this reason, of course, that sentence (3) would not be valued as an
illustration for the use of the verb send, and is not likely to be a valued entry in
the FrameNet annotation collection. FrameNet does not seek to annotate the
most frequent uses of a word. For a great many transitive verbs in English, the
most frequent occurrences have pronouns in the nuclear syntactic positions;
annotators are instructed to reject these in favor of examples whose compo-
nents are semantically relevant to the nature of the frame, without needing to
appeal to discourse context.
20. Details can be seen in the FrameNet public website:
 Valency issues in FrameNet 157

http://framenet.icsi.berkeley.edu/.
21. In the full-text annotation work mentioned earlier, all words in a sentence are
annotated. This would correspond to a full annotation set for each frame-
relevant word in the sentence.
22. This distinction corresponds to the distinction brought up by Thomas Herbst
during the conference as that between complement inventory and valency pat-
tern.
23. The structure of the system of frame-to-frame relations is set up, but the de-
tails have not been completed as of this writing. Many FrameNet frames are
elaborations of more abstract schemas of change, action, movement, experi-
ence, causation, etc., and the roles found in these are the ones that figure in
linking generalizations; many of the more refined frames can be seen as per-
spectives on the more abstract frames, in the way that buying is a subtype of
getting, selling and paying are kinds of giving, etc. Generalizations based on
inferences about who possesses what before and after the transaction depend
on the roles in the commercial transaction; generalizations about how syntactic
roles are assigned to the arguments depend on the more abstract inherited
schemas.
24. Annotation sets because the annotation for each targeted LU includes layered
descriptions specifying (1) the target lexical unit (which might be a discon-
tinuous character-string), (2) the syntactic labeling of the annotated phrases,
(3) the frame element labeling of these phrases, and (4) various other kinds of
information varying according the the target LU’s part of speech.
25. In general the category of “oblique nominal” (an NP with GF Dependent
rather than Object) is used as the second object of a ditransitive pattern, and as
nominal non-objects in expressions like ski Davos, shop Macys, walk the
plank, etc., where one can assume that a preposition has been omitted. By
making the same decision with “second objects” we recognize the similarities
between give someone a medal, present someone a medal, where a medal is
regarded as an oblique NP, and present someone with a medal, where the same
function is expressed with a PP: with present the preposition omission is op-
tional, with give it is obligatory.
26. Where suitable the definitions are taken from the Concise Oxford Dictionary,
for which we have permission from Oxford University Press.
27. A feature of the eventual lexical entries, not provided in the present state of
the project, is an abstract valency formula upon which the observed realization
patterns could be generated. Such formulas will be abstracted away from the
realizations found in the annotated sentences, where displacements and ellip-
ses are the product of syntactic processes independent of the purely lexical re-
quirements of the lexical target. It has not been possible to devise an automatic
way of deriving such valency formulas from the annotations, and the manual
work of drawing up such descriptions on the basis of the annotations has not
been done.
28. But see Narayanan et al. 2003.
29. At the level of granularity deemed relevant to the publisher’s intended market.
158 Charles J. Fillmore

30. Of course That counts is an acceptable sentence, but in the (still different)
sense of Validity, seen in Everybody’s vote counts.
31. A number of putative frame names introduced in this paragraph are tentative;
not all these frames exist in the current database.
32. See Ruhl (1989) and the research he praises.
33. In principle this means that to say that a word like remember has n senses is to
say that it belongs to n synsets in an idealized WordNet. Co-membership in a
single frame in FrameNet, however, is not limited – as are synsets – to words
in a single part of speech. Cf. Fellbaum (1988).
34. For each grammatical unit Gi there is a single corresponding meaning unit Mi.
35. In general it seems that-predicates requiring syntactic PP complements and
semantic propositional complements require extraposition with it in order to
satisfy both of these constraints.
36. A more complete inventory of missing valents and the conditions which call
for them can be found in Ruppenhofer (2005).
37. The INI and DNI are equivalent to what Allerton (1982) refers to as indefinite
deletion and contextual deletion, but without suggesting a layered grammatical
representation that includes a deletion operation.
38. For example, only agentively construable verbs participate in the imperative
construction, only passivizable verbs participate in the passive construction.
39. The point about (42) is the omission of the object; the subject omission as a
feature of the imperative form has already been discussed.
40. In many cases of deleted objects with instructional imperatives the possibili-
ties of valence alternation, which of course is lexically determined, does play a
role. Thus in stuff into a large pepper and stuff with ground pork both have the
direct object omitted, since the verb stuff allows both stuff x into y and stuff y
with x as two variant valency patterns. Pragmatically the omissions in instruc-
tional imperatives are like those in DNI (see below), since the identity of the
omitted entity has to be known in the context, but these are annotated as CNI
since the behavior is more determined by grammatical constructions than re-
lated to lexically specific information.
41. The omission of the contingency is limited to the limited case where the sub-
ject is that or it. Notice the unacceptable: *Success depends.
42. See the special issue on Associative Anaphora in Journal of Pragmatics
(1991, 31 (3): 311−440), where the general assumption is that a definite NP
that points to the existence in a text world of some specific kind of referent in-
vites the assumption that other things that typically accompany such an object
are likely to be present and permit referential pick-up.
43. It’s clear that the role represented here by of asthma is a core frame element,
since it gets the DNI interpretation in a sentence that mentions only the pa-
tient: It cured me.
44. The substructure of the interrupted phrase itself is determined by its own high-
est governor(s).
45. Grammatical tagging in FrameNet is intended to be theory-neutral, but where
that is impossible (as in the case of the contested analysis of “raising to ob-
 Valency issues in FrameNet 159

ject”) we choose the version from which conversion to alternative analyses is

most straightforward.

References

Allerton, David J.
1982 Valency and the English Verb. London: Academic Press.
Atkins, Beryl T. S., Charles J. Fillmore, and Christopher R. Johnson
2003 Lexicographic relevance: Selecting information from corpus evi-
dence. International Journal of Lexicography 16 (3): 251–280.
Charolles, Michel, and Georges Kleiber (eds.)
1991 Associative Anaphora. Special Issue of Journal of Pragmatics 31 (3).
Cruse, D. Alan
1986 Lexical Semantics. Cambridge: Cambridge University Press.
Fellbaum, Christiane
1988 WordNet: An Electronic Lexical Database, Cambridge: MIT Press.
Fillmore, Charles J.
1968 The case for case. In Universals in Linguistic Theory, Emmon Bach,
and Robert T. Harms (eds.), 1−88. New York: Holt, Rinehart, and
Winston.
1977 The case for case reopened. In Syntax and Semantics, Vol. 8: Gram-
matical Relations, Peter Cole, and Jerrold M. Sadock (eds.), 59–81.
New York: Academic Press.
1985 Frames and the semantics of understanding. Quaderni di Semantica 6
(2): 222−254.
1987 A private history of the concept of frame. In Concepts of Case, René
Dirven, and Günter Radden (eds.), Tübingen: Gunter Narr Verlag.
2003 Valency and semantic roles: The concept of deep structure case. In
Dependenz und Valenz / Dependency and Valency: Ein Internation-
ales Handbuch der Zeitgenössischen Forschung / An International
Handbook of Contemporary Research, Vol. 2, Vilmos Ágel, Ludwig
W. Eichinger, Hans-Werner Eroms, Peter Hellwig, Hans-Jürgen He-
ringer, and Henning Lobin (eds.), 457−474. Berlin/New York: Mou-
ton de Gruyter
Fillmore, Charles J., and Beryl T. Sue Atkins
1992 Towards a frame-based organization of the lexicon: The semantics of
RISK and its neighbors. In Frames, Fields, and Contrast: New Es-
says in Semantics and Lexical Organization, Adrienne Lehrer, and
Kittay Eva (eds.), 75−102. Hillsdale: Lawrence Erlbaum Associates.
1994 Starting where the dictionaries stop: The challenge for computational
lexicography. In Computational Approaches to the Lexicon, Beryl T.
Sue Atkins, and Antonio Zampolli (eds.), 349–393. Oxford: Oxford
University Press.
160 Charles J. Fillmore

Fontenelle, Thierry (ed.)

2003 FrameNet. Special issue of International Journal of Lexicography 16
(3).
Francis, Elaine J., and Laura A. Michaelis (eds.)
2003 Mismatch: Form-function Incongruity and the Architecture of Gram-
mar. Stanford: CSLI Publications.
Levin, Beth
1985 English Verb Classes and Alternations: A Preliminary Investigation.
Chicago: The University of Chicago Press.
Mel’cuk, Igor
1998 Collocations and lexical functions. In Phraseology: Theory, Analysis
and Aplications, Anthony P. Cowie (ed.), 23–53. Oxford: The Clar-
endon Press.
Mohit, Behrang, and Srini Narayanan
2003 Semantic extraction with wide-coverage lexical resources. In HLT-
NAACL 2003: Companion Volume, Marti Hearst, and Mari Ostendorf
(eds.), 64−66. Alberta, Canada.
Narayanan, Srini, Collin Baker, Charles Fillmore, and Miriam Petruck
2003 FrameNet meets the semantic web: Lexical semantics for the web. In
The Semantic Web – International Semantic Web Conference 2003,
Dieter Fensel, Katia Sycara, and John Mylopoulos (eds.), 771–787.
Berlin: Springer-Verlag.
Petruck, Miriam R. L.
1996 Frame semantics. In Handbook of Pragmatics 1996, Jef Verschueren,
Jan-Ola Östman, Jan Blommaert, and Chris Bulcaen (eds.), Philadel-
phia: John Benjamins.
Pustejovsky, James
1995 The Generative Lexicon. Cambridge: MIT Press.
Ruhl, Charles
1989 On Monosemy: A Study in Linguistic Semantics. New York: State
University of New York Press.
Ruppenhofer, Josef
2005 Regularities in null instantiation. Manuscript.
Sweetser, Eve
1990 From Etymology to Pragmatics: Metaphorical and Cultural Aspects
of Semantic Structure. Cambridge: Cambridge University Press.
Tesnière, Lucien
1959 Eléments de Syntaxe Structurale. Paris: Klincksieck.

http://www.darpa.mil/
http://www.ic-arda.org/
http://www.ldc.upenn.edu/
http://framenet.icsi.berkeley.edu/
Section 2
Cognitive issues and valency phenomena
Valency and cognition – a notion in transition

Gert Rickheit and Lorenz Sichelschmidt

Valency, one of the key notions in linguistics, is particularly suited to dem-

onstrate the development of this discipline during the past decades. The
way linguistics views itself has changed from a placement among the arts
or humanities towards cognitive or life sciences. Linguistic methods have
changed from introspection to experimentation, using sophisticated tech-
niques and the latest equipment. Even the subject of linguistics has
changed: while, in its early years, the discipline has focused on the struc-
ture of verbal utterances, contemporary linguistics embraces language us-
age as well as language users (Rickheit, Sichelschmidt, and Strohner 2002).
Presently, linguistics is further broadening its scope far beyond the cogni-
tive processes in the production and comprehension of verbal utterances as
it is venturing to embrace topics like the neurophysiological substrate of
language use, or human conversation as the art of successful information
interchange by verbal means. Along with the profile of linguistics as a sci-
ence, the notion of valency as a classical concept in linguistics has under-
gone dramatic changes. By tracing those changes over the decades, we
learn something about the development of linguistics, and hopefully, also
about its prospects.

1. Valency – the classical approach

The introduction of the term valency into linguistics is frequently attributed

to the French grammarian Lucien Tesnière (1959). Borrowing the term
from chemistry, Tesnière, in comparing words to atoms that combine with a
fixed number of other atoms of an opposite electrical charge, indeed laid
the foundations of what is referred to today as dependency grammar. In his
seminal book Éléments de syntaxe structurale, Tesnière (1959: 13−14)
explained the idea of a structural dependency of the elements of an utter-
ance as follows: “Les connexions structurales établissent entre les mots des
rapports de dépendance. Chaque connexion unit en principe un terme supé-
rieur à un terme inférieur. … L’ensemble des mots d’une phrase constitue
… une véritable hiérarchie.” [The structural connections form dependencies
164 Gert Rickheit and Lorenz Sichelschmidt

between the words. Basically, every connection binds a superior to a subor-

dinate element. … The words within a phrase … form a real hierarchy.]
In Tesnière’s conviction, the verb plays the central role in the structure
of an utterance. The verb requires particulars, which are to be specified
verbally by means of appropriate phrases or clauses – the so-called argu-
ments – or else have to be inferred from context. A verb like talk, for in-
stance, requires at least the specification of some speaker as a subject,
while a verb like donate requires specification of a donator, a donatee, and
a donation. So it is the verb which basically controls how many dependent
elements occur in a sentence. With this, valency, in its most basic quantita-
tive sense, refers to the capacity of a verb to take a specific number of de-
pendent units.
However, Tesnière was not the very first to think along these lines. The
Indian grammarian Panini (ca. 480 B.C.) is widely credited as the scholar
who first elaborated on structural dependencies between linguistic expres-
sions. In his book Astadhyayi, Panini used formal production rules to de-
scribe the structure of sentences and compound nouns in Sanskrit in a man-
ner quite similar to modern linguistic theories (Böhtlingk 1998).
Another precursor of the notion of valency can be found in early Euro-
pean psycholinguistics. The psychologist Karl Bühler, in his seminal book
Sprachtheorie (1934: 249) has elaborated on the valency of verbs on the
Latin example Caius necat leonem [‘Caius killed the lion’] in the following
way:

Wo immer ein Verbum die Komplexion regiert, dort und nur dort sind Leer-
stellen, in welche primär Caius und der Löwe eingesetzt werden können. ...
Warum provoziert das Verbum die Fragen wer und wen? Weil es der Aus-
druck einer bestimmten Weltauffassung ... ist, einer Auffassung, die Sach-
verhalte unter dem Aspekt des (tierischen und menschlichen) Verhaltens
begreift und zur Darstellung bringt.
[Where a verb governs a complex structure, there – and only there – are
slots which can be filled with Caius and the lion. … Why does the verb
evoke questions like who and whom? Because it expresses a certain per-
spective of the world …, a perspective which analyzes and represents situa-
tions according to the behaviour (of humans and animals).]

Following Tesnière, the notion of valency became well established in Euro-

pean linguistics – mostly in foreign language teaching, but also, in a way,
as a counterconcept to mainstream syntactic theory. Lyons (1981: 116), in
distinguishing between two principles of grammatical relations, depend-
ency and constituency, noted that “Chomskyan generative grammar has
opted for constituency, in this respect following Bloomfield and his succes-
 Valency and cognition 165

sors. Traditional grammar laid more emphasis on dependency”. However,

due to the fact that during the sixties, generative grammar was the dominant
paradigm in linguistic theorizing, it was not until the seventies that theo-
retical linguistics rediscovered the notion of valency. Since then, the de-
pendency framework has gained about as much recognition in theoretical
linguistics as it has had in applied linguistics. To date, numerous theoretical
and empirical papers have been published on diverse aspects of depend-
ency, and their number is steadily growing (see Ágel et al. 2003).
Since, in the view of dependency grammar, the faculty of controlling the
number of arguments in a sentence is an inherent property of the verb,
valency provides a means to subclassify verbs in terms of their dependent
units. These dependent units are basically equivalent, so that subject and
object specifications have equal status. Also, “this notion of valency does
not presuppose … that the dependents of a predicator are necessarily noun
phrases. What are traditionally referred to as adverbial complements of
time, place, etc., also fall within the scope of the definition of valency”
(Lyons 1981: 116−117). A few examples of possible verb classes based on
valency are listed below:

Table 1. Possible verb classes based on valency

verb class arguments English example

avalent 0 It was raining.

monovalent 1 Holmes yawned.
divalent 2 Holmes spotted Moriarty.
trivalent 3 Holmes handed the letter to Watson.

However, in attempts at subclassifying verbs according to their valency, a

fundamental problem arises from Tesnière’s division of arguments into
complements and adjuncts, which was widely adopted (see Somers 1987).
While complements (also termed actants) take the essential thematic roles
and thus are necessary to render a sentence grammatical, adjuncts (also
termed circumstantials) are of a more elaborative kind. Complements de-
pend in form on the governing verb; adjuncts, in contrast, are freely ap-
pendable to the valency structure. To illustrate, in the sentence Holmes
travelled from London to Geneva via Dover, only the arguments Holmes
and to Geneva would typically qualify as indispensable complements; the
others might (arguably) be regarded as adjuncts, so that, in effect, the verb
to travel would be classified as divalent. The problem with the dichotomy
is that the criteria for classifying an argument as a complement or an ad-
junct are anything but clear; after all, information which is essential to the
166 Gert Rickheit and Lorenz Sichelschmidt

interpretation of a sentence need not always be supplied explicitly. In con-

sideration of this, some researchers have proposed a more subtle distinction
as an alternative to the dichotomy (cf. Herbst and Roe 1996). Unfortu-
nately, though, a finer-grained classification is not a solution to the problem
of unclear criteria: a characterization of obligatory complements as
“Glieder, die in der Regel nicht weglassbar sind” [elements that usually
cannot be omitted] (Helbig 1992: 99) is far too vague to be of any help.
What is required, then, is a clarification of criteria – which can be accom-
plished only by more thorough inquiry into the way language users actually
utilize structural dependencies in producing or comprehending verbal utter-
ances.
Another issue that deserves consideration is the extent to which argu-
ments can be classified in terms of semantic qualities. Clearly, Tesnière’s
resort to syntactic case as an indicator of different semantic qualities of
complements (notably, nominative, dative, and accusative case to indicate
particulars about who, whom, and what, respectively) is inadequate when it
comes to handling other types of assertions (as in Holmes travelled from
London to Geneva via Dover), more complex dependencies, or implicit
complements. However, the incorporation of semantic qualities leads to a
different definition of valency. The valency of a verb, in this sense, is de-
termined by the constellation of the complements (not just by their num-
ber). Such a qualitative approach to structural valency, however, necessi-
tates the development of clear-cut criteria for a classification of arguments
in terms of their semantic roles – a point which shall be taken up in due
course. As an interim summary, valency, from a structural point of view,
can be regarded as the capacity of language elements to combine with par-
ticular dependent units for the formation of larger units.
The structural approach to valency has proved to be empirically fruitful.
Results from several studies suggest that the notion of valency has some
cognitive relevance. In a vintage memory experiment, for instance,
Wilczok (1973) read to her students a number of five-word sentences which
varied with respect to the valency of the verb: she compared divalent verbs
(as in The lady swallowed the tablets) to trivalent verbs (as in The lady sold
the tablets) so that each participant heard only one (randomly assigned)
version of each sentence, with versions balanced between two samples of
participants. In a subsequent cued recall test, the participants remembered
the divalent versions significantly better than the trivalent ones. In an ex-
tension of this study, Raue and Engelkamp (1977) also found that recall
decreased with increasing valency; in addition, they found that recall in-
creased along with the degree of semantic relatedness of the arguments.
 Valency and cognition 167

The notion of valency has also proved useful in the investigations of

language development. In a large-scale developmental study of child lan-
guage, Rickheit (1975; 1978) pursued the idea that dependency might be a
valid indicator of children’s linguistic abilities at primary school level, with
the valency hierarchy increasing in complexity in the course of language
development. In the study, 600 primary school children (aged 6;0 to 12;0
years) gave oral or written personal recollections and object descriptions.
These corpus materials were then transcribed and analyzed. For analysis,
Rickheit categorized valency as to three hierarchy levels:
− Primary constitutive verbal complements depend on the specific valency
of the verb (as in Fritz gab Peter das Buch [Fritz gave Peter the book]).
− Secondary constitutive verbal complements depend on the valency of
primary complements (as in Peter ist müde vom Reden [Peter is tired
from talking]).
− Tertiary constitutive verbal complements depend on the valency of sec-
ondary ones (as in Er wollte sie ins Krankenhaus bringen lassen [He
wanted to have her taken to hospital]).
The results of the empirical investigations were as follows: children clearly
preferred primary constitutive verbal complements over secondary or terti-
ary ones. In addition, children preferred uni- or divalent utterances over
more complex ones. The three most popular patterns in primary school
children’s oral language were: univalent (e.g., Wir haben viel gelacht [We
laughed a lot], 20%), divalent with an accusative-object complement (e.g.,
Er trägt eine Brille [He wears glasses], 12%), and divalent with an accusa-
tive-patient complement (e.g., Sie hat mich gerufen [She called me], 12%).
The proportion of these patterns remained relatively stable over the years
(45% for the 6-year-olds to 43% for the 9-year-olds), and so did their dis-
tribution. Altogether, the ten most frequent patterns accounted for 80% of
the syntactic structures used. Children’s oral and written language did not
differ with respect to syntactic pattern preference: the “top ten” patterns
were identical; there were, however, significant differences in the fre-
quency distributions.
These studies clearly show that valency is indeed a concept of relevance
to linguistic accounts of human language use. However, there are major
problems with the structural approach to valency. To reiterate, argument
classes are not unequivocally defined; neither purely syntactic criteria nor
optionality considerations based on introspection – or would intro-
speculation be more apt? – are sufficient for an adequate account. For an-
other, the proposed linguistic tests – above all, deletion or replacement tests
– are not unambiguous: depending on the importance granted to individual
knowledge and situational context, they may arrive at different answers.
168 Gert Rickheit and Lorenz Sichelschmidt

For instance, the verb to dwell is usually but not necessarily divalent; coun-
terexamples are occurrences of univalent usage like the catchphrase fea-
tured in some TV commercials of a European furniture dealer – “Wohnst
du noch oder lebst du schon?” [Still dwelling, or already living?] (IKEA
2005). Finally, the dependency approach and hence, the notion of valency,
merits some more thorough consideration of its extension beyond the verb.
From a structural point of view, other word classes also govern dependent
units. An adjective like similar, for example, can be considered divalent
because it requires two structures to be compared, likewise, a noun such as
fear requires some particulars about the experiencer and the cause.

2. A reappraisal of the structural approach

It was the linguist Charles Fillmore (1968) who opened a new vista for the
structural approach to dependency. In his so-called case grammar, Fillmore
foregrounded semantic aspects by focusing on the arguments, that is, on the
determined units rather than on the elements that govern these. The idea
was to define valency classes on the basis of the type of the arguments
rather than on their number. Thus, the quantitative point of view on valency
was supplemented by qualitative aspects, namely, the particular constella-
tion of complements. To illustrate, the verbs to dance and to die are both
univalent in that they demand one nominative phrase giving particulars on
who; they differ, however, in that to dance requires particulars on an agen-
tive entity while to die requires particulars on an experiencer. Thus, by
capturing the thematic roles of the participants (also referred to as theta
roles or case roles), the analysis was extended beyond syntactic case and
surface function into the semantic domain.
Linguistic theories generally proceed from the assumption that, since
people organize their knowledge in terms of relations between entities,
states, and events, there is a finite number of thematic roles (Chafe 1970,
Jackendoff 1987). The classic thematic roles are agent, patient (or experi-
encer), object, and instrument, plus a set of locational and temporal roles
like time, place, source, and goal. In some languages, these roles may have
distinctive morphosyntactic characteristics, such as unique case markings,
or restrictions on aspect or modality. Turkish, for instance, has a rich sys-
tem of syntactic cases which, in addition to the nominative, genitive, dative,
and accusative case, comprises locative, ablative, and illative case to spe-
cify location, source, and goal, respectively. However, thematic roles are in
principle conceived of as being universal and independent from surface
syntax. In fact, arguments can surface in a number of ways (see Hörmann
 Valency and cognition 169

1979: 230): if, in a sentence, the role of the agent is occupied, the instru-
ment may surface in a with-clause (e.g. Adrian opened the door with a
crowbar). However, if the role of the agent is not occupied, it can be as-
sumed by the instrument using a noun in the nominative case (e.g. The
crowbar opened the door).
It goes without saying that thematic roles are dynamic structures; after
all, an agent is only an agent at the bidding of and for a particular verb.
This fact has been exploited in order to classify verbs according to their
typical patterns of thematic roles – their valency structures, or case frames,
to use Fillmore’s (1968) terminology. For instance, the valency of the verb
to open has been assumed to unfold in a case frame that has an object as an
obligatory argument and an agent and an instrument as optional arguments.
Various sentence structures are compatible with this case frame; among
them The door opened, Adrian opened the door, and Adrian opened the
door with a crowbar.
In dependency grammar, such valency structures are regarded as the
syntactic-semantic basis of sentences (Welke 1995). Accordingly, valency
dictionaries for languages such as German, French, and – recently – Eng-
lish have been compiled (e.g. Helbig and Schenkel 1968; Herbst et al.
2004; Schumacher 1986) which list the verbal entries by their case frames.
The semantic view of valency has also found its way into theoretical
linguistics (e.g. Haegeman 1991; Jackendoff 1990). At that, the original
conception of arguments in terms of thematic roles has been modified in
diverse ways. Dixon (1991), for instance, has divided case frames into
some 50 verb classes, each of which has one to five distinct thematic roles.
On the other hand, Dowty (1991) has attempted to trace back the plethora
of thematic roles to only two thematic proto-roles – a proto-agent (which
involves characteristics like causation, perception, and volition) and a
proto-patient (which involves characteristics like effectiveness, responsive-
ness, or change of state). However, the basic idea that valency, conceived
of in terms of argument constellations, is a useful starting point for a com-
prehensive account of the semantics of sentences, has remained largely
unchanged.
Taken together then, case grammar as a semantic variety of the struc-
tural approach to valency has opened new avenues for linguistic research. It
has proved useful not only in describing the dependency relations that hold
among the elements of a sentence; it is also suited to explain phenomena
like the occurrence of an intransitive use of a transitive verbs (as in Paddy
doesn’t drink), so that certain arguments are left unspecified. Above all, the
structural approach to valency indisputably has some potential to system-
atically relate what – in transformational grammar – has been termed the
170 Gert Rickheit and Lorenz Sichelschmidt

surface and the deep structure of an utterance, thus smoothing the way to-
wards an in-depth study of linguistic semantics. On the other hand, the
notion of valency, as developed so far, has a few shortcomings: it is re-
stricted to structural aspects in that it attempts to provide a description of
the meaning of words and sentences – regardless of the fact that any verbal
utterance is awarded (at least part of) its meaning through its use in a par-
ticular situation. The notion of valency is restricted in that analyses of de-
pendencies proceed in a post-hoc fashion and hardly ever transgress sen-
tence boundaries – which fails to do justice to the fact that utterances are
produced and perceived in context. And it is largely built on linguists’ in-
tuitions about the functions of valency rather than on factual evidence about
its functionality in actual language use.

3. The functional approach

In the mid-seventies, linguistics – in its own view, the science of language

structure and language use – received new impetus from the blossoming
fields of psychology and cognitive science. As a consequence of the “cog-
nitive turn”, the structural orientation that had been prevalent in the domain
was now supplemented (or, in psycholinguistics, almost replaced) by a
functional orientation. In the course of this development, the subject matter
of linguistics has broadened so as to additionally comprise the language
user. Scientific interest gradually shifted away from verbal structures to-
wards mental structures. Research now focused on questions like how lan-
guage is represented in people’s minds, or how people actually produce and
comprehend verbal utterances. Along with this new approach, the canon of
linguistic methodology widened. Experimentation on cause-effect relation-
ships has played an increasingly important role in the discipline ever since
(Sichelschmidt and Carbone 2003).
As one side effect of the “cognitive turn”, the dependency idea was rein-
terpreted in functional terms. After all, semantics is not a matter of words
depending on each other but ultimately, a matter of concepts. In line with
this, valency, from a functional point of view, was reinterpreted as referring
to the capacity of concepts to combine with particular dependent concepts
for the formation of more complex ideas. In this sense, not only verbs,
nouns, or adjectives may serve as predicators that have valency, but con-
junctions or prepositions as well since these also govern the use of other
concepts. Engelkamp (1976) proposed three major conceptual classes of
predicators:
 Valency and cognition 171

− Attributive predicators are those that specify a feature, a state or an af-

filiation of an entity (as in the phrase an old book).
− Processional predicators are those that specify a change of a feature, a
state, or an affiliation of an entity (as in an intermittent light).
− Actional predicators are those that specify an action, a transaction or a
transformation (as in a heart-breaking song).
“It is important to note that the predicators reflect the structure of our world
knowledge,” Engelkamp (1976: 25) remarked. “These are the structures
that guide our thinking, calling forth expectations also about verbal infor-
mation.”
The metamorphosis of valency from a linguistic notion to a mental no-
tion was finalized by the psychologist Walter Kintsch. In an immensely
influential book on The Representation of Meaning in Memory, Kintsch
(1974) gave a comprehensive account of a dependency-based metalanguage
for the description of the meaning of verbal expressions in general. Mean-
ing, Kintsch argued, can be represented by structured sets of propositions.
Though no explicit reference is made to Tesnière, the metaphor at the bot-
tom of Kintsch’s propositional account closely corresponds to the classic: a
proposition is a sort of “meaning molecule” which comprises exactly one
predicator plus a number of arguments. A predicator can be expressed by a
verb, an adjective, a conjunction, a preposition, or the like, but more ab-
stract concepts like implication or consequence are also permitted. The-
matic roles such as agent, experiencer, instrument, object, source, or goal
are assigned as arguments; however, the propositional account also permits
hierarchical embedding – which means that a proposition may function as
an argument to a higher-level proposition. Kintsch advocated a functor-
argument notation system for propositions which is illustrated in the fol-
lowing examples (after Kintsch 1974; modified):

Table 2. Notation of propositions

Wording Propositions

The clown smiled. SMILE (CLOWN)

Cliff sold Linda a book. SELL (CLIFF, LINDA, BOOK)
Cliff sold Linda an interesting book. SELL (CLIFF, LINDA, INTERESTING
(BOOK))
The clown smiled and bowed. AND (SMILE (CLOWN), BOW (CLOWN))
Jane slept on the sofa. SLEEP (JANE, ON, SOFA)
The baby did not spill the milk. NOT (SPILL (BABY, MILK))
Not the baby spilled the milk. NOT (SPILL (BABY, MILK)) & SPILL ($,
MILK)
172 Gert Rickheit and Lorenz Sichelschmidt

It should be noted that Kintsch’s propositional account is meant to be truly

semantic: it is an abstraction from the actual wording (tense, for instance, is
not represented locally, nor is definiteness or indefiniteness of reference), it
includes resolution of ellipses or pronouns as a matter of course, and it
comprises inferences (for instance, in the affirmative proposition that
someone other than the baby spilled the milk). Another remarkable feature
of the propositional account is that it is easily expandable to larger texts.
Local coherence, the paramount characteristic that distinguishes text from
non-text, can be accommodated in the propositional account by evaluating
the recurrence of particular arguments, and thematic progression as well as
long-range dependencies can be treated by assigning a relevance weight to
each proposition, with the weight gradually decreasing along with progress
in the analysis of the text.
Several experiments have supplied empirical evidence as to the cogni-
tive status of propositions. In a timed reading study (Kintsch and Keenan
1973), participants had to read sentences which were displayed on a screen
one after another. The sentences had 16 words each (including punctua-
tion); they varied in the number of propositions (from 4 to 9) in the base
structure.
− Sample sentence with 4 propositions: Romulus, the legendary founder of
Rome, took the women of the Sabine by force.
− Sample sentence with 8 propositions: Cleopatra’s downfall lay in her
foolish trust in the fickle political figures of the Roman world.
Reading time for the sentences varied along with the number of proposi-
tions; on average, it took participants almost one extra second per proposi-
tion (the least square regression model was t = 6.34 + 0.94 nprop). Moreover,
the likelihood of recall of a proposition varied with the number of its argu-
ments and with its level in the recurrence hierarchy: propositions that ap-
peared early in a text and had many recurring arguments were recalled bet-
ter than those of a lower rank.
Also, the cognitive relevance of particular arguments has been put to
test (see Shapiro, Nagel, and Levine 1993). A memory experiment con-
ducted by Obliers (1985) has provided evidence against the assumption that
arguments are independently governed by the predicator. In the experiment,
participants were given trivalent sentences (e.g. The maid put the vase on
the desk) and were asked to underscore arguments of a particular type. In a
subsequent incidental recall test, neither the verb nor any of the thematic
roles played a central role in reactivating the sentences. Rather, focus on
agents led to an increased recall of agents, focus on patients increased recall
of patients plus objects, focus on objects did not increase recall at all, and
focus on locatives increased recall of locations as well as of patients. These
 Valency and cognition 173

observations point towards syntagmatic – and unidirectional – dependen-

cies among the various thematic roles, suggesting that combinations of the
predicator and one or more arguments form gestalt-like semantic units with
higher order meaning.
The propositional approach to semantic structure has subsequently de-
veloped into a comprehensive theory of text processing (Kintsch and van
Dijk 1978). According to this theory, text processing comprises the step-
wise, iterative extraction of propositions; these are temporarily stored in the
recipient’s working memory, where they are assembled to form a coherent
hierarchy, the so-called text base. This “bottom up” account of text process-
ing reveals several advantages, but also a few shortcomings of the proposi-
tional approach. On the positive side, predicator-argument constellations
are functionally useful as elements of a versatile metalanguage for the de-
scription of semantic structures; moreover, they appear to constitute cogni-
tively adequate processing units that can be tied down to the original notion
of valency. On the negative side, propositions require human expertise and
interpretation in order to accomplish reference and coreference resolution,
to handle indirect meaning in a plausible way, and to cope with non-
contingent constituents (as in Hank took his hat off – Hank took off his hat).
Most importantly, the cognitive mechanisms in the identification and ex-
traction of predicator-argument constellations need to be explored in detail.
In order to study the role of the predicator in on-line sentence process-
ing, Rickheit, Günther, and Sichelschmidt (1992) have taken advantage of
the flexibility of German word order. The authors hypothesized that, in case
of a verb-first expression, the verb – due to valency – will enable the reader
to mentally establish a relational link between arguments yet to be speci-
fied. In case of a verb-last expression, though, any argument noun phrases
will have to remain standing alone for some time since the information
necessary to integrate the arguments into a coherent conceptual structure is
supplied only when reading the verb. Hence, verb reading times should be
shorter with verb-first expressions than with verb-end expressions, whereas
noun reading times should be slightly prolonged. This hypothesis was
tested by comparing verb-first versions of 16 German sentences to verb-last
versions in a self-paced word-by-word reading experiment with 40 student
participants. For example, the verb-first clause …und so treiben die Bauern
das Vieh auf die Weide und… was compared to the verb-last clause …so
dass die Bauern das Vieh auf die Weide treiben und… [drive with the ar-
guments the farmers, the cattle, to the pastures].
174 Gert Rickheit and Lorenz Sichelschmidt

800
ms
Verb first
700 Verb last

600

500

400
V Det1 N1 Det2 N2 Prep Det3 N3 V

Figure 1. Word reading times (in ms) for trivalent verb-first and verb-last clauses

The analysis of word reading times (figure 1) showed that verb-first and
verb-last versions differed significantly in reading times for the verb and
for the last-mentioned noun: as predicted, the verb took longer to read when
it occurred after the arguments than when it occurred prior to them. On the
other hand, the last noun took longer to read in a verb-first expression than
in a verb-last expression. The latency profiles indicate that supplying read-
ers with the verb at an early point during reading indeed facilitates compre-
hension, whereas reading the arguments prior to the predicator is detrimen-
tal to processing because it increases working memory load. So far, the
expectations have been substantiated empirically. Contrary to the predic-
tions, however, verb fronting did not lead to a general increase in noun
reading times. So there is no empirical support for the idea that tying an
argument node to a relation link takes longer than merely installing it as a
stand-alone. Hence, the decelerating effect of verb fronting observed at the
last noun should perhaps be attributed to the fact that, in verb-first versions,
the last noun occurred at the end of the clause. This suggests another
mechanism to be operative in German sentence comprehension: readers
may want to postpone the cognitive integration of predicator and arguments
until attaining some degree of semantic saturation. In consequence, integra-
tive processes in comprehension are contingent on reading the last element
of a clause. This should become manifest in prolonged reading times for the
final element of an expression – which was exactly the pattern of results
found in the experiment.
 Valency and cognition 175

4. The procedural approach

With considerations like these, the discussion of valency has taken still
another turn. Dependency and valency have now become genuine mentalis-
tic notions. The basic idea is that valency, being (proverbially) in the mind
of the beholder, is assigned on-line and thus can be observed only by means
of scrutinizing the cognitive processes in language comprehension and
production. Pointedly, one might say that valency is not any longer a char-
acteristic of verbal expressions that may take an effect on processing but
rather, valency is an outcome of cognitive processes that may become
manifest in typical constellations of verbal expressions.
The renewed metamorphosis of the approaches to the notion of valency
– its transition from a functional to a procedural notion – coincided with the
emergence of constructionist thinking in psycholinguistics (Rickheit, Si-
chelschmidt, and Strohner 2002). Emphasis was now on knowledge-based
“top down” processes in text comprehension. The fundamental idea – remi-
niscent of Bühler’s (1934) – was that readers or listeners take a text as a
ground for activating knowledge which enables them to mentally construct
a comprehensive model of the states of affairs described in that text. The
cognitive scientist Philip N. Johnson-Laird (1983) has conceived of such
“mental models” as holistic, dynamic knowledge structures which, unlike
other forms of representation, have two important characteristics: mental
models represent the situation in question in such a way that the structure of
the representation depicts the structure of the represented, and they may, in
their representational function, go far beyond what is explicitly stated in the
text (see Rickheit and Sichelschmidt 1999; Zwaan 1999).
With respect to the procedural approach to valency, two concomitants of
the rise of constructionist theories are of relevance – the re-emergence of
the construct of cognitive schemata, and the placement of emphasis on
inference as a fundamental mechanism in language processing.
Cognitive schemata (Bartlett 1932) are abstract, stereotypical macro-
concepts of objects or situations which provide language users with a
framework for the organization of knowledge. Schemata are often thought
of as providing slots (variables) and default slot-fillers (values) which can
be specified in actual use. To illustrate, a schema of a car may comprise
general features of typical cars (such as ‘has an engine’, ‘has four wheels’,
‘has left-side steering’, ‘seats five’, and the like) which readers of the word
car may use to develop an enhanced representation of the object referred to.
Schemata serve important purposes in language processing; among them
selection (the filtering of information according to relevance) and instantia-
tion (the establishment of conceptual defaults). Experiments have demon-
176 Gert Rickheit and Lorenz Sichelschmidt

strated the importance of cognitive schemata as knowledge organization

units at various levels of processing. For instance, a question like How
many animals of each kind did Moses take on the Ark? will invoke some
biblical schema which may prevent recipients from detecting the anomaly
(it was not Moses who did). In contrast, a question like How many animals
of each kind did Mozart take on the Ark? which does not invoke a unitary
schema will make almost all recipients notice the anomaly (e.g. Sanford
and Garrod 1994).
From a procedural point of view, the notion of valency lends itself to in-
terpretation in terms of schematization: encountering a predicator will li-
cense the establishment of an appropriate argument frame, and accordingly,
the thematic roles can be filled with default concepts. In fact, there is em-
pirical evidence as to such a cognitive mechanism: using a single-word
priming technique, Ferretti, MacRae, and Hatherell (2001) have demon-
strated that verbs immediately activate knowledge of typical agents, pa-
tients, and instruments but not locations (e.g. arresting activates cop and
criminal). At that, the activation of agent and patient thematic role concepts
was modulated by syntactic cues. These findings have led the authors to
conclude that schematic world knowledge is tied tightly to on-line thematic
role assignment, and thus should be considered as part of thematic role
knowledge.
Inference, the other phenomenon that is relevant to valency, relates to
the fact that people often understand more than what is explicitly stated in a
text (see Rickheit and Strohner 2003). Knowledge-based schemata provide
a basis for the generation of inferences, that is, for the enrichment of mean-
ing by default. The current state of psycholinguistic research on inference
in text processing can briefly be summarized as follows: for one, inference
is an everyday cognitive mechanism which can be observed in any kind of
human information processing. For another, inference is not a uniform phe-
nomenon; researchers widely agree on a distinction between two broad
classes of inferences in text processing (cf. Rickheit, Sichelschmidt, and
Strohner 2002):
− Obligatory inferences are those that are required for text coherence (e.g.
suppositions about the referent of a pronoun). They occur frequently and
do not take much cognitive effort.
− Optional inferences are those that elaborate on the text (e.g. schema-
based speculations about the particulars of a situation). They occur less
frequently, depend on circumstances and take comparatively much cog-
nitive effort.
Since the latter inferences are at the core of constructionist theorizing (and
at the same time, subject of dispute), contemporary psycholinguistic re-
 Valency and cognition 177

search primarily pursues the question under which conditions, how, and
when in the course of processing people generate which optional infer-
ences.
In this context, the notion of valency is brought into play when address-
ing the issue of whether or not thematic roles can be inferred, and if so,
under which conditions, how, and when in the course of processing which
thematic role information is activated. Evidence from priming studies and
eye movement measurement has indicated that certain thematic roles such
as instruments can indeed be activated by inference. Garrod et al. (1990),
for example, have analyzed readers’ eye movements in order to detect any
differences in processing between four kinds of sentences:
− unspecific statements (e.g. He assaulted her with his weapon);
− specific statements (e.g. He stabbed her with his knife);
− statements with a specific predicator (e.g. He stabbed her with his
weapon);
− statements with a specific argument (e.g. He assaulted her with his
knife).
Eye movement patterns did not discriminate between the latter three condi-
tions; they were, however, significantly different from those observed with
unspecific statements. Apparently, thematic roles need not be made fully
explicit; an appropriate thematic role frame can be established by specific
predicators or arguments alike.
Other experiments have focused on inferences of other types of argu-
ments. Mauner, Tanenhaus, and Carlson (1995) have employed a “stop
making sense” reading task to investigate if people instantiate an agent
when reading short passives like The door was shut, in which no agent is
explicitly given. In the experiment, participants had to read sentences word
by word, either pressing a “yes” key to make the next word appear or a
“no” key to stop the procedure and indicate that they felt the sentence to be
unacceptable. The authors compared short passives like The door was shut
to reduce the noise to intransitives like The door shut to reduce the noise.
Intransitives, naturally, would have to be considered unacceptable because
a door, being an inanimate object, cannot act on purpose. Passives, on the
other hand, would make sense provided that readers inferred the presence
of an animate agent.
178 Gert Rickheit and Lorenz Sichelschmidt

30
∑%
25 passive
20
intransit

15

10

0
shut to reduce the noise

Figure 2. Cumulative % of “no” answers (after Mauner, Tanenhaus, and Carlson

1995)

The low percentage of “no” answers in the “passive” condition (figure 2)

indicates that people did indeed infer an agent when reading short passives.
In contrast, the sharp rise of “no” answers in the “intransitives” condition
after encountering a clause that implies purposeful action of some agentive
entity indicates that valency-related inferences are made on-line, that is,
immediately during reading.
Finally, a recent series of studies conducted by Knöferle et al. (2005)
shall be outlined which sheds still another light on the intricate relationship
of valency and cognition. In these studies, the authors have addressed the
question of whether the process of on-line thematic role assignment is in-
fluenced by nonverbal information. Participants in the experiments viewed
pictures of complex scenes (e.g. a fencer drawing a picture of a princess
who is taking a photograph of a pirate) while hearing German sentences
like Die Prinzessin malt offensichtlich den/der Fechter [Obviously, the
princess is painting / being painted by the fencer]. In these sentences, the
thematic role of the princess is initially ambiguous. The ambiguity can be
resolved only on hearing the second noun phrase: the accusative case de-
terminer den renders the princess an agent while the nominative case de-
terminer der renders the princess a patient. However, by relating the verb to
the depicted events, thematic role assignment can be facilitated. Analysis of
people’s eye movements, in particular, anticipatory fixations of the appro-
priate agent or patient entity, showed that verb-mediated visual event in-
formation allowed early on-line disambiguation. This was corroborated in a
 Valency and cognition 179

further experiment which demonstrated that with verb-final expressions,

intonation cues enabled role disambiguation even before people processed
the verb. Taken together, these findings suggest that the verb is just one
among several cues that contribute to the on-line establishment of an ap-
propriate thematic role frame. The results support a notion of valency
which includes nonverbal relationships as well as action and event sche-
mata in addition to arguments; they suggest that people, in the comprehen-
sion of verbal utterances, exploit a rich inventory of semantic categories
which go far beyond the linguistic domain.
With this, the notion of valency has eventually made its way from lin-
guistics into cognition, a transition from a structural concept to a procedural
one. Ultimately, valency has developed from a characteristic attributed to
verbal entities unto a characteristic inherent to the way language users per-
ceive situations. However, during these decades of transition, the notion of
valency has lost nothing of its value for the organization of human informa-
tion processing. Capturing the tendency of human beings to organize their
environment in terms of “belonging together”, in terms of partitioning
wholes or grouping elements so as to arrive at meaningful, workable, and
communicable units – this is perhaps the ulterior motive behind the linguis-
tic and psychological attempts to draw on valency as an explanatory con-
cept. And explanatory it is – provided that it has been given a comprehen-
sive explanation.

References

Ágel, Vilmos, Ludwig M. Eichinger, Hans-Werner Eroms, Peter Hellwig, Hans-

Jürgen Heringer, and Henning Lobin (eds.)
2003 Dependenz und Valenz – ein internationales Handbuch der zeitge-
nössischen Forschung. Berlin: de Gruyter.
Bartlett, Frederic C.
1932 Remembering. Cambridge: Cambridge University Press.
Böhtlingk, Otto
1998 Panini’s Grammatik. Hamburg: Buske.
Bühler, Karl
1934 Sprachtheorie. Die Darstellungsfunktion der Sprache. Jena: Fischer.
Chafe, Wallace L.
1970 Meaning and the Structure of Language. Chicago: University of
Chicago Press.
Dixon, Robert M. W.
1991 A New Approach to English Grammar on Semantic Principles. Ox-
ford: Oxford University Press.
180 Gert Rickheit and Lorenz Sichelschmidt

Dowty, David R.
1991 Thematic proto-roles and argument selection. Language 67: 547–
619.
Engelkamp, Johannes
1976 Satz und Bedeutung. Stuttgart: Kohlhammer.
Ferretti, Todd R., Ken MacRae, and Andrea Hatherell
2001 Integrating verbs, situation schemas, and thematic role concepts.
Journal of Memory and Language 44: 516–547.
Fillmore, Charles
1968 The case for case. In Universals in Linguistic Theory, Emmon Bach,
and Robert T. Harms (eds.), 10–88. New York: Holt, Rinehart and
Winston.
Garrod, Simon C., Edward J. O’Brien, Robin K. Morris, and Keith Rayner
1990 Elaborative inferences as an active or passive process. Journal of
Experimental Psychology: Learning, Memory, and Cognition 16:
250–257.
Haegeman, Liliane
1991 Introduction to Government and Binding Theory. Oxford: Blackwell.
Helbig, Gerhard
1992 Probleme der Valenz- und Kasustheorie. Tübingen: Niemeyer.
Helbig, Gerhard, and Wolfgang Schenkel
1968 Wörterbuch zur Valenz und Distribution deutscher Verben. Leipzig:
Verlag Enzyklopädie.
Herbst, Thomas, David Heath, Ian F. Roe, and Dieter Götz
2004 A Valency Dictionary of English. A Corpus-Based Analysis of the
Complementation Patterns of English Verbs, Nouns, and Adjectives.
Berlin: Mouton de Gruyter.
Herbst, Thomas, and Ian F. Roe
1996 How obligatory are obligatory complements? An alternative ap-
proach to the categorization of subjects and other complements in
valency grammar. English Studies 2: 179–199.
Hörmann, Hans
1979 Psycholinguistics. An Introduction to Research and Theory. 2d ed.
New York: Springer.
Jackendoff, Ray
1987 The status of thematic relations in linguistic theory. Linguistic In-
quiry 28: 369–412.
1990 Semantic Structures. Cambridge, MA: MIT Press.
Johnson-Laird, Philip N.
1983 Mental Models. Towards a Cognitive Science of Language, Infer-
ence, and Consciousness. Cambridge: Cambridge University Press.
Kintsch, Walter
1974 The Representation of Meaning in Memory. Hillsdale: Erlbaum.
 Valency and cognition 181

Kintsch, Walter, and Janice M. Keenan

1973 Reading rate and retention as a function of the number of the propo-
sitions in the base structure of sentences. Cognitive Psychology 5:
257–274.
Kintsch, Walter, and Teun A. van Dijk
1978 Toward a model of text comprehension and production. Psychologi-
cal Review 85: 363–394.
Knöferle, Pia, Matthew W. Crocker, Christoph Scheepers, and Martin J. Pickering
2005 The influence of the immediate visual context on incremental the-
matic role assignment: Evidence from eye movements in depicted
events. Cognition 95: 95–127.
Lyons, John
1981 Language and Linguistics. An Introduction. Cambridge: Cambridge
University Press.
Mauner, Gail, Michael K. Tanenhaus, and Gregory N. Carlson
1995 Implicit arguments in sentence processing. Journal of Memory and
Language 34: 357–382.
Obliers, Rainer
1985 Zur Revision prädikatzentrierter Satztheorien. Archiv für Psycholo-
gie 137: 175–200.
Raue, Burkhardt, and Johannes Engelkamp
1977 Gedächtnispsychologische Aspekte der Verbvalenz. Archiv für Psy-
chologie 129: 157–174.
Rickheit, Gert
1975 Zur Entwicklung der Syntax im Grundschulalter. Düsseldorf:
Schwann.
1978 Zur Syntax gesprochener und geschriebener Sprache acht- bis zehn-
jähriger Kinder. In Die Lernfelder des Lernbereichs Sprache in der
Primarstufe, Walter Popp (ed.), 150–174. Heidelberg: Winter.
Rickheit, Gert, Udo Günther, and Lorenz Sichelschmidt
1992 Coherence and coordination in written text: Reading time studies. In
Cooperating with Written Texts. The Pragmatics and Comprehen-
sion of Written Texts, Dieter Stein (ed.), 103–127. Berlin: Mouton de
Gruyter.
Rickheit, Gert, and Lorenz Sichelschmidt
1999 Mental models: Some answers, some questions, some suggestions. In
Mental Models in Discourse Processing and Reasoning, Gert Rick-
heit, and Christopher Habel (eds.), 3–49. Amsterdam: North-
Holland.
Rickheit, Gert, Lorenz Sichelschmidt, and Hans Strohner
2002 Psycholinguistik. Tübingen: Stauffenburg.
Rickheit, Gert, and Hans Strohner
2003 Inferenzen. In Psycholinguistk. Ein internationales Handbuch, Gert
Rickheit, Theo Herrmann, and Werner Deutsch (eds.). Berlin: Mou-
ton de Gruyter.
182 Gert Rickheit and Lorenz Sichelschmidt

Sanford, Anthony J., and Simon C. Garrod

1994 Selective processes in text understanding. In Handbook of Psycho-
linguistics, Morton A. Gernsbacher (ed.), 699–719. San Diego: Aca-
demic Press.
Schumacher, Helmut
1986 Verben in Feldern. Valenzwörterbuch zur Syntax und Semantik deut-
scher Verben. Berlin: de Gruyter.
Shapiro, Lewis P., H. Nicholas Nagel, and Beth A. Levin
1993 Preferences for a verb’s complements and their use in sentence proc-
essing. Journal of Memory and Language 32: 96–114.
Sichelschmidt, Lorenz, and Elena Carbone
2003 Experimentelle Methoden. In Psycholinguistik – ein internationales
Handbuch, Gert Rickheit, Theo Herrmann, and Werner Deutsch
(eds.), 115–124. Berlin: Mouton de Gruyter.
Somers, Harold
1987 Valency and Case in Computational Linguistics. Edinburgh: Edin-
burgh University Press.
Tesnière, Lucien
1959 Éléments de syntaxe structurale. Paris: Klincksieck.
Welke, Klaus
1995 Dependenz, Valenz und Konstituenz. In Dependenz und Valenz,
Ludwig M. Eichinger, and Hans-Werner Eroms (eds.), 163–175.
Hamburg: Buske.
Wilczok, Karin
1973 Satzbildungs- und Satzverarbeitungsprozesse mit semantisch unter-
schiedlich spezifizierten Verben. Unpublished diploma thesis, De-
partment of Psychology. Bochum: Ruhr University.
Zwaan, Rolf A.
1999 Situation models: The mental leap into imagined worlds. Current
Directions in Psychological Science 8: 15–18.
Valency grammar in mind
Rudolf Emons

0. Introduction

Why has valency grammar been so successful? Or to put the question more
modestly: why has valency grammar survived? My first rather general an-
swer is: it has been so successful because it has been a valency grammar
and not a valency theory.
The conference program mentions the concept of valency theory only
once, in the title of the concluding discussion: “The future of valency / Die
Zukunft der Valenztheorie”, suggesting perhaps that what in Anglo-Saxon
countries is just valency is valency theory in thorough and deep-thinking
Germany. In the past a seeming or apparent lack of theory building was a
decisive reason for many of our colleagues to hold valency grammar in not
too high esteem.
My aim is not to present different theoretical valency concepts here nor
to discuss anew if the concepts of Tesnière, Fillmore, Heringer and others
merit the denomination of theory or not. I rather want to provide some pro-
legomena for a real theoretical foundation of valency grammar. The theo-
retical foundation of valency grammar which I have in mind is not rooted in
linguistics as it is taught at the vast majority of our universities.
It is very welcome that some linguists in the field of cognitive linguis-
tics have taken important steps towards linking language and mind. How-
ever, it seems to me that these efforts still lack real interdisciplinarity. By
real interdisciplinarity I mean a program like the one that Nobel Prize win-
ning biologist Gerald Edelman has described in his book Bright air, bril-
liant fire – On the matter of the mind:

It is high time for another view of the mental, for a neuroscientific model of
the mind. What makes the one proposed here new is that it is based re-
morselessly on physics and biology. It is also based on the ideas of evolu-
tionary morphology and selection, and it rejects the notion that a syntactical
description of mental operations and representations … suffices to explain
the mind. Others have held similar positions but have not united them in a
single evolutionarily based theory, one that connects embryology, morphol-
ogy, physiology, and psychology. (Edelman 1992: 147)
184 Rudolf Emons

1. Colours

Edelman’s program is also a preamble to a new kind of linguistics.1 A shift

of focus takes place here. The focus is now on the construction of social
reality and on institutional facts. John Searle regards language as “essen-
tially constitutive of institutional reality” (1995: 59). Thus language is a
very important and a very complex social construct. This construct has to
be theoretically explained through the interplay of biological evolution and
of cultural evolution. I would like to explain and refine this program now.
Let me start with a quote from a work by the neuroscientist and evolu-
tionary anthropologist Terrence Deacon, whose theory on language devel-
opment plays a central role in my considerations:

Language is a social phenomenon. To consider it in purely formal, psycho-

logical, or neurobiological terms is to strip away its reason for being. Social
phenomena like language cannot be adequately explained without appealing
to a social evolutionary dynamic as well as a biological one. The source of
information that is used to “grow” a language lies neither in the corpus of
texts and corrections presented to the child, nor in the child’s brain to begin
with. It is highly distributed across myriad interactions between children’s
learning and the evolution of a language community. (Deacon 1997: 115)

This lesson about the reason for the existence of linguistic universals – and
predicates and their valency are such a universal – is more difficult to learn
than some linguists have thought so far. There are grammatical universals
that are, however, not just stored as such in our brains. So how does this
work?
Deacon cites as an example the development of colour terms in different
languages and societies and sheds new light on them. He takes the follow-
ing considerations as a starting point: the number of possible mappings of
colours and their linguistic terms is nearly unconstrained,

because: (a) the name of a color can be any combination of human vocal
sounds; (b) the human eye can see every gradation of color between certain
limits of wavelength; and (c) people can assign any term to any point on the
visible light spectrum. So any association between utterable sound and per-
ceivable light frequency is possible, in principle. (Deacon 1997: 116)

Reality, however, is quite different: “The mappings of colour terms to light

frequencies are not only limited, they are essentially universal in many
respects.” (Deacon 1997: 116). Prototype semantics shows:
 Valency grammar in mind 185

The best physical exemplars of particular color terms, irrespective of how

many are present in the language, are essentially the same when chosen
from an arbitrary set of color samples. In other words, though the bounda-
ries between one color and the next for which one has a term tend to be
graded and fuzzy, … color terms do apparently have something like a cate-
gory center; a best red or best green. Surprisingly, the best red and the best
green, whatever the terms used, are essentially agreed upon by people from
around the world. Though the words themselves are arbitrary and the colors
continually grade into one another, words are not arbitrarily mapped to
points on the color spectrum. They are universally constrained. (Deacon
1997: 117)

The explanation for this astonishing phenomenon lies in our brain physiol-
ogy and is – very briefly – as follows: different wavelengths of light acti-
vate different kinds of receptors in the human eye and cause different neu-
ronal responses in the brain for different colours.
The basis for universality lies in the interplay of sensory input and the
brainphysiological processing of this input. Sensory input as well as brain-
physiological processing are species-specific and are therefore obligatorily
the same for all members of this species. So all languages name colours in
practically the same way and cannot use the theoretical naming potential
mentioned above, because nobody can choose to perceive for example red
as green or vice versa. Colour blindness or red/green weakness therefore
are not problems of deficient language acquisition but are rooted in a bio-
logical defect.

