Cognitive Linguistics Croft Cruse (2004)

This page intentionally left blank

Cognitive Linguistics
Cognitive Linguistics argues that language is governed by general cognitive prin-
ciples, rather than by a special-purpose language module. This introductory text-
book surveys the field of cognitive linguistics as a distinct area of study, presenting
its theoretical foundations and the arguments supporting it. Clearly organized and
accessibly written, it provides a useful introduction to the relationship between
language and cognitive processing in the human brain. It covers the main top-
ics likely to be encountered in a course or seminar, and provides a synthesis of
study and research in this fast-growing field of linguistics. The authors begin by
explaining the conceptual structures and cognitive processes governing linguis-
tic representation and behavior, and go on to explore cognitive approaches to
lexical semantics, as well as syntactic representation and analysis, focusing on
the closely related frameworks of cognitive grammar and construction grammar.
This much-needed introduction will be welcomed by students in linguistics and
cognitive science.

william croft is Professor of Linguistics at the University of Manchester,

where he has taught since 1994. His primary areas of research are typology, cog-
nitive linguistics, construction grammar, language change and verbal semantics.
He has previously published in this series Typology and Universals, 1st edition
(1990), and 2nd edition (2003), and is also the author of Radical Construction
Grammar (2001).
d. alan cruse was Senior Lecturer in Linguistics at the University of

Manchester, until he retired in 2001. He taught semantics, pragmatics and
psycholinguistics, and his primary areas of research are lexical semantics and
metaphor. He has previously published in this series Lexical Semantics (1986),
and is also the joint editor of Handbook of Lexicology (2002).
CAMBRIDGE TEXTBOOKS IN LINGUISTICS
General editors: s . r . a n d e r s o n, j. b r e s na n, b. c o m r i e ,
w. d r e s s l e r , c . e w e n, r . h u d d l e s t o n, r . l a s s ,
d. lightfoot, j. lyons, p. h. matthews, r. posner,
s. romaine, n. v. smith, n. vincent
In this series:
p. h . m at t h e w s Morphology Second edition
b. c o m r i e Aspect
r . m . k e m p s o n Semantic Theory
t. bynon Historical Linguistics
j. allwood, l.-g. anderson and ö. dahl Logic in Linguistics
d. b. f r y The Physics of Speech
r. a. hudson Sociolinguistics Second edition
a . j. e l l i o t Child Language
p. h . m at t h e w s Syntax
a. redford Transformational Syntax
l . b au e r English Word-Formation
s. c. levinson Pragmatics
g. brown and g . y u l e Discourse Analysis
r . h u d d l e s t o n Introduction to the Grammar of English
r . l a s s Phonology
a. comrie Tense
w. k l e i n Second Language Acquisition
a. j. woods, p. fletcher and a. hughes Statistics in Language Studies
d. a. cruse Lexical Semantics
a. radford Transformational Grammar
m . g a r m a n Psycholinguistics
g . g . c o r b e t t Gender
h. j. giegerich English Phonology
r . c a n n Formal Semantics
j. l ave r Principles of Phonetics
f. r . pa l m e r Grammatical Roles and Relations
m. a. jones Foundations of French Syntax
a. radford Syntactic Theory and the Structure of English: A Minimalist Approach
r . d. va n va l i n, j r, and r . j. l a p o l l a Syntax: Structure, Meaning and Function
a. duranti Linguistic Anthropology
a . c r u t t e n d e n Intonation Second edition
j. k . c h a m b e r s and p. trudgill Dialectology Second edition
c. lyons Definiteness
r . k ag e r Optimality Theory
j. a. holm An Introduction to Pidgins and Creoles
c . g . c o r b e t t Number
c . j. e w e n and h. van der hulst The Phonological Structure of Words
f. r . pa l m e r Mood and Modality Second edition
b. j. b l a k e Case Second edition
e . g u s s m a n Phonology: Analysis and Theory
m. yip Tone
w. c r o f t Typology and Universals Second edition
f. c o u l m a s Writing Systems: an Introduction to their Linguistic Analysis
p. j. hopper and e . c . t r au g o t t Grammaticalization Second edition
l. white Second Language Acquisition and Universal Grammar
i. plag Word-formation in English
w. c r o f t and d. a . c r u s e Cognitive Linguistics
a. siewierska Person
a. radford Minimalist Syntax: Exploring the Structure of English
d. b üring Binding Theory
Cognitive Linguistics
W I L L I A M C RO F T
University of Manchester
and
D . A L A N C RU S E
University of Manchester
  
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press

The Edinburgh Building, Cambridge  , UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521661140
This publication is in copyright. Subject to statutory exception and to the provision of

relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
First published in print format 2004
- ---- eBook (NetLibrary)

- --- eBook (NetLibrary)
- ---- hardback

- --- hardback
- ---- paperback

- --- paperback
Cambridge University Press has no responsibility for the persistence or accuracy of s
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Contents
List of figures page xii

List of tables xiii
Preface xv
1 Introduction: what is cognitive linguistics? 1
Part I: A conceptual approach to linguistic analysis
2 Frames, domains, spaces: the organization of conceptual

structure 7
2.1 Arguments for frame semantics 7
2.2 Concepts: profile-frame organization 14
2.3 Some consequences of the profile-frame/domain distinction 16
2.4 Extensions of the basic profile-frame/domain distinction 22
2.4.1 Locational and configurational profiles 22
2.4.2 Scope of predication 23
2.4.3 Relationships between domains 24
2.5 Domains and idealized cognitive models 28
2.6 Mental spaces 32
3 Conceptualization and construal operations 40

3.1 Introduction 40
3.2 Attention/salience 46
3.2.1 Selection 47
3.2.2 Scope of attention (dominion) 50
3.2.3 Scalar adjustment 51
3.2.4 Dynamic attention 53
3.3 Judgement/comparison 54
3.3.1 Categorization 54
3.3.2 Metaphor 55
3.3.3 Figure-ground alignment 56
3.4 Perspective/situatedness 58
3.4.1 Viewpoint 59
vii
viii Contents
3.4.2 Deixis 59
3.4.3 Subjectivity 62
3.5 Constitution/Gestalt 63
3.5.1 Structural schematization 63
3.5.2 Force dynamics 66
3.5.3 Relationality (entity/interconnection) 67
3.6 Conclusion 69
4 Categories, concepts and meanings 74

4.1 Introduction 74
4.2 The classical model of category structure 76
4.3 The prototype model of category structure 77
4.3.1 Graded centrality 77
4.3.2 The representation of conceptual categories 81
4.3.3 Levels of categorization 82
4.3.4 Shortcomings of prototype theory 87
4.3.5 The frame-based account of prototype effects 91
4.4 A dynamic construal approach to conceptual categories 92
4.4.1 Category boundaries 93
4.4.2 Frames 95
4.4.3 Levels of categorization 96
4.5 The dynamic construal of meaning 97
4.5.1 Contextualized interpretation 98
4.5.2 Purport 100
4.5.3 Constraints 101
4.5.4 Construal 103
4.6 Structural and logical aspects of meaning 104
4.7 Part I: Concluding remarks 105
Part II: Cognitive approaches to lexical semantics
5 Polysemy: the construal of sense boundaries 109

5.1 Introduction 109
5.2 Full sense boundaries 110
5.2.1 Homonymy and polysemy 111
5.2.2 Entrenchment 111
5.2.3 Boundary effects 112
5.2.4 The nature of full sense units 115
5.3 Sub-sense units with near-sense properties 116
5.3.1 Facets 116
5.3.2 Microsenses 126
5.3.3 Ways-of-seeing 137
5.3.4 Semantic components and low-autonomy active zones 138
Contents ix
5.3.5 Contextual modulation 140

5.4 Autonomy: summary 140
6 A dynamic construal approach to sense relations I:

hyponymy and meronymy 141
6.1 Hyponymy 141
6.1.1 Introductory 141
6.1.2 Hyponymy and context 143
6.1.3 Relations between lexical items 146
6.1.4 Taxonymy 147
6.2 Lexical aspects of the part-whole relation 150
6.2.1 The part-whole relation 151
6.2.2 Meronymy 159
7 A dynamic construal approach to sense relations II:

antonymy and complementarity 164
7.1 Oppositeness 164
7.1.1 Aspects of the construal of oppositeness 164
7.1.2 Main varieties of opposite 165
7.1.3 Goodness-of-exemplar in opposites 166
7.2 Complementarity 167
7.2.1 Gradable vs. non-gradable construal of properties 167
7.2.2 Profiling against domains 167
7.3 Antonymy 169
7.3.1 A survey of antonym types 169
7.3.2 Monoscalar systems: polar antonyms 172
7.3.3 Bi-scalar systems 181
7.4 Variable construal of antonyms and complementaries 185
7.4.1 Absolute vs. relative construal 185
7.4.2 Scale features 189
7.5 Conclusion 192
8 Metaphor 193
8.1 Figurative language 193
8.2 The conceptual theory of metaphor 194
8.2.1 Introduction 194
8.2.2 Issues in the conceptual theory of metaphor 198
8.3 Novel metaphor 204
8.3.1 The life history of a metaphor 204
8.3.2 How do we recognize metaphors? 206
8.3.3 Blending Theory and novel metaphors 207
8.3.4 Context sensitivity 209
8.3.5 Asymmetry of vehicle and target 210
x Contents
8.4 Metaphor and simile 211

8.4.1 Two types of simile 211
8.4.2 Theories of the relation between simile and
metaphor 211
8.4.3 Metaphor-simile combinations 215
8.5 Metaphor and metonymy 216
8.5.1 Characterizing metonymy 216
8.5.2 Metaphor-metonymy relations 217
8.5.3 Types of indeterminacy 219
8.6 Conclusion 220
Part III: Cognitive approaches to grammatical form
9 From idioms to construction grammar 225

9.1 Introduction 225
9.2 The problem of idioms 229
9.3 Idioms as constructions 236
9.4 From constructions to construction grammar 247
10 An overview of construction grammars 257

10.1 Essentials of construction grammar theories 257
10.1.1 Grammatical representation: the anatomy of a
construction 257
10.1.2 The organization of constructional knowledge 262
10.2 Some current theories of construction grammar 265
10.2.1 Construction Grammar (Fillmore, Kay et al.) 266
10.2.2 Lakoff (1987) and Goldberg (1995) 272
10.2.3 Cognitive Grammar as a construction grammar 278
10.2.4 Radical Construction Grammar 283
10.3 Conclusion 290
11 The usage-based model 291

11.1 Grammatical representation and process 291
11.2 The usage-based model in morphology 292
11.2.1 Entrenchment and representation of word forms 292
11.2.2 Regularity, productivity and default status 295
11.2.3 Product-oriented schemas 300
11.2.4 Network organization of word forms 302
11.2.5 Conclusion 307
11.3 The usage-based model in syntax 308
11.3.1 Type/token frequency, productivity and
entrenchment 308
11.3.2 Product-oriented syntactic schemas 313
Contents xi
11.3.3 Relevance and the organization of construction

networks 318
11.3.4 The acquisition of syntax and syntactic change 323
11.4 Conclusion 326
12 Conclusion: cognitive linguistics and beyond 328
References 330
Author index 344
Subject index 347
Figures
2.1 RADIUS and CIRCLE page 15

2.2 NIECE 23
2.3 Domain structure underlying the concept of the letter T 26
2.4 Specific and nonspecific indefinites 35
2.5 Mental space diagram for example (29) 36
4.1 Boundaries of AIRPLANE, GLIDER and HANG GLIDER 90
7.1 A simplified monoscalar system 170
7.2 A disjunct equipollent system 170
7.3 A parallel equipollent system 170
7.4 An overlapping system 171
7.5 A full monoscalar system 173
10.1 The symbolic structure of a construction 258
10.2 The relation between form and function in a componential
syntactic theory 258
10.3 The relation between form and function in
construction grammar 259
10.4 Simplified generative grammar and construction grammar
representations of Heather sings 260
10.5 Elements, components and units of a construction 261
xii
Tables
3.1 Linguistic construal operations as instances of general

cognitive processes page 46
9.1 Types of idioms compared to regular syntactic expressions 236
9.2 The syntax-lexicon continuum 255
xiii
Preface
This book provides an overview of the basic principles and methods of

cognitive linguistics, in particular as they are applied to semantic and syntactic
issues. It is intended to be used as a textbook for a course on cognitive linguistics
for advanced undergraduates and postgraduate students, as well as functioning
as an introduction to this approach to language for linguists and researchers in
neighboring disciplines. Parts I and II may also function as a textbook for a course
on cognitive semantics, supplemented by case studies from the cognitive linguis-
tic literature. Part III may also function as introductory reading for a course on
construction grammar, followed by readings from the literature that delve into the
details of particular theories of construction grammar and the analyses of particular
constructions.
The chapters of the book were independently written, but jointly discussed.
Croft is responsible for chapters 1–3 and 9–12, and Cruse for chapters 4–8 (this
fact will no doubt be obvious to the reader). Cruse also contributed to §3.2.1, and
Croft to §8.2. Although we have written our chapters independently, the book
represents a single coherent perspective on cognitive linguistics. We agree on all
of the major points, and most of the minor ones; what minor disagreements remain
do not compromise the integrity of the analysis as a whole.
Croft would like to thank members of the linguistics and psychology departments
at the Max Planck Institute for Evolutionary Anthropology, Leipzig, Germany, and
Jóhanna Barð dal, Chuck Fillmore, Laura Janda, Paul Kay and Ron Langacker for
their comments on earlier versions of Part III, and Liliana Albertazzi and the
participants in the Workshop on ‘Which Semantics?,’ Bolzano, Italy, 1995 for
their comments on topics dealt with in Part I. Cruse would like to thank Liliana
Albertazzi and fellow-participants (George Lakoff, Ron Langacker and Len Talmy)
at the Summer School on Cognitive Semantics, Bolzano, Italy, 1999; Arie Verhagen
and the students at the LOT Winter School, Leiden, Holland, 2002; and members
of the Equipe Rhéma, University of Lyon, France, for their comments on various
topics dealt with in the book. Last but not least, we both thank the students of
successive classes on cognitive linguistics at the University of Manchester, who
used materials that eventually became the chapters presented here. Of course, all
responsibility for the final product remains with us.
xv
1
Introduction: what is
cognitive linguistics?
Cognitive linguistics is taken here to refer to the approach to the study of

language that began to emerge in the 1970s and has been increasingly active since
the 1980s (now endowed with an international society with biennial conferences
and a journal, Cognitive Linguistics). A quarter century later, a vast amount of
research has been generated under the name of cognitive linguistics. Most of the
research has focused on semantics, but a significant proportion also is devoted
to syntax and morphology, and there has been cognitive linguistic research into
other areas of linguistics such as language acquisition, phonology and historical
linguistics. This book can only outline the basic principles of the cognitive linguis-
tic approach and some of its more important results and implications for the study
of language. In this chapter, we briefly describe the major hypotheses of cognitive
linguistics (as we see them), and how we will develop these hypotheses in the rest
of the book.
We see three major hypotheses as guiding the cognitive linguistic approach to
language:
r language is not an autonomous cognitive faculty
r grammar is conceptualization
r knowledge of language emerges from language use
These three hypotheses represent a response by the pioneering figures in cognitive
linguistics to the dominant approaches to syntax and semantics at the time, namely
generative grammar and truth-conditional (logical) semantics. The first principle
is opposed to generative grammar’s well-known hypothesis that language is an
autonomous (indeed, innate) cognitive faculty or module, separated from nonlin-
guistic cognitive abilities. The second principle is opposed to truth-conditional
semantics, in which a semantic metalanguage is evaluated in terms of truth and
falsity relative to the world (or, more precisely, a model of the world). The third
principle is opposed to reductionist tendencies in both generative grammar and
truth-conditional semantics, in which maximally abstract and general representa-
tions of grammatical form and meaning are sought and many grammatical and
semantic phenomena are assigned to the ‘periphery’.
1
2 Introduction
Generative grammar and truth-conditional semantics are of course still vigorous

research paradigms today, and so cognitive linguists continue to present arguments
for their basic hypotheses as well as exploring more specific empirical questions
of syntax and semantics within the cognitive linguistic paradigm. Some of these
arguments will be presented in the course of this book. Here we describe in some-
what more detail the content of these three hypotheses and how they are manifested
in subsequent chapters.
The first hypothesis is that language is not an autonomous cognitive faculty. The
basic corollaries of this hypothesis are that the representation of linguistic knowl-
edge is essentially the same as the representation of other conceptual structures,
and that the processes in which that knowledge is used are not fundamentally
different from cognitive abilities that human beings use outside the domain of
language.
The first corollary is essentially that linguistic knowledge – knowledge of mean-
ing and form – is basically conceptual structure. It is probably not difficult to accept
the hypothesis that semantic representation is basically conceptual (though what
that entails is a matter of debate; see below). But cognitive linguists argue that
syntactic, morphological and phonological representation is also basically con-
ceptual. This might appear counterintuitive at first: sounds are physical entities,
and ultimately so are utterances and their formal structure. But sounds and utter-
ances must be comprehended and produced, and both of those processes involve
the mind. Sounds and utterances are the input and output of cognitive processes
that govern speaking and understanding.
The second corollary is that the cognitive processes that govern language use,
in particular the construction and communication of meaning by language, are
in principle the same as other cognitive abilities. That is, the organization and
retrieval of linguistic knowledge is not significantly different from the organization
and retrieval of other knowledge in the mind, and the cognitive abilities that we
apply to speaking and understanding language are not significantly different from
those applied to other cognitive tasks, such as visual perception, reasoning or motor
activity. Language is a distinct human cognitive ability, to be sure. From a cognitive
perspective, language is the real-time perception and production of a temporal
sequence of discrete, structured symbolic units. This particular configuration of
cognitive abilities is probably unique to language, but the component cognitive
skills required are not.
This position is sometimes taken as a denial of an innate human capacity for
language. This is not the case; it is only a denial of an autonomous, special-purpose
innate human capacity for language. It is of course reasonable to assume that there
is a significant innate component to general human cognitive abilities, and that
some of those innate properties give rise to human linguistic abilities that no other
Introduction 3
species apparently has. However, innateness of cognitive abilities has not been a
chief concern of cognitive linguists, who are more concerned with demonstrating
the role of general cognitive abilities in language.
The hypothesis that language is not an autonomous cognitive faculty has had two
major implications for cognitive linguistic research. Much cognitive linguistic re-
search has been devoted to elucidating conceptual structure and cognitive abilities
as they are seen to apply to language, in the effort to demonstrate that language can
be adequately modeled using just these general conceptual structures and cognitive
abilities. Part I of this book is devoted to explicating cognitive linguistic models
of cognitive structure and abilities (see also chapter 11).
Second, cognitive linguists appeal at least in principle to models in cognitive psy-
chology, in particular models of memory, perception, attention and categorization.
Psychological models of memory have inspired linguistic models of the organi-
zation of linguistic knowledge into frames/domains (chapter 2), and grammatical
knowledge in networks linked by taxonomic and other relations (see chapters
10–11 in Part III). Psychological models of attention and perception, especially
Gestalt psychology, have led to the explication of many conceptualization pro-
cesses in semantics (chapter 3, and see also the next paragraph). Finally, psycho-
logical models of categorization, in particular prototypes and graded centrality,
and more recent models of category structure, have had perhaps the greatest influ-
ence on both semantic and grammatical category analysis in cognitive linguistics
(chapter 3; see, e.g., Lakoff 1987, Taylor 1989[1997]).
The second major hypothesis of the cognitive linguistic approach is embodied
in Langacker’s slogan ‘grammar is conceptualization.’ This slogan refers to a more
specific hypothesis about conceptual structure, namely that conceptual structure
cannot be reduced to a simple truth-conditional correspondence with the world. A
major aspect of human cognitive ability is the conceptualization of the experience
to be communicated (and also the conceptualization of the linguistic knowledge
we possess). A major theme of the chapters in Part I of this book is that all aspects
of conceptual structure are subject to construal, including the structure of cate-
gories (chapter 4) and the organization of knowledge (i.e., conceptual structures;
chapter 2). In particular, it is argued that grammatical inflections and grammatical
constructions play a major role in construing the experience to be communicated
in specific ways (chapter 3). Part II of this book also explores and defends the
conceptualization hypothesis for a wide range of lexical semantic phenomena, in-
cluding topics widely discussed in cognitive linguistics (polysemy and metaphor)
and lexical semantic topics that have not generally been examined by cognitive
linguists (namely lexical relations such as antonymy, meronomy and hyponymy).
The third major hypothesis of the cognitive linguistic approach is that knowl-
edge of language emerges from language use. That is, categories and structures
4 Introduction
in semantics, syntax, morphology and phonology are built up from our cogni-
tion of specific utterances on specific occasions of use. This inductive process of
abstraction and schematization does not lose the conventionalized subtleties and
differences found among even highly specific grammatical constructions and word
meanings.
As we noted above, this hypothesis is a response to approaches to syntax and
semantics in which highly general and abstract schemas and categories, sometimes
claimed to be innately given, are assumed to govern the organization of linguistic
knowledge, and apparently idiosyncratic or anomalous patterns are relegated to
the periphery. Instead, cognitive linguists argue that the detailed analysis of subtle
variations in syntactic behavior and semantic interpretation give rise to a different
model of grammatical representation that accommodates idiosyncratic as well as
highly general patterns of linguistic behavior (see, e.g., the arguments in chapter 9).
In semantics, this model is manifested in Fillmore’s semantics of understanding
(chapter 2), and Cruse’s dynamic construal approach to categorization (chapter 4
and Part II; see also Croft 2000:99–114). In syntax, this hypothesis has given rise
directly to construction grammar as a new theory of syntax, and the usage-based
model, developed in greatest detail for morphology and phonology. These models
of syntax and morphology are described in Part III of this book.
PA RT I
A conceptual approach to
linguistic analysis
2
Frames, domains, spaces: the
organization of conceptual structure
2.1 Arguments for frame semantics
What is it that words denote, or symbolize as cognitive linguists usually

put it? A simple assumption that has guided much research in semantics is that
words denote concepts, units of meaning. Concepts symbolized by words such
as stallion and mare can be compared and contrasted with one another. Compar-
isons of words is the approach taken by structural semantics, which analyzes
types of semantic relations among words, including hyponymy and antonymy.
Some approaches to (lexical) semantics have proposed that word concepts such
as STALLION and MARE1 are not atomic. Many concepts can be broken down
into semantic features, so that STALLION is [EQUINE, MALE], and MARE is
[EQUINE, FEMALE]. Finally, in the logical tradition that underlies much work
in semantics, concepts are ultimately defined by their truth conditions: the con-
ditions under which one can say that a concept does, or does not, appropriately
apply to a situation in the world.
In this widespread approach to semantics, it is recognized that concepts do not
simply float around randomly in the mind. First, there are the relations between
words and their corresponding concepts described by structural semantics. But
there has been a strong feeling that concepts are organized in another way as well.
Certain concepts ‘belong together’ because they are associated in experience. To
use a classic example (Schank and Abelson 1977), a RESTAURANT is not merely
a service institution; it has associated with it a number of concepts such as CUS-
TOMER, WAITER, ORDERING, EATING, BILL. These concepts are not related
to RESTAURANT by hyponymy, meronymy, antonymy or other structural seman-
tic relations; they are related to RESTAURANT by ordinary human experience.
The concept of RESTAURANT is closely tied to the other concepts, and cannot
be isolated from the other concepts.
1 We follow the practice of Fillmore (1982a) and Langacker (1987) in using lower-case italics to
represent the word form, and capitals to represent the concept underlying the word meaning.
7
8 A conceptual approach to linguistic analysis
The need for another means for organizing concepts has been felt by researchers
in cognitive psychology and artificial intelligence as well as in various branches
of linguistics, and has led to a variety of similar proposals, each typically with its
own name. Among these names are: frame, schema, script, global pattern, pseudo-
text, cognitive model, experiential gestalt, base, scene (Fillmore 1985:223, n. 4).
The most influential version of this proposal in cognitive linguistics has been the
model of frame semantics developed by Fillmore. We present Fillmore’s theory
and arguments in this section, and turn to extensions of Fillmore’s ideas by other
cognitive linguists in later sections.2
Fillmore views frames not as an additional means for organizing concepts, but as
a fundamental rethinking of the goals of linguistic semantics. Fillmore describes his
frame semantic model as a model of the semantics of understanding, in contrast
to a truth-conditional semantics: the full, rich understanding that a speaker intends
to convey in a text and that a hearer constructs for that text. Fillmore argues that in
the analysis of linguistic meaning, understanding is the primary data; truth-value
judgments and judgments of semantic relations such as synonymy and implication
are derivative and theory-driven (Fillmore 1985:235). Fillmore’s frame semantics
brings linguistic semantics back to that primary data and does not exclude any of
it from consideration.
Fillmore uses a tool metaphor to describe the understanding process (Fillmore
1982a:112): a speaker produces words and constructions in a text as tools for a
particular activity, namely to evoke a particular understanding; the hearer’s task
is to figure out the activity those tools were intended for, namely to invoke that
understanding. That is, words and constructions evoke an understanding, or more
specifically a frame; a hearer invokes a frame upon hearing an utterance in order
to understand it.
Fillmore uses a wide range of examples to demonstrate that there are signifi-
cant phenomena in linguistic semantics that cannot easily be captured in a model
of structural semantics, semantic features and/or truth-conditional semantics. We
survey his arguments here.
The analysis of semantic features is often justified on the basis of lexical sets
that appear to be analyzable in terms of a simple set of features. For example,
the lexical set in (1) can be analyzed in terms of the features [MALE/FEMALE],
[ADULT/YOUNG], and [UNMARRIED]:
(1) [MALE] [FEMALE]

MAN WOMAN [ADULT]
BOY GIRL [YOUNG]
BACHELOR SPINSTER [UNMARRIED]
2 The basic sources for Fillmore’s ideas are Fillmore 1975, 1977 (an expanded version of the first
paper), 1982a, 1985, 1986. Unfortunately all of these are difficult to access.
Frames, domains, spaces 9
Yet our understanding of these concepts is more complex than this paradigm of
feature constrasts implies. The relation between man/boy and woman/girl is not the
same: for many people, the term girl is used for female humans at a significantly
higher age than the term boy is used for male humans (Fillmore 1982a:126).
Moreover, the attitudes towards the sexes that this linguistic behavior is assumed
to evoke has led to changes in the relationship and hypercorrection such that
the term woman is attested as being applied even to an eight-year-old girl (ibid.,
127). In a frame semantic analysis, man, boy, woman and girl evoke frames that
include not just the biological sexual distinction but also differences in attitudes
and behavior towards the sexes that would explain the traditional asymmetry in
the use of boy/girl and the more recent change in the use of woman, including
its hypercorrective use. Likewise, the difference between our understanding of
bachelor and our understanding of spinster involves much more than a simple
feature [MALE/FEMALE] (ibid., 131).
Many lexical contrasts contain semantic asymmetries that cannot be captured
by features (except in an ad hoc fashion), but lend themselves easily to a frame
semantic account. For example, the opposing terms used for the vertical extent of
an erect human being are tall and short, for vertical distance from a bottom baseline
(e.g. a branch of a tree) they are high and low, but for the vertical dimension of
a building they are tall and low (Fillmore 1977a:71). It would be difficult if not
impossible to come up with a unitary feature definition of tall that captured its
different contexts of use from high, and did the same for short vs. low. Instead,
one can simply describe the frames for humans, buildings and other objects, and
specify which words are used for vertical extent or distance in that frame.
Similarly, no simple unitary definitions would capture the contrast between the
adjectives live and alive given in (2)–(4) (Fillmore 1977a:76–77):
(2) a. Those are live lobsters.

b. Those lobsters are alive.
(3) a. Her manner is very alive.
b. She has a very alive manner.
(4) a. His performance was live.
b. He gave a live performance.
Moreover, one cannot define the features in terms of applicability to a semantic

class, such that the sense illustrated in (2a–b) applies to living things; this would
give an incorrect understanding to the theatre advertizing live naked girls than
the one intended (presumably, as opposed to naked girls on a film screen, not
dead naked girls; ibid.). In a frame semantic analysis, live and alive are simply
associated in different ways to three different frames: life in (2), personality in (3),
and mode of performance in (4). In other cases, there are outright lexical splits,
such as brother/brothers and brother/brethren, which represent a split in frames
including different plural forms; a unitary definition of brother would miss the
frame contrast (ibid., 76).
Fillmore notes that his frame semantic model shares significant properties with
lexical (semantic) field theory (Fillmore 1985:225–26; 1992:76–77). Lexical field
theory groups together words that are associated in experience, not unlike frame
semantics. However, lexical field theory differs from frame semantics in that words
are defined relative to other words in the same lexical field, whereas in frame
semantics, words are defined directly with respect to the frame. For example, in
lexical field theory, one would observe that large in the field of sizes of packages
of soapflakes is in contrast with jumbo, economy giant and family size and hence
describes the smallest size in the field, unlike uses of large in other lexical fields
(Fillmore 1985:227).
In frame semantics, the same observation can easily be captured: large labels
the smallest size in the SOAPFLAKES frame. But lexical field theory predicts that
the meaning of a word in a field can only be defined in contrast to neighboring
words in the field. Lexical field theory has difficulties if there are no neighboring
words, or a speaker does not know the neighboring words: it predicts that the term
has a different meaning. Fillmore notes that while German has a word for the
sides of a right angle triangle other than the Hypotenuse, namely Kathete, most
English speakers do not have such a word (ibid., 228–29). Yet the understanding
of English hypotenuse and German Hypotenuse is the same, provided the speaker
understands what a right angle triangle is. This is not a problem in frame semantics,
where the word concept is linked directly to the frame, in this case the RIGHT
ANGLE TRIANGLE frame.
Another argument in favor of a frame-based approach to lexical semantics are
words whose corresponding concepts inherently refer to other concepts extrinsic
to the concept denoted by the word. Some word concepts refer to a prior history of
the entity denoted. A scar is not just a feature of the surface of someone’s skin, but
the healing state of a wound; a widow is a woman who was once married but whose
husband has died (Fillmore 1977a:73). Other word concepts, especially for prop-
erties and actions, cannot be understood without understanding something about
the participant in the action or possessor of the properties: one cannot understand
gallop without knowing about the body of a horse, or hungry without understand-
ing the physiology of living things (ibid., 73–74). This is true of object concepts
as well: lap cannot be understood except in reference to a person’s posture and the
function of one’s lap in supporting another object (ibid.).
Another clear class of examples that requires reference to extrinsic entities are
deictic expressions that evoke the speech act situation (Fillmore 1982a:117). For
example, the past tense situates an event in a point or interval or time relative to
the speech act situation. The speech act situation, including its time of occurrence,
functions as the frame against which past time reference is profiled. Likewise, all
other deictic words and inflections, such as person deixis (I, you, he/she/it, we, they
and person-based agreement inflections) and spatial deixis (this, that, here, there),
evoke the speech act situation. Other types of grammatical words and inflections
also have meanings evoking the speech act situation. For example, the definite
articles the and a define the identity of the noun referent relative to the mutual
knowledge of speaker and hearer (the basically indicates mutually known, a not
mutually known, in most contexts). The meanings of the and a evoke the speech
act situation because they make reference to the mental states of speaker and hearer
(see also §3.4).
Above all, many word concepts cannot be understood apart from the intentions
of the participants or the social and cultural institutions and behavior in which
the action, state or thing is situated. For example, the concept VEGETARIAN
only makes sense in the frame of a culture in which meat-eating is common; the
concepts STRIKE or BORROW can only be understood in the frame of a culture
in which such actions occur (Fillmore 1982a:120). Even something as simple as
an apple core evokes a frame describing a particular way of eating apples: ‘an
apple-core is that part of the apple that somebody who eats apples the way most
of us do has left uneaten’ (Fillmore 1977a:73).
Another respect in which a word meaning makes reference to extrinsic entities is
that a word allows the speaker and hearer to focus their attention on only part of an
entire frame; no one word gives the full structure of the frame. The classic example
is the commercial transaction frame (Fillmore 1977a:58–59; 1977b); but a much
clearer case is the RISK frame (Fillmore and Atkins 1992). Fillmore and Atkins
identify the following elements of the RISK frame: Chance (uncertainty about the
future), Harm, Victim (of the Harm), Valued Object (potentially endangered by the
risk), Situation (which gives rise to the risk), Deed (that brings about the Situation),
Actor (of the Deed), (Intended) Gain (by the Actor in taking a risk), Purpose (of
the Actor in the Deed), Beneficiary and Motivation (for the Actor). The verb risk
occurs in many syntactic constructions, some of which are exemplified in (5a–e),
but none of them include all or even most of the elements of the RISK frame
(Fillmore & Atkins 1992: 83, 87, 89, 94, 96; all but the first are corpus examples):
(5) a. You’ve (Actor/Victim) risked your health (Valued Object) for a few cheap thrills
(Gain).
b. Others (Actor/Victim) had risked all (Valued Object) in the war (Situation).
c. She (Actor/Victim) had risked so much (Valued Object) for the sake of vanity
(Motivation).
d. Men (Actor/Victim) were not inclined to risk scalping (Harm) for the sake of
settlers they had never seen (Beneficiary).
e. I (Actor/Victim) didn’t dare risk a pause (Deed) to let that sink in (Purpose).
In a frame semantic analysis, any of the uses of risk evokes the entire RISK frame,
even if only part of that frame is overtly focused on by the construction in which
risk is used.
The semantics of understanding also allows Fillmore to account for linguistic
facts that do not lend themselves to a truth-functional analysis. For example, the
collocations in (6) could be reversed as in (7) without producing semantic anomaly
(Fillmore 1977a:75–76):
(6) a. A dog was barking.