2. Not seeing the trees for the wood

When we apply an evolutionary matrix to language acquisition and thereby

compare our species to other primate species that do not have language, we
will find that a major difference between us and them is that we are a sym-
bolic species, as Deacon so aptly puts it in the title of his book. Human
communication does not work along just indexical lines but symbolically.
We do not only know that certain tokens refer to certain objects but we also
establish logical relations between the linguistic signs. Language is a sys-
tem. Symbols are difficult to acquire. Nevertheless each child learns a lan-
guage. Universal grammar has every reason to ask here why this is so. With
knowledge of our brain physiology and of neuronal processes in our brain
the answer, however, will be quite different from the answer that universal
grammar is prepared to give. Deacon provides a fascinating answer:
186 Rudolf Emons

Symbol learning in general has many features that are similar to the prob-
lem of learning the complex and indirect statistical architecture of syntax.
This parallel is hardly a coincidence, because grammar and syntax inherit
the constraints implicit in the logic of symbol-symbol relationships. These
are not, in fact, separate learning problems, because systematic syntactic
regularities are essential to ease the discovery of the combinatorial logic
underlying symbols. (Deacon 1997: 136)

A quick learner who learns details very well has great difficulties with the
concept of symbolicity. He matches things to many indexical signs but he
does not need the qualitative leap that allows him to establish logical rela-
tions between the signs themselves. A less gifted learner, who does not
perceive details so well, paradoxically seems to have a great advantage in
this respect: he takes the bigger picture without bothering about details.
Children are learners of this latter type. It is easier for an immature brain to
acquire a symbol as such than it is for a more mature brain. Language struc-
tures themselves depend on biological evolution because they are adapted
to children’s learning structures. Valency structures are such adapted struc-
tures. Second language acquisition becomes more difficult than first lan-
guage acquisition because our neural cognitive resources have already been
used up for first language acquisition. We all know that language change
can take place at a much greater evolutionary rate than biological, genetic
change:

The relative slowness of evolutionary genetic change compared to language

change guarantees that only the most invariant and general features of lan-
guage will persist long enough to contribute any significant consistent effect
on long-term brain evolution. … For these reasons there is little possibility
for mental adaptations to specific syntactic structures. But there are many
features of languages, from the presence of words and sentence units to the
noun-part/verb-part distinction and many more subtle and idiosyncratic fea-
tures that are common to essentially every natural language. (Deacon 1997:
329)

Valency is such a general feature. Constant und invariant neural processes

must correspond to such constant and invariant language features. Now we
have to take a very careful look at the biological logic: linguistic universals
are “by their nature” (Deacon 1997: 333) very variable in their surface rep-
resentations. Therefore, those varying surface features have tended not to
develop specific neural supports in the brain. The separation of deep and
surface structure – no matter what form it takes – together with the assump-
tion of the universality of deep structures is virtually impossible from the
perspective of brain physiology and evolutionary psychology (Deacon
 Valency grammar in mind 187

1997: 333). In the light of symbolicity, the evolutionary development of

valency would have been like this: “The earliest symbolic systems would
necessarily have been combinatorial and would have exhibited something
like this operator-operand structure (and probably subject-predicate struc-
ture) right from the start. This is the minimum requirement to make the
transition from indexical to symbolic reference” (Deacon 1997: 334).
This is intimately related to the development of consciousness in gen-
eral, not only in human beings: “The ability to construct a scene related to
the value-category history of an individual marks the appearance of the
self” (Edelman 2004: 132). So valency is plausibly a deep structure and a
surface structure.

3. Human cognition

I would like to start this section by approaching the program of cognitive

linguistics from a certain angle. Michael Tomasello (1999) draws on work
in cognitive and functional linguistics like Langacker (1987) and at the
same time transcends this work by also integrating biological research.

The central theoretical point is that linguistic symbols embody the myriad
ways of construing the world intersubjectively that have accumulated in a
culture over historical time, and the process of acquiring the conventional
use of these symbolic artefacts, and so internalizing these construals, fun-
damentally transforms the nature of children’s cognitive representations.
(Tomasello 1999: 95−96)

Tomasello’s theoretical assumptions in this area are fundamentally differ-

ent from those of the primatologist Frans de Waal (2001). De Waal empha-
sizes the cognitive similarities between apes and humans, whereas
Tomasello emphasizes the cognitive differences between the species. I will
not go into any detail here. But de Waal’s study would also be part of the
framework of a new linguistics.
The valency grammarian will read Tomasello’s statements on verb is-
land constructions with great satisfaction:

As children begin to produce utterances that have more than one level of
organization, that is, as they begin to produce utterances with multiple
meaningful components, the most interesting question cognitively is how
they use those component parts to linguistically partition the experiential
scene as a whole into its constituent elements – including especially the
event (or state of affairs) and participants involved. And ultimately children
must also learn ways to symbolically indicate the different roles the partici-
188 Rudolf Emons

pants are playing in the event, such things as agent, patient, instrument, and
the like. (Tomasello 1999: 138)

He goes on:

The verb island hypothesis proposes that children’s early linguistic compe-
tence is comprised totally of an inventory of linguistic constructions of this
type: specific verbs with slots for participants whose roles are symbolically
marked on an individual basis …. At this early stage children have made no
generalizations about constructional patterns across verbs, and so they have
no verb-general linguistic categories, schemas, or marking conventions … .
To repeat, the inventory of verb island constructions – in effect a simple list
of constructions organized around individual verbs – makes up the totality
of children’s early linguistic competence; there are no other hidden princi-
ples, parameters, linguistic categories, or schemas that generate sentences.
This item-specific way of using language is not something that goes
away quickly. Indeed, in the view of many linguists, more of adult language
is item-specific than is generally realized, including idioms, clichés, habit-
ual collocations, and many other “non-core” linguistic constructions (e.g.,
How ya doing, He put her up to it, She’ll get over it; Bolinger, 1977; Fill-
more, Kay, and O’Connor, 1988). (Tomasello 1999: 139−140)

One very important point here is that according to Tomasello children start
from very concrete structures and proceed to more and more abstract con-
structions:

The mastery of verb island constructions is a major way station on the road
to adult linguistic competence – a kind of base camp that is the goal of the
early part of the journey but that, once reached, becomes only a means to
the end of more abstract and productive linguistic constructions. (Tomasello
1999: 141)

4. From head to foot

In the light of Deacon’s theory on language development this is a rather

traditional view of the development of syntax or – more specifically – of
the cognitive development of valency structures in children. I will therefore
try to turn Tomasello’s theory on its foot in this section. Tomasello seems
to get into some trouble with his traditional concept. Take the following
quotation: “Their [children’s] sentence-level constructions are verb island
constructions that are abstract with respect to the participants involved
(they have open participant slots) but are totally concrete with respect to the
 Valency grammar in mind 189

relational structure as expressed by the verb and syntactic symbols (word

order and grammatical case marking)” (Tomasello 1999: 140).
By any reasonable understanding of the meanings of abstract and con-
crete it is the relational structure that we would regard as abstract. Also the
view that idioms and habitual collocations are item-specific or concrete
would have to deal with the fact that patients with severe brain damage are
still able to reproduce those “specific” items despite the fact that these pa-
tients have otherwise very limited cognitive abilities and are no longer able
to perceive the specificity of certain situations.
So it seems to be necessary to turn Tomasello’s theory upside down in
this respect. I would like to suggest a new interpretation of the acquisition
of verb island constructions:
The base camp children start from does not consist of specific and indi-
vidual verb structures but rather of an abstract valency concept. Thus, the
child does not proceed from trees to forest or from verbs to valency, but it
is rather the other way round. This also seems to fit much better into
Tomasello’s general model of children’s cognition (1999: 180).

first stage young infants understanding others as animate beings

second stage nine months old understanding others as intentional agents
third stage four year olds understanding others as mental agents

This model also proceeds from more general to more and more specific
perceptions.

5. Valency grammar in mind?

I would like to begin this last section by putting a question mark behind the
main title of my article: valency grammar in mind? In addition, I would like
to raise two further questions in a similar form: valency theory in mind?
and: valency in mind? The logical sequence of these questions would be
theory, grammar, valency. What are the answers to these questions?

5.1. Valency theory in mind?

A really trivial answer is that someone who talks about a theory must have
this theory in mind in one form or the other. So it is “yes”.
A non-trivial answer can be given by making the following assumption:
someone is unable to utter anything whatever about valency theory nor
190 Rudolf Emons

does she consciously know anything about it, yet she has valency theory in
some innate or subconscious way in her mind: she cognizes valency theory.
Modern research in the field of brain physiology and consciousness shows
quite clearly that this assumption cannot be valid for a theory. So it is “no”
for this case.

5.2. Valency grammar in mind?

The trivial answer is the same as for the theory question.

A non-trivial answer can be given against the background of Deacon’s
assumptions on the connection between neuronal brain structures and rela-
tively variable language structures. The English and the German valency
grammars exhibit quite different structures, and so the assumption that we
have a valency grammar in mind seems rather improbable. So it seems to
be “no” again.

5.3. Valency in mind?

The interesting, non-trivial answer is “yes”. Valency is a universal concept

and it is precisely because of this universality that it is present in the neu-
ronal structures of our brains. This is of course restricted to our species, but
we will find forerunners of valency in other species, too.2
So let me return to the beginning of this article. People working in the
field of valency are invited to transcend the traditional linguistic limits.
They will then come to an interdisciplinary view of the concept of valency
and can happily combine their grammatical work with work in sociology
and biology. In the light of Edelman’s program the answer to the question
if there is a Valenztheorie or theory of valency will get a very clear answer:
“yes”, there is a modern theory of valency and it is a model for a new kind
of linguistics.

Notes

1. West has applied Edelman’s Theory of Neuronal Group Selection very fruit-
fully to valency grammar and language and linguistics in general, relating it to
knowledge and memory (2003: 272), and thus explaining the fuzziness of the
distinction between complements and adjuncts with reference to neuronal
processes within the brain. A different and more general focus on biological
and brain processes is adopted in the following essay.
 Valency grammar in mind 191

2. See Edelman (2004: ch. 9) for the causal interplay between brains, mental
imagery, concepts and meaning.

References

Deacon, Terrence
1997 The Symbolic Species: The Co-evolution of Language and the Brain.
New York: Norton.
De Waal, Frans
2001 The Ape and the Sushi Master: Cultural Reflections by a Primatolo-
gist. New York: Basic Books.
Edelman, Gerald
1992 Bright Air, Brilliant Fire – On the Matter of the Mind. New York:
Basic Books.
2004 Wider than the Sky: The Phenomenal Gift of Consciousness. London:
Penguin.
Langacker, Ronald
1987 Foundations of Cognitive Grammar. Stanford: Stanford University
Press.
Searle, John
1995 The Construction of Social Reality. London: Penguin.
Tomasello, Michael
1999 The Cultural Origins of Human Cognition. Cambridge, Mass.: Har-
vard University Press.
West, Jonathan
2003 What can valency tell us about linguistic theory? In Valency in Prac-
tice: Valenz in der Praxis, Alan Cornell, Klaus Fischer, and Ian F.
Roe (eds.), (German Linguistic and Cultural Studies 10.) Oxford:
Lang.
The acquisition of argument structure1

Heike Behrens

1. Introduction

The concept of valency or argument structure is a powerful one in linguis-

tics, although the current volume shows that there is still considerable de-
bate as to how to characterise the valency of any given verb exactly. But if
professional linguists and lexicographers encounter difficulties in defining
the relationship between a verb’s meaning(s) and its syntactic properties,
how can a two-year-old manage?
Research on child language has focussed on argument structure or logi-
cal and syntactic valency rather than on semantic valency, that is the speci-
fication of the semantics of the arguments. This reflects the anglophone
dominance in the field, but also emphasises the focus of interest, namely
the role of the verb in the clause and the syntactic positions it opens. Con-
sequently, I will follow this tradition and use the term argument structure
rather than valency to refer to the acquisition research.
Argument structure acquisition has been a popular topic for the past 25
years, with shifting focus of attention. In the 1980s, a number of deductive
accounts were proposed to explain which kind of knowledge helps children
to identify the arguments a verb requires. These approaches relied on con-
ceptual, semantic or syntactic cores, which could be universal and / or in-
nate, and assume modular levels of representation. I will summarise these
accounts under “bootstrapping” accounts, a metaphor used to explain how
children could use information on one level of representation in order to get
started (or to bootstrap in the technical sense) onto another level.
More recently, inductive learning accounts have gained popularity. In
this view, children accumulate knowledge through usage events and derive
generalizations about a given verb’s syntactic and semantic properties only
gradually. I will discuss such proposals under the heading “usage-based
accounts” because they assume that children gain their knowledge about
argument structure from observing the concrete usages of verbs in concrete
discourse situations. Two types of approaches are of interest here. First,
Construction Grammar accounts assume that the construction (a meaning-
ful form-function unit) is the primary source of information, from which
194 Heike Behrens

the properties of individual verbs can be derived. Second, there is research

on the discourse and informativeness factors which determine argument
realization in connected speech. These investigations use the concept of
Preferred Argument Structure.
Since much research is ongoing, and since new results especially about
crosslinguistic differences in argument structure are likely to lead to some
modifications of earlier accounts, this paper provides pointers to previous
and current research, rather than elaborating one of these aspects and theo-
ries in detail.

2. Deductive accounts of the acquisition of argument structure

The concept of argument structure assumes that verbs open up a number of

semantically and syntactically specified positions. Typically, a verb like put
opens positions for the putter, the thing being put and the location where
the thing is put, as in I put the book on the table, whereas a verb like see
opens two positions, namely the seer and the object being seen (I see a
boat), but not a position of the location of the seeing or the object. This
entails a relationship between events and the semantics of the verbs that
encode these events, as well as a semantics-to-syntax mapping for these
verbs.
Because argument structure seems such a logical and systematic con-
cept, it is not surprising that researchers have made use of this concept for
language acquisition. If argument structure is systematic, i.e., if there is a
predictable relationship between a verb’s semantics and the syntactic
frames it occurs in, this relationship could provide a stepping stone for lan-
guage learning because the verb “tells” the child about the linguistic items
it goes with. But before observing the systematic syntax-semantics link,
children could even make a connection between the event structure they
observe in the preverbal phase, and possible argument structure patterns.
Thus, if there is a systematic relationship between events, verb semantics
and verb syntax, there are three possible entryways into the linguistic sys-
tem. The assumption of such links leads to so-called bootstrapping ac-
counts that predict that children use knowledge of one level of representa-
tion to bootstrap another level.
 The acquisition of argument structure 195

2.1. Conceptual starting points

Dan Slobin (1985) proposed a conceptual account for the acquisition of

early syntactic relations. He argued that children all over the world will
construe similar event representations and build up similar concepts, which
will serve as the basis for linguistic encoding. That is, children learn to
categorise events in the preverbal stage and try to find the linguistic entities
that encode the participants in the event. For example, a common event is
the so-called Manipulative Activity Scene where an agent does something to
a patient. Children will form categories of such events and map them onto
two-argument verbs.

The central claim is that LMC [= language making capacity] constructs

similar early grammars from all input languages. The surface forms will, of
course, vary. What is constant are the basic notions that first receive gram-
matical expression, along with early constraints on the positioning of
grammatical elements and the way in which they relate to syntactic expres-
sion. (Slobin 1985: 1161)

Transitive sentences will thus denote Manipulative Activity Scenes, before

alternative, language-specific form-function mappings overrule this early
alignment. That is, deviant language-specific patterns should be learned
only in a second step.
Slobin’s view was criticised by Melissa Bowerman, and later withdrawn
by Slobin himself. Bowerman (1985) argues that children do not prefer
Manipulative Activity Scene in their early transitive sentences. She con-
cludes that there is no semantic basis for the acquisition of grammar. Based
on a larger body of typological research and cross-linguistic acquisition
studies, Slobin (1997, 2001) criticised his earlier views (Slobin 1985). He
argues that there is no evidence for privileged grammatizable notions. In-
stead, children seem to be able to pick up any form-function relationship.
In a different vein, several authors suggested that children start out with
conceptually simple, general-purpose verbs: Eve Clark (1993) formulated
the “light verb hypothesis”, and Anat Ninio (1999) the “pathbreaking verb
hypothesis”. These hypotheses claim that children start verb learning with
semantically light verbs like make, do, and put, which can serve as
pathbreaking verbs, and acquire semantically specific verbs only later.
Thus, children initially only need fairly unspecific conceptual notions that
serve a wide range of purposes, and differentiate their early unspecific con-
cepts through extended exposure to a target language. However, this view
was criticised from a crosslinguistic perspective because not all languages
have a light verb vocabulary; e.g., Tzeltal (Mayan) has a vast array of se-
196 Heike Behrens

mantically specific verbs, and it turns out that children from early on make
a number of specific distinctions. Brown (1998) shows that children from
early on have a rich and semantically rather specific inventory of eating
verbs, for example, verbs to mean the equivalent of eat soft things versus
eat tortillas or eat crunchy things. In a study of children’s early usage of
these verbs, Brown (2007) shows that children do not overgeneralise these
semantically specific verbs to mean something more general, but use them
adequately.
In sum, crosslinguistic analyses suggest that children make use of the
affordances of the language they acquire. If a language has light verbs, they
tend to be used frequently, although more specific verbs are used as well.
But if a language does not provide such verbs, children quickly build up a
rich repertoire of semantically specific verbs. To date, there is no evidence
for a privileged conceptual starting point for the acquisition of verb syntax
and semantics.
In the 1990s, two different approaches were much discussed. These
approaches focus more narrowly on the mapping of syntax and semantics
(cf. the review by Bowerman and Brown 2007). The so-called “semantic
bootstrapping hypothesis” predicts that innate linking rules help the child to
map the verb syntax onto already known verb semantics. The “syntactic
bootstrapping hypothesis”, in contrast, tries to explain how children can
acquire and refine their knowledge of verb semantics by paying attention to
the syntactic frames a verb is used in.

2.2. Semantic bootstrapping

Pinker (1994: 378) proposed that children use semantics to acquire syntax,
because meaning is constrained by semantic factors:

In the case of learning verb meanings, … not all logically possible constru-
als of a situation can be psychologically plausible candidates for the mean-
ing of a word. Instead, the hypotheses that a child’s word learning mecha-
nisms make available are constrained.

In his semantic bootstrapping hypothesis, Pinker (1989: 62) states that syn-
tactic argument structure is predictable from semantic structure via the
application of linking rules. The constraints on verb meaning interact with
syntax in a systematic way: in the mental lexicon, verbs have rich semantic
specifications. They project verb syntax by means of innate linking rules.
The verb hit, for example, calls for an agent and a patient argument. Link-
 The acquisition of argument structure 197

ing rules align the thematic roles to syntactically specified subjects and
objects as in (1):

(1) Bert hits Ernie.

AGENT PATIENT lexical representation
↓ ↓ linking-rules
SUBJECT OBJECT syntactic structure

The hypothesis predicts that verbs with high semantic typicality should
form the starting point for the acquisition of argument structure. For exam-
ple, verbs with high semantic transitivity should be most easily aligned
with transitive sentence structure.
Bowerman (1990) analysed English children’s early transitive construc-
tions and examined these predictions. She found that “best exemplars” are
not acquired first. Instead, verbs like have and see are among children’s
early transitive verbs. They are high-frequency verbs, but have a non-
prototypical linking between theta roles and syntactic structure, since the
subjects are not typical agents and the patients are not typical patients. In a
subsequent study, Bowerman (1996) examined the predictive power of
Pinker’s lexical rules for causative constructions. She analysed error pat-
terns in children’s encoding of causativity (as in I disappeared the ball) and
found that the group of verbs that show errors differs from the group for
which errors are predicted. In addition, Bowerman addresses the problem
of cutting back on overgeneralizations. Some English verbs have alterna-
tive valency patterns and can be used intransitively or transitively (2a, b):

(2) a. The stick breaks easily.

b. John breaks the stick.

But alternating patterns can also be overgeneralised as in the following

examples of non-alternating verbs (3a, b; Bowerman 1996: 454).

(3) a. Button me the rest.

b. I said her no.

Pinker (1989) proposed that broad-range linking rules, based on semantic

categories, provide the necessary conditions for alternation. In order to
account for the fact that some verbs do not alternate although they fit the
semantic pattern, a set of more specific narrow-range linking rules is in-
voked which provides the sufficient conditions for distinguishing alternat-
ing and non-alternating verbs. Bowerman’s (1996) summary of the data
198 Heike Behrens

makes it seem unlikely that acquisition of argument structure patterns can

be explained in terms of the interaction of broad- and narrow-range rules,
especially since there is no evidence that children adhere to strict semantic
groupings in the early stages of learning argument structure. Bowerman
argues that instead children work with overly general assumptions about
argument structure and have to learn to cut back on such errors.
One explanation for the low frequency of errors and their eventual dis-
appearance is pre-emption. This means that errors are blocked because
another verb or a related construction already occupies the semantic posi-
tion of the possible alternate. When children know the construction make
disappear, errors of the type I disappeared the cake will not occur because
the semantic position is already filled. Pre-emption predicts that verbs for
which the child knows the alternate construction should be less error-prone
than verbs for which the child does not know the alternate construction. But
a longitudinal study of two girls learning English showed that this is not the
case (Bowerman 1996: 463-464).
Instead, usage-based factors could account for the relative infrequency
of such errors, as well as for the disappearance of such errors, because re-
peated exposure to intransitive syntactic frames reduces the tendency to use
verbs transitively (see also MacWhinney 1987).
The investigation of how the verb meaning can help to narrow down
possible syntactic frames is just one side of the coin. If there is a predict-
able relationship between syntax and semantics, the process should work in
the other direction as well such that knowledge of syntax should help to
narrow down the possible meanings of a verb.

2.3. Syntactic bootstrapping

The syntactic bootstrapping hypothesis (Gleitman 1990) states that the syn-
tactic frames a given verb occurs in are more informative about its seman-
tics than a linking of the event itself to semantics, because any event is
open to several ways of highlighting event participants. We can encode a
“shopping” event from the perspective of the buyer (Peter buys a book), the
seller (Peter sells a book to Paula), or the object (The book cost Paula
10$). Observation of an event alone does not help us to identify the linguis-
tic perspective taken on an event. Moreover, Landau and Gleitman (1985)
showed that blind children acquired the semantics of different verbs of
vision, which demonstrates that the acquisition of verb semantics does not
depend on the observation of events, but on the exploitation of linguistic
structure. Gleitman (1990) predicts that the syntactic frames a given verb
 The acquisition of argument structure 199

occurs in are systematically linked to verb semantics. For this hypothesis to

work there needs to be a close alignment of argument structure and seman-
tics:

Verbs that describe externally caused transfer or change of possessor of an

object from place to place (or from person to person) fit naturally into sen-
tences with three noun phrases, for example, ‘John put the ball on the table’.
This is just the kind of transparent syntax/semantics relation that every
known language seems to embody. It is therefore not too wild to conjecture
that this relationship is part of the original presuppositional structure that
children bring to the language learning task. (Gleitman 1990: 30)

Subsequent research by Naigles, Gleitman and Gleitman (1993) showed

that children (mean age 2;9) can modify the verb meaning of familiar
words when they hear it in a novel frame; e.g., they are likely to interpret
the sentence The zebra goes the lion in a causative reading (‘the zebra
makes the lion go’) in analogy to other cases where this structure encodes
causativity. In a literature review on early verb knowledge, Naigles (2002)
states that “form is easy, meaning is hard”. Infants are good at processing
form patterns (segmental, prosodic, structural), less good in handling se-
mantic information. However, Naigles argues, later in development the
occurrence of a verb in different formats or syntactic frames helps the child
to narrow down the semantics (cf. also Naigles 1996).
Several authors found problems with the syntactic bootstrapping ac-
count. Pinker (1994: 382) criticised Gleitman’s strong reliance on syntactic
structure for the inference of verb meaning, because the other words in the
sentence carry meaning as well. He argues that in sentences like I filped the
delicious sandwich and now I’m full, the meaning of the pseudo-verb filp
can be inferred from the lexical knowledge of the other words in the sen-
tence without reliance on syntactic structure. This may explain why blind
children learn verb semantics without having access to visual information.
Furthermore, Pinker takes issue with Gleitman’s claim that meaning cannot
be learnt from observation of a word’s usage in concrete contexts. Rather,
for some semantically related verbs with the same argument structure, only
the context can disambiguate subtle semantic differences; e.g., real world
experience is needed to distinguish the manner of actions (e.g., open versus
close, tear versus break, Pinker 1994: 394). Pinker’s conclusion is that
Gleitman’s arguments are void if one assumes that children’s word mean-
ings are universally constrained such that they will not come up with non-
sensical hypotheses about word meanings. Then, context information will
provide sufficient information to derive the meaning distinctions between
different verbs.
200 Heike Behrens

Wilkins (2007) points out another problem with the Gleitman’s assump-
tion that the syntax-semantics alignment of verbs is part of the “original
presuppositional structure that children bring to the language learning
task”, because argument structure patterns are not the same crosslinguisti-
cally. Wilkins looked at the equivalents of the verbs look and put in Ar-
rente, an Australian Aboriginal language. In English, the perception verb
look is a classic example of a two-place predicate (agent and object), and
the transfer verb put is a classic example for a three-place predicate (agent,
object, location). In Arrente, however, by all linguistic tests, the verbs ar-
rerne [‘look’] and are [‘put’] are three-place predicates that open positions
for agent, object and location of the object put or seen. In the case of the
verb look, the resultant meaning can, for example, come close to the Eng-
lish verb find (to see something somewhere denotes ‘find it’). Thus, syntac-
tic bootstrapping accounts would fail with Arrente, because there is no
“natural” alignment between argument structure and verb semantics as
proposed by Gleitman. Nonetheless, in a corpus analysis of spoken Arrente,
Wilkins (2007) found that adults use these verbs in a different fashion: look
is used as a two-place predicate more often than put. For put, the locative
NP is realised more frequently than for look. Children follow this usage,
and have particular problems with the third argument for look. These find-
ings suggest that while there is no strict alignment between syntax and se-
mantics that would allow children to bootstrap from syntax to semantics, in
actual usage some argument structure realizations are more common than
others. Children may use such distributional differences to induce verb
meaning.
So far, we have seen that both syntactic and semantic bootstrapping do
not work in a deductive way: the link between semantics and syntax regard-
ing argument structure is not tight enough to allow full predictability. Con-
sequently, inductive accounts of language acquisition gained ground in the
past decade and can now be considered the dominant framework in acquisi-
tion.

3. Inductive accounts for language acquisition

The discussion so far has demonstrated that there seems to be little support
for theories that assume a tight link between verb semantics and syntax that
could be used to predict either syntax or semantics. But then how could
children acquire argument structure? Alternative theories known under
headings such as usage-based theories (Tomasello 2003) or emergentism
(Elman et al. 1996; MacWhinney 1999) focus on the learning and categori-
 The acquisition of argument structure 201

zation mechanisms itself. These theories assume that complex cognitive

patterns can be induced from noticing distributional properties of the input
language (Elman 2003). In addition to being able to use such probabilistic
cues as early as in infancy (Saffran 2003; Gomez and Gerken 2000), hu-
mans also demonstrate the ability to perceive the intention of others in a
concrete situation (Tomasello and Rakoczy 2003). If children are aware of
other people’s intention, however, this will help them to narrow down the
possible meaning of what is being said. That is, the concept of the child as
an intention reader replaces the Generative Grammar concept of the child
as an hypothesis tester. Several studies have shown how intention reading
contributes to early word learning; e.g., in an experiment a child and his/her
mother played with three novel and unnamed objects. The mother went out
and the child received another novel and unnamed object. When the mother
came back in, she looked at the four objects and exclaimed “Oh look! A
modi! A modi!”. 24-month-old children significantly associated the word
modi with the fourth object. They could not have done so by simple asso-
ciation but must have used social cognition, in this case their understanding
that people get excited about new things (Akthar, Carpenter, and Tomasello
1996).

3.1. Usage-based models of syntax

Usage-based accounts of acquisition assume that learning takes place by

generalising over concrete usage events (see Tomasello 2003 for a sum-
mary). They do not draw a distinction between universal and innate core
grammar, which is acquired by deduction, and the periphery, which has to
be learnt by induction. Instead, it is supposed that all properties of lan-
guages can be acquired from the input by powerful generalization abilities
in connection with social cognition. The plausibility of usage-based learn-
ing is supported by a growing body of research which shows that even in-
fants have a remarkable capacity for pattern recognition and statistical
learning, regardless of whether the patterns are semantically motivated or
not (see Saffran 2003 and Gomez and Gerken 2000 for a general introduc-
tion, and Newport and Aslin 2004 for more detail). Furthermore, research
in computational linguistics shows that grammatical categories as well as
information about constituency can be gained by data-driven parsing, with-
out supplying “rules” to the computer (Redington, Chater, and Finch 1998;
Keibel et al. 2006; Klein and Manning 2004). Finally, comparisons be-
tween child and input data show a close alignment between input patterns
and the structures attested in children, which suggests that children pay
202 Heike Behrens

close attention to the distributional properties of language use in the ambi-

ent language (Behrens 2006).
If acquisition is based on the evidence children get from the input, a
number of predictions follow. First, acquisition should be item-specific
because children have no access to a priori verb-general categories. Second,
cross-linguistic differences are expected: if different languages show dif-
ferent alignments of syntax and semantics in language use, this should be
reflected in acquisition.

3.1.1. Crosslinguistic variation

Recent investigations into “exotic”, non Indo-European languages revealed

that there is considerable variation both in terms of argument structure
proper and in terms of argument realization. In general, high and/or sub-
stantial variation makes deductive account less plausible, because phenom-
ena with large variability call for inductive learning processes.
Typological research has pointed out that semantic specificity has an
impact on argument realization, because in lexically-specific verbs, the
verb meaning may already incorporate some arguments. Compare, for ex-
ample, the verb kick with the construction push with foot. In kick, the in-
strument foot is incorporated in the verb meaning and need not be specified
as an extra argument. For push, in contrast, agent, object and instrument
need to be specified. Consequently, languages with a richly specified verb
lexicon tend to show more argument ellipsis than languages where the verb
lexicon is rather small and semantically more general (cf. Bowerman and
Brown 2007).

3.1.2. Item-specificity

In usage-based accounts for acquisition, the notion of verb-specificity has

become very relevant. It is argued that the syntax of early child language is
item-specific rather than abstract. This hypothesis has led researchers to
reconsider the units children operate with: rather than to assume that verbs
project syntactic structures based on their semantics, it is suggested that
children work out form-function alignments based on individual verbs, and
generalise over groups of verbs only later.
Tomasello (1992) analysed his daughter’s development of verb syntax
on a verb-by-verb basis. He did not find groups of transitive or intransitive
verbs that show similar syntactic behaviour, but rather that each individual
 The acquisition of argument structure 203

verb started out with its own, lexically-specific frame. At a given point in
time, the child used the verb cut only in the frame cut X, while the syntacti-
cally similar verb draw was used in a wider range of frames (draw X, draw
X on Y, draw on X, draw X for Y; Tomasello 2003: 117). These findings led
Tomasello to propose the verb island hypothesis. It states that the best pre-
dictor for a given verb’s use is not the use of other related verbs at the same
time, but the child’s previous use of that particular verb (Tomasello 1992:
256).
The item-specificity of early child language is related to the non-
productivity of these utterances: if a child uses direct objects only with
particular verbs, but not with all kinds of verbs that take direct objects, this
may indicate that these early constructions are frozen or (semi-)formulaic
concrete lexical units, rather than represented in an analytic or abstract
fashion. Indeed, Pine, Lieven, and Rowland (1998) found that in the speech
of twelve children learning English, the five most common slot-and-frame-
patterns like mommy X or want X accounted for an average of 70% of all
utterances containing verbs.
In the usage-based framework, the lexical robustness of early child ut-
terances is considered as evidence that children operate with prefabricated
“chunks” and do not generate utterances from scratch (Tomasello 2000).
Similar conclusions can be drawn from the behaviour of individual verbs:
Theakston et al. (2002) studied the used of word-forms of go (go, goes,
going, gone, went) in eleven British children. They found little evidence for
overlap of arguments across word forms. Instead, each word form seemed
to have its own frames. In addition, children’s use highly correlated with
adult usage.
But what is the advantage of analysing child language in such a pattern-
based approach? First, it directs the attention to the communicative function
of the utterance, not to the syntactic or semantic representation of words in
isolation. It is assumed that early formulae or patterns are linked by the
same communicative function. Second, pattern-based approaches assign a
different role to verbs. Rather than seeing verbs as the core elements that
project syntax, verbs constitute just one, albeit important, aspect of com-
municative units. It is in this respect that recent acquisition theory draws a
close connection to Cognitive Linguistics in general and Construction
Grammar in particular (Tomasello 1998).
204 Heike Behrens

3.2. Constructions as predictor for language learning

In Construction Grammar, constructions are defined as entities of variable

size, which are fixed pairings of form and meaning (see Fillmore and Kay
1993; Goldberg 1995 and 2006 for theory; Tomasello 1998 for acquisition).
Tomasello (1998) claims that early acquisition is more adequately de-
scribed in terms of constructions because the linguistic knowledge underly-
ing early child language is tied to lexical items rather than being abstract or
verb-general. Different structures need not be linked by rules, but could
represent independent schemata, which may be analyzed only partially.
There is no distinction between core and non-core-phenomena or between
universal and language-specific factors because all of language is acquired
bottom-up from language use (Tomasello 1992, 2003).
This approach differs crucially from the bootstrapping accounts de-
scribed above, because the construction approach does not rely on syntactic
or semantic primitives. It neither assumes the availability of abstract syn-
tactic categories like word class or thematic roles, nor the availability of a
detailed semantic analysis in terms of primitives that constitute the basis for
syntactic acquisition. Instead, children use larger or smaller units to convey
their communicative intention. In order to do so, it is not necessary that
they have abstracted all the component parts of the construction, just like
adults use idioms where the underlying structure remains opaque. What
constructivist approaches then have to account for is how and when linguis-
tic knowledge is abstracted. Linguistic creativity in both children and adults
shows that at some point in development they are able to go beyond what
they have heard and use their knowledge of the meaning of syntactic con-
structions to use lexemes in new and productive ways (cf. Fillmore and
Kay 1993).
In fact, linguistic productivity would break down completely if children
relied only on positive evidence and only used those constructions or ar-
guments they had actually registered in the ambient language (Bowerman
1996: 464; Goldberg 1995). Recent experiments from a usage-based per-
spective therefore focus on generalization mechanisms. Goldberg and col-
leagues undertook a number of studies that tested how Construction
Grammar can be used to predict the acquisition of argument structure
(Goldberg and Casenhiser 2005a, b; Goldberg, Casenhiser, and Sethu-
ranam, 2004; for summary and theoretical elaboration see Goldberg 2006).
In a training study, 51 5−7 year-old-children were trained with a new ar-
gument structure pattern of the form “NP NP novel verb” (e.g., the spot the
king moopoed) to encode appearance (the corresponding video showed a
spot appearing on the king’s nose). Within less than three minutes, the chil-
 The acquisition of argument structure 205

dren saw 16 videos representing five new verbs. One group saw them in a
skewed exposure (one video was shown eight times, the remaining four
videos twice each). The second group had a more balanced exposure (three
videos four times, two videos twice). It turned out that the group that was
exposed to a skewed distribution generalised the new pattern best. This
result confirms earlier findings from corpus studies that showed that within
a particular syntactic construction, the distribution tends to be biased in that
one verb represents a large number of tokens of that construction (Gold-
berg, Casenhiser, and Sethuranam 2004).
Based on these findings, Goldberg (2006) argues that the role of the
construction has an important impact on acquisition. But what exactly de-
termines the predictive power of verb-based constructions? At first glance,
it seems that within a construction, verbs still have the highest predictive
power because verbs are relational elements and therefore entail sentence
meaning (i.e., who did what to whom, Goldberg 2006: 104). In a set of
experiments, Goldberg and colleagues tried to test the relative contribution
of verbs versus constructions in light of the fact that many verbs are polyse-
mous such that their occurrence in different construction types is correlated
with different meanings. Thus, under which circumstances are verbs better
predictors, and under which circumstances are constructions the better pre-
dictor of sentence meaning? Goldberg argues that this is a matter of cue
validity, a concept adapted from the competition model by Bates and
MacWhinney (1987). This model hypothesises that all linguistic structures
represent different formal and functional cues. Acquisition sequences are
determined by the cue cost, the effort it takes to detect and process this cue
(e.g., affixes are easier to detect and to segment than stem changes) and cue
validity, the degree to which this cue is a reliable cue for this phenomenon
(e.g., morphological paradigms with a 1:1 form-function correspondence
have higher cue validity than paradigms with a high degree of syncretism
and ambiguity). Regarding verb semantics, highly polysemous verbs like
get have low cue validity regarding meaning. Here, the construction type
can help to disambiguate possible readings and thus has a higher cue valid-
ity for meaning (see example 4; Goldberg 2006: 106).

(4) a. Pattern VOL: (Subj) V Obj OBlpath/loc Æ caused motion

Pat got the ball over the fence.
b. Pattern VOO: (Subj) V Obj Obj Æ transfer
Pat got Bob a cake.

Based on corpus analyses and experiments, Goldberg and colleagues con-

clude that for early acquisition, verb-argument constructions (compatible to
206 Heike Behrens

the verb-island constructions or slot-and-frame patterns discussed above)

are better predictors of sentence meaning than the verb in isolation. This
also holds for generalization in later stages of acquisition: the argument
frame is at least as good a predictor of sentence meaning as the verb itself
in isolation, because many high-frequency verbs are polysemous and have
low cue validity for meaning (see Goldberg 2006: 105−126 for a summary
of the results).
In sum, constructivist accounts point to the primary nature of the con-
struction as the main conveyor of meaning since we talk in utterances, not
in isolated words. These accounts also tend to be inductive, because they
assume a usage-based vantage point where general learning mechanisms as
well as social cognition regarding the intention of the other speaker allow
children to induce linguistic knowledge on increasingly complex and ab-
stract levels.
The impact of language use on argument structure is also studied in a
different research tradition that investigates the influence of the discourse
context on argument structure.

4. Preferred Argument Structure (PAS)

The studies reported so far focussed on a “context neutral” perspective of

argument structure: which arguments does a given verb with a particular
semantics call for? But in concrete connected speech or discourse, argu-
ments can be dropped or provided for a number of reasons. First, there are
language-specific structural reasons because some languages like Chinese
show the phenomenon of topic drop: a topic once established needs not be
encoded again, unless the topic changes. Second, there are various factors
that influence context-dependent ellipsis. An argument can be assumed as
“given”, for example, because it is visible and can be pointed to or looked
at. Furthermore, arguments need not be realised lexically, but can be en-
coded as pronouns or affixes on the verb. These factors determine argument
realization. Discourse studies have shown that “givenness” in previous
discourse is likely to lead to ellipsis or pronominal realisation, whereas
“newness” is more likely to lead to encoding by a full NP. Several re-
searchers are interested in “Preferred Argument Structure” and look at the
structural (DuBois 1987) and discourse-pragmatic factors (Clancy 1997)
that determine the number and nature of arguments that are realised in a
particular language or a particular genre.
The concept of Preferred Argument Structure can be applied to child
language. What is the effect of ellipsis, or pronominal versus lexical encod-
 The acquisition of argument structure 207

ing in the adult language on acquisition? How does a child learn which
arguments to provide and when? It is a common feature of early child lan-
guage that arguments are omitted. For example, utterances often lack the
subject as in want milk. Allen (2000: 484f.) identifies three explanations for
this phenomenon. The first comes from a Generative Grammar perspective
and hypothesises that children’s grammar is consistent with adult grammar.
In a parameter-setting version of Generative Grammar, children may as-
sume that arguments are dropped unless positive evidence in the adult lan-
guage tells them that they should be provided. Thus, the innate state would
be that the child is equipped with knowledge about the circumstances under
which arguments may be dropped (e.g., Hyams 1986). Second, perform-
ance factors are held responsible for argument omission. Researchers as-
sume that children know the argument structure of a verb, but that their
processing capacities are insufficient to handle all arguments. Thus, their
representation of argument structure is adult-like, but provision of argu-
ments is hindered by performance restrictions (e.g., Valian 1991). Thirdly,
discourse-pragmatic accounts investigate which situational factors lead to
the provision of arguments, without assuming that children’s knowledge is
adult-like (Clancy 1993).
Allen (2000) examined eight features of discourse-pragmatic promi-
nence which contribute to the relative informativeness of arguments in the
speech of four Inuktitut-speaking children aged 2;0 to 3;6. The “informa-
tiveness features” include knowledge features as well as confusion factors.
For example, if one wants to talk about an object that is absent in the physi-
cal context, it must be realised as an argument unless it has already been
established as the discourse topic. Likewise, one needs to realise arguments
that one asks questions about. But “confusion features” also lead to the
provision of arguments; e.g., if there are two or more possible referents in
the discourse context, the intended referent has to be encoded overtly.
Inuktitut, a Inuit language spoken in Northern Canada, allows for mas-
sive argument ellipsis, and children between 2;0 and 3;6 years of age only
provide about 18% of all arguments (Allen and Schröder 2000). When they
do, their provision of arguments follows the predictions of DuBois’ Pre-
ferred Argument Structure in that there is no more than one new argument
per clause, and in that lexical arguments (as opposed to demonstratives or
affixes) tend to encode new arguments (Allen and Schröder 2000). But the
rampant omission of arguments in adult language raises the question why
children provide arguments at all. Logistic regression analyses showed that
argument provision by Inuktitut children is not random (Allen 2000). A
model containing all eight features of pragmatic prominence is significantly
more accurate at predicting which arguments will be overtly represented
208 Heike Behrens

than a model containing none of these features. The presence of informa-

tiveness features also explains the overproductions of some types of argu-
ments in early child language, as well as the omission of uninformative
arguments by children where adults provide pronouns.

5. Summary and discussion

Generally speaking, inductive and deductive accounts can be distinguished

by their vantage point: inductive accounts see linguistic categories as pro-
babilistic concepts. For example, the “usual” case is for verbs with a transi-
tive meaning to take two arguments, and for verbs of transfer to take three
arguments. Deductive accounts assume that linguistic categories have a
semantic or symbolic core, which is considered to be absolute such that
children could make use of the link between the semantic and syntactic
core in order to bootstrap another level of linguistic representation.
The semantic bootstrapping account comes closest to the traditional
notion of valency. Valency in its core is a “projection” account: the verb
exercises control over the arguments it occurs with. Consequently, there
should be a systematic link between verb semantics and verb syntax that
could be exploited in language learning since it would allow the child to
predict the properties of semantically or syntactically similar verbs. The
semantic bootstrapping account strives for full predictability of syntax on
the basis of semantics (e.g., narrow and broad range linking rules) because
it is assumed that learners are hypothesis-testing, thus grammar and seman-
tics needed to be constrained in order to protect the learner from generaliz-
ing overly general grammars.
The syntactic bootstrapping account focuses on how children can use
their syntactic knowledge for possible verb semantics. Syntactic bootstrap-
ping cannot be the starting point for acquisition because it requires that
children have built up some lexical as well as structural knowledge in order
to deduce semantics based on structure.
More recently, the role of the construction has been emphasised in an-
other framework, usage-based models of language. These inductive models
are more lenient because they rely on probabilistic, not absolute cues. Since
learners are assumed to be conservative, not hypothesis-testing, they will
only generalise on the basis of positive evidence. They start out with lexi-
cally-based utterance schemas in order to encode their intentions and ab-
stract semantic and syntactic components only gradually. It is important to
note that constructions are defined as form-function units, thus form and
function are equally important. The starting point need not be the semantic
 The acquisition of argument structure 209

or syntactic “core” from which the periphery is acquired; instead, the core
components would be the results of generalization over repeated experi-
ence. Research in this tradition focuses more narrowly on the exact learning
processes that lead to more schematic and later fully abstract representa-
tions.
In usage-based models we observe a shift of attention from the role of
the verb to the role of the syntactic frame or construction. This is psycho-
logically plausible because humans communicate in order to convey inten-
tions, and they do so using utterances, not words (Tomasello 2000). Thus,
utterances are the primary source of information from which words and
syntactic operations that combine them can be isolated or abstracted. For
this to happen, there needs to be repetition and variation: repeated exposure
leads to the entrenchment of that particular structure. However, without
variation this structure would be unanalyzed and frozen, and productivity
would break down. Variation in the structure is needed to acquire more
general and abstract schemata; e.g., if a given verb is only used with prepo-
sitional phrases denoting location, the learner will probably not generalise
this frame to manner information as well. Thus, a model that integrates
both entrenchment and variation leads to more sophisticated mental models
that allow for (frequency-based) generalisations and help to explain devel-
opmental as well as diachronic language change (cf. Bybee 2005).
One of the key problems is to determine in more detail how repetition
and variation interact. Bybee (2005) alludes to exemplar-based models of
language, which assume that each usage-event is an exemplar that acts on
our representation because it leaves a memory trace. This theory thus relies
on concrete (= substantial) usage that is stored. It is as yet not known
whether we simply store more and more tokens upon repeated usage, or
whether we store more repeated information on a more general and abstract
level when available, or whether we do both. The latter is conceivable since
first results suggest that we have access to multiple levels of specificity
(Bybee and Scheibman 1999).
And finally, research on the exact nature of storage in the mental lexi-
con is required. Elman (2004) refutes the classic perspective of the mental
lexicon as that of a “dictionary” in long-term memory with a passive stor-
age for semantic and structural information. Alternatively, he proposes a
dynamic model of the mental lexicon based on previous experience. With
each new experience with words, the mental space of the lexicon is refined
and redivided; e.g., each new exposure to the word child in context acts on
our existing representation of the concept ‘child’. We do not simply re-
trieve a fixed word meaning from memory in order to process the new sen-
tence. Elman (2004: 305) proposes that there is a continuum from learning
210 Heike Behrens

words to learning constructions: “Thus, knowledge of constructions is a

straightforward extension, by generalization, of knowledge of groups of
words that behave similarly”. From a usage-based perspective, children’s
and adult’s representations can be seen as a dynamic mental inventory of
lexical items and constructions.

Notes

1. My interest in and knowledge about this topic goes back to the many intense
and lively discussions in the Argument Structure Project at the Max-Planck-
Institute for Psycholinguistics in Nijmegen in the mid 1990s. In particular, I
would like to thank Shanley Allen, Melissa Bowerman, Penny Brown,
Paulette Levy and David Wilkins for discussions on this topic. The inspiration
for usage-based acquisition research came from many discussions at the Max-
Planck-Institute for Evolutionary Anthropology in Leipzig, most notably with
Mike Tomasello, Elena Lieven and Kirsten Abbot-Smith.