b. A hound was baying.
(7) a. A dog was baying.
b. A hound was barking.
In other words, the difference between (6) and (7) cannot be accounted for by
semantic constraints. But the examples in (6a–b) sound much more natural because
the noun and the verb in each sentence both evoke the same frame.
Likewise, a truth-conditional semantics cannot capture many aspects of our
understanding of (8) (Fillmore 1985:230–31):
(8) My dad wasted most of the morning on the bus.
Fillmore notes that choosing father or dad (without the possessive) would express
a different relationship between the speaker and the speaker’s father; the morning
is understood to be defined against the frame of the working day (i.e, around 8am
to noon) rather than the calendar day (midnight to noon); waste frames the use of
time very differently from spend; and on the bus frames the speaker’s location in
terms of the bus being in service, rather than simply a physical container (which
would be evoked by in the bus).
A truth-conditional model also cannot account for the anomaly of frames that
are appropriate at one time of utterance but not at another because the world has
changed in the meantime. Fillmore uses the contrived example in (9), noting that
it could be said in 1984 but not in, say, 1919 (Fillmore 1985:238–39):
(9) During World War I, Ronald Reagan’s birth mother dropped his analog watch
into the sound hole of the acoustic guitar.
Such a sentence could be uttered in 1984, because World War II had occurred,
allowing the 1914–18 war to be renamed World War I; medical technology had
allowed the dissociation of the birth mother from the genetic mother (who donates
the egg); and electric guitars and digital watches had been invented. None of these
framings of the objects, persons or events was available in 1919, and so (9) would
be an impossible utterance at that time, even if true retrospectively.
Finally, frame semantics offers a natural account of a number of problematic
phenomena that seem to be caught between semantics and pragmatics, including
the nature of text coherence. A large class of presuppositional phenomena appear

to be tied to specific words, such as regret in (10) (Fillmore 1985:249):
(10) John regretted signing the letter.
Example (10) denotes (or entails) a particular mental state of John, namely his
feeling of regret, but (in one analysis) is said to presuppose that John did sign
the letter. The problem for truth-conditional semantics is that if John did not sign
the letter, (9) has no truth conditions. In order to preserve truth conditions one
may shunt the presupposition problem off to pragmatics. This seems odd since
the presupposition is associated with a particular word and its meaning, which
is semantic. But presuppositions display a further type of peculiar behavior, in
negative sentences such as (11):
(11) John didn’t regret signing the letter.
In one interpretation, the entailment is negated – John does not have any regrets –
but the presupposition is not – John signed the letter. However, there is another
interpretation of (11), namely that John did not regret signing the letter because
he did not sign it (Fillmore 1985:251).
Fillmore argues that the behavior of presuppositions can be easily accounted
for in a frame semantic analysis. The concept REGRET includes in its frame the
accomplishment of an action towards which the regretter has his/her regrets. If
the action is absent from the frame, understanding of the positive sentence fails.
Negation, on the other hand, can negate either the concept denoted or the frame
itself. Negating the state of affairs in the frame preserves the rest of the frame,
including the action that could have led to the regrets. This is the first interpretation
of (11) described above. Negating the entire frame, on the other hand, also negates
the action that could have led to the regrets. This is the second interpretation of
(11): the speaker denies the framing of the situation as including the action of John
having signed the letter.
We may compare (11) to other examples of frame negation, such as the one in
(12) (Fillmore 1985:245, from Wilson 1975:138):
(12) You didn’t spare me a day at the seaside: you deprived me of one.
In (12), the speaker denies the positively evaluated framing of the action as sparing
her, and replaces it with the negatively evaluated framing of depriving her.
Finally, the semantics of understanding plays a major role in text understand-
ing. For example, the well-known example of initial definite reference, e.g. the
carburetor in (13) (described as ‘bridging’ by Clark and Haviland [1977] and
‘evoked’ by Prince [1981a]), are in fact due to the frame evoked by the first
sentence (Fillmore 1977a:75):
(13) I had trouble with the car yesterday. The carburetor was dirty.
The car in the first sentence evokes a frame that allows the hearer to identify which
carburetor of the millions in the world the speaker was referring to. But frame
semantics contributes more than the resolution of definite reference to the analysis
of the coherence of texts. Fillmore contrasts (13) with (14) (ibid.):
(14) I had trouble with the car yesterday. The ashtray was dirty.
The second sentence in (14) is incoherent with the first, even though the definite
reference can be resolved (most cars have ashtrays). The reason for this is that there
is nothing evoked in the frame of having trouble with the car that has anything to
do with the ashtray – unlike (13), because dirty carburetors do cause problems for
cars.
Fillmore’s arguments present a wide range of data that justify the introduction
of frames to the analysis of linguistic semantics, and the replacement of a truth-
conditional semantics with a semantics of understanding. In the following sections,
we lay out more systematically the frame semantic model and follow its further
development in cognitive linguistics.
2.2 Concepts: profile-frame organization
In the preceding section, we described the frame as a coherent region

of human knowledge, or as a coherent region of conceptual space. The question
immediately arises: How does one identify a coherent region of conceptual space,
differentiating it from other regions? An a priori approach to this question, us-
ing one’s own intuitions to identify frames, would be highly subjective. A more
empirical approach to this question is to identify frames based on the words and
constructions of a human language such as English. This approach is taken by
Langacker (1987), which we will use as our starting point.
Langacker illustrates his approach to the problem with the meaning of the word
radius. The word form radius symbolizes (denotes) the concept RADIUS. We
begin here by assuming that concepts correspond to meanings of linguistic units
(words, complex expressions or constructions). One may also assume that concepts
exist that do not correspond to linguistic meanings. However, one would have the
same problems trying to identify concepts independent of linguistic meanings as
trying to identify frames independent of linguistic meanings, namely the lack of an
empirical basis for doing so. For this reason, we will restrict ourselves to concepts
corresponding to actual linguistic meanings in this chapter.
The first sense for radius in the American Heritage Dictionary is ‘a line segment
that joins the center of a circle with any point on its circumference.’ A RADIUS is
a line segment, but not any line segment: the line segment is defined relative to the
structure of the circle. In other words, one can understand RADIUS only against
a background understanding of the concept CIRCLE, which can be geometrically
illustrated as in Figure 2.1.
RADIUS
CIRCLE
Figure 2.1 RADIUS and CIRCLE
In other words, the concepts RADIUS and CIRCLE are intimately related, and
this relationship must be represented in conceptual structure. Langacker describes
the relationship between RADIUS and CIRCLE as one of a concept profile against
a base. The profile refers to the concept symbolized by the word in question. The
base is that knowledge or conceptual structure that is presupposed by the profiled
concept. Langacker also uses the term domain for the base (this term is also used
in Lakoff 1987). This is identical to Fillmore’s frame (§2.1): ‘by the term “frame”
I have in mind any system of concepts related in such a way that to understand any
one of them you have to understand the whole structure in which it fits’ (Fillmore
1982a:111). The term ‘profile’ has also come to be used as a verb to describe
the relationship between word form and word meaning (profile+base): e.g. radius
profiles a particular line segment in the CIRCLE base/domain/frame.
A concept profile is insufficient to define a word concept, because it presup-
poses other knowledge in its definition, namely its base. But a single base, such
as CIRCLE, is a complex conceptual structure that includes a wide range of con-
cept profiles, such as RADIUS, ARC, CENTER, DIAMETER, CHORD and so
on. Hence the base alone is insufficient to define a linguistic concept either. The
conclusion that follows from this is that the meaning of a linguistic
unit must specify both the profile and its base. This is identical
to Fillmore’s conclusion regarding concept frames.
The fact that a base supports multiple concept profiles is what makes the base a
domain, in the intuitive sense: several different concept profiles have it as a base.
We can now define a domain as a semantic structure that functions
as the base for at least one concept profile (typically, many
profiles). As Taylor (1989[1997]:84) notes, ‘In principle, any conceptualization
or knowledge configuration, no matter how simple or complex, can serve as the
cognitive domain for the characterization of meanings.’ We may now say that the
domain CIRCLE includes the concepts of an arc, a diameter, a radius, a chord and
so on.
The canonical example of a profile-base relation is the part-whole relation: all
agree that a concept such as ARM cannot be defined without reference to BODY. A
similar class of concepts are kin terms such as daughter. The concept DAUGHTER
presupposes the concept PARENT, and the particular type of kin relationship that
holds between them. The concept NIECE presupposes other kinship concepts, and
more complex kin relationships. In other words, the base against which a profile is
defined can be more complex than just the whole of which some entity is a part. In
some cases, one cannot always find a single simple word to describe the base: for
NIECE, perhaps the best description of the base is KINSHIP SYSTEM, or some
part of that system (see §2.4).
But it is not only relational nouns that represent a concept profile against a
base, as we saw in §2.1. Consider another example, the word weekend (Fillmore
1985:223–24). The concept WEEKEND can only be understood against a whole
background system of the calendrical cycle, defined partly by natural phenomena
(the sequence of day and night) and cultural conventions (the seven-day week
cycle, and its division into working days and nonworking days). Likewise, the
concept BUY can only be understood against a background knowledge of the
commercial transaction situation. Different aspects of the commercial transaction
are profiled by BUY, SELL, PAY, COST and so forth. Such domains/frames cannot
be readily represented in a geometric form in the way that RADIUS and CIRCLE
are represented in Figure 2.1, although schematic diagrams are often resorted to
in cognitive linguistics in order to represent the complex interconnectedness of
concepts in domains or frames.
In fact, no concept exists autonomously: all are understood to fit into our general
knowledge of the world in one way or another. What matters for semantic analysis is
the profile-base relation, and the relationships between bases and domains. Some of
the corollaries of this analysis of word meaning into profile and base/frame/domain
will be explored in the following section.
2.3 Some consequences of the profile-frame/domain

distinction
The terms frame (Fillmore), base (Langacker) and domain (Fillmore,

Lakoff, Langacker) all appear to identify the same theoretical framework, as de-
scribed in the preceding sections. Fillmore describes this framework as frame
semantics, and this term has entered into more general usage among cognitive
linguists. However, the terms frame and domain continue to compete for usage,
and base is also used among cognitive grammarians. We will use the terms frame
and domain interchangeably here. Nevertheless, there are still other terms that have
been proposed to describe types of semantic analyses that bear a strong affinity
to frame semantics. We mention three influential theories here, which originated
in artificial intelligence (scripts), cognitive psychology (the ‘theory theory’) and
sociology (communities).
The examples of frames given above appear to be largely static in character.
But this is not necessary: a frame is any coherent body of knowledge presupposed
by a word concept. In particular, frames can include dynamic concepts, that is,
extending through time. For example, PURIFIED presupposes in its frame a prior
impure state of the entity which is then changed by some process; in contrast,
PURE does not presuppose anything about prior states and processes. Of course,
process terms such as RUN or BUY presuppose a sequence of events and prior and
posterior states. The term script is often used for a frame/domain with a sequence
of events, following Schank and Abelson (1977). They use the term to describe a
canonical sequence of events presupposed by a social activity such as going to a
restaurant. We subsume scripts under frames/domains.
Another theoretical construct that can be understood as a type of frame or domain
is the so-called ‘theory theory’ of categorization found in cognitive psychology.
Advocates of the theory theory argue that our understanding of categories such
as HORSE or HAMMER is based not on perceptual features but on theories of
biological kinds and artifacts respectively (Murphy and Medin 1985). For instance,
we have at least a folk theory of biological kinds that indicates that individuals
of the same category (e.g. HORSE) are members of that category by virtue of
descent and reproduction, and perceptual similarity of horses (and the distinctness
of individuals of other species) are a result of those basic biological patterns.
Likewise, hammers are defined by the fact that they are manufactured by human
beings for a particular function, and perceptual similarity of hammers (and the
distinctness of other kinds of artifacts) are a result of their intended function. In
frame semantic terms, the base for HORSE includes the ‘theory’ of biological
kinds and the base for HAMMER includes the ‘theory’ of artifacts (see Fillmore
1986a:54).
Fillmore also uses the notion of framing to describe differences in the com-
munity or social domain of use of a word (Fillmore 1982a:127–29). For exam-
ple, he notes that in the legal domain, that is, the community that engages in
legal activity, the concepts of MURDER and INNOCENT differ from those con-
cepts used outside that domain/community. In the legal domain, MURDER is pro-
filed in a frame/domain where it contrasts with MANSLAUGHTER, but outside
that domain, MURDER is profiled in a domain lacking that contrast. In the legal
domain, INNOCENCE is profiled against a frame in which innocence and guilt are
the result of judgements in a trial (and in fact, guilt can be established only after
the completion of the trial). Outside that domain, INNOCENT is profiled against a
frame in which innocence and guilt are defined by whether the person in question
committed the crime or not. Other concepts such as FLIP STRENGTH exist only
in a specialized community, in this case publishers of pornography (the interested
reader may turn to Fillmore 1982a:12 for further details). Hence, frame semantics
is being extended to describe differences that appear to be defined on social rather
than conceptual grounds. But there is a link between the two. Communities are
defined by the social activities that bind the members together. Clark argues that
communities involve the possession of shared expertise among their members: the
specialized knowledge that is acquired by engaging in the activities that define the
community (Clark 1996:102–4). This shared expertise is the conceptual structure
that is found in the frame/domains of the concepts symbolized by the specialized
vocabulary used by members of the community.
The distinction between profile and frame/domain is a useful tool for analyzing a
number of interesting semantic questions. In particular, some distinctions in word
meaning apply not to the profiled concept – what is usually thought of as ‘the
definition’ of a word – but to its frame/domain.
For example, some concepts appear to denote the same thing in the world but
profile it against a different frame. For example, LAND and GROUND denote
(profile) what seems to be the ‘same thing,’ but against different frames: LAND
describes the dry surface of the earth in contrast with SEA, while GROUND de-
scribes the dry surface of the earth in contrast with AIR (Fillmore 1982a:121).
The frame chosen by one word or another allows one to make different inferences:
Fillmore notes that a bird that spends its life on land does not go in the water, but
a bird that spends its life on the ground does not fly (ibid.). Langacker offers the
example of ROE and CAVIAR, both being fish eggs: ROE is profiled against the
frame/domain of the reproductive cycle of fish, while CAVIAR is profiled against
the frame/domain of food preparation/consumption (Langacker 1987:164–65).
Another example is FLESH, profiled against the frame/domain of the body’s
anatomy, vs. MEAT, profiled against the frame/domain of food. The semantic
difference is reflected in the collocations flesh and bones, describing an emaciated
body, and meat and potatoes, describing a bland but filling type of meal (contrast
meat and bones and flesh and potatoes).
The alternative framing of the same profile is particularly common with terms
that are evaluative in character. For example, STINGY profiles one end of a scale,
the opposite of which is GENEROUS; while THRIFTY appears to profile the
same end of the same scale, and its opposite end is profiled by WASTEFUL
(Fillmore 1982a:125). The difference is the orientation of the associated evaluative
scale: the evaluation of STINGY-GENEROUS is the inverse of that for THRIFTY-

WASTEFUL. Of course, a speaker may choose to frame someone as either
STINGY or THRIFTY. In other words, how an experience is framed is a mat-
ter of construal: it depends on how the speaker conceptualizes the experience to
be communicated, for the understanding of the hearer. This is only one example of
the construals that pervade human conceptualization of experience (see chapter 3).
Another type of evaluative framing effect is more indirect, as in the example
FETUS vs. UNBORN BABY, terms used by opposing sides of the debate on
abortion. FETUS profiles the entity in question against a more general MAMMAL
frame: any mammal’s unborn progeny may be called a fetus. This frame makes
abortion appear less morally repugnant, since it is widely accepted in society that
animals can be killed for certain purposes. The complex phrase UNBORN BABY
exploits two frames. BABY profiles the same entity against the more specific
HUMAN frame: we prototypically use baby only for human offspring. Both BABY
and UNBORN profile the entity against its projected later lifestage, namely after
birth. These frames make abortion appear more repugnant, since killing humans is
accepted only under quite restricted circumstances (e.g. war and self-defense), and
all agree that once a fetus is born, it is a human being. The difference in framing
the entity denoted by fetus or unborn baby therefore orientates (or biases, to frame
it differently!) the hearer towards the political stance on abortion adopted by the
speaker.
The above examples all illustrate different words that profile the same concept
but in subtly different frames. There are also examples where a single word is
usually analyzed as polysemous – having distinct albeit related meanings – but
where those meaning differences are more due to differences in frame rather than
differences in profile. For example, a word such as mouth describes roughly the
same concept profile but with different frames:
(15) mouth: BODY, BOTTLE, CAVE, RIVER
In the examples of frames for mouth in (15), mouth can be thought of as denoting
the same type of profile, namely the opening to a container (however, a cave may
have several openings to the earth’s surface, and the container of a river is defined
by both the riverbed and gravity). The word mouth is generally considered to be
polysemous, that is, it has a sense for each of the profile-base pairings (senses that
may not share the same word in other languages). In other words, the profile alone
is insufficient in defining the senses of mouth.
The profile-frame/domain distinction is particularly useful in understanding
the nature of semantic differences between words and their apparent translation
equivalents in different languages. The profile-frame/domain distinction may shed
light on some aspects of why translation is difficult and often unsuccessful.
One can find frame-based semantic contrasts across languages that are similar to
those found within a single language such as LAND/GROUND. Fillmore contrasts
the English word concept LUKEWARM with the Japanese word concept NURUI.
Both concepts profile the state ‘at room temperature,’ but for English speakers
LUKEWARM is used for liquids that are ideally hot or ideally cold, whereas
for some Japanese speakers NURUI is used only for liquids that are ideally hot
(Fillmore 1982a:121).
Sometimes linguistic differences across languages represent differences in how
much information is specified in the frame, rather than something about the inherent
structure of the profiled concept. English river profiles a more or less permanent
flow of water in a natural channel, presupposing a frame specifying the relevant
topographical features. French contrasts rivière and fleuve. The concepts for both
French words is essentially the same as the concept for the English word – a natural
flowing waterway – but the frame is more specific than the English frame for each
word: fleuve specifies in addition that it is a major waterway that flows into the sea,
unlike rivière. Hence a translation of fleuve as river is partly accurate (the profile),
but not completely so (its frame). Another example is English eat, which profiles
the process of consuming food. German contrasts essen and fressen: both describe
the process of consuming food also, but the former specifies that the eater is human
and the latter that the eater is an animal (nonhuman). This is a framing effect and
is therefore subject to construal (§3.3). That is, one can use the term fressen to
describe the action of a human being, but it frames that action much differently
than essen, leading to a construal of the action as being animal-like (crude, sloppy,
etc.).
The nature of word meaning across languages is sometimes obscured by analysts
who do not distinguish between profile and frame in their word definitions. For
example, some languages are described as having words that correspond to whole
sentences in English. Two candidate examples are given in (16)–(17), the first from
a native American language and the second from a European language:
(16) Alabama ispaspaakáhmit ‘to be shaped into a patty, shaped like a biscuit (said
of the shape of the mixture of brains and moss used for curing hides)’ (Sylestine
et al. 1993:203)
(17) Swedish tura – ‘sitting on the boat going back and forth between Helsingborg
and Helsingør’ (Karina Vamling, pers. comm.)
By distinguishing profile from frame/domain, we may give a more straightfor-

ward semantic analysis. The concepts profiled are pretty simple: ispaspaakáhmit
profiles a shape and tura profiles sitting. But the frames in which the concepts are
situated are very specific (in [17], staying on the boat means paying only one fare
and drinking duty-free alcohol). A similar English example would be genuflect.

Genuflect profiles more or less the same concept as kneel, a bodily movement –
but in a highly specific frame, namely Catholic religious practice. But even less
culturally specific word concepts can be fruitfully divided into profile and frame.
A radius profiles a line segment; the difference between radius and line segment
is found in their frames.
The profile-frame distinction also allows for a natural analysis of words in
languages that have been described as ‘untranslatable’ in a more profound way
than the examples of fleuve or fressen given above. For example, Geertz gives a
lengthy explication of Javanese rasa, part of which is quoted below:
Rasa has two primary meanings: ‘feeling’ and ‘meaning’. As ‘feeling’ it is one
of the traditional Javanese five senses – seeing, hearing, talking, smelling and
feeling, and it includes within itself three aspects of ‘feeling’ that our view of
the five senses separates: taste on the tongue, touch on the body, and emotional
‘feeling’ within the ‘heart’, like sadness and happiness. The taste of a banana is
its rasa; a hunch is a rasa; a pain is a rasa; and so is a passion. As ‘meaning’,
rasa is applied to words in a letter, in a poem, or even in common speech to
indicate the between-the-lines type of indirection and allusive suggestion that is
so important in Javanese communication and social intercourse. And it is given
the same application to behavioral acts generally: to indicate the implicit import,
the connotative ‘feeling’ of dance movements, polite gestures, and so forth. But
in this second, semantic sense, it also means ‘ultimate significance’ – the deep-
est meaning at which one arrives by dint of mystical effort and whose clarifi-
cation resolves all the ambiguities of mundane existence [etc.]. (Geertz 1973:
134–35)
Basically, understanding the meaning of rasa presupposes understanding large

portions of Javanese culture and worldview. In frame semantic terms, the concept
RASA presupposes a frame consisting of much of Javanese culture. Examples of
this type can also be found closer to home. For example, a translator dealing with
a twentieth-century German philosophical work, writes about the term Bildung:
‘Bildung is translated by “culture” and related forms such as “cultivation”, “culti-
vated” . . . The term has the flavor of the late-eighteenth and nineteenth centuries
and played a key role throughout German-speaking Europe’ (Gadamer 1989:xii).
That is, BILDUNG is profiled against a frame of the culture of the German intel-
lectual elite stretching back almost two centuries.
In other words, the reason that words such as rasa and Bildung are ‘untranslat-
able’ is because of the culture-specific character of the frame/base against which
the concept is profiled. Translating rasa as feeling or meaning, or Bildung as
culture, approximates the profile of the concept but does not have the same frame
at all.
2.4 Extensions of the basic profile-frame/domain distinction
The distinction between profile and domain/frame is a fundamental one

in the theory of semantics used in cognitive linguistics. It has nevertheless proved
to be insufficient in itself to capture a number of important semantic phenomena,
and the basic theory has been elaborated in several directions.
2.4.1 Locational and configurational profiles

One extension of the frame semantic model recognizes two different kinds
of profiles. Consider again the SPACE domain. A concept like RECTANGLE is
profiled in the SPACE domain. Note that an octagon is an octagon wherever it is lo-
cated in space. What matters for the profile of RECTANGLE is simply the number
and configuration of line segments forming the sides. The profile for RECTAN-
GLE contrasts with the profile of a spatial concept such as HERE. HERE profiles
a location in SPACE, one that is defined with respect to the position of the speaker.
You cannot move the profiled location without changing the concept. The same
constraint applies to a concept like MOUNT TAMALPAIS. This concept also pro-
files a location in SPACE; another mountain in another location is not, nor ever will
be, MOUNT TAMALPAIS (in contrast to MOUNTAIN, which is a topographi-
cal configuration that can be located anywhere). These are two different kinds
of profiles: RECTANGLE has a configurational profile and HERE or MOUNT
TAMALPAIS has a locational profile (Langacker 1987:153; Clausner and Croft
1999:7–13).3
Not every frame/domain can support both kinds of profiles. Color words, for
example, specify regions in the HUE scale; if one moves to a different location on
the HUE scale, then the concept changes, for example from RED to YELLOW.
But there is no configurational profile on the HUE scale. This is not a fact about all
one-dimensional scalar domains. For example, the domain of (musical) PITCH has
both locational and configurational profiles. For example, particular notes such as
C-SHARP, or more precisely one specific note such as C# (the C-sharp an augmen-
ted octave above middle C), profiles a single location on the pitch scale. However,
a musical interval such as OCTAVE has a configurational profile: an OCTAVE
is an octave wherever it occurs as long as the pitch interval is correct (Clausner
and Croft 1999:10). More generally, measurable one-dimensional scalar domains
such as PITCH, LENGTH and so on allow for both locational and configurational
3 Langacker argues that the locational-configurational distinction applies to domains, but Clausner and
Croft demonstrate that the same domain can support locational and configurational profiles.
profiles. Antonymic adjectives such as TALL/SHORT (see chapter 7) profile a

particular location or direction on a scalar domain. Units of measurement on the
scale, such as INCH or FOOT, profile configurations: an inch is the same interval
no matter the locations subsumed under the measured interval.
2.4.2 Scope of predication

In §2.2, we used the example of NIECE, demonstrating that its proper
definition presupposes the system of kinship relations. But we do not need the
entire kinship system in order to understand the concept NIECE. Only a small part
of it is necessary as represented in Figure 2.2 (the gender-neutral square symbol
is used because the intervening kin for NIECE may be male or female).
= EGO
NIECE
Figure 2.2 NIECE
The relevant part of the kinship system for defining NIECE is called the scope of
predication for the concept (Langacker 1987:118–19; renamed immediate scope
[Langacker 1999:49]).
An example of different scopes of predication can be found in the behavior of
human body parts such as the following parts of the arm (Langacker 1987:119):
KNUCKLE ⊂ FINGER ⊂ HAND ⊂ ARM ⊂ BODY. Each one has its imme-
diate successor as its scope of predication. Possessive constructions referring to
wholes within the scope of predication are acceptable, but if the whole is beyond
the scope of predication, then the sentence is odd (Langacker 1987:119; but see
§6.2.1.7):
(18) a. A body has two arms.

b. A hand has five fingers.
c. A finger has three knuckles and a fingernail.
d. ?An arm has five fingers.
e. ??A body has twenty-eight knuckles.
Nested scopes of predications can be generalized to nesting of frames/domains

in general.
2.4.3 Relationships between domains

Much more complex is the elaboration of the relationships among do-
mains – not surprisingly, since this touches on the organization of human knowl-
edge in the mind.
An important fact about profiles and frames/domains is that one can have suc-
cessive chains of profile-frame relations. The concept RADIUS can only be under-
stood in terms of CIRCLE, as noted above. But the concept CIRCLE can itself only
be understood in terms of (two-dimensional) SPACE. That is, the word circle pro-
files CIRCLE against the SPACE frame. In other words, a concept that functions
as the frame/domain for other concepts is itself a profile for another conceptual
frame/domain. In other words, whether a conceptual structure is the profile or
frame/domain is a matter of construal (see §§2.3, 3.2).
The chain of profile-frame relations does eventually bottom out, when we reach
directly embodied human experience. SPACE is a good candidate for a directly
embodied human experience. Langacker calls domains rooted in directly embod-
ied human experience basic domains (Langacker 1987:148); he calls nonbasic
domains abstract domains. A major theme of Lakoff and Johnson’s cognitive lin-
guistic research is that even our most abstract knowledge is ultimately grounded
in directly embodied human experience (Lakoff and Johnson 1980, chapter 12;
Johnson 1987; Lakoff and Johnson 1999). Other examples of basic domains be-
sides SPACE are MATERIAL, TIME, FORCE and a host of perceptual and bodily
sensations (COLOR, HARDNESS, LOUDNESS, HUNGER, PAIN etc.). There
are also emotional and other mental states and processes, and also social proper-
ties, relations and processes, that do not presuppose other domains. Exactly which
mental and social domains are basic depends on one’s theory of mind and social
interaction, and so we will not make any specific proposals here.
The relation between an abstract domain and the basic domain it presupposes
is not a taxonomic relation (or, as Langacker calls such relations, a schematic
one). It is a relationship of concept to background assumption or presupposition.
This distinction is sometimes obscured by the English language. For example, the
word shape as a mass noun stands for the domain, but as a count noun (a shape)
it is a more general or schematic concept subsuming [CIRCLE], [SQUARE],
[TRIANGLE] and so on. A more general or schematic concept is not the domain
for the particular concept; in fact, a schematic concept is itself profiled in the
same domain as its instantiation. As will be seen below, it is not always easy to
distinguish a taxonomic relation from a profile-domain relation.
Langacker argues that some domains involve more than one dimension
(Langacker 1987:150–51). An obvious case is space, which involves three di-
mensions (some concepts such as CIRCLE need only two dimensions for their
definition; others such as LINE need only one). Many physical qualities that are
grounded in the experience of sensory perception, such as TEMPERATURE and
PITCH, are one-dimensional. Others, such as COLOR, can be divided into HUE,
BRIGHTNESS and SATURATION. Generally, dimensions of a domain are all si-
multaneously presupposed by concepts profiled in that domain. This is the critical
point: a concept may presuppose several different dimensions at once.
In fact, a concept may presuppose (be profiled in) several different domains.
For example, a human being must be defined relative to the domains of physical
objects, living things and volitional agents (and several other domains, e.g. emo-
tion). The combination of domains simultaneously presupposed by a concept such
as HUMAN BEING is called a domain matrix. Langacker makes the important
point that there is in principle only a difference of degree between dimensions of a
domain and domains in a matrix (Langacker 1987:152). In practice, we are more
likely to call a semantic structure a domain if there are a substantial number of
concepts profiled relative to that structure. If there are few if any concepts profiled
relative to that structure alone, but instead there are concepts profiled relative to
that structure and another one, then those structures are likely to be called two
dimensions of a single domain. The term ‘domain’ implies a degree of cognitive
independence not found in a dimension (see also §5.3.1).
The domain structure presupposed by a concept can be extremely complex. Let
us now consider how one would define what seems to be a kind of physical object,
the letter T. It is directly defined as a letter of the alphabet; its base (domain) is
hence the alphabet. The alphabet is itself an abstract domain presupposing the
notion of a writing system – it is not just an instance of a writing system, since
the latter involves not just a set of symbols such as an alphabet but also the means
of putting them together, including the direction of letters on a page, spaces for
words and so on. The domain of writing systems in turn presupposes the activity of
writing. The activity of writing must be defined in terms of human communication,
which presupposes the notion of meaning – perhaps a basic domain, since the
symbolic relation appears not to be reducible to some other relation – and visual
sensations, since writing is communication via visually perceived inscriptions,
rather than auditorily or through gestures. And since writing is an activity, the
domains of time and force or causation (both basic domains; see §3.5) are also
involved in the domain matrix of writing, since the letter T is the product of an
activity. Since writing is a human activity, it presupposes the involvement of human
beings. Human beings are living things with mental abilities, such as volition,
intention, cognition and emotion (themselves dimensions of the mental domain or
better domains in the matrix of the domain of the mind). Living things in turn are
physical objects endowed with life. Physical objects possess material existence and
are spatial entities (although material objects always have spatial extent, spatial
objects like geometric figures can exist without material embodiment).
A diagram exhibiting all of the basic-abstract domain relations presupposed
in defining the concept of the letter T is given in Figure 2.3 (based on Croft
1993[2002]:170, Fig. 2.1; the profiled concept is given in boldface, and the basic
domains are given in capitals).
alphabet
writing system
writing
communication VISION
TIME FORCE human beings
living things MIND
LIFE physical objects
MATTER SPACE
Figure 2.3 Domain structure underlying the concept of the letter T
From this, it can be seen that it is incorrect to describe the concept of the letter
T simply as belonging to the domain of writing, as a typical informal theory of
domains would most likely have it. The vast majority of concepts belong to abstract
domains, which are themselves profiled in complex domain matrices, often also
abstract and so ultimately presuppose a large array of basic domains that can be
called a domain structure (Croft 1993[2002]:169; this corresponds to Langacker’s
maximal scope [Langacker 1999:49]).
It is not easy to distinguish profile-base relations from taxonomic/schematic
relations (that is, type vs. instance). For example, is writing an instance of human
communication, or is writing an instance of an activity that can only be understood
in terms of the goals of human communication? Figure 2.3 assumes that the
latter is a more accurate description. Likewise, since writing is an instance of
human activity, human activity does not appear as its domain, but the various
domains that it presupposes – time, change, force, volition – do appear, because
anything presupposed by a human activity will be presupposed by any instance
of it.
It is also difficult to determine direct vs. indirect reference to a domain. The
definition of an arc does not directly presuppose two-dimensional space, but rather
it presupposes a circle which in turn presupposes two-dimensional space. Thus,
an arc is not directly a two-dimensional object per se, but only such by virtue of
being a part of a circle. Likewise, the letter T is not directly a shape, but only such
by virtue of being a letter of the alphabet. But in fact, is the letter T a shape by
virtue of being a letter of the alphabet, or by virtue of being the physical product of
the activity of writing? Figure 2.3 assumes that it is best described as the former,
since the set of symbols is a set of shapes.
Another similar problem in this example is the location of the domain of mental
ability. The activity of writing is a volitional, intentional activity, so it presupposes
the domain of mental ability. But mental ability is presupposed by writing because
writing presupposes human involvement, and the human involvement involves
volition and intention. Determining the exact structure of the array of domains
upon which a profiled concept is based requires a careful working out of the
definitions of concepts.
A further complication in the relation between profiles and domain matrices is
that a word sometimes profiles a concept in only one of the domains in the domain
matrix, or even just a domain deeply nested in the domain structure. The contrast
can be illustrated by the concepts PERSON and BODY. PERSON is profiled against
the abstract domain of HUMAN BEING (along with MAN, WOMAN etc.). The
concept of HUMAN BEING is in turn profiled against the domain matrix of
LIVING THING + MIND: human beings are living things with certain mental
states and abilities (recall the classical definition of man as a rational animal).
LIVING THING is in turn profiled against the domains of PHYSICAL OBJECT
and LIFE: living things are physical objects endowed with life. The concept BODY
represents a person’s physical reality (alive or dead). Its base is nevertheless still
the abstract domain of HUMAN BEING (or more precisely ANIMAL), but it
profiles just the PHYSICAL OBJECT domain in the domain structure underlying
HUMAN BEINGS. Contrast BODY with SOUL, which profiles a nonphysical
domain of a human being, what we have called MIND for convenience; or with
CORPSE, which profiles the PHYSICAL OBJECT domain but also profiles a
particular region in the LIFE domain, namely DEAD.
2.5 Domains and idealized cognitive models
An important insight of Fillmore and Lakoff in their early work on

frames/domains is that the knowledge represented in the frame is itself a con-
ceptualization of experience that often does not match the reality. The example
most cited to illustrate this point in the cognitive literature (not just linguistics but
also philosophy, psychology and artificial intelligence) is the concept of BACH-
ELOR (Fillmore 1975:128–29; 1977a:68–70). A simple conceptual analysis of
BACHELOR is an ADULT UNMARRIED MALE. This definition may suit most
normal cases. But there are a number of cases where speakers react with uncertainty
as to whether the person involved is a bachelor or not:
(19) a. The Pope

b. Tarzan
c. An adult male living with his girlfriend
d. A male homosexual
e. A male homosexual living with his boyfriend
f. A seventeen-year-old living on his own, running his own Internet firm, and
dating several women. [cf. a seventeen-year-old living with his parents and
going to school, who virtually all agree is not a bachelor]
The apparent problem with the simple definition of BACHELOR as ADULT

UNMARRIED MALE is not that the definition is too simple (but see below).
There is a sense in which the meaning of bachelor really is just ‘adult unmarried
male’. It is just that the concept BACHELOR is profiled against a frame that
does not accommodate the variety of actual social statuses found in the real world
(Fillmore 1977a:69; 1982a:117–18). The frame for BACHELOR represents an
idealized version of the world that simply does not include all possible real-world
situations. Lakoff calls such a frame an idealized cognitive model (ICM; Lakoff
1987, chapter 4).
The analysis proposed in the preceding paragraph makes it look as if the informa-
tion in the frame for the idealized cognitive model – just ADULT UNMARRIED
MALE – is simpler than the reality that often does not match the idealization. But it
is not. The ICM for BACHELOR, or rather the ICMs for ADULT, UNMARRIED
and MALE have to include much more information than is usually associated with
those labels. The ICM for ADULT must include reference to living arrangements,
relationships to parents, and occupational activity (see [19b] and [19f]). The ICM
for UNMARRIED must include a life history sequence in which adolescence is
followed by an absence of lasting sexual relationships and then followed by mar-
riage, without taking vows of celibacy (see [19a] and [19c]). The ICM for MALE
must also include sexual orientation (with an eye to reproduction; see [19d] and
[19e]). In other words, the ICMs for BACHELOR are going to be as detailed and
as hedged as reality in order to describe the ‘ideal’ life history and lifestyle that is
implied by BACHELOR.
Searle (1979) argues that in fact the frame for any word concept is going to
be infinitely complex. Searle is interested in what he calls the background as-
sumptions for defining the literal meaning of words; in frame semantic terms, the
background assumptions are the frame(s) for understanding the literal meaning of
a linguistic expression. Searle argues that the sort of background knowledge that
is relevant to a linguistic expression’s meaning cannot be enumerated in such a
way that all contexts of use can be predicted. That is, a basically infinite set of
background assumptions are required to characterize the literal meaning of an ut-
terance, and hence its appropriate use in context. Consider the following example
(Searle 1979:127):
(20) Give me a hamburger, medium rare, with ketchup and mustard, but easy on the
relish.
We assume we understand what the meaning of this request is; we invoke a back-
ground frame of fast food restaurants, the ordering and serving of food, how a
hamburger is cooked and garnished, and so on. But there is more to it than that:
Suppose for example that the hamburger is brought to me encased in a cubic

yard of solid lucite plastic so rigid that it takes a jack hammer to bust it open, or
suppose the hamburger is a mile wide and is ‘delivered’ to me by smashing down
the wall of the restaurant and sliding the edge of it in. (Searle 1979:127)
These situations are admittedly unlikely to be encountered in real life, in the way
that unmarried men living with their girlfriends or homosexual men commonly are
encountered. Nevertheless, in the frame for ordering a hamburger we would want
to represent the assumptions that it is not supposed to be too large, nor encased in
solid lucite plastic, nor any of an indefinitely large number of other things that one
could do to a hamburger.
Langacker makes a similar observation with a similar type of example, given in
(21) (Langacker 1988:16):
(21) He is barely keeping his head above the water.
We may think we know what this sentence means, but
imagine a race over the ocean by helicopter, where the contestants must transport
a severed head, suspended by a rope from the helicopter, from the starting line to
the finish; a contestant is disqualified if the head he is carrying ever dips below
the water’s surface. (Langacker 1988:16–17)
In other words, we have to bring to bear our full knowledge of the way the world
is or, more accurately, the way we expect the world to be, in order to describe the
precise meaning of an utterance.
Another way of saying this – the more common way of saying it in cognitive
linguistics – is that we have to call on our encyclopedic knowledge in order to
properly understand a concept. Some semanticists have argued that only a small
subset of our knowledge of a concept needs to be represented as the linguistic
meaning of a word; this is known as the dictionary view of linguistic meaning.
But the frame semantic model of linguistic meaning highlights the failings of
the dictionary view (Fillmore 1982a:134; 1985:233). The dictionary view fails
because it generally describes only the concept profile, or at best a very simplified
version of the concept frame implicit in a concept profile (see Haiman 1980 for
further arguments; see also Quine 1951[1961]). Once one begins to specify the
conceptual structure of the frame that supports the concept profile for a word or
linguistic expression, the semantic structure quickly expands to encompass the
total (encyclopedic) knowledge that speakers have about the concept symbolized
by the word or construction.
Of course, encyclopedic knowledge is all interconnected in our minds. If the
meaning of a word includes the frame as well as the profile, then one must abandon
the concept of word meanings as small discrete chunks of conceptual structure.
Langacker proposes an alternative model of the meaning of a word as an access
node into the knowledge network (Langacker 1987:161–64):
The entity designated by a symbolic unit can therefore be thought of as a point of

access to a network. The semantic value of a symbolic unit is given by the open-
ended set of relations . . . in which this access node participates. Each of these
relations is a cognitive routine, and because they share at least one component the
activation of one routine facilitates (but does not always necessitate) the activation
of another. (Langacker 1987:163)
A word meaning is therefore a perspective on our knowledge of the world, as

seen through the concept profiled by the word. This view of word meaning is not
that different from the view of a conceptual category in cognitive psychology as a
means of accessing further information about the individual categorized. This view
of word meaning also highlights how choosing a word is a way of construing the
relationship between the experience being communicated and the interlocutors’
existing knowledge.
In the example of the ICM for the word bachelor, the deviations from the ICM
were all examples in which it is not clear whether bachelor is applicable to those
cases at all. For other words, a modifier is appended to the word to indicate deviation
from the ICM. For example, Lakoff describes the ICM for mother as involving a
cluster of several different ICMs (Lakoff 1987:74–76):
(22) BIRTH: the person giving birth is the mother

GENETIC: the female who contributed the genetic material is the mother
NURTURANCE: the female adult who nurtures and raises a child is the mother
of that child
MARITAL: the wife of the father is the mother
GENEALOGICAL: the closest female ancestor is the mother
The cluster ICM (as Lakoff names it) is essentially a domain matrix. Thanks
both to modern medicine and to traditional social arrangements, the real world
has many cases where only parts of the cluster model for MOTHER applies to
particular individuals. These deviations from the cluster ICM are indicated by
conventional compounds and adjective + noun expressions:
(23) a. stepmother: fits the NURTURANCE and MARITAL models but none of the
others
b. foster mother: fits the NURTURANCE model but none of the others
c. birth mother: fits the BIRTH model but none, or not all, of the others
d. genetic mother: fits the GENETIC model but not all of the others
e. unwed mother: fits (probably) all but the MARITAL model [etc.]
Nevertheless, one might still obtain varying results if asking of individuals falling
under any of the categories in (23) whether she is the ‘real mother’ of the child
(see chapter 5).
In other cases, there is clearly an ICM but linguistic convention has allowed the
word, unmodified, to describe situations that lack some of the properties of the
ICM. Fillmore gives the example of the ICM for breakfast, which has as its frame
a cycle of meals, and profiles ‘the one which is eaten early in the day, after a period
of sleep, and for it to consist of a somewhat unique menu’ (Fillmore 1982a:118).
But you can work through the night and have eggs, toast and so on at sunup and
call it breakfast; you can sleep till 3pm, get up and have eggs, toast and so on
and call it breakfast; and you can sleep through the night and in the morning have
chocolate cream pie and a martini and call it breakfast (ibid., 118–19). Each of
these cases lacks one feature of the ICM for BREAKFAST. One can also call a
meal breakfast that lacks both ‘early in the day’ and ‘after a period of sleep’ too:
restaurants exist that serve breakfast all day (ibid.; the menu feature appears to be
more important than the other two).
Another example similar to BREAKFAST is the ICM for lie (Coleman and
Kay 1981). The ICM for LIE, such that a speaker S telling an addressee A the
proposition P is a lie, is:
(24) a. P is false.
b. S believes P to be false.
c. In uttering P, S intends to deceive A.
Coleman and Kay performed an experiment with stories designed to test every
combination of the features listed in (24), and found that, in general, the situations
with more of the three properties (24a/b/c) tended to be described by experimental
subjects more often as lies than situations with fewer of the properties. Two situa-
tions (at least) have conventional expressions that indicate their deviation from the
ICM. Polite social lies such as saying What a lovely party! or How nice to see you!
can be said in circumstances in which (24a/b) hold but (24c) does not. The other
situation can be illustrated with the exchange in (25) (Coleman and Kay 1981:29):
(25) John: Where are you going?

Mary: [out to buy John’s birthday present] We’re out of paprika.
In the situation in (25), (24a/b) do not hold but (24c) does (just the opposite of social
lies); in this situation an English speaker could say that Mary is being economical
with the truth.
In the case of breakfast and lie, the word profile extends to a range of situations
whose features vary. Nevertheless, there appears to be agreement as to the situation
that counts as the ICM for these words. ICMs thus give rise to judgements of graded
centrality to members of a category, a phenomenon that is usually described as
prototype effects (see Lakoff 1987 and chapter 4).
2.6 Mental spaces
Semantic frames/domains represent one of the two major organizing prin-

ciples for conceptual structure. The other important organizing principle is the one
illustrated by the examples in (26):
(26) a. Gina bought a sports car.

b. Giorgio believes that Gina bought a sports car.
c. Paolo believes that Gina bought a pickup truck.
d. Gina wants to buy a sports car.
e. Gina will buy a sports car.
f. If Gina buys a sports car, then she will drive to Paris.
In (26a), a situation is asserted (profiled), evoking the frame/domain of com-

mercial transactions. In (26b), the same situation is represented, but as a belief
rather than a fact. Example (26c) demonstrates that such beliefs may be at vari-
ance with the facts, and with other beliefs. In (27d), the same apparent situation
is also represented, but it has a different status: the event has not taken place, it is
only something in Gina’s mind. In fact, even the sports car may exist only in Gina’s
mind. Example (26e) is more similar to (26d) than (26a), even if it is a prediction
about the real world: the event has not taken place. Finally, in example (26f) the
event is again hypothetical, and so is the event described in the consequent clause.
In a truth-conditional semantics, (26a) is unproblematic, but (26b–f) are. The
situation ‘Gina has bought a sports car’ is false in (26d–e), but not necessarily false
in (26b–d) or even (26g). One must be able to distinguish between the status of
situations depending on whether they are true in the real world, or whether they
are only true in someone’s beliefs or desires, or true at another time in the real
world.
In a truth-conditional semantics, the standard way of representing the status of
situations is as possible worlds: there is the real world, and then there are worlds
with situations that are possible but not (necessarily) actual. Possible worlds are
then identified with a person’s beliefs or wishes or some other mental attitude. Pos-
sible worlds pose metaphysical problems for many people, however. Do possible
worlds exist? If so – or especially if not – where are they?
Fauconnier (1985, 1997; see also Fauconnier and Sweetser 1996) proposes an
alternative model of representing the status of knowledge that is metaphysically
more attractive and allows for elegant solutions to a number of problems in semantic
and pragmatic analysis. Fauconnier replaces the notion of a possible world with
that of a mental space, and argues that the mental space is a cognitive structure.
That is, the allocation of a situation to ‘Gina’s desire,’ ‘Paolo’s belief’ or ‘The
hypothetical situation’ is done in the mind of the speaker (and hearer), not in
some as yet unclear metaphysical location. Fauconnier then proposes a set of
principles for the interpretation of utterances and the assignment of situations to
the appropriate mental space. We briefly present Fauconnier’s model and a number
of examples here; the reader should consult his work for detailed arguments in favor
of his model over truth-conditional approaches to the same phenomena.
Utterances such as (26a) are normally construed as situating events or states in
a base space (Fauconnier 1997:38–39), normally the present reality (more pre-
cisely, the mutually known world of the interlocutors; in Fauconnier 1985 this is
called the reality space). Utterances such as (27b–f) have elements that Fauconnier
describes as space builders: included in their meaning is the setting up of a new
space different from the base space and linked to it. Space builders include a
wide range of semantic phenomena corresponding not only to possible worlds in
logical semantics but also a variety of other operators, including temporal expres-
sions ([27a]; see Fauconnier 1985:29–30, 33–34; Fauconnier 1997, chapter 3),
image or ‘picture noun’ contexts ([27b]; Fauconnier 1985:10–12), fictional situ-
ations ([27c]; ibid., 73–81), games and other systems ([27d]; ibid, 31), negation
and disjunction ([27e–f]; ibid., 92, 96–98) and the separate cases in quantification
([27g]; see Fauconnier 1986):
(27) a. In 1770, France was a monarchy.

b. In the photo, she has black hair.
c. In the movie, Ian McKellen is Gandalf.
d. In this game, aces are low.
e. I don’t have a car.
f. Either you take a cab or you walk home.
g. Every guest got a receipt.
All of these examples have in common the building of a mental space in which
a situation is held to be ‘true’ in that space only. More generally, we can say
that just as words and constructions evoke semantic frames/domains, words and
constructions also build spaces; at the very least, they ‘build’ or evoke the base
space. The relevant word or construction then conventionally specifies that the
asserted situation holds in the appropriate space.
Between the base space and any built space, there must be a mapping of the
elements found in each space. Many interesting and puzzling semantic and prag-
matic phenomena are a product of the possible mappings between spaces. We
may divide the phenomena found in the mapping into two parts. First, what do
the named elements of the built space (e.g. Gina and sports car) correspond to,
if anything, in the base space? Second, what conceptual structures from the base
space also occur in the built space(s), and vice versa? We begin with the first
question.
In (26d), it seems straightforward to say that the person named Gina in the desire
space built by Gina wants . . . is mapped onto Gina in the base space. But the object
described as a sports car may or may not correspond to anything in the base space:
Gina may have seen a particular car on the lot, or she may not have any specific
car in mind. This is the distinction between a specific and nonspecific reading,
respectively, of a sports car in (26d). The specific and nonspecific readings are
represented in Figure 2.4.
Fauconnier crucially distinguishes between roles and values in mappings be-
tween spaces. A role is a linguistic description describing a category; a value is an
individual that can be described by that category. Roles can be a category or type
with various instances or tokens; sports car is such a role, since there are many
instances (values) of sports cars. A role can also be a category that is filled by a
single individual at one time but by different individuals over time; the President
of the United States is an example of such a role. Roles and values are specific to
a single mental space, and all counterpart relations between roles and values in
different spaces must be established cognitively by the interlocutors.
specific reading nonspecific reading
sc sc sc sc
x x' x'
G R G
R
sc: sports car (role)

R: reality
, space
G: Gina s want space
Figure 2.4 Specific and nonspecific indefinites
Figure 2.4 easily represents the difference between the specific and nonspecific
readings. In the specific reading of (26d), the value x of a sports car in Gina’s
want space G has a counterpart value x in reality (the car she saw on the lot). In the
nonspecific reading, there is no counterpart value in reality: she imagines a sports
car she wants, but has not identified it with any existing car.
One of Fauconnier’s central insights is that many puzzling semantic phenomena
are the result of the fact that a value in one space can be described by the role its
counterpart in another space has, even if that role is invalid for the value in the
first space. This is the Access Principle (Fauconnier 1997:41; it is called the ID
Principle in Fauconnier 1985:3). For example, (28) is not contradictory (Fauconnier
1985:29):
(28) In 1929, the lady with white hair was blonde.
The value in the 1929 temporal space – the blonde girl – is being described with
a role, lady with white hair, from the base space (the current reality).
Armed with the distinctions between mental spaces, between roles and values
within spaces and across spaces, and the Access Principle, Fauconnier goes on to
explain a wide range of semantic and pragmatic phenomena using these distinc-
tions. Only a selection of these can be described here.
The phenomenon described as referential opacity is illustrated in 29:
(29) Oedipus wants to marry his mother.
In the Greek myth, (29) is true under one reading (his mother = ‘the person you
and I know is Oedipus’ mother’) but false under another (his mother = ‘the person
Oedipus believes is his mother’). This distinction is due to the fact that Oedipus
does not know that Jocasta is his mother. In mental space terms, the individual value
named Jocasta does not fill the role his mother in Oedipus’ belief space, although
she does in reality space; see Figure 2.5 (adapted from Fauconnier 1985:49).
om om
j j′
R O
,
om: Oedipus mother (role)
j: Jocasta (value in R)
j′: Jocasta (value in O)
R: reality space
,
O: Oedipus belief space
Figure 2.5 Mental space diagram for example (29)
In the true reading of (29), the description his mother for j in R is used for the value
j in O by the Access Principle. The false reading of (29) uses the description his
mother in O, but it does not apply to j in O. (A similar analysis can be applied to
referential/attributive ambiguities; Fauconnier 1985:159–60.)
Only a few further examples can be given of how Fauconnier’s model handles
a variety of complex reference and identity phenomena (Fauconnier 1985:45, 36,
32, 39, 31, 155):
(30) a. Rose is blonde, but George thinks she’s a redhead.

b. Hitchcock saw himself in that movie.
c. I didn’t buy a car. Otherwise, I would drive it to work.
d. Your car is always different.
e. If I were a millionaire, my VW would be a Rolls.
f. Hesperus [the Morning Star] is Phosphorus [the Evening Star].
In (30a–b), the pronoun identifies a value in a built space (George’s beliefs,

the movie) by referring to its counterpart in the base space (the blonde Rose,
Hitchcock). In (30c), the pronoun refers to a value in a negative space which has
no counterpart in reality (otherwise . . . would evokes the same negative space; see
example [32] below). In (30d), your car refers to a role and the predicate describes
its changes in value over a sequence of temporal spaces. In (30e), the value filling
the role a Rolls in the counterfactual space is identified with the value filling the
role my VW in reality. In (30f), a classic philosophy example, two distinct values
in the prior reality space are identified as one value in the current reality space.
The second set of phenomena that Fauconnier explores is what conceptual struc-
tures from the base space also occur in the built space(s), and vice versa. For
example, in (26d), how much of our knowledge of reality should be attributed to
the hypothetical space? Obviously, one cannot attribute to the hypothetical space
the real-world fact that Gina has not bought a sports car; that is precisely what is
asserted in the hypothetical space. On the other hand, at least other information
about Gina, and about sports cars, not to mention much other knowledge about the
world, may be attributed to the hypothetical space.
Fauconnier first addresses the question of presuppositions (Fauconnier 1985,
chapter 3). As noted in §2.1, presuppositions are situations that are part of the
frame of a concept, but are not asserted. The question is, what is the relationship of
presuppositions in a built space to those in the base space? For example, consider
the sentences in (31) (Fauconnier 1985:89–90):
(31) a. If Max has gone to the meeting, then Max’s children are alone.
b. If Max has children, Max’s children are American.
The phrase Max’s children presupposes that Max has children; that is, a referring
expression presupposes the existence of its referent(s). The traditional pragmatic
analysis is that one must determine the presupposition of the whole sentence from
the presuppositions of its parts (presupposition projection; see, e.g., Levinson
1985:191–225). In (31a), the presupposition that Max has children ‘projects’ to the
base space. But in (31b), it does not project because it is asserted in the antecedent
clause: Max may or may not have children in the base space (reality).
Fauconnier instead introduces the principle of presupposition float: ‘informally:
a presupposition floats up [from a built space to its base space] until it meets itself or
its opposite’ (Fauconnier 1997:61). In example (31a), the built space presupposes
that Max has children but does not assert it. Hence the presupposition can float to
the base space. In example (31b), however, the built space asserts that Max has
children, and hence the presupposition cannot float beyond it to the base space.
Two more complicated examples are given in (32)–(33) (Fauconnier 1985:95,
93):
(32) It is possible that John has children, and it is possible that John’s children are
away.
(33) Luke believes it’s raining and hopes that it will stop raining.
Examples (32)–(33) demonstrate that space builders may build the same space
or related nonreal mental spaces. In (32), it is possible in the second conjunct
can be construed as evoking the same possibility space that was built in the first
conjunct. In this case, the presupposition that John has children is asserted in the
hypothetical space in the first conjunct and therefore it does not float to the base
space. Example (33) demonstrates that certain built spaces are related in privileged
ways that allow presuppositions to float (see also McCawley 1981[1993]:415–30).
A hope for some situation can be built on one’s beliefs. Hence the presupposition
that it is raining in the second clause of (33) is built on the assertion in the first clause
(and therefore does not float to the base space). Reversing the relation between the
two spaces fails, because beliefs cannot be built on hopes:
(34) ??Luke hopes that it is raining and believes that it will stop raining.
Fauconnier then turns to the question of counterfactual conditionals, as in (35)

(Fauconnier 1985, chapter 4; 1997, chapter 4):
(35) If Boris had not come, Olga would have come anyway.
A counterfactual conditional builds a space in which the antecedent clause (Boris

does not come) is explicitly the opposite of the base space (Boris did come). Again,
the question is, what structures in the base space are also found in the built space?
Fauconnier argues that previous, truth-conditional approaches to counterfactuals
transfer as much of the structure of the base space as possible to the counterfactual
space (Fauconnier 1985:118). Fauconnier instead advocates an analysis in which
only the structure relevant for the counterfactual reasoning is transferred to the
counterfactual space. He argues that the great flexibility of counterfactuals pre-
cludes transferring too much of the structure of the base space to the counterfactual
space; compare (36a–d) (ibid.):
(36) a. If Napoleon had been the son of Alexander, he would have been Macedonian.
b. If Napoleon had been the son of Alexander, he would have won the battle of
Waterloo.
c. If Napoleon had been the son of Alexander, he would not have been Napoleon.
d. If Napoleon had been the son of Alexander, Alexander would have been
Corsican.
Fauconnier writes:
It would not make sense to evaluate the ‘absolute’ truth of any of these statements,
but they can all be used to make some point, which requires only very partial
structuring of H [the counterfactual space] . . . such examples suggest that there is
no general linguistic algorithm for going from R [the base space] to H. (Fauconnier
1985:118)
In Fauconnier’s more recent work (Fauconnier 1997, chapter 6; Fauconnier and

Turner 2002; see also Coulson 2000), he and Turner have emphasized the fact
that information from two different spaces, such as those in the counterfactual
conditionals in (36a–d), is blended in the resulting space, and that this blending
process occurs in a much wider range of contexts than counterfactual conditionals.
For example, (37) blends elements of a voyage by the catamaran Great America II
from Boston to San Francisco in 1993 with a voyage by the clipper Northern Light
on the same route in 1853 (Fauconnier 1997:155–56; see Fauconnier and Turner
1994):
(37) At this point, Great America II is barely maintaining a 4.5 day lead over Northern
Light.
Obviously, Northern Light is nowhere to be seen in 1993, but the blend of the
1853 temporal space and the 1993 temporal space in (37) ‘makes a point’ about
the progress of Great America II. In Fauconnier and Turner’s blending theory,
example (37) evokes four mental spaces: two input spaces (in [37], the 1853 and
1993 temporal spaces); a generic space, which abstracts the commonalities from
the two spaces (the route of travel, distance traversed, time taken etc.) and thereby
defines the cross-space mapping between the elements in the two input spaces; and
a blended space, which creates a novel expressive effect, in this case an image of
a race between the current boat and a boat from the nineteenth century. Fauconnier
and Turner (2002) argue that blending is a process of space mapping that pervades
human reasoning, and explore the phenomenon of blending in a wide range of
phenomena, most notably metaphor.
At this point, blending theory has moved quite a distance from mental space
theory. Mental space theory illustrates how utterances evoke not just semantic
frames but also spaces representing the status of our knowledge (beliefs, desires,
hypotheticals, counterfactuals) relative to reality, how language uses links between
different spaces in referring to individuals, and how knowledge can float between
spaces. Blending theory has shifted the focus to how information from two spaces,
construed broadly to include domains, is combined to produce novel conceptual
structures. This aspect of blending theory is discussed with respect to metaphor in
§8.3.3. In this chapter we have focused on the fact that the original mental space
theory describes a significant dimension for the structuring of our conceptual
knowledge orthogonal to semantic frames/domains, and offers solutions to many
semantic and pragmatic problems in addition to those illustrated in this section.
3
Conceptualization and
construal operations
3.1 Introduction
In chapter 1, we noted that one of the basic hypotheses of cognitive linguis-

tics is that, in Langacker’s words, semantics is conceptualization. This hypothesis
challenges the view that semantics is purely truth-conditional. We have already
seen in chapter 2 examples of semantic interpretations of linguistic expressions
that go beyond truth-conditional semantics. Situations can be framed in different
ways – e.g., my dad vs. dad vs. father and waste time vs. spend time in My dad
wasted most of the morning on the bus (see §2.1) – and these ways convey to the
hearer different conceptualizations of the relationship between the speaker and the
speaker’s father, of the positive or negative quality of the situation being described,
and even of the nature of the situation being described (characterizing time in terms
of money).
Framing is pervasive in language: as we argued in chapter 2, all linguistic units
evoke a semantic frame. Yet framing is but one example of the ubiquity of concep-
tualization in linguistic expression. All aspects of the grammatical expression of
a situation involve conceptualization in one way or another, including inflectional
and derivational morphology and even the basic parts of speech. Whenever we utter
a sentence, we unconsciously structure every aspect of the experience we intend to
convey. The purpose of this chapter is to describe the range of conceptualization
processes or construal operations that human beings employ in language.
The role of conceptualization in language is clearest when a single lan-
guage provides alternative expressions for what appears to be truth-functionally
equivalent situations. The example of framing given above contrasts different
lexical expressions – dad/father and spend/waste – that otherwise appear to
be truth-functionally equivalent. (We deliberately hedge on the phrase ‘truth-
functionally equivalent’ because it often turns out that there are some situations
that so favor one conceptualization over another that the other expression
is unacceptable and so the two expressions are not always judged as truth-
functionally equivalent.) But it is equally easy to find examples of inflectional and
40
Conceptualization and construal operations 41
derivational contrasts between otherwise approximately truth-functionally equiv-

alent expressions:
(1) a. leaves on the tree

b. foliage-Ø on the tree
(2) a. Conor lives in New York City.
b. Conor is living in New York City.
(3) a. The chimney is above the window.
b. The window is below the chimney.
(4) a. Something moved in the grass.
b. There was a movement in the grass.
(5) a. The car brushed the bicycle.
b. The bicycle was brushed by the car.
(6) a. There was Sam sitting on the floor.
b. Sam was sitting on the floor.
Examples (1a–b) and (2a–b) differ in the choice of nominal and verbal inflection,
(1a–b) by plural count noun and mass noun, and (2a–b) by the choice of a simple vs.
a progressive form. Examples (3a–b) differ in the choice of a function word, in this
case a preposition, and a reversal of subject and prepositional complement choice.
Examples (4a–b) differ derivationally in part of speech, between a verb and its
derived noun. Examples (5a–b) and (6a–b) differ in the grammatical construction
used to describe the scene, active vs. passive voice in (5) and presentational vs.
ordinary declarative in (6). All of these sentences seem to be truth-functionally
equivalent. But English is not being unnecessarily profligate here: the a and b
members offer a different conceptualization of the experience in every case. These
and other examples will be explicated in the following sections.
Similar evidence is found in cases when the same word can be used with two
different inflections, derivations or constructions, and there are subtle but definite
conventional truth-functional differences in the two uses:
(7) a. We have chocolate-Ø for dessert.

b. We have chocolates for dessert.
(8) a. Ira is a nuisance.
b. Ira, stop being a nuisance!
(9) a. Timmy is in front of the tree.
b. Timmy is behind the tree.
(10) a. Jill is fussy.
b. Jill is a fussbudget. (Bolinger 1980:79)
(11) a. The dog chewed the bone.
b. The dog chewed on the bone.
In (7a), the mass noun refers to a substance, but in (7b), the same noun as a count
noun refers to an object which is covered with the substance but may or may not
be filled with the same substance. In (8a), the simple present describes a behav-
ioral trait of Ira’s, while (8b) describes a particular activity of his. Examples (9a)
and (9b) differ in Timmy’s position relative to the speaker and the tree; to make
(9a) and (9b) truth-functionally equivalent, the speaker would have to move to the
other side of the tree. Example (10a) describes a behavioral trait of Jill, but (10b)
describes the same trait as a constant (and annoying) aspect of Jill’s personality.
In example (11a), the object role of the bone indicates that the bone itself is being
affected by the dog’s action, but in (11b), the oblique role indicates that it is only
meat and gristle on the bone that is being affected.
The examples in (7)–(11) are equally indicative of the role of construal in lan-
guage as the examples in (1)–(6). The truth-functional differences in (7)–(11) are
indicative of favored conceptualizations that have led to the extension of a partic-
ular construal to a situation that does not (easily) allow for alternative construals.
For example, the fact that certain sweets come in individuated units allows for a
construal of the substance noun chocolate as a count noun (compare I’d like an
orange juice, please), and chocolate as a count noun is extended to other such
sweets where only the (perceptually salient) outer surface is made of chocolate.
But both a chocolate and an orange juice share the construal as an individuated
unit, even if the link to the substance is more tenuous in the former case. In fact, in
cognitive linguistics conceptualization is the fundamental semantic phenomenon;
whether alternative construals give rise to differences in truth conditions or not is
a derivative semantic fact.
In some cases where construal is accompanied by a truth-functional seman-
tic shift in meaning, English allows speakers to express the construal-plus-shift
overtly, as in:
(12) Stop acting like a nuisance!

(13) I’d like a glass of orange juice, please.
Example (12) uses the process verb act and a manner construction to denote
the activity and (13) uses a container word glass and the partitive construction to
denote the individuated amount of juice. This is not always possible in English; for
example, a chocolate does not have an obvious overt expression of its individuation,
and there is no simple overt expression of the different types of chewing actions
in (11a–b).
Other languages require overt expression of construal-plus-shift, in contrast to
English. For example, Vietnamese cannot simply use a mass noun in a count noun
construction or vice versa, as English does in (7); and Russian requires overt
marking in the following examples were English does not:
(14) a. soloma ‘straw’ [mass]

b. solom-ink-a ‘a straw’ [count (American English)]
(15) a. On kričal. ‘He cried/was crying’ [multiple times]
b. On krik-nu-l. ‘He cried’ [once]
In (14b), Russian requires the singulative suffix -ink before the gender-case in-
flection -a in order to use the mass noun stem solom- in a count construction; En-
glish simply uses straw in either construction (with a concomitant truth-conditional
semantic shift). In (15b), Russian requires the semelfactive suffix -nu before the
past inflection -l in order to use the extended-activity verb stem krik- in the once-
only construal; English simply uses cry in the simple past or the past progressive
(with the relevant semantic shift).
Thus, there is cross-linguistic and language-internal variation as to whether
construal plus truth-conditional semantic shift is expressed covertly or overtly.
We will call the former coercion and the latter conversion.1 We see no signifi-
cant difference between coercion and conversion in conceptual semantic terms; in
both cases, the truth-conditional semantic shift that accompanies the construal is
conventionalized in the language, and cannot be assumed to carry over to other
languages or even other words in the same language. In both cases, what we are
interested in is the construal process itself, and this is a part of both conversion
and coercion.
There are many construal operations that have been identified by cognitive
linguists, and by other linguists who take a conceptualist approach to linguistic
semantics. There have also been various proposals for grouping together construal
operations that appear to be related. The two most comprehensive classifications
are those of Talmy and Langacker. Talmy proposes a four-way classification under
the name of imaging systems (Talmy 1977, 1978a, 1988a,b), given in (16):2
(16) I. Structural Schematization
II. Deployment of Perspective
III. Distribution of Attention
IV. Force Dynamics
Langacker surveys a wide range of construal operations under the rubric of
focal adjustments (Langacker 1987, §3.3); his classification of focal adjustments
1 There is no uniformity in terminology in the literature, unfortunately: ‘coercion,’ ‘conversion’ and

‘shift’ have all been used for the covert case.
2 Talmy somewhat alters this classification in the version found in Talmy (2000), and changes their
name to schematic systems. The first three categories are basically the same (the first is renamed
configurational structure), while force dynamics is dropped entirely and a system called ‘Domain’,
consisting solely of the domains of space and time, is added. The ‘Domain’ category includes the
construals represented by noun and verb (see §§3.2, 3.5). We believe that force dynamics should be
retained as a construal system; see Table 3.1.
is given in (17):
(17) I. Selection
II. Perspective
A. Figure/Ground
B. Viewpoint
C. Deixis
D. Subjectivity/Objectivity
III. Abstraction
Talmy’s and Langacker’s classifications have a number of features in common:

for example, both have categories of perspective, and Talmy’s attentional imaging
system includes Langacker’s selection and abstraction focal adjustments. These
classifications are not comprehensive, however. Fillmore’s framing is a construal
operation, for example, but does not obviously fall under the categories in either
Talmy’s or Langacker’s classifications. Lakoff and Johnson’s (1980) theory of
metaphor is an example of another widespread type of linguistic conceptualization
that is not discussed explicitly by either Langacker or Talmy. Langacker himself
makes use of other construal operations, such as scanning, comparison and the
entity/interconnection distinction, that he does not include in his classification of
focal adjustments.
There is another theoretical construct in cognitive linguistics which imposes a
conceptualization of experience, namely image schemas (Lakoff 1987; Johnson
1987; Lakoff and Turner 1989; Johnson 1987; Clausner and Croft 1999). Image
schemas are defined as schematic versions of images. Images are representa-
tions of specific, embodied experiences (see Fillmore 1975:123; 1977a:73–74).
Domains that give rise to images are described as embodied (Lakoff 1987:267;
Johnson 1987:19–23) or grounded (Lakoff and Turner 1989:113). These domains
are all basic domains as defined in §2.4. Image schemas are not specific im-
ages but are schematic. They represent schematic patterns arising from imag-
istic domains, such as containers, paths, links, forces, and balance that recur
in a variety of embodied domains and structure our bodily experience (Lakoff
1987:453; Johnson 1987:29). Image schemas are also not specific to a partic-
ular sensory modality (Lakoff 1987:267; Johnson 1987:24–25). Image schemas
structure our bodily experience (Talmy 1972, 1977, 1983), and they structure our
non-bodily experience as well, via metaphor (Lakoff 1987:453; Johnson 1987:29;
see chapter 8). This definition clarifies the seemingly contradictory description of
image schemas sometimes found: image schemas are ‘abstract’ in one sense of
that word – they are schematic – but not ‘abstract’ in another sense of that word –
they are embodied.
An inventory of image schemas drawn from Johnson 1987 and Lakoff and
Turner 1989 is given in (18) (based on Clausner and Croft 1999:15; the headings
and items in italics were added by Clausner and Croft).
(18) SPACE UP-DOWN, FRONT-BACK, LEFT-RIGHT,
NEAR-FAR, CENTER-PERIPHERY, CONTACT
SCALE PATH
CONTAINER CONTAINMENT, IN-OUT, SURFACE,
FULL-EMPTY, CONTENT
FORCE BALANCE, COUNTERFORCE, COMPULSION,
RESTRAINT, ENABLEMENT, BLOCKAGE,
DIVERSION, ATTRACTION
UNITY/MULTIPLICITY MERGING, COLLECTION, SPLITTING,
ITERATION, PART-WHOLE, MASS-COUNT,
LINK
IDENTITY MATCHING, SUPERIMPOSITION
EXISTENCE REMOVAL, BOUNDED SPACE, CYCLE,
OBJECT, PROCESS
Image schemas are also construals of experience, though they exhibit some of
the characteristics of domains as well (see §3.5).
In this chapter, we present all of the construal operations and image schemas
discussed by cognitive linguists under a new classification.3 This classification is
given in Table 3.1 on page 46.
A chief aim of this classification is to demonstrate the close relationship be-
tween construal operations proposed by linguists and psychological processes
proposed by cognitive psychologists and phenomenologists. If linguistic construal
operations are truly cognitive, then they should be related to, or identical with,
general cognitive processes that are postulated by psychologists. In fact, most if
not all of these construal operations are special cases of general cognitive processes
described in psychology and phenomenology. This view follows from the basic
hypothesis of cognitive linguistics that language is an instance of general cognitive
abilities.
The classification of construal operations in Table 3.1 is not intended to be
a reduction of construal operations to just four processes. The various construal
operations listed under the four headings are all distinct cognitive processes. The
analysis we propose is that the various construal operations are manifestations of
the four basic cognitive abilities in different aspects of experience. The remainder
of this chapter describes and illustrates the construal operations under these four
headings.
3 An earlier version of this classification is presented in Croft and Wood 2000.

Table 3.1 Linguistic construal operations as instances of general cognitive

processes
I. Attention/salience
A. Selection
1. Profiling
2. Metonymy
B. Scope (dominion)
1. Scope of predication
2. Search domains
3. Accessibility
C. Scalar adjustment
1. Quantitative (abstraction)
2. Qualitative (schematization)
D. Dynamic
1. Fictive motion
2. Summary/sequential scanning
II. Judgement/comparison (including identity image schemas)
A. Categorization (framing)
B. Metaphor
C. Figure/ground
III. Perspective/situatedness
A. Viewpoint
1. Vantage point
2. Orientation
B. Deixis
1. Spatiotemporal (including spatial image schemas)
2. Epistemic (common ground)
3. Empathy
C. Subjectivity/objectivity
IV. Constitution/Gestalt (including most other image schemas)
A. Structural schematization
1. Individuation (boundedness, unity/multiplicity, etc.)
2. Topological/geometric schematization (container, etc.)
3. Scale
B. Force dynamics
C. Relationality (entity/interconnection)
3.2 Attention/salience
The process of attention is a well-known basic phenomenon in cog-

nitive psychology. Attention appears to be closest to what Chafe (1994:26–30)
calls the focus of consciousness. Attention comes in degrees and is usually mod-
eled in terms of degree of activation of conceptual structures in a neural network
model of the mind. The phenomenon of attention focuses on the human cogni-
tive ability involved, but there are also natural properties of phenomena in the
perceived world that lend themselves to being attended to by human beings, and
these properties are said to enhance those phenomena’s salience to human beings’
attention.
Attention is a complex psychological ability whose different aspects can be most
easily illustrated by visual ability: one can select one object or another to focus
one’s attention on; focus of attention is surrounded by a scope of attention; one
can take a more coarse-grained or more fine-grained view of a scene; and one can
fix one’s gaze on a scene or move one’s eye over it. These four aspects of attention
are found across all domains of thought.
3.2.1 Selection
The focal adjustment of selection is our ability to attend to parts of our
experience that are relevant to the purpose at hand and ignore aspects of our
experience that are irrelevant. The phenomenon of profiling a concept in a semantic
frame, described in detail in chapter 2, is an example of selection. In most cases,
different words in a semantic frame or domain focus our attention on the different
elements in the frame, for example radius, arc, circumference in the CIRCLE
frame. In other cases, derivational morphology shifts the profile, as in writer,
whose -er suffix shifts the profile of write from process to agent. The participant
that the -er suffix selects is not fixed to a single participant role but depends on
salience, manifest both in conventionalized forms such as stapler (the instrument)
or in novel forms such as clapper (Jane T., describing a lamp that turns on when
you clap your hands).
Selection of the profile by a single underived word stem is also flexible and
subject to construal. For example, many English nouns are also used as verbs
(Clark and Clark 1979): pan can be construed as profiling either a metal object or
a process in the GOLDSEEKING frame. Both the process and the metal object
are salient in this frame, hence the choice of one word for both. Likewise, British
English speakers can construe bin as profiling either a wastebasket or the action
of tossing something into the wastebasket.
Such examples are not usually analyzed as examples of construal since the
profile is of course central to a word’s meaning and any shift in profile has truth-
functional consequences. However, two semantic processes that involve subtler
and/or more systematic shifts in profile lend themselves to a construal analysis.
The first example is the highlighting of different facets (see chapter 5) or domains
in a domain matrix, as in (19)–(22):
(19) a. Where is the Sunday Times? (physical object or tome)

b. Have you read the Sunday Times? (semantic content or text)
(20) a. Paris is a beautiful city. (location)
b. Paris closed the Boulevard St. Michel. (government)
c. Paris elected the Green candidate as mayor. (population)
(21) a. The Chronicle costs a dollar. (tome)
b. The Chronicle called for his resignation. (editor)
c. The Chronicle went bankrupt. (company)
(22) a. The window is dirty. (pane)
b. She came in through the bathroom window. (opening)
A newspaper, book or other embodied text is simultaneously a physical tome

and a meaningful text. But (19a) selects the physical object facet and (19b) only the
text facet. The range of possible profiled facets for a word can be quite wide. For
example, (20c) selects the voting members of the population, but All Paris turned
out to see the king selects a wider (but still incomplete) set of the population. The
government in (20b) is the city government; but Paris opposes any reform of the
Common Agricultural Policy profiles the national government by virtue of Paris
being the capital of France. It is not clear whether different facets count as different
senses; see §5.3.1 for further discussion.
The second example is the phenomenon of metonymy. Metonymy is, loosely,
the use of a word to denote a concept other than its ‘literal’ denotation. Examples
of metonymy include the following (Nunberg 1995:115; Langacker 1991b:189):
(23) That french fries is getting impatient.