References

Akthar, Nameera, Malinda Carpenter, and Michael Tomasello

1996 The role of discourse novelty in early word learning. Child Develop-
ment 67: 635–645.
Allen, Shanley E. M.
2000 A discourse-pragmatic explanation for argument representation in
child Inuktitut. Linguistics 38: 483–521.
Allen, Shanley E. M., and Heike Schröder
2000 Preferred argument structure in early Inuktitut spontaneous speech
data. In Preferred Argument Structure: Grammar as Architecture for
Function, John W. DuBois, Lorraine E. Kumpf, and William J.
Ashby (eds.), 301–338. Amsterdam: Benjamins.
Bates, Elizabeth, and Brian MacWhinney
1987 Competition, variation and language learning. In Mechanisms of
Language Acquisition, Brian MacWhinney (ed.), 157−193. Hillsdale,
NJ: Erlbaum.
Behrens, Heike
2006 The input-output relationship in first language acquisition. Language
and Cognitive Processes 21: 2–24.
Bowerman, Melissa
1985 What shapes children’s grammar? In The Crosslinguistic Study of
Language Acquisition. Vol. 2: Theoretical Issues, Dan I. Slobin (ed.),
1257−1319. Hillsdale, NJ: Erlbaum.
 The acquisition of argument structure 211

1990 Mapping thematic roles onto syntactic functions: Are children helped
by linking rules? Linguistics 28: 1253–1290.
1996 Argument structure and learnability: Is a solution in sight? In Pro-
ceedings of the Twenty-Second Meeting of the Berkeley Linguistics
Society (=BLS 22), Jan Johnson, Matthew L. Juge, and Jeri L. Mox-
ley (eds.), 454–468. Berkeley: Berkeley Linguistics Society.
Bowerman, Melissa, and Penelope Brown
2007 Introduction. In Crosslinguistic Perspectives on Argument Structure:
Implications for Language Acquisition, Melissa Bowerman, and
Penelope Brown (eds.). Mahwah, NJ: Erlbaum.
Brown, Penelope
1998 Children’s first verbs in Tzeltal: Evidence from the early verb cate-
gory. Linguistics 36: 713–753.
2007 Verb specificity and argument realization in Tzeltal child language:
Implications for language acquisition. In Crosslinguistic Perspectives
on Argument Structure: Implications for Language Acquisition,
Melissa Bowerman, and Penelope Brown (eds.). Mahwah, NJ: Erl-
baum.
Bybee, Joan L.
2005 From usage to grammar: The mind’s response to repetition. Manu-
script: University of New Mexico, Albuquerque.
Bybee, Joan L., and Joanne Scheibman
1999 The effects of usage of degrees of constituency: The reduction of
“don’t” in English. Linguistics 37: 575–596.
Clancy, Patricia
1993 Preferred argument structure in Korean acquisition. In Proceedings of
the 25th Annual Child Language Research Forum, Eve V. Clark
(ed.), 307–314. Stanford: CSLI.
1997 Discourse motivations of referential choice in Korean acquisition. In
Japanese / Korean Linguistics 6, Ho-min Sohn, and John Haig (eds.),
639–659. Stanford, CA: CSLI.
Clark, Eve V.
1993 The Lexicon in Acquisition. Cambridge: Cambridge University Press.
DuBois, John W.
1987 The discourse basis of ergativity. Language 63: 805–855.
Elman, Jeffrey L.
2003 Generalization from sparse input. Proceedings of the 38th Meeting of
the Chicago Linguistics Society. Chicago: Chicago University Press.
2004 A different view on the mental lexicon. Trends in Cognitive Science
8: 301–306.
Elman, Jeffrey L., Elizabeth A. Bates, Mark H. Johnson, Annette Karmiloff-Smith,
Domenico Parisi, and Kim Plunkett
1996 Rethinking Innateness: A Connectionist Perspective on Development.
Cambridge, Mass.: MIT Press.
212 Heike Behrens

Fillmore, Charles J., and Paul Kay

1993 Construction Grammar Coursebook: Chapters 1 thru 11. University
of California at Berkeley: Department of Linguistics.
Gleitman, Lila R.
1990 The structural sources of verb meaning. Language Acquisition 1: 3–
55.
Goldberg, Adele E.
1995 Constructions. Chicago: Chicago University Press.
2006 Constructions at Work: The Nature of Generalization in Language.
Oxford: Oxford University Press.
Goldberg, Adele E., and Devin M. Casenhiser
2005a The role of prediction in construction learning. Journal of Child Lan-
guage 32: 407–426.
2005b Fast mapping between a phrasal form and meaning. Developmental
Science 8: 500–508.
Goldberg, Adele E., Devin M. Casenhiser, and Nitya Sethuranam
2004 Learning argument structure generalizations. Cognitive Linguistics
15: 289–316.
Gomez, Rebecca L., and Louann Gerken
2000 Infant artificial language learning and language acquisition. Trends in
Cognitive Science 4: 178–186.
Hyams, Nina
1986 Language Acquisition and the Role of Parameters. Dordrecht: Reidel.
Keibel, Holger, Jeffrey L. Elman, Elena Lieven, and Michael Tomasello
2006 From words to categories. University of Freiburg: Unpublished
Manuscript.
Klein, Dan, and Christopher Manning
2004 Corpus-based induction of syntactic structure: Models of dependency
and constituency. Proceedings of the 42nd Annual Meeting of the As-
sociation for Computational Linguistics (ACL 2004).
Landau, Barbara, and Lila R. Gleitman
1985 Language and Experience: Evidence from the Blind Child. Cam-
bridge, Mass.: Harvard University Press.
MacWhinney, Brian
1987 The competition model. In Mechanisms of Language Acquisition,
Brian MacWhinney (ed.), 249–308. Hillsdale, NJ: Erlbaum.
MacWhinney, Brian (ed.)
1999 The Emergence of Language. Mahwah, NJ: Erlbaum.
Naigles, Letitia R.
1996 The use of multiple frames in verb learning via syntactic bootstrap-
ping. Cognition 58: 221–251.
2002 Form is easy, meaning is hard: Resolving a paradox in early child
language. Cognition 86: 157–199.
 The acquisition of argument structure 213

Naigles, Letitia, Lila R. Gleitman, and Henry Gleitman

1993 Children acquire word meaning components from syntactic evidence.
In Language and Cognition: A Developmental Perspective, Esther
Dromi (ed.), 104–140. Norwood, NJ: Ablex.
Newport, Elissa L., and Richard N. Aslin
2004 Learning at a distance: I. Statistical learning of non-adjacent depend-
encies. Cognitive Psychology 48: 127–162.
Ninio, Anat
1999 Pathbreaking verbs in syntactic development and the question of
prototypical transitivity. Journal of Child Language 26: 619–653.
Pine, Julian M., Elena V. M. Lieven, and Caroline F. Rowland
1998 Comparing different models of the English verb category. Linguistics
36: 807–830.
Pinker, Steven
1989 Learnability and Cognition: The Acquisition of Argument Structure.
Cambridge, Mass.: MIT Press.
1994 How could a child use verb syntax to learn verb semantics? Lingua
92: 377–410.
Redington, Martin, Nick Chater, and Steven Finch
1998 Distributional information: A powerful cue for acquiring syntactic
categories. Cognitive Science 22: 425–469.
Saffran, Jenny R.
2003 Statistical language learning: Mechanisms and constraints. Current
Directions in Psychological Science 12: 110–114.
Slobin, Dan I.
1985 Crosslinguistic evidence for the language-making capacity. In The
Crosslinguistic Study of Language Acquisition. Vol. 2: Theoretical
Issues, Dan I. Slobin (ed.), 1157–1249. Hillsdale, NJ: Erlbaum.
1997 The origins of grammaticizable notions: Beyond the individual mind.
In The Crosslinguistic Study of Language Acquisition. Vol. 5: Ex-
panding the Contexts, Dan I. Slobin (ed.), 265–323. Mahwah, NJ:
Erlbaum.
2001 Form-function relations: How do children find out what they are? In
Language Acquisition and Conceptual Development, Melissa Bow-
erman, and Steve Levinson (eds.), 406–449. Cambridge: Cambridge
University Press.
Theakston, Anna L., Elena V. M. Lieven, Julian M. Pine, and Caroline F. Rowland
2002 Going, going, gone: The acquisition of the verb “go”. Journal of
Child Language 29: 783–811.
Tomasello, Michael
1992 First Verbs: A Case Study of Early Grammatical Development. Cam-
bridge: Cambridge University Press.
1998 The return of constructions. Review essay on: Goldberg, A., 1995
‘Constructions: A construction grammar approach to argument struc-
ture’. Journal of Child Language 25: 443–484.
214 Heike Behrens

2000 Do young children have adult syntactic competence? Cognition 74:

209–253.
2003 Constructing a Language: A Usage-Based Account of Language
Acquisition. Cambridge, Mass.: Harvard University Press.
Tomasello, Michael, and Hannes Racoczy
2003 What makes human cognition unique? From individual to shared to
collective intentionality. Mind and Language 18: 121–147.
Valian, Virginia
1991 Syntactic subjects in the early speech of American and Italian chil-
dren. Cognition 40: 21–81.
Wilkins, David P.
2007 Same argument structure, different meanings: Learning ‘put’ and
‘look’ in Arrernte. In Crosslinguistic Perspectives on Argument
Structure: Implications for Language Acquisition, Melissa Bower-
man, and Penelope Brown (eds.). Mahwah, NJ: Erlbaum.
Section 3
Contrastive aspects of valency


Valency and the errors of learners of English and

German

Ian Roe

1. Introduction

In contrast to the majority of contributions to the current volume, this essay

is concerned primarily with valency not as a theoretical construct but as a
learning (and teaching) tool. The first English valency dictionary (Herbst et
al. 2004) has recently been published, the German tradition in this field is
already well established both in terms of grammar and dictionaries, al-
though only the recently published VALBU (Schumacher et al. 2004)
comes close to meriting the term “user-friendly” and there is as yet no
valency dictionary targeted at English-speaking learners.1 The current essay
seeks to investigate some of the areas that need to be addressed by a user-
friendly valency dictionary for learners by drawing some comparisons, on
the basis of a small analysis of errors, between the linguistic problems en-
countered by German and English advanced learners of the other language.
Two questions need to be asked: what proportion (and what types) of errors
might be seen as valency-specific? And what different conclusions might
the compilers of English and German valency dictionaries need to draw?

2. Brief analysis of errors

2.1. German learners of English

A tempting short-cut to painstaking analysis in this field is, on the surface

at least, Geoff Parkes’s Mistakes Clinic (2001), which offers an analysis of
over twenty years of work submitted by German learners of English, in-
volving 4,000 errors of vocabulary and 5,500 errors of grammar and syn-
tax. It is interesting to note that, in a list of the top twelve grammar mis-
takes, the most frequent is the use of prepositions, the fourth most frequent
is confusion between ing- and infinitive-constructions. However, although
Parkes’s booklet provides useful examples and lists and is undoubtedly
helpful for teachers of English, for our present purposes there is no attempt
218 Ian Roe

to distinguish complementation errors and other types of syntactical or

grammatical error. As one example: where more detailed information is
provided on prepositions by giving a top 20, the first four are all errors
concerning adjuncts (e.g.*leave school with 16, *in the last moment), but
the next five are valency-specific (which preposition is needed after good,
contact, pay, typical, depend) and errors with explain, connection, discuss,
react, knowledge are listed at 13 and 16-19 respectively (Parkes 2001:
125).
For a less extensive but, in the present context, more differentiated
analysis I have used a small corpus of translations into English by German
exchange students together with essays in English by German native-
speakers.
Conflating the analysis of two sets of advanced translations, one finds
that of 189 errors of syntax,
− 66 (one third of the total) were valency driven errors, of which 44 in-
volved verbs, 13 nouns, and 9 adjectives;
− 123 were general errors (e.g. tense or aspect, misuse of articles, an ad-
jective used instead of an adverb, word order, etc.).
The collection of essays produced a slightly different picture:2 almost half
(53) of the 114 errors were valency, and a rather higher percentage (21 of
53) were errors of noun valency, but that was perhaps inevitable in essays,
as the errors occurred with nouns such as argument, question, discussion,
criticism.
Taking the two sets together, 119 of 303 errors (39%) were valency-
specific, and verbs accounted for 59% of the valency errors. Of the 119
valency-specific errors, 36 involved the wrong choice of clause, especially
between infinitive and ing-constructions, 47 were wrong choice of preposi-
tion; there were also 13 examples of the wrong choice between noun phrase
and prepositional complement (*She returned me my change), although the
majority of these latter errors involved problems with apposition (*the
province Carinthia, *the state Israel). Of the 184 non-valency errors, 64
involved the wrong choice of tense or aspect (including wrong choice of
simple or continuous [25], and incorrect use or omission of the auxiliary do
[17]), 21 wrong prepositions, 25 wrong uses of the article, and 22 word-
order errors (mainly the position of adverbials).

2.2. English learners of German

The parallel German corpus consists of two sets of translations into Ger-
man by British final-year university students. The first set was written a
 Valency and the errors of learners of English and German 219

number of years ago by 33 final-year students with, in most cases, nine

years of German, including five or ten months in Germany or Austria. Dic-
tionaries were not allowed in this earlier examination. Leaving aside errors
of vocabulary, the 155 valency-specific errors can initially be analysed as
follows:
− 126 errors of verb valency (plus a further 7 instances of the nominative
used as an object);
− 14 problems of noun valency, although mainly the use of a plural sub-
ject with a singular verb;
− 1 adjective (befürchtet, which might therefore be seen as verbal);
− and 7 errors in the use of collocation (what Ágel [2003: 33] refers to as
“syntaktische Ausdrucksvalenzträger”): zu tun haben mit, Bescheid wis-
sen.
It is important, however, to note that there was a large variety of errors that
were not valency-specific. Such general grammatical errors numbered 541
in total,3 or 16.5 errors per student: arguably more, as multiple occurrence
of identical or very similar errors was counted only once. The figure of 541
contrasts with 155 valency-specific errors (4.7 per candidate). Inevitably
these figures are capable of variation, as analysis becomes difficult or al-
most impossible with a sentence such as the following: Die neun Millionen
Wörter, die über das Thema geschrieben worden, würden von ein Kompu-
ter gezahlt, hatten in Zeitungen erschienen während des letzten zwölf Mo-
natens, in lautere Reden von Politiker und das Thema wurde von Fernseher
behandelt, haben nicht gewirkt.4
The second set is from a more recent examination (2003) taken by 24
students (again with a minimum of 8-9 years of German plus 5-10 months
abroad). Dictionaries were allowed with this later examination. Again leav-
ing aside errors of vocabulary, including wrong mode of address or confu-
sion of wann and als, the 104 valency-specific errors can initially be ana-
lysed as follows:5
− 54 errors can be seen as verb valency; a further 6 might be seen in this
category, but might be just confusion of cases (use of nominative, also
one dative, as object; accusative used as subject of a passive sentence);
− 27 are noun valency: problems with Gedanke, Idee, Art/Sorte, Ort/Stelle
(mainly arising from difficulty in producing a German equivalent of the
best place to go);
− 1 possible example of adjective valency (the need for a complement
with für with verantwortlich);
− 16 errors in the use of collocations such as satt haben, leid sein, hungrig
sein.
220 Ian Roe

Alongside these 104 errors in total (4 per candidate), which will be exam-
ined in more detail below, there were again a much larger number of non-
valency-specific errors: the total of 339 general grammatical errors repre-
sents almost 14 per student.6 This figure contrasts with 16.5 errors per stu-
dent in the earlier set analysed, but the lower figure is explained to a con-
siderable extent by the availability of dictionaries, which undoubtedly pro-
duced fewer errors in matters of gender and plural; if one discounts such
errors, the error-count per candidate is approximately the same.7 A small
number of mistakes were difficult to categorise but were clearly not valen-
cy-specific (*keine Gedanke, *sie haben ein deutschen Fest gefeiert),8
whilst as in the previous batch some sentences made an overall analysis
difficult: *Man sollte zu dieser Zeit des jahres Deutschland als das zuletz-
ten Land hinzufahren halt, oder? or *Das Speise dorthin findest du nichts
in einem erste-klasse Restaurants, jedoch wenn du kalt oder hungrish wä-
rest denn dies wie das Beste überall schmecken würde.9
Interesting by comparison was the very low occurrence of what one
might rather unkindly categorise as “gibberish sentences” in the English
written by German learners, an obvious consequence of the much longer
and more constant exposure to the foreign language that marks out German
learners of English from their British counterparts.10

3. Analysis of valency-specific errors in German

Valency-specific errors therefore accounted for 23% of grammatical and

syntactical errors (compared to 39% with the German learners of English).
If we collate the two analyses of the 252 valency-specific errors we find the
following types of errors in particular:

Table 1. Types of valency-specific errors

number type of error example

8 omission of obligatory comple- with verbs such as beantwor-
ments ten, beschuldigen, fühlen
48 wrong case of noun complement accusative used after sein or
werden
21 noun phrase instead of preposi- antworten + ACC
tional complement
14 prepositional complement used with entstehen, geben
instead of noun phrase
 Valency and the errors of learners of English and German 221

1 adverbial complement used in- entstammen

stead of noun phrase
56 wrong choice of preposition, or antworten zu/an
preposition used instead of ad-
verb phrase
10 correct preposition but wrong geraten in, denken an
case
19 wrong choice of clause comple- sie erscheinen, dass …
ment
10 wrong type of complement erweisen and beweisen used
with infinitive clause
5 wrong use of clause realisation of es handelt sich um, dass …
prepositional complement11
7 wrong use of or omission of re- beweisen, erweisen
flexive
43 wrong choice from semantically nennen for benennen
linked verbs12

4. Comparison of valency-specific errors in English and German

This analysis of the various sets of errors indicates some common ground
between the errors made by learners of English and German but quite a
number of key differences. The most obvious areas of common ground
(speaking of types of problems rather than specific individual problems
encountered) were:
− the choice of prepositional complement (almost 40% in English, 20+%
for German);
− the choice of clause complement (30% for English, 12% in German);
− the choice between noun phrase and prepositional complement (includ-
ing problems with apposition such as the state Israel or die Art Gericht)
(11% for English, 14% for German).
It is especially worth noting that these three areas accounted for approxi-
mately 80% of the valency-specific errors of German learners of English.
A further area of common ground, but accounting for a relatively small
number of problems, indeed far fewer than might have been expected, was
the problem of the optionality of the complement (in constructions with
remind, allow, beantworten, fühlen, vorstellen).13
222 Ian Roe

Problems for German learners of English arose from:

− one specific aspect within the choice of structures already referred to,
namely that between +to-INF and +of V-ing, especially with nouns such
as aim, intention, chance;
− the need in English for use of a semantically empty noun as a correlate
or stylistic filler, most particularly the fact (that). Empirical evidence
suggests that this should be listed as common ground, but it was less of
a problem in the German material analysed, either, as suggested above,
because it was possible to avoid by the English-influenced use of nouns
such as Tatsache, Situation, or because the relatively limited corpus re-
sulted in only a few errors of the kind *es handelt sich um, dass … . A
further use or non-use of correlates caused occasional problems for the
learners of English, as revealed by errors sich as *Important was that /
*I find interesting that / *Can I point it out to you that / *I cannot stand
that you do that;
− the choice of tense/aspect, which will be considered below.
Vilmos Ágel (2004: 27–28) has argued for an integration of word order into
valency, but in the material analysed there was little evidence of valency-
specific problems arising from restricted word order in English (i.e. errors
of the type *I gave the book him or *To Frankfurt he went). Problems that
might be seen as valency-driven were the result of incorrect use of the aux-
iliary do, but otherwise it was the position of English adverbials that most
frequently resulted in word-order error.
Specific problems for English learners of German were:
− most obviously, and perhaps not surprisingly, the case system.14 Some
of these errors were valency-specific, not least the choice of case with
prepositional complements, the incorrect use of the accusative after sein,
and also the very English tendency to assume that any noun phrase after
the verb should be an object, resulting in errors of the kind *Es besteht
keinen Grund, daran zu zweifeln;
− one particular aspect of that same overall problem is the relative inflexi-
bility of the passive in German (a particular problem for English learn-
ers but less so for speakers of other Western European languages). By
comparison, the passive was only occasionally a problem for German
learners of English, with errors resulting from interference from German
constructions involving the impersonal passive − *It has to be tried to …
/ *There should not be made any attempt to …;15
− the choice between semantically related words requiring alternative
structures (antworten / beantworten) also seemed much more of a prob-
lem for British learners of German, a further reflection perhaps of the
 Valency and the errors of learners of English and German 223

greater linguistic subtlety or awareness of German learners of English,

as noted earlier in the context of “gibberish” sentences.
Alongside those valency-specific errors one must, however, also note the
large number of problems arising from other aspects of the German mor-
phological system, whether case-driven or convention-driven (e.g. strong
and weak adjective endings).

5. Valency errors and the consequences for a valency dictionary

Despite the last comment in the previous section it is important to note that
the number of valency-specific errors is still numerically high for English-
learners of German; the complexity of German morphology, however, pro-
duces a much higher number of learner errors overall and in the process
also a lower percentage of valency-specific errors. In respect of both lan-
guages it may be argued that valency-specific errors frequently relate to
structure and are consequently more likely to detract from comprehensibil-
ity than (in German) incorrect plurals, gender or adjective endings, or (in
English) incorrect word order or problems of tense and aspect.
The vital role of valency in learners’ errors is not to be questioned,
therefore, and it remains to ask, on the basis of the evidence analysed so
far, what demands one should make of a good valency dictionary. That is of
course to assume that our targeted learners do use dictionaries, although
there is the frequent frustration for teachers that they do not, or that they
rarely use the half of a bilingual dictionary in which the entries are in the
target language. Petra Bräunling (1989) and more recently Monika Bie-
lińska (2003) have underlined the even greater difficulty of getting FL
learners or even FL teachers to use a valency dictionary. Leaving such cau-
tion aside, however, we may suggest that a valency dictionary should:
− concentrate on key difficult words, especially verbs.16 This might be
seen as stating the obvious, but some German valency dictionaries have
taken a different line;17
− give cross references between semantically related words with different
valency. English comprise or German beantworten do not themselves
merit an individual entry, but users can be referred to the entries for
consist and antworten respectively. A glossary or index of target-
language words might be considered important here and even an index
of cross references in the source language.18
Moving on to what are arguably the really crucial aims, a valency diction-
ary must provide:
224 Ian Roe

− very detailed analysis of prepositional complements, especially where

alternative realisations are possible: agree + to/with; care + for/about;
warten + auf/bis; bestehen + aus/in/auf. The evidence analysed does not
provide justification for detailed analysis of clause realisation of Ger-
man prepositional complements, but it is possible that a more detailed
analysis than was possible with the available corpus would suggest oth-
erwise;
− detailed analysis of clause complementation, with specific reference to
particular sources of error or confusion. In English this problem is fur-
ther compounded by confusion between clause complements and prepo-
sition-plus-present participle structures. Recently Klaus Fischer has un-
derlined the flexibility of complement clauses in English (2004: 223–
224), the fact that a number of English clause-types are not found in
German: to the list discussed by Fischer (including especially +for N
+to-INF) one could add in particular the bolt-on infinitive such as she
had no one to turn to. This flexibility is not often a source of syntactical
error, and hence does not feature in the analysis of errors, but it can pro-
duce stilted English in sentences such as She had no one to whom she
could turn or It is rare that a house is built in two weeks (instead of for
a house to be built …). A German dictionary for English-speaking
learners clearly needs to underline the lack of flexibility in this respect,
and the incorrectness of a structure such as *Deutschland ist nicht der
beste Ort hinzufahren. A valency dictionary, which should be concen-
trating on key words, can play an important role in sensitising learners
to wider stylistic possibilities rather than simply concentrating on
grammatical and syntactic accuracy.19
Two areas that are more specific to only the one side of our present Ger-
man-English comparison:
− especially in English one may ask to what extent tense or aspect are part
of valency. Some areas might be expected to be covered in our ideal no-
tion of a valency dictionary, such as the wrong choice of tense in indi-
rect speech or as if clauses, for instance, but other aspects are not obvi-
ously valency-driven. Thomas Herbst and I have demonstrated (Herbst
and Roe 1996) how the optionality of complements is affected by ques-
tions of tense and aspect and one might consider in more detail the ex-
tent to which other issues of tense are important: She has written sounds
fine in context, whereas She wrote without an object seems less likely.
There is however a very delicate question of balance involved here, as
any move from lexeme-based valency to word-form valency can easily
render a valency dictionary in any manageable volume impossible;20
 Valency and the errors of learners of English and German 225

− on the German side of the equation there is the question of the extent to
which the seemingly endless problems caused by case and morphology
can be integrated into a valency description. Clearly valency cannot be
used to teach adjective endings or the plural of German nouns, or that
mit is followed by a noun in the dative, and the current analysis suggests
that even advanced learners of German make a large number of errors
that require other teaching and learning approaches alongside that of
valency grammar. Nevertheless the use of a valency approach can place
great emphasis on case use in paradigmatic situations such as the func-
tion of the dative, the correct choice of case with prepositional comple-
ments, or the greater restrictions on the semantic role of cases (as in the
relative inflexibility of passive transformations in German).
Some of these aspects could of course be more fully treated in a valency
dictionary aimed at a particular linguistic group of learners: an English
dictionary for German learners, for example, as opposed to one for all
learners of the language. In such a targeted valency dictionary, problems of
interference can be fully addressed, for example with warnings of the kind
“note that you may NOT say …”. Such an approach is envisaged in the
proposed German dictionary for English learners referred to earlier (see
note 1); to discuss it in the present context would, as Theodor Fontane or
Günter Grass might have put it, be “ein zu weites Feld”.21

Notes


1. A project for such a dictionary is underway, coedited by myself, Alan Cornell

and Klaus Fischer; for an outline of the project, see Cornell and Roe 1999.
2. I am grateful to Alan Cornell for supplying this material.
3. These general grammatical errors include 139 wrong verb conjugations, 62
wrong plurals, 88 wrong genders, 36 incorrect choices of preposition in an ad-
verbial phrase, 36 wrong cases after prepositions in adverbial phrases, 112
other wrong endings, 48 word order errors, and 2 misuse of articles.
4. Or, from a number of similar examples:
Eine Frau aus Battersea glaubte, daß die Teilung der Konservativen aufgrund
des Lusts Margaret Thatchers nochmal an die Stellung Premierministers sei ...
jemand anderen dachte es hatte etwas mit die gleichen arbeit stunden für das
gleicher geld für alle zu tun.
5. 4 per student, though some with an obvious native-speaker background had no
valency-specific errors of any kind.
6. These non-valency-specific errors include 43 wrong verb conjugations, 43
wrong plurals, 22 wrong genders, 40 wrong cases after prepositions, 73 other

wrong endings, 12 errors relating to relative pronouns, 29 incorrect uses of ar-
226 Ian Roe


ticles, 27 incorrect choices of preposition in adverbial adjuncts, and 48 word-

order errors.
7. 11.5 errors per candidate in the later (2003) examination compared to 11.8 in
the earlier set, but the later translation was undoubtedly less difficult.
8. As Anke Lüdeling et al. (2005: 2–3) underline, categorisation of errors can
indeed be problematic, although an experienced teacher accustomed to seeing
mistakes by students with the same native language is usually able to make an
accurate assessment of the type of mistake involved.
9. Or, again from many similar examples:
Wenn du zum Weihnachten Kotzen finden und kann nicht es nochmal feiern
dann fahrst du nicht nach Deutschland, oder?
Ein Weihnachtsmarkt hat alles, dass Sie alle Hoffnung immer aufgeben, in
Dezember in einer typischen britischen Hauptstraße zu finden.
10. Indeed the degree of opaqueness is arguably less even in those English sen-
tences that do pose problems of comprehension:
… the respectable time historian Karl-Friedrich Bracher pointed at the power
of seduction of how to handle certain terms linguistically.
“Reading Haider and thinking of Hitler” was even said on banners.
11. The number of such errors is arguably much higher, as the problem was often
partly avoided through stylistically unacceptable but grammatically just ac-
ceptable sentences of the type ?es hat zu tun mit diesem Ding, wo wir alle die
selben Stunden arbeiten für das gleich Lohn.
12. Arguably some of these might be seen as errors of vocabulary rather than of
valency. Furthermore, a small number of individual problems merit a brief
mention. The German equivalent of answer was a particular problem, as de-
spite the best efforts of lecturers and teachers antworten or beantworten was
used wrongly by almost two-thirds of the candidates; especially antworten was
used with either an accusative complement or with the wrong prepositional
complement; frequent problems also occurred with passive constructions in-
volving stellen or fragen; and many problems arose from the English sen-
tences Germany is not the best place to go at this time of year, and also and
perhaps more understandably, The market place contains everything you al-
ways despaired of finding.
13. Although it would not normally be considered an issue of valency, one may
also note problems with the use or non-use of articles, despite the great simi-
larities between English and German in this area of grammar.
14. See most recently Fischer (2004).
15. The theoretically interesting question of which elements in English can be-
come the subject of a passive construction does not cause problems for foreign
learners – presumably because they do not venture as far as complexities such
as Which film to see has not yet been talked about.
16. The lack of problems encountered with the valency of adjectives was particu-
larly noticeable in the material analysed, although this may be a result of the
restricted corpus.
17.

See especially Sommerfeldt and Schreiber (1996).
 Valency and the errors of learners of English and German 227


18. Bielińska (2003: 244–245) suggests that the use of valency dictionaries for
learners may only become possible once the learner has come to the FL word
via a bilingual dictionary.
19. The “sensitising” role of valency dictionaries is underlined by Alan Cornell
(2003: 142).
20. This danger was underlined at one point of the conference in Erlangen when a
participant was heard to comment that one might need one volume simply for
the verb put. For an indication of the breadth of detail that is potentially possi-
ble, see Willems and Coene (2003) on glauben; also Kolde (2004).
21. See the example entry for antworten in Roe (2003: 196–197).


References

Ágel, Vilmos
2000 Valenztheorie. Tübingen: Narr.
2003 Wort- und Ausdrucksvalenz(träger). In Valency in Practice: Valenz
in der Praxis, Alan Cornell, Klaus Fischer, and Ian F. Roe (eds.),
17–36. Oxford/Bern/Berlin/Bruxelles/Frankfurt M./New York/Wien:
Lang.
2004 Prinzipien der Valenztheorie(n). In Valenztheorie: Bestandsaufnah-
me und Perspektiven, Speranţa Stănescu (ed.), 11–30. Frankfurt M.:
Lang.
Bielińska, Monika
2003 Valenzwörterbücher – das Ideal und das Leben. In Valency in Prac-
tice: Valenz in der Praxis, Alan Cornell, Klaus Fischer, and Ian F.
Roe (eds.), 241–258. Oxford/Bern/Berlin/Bruxelles/Frankfurt M./
New York/Wien: Lang.
Bräunling, Petra
1989 Umfrage zum Thema Valenzwörterbücher. Lexicographica 5: 168–
177.
Cornell, Alan, and Ian F. Roe
1999 A valency dictionary for English-speaking learners of German. In
From Classical Shades to Vickers Victorious: Shifting Perspectives
in British German Studies, Steve Giles, and Peter Graves (eds.),
153–170. Bern: Lang.
Cornell, Alan, Klaus Fischer, and Ian F. Roe
2003 Valency in Practice: Valenz in der Praxis. Oxford/Bern/Berlin/
Bruxelles/Frankfurt M./New York/Wien: Lang.
Cornell, Alan
2003 Valency for learners of German: How do the customers feel? In
Valency in Practice: Valenz in der Praxis, Alan Cornell, Klaus
Fischer, and Ian F. Roe (eds.), 127–143. Oxford/Bern/Berlin/Bru-
xelles/Frankfurt M./New York/Wien: Lang.

228 Ian Roe


Fischer, Klaus
1997 German-English Verb Valency. Tübingen: Narr.
2004 Deutsche und englische Ergänzungssätze: Zwei typologische Ano-
malien? In Valenztheorie: Bestandsaufnahme und Perspektiven,
Speranţa Stănescu (ed.), 213–236. Frankfurt M.: Lang.
Herbst, Thomas, David Heath, Ian F. Roe, and Dieter Götz
2004 A Valency Dictionary of English. A Corpus-Based Analysis of the
Complementation Patterns of English Verbs, Nouns and Adjectives.
Berlin/New York: Mouton de Gruyter.
Herbst, Thomas, and Ian F. Roe
1996 How obligatory are obligatory complements? – An alternative ap-
proach to the categorization of subjects and other complements in
valency grammar. English Studies 77: 179–199.
Kolde, Gottfried
2004 Gehört der Heckenausdruck (so) (ei)ne Art (von) X ins Valenzwör-
terbuch? In Die Valenztheorie: Bestandsaufnahme und Perspektiven,
Speranta Stănescu (ed.), 133–146. Frankfurt M.: Lang.
Lüdeling, Anke, Maik Walter, Emil Kroymann, and Peter Adolphs
2005 Multi-level error annotation in learner corpora.
http://www.linguistik.hu-berlin.designato.de/korpuslinguistik/
projekte/falko/FALKO-CL2005.pdf (15 August 2005)
Parkes, Geoff
2001 The Mistakes Clinic for German-speaking Learners of English.
Southampton: Englang Books.
Roe, Ian F.
2003 Layout of valency dictionary entries: Theory and practice. In
Valency in Practice: Valenz in der Praxis, Alan Cornell, Klaus
Fischer, and Ian F. Roe (eds.), 187–209. Oxford/Bern/Berlin/Bru-
xelles/Frankfurt M./New York/Wien: Lang.
Schumacher, Helmut, Jacqueline Kubczak, Renate Schmidt, and Vera de Ruiter
(eds.)
2004 VALBU – Valenzwörterbuch deutscher Verben. Tübingen: Narr.
Sommerfeldt, Karl-Ernst, and Herbert Schreiber
1996 Wörterbuch zur Valenz etymologisch verwandter Wörter. Tübingen:
Niemeyer.
Stănescu, Speranţa
2004 Die Valenztheorie: Bestandsaufnahme und Perspektiven. Frankfurt
M.: Lang.
Willems, Klaas, and Ann Coene
2003 Argumentstruktur, verbale Polysemie und Koerzion. In Valency in
Practice: Valenz in der Praxis, Alan Cornell, Klaus Fischer, and Ian
F. Roe (eds.), 37–63. Oxford/Bern/Berlin/Bruxelles/Frankfurt M./New
York/Wien: Lang.

Temporary ambiguity of German and English term
complements1

Klaus Fischer

1. Introduction

In the assessment of the contrasts between English and German, there is a

consensus that German is more semantically transparent or overspecified
than English, has preserved more grammatical complexity and is thus a
grammatically more mature language. One reason for the contrasts is seen
in the contact-rich history of English which has lost much of its Germanic
grammatical inheritance, possibly because of imperfect acquisition by
Scandinavian settlers (McWhorter 2002: 253–265; Olga Fischer 1992:
207–208, but see Fennell 2001: 92; Thomason and Kaufman 1988: 275–
306).
Comparisons of the two languages by John Hawkins (1986, 1992), John
McWhorter (2002), Werner Abraham (2003) and Frans Plank (1984) have
understandably concentrated on the extent and systematic nature of the
contrasts. The perhaps unavoidable danger is that data that do not quite fit
the broad picture but nevertheless inform the English-German contrasts, are
ignored or not given the prominence they deserve.
In this article, I would like to adjust the picture that John Hawkins in
particular has drawn but that is also more or less explicitly present in other
accounts. I will address the claim that the English subject and direct object
are syntactically and semantically more ambiguous than the German sub-
ject and direct object. I will also investigate German verb position that is
linked to these claims.

2. Overspecification, complexity and maturity

Semantic transparency or overspecification refers to the degree of explicit

expression that can be decoded contrary to implicit information that can be
inferred (cf. Sperber and Wilson 1995: 1–117). For instance, German case-
marked noun phrases contain information that English caseless noun
phrases taken in isolation do not. The same applies to German directional
230 Klaus Fischer

adverbs (for instance dorthin) in comparison with their English counter-

parts (for instance there) that do not indicate direction explicitly.
Grammatical complexity is measured by the complexity of the gram-
matical description (Dahl 2004: 27; 49–50; McWhorter 2003: 219–220).
For instance the grammatical description of German noun phrases is more
complex than that of English noun phrases because of grammatical gender
and case.
Grammatical maturity (Dahl 2004: 111–162) is complexity seen as an
evolutionary product, as the result of grammaticalisation processes. Fu-
sional morphology such as German case, gender and number morphology
is the outcome of a staged development from ultimately free combinations
of lexical items.
Semantic transparency or overspecification on the one hand and gram-
matical complexity or maturity on the other are not identical. Grammatica-
lisation often leads to forms with very abstract meaning:

(1) Sie hat ihm die Hausaufgaben erklärt/gemacht.

She explained the homework to him/did the homework for him.

The fusional dative form is more mature than its periphrastic prepositional
counterparts but less semantically transparent: it only limits the semantic
role to an abstract role that has been labelled Betroffener [‘concerned’]
(Wegener 1985: 275) while the prepositional phrases differentiate between
goal on the one hand and beneficiary/substitute on the other.2
John Hawkins generally discusses semantic transparency with a matur-
ity bias: he stresses the information that mature German forms achieve but
neglects to mention the semantic transparency of less mature English struc-
tures.

3. Corpus

The statistical data and most example sentences in this article refer to the
following small corpus of communicative translations, using the abbrevia-
tions indicated in brackets:
− J.K. Rowling: Harry Potter and the Philosopher’s Stone (1997), ch. 1 (=
HP; 4581 words), and the German translation by Klaus Fritz (1998) (=
HPD; 4825 words);
− Cornelia Funke: Der Herr der Diebe (2000), ch. 1 and 2 (= HD; 4162
words), and the English translation by Oliver Latsch (2002) (= HDE;
4390 words);
 Temporary ambiguity of German and English term complements 231

− Detlef Gürtler: Wohlstand für alle. Das Erfolgsgeheimnis der sozialen

Marktwirtschaft. In Deutschland. Forum für Politik, Kultur, Wirtschaft
und Wissenschaft 3, 2003, S. 8–11 (= D; 1798 words), and the English
translation (= DE; 2179 words).
Thus my findings apply to written texts in the first instance.

4. Processing of argument-predicate structures

4.1. Hawkins’s psycholinguistic processing model

The background for Hawkins’s psycholinguistic processing model is his

processing theory which concentrates on the rapid recognition of constitu-
ent structure. I will ignore A Performance Theory of Order and Constitu-
ency (Hawkins 1994) here as it concludes – albeit after some discussion –
that English and German recognition of constituency structure works in
parallel: English verbs construct the verb phrase, so does German case
(1994: 397). In valency terms: case signals that a clause is to be constructed
rather than something else. I suppose the idea, though debatable, is unprob-
lematic from a valency point of view as whenever something is said the
default assumption is that this is going to be a clause.
Assuming constituents have been established, they have to be matched
with semantic predicate frames. The task takes on different facets depend-
ing on the verb position (see figure 1). As predicate frames are activated
early in English, there is time to provisionally attach constituents to a
predicate frame as they are encountered successively and later revise the
decision. English therefore tolerates temporary ambiguity resulting from
multiple predicate frames being associated with individual verbs and from
lack of morphological argument differentiation. For instance, the initial
string My guitar broke is compatible with three frames (see table 1) and
might mislead or “garden-path” a listener. Because the German equivalents
of break (brechen/zerbrechen/zerreißen) allow fewer predicate frames
and/or have tighter semantic restrictions, no garden-pathing occurs.
232 Klaus Fischer

PROCESSING MODEL 1: VERB-EARLY LANGUAGES (E.G. ENGLISH)

................ V ...................................................
-----------> | ------------------------------------> |
A | C |
| |
B D

PROCESSING MODEL 2: VERB-FINAL LANGUAGES (E.G. GERMAN)

...............................................................................………. V
-------------------------------------------------------------------> |
A |
|
B
(C)
D
A: look-ahead period
B: activation of predicate frames (all in parallel)
C: decision course: one frame provisionally selected
D: final decision on one frame
Figure 1. Processing in verb-early and verb-final languages according to Hawkins
(1992: 122–124)

Table 1. Predicate frames of break according to Hawkins (1992: 121)3

Frame 1. NP - V - NP John broke my guitar.

[Agent] [Patient] John hat meine Gitarre zerbro-
chen.
Frame 2. NP - V My guitar broke.
[Patient] Meine Gitarre ist zerbrochen.
Frame 3. NP - V - PP A string broke on my guitar.
[Patient] [Locative] An meiner Gitarre ist eine Saite
zerrissen.
Frame 4. NP - V - NP My guitar broke a string.
[Locative] [Patient] *Meine Gitarre brach/zerriß
eine Saite.
Frame 5. NP - V - NP My guitar broke a world record.
[Instru- [Patient] Meine Gitarre hat einen Welt-
mental] rekord gebrochen.

John Hawkins sees German as having basic verb-final order. This puts in-
terpreters of German sentences in quite a different position from interpret-
ers of English sentences: the early English verb position allows time to
 Temporary ambiguity of German and English term complements 233

decide between the different semantic frames while further phrases are
encountered. For the same reason phrases do not have to carry information
as to their thematic roles as there is time to address any temporary ambigu-
ity.4 Interpreters of German sentences only encounter the verb-induced list
of semantic frames at the end of the sentence and have to take a quick deci-
sion before the next sentence starts. To ensure this rapid interpretation at
the end, German case complements possess information that limits the pos-
sible scenario. Hawkins also maintains that German verbs are associated
with a smaller set of sentence frames than English verbs. As a result, Ger-
man clauses are claimed to possess less temporary ambiguity or garden
path structures.
Hawkins (1992: 128) stipulates the following desiderata for verb-final
languages:
− argument differentiation: phrases should indicate their thematic role so
that they can be quickly mapped onto predicate frames;
− predicate frame differentiation: a low number of possible predicate
frames helps quick recognition of the intended frame from surface struc-
ture;
− clear argument-predicate attachment: there should be no ambiguity as
to which predicate a phrase is to be attached to by avoiding “argument
trespassing” (raisings, wh-extractions).
Hawkins sees the three desiderata realised in German.
However, argument differentiation contains two steps: first, the identifi-
cation and interpretation of lexical and formal syntactic means (position,
morphology, prosody) which allows allocation of phrases to syntactic func-
tions (see steps (a) to (c) in figure 2), and second, the allocation of a the-
matic role on this basis (see step (d) in figure 2).

(a) Recognition and interpretation of lexical (e.g. prepositions) and formal

syntactic means (position, morphology, prosody)
|
(b) Recognition of constituent structure
|
(c) Recognition of syntactic functions
|
(d) Allocation of thematic roles
|
(e) Identification of scenario
Figure 2. Syntactic argument differentiation

Hawkins discusses (a) in relation to constituent structure but ignores it in

relation to syntactic functions (cf. Fischer 2005a). In his discussion of tem-
234 Klaus Fischer

porary ambiguity he does not mention that the mapping of case forms onto
syntactic functions is not always straightforward and therefore a source of
temporary ambiguity: his comparison of German and English argument
differentiation is only based on steps (d) to (e), which skews the result.
There are other problematic aspects of Hawkins’s processing model that
I can only briefly mention here: the failure to establish a link between verb
meaning and complementation, that is to differentiate between inherent and
combinatorial verb meaning (cf. Engel 1988: 358), and the consequent lack
of differentiation between basic and extended or derived valency struc-
tures.5 This cavalier treatment of valency leads to the doubtful idea of par-
allel activation of predicate frames (like a drop-down menu): hearers might
realistically just construct one frame on the basis of all available informa-
tion and, if problems are encountered, reconsider.

4.2. Ambiguity of English and German term complements

To what degree are German and English term complements ambiguous on-
line? Hawkins’s model suggests that English could afford a more ambigu-
ous mapping of term complements onto syntactic functions and thematic
roles than German.
English noun phrases in the narrow sense are not case-marked and
therefore are ambiguous as to their syntactic function. However, a number
of pronouns are case-marked as subjective or objective, third person per-
sonal pronouns to a higher degree than their German counterparts:

Table 2. Morphological marking of the differentiation between subject and direct

object

Personal pronoun Noun

Singular Plural Singular Plural

1. 2. 3.M 3.F 3.N 1. 2. 3. M F N
German 2 2 2 2 2 2
English 2 2 2 2 2

The profile of English and German subject-direct object differentiation

shows a graded difference that – with the exception of English second per-
son pronouns – complies with typological hierarchies. Typologically un-
marked forms are more likely to indicate subject-direct object differentia-
tion than marked forms (Croft 2003: 130, 156, 161; < is less marked than):
 Temporary ambiguity of German and English term complements 235

personal pronoun < noun

singular < plural
1st/2nd person < 3rd person (in respect of number and animacy)
masculinum (M) < femininum (F) < neutrum (N)6
human < animate < inanimate

The lack of morphological subject-direct object differentiation in German

noun phrases – it applies to feminine and neuter nouns in the singular and
nouns of any gender in the plural (see table 2) – is no oddity or failure but
typologically unremarkable. It has, however, severe consequences for on-
line differentiation (see table 3).

Table 3. Morphological differentiation of subjects from direct objects

German English
NP (without proper nouns) 25.0% 0.0%
Pronoun 50.9% 64.0%
Total (including proper nouns 34.9% 36.6%
and subject clauses)

Only 25% of all German noun-subjects were marked as non-accusatives in

my corpus, for German pronoun-subjects the figure is 50.9%.7 Overall,
only 34.9% of German subjects were morphologically differentiated from
accusative complements. This constitutes a considerable potential for tem-
porary ambiguity.
As English pronouns mark the subject-direct object differentiation more
strongly – 64.0% of pronoun subjects were marked as such –, a surprising
36.6% of English subjects were case-marked in my corpus. This means that
English, which is not considered a case language, shows slightly higher
morphological subject-direct object-differentiation than German.
The partial lack of subject-direct object-differentiation has very different
consequences for on-line processing in both languages. Let us first turn to
English: English subjects are easily identified in relation to verb position.
The subject-verb axis is as fundamental to modern English sentence con-
struction as verb-second is to German. Though inversion in English de-
clarative main clauses, possibly a remnant of former English verb-second,
has some text frequency – in my corpus, 5.04% (45/893) of declarative
main clauses were inverted – it is limited to particular structures that ex-
clude ambiguity: 4.37% (39/893) concerned preposed citations, of the re-
maining six cases five concerned the fronting of an adverb or a locative
phrase headed by a preposition and one a qualitative as-phrase.8 In three of
236 Klaus Fischer

these just the auxiliary was inverted, leaving three instances of main verb
inversion. Below I give examples for the different types of inversion (sub-
jects underlined):9
− preposed citation: 4.37% (39/893)

(2) ‘It’s – it’s true?’ faltered Professor McGonagall. (HP: 15)

− other preposed phrases 0.67% (6/893)

a. auxiliary inversion (3/893)

(3) But only in very few of them [countries] do these organizations

represent a dynamic element in the economy. (DE: 10)
(4) … for neither as a cat nor as a woman had she fixed Dumbledore
with such a piercing stare … (HP: 14)

b. main verb inversion (3/893)

(5) Inside, just visible, was a baby boy, fast asleep. (HP: 16)
(6) In its window, between coffee machines and toasters, stood a few
toys. (HDE: 16)

An initial English noun phrase can thus be interpreted as subject or non-

subject as soon as the next phrase is encountered:

Table 4. Disambiguation of English term complements

(7) Peter (,) Claire was (desperately) looking for (not Paul).

NP1 NP2 Vaux Adv V

pause,
new start
NP1 shown as NP2 shown as subject
non-subject

The identification of English direct objects is similarly unproblematic. This

is not what Hawkins’s model suggests. Because of the early verb position,
English clauses could tolerate a degree of complement permutation and
ensuing ambiguity. However, they do not.
The topological grammaticalisation of English term complements makes
case marking superfluous. It is tempting to see case marking in contempo-
 Temporary ambiguity of German and English term complements 237

rary English as redundant overspecification. However, English case mor-

phology has partially lost, or is about to lose, its relation to syntactic func-
tions as insecurities in case assignment and the tendency to use objective
case as default show (Who has done this? Me. Coll.: Me and Peter had a
good time. [pseudocorrect usage: between you and I]).
The situation is quite different in German. While the preverbal position
is a constructional sign for subjecthood in English, it is only a weak indica-
tor of subjecthood in German: just 56.6% of German subjects occupied the
first position in declarative main clauses as opposed to 37.9% in the third
position. If a preverbal German noun phrase belongs to the 64.9% without
definitive case marking, it can only be interpreted as a subject with a cer-
tain element of doubt and the interpretation might have to be revised. After
all 10.4% of noun phrase-accusative complements occur before the subject
where their identification is not topologically supported.10 This means in
effect that the majority of German sentences, quite unlike English sen-
tences, feature an inbuilt ambiguity, forcing greater context reliance:

Table 5. Disambiguation of German term complements

(7’) Peter hat Claire (verzweifelt) gesucht (nicht Paul).

NP1 Vaux NP2 Adv V

no clear disambiguation
(not even with contrastive stress)

This creates obvious problems for English learners of German who might
treat the preverbal position as a structural sign for subjecthood rather than a
weak indicator for it and therefore ignore other clues (see table 6).

Table 6. Topological marking of the subject

English German
NP (Adv) V NP V
preverbal position: constructional sign preverbal position: weak indicator for
for subjecthood subjecthood
position decoded position one of the clues that is in-
volved in inference

Language learning: interference

238 Klaus Fischer

If an initial noun phrase is not morphologically marked as subject or accu-

sative complement, agreement or the case marking of a later noun phrase
can lead to formal disambiguation. But in many of these cases, i.e. if it is
the second noun phrase in a main clause or verb agreement in a subordinate
clause that resolves the temporary ambiguity, disambiguation is later than
in English. Just considering the marked “accusative complement before
subject” order, my corpus featured 29 cases – this represents 2.0% of all
finite clauses – where temporary ambiguity was not resolved with the fol-
lowing constituent. In 10 cases – 0.7% of all finite clauses – there was no
formal resolution of the ambiguity (SUB = subject, AKK = accusative com-
plement):

(8) Dass das Meer … alles verschluckte: die Häuser und Brücken,
Kirchen und Paläste,
a. die [AKK] die Menschen [SUB] dem Wasser so frech aufs Ge-
sicht gebaut hatten. (HD: 7)
b. die [SUB] die Menschen [AKK] so lange erfreut hatten.

Obviously, the ambiguities are resolved using non-formal means, namely

verb meaning. Also, it is no accident that it is the contrast between subject
and direct object that tends not to be marked cross-linguistically as the
animate-inanimate opposition and others reduce ambiguity: the subject is
more likely to be animate, definite or a pronoun than is the direct object.
And permutations of German complements, while making on-line recogni-
tion of syntactic functions more difficult than in English, have well-known
discourse advantages while the rapid recognition of English term comple-
ments has verbose11 structures or, in Abraham’s words, sperrige Vertex-
tungsmittel [‘unwieldy discourse organisers’] such as cleft sentences as a
consequence (2003: 67).
But this does not detract from the fact that neither English nor German
behave in the way that Hawkins’s processing model suggests: in an impor-
tant respect German clauses display regular temporary ambiguities which
do not just irritate learners of German but catch out native readers as well.

5. Correspondence between syntactic functions and thematic roles

While Hawkins failed to discuss the mapping of phrases onto syntactic

functions, presumably because he sees them as unproblematic, he discusses
in depth syntactic restrictions and the mapping of syntactic functions onto
thematic roles in his 1986 monograph. A number of German verbs are
 Temporary ambiguity of German and English term complements 239

shown to have tighter semantic restrictions (1986: 28–35). Also, as a result

of the collapse of the English case system, former English dative and geni-
tive complements were turned into subjects and direct objects, mapping
non-prototypical thematic roles onto these two syntactic relations and mak-
ing them thereby less semantically transparent (1986: 53–73). The ensuing
ambiguity of the English subject and direct object relation was compen-
sated for by the early verb position, giving the hearer time to allocate the-
matic roles. Hawkins sees as another result of these changes that the Eng-
lish subject and direct object form supersets to the German subject and
direct object. As a consequence, English subjects and direct objects should
have higher text frequency than their German counterparts.
I established the text frequency of German and English term comple-
ments on the basis of the form-oriented classification widely used in Ger-
man valency research. For English I used topology as the form equivalent
of case (see endnote 1). Both the absolute number of each complement and
the frequency per verb phrase or clause are indicated (see table 7). The
statistics show that explicit English subjects have only a slightly higher text
frequency while German accusative complements were more frequent than
English direct objects. This surprising result even holds – at least propor-
tionally – if markers of inherent reflexivity are discarded, of which the
German texts had 77, the English one. This lowers the number of German
accusative complements to 748 (= 0.460 per verb phrase) vs. 763 English
direct objects (= 0.421 per verb phrase).

Table 7. Frequency of explicit term complements per verb phrase/clause (includ-

ing dummy complements [es, it, there] and complement clauses; E = Er-
gänzung [complement])12

German English

Case complements Topological complements

Subjekt13 0,889 0,927 subject
1296/1457 1356/1463
Akkusativ-E 0,508 0,421 direct object
825/1625 764/1814
Dativ-E 0,090 0,025 indirect object
146/1625 45/1814 (only noun phrases)
Genitiv-E 0,0006
1/1625
Präpositiv-E 0,120 0,116 prepositional object
195/1625 210/1814
240 Klaus Fischer

More interesting than the statistical data per se are the underlying reasons.
The reason for the lower frequency of German subjects was a preference
for co-ordinated clauses with subject deletion while English either used
subordinated structures or, less often, was more reluctant to delete the sub-
ject in co-ordinated structures:
a) English subordinate clause vs. German co-ordinate clause with subject
deletion (subjects underlined):

(9) When Dudley had been put to bed, he went into the living-room …
(HP: 10)
Er brachte Dudley zu Bett und ging dann ins Wohnzimmer …
(HPD: 10)

(10) „Uns ist etwas verloren gegangen“, sagte die Frau und schob ihm
ein Foto über den Schreibtisch. (HD: 9)
‘This is what we’ve lost,’ said the woman as she pushed the pho-
tograph across the desk. (HDE: 9)

b) English co-ordinate structure without vs. German co-ordinate structure

with subject deletion:

(11) Aber Prosper hatte ihm das Stehlen verboten und schimpfte ihn
jedes Mal fürchterlich aus, wenn er ihn dabei erwischte. (HD: 19)
Prosper had forbidden his brother to steal anything and he told
him off very harshly every time he caught him. (HDE: 19)

(12) „Du hast immer Hunger“, stellte Prosper fest, öffnete die Tür …
(HD: 16)
‘You’re always hungry,’ Prosper smiled. He opened the door …
(HDE: 16)

c) In addition, subject ellipsis was more frequent in the German than in the
English texts. German subject ellipsis vs. English lack of subject ellipsis:

(13) „Tja, wäre schön gewesen.“ (HD: 20)

‘Well, it would have been nice.’ (HDE: 19)

The higher text frequency of English subjects was not a result of more Eng-
lish verbs having a subject in their valency – all the verbs in the corpus
required at least a dummy subject – but was brought about by a reluctance
 Temporary ambiguity of German and English term complements 241

to delete subjects, incidentally leading to greater semantic transparency of

the English texts.
A more faithful English form-argument correspondence is not limited to
the subject. Rohdenburg (1990: 17; 1991: 18–25) has shown that a number
of German indefinite deletions cannot be replicated in English (direct ob-
jects underlined):

(14) Alkohol macht müde.

a. *Alcohol makes tired.
b. Alcohol makes you tired.

(15) Sie befahl/forderte dazu auf, das Gelände zu verlassen.

a. *She ordered/requested to leave the premises.
b. She ordered/requested people to leave the premises.

Krone (2003: 104; 231–233) found in a study of German and English foot-
ball match commentaries that German commentators are much more prone
than English commentators to delete direct objects referring to the ball or to
standard situations. Regularly German constructions with two complements
corresponded to English constructions with three (x flankt nach innen, lupft
in den Strafraum, köpft zu Ziege vs. X plants the ball in towards the penalty
area, heads it to the far side, nudges it out of play).
What about the semantic flexibility of the English subject? There was
some evidence that English verbs have indeed less tight semantic restric-
tions (subjects underlined):

(16) … the sign that said “Privet Drive” … (HP: 8)14

a.? das Zeichen, das „Ligusterweg“ sagte
b. … das Schild mit dem Namen „Ligusterweg“ … (HPD: 7)

Also some English verbs allow transitive and intransitive use while their
German counterparts require a reflexive marker:

(17) … Hagrid’s shoulders shook … (HP: 17)

a. Hagrids Schultern schüttelten sich
b. … Hagrids Schultern zuckten … (HPD: 21)

The texts in both languages featured a number of secondary subjectisations

though it was difficult to delineate them from metaphorical usage. Count-
ing generously, there were nine of them in the English and seven in the
German texts, none as dramatic as the examples documented in Rohden-
242 Klaus Fischer

burg (1974) that Hawkins quotes (1986: 60–61). The semantic role alloca-
tions indicated below demonstrate the unusual status of the subjects. I have
first listed thematic roles, which are largely situational, and second in italics
construction-induced roles.15 For instance, an agent or non-agent (thematic
roles) might be construed as exercising control over the action or process,
that is as an Agentive (construction-induced role) (cf. Fischer 1997: 55–65,
Ickler in this volume):16

(18) [ein Rollladen, breit und rostig], verschloss [die Eingangstür]. (HD: 23)
[Instrument, Agentive] [Patient, Objective]
[the entrance] was blocked off [with rusty shutters]. (HDE: 22)
[Patient, Objective] [Instrument, Co-presentive]

(19) [Auf seiner Nase] schälte [sich] [ein Sonnenbrand]. (HD: 9)

[Locus, Locative] [Objective] [Cause/Patient, Agentive]
[His nose] was peeling [from sunburn]. (HDE: 9)
[Locus/Patient, Agentive] [Cause, Originitive]

But it was in the German texts that a number of marked subjects of verbs
with active language constructions could be found (relevant subjects under-
lined, dative complements in bold):

(20) Uns ist etwas verloren gegangen. (HD: 9)

This is what we’ve lost.17 (HDE: 9)

(21) … since Madam Pomfrey told me she liked my new ear muffs.
(HP: 14)
… seit Madam Pomfrey mir gesagt hat, ihr gefielen meine neuen
Ohrenschützer. (HPD: 16)

These subjects are not first complements (or in other words: the last com-
plements to be bound by the verb; cf. Zifonun et al. 1997: 1303) and feature
non-prototypical thematic roles, making the German subject more abstract
and ambiguous in Hawkins’s sense.
Why are German accusative complements (direct objects) so numerous?
One important reason is prefix verbs, especially verbs with the prefix an-
(accusative complements and corresponding prepositional phrases under-
lined):

(22) … der sie von einem vorbeifahrenden Boot ankläffte … (HD: 13)
(22e) … barking at them from a passing barge … (HDE: 13)
 Temporary ambiguity of German and English term complements 243

(23) Mr Dursley blinked and stared at the cat. (HP: 8)

(23e) Mr. Dursley blinzelte und starrte die Katze an. (HPD: 7)

What Nichols (1986: 84) has called “headward migration” (of an adposi-
tion) has inflated the number of German accusative complements: the se-
mantic transparency has, so to speak, moved from the complement to the
verb. The thematic role of direction (for instance: auf etwas/wohin starren)
is frequently mapped onto the accusative complement (etwas anstarren),
turning it into the construction-induced objective role in the process. As a
result German accusative complements quite regularly have English prepo-
sitional objects as their counterpart. While German prefix verbs are a ma-
ture phenomenon, they increase the ambiguity of accusative complements
in Hawkins’s sense (that is in relation to thematic roles, not in relation to
construction-induced roles).
Some of Hawkins’s semantic claims could be confirmed. But his subset
relations, i.e. that the German subject and direct object are subsets of their
English counterparts, had to be refuted. In addition, several of the phenom-
ena that emerged as explanatory for relevant statistical data do not feature
at all in Hawkins’s typological account of the German-English contrasts. It
seems that the typological profiling of languages should beware of possibly
biased system accounts and supplement these with text-based analyses. My
corpus-based argument thus has wider application: it makes a case for a
“typology of parole”.