(24) They played lots of Mozart.
(25) She heard the piano.
(26) I’m in the phone book.
A cognitive linguistic analysis of metonymy is the ability of a speaker to select

a different contextually salient concept profile in a domain or domain matrix than
the one usually symbolized by the word. In (23)–(24), it is the nouns (french fries
and Mozart) whose concept profiles are shifted. The evidence for this analysis is
found in the grammar of the sentences. In (23), although french fries is plural, the
demonstrative modifying it and the verb are singular, indicating that french fries
is profiling the single individual who has ordered french fries. In (24), although
Mozart is countable, the quantifier modifying it is used with mass nouns, indicating
that Mozart is profiling the abstract uncountable music of Mozart.
In (25)–(26), however, Langacker argues that it is the verbs whose concept pro-
files are shifted. This shift in verb concept profile is an example of what Langacker
describes as an active zone analysis. In an active zone analysis the relational
predication – a verb, adjective, adverb or preposition – adjusts its meaning to
accommodate its semantic argument, and incorporates the ‘literal’ argument as its
active zone. Thus, in (25) the meaning of heard is ‘S b j heard the sound of Obj’ –
compare She heard the sound of the piano – and the active zone in the verb meaning
is the sound emitted by the object referent. In (26) the meaning of be in (the phone
book) is ‘S b j’s name is printed in Loc’, and the active zone is a name and its
relation to the subject referent.
Langacker argues that the difference between (27a–c) is due to alternative con-
struals of the adjective, not a syntactic alternation:
(27) a. To play Monopoly is fun.

b. Monopoly is fun to play.
c. Monopoly is fun.
In transformational analyses, be fun in (27b) is semantically identical to be fun in

(27a), and (27b) is syntactically derived by a movement of Monopoly to the main
clause subject position (‘Tough-movement’). In Langacker’s analysis, (27b) is not
syntactically derived from (27a). Instead, be fun in (27b) is semantically distinct
from be fun in (27a); (27a) takes an entity as its subject and has the activity carried
out on the subject referent as its active zone. Support for the active zone analysis
is suggested by (27c): the clause from which the subject would be moved in the
syntactic analysis is absent, and yet (27c) is acceptable, with an interpretation of
be fun as in (27b).
The active zone analysis of (25)–(26) is the opposite of the traditional metonymy
analysis, in which it is the noun phrase that shifts profile (the piano to the sound
of the piano, I to my name). The verb (adjective, preposition) semantic shift is
quite different from the usual examples of metonymy on verb phrases, where, for
example, go to the bathroom is metonymic for ‘perform certain bodily functions’,
and pick up the phone in If you want to find out, you’ve only got to pick up the phone
is metonymic for a more complex activity. In these latter examples, the profile is
shifted from one event to an associated event in the same semantic frame. The
semantic shift of the relational predicate in the active zone analysis for (25)–(26)
serves a different function.
Nunberg (1995) argues for a distinction between noun and verb semantic shift
(the latter equivalent to the active zone analysis), or ‘predicate transfer’ as he calls
it, using evidence from grammatical behavior such as that cited above for (23)–
(24). Nunberg argues that one advantage of the active zone analysis is that in a
sentence such as Roth is Jewish and widely read, one can account for the lack of
zeugma by analyzing widely read as taking the author as subject, not his books;
hence the subjects are truly coreferential (Nunberg 1995:122–23); but see §5.3.1
for an alternative analysis of the same phenomenon in facets.
Whichever is the best analysis for examples such as (25)–(26), the phenomenon
of profile shift (semantic shift) appears to be a function of salience. Langacker
notes that in (25)–(26), and (27b–c) as well, the effect of the semantic shift is
to allow a more salient entity to be the semantic as well as syntactic argument
of the verb (I instead of my name, the piano instead of the sound of the piano,
Monopoly instead of to play Monopoly; Langacker 1991b:193). Nunberg argues
that the primary ‘pragmatic’ constraint on predicate transfer is the noteworthiness
of the relationship of the predicate to its argument in the context – in cognitive
linguistic terms, its salience in the semantic frame. For example, I’m in the Whitney
Museum (said by an artist about her painting) confers a noteworthy property to the
artist of being represented in a major museum; the same artist saying ??I’m in the
second crate on the right does not (Nunberg 1995:113). Nunberg also notes that
noteworthiness is subject to construal: a jealous painter might say Those daubers
get one-person shows while I’m relegated to a crate in the basement (ibid., 129,
n. 7).
3.2.2 Scope of attention (dominion)

The second aspect of attention is that the focus of attention – what is
selected – is surrounded by a scope of attention, that is, a periphery of consciousness
where entities are accessible to attention (Chafe 1994:29). We saw an example of
scope of attention in §2.4, example (18e), repeated in (28):
(28) ??A body has twenty-eight knuckles.
In the scope of predication, the domains immediately presupposed by a profiled

concept are accessible in a way that more indirectly presupposed domains are not
(see §2.4, example [18]). This is a matter of construal; the scope of predication
can shift in the appropriate context as in (29):
(29) A: We’ve found every bit of the body, sir – even the knuckles.
B: How many did you find?
A: Twenty-seven, sir.
B: Come on, now! How many knuckles does a body have?
A: Oh, you’re right, sir. Twenty-eight.
Another example of a grammatical constraint that makes reference to the scope

of attention has to do with a combination of locative expressions specifying a
location:
(30) The money is in the kitchen, under the counter, in the lefthand cabinet, on the top
shelf, behind the meat grinder.
Each locative expression profiles an entity in the scope defined by the preced-
ing locative expression (i.e., the locative expression defines successively nar-
rower search domains; Langacker 1987:286). Scrambling the order of locative
expressions creates cognitive chaos:

(31) The money is on the top shelf, in the kitchen, under the counter, behind the meat
grinder, in the lefthand cabinet.
Yet another example of scope of attention is the notion of accessibility of a

referent in discourse (Ariel 1990; Gundel et al. 1993; Chafe 1994). Consider the
passage given in (32), from the Pear Stories narratives (Chafe 1980):
(32) And then definitely when he’s up there,
. . a kid comes by on a bicycle.
From the direction where the goat man left,
okay?
A–nd . . u–m the bicycle’s way too big for the kid.
I’m giving you all these details.
I don’t know if you want them.
A third person pronoun such as them construes the referent as being in the focus
of attention of the hearer, which is appropriate in (32) because the details have
just been mentioned. However, when details is first uttered, the details are not in
the focus of attention – they have not been mentioned as such – but they are in
the scope of attention – the description in the preceding intonation units has been
about the details of the film. The choice of the proximal demonstrative adjective
these details construes the details as being in the hearer’s scope of attention but
not in focus (Gundel et al. 1993:275).
Langacker proposes a highly generalized concept of scope in terms of the do-
minion made accessible by a reference point which functions as the (initial)
focus of attention (Langacker 1999, chapter 6). Langacker argues that reference
point and dominion constitute the construal underlying the possessive construc-
tion: the ‘possessor’ in examples such as my watch, your anxiety and Lincoln’s
assassination functions as a reference point to establish a dominion in which the
appropriate referent of the head noun can be selected. Langacker also extends his
analysis to metonymy. For example, in (20c) above, the speaker focuses the hearer’s
attention on the city of Paris by using Paris. Paris then functions as a reference
point whose scope or dominion includes its inhabitants; the inhabitants of Paris
are therefore accessible as the subject referent for the predicate. Choosing The
people of Paris instead of Paris in (20c) would construe the situation differently,
putting the people in the focus of attention instead of in the dominion of another
focus.
3.2.3 Scalar adjustment

The third aspect of attention is an adjustment of the scale of attention. It
can be illustrated with a visual example (Talmy 1983:238):
(33) a. She ran across the field.

b. She ran through the field.
Examples (33a–b) could describe the same scene, but (33b) invites the hearer
to attend to the thickness of the vegetation in the field by using a preposition
requiring a three-dimensional volume; (33a) instead construes the field as a two-
dimensional surface without thickness. To describe the conceptualization involved
here, a metaphor of magnification or granularity is often used. Example (33a)
offers a coarse-grained view of the field, seen as if from a distance so that the
thickness of whatever covers the field is invisible to us. Example (33b) offers a
fine-grained view of the field, as if our view was magnified to reveal its thickness.
Examples (34a–c) provide a more elaborate example of granularity:
(34) a. We drove along the road.
b. A squirrel ran across the road.
c. The construction workers dug through the road.
In (34a), the road is viewed at such a coarse grain that it is conceptualized

as a line, which provides a path for movement. In (34b–c), the same difference
in granularity is found as in (33a–b): in (34b) the road is construed as a two-
dimensional surface that can be traversed, and in (34c), the road is construed
as a three-dimensional volume whose depth in this case can be an obstacle. At a
coarse-grained view, the road’s width and depth are reduced and lost, and the road is
merely one-dimensional. At a more fine-grained view, or a greater magnification,
the width of the road becomes visible, so to speak, and at a still finer-grained
view/magnification, the depth becomes visible as well.
Examples (33)–(34) illustrate quantitative scalar adjustment:4 a construal of
an object by adjusting the granularity of the scalar dimensions, in this case the three
spatial dimensions. Scalar adjustment is found in other measurable dimensions as
well. For example, part of the difference in the construals of (2a–b) and (8a–b) in
§3.1 is due to temporal scalar adjustment. In both cases, the progressive evokes a
finer-grained scale than the simple present. In (8a), Ira’s nuisance extends over a
long period of time, possibly his entire lifetime; but in (8b), Ira’s nuisance extends
only over a brief time, that of his objectionable activity. In (2a), Conor’s time
in New York is conceived of as permanent, or at least construed as long-term,
while in (2b), it is construed as short-term or temporary in the context of his
lifetime.
Langacker also includes what he calls schematization under the same category.
Schematization, that is, viewing something by means of a more encompassing
4 Langacker calls this ‘abstraction,’ but that term is used for such a wide range of theoretical concepts,
even in cognitive linguistics, that we choose a more precise term here.
category, is a qualitative scalar adjustment: rather than losing a measurable scale

or dimension, one loses irrelevant properties. For instance, triangle specifies the
number of sides of the shape, but the more schematic polygon, which could be
used to describe the same shape, is vague as to the number of sides it possesses
(Langacker 1987:135). A word or construction that is vague rather than ambiguous
with respect to a semantic property is an example of schematization. Possessing
an indeterminate or nonrestrictive property is not the same as lacking the property:
polygons have sides, even if the concept of a polygon is indeterminate with respect
to that feature. Schematization is a phenomenon of attention: the concept of a
polygon ignores the number of sides to any particular subtype of polygon, in
contrast to the concept of a triangle. One can also find differences in schematization
in frames as well as profiles. To take an example from §2.3, the frame for the
English word eat is vague as to whether the eater is human or animal, unlike
German essen and fressen. More generally, the choice of words at different levels
of categorization (see chapter 4) to refer to an object will construe the object at
different levels of schematization.
3.2.4 Dynamic attention

Focus, scope and scale of attention are all static construals of a scene.
The fourth aspect of attention is that it can be dynamic: one’s attention can move
across a scene. That this is a matter of conceptualization, not just a fact about the
world, can be seen in (35) (see Talmy 2000, chapter 2):
(35) The road winds through the valley and then climbs over the high mountains.
The road is not actually going anywhere, but it is conceptualized as if it is: the
mind’s eye, so to speak, represents one as going along the road. Talmy describes
this as fictive motion, because it is a construal of a static scene in dynamic terms.
Of course, most of the time speakers construe static scenes statically and dynamic
scenes dynamically; this underlies the distinction between state and process in
predicates (see §3.6). But example (35) and many similar examples demonstrate
that this semantic property is subject to construal.
Langacker also makes use of the static/dynamic attentional contrast in construal,
but for a different phenomenon. One of the basic conceptual distinctions between
predicates (prototypically verbs) and arguments or modifiers (nouns and adjec-
tives), according to Langacker, is the mode of scanning of the scene. Langacker
distinguishes summary scanning, a holistic conceptualization of a scene in its
entirety, and sequential scanning, a scanning of a scene in conceived time, which
is not the same as objective time (Langacker 1987:144–45, 248–49). For exam-
ple, when a verb predicates an action as in Boston Bridge collapsed, the event is
scanned sequentially, over time. In contrast, when the verb is nominalized in a

referring expression such as the collapse of Boston Bridge, the event is construed
summarily as a whole unit without being scanned through time, even though the
event occurred objectively through an interval of time (compare move vs. movement
in examples [4a–b] in §3.1). Summary scanning is the norm for nouns denoting
objects when used as referring expressions, such as the tree or the lamp, and for
modifiers such as tall in the tall tree.
Summary/sequential scanning is not the same as fictive motion. Fictive motion
represents the state/process construal, but summary/sequential scanning underlies
the difference between sentence predication and nonpredicated states of affairs.
For Langacker, The road is in the valley involves sequential scanning because it is
predicated, but The road winding through the valley involves summary scanning
because the (fictive) motion of the road is not predicated.
3.3 Judgement/comparison
Kant describes judgement, which he considers a fundamental cognitive

faculty, as a particular kind of comparison: ‘judgement in general is the faculty
of thinking the particular as contained under the universal’ (Kant 1790[1952]:18).
Husserl, generalizing over the Western philosophical tradition from Aristotle on-
wards, gives judgement the more general meaning of a comparison between two
entities: ‘the most general characteristic of the predicative judgement is that it
has two members: a “substrate” (hypokeimenon), about which something is af-
firmed, and that which is affirmed of it (kategoroumenon)’ (Husserl 1948[1973]:14,
emphasis original). Langacker also considers comparison to be a fundamental
cognitive operation (Langacker 1987:103–5). Thus we may link the fundamen-
tal philosophical concept of judgement to the cognitive psychological process of
comparison.
3.3.1 Categorization
Perhaps the most fundamental judgement of comparison is categorization,
which was described in terms of framing in chapter 2. The act of categorization –
applying a word, morpheme or construction to a particular experience to be com-
municated – involves comparison of the experience in question to prior experiences
and judging it to belong to the class of prior experiences to which the linguistic
expression has been applied. There are many ways in which a situation can be com-
pared and judged to be like a prior experience. As we saw in §2.3, the choice of a
linguistic category based on comparison to a prior situation frames – construes – the
current situation in different ways, as in fetus vs. unborn baby or thrifty vs. stingy.
In addition to the flexibility of framing a situation by comparing it to one or an-
other prior situation, speakers also have the flexibility of comparing the current
situation to a prior one and in effect redefining the frame. For example, upon en-
tering a holding pattern over Milan airport, a pilot said We’ll be on the path they
call a racetrack; that’s essentially a circle with two straight sides – a significant
reconceptualization of the category CIRCLE.
Langacker describes the comparison process between the current situation and
the category to which it is assigned as sanction (Langacker 1987:66–71). He
recognizes a gradient between full sanction – unproblematic subsumption of the
new situation – and partial sanction – a more creative extension of the category
to the current situation. Categorization involves schematization (§3.2) as well as
judgement: in comparing the new experience to prior ones and categorizing it
in one way over another, we attend to some characteristics and ignore others.
The pervasiveness of construal in the process of categorization has already been
discussed in chapter 2 and will be described in detail in chapter 4.
3.3.2 Metaphor
Another construal operation widely discussed in cognitive linguistics,
metaphor, also involves judgement or comparison. Metaphor involves a relation-
ship between a source domain, the source of the literal meaning of the metaphor-
ical expression, and a target domain, the domain of the experience actually being
described by the metaphor. For example, to waste time involves comparing TIME
(the target domain) to MONEY (the source domain) in the metaphor represented
by the Lakoffian formula t i m e i s m o n e y (Lakoff and Johnson 1980). Time is
construed as a valuable asset that is possessed by human beings and can be ‘used’
in the same way that money is.
The choice of metaphor to describe a situation in a particular domain construes
the structure of that domain in a particular way that differs depending on the
metaphor chosen. For example, the metaphor in stockmarket crash construes the
low level of the market as abnormal, the result of defective operation, whereas
a high (or rising) market is normal. On the other hand, stockmarket correction
construes the low level of the market as normal, its correct level, whereas the high
level is abnormal.
The exact relationship between the source and target domains in a metaphorical
expression is a matter of debate within cognitive linguistics. Metaphor, like cate-
gorization, is sufficiently important for conceptualization to merit its own chapter
in this book (chapter 8), and is discussed in greater detail there.
3.3.3 Figure-ground alignment

A third example of comparison as a linguistic construal is figure-ground
alignment. Figure-ground alignment appears to be strongly influenced by objective
properties of the scene, although they can be overridden in various ways (that is,
it is subject to construal). The figure-ground distinction is derived from Gestalt
psychology (e.g. Koffka 1935, ch. 5), introduced into cognitive linguistics by
Talmy (Talmy 1972, 1983, 2000).
Talmy uses the figure-ground relation to account for the expression of spatial
relations in natural language. All spatial relations in language – both location (36)
or motion (37) – are expressed by specifying the position of one object, the figure,
relative to another object, the ground (sometimes more than one ground object,
as in [38]–[39]):
(36) The book [figure] is on the floor [ground].

(37) Sheila [figure] went into the house [ground].
(38) The Isaac CDs [figure] are between Compère [ground] and Josquin [ground].
(39) Greg [figure] drove from San Rafael [ground] to Trinidad [ground] in five
hours.
The figure and ground are asymmetrical. Although near is a spatially symmet-
rical preposition, (40b) sounds odd compared to (40a) (Talmy 2000:314):
(40) a. The bike is near the house.

b. ??The house is near the bike.
Likewise, there is no preposition that functions as the inverse of in in (41a) (from

Leonard Talmy), because the figure-ground orientation is quite unnatural:
(41) a. There’s a crocodile in the water.

b. ??There’s water ‘being-a-suspending-medium-for’ the crocodile.
Talmy identifies the following properties of objects that favor figure or ground
construal, in the narrower domain of spatial relations (based on Talmy 1983:230–
31; see Talmy 2000:315–16):
(42) Figure Ground
location less known location more known
smaller larger
more mobile more stationary
structurally simpler structurally more complex
more salient more backgrounded
more recently in awareness earlier on scene/in memory
Nevertheless, figure-ground relations can be manipulated. The same object can

function as figure in one context and ground in another, as in (43a–b); and the
favoring contexts can also be overridden for the opposite figure-ground construal,
with appropriate contextualization, as in (44):
(43) a. The cat [figure] is on the table [ground].

b. I found a flea [figure] on the cat [ground].
(44) [The speaker is composing a scene for a photograph:]
I want the house [figure] to be behind Susan [ground]!
Figure-ground relations are found in other domains, including relations be-

tween events (Talmy 1978; Croft 2001, chapter 9). The main (figure)-subordinate
(ground) event relation is construed asymmetrically in (45a), compared to the
symmetrical coordinate event relation in (45b):
(45) a. I read while she sewed.

b. I read and she sewed.
The event in the ground/subordinate clause is conceptualized as the basis or

ground – i.e., a cause or precondition – for the event in the figure/main clause.
Figure-ground asymmetry may lead to outright anomaly, as in example (46) (Talmy
2000:325):
(46) a. He dreamed while he slept.

b. *He slept while he dreamed.
The two events could be coextensive, but since dreaming is contingent on sleeping,
sleeping must function as the ground and therefore (46a) is acceptable while (46b)
is not.
For most figure-ground subordinators, there is no natural inverse for the figure-
ground relation specified by the subordinator (Talmy 2000:326):
(47) a. She slept until he arrived.

b. ??He arrived ‘immediately-and-causally-before-the-end-of’ her sleeping.
(48) a. We stayed home because he had arrived.
b. ??He arrived ‘to-the-occasioning-of-(the-decision-of)’ our
staying home.
In a few cases, it is syntactically simple to construct both a semantic relation and

its inverse, for example before and after. However, there is a difference in construal
depending on the choice of event as figure ([49a–b], from Croft 2001:331):
(49) a. After Tom resigned, all hell broke loose.

b. Tom resigned before all hell broke loose.
In (49a), Tom’s resignation is presumed to let loose the forces of chaos; whereas in
(49b), Tom succeeded in cutting out when he saw what was happening (or perhaps
before the consequences of his actions became apparent to everyone).
Similarly, simultaneous subordinators such as when are temporally symmetrical,

but inappropriate choice of figure and ground events leads to conceptual peculiarity
(compare [50b] to [50a], from Croft 2001:330):
(50) a. When Jerry was chair of the department, everything was all right.
b. ??When everything was all right, Jerry was chair of the department.
In (50a), the healthy state of affairs is presumed to be due to Jerry’s chairmanship.

Example (50b) on the other hand is odd, making Jerry look like an opportunist
who has the extraordinary ability to take advantage of a healthy state of affairs to
assume the chairmanship of the department.
Figure-ground alignment is an example of comparison in that the two elements
of the scene are compared to each other; but unlike categorization and metaphor, the
judgement is one of contrast rather than similarity. Langacker also explicitly links
figure-ground to comparison, arguing that the typical figure-ground alignment falls
out of his model of comparison as cognitive events of scanning a scene (Langacker
1987:121–22).5
3.4 Perspective/situatedness
Perspective, especially deixis, is perhaps the most obvious and most com-
mented upon of the construal operations. Particularly for spatial descriptions, per-
spective is essential, and its dependence on the relative position and viewpoint of
the speaker is well known. But perspective is also found in nonspatial domains:
we have a perspective based on our knowledge, belief and attitudes as well as
our spatiotemporal location. The closest cognitive property to perspective taken
broadly is probably the philosophical notion of our situatedness in the world in a
particular location – where location must be construed broadly to include temporal,
epistemic and cultural context as well as spatial location. This broad interpreta-
tion of location is related to what the phenomenological philosopher Heidegger
calls Being-in-the-world. Heidegger argues that Being-in-the-world is more than
simple spatial inclusion; rather, it is the fundamental situatedness of existence in
5 Despite this, Langacker subsumes figure-ground under perspective among his focal adjustments. In
fact, Langacker himself gives an argument to show that figure-ground alignment is conceptually
independent of foreground-background perspective (see §3.4 below). Langacker also argues that
figure-ground is conceptually distinct from focus of attention, which suggests that the figure-ground
distinction does not belong under the general category of attention either (pace Talmy 1988a:195;
Talmy does not include figure-ground in the schematic systems in Talmy 2000).
Langacker makes greater use of the concepts trajector and landmark. A trajector is defined as
the figure in a relational profile (Langacker 1987:217; see §3.5 for the definition of relationality);
landmarks function as grounds to the trajector.
all respects (Heidegger 1927[1962]:79–80; cf. Dreyfus 1991:40–45). That is, we

are always already in a situation and construing it from some perspective.
3.4.1 Viewpoint
It is easiest to begin illustrating perspectival construals with spatial ex-
amples. Langacker proposes viewpoint as a focal adjustment with two subtypes:
vantage point and orientation (Langacker 1987:122–26). Vantage point was il-
lustrated in (9a–b) in §3.1: the description of Timmy’s position as being in front
of the tree or behind the tree depends on the vantage point of the speaker.6 A
particular vantage point imposes a foreground-background alignment on a scene
(ibid., 124–25).7 Alternative construals of Timmy’s position are achieved simply
by the speaker moving to another position – that is, the linguistically expressed
spatial relation is dependent on the speaker’s situatedness. Vantage point is sensi-
tive to construal; in (51), it is the vantage point of the addressee (at the relevant
future time) that is used to interpret behind:
(51) Follow my instructions carefully. Enter the woods by the south gate. Follow the
path until you come to the big oak tree. You will find the box behind it.
Orientation refers to the vertical dimension, defined by a person’s canonical up-

right position. One example of orientation is the choice of above and below in
(3a–b) in §3.1 (along with a switch in figure-ground alignment): the actual
chimney-window orientation described in (3a–b) is relative to the canonical ori-
entation of the speaker. Alternative construals for orientation are much rarer, since
we rarely go around standing on our heads or hanging from our feet.
3.4.2 Deixis
Deixis is the phenomenon of using elements of the subject’s situated-
ness – more specifically, the subject qua speaker in a speech event – to designate
something in the scene. Deixis has been widely studied (see, e.g., Levinson 1983,
chapter 2), and we focus on deixis as construal here. Person deixis – the pronouns
6 This analysis applies only to the situational use of in front of/behind. In sentences such as (9a–b), the
choice of preposition is determined purely situationally, by the relative positions of speaker, Timmy
and tree. In a sentence such as The cat is in front of the house, there is available another interpretation
in which the house has an inherent orientation such that the side with the main entrance is the front
side, regardless of the speaker’s relative position. The remarks in this paragraph refer only to the
purely situational interpretation.
7 A number of linguists have argued that clausal subordination represents a foreground-background
distinction, but it appears to be better analyzed as a figure-ground distinction (Talmy 1978, 2000;
Reinhart 1984; Croft 2001).
I, you, he/she/it, we and they – are only defined relative to who is speaking, and this
variation is an example of alternative construals defined by the speech act situa-
tion (§2.1). Likewise, deictic demonstratives such as this and that, and deictic time
reference such as present and past tense, are only defined relative to the location
and time of the speech event.
In addition to the relativity of spatiotemporal reference to the situation or per-
spective of the speech event, it is possible to construe another time and place as
the deictic center. Examples (52a) and (52b) are construed with a deictic point of
view at a place and time in the narrative; (52c) is bizarre because the point of view
clashes with the reported information (Fillmore 1982b:262–63):
(52) a. He was coming up the steps. There was a broad smile on his face.
b. He was going up the steps. There was a wad of bubblegum on the seat of his
pants.
c. He was coming up the steps. There was a wad of bubblegum on the seat of his
pants.
Likewise, the use of the so-called narrative present in (53b), or the ‘sportcaster’s
present’ in (54b), presents a construal of the time of the linguistic event, which
has the effect of bringing the reported event conceptually closer to the listener
(this construal also involves a ‘moving’ deictic center, moving with each reported
event):
(53) a. He came up behind me, I stopped suddenly, and he rammed into me.
b. He comes up behind me, I stop suddenly, and he rams into me.
(54) a. He hit the ball and the first baseman missed it.
b. He hits the ball – the first baseman misses it . . .
In other words, deictic elements often display two layers of conceptualization:

one relative to the situatedness of the speech act participants, and another con-
strual that displaces the actual situatedness of the interlocutors to another time and
place.
The situatedness of the speech act participants affects the structure of utterances
in another, more profound way. The formulation of utterances is dependent on the
shared knowledge, belief and attitudes of the interlocutors, what is often called
the interlocutors’ common ground (e.g. Clark 1996; Langacker [1987:127] calls
it the epistemic ground). As Clark points out in many different contexts, what we
choose to express in utterances and how we express it is determined to a great extent
by what we assume is or is not part of the common ground; the common ground
provides us with an epistemic perspective situating the speaker and the hearer.
The simplest example of epistemic perspective is the use of the definite and
indefinite articles. Examples (55a–b) represent an alternative construal of what the
hearer knows:
(55) a. Did you see a hedgehog?

b. Did you see the hedgehog?
Example (55a) construes the hedgehog as unknown to the hearer, while (55b)
construes it as part of their common ground. The construal in (55b) could also be
used in a context when the hearer in fact does not know about the hedgehog, as
a way of surprising the hearer with the discovery by manipulating the epistemic
deictic construal.
But in fact the wholesale structuring of clauses is determined by the epistemic
perspective of common ground. This structuring often goes under the name of in-
formation structure or information packaging (Lambrecht 1994). Examples (6a–b)
in §3.1 illustrate alternative construals of a scene based on a difference in informa-
tion structure. The presentational sentence in (6a) (a subtype of what are also called
thetic or sentence-focus structures; Lambrecht 1994:177) presents all of the infor-
mation as part of the assertion directed to the hearer. The ordinary topic-comment
(categorical or predicate-focus) sentence in (6b) presents the subject referent as part
of a presupposition of current interest (and therefore part of the interlocutors’ com-
mon ground) and only the predicate as part of the assertion (Lambrecht 1994:121).
In fact, information structure is one of the clearest syntactic examples of alterna-
tive construals of what is the same scene from a truth-conditional point of view.
Another example of perspectival construal that has been discussed more by prag-
matically oriented linguists than by cognitive linguists is the notion of empathy
(Kuno and Kaburaki 1977). By empathy, Kuno and Kaburaki mean the partici-
pant in the reported event whose perspective is taken by the speaker. Empathy is
perspectival and as such is subject to alternative construals. Kuno and Kaburaki
argue that empathy is involved in the semantics of a number of grammatical con-
structions. For example, the choice of describing Bill as John’s brother, that is, as
a person anchored by his relation to John, instead of as Bill independent of John,
implies that the speaker empathizes more with John than Bill. Likewise, choos-
ing the passive Bill was hit by John, instead of the active John hit Bill, implies
speaker empathy with the subject referent, namely Bill (compare [5a–b] in §3.1).
Example (56c) is therefore odd compared to (56a–b), because of a clash in con-
strued empathy: the speaker uses a description of Bill that implies empathy with
John, combined with a grammatical voice that implies empathy with Bill (Kuno
1987:203–6):
(56) a. Then Johni hit hisi brother.

b. Then Bill was hit by John/his brother.
c. ??Then John’s brotheri was hit by himi .
Some linguists have argued that the distinction between the grammatical re-
lations of subject and object is one of deixis or empathy. DeLancey (1981), in
his analysis of split ergativity and subjecthood, argues that the unmarked subject
category involves construing the orientation of the speaker towards the temporal
and causal beginning of the event. On the other hand, Langacker (1991a:305–
17) defines subjecthood in terms of the most prominent figure, which combines
the construal operations of attention (most prominent) and judgement (figure).
Whichever is the correct analysis – and it is possible that different languages use
different construals for subjecthood – the point is that fundamental grammatical
categories such as subject, which are treated as ‘meaningless’ in some syntactic
theories, represent a construal of the referent in the situation described by the ut-
terance, and the construal analysis can predict patterns of (un)acceptability such
as that found in (56a–c).
3.4.3 Subjectivity
The last construal operation under perspective in Table 3.1 is Langacker’s
notion of subjectivity/objectivity. This refers to how one conceptualizes a scene
that includes the speaker herself/himself. Two simple if restricted examples illus-
trate the alternative construals (Langacker 1987:131):
(57) [said by mother to child:]

a. Don’t lie to me!
b. Don’t lie to your mother!
Example (57a) represents the more common subjective construal of the speaker
using a deictic personal pronoun, defining her identity relative to the speech act
situation. Example (57b) involves objectification: the speaker describes herself in
terms independent of the speech act situation.
It is also possible to subjectify reference to an entity (Langacker 1987:132):
(58) That’s me in the top row. [said when examining a photograph]
In (58), an entity that is not the speaker, namely the physical image in the pho-
tograph, is described using a deictic expression (me), as is in fact common in
so-called picture noun contexts (see also §2.6). Another, more common example
of subjectification is the construal implied when using certain spatial expressions
that can leave a ground object unexpressed (Langacker 1991b:326, 328):
(59) a. Vanessa is sitting across the table from Veronica.

b. Vanessa is sitting across the table from me.
c. Vanessa is sitting across the table.
Example (59c) is an instance of subjectification of the across phrase: (59c) can

only refer to the situation where Vanessa is sitting across the table from the speaker
(unlike [59a]), yet explicit reference to the speech act participant is absent (unlike
[59b]). Langacker argues that many grammatical expressions, and also the pro-
cess of grammaticalization in language change, crucially involve subjectification
(Langacker 1991b, chapter 10; 1998); unfortunately, space prevents us from pre-
senting Langacker’s analyses here.
Perspectival construals all result from our being in the world in a particular
location and manner. From a purely bodily point of view, we are in a particular
spatial location in the world (vantage point) and in a canonical upright orientation.
From a communicative perspective, we are situated as participants in the speech
event, which defines our spatial and temporal location and our roles in the speech
event (deixis). Our roles in the speech event, however, also define the status of the
situation to be communicated in speaking (epistemic deixis), our attitude towards
it (empathy), and our presentation of ourselves in that situation (subjectivity).
3.5 Constitution/Gestalt
The construal operations to be described in this section represent the

conceptualization of the very structure of the entities in a scene. These construal
operations represent the most basic level of constituting experience and giving
it structure or a Gestalt, as described by Gestalt psychologists (Koffka 1935;
Wertheimer 1923[1950]) and phenomenologists such as Husserl (who uses the
term ‘constitution’ in a similar context; see Husserl 1948[1973]). For example,
many of the principles of Gestalt psychology such as proximity, bounding and
good continuation are analyses of how human minds construe a single complex
object from seemingly fragmented perceptual sensations. In cognitive linguistics,
the most detailed discussion of constitutive construals is by Talmy, under his
imaging systems of structural schematization (Talmy 1988a) and force dynamics
(Talmy 1988b; both revised and expanded in Talmy 2000). In addition, most of
the image schemas described by Lakoff, Johnson and Turner are construals of the
structure of entities.
3.5.1 Structural schematization

Structural schematization describes the conceptualization of the topo-
logical, meronomic and geometrical structure of entities and their component parts.
The subgrouping in Table 3.1 attempts to classify the wide range of structural
schematizations (for a more complete inventory, see Talmy 2000:47–68). The first
subgroup, individuation, includes whether or not entities are individuated (bound-

edness), and if so, their unity and relation to their parts, and their multiplicity if
more than one individual is construed.
Such basic structural properties of entities are manifested in the choice of a count
noun, mass noun or pluralia tanta form for nouns, and aspectual inflections for
verbs. Even these properties are a matter of construal. For example, boundedness
is not simply a spatial or material property. A person, star and island represent
individuals bounded spatiotemporally; but a team, constellation and archipelago
are also bounded entities (count nouns) where the speaker has construed them as
whole units with distinct parts; even nothingness can be bounded as in hole or
intermission (Langacker 1987:200–1).
In many cases there are alternative expressions for what appear to be the same
entities that differ in their construal of structure (see especially Wierzbicka 1985).
Examples (1a–b) in §3.1, the count noun leaves vs. the mass noun foliage, is one
case. Foliage construes the entity as a relatively homogeneous substance, without
clear boundaries (a mass of foliage can be borne on several trees). Leaf construes
the entity as a bounded individual, which in turn is part of a single tree; leaves
multiplies the individual, making it truth-conditionally comparable to foliage in the
right contexts. Either construal is available through the lexicon of English. There is
also a quantitative scalar adjustment involved: leaves evokes a more fine-grained
construal than foliage. Similarly, chocolate as a mass noun is a homogeneous,
unbounded substance, while a chocolate is a bounded individual with internal
structure, the characteristic construal of a count noun (conventionalized in this
case; see [7a–b]). Again, the mass noun represents a more coarse-grained scalar
adjustment. Countability also interacts with qualitative scalar adjustment: chair
construes the entity as individuated and of a specific type; furniture construes it
as an abstract mass along with tables, sofas, beds and so on in a coarse-grained
schematization.
The bounded/unbounded structural schematization also applies to states and
processes, as in (8a–b) from §3.1 (Croft 1998a, in prep.). The simple tense/aspect
in Ira is a nuisance construes Ira’s behavior as a temporally unbounded behavioral
trait of Ira that abstracts away from individual instances of nuisance behavior on
Ira’s part. On the other hand, the progressive in Ira, stop being a nuisance! provides
a finer-grained scalar adjustment that construes an individual bounded action of
Ira’s.
Image schemas such as containers or surfaces represent a construal of a more
specific topological or geometric structure of objects. Herskovits 1986 provides a
detailed and insightful construal analysis of geometric structure, only two examples
from which can we present here. There are natural construals of objects that lend
themselves to being containers or flat objects, such as in the box or on the carpet;
but there are also many examples of alternative construals of objects (Herskovits
1986:76):
(60) a. There is milk in the bowl.

b. There is dust on the bowl.
If there is a lot of dust or a few drops of milk, the actual spatial configuration
of figure and ground in (60a) and (60b) is not that much different. But since the
function of bowls is to contain potable liquids, the bowl is construed as a container
with in in (60a), and since dust is thought of as an extraneous substance, the bowl
is construed as a surface with on in (60b).
The geometric construal of an object often requires selective attention (Her-
skovits 1986:65, 67):
(61) a. She is under the tree.

b. The cat is under the table.
c. One could see the shiny silver carp under the water.
In (61a), she is unlikely to be under the ground or inside the trunk (though such
a construal is possible if, say, the suspect is leading the detective to the location
of the body of the murder victim). The usual construal selects only the lower
surface of the foliage as the underside of the object. In (61b), selection ignores
the table legs, and scalar adjustment reduces the tabletop to a two-dimensional
surface. In (61c), selection (driven by encyclopedic knowledge) profiles only the
top surface of the water to specify the figure-ground relationship (and all of these
examples are construed relative to the speaker’s canonical upright orientation, of
course).
Another image schema that imposes a structure, this time more typically as-
sociated with properties, is the scale image schema, which provides a gradable
dimension to a domain, which may or may not be measurable. The ways in which
an entity is construed as possessing a scale or multiple scales are described in
detail in §7.4. Here we simply note that the same domain may be construed with
a scale (in contrast to a polar construal, as in [62a–b] and [63a–b]), or construed
as calibratable, as in (64), a domain not usually considered measurable:
(62) a. Sally’s pregnant.

b. Sally’s very pregnant.
(63) a. Here is a used washing machine.
b. Let me offer you this slightly used washing machine for only
$300!
(64) a. This Sauternes has a fragrant bouquet.
b. The bouquet of the Fargues is twice as fragrant as that of the Climens.
3.5.2 Force dynamics

A second major category of constitutive construals is the force dynamic
model of the conceptualization of events (Talmy 1976, 1988b, 2000). The force
dynamic model is a generalization of the notion of causation, in which processes
are conceptualized as involving different kinds of forces acting in different ways
upon the participants of the event. The examples in (65) illustrate some of the
force-dynamic patterns that Talmy analyzes:
(65) a. I kicked the ball.

b. I held the ball.
c. I dropped the ball.
Example (65a) represents the prototypical causative type: an antagonist (the causer)
forces an agonist (the causee – the ball) that tends towards rest to move. Example
(65b) extends the notion of causation to maintaining a rest state: the antagonist
resists the agonist’s tendency to move. Example (65c) further extends to notion
of causation to enablement: the antagonist acts in a way that allows the agonist to
exert its tendency towards motion.
Croft (1991, 1998b, in prep.) argues that the force-dynamic structure of events
largely determines the encoding of subject, object and oblique arguments of pred-
icates. For instance, the choice of for in I baked brownies for Mary vs. with in I
beat the eggs with a fork is determined by the fact that Mary is the endpoint, the
beneficiary of the baking event, while the fork acts upon the eggs and is therefore
an intermediate participant in the force-dynamic chain. The difference in degree
of affectedness of chew the bone and chew on the bone in (11a–b) in §3.1 is a
(conventionalized) consequence of the alternative construals of the degree of af-
fectedness of the bone evoked by the object-oblique contrast.
Different choices of verbs, or different voice forms, or different argument-
linking constructions, express different conceptualizations of the force-dynamic
structure of the event. For example, (66a) construes the situation as force-
dynamically neutral (being a static situation), but (66b) construes the situation
as having a force-dynamic value of resisting the effects of some (unspecified)
force-applying process.
(66) a. The bowl was on the table.