6. Verb position

If the two central term complements of subject and direct object are, at least
in a number of respects, more syntactically and semantically ambiguous in
German than in English, but if German nevertheless allows discourse-
induced permutations, then Hawkins’s semantic processing model seems to
have little relevance for modern German structures. Alternatively, one can
question Hawkins’s classification of German as a verb-final language.
Obviously, German has verb-object-verb order (V2 …V) but can one of
the two verb positions be established as basic or unmarked? The argument
for verb-final is that the constituent order in German subordinate clauses
mirrors the order by which complements are bound by the verb. But this
grammaticographic perspective collides with the typological demand that
basic orders be established in unmarked contexts. For clauses this is the
declarative main clause in the indicative present active. German emerges as
having “verb-object” as the unmarked order. This is borne out statistically:
244 Klaus Fischer

Table 8. Verb positions in declarative main clauses (Herr der Diebe ch. 1+2)18

V2 V2 … Vpart v2 … V

281 56 86
(66,4%) (13,2%) (20,3%)

However, if all the different sentence types are considered, the picture is
less clear. The main verb is in verb-second position in 45.8% of all clauses
as opposed to 42% in verb-final position:

Table 9. Verb positions in Herr der Diebe ch. 1+2

V1 V1 … Vpart v1 … V V2 V2 … Vpart v2 …V …V

15 4 10 281 56 86 161
(2,4%) (0,7%) (1,6%) (45,8%) (9,1%) (14,0%) (26,3%)

It is also worth mentioning that verb-final features in all the sentence types
while verb-first and verb-second define sentence types: the addition of all
the clauses that use verb-final in some form adds up to a small majority of
51.7%.19

7. Conclusion

The German subject and accusative complement create considerable tem-

porary ambiguity both because of ambiguous formal marking and the vari-
ety of thematic roles that can be associated with each of the two syntactic
relations. The positional marking of the English subject and direct object is
far less ambiguous than case marking in German. Somewhat surprisingly,
the subject-direct object-opposition is less frequently morphologically
marked in the case language German than in the non-case language Eng-
lish.
While there is evidence to support Hawkins’s claim that the English
subject and direct object have less tight semantic restrictions – at least with
a number of frequently used verbs – and readily allow the mapping of a
variety of thematic roles onto the two syntactic relations, the German sub-
ject and accusative complement have two quite frequently used sources of
non-prototypical thematic role assignment: active language verb usages for
the subject and prefixed verbs for the accusative complement (direct ob-
 Temporary ambiguity of German and English term complements 245

ject). It is not a forgone conclusion whether the English or the German

syntactic relations are more semantically transparent.
My findings are at odds with Hawkins’s claim of greater argument dif-
ferentiation in German because of its verb-final character. While the dis-
cussion about a basic verb position in German can probably be seen as fu-
tile as there is no speaker choice between the positions, typological mark-
edness suggests that verb-second should be seen as the basic verb position.
Statistically nearly 50% of German clauses in a narrative text showed the
whole main verb in an early position. This somewhat relieves the need for
German argument differentiation. Hawkins (1994) argued that German and
English recognition of constituent structure runs in parallel left to right
(rather than right to left in German as in true verb-final languages such as
Japanese). More often than not, German and English scenario recognition
works in parallel as well, starting with an early verb offering a basic sce-
nario, which will be confirmed or changed by evidence from later comple-
ments.
If German case counts as more mature than the grammaticalised posi-
tions of the English subject and direct object, then the less mature English
structures achieve greater syntactic-semantic transparency. Rohdenburg
(1991, 1992) and Fischer (1997: 297–315) have shown that greater English
semantic transparency applies to a number of constructions: for instance,
the development of Old English deverbal nouns into gerunds (see Fanego
2004) ultimately led to a more complex and semantically more transparent
system of English non-finite clauses than the German one.20 Though the
increase in English non-finite clauses – a change that very much informs
the verbal character of Modern English21 – was probably not driven by
semantic transparency, it certainly has it as its side effect.
John McWhorter (2002: 250) might be right that the loss of complexity
in the Middle English period has not been compensated by a concurrent or
more recent increase in complexity. But the mature English structures that
there are should be given due place in a typological assessment as should
the semantic transparency of English and German structures, both mature
and immature. A typological assessment of individual languages should
take a comprehensive rather than selective approach. But not all structures
are of equal importance in determining the character of a language: estab-
lishing the frequency of structures helped adjust typological claims about
German and English. An encompassing typological assessment cannot af-
ford to ignore texts.
While the lexically-centred valency approach with its bottom-up per-
spective naturally militates against taking typological shortcuts, it can also
lead to typological abstinence (“lost in lexicography”). It is desirable that
246 Klaus Fischer

the recent re-discovery of a typological valency perspective (Ágel 2000;

Fischer 2003), a perspective that was so important to Lucien Tesnière, the
creator of valency as a linguistic framework, is taken forward.

Notes

1. Term complements are complements that refer to entities rather than localities
or characterisations, so they are the subject and objects including prepositional
objects (cf. Zifonun et al. 1997: 1065–1099). I will limit my discussion to the
subject and direct object as the most central term complements. The use of the
term direct object does not imply passivisability. The direct object is rather de-
fined on the basis of proform replacement (both languages) and position (Eng-
lish): it is the Akkusativergänzung (accusative complement) of German valen-
cy work and the direct complement as defined in Fischer (1997: 99ff.): the
only, or if there is more than one the second, postverbal NP in unmarked order
that allows replacement by a personal pronoun in the objective case.
I would like to express my gratitude to John Hawkins, from whose ideas
and analyses, always lucidly expressed, I have benefited considerably. I would
also like to thank John McWhorter and Ian Roe for having commented on a
draft version of this article.
I am indebted to London Metropolitan University and the British Arts and
Humanities Research Board (now: Arts and Humanities Research Council) for
supporting my research by granting me two sabbatical terms (Research Leave
Scheme Award RL/AN6564/APN16978).
2. Olga Fischer implies that the replacement of the Old English object genitive
by (ultimately) several prepositions led to “finer semantic role distinctions”
(1992: 234).
3. One can add a further frame that has a similar relationship to Frame 5 as
Frame 3 has to Frame 4:
Frame 6. NP - V - NP -- NP I broke a world record
[Agent] [Patient] [Instru- with my guitar.
mental] Ich habe mit meiner
Gitarre einen Weltre-
kord gebrochen.

4. Hawkins (1986, 1992) uses the term semantic role which I reserve as a generic
term for different concepts of roles.
5. See the lexically induced hierarchisation of predicate frames in Willems and
Coene (2003) and the concept of a structured Satzmusterparadigma [‘para-
digm of sentence patterns’] derived from inherent verb meaning in Coene
(2004: 562). For Grundvalenz [‘basic valency’] see Welke (1988: 27) and
Fischer (2003: 27).
6. The gender hierarchy is sometimes given as “M, F < N” (e.g. Hawkins 2004:
72) as there are languages where F is less typologically marked than M. How-
 Temporary ambiguity of German and English term complements 247

ever, morphological marking and distribution in German support the original

hierarchy.
7. Ulrich Heid has kindly drawn my attention to Evert (2004) and Evert, Heid,
and Spranger (2004) who present data on the ambiguity of German noun
phrases (64605 tokens of common nouns) derived from an automated analysis
using the Negra corpus of German newspaper texts. Only 22.03% of noun
phrases were unambiguously marked for case, 21.40% did not contain any
case information at all, the rest was ambiguous between two or three cases.
Looking at candidates for nominative noun phrases, only 9.48% were marked
as non-accusatives. Though the data are not entirely comparable with mine –
neither syntactic function nor actual cases were ascertained – they point at
even greater ambiguity in journalistic texts than in the mainly narrative texts
which I have investigated and thus confirm my findings.
8. While many of my findings will be valid for spontaneous spoken text, this
does not apply in this case as inversions in declarative main clauses are largely
restricted to formal and/or narrative texts. Those that do occur in informal
speech are formulaic (Here comes the postman). (Cf. Quirk et al. 1985: 1379–
1383)
9. Note that preposing of the non-subject phrases in examples (3) and (4) forces
inversion, which thus constitutes a separate construction here, not just a con-
structional variant of a corresponding XSV-clause. The reason for this is the
restrictive or negative adverb in the preposed phrase which introduces a scope
restriction and the way in which the scope restriction is used: the statement is
by implication rather about the excluded cases than about those in the scope
restriction. Compare with: But in some of the countries these organizations
represent a dynamic element in the economy. Inversion is optional, at least in
principle, in the other examples (‘It’s – it’s true?’ Professor McGonagall fal-
tered), but cannot be realised in (5) to avoid ending a sentence with was.
10. In spontaneous spoken language this figure is likely to be lower.
11. The term – other than Abraham’s assessment – does not imply a value judge-
ment (see Dahl 2004: 60).
12. Complement labels for German are based in part on Engel and Schumacher
(1978), Engel (1988), in part on Zifonun et al. (1997). For English comple-
ment labels see Fischer (1997: 94–151).
13. The frequencies for Subjekt and subject were calculated on the basis of finite
clauses, for all other complements nonfinite clauses were considered as well.
14. Italics in the original text of examples (16) and (16b) were replaced by quota-
tion marks.
15. The thematic roles are the Fillmore-type thematic roles used by Hawkins (see
table 1), which capture a linguistic perspective only to a very limited degree. I
have slightly changed some of the labels to emphasise this and to reserve la-
bels ending in -ive for the construction-induced roles.
16. The dual role allocation demonstrates the problematic status of the various
concepts of semantic roles. It is also meant as an implicit criticism of Haw-
248 Klaus Fischer

kins’s analysis of secondary subjectisations that solely relies on thematic roles

and fails to address the semantic contribution of the construction.
17. The translation is rather free. (20) simply means: We’ve lost something.
18. V = main verb, v = auxiliary verb, Vpart = separated prefix.
19. I have only presented statistics on verb positions for a narrative text with a fair
amount of dialogue, but a separate count for a journalistic text confirmed the
distribution (see Fischer 2005b: 246, footnote 37).
20. The increase in verbal complexity was matched by a decrease in nominal
complexity as compact nominal structures were unravelled.
21. See the greater number of verbs in the English corpus texts (table 7).

References

Abraham, Werner
2003 Faszination der kontrastiven Linguistik ‚DaF‘: Der Parameter
‚schwere/leichte‘ Sprache unter typologischer Sicht. In Deutsch von
außen, Gerhard Stickel (ed.), 34–73. (Jahrbuch 2002 des Instituts für
Deutsche Sprache.) Berlin/New York: Walter de Gruyter.
Ágel, Vilmos
2000 Valenztheorie. Tübingen: Gunter Narr.
Coene, Ann
2004 Valenz und verbale Monosemie. Deutsche Kognitionsverben im
Lichte der strukturell-funktionellen Semantik und Koerzionstheorie.
Ph. D. diss., Faculteit der Letteren en Wijsbegeerte, Universiteit
Gent.
Croft, William
2003 Typology and Universals. 2d ed. Cambridge: Cambridge University
Press.
Dahl, Östen
2004 The Growth and Maintenance of Linguistic Complexity. Amsterdam:
John Benjamins.
Engel, Ulrich
1988 Deutsche Grammatik. Heidelberg: Julius Groos.
Engel, Ulrich, and Helmut Schumacher
1978 Kleines Valenzlexikon deutscher Verben. 2d ed. Tübingen: Gunter
Narr.
Evert, Stefan
2004 The statistical analysis of morphosyntactic distributions. In Fourth
International Conference on Language Resources and Evaluation,
Maria Teresa Lino, Maria Francisca Xavier, Fátima Ferreira, Rute
Costa, and Raquel Silva (eds.), 1539–1542. (Proceedings 3.) Paris:
ELRA.
Evert, Stefan, Ulrich Heid, and Kristina Spranger
2004 Identifying morphosyntactic preferences in collocations. In Fourth
 Temporary ambiguity of German and English term complements 249

International Conference on Language Resources and Evaluation,

Maria Teresa Lino, Maria Francisca Xavier, Fátima Ferreira, Rute
Costa, and Raquel Silva (eds.), 907–910. (Proceedings 3.) Paris:
ELRA.
Fanego, Teresa
2004 The rise and development of English verbal gerunds. Diachronica 21
(1): 5–55.
Fennell, Barbara
2001 A History of English: A Sociolinguistic Approach. Oxford: Black-
well.
Fischer, Klaus
1997 German-English Verb Valency. A Contrastive Analysis. Tübingen:
Gunter Narr.
2003 Verb, Aussage, Valenzdefinition und Valenzrealisierung: Auf dem
Weg zu einer typologisch adäquaten Valenztheorie. In Valenztheo-
rie. Neuere Perspektiven, Klaas Willems, Ann Coene, and Jeroen
Van Pottelberge (eds.), 14–64. (Studia Germanica Gandensia 2003–
2.) Akademia Press: Gent.
2005a Semantic transparency and ‘parole’: some statistical observations on
German and English sentences. In Getting into German. Multidisci-
plinary Linguistic Approaches, John Partridge (ed.), 181–217. Bern:
Peter Lang.
2005b Semantische Transparenz deutscher und englischer Satzstrukturen.
In Germanistentreffen Deutschland Großbritannien Irland. Dresden
2004, 219–250. Bonn: DAAD.
Fischer, Olga
1992 Syntax. In The Cambridge History of the English language. Vol. 2:
1066–1476, Norman Blake (ed.), 207–408. Cambridge: Cambridge
University Press.
Hawkins, John A.
1986 A Comparative Typology of English and German. Unifying the Con-
trasts. London/Sydney: Croom Helm.
1992 A performance approach to English/German contrasts. In New De-
partures in Contrastive Linguistics / Neue Ansätze in der Kontrasti-
ven Linguistik, Christian Mair, and Manfred Markus (eds.), 115–136.
(Innsbrucker Beiträge zur Kulturwissenschaft. Anglistische Reihe 4.)
Innsbruck: Universität Innsbruck.
1994 A Performance Theory of Order and Constituency. Cambridge:
Cambridge University Press.
2004 Efficiency and Complexity in Grammars. Oxford: Oxford University
Press.
Ickler, Irene
2007 Sentence patterns and perspective in English and German. This vol-
ume.
250 Klaus Fischer

Krone, Maike
2003 The language of football. A contrastive study of syntactic and se-
mantic specifics of verb usage in English and German match com-
mentaries. Ph. D. diss., Department of Language Studies, London
Guildhall University.
McWhorter, John
2002 What happened to English? Diachronica 19 (2): 217–272.
Nichols, Johanna
1986 Head-marking and dependent-marking grammar. Language 62: 56–
119.
Plank, Frans
1984 Verbs and objects in semantic agreement: Minor differences between
languages that might suggest a major one. Journal of Semantics 3:
305–60.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik
1985 A Comprehensive Grammar of the English Language. London/New
York: Longman.
Rohdenburg, Günter
1974 Sekundäre Subjektivierungen im Englischen und Deutschen. Verglei-
chende Untersuchungen zur Verb- und Adjektivsyntax. (PAKS Ar-
beitsbericht 8.) Bielefeld: Cornelsen-Velhagen & Klasing.
1990 Aspekte einer vergleichenden Typologie des Englischen und Deut-
schen. Kritische Anmerkungen zu einem Buch von John A. Haw-
kins. In Kontrastive Linguistik, Claus Gnutzmann (ed.), 133–152.
(Forum angewandte Linguistik 19.) Frankfurt M.: Peter Lang.
1991 Weitere Betrachtungen zu einer vergleichenden Typologie des Engli-
schen und Deutschen. (Series A Paper No. 302.) Duisburg: Linguis-
tic Agency University of Duisburg.
1992 Bemerkungen zu infiniten Konstruktionen im Englischen und Deut-
schen. In New Departures in Contrastive Linguistics/Neue Ansätze in
der Kontrastiven Linguistik, Christian Mair, and Manfred Markus
(eds.), 187–207. (Innsbrucker Beiträge zur Kulturwissenschaft. Ang-
listische Reihe 4.) Innsbruck: Universität Innsbruck.
Sperber, Dan, and Deirdre Wilson
1995 Relevance. Communication and Cognition. 2d ed. Oxford: Black-
well.
Thomason, Sarah Grey, and Terence Kaufman
1988 Language Contact, Creolization, and Genetic Linguistics. Berkeley,
CA: University of California Press.
Wegener, Heide
1985 Der Dativ im heutigen Deutsch. Tübingen: Gunter Narr.
Welke, Klaus M.
1988 Einführung in die Valenz- und Kasustheorie. Leipzig: VEB Biblio-
graphisches Institut.
 Temporary ambiguity of German and English term complements 251

Willems, Klaas, and Ann Coene

2003 Argumentstruktur, verbale Polysemie und Koerzion. In Valency in
Practice / Valenz in der Praxis, Alan Cornell, Klaus Fischer, and Ian
Roe (eds.), 37–63. Genf: Peter Lang.
Zifonun, Gisela, Ludger Hoffmann, Bruno Strecker, Joachim Ballweg, Ursula
Brauße, Eva Breindl, Ulrich Engel, Helmut Frosch, Ursula Hoberg,
and Klaus Vorderwülbecke
1997 Grammatik der deutschen Sprache. Berlin/New York: Walter de
Gruyter.
Sentence patterns and perspective in English and
German

Irene Ickler

1. Sentence pattern and verb valency

It is my aim to show that sentence patterns can be related to specific func-

tional meanings which can to a certain extent be described independently of
the verbal lexeme. The functional meaning of the sentence pattern coordi-
nates with verb valency in such a way that it either corresponds to part of
the verbal meaning or adds to the (compatible) verbal meaning. Verbs can
be compatible with different sentence patterns which is not surprising be-
cause sentence patterns are linked and related to one another. Verbal word
formation may further reflect the functional meaning of a type of sentence
pattern and so make the verb correspond to it.1

2. A systematic change of perspective

The meaning conveyed by the sentence pattern has to do with the perspec-
tive chosen by the speaker. Languages provide means to vary these per-
spectives in a systematic way.
I want to concentrate on one special systematic alternation of perspec-
tives that is omnipresent and similar in English and German. It is the con-
verse relation between two entities regularly referred to in verb comple-
ments: the Dynamic and the Static Entity. At any one time only one can be
mentioned in a superior syntactic position. The relative position and status
of these entities in a specific perspective is indicated by the grammar of the
sentence pattern as a unit, especially by the combination of word order,
cases, prepositions, prefixes and verb particles as well as the form of the
perfect tense. It is important to stress that these formal features of a sen-
tence pattern are more or less ambiguous if studied separately. Only in a
combined construction are they able to convey functional meaning. In this
approach the lexical meaning of the verbal stem or the complement nouns
will only be drawn upon if the rest of the construction is ambiguous.
254 Irene Ickler

The criterion of passivizability will help to distinguish the two above-

mentioned entities from the third main entity typically encoded as a verb
complement: the Causative Entity.

2.1. Examples

Before I define these three basic semantic roles I would like to cite a set of
examples that immediately demonstrates what I mean:

(1) a. Sie Desinfektionsmittel auf die Wunde. (sprühte)

She disinfectant on the wound. (sprayed)
X Y Z

X is the Causative Entity that always fills the subject position if and only if
the sentence is passivizable. Y is the Dynamic Entity which is found here in
a position that is relatively central and superior to Z’s position. Z is the
Static Entity. Please note that there is no need to look at the verb in order to
identify the Dynamic and the Static Entity. It is the divalent local preposi-
tion auf/on that clearly assigns a Static Entity to the NP following it and a
Dynamic or potentially Dynamic Entity to the NP preceding it.

(1) b. Sie die Wunde mit Desinfektionsmittel. (besprühte)

She the wound with disinfectant. (sprayed)
X Z Y

In sentence (1b) now Z, the Static Entity, is placed in the superior position.
The non-local preposition mit/with introduces a prepositional object that
carries the Dynamic Entity Y. So here the preposition with is converse to
the local preposition in (1a).

(2) a. Luft in den / dem Ballon. (strömt / ist)

Air into / in the balloon. (streams / is)
Y Z

In (2a) the Dynamic Entity Y fills the subject position because there is no
Causative Entity to claim its right to the highest position.

(2) b. Der Ballon Luft. (verliert / enthält)

The balloon air. (loses / contains)
Z Y
 Sentence patterns and perspective in English and German 255

In (2b) we need the verb to tell us that the transitive-looking sentence is in

fact non-passivizable, so that there is no Causative Entity to fill the subject
position. But once that is found out, it is quite easy to conclude the func-
tional meaning of this sentence pattern: it is the pattern of ‘Having or In-
corporating and opposite’ (pattern 10) with the Static Entity Z superior to
the Dynamic Entity Y.
The change of perspective in the sentences (a) vs. (b), where either Y or
Z is central, is systematically indicated in speech. Below I list more sen-
tence patterns that relate to one another in this respect.

2.2. The three basic perspective roles

Static Entity Z: the Static Entity Z has (= is represented as having) a stable

local position relative to the other entities and to the speaker’s standpoint: Z
does not appear, does not disappear nor does it move from one place to
another.
Dynamic Entity Y: the Dynamic Entity Y has no stable local position. In
the representation of a process Y moves relatively to Z and to the speaker’s
standpoint: Y appears, disappears and moves from one place to another. In
the representation of a state Y’s position is seen as the more mobile or more
uncertain in relation to Z’s position.
Causative Entity X: the Causative Entity X acts either primarily on Z or
primarily on Y. If X acts primarily on Z, it affects Z. If X acts primarily on
Y, it effects Y, that is it either produces Y, extinguishes Y or it moves Y
from one place to another.2

2.3. The specific perspective roles of the Static Entity Z and the Dynamic
Entity Y

Y and Z are very basic semantic roles, but they are linguistic and perspec-
tive roles, as can be seen in mother beside father, where mother is set as the
Dynamic Entity Y and father as the Static Entity Z, and the other way
round in father beside mother. Y and Z each assume different specific per-
spective roles depending on the relative status they hold in a perspective:
− If Z has a more peripheral status in the perspective it is seen as a loca-
tion (which may be split into source, path, goal and position). This loca-
tion is not holistically affected by an action, process or state.
256 Irene Ickler

− If Z has a more central status it is seen as the entity that has, acquires or
loses something or that changes a feature. This entity is seen as holisti-
cally affected by the action, process or state.
− If Y has a more peripheral status in the perspective it is seen as a feature
added to or taken from Z.
− If Y has a more central status it is seen as an entity whose (change of)
location or existence is of prime interest.

3. Hierarchy of grammatical functions

In a sentence pattern the specific perspective roles are assigned different

grammatical functions according to their status (central or not) as well as
their basic role (X, Y or Z).
Here I only want to refer to a section of the hierarchy of grammatical
functions that shows the accessibility for grammatical regularities like pro-
nominalization, passivization, congruence, case-marking, reflexivation and
others:

(3) Subject > Direct Object > Indirect Object > Prepositional Objects/
Local Adverbials

As we shall see later, the indirect object can be seen as a threshold between
the central functional complements that indicate different semantic roles in
different patterns and the marginal (oblique) complements that indicate one
specific semantic role.
The assignment of perspective roles and grammatical functions will be
demonstrated in the following list of sentence patterns and stated again at
the end of this paper.

4. List of sentence patterns and their functional meaning

In the following list twelve sentence patterns are marked by the grammati-
cal functions of their complements, the assigned basic semantic roles, a
convenient but not precise label for their functional meaning and the ex-
plicit functional meaning in brackets. They are exemplified by English and
German model sentences and further verbs that fit the patterns. The pat-
terns are not only listed but also linked to one another, i.e. I will mention
connections to already introduced patterns.
 Sentence patterns and perspective in English and German 257

I will start with the sentence pattern of “Transfer”. This functional

meaning is well-marked by the specific combination of complements.
Three main types of verbal lexemes fit into this frame: 1, simplex verbs
whose lexical meanings correspond totally or partly to the functional mean-
ing of the pattern (e.g. put, send); 2, simplex verbs that do not carry this
functional meaning intrinsically but that can be understood as compatible
(e.g. crash, rub). The advantage here is that the verbal lexemes can add a
great variety of modifying meaning to the functional meaning of the pat-
tern; 3, particle or phrasal verbs that carry a part of the functional meaning
in their particle. We will observe these complex verbs as we proceed.

I SU + DO + L: ‘transfer’
X Y Z
(X causes Y to move/stay relatively to Z)
a. so. puts sth. somewhere:
drive, roll, crash, pack off, bring away, send on, take down, carry
off, drive back, drag out
b. so. removes sth. from somewhere:
get, shake, throw off, exclude, extract, take out/off, sweep away,
fish out, pull back, pick up
German:
a. jmd. tut etw. irgendwohin:
bringen, kleben, einparken in, abschieben aus, umleiten nach,
anbauen an, aufladen auf
b. jmd. nimmt etw. irgendwoher:
zurückziehen, wegreißen, abwischen von, aufsammeln von, aus-
pressen aus, entnehmen aus

The verbal particle specifies the directional movement of Y in relation to a

Z which it implies. It also changes the status of the locational complement
from obligatory to optional. The preposition of this complement is obliga-
torily di-valent and specifies a local relation between Y and Z. As periph-
eral Z splits up into source, goal, etc., the particle and the preposition can
bind different aspects of Z. Interestingly, in English there seems to be a
barrier to the particle and the preposition being congruent, whereas the
opposite is the case in German.
The second sentence pattern is similar to the first, only the Static Entity
Z is not merely marked as a location for Y but as a person who may receive
Y. This special status makes Z the third entity in the speaker’s focus of
interest: it can be assigned a central position as an indirect object. Alterna-
tively it can be assigned a more marginal position as a prepositional object:
258 Irene Ickler

II i) SU + IO + DO
X Z Y
ii) SU + DO + PO to: ‘give’
X Y Z
(X causes Y to be at Z’s disposal and opposite)
a. so. gives so. sth. / so. gives sth. to so.:
bring, offer, tell, show, hand on/over/up/down, send off, present
(to), allow (to), supply (to)
b. so. charges so. sth. / so. takes sth. from so.:
fine, deny, refuse, save, spare, envy, ask (of), demand of/from,
steal from, take away from
German: i) SU + IO + DO
ii) SU + DO + PO an/zu
a. jmd. übergibt jmdm. etw./jmd. übergibt etw. an jmdn.:
(weiter-)geben, anbieten, vorstellen, zuflüstern, aufbürden, lie-
fern (an), bringen (zu)
b. jmd. nimmt jmdm. etw./jmd. nimmt etw. von jmdm.:
abnehmen, wegnehmen, ausziehen, vorenthalten, entlocken, steh-
len (von), kaufen von

Patterns i) and ii) differ slightly in perspective: in i) two personal individu-

als, the agent X and the receiver Z, are typically placed near each other in a
central position. Where the agent acts, the receiver is expected to re-act in a
mentally receptive way. This sentence pattern is twofold passivizable,
which means that both objects can be made subject.3 In ii) Z has more local
features. The mentally re-acting feature is neutralized. Hence I would pre-
fer to call Z ‘addressee’ in this construction.
Of course there are syntactic restrictions to the English intern IO, which
is why nearly all verbs that fit into construction i) also fit into ii). But many
verbs with a similar meaning are more or less restricted to construction ii);
for example, verbs like shout to or buy from make do with the neutralized
‘addressee’-construction. In German more fine-tuning is possible because
of the dative case-marking combined with productive particle verbs like in
zurufen or abkaufen. Here the particles are often used to make verbs cor-
respond to this special pattern and therefore then the complements are
obligatory (otherwise their function is similar to that in pattern I).
The third sentence pattern offers the possibility of expressing a perspec-
tive that is converse to both patterns I and II: here the Static Entity Z takes
on the central position of the direct object.
 Sentence patterns and perspective in English and German 259

III SU + DO + PO with (of): ‘provide with’

X Z Y
(X causes Z to have Y and be holistically affected by this and op-
posite)
a. so. provides so./sth. with sth.:
fill (up/in), deck (out), cover (up), surround, (over-/under-)stock,
load (up/down), paste over
with PO of: remind of, inform of, notify of, suspect of, accuse of
b. so. rids so./sth. of sth.:
deprive, rob, rid, relieve, clear of/from, discharge of/from, exon-
erate of/from, absolve from
German: SU + DO + PO mit (GO)
a. jmd. versieht jmdn./etw. mit etw.:
(be-/ver-/ab-/an-/auf-/aus-)füllen, (be-/zu-)decken, einhüllen,
überziehen, untermauern
with GO: beschuldigen, bezichtigen, überführen
b. jmd. befreit jmdn. von etw.:
freisprechen, erlösen, entlasten, entbinden, reinigen, räumen,
heilen
with GO: berauben, entledigen, entwöhnen, entheben

German pattern III entails quite regular and motivated verbal prefigation
(be-, ver-, über-, ent-) that makes verbs correspond to the functional mean-
ing of this pattern type. Strikingly, here the stressed local verb particles like
ein-, aus-, ab-, an- in German or in, out, up, down in English have a some-
what different meaning as compared to patterns I and II. In pattern III they
do not indicate a movement of the direct object entity, which is Z (e.g. to
fill up a bottle, to fill in a hole). Instead, an aspectual meaning is added to
them: the holistic and telic meaning of the sentence pattern is intensified or
modified by them. In addition, they imply Y by stating its local relation to
Z. So these particles in their converse variant can also be said to corre-
spond to the sentence pattern. The third complement carrying Y is optional
here, though the special form of GO/PO of is mostly only contextually
optional.4
If we now reduce the trivalent sentence patterns I, II and III to just a di-
valent pattern of SU + DO, the structural meaning will be more ambiguous,
because what we lose are the less central complements that disambiguate
the meaning of the pattern. Still I find it justifiable to distinguish two pat-
terns: pattern IV with a Dynamic Entity Y in the DO and pattern V with a
Static Entity Z in the DO (e.g. to wipe off crumbs vs. to wipe off the table;
260 Irene Ickler

to supply food vs. to supply shops; similar in German with the verbs ab-
wischen; liefern vs. beliefern).

IV SU + DO: ‘effect’
X Y
(X causes Y to exist or to emerge somewhere and opposite)
a. so. produces sth. / so. takes sth. in / so. extinguishes sth.
make, say, construct, set up, make up/out, bring up/about/along,
erase, absorb, dispel
b. so. stores sth. up / so. removes sth. (can be extended to sentence
pattern I):
write down, put away/across/out/off, box in/up, bottle, imbed, in-
volve, exclude, give out
German:
a. jmd. bringt etw. hervor / jmd. nimmt etw. ein / jmd. löscht etw.
aus:
formen, erschaffen, aufbauen, aussagen, abgeben, aufnehmen,
herstellen, (ver-)tilgen
b. jmd. speichert etw. / jmd. entfernt etw. (can be extended to sen-
tence pattern I):
aufschreiben, einkellern, entsenden, auslagern, ab-/weg-/los-
schicken, hin-/herbringen

V SU + DO: ‘affect’
X Z
(X affects Z holistically)
a. so. touches so. / so. treats sth.:
caress, kiss, love, redden, clean, steam, damage, overbid,
underestimate, wash (out)
b. so. (dis-)arms so. (can be extended to sentence pattern III):
crown, spice, colour, cork (up), roof in/over, plaster (up/over),
depopulate, weed, heal
German:
a. jmd. berührt jmdn. / jmd. bearbeitet etw.:
streicheln, küssen, röten, dämpfen, be-/anlächeln, verändern,
überfahren, durchbohren
b. jmd. ent-/bewaffnet jmdn. (can be extended to sentence pattern
III):
krönen, (be-)wässern, benachteiligen, verkorken, unterkellern,
überdachen, einfetten
 Sentence patterns and perspective in English and German 261

In spite of lexicalization there are typical prefixes and particles for each
pattern. There tend to be more particle verbs in IV and more simplex and
prefixed verbs in V (e.g. Weizen anbauen / grow wheat vs. den Acker be-
bauen / till the soil). In addition, denominal verbs in IV encapsulate Z (as
in einsargen / to coffin) and denominal verbs in V encapsulate Y (as in
bewaffnen / to arm).
If we reduce sentence pattern I by the Causative Entity X we get pattern
VI (e.g the boy rolls the ball across the road vs. the ball/boy rolls/goes
across the road). The Dynamic Entity Y advances into subject position and
the Static Entity Z remains in its local adverbial. This pattern can offer a
different perspective to sentence pattern V (e.g. the boy crosses the road),
where the subject refers to a Causative Entity and the Static Entity has
advanced from location to holistically affected entity.

VI SU + L: ‘move/stay’
Y Z
(Y moves/stays relatively to Z)
a. so./sth. moves somewhere:
go, come, crash, lie (down), sit (down), stream, drip, go
off/out/back, run off/away from
b. so./sth. is somewhere:
be (situated), stay, hang, remain, stick, live, lodge, occur, exist,
sit, stand, lie
German:
a. jmd./etw. bewegt sich irgendwohin:
gehen, kommen, tanzen, sich (hin-)setzen, einsteigen in, ausstei-
gen aus, durchfahren durch
b. jmd./etw. befindet sich irgendwo:
sein, sich befinden, sich aufhalten, bleiben, stecken, wohnen, vor-
kommen, stehen, sitzen

This pattern is not passivizable. The German verbs in (a) form their perfect
tense with the auxiliary sein, even those that do not do so in other sentence
patterns like tanzen, krachen, schnaufen, tröpfeln, or those that can also be
used with transitive sentence pattern I, such as fahren, fliegen, rollen,
krachen.
To indicate the distinction I would like to point out the similar sentence
pattern VII, which is passivizable and in which the German verbs form
their perfect tense with the auxiliary haben.5 Here pattern I is not reduced
by the Causative Entity X but by the Dynamic Entity Y.
262 Irene Ickler

VII SU + PO: ‘aim at’

X Z
(X directs his action towards Z and may affect Z non-holistically
because of that)
laugh at, look after, object to, rely on, comment on, believe in,
break in on, dwell upon
German:
lachen über, aufpassen auf, anspielen auf, raten zu, glauben an,
greifen nach, schwelgen in

The prepositions are similar to those in patterns I and VI, but instead of
being divalent with the Dynamic Entity Y preceding them and the Static
Entity Z following them, they are monovalent. We may think of Y as being
the action itself, having melted into the verbal meaning, thereby attaching
the preposition more closely to the verb and promoting Z from location to
non-holistically, indirectly affected object.6
If we reduce sentence pattern II by the Causative Entity X we get sen-
tence pattern VIII (e.g. someone gave it to me vs. it belongs to me). Again
the Dynamic Entity Y advances to the subject position and the Static Entity
Z remains in its position.

VIII SU + IO (PO to): ‘come to/ belong to’

Y Z
(Y comes to / is with a mental Z and opposite)
a. sth. occurs to so. / sth. belongs to so.:
happen to, appear to, go to, come (in handy) to, appeal to, be
useful to, correspond to
b. sth./(so.) springs from so.:
come from, escape from, slip (away) from, issue from, (fail so.,
escape so., slip so.)
German: SU + IO (NP dat)
a. etw./jmd. erscheint jmdm. / etw. gehört jmdm.:
begegnen, widerfahren, unterlaufen, zustoßen, passen, einfallen,
(aus-)reichen, auffallen
b. etw./jmd. entgeht jmdm. / etw. fehlt jmdm.:
entfallen, entfahren, entgleiten, entweichen, entwischen, weglau-
fen, abgehen, ausweichen

This pattern is not passivizable and most German verbs form their perfect
tense with the auxiliary sein.
 Sentence patterns and perspective in English and German 263

To indicate the distinction I would like to point out a similar sentence

pattern which is passivizable and in which the German verbs form their
perfect tense with the auxiliary haben. Here pattern II is not reduced by the
Causative Entity X but by the Dynamic Entity.

IX SU + IO (PO to): ‘pay attention to’

X Z
(X directs his/her attention towards a mental Z and may initiate
an interaction with Z because of this)
attend to, listen to, talk to, speak to, read to, write to, preach to,
pray to, lie to, smile to
German: SU + IO (NP dat)
helfen, danken, gratulieren, zuhören, zustimmen, zulächeln,
nachspionieren, vortanzen

If we reduce sentence pattern III by its Causative Entity X, we have pattern

X with the Static Entity Z in subject position. Interestingly the Dynamic
Entity Y can be stated in different syntactic positions, which means that
there are subvariants:

X i) SU + DO
Z Y
ii) SU + PO with/of: ‘incorporate/have’
Z Y
(Z has/gets Y and opposite)
a. so. has sth. / so. gets sth. (also intellectually):
with DO: possess, know, deserve (praise), contract (a bad habit),
lack, need, want (care)
with PO: be full of, cope with, become possessed with, do
with(out), be free of
b. sth. contains sth. / sth. fills with sth.:
with DO: comprise, seat, house, fit, admit, emit, dispense, give,
hold, shed, lose, leak, seep
with PO: fill with, charge up with, swarm with, buzz with, crawl
with, be clear of
264 Irene Ickler

German: i) SU + DO
ii) SU + PO mit / GO
a. jmd. hat etw. / jmd. bekommt etw. (auch geistig):
with DO: besitzen, bekommen, erhalten, erfahren, ermessen, ken-
nen, merken, verlieren
with GO: sich erfreuen, sich bewußt sein, bedürfen, entraten, er-
mangeln, voller Y sein
with PO: reichen mit, begabt sein mit, von sich geben, frei
sein/werden von, voll sein von
b. etw. enthält etw. / etw. füllt sich mit etw.:
with DO: beinhalten, (um-)fassen, einschließen, ergeben, betra-
gen, wiegen, messen, kosten
with GO: ermangeln, voller (Y) sein/liegen/hängen,
with PO: sich füllen/anreichern/aufladen mit, vollhängen von,
wimmeln von

The transitive-looking pattern is in fact not passivizable. Pattern X offers a

real converse to patterns VI (e.g. water drips into/out of the basin vs. the
basin fills with water / leaks water) and VIII (e.g. the ring belongs to Mary
vs. Mary possesses a ring).
If we now reduce the divalent patterns IV und V by the Causative Entity
we get the two parallel monovalent patterns XI and XII. They are distin-
guished semantically by the fact that either the Dynamic or the Static En-
tity has advanced into subject position. Please compare the sentences in
(4).

(4) a. sth. has appeared b. so. has laughed

(ist erschienen) (hat gelacht)
so. has arrived sth. has reddened
(ist angekommen) (hat sich gerötet)
steam has developed the water has steamed
(ist entstanden) (hat gedampft)
water has leaked out the tank has leaked
(ist ausgelaufen) (hat geleckt)
SU SU
Y Z

In (a) the subject entity has changed its local position or has come into
existence. In (b) the subject entity has not changed its position; instead it is
assigned a certain feature.
 Sentence patterns and perspective in English and German 265

XI SU: ‘exist/appear/move’
Y
(Y is or appears somewhere and opposite)
a. sth./so. appears/disappears:
come (about/off/up), emerge, arise, break out, spring up, go
(away/by/out), thaw (off)
b. sth./so. moves on/stays (can be extended to sentence pattern VI):
go (up/down/on/in), climb (up/down), drip (out), flow down/off,
drain off, start off, be on
German:
a. etw./jmd. erscheint/verschwindet:
entstehen, (an-)kommen, (er-)wachsen, aufkommen, austreten,
(weg-)gehen, (ab-)schmelzen
b. etw./jmd. bewegt sich fort/bleibt (can be extended to sentence
pattern VI):
(auf-/an-/ab-)steigen, auf-/ab-/fahren, (ab-)sinken, ausgehen,
da/an/aus/weg/vorbei sein

XII SU: ‘characterization’

Z
(Z has or changes a feature / gives off or takes on something)
a. so. laughs / sth. foams:7
sleep, talk, read, sing, box, hunt, steam, sprout, hay, curse,
dream, joke, drip, leak
b. so. blushes / sth. reddens / sth. is red:
age, freeze, bleach, defrost, drain, mold, pale, open (up), close
(down), shut (up), cool (down)
German:
a. jmd. lacht / etw. schäumt:
schlafen, wursten, knallen, tropfen, abtauen (Kühlschrank), aus-
dunsten, aufheulen, abfärben
b. jmd. errötet / etw. rötet sich / etw. ist rot:
altern, (ge-)frieren, trocknen, (er-)bleichen, veralten, aufbersten,
(zer-)brechen, auslaufen

All German verbs in XI form their perfect tense with the auxiliary sein, but
so do the transformative verbs in (XIIb). There tend to be more particle
verbs in XI and more simplex but denominal and deadjectival verbs in XII.
The particles in XI imply Z and those in XII have the converted, aspectual
meaning. The denominal verbs in XII encapsulate Y (as in smoke > emit
266 Irene Ickler

smoke, haaren > Haare verlieren) and those in XI encapsulate Z (as in to

emplane, sich einschiffen).

5. Conclusion

5.1. Perspectives and their sentence patterns

In the following survey the perspectives I discussed are classified accord-

ing to the presence of a Causative Entity (+X: “Causative Perspective” and
–X: “Process/State Perspective”) and according to the syntactic priority of
Dynamic versus Static Entity (Y > Z: “Existence Perspective” and Z > Y:
“Characterization Perspective”). These types are subclassified into their
sentence patterns. Converse patterns are linked with a line.

I. CAUSATIVE PERSPECTIVE

A. EXISTENCE PERSPECTIVE B. CHARACTERIZATION PERSPECTIVE

1 SU + DO + L ‘transfer’
X Y Z 3 SU + DO + PO (/GO) ‘provide with’
X Z Y
2 SU + DO + IO ‘give’
X Y Z

4 SU + DO ‘effect’ 5 SU + DO ‘affect’
X Y X Z

II. PROCESS/STATE PERSPECTIVE

A. EXISTENCE PERSPECTIVE B. CHARACTERIZATION PERSPECTIVE

6 SU + L ‘move’
Y Z 10 SU + DO (PO/GO) ‘have’
Z Y
8 SU + IO ‘belong to’
Y Z

11 SU ‘existence’ 12 SU ‘characterization’
Y Z

The left side (which has to do with BEING somewhere) and the right side
(which has to do with HAVING a feature), each combine with their typical
 Sentence patterns and perspective in English and German 267

variants of particles and prefixes in complex verbs. There is also a typical

distribution of the auxiliaries sein and haben in German perfect tense.8

5.2. The naming of the specific perspective roles

Semantic roles have been identified and variously classified in many

splendid studies, first and foremost by Charles Fillmore. The following
open list of specific perspective roles is familiar but I would like to empha-
size its anchorage in the syntactic construction of a sentence pattern
through the three basic perspective roles:

SYNTACTIC FUNCTION BASIC PERSPECTIVE SPECIFIC PERSPECTIVE ROLE

ROLE
SU X ‘Causative’ Agent
Y ‘Dynamic’ Existing Entity
Z ‘Static’ Characterized Entity
DO Y ‘Dynamic’ Effected Entity
Z ‘Static’ Affected Entity
IO, PO to/an Z ‘Static’ Recipient, Addressee
L, PO at/on/in Z ‘Static’ Locative Entity,
Indirectly Affected Entity
PO with/mit/of/von/GO Y ‘Dynamic’ Ornative Entity

Notes

1. This attempt is a development of Ickler (1985; 1990; 1993). My basic ap-

proach owes to or was encouraged by Hans Altmann as well as works by Fill-
more (1988), Goldberg (1995), Langacker (1987; 1991), Wierzbicka (1988),
Zaima (1987) and others.
2. Please note that “effect” is used in a broader sense than usual, also including
the causation of an entitity’s appearance or disappearance relative to a stand-
point. By comparison, “affect” does not mean change of location or existence,
but change of feature.
3. The dative-passive is much more restricted in German than in English and it
requires its own auxiliary kriegen/bekommen. The use of dative in German
may be extended to “mental” entities like in dem Text etwas entnehmen, je-
manden einer Richtung zuordnen, but here no dative-passive is possible.
268 Irene Ickler

4. A verb having an optional complement just means that this verb is compatible
with more than one sentence pattern. These patterns differ in length and func-
tional meaning, but are, of course, related in form and meaning. Please see
patterns IVb) and Vb) for examples.
5. In German the subjectless passive which basically can be applied to all actions
can easily be applied here with all complements and their prepositions, com-
pare: Dazu wird (von dem Anwalt) geraten. vs. *Dazu wird (von den Kindern)
gegangen. Similar in English: It was objected to (by many members). vs. *It
was gone to (by the children).
6. Compare an etw. arbeiten (non-holistically affected) and etw. bearbeiten (ho-
listically affected), an einem Roman schreiben (non-holistically affected) and
einen Roman schreiben (effected). Similar in English: be at sth. and do sth.
7. Many verbs in this pattern can be combined with an optional direct object, that
is, they are also compatible with patterns 4, 5, or 10, e.g. sing (a song), hunt (a
deer), leak (oil).
8. Typical complex verbs on the left side are combined with local adverbial par-
ticles like in, out, off, up, down / ein, aus, ab, an and prefixes like con-, ex-,
dis- / er-, ver- and typical complex verbs on the right side are combined with
aspectual particles (homophonous to the ones on the left side) or prefixes like
over-, under- / be-, ver-.

References

Fillmore, Charles
1988 The mechanisms of “construction grammar”. In Proceedings of the
Fourteenth Annual Meeting of the Berkeley Linguistic Society, 35–
55.
Goldberg, Adele
1995 Constructions. A Constructional Approach to Argument Structure.
Chicago: University of Chicago Press.
Ickler, Irene
1985 Verb und Verbzusatz. Zur grammatischen Beschreibung von Parti-
kelverben und partikelverbähnlichen Strukturen. M.A. thesis, De-
partment of German Philology, University of Munich.
1990 Kasusrahmen und Perspektive. Zur Kodierung von semantischen
Rollen. Deutsche Sprache 1: 1–37.
1993 Kasusrahmen und Perspektive im Deutschen und Englischen. Ger-
manistische Linguistik 119–120: 151–200.
Langacker, Ronald
1987 Foundations of Cognitive Grammar. Vol. 1. Stanford: Stanford Uni-
versity Press.
1991 Foundations of Cognitive Grammar. Vol. 2. Stanford: Stanford Uni-
versity Press.
 Sentence patterns and perspective in English and German 269

Wierzbicka, Anna
1988 The Semantics of Grammar. Amsterdam/Philadelphia: John Benja-
mins Publishing Company.
Zaima, Susumu
1987 “Verbbedeutung” und syntaktische Struktur. Deutsche Sprache 1:
35–45.
Contrasting valency in English and German

Brigitta Mittmann

In 2004, two large-scale monolingual valency dictionaries appeared – one

for English (VDE) and one for German (VALBU). Now that most of the
valency structures of these two languages have been recorded, the question
of contrasting these structures in a bilingual reference work has become
even more interesting. However, contrasting valency structures and patterns
across languages is a complex task.
This contribution to the present volume explores different approaches to
comparative valency research. It seems that for certain applied purposes –
and not just for translating or lexicography – an approach which takes the
text production situation into account and uses frames or constellations of
semantic roles as tertium comparationis is the most useful one. More spe-
cifically, this paper looks at a set of verbs that are highly relevant for a
particular text type. It uses them as a background for some observations on
contrastive valency studies, on text types and text production, perspective,
morphological forms, and dictionaries.

1. The “monosemy problem”

During the conference that this collection of articles is based upon, various
speakers expressed the view that words are essentially monosemous. When
comparing two languages, it is difficult to maintain this position, especially
if one connects this notion with the idea – well-known from structuralist
semantics – that the meanings of words are delimited by the meanings of
other words. Imagine a two-dimensional model of the words of a language
which is similar to a jigsaw puzzle with larger and smaller pieces. The lar-
ger pieces are those words which take on more metaphorical and meto-
nymical Lesarten, and the smaller pieces are those words which have fewer
of these metaphorical extensions. The problem is that once one starts com-
paring two languages, the situation becomes much more complex, for when
the two jigsaw puzzles are put on top of each other, they do not match.
There are plenty of intersections and large pieces from one language corre-
spond to many parts of pieces from the other one. And of course the situa-
272 Brigitta Mittmann

tion becomes yet more complex because word meaning is not just two-
dimensional, but multi-dimensional.1
For this reason, it is not surprising that for practical purposes, compilers
of bilingual valency dictionaries have treated words as polysemous and
have also added further details about the selectional restrictions of the verbs
in question (cf. e.g. Bianco 1996). Another way to come to terms with this
complexity is to make bilingual valency dictionaries monodirectional, i.e.
to use a selection of items from one language as a starting point, describe
them in detail and give potential equivalents in the other language for these
usages, rather than accounting for both languages at the same time
(Schumacher 1995: 294).

2. Finding equivalents

Another problem is that of finding equivalents for the lexemes whose

valency is to be compared. Fischer states that “[c]ontrastive valency dic-
tionaries show syntactic and semantic parallels and contrasts in the envi-
ronment of words that regularly occur as translations of each other” (1997:
217) and it seems that he used the same principle in a part of his own work.
This is suitable for the aims that he pursues and it can show similarities and
differences between the behaviour of words which are likely to be pre-
sented as potential equivalents in textbooks or language lessons. However,
this does not necessarily mean that these are the equivalents that one should
aim for in a given foreign language text production or L1–L2 translation
situation, as most words can be translated in very different ways depending
on the respective textual context and on factors such as register.
For example, Fischer (1999: 244–245) gives a long list of German and
English verbs with prepositional complements together with equivalents
from the other language which take a direct object or dative/accusative
object respectively. This enables him to demonstrate potential differences
between structures in the two languages. However, in several cases the
proposed contrast is only true of the words chosen, but not of other poten-
tial equivalents. For example, Fischer contrasts German an etw. zweifeln
with English doubt sth. Unlike zweifeln, the verb doubt does not take any
prepositional complement. However, besides an etw. zweifeln German also
has etwas bezweifeln, which – like doubt – can take an object without a
preposition. It is not quite clear why Fischer discusses only zweifeln in this
context. He is not alone in this preference for the non-prefixed form (i.e.
zweifeln rather than bezweifeln). There seems to be a general tendency in
German valency dictionaries and the core vocabulary lists that they are
based upon to describe the prefixless verbs even though both are similar in
 Contrasting valency in English and German 273

meaning and there is hardly any significant difference in their frequency of

occurrence. In fact, the lexeme bezweifeln occurs slightly more frequently
in the large-scale written corpus (Korpus W – Archiv der geschriebenen
Sprache of the Cosmas II project).
Similarly, Fischer contrasts in etw. einsteigen with enter sth. However,
in etwas einsteigen might just as well be translated as get into (or even
onto, if it is a translation of in einen Bus einsteigen), go into, climb into,
etc., depending on the context and the register of the text. All of them are
more similar to in etw. einsteigen as regards their valency.
One potential solution for this problem might be frequency: if there are
verbs which can – broadly speaking – be considered synonymous, then the
one that occurs more frequently should be chosen. However, frequency is
also dependent on forms of the verb: while zweifeln and zweifelst are more
frequent in the Archiv der geschriebenen Sprache in Cosmas II than be-
zweifeln and bezweifelst, the situation is reversed for (be)zweifle,
(be)zweifelt, and (be)zweifelte. A second problem is that, ideally, what
should be considered is not the frequency of the lexeme as a whole, but that
of the lexical unit, i.e. the use of a particular meaning of a word. This, how-
ever, is a prohibitively time-consuming task.
Moreover, frequencies are also highly register-dependent. A few simple
online researches in the British National Corpus support Fischer’s choice in
that they show that for the BNC as a whole, the forms of enter occur more
frequently than those for get into and go into put together.2 However, if one
looks only at the frequencies for the conversations of the ‘spoken demo-
graphic’ part of the BNC only, one will find that the situation is reversed.3
Here, the verbs with prepositional phrases as complements occur more
frequently. And although the BNC as a whole contains mostly written Eng-
lish, most people arguably tend to use and be exposed to spoken English
much more frequently than to the written registers.
Curcio’s Kontrastives Valenzwörterbuch der gesprochenen Sprache
Italienisch-Deutsch respects register specificities to a certain extent. She
studies the 1000 most frequent verbs in spoken Italian (1999: 8) and arrives
at potential German equivalents through translating relevant sentences
taken from the corpus of the Lessico di frequenza dell’italiano parlato
(LIP), which she uses as material for her research. While this is a suitable
method for obtaining a variety of potential equivalents, one has to bear in
mind that these counterparts from the other language may show signs of
“translationese”. In other words, they may not be the words that would be
used in the same situations in natural spoken German. Moreover, if single
sentences (rather than whole texts) are translated, this may influence the
274 Brigitta Mittmann

theme-rheme arrangement of the sentences and the selection of the verbs

resulting from this.4
In Curcio’s case, it was necessary to use a translation corpus, as no suit-
able parallel corpus of spoken German existed. If, however, such a parallel
corpus is available, there is a more elegant solution to the problem. It will
then be possible to identify what Jürgens (following Klix) has called
Geschehenstypen (Krone 2003: 107), that is situations – reflected in sen-
tences – which differ in many respects, but have identical constellations of
semantic roles. This makes it possible to look for the tertium comparationis
at the semantic role level. The following pages describe such an alternative
approach. As it will become clear, this is very similar to the approach cho-
sen in FrameNet, as it compares many different words which evoke the
same frame.