b. The bowl stayed on the table.
The alternative transitive and intransitive constructions in (67a–b) construe the

event as externally caused or as self-contained (which allows [67b] to be used if
no external agent or force is manifest as well as when the speaker wishes only to
construe the event as such):
(67) a. She opened the door.

b. The door opened.
Force and resistance play a role in the construal of semantic domains other than
causation. For example, Talmy (1988b) and Sweetser (1990, chapter 3) argue that
deontic modals such as may and must in (68a–b) construe the deontic modality
as letting causation or the absence of resistance ([68a]; compare [65c]) vs. the
application of force ([68b]; compare [65a]):
(68) a. You may leave.

b. You must leave.
Sweetser extends this analysis metaphorically to the epistemic meanings of the

modals in (69):
(69) a. She may be ill.

b. She must be ill.
May in (69a) indicates the absence of resistance from concluding that the propo-
sition She is ill is true, while must in (69b) forces one to the conclusion that She is
ill is true.
3.5.3 Relationality (entity/interconnection)

Finally, an even more fundamental constitutive property of entities
is subject to construal. Many semanticists distinguish between relational and
nonrelational entities. A relational entity inherently implies the existence of an-
other entity. For example, an adjectival concept such as ROUND cannot be con-
ceived of without reference to something that is round, and a verbal concept such
as RUN cannot be conceived of without reference to a runner. A nonrelational
entity can be so conceived: for example, a nominal concept such as TABLE can
be conceived of without reference to another entity.
Langacker argues that the difference between nouns (‘things’ in his conceptual
terminology) and adjectives or verbs is that the latter are relational and the for-
mer are not (Langacker 1987:214–17). Thus, in Langacker’s conceptual scheme,
verbs (‘processes’) are construed as relational and sequentially scanned (see §3.2);
adjectives and other modifiers (‘atemporal relations’) are construed as relational
but summarily scanned; and nouns (‘things’) are construed as nonrelational and
summarily scanned.
Langacker’s definition of relationality rests on his definition of things/nounhood.
He argues that nounhood construes a concept as a region or ‘set of intercon-
nected entities’ (Langacker 1987:198); entities are nonrelational. Contrasting
nonrelational things with relational concepts, Langacker describes a relational

concept as profiling the interconnections between entities while a noun profiles
the entities that are interconnected (ibid., 216). For example, one can think of the
(nonrelational) noun circle as profiling the points (entities) that make up the circle,
while the (relational) adjective round profiles the interconnections that define the
circle’s curvature.
Langacker’s definition of relational vs. nonrelational suggests that a construal
as an entity (or set of entities) is a better definition of nounhood than functioning
as a region. A verbal or adjectival concept such as round involves a region also, but
profiles its interconnections instead of its entities. Moreover, Langacker’s example
of circle vs. round also reveals that entity and interconnection are not givens,
but instead involve conceptualization. The noun roundness does not profile the
points/entities making up a round thing; it does not mean the same thing as circle,
or even round object. Instead, roundness construes the shape as an entity, rather
than as interconnections as the adjective round does. In fact, Langacker notes that
anything can be construed as an entity, including interconnections (ibid., 198).
Conversely, a predicate nominal construction like be a circle – which in many
other languages is simply the word ‘circle’ inflected more or less like a verbal
predicate – construes the set of entities as an interconnection, namely the relation
of being an instance of the type defined by the noun (Croft 1991:69–70).
Examples (10a–b) in §3.1 illustrate the entity-interconnection construal. Jill is
fussy construes the trait as relational, and thus introduces a degree of separation
between the behavioral trait and the person (expressed as subject). Be a fussbudget
construes the trait as nonrelational, and therefore saying that Jill is a fussbudget
can only be construed as membership of a category of persons who are defined by
this behavioral trait. Hence (10b) makes out Jill as a more problematic child than
does (10a).
Constitutive construal operations are somewhat different from the construal
operations resulting from the cognitive abilities of attention, comparison and per-
spective. Constitutive construal operations provide a structure to the experience
being communicated. As such, they are not unlike domains, as has been noted for
image schemas, most of which are constitutive construal operations (Clausner and
Croft 1999:16–25). There are many words denoting concepts that must be profiled
in an image-schematic domain, such as more, very, in, part, alike, force (see §2.2;
in fact, such words make up Parts I and II of Roget’s Thesaurus). Image schemas
also have a complex internal structure, like domains:
. . . image schematic gestalts have considerable internal structure – they are

not undifferentiated. On the contrary, it is the organization of their structure
that makes them experientially basic meaningful patterns in our experience and
understanding. The schema for these gestalts have parts and dimensions that stand
in various relationships that allow us to make sense of our experience. (Johnson
1987:61)
Johnson notes that many image schemas are experienced together and describes
this as a superposition of schemas, using the example of things we co-experience
as both near us and central to our vantage point vs. things far away and peripheral:
‘The center-periphery schema is almost never experienced in an isolated or
self-contained fashion . . . Given a center and a periphery, we will also experience
the near-far schema as stretching along our perceptual or conceptual perspec-
tive’ (Johnson 1987:125). The superimposition of image schemas is identical to
the combination of domains in a domain matrix. In fact, image-schematic domains
are usually combined in a matrix with ordinary domains. For example, our expe-
rience of degrees of weight combines the SCALE image schema(tic domain) with
another basic domain, WEIGHT. It is very difficult to separate WEIGHT and
SCALE, but in this respect, WEIGHT and SCALE represent the tightest relation-
ship between domains in a domain matrix, that is, what Langacker describes as
dimensions of a domain (see §2.4).
The analysis of (constitutive) image schemas as image-schematic domains is
not incompatible with their function as construal operations, because domains
themselves are construals, framing the experience to be communicated in a certain
way. What makes image schemas worthy of separate treatment here is their per-
vasiveness in experience: to be communicated, our experience must be construed
in terms of basic structure, scales and force dynamics.
3.6 Conclusion
As the preceding discussion demonstrates, any sentence involves a myriad

of construals of the experience to be communicated. Everything from the choice
of words and their part of speech to the various inflections and constructions that
make up the grammatical structure of an utterance involves conceptualization.
Even fundamental conceptual properties such as the categorization of experiences
and their basic structure is subject to construal. This fact raises two questions, one
more technical and one more philosophical.
The more technical question is: how do the construal operations interact? Two
general observations can be made about the interaction of construal operations.
The first is that construal operations can be nested or iterated (Talmy 2000: 84–
88; Herskovits 1986:57–59; Langacker 1987:138–46). Talmy illustrates nesting
with the examples of a gradually built up set of structural schematizations in (70)
(Talmy 2000:84):
(70) a. The beacon flashed (as I glanced over).

b. The beacon kept flashing.
c. The beacon flashed five times in a row.
d. The beacon kept flashing five times at a stretch.
e. The beacon flashed five times at a stretch for three hours.
In (70a), flash is construed as a single individuated event. In (70b), the event is

construed as iterated. In (70c), the iterated event is bounded by the number of
iterations; (70d) iterates the bounded iterated event of (70c), and (70e) bounds the
iterated complex event of (70d).
Herskovits represents nested geometric construal operations as semantic func-
tions, as in (71) (Herskovits 1986:59–60):
(71) a. The bird is in the bush.

b. Included(Part(Place(Bird)), Interior(Outline(VisiblePart(Place (Bush)))))
The formula in (71b) describes the most natural construal of (71a): a scene in
which (at least) part of the spatial region occupied by the bird is included – the
meaning of in – in the interior space defined by the outline of the visible part of
the spatial region occupied by the bush.
The second general observation is that the layers of construal operations must
yield a conceptually unified construal of the meaning of the utterance (Croft
1993[2002]:163, 194–99). Croft argues that all of the concepts in a single clause
must be construed as part of a single unified domain.8 Examples (72)–(73), for
example, must be construed wholly in the domain of emotion and semantic content
respectively; this requires a metaphorical construal of the spatial preposition in in
(72) and a metonymic construal of the human proper name Proust in (73) (ibid.,
195):
(72) She’s in a good mood.

(73) Proust is tough to read.
In (72)–(73), the alternative construal of one word in the clause (in, Proust) is
driven by the normal or ‘literal’ construal of another word or phrase in the clause
(mood, read). However, all that matters is that the entire clause is construed in a
single domain. It is possible that alternative construals are available for the entire
8 In the original 1993 paper, Croft argues that only immediate clause dependents must be conceptu-
ally unified; the internal structure of argument phrases may be unified around a different domain.
Nunberg’s analysis of metonymy suggests that even argument phrases must conform to the unity of
domain (Nunberg 1995; see also §3.2.1).
clause, leading to ambiguity as in (74), or a construal that is outside the ‘literal’

meanings of any of the words, as in (75) (Croft 1993[2002]:198, 199):
(74) This book is heavy. (physically weighty or emotionally powerful)

(75) The newspaper went under. (construed in the domain of business activity)
The conceptual unity of domain is only one of three conceptual unities that a
clause must obey. The referents in a clause must be construed as belonging to a
single place and time in a mental space (Fauconnier 1985; Croft 1993[2002]:200).
Thus in (76), the hearer must construe the referent of her sister to exist in Margaret’s
belief space (whether or not Margaret has a sister in reality, or whether or not the
description her sister applies to the referent in Margaret’s belief space), and must
construe the referent of a car to exist in the same space (again, whether or not a
specific car fitting the description exists in reality):
(76) Margaret believes that her sister bought a car.
The referents in a clause must also conform to a single instantiation of the event
plus participants (Croft 1993[2002]:201). This unity of selection accounts for the
interaction between verbal aspect, noun countability and adverbial construals, as
in (77)–(78):
(77) a. Sally drinks wine.

b. Sally spilled wine on the carpet.
(78) a. Dan wrote the letters in two hours.
b. Dan wrote letters for two hours.
In (77a), the situation is construed as generic (hence unbounded and not referring
to a specific event); so wine must be construed as referring to the type. In (77b),
the situation is construed as specific and bounded, so wine must be construed
as referring to a specific bounded amount of the liquid. In (78a), the situation is
construed as bounded by the definite noun phrase and the compatible adverbial
phrase in two hours; so write is construed as bounded (telic). In (78b), the situation
is construed as unbounded and specific by the adverbial phrase for two hours, so
the unbounded bare plural letters is construed as specific (not generic as in I hate
to write letters), and write is construed as specific and unbounded.
The more philosophical question that construal raises is, what is the relation-
ship between language, thought and experience? Are there any constraints on the
relationship, and in which direction do the constraints operate: from language to
thought to experience, or the reverse direction, or both?
In a number of places in this chapter, we have referred to the typical construal
of a particular experience: for example, pregnancy is typically construed as not
gradable, a smaller, movable object is typically construed as figure, an action is
typically scanned sequentially, and so on. The nature of our experience in many
instances favors certain construals over others. It is these widespread typical or
default construals that have led semanticists to posit a more rigid model of the
mapping from linguistic meaning to experience. Cognitive linguists emphasize that
flexibility is necessary for understanding conceptualization, and this flexibility is
due to the nature of the human mind as it engages with the world. But experience
does constrain human conceptualization to some degree, rendering some construals
difficult and others almost impossible.
It appears that the pervasive role of conceptualization in linguistic expression
entails a relativistic approach to the relation between language and thought: the
way we conceive our experience is determined by the grammatical structure of our
language. This strongly relativistic formulation is not generally found in cognitive
linguistics, however.
Langacker argues that language-specific semantic structure must be distin-
guished from a universal conceptual structure (Langacker 1976). Langacker rejects
the claim that ‘semantic structure can, in some unclear but hopefully straightfor-
ward way, be related directly to thought and cognition, i.e. the structures manip-
ulated in cognition are essentially the same as the semantic structures underlying
sentences’ (ibid.). In discussing an example from Whorf, who compares the English
He invites people for a feast to its nearest Nootka equivalent, which literally trans-
lates into something like ‘He goes for eaters of cooked (food)’ (ibid., 342–44),
Langacker suggests that both could be expressing the same cognitive experience,
but employ different semantic structures to express the experience. As Lakoff puts
it, ‘experience does not determine conceptual systems, [it] only motivates them’
(Lakoff 1987:310).
Langacker calls language-specific semantic structures conventional imagery.
The semantic representations of Cognitive Grammar, many examples of which
have been used in this chapter, are intended to describe this conventional imagery,
not the presumably universal cognitive structures that these conventional images
construe. Langacker uses the example of the expression of bodily states (Langacker
1976:345) to illustrate his approach: English speakers say ‘I am cold’, whereas
French speakers say literally ‘I have cold’ and Modern Hebrew speakers say ‘It
is cold to me.’ In Langacker’s view, ‘these expressions differ semantically even
though they refer to the same experience, for they employ different images to
structure the same basic conceptual content’ (Langacker 1987:47).
The question remains as to the status of these conventional images in thought.
Langacker suggests that it is relatively ephemeral:
When we use a particular construction or grammatical morpheme, we thereby

select a particular image to structure the conceived situation for communicative
purposes. Because languages differ in their grammatical structure, they differ in

the imagery that speakers employ when conforming to linguistic convention. The
relativistic view does not per se imply that lexicogrammatical structure imposes
any significant constraints on our thought process – in fact I suspect it to be
rather superficial (cf. Langacker 1976). The symbolic resources of a language
generally provide an array of alternative images for describing a given scene, and
we shift from one to another with great facility, often within the confines of a
single sentence. The conventional imagery invoked for linguistic expression is
a fleeting thing that neither defines nor constrains the contents of our thoughts
(Langacker 1991b:12).
Slobin (1991) suggests a somewhat similar view, which he describes as ‘thinking

for speaking’: the conceptualization of experience found in grammar is relevant
for communication, but not necessarily relevant for other cognitive activities.
However, not all construals are conventional; there are many cases of novel
language use that will represent novel construals, and all conventional construals
began as novel. Even some conventional construals are still perceived as ‘loaded.’
Croft (2001, chapter 3) suggests a stronger role for active construal which he calls
the conventional universalist position. When a grammatical structure is used for
the first time (or the first few times) to construe an experience, it does influence the
way speakers think of that experience. But as the extension of that grammatical
expression to the new experience becomes conventionalized – that is, it becomes
the normal or even the only way to talk about the experience – then the original
construal no longer constrains how speakers think of that experience. Evidence
for the conventional universalist position includes the fact that the extended con-
struction displays grammatical behavior incompatible with the original construal
but applicable to the experience being communicated. For example, in French the
expression of bodily states is construed as possession; compare (79a) to (79b):
(79) a. J’ai froid. ‘I am cold’ [lit. ‘I have cold’]
b. J’ai une voiture. ‘I have a car.’
But bodily states can be construed as gradable, which is incompatible with pos-
session (Croft 2001:115):
(80) a. J’ai très froid. ‘I am very cold’ [lit. ‘I have very cold’]
b. *J’ai très une voiture. [lit. ‘I have very a car’]
Construal is a central aspect of language and its relation to thought; but it is

constrained by convention as well as by the experience itself. This hypothesis
guides the approach to categorization in the next chapter.
4
Categories, concepts and meanings
4.1 Introduction
The act of categorization is one of the most basic human cognitive ac-
tivities. Categorization involves the apprehension of some individual entity, some
particular of experience, as an instance of something conceived more abstractly
that also encompasses other actual and potential instantiations. For instance, a
specific animal can be construed as an instantiation of the species DOG, a specific
patch of color as a manifestation of the property RED, and so on. We shall call
this abstract mental construct a conceptual category. Conceptual categories can
be regarded as cognitive tools, and are usually credited with a number of general
functions:
(a) Learning. Experiences never recur exactly: our ability to learn from past
experience would be severely impaired if we could not relate the present to sim-
ilar aspects of past experience, that is, by putting them into the same conceptual
categories.
(b) Planning. The formulation of goals and plans to achieve them also requires
knowledge to be disassociated from individuals and packaged into concepts char-
acterizing categories of entities.
(c) Communication. Language works in terms of generalities, that is, in terms
of categories. Any linguistic expression, however detailed, in the end represents
only a category of referents.
(d) Economy. Knowledge does not (all) need to be related to individual mem-
bers: a significant amount can be stored in relation to groups of individuals. New
knowledge gained on the basis of interaction with one or more individuals can be
easily generalized to other members of category. Conversely, knowing, on the basis
of a limited number of criteria, that an individual belongs to a particular category,
can give access to a much wider range of information about that individual.
There is an important distinction to be made between generic concepts like
CAT and TERRORIST, and individual concepts like TONY BLAIR and CLEOPA-
TRA. The process of categorization presupposes a more basic one, namely, that of
74
Categories, concepts and meanings 75
classifying particular experiences as experiences of one and the same individual

entity. In both cases a Gestalt is formed, but they are different sorts of Gestalt.
Both are construals, in our sense. An adult human has knowledge about thousands
of individual items, not just persons, but also objects, places and so on. A large
proportion of things communicated about and consciously entertained are individ-
ual in nature: consciousness is largely inhabited by individuals. Generic concepts
mostly function to identify and/or characterize individuals. A particular individual
concept is also a bundle of knowledge, perhaps very rich, perhaps sketchy in the
extreme. An individual concept is not itself a final construal, as it is capable of
almost unlimited modulation, particularly via the descriptive content of definite
referring expressions (e.g. that shifty-looking character standing beside the piano).
Such content has a dual function: it contributes to the narrowing down of the search
space in which the referent is located, and it modulates the eventual construal of
the individual concept. Most experimental work and theorizing, both in cognitive
psychology and cognitive linguistics, has been concerned with generic concepts
rather than individual concepts.
Conceptual categories can be viewed from several different perspectives, which,
although connected, should be clearly distinguished. We shall be mainly concerned
with three of these. Firstly, conceptual categories can be viewed as collections of
individuals. The properties of collections are distinct from the properties of the
individuals that constitute them. The two properties that will concern us most
are, first, category boundaries, and second, graded centrality, that is, the fact that
a category typically has a core tapering to a periphery. Secondly, we can look
at a conceptual category from the point of view of the individuals that make up
the category: how can we characterize them, and how can we distinguish them
from members of other categories? Thirdly, there is the question of the level of
categorization. This is partly a matter of inclusiveness – some categories include
others as subcategories – and hence is a relative property, but as we shall see,
there are grounds for proposing absolute levels with definable characteristics. We
shall treat level of categorization as being determined by the type and quantity of
information in the characterization of members of the category.
The view of conceptual categories as fixed cognitive entities with stable as-
sociations with linguistic expressions has been, and still is, the dominant one in
cognitive psychology and linguistics. However, more recently, a dynamic picture
of concepts is emerging, in which they are viewed as being created at the moment
of use. On this view, all aspects of conceptual categories are subject to construal.
This is the view that will be adopted in this book, alongside a parallel view of
word meaning. Before expounding the dynamic construal approach, we first sur-
vey the theories of conceptual structure that have had the greatest influence on the
development of cognitive linguistics.
4.2 The classical model of category structure
The so-called classical model of conceptual categories defines them in

terms of a set of necessary and sufficient features. The features are necessary
in that no entity that does not possess the full set is a member of the category, and
they are sufficient in that possession of all the features guarantees membership.
Thus, the category COLT may be defined by the features [EQUINE], [MALE],
[YOUNG]. This basic idea is of great antiquity, but the immediate inspiration
for its adoption by psychologists such as Collins and Quillian (1969) was its use
in structuralist semantics, and later by Katz and Fodor (1963). In the Katz and
Fodor system, some features were binary and others not. Binary features had only
two values, present or absent. For instance, the definition of FILLY would dif-
fer from that of COLT in the value of the feature denoting sex: COLT would
be [EQUINE], [MALE+], [ADULT−], whereas FILLY would be [EQUINE],
[MALE−], [ADULT−]; MARE would be defined by the features [EQUINE],
[MALE−], [ADULT+], and STALLION by [EQUINE], [MALE+], [ADULT+].
Non-binary features such as [EQUINE] belonged to sets of ‘antonymous n-tuples,’
only one of which may be present in any individual (that is to say, a combination
of features such as [EQUINE], [CANINE], [FELINE] would be impossible). This
picture of category structure is typically accompanied by the ‘nesting assump-
tion’ (Hahn and Chater 1997:47), which states that a subordinate concept, such
as ROBIN, contains as part of its definition the features defining a superordinate
concept such as BIRD (in construction grammar, this is called ‘inheritance’ – see
§10.2.1). Collins and Quillian (1969) incorporated the feature definition of con-
cepts, together with the nesting assumption, in their proposal for a hierarchical
model of semantic memory.
The classical model establishes a clear and rigid boundary to a category. Inclu-
sion relations between categories are also captured, but no account is possible of
absolute levels of categorization (see below). It is important to note that a classical
definition of an entity is not a full description of it or its place in the world: one has
only to think of one’s experience and knowledge of, say, a kitten, in comparison
with the definition [FELINE][DOMESTIC+][ADULT−].
The difficulties faced by the classical model of conceptual categories are many.
Three frequently cited shortcomings have provided the major motivation for
the development of alternative theories. Firstly, for many everyday concepts, as
Wittgenstein pointed out with his well-known example of GAME, adequate defi-
nitions in terms of necessary and sufficient features are simply not available. Fur-
thermore, as Fillmore (1975) pointed out in connection with the noun bachelor,
even for those concepts that seem to have definitions, the definitions typically
hold only within a specific domain (see discussion of bachelor in §2.5). Secondly,
what is here called ‘graded centrality’ constitutes a problem; that is, the fact that
some members of a category are judged ‘better’, or ‘more representative’ of the
category than others: in a classical category, all members are equal. Thirdly, the
classical model can offer no account of why category boundaries, in practice, seem
to be vague and variable (they are frequently described as ‘fuzzy’, but our account
will be somewhat different). A model of category structure is supposed to provide
a basis for an account of how we use categories in remembering, planning, reason-
ing and so on. A classical definition is not a very efficient vehicle for this purpose,
because the information it contains is too sparse.
Several theories of the nature of natural categories have been proposed, mostly
in the psychological literature, but the theory that has had the most influence on
the development of cognitive linguistics is undoubtedly prototype theory, to which
we now turn.
4.3 The prototype model of category structure
The pioneering experimental and theoretical work on prototype theory

was carried out by Rosch and her co-workers (see Rosch 1973, 1978; Rosch and
Mervis 1975), although this built on earlier insights, notably Wittgenstein 1953
and Brown 1958.
4.3.1 Graded centrality

Not all members of a category have the same status within the category.
People have intuitions that some category members are better examples of the
category than others. Members that are judged to be the best examples of a cat-
egory can be considered to be the most central in the category. There has been
a considerable amount of experimental work by cognitive psychologists on the
notion of Goodness-Of-Exemplar (henceforward GOE). The most basic experi-
mental procedure is simply to present subjects with a category and a list of putative
members of the category and to ask them to assign to each member a numerical
score from 1 to 7 according to how good an example it is, with 1 designating a
very good example, and 7 a very poor example or not an example at all. Subjects
reportedly have no difficulty grasping what is required of them. Furthermore, pro-
vided the subjects are drawn from a more-or-less uniform speech community, the
results cluster strongly around particular values (in other words, subjects are not
responding at random). Combining the results from a large number of subjects
allows the identification of the best examples of categories: these are typically
referred to as the prototypes or prototypical members of the category. So, for
instance, if the category was VEGETABLE, the ratings of various items (by British
subjects) might be as follows (these scores represent the ratings of one of the
authors):
GOE rating
LEEK, CARROT 1
BROCCOLI, PARSNIP 2
CELERY, BEETROOT 3
AUBERGINE, COURGETTE 4
PARSLEY, BASIL 5
RHUBARB 6
LEMON 7
GOE ratings may be strongly culture dependent. (Familiarity is undoubtedly a

factor influencing GOE scores, but the scores cannot be reduced to familiarity.)
For instance, in a British context (say, a typical class of undergraduates), DATE
typically receives a GOE score of 3–5 relative to the category of FRUIT, but an
audience of Jordanians accorded it an almost unanimous 1.
The significance of GOE scores is enhanced by experiments showing that they
correlate to a significant degree with a number of independent properties. The
following is a selection of these properties (‘a high GOE score’ means one that is
close to 1).
(i) Frequency and order of mention. When subjects are asked to list as many
examples of a given category as possible, usually within a time limit, the overall
frequency of mention of an item shows a strong correlation with its GOE score,
while the average position of an item in lists correlates inversely with GOE.
(ii) Order of learning. By and large, children learn prototypical members of
categories before more peripheral members. (This may, however, simply be a
function of the frequency of words addressed to them.)
(iii) Family resemblance. Items with a high GOE rating have a higher degree of
family resemblance (measured by sharing of features) to other category members
than items with low GOE ratings, and a lower degree of resemblance to members
of other categories.
(iv) Verification speed. In typical experiments, subjects see two words flashed
onto a screen. Their task is to answer as quickly as possible ‘Yes’ if the second
word denotes a member of the category designated by the first word, and ‘No’
if it does not (e.g. VEGETABLE: CARROT, VEHICLE: CHAIR). The subjects’
speed of response is measured. It is found that responses are faster to items with
a higher GOE score.
(v) Priming. Priming experiments frequently use the lexical decision task:
subjects are presented with a string of letters and have to say as quickly as possible
whether or not the string forms a word. Presenting a semantically related word, or
the same word, before a test item has the effect of speeding up subjects’ responses:
this phenomenon is known as priming. The relevant case here is when the prime
is a category name, like FRUIT. The degree of speeding up is the priming effect.
The priming effect correlates with the GOE score of the category member, that is,
for Britons, FRUIT will speed up the response to APPLE to a greater degree than
the response to, for instance, DATE.
Psycholinguistic variables such as verification speed and priming are regarded
as particularly significant correlates of GOE because they are not under conscious
control and therefore can be claimed to reveal underlying properties of categories.
There has been some dispute in the literature regarding the relationship between
the GOE of an item and its degree of membership (henceforward DOM) in
the category. Some say that, in giving GOE scores, subjects are in fact giving
DOM scores. However, this is misleading. What they were asked to do was to
rate items as to how good they were as members of particular categories. Saying
that they were giving DOM ratings is a subsequent interpretation. Those who
object to the equation of GOE and DOM (for instance, Lakoff [1987:45], Pulman
[1983], Cruse [1992b]), point to examples like OSTRICH in the category BIRD.
There is no doubt, they say, that an ostrich is a fully paid-up member of the
BIRD category, but also undeniably has a low GOE, hence the two parameters
must be independent. Ungerer and Schmid (1996) claim not to see a problem, but
they do not throw any light on the matter. Taylor (1989[1997]) claims that both
assessments of OSTRICH are DOM judgements, but they are made with respect
to differently construed categories. An ostrich is judged a full member relative to
an ‘expert’ category, which has clear membership criteria; the graded membership
judgement is made relative to the everyday category BIRD, which does not have
clear membership criteria. This is ingenious, and we are sympathetic to the appeal
to different construals of categories denoted by the same lexical item, but Taylor’s
account does not stand up to close scrutiny.
The first point to make is that yes/no judgements and graded judgements co-
exist as alternative construals in many semantic domains. Take, for example, the
case of dead and alive. The domain of what might be called ‘vital status’ is often
construed dichotomously: saying John is dead normally commits one to the truth
of John is not alive. But it also possible to say John is more alive than Mary. This
does not change the domain, but reconstrues it as a gradable scale. The same is
true of category membership. In the case of BIRD (whether construed as an expert
category or an everyday one), anything on the right side of the boundary is in the
category, but at the same time, variable centrality allows a gradable construal of
some things as more in the category than others, hence there is some legitimacy
in interpreting GOE as DOM.
At the same time, there is something counterintuitive about saying that an ostrich
is, say, only 30% a bird, and perhaps the term DOM should be reserved for a
distinctive property. We may think of a category as a container (i.e., a result of
the imposition of the Lakoffian CONTAINER image-schema on a domain). What
would we normally understand by a description of an object as 30% in a container?
Surely something like a teddy bear that is partly in and partly out of the toy box,
rather than one that is nearer the side of the box than the middle? There is a
category equivalent of this picture. When we say that, for instance, a priest is to
some extent a social worker, we are effectively placing him part-in, part-out of
the category (cf. Lakoff 1973). That is to say, we are construing the categories
PRIEST and SOCIAL WORKER as partially overlapping. This is surely a more
useful conception of DOM. (NB: the view expressed here regarding DOM is
significantly different from that in Cruse 1992.)
Two problems may be signaled here in connection with GOE experiments and
results. One concerns the meaning attributed, in the context of the experiments, to
expressions such as How good is X as an example of category Y? How, exactly, is
How good? interpreted? Used in actual contexts, good and better do not normally
give rise to communicative problems. For instance, the goodness of a dog, if
thought of as a pet for a young child, is different from what it would be if it was
thought of as providing security for a house, or as contributing to the life of a
farm. This does not destroy the notion of GOE, but suggests that truly significant
results would require specific construals of both the categories being judged and
the meaning of good. Various at least partially distinct notions of goodness can
be teased out (see, for instance, Lakoff 1987:84–90). The following are the main
types:
(i) Typicality/representativeness. This indicates how accurate/useful an idea of a
category one would get from familiarity with only one subcategory. This dimension
has a clear relation to frequency. Lakoff points out that we are much more likely
to generalize properties from typical to non-typical members than vice versa. In
certain cases, a known individual member may be assumed to be typical of a class
(for instance, if a person has limited experience of the class).
(ii) Closeness to an ideal. This is related to what in Cruse 1990 was called
‘quality.’ The example given there was that of emeralds. The most highly valued
emeralds have a deep, intense color, and are without flaws; but these are also the
most rare (and the bigger, the rarer), so they are in no way typical. As Lakoff
points out, ideals operate in many domains. They may be represented by a set of
abstract properties, as in the case of emeralds, or they may be centered around an
individual (called by Lakoff a ‘paragon’).
(iii) Stereotypicality. This is interestingly different from typicality, but a fully
convincing explanation of the difference is not yet available. Lakoff’s account
(1987:85–6) is suggestive, but not fully explanatory. Lakoff says that the use of
typicality is usually unconscious and automatic, whereas the use of stereotypes is
conscious: this is plausible. He also says that stereotypes change, but typicality is
constant over a person’s lifetime. However, typicality changes as reality changes
(think of a typical car or computer or camera), whereas a stereotype can persist
in the face of change. Stereotypes are also typically associated with evaluative
features.
There is another problem. A lot of the classical experiments deal with sub-
categories as category members, rather than individuals: for example, subjects
are given a category such as FRUIT, and a range of fruit types such as APPLE,
STRAWBERRY, MANGO, PASSION FRUIT, DATE, OLIVE and so on for GOE
scoring. Other experiments involve individuals: for instance, the work on proto-
typical colors by Rosch (Heider 1971, 1972, and Berlin and Kay 1969) and others;
also experiments with young children typically use individual items, not category
names. It does make a difference. Using categories as examples of other categories
suppresses properties that can enter into the notion of goodness. This is true of
quality as described above. Another example is the property of well-formedness:
it is all very well saying that an apple is the best example of a fruit, but what if it is
a rotten apple? As far as individuals are concerned, well-formedness is yet another
variety of goodness.
4.3.2 The representation of conceptual categories

Prototype theory comes in two main versions (according to the psychol-
ogist Hampton 1997). In both versions, the linked notions of graded centrality and
best examples occupy a central place. (Linguists tend not to distinguish clearly
between the two versions.) One version represents a concept in terms of a list of
the attributes of category members. This resembles a classical definition except
that the features of a prototype representation are not required to be necessary
and sufficient (although neither of these is necessarily ruled out for individual
features). The centrality of an item in the category depends on how many of the
relevant set of features it possesses: the more it possesses, the better an example
of the category it will be. A feature is justified if, other things being equal, its
presence leads to a higher GOE rating. In some versions, features can be weighted
according to their contribution to centrality, and such versions will set an overall
weighting score as a qualification for category membership. It is possible that no
existing member of a category possesses all the prototype features. In such a case,
the core of the category is represented by the member or members with the highest
feature count (the actual prototype will then be an idealization represented by the
full set of features).
Another version of prototype theory depends on the notion of similarity. A

concept can be thought of as represented by an ideal exemplar, and membership and
centrality of other items can be defined in terms of their similarity to the prototype.
Hampton emphasizes that the two versions of prototype theory are not equivalent.
Simple concepts, such as those involving color or shape, for instance, are better
served by the similarity approach, whereas complex concepts, such as BIRD or
VEHICLE, can best be handled by the feature-list model (see Hampton 1997:88).
The feature-list version of prototype theory accounts for Wittgensteinian categories
such as GAME, for which there is no definition in terms of necessary and sufficient
features, but which show family resemblance relations. The fact that there are no
necessary and sufficient features is also consistent with the fact that when subjects
are asked to supply attributes that characterize a category, they do not confine
themselves to attributes possessed by all members of the category, but also give
features that a significant majority possess. Hampton says that both versions of the
theory give a satisfactory account of the existence of borderline cases. However,
it is difficult to see why a system where category membership depends on the
presence of X features out of a list of Y will generate more borderline instances
than one where membership depends on the presence of Y features out of a list
of Y. On the other hand, one can see how a similarity-based system will throw up
borderline cases, especially if the similarity dimensions vary continuously.
4.3.3 Levels of categorization

Prototype theory also provides an account of levels of categorization.
Categories occur at different levels of inclusiveness, with more specific ones nested
within more inclusive ones:
(1) a. vehicle – car – hatchback

b. fruit – apple – Granny Smith
c. living thing – creature – animal – dog – spaniel
d. object – implement – cutlery – knife – bread knife
e. object – item of furniture – table – card table
Normally, one level of specificity in each set, called the basic (Rosch et al.
1976) or generic (Berlin et al. 1973) level of specificity, has a special status,
and importance. (The basic level items in [1] are printed in bold italic.) Apart
from the basic level, two further levels of specificity with different characteristics
are usually identified: superordinate level and subordinate level. These are not
defined simply by their position in the chain – there are substantive characteristics
that distinguish one level from another. (For an extended discussion of hierarchical
structure in concepts, see Murphy and Lassaline 1997.)
4.3.3.1 Basic level categories

The principal distinguishing characteristics of basic level items are as
follows:
(i) It is the most inclusive level at which there are characteristic patterns of behavioral
interaction.
To appreciate this point, imagine one is asked to mime how one behaves with,
say, a dog: this is not too difficult, most people would mime, for instance, patting
and stroking the dog. But suppose one were asked to mime how one behaves with
an animal: this is very difficult unless one knows what kind of animal it is. The
same is true of furniture relative to chair, and spoon relative to cutlery.
(ii) The most inclusive level for which a clear visual image can be formed.
A similar effect can be observed if one is asked to visualize a member of a

category: it is easy to form a mental image of a non-specific dog, chair or apple, but
virtually impossible to do so for animal, furniture or fruit, without being specific.
It is also the level at which picture-word matching is most rapid.
(iii) The most inclusive level at which part-whole information is represented.
This includes relations between parts. For most superordinate artifactual cate-
gories, such as TOOL, CUTLERY, CLOTHES or FURNITURE, there is no com-
mon part-whole structure for members. Biological superordinate categories show
more regularity in part-whole structure, but there is much less commonality in the
relations between the parts.
(iv) The level used for everyday neutral reference.
A chain of specificity, like those illustrated in (1) above, provides a range of

terms potentially usable for reference to an individual entity. Thus a particular dog
can be simultaneously a spaniel, a dog and an animal. However, unless there is a
specific communicative need, the basic level term will be used for reference (see
Cruse 1977 for more details):
(2) A: I can hear something outside.