3. Verbs in abstracts

When translating scholarly abstracts, translators may find that there is a

certain group of verbs which have very similar meanings in this context,
but for which bilingual dictionaries do not give enough reliable translations
to keep the translators from repeating the same words all over. Amongst the
verbs belonging to this group there are, for example, sich befassen mit and
sich einer Sache widmen, as used in the following examples:

(1) Nicola Würffel beschäftigt sich so mit der Methode des lauten
Denkens als einem Mittel, um die Effizienz von Lernsoftware zu
bestimmen.
‘In this context, Nicola Würffel discusses “think aloud protocols”
as a means of determining the efficiency of pedagogical soft-
ware.’
(2) Die vorliegende Arbeit befasst sich mit den spezifischen Proble-
men der Behandlung von Kollokationen im zweisprachigen Wör-
terbuch.
‘The present study deals with the specific problems connected
with the treatment of collocations in bilingual dictionaries.’
(3) Die Arbeit widmet sich also einem der markantesten Problembe-
reiche auf dem Gebiet von Theorie und Methodologie des zwei-
sprachigen Wörterbuchs.
‘Thus, the book deals with one of the most prominent problem
areas concerning the theory and methodology of bilingual diction-
aries.’
 Contrasting valency in English and German 275

(4) Claus Gnutzmann widmet sich mit „Englisch als globale lingua
franca“ dem Problem der Herausbildung und Bedeutung des
Englischen als Globalsprache und den sich daraus ergebenden
didaktischen Konsequenzen.
‘In his contribution on “English as a global lingua franca”, Claus
Gnutzmann deals with the problems of the development of Eng-
lish into, and its significance as a global language, as well as with
the consequences for teaching English which result from this.’

In abstracts, there can be very many verbs of this kind, like sich beschäfti-
gen mit, behandeln, sich widmen, eingehen auf, beschreiben, darstellen,
erörtern, untersuchen, and so on. These verbs can be grouped into various
subcategories. Verben in Feldern (1986: 587–597, 709–718, 601–605), for
example, mentions the following categories: (i) “Verben der geistigen Be-
schäftigung” [verbs of mental activity] like es zu tun haben mit, sich zu-
wenden, ansprechen, sich beschäftigen mit, sich befassen mit, sich konzen-
trieren auf, sich auseinandersetzen mit, eingehen auf, behandeln; (ii) “Ver-
ben des Diskutierens” [verbs of discussion] like diskutieren, debattieren,
erörtern, besprechen; and (iii) “Verben des Untersuchens” [verbs of analy-
sis] such as untersuchen, analysieren, erforschen. One could add a fourth
group of verbs of description like beschreiben or darstellen.
Despite these subcategories, it is interesting to note that in abstracts they
are all used to mean ‘deal with’, or perhaps even ‘write about (in some
detail)’. The text type determines a particular semantic interpretation.

4. General bilingual dictionaries

Traditional bilingual dictionaries do not offer much help when one is trans-
lating these words. For example, the fourth edition of the Collins
Großwörterbuch Englisch (1999) offers the following translations for sich
mit etwas befassen: first, the ubiquitous to deal with something, then to look
into something, to attend to something, and to work on something. With all
of the latter three, the reader may feel uncomfortable in this particular con-
text.

befassen
1 vr a (= sich beschäftigen) sich mit etw ~ to deal with sth; mit Problem,
Frage auch to look into sth; mit Fall, Angelegenheit auch to attend to sth; mit
Arbeit auch, Forschungsbereich etc to work on sth; (...)
276 Brigitta Mittmann

In example (2) mentioned earlier, den spezifischen Problemen der Behand-

lung von Kollokationen im zweisprachigen Wörterbuch is the topic that the
author writes about, so in a sense it can be interpreted as a Problem, a
Frage, an Angelegenheit or even a Forschungsbereich, but neither of the
verbs offered here seems quite appropriate. If one adopts the usual strategy
of looking up another German word that can be used synonymously in the
same context, this may be even less helpful. For sich widmen, for example,
the reader is faced once again with to attend to, and also with to devote
oneself to and to apply oneself to.5

widmen
2 vr +dat to devote oneself to; (= sich kümmern um) den Gästen etc to attend to;
einem Problem, einer Aufgabe to apply oneself to, to attend to; (...)

For a non-native user of English it will not become clear from this which of
these suggestions can be used in translating In Kapitel 6 beschäftigt sich
der Autor mit Syntax, In Kapitel 6 widmet sich der Autor der Syntax or even
Kapitel 6 beschäftigt sich mit der Syntax.
The obvious solution for a dyed-in-the-wool corpus-linguist is to build
her own corpus of abstracts. The corpus that was used here consists of eight
reviews of linguistics books written by native speakers of English as con-
tributions to the LinguistList. All of these reviews contain a section in
which there is an abstract of the books’ contents, and since many of the
books are collections of articles, the abstract sections are often rather long
and contain many of the verbs needed. For this reason, a very small corpus
serves the current purpose.

Table 1. Verbs in the Reviews Corpus

31 discuss
19 examine
10 explore, offer
9 demonstrate
8 describe
7 outline, present
5 cover, consider
4 provide, investigate
3 deal with
2 focus on, look at
1 analyse, chart, detail, pay attention to, pin-point, put forward, share,
tackle, trace
 Contrasting valency in English and German 277

The most frequent verb in these abstracts was discuss, followed by exam-
ine, explore, offer, and many others. Deal with is also among them, but it is
not one of the more frequent ones. These verbs were compared with those
found in a small parallel corpus of German reviews.6 The words which are
used in the English abstracts seem to be more verbs of investigating and
exploring than their German counterparts which in turn tend to be what are
called “Verben der geistigen Beschäftigung” [verbs of mental activity (di-
rected towards an object)] in Schumacher’s Verben in Feldern. And while
some of the verbs in the list, like present or put forward, can be said to
describe quite literally what the author is doing in the text, this does not
apply to others like discuss, examine, and explore. They only mean some-
thing like ‘produce text about’ in this context, though saying that they
“mean something like ‘produce text about’” is not very helpful. It is much
more appropriate to describe their similarity with the help of semantic
roles, or, for those who use a frame semantics approach, to say, that in
scholarly abstracts, these verbs evoke the same frame, with the following
frame elements, or roles: an [AUTHOR], a [TEXT], and a [TOPIC]. Not all of
these have to be present in the sentences, as one can see in other sentences
in the corpus where either the subject is not the [AUTHOR], but the [TEXT],
as in examples (2) and (3) above, or where the verbs mentioned above are
replaced by constructions like consists of, is concerned with, or has to do
with. There are also other German constructions expressing the same frame:
es geht um, eine Rolle spielen, im Mittelpunkt stehen.

5. Choosing a perspective – textual aspects

The choice of verb – and the choice of subject and voice – and thus also the
choice of semantic roles and their order in the sentence – determine the
perspective that is expressed in the sentence. It seems, however, that there
is also a textual component in the selection. If one looks at the reviews in
the corpus (and especially if one has previous experience of writing or
translating abstracts) one gains the impression that there is something like a
“sentence construction mechanism” underlying them. There are several
reviews in it which consist in part of a long series of mini-abstracts, like the
following:
278 Brigitta Mittmann

(5) [The first chapter (3–17), “American English: its origins and
history,”]text [by Richard W. Bailey,]author examines [the genesis of
American English varieties]topic through the lens of settlement
history. [Bailey]author demonstrates [that the American English
lexicon comes from a complex social situation, where Amerin-
dian, European, and African languages and peoples coexist-
ed]topic. [He]author also offers [a brief account of early nineteenth
century debates regarding the value of American English as a
marker of national identity]topic.
[In Chapter 2 (18–38), “American English and its distinctive-
ness,”]text [Edward Finegan]author addresses [the actual and per-
ceived differences between American and British English varie-
ties]topic. [Finegan]author examines [variations in American and
British pronunciations (represented with the International Pho-
netic Alphabet and pronunciation-based respellings), lexical
items, grammar, semantics, discourse, and orthography]topic.
[Chapter 3 (39–57), “Regional Dialects,”]text [by William A.
Kretzschmar, Jr.,]author points out [the problems with broad gen-
eralizations regarding regional speech]topic, yet acknowledges that
Americans are justified in thinking that persons from distinct ar-
eas speak English differently. [Kretzschmar]author presents [his-
torical origins of and linguistic examples from U.S. regional dia-
lects]topic using maps and tables, including an explanation of the
creation and use of these scholarly tools. ...
(Shuttlesworth, <LINGUIST List 16.843>, 17-MAR-2005)

Even such a short stretch – and the review in question contains many more
of these mini-abstracts – shows how many verbs there are in this text type
which have the meaning ‘to deal with’ in this context. There is a limited
number of sequences or role constellations which occur with the verbs in
question.7 Firstly, there is the sequence [AUTHOR]subj + VERB + [TOPIC], as
exemplified in (6). This sequence can be expanded by an introductory ad-
junct ([in TEXT]adju), as in (7).

(6) Finegan examines variations in American and British pronuncia-

tions.
(7) In Chapter 2 (18–38), “American English and its distinctiveness,”
Edward Finegan addresses the actual and perceived differences
between American and British English varieties.
 Contrasting valency in English and German 279

Alternatively, the [TEXT] can be the subject of the sentence, with an op-
tional postmodification naming the [AUTHOR]. This results in the structure
[TEXT]subj + ([by AUTHOR]postm) + VERB + [TOPIC]:

(8) Chapter 3 (39–57), “Regional Dialects,” by William A. Kretzsch-

mar, Jr., points out the problems with broad generalizations re-
garding regional speech.

Differences between the semantic roles become apparent in passivization:

the [AUTHOR] appears as a by-phrase in a passive sentence, whereas the
[TEXT] – if it does appear – is an adverbial of place. In many cases, how-
ever, neither are present:

(9) The shift to English by immigrants is also examined, as are the

classroom and untutored means immigrants use to learn English
and the social, economic, and personal barriers to their success.
(10) In Chapter 2 sociological approaches to the relationship between
language and society are reviewed.

In German, the [TEXT] can also appear after the preposition mit, as in ex-
ample (11).

(11) [Claus Gnutzmann]subj widmet sich [mit „Englisch als globale

lingua franca“] [dem Problem der Herausbildung und Bedeu-
tung des Englischen als Globalsprache und den sich daraus
ergebenden didaktischen Konsequenzen].
‘In his contribution on „English as a global lingua franca“, Claus
Gnutzmann deals with the problems of the development of Eng-
lish into, and its significance as a global language, as well as
with the consequences for teaching English which result from
this.’

According to the analysis of the Valency Dictionary of English – which will

be followed here – these adverbials are of course not complements of the
verb, but adjuncts. However, it is remarkable – and a computer which
might one day translate or produce sentences like these would have to
know this – that both the [TEXT] and the [AUTHOR] can be subject of the
sentence, but only the [TEXT] can also be part of a prepositional phrase
which in English usually functions as an adverbial of (more or less meta-
phorical) place, whereas the [AUTHOR] can appear in the by-phrase in pas-
sive sentences.
280 Brigitta Mittmann

Psycholinguistically, it would be interesting to know if when writing

texts like these abstracts, we first select the constellation of semantic roles
for the sentence or the verb itself. The latter seems to be more in line with
the ideas of most valency theorists, but some findings from analysing the
spoken language – such as the verbless constructions in football commen-
taries mentioned by Krone (2003: 123) – would seem to contradict this.
There is, of course, always the possibility of some kind of “co-selection”
going on.
The selection of roles and the question of choosing a sequence seems to
follow certain textual principles as well: firstly, as in example (5), there can
be a tendency to impose a structure on the text and increase the cohesion of
the text by starting paragraphs with the same semantic role sequence in the
first sentence. Here, [TEXT] is always at the beginning of the sentence,
twice as subject, once in the adverbial. On the other hand, with so many
sentences in which the same role constellation is mentioned over and over
again, there is of course also the principle of varying sentence structure and
of avoiding too many similar and thus boring sentences. (This does not
show up so much in this section, but it becomes quite clear in other re-
views.)

6. Cross-linguistic differences in perspective

However, while these textual aspects are interesting from a text production
point of view, they are not relevant for bilingual lexicography. What is
more interesting in this context are the lexical aspects of sequencing roles
and some other properties of the verbs studied here. There must be many
cases like the following where the perspective (in active clauses at least)
can be lexicalised in different ways in different languages:

[THEMA] steht im Mittelpunkt [eines TEXTES]

[AUTHOR] focuses on [TOPIC]

If we compare German im Mittelpunkt stehen with English focus on, we see

that in German one makes a statement about something being located
somewhere, whereas in English we talk about the activity of a person who
is not normally mentioned in the equivalent German sentences. Cases like
these can be very interesting, even though some researchers deliberately
exclude them from their research because of their difference in structure.
Here, once again, meaning-based approaches like FrameNet can be very
 Contrasting valency in English and German 281

useful, as they can compare many different words which evoke the same
frame.
And indeed, if we look at the FrameNet homepage, we can find a frame
which corresponds exactly to the set of verbs and other constructions which
are at the heart of this paper:8 if we look up the verb discuss in the lexical
index (we will recall that it was the most frequent one of the set to be found
in the corpus), we arrive at the frame labelled TOPIC which has the so-
called “core frame elements” [COMMUNICATOR], [TEXT] and [TOPIC].
There are also some “marginal frame elements”, namely [DEGREE], [MAN-
NER], and [STATUS], which will not concern us here. And, as was pointed
out earlier, although [COMMUNICATOR], [TEXT] and [TOPIC] are labelled as
“core frame elements”, they are not necessarily realised as complements of
the verb in actual sentences, but may either be left out or be realised as
adjuncts. The following examples are given for the frame TOPIC:

(12) Chapter 5 discusses the issue of transubstantiation.

(13) Smither’s essay is about plane spotting.
(14) Ostrovsky addresses monetary policy in Chapter 5.
(15) This book is mostly about particle physics.

The following lexical units are listed at the bottom of the page describing
the frame: the prepositions about, on, concerning and regarding, the verbs
address, concern, cover, discuss, dwell_(on),9 and treat, and the nouns sub-
ject, theme, and topic. As the research on the mini-corpus has shown, this
list could be extended considerably. At present, of course, because the
FrameNet project is still incomplete, there do not seem to be entries for
many lexical units as yet. One that is missing, for example, is that for ex-
plore.

7. The new valency dictionaries

While the FrameNet database is, in theory, infinite and can accommodate
many nuances of meaning, this is not the case with printed dictionaries.
Nevertheless, both the VALBU and the VDE are much better at describing
the ‘deal with’ set of verbs than the Collins Großwörterbuch – if, that is,
they include these verbs in the macrostructure at all. Because of the fact
that both dictionaries had to concentrate on a limited number of words,
some had to be left out. Thus, for example, there is no entry for explore in
the Valency Dictionary of English. However, for example, the VDE expli-
citly describes the ‘write about’ or ‘deal with’ use of a number of verbs and
also mentions the fact that cover, deal with, describe, discuss, and treat can
282 Brigitta Mittmann

also be used with a [TEXT] functioning as grammatical subject. Neverthe-

less, there is also some, albeit limited, scope for improvement: for examine
and provide, there is only an example in which the subject is a [TEXT] –
and no mention in the note block – and for look at in the meaning ‘deal
with’ there is not even an example. Even though the relationship between
author and text is a well-known source of metonymies, it is not self-evident
for non-native speakers that these verbs can be used with non-human sub-
jects.
VALBU is very systematic in its treatment of these verbs. For instance,
the “Belegungsregeln” [semantic scope] for the Nominativergänzung for
untersuchen specify that it can be “derjenige, der etwas analysiert: Per-
son/Institution/[auch geistiges Produkt]” [someone analysing something:
person/institution/also “intellectual product”] and an example mentions an
academic paper (2004: 761).
Both the VDE and the VALBU would treat cases like In Kapitel 7 be-
fasst sich der Autor mit Valenzgrammatik and Kapitel 7 befasst sich mit
Valenzgrammatik as the same meaning of the verb sich befassen (see ex-
ample [2]). No new lexical unit is established. This is, of course, very sen-
sible from a practical point of view – if one looks through just a few pages
of either dictionary, one will come across a multitude of verbs with differ-
ent lexical classes of subjects.

8. Large corpora reveal importance of morphological forms

This micro-study of the TOPIC frame shows how useful small specialised
corpora can be. However, in order to find out about the distribution of these
uses of the verbs in other text types, it is necessary to look at larger and
more balanced corpora like the British National Corpus. In a random selec-
tion of 50 (out of a total of 1091) sentences for looks at (please note the
inflectional form) almost half of the examples had the frame element con-
stellation ([AUTHOR]) + ([TEXT/INTELLECTUAL PRODUCT]) + [TOPIC]. Two
thirds of them had [TEXT] as subject. For explores, the figures are even
higher: three quarters of the examples had the frame element constellation
discussed here. Once again, two thirds of these had [TEXT/INTELLECTUAL
PRODUCT] as subject.
This means that if one encounters the third person singular present tense
of these verbs, it is very likely that one will also encounter this particular
meaning and semantic role constellation, and also this particular choice of
subject. Thus, this morphological form is linked closely to this frame and
use of the verb, even though if one asked a native speaker to produce sen-
tences with looks at and explores, these are likely to be more like the sen-
 Contrasting valency in English and German 283

tences in pre-corpus editions of learners’ dictionaries. Native speakers of

German who were asked to produce sentences with behandeln, sich befas-
sen, sich widmen and untersuchen gave, amongst others, Der Arzt behan-
delt die Magenverstimmung des Patienten, Die Rechtsanwältin befasst sich
mit der Akte des Delinquenten, Der Großvater widmet sich seinem Enkel,
Der Großvater widmet sich der Musik, and Der Geografieprofessor unter-
sucht die vulkanischen Erscheinungen der Rhön. The only one that came
close and fitted the TOPIC frame was Der Professor behandelt das Thema
in der Vorlesung.

9. Summary

Thus, to sum up, the perfect bilingual reference work on valency would be
a comprehensive database in which the tertium comparationis lies at the
frame or semantic role level and which includes all verbs, nouns and adjec-
tives of the languages in question (or at least those that occur reasonably
frequently in a large corpus). This database should also take into account
certain other constructions like im Mittelpunkt stehen (or, in fact, take into
account) and should give information about the frequencies of synonymous
patterns, perhaps even in different text types. So one must hope that it will
be possible to extend the project that Boas (2001; 2002) has described in
his articles on creating a bilingual version of FrameNet. It would certainly
be very useful for the material described here.

Notes

1. Other authors have also drawn attention to this, e.g. Schumacher (1995: 294).
2. At http://sara.natcorp.ox.ac.uk/lookup.html (August 5, 2005). These were sim-
ple formal queries disregarding the meanings that these lexemes can take on.
This is crude but can still give an indication of general tendencies.
3. The spoken demographic part of the BNC contains around four million words
of spontaneous conversation.
4. In another part of her book, Curcio (1999: 149) remarks that in 98,6% of sen-
tences the Italian subject corresponded to the German subject in the transla-
tions. It is quite possible that this figure was influenced by the translation
process.
5. The writers of this dictionary seem to assume that the user will understand that
vr+dat refers to constructions like sich einer Sache widmen. In fact, it is likely
that even well-trained dictionary users will overlook this code.
6. This corpus contains one abstract and seven reviews. For details see bibliogra-
phy.
284 Brigitta Mittmann

7. Note that with this group of verbs, the [TEXT] cannot be the object of the ac-
tive clause (or the subject of the corresponding passive clause). Verbs like
write belong to a different group of verbs, even though they may have the
same role constellation in examples like J.K. Rowling has written a new book
about Harry Potter (invented example) or A new book has been written about
Harry Potter (invented example) where the role constellations are [AU-
THOR]subj + VERB + [TEXT] + ([on/about TOPIC]) and [TEXT]subj VERBPASS
([on/about TOPIC]) respectively.
8. http://framenet.icsi.berkeley.edu/ (March 25, 2005).
9. In contrast to the other verbs mentioned so far, dwell (on) is of course used to
indicate that the [AUTHOR] spends too much time dealing with the [TOPIC].

References

Bianco, Maria Teresa

1996 Valenzlexikon Deutsch-Italienisch. Dizionario della valenza verbale.
(Deutsch im Kontrast 17.) Heidelberg: Julius Groos.
Boas, Hans C.
2001 Frame semantics as a framework for describing polysemy and syn-
tactic structures of English and German motion verbs in contrastive
computational lexicography. In Proceedings of the Corpus Linguis-
tics 2001 conference, Paul Rayson, Andrew Wilson, Tony McEnery,
Andrew Hardie, and Shereen Khoja (eds.), 64–73. (Technical Pa-
pers, 13.) Lancaster: University Centre for Computer Corpus Re-
search on Language.
2002 Bilingual FrameNet dictionaries for machine translation. In Proceed-
ings of the Third International Conference on Language Resources
and Evaluation. Vol. 4. Manuel González Rodriguez, and Carmen
Paz Suárez Araujo (eds.), 1364–1371. Las Palmas, Spain.
Curcio, Martina Lucia
1999 Kontrastives Valenzwörterbuch der gesprochenen Sprache Italie-
nisch-Deutsch. Grundlagen und Auswertung. (amades 3.) Mann-
heim: IDS.
Fischer, Klaus
1997 German-English Verb Valency: A Contrastive Analysis. (Tübinger
Beiträge zu Linguistik 422.) Tübingen: Narr.
1999 Englische und deutsche Satzstrukturen: Ein valenztheoretischer Ver-
gleich mit statistischen Anmerkungen. Sprachwissenschaft 24 (2):
221–255.
Herbst, Thomas, David Heath, Ian F. Roe, and Dieter Götz
2004 A Valency Dictionary of English. A Corpus-Based Analysis of the
Complementation Patterns of English Verbs, Nouns and Adjectives.
Berlin/New York: Mouton de Gruyter. [= VDE]
 Contrasting valency in English and German 285

Krone, Maike
2003 Valenzstrukturen in deutschen und englischen Fußballreportagen am
Beispiel von Freistößen. In Valency in Practice. Valenz in der
Praxis, Alan Cornell, Klaus Fischer, and Ian F. Roe (eds.), 105–126.
Oxford/Bern/Berlin/Bruxelles/Frankfurt M./New York/Wien: Peter
Lang.
Schumacher, Helmut
1986 Verben in Feldern. Valenzwörterbuch zur Syntax und Semantik deut-
scher Verben. Berlin/New York: Walter de Gruyter.
1995 Kontrastive Valenzlexikographie. In Deutsch als Fremdsprache. An
den Quellen eines Faches. Festschrift für Gerhard Helbig zum 65.
Geburtstag, Heidrun Popp (ed.), 287–315. München: Iudicium.
Schumacher, Helmut, Jacqueline Kubczak, Renate Schmidt, and Vera de Ruiter
(eds.)
2004 VALBU – Valenzwörterbuch deutscher Verben. Tübingen: Narr. [=
VALBU]
Terrell, Peter, Veronika Schnorr, Wendy V.A. Morris, and Roland Breitsprecher
1999 Collins Großwörterbuch Deutsch-Englisch, Englisch-Deutsch. 4th
ed. Glasgow: Collins.

Corpora

The British National Corpus is a collaborative inititative carried out by Oxford

University Press, Longman, Chambers Harrap, Oxford University Computing
Services, Lancaster University’s Unit for Computer Research in the English
Language, and the British Library. The project received funding from the UK
Department of Trade and Industry and the Science and Engineering Research
Council and was supported by additional research grants from the British Acad-
emy and the British Library. For more details see http://info.ox.ac.uk/bnc/.
Cosmas-II-Projekt: Korpus W – Archiv der geschriebenen Sprache. For more de-
tails see http://www.ids-mannheim.de/cosmas2/ (August 2, 2005).
LINGUIST List issues 16.24 (Callahan), 16.163 (Willoughby), 16.314 (Unsworth),
16.467 (Hewitt), 16.577 (Maxwell), 16.601 (Wilcox), 16.757 (Henderson),
16.843 (Shuttlesworth).
See http://www.linguistlist.org/pubs/reviews/index.html (March 26, 2005).
Mini-Corpus of German Reviews and Abstracts. Contents: Jekaterina Boutina-
Koller’s abstract of her PhD thesis on Kollokationen im zweisprachigen
Wörterbuch plus reviews by Claus Altmayer (jg-06-2), Dieter Kranz (jg-06-2),
Haymo Mitschian (jg-09-1) and Guido Oebel (jg-09-3) from Zeitschrift für In-
terkulturellen Fremdsprachenunterricht (http://zif.spz.tu-darmstadt.de/earchiv.htm;
downloaded on April 3, 2005) as well as reviews from Philologie im Netz
(http://www.fu-berlin.de/phin/welcome.html, downloaded on April 3, 2005) by
Christiane Maaß (PhiN 29/2004: 79), Susanne Mühleisen (PhiN 9/1999: 38)
and Richard Waltereit (PhiN 4/1998: 53).
Valency in a contrastive perspective: Structure and
use

Stig Johansson

1. Introduction

If we want to study verbs in a contrastive perspective, we can compare

groups such as modal auxiliaries or mental process verbs. But how do we
know what forms to contrast? We know that modality is expressed by other
means than by modal auxiliaries, so a comparison of modal auxiliaries is
clearly insufficient. Mental processes in English are often expressed by
constructions containing the noun mind, and these go far beyond idiomatic
forms like keep in mind and make up one’s mind (cf. Johansson 1998: 16–
18). Compare:

(1) in my mind it was as if jeg følte det som

[‘I felt as if’]
my mind fills with jeg tenker på
[‘I think of’]

(2) sånn jeg tenker the way my mind works

[‘the way I think’]
jeg tenker på alt annet my mind fills with quite
[‘I think of everything else’] other things

In (1) we find constructions with mind translated by Norwegian mental

process verbs, in (2) Norwegian mental process verbs are rendered by con-
structions with mind in the English translation.
To identify what forms may be associated across languages, it is useful
to turn to multilingual corpora. I will give some examples illustrating how
such corpora can be used in a cross-linguistic study of verbs. My main
claim is that this sort of comparison makes it possible to contrast both
structure and use.
288 Stig Johansson

2. Material and method

The corpora I will refer to are the English-Norwegian Parallel Corpus

(ENPC), the source of the examples above, and the Oslo Multilingual Cor-
pus (OMC). The ENPC is a bidirectional translation corpus containing
English and Norwegian original texts, both fiction and non-fiction, and
their translations into the other language. The different parts are balanced to
facilitate a comparison across languages. With this kind of model we can
see both what forms are associated by translators, and we can control for
translation effects (see Johansson 1998). The OMC is an umbrella term for
a group of corpora developed at the University of Oslo, including a corpus
of English, German, and Norwegian texts.1
When we compare verbs in a multilingual corpus, it is striking how
great differences we may find even with cognates (or prototypical equiva-
lents) in closely related languages. For example, Åke Viberg (1996; 2002)
has shown that the cognate verb pairs go/gå and give/ge in English and
Swedish correspond to each other only in about a third of the cases. I will
now go on to examine correspondences for two verbs: English spend in
expressions of time and a special construction containing the Norwegian
verb hende [‘happen’]. By correspondences I mean the forms that are asso-
ciated by translators.

3. The verb spend in expressions of time

The verb spend in expressions of time is an example of the “time is money”

metaphor (cf. Lakoff and Johnson 1981: 7−9). Cross-linguistically, it has
interesting correspondence patterns. Figure 1 summarizes the overall distri-
bution in the fiction texts of the ENPC of spend and its most frequent cor-
respondence in Norwegian, tilbringe [‘pass time’; cf. German verbrin-
gen/zubringen]. We see that there is a wide difference in frequency in the
original texts, suggesting that there are major differences in use between
the English and the Norwegian verb, though the differences are evened out
in translations. Tilbringe is much more common in texts translated from
English than in Norwegian original texts; the same pattern has been found
for Swedish tillbringa in a study by Gellerstam (1996: 59). For spend in
expressions of time, we find the opposite effect. In other words, the ways of
referring to passing/spending time appear to be influenced by the source
language.
 Valency in a contrastive perspective: Structure and use 289

140
120
100
80 Orig
60 Trans
40
20
0
spend tilbringe

Figure 1. The overall distribution of English spend and Norwegian tilbringe in the
fiction texts of the ENPC (30 texts of each type).

What correspondences do we find for English spend? At the time of the

study I had 16 English original texts available with translations into Nor-
wegian and German, and I decided to focus on this material. The material
was not very large, but sufficient to show some very clear tendencies.
There are congruent correspondences, which preserve the syntax of the
English original, as well as cases of restructuring (see the overview of
complementation patterns in table 1 and the summary of correspondences
in table 2.) As shown in table 1, spend takes a temporal complement in the
form of a noun phrase, typically followed by an adverbial and/or a verb in
the -ing form.

Table 1. The distribution of complementation patterns of spend in 16 English fic-

tion texts

spend + NPtemp 2
spend + NPtemp + ADVplace 21
ADVaccomp 6
ADVmanner 4
ADVplace+accomp 2
ADVplace+manner 1
ADVplace+ V-ing 2
spend + NPtemp + V-ing 28
Total 66
290 Stig Johansson

Table 2. Correspondence patterns for spend in German and Norwegian translations

of 16 English fiction texts

German Norwegian

spend + NPtemp bleiben bli over

[‘stay over’]
sein gjennomgå
[‘go through’]

spend + NPtemp + ADVplace verbringen (13) tilbringe (11)

intr/refl verb (7) intr verb (7)
other (1) other (3)
ADVaccomp verbringen (4) tilbringe (2)
intr verb (1) intr verb (2)
other (1) other (2)
ADVmanner verbringen (2) tilbringe (1)
zubringen (1)
nutzen (1) bruke (3)
verwenden (1)
intr verb (1) intr verb (2)
ADVplace+accomp verbringen (1) tilbringe (1)
pass verb (1) pass verb (1)
ADVplace+manner intr verb (1) tilbringe (1)
ADVplace + V-ing verbringen (1)
intr verb (1) intr verb (2)

spend + NPtemp + V-ing verbringen (13) tilbringe (9)

zubringen (2)
verwenden (2) bruke (9)
V + ADV (10) V + ADV (8)
intr verb (1)
other (1) other (1)

Total 66 66

3.1. Congruent correspondences

Congruent correspondences with tilbringe or verbringen/zubringen are

found with most of the patterns in table 2. These are the “standard” trans-
lations which are typically listed first in bilingual dictionaries for spend in
expressions of time. Examples:2
 Valency in a contrastive perspective: Structure and use 291

(3) He liked Sir Bernard Hemmings, but it was an open secret inside
“Five” that the old man was ill and spending less and less time in
the office. (FF1)
Er mochte Sir Bernard Hemmings, aber es war in ”Fünf” ein of-
fenes Geheimnis, daß der alte Mann krank war und immer weni-
ger Zeit im Büro verbrachte.
Han likte Sir Bernhard Hemmings, men det var en åpen hemme-
lighet i ”Fem” at den gamle mann var syk og tilbrakte mindre og
mindre tid på kontoret.

(4) I spent most of the time sobbing in the protecting darkness of the
great cathedral, only half conscious of the endless stream of tour-
ists shuffling past. (ABR1)
Die meiste Zeit verbrachte ich damit, im schützenden Dunkel der
großen Kathedrale zu schluchzen, wobei ich mir des endlosen
Stroms der vorbeischlürfenden [sic] Touristen nur halb bewusst
war.
Jeg tilbrakte det meste av tiden med å hulke i det beskyttende
mørket i den store katedralen, bare halvt oppmerksom på den
endeløse strømmen av turister som subbet forbi.

Other congruent correspondences have the verbs nutzen or verwenden in

German and the verb bruke [‘use’] in Norwegian, as in:

(5) Look Brian, I’ve spent two years on that investigation. (FF1)
Hören Sie, Brian, ich habe zwei Jahre auf diese Nachforschun-
gen verwendet.
Hør nå, Brian. Jeg har brukt to år på denne etterforskingen.

(6) I actually spend time thinking about this. (MA1)

Ich verwende tatsächlich Zeit darauf, über diese Frage nachzu-
denken.
Jeg bruker faktisk tid på å tenke ut dette.

As shown in table 2, this type was only recorded with manner adverbials
and -ing complements.
292 Stig Johansson

3.2. Restructuring

In spite of the translation effect established in the ENPC, there is a lot of

restructuring. Most typically, we find intransitive or reflexive verbs: sich
aufhalten [lit. ‘keep oneself’], bleiben [‘stay’], verweilen [‘stay’]; bli
[‘stay’], bo [‘live, stay’], oppholde seg [lit. ‘keep oneself’], sitte [‘sit’],
være [‘be’]. Some examples of restructuring are:

(7) She informed us that she planned to spend that night, then go to
church with us, and be back in Des Moines by suppertime. (JSM1)
Sie teilte uns mit, daß sie vorhatte, die Nacht zu bleiben, dann mit
uns in die Kirche zu gehen und zum Abendessen wieder zurück in
Des Moines zu sein.
Hun kunngjorde at hun aktet å bli over [‘stay’] en natt, gå i kirken
med oss neste morgen, og være tilbake i Des Moines til kvelds.

(8) I might even delve deeper into natural history and say, “The peri-
odical cicada spends six years as a grub underground, and no
more than six days as a free creature of sunlight and air.” (RD1)
Kann sein, daß ich mich sogar noch eingehender mit der Natur-
geschichte befassen und sagen würde: “Die sich häutende Zikade
bleibt im Puppenzustand sechs Jahre lang im Verborgenen und
verbringt nicht mehr als sechs Tage als freies Insekt in Licht und
Luft.”
Jeg kunne trukket fram andre ting fra zoologien også: “Sikadens
livssyklus er slik at den lever [‘lives’] seks år som larve under
jorda, men bare seks dager som et fritt vesen i sola og lufta.”

(9) “But I spent the night at Rose’s.” (JSM1)

“Aber ich hab heut nacht bei Rose geschlafen.”
“Men jeg har jo ligget over hos Rose.” [lit. ‘lie over’]

(10) Since the age of eighteen, he’d spent an accumulated nine years
in jail. (SG1)
Seit seinem achtzehnten Lebensjahr hatte er alles in allem neun
Jahre im Gefängnis verbracht.
Siden attenårsalderen hadde han sittet inne i tilsammen ni år.
[lit. ‘sit inside’]

Example (7) is one of the rare cases where spend has no further comple-
mentation apart from the temporal NP. Both the Norwegian and the Ger-
 Valency in a contrastive perspective: Structure and use 293

man translators have opted for intransitive verbs, and the same applies to
(8), where the original has an adverbial of manner and a place adverbial (in
addition to the temporal NP). In (9) the German translation has the intran-
sitive verb for ‘sleeping’ (if you sleep in a place, you are there), while (10)
has a congruent translation with verbringen. In both of these cases, the
Norwegian translator has chosen a lexicalised expression for ‘staying the
night’ and ‘being in prison’.
The most interesting pattern is found where there is an -ing complement
in the English original. In these cases there is often no verb at all cor-
responding to spend, and its place is taken by the complementing verb,
which is so to speak “raised” to the superordinate clause, as in:

(11) After leaving school at sixteen, Rawlings had spent ten years
working with and under his Uncle Albert in the latter’s hard-
ware shop. (FF1)
Nach seinem Schulabgang im Alter von sechzehn hatte Rawlings
zehn Jahre in der Eisenwarenhandlung seines Onkels Albert
gearbeitet.
Rawlings hadde sluttet på skolen da han var seksten år og siden
arbeidet i ti år sammen med og under sin onkel Albert som drev
jernvarehandel.

(12) We spent a lot of the time driving, in our low-slung, boat-sized

…. (MA1)
Die meiste Zeit fuhren wir in unserem niedrigen, bootsförmigen
Studebaker herum … .
Mye av tiden kjørte vi bil, en lav Studebaker, … .

(13) Nights on end she spends flying, beyond the reach of all that
threatens her by day. (ABR1)
Ganze Nächte hindurch fliegt sie dahin, unerreichbar für alles,
das sie tagsüber bedroht.
Natt etter natt flyr hun, utenfor rekkevidde av alt det som truer
henne om dagen.

(14) He spent pleasurable hours dithering over questions of punctua-

tion. (AT1)
Er grübelte vergnügliche Stunden lang über Interpunktionspro-
bleme nach.
Han tilbrakte koselige timer med å gruble over tegnsettingen.
294 Stig Johansson

In (11) to (13) the German and Norwegian translators have opted for “rais-
ing”. The same type of restructuring is found in the German translation of
(14), while the Norwegian translator has relied on the “standard” transla-
tion, a form of tilbringe.
Apart from the correspondence types I have commented on above, there
are other more sporadic renderings; for a more detailed account, see Jo-
hansson (2002).

3.3. Summing up: Spend in a contrastive perspective

We clearly have structures in German and Norwegian which correspond

closely to the English original, but there are also many cases of restructur-
ing. Tilbringe and verbringen/zubringen are possible choices in translating
spend. To what extent, and in what contexts, are congruent structures over-
used as compared with their distribution in original texts in the target lan-
guage? As the English-German-Norwegian subcorpus has not been devel-
oped sufficiently, I will restrict my remarks to the ENPC; see table 3.

Table 3. The distribution of complementation patterns of Norwegian tilbringe

[‘pass time’] in original and translated fiction texts of the ENPC (30 texts
of each type)

Original Translation

tilbringe + NPtemp + ADVplace 19 26

ADVaccomp 2 6
ADVmanner 0 9
ADVplace + med [‘with’] + V-inf 0 1

tilbringe + NPtemp + med/til [‘with, to’] + V-inf 1 11

Total 22 53

We see that tilbringe is more than twice as common overall in translated

texts as in original Norwegian texts. Moreover, the overuse is found par-
ticularly with adverbials of manner and with infinitive complements (trans-
lating spend + V-ing). These types were found only once in original texts,
as compared with twenty examples in texts translated from English. For
translators, these results should be taken as an indication that they should
look for more creative renderings than the “standard” translation with til-
bringe. Such renderings are amply illustrated in the corpus material. I will
 Valency in a contrastive perspective: Structure and use 295

come back later to what conclusions we can draw from this example as
regards valency in a contrastive perspective.

4. The Norwegian det + hende construction

Formally, the det + hende construction consists of the dummy subject det
[‘it’], a form of the verb hende [‘happen’], and a complement clause intro-
duced by at [‘that’], though the conjunction is often omitted. This clause is
placed at the end and cannot be fronted:

(15) Det hender at [‘it happens that’] Elsa går på en utstilling, hvis
Håkon kan være hjemme og se til barna. (BV1)
Occasionally Elsa goes to an art exhibition if Håkon can stay at
home and look after the children.
Cf. *At Elsa går på en utstilling hender [‘That Elsa goes to an art
exhibition happens’].

The construction can be expanded by modal auxiliaries and adverbials, as

will be shown below, but generally it is unexpanded. The correspondences
show very clearly that the bare det + hende construction is an expression of
usuality, following the analysis of Halliday (2004), who regards usuality as
a type of modality.

4.1. The bare det + hende construction

Most commonly, the English correspondence is a frequency adverbial, as in

(15) above. The forms found most often in the material are sometimes and
occasionally. Judging by these forms, the det + hende construction typi-
cally denotes a low degree of usuality. These are some more examples:

(16) Det har hendt at [‘it has happened that’] du har sett på meg med
akkurat det blikket. (OEL1)
Sometimes you’ve looked at me in exactly that way.

(17) Det hendte [‘it happened that’] han kom helt ut på kaia før han
husket hodeplagget. (HW1)
Sometimes he got all the way out on the wharf before he remem-
bered his headgear.
296 Stig Johansson

(18) Det hender vi slåss. (MA1T)

Once in a while we fight.

In addition to frequency adverbials on their own, we find combinations of

frequency adverbials and modal auxiliaries denoting habit or possibility
(would, used to, can, could, may, might), as in:

(19) Det hendte at gamle venner kom innom, tykke menn i flekkede
tweeddresser, kvinner med usminkede ansikter. (AB1T)
Occasionally friends from the old days would drop in, fat men in
stained tweed suits, women with unadorned faces.

(20) Det hender at man ser [lit. ‘it happens that one sees’] dådyr mel-
lom trærne. (RR1T)
Sometimes fallow deer can be seen among the trees.

Sometimes the det + hende construction corresponds to a modal auxiliary

on its own, as in:

(21) Når Sofies mor var sur for et eller annet, hendte det at hun kalte
huset de bodde i for et menasjeri. (JG1)
Whenever Sophie’s mother was in a bad mood, she would call the
house they lived in a menagerie.

(22) Når kvinner oppdrar gutter aleine, hender det at de kommer inn i
den voksne verdenen ansiktsløse. (ROB1T)
When women, even women with the best intentions, bring up a boy
alone, he may in some way have no male face, or he may have no
face at all.

The modals represented are would, may, and could, i.e. the same types of
modals as were found in combinations with frequency adverbials.

4.2. Adverbial expansions

In all the cases we have seen so far, the hende-clause has disappeared in the
translation. This correspondence type is also found where there is an adver-
bial expansion in the Norwegian material, as in:
 Valency in a contrastive perspective: Structure and use 297

(23) Men det hendte aldri at [lit. ‘it happened never that’] jeg hilste
først. (EHA1)
But I never greeted them first.

(24) Det hendte bare en eneste gang at [lit. ‘it happened only a single
time that’] hun ikke kunne leksa det året. (PEJ1)
Only once, that whole year, did she not know her lesson.

In (23) there is a frequency adverbial denoting that something does not

occur, in (24) we find an adverbial referring to a single time in the past.
There are some examples where an expanded det + hende construction
corresponds to a matrix clause with happen. Apart from a single example,
these are found in translations from Norwegian. Examples:

(25) “Det har hendt før at en kunstner forblir ukjent,” sa han så, “men
aldri for å dukke opp igjen som et geni på linje med de aller stør-
ste.” (JW1)
“It has happened before that an artist has remained unknown,”
he went on, “but never before to emerge again as a genius on a
par with the very greatest.”

(26) Det hendte i Tuv som andre steder at ungdommen fant seg
kjærester når det var sommer og midnattssol og lyse netter. (PEJ1)
It happened in Tuv, as it did in other places, that young people
fell in love in summer, when the days were longest and the nights
were bright with the midnight sun.

(27) Det hendte han satte seg på kjøkkenet sammen med pikene, stjal
seg til en kopp nypete og fortalte bløte vitser som alle hadde hørt
før. (BV1)
It sometimes happened that he sat down in the kitchen with the
girls, helped himself to a cup of rose-hip tea and told silly jokes
which they had both heard before.

(28) Det hendte at Maria forsøkte seg på en sigarett, den gamle pianis-
ten blunket til dem og spilte revyviser. (BV1)
It might happen that Maria would try a cigarette, and the old
piano-player would wink at them and play tunes from the old mu-
sicals.
298 Stig Johansson

Note that, in all these cases, the English happen clause has an expansion: in
(25) a time adverbial, in (26) a place adverbial, in (27) a frequency adver-
bial, and in (28) a modal expansion.

4.3. Modal expansions

Det + hende constructions can also be expanded by modal auxiliaries, usu-

ally by the present-tense form kan [‘can’], less often the past-tense form
kunne [‘could’]. These constructions express possibility, as in:

(29) Det er ikke noe farlig, men det kan hende du mister [lit. ‘it can
happen you lose’] litt av håret ditt, Herman. (LSC1)
It isn’t anything serious, but you might lose a little of your hair,
Herman.

(30) Etter et par timer kunne det hende [lit. ‘could it happen’] at en og
annen fant ut at han skulle handle litt. (HW1)
After a few hours of that, one of them might even remember that
he was supposed to do some shopping.

(31) Nåja, hvis du må reise, kan det hende jeg blir. (PDJ3T)
Well, if you have to go, maybe I’ll stay on.

(32) Kanskje var jeg [lit. ‘perhaps was I’] ganske enkelt lei av å komme
og gå. Det er forferdelig alltid å være i overgangen. Det kan også
hende at [lit. ’it can also happen that’] jeg ville smake på denne
verden … . (BO1T)
It may simply have been that I had grown tired of coming and go-
ing. It is terrible to forever remain in-between. It may also have
been that I wanted to taste of this world … .

(33) Det kan hende at Robert M. Turner hadde gitt kelneren inntrykk
av at … . (FC1)
It could be that Robert Turner had given the waiter the impression
….
 Valency in a contrastive perspective: Structure and use 299

(34) Kan hende det intime vennskapet mellom Scott og Wilson, blir den
tilleggsbelastning på det psykiske plan som knekker Shackleton.
(KH1)
Perhaps the intimate friendship between Scott and Wilson became
the last mental straw which broke Shackleton’s back.

The English correspondences vary. Most often, there is no matrix clause in

English, and the meaning is conveyed by a possibility modal (may, might,
could), as in (29) and (30), less commonly by a modal adverb, as in (31).
Where the English correspondence has a matrix clause in the material, the
verb is be rather than happen, as in (32) and (33), testifying to the weaken-
ing of the meaning of the Norwegian verb. A further development is shown
in (34) where the fixed sequence kan hende [‘can happen’] is an adverbial
corresponding to English perhaps. The same development is found with the
more common kanskje [lit. ‘can happen’], shown in the opening of (32).
Although it + happen constructions expanded by a possibility modal
were not found as correspondences of similar Norwegian constructions,
they are by no means excluded in English, as we shall see later (section 5).
They seem, however, to be a less common option than in Norwegian.

4.4. Summing up: The det + hende construction

From the study we can draw the following conclusions (for a more detailed
account, see Johansson 2005). There is generally no matrix clause in Eng-
lish. The correspondences clearly show that the bare det + hende construc-
tion denotes low usuality, most typically being rendered by an English fre-
quency adverbial or by a combination of a frequency adverbial and a modal
auxiliary denoting habit or possibility. Where the Norwegian construction
is expanded, the meaning is guided by the expansion: with a frequency
adverbial like aldri [‘never’], as in (23), it means that something does not
occur; with en gang [‘once’], as in (24), that it occurs once; with a time
adverbial like før [‘before’], as in (25), the reference is to a particular time;
and with modal expansions with kan/kunne, it denotes possibility. What is
most striking cross-linguistically is that the det + hende construction only
exceptionally corresponds to happen in English, chiefly in translations and
when the construction is expanded by an adverbial.
300 Stig Johansson

5. A comparison with the English it + happen construction

English has a construction that is superficially similar to the it + hende con-

struction, but the conditions of use appear to differ greatly. The bare it +
happen construction is found only once in the English original texts of the
ENPC. The Norwegian correspondence is tilfeldigvis, which means that
something happens by accident and is quite different from the det + hende
construction:

(35) Intuition was telling me to turn this guy down, but it happens that
the rent on my apartment was due the next day. (SG1)
Min intuisjon fortalte meg at jeg burde avvise oppdraget, men
husleien min forfalt tilfeldigvis neste dag.

In addition, there were a handful examples of expanded constructions,

where the meaning is guided by the expansion; see the comments on (25–
28) above.
The ENPC material is sufficient to show that the English and Norwe-
gian constructions are quite different both in frequency and meaning. As
the English construction was only rarely found in the parallel corpus, I have
explored its use more fully with reference to the Oxford English Dictionary
(OED) and the British National Corpus (BNC).
The OED has examples of happen used “impersonally, with or without
it” from the end of the Middle English period. The earliest examples in the
entry for happen contain a form of happen + sa (i.e. so) + a nominal clause.
A text search in the OED quotations for the sequence happens that revealed
that the great majority of the examples conform to these patterns:
− it + so + happens + that-clause
− it + frequency adverbial (sometimes, rarely, often, frequently, etc.) +
happens + that-clause
Besides, there were some combinations with other adverbials (easily,
hardly, then, unfortunately) and a few instances of the simple it + happen
construction (chiefly found in if-clauses).
As the OED examples represent different time periods and can only be
studied within the context of a single sentence, I turned to the BNC to ex-
amine the present-day English use of the it + happen construction more
fully. The great majority of the present-tense examples are expanded and
conform to the two most common patterns found in the OED, as in:

(36) It just so happens that Berlin has expressed an interest in a loan

show. (A4A 28)
 Valency in a contrastive perspective: Structure and use 301

(37) It often happens that young children find it enormously difficult to

“surrender power”. (AM6 987)

The former refers to a particular situation, the latter to something occurring

repeatedly. Where the bare it + happen construction occurs, it seems
equivalent to it so happens that, as in:

(38) It happens that my father is one of the top people in what is known
as Work Study. (FEU 45)

In the past tense the great majority of the instances are combinations with
so, but there are also a good number of bare constructions, as in:

(39) It happened that I called at Beatrice’s house the last time Aunt
Nessy visited there – the time before she was banished. (AC7 940)

Sequences with the base form happen are of necessity expanded. The ma-
jority contain a possibility modal, usually can or may, as in:

(40) It can happen that when an assistant is helping somebody to get

dressed, the person suddenly gets violent. (B32 604)

To sum up, the main uses of the it + happen construction in English are: (1)
in combination with so, and less commonly in unexpanded constructions, it
is used to refer to a particular situation; (2) in combination with a frequency
adverbial, it refers to something occurring repeatedly; and (3) in combina-
tion with a modal auxiliary, sometimes expanded by a frequency adverbial,
it refers to the likelihood of a situation. Unlike Norwegian, English does
not have an unexpanded construction expressing usuality. Bare it + happen
constructions refer to a single situation. Though the det + hende and the it
+ happen constructions are syntactically quite parallel, they have devel-
oped in different directions in Norwegian and English.

6. A note on similar constructions in other languages

Clauses of the det + hende type are found in other languages. Swedish
hända is used in much the same way. An example from the OMC, with a
sentence in Norwegian and translations into three other languages, may
serve as a further illustration:
302 Stig Johansson

(41) Det hendte at Dagnys bok eller sytøy ble søkk borte. (HW2)
Es kam vor, dass Dagnys Buch oder Nähzeug verschwunden war.
Sometimes Dagny’s book or sewing disappeared.
Il arrivait que le livre de Dagny, ou son ouvrage de couture, dis-
paraisse complètement.

Judging by the material in the OMC, French generally uses an opening

clause with arriver in such cases, although there are also many instances
containing a frequency adverbial like parfois in the main proposition. The
German translations are more evenly divided between constructions with a
lexical verb, most typically vorkommen, and a frequency adverbial like
manchmal. The English translations regularly contain a frequency adver-
bial. We may conclude that opening clauses of the hende type occur in a
number of languages, although the extent to which they occur and the pur-
poses for which they are used may vary.
Why does English stand out from the other languages I have referred to?
The connection of the it + happen construction with a single situation may
be a reflection of the historical origin of the verb. According to the OED,
happen means “to come to pass (orig. by ‘hap’ or chance)”, which suggests
a reference to a single event. As already mentioned, the earliest examples
of the it + happen construction are combinations with a form of so. The
connection with a single event is quite clear with happen to + infinitive, as
in she happened to do it, which typically corresponds to tilfeldigvis [‘by
chance’]; cf. the rendering of the simple it + happen construction in (35)
above.3 To express usuality, happen must be combined with a frequency
adverbial.
Another relevant question is why we find forms such as the det + hende
and the it + happen constructions. Syntactically, it is natural to analyse
them as matrix clauses. From the point of view of function, they can be
viewed as clausal prefixes or utterance launchers. I will mention three pos-
sible motivations for these. In the first place, clausal prefixes allow expan-
sions of different kinds. We can, for example, easily add a stance adverbial
such as of course. Secondly, they seem to serve as thematisation devices.
Note the initial position of the frequency adverbials in examples (15–20).
Thirdly, it looks as if clausal prefixes of this kind could also have a func-
tion on the text level and introduce a longer stretch of text:
 Valency in a contrastive perspective: Structure and use 303

(42) Det hendte at mamma dasket henne bak med håndflaten. Et lite
klask. Men det var fordi hun ville at Tora skulle vite at hun var lei
seg. De klaskene var aldri vonde.
Det var ikke ofte mamma slo. Bare når hun måtte. Tora torde grå-
te når mamma slo. (HW1)
Sometimes Mama swatted her on the bottom with the palm of her
hand. A little swat.
But that was because she wanted Tora to know she was aggra-
vated. The swats never hurt. Mama didn’t hit her often. Only when
she had to. Tora wasn’t afraid to cry when Mama hit her.

A clearer example of the text-structuring function, with a similar type of

construction, is the opening of the Christmas gospel (hende and happen are
not used in these translations, but they could easily be inserted):

(43) Det skjedde i de dager at …

And it came to pass in those days that …
Factum est autem in illis diebus …
Εγένετο δε εν ταίς ήμέραις ...

All the versions have the same type of opening. Note, finally, that the struc-
tures I have dealt with have idiom-like features. In examining the material
for this paper, I did not find any examples where the matrix clause was
negated (i.e. not: det hender ikke at …; it does not happen that …). Some
related expressions, such as Norwegian kanskje and English maybe, have
gone all the way and become invariable single words (cf. section 4.3).

7. Conclusion

What do I mean by structure and use in the title of my paper? Both in the
case of spend and tilbringe in expressions of time and the det + hende and
the it + happen constructions, we clearly have equivalent structures in Eng-
lish and Norwegian, but there are differences in use. Tilbringe is a possible
choice in Norwegian, but studies of Norwegian original texts show that
some other form is often preferred. With hende and happen, we find ex-
actly the same syntactic structures, but the meaning of the unexpanded
constructions is totally different. When they are expanded, however, they
can be used to express the same meaning: it sometimes happens that … /
det hender iblant at …, it can happen that … / det kan hende at …, it hap-
304 Stig Johansson

pened once that … / det hendte en gang at … – here we have similar forms
in both languages, though the conditions of use appear to differ. If by
studying valency in a contrastive perspective we mean a structural com-
parison, it is clearly insufficient.
The conclusion I would like to draw is that cross-linguistic studies
should not be limited to a comparison of structures. We have to consider
conditions of use, including preferred ways of putting things (cf. Kennedy
1992). Such a study can best be done with reference to multilingual cor-
pora.

Notes

1. For more information on the corpora, see our websites:

http://www.hf.uio.no/ilos/forskning/forskningsprosjekter/enpc/ (ENPC) and
http://www.hf.uio.no/forskningsprosjekter/sprik/korpus/index.html (OMC).
2. Corpus examples are accompanied by a text code. Text codes ending in T
represent translations. For more information on the texts, see our websites.
3. Happen to + infinitive and the it + happen construction are, however, not
equivalent, though space does not allow a discussion of this matter in the pre-
sent paper. The bare it + happen construction frequently suggests that some-
thing is surprising or interesting and deserves special notice.

References

Gellerstam, Martin
1996 Translations as a source for cross-linguistic studies. In Languages in
Contrast. Papers from a Symposium on Text-based Cross-linguistic
Studies, Lund 4-5 March 1994, Karin Aijmer, Bengt Altenberg, and
Mats Johansson (eds.), 53–62. (Lund Studies in English 88.) Lund:
Lund University Press.
Halliday, M. A. K.
2004 An Introduction to Functional Grammar. 3d ed. Revised by Chris-
tian M. I. M. Matthiessen. London: Arnold.
Johansson, Stig
1998 On the role of corpora in cross-linguistic research. In Corpora and
Cross-linguistic Research: Theory, Method, and Case Studies, Stig
Johansson, and Signe Oksefjell (eds.), 3–24. Amsterdam/Atlanta,
GA: Rodopi.
2002 Towards a multilingual corpus for contrastive analysis and transla-
tion studies. In Parallel Corpora, Parallel Worlds. Selected Papers
from a Symposium on Parallel and Comparable Corpora at Uppsala
 Valency in a contrastive perspective: Structure and use 305

University, Sweden, 22-23 April, 1999, Lars Borin (ed.), 47–59. Am-
sterdam/New York: Rodopi.
2005 Some aspects of usuality in English and Norwegian. In Semiotics
from the North. Nordic Approaches to Systemic Functional Linguis-
tics, Kjell Lars Berge, and Eva Maagerø (eds.), 69–85. Oslo: Novus.
Kennedy, Graeme
1992 Preferred ways of putting things with implications for language
teaching. In Directions in Corpus Linguistics. Proceedings of Nobel
Symposium 82, Stockholm, 4-8 August 1991, Jan Svartvik (ed.), 335–
373. Berlin/New York: Mouton de Gruyter.
Lakoff, George, and Mark Johnson
1981 Metaphors We Live By. Chicago: University of Chicago Press.
Viberg, Åke
1996 Cross-linguistic lexicology. The case of English go and Swedish go.
In Languages in Contrast. Papers from a Symposium on Text-based
Cross-linguistic Studies, Lund 4-5 March 1994, Karin Aijmer, Bengt
Altenberg, and Mats Johansson (eds.), 151–182. (Lund Studies in
English 88.) Lund: Lund University Press.
2002 The polysemy of Swedish ge ‘give’ from a crosslinguistic perspec-
tive. In Proceedings of Euralex 2002, Anna Braasch, and Claus
Povlsen (eds.), 669–682. Copenhagen University.

The English-Norwegian Parallel Corpus (ENPC):

http://www.hf.uio.no/ilos/forskning/forskningsprosjekter/enpc/
The Oslo Multilingual Corpus (OMC):
http://www.hf.uio.no/forskningsprosjekter/sprik/korpus/index.html
Section 4
Computational aspects of valency analysis
Valency and automatic syntactic and semantic
analysis

Dieter Götz

1. Putting valency information to use

In the Valency Dictionary of English (2004), the lexicographic treatment of

verbs contains several sections of information – a complement inventory, a
pattern and examples section and a note block. Basically, complements are
described in formal terms: the dictionary lists patterns such as + NP + N for
I’ll call you a cab or He called me a nuisance, where N stands for noun
phrase and P indicates the ability of a complement to function as the subject
of a passive clause. Thus, when looking up call in a string like I’ll call you
a cab you would have to consult the section call + NP + N in the entry for
call – as opposed to call + N, where you would find I called my brother.
If I’ll call you a cab were your query, then the dictionary example
Olivia will be able to call you a cab, within the section call + NP + N,
would suggest that you were about to hit your target, and this example
would lead you to a note ( > A, B, C etc.). For the meaning of call in I’ll
call you a cab the note reads as follows:1

(1) A personI can call another person or a service such as the po-
lice, the fire brigade etc.II or call for themII, i.e. attract their
attention and ask them to come.

With this kind of information it should be possible to devise a program

which could deal with e.g. I’ll call you a cab as a dictionary query, and
produce a syntactic analysis of the vicinity of the verb, plus the note, and
plus an example. In other words, it would provide a syntactic-semantic
analysis of the sentence or of parts of the sentence. The kind of note given
to describe the semantic and lexical properties of valency complements in
VDE thus seems to provide the ideal basis for computational processing.
Obviously, the notes in VDE were designed with the aim of providing
information on possible complements in a way that can easily be under-
stood by foreign speakers of English, i.e. human users. This means that in
order to make them directly applicable to computational analysis, some of
310 Dieter Götz

the VDE notes may have to be modified by applying a fixed number of

categories more stringently. However, this could be done relatively easily,
for instance the existing note covering cases such as I call you Peter could
be rewritten in the following form:

(2) If a personI calls another personII a word that is a nameIII, they

will use that name for addressing and referring to the other person.

It should be understood that the considerations that follow are intended as a

linguist’s contribution towards a program, a program which might be used
for automatic sentence analysis. These considerations are based on the gen-
eral principles of VDE and take a slightly modified version of the valency
dictionary as its basis (i.e. modified VDE-style entries). In particular, it
should be noted that for reasons of simplicity some of the information actu-
ally provided by VDE – concerning passivizability, for instance – will be
ignored.

2. Preliminaries

The query, still I’ll call you Peter, needs tagging. The tagged version, e.g.

(3) I[N] ’ll[mod] call[verb] you[N] Peter[N]

would enable the machine to stop at the asterisk below, in the entry for call,
just before the example sentence and the letter for the respective meaning:

(4) call + N + N*
Give her another year and I reckon Olivia will be able to call you
a cab as well as any doorman in Britain. >A
I’ll call you Den. >B
I wasn’t really what you’d call a public schoolboy – I wasn’t from
the same social strata as the other kids. >C

These three options are all of the type call + N + N. On a lower syntactic
level they are, of course, not quite that ambiguous, the first is IO + DO, the
second and third are DO + OComp.
Taking syntactic functions into account makes it necessary to rewrite the
notes. Note (2) above would now be:
 Valency and automatic syntactic and semantic analysis 311

(5) >B: If a SubjN1: person calls a DON2: person a OCompN3:

name, they will use that name for addressing and referring to the
other person.

Given one pre-verbal N and two postverbal Ns, the machine can now de-
cide whether I is a suitable candidate for the subject (matching ‘person’),
you a suitable candidate for a DO (matching ‘person’) and Peter a suitable
candidate for ‘name’. On the basis of note >B, as in (6) above, syntactic
analysis and the semantic ranges of the Ns mutually confirm each other.
The machine can now present a syntactic analysis, a simplified note giving
the gist of note >B, as in above (6), and an example, in this case, I’ll call
you Den. It would, of course, exclude presenting call you a cab (because a
cab is not a person) and call you a public schoolboy (because a schoolboy
is not a name).

3. Range indicators

It is obvious that quite a number of range indicators are necessary or possi-

ble: relatively specific ones like pilot, fairly general ones like business (He
ran the business) and general ones like something/somebody for He looked
at N. In order to facilitate matching, the machine might be equipped with a
smallish internal dictionary of the following type:

(6) nose: part of body

dexterity: quality
elephant: animal
anger: state of mind
officer: person, authority

It is known that defining vocabularies of about 3000 words can cope with
up to two thirds of a text. This means that a lot of the material under analy-
sis can be treated by internal resources of the program (but see below).
Range indicators are not confined to one word. Thus, a subset of the
things that can be opened could be given by “open + DON2: structure,
door, gate, barrier, window”. If there is sufficient corpus evidence, even
more details could be given: the use of call as in I’ll call you Peter and He
called me a liar very often shows a pronoun as DO, which would yield
“DON2: person, me, you, him, her, us, them”.
312 Dieter Götz

In case of e.g. cut + N + with N (Someone cut the bread with a blunt
knife), we would have something like

(7) … cut a DON2: food, bread, meat, sausage, cheese, vegetable,

wood, string, chord, paper, cloth withN3: instrument, knife, scis-
sors.

What if the query were Someone has cut the salami with a sharp dagger?
The machine might, since it cannot understand dagger, put dagger to
WordNet or similar electronic lexicographic enterprises, and check para-
phrase, synonyms, hyperonyms, meronyms etc. related to dagger. It will be
able to understand dagger as soon as it meets knife or instrument as seman-
tically related words. For some purposes, Windows Thesaurus might do.
An assembly of WordNet, BNC, The Bank of English, FrameNet,
wortschatz.uni-leipzig and the Oxford English Dictionary will certainly
help in most cases.
The machine should perhaps have a command “search for collocates”
outside the valency pattern, to the left and to the right (<COLL, COLL>,
see below open in the meaning of ‘rain’).

4. Trials

The following is a trial run for the verbs open and admit with the respective
VDE entries for open and admit rewritten in a short form (after the exam-
ple, in bold, and abbreviated from a full meaning description as e.g. in [5]
above).

(8) M
Suddenly the kitchen door opened and Alfred was standing there. N1Subj: struc-
ture, door, gate, barrier, window
A flash of thunder pierces the complacent, brown layer of smog. The Heavens
open. A real rain falls.
<COLL; heavy, torrential, storm, rain, thunder; N1Subj: cloud, heavens,
sky; COLL> heavy, torrential, storm, rain, thunder
D1
+N
Fresh air is important throughout the day, and remember to open a window while
you carry out any indoor exercises.
N1Subj: person, force, wind, gale; N2DO: structure, window, door, gate,
barrier
He opened his eyes carefully. He was in different surroundings; he was sure of
that, at least.
N1Subj: person, animal; N2DO: eyes
 Valency and automatic syntactic and semantic analysis 313

Wade heard the bottle being opened.

N1Subj: person, instrument; N2DO: container: bottle, can, box
My key opens both doors.
N1Subj: instrument, key, card; N2DO: structure, door, gate, lock
The Princess Royal has opened an exhibition of British life in the Ukrainian
capital, Kiew.
N1Subj: person, authority, politician; N2DO: event, show, exhibition, pre-
sentation, display
The M1, Britain’s first motorway, opened in the late 1950s to speed traffic be-
tween London and the Midlands.
N1Subj: person, authority, politician; N2DO: structure, building, road
Through the surgery, she heard that the Stroke Association was opening a local
advice centre and applied for a job.
N1Subj: person, institution; N2DO: structure, institution
At least one person was killed when security forces opened fire to break up a
disturbance in one village.
N1Subj: person, weapon; N2DO: fire
D2
+ by V-ing
President Mitterand opened by posing a number of questions.
N1Subj: person
D3
+ into N
The front door opens into a hall.
N1Subj: structure, window, door, gate; intoN2PC
D4
+ onto N/on to N
A wooden door in the stone wall opens onto a grassy terrace.
N1Subj: structure, window, door, gate; onto N2PC1
From here french windows open on to a small garden.
N1Subj: structure, window, door, gate; on to N2PC1
D5
+ to N
It may be that the gardens continue to open to the public.
D6
N1Subj: structure, building, event, institution; PC1: public, people, visitor
+ with N/N V-ing
This week’s concert opened with the London premiere of John Casken’s “Darting
the Skiff”.
N1Subj: event, performance, show, exhibition, presentation, display, event;
PC1
The meeting opened with everyone giving their reasons for attendance.
N1Subj: event, performance, show, exhibition, presentation, display
D7
+ ADV
A year ago Friday, we opened. It’s been hard.
N1Subj: person, institution, business; ADV: time, year, month, week, day,
hour
There is simply no consistency about the museums of the world, or even of one
given country. They all open at different times, on different days, during different
seasons.
314 Dieter Götz

N1Subj: person, institution, business; ADV: time, year, month, week, day,
hour
The film is due to open in London at the end of the year.
N1Subj: event; {ADV1: place; ADV2: time}
Some government offices open on alternate Saturdays.
N1Subj: person, institution, business; ADV: time, year, month, week, day,
hour
The country’s first National Bottle Museum has opened in Barnsley, Yorkshire.
N1Subj: event, show, presentation, exhibition; {ADV1: place; ADV2: time}
D8
+ ADV: QUAL
The door opened easily.
N1Subj: door, gate, window, container; ADV: quality
T1
+ N + by V-ing
He opened his speech by praising the Russian Federation President, Mr Boris
Yeltsin.
N1Subj: person; N2DO: event, performance; ADV: by verb-ing
T2
+ N + to N
She refused to open her books to the auditors.
N1Subj: person, institution, organisation; N2DO: document, books; to N3:
inquiry, examination, investigation, examiner, investigator
Mr Jackson appears to be ready to open his doors to business leaders.
N1Subj: person; N2DO: door; to N3: person
He decided to open his home to paying guests.
N1Subj: person, institution; N2DO: home, place, property, building; to N3:
person
I know he had opened his heart to me and that I had found a place there.
N1Subj: person; N2DO: heart; to N3: person
T3
+ N + with N/N V-ing
You could open the door with a credit card.
N1Subj: person; N2: door, gate, lock; with N3: key, card

The vocabulary used for describing the ranges of N1Subj, N2DO etc. is,
apart from necessarily specified ranges like heart, fire, fairly general. Writ-
ing an internal dictionary, as sketched above in (7) should therefore not
prove too difficult, particularly if there is the WordNet option or another
thesaurus option. Here is the trial run for admit:

(9) D1
+N
Should she force him to admit the truth?
N1Subj: person, institution, organisation; N2DO: truth, crime, mistake, guilt,
responsibility
Greycoat Commercial Estates and associated companies finally admitted defeat and
sold their land interests to the GLC on 29 March 1984.
N1Subj: person, institution, organisation, military leader; N2DO: defeat, lose
 Valency and automatic syntactic and semantic analysis 315

Apparently, Barker admitted his mistake and apologised to Mujtaba afterwards.

N1Subj: person, institution, organisation; N2DO: truth, crime, mistake, guilt,
responsibility
No-one has admitted responsibility for the murder.
N1Subj: person, institution, organisation; N2DO: truth, crime, mistake, guilt,
responsibility
He admitted each of the delegates himself.
N1Subj: person, institution, organisation, rule, right; N2DO: person, things
Each ticket admits two people and is valid until the end of October. N1Subj: per-
son, institution, organisation, rule, right; N2DO: person, things
I lay in my pallet waiting for sleep, with my window open to admit the bright au-
tumn air.
N1Subj: structure; N2DO
D2
+ V-ing
So far no group has admitted carrying out the murder.
N1Subj: person; N2DO: doing something, truth, crime, mistake, guilt, respon-
sibility
But if he had something to do with it, why’d he admit being here?
N1Subj: person; N2DO: truth, crime, mistake, guilt, responsibility; N2DO:
doing something wrong, truth, crime, mistake, guilt, responsibility
D3
+ (that)-CL
I have to admit that I have bad handwriting, but that is not a moral fault of mine.
N1Subj: person; DO: that-clause
Philip admits he can’t walk past a bookshop without going in.
N1Subj: person; DO: clause
I have to admit, sir, there’s one thing that worries me.
N1Subj: person; DO: clause
D4
wh-CL
Perhaps, he muses, Milligan was terrified to admit how much pleasure he was
missing out on.
N1Subj: person; DO: wh-clause
I am ashamed to admit what a relief this was.
N1Subj: person; DO: wh-clause
Some of the fur traders have been bold enough to admit why their industry has
been hit.
N1Subj: person; DO: wh-clause
D5
QUOTE / SENTENCE
“I must admit, when we got to Sydney I really didn’t feel very well at all,” she said.
N1Subj: person; DO: clause
“I do not know yet,” she admitted.
N1Subj: person; DO: clause
D6
+ of N/V-ing
There will be slow growth and greater unemployment for years: our economic
problems admit of no other solution.
N1Subj: person; of N2
If the link really were necessary, it would admit of no exceptions.
316 Dieter Götz

N1Subj: person; of N2
Not only do both works admit of being read either exoterically or esoterically: both
works express precisely similar attitudes towards eternal life.
N1Subj: person; of V-ing
D7
+ to N/V-ing
Your father did not admit to his blindness and your mother, long after his death,
continued to behave as if he had not died.
N1Subj: person; to N2
She described herself as an emotional person easily moved to laughter or tears and
admitted to being rather shy.
N1Subj: person; to V-ing
T1
+ N + as N
The United Nations has voted to admit Namibia as its one-hundred and sixtieth
member, one month after it gained independence from South Africa.
N1Subj: person, institution, organisation; as N2
T2
+ N + into N
Even Galiani admitted more of social forces into his utility theory than modern
theorists would allow.
N1Subj: person, institution, organisation, structure; N2DO; into N3
The side arcades which with their tall arches above admit as much light into the
nave as is possible.
N1Subj: person, institution, organisation, structure; N2DO; into N3
You will be admitted into the hospital either on the day of the procedure or possi-
bly the night before.
N1Subj: person, institution, organisation, structure; N2DO; into N3
T3
+ N + to N
He is also in favour of women being admitted to his club, the United Oxford and
Cambridge University Club.
N1Subj: person, institution, organisation; N2DO; to N3
Wu Man, a brilliant young virtuoso, was among the first group admitted to the
Beijing Conservatory after the Cultural Revolution.
N1Subj: person, institution, organisation; N2DO; to N3
Six people are reported to have been admitted to hospital with bullet wounds or
injuries from bomb explosions.
N1Subj: person, institution, organisation; N2DO: person; to N3
+ N + to N
He may never have admitted this even to himself.
N1Subj: person, institution, organisation; toN2: person, institution, organisa-
tion
T4
+ to N + (that)-CL
We don’t admit to ourselves that we’re playing games with our children.
N1Subj: person, institution, organisation; toN2: person, institution, organisa-
tion; DO: that-clause
I’m talking about the people who admitted to me they were guilty. N1Subj: per-
son, institution, organisation; toN2: person, institution, organisation; DO:
clause
 Valency and automatic syntactic and semantic analysis 317

T5
+ to N QUOTE / SENTENCE
“I absolutely cannot compete with it all, or be natural or cheerful, when they won’t
treat me like a human being,” he admitted to his mother.
N1Subj: person, institution, organisation; toN2: person, institution, organisa-
tion; DO: clause
Q
+ N + as N + to N
The Foreign Ministers of Lithuania, Latvia and Estonia say they have asked for the
three Baltic states to be admitted as observers to the thirty-five nation human rights
meeting taking place in Copenhagen.
N1Subj: person, institution, organisation; DON2: person, institution, organi-
sation; as N3; to N4

5. Comment on the trial runs

100 occurrences of admitted, admits from the BNC, randomised, were

checked against the above presentation. There were no occurrences of ad-
mit* which were not captured by the re-written dictionary entry. Unfortu-
nately, however, checking 200 randomised results for the queries opens and
opened had proved less satisfactory. There is no denying it, some meanings
are simply missing, or, at least, not sufficiently illustrated by examples,
such as e.g. open plus book, envelope, mail, open plus account, open plus
part of body (abdomen), open plus something that is wrapped up or folded,
open plus something in order to get to its interior parts.
Some of these meanings can of course be added by expanding the
ranges. Thus, in the appropriate line above there should also be bag and
case, answering for purse and wallet, and adding wings, hatchet and border
at the appropriate places would help as well. The next 50 randomised re-
sults yielded no new puzzles, except perhaps that the pores opened. Revi-
sion is necessary, but it should not prove too difficult.
Now suppose you had He opened his fist. The machine would, at the
present stage of instructions, be baffled. In such a case it could look up fist
in WordNet:

(10) fist, clenched fist (a hand with the fingers clenched in the palm (as
for hitting))

and go from there to

(11) clenched, clinched (closed or squeezed together tightly) “a

clenched fist”; “his clenched (or clinched) teeth”
318 Dieter Götz

which is where it would note one of the key concepts for open, namely
closed. It could then copy the line and present it for inspection. To repeat
the method: a word will be checked section by section of the WordNet en-
try until parts of the meaning description are found. For this, the machine
would have to take words from the meaning block and check it against the
outside dictionary or thesaurus information.
Unfortunately however, there are occurrences of open, particularly fre-
quent metonymic and metaphoric usages, where this method fails. Decod-
ing I opened the whisky requires finding bottle of, cask of in WordNet, but
these collocations are not listed. It would require searching a corpus for
collocations of the type container + whiskey.
We may encounter quite a number of similar verbs with a multitude of
possible meanings. In these cases, the procedure might be to have two
meaning blocks. One of them would be a “usage block”, with notes as al-
ready presented. These notes would be presented in case of alleged cer-
tainty or high probability (as in The gates opened and the chariot went
through). Incidentally, this kind of notes would be necessary anyhow if we
wanted to add multilingual translation equivalents. From this usage block
the machine would gather notes for presentation like:

(12) A door, window, etc. can open or be opened, i.e. … or

A container such as a tin can be opened, i.e. … or
A bag, a case, a chest can be opened, i.e. …

The other block would be a kind of “general meaning block”, reverted to by

default and presented as a whole to the human user:

(13) Open can mean ‘become open, become no longer closed’. Most
things that can be said to be closed can be opened.
(i) A door, window etc. can open or be opened, i.e. …
(ii) A container such as a tin can open or be opened, i.e. …
(iii) A bag, a case, a chest can open or be opened i.e. …
(iv) A lock can open or be opened i.e. …
(v) You can open something that is written: a book or a let-
ter, i.e. …
(vi) You can open an account at a bank, i.e. …
(vii) Something that is folded or wrapped up can open or be
opened, i.e. …
(viii) You can open something in order to access its interior
parts, i.e. …
Open can mean ‘start working, functioning, taking place’. etc…
 Valency and automatic syntactic and semantic analysis 319

With this general meaning information the human user could decode occur-
rences like They opened his heart, The buzzard opened his wings, I opened
the umbrella, I opened the sherry, I opened the watch.

6. Other applications

Detailed notes might open up other applications. A machine might be able

to answer yes to a question like Did the show take place? when given The
Princess Royal opened the exhibition. And this could lead to more effective
search machines. Else, you might put all notes together in a single file.
There would be enough material to investigate issues like “other roles than
agent in subject position” or “the semantics of direct objects with verbs of
motion”, or “nouns after verbs of saying”.

Notes

1. The description uses Roman superscripts to indicate parts of the pattern in the
order in which they normally appear, and specifies the semantic range to
which these parts belong.

References

Herbst, Thomas, David Heath, Ian Roe, and Dieter Götz (eds.)
2004 A Valency Dictionary of English. A Corpus-Based Analysis of the
Complementation Patterns of English Verbs, Nouns and Adjectives.
Berlin/New York: Mouton de Gruyter.
Simpson, John, and Edmund Weiner (eds.)
1989 The Oxford English Dictionary. 2d ed. Oxford: Oxford University
Press.

The Bank of English: http://www.titania.bham.ac.uk/

British National Corpus: http://info.ox.ac.uk/bnc/
Framenet: http://framenet.icsi.berkeley.edu/
WordNet: http://wordnet.princeton.edu/
Wortschatz Lexikon: http://wortschatz.uni-leipzig.de
Handling valency and coordination in Database
Semantics 1
Roland Hausser

1. Sign-based vs. agent-based approaches to language

Most linguistic approaches are sign-oriented in that they analyze expres-

sions of natural language as objects, fixed on paper, magnetic tape or by
electronic means. They abstract away from the aspect of communication
and analyze signs as hierarchical structures which are represented as trees
and formally based on the principle of possible substitutions.
Database Semantics (DBS), in contrast, is agent-oriented in that it ana-
lyzes signs as the result of the speaker’s language production and as the
starting point of the hearer’s language interpretation. Inclusion of the
agents’ production and interpretation procedures requires a time-linear
analysis which is formally based on the principle of possible continuations.
The goal of Database Semantics is a theory of natural language commu-
nication which is complete with respect to function and data coverage, of
low mathematical complexity, and suitable for an efficient implementation
on the computer. The central question of Database Semantics is: how does
communicating with natural language work?
In the most simple form, DBS answers this question as follows: natural
language communication takes place between cognitive agents. These have
interfaces for non-verbal recognition and action at the level of context, and
verbal recognition and action at the level of language. Each agent contains
a database in which contents are stored. These contents consist of the
agent’s knowledge, memories, current recognition, intentions, plans, etc.
Contents are read into and out of the database by means of a time-linear
algorithm. Cognitive agents can switch between the speaker- and the
hearer-mode (turn-taking).
In a communication procedure, an agent in the speaker-mode codes
content from its database into signs of language which are realized exter-
nally via the language output interface. These signs are recognized by an-
other agent in the hearer-mode via the language input interface, their con-
tent is decoded, and stored in the second agent’s database. This procedure
is successful if the content coded by the speaker is decoded and stored
equivalently by the hearer.
322 Roland Hausser

2. Complementation

In Valency Theory (cf. Ágel 2000) there is a basic distinction between

valency carriers and valency fillers. Valency carriers have certain slots for
which there must be suitable fillers in order for a sentence to be grammati-
cal and complete. For example, the verb MPQY is a valency carrier with
valency positions for two nominal fillers, one serving as the subject, the
other as the object.
In Dependency Grammar, the relations between a valency carrier and its
fillers in a sentence are shown in the form of a tree, called dependency
graph or stemma. Consider the following dependency graph of the sentence
,WNKCMPQYU,QJP:

Example 1. Representing dependency relations as a tree

Supplying a valency carrier with suitable fillers is called complementation.

In Database Semantics, the grammatical structure of a sentence is repre-
sented as a set of proplets rather than a tree. Proplets are non-recursive
(flat) feature structures in the sense that attributes may not take feature
structures as values.

Example 2. Representing dependency relations as a set of proplets

The relations between the valency carrier know and its fillers Julia and
John are coded by values for certain attributes. More specifically, the prop-
let know has the attribute arg with the values ,WNKC ,QJP, and the proplets
 Handling valency and coordination in Database Semantics 323

Julia and John have the attribute fnc with the value MPQY (bidirectional
pointering).
Proplets belonging to the same proposition share a common proposition
number (here prn: 6). Each proplet is an autonomous item which may be
stored anywhere in memory without affecting the coding of grammatical
relations. Thus, the grammatical relations characterized by sign-oriented
approaches in the form of a tree are recoded equivalently in Database Se-
mantics by means of values of certain attributes. They establish a bidirec-
tional pointering between proplets of valency carriers and their filler prop-
lets, regardless of where they are located in storage.
There is a total of eleven attributes in the proplets of example (2), the
values of which have the following properties:
1. Surface attribute: UWT
Each proplet has a unique surface attribute. Its value is the language
dependent surface of a proplet, needed for lexical lookup in the hearer-
mode. After lexical lookup, the sur value is usually omitted.
2. Core attributes: PQWPXGTDCFL
Each proplet has a unique core attribute, which gets its value from the
lexicon. From a sign-theoretic point of view,2 the core values may con-
sist of a concept, a pointer, or a marker, which corresponds to the sign
kinds of symbol, indexical and name. While a XGTD-attribute can only
take a concept as its value, an CFL-attribute can take a concept or a
pointer. The most general kind is the PQWP-attribute, which can take a
concept, a pointer, or a marker as value.3
3. Continuation attributes: HPE, CTI, OFF, OFT
Each proplet has several continuation attributes, which get their values
by copying during the composition of proplets (cf. example [3]). The
values consist of characters (char), which represent the names of other
proplets. In complete propositions, the values of HPE, CTI, and OFF must
be non-NIL, while that of OFT may be NIL.
Additional continuation attributes are RE (previous conjunct) and PE
(next conjunct) in verbal proplets. They are used for connecting proposi-
tions in a time-linear sequence.
4. SynSem attributes: ECV, UGO
Each proplet has the SynSem attributes ECV and UGO. They get their val-
ues in part from the lexicon, for example PO (for name), and in part by
copying.
5. Book-keeping attributes: RTP (proposition number)
Each proplet has one or more book-keeping attributes, which get their
values by the control structure of the parser and consist of numbers (in-
tegers). In connected proplets, these values must be non-NIL. Additional
324 Roland Hausser

book-keeping attributes are KF[ (identity), YTP (word number), and VTE
(transition counter).
Coding grammatical relations in terms of attributes and values is suitable
not only for the treatment of complementation, i.e. obligatory and optional
relations between a carrier and its fillers which are restricted in some
grammatical way (for example agreement), but also for the treatment of
adjuncts, which are usually excluded by lexical approaches to Valency
Theory (cf. Herbst 1999; Herbst et al. 2004).
In other words, the handling of grammatical relations in Database Se-
mantics applies not only to the traditional valency relations, but to functor-
argument structures in general. Treating valency relations as an instance of
functor-argument structure has several advantages, one of them being that
valency relations are supplied with the standard semantic interpretation of
functor-argument structure.

3. Basic model of communication in Database Semantics

Using the reconstruction of complementation in Database Semantics, we

may illustrate the basic functioning of natural language communication. An
agent in the hearer-mode receives a sequence of unanalyzed surfaces as
input. During lexical lookup, these surfaces are matched with correspond-
ing lexical proplets. Lexical proplets are unconnected in that the attributes’
coding relations to other proplets have no values yet.

Example 3. Coding valency structure in the hearer-mode

 Handling valency and coordination in Database Semantics 325

After lexical lookup, the proplets are connected by syntactic-semantic pars-

ing. This strictly time-linear procedure is performed by an LA-grammar
called LA-hear. It is based on copying values (here indicated by arrows)
and by providing values to the book-keeping attributes (here the prn value
22).
The result of syntactic-semantic parsing in the hearer-mode is a(n unor-
dered) set of proplets. For purposes of indexing and retrieval, these proplets
are stored at the end of alphabetically ordered token lines. Each token line
begins with the name of the core value and is followed by all proplets con-
taining this core value. In this way, the content coded by the natural lan-
guage expression ,WNKC MPQYU ,QJP is stored in the database of the hearer
(see hearer on the left):

Example 4. Transfer of content from speaker to hearer

When the hearer turns into a speaker (see speaker on the right), stored con-
tent is activated by means of a time-linear navigation. Let us assume, for
example, that Julia has been activated by the agent’s control structure as
326 Roland Hausser

the navigation-initial proplet. It has the attribute fnc with the value MPQY
and the attribute prn with the value 22. Based on these values, the continua-
tion proplet know is being activated next (see arrow from Julia to know).
The first value of its arg attribute, i.e. ,WNKC, confirms the legality of the first
navigation step. The second value, i.e. ,QJP, provides the information for
continuing the navigation by activating a third proplet (see arrow from
know to John).
Such a navigation through stored content serves as the basic model of
thought in Database Semantics. Based on the grammatical relations be-
tween proplets (intrapropositional navigation) and between propositions
(extrapropositional navigation), the navigation uses the standard retrieval
mechanism of the database.
Because proplets normally provide more than one possible successor,
the navigation algorithm, called LA-think, must make choices. The most
basic solutions are either completely random choices or completely fixed
choices, based on some predefined schema. For rational behavior, however,
the LA-think grammar must be refined into a control structure which
chooses between continuation alternatives based on the evaluation of exter-
nal and internal stimuli, the frequency of previous traversals, learned pro-
cedures, theme/rheme structure, etc.
For present purposes, we assume the predefined schema of a standard
navigation, starting with the verb and continuing with the arguments in
their given order. This navigation may be represented schematically as
VNN, with V representing the verb proplet, the first N the subject, and the
second N the object. In principle, any such navigation through the word
bank is independent of language. However, in cognitive agents with lan-
guage, the navigation serves as the speaker’s conceptualization, i.e., as the
speaker’s choice of what to say and how to say it.
A conceptualization defined as a time-linear navigation through content
makes language production relatively straightforward: if the speaker de-
cides to communicate a navigation to the hearer, the concept names (i.e.,
values of the core attributes) of the proplets traversed by the navigation are
translated into their language-dependent counterparts and realized as exter-
nal signs. In addition to this language-dependent lexicalization of the uni-
versal navigation, the system must provide
1. language-dependent word order,
2. function word precipitation, and
3. word form selection for proper agreement.
This process is handled by language-dependent LA-speak grammars in
combination with language-dependent word form production. For example,
the word form CVG is produced from an eat proplet the UGO attribute of
 Handling valency and coordination in Database Semantics 327

which contains the value RCUV.

Because word form selection for proper agreement involves a large
amount of morphosyntactic detail, the language-specific production from
the VNN navigation of our example is characterized by the following sim-
plified format:

Example 5. Schematic production of ,WNKCMPQYU,QJP. (speaker-mode)

The letter ‘i’ stands for the number of the sentence produced. The letters P,
HX and R are abstract surfaces for name, finite verb, and punctuation (here
full stop), respectively.
The derivation begins with a navigation from V to N, based on LA-think.
In line i.1, the N is realized as the P ‘,WNKC’, and in line i.2 the V is realized
as the HX ‘MPQYU’ by LA-speak. In line i.3, LA-think continues the naviga-
tion to the second N, which is realized as the P ‘,QJP’ by LA-speak. Finally,
LA-speak realizes the R ‘.’ from the V proplet (line i.4).
The time-linear switching between LA-think and LA-speak is motivated
not only by psychological considerations,5 but also by computational effi-
ciency. The reason is that realizing the surfaces of proplets as soon as pos-
sible (instead of navigating to the end of the proposition first) results in a
more restricted set of candidates for matching by the LA-speak rules than
having to consider the proposition’s complete set of proplets.
The method of production shown in (5), based on an underlying VNN
navigation, can be used to realize not only an SVO surface (subject-verb-
object) as in the above example, but also an SOV and (trivially) a VSO
surface (Greenberg 1963). The ordering and lexicalization are specified by
the rules of an LA-speak grammar, whereby the design of these grammars is
supported conceptually by abstract derivations like (5).
328 Roland Hausser

4. Treating adjuncts

Having outlined the basic mechanism of natural language communication

in Database Semantics with a reanalysis of complementation, let us turn
next to the treatment of adjuncts. Herbst (1999) presents the following ex-
ample, shown here in simplified form:

Example 6. Dependency graph (stemma) with two-place verb and adjunct

The optional character of the adjunct is indicated by the dotted line. In Da-
tabase Semantics, the same example is represented as the following set of
proplets:

Example 7. Corresponding representation in Database Semantics

There are two adjuncts, the adnominal modifier QVJGT and the adverbial
modifier [GUVGTFC[. The bidirectional pointering is between the modified
(mdd), i.e. CTVKUV and XKUKV, and the modifier (mdr), i.e. QVJGT and [GUVGTFC[,
respectively. The optional character of modifiers is treated as a property of
the mdr attribute (typing), which may have the value NIL.

5. The treatment of certain function words (translatives)

One of the basic distinctions in Tesnière (1959) is between mot plein and
mot vide, which may be translated as content word and function word, re-
spectively. Examples of function words are determiners and auxiliaries,
 Handling valency and coordination in Database Semantics 329

which are analyzed as “translatives”.

In Database Semantics, function words are fused with associated content
words during interpretation in the hearer-mode (absorption) and extracted
during production in the speaker-mode (precipitation). Consider the follow-
ing example illustrating absorption of a determiner and two auxiliaries in a
hearer-mode derivation:

Example 8. Parsing 6JGNKVVNGFQIJCUDGGPDCTMKPI in the hearer mode

330 Roland Hausser

Because Database Semantics is strictly surface compositional, all word

form surfaces in the input expression are lexically analyzed (see lexical
lookup). The lexical analysis of function words is special, however, in that
their core values are substitution values, e.g. n_1. During the LA-hear deri-
vation, the substitution value of a function word is replaced by the associ-
ated content word.
Consider the time-linear derivation of VJG NKVVNG FQI in (8). In combina-
tion step 1, the adnominal NKVVNG is copied into the mdr slot of the, and the
substitution value n_1 of the is copied into the mdd slot of little. In combi-
nation step 2, the two occurrences of the substitution value n_1 in the first
two proplets are replaced by the core value of dog. Then the third proplet is
discarded. In this way, all relevant grammatical relations have been coded
into the first two proplets.
Similarly in the time-linear derivation of the complex verb: in combina-
tion step 3, the core value FQI is copied into the arg slot of the lexical
analysis of JCU, and the substitution value v_1 is copied into the fnc slot of
the dog proplet. In combination step 4, the two occurrences of the substitu-
tion value v_1 are replaced by the substitution value v_2 of the lexical
analysis of DGGP. In combination step 5, finally, the occurrences of v_2 are
replaced by the core value DCTM of the last proplet. Then the last proplet is
discarded. In the course of this derivation, the semantic contribution of the
auxiliaries is coded into the sem slot of the verb (not shown here).6
The three proplets derived in this way can now be stored in the data
base. Based on the values of the attributes arg and mdr, the proplets may be
activated in a VNA navigation. In the speaker-mode, the original surface
may be reconstructed as shown in the following derivation:

Example 9. Schematic production of 6JGNKVVNGFQIJCUDGGPDCTMKPI.

 Handling valency and coordination in Database Semantics 331

Here, F CP P CZ PX, and R stand for determiner, adnominal, auxiliary,
non-finite verb, and punctuation (full stop), respectively.

6. Coordination

The most important construction of natural language besides valency in

particular and functor-argument structure in general is coordination. For
Database Semantics, this presents the task of integrating the treatment of
valency and coordination in a unified, functional system. The solution is
illustrated below using a simple example of a nominal conjunction with the
function word (junctive) CPF, serving as the subject:

Example 10. Lexicalization of6JGOCPVJGYQOCPCPFVJGEJKNFUNGGR

The task of connecting these isolated proplets in a time-linear derivation

(similar to example [8]) raises two basic questions. The first is how to build
the grammatical relations between the conjuncts. In other words: what is
the “connexion” between the elements of a conjunction in DBS? Consider
the following solution:

Example 11. Relations within a nominal conjunction


Each conjunct specifies its predecessor in the pc (previous conjunct) and its
successor in the nc (next conjunct) attribute. These attributes receive their
values by downward copying in the hearer-mode, and are used for upward
retrieval during conceptualization. The kind of conjunction, for example
CPF versus QT, is indicated after the concept value of the first conjunct (here
OCP). This treatment of coordination is strictly surface compositional and
332 Roland Hausser

time-linear, in contrast to transformational systems such as Hellwig (2003).

The second question is how to integrate such a conjunction into the
valency or functor-argument structure of a proposition. Consider the fol-
lowing solution:

Example 12. Relating a nominal conjunction to the valency structure


This analysis specifies the grammatical relations of a conjunction in a way
which is as complete as necessary and as parsimonious as possible: only the
first conjunct man specifies the verb in its HPE-slot, and the verb specifies
only the first conjunct in its CTI-slot. For retrieval, this has the following
consequence.
When searching for UNGGRKPI EJKNF, for example, the child proplet in
question merely indicates that it is part of a conjunction (RE YQOCP); in
order to determine the associated verb, LA-think has to navigate to the first
element of the conjunction, i.e. man, and check whether or not its HPE value
is UNGGR. The verb proplet sleep also indicates that its argument is a con-
junction (CTI OCP ). Therefore, the search for a non-initial conjunct is
attempted only if the proplet belongs to a proposition which actually con-
tains a conjunction.
The (re)production of the input sentence is based on a VNNN se-
quence.7 The following derivation uses the new abstract surface EP, for
conjunction:
 Handling valency and coordination in Database Semantics 333

Example 13. Production of6JGOCPVJGYQOCPCPFVJGEJKNFUNGRV

The conjunction is lexicalized from the first conjunct in line i.6. Due to this
late realization, it appears between the penultimate and ultimate conjunct.
A second kind of intrapropositional conjunction are verbal conjunctions,
as in ,QJPDQWIJVEQQMGFCPFCVGVJGRK\\C The relations within a verbal
conjunction are similar to those in a nominal conjunction. This approach
works also for the combination of conjunctions as in VJGOCPVJGYQOCP
CPF VJG EJKNF DQWIJV EQQMGF CPF CVG VJG UVGCMU VJG RQVCVQGU CPF VJG
DTQEEQNK. The functor-argument structure of this example is that of VJG
OCPDQWIJVVJGUVGCMU.

7. Conclusion

It has been shown in which way the notions of Dependency Grammar

based on Valency Theory, such as valency carrier (functor) and valency
filler (argument), adjunct (modifier), mot plein (content word), mot vide
(function word, translative, junctive), connexion (grammatical relation),
etc., have counterparts in Database Semantics. At the same time it became
apparent that the realization of these notions and their location in the over-
all theories is different in the two approaches. This is mainly because De-
pendency Grammar and Valency Theory are sign-oriented while DBS is
334 Roland Hausser

agent-oriented.
To further clarify the notions and distinctions in Database Semantics as
compared to Dependency Grammar and Valency Theory, consider the fol-
lowing hierarchy:

Example 14. Hierarchy of notions and distinctions in Database Semantics

At the root of the tree there is the time-linear concatenation (level 0) of

word forms which are in relations to each other (level 1). This most basic
structural property of natural language is realized by the time-linear algo-
rithm of LA-grammar (Hausser 1992) in its variants of LA-hear, LA-think,
and LA-speak (Hausser 2001).
The word forms are divided into content words and function words
(level 2). The content words are divided into the three basic kinds of signs,
namely symbol, indexical, and name (level 3). In the branch of relations,
the kinds of signs serve the “vertical” relation of reference, implemented as
a matching between the levels of language and context.
 Handling valency and coordination in Database Semantics 335

The kinds of signs (level 3) are correlated with the parts of speech (level
4). Symbols can be verbs, adjectives, or nouns, indexicals can be adjectives
or nouns, and names can be nouns only. This is shown graphically by the
lines relating the kinds of signs and the parts of speech. In the branch of
relations, the parts of speech serve the “horizontal” relations of functor-
argument structure and coordination.8
The structures shown above the dotted line separating level 4 and 5 are
universal: all natural languages are based on a time-linear concatenation of
word forms, the distinction between content and function words, the three
kinds of signs, the three parts of speech, the vertical relation of reference,
and the horizontal relations of functor-argument structure and coordination.
In Database Semantics, these structures are realized in the form of an artifi-
cial agent with interfaces for recognition and action at the context and the
language level, the data structure of a word bank (database) containing
proplets, and the algorithm of LA-grammar for reading content into and out
of the database.
The structures shown below the dotted line are language-dependent. For
the verb forms of the Indo-European languages, for example, this holds for
the genus, modus, and tempus verbi (levels 5, 6, and 7), the valency struc-
ture of the verbs (level 8), as well as the person and number distinction
(level 9). For the adjectives, it holds for the distinction between adnominal
and adverbial use and for synthetic comparation. For the nouns, it holds for
the different case systems, and the number and gender distinctions (which
are missing, for example, in Korean). Language-dependent is also whether
the coding of grammatical relations and distinctions is handled analytically
by means of functions words (e.g. junctives, translatives) or synthetically in
terms of morphology. In Database Semantics, these aspects are treated by
language-dependent LA-grammars with a suitable lexicon, restrictions on
variables for handling agreement, and a rule system for handling word or-
der.

Notes

1. This paper benefited from comments by Jae Woong Choe (Korea University,
Seoul), Besim Kabashi (Friedrich-Alexander-University, Erlangen), Haitao
Liu (Communications University of China, Beijing), and Brian MacWhinney,
(Carnegie Mellon University, Pittsburgh).
2. See Hausser (1999), chapter 6.
3. Seventh Principle of Pragmatics (PoP-7). See Hausser (1999: 107).
4. For functor, argument, modified, and modifier, respectively.
336 Roland Hausser

5. The principle of incrementality emphasizes the extent to which interpretation

occurs in real time as new words are being heard (cf. MacWhinney 1987). The
linkage of valency relations to incremental parsing found in the LA approach
also provides a comprehensive approach to the aspects of the language learn-
ing problem known as the logical problem of language acquisition. In particu-
lar, MacWhinney (2004; 2005) has shown that children can induce the correct
set of valency relations in their target language by focusing on meaningful re-
lations in main clauses, along with their verbal complements.
6. For a more detailed analysis of the major constructions of English see Hausser
(2006).
7. Note that VNNN can represent a three-place proposition like John gave Mary
a flower or a one-place proposition with a nominal conjunction as in (12). The
notation may be disambiguated by means of subscripts.
8. Treating coordination as a bona fide grammatical relation like functor-
argument structure, handled in terms of the attributes and values of proplets, is
in contrast to Lobin (1993a: 176): “the best way of dealing with coordination
in syntax is not to deal with it at all, but ‘process it away’ immeadetly [sic].”
See also Lobin (1993b).

References

Ágel, Vilmos
2000 Valenztheorie. Tübingen: Gunter Narr.
Greenberg, Joseph H.
1963 Some universals of grammar with particular reference to the order of
meaningful elements. In Universals of Language, Joseph H. Green-
berg (ed.), 73–113. Cambridge, Mass.: MIT Press.
Hausser, Roland
1992 Complexity in left-associative grammar. Theoretical Computer Sci-
ence, 106 (2): 283–308.
1999 Foundations of Computational Linguistics, Human-Computer Com-
munication in Natural Language. 2d ed. 2001. Berlin/New York:
Springer-Verlag.
2001 Database semantics for natural language. Artificial Intelligence 130
(1): 27–74.
2006 A Computational Model of Natural Language Communication: In-
terpretation, Inference, and Production in Database Semantics, Ber-
lin/New York: Springer-Verlag.
Hellwig, Peter
2003 Dependency unification grammar. In Dependency and Valency. An
International Handbook of Contemporary Research, Vilmos Ágel,
Ludwig M. Eichinger, Hans-Werner Eroms, Peter Hellwig, Hans-
Jürgen Heringer, and Henning Lobin (eds.), 593–635, Berlin/New
York: Mouton de Gruyter.
 Handling valency and coordination in Database Semantics 337

Herbst, Thomas
1999 English valency structures – A first sketch. Erfurt Electronic Studies
in English (EESE) 6:
http://webdoc.gwdg.de/edoc/ia/eese/artic99/herbst/6_99.html.
Herbst Thomas, David Heath, Ian F. Roe, and Dieter Götz (eds.)
2004 A Valency Dictionary of English: A Corpus-Based Analysis of the
Complementation Patterns of English Verbs, Nouns and Adjectives.
Berlin/New York: Mouton de Gruyter.
Lobin, Henning
1993a Linguistic perception and syntactic structure. In Functional Descrip-
tion of Language, Eva Hajicova (ed.), 163–178. Prague: Charles
University.
1993b Koordinations-Syntax als prozedurales Phänomen. Tübingen:
Gunter Narr.
MacWhinney, Brian
1987 The competition model. In Mechanisms of Language Acquisition,
Brian MacWhinney (ed.), 249–308. Hillsdale, NJ: Lawrence Erl-
baum.
2004 A multiple process solution to the logical problem of language ac-
quisition. Journal of Child Language 31: 883–914.
2005 Item-based constructions and the logical problem. ACL 2005: 46–54.
Tesnière, Lucien
1959 Éléments de Syntaxe Structurale, Paris: Editions Klincksieck.
Pronominal clitics and valency in Albanian:
A computational linguistics perspective and
modelling within the LAG-Framework1
Besim Kabashi

1. Theoretical aspects: Pronominal clitics and valency

When reading Albanian texts, elements like e, i, ia cannot be overlooked. If

they immediately precede the verb or if they are part of the verb form in
non-negated imperative sentences, they are instances of trajtat e shkurtra të
përemrave vetorë [short forms of personal pronouns; Domi 1995; 1997].
Buchholz and Fiedler (1987) call them Objektszeichen [object signs], Kal-
luli (1995) names them clitics. Like Newmark, Hubbard, and Prifti (1982),
we will use the term pronominal clitics (pCls).

1.1. Formal properties of pronominal clitics in Albanian

The following table gives an overview of the Albanian pCl forms:

Table 1. The personal pronouns in Albanian (left column) with their clitic forms
(right column)

1st Person 2nd Person

Singular Plural Singular Plural
Nom. unë - ne - ti - ju -
Dat. mua më neve na ty të juve ju
Acc. mua më ne na ty të ju ju

3rd Person
Singular Plural
Masculine Feminine Masculine Feminine
Nom. ai - ajo - ata - ato -
Dat. atij i asaj i atyre u atyre u
Acc. atë e atë e ata i ato i
340 Besim Kabashi

PCls in Albanian indicate the person and number of the respective objects
of the verb. They occur in dative and accusative case: dative pCls can be
combined with accusative pCls. In most cases this results in amalgamated
forms (crasis). For example më and e amalgamate to ma. The combination
of na with e, i, or u does not involve amalgamation but concatenation, in
which case the dative precedes the accusative: na e, na i, and na u.
There are two morphosyntactic types of pCls: bound and free. Bound
forms occur within positive (non-negated) imperatives after the verb stem
(enclitic position). In the plural they appear between the verb stem and the
suffix ni, cf. the following example:

Example 1. Pronominal clitics as bound forms2

(1) sill VStem No pCl; Sg; ‘bring’
a. sille V Stem +pCl A pClA, Sg; ‘bring it’
b. sillma VStem+pClD+A pClD+A, Sg; ‘bring me it’
c. sillni VStem+ni No pCl; Pl; ‘bring’
d. silleni VStem+pClA+ni pClA, Pl; ‘bring it’
e. sillmani VStem+pClD+A+ni pClD+A, Pl; ‘bring me it’

In cases with negation particles, pCls cannot occur as bound forms, e.g.
Mos e sill! [‘(You: Sg) do not it bring!’, i.e. ‘Do not bring it!’] and Mos e
sillni! [‘(You: Pl) do not it bring!’, i.e. ‘Do not bring it!’]. As free forms,
pCls always precede the finite verb (proclitic position).3 Word order in
Albanian is relatively free. Therefore, both subject and objects may appear
in front of the verb complex, which allows for a large number of different
sentence patterns for a given verb; the order of elements within the verb
complex is fixed. If all of these elements occur, they are in the following
sequence: negation, future marker/modal verb, subjunctive particle, pCls,
finite and non-finite verb. The subject can be deduced from verb inflection
and thus be left out.
PCls in Albanian appear either in addition to objects in the sentence (ob-
ject doubling) or they replace objects in the sentence, so that the objects
themselves can be left out (object elimination).
 Pronominal clitics and valency in Albanian 341

1.2. Object doubling

The following example (2b) shows the doubling of the accusative object:

Example 2. Accusative object doubling4

(2) S+V
Ne shohim.
Subject: Verb: Itr
N Pl1 Pl1 Ind Prs Act Nad
we see
‘We (are able to) see.’

S+V+OA
a. Ne shohim studentët/ata.
Verb: Tr O: A
A Pl3 M
we see the students/them

S+pClA+V+OA
b. Ne i shohim studentët/ata.
pCl: A
Pl3
we them see the students/them
‘We see the students/them.’

The following example shows the doubling of dative (3, 3a, and 3b) and
accusative objects (3a and 3c).

Example 3. Object doubling in dative and accusative5

(3) S+pClD+V+OD+OA
Ne u dhamë studentëve/atyre librat/ato.
Subject: pCl: u Verb: Tr O: D O: A
N Pl1 D Pl3 Pl3 Ind Aor Act D Pl3 A Pl Det
Nad
we them gave the students/ the books/
them them
‘We gave the students/them the books/them.’

S+pClD+A+V+OD+OA
a. Ne ua dhamë studentëve/atyre librat/ato
pCl: u+i
D Pl3 +
A Pl3
we them + gave the students/ the books/
them them them
‘We gave the students/them the books/them.’
342 Besim Kabashi

S+pClD+A+V+OD
b. Ne ua dhamë studentëve/atyre
we them + gave the students/
them them
‘We gave the students/them them.’

S+pClD+A+V+OA
c. Ne ua dhamë librat/ato
we them + gave the
them books/them
‘We gave them the books/them.’

Sentence (3) has two objects, one in dative and one in accusative case, and
one pCl in dative case, which doubles the dative object. Whenever the verb
has a dative valency, the dative pCl cannot be left out, regardless of the
presence of a dative object in the sentence. In sentence (3a), both objects
are doubled by means of the respective pCls. Only one object is doubled by
the amalgamated pCl in sentences (3b) and (3c). Object doubling is also
possible in imperatives, e.g. Sillma librin! [‘Bring+me+it the book!’,
‘Bring me the book!’], i.e. doubling of accusative object.
When the verb has a first or second person accusative or dative object,
the pCls of the first and second person cannot be omitted, e.g. Studentët të
kuptojnë ty. or Studentët të kuptojnë. [‘The students understand you’]. The
form *Studentët kuptojnë ty. is ungrammatical.

1.3. Object elimination

If a pCl occurs, the corresponding object can be left out.6 This phenomenon
has been called Objektseliminierung [object elimination] by Buchholz
(1977), and Buchholz and Fiedler (1987). As there is no established Eng-
lish term, we will use the term object elimination as a translation of Ob-
jektseliminierung.7 Example sentences where the objects are left out are
(3b) and (3c). A sentence consisting only of pCl(s) and verb can be syntac-
tically well-formed, cf. sentences (4a) and (4b) in the following example,
where the subject is optional.

Example 4. Object elimination

(4) *S+V+OD+OA
*Ne _ dhamë studentëve/atyre librat/ato.
we gave the students/them the books/
them
*‘We gave _ the students/them the books/them.’
 Pronominal clitics and valency in Albanian 343

S+pClD+A+V
a. Ne ua dhamë.
we them + gave
them
‘We gave them to them.’

S+pClA+V
b. Ne i shohim
them see
‘We see them.’

The following patterns8 are possible for doubling and elimination of

accusative complements: S+V+OA (object without pCl), S+pClA+V+OA
(object doubling) and S+pClA+V (object elimination). For dative
complements, the following patterns apply: S+pClD+V+OD (object
doubling) and S+pClD+V (object elimination). As shown in the examples
above, these patterns can be combined with each other.