B: It’s just a dog/?spaniel/?animal
Basic level terms (i.e., terms whose default construals are basic level categories)
are often felt by speakers to be the ‘real’ name of the referent. Cross-linguistic
studies have shown that they tend to be shorter than terms at other levels, nor-
mally monomorphemic, and are original in the sense of not being borrowed by
metaphorical extension from other domains (Berlin et al. 1973). They are also
more frequently used by parents in talk to children, and thus, not surprisingly, they
are the first words children learn.
(v) Individual items are more rapidly categorized as members of basic level categories
than as members of superordinate or subordinate categories.
A spaniel, say, in a photograph, will be more rapidly categorized as a dog, than

as an animal or a spaniel.
All these properties can plausibly be seen as consequences of what Murphy and
Lassaline (1997:106–7) call the ‘differentiation explanation’ for basic level prop-
erties. Basic level categories represent the best categories into which the immediate
superordinate category can be divided, in terms of:
the degree of difference between members of the category and members of
neighboring categories.
internal homogeneity, i.e., the degree to which members of the category resemble
one another.
informativeness, i.e., how much additional information can be accessed over
and above what the superordinate term gives access to.
Consider the terms animal, dog and spaniel. The category ANIMAL is satisfac-
torily distinct from neighboring categories such as BIRD, FISH and INSECT, and
is also informative, but the degree of resemblance between members is less than
for the category DOG. The category SPANIEL has a high degree of resemblance
between members, but distinctiveness from members of neighboring categories
and extra informativeness are low. The category DOG scores highly on all criteria.
4.3.3.2 Superordinate level categories

Superordinate categories have the following characteristics (NB: the term
‘superordinate’ is not here used in the purely relational sense of ‘hyperonymic’):
(i) Superordinate categories are less good categories than basic level categories,
because although members are relatively distinct from members of neighboring
categories, within-category resemblance is relatively low.
(ii) Superordinate categories have fewer defining attributes than basic level categories.
In experiments by Rosch et al. (1976), where subjects were asked to list the
attributes of basic level items in a superordinate category, few attributes were
generated that could serve as defining attributes of the superordinate category.
However, as suggested in Cruse 1992b, this is perhaps not the best way to elicit
attributes of superordinate categories. The most salient attributes of a category are
those that differentiate it from other members of its default contrast set. The salient
features of a horse are those that distinguish a horse from other animals. Subjects
asked to list the attributes of a horse are unlikely to produce ‘has bones,’ ‘breathes’
and so on, because these are shared by other members of the contrast set; they are
more likely to mention ‘can be ridden,’ ‘has a mane,’ ‘has a long tail,’ ‘neighs,’ and
so on. The only way to get significant attributes of ANIMAL is to set up contrasts
with categories such as FISH, PLANT, INSECT. The same is true of FURNITURE:
it is no use looking at informants’ responses to chair, table and the like. More
revealing would be to ask what features distinguish items of furniture from, say,
curtains, carpets, appliances, fireplaces and windows. Thinking of furniture in
this way suggests that items of furniture are prototypically hard (unlike carpets),
mobile (unlike fireplaces) and are places where things happen (unlike appliances,
which are for doing things with). (For a similar, but independent, analysis of
the category FURNITURE, see Bolinger 1992). However, it remains true that
characteristic features of superordinate categories are fewer, and, as a consequence,
family resemblance relations are less marked.
(iii) Immediate superordinates of basic level categories often have a single-attribute

relation to a higher superordinate category (think of FOOTWEAR in relation to
SANDAL, UNDERWEAR in relation to VEST).
(iv) Linguistically, names for superordinate categories are often mass nouns when
basic level terms are count nouns.
Examples of this are crockery (cups and plates), cutlery (spoons and forks),
furniture (tables and chairs), footwear (boots and shoes), (computer) hardware
(hard disks and modems). Less frequently mentioned are cases where the con-
verse is true: the superordinate is a count noun and the basic level term a mass
noun: metals (iron and copper), beverages (beer and wine), spices (pepper and
coriander). There is never a discrepancy in this respect between basic-level and
subordinate-level terms. Superordinate terms are also frequently morphologically
complex and/or polysyllabic.
4.3.3.3 Subordinate level categories

Subordinate level categories have the following characteristics:
(i) They are less good categories than basic level, because although members have
high mutual resemblance, they have low distinctiveness from members of neigh-
boring categories.
(ii) They are much less informative relative to their immediate hyperonymic category,
hence, when subjects are asked to list distinctive attributes, the lists differ very
little from the lists given for the hyperonymic basic level items.
(iii) They are frequently polymorphemic, the most common pattern being modifier-
head (e.g. teaspoon, rocking-chair).
This is taken by, for example, Ungerer and Schmid (1996) to indicate that they
are distinguished from basic level by a single property, rather than encyclope-
dically (e.g. teaspoon, rocking-chair). However, a distinction needs to be made
between naming practices and conceptual content: the ‘single property’ is a matter
of naming, while there are virtually always unencoded encyclopedic distinctive
characteristics. For instance, although spaniel is a single-morpheme word, and
long-tailed tit is a complex expression incorporating a single property (the posses-
sion of a long tail), the extra specificity in each case over the basic level category
is encyclopedic.
The above account is close to the account given by cognitive psychologists
such as Murphy and Lassaline (1997). Anthropological linguists have also made
extensive studies of the hierarchical organization of categories (see, for instance,
Brown 2002). Their approach differs in many ways from that of the psycholo-
gists. Firstly, they have a strong cross-linguistic orientation. Secondly, the most
extensive studies have been of biological kinds (a distinction is usually made be-
tween ‘folk-classifications’ and ‘expert systems’: most studies are of the former):
some (e.g. Atran 1990) claim that only biological kinds are truly hierarchized,
and among biological kind concepts only ‘general purpose’ categories such as
animal, dog, spaniel, beech, copper beech, bush and so on are hierarchized, but
not utilitarian categories such as vegetable, weed or pet. Thirdly, they recognize
a greater ‘depth’ of hierarchization, and use different terms for the levels. The
following is an example (the equivalent psychological categories are given in
brackets):
(3) beginner plant

life form bush (= superordinate level)
generic rose (= basic level)
specific hybrid tea (= subordinate level)
varietal Peace
The properties attributed to the generic level do not significantly add to what was
said above concerning the basic level. The remarkable constancy of hierarchical
structuring across a wide variety and degree of complexity of cultures suggests
that it is a cognitive universal and probably innate. There is a dispute among
anthropologists as to the underlying motive force for the evolution of classificatory
systems. One school holds that it is driven mainly by utilitarian considerations:
the categories evolved because they were an aid to survival. The other school
holds that the evolution of the systems for classifying biological kinds is driven
by intellectual curiosity. There are a number of reasons for the latter claim. The
systems are remarkably similar across the world, even though the cultures and
their environments differ markedly; many cases are found of distinctions that have
no utilitarian value in the culture that uses them; they tend to coincide to a high
degree with scientific classifications. Brown (2002), while acknowledging the force
of these arguments, points out nonetheless that hunter-gatherer societies typically
have far fewer categories than settled agrarian societies, and suggests that there is
a functional reason for this. Small agrarian communities are typically larger than
hunter-gatherer communities, and when harvests fail, it is very important for them
to have access to alternative food sources, hence a detailed knowledge has survival
value. Hunter-gatherer societies, on the other hand, are typically much smaller,
and their essentially mobile lifestyle makes them much less dependent on food
available in a particular locality; hence detailed knowledge of local flora and fauna
is of less value.
4.3.4 Shortcomings of prototype theory

A number of problems have been pointed out in connection with prototype
theory. Only the major ones are presented here.
4.3.4.1 Simplistic nature of feature list

A major criticism of the prototype model of category structure is that
a simple feature list, even with a relaxation of the requirement that features be
necessary and sufficient, is far too simplistic. Even more sophisticated versions
such as Barsalou’s (1992b) model, based on frames (in the sense of structured lists
of dimensions and values), fail to capture the full range of properties linked in
complex chains of association and causation involved in a typical ‘rich’ concept
such as a natural kind concept. There are various aspects to this excessive simplicity.
One is that it cannot handle context sensitivity. Studies have shown that what is
chosen as the best example of a category can be influenced by indicating a context
for the judgement (Barsalou 1987). Labov (1973) also showed that the boundaries
between adjacent categories can be affected by adding contextual features. Another
concerns the relation between the number of features present and GOE. This
relation is not a simple one of counting how many features are present. There is also
interaction between the features: the effect of the presence of one feature depends
on the presence and values of other features. To give a very simple example, the
feature WOODEN lowers the GOE of a spoon if it is small, but not if it is large.
There must be thousands of interactions like this, some involving several features
simultaneously.
4.3.4.2 The ‘odd number paradox’

The ‘odd number paradox’ has also been put forward as a problem for
prototype theory. Armstrong et al. (1983) found that people will grade ODD NUM-
BERS for centrality, even though the category ODD NUMBER has a clear defi-
nition in terms of necessary and sufficient features. Their proposed solution, the
so-called ‘dual representation’ hypothesis, combines the prototype approach and
the classical approach (Smith et al. 1974). The idea is that concepts have two repre-
sentations, which have different functions. There is a ‘core’ representation, which
has basically the form of a classical definition. This representation will govern the
logical properties of the concept. The other representation is some sort of prototype
system which prioritizes the most typical features, and whose function is to allow
rapid categorization of instances encountered. With this set-up, the odd-number
effect ceases to be a puzzle. However, this conjunction of two theories inherits
most of the problems of both of them: in particular, it reinstates a major problem
of the classical theory that prototype theory was intended to solve, namely, the fact
that for a great many everyday concepts there is no available core definition.
4.3.4.3 Problems with features

There is a problem that afflicts all models of conceptual structure that
traffic in features. What are they, and where do they come from? In most accounts
they seem to be simply other concepts. In other words, concepts are just points
in a concept network. This is curiously reminiscent of the structuralist character-
ization of word meaning (see, for instance, Lyons 1963, 1968). But in that case,
the conceptual system is hermetically sealed off from the world, and it is difficult
to see what explanatory power it can have in terms of human mental activity. To
be really explanatory, the features will have to be ‘grounded in a subsymbolic
level’ (Hampton 1997:91), that is, will have to be, or be shown to relate system-
atically to, non-linguistic features drawn from perception, action, intention and so
forth. Hampton points out an even more vicious circularity. If we map BIRD onto
HEAD by means of a ‘has a’ link, we will lose a lot of information if we do not
make it clear that the head in question is not, say, an elephant’s head, but a bird’s
head.
4.3.4.4 Contrasting categories

Contrasting categories such as CAT, DOG, LION, CAMEL and so on
pose a problem for prototype theory. There is really no explanation for the mu-
tual exclusion relation that holds between them. Such an explanation is logically
impossible unless we incorporate into prototype representations features with neg-
ative weighting (so, for instance, ‘has soft fur,’ ‘purrs when stroked,’ ‘moves by
hopping,’ ‘larger than average human,’ ‘has scales’ would all have negative weight-
ing for dog). It does not appear that this strategy has ever been followed.
4.3.4.5 Boundaries in prototype theory

Prototype theorists have paid insufficient attention to the question of cate-
gory boundaries and their location, and the same is true of many cognitive linguists.
In the following quote Langacker appears to deny the existence of boundaries:
‘There is no fixed limit on how far something can depart from the prototype and
still be assimilated to the class, if the categorizer is perceptive or clever enough to
find some point of resemblance to typical instances’ (Langacker 1991:266).
Lakoff acknowledges their existence, but devotes little space to discussing them.
(Hampton [1991] is one of the few psychologists to propose a version of the
prototype model in which category boundaries are explicitly recognized.) However,
a boundary is arguably the most basic of all the properties of a category. A category
is like a container: one of its major functions is to divide the objects in the world
into those things that are in it and those things that are not in it. This function
cannot be fulfilled without a boundary.
The notion of category boundary hardly needs detailed justification. If A says
That is an X and B says No, it isn’t, then either they perceive the referent of that
differently or they disagree as to the location of the boundary of the category X.
Certain adjectives such as artificial, as in artificial cream, and fake as in a fake
Monet, indicate that the referent does not fall into the category designated by the
noun; regular as in Lakoff’s (1973) Mark Spitz is a regular fish (Mark Spitz was
an Olympic swimmer), for appropriate interpretation, requires the hearer to realize
that Mark Spitz is not, in fact, a fish.1
Apparently well-formed but nonetheless unacceptable inference patterns like
the following can only be explained in terms of boundary location:
(4) a. A car seat is a kind of seat.
b. A seat is an item of furniture.
c. ?A car seat is an item of furniture.
One explanation (due to Hampton – a slightly different account is given in

chapter 6) is that for An X is a Y to be acceptable, it is enough that a prototypical
X should fall within the category Y; it is not necessary for all X’s to fall within
the category. Hence, the pattern of acceptability seen in (4) can be explained by
the disposition of boundaries in Figure 4.1 (heavily shaded areas denote prototype
cores):
1 It should be emphasized that the existence of a boundary does not entail the existence of a ‘core
definition’ with necessary and sufficient criteria, as proposed in what Hampton (1997:93) calls the
‘binary view’ put forward by Smith et al. (1974).
airplane
glider
hang
glider
Figure 4.1 Boundaries of AIRPLANE, GLIDER and HANG GLIDER
Here, the prototype of GLIDER falls within the category boundary of AIR-
PLANE, but HANG GLIDER falls outside it.
There is a psycholinguistic correlate of the position of a boundary, involving
speed of response to a categorization task. If subjects are asked whether or not an
item belongs to a given category, the speed of response depends on how near the
item is to the category boundary: the closer it is, the slower will be the response.
Hence, in (5), the bold items will be responded to slowest:
(5) (FRUIT) apple . . . . . . tomato . . . . . . potato

(VEHICLE) car . . . . . . bicycle . . . . . . chair
(WORD) hand . . . . . . malk . . . . . . pkhq
Another indication of marginal status is the following. Take the case of SHOE as
a member of the category CLOTHES. One would not hesitate to say of a suitcase
that contained underwear, shirts, socks, jackets and trousers, but nothing else, that
it ‘contained only clothes.’ However, one would hesitate to say of a suitcase full
of shoes, that it contained only clothes; on the other hand, there would be no
such hesitation if the contents were all shirts. This is arguably a function of the
marginal status of shoes in the category of clothes compared with shirts. It appears
that we construe the category CLOTHES differently, that is, more generously,
when confronted with a variety of types of clothes, including marginal ones; when
confronted with a collection of shoes, we construe the categories CLOTHES and
SHOES as mutually excluding.
The location of the boundary of a category is independent of its prototype, that
is to say, two categories may have the same prototype but different boundaries;
likewise, two categories may have the same boundaries but different prototypes.
Take the French word corde and its default English translation rope. A questioning
of native speakers of the two languages suggests that the prototypes of the two
categories are very close: both put forward the same sorts of thing as best examples.
However, their boundaries differ. Le Petit Larousse defines ficelle (‘string’) as ‘une
corde mince’; a parallel definition of string as ‘a thin rope,’ would seem very odd.
That is to say, ficelle falls within the (default) boundary of the category CORDE,
but string falls outside the boundary of the category ROPE. The converse case,
of identical boundary but different core, is perhaps exemplified by courage and
bravery in English. It would be hard to think of an act that was a manifestation
of courage but not of bravery, or vice versa. But their core regions are arguably
distinguishable. Student informants were asked to give a relative rating of (6a) and
(6b) as (i) an example of a brave act and (ii) an example of a courageous act.
(6) a. A person jumps into a fast-flowing river in an attempt to save someone who
has fallen in.
b. A person risks his/her career and livelihood by exposing malpractice and in-
justice at the heart of government.
There was substantial agreement that (6a) was the better example of bravery
and (6b) the better example of courage.
A fundamental problem with boundaries is that they do not arise naturally from a
prototype representation. Even in Hampton’s version of the model, the boundaries
are simply stipulated in an arbitrary fashion. Prototype theorists typically say that
natural conceptual categories have fuzzy boundaries. Indeed, this is one of the
main arguments against the classical model. Claimed pointers to fuzziness are, for
instance, the fact that different subjects make different judgements as to the location
of boundaries, and the same subject will make different judgements under different
contextual conditions. Even the psycholinguistic experiment quoted above yields
a borderline region rather than a sharp line. However, it should be pointed out that
even a fuzzy boundary has a location. The notion of a fuzzy boundary will be
critically examined below.
4.3.5 The frame-based account of prototype effects

A simple list of features is inadequate as a representation of a conceptual
category. The notion of frame as described in chapter 2 offers a more satisfactory
picture of a concept as a complexly structured body of interconnected knowledge.
This picture allows a more flexible account of such matters as graded centrality.
Graded centrality can be seen as a matter of the goodness of fit between the
perceived features of some individual, and one or more aspects of the frame that
characterizes an ideal individual in a category.
Three ways in which this occurs can be identified. First, there is the question of
the convergence between the individual and the profiled region of the frame. Take
the case of car and tractor within the category of VEHICLE. Most informants
award car a somewhat higher GOE score than tractor. The reason appears to be
that the ideal vehicle is designed for travel along roads, rather than across fields,
hence there is a better fit between CAR and VEHICLE than between TRACTOR
and VEHICLE. The second type of graded centrality involves items that do have
a traditional definition, like bachelor. In this case, graded centrality can arise
from similarity between the ideal background domain and the actual background
of the individual. As we have seen, the definition operates against a set of cul-
tural background assumptions concerning marriageability: the reason we regard a
(Roman Catholic) priest as not a very good example of a bachelor, even though he
satisfies the basic definition, is that our background assumptions about priests do
not fit our assumptions about an ‘ideal’ bachelor. A third case is when a concept
is characterized by a cluster of ICMs, as in the case of MOTHER (§2.5). Here,
the ICMs behave like features, in that the more of the members in the ideal cluster
are present in a particular instance, the more central the instance is within the
category.
4.4 A dynamic construal approach to conceptual categories
Most views on the nature of categories have had in common a belief

in a constant underlying mental representation of some kind for each category.
However, more recently, a new approach to categories has emerged that challenges
this assumption. For instance, Smith and Samuelson pass a harsh judgement on the
‘fixed categories’ assumption: ‘These foundational ideas of stable categories and
stable concepts, however, have led to little progress. Instead, a steady succession
of theories of concepts have been offered, rejected, resurrected and rejected again’
(Smith and Samuelson 1997:163).
Smith and Samuelson quote a number of experimental results in support of a
proposal that the notion of fixed categories with permanent representations is a
myth. Among these are Barsalou’s (1983) experiments involving ad hoc categories
with no conventional names, like ‘things on a desk that could be used to pound
in a nail,’ or, ‘things to take on a picnic.’ Subjects readily formed new and con-
textually coherent categories that showed the same characteristics as established
categories, including graded centrality and characteristic features. They propose
instead, that categories are inherently variable, and created on-line as and when
needed. This general line of approach is endorsed by Whittlesea (1997), who argues
that there is no hard evidence for the existence of a separate system of abstracted
knowledge, and that the alleged properties of fixed concepts can be given other
explanations.
According to Smith and Samuelson, the elements out of which a concept is
created are past history, recent history, current input. On the topic of past history,
that is to say, accumulated memories of previous experiences, they point out that
each experience has a permanent effect on our ‘ways of knowing,’ and further have
this to say:
Critically, the accrual of these long-term changes provides a source of stability

in a continually changing system. If there are statistical regularities, patterns, in
our experiences that recur over and over again, then as each moment of knowing
is laid on the preceding moments, weak tendencies to behave and to think in
certain ways will become strong tendencies – sometimes so strong that they
will not be easily perturbed and thus might seem fixed. (Smith and Samuelson
1997:175–6)
What is recorded on each past experience will include such things as salient
contextual factors, perceived and inferred relations (causal and other) with other
things, accompanying language and so on. The second element is immediately
preceding mental activity. They adduce the ubiquitous effect of priming as an
example of this. More particularly for concept formation they claim:
[T]here is a pull for coherence from one thought to the next one, for the meaning
of an event to depend on its place in a stream of events. If we think first about
eating and then about frogs we will think differently than if we think first about
ponds and then about frogs. (ibid.)
The final element is a construal of immediate context, including linguistic, per-

ceptual, social, psychological aspects, including current goals and plans, inferences
and expected outcomes, perceived causal relations and so on.
Let us now look briefly at how this dynamic view of concepts impinges on the
three key features of concepts, namely, boundaries, frames and levels.
4.4.1 Category boundaries

It is not difficult to find examples of different placement of category
boundaries in construals of a word in different contexts. Take the example of pet in
English. Nowadays, there are electronic devices that mimic certain characteristics
of animals, except that they are less demanding and less messy: they are sometimes
called cyberpets. Suppose we ask whether these objects are pets or not:
(7) Is a cyberpet a pet?

When this question is put to a typical class of undergraduates, a typical result is that
a minority, but a significant minority, answer Yes, while the majority say No. This
is, of course, a typical ‘fuzzy’ result. Now suppose the question in (8) is asked:
(8) Is a cyberpet a real pet?
The response this time is overwhelmingly No, because the word real encourages
a particular construal of the position of the category boundary. On the other hand,
suppose a scene is set such as the following: an educational psychologist, say, is
advising the parents of a child with behavioral problems, and says (9):
(9) I advise you to get her some kind of pet – even an electronic one might be
beneficial.
In this case, no one in a typical class finds anything anomalous in the psychologist’s
utterance, even though pet is used to include the electronic variety. The expressions
some kind of and even in the context encourage us to construe a broader category
of pets.
Another example is dog in (10)–(12):
(10) A dog has four legs.
At first sight this seems an obvious truth. But what about dogs that have lost one
or more legs in an accident? It seems that when we interpret (10), we construe the
category of dogs to include only well-formed dogs. Yet another construal of the
boundaries of the category of dogs is illustrated in (11):
(11) Dogs are mammals.
Here we construe a category appropriate to biological discourse, which includes

three-legged dogs and wild dogs. Consider also cases like (12):
(12) A dog makes an excellent companion for an old person.
Here we construe a category appropriate to human social behavior, which includes

only pet dogs.
As a final example, consider the construals of bird in (13) and (14):
(13) I wish I could fly like a bird.
(14) We get lots of birds in our garden.
For (13), we must exclude flightless birds and injured birds incapable of flight
from our construal of bird. In (14) we are constrained to interpret bird (if uttered
by an inhabitant of a typical Manchester suburb) as ‘most familiar type of small
garden bird,’ on the assumption that no one would expect to see ostriches or eagles
in their garden.
As we have seen, one of the perceived inadequacies of the classical model of

category structure was that it entailed sharp boundaries, whereas natural categories
were claimed to have fuzzy boundaries. However, the notion of a fuzzy boundary
needs reexamining. Claimed pointers to fuzziness are, for instance, the fact that
different subjects make different judgements as to the location of boundaries,
and the same subject will make different judgements under different contextual
conditions. But all the evidence for fuzziness involves reactions to isolated lexical
items, rather than construals in specific contexts. While the category boundary
construed in response to a lexical item can vary with context, there is no reason
to suppose that there is anything fuzzy about the different construed boundaries.
A boundary is a line of demarcation between ‘inside’ and ‘outside.’ According to
the dynamic construal approach, it is in principle sharp. However, we can have
various degrees of knowledge about a boundary. For instance, we may only know
that it is located within a certain range of possibilities. Uncertainty as to location
is perfectly compatible with the sharpness of a boundary.
Consider the boundary between ‘alive’ and ‘dead.’ The decision criteria vary
according to context and according to what we are talking about. In the case of
human beings, the boundary is a matter of dispute. Consider, too, the ‘human
being’ boundary in connection with the debates on abortion. In both these cases,
the location of the boundary is a matter of dispute and uncertainty, but is not vague,
certainly not to the disputants. The debate presupposes that there is a dichotomous
construal with a determinate boundary. Very often a boundary construal serves
only to categorize specific individuals as inside a category or not, that is, only
a local boundary needs to be construed rather than a complete delimitation of
the category. Lakoff (1987) makes a special case for the fuzziness of categories
such as TALL MAN, which involve a graded property. However, on the dynamic
construal approach, contextualized occurrences even of categories of this type
involve a specific construed reference point on the relevant scale (see chapter 8 for
more detailed discussion).
In conclusion, it is arguable that we do not need the notion of fuzzy bound-
ary: everything can be accounted for by variable construal of a normal, that is,
determinate, boundary.
4.4.2 Frames
Frames/ICMs (in some cases cluster ICMs) are presented by Fillmore
and Lakoff as more-or-less invariant structures having a stable association with
lexical items, which allow for variable boundary construal, presumably in terms
of the goodness-of-fit required between perceived reality and aspects of the frame.
However, although the frame may be relatively more stable than the boundaries, the
dynamic construal approach allows also for variable construal of the frame itself.
The experiments of Barsalou, reported above, where ad hoc categories are formed
that have all the characteristics of established categories, suggest that frames may
be construed on-line. Also, the type of variation shown by the category DOG in
dogs and other pets and Dogs are mammals seems more convincingly explained
by a modulation of the frame, rather than an adjustment in the degree of fit with a
constant underlying the DOG frame. In any case, whether or not frames are subject
to construal, the mechanism of boundary placement is still in need of elucidation.
4.4.3 Levels of categorization

Given that level status is a function of content and relations between
contents, it would not be surprising if variation in level construal were to be
observed between speakers, and within the usage of a given speaker at different
times and in different contexts.
It is not difficult to find cases where different speakers apparently assign items
to different levels. Take the case of categories associated with the word bird. For
some speakers, bird denotes a category at superordinate level which we may call
AVES. This has ANIMAL, FISH, INSECT and so on as sister categories, and as
subcategories at basic level we will find SPARROW, THRUSH, BLACKBIRD and
so on. For other speakers, the default denotation of bird is a basic level category
that contains familiar garden birds, which has as sister categories not only CAT
and DOG, but also less familiar birds such as TURKEY, OSTRICH and EAGLE
(Jolicoeur et al. 1984). Individual bird species such as THRUSH and BLACKBIRD
are subcategories of BIRD at subordinate level. The first picture makes more sense
biologically, as the species are aligned at basic level. But many speakers argue
strongly that, for them, the difference between, say, a sparrow and a thrush is
‘more like’ that between a collie and a spaniel, than like that between a cat and
a dog. Speakers for whom the default construal of bird is BIRD presumably also
have a construal of bird as AVES. Equally, speakers (like the present author) for
whom thrush denotes a basic level category, and whose default construal of bird
is AVES are capable of operating with the other system, if the occasion demands.
Now, the question is: What is the difference between the basic level THRUSH
and the subordinate level THRUSH? It seems that an important factor is richness
of content, in terms of knowledge, memories, connections and so forth. Basic level
categories ideally have rich content and clear differentiation from sisters. While a
competent naturalist will have a relatively rich representation of items such as blue-
tit, swallow, thrush and so on, town-dwellers may well know the names, but very
little else about the different birds, so they will not form satisfactory basic level
items: the names will be little more than placeholders for potential knowledge.
They may be able to form an image of a generic (garden) bird, but not have enough
experience or knowledge to be able to visualize individual species; their patterns
of behavior will be very much the same towards all garden birds. Insofar as they
have both a superordinate level and a basic level construal of bird, these will not
be different level construals of the same category, but two different categories that
fit at two levels.
What happens in the case of a speaker who can operate with two different
systems? Consider the case of a dog-breeder who in his work environment exhibits
basic level characteristic behavior in respect of categories such as SPANIEL,
COLLIE, ALSATIAN, TERRIER. What happens when he converses with a non-
specialist? Presumably many, at least, will adjust themselves cognitively to the
new situation, and revert to the societal default construal of the terms. But do
they actually change their categories? Or do they simply construe a new level for
them? If they actually restructure the categories, for instance by backgrounding
aspects of knowledge that are highly relevant in a professional setting, then they
are effectively creating new conceptual categories. It would seem reasonable to
assume that level cannot be construed independently of content, that is to say, any
observed movement up or down a taxonomic hierarchy will be a consequence of
different construals of the category denoted by a lexical item.
4.5 The dynamic construal of meaning
A major requirement of a satisfactory account of the relation between

words and meanings is to integrate in a coherent picture both the appearance of
determinate structural properties in the lexicon and, at the same time, the apparently
infinite flexibility of meaning in context. A fairly standard way of attempting this
is to locate structure in the lexicon (or at least infer it from lexical entries) and
account for variability by means of pragmatic rules and principles. In this book, an
alternative approach is explored, whereby neither meanings nor structural relations
are specified in the lexicon, but are construed ‘on-line,’ in actual situations of use.
This is not a new idea. It was first suggested within linguistics by Moore and
Carling (1982), and it is not uncommon now among cognitive linguists (for instance
Lakoff and Sweetser [1994] and Croft [2000]); there are also parallel proposals
regarding concepts within cognitive psychology, as we have seen. But proponents
of this approach in the past have typically not applied it to the sorts of problems
that engaged the attention of structural linguists, such as sense relations, lexical
fields, componential analysis and so on (polysemy has been extensively studied,
but arguably such studies pay insufficient attention to structural features such as
sense boundaries). It is not of course denied that the linguistic expressions provide
a vital component of the raw material required for the construal of meaning. But,
as we shall see, they represent only one component among several. On this view,
words do not really have meanings, nor do sentences have meanings: meanings are
something that we construe, using the properties of linguistic elements as partial
clues, alongside non-linguistic knowledge, information available from context,
knowledge and conjectures regarding the state of mind of hearers and so on.
Our account of word meaning will incorporate, albeit in an adapted form, the
basic insights of the dynamic construal picture of conceptual categories. It should
be borne in mind, however, that concepts are not necessarily equatable with con-
textually construed meanings, or, as we shall call them, interpretations. Consider
the following sentences:
(15) Dogs are not allowed in this building.
(16) I like cats, but I can’t stand dogs.
In any situated use of these sentences, the interpretation of dog will be the
same as the concept DOG construed in the same context. However, cases like this,
although widespread, are probably in the minority. Take sentence (17), said in
reference to the family dog, whom someone forgot to feed at mid-day:
(17) Oh, look: that poor animal hasn’t had anything to eat since this morning!
Here we can say that the word animal causes the construal of an appropriate
conceptual category ANIMAL. However, the fully construed meaning involves
an individual concept, namely the family dog, which is itself further construed in
response to contextual factors (which include, among many others, the fact that
the word animal was used in the referring expression, rather than the word dog).
Even without statistical evidence, it seems a safe guess that the bulk of everyday
communication ultimately concerns individual things or people, rather than classes
of individuals.
There are four basic notions in the present account of meaning, namely, contex-
tualized interpretation, purport, constraints and construal and they will be discussed
in that order.
4.5.1 Contextualized interpretation

There is something special about a word actually used in a living context.
As Wittgenstein says, ‘Every sign by itself seems dead. What gives it life? In use it
is alive. Is life breathed into it there? – Or is the use its life?’ (Wittgenstein, quoted
in The Guardian, September 7, 2001).
We shall say that ‘life’ is breathed into a sign when it is given a contextual-
ized interpretation. An isolated sign certainly has semantically relevant properties,
semantic potential, and these properties have an influence on eventual interpreta-

tions, but they are to be distinguished from the interpretations themselves. (Any
attenuated intuitions of meaning that we get from isolated words can be attributed
to default construals of some kind.) Let us look at a couple of concrete examples
that illustrate this point (the following examples are taken from the novel The
Breaker, by Minette Walters):
(18) Bertie was lying on the doorstep in the sunshine as Ingram drew the jeep to a halt
beside his gate . . . The dog raised his shaggy head and thumped his tail on the
mat before rising leisurely to his feet and yawning.
(19) Bibi . . . sat cross-legged on the floor at Tony’s feet . . . nervously [she] raised her
head.
It is not necessary to dwell on these passages in detail. They are cited simply to
draw attention to the ‘deadness’ of the individual signs, in contrast to the vividness
of the interpretations we construct. Think of what a dictionary entry will tell us
about the meanings of, for instance, thump, raise, rise, and compare this with the
detailed picture the words evoke in the passages quoted. And contrast raise in the
dog raised his shaggy head and nervously she raised her head. These are different
actions, not simply because one is performed by a dog and the other by a girl:
the dog’s head actually moves to a higher position relative to the ground; the girl
merely tilts her head so as to look upwards.
Of course, when we construe these scenes, we draw on our stored knowledge
of the behavior of girls and dogs in different circumstances: if Bibi had been lying
face-down on the floor with her chin on a book, say, and someone had asked to see
the book, then the head-raising referred to in Bibi raised her head would have been
much closer to that referred to in The dog raised his head. Clearly these subtleties
are not inherent in the word raise, but they are part of construed meaning in context,
and are construed as a direct result of the occurrence of the word raise (cf. the
discussion of Fillmore’s semantics of understanding in §2.1).
When we encounter a piece of language in the course of normal communication,
there is an instant of comprehension, a kind of crystallization of the perception of
meaning – we know what somebody has said (or written etc.). This is similar to
our recognition of a familiar face, or when we realize that what we are seeing is a
dog and so on. In the case of the face, we do not merely recognize whose face it
is, but at the same instant we see perhaps that the person is tired, or worried, and
the hair is windblown and so on. On further reflection, we might infer what the
person has been doing, or what the cause of worry is. The processing can continue
indefinitely, but there is nonetheless a prior moment of recognition.
Something similar happens when we encounter a piece of language. We recog-
nize in an instant what has been said, but we can go on working out consequences
and further inferences indefinitely. It is what constitutes the focus of our attention
at the moment of understanding that is referred to here as the interpretation of
an expression. Phenomenologically, it is a fairly clear-cut event. It will be use-
ful to distinguish pre-crystallization processes, processes preceding and leading
up to crystallization, and post-crystallization processes. In many approaches to
meaning, there is a determinate starting point for the process of constructing an
interpretation, but an indeterminate end point. For example, in Relevance Theory
(Sperber and Wilson 1986), the starting point is an explicature, and the end point
(insofar as there is one) is an indeterminate series of implicatures of diminishing
strengths. The present model of comprehension has an indeterminate starting point
(a purport) and a determinate end point.
An interpretation resembles a picture in that it is not susceptible of finite charac-
terization in terms of semantic features, or whatever. Any features are themselves
construals. Of course, a meaning must in some sense have a finite neural represen-
tation, but the elements out of which the representation is composed are more like
the pixels underlying a picture on a computer screen: the resulting experienced
picture is a Gestalt and so is an interpretation. The nature of this experience is still
mysterious.
Notice that the above characterization focuses on the hearer. Presumably there
is something in the mind of the speaker which precedes the utterance, but there is
a sense in which a speaker does not know what they have said until they process
their own utterance. A major task for the speaker is to devise an utterance that will
lead to the desired interpretation forming in the hearer’s mind.
4.5.2 Purport
Each lexical item (word form) is associated with a body of conceptual
content that is here given the name purport. Purport is part of the raw material
contributed by the word to processes of construal of an interpretation (the other
part being a set of conventional constraints). A purport does not correspond to
any specific interpretation, even an abstract one, nor does a word, in general, have
a stable association with specific conceptual categories. At the same time, there
is an intuitive sense of coherence among most of the uses of a word. This can
undoubtedly be partly explained by the constant association between word form
and purport.
Purport may consist of a relatively coherent body of content, or it may display
relatively disjunct parts (as in traditional ‘homonymy’); or, indeed any intermediate
degree of coherence or lack of it. (We do not say of a word such as bank, which
has a disjunct purport, that it has two [or more] purports. There is not a great
deal of harm in saying this for extreme cases, except that it obscures the fact
that there is a continuum of ‘disjunctness’ on which these cases occupy extreme

positions.)
Purport is not to be thought of as a variety of construed meaning. Purport is
to interpretation as egg is to omelette, or flour to bread: it is of a different onto-
logical category. Purport is an ingredient of meaning, not a constituent. It cannot
be explained, in general, as an abstract, or superordinate meaning, which becomes
specified in context. Interpretations are not contextual specifications of purports,
they are transformations.
Purport is some function of previous experiences of (construed) occurrences
of the word in specific situations. As such, it is continually developing: every
experience of the use of a word modifies the word’s purport to some degree.
4.5.3 Constraints
Of course, the construal of interpretations is not unconstrained: the con-
straints are many and varied. They vary in strength and may reinforce one another
or cancel one another out (for examples of conflicting constraints, see the dis-
cussion of facets and microsenses in chapter 5). They can also be overcome by
cognitive effort, but the stronger the constraint, the greater the cognitive effort
required to impose a construal that defies the constraint. They also vary in their
stability under change of context. I have classified constraints informally under a
number of headings.
4.5.3.1 Human cognitive capacities

One very basic type of constraint is represented by the nature of the human
cognitive system. I am thinking here both of positive aspects, such as the universal
tendency to impose a figure-ground structure, or other Gestalt principles, such as
closure and so on, as well as of negative aspects such as memory and attentional
limitations.
4.5.3.2 Nature of reality

Another type of constraint is what we shall naively call ‘the nature of
reality.’ Some aspects of experience naturally lend themselves more readily to
construal in certain respects and less readily to construal in other respects. To take
a simple example, we have the choice when speaking of attributes, to construe
them either as being present or absent, or as being present to varying degrees.
Take, for example, the state of being married: this is much easier to construe as
a dichotomy (married:single) than as a matter of degree (very married, slightly
married). On the other hand, if we are thinking of linear spatial extent, this is much
easier to construe as a gradable scale than as a dichotomy. For similar reasons, it
is difficult not to construe a sense boundary between the two meanings of bank,
and this has something to do with the raw/brute reality concerning river banks and
money banks.
4.5.3.3 Convention
Another very important constraint is convention: how the society in which
we live habitually construes situations and uses words, and so on (see Lewis 1969,
Clark 1996 and Croft 2000 for the theory of convention and its limits). There are
two aspects of convention. One is the mapping between word forms and regions
of conceptual content. This in itself constitutes a constraint: the principal source
of difference in semantic potential between one word and another arises from
difference in associated purport. The other aspect of convention is a limitation of
the possibilities of construal of a particular purport. Certain construals, because
of the strength of the constraints, will acquire a special default status, and extra
cognitive effort will be required to impose a different construal. Conventional
constraints are frequently context sensitive, that is to say, given a particular purport
in a particular context (or, more likely, a context-type) conventional constraints
may favor certain construals over others.
4.5.3.4 Context
Last, but not least, there are contextual constraints. These, in general,
correspond to what Clark (1996) calls common ground.
(a) Linguistic context