1.4. Pronominal clitics as valency fillers

Sentences (3b-c) and (4a-b) show the ability of pCls to function as valency
fillers, as they can replace objects. The grammatical information of the pCl
is sufficient to fill the valency of the verb; only the lexical content is miss-
ing, which has to be recoverable from the linguistic or extralinguistic con-
text. Thus, if pCls occur, objects can be left out without making the sen-
tence ungrammatical.
If objects are left out, the omission of pCls leads to the selection of a
different valency pattern. This can (but does not have to) indicate the use of
the verb in a different sense, cf. the difference of valency and meaning be-
tween sentences (2) and (2a), (2) and (2b), as well as in the following ex-
ample adapted from Buchholz, Fiedler, and Uhlisch (1993):

Example 5. Different patterns/meanings of the verb flas9

(5) S+V
Ai flet.
Subject: Verb: Itr
N Sg3 M Sg3 Ind Prs Act Nad
He speak
‘He speaks.’
344 Besim Kabashi

S+pClD+V+OD
a. Ai i flet atij.
pCl: Verb: Tr O: D
D Sg3 D Sg3 M
He him scold him
‘He scolds him.’

S+pClD+A+V+OD+OA
b. Ai ia flet atij një libër.
pCl: A Sg1 M Undet
D Sg3 +
A Sg3
He him + it promise him one book
‘He promises him a book.’

The role of pCls in the context of valency is different from the substitution
of objects by pronouns, despite the fact that the pCls are short forms of
personal pronouns. Personal pronouns, however, offer a more precise
description of case and number (first and second person plural) and gender
(third person singular) than pCls. First of all, the position of the pronoun is
the same as the position of the object if a pronoun substitutes the object. A
pCl has a fixed position in the verbal complex, regardless of the position of
the object it replaces. While pronouns always replace objects, pCls are
capable of either eliminating or doubling the objects they refer to. The
appearance of the dative pCl is obligatory while the substitution of the
dative object by pronouns is optional. The substitution of objects by
pronouns does not change the structure of the sentence (which means the
sentence matches to the same pattern), i.e. the OA is only realized by a
pronoun instead of a noun phrase, but the pattern S+V+OA remains the
same. If a pCl appears, however, the pattern is changed, e.g. from S+V+OA
to S+pClA+V+OA or to S+pClA+V. As shown above, pCls and pronouns
can (and in some cases must, cf. section 1.2.) occur together in a sentence.
As the dative pCl cannot be left out without making the sentence
ungrammatical, it always shares the dative valency slot with the dative
object if the object is present. The pCl fills the valency slot itself, if the
dative object is eliminated. In the case of the accusative, there is one more
option: the object alone fills the valency slot of the verb, the object and pCl
share the valency slot, or the pCl alone fills the valency slot. Thus in this
analysis, there is a minimum number of required complements of 1 and a
maximum number of required complements of 2 for the dative and the
accusative valency slot (by analogy to Herbst et al. [2004], where the
minimum and maximum valency of each verb is indicated); cf. the patterns
S+V+OA, S+pClA+V and S+pClD+V for minimum, and S+pClA+V+OA and
 Pronominal clitics and valency in Albanian 345

S+pClD+V+OD for maximum complements.

The fact that the dative pCl is obligatory even if the dative object is also
present can be accounted for using the concept of structural necessity
presented by Herbst (1999), which means that it is obligatory only on the
syntactic level, not as a lexical property. As pCls can be correctly predicted
by a set of rules, they are no idiosyncratic properties of lexical units and
thus need not be specified in the valency frame of a lexical unit. It is
enough to specify valency slots such as dative or accusative. Object
doubling is always optional on the syntactic level; here semantic factors
(needed for lexical content) as well as pragmatic factors (emphasis) decide
whether doubling takes place or not. Additional functions and properties of
pCls will not be discussed here.10
The properties of the pCl in the context of valency can be summarized
as follows: in case of object elimination, the pCls indicate the verb valency
and function as valency fillers, whereas in case of object doubling, both
pCls and objects share the valency slot of the verb.

2. Practical aspects: The computational model

The formalism used in the following computational model is Left Associa-

tive Grammar (LAG).11 LAG was developed as a formalism for the SLIM
(Surface compositional Linear Internal Matching) language theory in-
tended to model and reconstruct natural language communication on a
computer.

2.1. The formalism

The formalism works according to the principle of possible continuations

whereby every grammar rule concatenates the sentence start read so far
with the next word. The result of this concatenation becomes the new sen-
tence start to be concatenated with the next word again; this procedure is
repeated until the end of the input is reached. For example the input a b c d
is parsed in the steps a+b → ab; ab+c → abc; abc+d → abcd, summarized
as (((a+b)+c)+d) → abcd. Ambiguities are handled by tracing several
derivation paths in parallel. The most varied sentence types can be modeled
with linear effort regarding complexity and parse time, because only the
possible continuations matter in each LAG rule.
346 Besim Kabashi

2.2. An example

Below, an algorithm for the treatment of dependencies between verb, pCls

and objects is outlined using the sentence Ai na i dha librat. [‘He gave us
the books’].12 It starts with the first word form and the matching rule
Subject (start rule).

AiN Sg3 M

n RULE: Subject
LVF = <Subject>;
Follow RULE pCl_D;

The subject is added to the list of valency fillers (LVF). In case of an

omitted subject, the algorithm starts with the next rule, i.e. the parser
searches for a rule with a matching start pattern. Follow means
continuations, i.e. the next applicable rule, in this case pCl_D. When
reading a word form, information required for recognition of the verb
complex and for dealing with valency and congruency is provided by the
lexicon or a morphological analysis component.

Ai + na D Pl1

o RULE: pCl_D
LVF = <Subject, pCl_D_Object_elimination>;
Follow RULE pCl_A;

pCl_D is added to the LVF and labeled Object_elimination because the

dative object can be left out. Since at this point it is not clear yet whether
the actual pattern is elimination or doubling, the modification of the attri-
bute to pCl_D_Object_doubling can be done when a matching object is
encountered. If no matching object is found, the attribute is just left at the
value set here. Here, two non-amalgamated (concatenated) pCls are used to
demonstrate the canceling (filling) of valencies (valency slots). Amalga-
mated pCls have to be analyzed in the morphology component.

Ai na + i A Pl3

p RULE: pCl_A
LVF = <Subject, pCl_D_Object_elimination, pCl_A_Object_elimination>;
Follow RULE finVerb;
 Pronominal clitics and valency in Albanian 347

pCl_A is added to the LFV just like pCl_D in the previous rule, also with
the Object_elimination label.

Ai na i + dha Sg3 Aor Ind Act Nad <<D, A>, <A>>

q RULE: finVerb
LVF = <pCl_D_Object_elimination, pCl_A_Object_elimination>;
Follow RULE Accusative_object;

The verb is read in and checked for agreement with the subject. The subject
valency in LVF is canceled, the pCls, however, remain in the LVF because
they can share their valency slot with an object and thus cannot be canceled
before the respective object has been read or the sentence is finished. Be-
cause pCls always precede the verb, a (minimum) valency pattern that se-
lects one or more of the possible lexical readings can be already con-
structed at this point.

Ai na i dha + librat A Pl3

r RULE: Accusative_object
Replace pCl_A_Object_elimination by pCl_A_Object_doubling;
LVF = <pCl_D_Object_elimination, pCl_A_Object_doubling>;
Follow RULE Punctuation;

An accusative object is read in and modifies the pCl_A_Object_elimination

entry in the LFV to pCl_A_Object_doubling.

Ai na i dha librat + .

s RULE: Punctuation
LVF = <pCl_D_Object_elimination, pCl_A_Object_doubling>;
RESULT = pCl_D_Object_elimination, pCl_A_Object_doubling;

The end of the sentence has been reached, the result is in LVF: object
doubling for accusative and object elimination for dative.
Only the path that actually parses this sentence is shown above; other
possible pathes were left out. After various rule applications, several con-
tinuations (parallel paths) would be possible and would have to be tried out
by the algorithm. For example, after the rule finVerb the sentence may con-
tinue with punctuation, dative object, accusative object, a preposition, an
adjunct etc.
348 Besim Kabashi

As demonstrated in the example model, valencies can be canceled im-

mediately when a potential valency filler has been read. Another possibility
is end canceling, where the properties of all word forms read are collected
in attribute-value matrices and are not canceled until a punctuation mark
signals the end of the sentence. This has the advantage of transparency
when handling constructions such as subclauses with the verb in final posi-
tion, but one disadvantage is that ungrammatical constructions may not be
rejected before they have been completely parsed.
When reading amalgamated pCl word forms such as ma, ta, t’ia etc., it
is important to read and process the individual parts of the morphosyntactic
information, e.g. ma (mëD Sg1 + eA Sg3). In this way, an amalgam can replace
one object (object elimination) and double another at the same time, cf.
sentences (3b) and (3c).
The following figure shows an example of a morphological analysis of
the imperative one-word-sentence sillmani. As shown under the Clitic at-
tribute, the enclitic form ma has been recognized as an amalgam of më and
e.13

Figure 1. Morphological analysis of the word form sillmani from Kabashi (2003)

Here, valencies from the corresponding attribute are canceled with the
matching cases from the clitic attribute Declension. It is necessary to check
the information of lexical entries and morphological analysis to be able to
select or construct the correct valency pattern, particularly with regard to
the pCls that may be used with a verb. Information on possible valency
patterns of a verb comes from its lemma in the base-form lexicon.
 Pronominal clitics and valency in Albanian 349

Figure 2 shows the result of syntax analysis and valency handling of the
sentence Ai na i dha librat. treated above. The dative object is missing
there, cf. the FilledValencyFromObjects attribute. A dative object was ex-
pected according to the verb’s valency pattern but was not present. A dative
pCl was found and thus the sentence can be analyzed as well-formed, cf.
the FilledValencyFromClitics and the DativeSlot attributes. On the other
hand the fact that both an accusative pCl and an accusative object are
found, leads to object doubling, cf. the corresponding attributes, FilledVa-
lencyFromObjects and AccusativeSlot. The Index attribute indicates the
position of a word form in the analyzed sentence. The actual pattern is de-
rived from the attributes Clitic_D, Clitic_A, Verb, and Object_A. The
Meaning attribute contains the meaning of the verb in the currently selected
pattern.

Figure 2. Syntactical analysis of the sentence Ai na i dha librat.

350 Besim Kabashi

3. Conclusion

Acting as object substitutes (object elimination), pronominal clitics deter-

mine the verb’s valency and assume the role of valency fillers. They play
an important role in distinguishing between various possible valency
frames of a verb and thus between different meanings. In the case of object
doubling, they merely function as semantic and pragmatic markers for their
respective object.
PCls supply grammatical information that can be very useful both in
natural language communication as well as in natural language processing,
e.g. for processing discontinued sentences during turn-taking in dialogue
analysis.
In spite of the treated phenomenon’s complexity, an efficient implemen-
tation of a parser is possible using the LAG formalism. As shown in the
algorithm, canceling (filling) of verb valencies (valency slots) is easily
solved despite the multitude of possible combinations and the consequently
large number of sentence patterns.

Notes

1. For comments on the draft of this paper I would like to thank Jörg Kapfer,
Matthias Bethke, and Peter Uhrig (all Friedrich-Alexander-Universität Erlan-
gen-Nürnberg). Only the properties relevant to verb valency and computa-
tional modeling will be treated here. For further information on pronominal
clitics left out here, see Buchholz (1977), Buchholz and Fiedler (1987), Domi
(1995; 1997), Kallulli (1995), and Newmark, Hubbard, and Prifti (1982).
2. Here, the following abbreviations are used: A=Accusative, D=Dative, pClD+A=
Amalgam of pClD and pClA, Pl=Plural, Sg=Singular, and V=Verb. Forms like
silleni have alternatives in the form sillnie.
3. In subjunctive clauses, pCls are positioned after the subjunctive and future
particles and precede the finite verb. In this case the pCl can be combined with
these particles in one word, e.g. the amalgam t’i consisting of the subjunctive
particle të and the pCl i.
4. Act=Active, Ind=Indicative, Itr=Intransitive, M=Masculine, Nad=Not admira-
tive, N=Nominative, O=Object, Prs=Present, S=Subject, and Tr=Transitive.
5. Aor=Aorist (definite past), and Det=Determined.
6. An exception is the reflexive use of verbs, e.g. Ai e(pCl: A) lavdëron vetën(O: A,
Reflexive) [‘He (him) praised himself.’, i.e. ‘He praised himself.’], vs. Ai e
lavdëron. [‘He (him) praised.’, i.e. ‘He praised him.’].
7. This term is described in Buchholz and Fiedler (1987) as Vertretung des Ob-
jekts (which might be translated as object replacement or object substitution).
 Pronominal clitics and valency in Albanian 351

8. As we focus on the influence of pronominal clitics on verb valency, only di-

rect and indirect objects (without prepositions) are treated here.
9. Undet=Undetermined. Other patterns of the verb flas, e.g. pCl(s) + flas +
preposition, and other meanings not presented in Buchholz, Fiedler, and Uh-
lisch (1993) are not treated here. For more patterns/meanings cf. Kostallari
(1980), where the pattern/meaning from sentence (5b) is marked as a dialect
form. Cf. also Qesku (1999). Toçi (2002) does not list this pattern.
10. Buchholz and Fiedler (1987: 445–446) have described a group of verbs which
cooccur with „‚pleonastisch‘ verwendete Objektszeichen“ [“pleonastically
used” pronominal clitics]. These pCls are „nicht ... systematisch aus einem zur
Grundstruktur gehörenden dir[ekten] Obj[ekt] ableitbar“ [not derived in a sys-
tematic way from a direct object belonging to the base structure]. Thus sen-
tences in which only one dative pCl would be expected can have two pCls or
an amalgamated one consisting of a dative and an accusative pCl, like for ex-
ample Ia(pCl: D+A) hipi(V) kalit(O: D). [‘He mounted the horse’]; in this example
the subject is left out. These verbs cooccur only with specific lexical entries
(and so they must be marked in the lexicon to be considered during automatic
syntactic and semantic analysis). According to Buchholz and Fiedler (1987:
445), this group contains, among others, the following verbs: arrin, del, fillon,
hipën, hyn, kërcen, mbath, merr, nis, pëlcet, shtron, and thotë.
PCls can also occur as ethical datives, cf. Buchholz and Fiedler (1987:
447–448). In this case, the verb valency must be treated differently as well.
11. For the formal definition of LAG see Hausser (1992; 2001a).
12. The version illustrated here is only a simple LAG. Information about valency
in Database Semantics (DBS) can be found in the article Handling valency
and coordination in database semantics by R. Hausser in this volume. For in-
formation on DBS see Hausser (2001b).
13. During the syntactic analysis both pCls would fill the corresponding attribute
in the sentence structure while reading the verb.

References

Buchholz, Oda
1977 Zur Verdoppelung der Objekte im Albanischen. Linguistische Stu-
dien, Reihe A, Arbeitsberichte 34. Berlin: Akademie der Wissen-
schaften der DDR.
Buchholz, Oda, and Wilfried Fiedler
1987 Albanische Grammatik. Leipzig: VEB.
Buchholz, Oda, Wilfried Fiedler, and Gerda Uhlisch
1993 Wörterbuch Albanisch–Deutsch. München: Langenscheidt.
Domi, Mahir (ed.)
1995 Gramatika e gjuhës shqipe. Vëllimi I – Morfologjia [Grammar of the
Albanian Language. Vol. 1: Morphology]. Tiranë: Akademia e
Shkencave e Republikës së Shqipërisë.
352 Besim Kabashi

1997 Gramatika e gjuhës shqipe. Vëllimi II – Sintaksa [Grammar of the

Albanian Language. Vol. 2: Syntax]. Tiranë: Akademia e Shkencave
e Republikës së Shqipërisë.
Hausser, Roland
1992 Complexity in left-associative grammar. In Theoretical Computer
Science. 106 (2): 283–308.
2001a Foundations of Computational Linguistics. Human-Computer Com-
munication in Natural Language. 2d ed. Berlin/New York: Springer.
2001b Database semantics for natural language. In Artificial Intelligence
130: 27–74.
2007 Handling valency and coordination in database semantics. This vol-
ume.
Herbst, Thomas
1999 English valency structures – A first sketch. Erfurt Electronic Studies
in English (EESE).
http://webdoc.gwdg.de/edoc/ia/eese/artic99/herbst/6_99.html.
Herbst, Thomas, David Heath, Ian F. Roe, and Dieter Götz (eds.)
2004 A Valency Dictionary of English. A Corpus-Based Analysis of the
Complementation Patterns of English Verbs, Nouns and Adjectives.
Berlin/New York: Mouton de Gruyter.
Kabashi, Besim
2003 Automatische Wortformerkennung für das Albanische. Master’s
thesis, Computational Linguistics. Universität Erlangen-Nürnberg.
Kallulli, Dalina
1995 Clitics in Albanian. (Working Papers in Linguistics 24.) Trondheim:
University of Trondheim.
Kostallari, Androkli (ed.)
1980 Fjalor i gjuhës së sotme shqipe [Dictionary of Contemporary Alba-
nian Language]. Tiranë: Akademia e Shkencave e RPS të Shqipërisë.
Newmark, Leonard, Philipp Hubbard, and Peter Prifti
1982 Standard Albanian. A Reference Grammar for Students. Stanford:
Stanford University Press.
Qesku, Pavli
1999 Fjalor Shqip–Anglisht. Albanian–English Dictionary. Tiranë: EDFA.
Toçi, Fatmir (ed.)
2002 Fjalor i shqipes së sotme [Dictionary of Contemporary Albanian]. 2d
ed. Tiranë: Toena.
The practical use of valencies in the Erlangen
speech dialogue system CONALD

Günther Görz and Bernd Ludwig

1. Motivation

Within the broad spectrum of applications of computational linguistics the

conversational computer is often regarded as the ultimate challenge: “The
conversational computer paradigm provides a way to articulate the proper-
ties and challenges of natural language applications and the ways those
challenges are being addressed within the field of computational linguis-
tics (Cole et al. 1996)” (Resnik and Klavans 2003: 376).

2. Applications of language understanding

Linguistic computer applications do not just involve language per se, but
also interactions between linguistic knowledge and other areas of knowl-
edge. These interactions pose the main challenge for any more or less gen-
eral approach to natural language understanding: constructing the meaning
of a natural language phrase or sentence is guided by different principles
than constructing sentences for a formal (artificial) language used for the
representation of knowledge in a computing environment. There are two
main differences:

2.1. Natural versus formal languages

Formal languages do not encode implicitly semantic relations between

parts of a sentence. In contrast, in the natural language example

(1) I want to watch a thriller tonight.

grammatical markers (subject, object, and attribute/adverbial) are used to

establish relations between the described intention want to watch, the
354 Günther Görz and Bernd Ludwig

agent I who has this intention, the TV programme thriller, and the indica-
tion of a time span (tonight) during which the programme should be on air.
In formal languages, such relations are non-ambiguous because each
constituent of a sentence serves unambiguously as a functor or argument
of another constituent:

want-to-see (I, thriller, tonight)

This functor-argument-structure defines precisely and unambiguously

tonight as the time span when want to see should take place. This would
probably be the standard interpretation for the natural language sentence as
well.
However, depending on the content of the sentence, grammatical mark-
ers may be insufficient for avoiding ambiguities, as in the example:

(2) I want to watch the thriller at 8 pm.

In this sentence, there are no grammatical rules preventing the preposi-

tional phrase at 8 pm from being attached to the noun phrase the thriller.
As a consequence, it depends on the evaluation of both phrases which
reading is the correct one. If there was a thriller recorded the day before it
could be just the one the user wants to see, not necessarily a thriller broad-
cast when the utterance is made.

2.2. Context-dependent evaluation

The need to evaluate phrases in order to understand them suggests the

execution of algorithms in order to make the evaluation effective. Such
algorithms apply the information contained in phrases for input and com-
pute output that is, in its turn, used to determine all the possible interpreta-
tions of an utterance and to react appropriately. A consequence of this
approach is that the ability to understand an utterance are determined and
limited by the ability to process given information with the help of prob-
lem-specific algorithms.
As an example, consider an intelligent interface to a TV set that tries to
propose programmes to the user that match certain user-defined criteria. A
programme is described formally by a filled data structure such as the
following:
 The practical use of valencies in CONALD 355

<programmes>
<IsNowNext/>
<TransportStreamId>1073</TransportStreamId>
<OriginalNet-
workId_ServiceId>93787</OriginalNetworkId_ServiceId>
<EventId>43932</EventId>
<StartTime>2005-07-28 12:00:00</StartTime>
<Duration>00:30:00</Duration>
<RunningStatus>1</RunningStatus>
<Pil>238764127</Pil>
<Title>Eisenbahnromantik</Title>
<ShortInfo>Im Zug von Bratislava nach Ungarn</ShortInfo>
<LanguageCode>7693668</LanguageCode>
<ExtendedInfo>Der "Eisenbahnromantik"-Sonderzug fährt in
die schöne Slowakei und nach Ungarn. Gefahren wird mit
Diesel-, Dampf- und elektrischen Zügen. Der erste Teil der
Reise führt uns über Bratislava, Zvolen, Poprad-Tatry an
die ungarische Grenze.
</ExtendedInfo>
<ExtendedLanguageCode>7693668</ExtendedLanguageCode>
</programmes>

In order to evaluate the user query Are there any documentaries about
travelling now?, three different types of pragmatic evaluation have to be
distinguished:
− documentaries: look for programmes of this genre!
− about travelling: which programmes cover this topic?
− now: the programme should be being broadcast at the time of speaking.
Three different algorithms have to be employed for evaluating each of the
above types:
− data base lookup to find the right genre,
− analysis of the ExtendedInfo field to find the right content,
− temporal reasoning to find the right time interval.
This is a typical situation complex software systems have to deal with. As
a consequence, it is not practicable to rely on a single formal language to
cover all types of evaluations. On the other hand, how can one neverthe-
less implement modules for natural language analysis that can be config-
ured for different applications and therefore be used in contexts with to-
tally different and often unpredictable types of evaluations?
Dialogue systems, e.g. for assisting human users in performing practi-
cal tasks, are a typical example of the variety of scenarios and applications
just addressed: in cooperation with some technical application − e.g. a
database system providing information of some kind − railway timetable,
weather, stock market, theatre programmes, etc., a system to order mer-
chandise, a system to control devices − a goal expressed by the user, usu-
356 Günther Görz and Bernd Ludwig

ally in spoken language, has to be achieved, if possible. If not, the system

should be able to explain why and provide further help.
Aside from dialogue systems, there are of course many other applica-
tions of computational linguistics such as
− information retrieval (“the Google challenge”),
− automatic machine translation,
− automatic text summarization,
but also
− evaluating linguistic hypotheses, as far as they are fully formalized,
− simulating human language processing,
and many more.

3. Grammar and valencies in natural language understanding

The solution we propose is that any formal representation of natural lan-

guage within a natural language understanding system must resort to a
formal language that is able to represent explicitly what is entailed expli-
citly and implicitly in a natural language utterance. This language must be
sufficiently expressive to cover valencies, (generalized) quantifiers, modi-
fiers, modalities, and coordination operators that are not available in al-
most every formal (logical) language due to limitations in computability
and decidability.
In a second step towards understanding, a statement that represents the
semantics of a natural language sentence or utterance has to be translated −
by applying evaluation as in the example above, not just syntactical trans-
formations − into an expression of the formal language(s) the underlying
computational environment works with.
In this paper, we will focus the discussion on the use of valencies and
address only marginally the broader problem of dialogue analysis and gen-
eration. For a better understanding of the role of valencies in our work, we
first present the general framework and will then point out how valency
plays an essential role in its linguistic analysis component.

3.1. The Erlangen dialogue system CONALD

Our research on language understanding is embedded in work on dialogue

understanding. It aims at systems for rational dialogues based on a prag-
matic approach. So, our overall perspective is that of language as action in
which speech acts play a central role. As a consequence, the traditional
 The practical use of valencies in CONALD 357

“computational linguistics pipeline” − syntactic derivation, semantic reso-

lution and inference, and, eventually pragmatic analysis − is turned upside
down. Pragmatics gains control in that processing is controlled by a dia-
logue management module.
CONALD is a spoken language dialogue system which enables interac-
tion between users and a technical system in dynamic environments.1 The
primary goal of its design is to achieve quick configurability for various
applications. It combines deep syntactic and semantic analysis, discourse
processing, and language generation and features a complex semantics-
pragmatics interface in the sense of Brietzmann and Görz (1982). Seman-
tics is defined in terms of an extended version of discourse representation
theory (DRT). Discourse and application pragmatics are considered inde-
pendent; user utterances affect application pragmatics when their speech
acts are executed. They can be specified in a script language interpreted by
the dialogue manager.
In dynamic environments, changes can be consequences of user re-
quests as well as of external events in the application. The design of our
system allows configuration for a wide range of applications (like house-
hold applications, the automotive environment, medical purposes, etc.).
The user may give commands in natural language by speaking into a mi-
crophone.
The parser transforms the user’s utterance into a semantic representa-
tion, from which the dialogue system derives a goal to change the envi-
ronment. A plan to achieve this goal is computed and executed by a group
of agents. Under the assumption of a closed world, the overall system fea-
tures hierarchical planning, plan execution, and plan observation by sev-
eral agents with different responsibilities and capabilities. In this way,
system knowledge is distributed and organized hierarchically correspond-
ing to the tasks each agent has to execute.
Agents are organized in a hierarchy of layers. On top there is the Assis-
tance System (see figure 1) which plays the role of an interface between
the application and the dialogue system. The main task of the Assistance
System is to achieve the user’s goals by computing and executing high-
level plans.
Negotiation between system and users is handled by a dialogue man-
ager. For user utterances, input from a speech recognizer is processed by a
parser resulting in a representation of the meaning of the utterance.
358 Günther Görz and Bernd Ludwig

Figure 1. The CONALD system architecture

4. Incremental semantic composition

If we want human-computer dialogues to be natural, we must allow hu-

mans to talk to the computer as they do to humans. Spontaneous speech is
often incomplete or incorrect, full of interruptions and self-corrections,
leading to ungrammatical input to the parser. Additionally, given the error
rates of speech recognizers, even with correct input the speech recognizer
may produce an output which is not grammatical. Apart from that, parsing
German input is difficult, as German is a language with fairly free word
order, also allowing for discontinuous constituents. Therefore, the gram-
mar cannot rely only on linear input sequences as its main concept. We try
to overcome these problems by the design of a two-phase parsing process
(as presented in Bücher, Knorr, and Ludwig 2002).
First, the speech recognizer’s output is segmented into chunks (Abney
1991). These have to be translated into constraints for a (partial) descrip-
tion of a system state. For that purpose, an approach motivated by depend-
ency theory is applied: valencies for the syntactic head of each chunk are
analyzed if they can serve as the dependent for some other chunk (its re-
gent). Dependent and regent have to meet three classes of criteria in paral-
lel: syntactic constraints, semantic constraints (is the semantic part of the
valency satisfied?), pragmatic constraints (can a constraint in the applica-
tion domain be derived from the triple regent, thematic role, and depend-
 The practical use of valencies in CONALD 359

ent?). For the example utterance I want to watch a thriller tonight! the
parser computes the following analysis:

Chunk Semantics (informal)

I want to watch assistance task

a thriller object for task
tonight A time span for task

4.1. Applying case frames to chunks

The three chunks shown above are connected by semantic relations which
have to be identified during the second phase of the parsing process. It
relies on a kind of dependency grammar which for each chunk of phase 1
gives a list of possible syntactic functions the chunk may have:

C1 has C2 → <synfunc>
<constraint equation>

example: VP has PP → adverbial

NP has PP → attribute
VP has NP → subject
NP agr case = nom,
NP agr num = VP agr num.

The options are constrained by the morphological features of the chunk,

e.g. an NP-chunk functions as subject only if it is in the nominative case.
For each chunk there is a case frame for its semantic head that stores
information about the valencies.2 The valencies of each chunk are filled by
combining it with other chunks, e.g. building a VP from a verb and an NP
that functions as its direct object, or expanding a VP by an adverb. The
suitability of the combination of two chunks is determined by the semantic
constraints of the application ontology. Take the case frame for sehen:

infinitive: sehen

syntactic function thematic role EWN concept

subject involved-agent: Person1
adverbial involved-timespan: TimeInterval1
360 Günther Görz and Bernd Ludwig

From the case frame we derive hypotheses about possible fillers of a com-
plement position of a chunk using the syntactic functions. Whether a hy-
pothesis is satisfiable is determined by the concepts of the chunks. If they
fit, the DRS can be computed.
In our example, the VP want to see can be combined with the NP a
thriller and the adverbial tonight since in the case frame of sehen there are
valencies allowing semantic relations to be established.

5. Building a case frame database

We use our approach to semantics construction in different applications.

As a consequence, we have gathered a huge amount of semantic defini-
tions (i.e. taxonomic chains) and case frames (i.e. thematic roles) defined
by these applications. Some of these data are specific to a given applica-
tion, whereas others are used by several applications. This created the need
for a tool that enables efficient storage and easy and fast access and that
prepares the data required by the parser.

Figure 2. A screenshot of the valency editor

 The practical use of valencies in CONALD 361

For this purpose, we have developed a lexicon tool that permits the editing
of semantic data, checks their coherence according to the algorithm pre-
sented in section 4, and visualizes them as well (see figure 2). The tool
depends on the following resources as a basis for its data:
− the EuroWordNet (EWN) ontology,
− the SUMO ontology (a generic reference ontology), and
− semantic lexica.
In this respect, it is worth highlighting the differences between our frame
data base and FrameNet (Baker, Fillmore, and Lowe 1998). FrameNet is
an online lexical resource for English based on the principles of frame
semantics and supported by corpus evidence.3 It can serve as a dictionary,
for it includes definitions and grammatical functions of the entries. And
hence entries are linked to the semantic frames in which they participate,
FrameNet can serve as a thesaurus as well.
However, the information provided by FrameNet is not sufficiently
formalized to be directly applicable within our system; in other words, it is
not possible to use FrameNet to parse utterances directed to the system or
to construct semantic representations for them. So, from a practical point
of view, what we need is a formal specification for the information repre-
sented in FrameNet and which, on the one hand, can directly be encoded in
Description Logic (which is the logical framework we use), and on the
other hand, can be used with an efficient inference mechanism.
Another difference is that the current FrameNet is basically constructed
for the English language and hence can be used only in systems based on
English. Since our application is multilingual, our representation scheme is
based on the ILI-representation of EWN, which makes our tool language
independent.

6. Conclusions

To conclude, there are some points discussed in other contributions in this

volume which are important to our work:
− first of all, data-orientation in general, which by the way is common in
the speech recognition community, and in connection with that,
− the importance of storage of patterns in the lexicon, (which is, among
other properties, a prerequisite for what in Artificial Intelligence is
called case-based reasoning), and is perfectly compatible with the lexi-
calist character of our grammar,
− an emphasis on pragmatics and context, i.e., in our case, the conceptual
domain model, the discourse context maintained in the dialogue man-
362 Günther Görz and Bernd Ludwig

ager, and the situation context represented in the application system

providing the key to disambiguation.
Furthermore, the success of any application system depends crucially on
the availability of appropriate and comprehensive linguistic resources.
Although this sounds trivial, the actual situation we face − at least for Ger-
man − is not that easy.
It may improve within one or two years with the availability of a new
generation of resources like GlobalWordNet and German FrameNet. In the
long run, we would like to have access to a powerful valency lexicon data-
base − in particular considering that multilinguality is of increasing impor-
tance − for the technical aspects of which we can give some advice from
the viewpoint of computer science:
− As far as the formal representation of the lexicon is concerned, the ex-
pressivity of the language is of extreme importance. By that, we do not
want to emphasize encoding in some XML language − which is state of
the art − but rather its expressivity in logical terms, i.e. whether it can
express conjunction, disjunction, negation, quantification, subsumption
and inheritance, and whether it provides means to express non-
monotonic notions like defaults, a notion which lies beyond standard
first-order logic. And we must be aware that in a real, i.e. empirically
based lexicon, we will be confronted with inconsistency, and we will
have to deal with it.
− A minor, but nevertheless important issue is the interface, i.e., what is
the expressitivity of the query language? And, furthermore: is access
strictly sequential or is it possible to have parallel access?

Notes

1. Acknowledgment: Our work is supported by the Bavarian Research Associa-

tion FORSIP. We would like to thank the current and former members of our
research group for their contributions: Kerstin Bücher, Martin Klarner,
Yuliya Lierler, Peter Reiss, Bernhard Schiemann, and Iman Thabet.
2. The term case is used as it is by Fillmore (1968) meaning thematic roles. The
term valency is used here in a broader sense: it includes not only obligatory
elements needed to make a phrase syntactically complete; more than that, the
case frames list all semantically and pragmatically suitable modifications and
their syntactic representations, e.g. attributes for nouns or adverbials for verbs.
3. http://www.icsi.berkeley.edu/framenet/.
 The practical use of valencies in CONALD 363

References

Abney, Steven
1991 Parsing by chunks. In Principle-based Parsing, Robert Berwick,
Steven Abney, and Carol Tenny (eds.), 257–278. Kluwer:
Dordrecht.
Baker, Collin F., Charles J. Fillmore, and John B. Lowe
1998 The Berkeley FrameNet project. In Proceedings of the COLING-
ACL 1998, 86–90. Montreal.
Brietzmann, Astrid, and Günther Görz
1982 Pragmatics in speech understanding – revisited. In Proceedings of
the Ninth International Conference on Computational Linguistics,
Ján Horecký (ed.), 49–54. Amsterdam: North-Holland.
Bücher, Kerstin, Michael Knorr, and Bernd Ludwig
2002 Anything to clarify? Report your parsing ambiguities! In Proceed-
ings of the 15th European Conference on Artificial Intelligence,
Frank van Harmelen (ed.), 465–469. Lyon.
Cole, Ronald A., Joseph Mariani, Hans Uszkoreit, Annie Zaenen, and Victor Zue
(eds.)
1996 Survey of the State of the Art in Human Language Technology. New
York: Cambridge University Press.
Fillmore, Charles J.
1968 The case for case. In Universals in Linguistic Theory, Emmon Bach,
and Robert T. Harms (eds.), 1–88. New York: Holt, Rinehart, and
Winston.
Resnik, Philip, and Judith L. Klavans
2003 Applications of language technology. In International Encyclopedia
of Linguistics, 2d ed. Vol. 2. William J. Frawley (ed.), 376–377. Ox-
ford: Oxford University Press.
Valency data for Natural Language Processing:
What can the Valency Dictionary of English
provide?

Ulrich Heid

1. Introduction

1.1. Objectives of this paper

This paper addresses two questions: a specific one and a more general one.
The specific question could be phrased as follows: “A Valency Dictionary
of English (VDE, Herbst et al. 2004) is available in a printed form and in
the underlying electronic format. Could this dictionary be of use for Natural
Language Processing (NLP)?” In trying to give an answer to this specific
question, one is confronted with a second, more general issue: “What are
requirements with respect to a valency dictionary for NLP and which data,
which representation and which degree of detail are expected?”
We will try to briefly address both issues, starting with the more general
one. Our views on NLP valency dictionaries will be influenced by the
grammatical theory of Lexical Functional Grammar (LFG, Bresnan 1982a),
as will be the answer to the specific question: we report in fact about an
experiment in which an attempt was made to convert data from an early
version of the VDE (July 2004) into the form and format of LFG and to use
the result as a valency dictionary for an existing English LFG grammar.
This experiment was assessed by means of an automatic analysis of the
25,000 example sentences contained in the VDE.1

1.2. Valency data in Natural Language Processing

All current symbolic approaches to syntactic analysis in Natural Language

Processing (henceforth: NLP) rely on valency data; and in addition, many
hybrid systems, which combine symbolic and statistical processing, do so
in one way or another as well.
366 Ulrich Heid

Valency (often called subcategorization in NLP work) is understood,

here, as syntactic complementation, possibly related to predicate-argument
structures. Thus, a predicate (a verb, adjective or noun) is described with
respect to its capacity to have arguments, and in particular with respect to
the syntactic form by which these are realized in sentences. To describe the
behaviour of the verb [to] substitute, as used in they substituted bricks for
the expensive granite stone, the fact will be mentioned that a subject and
two complements show up in the syntactic environment of this verb.
NLP-oriented grammatical theories and the respective coding formal-
isms for lexical knowledge diverge considerably as to the details of valency
description. However, they are unanimous in suggesting that the lexical
description of predicates should go hand in hand with a grammatical rule
that allows a system to derive a (syntactic and/or semantic) representation
of a sentence which singles out the predicate and its arguments. These basic
assumptions are true, irrespective of whether the representation follows
rather a line of constituent structure2 or of predicate-argument structure
(such as LFG’s F-structures), or whether it is inspired by dependency struc-
tures (cf. TesniÂre 1959; Mel’čuk 1988).
A third component of valency-based implementations, besides lexical
entries and grammatical rules, is a constraint system, which checks that
only sentences are accepted (or generated) which contain all obligatory
arguments of the predicate in question, and only these.3 More details about
approaches to valency in NLP can be found in the article by Roland
Hausser in this volume.

2. Elements of an expectation horizon for a valency dictionary for

NLP

In this section, a few NLP-related requirements for the lexicographic de-

scription of valency properties of words will be formulated. Starting from
the current state of the art, we address needs with respect to the predicates
and to a few specific valency phenomena to be covered; we also discuss the
representation of valency data in the lexicon and preferences related to
certain valency phenomena.

2.1. Types of predicates

In line with Tesnière’s notion, the common understanding of linguists, lexi-

cographers and NLP resource developers is that certain types of predicates
 Valency data for natural language processing: What can the VDE provide? 367

are to be considered as valency-bearers. These include verbs, adjectives (as

in [1]) and (mostly derived) nouns (cf. [2]).

(1) He is proud of this exceptional success.

(2) John’s proposal to postpone the meeting was accepted.

For several languages, it has been observed that some (especially verbal)
predicates can be “multiword expressions”, i.e. composed of several lex-
emes. This is true of certain multiword idiomatic expressions such as Ger-
man in der Lage sein (‘be able to’), as in (3). Unless we consider in der
Lage sein as a multiword predicate, it is hard to see how to assign it or its
components a valency description. The expression requires a prepositional
phrase with zu (or a pronominal adverb, dazu) or an infinitival with zu, cf.
(4).

(3) Hans ist in der Lage die Aufgabe zu lösen.

Hans is in the position the task to do
‘Hans is able to do the task’
(4) Hans ist zu allem in der Lage.
Hans ist dazu in der Lage.
*Hans ist in der Lage.

Some multiwords which lexicography and linguistics would rather classify

as collocations (and not as idiomatic expressions) seem equally to have
subcategorized complements as a whole. German examples include in Er-
fahrung bringen (‘understand’), in Rechnung stellen (‘take into account’) or
zu Papier bringen (‘write down’), which all can have an object clause (in-
troduced by da² or by a wh-word), even though a sentential complement is
restricted with the noun Erfahrung (no wh-clause) and impossible with
Rechnung and Papier. If one counts the potential to take a subject clause
(again with da² or wh-) among the valency properties of a predicate, Ger-
man collocations like zum Ausdruck kommen (‘be expressed’), in Verges-
senheit geraten (‘fall into oblivion’), zum Vorschein kommen (‘appear’)
may be added to the inventory of multiword expressions with predicate
character. It seems that little is known, at least for German, but also to
some extent for English, about such phenomena; but we suggest for them
an NLP treatment as valency-bearing predicates, simply because they
would otherwise block automatic analysis.
368 Ulrich Heid

2.2. Levels of linguistic description in a valency dictionary

Valency phenomena are often seen as concerning the interface between

semantics and syntax. They involve argument structure (and thus the se-
mantic description of the predicates) and at the same time the syntactic
insertion of predicates into sentences. There is no agreement, however, as
to whether this implies that lexical description and representation have to
make explicit reference to both levels of description, syntax and semantics.
Some approaches would tend to predict syntactic properties from a seman-
tic description, others restrict themselves to a syntactic description, yet
others cover both levels explicitly.
Many NLP grammars and lexicons mainly use descriptive devices per-
taining to constituent structure. This is true of HPSG and of Tree-Adjoining
Grammar. Another example is work on French in the framework of lexicon
grammar (cf. Gross 1975), which uses abbreviations for noun phrases (N)
and indexes as well as positional coding to indicate grammatical functions,
as illustrated in (5):

(5) a. N0 donner N1 à N2
b. Jean donne la clef à son amie.
‘Jean gives the key to his girlfriend.’

Lexical Functional Grammar (LFG, Bresnan 1982a) is characterized by the

use of two interrelated representations of sentences, one at the level of con-
stituency (C-structure) and one at the level of predicate-argument structures
(F-structures). LFG’s lexical description makes reference to F-Structures:
valency descriptions in lexical entries involve grammatical functions (sub-
ject, object, indirect object, etc.) and a notation which ensures a mapping to
logic-like predicate-argument structures (cf. Zaenen 1988). In the grammar,
a mapping between grammatical functions and grammatical categories is
defined. An LFG entry for donner equivalent to (5) above, is given in (6)
below:4

(6) donner, V:
(↑ PRED) = “donner<(↑ SUBJ)(↑ OBJ)(↑ INDIRECT-OBJ)>”

One of the most detailed and most explicit recent approaches to the lexical
description of argument structure and valency is that of Frame Semantics
(Baker, Fillmore, and Cronin 2003; Petruck 1996). It involves descriptive
devices from three layers, namely constituents (NP, PP, etc.), grammatical
functions (Ext[ernal argument], Obj[ect], Dep[endent]) and semantic roles
 Valency data for natural language processing: What can the VDE provide? 369

(‘frame elements’ in Frame Semantics). In (7), we show an extract from

FrameNet’s valency tables5, for a reading of the English verb [to] substi-
tute, as exemplified in (8).

(7) Agent New Old

CNI NP PP[for]
-- Ext Dep
NP NP PP[for]
Ext Obj Dep

(8) a. A young lady now substitutes for our former head of de-
partment.
b. They substituted bricks for the expensive granite stone.

FrameNet’s valency representation is attractive for NLP dictionaries, be-

cause it is rather explicit, remains close to observable phenomena and does
not assume knowledge about correspondences between the three layers,
which would otherwise need to be coded outside the dictionary.6
Because of these properties, FrameNet valency tables can also be used
for the encoding of exceptions, of valency variation (e.g. that-clause vs.
infinitive vs. noun phrase), for the comparison of near synonyms and mor-
phologically related items (which would share frame elements, cf. substi-
tute vs. replace, read vs. readable), or for relating single verb entries and
entries for multiword predicates which are semantically close to each other.
(9) is an example of the latter (cf. also Ruppenhofer, Baker, and Fillmore
2004). In our view, the three-layered representation is particularly flexible
and particularly interesting for NLP.

(9) a. John proposed to postpone the meeting.

b. John made the proposal to postpone the meeting.
c. predicates↓ / roles→ SPEAKER MESSAGE
propose SUBJ / NP COMP / to-inf
make + proposal SUBJ / NP COMP / to-inf

2.3. Phenomena to be captured in an electronic valency dictionary

Above, in section 2.1, the needs of NLP in terms of predicate types were
addressed; we now turn to complementation types and to the coocurrence
of complementation types with other lexical properties.
370 Ulrich Heid

2.3.1. Complementation phenomena

There is a huge literature on complementation facts to be covered by

valency dictionaries, and we do not aim, in this section, to repeat the ele-
ments of the “standard” inventory. We only wish to point to a small number
of facts which sometimes are not covered by dictionaries, and which merit
treatment. We understand the following list as a proposal for additions to
the usual typologies.
LFG’s complementation theory includes a theory of control (cf. Bresnan
1982b) which predicts the subject of dependent infinitives. Some predicates
seem not to fall under any predictable generalizations, and thus LFG dic-
tionary entries typically include statements about the subject of the infini-
tive (cf. promise vs. command vs. propose). This proves vital for syntactic
and semantic analysis.
Other phenomena which are sometimes not treated in sufficient detail
are subcategorized adverbials (VDE has she acted strangely), subject and
object predicatives and details on NP complements which are not objects.
The latter include for example measure phrases (the sessions lasted six
hours), complements of verbs like German enthalten, erhalten, verlieren
(‘contain’, ‘get’, ‘lose’) and idiosyncratic cases like that of the German
verb halten (von jmdm/etwas ‘to have an opinion about sb/sth’), which
requires a quantifying noun phrase as a complement, but a lexically very
restricted one (er h¼lt viel/wenig/nichts/eine Menge von ihr, ‘he thinks a lot
of her / does not think much of her’).
From an NLP perspective, it would be useful if valency dictionaries
could address such phenomena.

2.3.2. Contextual preferences related to valency

We know rather little, so far, about contextual preferences related to

valency properties of predicates, i.e. about conditions under which certain,
possibly exchangeable, valency patterns are particulary likely. These in-
clude the interaction with collocations, i.e. the fact that alternative valency
constructions of a given verb are unevenly distributed across certain nouns
which appear frequently as complements of this verb. Klotz (2000) has
numerous examples of this phenomenon, and in table 1 we summarize a
few of his findings (Klotz 2000: 178−179) for the verb [to] ask:
 Valency data for natural language processing: What can the VDE provide? 371

Table 1. Distribution of alternative valency structures of ask across certain nouns

Pattern favour permission asylum

ask sbdy sth + ?? ??
ask sbdy for sth + + +
ask sth from sbdy ?? + ??
ask sth of sbdy + ?? ??

On the basis of work towards VDE, Klotz has shown the use of corpus data
for the identification of such preferences. According to this data,7 many
examples show up for the combinations marked with “+”, and very few or
none for those marked with “??”. This knowledge is central for language
generation and equally useful to guide expectations in the analysis of tex-
tual data.
Other preferences concern the cooccurrence of certain subcategorization
patterns with certain tense and/or modalization forms (e.g. the possibility or
not of dependent wh-interrogatives) or the position of German subject
clauses with adjectives (cf. Heid and Kermes 2002). In fact, a subject
clause of a German adjective may either be topicalized, as in (10a) or ex-
traposed (10b):

(10) a. Daß er kommt ist klar (‘that he is coming is clear’)

b. Es ist klar, daß er kommt (‘it is clear that he is coming)
(11) Ob er kommt, ist unsicher/unklar/zweifelhaft
(‘whether he is coming is unclear …’)

Across large amounts of German newspaper text, extraposed cases come

out as more frequent than topicalized ones, if counted over all adjectives.
However, certain classes of adjectives (e.g. unklar, unsicher, zweifelhaft,
etc., ‘unclear’, ‘unsure’, ‘doubtful’) show clear preferences for topicaliza-
tion of their subject indirect question clauses (cf. 11). This seems to have to
do with their semantics, or it is explainable through the relationship be-
tween their semantics and information structure under topicalization; but
for language learners (and obviously for NLP programs such as text gen-
eration), an explicit indication in the dictionary would be preferable.
372 Ulrich Heid

2.4. Regularity and idiomaticity in valency descriptions

Some of the phenomena discussed in this section could be classified as

subregularities, or as effects of idiomatization. Some of them are preferen-
tial in nature, rather than absolute. And they constitute contextual parame-
ters of the valency description, which may be identified in large text cor-
pora, at least as tendencies.
To our knowledge, these phenomena have not yet been studied in much
detail, nor in large quantities, which is why dictionaries only sporadically
mention them.
For NLP applications, and specifically for the analysis of large amounts
of text and for natural language generation, it would however be desirable
if valency dictionaries covered the phenomena we touched upon in this
section. More generally, it seems that a large part of the regularity in
valency behaviour of lexical items has been analyzed and described, at least
for the major languages. But the domain of idiomaticity, or of local
subregularities, needs to be explored in more depth and breadth, because it
closely interacts with valency. In this sense, part of the present expectation
horizon may seem ambitious, or it may be rather a work programme for the
future.8

3. Comparing VDE and an English LFG grammar and lexicon

In this section, we will address the question whether the new Valency Dic-
tionary of English (Herbst et al. 2004, VDE) could be used in Natural Lan-
guage Processing. We use the expectation horizon discussed in section 2 in
order to assess this, first in a broad quantitative form, then in qualitative
terms. This assessment is the result of work by Spohr (2004), to which we
add a few comments motivated by later work on an internal version of the
VDE. The VDE itself is presented and discussed in articles by Herbst,
Götz, Götz-Votteler and Klotz in this volume.

3.1. An experiment into dictionary reuse

The authors of the VDE kindly provided us, in spring 2004, with a prefinal
marked-up WORD version of the full text of the VDE, along with explana-
tions of its macro- and microstructure, its abbreviatory conventions etc.9
Dennis Spohr then embarked on an experiment the objective of which was
to feed an NLP grammar of English with data from the VDE. We used an
 Valency data for natural language processing: What can the VDE provide? 373

LFG grammar of English from PARC (Palo Alto Research Centre, version
of 21.07.2004), designed using the Xerox Linguistic Environment, XLE, a
platform and tool for parsing and generation developed from the LFG
Grammar Writer’s Workbench (cf. Maxwell and Kaplan 1996).10
The grammar and its lexicon are part of the multilingual Pargram pro-
ject, the aim of which is to provide broad coverage NLP grammars and
lexicons for several languages.11 The PARC grammar of English comes
with its own lexicon of verbal subcategorization. We wanted to check how
the grammar performed, if we substituted the VDE for the existing PARC
lexicon. To assess the results, automatic syntactic analyses of the example
sentences of the VDE were produced. VDE contains over 25,000 example
sentences, and we tested the performance of the grammar on this corpus if
using the VDE lexicon, compared to the grammar with its existing lexicon.
This test would allow us to use a substantial amount of VDE data in syntac-
tic analysis, and to thoroughly assess the VDE against an NLP-oriented
expectation horizon. A similar experiment has been reported on, for Dan-
ish, by Asmussen and «rsnes (2005).
In order to be combinable with the NLP grammar, the marked-up
WORD file of VDE was first metalexicographically analyzed and trans-
formed into an XML representation. From there, a semi-automatic mapping
into LFG-style lexical entries (cf. example [6] above) was performed (in
fact into instances of templates which encode subcategorization patterns).
The mapping from the printed-style dictionary to an XML version was not
a trivial one, as the VDE makes use, among other things, of lexicographic
text condensation devices: for example, in the notation of alternative
valency patterns, the scope of the alternation symbol is not always auto-
matically derivable; the string “about wh-CL/wh to-INF” needs to be trans-
lated into “about wh-CL OR about wh to-INF”: in this example the “OR”-
symbol groups a single-string expression (wh-CL) and a two-string expres-
sion (wh to-INF), which makes an automatic expansion less easy. Simi-
larly, open lists of lexical constraints (e.g. a complement lexicalizable as
something/little/what etc.) or approximative preference statements (usually,
normally, often) cause difficulties in the mapping.
A by-product of this two-step procedure is a version of the (prefinal)
VDE represented in XML, according to the CONCEDE DTD12 (cf. Spohr
2004: 53−55). A major innovation of this XML version of the VDE over
the original is the fact that it contains explicit links between VDE’s descrip-
tive indications (valency indications, verb senses, etc.) and its example
sentences. Being able to relate example sentences with readings, and spe-
cific lexicographic indications with examples that illustrate them, contrib-
utes to a clearer addressing structure and to a richer microstructure. In an
374 Ulrich Heid

electronic dictionary, ideally all indications should be illustrated with ex-

ample sentences. VDE has gone a long way along these lines, and the XML
version completes VDE’s intentions also on formal grounds, linking expli-
citly most of the 25.000 examples.13
More generally, for computational linguistics and knowledge represen-
tation (or knowledge processing), one would dream of many more links
between the descriptive elements of the VDE entries. Patterns, subcategori-
zation types, senses, semantic roles: all of these could ideally be part of a
two-layered network-like structure. One layer would serve to describe and
to (formally) constrain the descriptive vocabulary (the “attributes” and
“value names” used by the lexicographers, and their interrelationships).
The other layer would contain the actual entries, their links and relations
among each other, and their links with the definitional layer.
We have worked out a prototype of such a structure for a collocations
dictionary (cf. Heid and Gouws 2006; Spohr and Heid 2006) and it allows
for interesting new kinds of queries, including ad-hoc queries not based on
a particular entry (e.g. “give me all entries having a combination of proper-
ties x and y”). We think that a major task for both NLP and human-use
lexicography of the coming years is to turn list-like dictionaries into net-
work-like ones, by adding explicit relational data. VDE seems to have the
ingredients to undergo such a transformation, even though this would still
require some additional work.

3.2. Experimental results: Using the VDE for parsing

As mentioned above, the 25,000 example sentences of VDE were used as a

test corpus, parsed with the existing PARC lexicon and grammar of Eng-
lish, and then with the VDE and an appropriately adapted grammar. The
overall quantitative result shows a positive effect of VDE, but it does not
look extraordinary at first sight. If the old PARC grammar produced at least
one analysis for 78.5% of the sentences, the new grammar-cum-VDE goes
up to 83.5% (Spohr 2004: 42): an increase in coverage by 5%. What is
counted here is the number of sentences which receive at least one analysis
(likely several analyses). This figure is indeed correlated with the quality of
the subcategorization dictionary, because sentences for which there is no
subcategorization entry available do not get any complete analysis, but at
most a partial one.
A second, perhaps even more relevant figure concerns those sentences
which got no analysis with the PARC grammar (over 4,500 sentences). An
improvement through the use of VDE is visible here, since 27.59% of these
 Valency data for natural language processing: What can the VDE provide? 375

do in fact receive an analysis if the VDE dictionary data are used.

Reasons why still a substantial portion of VDE’s example sentences do
not receive an analysis include, among others, the properties of these ex-
ample sentences themselves: in fact, many of them are quite long (40 words
not being exceptional) and rather complex; most of them have been taken
from the Bank of English corpus, and they display the typical complexity of
published written English. Other sentences are in fact dialogues: the gram-
mar does not assign one common analysis to a sequence of turns but one
per turn, while still counting, for the purpose of the quality assessment, the
dialogue as one (failed) analysis task. In this sense, the above figures can
be seen as the lower bound of the outcome of our experiment.
In conclusion, we have shown that VDE could indeed serve as a valency
dictionary for a formal grammar, as its use leads to an acceptable syntactic
analysis coverage of its own example sentences.14 VDE is rich enough to
provide good quality analyses, and it is richer than one of the most detailed
NLP grammars and lexicons of English available, PARC’s English LFG
grammar and lexicon.

3.3. Learning from the experiment: A phenomenological comparison be-

tween VDE and Lexical Functional Grammar

Coming back to the expectation horizon discussed in section 2, we will

now comment on parallelisms and differences between VDE and LFG. We
do so by addressing in turn the treatment units covered, the levels of de-
scription available and the valency phenomena and preferences addressed
in both resources.
Treatment units. The VDE covers nouns, verbs and adjectives; NLP
systems need valency information for all three of these word classes, and
only the most recent printed valency dictionaries, such as VDE or VALBU
(Schumacher et al. 2004), go beyond verbs. VDE is much more detailed
and complete than the version of LFG we used, as far as idioms and collo-
cations and their valency properties are concerned. It provides a substantial
number of entries for such items; the current LFG grammar, on the other
hand, does not contain rules to analyze all of them. This problem is both a
theoretical and a practical one, as the grammatical analysis of idioms is still
a difficult task for NLP systems (cf. Sag et al. 2002). Thus, not all of
VDE’s entries for multiwords could be transformed into LFG.
376 Ulrich Heid

Levels of description. VDE and LFG diverge as far as the level of linguistic
description is concerned at which valency patterns or subcategorization
frames are formulated: VDE uses grammatical categories (NP, AP, PP,
etc.) and LFG uses grammatical functions. Consequently, the mapping
from VDE to LFG was not an automatic one, as it required linguistic
knowledge not explicit in VDE. More precisely, VDE is in some places
underspecified with respect to the requirements of LFG, while at the same
time being more specific on a few particular phenomena.
A lack of specificity is encountered, for example, with respect to LFG’s
treatment of complements of three-place verbs. While LFG distinguishes
the grammatical functions “Object” and “Object2” (or: direct vs. indirect
object), our (preliminary) version of VDE had a distinction between pas-
sivizable and non-passivizable NP complements, which is good for distin-
guishing “Objects” from other NP complements, but does not help, for
example, to distinguish the two complements of give which are both pas-
sivizable. Similarly, non-passivizable NPs needed to be further split into
predicative ones (he died a rich man) as opposed to nominal obliques (to
ski Tahoe; the balloon lost air, etc.).
On the other hand, distinctions from VDE in the domain of sentential
complements would merit being expressed in LFG grammars, too: for ex-
ample, VDE distinguishes between that-complements, wh-complements
and direct speech. English verbs seem (as German ones) to have specific
requirements as to which of these they allow. This distinction is relevant in
sentence generation and in (machine) translation; current LFG grammars
tend to map all these complements to a grammatical function
“COMP(lement)” (or to a sentential “OBJect”), which is underspecified
with respect to constituency. The same holds for finite vs. infinite indirect
questions (they discussed who should do it vs. they discussed about how to
do it). Finally, the PARC grammar did not capture, before our experiments,
passivizable prepositional objects (freedom was marched for).
In general, taking into account both grammatical functions and gram-
matical categories seems to be an ideal way of modeling valency patterns
for NLP, in terms of detail and precision. However, such an approach
would lead also to redundancy wherever it explicitly states twice what
could be predicted from one level of description and from general princi-
ples.
VDE includes a semantic description in terms of semantic roles, at least
for a considerable number of items. A general use of this device in all en-
tries, in addition to grammatical categories and grammatical functions,
would bring the valency description close to that of FrameNet (Baker,
Fillmore, and Cronin 2003). The advantages of such a three-way descrip-
 Valency data for natural language processing: What can the VDE provide? 377

tion have been discussed above in section 2.2.

Valency phenomena covered. Above, in section 2.3, we mentioned a few
facts of complementation which need to be described in an NLP dictionary.
VDE has an excellent treatment of subcategorized adverbials (she acted
strangely etc.) and of passivizability (cf. freedom was marched for). In both
cases, VDE descriptions are much more fine-grained than the analyses
available in the original LFG grammar and lexicon. Consequently, the
grammar was enhanced to capture these phenomena.
A similar case was observed with verbs taking complex (i.e. multiword)
prepositions; the authors of VDE consider, for example, the preposititional
phrase in favour of X as subcategorized by the verb [to] argue, whereas the
assumption in the original LFG grammar was that multiword prepositions
would never be subcategorized. Here, clearly VDE throws up phenomena
which have been overlooked in the PARC resources.
In the domain of control, VDE is less fine-grained than expected by
LFG. In fact, the distinction between subject control and non-subject con-
trol in three-place verbs (persuade vs. promise) is not present in the version
of VDE we analyzed.
Lexical and contextual constraints on valency. Being based on corpus
evidence, VDE contains a substantial amount of data showing preferences
or constraints attached to valency properties of predicates. In our experi-
ments, quantification phrases (with predicates of duration, weight, length
etc.) provided an example of the richness of such constraints, but also of
the difficulty of capturing them in a formal way. VDE uses a subclassifica-
tion of these measure phrases which is far more fine-grained than that of
LFG, distinguishing, for example, quantifying NPs (of the kind [to] weigh
50 lbs, [to] last 30 years) from NPs without explicit quantification ([to] last
a lifetime). For many predicates, the full range of possibilities is not fully
explored. We think that further semi-automatic data extraction work, on the
basis of very large corpora, is needed here to complete the picture.

4. Making a good thing even better

The experiment described in the previous section clearly shows that the
VDE could indeed serve as a starting point for a detailed large-scale NLP
dictionary of English valency.
Even if we could not document, in this paper, a comparison with COM-
LEX,15 it seems that the entries of VDE contain more detail than those of
COMLEX: multiword predicates, subcategorized adverbials, subcatego-
rized quantifying phrases, to name but a few. VDE shares some properties
378 Ulrich Heid

with FrameNet (e.g. the use of semantic roles in some entries), but it differs
from FrameNet in that it does not address the semantic aspects of valency
in the same detail and coverage, and – obviously – in terms of the coverage
of the vocabulary: as FrameNet is being created according to a framewise
procedure, some high frequency items may be absent from FrameNet
which are dealt with in the VDE.
In addition, we see VDE as a source of lexical information which lends
itself easily to a systematic exploration of valency-related phenomena, such
as variation in valency patterns, interrelationships between collocations and
valency, etc. A fully formalized representation of VDE, perhaps one which
would take Spohr’s (2004) DTD or a similar modeling as a starting point,
would allow to turn VDE into a network-like data source which could be
explored in many and quite flexible ways.
Finally, to further improve on the side of contextual preferences, one
could imagine using the VDE as a starting point for specialized corpus
exploration. To this end, substantial amounts of illustrative examples for a
given valency pattern would need to be collected and grouped. For exam-
ple, the actual form and distribution of quantifying complements (cf. [to]
last 50 years, [to] last a lifetime) could in this way be approximated, or
subcategorized adverbials could be listed. Obviously, this would not allevi-
ate the problem of how to generalize from the observed corpus data, but the
mere fact of having quantifiable data on such phenomena available would
already be an advantage. In a similar way, one could envisage providing
corpus frequency data for the different valency patterns a predicate can
have. Approximative data of this kind (in fact probabilities) have been pro-
vided, for example, by Schulte im Walde (2002), and a more fine-grained
version could be the result of VDE-based corpus analysis.
Procedures to extract examples of valency patterns from text have been
discussed in the NLP community for almost 20 years (see the overview by
Schulte im Walde forthcoming), mostly with the objective to automatically
learn verb subcategorization. Having VDE as a starting point, details about
preferences, about frequency distributions etc. seem to be within reach.
Instead of having to identify both at a time, valency patterns and the lexical
and contextual properties associated with them, one could concentrate on
the latter and include such contextual data, for example, in an electronic
version of VDE.
In conclusion, VDE shows considerable affinity with NLP, even though
it was not conceived with the use by automatic systems in mind. But the
presence of a clear descriptive programme, its richness in details and its
reproducible internal structure contribute to its multifunctionality.
 Valency data for natural language processing: What can the VDE provide? 379

Notes

1. The experiments were carried out by Dennis Spohr (cf. Spohr 2004), and the
author should like to thank him in particular for making the results of his work
available. All errors and inconsistencies in the present article are of the re-
sponsibility of the author.
2. As for example in Head-driven Phrase Structure Grammar (HPSG, cf. Pollard
and Sag 1994), Tree-Adjoining Grammar (cf. Joshi 1985) or in C-structures of
Lexical Functional Grammar (LFG).
3. Obviously, this leaves space for adjuncts (or: modifiers), but it raises the issue
of handling optional arguments; typically, optionality gives rise to the disjunc-
tive formulation of two or more (sub-)entries.
4. Uparrows are variables for the insertion of the entry into a grammar.
5. The top line of the table contains names of frame elements; each valency indi-
cation occupies two lines, one where phrase types are indicated and a func-
tional one. The abbreviation CNI stands for ‘contextual null instantiation’.
6. On the contrary, Spohr et al. (2007) have shown that valency descriptions can
be derived from corpus data annotated with Frame Semantics roles, grammati-
cal functions and phrase types: the interrelationships between the three layers
then come out as preferred cooccurrences between facts from these layers, and
variation can be captured without any extra effort.
7. Material from the Bank of English, cf. (as of 22.02.2007):
http://www.cobuild.collins.co.uk
8. Part of the author’s own ongoing work is devoted, among others, to the devel-
opment of methods for identifying contextual parameters along with the ex-
traction of data from corpus text; cf. the project B3 in the framework of the
DFG-funded Special Research Centre SFB-732, URL (02.03.2007):
http://www.uni-stuttgart.de/linguistik/sfb732/
9. We should like to thank Thomas Herbst and Dieter Götz, as well as all of their
team for making the VDE available to us in a prefinal WORD version. Only
this cooperation allowed us to assess the data in detail.
10. XLE and the grammar are property of Xerox and PARC; IMS Stuttgart has
used both under academic usage licenses.
11. The languages include English, German, French, Norwegian, Korean, Japa-
nese, Urdu, which all use parallel LFG-based methodology (cf. the URL (as of
22.02.2007): http://www2.parc.com/istl/groups/nltt/pargram/).
12. The CONCEDE project (Erjavec et al. 2003) provides a DTD into which the
project transformed monolingual definition dictionaries. Spohr’s version of the
VDE DTD is extended with respect to the standard CONCEDE DTD.
13. An implicit link is obviously provided in the text version of VDE anyway, the
XML version just makes it explicit.
14. One could have made more tests, e.g. with parts of the BNC. As this would
have required filtering BNC data (to select relevant sentences), such a test
would have gone beyond the experimental setup used here.
15. http://nlp.cs.nyu.edu/comlex/index.html
380 Ulrich Heid

References

Asmussen, Jørg, and Bjarne Ørsnes

2005 Valency information for dictionaries and NLP lexicons: “Adapting
valency frames from The Danish Dictionary to an LFG lexicon”. In
Papers in Computational Lexicography COMPLEX 2005, Ferenc
Kiefer, Gábor Kiss, and Júlia Pajzs (eds.), 28–39. Budapest: Linguis-
tic Institute, Hungarian Academy of Sciences.
Baker, Collin F., Charles J. Fillmore, and Beau Cronin
2003 The structure of the FrameNet Database. International Journal of
Lexicography 16: 281–296.
Bresnan, Joan (ed.)
1982a The Mental Representation of Grammatical Relations. Cambridge,
MA.: The MIT Press.
Bresnan, Joan
1982b Control and complementation. In The Mental Representation of
Grammatical Relation, Joan Bresnan (ed.), 282–390. Cambridge,
MA.: The MIT Press.
Erjavec, Tomaz, Roger Evans, Nancy Ide, and Adam Kilgarriff
2000 The CONCEDE model for lexical databases. ITRI Technical Report
Series, ITRI-00-22. Brighton: University of Brighton.
Götz, Dieter
2007 Valency and automatic syntactic and semantic analysis. This volume.
Götz-Votteler, Katrin
2007 Describing semantic valency. This volume.
Gross, Maurice
1975 Méthodes en Syntaxe. Paris: Hermann.
Hausser, Roland
2007 Handling valency and coordination in database semantics. This vol-
ume.
Heid, Ulrich, and Rufus H. Gouws
2006 A model for a multifunctional electronic dictionary of collocations.
In Proceedings of the 12th Euralex International Congress, Elisa
Corino, Carla Marello, and Cristina Onesti (eds.), 979–988. Alessan-
dria: Edizioni dell’Orso.
Heid, Ulrich, and Hannah Kermes
2002 Providing lexicographers with corpus evidence for fine-grained syn-
tactic description: Adjectives taking subject and complement clauses.
In Proceedings of the 10th EURALEX International Congress, Anna
Braasch, and Claus Povlsen (eds.), 119–128. København: CST/KU.
Herbst, Thomas
2007 Valency complements or valency patterns? This volume.
Herbst, Thomas, David Heath, Ian Roe, and Dieter Götz
2004 A Valency Dictionary of English. Berlin/New York: Mouton de
Gruyter.
 Valency data for natural language processing: What can the VDE provide? 381

Joshi, Aravind
1985 Tree adjoining grammars. How much context-sensitivity is required
to provide reasonable structural descriptions? In Natural Language
Parsing: Psychological, Computational and Theoretical Perspec-
tives, David Dowty, Lauri Karttunen, and Arnold Zwicky (eds.),
206–250. Cambridge: Cambridge University Press.
Klotz, Michael
2000 Grammatik und Lexik. Studien zur Syntagmatik englischer Verben.
Tübingen: Stauffenburg.
2007 Valency rules? The case of verbs with propositional complements.
This volume.
Maxwell, John T., and Ronald Kaplan
1993 The interface between phrasal and functional constraints. Computa-
tional Linguistics, 19 (4): 571–590.
Mel’čuk, Igor A.
1988 Dependency Syntax: Theory and Practice. Albany, N.Y.: SUNY
Press.
Petruck, Miriam R. L.
1996 Frame semantics. In Handbook of Pragmatics, Jef Verschueren, Jan-
Ola Östman, Jan Blommaert, and Chris Bulcaen (eds.), 1–13. Am-
sterdam: Benjamins.
Pollard, Carl, and Ivan A. Sag
1994 Head-Driven Phrase Structure Grammar. Chicago: University of
Chicago Press.
Ruppenhofer, Josef, Collin F. Baker, and Charles J. Fillmore
2002 Collocational information in the FrameNet Database. In Proceedings
of the 10th EURALEX International Congress, Anna Braasch, and
Claus Povlsen (eds.), 359–370. København: CST/KU.
Sag, Ivan A., Timothy Baldwin, Francis Bond, Ann Copestake, and Dan Flickinger
2002 Multiword expressions: A pain in the neck for NLP. In Computa-
tional Linguistics and Intelligent Text Processing: Third Interna-
tional Conference: CICLing-2002, Alexander Gelbukh (ed.), 1–15.
Heidelberg/Berlin: Springer.
Schulte im Walde, Sabine
2002 Evaluating verb subcategorization frames learned by a German sta-
tistical grammar against manual definitions in the Duden dictionary.
In Proceedings of the 10th EURALEX International Congress, Anna
Braasch, and Claus Povlsen (eds.), 187–198. København: CST/KU.
forthc. The induction of verb frames and verb classes from corpora. In Cor-
pus Linguistics. An International Handbook, Anke Lüdeling, and
Merja Kytö (eds.). Berlin: Mouton de Gruyter.
Schumacher, Helmut, Jacqueline Kubczak, Renate Schmidt, and Vera de Ruiter
(eds.)
2004 VALBU – Valenzwörterbuch deutscher Verben. Tübingen: Narr.
382 Ulrich Heid

Spohr, Dennis
2004 Using A Valency Dictionary of English to enhance the lexicon of an
English LFG grammar. M.A. study, Institut für maschinelle Sprach-
verarbeitung, Universität Stuttgart.
Spohr, Dennis, and Ulrich Heid
2006 Modeling monolingual and bilingual collocation dictionaries in De-
scription Logic. In Proceedings of the EACL Workshop on Multi-
words and Multilinguality – ACL 2006, Paul Rayson, Serge Sharoff,
and Svenja Adolphs (eds.), 65–72. Trento: IRST/ACL
Spohr, Dennis, Aljoscha Burchardt, Sebastian Padó, Anette Frank, and Ulrich Heid
2007 Inducing a computational lexicon from a corpus with syntactic and
semantic annotation. In Proceedings of the Seventh International
Workshop on Computational Semantics (IWCS-7), Jeroen Geertzen,
Elias Thijsse, Harry Bunt, Amanda Schiffrin (eds.), 210–222. Til-
burg: University of Tilburg.
Tesnière, Lucien
1959 Eléments de Syntaxe Structurale. Paris: Klincksieck.
Zaenen, Annie
1988 Lexical information in LFG. An overview. Ms. University of Essex.
Subject index
ablative, 88, 90, 168 allocation, 233, 242, 247
abstraction, 29, 31, 60, 96, 172 apposition, 218, 221
accusative, 9, 25, 56, 86, 88-97, 166- appositive, 135
168, 178, 219-220, 222, 226, argument, 3, 10, 18-22, 30, 37, 48,
235, 237-239, 242-244, 246-247, 60, 69, 75, 77, 118, 121-123,
272, 340-345, 347, 349-351 126, 135, 144, 147-148, 152,
actant, 3-4, 37, 133, 165 157, 164-169, 171-174, 176-177,
active, 19, 27, 32, 37, 51, 131, 136, 179, 193-196, 200, 202-204, 206-
142, 242-244, 280, 284, 350 208, 231, 233-234, 241, 245,
actor, 37, 45 326, 332-333, 354, 366, 368, 379
addressee, 10, 258, 267 argument structure, 13, 27, 29, 67-
adjectival, 103, 265 68, 72, 74-78, 193-194, 196-200,
adjective, 4-5, 7-9, 11-13, 15, 89, 202, 204, 206-207, 210, 324,
101-114, 130, 135, 137, 139, 331-333, 335-336, 354, 366, 368
141-142, 148, 152-153, 168, 170- Preferred Argument Structure,
171, 218-219, 223, 225-226, 283, 194, 206-207
335, 366-367, 371, 375 aspect, 53-54, 142-143, 152, 156,
adjunct, 10, 15, 86-87, 92, 94, 97, 168, 218, 222-224, 259, 265, 268
133, 135, 165, 190, 218, 226, attachment, 233
278-279, 281, 324, 328, 333, avalent, 5, 52-53, 62, 165
347, 379
adposition, 77, 243 beneficiary, 48, 134, 230
adverb, 4, 6-7, 53, 87, 130, 142, 148, bootstrapping, 193-194, 196, 198-
152, 218, 230, 235, 247, 299, 200, 204, 208
359, 367 British National Corpus (BNC), 20-
adverb phrase, 76, 86, 88, 221, 225 22, 24, 29-30, 32, 41-45, 48, 101-
adverbial, 41, 133, 165, 218, 221- 103, 119-120, 129, 155, 273,
222, 226, 256, 261, 268, 279- 282-283, 300, 312, 317, 379
280, 289-291, 293-302, 328, 335,
353, 359-360, 362, 370, 377-378 case, 9, 19, 25-26, 37-38, 48, 56, 85-
affected, 22, 44, 88, 92, 224, 255- 96, 131, 155, 166, 168-169, 178,
256, 259, 261-262, 267-268 189, 219-223, 225, 229-231, 233-
affirmative, 60, 172 239, 244-247, 253, 256, 258,
affix, 61, 205-207 335, 340, 342, 344, 348, 359-362
affixation, 76 case frame, 19, 155, 169, 359-360,
agent, 26-27, 38-39, 43-48, 51, 55, 362
61, 67, 77, 86, 91, 124-125, 131, category, 3-6, 8, 11-12, 15, 26, 38,
133-134, 146-147, 156, 158, 168- 40-41, 43, 48, 51, 55, 61, 69, 76,
169, 171-172, 176-178, 188-189, 117, 133, 125, 153-154, 156-157,
195-197, 200, 202, 232, 242, 179, 185, 187, 188, 195, 197,
246, 258, 267, 319, 321, 324- 201-202, 204, 208, 219, 275,
326, 334-335, 354, 357, 359, 369 310, 368, 376
Albanian, 339-340 categorisation, 117, 141, 226, 371
384 Subject index

CGEL (A Comprehensive Grammar constituent, 4, 7-8, 55, 133, 135,

of the English Language), v, 102- 137, 144-146, 149-150, 173, 187,
103 231, 233, 238, 243, 245, 354,
Chinese, 206 358, 366, 368
chunk, 203, 358-360 construction, 4-12, 22, 28-29, 45-46,
circonstant, 133 51, 53-56, 59-60, 63, 67-80, 90,
circumstantial, 148, 165 97, 102, 110-111, 118-122, 131,
cleft, 238 134-136, 139, 145-146, 150, 152,
clitic, 61, 76, 339-340, 343, 348-351 156, 158, 175-176, 184, 188,
cluster, 106 193, 197-198, 202-206, 208-210,
Cobuild, 44, 46, 118, 379 217-218, 221-222, 226, 235, 237,
collocation, 16-17, 20, 31, 70, 135, 241-243, 245, 247-248, 253, 258,
188-189, 219, 274, 318, 367, 267, 277, 280-281, 283, 287-287,
370, 374-375, 378 295-304, 331, 336, 348, 360, 370
commutation, 60 construction grammar, 28, 32, 69-70,
complement, 6-13, 15, 18-27, 29, 32, 79, 193, 203-204
37-41, 43, 48, 67, 86-87, 89-93, context, 30, 43, 59, 70-71, 76-78, 95-
95-97, 103, 108, 110, 113, 117- 96, 106, 118-119, 131, 136, 140,
119, 121, 123-125, 127, 142-144, 148, 150, 152, 154, 156, 158,
147, 150-151, 154-155, 158, 165- 164, 167, 170, 199, 206-207,
168, 190, 218-222, 224-226, 229, 209, 237, 243, 272-273, 276-278,
233-239, 241-247, 253-254, 256- 294, 321, 334-335, 343, 355,
259, 268, 272-273, 279, 281, 361-362, 381
289, 291-295, 309, 336, 343-345, context-dependent, 206, 354
360, 366-367, 370, 373, 376, 378 contextual, 88, 137, 148, 158, 370,
complement inventory, 19-23, 27, 372, 377-379
32, 136, 157, 309 contextually optional, 20, 259
complementation, v, 13, 15, 102, contrastive, 215, 237, 271, 287, 294-
119-121, 123-127, 142, 218, 224, 295, 304
234, 289, 292, 294, 322, 324, coordination, 146, 321, 331, 335-
328, 366, 369-370, 377 336, 351, 356
component, 15, 18, 23, 25, 38-41, core (see also core frame elements),
130, 132, 137, 145, 147, 152, 17, 41, 77, 133-135, 147, 188,
156, 187, 204, 208-209, 277, 193, 201, 203-204, 208-209
346, 356, 366-367 crosslinguistic, 194-196, 200, 202
comprehension, 30, 163, 174-175,
179, 226
congruence, 256-257, 289-291, 293- Danish, 37, 48, 51-54, 57-62, 373
294, 346 dative, 9, 56, 73, 87-88, 90, 93, 96,
conjunct, 6, 12, 87, 101-102, 111, 144, 166, 168, 219, 225, 230,
114, 170-171, 295, 323, 331-333, 239, 242, 258, 267, 272, 340-
336, 362 347, 349-351
connexion, 163, 331, 333 declarative, 27, 32, 103, 235, 237,
constituency, 164, 201, 231, 368, 243-244, 247
376 deletion, 158, 167, 240-241
 Subject index 385

dependence/dependency, 3-6, 7-8, etymon, 56, 74

11, 15, 48, 71, 87, 163-173, 175, experiencer, 38, 48, 55-56, 73-74,
316, 322, 328, 333-334, 346, 89, 91, 102-104, 108, 156, 168,
358-359, 366 171
dependent, 4-5, 7, 15, 71, 111, 130, experiment, 31, 55, 67, 163, 166,
135, 144-146, 151-152, 155, 157, 170, 172-175, 177-179, 201, 204-
164-166, 168, 170, 273, 358, 205, 365, 372-377, 379
370-371 extraposition, 102-104, 107, 114,
diachronic, 51, 56, 68, 71, 73, 78, 146, 158
85, 90, 209
dialect, 278-279, 351 foreign language, 15-17, 164, 220,
dictionary, 3-4, 8, 15, 19, 26-27, 29- 272
30, 32, 38, 48, 52, 62, 85, 96-97, formal grammar, 375
101, 112, 114, 118-119, 121-122, frame, 19, 26, 30, 60, 79, 129-146,
127, 129, 139, 153, 157, 169, 149, 151-158, 169, 176-177, 179,
209, 217, 219-220, 223-225, 227, 194, 196, 198-199, 203, 206,
271-272, 274-275, 279, 281-283, 209, 231-234, 246, 257, 271,
290, 300, 309-312, 314, 317-318, 274, 277, 281-283, 345, 350,
361, 365-366, 368-375, 377, 379 359-362, 368-369, 376
ditransitive, 67, 74, 77, 157 frame element, 129-135, 144-146,
divalent, 19-20, 28, 41, 52, 55, 57, 151-152, 154, 157-158, 277, 281-
73, 110, 165-168, 254, 259, 262, 282, 369, 379
264 core, 133, 281
peripheral, 133
effected, 22, 267-268 Framenet, 25-26, 32, 48, 129-136,
elimination, 340, 342-343, 345-348, 138, 140-142, 144, 146-147, 149-
350 151, 153-158, 274, 280-281, 283-
ellipsis, 7, 13, 145-146, 157, 172, 284, 312, 361-362, 369, 376, 378
202, 206-207, 240 French, 51, 53-54, 56-57, 63, 68, 73-
emergence, 78, 175 74, 77, 163, 169, 302, 313, 368,
emergentism, 30, 32, 200 379
English, 3, 5-6, 9, 15, 19, 27, 30, 32, frequency, 28, 62, 67, 95, 108-109,
37-38, 44, 46-48, 51, 53, 58-60, 113, 134, 167, 184, 197-198,
67-67, 97, 101-103, 105, 114, 206, 209, 235, 239-240, 245,
117-119, 121-122, 127, 130, 136, 247, 273, 283, 288, 295-302,
145-146, 151, 153-154, 156, 165, 326, 378
169, 190, 197-198, 200, 203, frequent, 30, 52, 54, 62, 67, 91, 95,
217-218, 220-226, 229-241, 243- 101, 108-109, 112-114, 124, 134,
248, 253, 256-259, 267-268, 271- 156, 163, 167, 176, 196, 200,
273, 275-281, 287-290, 293-295, 217, 222-223, 226, 239-240, 243-
298-303, 309, 312, 336, 342, 244, 273, 277, 281, 283, 288,
361, 365, 367, 369, 372-377, 379 300, 304, 318, 370-371
ergative, 24, 44-46, 89 functor, 171, 324, 331-333, 335-336,
error, 197-198, 217-226, 358, 379 354
386 Subject index

garden path, 233 idiom, 11, 69, 188-189, 204, 303,

generalisation, 27, 29-31, 72, 76, 79, 375
141, 157, 188, 193, 197, 201, idiomatic, 17-18, 25, 287, 367
204, 206, 209-210, 278-279, 370 idiomaticity, 31, 372
generative grammar, 164-165, 201, idiomatization, 372
207 idiom principle, 15, 17
genitive, 87-97, 135, 168, 239, 246 idiosyncrasy, 88, 90
German, 3, 15-16, 31, 37, 77, 85-90, idiosyncratic, 15-18, 25, 27, 29, 186,
92-93, 95-97, 169, 173-174, 178, 345, 370
183, 190, 217-226, 229-235, 237- imperative, 146-147, 158, 339-340,
247, 253, 256-265, 267-268, 271- 342, 348
274, 276-277, 279-280, 283, 288- imperfect, 53-54, 229
291, 293-294, 302, 358, 362, incremental, 336, 358
367, 370-371, 376, 379 incremental semantic composition,
Germanic, 88, 229 358
gestalt, 29, 173 Indo-European, 88, 90, 96, 202, 335
goal, 48, 51, 60, 117, 130, 134, 140- instrument, 88, 156, 168-169, 171,
141, 156, 168, 171, 188, 230, 176-177, 188, 202, 232, 242,
255, 257, 321, 355, 357 246, 312-313
government, v, 48, 130, 137, 314 intransitive, 11-12, 58-59, 169, 177-
governor, 3-5, 11, 13, 122, 133, 144, 178, 197-198, 202, 241, 292-293,
146, 158 350
grammar, 3, 5, 9, 17, 26, 28, 32, 37, Italian, 273, 283
44, 46, 53, 57, 60, 62, 68-70, 72, item-specific, 31, 188-189, 202-203
75, 79, 85, 96, 101-103, 107-109,
118, 163-165, 168-169, 183, 185- Japanese, 245, 379
187, 189-190, 193, 195, 201,
203-204, 207-207, 217, 225-226, Left Associative Grammar (LAG),
253, 278, 322, 325-327, 333-335, 325, 334-335, 339, 345, 350-351
345, 356, 358-360, 361, 365, language acquisition, 28-30, 185-
368, 372-377, 379 186, 193-198, 194, 200-208, 210,
grammatical function, 68, 71, 74, 229, 336
129, 135-136, 256, 361, 368, Latin, 9, 56, 60, 62, 74-75, 164
376, 379 LDOCE, 30, 41-42, 45, 48
grammatical relation, 132, 164, 323- learned, 17-18, 117, 131, 195, 326
324, 326, 330-333, 335-336 learner, 16, 30, 186, 208-209, 217-
grammaticalisation, 51, 55, 57-58, 218, 220-227, 237-238, 283, 371
60-63, 67-80, 230, 236 learning, 15, 29, 117, 184, 186, 193-
196, 198-204, 206, 208-210, 217,
225, 237, 336, 375
head, 5-9, 11, 13, 17, 26, 41, 59, 74, lemma, 141, 348
117, 133, 135, 137, 150, 152, letter, 12, 79, 133, 139, 145, 165,
188, 241, 358-259, 379 310, 318, 327
hierarchisation, 246 lexeme, 4-5, 8, 12-13, 28, 70, 87, 96,
hierarchy, 4, 9, 46, 164, 167, 172- 121-122, 204, 224, 253, 257,
173, 234, 246-247, 256, 334, 357 272-274, 283, 367
 Subject index 387

Lexical Functional Grammar (LFG), nominalization, 142, 256

365-366, 368, 370, 372-373, 375- nominals, 43
377, 379 Norwegian, 287-297, 299-301, 303,
lexical unit, 3-4, 8-9, 11, 15-16, 18- 379
20, 25-28, 44, 121-122, 124, 129- noun, 5-9, 11-13, 15, 20-21, 23, 32,
132, 134, 137, 144, 147, 151, 39-41, 71, 77, 87, 92, 97, 104,
155, 157, 203, 273, 281-282, 345 119, 130, 135, 137, 139, 141-
lexicalization, 261, 326-327, 331 142, 145-146, 148-149, 152-153,
lexicography, 27, 139, 245, 271, 155, 164-165, 168-170, 173-174,
280, 367, 374 178, 186, 199, 218-222, 225,
lexicon, 4, 6, 27-28, 53, 55, 57, 60, 229-230, 234-239, 245, 247, 253,
62, 75, 130, 140, 153, 155, 196, 281, 283, 287, 289, 309, 311,
202, 209, 278, 323, 335, 346, 319, 323, 335, 344, 354, 362,
348, 351, 361-362, 366, 368, 366-371, 375
372-375, 377
locative, 7-8, 12, 52, 55, 88, 146,
168, 172, 200, 232, 235, 242, 267 object, 4-6, 9-11, 38-39, 43, 51-52,
54-57, 62, 73-74, 80, 86, 88-89,
meaning, 10, 12-13, 21, 28-30, 32, 91-93, 96, 110-111, 114, 122,
38, 40-41, 48, 51, 57, 60-61, 67, 131, 134-137, 142-144, 147, 149,
69-71, 74-76, 78, 86-88, 90-93, 152, 157-158, 165, 167-169, 171-
95-96, 107, 109, 112-113, 117- 172, 197-203, 219, 222, 224,
119, 123-124, 126-127, 129-131, 229, 234-239, 241-246, 254, 256-
133, 139-143, 145, 153-154, 158, 259, 262, 268, 272, 284, 319,
170-171, 173, 176, 179, 187, 322, 326-327, 339-351, 353, 359,
189, 191, 193, 196, 198-202, 367-368, 370, 376
204-206, 208-209, 230, 234, 238, objectification, 91
246, 253, 255-259, 262, 265, obligatoriness, 38
268, 271-274, 278, 280-283, 299- obligatory, 6-7, 13, 18, 20, 22, 32,
300, 303, 309-310, 312, 317-319, 133, 151, 157, 166, 169, 176,
336, 343, 349-351, 353, 357, 362 220, 257-258, 324, 344-345, 362,
mental lexicon, 27-28, 196, 209 366
mentalistic, 175 oblique, 133, 135, 144, 149, 157,
metaphor, 37, 67, 171, 193, 288 256, 376
metaphoric, 318 OED, 300, 302
metaphorical, 77, 241, 271, 279 omissibility, 137, 146
modifier, 7, 13, 97, 135, 149, 153, omission, 146-148, 157-158, 207-
328, 333, 335, 356, 379 208, 218, 220-221, 343
monovalent, 20-21, 61, 110, 165, operator, 71, 187, 356
262, 264 optional, 5-6, 12-13, 18, 20-22, 32,
multilingual, 287-288, 304, 318, 86-87, 95, 101, 147, 157, 169,
361-361, 373 176-177, 247, 257, 259, 268,
279, 324, 328, 342, 344-345, 379
Natural Language Processing (NLP), optionality, 20-22, 38, 167, 221, 224,
350, 365-379 379
Neogrammarians, 85 overgeneralizations, 197
388 Subject index

participant, 40-41, 43-47, 130-132, preference, 127, 167, 240, 272, 366,
134, 137-138, 168, 187-188, 195, 370-371, 373, 375, 377-378
198 preposition, 6-8, 11-13, 52, 54, 59,
particle, 52, 87, 96, 145, 151, 156, 71, 75, 88-89, 94, 96, 120, 130-
253, 257-259, 261, 265, 267-268, 131, 133, 136, 144-145, 150-152,
281, 340, 350 156-157, 170-171, 209, 217-218,
passive, 24, 32, 37, 46, 57, 132, 136, 220-222, 224-226, 230, 233, 235,
145-147, 150, 156, 158, 177-178, 239, 242-243, 246, 253-254, 256-
209, 219, 222, 225-226, 267-268, 257, 262, 268, 272-273, 279,
279, 284, 309 281, 347, 351, 354, 367, 376-377
passivisability, 151, 246, 254, 310, presupposition, 143, 199-200
377 probabilistic, 132, 201, 208
passivization, 144, 256, 279 processing, 70, 173-177, 179, 185,
patient, 38-39, 43-47, 86, 149, 156, 199, 207, 231-232, 234-235, 238,
158, 167-169, 172, 176, 178, 243, 309, 350, 356-357, 365,
188-189, 195-197, 232, 242, 246, 372, 374
283 pronoun, 10, 87-89, 94-95, 97, 101-
pattern, 8, 15, 18-25, 27-29, 32, 37, 102, 104-105, 108, 114, 132,
51-63, 67-69, 72-78, 85-86, 88- 135, 146, 156, 172, 176, 206,
90, 93, 101-102, 107-108, 110- 208, 225, 234-235, 238, 246,
114, 118-119, 122, 127, 129, 311, 339, 344
132, 134-137, 141-142, 144-145, proplet, 322-328, 330-332, 335-336
156-158, 167, 169, 174, 177, proposition, 10, 75, 78, 117-118,
188, 194-195, 197-201, 203-206, 121, 123-126, 151, 158, 171-173,
246, 253, 255-268, 271, 283, 302, 323, 326-327, 332-333, 336
288-290, 293-294, 300, 309, 312, proto-role, 169
319, 340, 343-344, 346-351, 361,
370-371, 373-374, 376, 378
periphery, 17, 133, 201, 209 raising, 10-11, 135, 151, 158, 233,
perspective role, 255-256, 267 294
phrase type, 129, 135-136, 379 range indicator, 311
phraseological, 31 receiver, 144, 258
polysemy, 28, 95, 129, 139, 141, recipient, 39, 43, 67, 77, 132, 156,
144, 154 173, 176, 267
polyvalency, 85, 87, 95-96 regent, 358
polyvalent, 96 register, 134, 147, 152, 204, 272-273
possessor, 55, 149, 199 regrammaticalisation, 63
postmodification, 279 regularity, 127, 142, 186, 256, 372
predicate, 11, 18, 37, 75, 103, 110, Romance, 75
119, 135, 149, 151-152, 158, rule, 9, 12, 16-18, 21, 28-31, 87,
184, 187, 200, 231-234, 246, 117, 127, 164, 196-198, 201,
366-370, 377-378 204, 208, 327, 335, 345-347,
predication, 10, 91, 151, 154 354, 366, 375
predicator, 133, 165, 170-174, 176-
177
 Subject index 389

scenario, 130, 134, 139, 155, 233, substitution, 321, 330, 344, 350
245, 355 Swedish, 58, 288, 301
scene, 131, 139, 178, 187, 195 synchronic, 51, 68, 70, 85
schema, 67-69, 71-73, 75-76, 78, 80,
86-87, 130, 145, 153, 155, 157, text type, 271, 275, 278, 282-283
175-176, 179, 188, 204, 208-209, thematic role, 19, 38, 131, 165, 168-
326-327, 30 169, 171-173, 176-179, 197, 204,
scheme, 120, 133, 246, 361 233-234, 238-239, 242-244, 247-
script, 18, 74, 79, 155, 357 248, 358-360, 362
sentence pattern, 246, 253, 255-263, theme, 32, 39, 48, 93, 96, 130, 133-
265-268, 340, 350 134, 156, 274, 281, 326
speech dialogue system, 353 Transformational Grammar, 169
s-selection, 117 transitive, 4-5, 9, 11-12, 51-52, 58-
storage, 18, 28-30, 117, 127, 209, 59, 67, 74, 77, 137, 156-157,
323, 360-361 169, 177-178, 195, 197-198, 202,
structure, 7, 10, 12, 15-16, 27-28, 37, 208, 241, 255, 261, 264, 292-
44, 51, 67, 69, 72, 75, 86, 88-89, 293, 350
91-92, 95, 102-103, 110, 114, translation, 16, 54, 63, 138, 218,
117, 129, 132, 139-141, 150-158, 226, 230-231, 248, 272-275, 283,
163-165, 167-173, 175, 186-190, 287-290, 292-294, 296-297, 299,
193-194, 196-202, 204-205, 208- 301-304, 318, 342, 356, 376
209, 222-224, 230-231, 233-235, translative, 328-329, 333, 335
238, 240, 243, 245-246, 248, trivalent, 19, 24, 38, 57-58, 67, 93,
271-272, 279-281, 287, 294, 303- 110, 165-166, 172, 174, 259
304, 311-313, 315-316, 321-326,
332, 335, 344, 351, 354, 366, universal, 71, 168, 184-186, 190,
368, 371-374, 378-379 193, 199, 201, 204, 326, 335
subcategorisation, v, 3-5, 117-118, universality, 185-186, 190
136, 156, 366, 371, 373-374, usage-based, 41, 80, 193, 198, 200-
376, 378 204, 206, 208-210
subclassification, 108, 377
subject, 5, 9-11, 19, 21, 27, 32, 37- VALBU (Valenzwörterbuch
39, 43-47, 51-52, 55-57, 62, 67, deutscher Verben), 15, 25-27, 40,
69, 73, 86, 89-91, 93, 102-110, 48, 217, 271, 281-282, 375
112-114, 119, 130-132, 135-137, valency/valence, v, 1, 3-6, 8-9, 11-
144-147, 149, 152, 154, 156, 13, 15-23, 25, 27,-32, 37-38, 41,
158, 164-165, 187, 197, 207, 48, 51-54, 58, 62-63, 79, 85-90,
219, 226, 229, 234-248, 254-256, 92-93, 95-96, 101, 112, 117-118,
258, 261-264, 268, 277, 279-280, 121-122, 126, 129-132, 134, 137,
282-284, 295, 309, 311, 319, 142, 144-145, 148, 151-152, 153-
322, 326-327, 331, 340-343, 346- 155, 157, 158, 163-171, 173,
347, 350-351, 353, 359, 366-368, 175-179, 183-184, 186-190, 193,
370-371, 377 208, 217-226, 231, 234, 239-240,
subjecthood, 237 245-246, 253, 271-273, 280, 283,
substitute, 7, 110, 230, 344, 350, 287, 295, 304, 309, 321-322,
366, 369, 373 324, 331-356, 339, 342-351, 353,
390 Subject index

356, 358-360, 362, 365-373, 375- variation, 32, 95, 134, 145, 202, 209,
379 219, 278, 369, 378-379
carrier, 15, 25, 118, 322-323, 333 VDE (A Valency Dictionary of
dictionary (see also VALBU, English), 15, 19-28, 32, 38, 40-
VDE), 19, 32, 38, 48, 52, 62, 85, 43, 48, 101-102, 104, 110-114,
97, 101, 114, 118-119, 121-122, 118-121, 122-125, 127, 271, 279,
127, 217, 223-225, 279, 281, 281-282, 309-310, 312, 365, 370-
309-310, 365-366, 368-369, 372, 379
375 verb, 3-6, 8-13, 15, 18-26, 28-32, 37-
filler, 322, 333, 343, 345-346, 41, 43-48, 51-63, 67-68, 70, 72-
348, 350 74, 76-78, 85-97, 102-103, 110,
pattern, 15, 18-19, 23, 25, 27, 29, 117-127, 130, 132-137, 139-145,
32, 37, 51-55, 57-62, 67-69, 72- 147-152, 156, 158, 163-176, 178-
78, 90, 101-102, 110-114, 119, 179, 186, 188-189, 193-200, 202-
127, 129, 134, 136, 142, 144- 209, 218-223, 225, 227, 229,
145, 157-158, 197, 312, 343, 231-236, 238-246, 248, 253-263,
347-349, 370, 373, 376, 378 265, 267-268, 271-284, 287-293,
qualitative, 62, 166, 168 295, 299, 302, 309-310, 312,
quantitative, 20, 52, 62, 86, 164, 314, 318-319, 321-323, 326-328,
168 330-333, 335-336, 339-351, 359,
semantic, 20, 37-38, 40-41, 46- 362, 366-367, 369-370, 373, 375-
47, 145, 193 378
shift, 85, 88-90 verb island construction, 187-189,
syntactic, 18, 37-38, 117-118, 206
144, 193
Author index
Abney, Steven, 358 Casenhiser, Devin M., 67, 204-205
Abraham, Werner, 86, 229, 238, 247 Chafe, Wallace L., 168
Ágel, Vilmos, vi, 31, 85, 97, 165, Chomsky, Noam, 10, 17-18, 117-
219, 222, 246, 322 118, 121, 164
Akthar, Nameera, 201 Clancy, Patricia, 206-207
Allen, Shanley E. M., 207, 210 Clark, Eve V., 195
Allerton, David J., vii, 37-38, 120, Coene, Ann, 227, 246
158 Cole, Ronald A., 353
Andersen, Henning, 60-63, 75-76, 80 Collins, Peter, 44, 46, 107, 275, 281
Asmussen, Jørg, 373 Cornell, Alan, 225, 227
Atkins, Beryl T. Sue, 130, 136, 144 Croft, William, 32, 46, 60, 68-70,
73, 79, 234
Cronin, Beau, 368, 376
Baker, Collin F., 361, 368-369, 376 Cruse, David Alan, 18, 32, 68-69,
Bartlett, Frederic C., 175 79, 121, 155
Bates, Elizabeth A., 205 Crystal, David, 4, 8
Behaghel, Otto, 97 Curcio, Martina Lucia, 273-274, 283
Behrens, Heike, vi, 193, 202
Bianco, Maria Teresa, 272
Biber, Douglas, 102-103, 107-108 Dahl, Östen, 230, 247
Bielińska, Monika, 223, 227 Deacon, Terrence, 184-188, 190
Bisang, Walter, 72, 78, 80 Dederding, Hans-Martin, vi, 31
Boas, Hans C., 283 Delbrück, Berthold, 97
Böhtlingk, Otto, 164 Denison, David, 11, 79
Bowerman, Melissa, 195-198, 202, Dijk, Teun A. van, 173
204, 210 Dixon, Robert M. W., 169
Braune, Wilhelm, 97 Domi, Mahir, 339, 350
Bräunling, Petra, 223 Donhauser, Karin, 97
Bresnan, Joan, 365, 368, 370 Dowty, David R., 169
Brietzmann, Astrid, 357 DuBois, John W., 206-207
Broschart, Günter, 51 Durme, Karen van, 62
Brown, Penelope, 196, 202, 210 Dürscheid, Christa, 97
Buchholz, Oda, 339, 342-343, 350- Durst-Andersen, Per, 54
351
Bücher, Kerstin, 358, 362
Bühler, Karl, v, 164, 175 Ebert, Robert Peter, 97
Bybee, Joan L., 28, 70, 80, 209 Edelman, Gerald, 183-184, 187, 190-
191
Elman, Jeffrey L., 200-201, 209
Carbone, Elena, 170 Emons, Rudolf, vi-vii 18, 183
Carlson, Gregory N., 177-178 Engberg-Pedersen, Elisabeth, 60
Carpenter, Malinda, 201 Engel, Ulrich, v, 15, 19, 31-32, 234,
Carroll, Lewis, 51, 59 247
392 Author index

Engelen, Bernhard, 19 Groot, Albert W. de, v-vi

Engelkamp, Johannes, 166, 170-171 Gross, Maurice, 55, 57, 368
Erdmann, Peter, 102 Günther, Udo, 173
Erjavec, Tomaz, 379
Evert, Stefan, 247 Habermann, Mechthild, vi, 85
Haegeman, Liliane, 18, 48, 169
Hair, Joseph F., 106
Fanego, Teresa, 245 Halliday, Michael A. K., 80, 295
Fellbaum, Christiane, 158 Harder, Peter, 60
Fennell, Barbara, 229 Harris, Zellig S., 6
Ferretti, Todd R., 176 Haspelmath, Martin, 71
Fiedler, Wilfried, 339, 342-343, 350- Hatherell, Andrea, 176
351 Hausmann, Franz Josef, 16
Fillmore, Charles John, vi, 9, 19, 25, Hausser, Roland, vi, 321, 334-336,
31-32, 38, 69-70, 79, 97, 129- 351, 366
130, 136, 144, 156, 168-169, Hawkins, John A., 229-234, 236,
183, 188, 204, 247, 267, 361- 238-239, 242-247
362, 368-369, 376 Heath, David, vi, 31
Finegan, Edward, 278 Heid, Ulrich, vi, 247, 365, 371, 374
Fischer, Klaus, vi, 224-226, 229, Heine, Bernd, 75
233, 242, 245-248, 272-273 Helbig, Gerhard, v-vi, 15, 18, 25-26,
Fischer, Olga, 75, 80, 229, 246 31, 37-39, 118, 166, 169
Fontenelle, Thierry, 130 Hellwig, Peter, 332
Francis, Elaine J., 145 Heltoft, Lars, 60, 63
Francis, Gill, 102, 118 Herbst, Thomas, v-vii, 3, 9, 15, 18,
François, Jacques, 51 29, 31-32, 101, 104, 110-114,
Fraser, Norman M., 7 117-118, 136, 157, 166, 169,
Fromkin, Victoria, 10 217, 224, 324, 328, 344-345,
365, 372, 379
Heringer, Hans-Jürgen, 15, 183
Garrod, Simon C., 176-177 Herslund, Michael, 54
Gazdar, Gerald, 118 Hilpert, Martin, 68, 79
Geisler, Hans, 63 Himmelmann, Nikolaus P.,70-72,
Gellerstam, Martin, 288 76-77
Givón, Talmy, 47, 72 Hoffmann, Sebastian,80
Goldberg, Adele E., 9, 27, 31, 60, Hopper, Paul J.,51, 61, 68, 71-72,
67-69, 74, 77, 79, 204-206, 267 74, 76, 79
Görz, Günther, vi, 353, 357 Hörmann, Hans, 168
Götz, Dieter, vi, 309, 372, 379 Householder, Fred. W. Jr., 102
Götz-Votteler, Katrin, v-vi, 37, 47, Hubbard, Philipp, 339, 350
372 Huddleston, Rodney, 5-8, 10, 102-
Gouws, Rufus H., 374 103, 107, 109, 117-118, 122
Goyens, Michèle, 62, 73-74, 80 Hunston, Susan, 102, 118
Greenberg, Joseph H., 327
Greule, Albrecht, 85, 97 Ickler, Irene, vi, 242, 253, 267
Grimm, Jacob, 91 Israel, Michael, 68
 Author index 393

Jackendoff, Ray, 168-169 MacRae, Ken, 176

Johansson, Stig, vi, 287-288, 294, MacWhinney, Brian, 32, 198, 200,
299 205, 335-336
Johnson, Christopher R., 136 Manning, Christopher, 201
Johnson, Mark H., 288 Manning, Elizabeth, 102, 118
Johnson-Laird, Philip N., 175 Matthews, Peter H., v-vi, 3, 11, 15,
Joshi, Aravind, 379 102, 120
Mauner, Gail, 177-178
Kabashi, Besim, vi, 335, 339, 348 Maxwell, Hugh, 85
Kallulli, Dalina, 350 Maxwell, John T., 373
Kaltenböck, Gunther, 102 McGlashan, Scott, 7
Kaplan, Ronald, 373 McWhorter, John, 229-230, 245-246
Kaufman, Terence, 229 Meillet, Antoine, 76
Kay, Paul, 9, 31, 69-70, 79, 188, 204 Mel’čuk, Igor A., 152, 366
Keenan, Janice M., 172 Michaelis, Laura A., 145
Kemmer, Suzanne, 68, 79 Milligan, Thomas R., 97, 315
Kennedy, Graeme, 304 Mindt, Ilka, vi, 101, 104-108, 114
Kermes, Hannah, 371 Mittmann, Brigitta, vi, 271
Kintsch, Walter, 171-173 Mohit, Behrang, 154
Kirchmeier-Andersen, Sabine, 62
Klavans, Judith L., 353
Klein, Ewan, 201 Nagel, H. Nicholas, 172
Klotz, Michael, vi, 18, 24, 29, 32, Narayanan, Srini, 154, 157
48, 117, 370-372 Newmark, Leonard, 339, 350
Knöferle, Pia, 178 Nichols, Johanna, 243
Knorr, Michael, 358 Noël, Dirk, vi, 60-61, 63, 67, 80, 118
Koch, Peter, 56
Kolde, Gottfried, 227
Kolvenbach, Monika, 97 O’Connor, Mary C., 9, 31, 69-70,
Korhonen, Jarmo, 85 188
Kostallari, Androkli, 351 Obliers, Rainer, 172
Krefeld, Thomas, 51 Ørsnes, Bjarne, 373
Krone, Maike, 241, 274, 280

Pagliuca, William, 70
Lakoff, George, 79, 288 Parkes, Geoff, 217-218
Langacker, Ronald W., 79, 187, 267 Paul, Hermann, 97, 236-237
Leech, Geoffrey, 31 Perkins, Revere, 70
Lehmann, Christian, 63, 72 Petruck, Miriam R. L., 155, 368
Lenz, Barbara, 97 Plank, Frans, 229
Levin, Beth A., 53, 156 Pollard, Carl, 379
Lobin, Henning, 336 Postal, Paul M., 10
Lowe, John B., 361 Prifti, Peter, 339, 350
Ludwig, Bernd, vi, 353, 358 Pullum, Geoffrey K., 5-8, 10, 102-
Lüdeling, Anke, 226 103, 107, 109, 117-118, 122
Lyons, John, 164-165 Pustejovsky, James, 153
394 Author index

Qesku, Pavli, 351 Sperber, Dan, 229

Quirk, Randolph, 9-11, 13, 102-104, Spohr, Dennis, 372-374, 378-379
107-108, 118, 120, 247 Spranger, Kristina, 247
Sridhar, Shikaripur N., 43
Raue, Burkhardt, 166 Strohner, Hans, 163, 175-176
Rausch, Georg, 97 Sweetser, Eve, 154
Resnik, Philip, 353
Rickheit, Gert, vi, 48, 163, 167, 173, Tabor, Whitney, 70
175-176 Tanenhaus, Michael K., 177-178
Roe, Ian F., vi, 166, 217, 224-225, Tesnière, Lucien, v, 3-5, 13, 37, 62,
227, 246 133, 163-166, 171, 183, 246,
Rohdenburg, Günter, 241, 245 328, 366
Rosier, Irène, 6 Thomason, Sarah Grey, 229
Ruhl, Charles, 158 Thompson, Sandra A., 51, 61
Ruppenhofer, Josef, 155, 158, 369 Toçi, Fatmir, 351
Tomasello, Michael, 29-31, 187-189,
Sag, Ivan A., 375, 379 200-204, 209-210
Sanford, Anthony J., 176 Trask, R. Lawrence, 3-4
Saussure, Ferdinand de, 17 Traugott, Elizabeth Closs, 68, 70-72,
Scheibman, Joanne, 209 74, 76, 79
Schenkel, Wolfgang, v, 15, 18, 25,
39, 169 Uhlisch, Gerda, 343, 351
Schøsler, Lene, vi, 48, 51, 62-63, 68,
73-80 Verhagen, Arie, 68
Schreiber, Herbert, 226 Viberg, Åke, 288
Schröder, Heike, 207
Schrodt, Richard, 93, 97 Waal, Frans, de 187
Schüller, Susen, 15 Wegener, Heide, 230
Schulte im Walde, Sabine, 378 Welke, Klaus M., 169, 246
Schumacher, Helmut, v, 15, 169, West, Jonathan, 190
217, 247, 272, 277, 283, 375 Wiemer, Björn, 72, 78
Searle, John, 184 Wierzbicka, Anna, 267
Seppänen, Aimo, 110 Wilczok, Karin, 166
Sethuraman, Nitya, 67 Willems, Klaas, 227, 246
Shapiro, Lewis P., 172 Wilson, Deirdre, 229, 299
Sichelschmidt, Lorenz, vi, 48, 163,
170, 173, 175-176 Zaenen, Annie, 368
Sinclair, John, 17, 31, 102, 118 Zaima, Susumu, 267
Skafte Jensen, Eva, 58 Zifonun, Gisela, 242, 246-247
Somers, Harold, 165 Zwaan, Rolf A., 175
Sommerfeldt, Karl-Ernst, 226 Zwicky, Arnold, 7