This is one component (the actional basis) of what Clark calls personal
common ground. We can distinguish three aspects:
(i) Previous discourse, that is, what has been said immediately prior to a given
utterance, obviously constitutes a powerful source of constraints.
(ii) Immediate linguistic environment: the phrase or sentence in which a word
appears strongly constrains its construal. To give an obvious example, we would
tend to interpret bank differently in (20) and (21), simply because of the immediate
linguistic context:
(20) We moored the boat to the bank.
(21) I’ve got no money – I’ll call in at the bank on the way home.
(iii) Type of discourse: under this heading are included such matters as genre
(whether we are dealing with a poem, novel, textbook, newspaper report, personal
letter, friendly conversation, police interrogation etc.), register (whether formal
or informal, if formal, whether technical or non-technical, if informal, whether
jocular etc.) and field of discourse (legal, ecclesiastical, sporting, political etc.).
(b) Physical context

Also important is what the participants can see, hear and so on in their im-
mediate surroundings. This corresponds to Clark’s perceptual basis for personal
common ground.
(c) Social context

This refers to the kind of situation the participants are in and the social
relations between them (including power relations). These can strongly influence
construal.
(d) Stored knowledge

All utterances are processed against the background of a vast store of re-
membered experiences and knowledge, which is capable of affecting the likelihood
of particular construals. For instance, in the case of bank above, our interpretation
of (21) will be different if we know that the speaker is an eccentric recluse, and
keeps his money in a box in the bank of a river. Items (iii) and (iv) are components
of Clark’s communal common ground.
4.5.4 Construal
Construal, in the sense introduced by Langacker, is the central notion in
this account of lexical semantics. It is by means of a series of processes of construal
that an essentially non-semantic purport is transformed into fully contextualized
meanings. For a detailed discussion of construal processes see chapter 3.
4.5.4.1 Chains of construal and pre-meanings

As pictured here, the process of construing a meaning is not accomplished
in a single stage, but is a result of a series of elementary processes, some of which
are serially ordered, others proceeding in parallel. In many cases, as we shall see,
stages in the construal process intermediate between purport and interpretation
may have important semantic properties. Such intermediate stages will be referred
to as pre-meanings. For instance, the construal of boundaries creates pre-meanings
that have logical properties and that are independent of subsequent or concomitant
construal processes (see, for instance, the discussion of autonomous sense units in
chapter 5, and category inclusion in chapter 6).
4.5.4.2 Default construals

Conventional constraints play a vital role in stabilizing language usage
within a speech community, indeed, of making communication possible. They are
represented in the minds of individual speakers, but their origin lies outside the
individual, in the speech community. Different conventional constraints vary in
strength. A weak constraint will yield no more than a favorite, or most likely con-
strual, which can be easily overridden by contextual constraints; a strong constraint
will require heavy countervailing pressure to be overcome. Different aspects of
construal may be independently subject to conventional constraints: the result of
a constraint will rarely be a fully construed meaning, and is much more likely to
be a pre-meaning involving, say, only a boundary placement and leaving scope
for further enrichment by construal. Conventional constraints are not necessarily
independent of context: some may operate only in certain context types. They can
also operate at different levels of specificity, with narrow scope constraints refining
the work of wider scope constraints. Some constraints will be defeasible, but will
govern some aspect of construal if there is insufficient or no indication from the
context as to required construal. The outcomes of these constraints will function as
default construals. Default construals can also be context dependent. It is probably
default construals that give the illusion of fixity of meaning.
4.6 Structural and logical aspects of meaning
One of the main aims of the dynamic construal approach to word meaning
is to achieve a unified account of both hard and soft aspects of word meaning, both
flexibility and rigidity, and to locate the origins of these at first sight contradic-
tory properties. The ‘hard’ properties include sense relations such as hyponymy,
incompatibility, meronymy and antonymy, and the existence of structured lexical
sets (word fields), as well as logical properties such as entailment. For instance,
it is generally accepted that It’s a dog entails It’s an animal. But what does this
mean? And how can it be the case if dog and animal do not have fixed mean-
ings? It will be argued here that such properties properly belong to pre-meanings
resulting from the construal of boundaries, principally, and scales and reference
points. Confident assertions regarding entailment in decontextualized sentences
can be attributed to default construals of boundaries. Hence, flexibility comes
from the nature of purport and the sensitivity of construal processes to contex-
tual factors; conventional constraints ensure that contextual variability remains
within certain limits. Rigidity comes from the operation of image schemas such as
the container schema and the scale schema. Boundary construal can also account
for the appearance of componentiality in word meaning, without the necessity
of assuming that semantic features are permanent elements of the meaning of
a word.
The principle of compositionality states that the meaning of a complex expres-

sion is a compositional function of the meanings of its parts. The view of word
meaning expounded in this book obviously has repercussions on the interpretation
of this principle and its validity. The first point to make is that if the ‘parts’ are taken
to be words, then since, on the present approach, words do not have meanings,
then the principle is uninterpretable. However, we could take ‘meanings’ to be
‘construed meanings,’ in which case we need to consider a slightly revised version
of the principle:
(22) The construed meaning of a complex expression is a compositional function of
the construed meanings of its parts.
This, too, is untenable, however, as it does not allow an adequate role for context
in the construal of the meaning of the complex. The principle could be further
modified to take account of this:
(23) The meaning of a complex expression is the result of a construal process one of
the inputs to which are the construals of its constituent parts.
This allows context to act at two levels: the initial construal of the word meanings,
and at the level of the whole expression. But this is a long way from the original
principle of compositionality. It is more reminiscent of cookery. Is cookery a
compositional art? Certainly, the final result is determined by (a) the ingredients
and (b) the processes applied, so there is an element of compositionality. But it is
not what the proponents of the principle usually have in mind. If we think of global
construals, then they are almost certainly compositional only in the cookery sense.
But there may be aspects of meaning that do obey the classical principle, at least up
to a point. Logical properties are determined by boundary placements, so perhaps
the pre-meanings created by boundary construals behave in the classical way?
This seems plausible: when we construe red hats, we construe a category of red
things and a category of hats, and it seems inescapable that the resulting category
will be the intersection of these two categories. This is only valid, however, for
class membership: we have no guarantee, for instance, that the result will be either
prototypical hats nor that their color will be a prototypical red. In other words, for
certain aspects of meaning, at certain levels of construal, classical compositionality
holds, but not for all aspects or levels.
4.7 Part I: Concluding remarks
With this chapter we conclude the exposition of the basic principles and
key concepts underlying the enterprise of cognitive linguistics as we see it. In
the chapters that follow, these principles and concepts will be refined, expanded
and further illustrated through their application, first, to aspects of word meaning
in Part II, and second, to grammar in Part III. We aim to show that the cognitive
approach to language not only opens up new aspects of language, but also addresses
the traditional concerns of grammarians and semanticists in a more satisfying
fashion.
PA RT I I
Cognitive approaches to
lexical semantics
5
Polysemy: the construal of
sense boundaries
5.1 Introduction
Polysemy is understood here in a broad sense as variation in the construal

of a word on different occasions of use. It will be treated here as a matter of isolating
different parts of the total meaning potential of a word in different circumstances.
The process of isolating a portion of meaning potential will be viewed as the
creation of a sense boundary delimiting an autonomous unit of sense. For instance,
in (1), the meaning ‘river bank’ is as it were fenced off from the rest of the word’s
potential, and presented as the only functionally relevant portion. The fact that
bank can also refer to a financial institution is suppressed:
(1) John moored the boat to the bank.
The operative factor in this case is of course the immediate linguistic context.
Polysemy is interpreted in this chapter somewhat more broadly than the traditional
lexicographic acceptation involving distinct, established senses, but it includes the
traditional view as a special, perhaps in some ways prototypical, case.
On the present account, bounded sense units are not a property of lexical items
as such; rather, they are construed at the moment of use. (The notion of sense
boundaries that are sharp, but subject to construal, distinguishes the present account
from that of Cruse 2000, and also from accounts such as Deane 1988 and Geeraerts
1993, which question the existence of boundaries.) When we retrieve a word from
the mental lexicon, it does not come with a full set of ready-made sense divisions.
What we get is a purport, together with a set of conventional constraints. However,
in particular cases there may be powerful stable constraints favoring the construal
of certain sense units. If the permanent constraints are pushing very strongly in
one direction, a correspondingly strong countervailing pressure will be necessary
to go against them; if the permanent constraints are weak, whether a boundary is
construed or not will depend on other, mainly contextual, factors. We can portray
the total meaning potential of a word as a region in conceptual space, and each
individual interpretation as a point therein. Understood in this way, the meaning
potential of a word is typically not a uniform continuum: the interpretations tend
109
110 Cognitive approaches to lexical semantics
to cluster in groups showing different degrees of salience and cohesiveness, and

between the groups there are relatively sparsely inhabited regions.
The phenomenon can be illustrated briefly using the well-known example of
bank in English. There are various uses of this word related to the notion of the
collection and custody of money or some other commodity, as in a high street
bank, The World Bank, a blood bank, a sperm bank, a data bank and so on. These
form a cluster. There is an intuitively clear discontinuity separating all these from,
for instance, We moored the boat to the bank, The banks of the stream were covered
with brambles, He slid down the bank into the water, I know a bank whereon the
wild thyme blows (Shakespeare) and Ye banks and braes o’ bonny Doon (Burns).
This forms a natural ‘fault line’ which facilitates the construal of a boundary.
Within the ‘custodial’ cluster we can find subclusters. One such cluster has to do
with the custody of money. Even this has subclusters. The exact significance of
bank in terms of images evoked, implicatures and so on is different in She works
in a bank, I must hurry – I want to get to the bank before it closes and I got an
unpleasant letter from my bank this morning.
The distinct sense units that fall under the heading of polysemy are not, in
general, interpretations as these were defined in chapter 4: mostly they are what
were called pre-meanings, that is, units that appear somewhat further upstream in
the construal process than full-fledged interpretations. Recall that an interpretation
may be the result of a chain of construal processes: a pre-meaning is still subject to
further construal. It is, however, more elaborated than ‘raw’ purport. Pre-meanings
can appear at different stages in the construal chain, the result being that one set
of pre-meanings can be nested inside a larger pre-meaning, and several stages of
nesting can occur. (This is not to suggest that all the units construable from a single
purport can be arranged in a single hierarchy.)
There are several aspects to the partitioning of word meaning. One is the nature
of the distinct units that appear; another is the nature of the differentiating factors
separating adjacent units; a third is the nature of the boundary – what are the
consequences of the boundary and how does it reveal itself? All these aspects are
important and interesting. We shall start by considering full sense boundaries and
the units they delimit.
5.2 Full sense boundaries
Full sense boundaries delimit the sort of sense units that include those
that are the stock-in-trade of traditional dictionaries. For us, the class is wider:
traditional dictionaries include only those sense units that are well entrenched in
Polysemy 111
the language and are strongly supported by conventional constraints; however, the
same characteristics can be found in nonce construals.
5.2.1 Homonymy and polysemy

It is usual for dictionaries to make a distinction between homonymy and
polysemy in the ordering of entries on the page: homonymous senses are given
separate main headings, that is to say, they are treated as separate words that have,
accidentally, the same spelling and/or sound. Polysemic senses are listed under
a single main heading and are treated as ‘different meanings of the same word.’
The distinction between polysemy and homonymy will not figure prominently in
the present account, because it has few, if any, consequences in terms of boundary
effects, or the nature of the delimited units.
The distinction can be viewed either diachronically or synchronically. The more
traditional distinction is the diachronic one: homonymous units are derived from
distinct lexical sources, and their orthographical/ phonological identity is due either
to the loss of an original distinction due to language change, or to borrowing,
whereas polysemic units are derived from the same lexical source, being the result
of processes of extension such as metaphor and metonymy.
The diachronic distinction between homonymy and polysemy is a yes/no matter,
and is a question of historical fact, resolvable in principle, if not always in practice.
The synchronic distinction is less firmly based, and is a matter of degree. The
question is whether there is a felt semantic relationship between two interpretations
of a word or not. The question can be more precisely formulated (but still not very
precisely) as follows: is one interpretation a plausible semantic extension of the
other? There are several problems with this characterization. One is that there are
degrees of plausibility: where do we draw the line? Another, perhaps more serious,
is what exactly ‘plausible’ means. There is a difference between (i) ‘I can sense a
connection between the two meanings,’ (ii) ‘I understand your explanation of how
one meaning gave rise to the other’ and (iii) ‘If I had never met meaning B before,
only meaning A, I would understand the word in sense B if I encountered it in a
suitable context.’
5.2.2 Entrenchment
The interests of lexicographers are necessarily focused on aspects of word
meaning that are strongly supported by stable, mainly conventional constraints,
and that have attained some sort of default status. They are more likely to recognize
gross distinctions than subtle ones. This is not a criticism. The possible readings
of any word are nondenumerable. A dictionary can only offer a finite list, so a
high degree of selectivity is inevitable. In the absence of clear criteria, it is also
inevitable that different dictionaries will make different distinctions, especially
where more subtle distinctions, or ones less clearly licensed by convention are
concerned. As far as the present discussion is concerned, there is no difference
between entrenched readings and nonce readings in respect of boundary properties
or the nature of delimited units.
5.2.3 Boundary effects

A sense boundary reveals its presence in a variety of ways. These can
be viewed as different kinds of autonomy of the delimited units. By autonomy is
meant the ability of a unit to behave independently of other units that might be
construed in the same context.
5.2.3.1 Antagonism: attentional autonomy

The feature that distinguishes full sense units from other types of unit is
antagonism. Basically this means that two units are mutually exclusive as foci of
attention. They are in competition, and if one is at the focus of attention, the other
is excluded. This can be clearly felt in the following examples:
(2) We finally reached the bank. (margin of river, financial institution)

(3) Mary was wearing a light coat. (light in color, light in weight)
(4) He studies moles. (animals, skin defects, industrial spies)
(5) If you don’t do something about those roses, they will die. (bushes, flowers)
There are two main consequences of antagonism. The first is the well-known
identity constraint:
(6) Mary was wearing a light coat; so was Jane.
There are strong constraints favoring the construal of two autonomous pre-
meanings, namely, ‘light in color’ and ‘light in weight.’ Because the two readings
of light are antagonistic, we are strongly constrained to construe the same reading
of light in both conjuncts. Two points may be noted here. The first is that the identity
constraint is not absolute: it is possible to construe (6) with different readings for
light in the two conjuncts, but it requires more cognitive work, and the penalty is
a sense of wordplay. The second point is that if we construct the sentence without
verb-phrase anaphora, while there is still arguably some pressure to use a single
reading of light, it is much weaker:
(7) Mary was wearing a light coat; Jane was wearing a light coat, too.
Polysemy 113
The second consequence of antagonism appears when two antagonistic readings

are simultaneously evoked in connection with a single occurrence of a word. The
effect is variously referred to as punning, zeugma or syllepsis. (For some scholars
these are not identical, but they all have the same source. We shall not pursue
the differences here.) It is intuitively instantly recognizable, creating a species of
cognitive tension and often a comical effect, and is presumably due to competition
between the readings:
(8) John and his driving license expired last Thursday.

(9) Dogs can become pregnant at twelve months, but usually live longer than bitches.
The so-called ‘zeugma test’ has been criticized on the grounds that it does not
give consistent results, in that cases occur where an identical semantic contrast
will in some contexts produce zeugma and in other contexts will not, and, most
damagingly, contexts can be found where even the prototypical examples of alleged
ambiguity will not yield zeugma. One type of example is illustrated in (10):
(10) (B is doing a crossword puzzle)

A: The answer to 21 across is bank.
B: But the clue doesn’t say anything about money.
A: True, but not all banks are money banks, you know.
The first occurrence of bank in A’s response to B is not felt by most people to
be zeugmatic, although it ought to be, if bank is truly ambiguous and the zeugma
test is reliable (so the argument goes). It ought to be clear that this is no problem
for the approach adopted here. Whether or not a sense boundary appears in par-
ticular cases is a matter of construal: ‘consistency’ would only be expected if the
boundary was an inherent property of the word, which our approach denies. Now,
in the case of bank, there are certainly strong conventional constraints favoring the
construal of a boundary, which means that in the majority of contexts a boundary
will be construed. However, in (10), no boundary is construed (the two readings
are ‘unified’ – see discussion below), and the above occurrence of bank is not
ambiguous. This occurs only in a very small range of context types. (Incidentally,
any two meanings can be unified in this way.) However, the fact that it occurs does
not constitute a valid reason for questioning the validity of the test.
5.2.3.2 Relational autonomy

Antagonism is indicative of the most marked level of autonomy. There
are other symptoms, but these also appear at lower levels of autonomy and are
therefore not diagnostic for full sense boundaries. One of these symptoms is the
possession by two readings of distinct sets of sense relations. This feature is com-
plicated by the fact that sense relations are themselves basically context-sensitive
construals; for this reason, the more stable the relations are, and the more sup-
ported by conventional constraints, the stronger the evidence they provide for the
autonomy of readings, and hence for the presence of a boundary. An example is
light in (11):
(11) Mary is wearing a light coat.
Here light has two distinct antonyms, dark and heavy. Cases such as light are
to be distinguished from cases such as old. Old might be held to have two distinct
antonyms in an old car (new) and an old man (young). However, there are good
reasons for viewing new and young as jointly constituting the antonym of old. New
and old have a common meaning component ‘has been in existence a relatively
short time’ which forms a satisfactory antonymic partner to an interpretation that
covers the range of possibilities for old, namely, ‘has been in existence a relatively
long time.’ The unity of old is supported by the lack of zeugma in (12):
(12) John’s car is almost as old as he is.
When the relationships in question are of a taxonomic or meronomic variety,

what is important is that purportedly autonomous readings should belong to dif-
ferent lexical fields. The two pre-meanings of bank are good examples. Thus,
bank in the bank of the river has source, bed and mouth as co-meronyms, but
no obvious hyponyms or hyperonyms, whereas in the bank in the High Street,
it has co-hyponyms like building society and insurance company, and financial
institution as a hyperonym; as for meronyms, they would presumably be different
departments of the bank.
5.2.3.3 Compositional autonomy

Compositional autonomy will assume greater importance in connection
with lower levels of autonomy, but it also applies clearly to units delimited by full
sense boundaries. It refers to the fact that in a compositional process – for instance
the modification of a noun by an adjective – one of the participating elements will
engage with, or take as its scope, only a portion of the meaning of the other. That
portion will be said to display compositional autonomy, and it will be assumed
that a boundary has been drawn between it and the rest of the meaning.
The workings of compositional autonomy are very obvious in the case of units
delimited by full sense boundaries. For instance, in a steep bank, the adjective
steep in effect ignores completely the meaning ‘financial institution’ – it is as if it
did not exist; likewise, in a high-street bank, the meaning ‘edge of river’ is cut off
and plays no part.
Polysemy 115
5.2.4 The nature of full sense units

A question that arises in connection with full sense units is this: What is
it about them that makes them antagonistic? An obvious, but not quite adequate,
answer is that they are typically very different semantically. They tend, for in-
stance, to have few components in common, to belong to different domains, to
be of different ontological types and so on. The most prototypical cases show se-
mantic distance on one or more of these dimensions, and generally speaking, the
greater the semantic distance between two readings, the more likely they are to be
antagonistic. However, this is not the whole story, because some very close read-
ings are antagonistic. Take the case of month, which maps onto two sense units,
each clearly defined, each referring to a period of time: ‘period of four weeks’
and ‘calendar month.’ When we use the word month, we must have one or other
of these units in mind: a month’s supply of tablets normally means 28 days; the
first three months of the year would almost certainly be construed as referring to
calendar months. It would hardly be possible to find an ambiguous word whose
component units were closer.
A more revealing approach (but which is still in some ways mysterious) is to
say that antagonistic readings are readings that resist unification. There are three
relevant modes of unification: (i) the assimilation of two items as parts of the same
whole; (ii) the inclusion of two classes as subclasses of a superordinate class; (iii)
englobement – the encompassing of two disparate items as components of a
global Gestalt. On this view, full senses are readings that strongly resist any kind
of unification. Take the case of bank. It is hard to think of the different sorts of bank
as parts of a whole, or as united into a global Gestalt. We might think of a very
general category to which they both belong, such as ‘entity,’ or even ‘location,’
but this is not good enough, because it does not distinguish banks from non-banks.
Actually, it is possible to unify the two readings of bank, in a very restricted range
of contexts (which is why we cannot say that antagonistic readings can never
be unified). Recall the example cited above, which most speakers do not find
anomalous:
(13) Not all banks are money banks.
The first mention of bank here represents a unification of the two units. The
context in this case enjoins us to find a distinctive unifying property for the two
kinds of bank. As it happens, this is not too difficult, although it requires a con-
ceptual shift to a metalinguistic level. The unifying factor is that both concepts are
designated/mapped onto by the same word form.
5.3 Sub-sense units with near-sense properties
We have taken antagonism between sense units to be indicative of the

highest degree of autonomy. However, there also exist sense units that resemble
full senses in many ways, but do not display antagonism, or do so only in restricted
contexts (in the same way that full senses can only be unified in restricted contexts).
The fact that these cases do not show antagonism means that the units in question
can be unified. The different types can be conveniently classified on the basis of
the type of unification involved. Units that have a significant degree of autonomy,
but can be unified into a superordinate category will be termed ‘microsenses’ (the
present account is based mainly on Cruse 2002a); units that have a significant
degree of autonomy, but can be unified to form a global Gestalt, will be termed
‘facets’ (see Cruse 2000a, 2000b, especially the latter). These appear to be the only
possibilities. Although parts of a whole can function independently for composi-
tional purposes (see below), there do not seem to be any part-whole counterparts
to microsenses.
5.3.1 Facets
Facets are distinguishable components of a global whole, but they are not
capable of being subsumed under a hyperonym.
5.3.1.1 Introduction
Although facets display a significant degree of autonomy, they are not
generally considered to represent polysemy in the traditional sense; for instance,
they are rarely given separate definitions in dictionaries. Prototypical examples
of words with facets are the readings of bank (in the ‘financial’ sense) and
book:
(14) bank = [PREMISES] The bank was blown up.
[PERSONNEL] It’s a friendly bank.
[INSTITUTION] The bank was founded in 1597.
(15) book = [TOME] a red book
[TEXT] an interesting book
The items in (16) illustrate cases of words possessing facets analogous to
[TOME] and [TEXT]:
(16) a. letter: a crumpled letter

a moving letter
b. CD: an indestructible CD
a beautiful CD
Polysemy 117
c. film: a 16mm film

a sad film
d. speech: a deafening speech
an incomprehensible speech
Sentences (17a–c) illustrate three facets of Britain:
(17) a. Britain today lies under one meter of snow. [LAND]

b. Britain is today mourning the death of the Royal corgi. [PEOPLE]
c. Britain declares war on North Korea. [STATE]
Other possible examples of facets are: mother [CARE-GIVER], [BIRTH-

GIVER], and chicken [BIRD], [FOOD].
It should be borne in mind that facets as such are not meanings, but pre-meanings,
and are both the result of construal processes and at the same time the subject of
further construal. I shall illustrate the properties of facets using the [TOME] and
[TEXT] facets of book. (In Cruse 2000b evidence was presented for the existence
of a third facet of ‘book,’ which might be characterized as [PHYSICAL TEXT]
[picked out by such predicates as ‘badly printed’/ ‘poor lay-out,’ ‘in Cyrillic,’ ‘hard
on the eyes’ etc.], and which differs from the [TOME] facet in being, for instance,
two-dimensional, rather than three-dimensional etc. While only a few facets may
be strongly favored by conventional constraints, in principle there is no reason to
believe that the facets of a sense constitute a determinate set.)
5.3.1.2 Autonomy in facets

The facets of book show their autonomy clearly in two of the three main
ways discussed above:
(a) Relational autonomy

Facets show full relational independence, in that each facet may partic-
ipate in its own sense relations, independently of other facets. For instance, in
most contexts, the hyponyms of book form two parallel taxonomies, but there is
no relationship between the two taxonomies:
(18) Some of the books we read were novels and the others were biographies.
In (18), book is construed as the facet [TEXT], which stands in a hyperonymic

relation to the incompatible hyponyms novel and biography.
(19) Some of the books were paperbacks, most were hardbacks.
In (19), book is construed as the facet [TOME], which stands in a hyperonymic

relation to the incompatible hyponyms paperback and hardback.
(20) *Some of the books [GLOBAL] were hardbacks, the rest were novels.
As (20) illustrates, novel and hardback do not coordinate in a context that expects
items to be construed as incompatibles within a single domain. The normality
of generic decontextualized statements such as (21a) and (21b) shows that the
potentiality for the construal of facets for book is well supported by conventional
constraints:
(21) a. A novel is a kind of book.

b. A paperback is a kind of book.
(b) Compositional autonomy

Predicates can apply to facets independently:
(22) a red book; a dusty book
In these phrases, the adjective modifies only the [TOME] facet of book. In (23),
on the other hand, the adjectives modify only the [TEXT] facet:
(23) an exciting book; a difficult book
This in itself is only a relatively weak indication of autonomy. More striking is

the ambiguity of a phrase with no ambiguous words and no syntactic ambiguity.
Any predicate that can apply to either facet separately gives rise to an ambiguity
not attributable either to lexis or syntax:
(24) a. two books

b. two books in one
c. a new book
d. a long book
Example (24a) can designate either two copies of the same text (i.e. two ‘tomes’)
or two texts; (24b) is interesting because the numerals two and one, respectively,
modify different facets (yet there is no zeugma); a new book may be a new copy
of a very ancient text, or a copy (whether in pristine condition or not) of a recently
composed text; a long book may be one with lots of words in it, or it may have a
non-canonical physical shape (notice that two long books or two new books have
to be long or new in the same way).
In an appropriate context, a question containing book can be truthfully answered
both in the affirmative and the negative, that is to say, responses can be relative to
one facet to the exclusion of the other. Consider (25):
(25) Do you like the book?
Out of context this is likely to be interpreted as referring to the [TEXT] facet

only, but it is not difficult to envisage a scenario in which it could be construed as
Polysemy 119
referring equally to the [TOME] facet. In such a case, the potential exists for both
Yes and No to be true answers.
(c) Autonomous ‘cores’

An interesting manifestation of autonomy is worth mentioning. Consider
(26) and (27):
(26) I’m not interested in the contents, etc., I’m interested in the book itself.
(27) I’m not interested in the binding, etc., I’m interested in the book itself.
Both (26) and (27) are normally interpretable and involve different readings of
the book itself, namely, [TOME] in (26) and [TEXT] in (27). This will be taken to
indicate that book has two ‘cores’. Compare, however, the same procedure applied
to novel:
(28) ?I’m not interested in the plot, etc., I’m interested in the novel itself.
(29) I’m not interested in the binding, etc., I’m interested in the novel itself.
It seems that it is easy to isolate the [TEXT] facet of novel, but more difficult
to construe an autonomous [TOME] facet. This is also suggested by the relative
oddness (at least in the absence of more specific context) of a red novel, a dusty
novel, a shiny novel, compared with a red book, a dusty book and a shiny book
(we shall return to the ‘novel problem’ below).
(d) Some miscellaneous aspects of autonomy

A textless tome and a tomeless text can both be designated by book:
(30) I’ve got a book to write the minutes of the meeting in.
(31) A: How’s your book going?
B: Oh, it’s all in here [pointing to head], but I haven’t written anything down yet.
This, and the independent countability of facets can be taken as an indication of a

certain degree of referential distinctness.
Also, each facet can have independent metaphorical extensions. In order to
interpret (32), it is only necessary to access the [TOME] facet:
(32) a book of matches
Finally, each facet can have an independent proper name. Middlemarch is the
name of a text, not of a physical object, nor yet of a text-physical object complex.
(e) Attentional autonomy in facets

It is extremely hard to demonstrate attentional autonomy with facets. In
most situations where zeugma might be expected (if facets were full senses), it
does not arise:
(33) It’s a very helpful book, but rather heavy to carry around.
One possible case, however, is (34):
(34) ?John wrote a red book.
What is written is the text, but what is red is the physical object, and combining
the two seems to be cognitively difficult. It is not clear what the conditions are for
the appearance of zeugma here, since it does not appear to be present in (35):
(35) John wrote that red book on the top shelf over there.
However, in the vast majority of contexts, simultaneous activation of distinct

facets does not lead to zeugma.
5.3.1.3 Unifying facets

On the above evidence, one might be tempted to say that the facets of
book are separate senses, related to separate concepts. However, the distinctness of
facets is sometimes overestimated. For instance, the Oxford Advanced Learner’s
Dictionary gives two separate entries for the facets of book: they are listed as
1(a) and 1(b), and are thus not distinguished from senses. It was argued in Cruse
1986 that they were separate senses. However, there is evidence that there exists a
global sense ‘book,’ corresponding to a global concept BOOK, which represents
a unification of the two facets, and which justifies the claim that the facets do not
have the full status of lexical senses. A summary of this evidence follows. (Notice
that facets cannot be unified as hyponyms of a hyperonymic category: there is no
category that subsumes ‘tomes’ and ‘texts.’ Nor are they parts of a whole in the
normal sense.)
(a) Prototypical co-occurrence

If we think of book as a basic-level item, it is clear that the prototype has
both facets: although either can exist without the other, partnerless facets are, to
say the least, peripheral. Naive subjects learn about the dual nature of book with
initial surprise (but ready acceptance, nonetheless).
(b) Joint compositional properties

One aspect of joint compositional properties is that there are predicates
that attach themselves to both facets simultaneously:
(36) to publish a book
It is not possible to publish something that does not comprise both a text and some
physical manifestation.
Polysemy 121
Another aspect of joint compositional properties has already been mentioned,

namely, serial composition without zeugma. In other words, facets do not show
antagonism in circumstances where senses would:
(37) This is a very interesting book, but it is awfully heavy to carry around.
In (37), interesting modifies the [TEXT] facet, and heavy to carry around the
[TOME] facet.
(c) Joint lexical relations

The global reading (which includes both facets) has its own sense rela-
tions. An example of this is the hyperonym/hyponym relation between publication
and book (also, between educational establishment and school, and tourist accom-
modation and hotel, and the incompatibility relation between building society and
bank).
(d) Global reference

Definite noun phrases such as the red book, or that friendly hotel arguably
refer to the relevant global entity, rather than purely to the facet targeted by the
adjective.
(e) Joint extensions

There are extensions of sense that require both facets to be taken into
account for them to be intelligible:
(38) I can read him like an open book.
To interpret read in (38), we must access our knowledge of how texts are processed;
to interpret open, we need to access knowledge of books as physical objects.
(f) Joint nameability

The Lindisfarne Gospels (a medieval text) is the name of a global
[TEXT]+[TOME] entity.
5.3.1.4 Why are facets not full senses?

In the case of bank, the conceptual representation provides strong pressure
for a split reading, and in the case of teacher in (39) and (40), which exhibits none of
the symptoms of autonomy, the conceptual representation provides strong pressure
for a unified reading.
(39) Our teacher is on maternity leave. (construed as ‘female teacher’)

(40) Our teacher is on paternity leave. (construed as ‘male teacher’)
In the case of book, it appears that the stable constraints do not push us strongly
in either direction, so we are (relatively) free to construe the facets either as au-
tonomous or as unified in response to other constraints, such as contextual or com-
municative ones. This is not so much because conceptual constraints are absent, or
inoperative, but because they act in contrary directions, effectively canceling one
another out.
One obvious factor pointing toward an autonomous manifestation of facets is
‘conceptual/semantic distance.’ As we have already noted, the more similar two
readings are, other things being equal, the easier it will be to unify them, and the
more difficult to construe a boundary between them; conversely, the more different
they are, the more difficult it will be (generally speaking) to unify them and the
easier it will be to construe them as autonomous. Clearly, a difference of basic
ontological type, for instance, ‘concrete’ vs. ‘abstract,’ represents a substantial
semantic distance, and will be expected to constitute a constraint opposing unifi-
cation. The facets of book, and those of every multifaceted word that have been
identified, are of distinct ontological types, and therefore it is no surprise that they
show autonomous tendencies.
The problem with facets is rather one of explaining why they are so easily
unified, given their conceptual distinctness. One possibility is that there is a coun-
terconstraint operative here. First of all, the facets typically co-occur in a range
of significant contexts – in fact, they prototypically co-occur; secondly, they do
not simply co-occur, rather, they operate in a kind of functional symbiosis. The
only reason for having the [TOME] is to concretely manifest the [TEXT]; the
text is useless without some physical manifestation. In Langackerian terms, we
can say that they are jointly profiled against a single domain matrix. This acts as
a strong constraint favoring unity. So, in the case of facets, we have significant
constraints favoring a unified construal and constraints favoring the construal of a
boundary. When this state of affairs obtains, whether a boundary is construed or
not in particular circumstances will depend on other factors. For instance, in (28)
and (29) above, the linguistic context motivates the hearer strongly to look for a
reading excluded by the first conjunct, i.e. to construe a boundary.
5.3.1.5 The novel problem

Let us turn now briefly to the novel problem. Why is it that, out of context,
a red book is normal, but a red novel is somewhat odd? It seems that construing a
[TOME] facet for novel is difficult. There are certain usages that at first sight seem
to point to the existence of a [TOME] reading for novel (in that they are normal
even out of context):
Polysemy 123
(41) a. A novel of some three hundred pages

b. A thick novel with many colored illustrations
c. A paperback novel
However, these are not really counterexamples to the general inaccessibility of

the [TOME] reading, because the physical features can be construed as giving
information about the text of the novel (cf. Kleiber 1996). But even a red novel can
be normalized in a sufficiently (and appropriately) elaborated context (a special
intonation is required to get the correct reading):
(42) All the novels are on the right and the travel books are on the left. Incidentally, I
want to show you something – pass me that red novel on the top shelf.
Cases like (43) are presumably similar (cf. example 1 in Fauconnier 1994:143):
(43) A: Pass me the Keats and the Wordsworth: Keats is red and Wordsworth is
green.
B: I can see a blue Keats, but not a red one.
These examples are convincing enough, but are hard to invent and clearly require
special contextual justification. But there are still facts about novel that need to
be explained. For instance, as mentioned earlier, why are the following odd out of
context, while the equivalent expressions with book are not?:
(44) a. ?a red novel
b. ?a dirty novel (in the physical sense)/?This novel is dirty
c. ?a dusty novel
Furthermore, certain contexts that are clearly ambiguous with book require a
considerable cognitive effort to see as ambiguous with novel:
(45) a new novel
two novels
Perhaps most strikingly, as has been already noted, the X itself construction has
two readings with book, but only one with novel:
(46) I’m not interested in the typography or the cover design, I’m interested in the
novel itself.
(47) ?I’m not interested in the plot or the characters, I’m interested in the novel itself.
At the very least, we can say that the [TOME] reading of novel, in spite of the
fact that it informs a good deal of our everyday interaction with novels, for one
reason or another, is much more difficult to construe than that of book.
Why, then, does novel behave in a different way from book? One approach to
answering this question is to look at features of meaning that distinguish items
that behave like book from those that behave like novel. It might be argued, for
instance, that what is distinctive about a novel is its text, not its physical format;
one might even say that the immediate default contrasts of novel are other literary
genres, like poem or short story. In the case of book, on the other hand, the facets
are perhaps more equally balanced. There are two sorts of contrast: first, with other
everyday physical objects such as clock, vase, ornament or in-tray, and second,
with other text-types, such as newspaper, magazine, brochure, directory.
There is a certain intuitive plausibility about this explanation, but on close exam-
ination, the matter is much more complex. Take the case of dictionary. The most
distinctive property of a dictionary is the text it contains. Yet dictionary behaves
not like novel but like book in that, for instance, two dictionaries is ambiguous
between ‘two different dictionary-texts’ and ‘two copies of the same text.’ At this
point one might point out that while a novel does not have a distinctive format,
a dictionary does, at least prototypically, and this might account for the higher
salience of the [TOME] facet. There are two objections to this. One is that there
are other book-like entities with distinctive formats that behave like novel: one of
these is thesis, which at least in a British university has a highly distinctive for-
mat. Yet two theses can only mean two different texts, and not two copies of the
same thesis. The other objection is that novels do prototypically have a distinctive
format, or at least there are possible book formats that are unlikely to be novels:
think of the characteristic large-format art books, or atlases.
A different line of argument is to say that there is no novel problem: novel behaves
as one might expect. The problem lies with dictionary in that it unexpectedly
behaves like book in spite of being defined by its contents. At this point, the
example of bible is perhaps relevant: a bible is defined by its contents, but two
bibles normally refers to two copies of the same text. This is presumably because
the text of a bible is (in everyday experience) unique – there are no others of the
same type. There is some plausibility in the suggestion that, to an ordinary person,
the same is true of ‘The Dictionary.’ That is to say, there is a naive assumption
that there is only one text, and this forces a [TOME] interpretation of plurality
(something similar would occur with, for instance, two David Copperfields). (The
fact that one says Look it up in the dictionary at least as readily as Look it up in a
dictionary is confirmatory evidence.)
There is still a problem, though. Consider the following cases:
(48) a. a tall secretary

b. a burly barman
c. a fair-haired professor of linguistics
Just as novel is defined by its text, secretary, barman and professor of linguistics
are defined by their jobs, yet there appears to be no prohibition on the use of
Polysemy 125
‘irrelevant’ adjectives, as there is with novel and thesis. (It may be that this is an
aspect of the wider problem of restrictions on active zones: why, for instance, can
we say Mary is fair, meaning that she has fair hair, but not ?Mary is blue [without
elaborate contextualization], meaning that she has blue eyes?)
5.3.1.6 Near neighbors of facets

Many word meanings represent an association between ontologically dis-
tinct components, and it is in the domain of such associations that facets appear.
However, the degree of integration of the components varies, and it appears that
the construal of facets is favored by an intermediate degree of integration. Where
there is a lower degree of integration, the components behave more like full senses,
and where integration is higher, the facets lose their autonomy. Consider the no-
tion of a punch on the nose. Here we have an action (concrete) associated with
a sensation (mental). However, signs of facethood are absent, since intense, for
instance, which is normal with pain, is not normal with punch on the nose:
(49) *an intense punch on the nose
This would be contrary to expectations if action and sensation could be construed

as facets. Somewhat more facet-like is factory:
(50) a. The factory was blown up. [PREMISES]
b. The whole factory came out on strike. [PERSONNEL]
c. (50%?) The factory that was blown up came out on strike.
However, out of a typical class of student informants, only about half will accept
(50c) as normal. In contrast, there is unanimity concerning the normality of (51c):
(51) a. A red book
b. A funny book
c. You’ll find that red book on the top shelf very funny.
This suggests that the components of factory are less integrated than those of
book. However, there seem to be cases of dual-nature concepts that are even more
integrated than book, and that show significantly less facet-like behavior. One ex-
ample is woman (many words referring to human beings are similar). Although
informants are intuitively not happy to put woman into the same semantic category
as book, it does shows some signs of facethood (the facets will be referred to as
[BODY] and [MIND], for convenience, but this carries no philosophical implica-
tions). For instance, there are adjectives that attach themselves to one facet or the
other:
(52) a. A tall woman [BODY]
b. An intelligent woman [MIND]
There is some evidence of relational autonomy in that, for instance, the

meronyms are different; and there are separate classificatory systems for body
types and personality types. And consider the following scenario. Suppose we
wired Mary up to a computer which mapped accurately every neuronal connection
in her brain, producing a replica of her mind that one could communicate with, but
unfortunately destroyed the original patterns in the process. We would then have a
body and a mind. Which one would be the ‘real’ Mary? Arguably, one could refer
to either entity as Mary. But other indications of autonomy seem to be weak. For
instance, it is hard to find any ambiguous contexts explicable only by appeal to
facets; and woman (for many informants) fails the X itself test:
(53) I’m not interested in the woman’s body, I’m interested in the woman herself.
(54) ?I’m not interested in the woman’s mind or personality or feelings, I’m interested
in the woman herself.
Another example concerns the verb weigh. This involves a physical action allied
to a mental action:
(55) a. John weighed the potatoes with trembling hands. [PHYSICAL ACTION]
b. John weighed the potatoes accurately. [MENTAL ACTION]
The integration here shows up in the lack of truth-conditional independence in

the two putative facets. In answer to the question Did John weigh the potatoes? one
can only base one’s answer on the global reading; that is to say, it is not sufficient
merely, for instance, to put some potatoes onto a scale to be able to truthfully say one
has weighed them. Likewise, there is no analogue of a tomeless text, or a textless
tome. Nor does, for instance, John weighed the potatoes calmly seem ambiguous
in the way that John writes beautifully is, or even Did you write the letter? (i.e. ‘Did
you compose the text?’/‘Did you produce the physical inscription?’) It is not being
suggested that there is anything unusual about woman or weigh. There are grounds
for believing that it is facets that represent the atypical case. Facets can be viewed
as occupying a relatively restricted position in a theoretical space constituted by
two dimensions: integration and ontological distinctness, both characterized by
variation in degree. With reduced integration, facets are simply polysemic senses
related by metonymy; with increased integration, facets lose their independence.
A reduction of ontological distinctness also leads to a loss of autonomy, giving
rise to normal sister parts of complex objects or events.
5.3.2 Microsenses
Microsenses are distinct sense units of a word that occur in different
contexts and whose default construals stand in a relation of mutual incompatibility
Polysemy 127
at the same hierarchical level, rather like, for instance, the names of animals such
as cat, dog, sheep, cow, pig, horse. Typical examples of words with microsenses
are knife and card. Consider how these words are understood in the following
contexts:
(56) John called the waiter over to his table and complained that he had not been given
a knife and fork.
(57) The attacker threatened the couple with a knife.
(58) I got a card the other day from Ralph, who’s on holiday in Tenerife.
(59) Let me give you my card; let me know as soon as you have any news.
These are to be contrasted with the readings of knife and card that appear in (60)
and (61), respectively, which are hyperonymic to the readings in (56)–(59) above:
(60) You can buy any kind of knife there.

(61) The box was full of cards of various sorts.
Words like knife and card have a hyperonymic reading and a cluster of hypony-
mous readings, whose default construals are sister incompatibles. All of these
units exhibit a significant degree of autonomy. It is the specific units, which are
here termed microsenses, that do the bulk of the ‘semantic work’ of the lexeme.
The hyperonymic construal has a secondary role. It requires positive contextual
pressure for activation: it is never the default selection (for instance, in [60] and
[61] any kind of and of various sorts are contextual triggers for the hyperonymic
reading). Furthermore, it is relatively ill-defined, and does not have an established
place in any lexical field. The words knife and card show what may be termed
default specificity, that is, when we encounter them, our first assumption is that
one of the specific construals is intended, and we look for evidence as to which
one.
A natural reaction at this point is to say that this is a purely pragmatic matter – the
different construals of knife and card in the above examples are no different from
those of friend in (62) and (63) (i.e. ‘female friend’ and ‘male friend,’ respectively):
(62) My best friend married my brother.

(63) My best friend married my sister.
However, there are significant differences. Firstly, in (62) and (63) we see merely
different contextual construals of a purport that is essentially neutral with respect
to sex. The readings ‘female friend’ and ‘male friend’ do not exhibit autonomy.
The neutral purport is inferentially enriched in different ways in (62) and (63) in
response to the contextual elements married my brother and married my sister.
In the absence of contextual pressure, the neutral reading of friend will be con-
strued. This type of construal, which does not involve the creation of autonomous
intermediate units of sense, will be termed contextual modulation. Microsenses,

on the other hand, constitute autonomous sense units, and their hyperonymic read-
ing requires contextual pressure.
In order to substantiate the above claims regarding the special status of mi-
crosenses it is necessary to lay out the arguments in detail. First we need to es-
tablish that microsenses are not merely contextual modulations; then we need to
distinguish them from both full senses and facets.
5.3.2.1 Why microsenses are not contextual modulations

The argument that microsenses are not contextual modulations has two
strands: the first involves evidence that the specific readings have too much au-
tonomy, and the second involves evidence that the hyperonymic reading does not
have default status.
Evidence for the autonomy of the specific construals comes from a number
of sources. First, they can show relational autonomy, in that each has its own
independent set of sense relations. The microsenses of knife, for instance, give rise
to construals belonging to different taxonomies, and which therefore have different
hyperonyms, hyponyms and co-hyponyms:
(64) cutlery: knife, fork, spoon

weapon: knife, gun, cosh, grenade
instrument: knife, scalpel, forceps
(garden) tool: knife, spade, fork, trowel, rake
(DIY) tool: knife, screwdriver, hammer, plane
A second indication of the salient individuality of microsenses is that they

exhibit truth-conditional autonomy. An indication of this is when a yes/no question
can be truthfully answered No, based on one microsense, when the hyperonymic
reading, or another microsense, would require a positive answer. The following
are examples:
(65) Mother: (at table; Johnny is playing with his meat with his fingers): Use your
knife to cut your meat, Johnny.
Johnny: (who has a pen-knife in his pocket, but no knife of the proper sort) I
haven’t got one.
(66) Tom: (who has a football under his arm): Let’s play tennis.
Billy: Have you got a ball?
Tom: No, I thought YOU had one.
Of course it is the case here that context makes clear what sort of ball is relevant,
and this means that if there were no appropriate microsense available for selection,
contextual modulation would lead us to much the same meaning. It is important to
Polysemy 129
emphasize, however, that truth-conditional autonomy is not an automatic property

of contextually enriched readings. For instance, in (67), although context makes
clear what sort of car would be appropriate, B’s answer cannot be based on the
contextually specified reading:
(67) A: (There are 6 people to transport) Do you have a car?

B: (Who has a 2CV) *No.
Yes, but it’s too small to take us all.
Microsenses also give rise to an identity constraint. There is a palpable pressure

to interpret the second conjunct in each of the following sentences with the same
reading of the word in bold as was chosen for the first conjunct:
(68) a. John sent a card; so did Mary.

b. John has some equipment; so has Bill.
c. John needs a knife; so does Bill.
The constraint appears to be stronger in some cases than others: this may be
a reflection of degree of autonomy. But the contrast with (69a) and (69b), where
there is no constraint, is clear:
(69) a. John has a cousin; so has Mary.

b. John has a car; so has Mary.
An important aspect of the individuality of microsenses is that they function

as basic level items in their home domains. For instance, at table, knife is a basic
level item, alongside spoon, fork and so on (the relevant reading of fork is also
a microsense), and displays all the characteristic properties of basic level items.
This is not any sort of knife, but the specific variety that falls under the hyperonym
cutlery (at least in British English). Although there are other sorts of knives, in
the appropriate context there is no need for a specifying epithet: and, indeed, most
speakers would be hard put to supply one. In contrast to microsenses, there are no
cases where contextually modulated readings like those in (62) and (63) function
as basic level items.
The hyperonymic readings of microsense complexes (i.e. the combination of
hyperonym plus a set of microsenses) are distinct in a number of ways from those
of contextual modulations. One of the most striking properties of hyperonyms
which subsume a set of microsenses is their default specificity. Words like knife,
card and equipment have a strong preference for specific use, and this renders them
odd (to varying degrees) in a minimal context:
(70) a. Do you have any equipment?

b. Do you possess a card?
c. Do you have a knife?
The examples in (70) may be contrasted with those in (71):

(71) a. Do you have any children?
b. Do you have a car?
In (71a-b) there is no pressure for specificity, and they are not (automatically)
anomalous in contexts that do not sanction a more specific interpretation. A corol-
lary of the property of default specificity is that the hyperonymic readings of words
such as knife require overt contextual pressure for their activation: the mere ab-
sence of specifying pressure is not enough. Thus, although the questions in (70)
are to some degree odd, those in (72) and (73) are not:1
(72) Do you have a knife/card/ball of any kind?
(73) a knife wound; a knife-sharpener
The hyperonyms of sets of microsenses seem to correspond to somewhat non-

prototypical concepts. For instance, they do not seem to have clear relational
properties, or at least not readily accessible ones. In spite of the fact that a knife
is intuitively an everyday object, speakers are notably hesitant in suggesting ei-
ther a hyperonym or other member of the same contrast set that applies to all
knives, whereas they have no problem with, say, dog or, indeed, with one of the
microsenses of knife. Furthermore, in terms of the distinction between subordinate
level, basic level and superordinate level conceptual categories, the hyperonymic
reading of knife seems to be associated with the latter type of category. There are
resemblances between the inclusive category KNIFE and typical superordinate
level categories such as FURNITURE. For instance, both have a rather schematic,
impoverished nature, which is different from the rich content of typical basic level
categories. Also, they are not associated with clear visual images or patterns of
behavioral interaction.
It is a curious fact that, while (linguistically untrained) speakers readily accept
the dual nature of multifaceted words like book, they are reluctant to accept the
composite nature of knife. It seems that the unity of the concept is more salient than
1 The account of default specificity given in Cruse 2000a is not quite correct. There it was suggested
that questions like those in (70) are odd if the context does not sanction the selection of one of the
microsenses. However, it was also suggested that there are situational contexts, for instance camping,
for which there is no corresponding microsense. It would seem to follow from this that questions
like those in (70) are invariably odd in such contexts. However, this is not true: (i) seems normal:
(i) (in a camping context) I need to cut this rope. Do you have a knife?
Relevance considerations here narrow down the range of suitable knives: they have to be capable
of cutting the rope in question. But this looks like contextual modulation rather than the selection
of a microsense. It seems therefore that the default specificity constraint is satisfied by any sort of
specification, whether or not it involves a microsense, and it is the unrestricted reading of the hyper-
onym that requires overt contextual justification.
Polysemy 131
its components. Speakers readily offer unified dictionary-type definitions such as

‘an implement with a handle and a blade used for cutting.’ However, the initial
intuitive plausibility of this definition diminishes slightly on closer examination.
Firstly, it does not differentiate knives from saws, chisels and other cutting imple-
ments, and secondly, it contains several words with microsenses just like those of
knife, so it must be interpreted as something like: ‘an implement with a knife-type
blade and a knife-type handle, used for knife-style cutting’; in other words, we
need to know what a knife is before we can make intuitively satisfying sense of
the definition.
5.3.2.2 Why microsenses are not full senses or facets

Microsenses are not distinct senses because they are unifiable under a
hyperonym, and are therefore not, by the usual criteria, antagonistic. The reason
microsenses are not facets is because facets are of different ontological types and
cannot be subsumed under a hyperonym.
5.3.2.3 Microsenses and facets compared

Microsenses are in some ways a kind of converse of facets in that, whereas
for facets, the difference of ontological type would lead us to expect them to
be autonomous, but the default construal is in fact a unified one, in the case of
microsenses, their close similarity would be expected to produce unified readings
but, in fact, it is the specific readings that are in some sense the default case. Hence,
for microsenses, we need to identify a strong countervailing constraint favoring
the placement of boundaries. The answer once again seems to be the domains in
which the microsenses operate. The notion of domain was invoked to unify facets,
whereas with microsenses, it is the different domains in which the microsenses
operate that confer distinctness. Thus, semantic distance between readings may be
counterbalanced by a strong tendency toward joint relevance, whereas semantic
closeness can be counterbalanced by distinctness of habitual contexts. (Remember
that in the case of both facets and microsenses, the countervailing constraints
are sufficiently closely balanced for contextual constraints to have a significant
effect.)
5.3.2.4 A Langackerian explanation for microsense behavior

Two main lines of explanation will be explored here. The first is broadly
Langackerian in spirit (cf. Tuggy 1993). Within this framework there are two possi-
ble approaches to microsense complexes, which are not necessarily contradictory.
The first approach is based on different levels of entrenchment (which will be taken
to be equivalent to the strength of conventional constraints) for the hyperonymic
reading and the specific readings, as illustrated in (74):
(74) cousin: hyperonym: heavily entrenched; rich

specifics: pragmatically generated
knife: hyperonym: lightly entrenched; sparse
specifics: heavily entrenched; rich
card (paper vs. plastic): hyperonym: pragmatically generated
specifics: heavily entrenched
bank: hyperonym: non-existent
specifics: heavily entrenched
There is a graded shift of emphasis from hyperonym to hyponym(s); although

only four stages are shown, the scale is to be thought of as continuous. The rep-
resentation of card needs further comment. The distinction between the different
sorts of ‘paper’ cards would be like that of knife. What is illustrated in (74) is the
distinction between paper and plastic cards. Many speakers report that they find it
difficult to interpret sentences like (75) to include both types:
(75) The box contained various types of cards.
However, plastic and paper cards can be united in special contexts such as (76):
(76) The box contained a variety of plastic and other sorts of cards.
A distinction is being made here between hyperonymic readings that must be

‘overtly licensed’ by the context (as with knife) and those that must be ‘contextually
coerced’ (as with card); the latter are considered not to be entrenched, but prag-
matically generated. (Again, of course, we are dealing with a continuous scale, this
time of resistance to unification, and the cutoff point is to some extent arbitrary.)
To what extent can the above picture throw light on the nature of microsense
complexes? Let us assume that every occurrence of a use of a word with a particular
reading increases that reading’s entrenchment, and that increasing entrenchment
enhances the permanence, salience and accessibility of the reading.
Clearly, if a range of specific readings is more accessible than the hyperonymic
reading, then there will be a preference for a specific reading, if there is one that
fits the context. More can be explained if we assume that each reading represents
a point in a profile-base chain, or, thinking in terms of the Langackerian notion of
domain matrix, a point of intersection of a number of profile-base chains. Now it
may be presumed that the profilings against different bases can differ in salience.
Since there is a common hyperonym for all the members of a set of microsenses,
then they must all have in common a profiling against the same base. But let us
suppose that that is not the most salient profiling for any of them: each has a most
salient profiling against a different base (thinking of cutlery, weapons, surgical
instruments etc.).
Polysemy 133
This will explain the seeming anomaly of sets of incompatibles sharing the same
word form. These readings are not really in competition with one another – the
sister units are only half-sisters. It also plausibly motivates their truth-conditional
autonomy. If the different microsenses were not domiciled in different domains, it
would be difficult to explain why, say, child does not develop microsenses corre-
sponding to ‘boy’ and ‘girl,’ since it must frequently be used with such reference.
It is possible, then, that mere frequency of occurrence is not enough to cause auto-
nomy to develop, but that some other differentiating factor must be present. In the
case of microsenses, this extra factor could plausibly be occurrence in different
domains; in the case of facets, although they are not differentiated by domain
(sister facets tend to occur together), they are differentiated by ontological type.
What does this account fail to explain? Two apparently contradictory things.
Firstly, it does not explain why immediate intuition goes for the hyperonymic
reading. In other words, people need convincing that the word shows default
specificity: unlettered intuition is aware more of the unity of ‘knife’ than of its
plurality. Secondly, given that this is so, what would we expect in a context that
does not sanction the selection of one of the specifics? Would we not expect the
hyperonymic reading to emerge? But it does not – it only appears when it is overtly
sanctioned, that is, one component of default specificity is not accounted for.
There is another way of looking at microsense complexes within an overall
Langackerian approach. The above approach implicitly pictures microsenses as
developing from an initial unified reading by ‘entrenchment + differentiating
factor.’ Perhaps this might explain the intuition of unity of words like knife, but it
is not a wholly plausible scenario. A more plausible scenario is that we start with
distinct readings that then become united. Langacker’s network model works like
this (cf. Langacker 1991: 266–71). We start with knife being used to designate a
particular kind of knife; then this is metaphorically extended to refer to a distinct
type of implement, with sufficient resemblance to the original to justify the use of
the same word; then a hyperonymic reading develops, which subsumes these two.
In this case, without taking differential entrenchment into account, the difference
in ‘richness’ of the resulting concepts may be governed by how much common
content can be abstracted from the specific readings. Obviously, if child developed
this way, then the common abstracted content would be rich, but in the case of
knife, much less so.
However, the ‘bottom-up’ scenario fails to explain the intuitive primacy of the
hyperonymic reading of knife. One possible explanation for this is that metalin-
guistic functions of language obey different rules to ‘actual situated use.’ Perhaps
for metalinguistic purposes, semantic similarity is the stronger constraint, whereas
in everyday use, domain allegiance is more important. Perhaps there is enough se-
mantic commonality in the various microsenses of a word like knife to fuel a unified
response in reflective metalinguistic use, but not enough to counteract the domain
distinctness in normal situated use. If so, then we can count a ‘metalinguistic’

frame as one of the ‘overt contextual sanctions’ for a hyperonymic construal.
5.3.2.5 A ‘latency’ approach

There is another phenomenon, which exhibits a number of resemblances
and parallels with the phenomenon of microsenses, but which is not easily assim-
ilated to the above account of microsenses. Since this second phenomenon cannot
be sharply distinguished from microsenses, an account of the latter that throws no
light on the former is necessarily incomplete. The phenomenon in question is what
Matthews (1981:125) calls latency.
A prime example of latency is the following:
(77) Look out, Mary’s watching!
Here a specific direct object has to be recovered from context by the hearer. Notice
that the identity constraint is operative here:
(78) John’s watching; so is Bill.
This is to be contrasted with, for example (79), where there is no identity con-
straint:
(79) John’s reading; so is Bill.
Furthermore, different readings have truth-conditional autonomy:
(80) Is John watching?
In answering (81), one is only obliged to answer with respect to the latent direct
object. Suppose the latent element is us; then we can answer ‘No’ to the question, as
long as John is not watching us, and it does not matter if he is watching something
else. Notice, too, that the autonomy can be overridden in contexts that make the
objects explicit; (81) is not zeugmatic:
(81) John is watching Mary, and Bill, Susan.
This example of latency is of course syntactic in nature, and the latent element is
a separate word. Perhaps a little closer to present concerns is the example of cub:
this has a kind of default specificity, not unlike that of knife, which requires us to
identify which kind of cub is being talked about, if it is not made explicit. Thus,
for example, if an identification cannot be made, the result is odd:
(82) I was walking through the woods this morning and I spotted a cub.
Similarly (83) is odd if what I saw were a lion cub, a fox cub and a bear cub.
Polysemy 135
(83) I saw three cubs this morning.
The identity constraint operates here, too:
(84) John saw a cub, and so did Mary.
And different readings have independent truth-conditions. The question Did you
see any cubs? can be truthfully answered No if the subject of conversation/common
interest is foxes, and the speaker did not see any fox cubs, irrespective of whether
he/she saw any other kind of cub. Specifying the type of cub explicitly, removes
the restrictions and allows a hyperonymic reading:
(85) The lioness and the vixen were playing with their cubs.
This sort of behavior is exhibited by a wide range of words, such as handle,

wheel, cover, blade, lid, leg, thigh, top, patient, to name but a few. (Is it possible
to make sense of John is a patient, without knowing whose patient he is? There is
no such thing as a patient ‘tout court.’)
There are some obvious resemblances between these examples of latency and
microsenses. As we have seen, with latent elements, we find default specificity,
a hyperonymic reading accessible only under contextual pressure, identity con-
straints, independent truth-conditions. All this looks very much like what we find
with microsenses. A possibility, therefore, is that knife-type words and words
that show latency basically exemplify the same phenomenon. That is to say, mi-
crosenses might correspond to deleted specifying expressions that must be re-
covered from the context before an interpretation can be carried out. This is a
different explanation from the one offered above for microsenses, especially if
we were thinking in terms of the meaning of knife as a single mapping onto a
complex concept. It is much closer to pure pragmatics, since there will only be a
mapping to the hyperonymic concept, and the detailed readings will be left to con-
textual constraints, as opposed to being selected from entrenched alternatives (i.e.,
having a strong conventional component). It would still not be simple contextual
modulation, however, because of the obligatory nature of specification.
There are a number of possible objections to the latency analysis for microsenses.
First of all, although in some cases an appropriate specifying expression can be
found corresponding to a microsense (for instance, postcard, Xmas card and so
on), in many cases there appears to be no readily available qualifier. This is true,
for instance, of the knife and fork used at table and, for many speakers, of the card
bearing one’s name, address and so on that one gives to people one meets. This
seems to argue against saying that when we speak of knives and forks at table we are
deleting a specifying epithet. The fully rounded basic-level status of these concepts
also seems distinct from the possible interpretations of cub. Another objection is
the ‘openness’ of latency. If we leave everything to pragmatics, then the list of

possibilities is open-ended, whereas the range of possible readings of knife-type
words seems to be too highly constrained: for instance, the list of possible implicit
qualifiers for ball includes tennis and golf, but excludes small, soft, green and
rubber.
However, when we look at the cases of latency suggested above, we find that in
many cases, at least, the range of choice for specification is not completely open.
It is true that the possible direct objects of watch in (78) and (79) seem to be un-
constrained except by the meaning of watch together with pure contextual factors.
But other examples vary in the freedom with which the implicit qualifier can be
implemented. For instance, for thigh, any type of animal that has thighs is possible,
and the list is open in the sense that a newly discovered animal would instantly be
added to the list, but, for example, plump or long or sunburnt are excluded. For
cub, the restrictions are even greater: only types of animal are allowed, not young,
small, brown, or whatever, and among animals, only certain ones count (not, for
example, cat, dog, sheep, horse etc.). These cases begin to look not very different
from the case of ball, cited earlier as an example of a microsense complex. (There
is a curious tie-up, both for ball and cub, with the notion of ‘kind of,’ discussed in
Cruse 1986, Cruse 2002b and chapter 6: a tennis ball is a kind of ball, but a large
ball is not, and the qualifiers of cub have to be kinds or types of animal.)
In other words, the more restricted the choice, the more difficult it is to dif-
ferentiate the latency account from the microsense account. In the case of sentir
it appears literally impossible to decide. In certain contexts this either means
‘to smell’ or ‘to taste’:
(86) Jean peut sentir l’ail.
This is to all intents and purposes ambiguous: French informants say they have to
intend one meaning or the other, one cannot be non-committal. Furthermore, there
is a strong identity constraint in (87):
(87) Jean peut sentir l’ail; Marie aussi.
However, the two readings can coordinate without zeugma, provided the sense
modalities are made explicit:
(88) Jean sentait l’odeur du citron et le goût de l’ail.
So perhaps the best solution is to propose a continuum ranging from the pure
latency of watch, where the choice of ‘filler’ is purely pragmatic, through cases
like cub where the choice is open along one dimension, but constrained in others,
to cases like knife, where the choice is restricted to a range of entrenched units.
Polysemy 137
But we are still lacking a satisfactory explanation for the difference between the
types.
5.3.3 Ways-of-seeing
What are called here ways-of-seeing are derived from what Pustejovsky
calls qualia roles (Pustejovsky 1995). These are in some ways analogous to the-
matic roles, but instead of detailing ways arguments may attach to a verb, they
govern ways in which predicates can attach themselves to nouns. Pustejovsky
proposes four qualia roles:
The constitutive role: this describes the internal constitution of the object, and
refers to such matters as material, weight, parts and components.
The formal role: the features under this heading serve to distinguish the ob-
ject from other objects within a larger domain, and refer to such matters as
orientation, magnitude, shape, color and position.
The telic role: this describes the purpose and/or function of the object.
The agentive role: the matters referred to here are, for instance, how the object
comes into being, whether an artifact, a natural kind, created by whom or
what, what causal chain leads up to it, and so on.
Qualia roles are reconceptualized here as ways-of-seeing (henceforth WOS) as
follows:
The part-whole WOS: views an entity as a whole with parts (e.g. a horse, as
viewed by a vet).
The kind WOS: views an entity as a kind among other kinds (e.g. a horse as
viewed by a zoologist).
The functional WOS: views an entity in terms of its interactions with other
entities (e.g. a horse as viewed by a jockey).
The life-history WOS: views an entity in terms of its life-history, especially its
coming into being (e.g. a book as viewed by an author or publisher).
The degree of autonomy of ways-of-seeing is less than that of facets or mi-
crosenses. Ways-of-seeing do not correspond to distinct concepts, and they are
not referentially distinct: they represent different ways of looking at the same
thing. There is some limited evidence of relational autonomy, which is one fea-
ture that differentiates them from the active zones considered below. Take the
example of hotel. This can be viewed as a piece of real estate, in which case it
will contrast with houses, offices, factories and so on. Or it can be viewed as a
type of accommodation, in which case it will contrast with B & B, Youth Hostel
and so on. Ways-of-seeing provide a possible explanation for certain cases of
ambiguous phrases with no ambiguous words and univocal syntax, that is to say,
they apparently show compositional autonomy:
(89) a complete soldier

Part-whole WOS:‘a soldier with no body parts missing’ (imagine alien visitors to
Earth collecting human specimens)
Kind WOS: ‘soldier with all the conventional qualities of a soldier’
(90) an expensive hotel
Kind WOS: ‘a hotel that is/was expensive to buy’
Functional WOS: ‘a hotel that is expensive to stay at’
Life-history WOS: ‘a hotel that is/was expensive to build’
(91) a delightful house
Part-whole WOS: ‘a house that is delightful to look at (due to the harmonious
distribution of its parts)’
Functional WOS: ‘a house that is delightful to live in’
There are problems in connection with ways-of-seeing. They are less well sup-
ported than facets or microsenses. It is not clear, for instance, that autonomy is
greater between perspectives than within ways-of-seeing. A number of cases of
WOS-like differential construals are difficult to assign unambiguously to one of
the proposed possibilities. Also, the number proposed seems to be largely arbi-
trary. At the same time, something of the sort is clearly needed: there is a need for
a construed pre-meaning unit intervening between facets and the purely composi-
tionally relevant features and active zones. It would be entirely consistent with the
approach adopted here to refrain from enumerating possible ways-of-seeing, and
to treat the above examples merely as particularly salient possibilities, where it is
relatively easy to construe a (minor) sense boundary.
5.3.4 Semantic components and low-autonomy active zones

The lowest (positive) degree of autonomy is shown by certain active
zones (see §3.2.1) and semantic components: these are units whose sole mani-
festation of autonomy is compositional; they do not constitute pre-meanings in
the full sense. Semantic components are intensional in nature: they are parts of
a more inclusive sense which are compositionally active; active zones have an
extensional/referential basis: they are parts of something which are isolated for
compositional purposes.
As an example of semantic components, take the case of an overworked stallion,
and the two readings ‘overworked in terms of stud duties’ and ‘overworked in terms
of pulling carts.’ We may assume an analysis of ‘stallion’ as [EQUINE][MALE].
One possible explanation for the ambiguity is that there are two possible sense
units to which the modifier ‘overworked’ may be attached:
Polysemy 139
(92) ([EQUINE][MALE])(overworked)
[EQUINE]([MALE](overworked))
That is, the animal is either overworked as a male or as a horse, generally. In other
words, a semantic unit [MALE] may be construed as separable, and can function
as scope for the predicate [OVERWORKED]. (Notice that the same construal is
not so easy for an overworked husband.) The two readings can coordinate without
zeugma:
(93) Dobbin is overworked both as a begetter of foals and as a puller of haycarts.
Another example involves negation.
(94) It is not a mare.
Here the construal of a boundary separating off [FEMALE] would seem to be

the default reading, that is, the sentence would normally be taken to mean that the
animal in question was a stallion that is, only the feature [FEMALE] is affected
by the negation, the feature [EQUINE] being presupposed. However, in a suitable
context where femaleness was apparent, the animal might well be a cow, which
would involve negation taking the whole of ‘mare’ as its scope and [FEMALE]
being presupposed. In yet another context, what is presupposed might be ‘possible
birthday presents.’
Possibly similar cases are ‘an old friend’:
(95) [HUMAN] (old)

|
[RELATIONSHIP] (old)
Here again, negation would preferentially apply only to the [RELATIONSHIP]

component.
(96) The astronaut entered the atmosphere again.

either: ([BECOME] [IN]) again
or: [BECOME] ([IN] again)
(97) I almost killed him.
either: almost ([CAUSE][BECOME][DEAD])
or: [CAUSE][BECOME] (almost [DEAD])
In some approaches to semantics, such semantic components are held to be

inherent properties of lexical items, indeed, permanent constituents of their mean-
ing. In the current approach, they are not fixed, but are construed as and when
needed. Of course some construals are inherently easier than others, and some are
more subject to conventional constraints than others (there is obviously a relation
between these two). The delimitation of [MALE] in ‘stallion’ is so well entrenched
that ‘mare’ is the default construal of ‘not a stallion.’ But, for instance, it is not
difficult to devise a scenario where not a horse will not be construed as ‘some non-
horse animal,’ but as ‘some non-horse means of transport’: the particular situation
will determine what is salient, and the possibilities would seem in principle to be
open.
Some active zones have the same sort of degree of autonomy as semantic com-
ponents, that is, they manifest themselves only in compositional terms. Like com-
ponents, they can give rise to ambiguity, as in red eyes (‘bloodshot eyes’; ‘pupil
is red, as in a photograph’ [potentially] ‘the iris is red’) and a red pencil (‘writes
red’; ‘red on the outside’). Notice that although the parts of a pencil selected as
the domain of relevance of the adjective red are readily distinguishable in reality,
and can be pointed to (unlike [MALE] and [HORSE] for stallion), a definite noun-
phrase like that red pencil, whichever interpretation is given, does not refer to the
part in question, but to the whole pencil.
5.3.5 Contextual modulation

With contextual modulation, as in the ‘male cousin’ reading in My cousin
married an actress, we come to the end of the scale of autonomy, the zero point.
Different contextual modulations certainly represent different construals and are
not in any sense less important communicatively than different microsenses or
facets. What distinguishes them from all the other cases we have considered is that
they do not require the construal of distinct autonomous pre-meanings as part of the
total construal process. The specifying features of different contextual modulations
are, as it were, contributed by the context, not selected, or their creation triggered,
by context.
5.4 Autonomy: summary
The construal of autonomy in a sense unit is the result of complex inter-

actions among a variety of constraints, and cannot, in general terms, be regarded
as having an inherent association with specific lexical items. Although, in specific
cases, the strength of contextual constraints may give the illusion of inherentness,
the balance between conventional, cognitive and contextual constraints is in fact
continuously variable across instances, and an assumption of inherentness does
not give a satisfactory general account of variation. At the same time, autonomy
itself is a variable, rather than an all-or-nothing property. No adequate account of
variable word meaning can afford to ignore the property of autonomy.
6
A dynamic construal approach to sense
relations I: hyponymy and meronymy
Sense relations such as hyponymy, incompatibility and antonymy have been a

topic of lively interest for lexical semanticists since the structuralist period (see, for
instance, Coseriu 1975; Geckeler 1971; Lyons 1963, 1968; Cruse 1986). Although
Lyons (1968) declared that all sense relations were context dependent, they have
almost universally been treated (including by Lyons himself) as stable properties
of individual lexical items.
Cognitive linguists, for the most part, have had very little to say on the topic.
In this chapter and the next, we reexamine a number of sense relations from the
standpoint of the dynamic construal approach to meaning. Sense relations are
treated as semantic relations not between words as such, but between particular
contextual construals of words. We hope to show both that sense relations are a
worthwhile object of study (even for cognitive linguists) and that the dynamic
construal approach can throw new light on their nature. This chapter focuses on
hyponymy and meronymy.
6.1 Hyponymy
6.1.1 Introductory
The following are examples of linguistic expressions whose semantic
well-formedness depends on hyponymy (X is hyponymous to Y):
(1) Xs are Ys (Koalas are marsupials)
(2) Xs and other Ys (Koalas and other marsupials)
(3) Of all Ys, I prefer Xs. (Of all fruit I prefer mangoes.)
(4) Is it a Y?
Yes, it’s an X. (Is it a tit? Yes, it’s a coal-tit.)
(5) There was a marvelous show of Ys: the Xs were particularly good. (There was a
marvellous show of flowers: the roses were particularly good.)
All the above examples involve nouns. A similar relation can be found between
items belonging to other parts of speech:
141
(6) Did she hit him?

Yes, she punched him in the stomach.
(7) Is your new skirt red?
Yes, it’s a maroon velvet.
However, pairs of lexical items related by hyponymy are far more frequently found
among nouns than among adjectives or verbs.
In principle the relation of hyponymy is a simple one (contrary to the view
expressed in Cruse 2002b), and can be seen as an instance of the operation of
the Lakoffian container image schema (see, for example, Lakoff 1987:271–73).
Hyponymy can be regarded as simple class inclusion:
(8) a. the class of dogs is a subset of the class of animals

b. the class of instances of water is a subset of the class of instances of liquid
c. the class of instances of murdering someone is a subset of the class of instances
of killing someone
d. the class of scarlet things is a subset of the class of red things
The notion of inclusion can also be pursued intensionally. This yields some
insights and is more directly related to semantic concerns, since we are dealing
with meanings. It therefore promises greater explanatory power, but at the same
time raises some problems. To characterize hyponymy intensionally, we need to
say something like:
(9) If X is a hyponym of Y, then the semantic content of Y is a proper subpart of the

semantic content of X.
In one sense this is obvious enough. For instance, if we assume that the relation
of hyponymy applies to semantically composite expressions as well as to single
lexical items (this convention is not universally followed, but it is harmless), then
a red hat is hyponymous to a hat in that the class of red hats is a proper subset of
the class of hats. Intensionally, we can see that the meaning ‘a red hat’ contains
as a proper part, the meaning ‘hat’ (taking a naive view of compositionality). This
picture seems readily transferrable to single lexical items such as stallion and
horse: the meaning ‘horse’ is a proper part of the meaning ‘stallion,’ which can be
analysed as ‘horse’ + ‘male.’ (This does not conflict with what was said in chapter
5, as long as ‘male’ and ‘horse’ are viewed as potentially autonomous components
of a particular construal of ‘stallion,’ and not as inherent properties of the words.)
One way of testing for meaning inclusion is in terms of entailment relations be-
tween sentences (more properly, propositions) containing the relevant lexical items
in corresponding structural positions. One formulation, which is not restricted to
nouns, goes as follows:
(10) X is a hyponym of Y iff F(X) entails, but is not entailed by F(Y).

Hyponymy and meronymy 143
This sort of definition, but not in this exact form, was first put forward by Lyons
(1963). Here, F(–) is a sentential function satisfied by X and Y.
Thus, the fact that It’s a dog unilaterally entails It’s an animal indicates that
dog is a hyponym of animal. The reasoning behind this definition is obvious. The
entailment definition has the advantage that it does not require us to specify what
the components A, B and C actually are. Unfortunately, it also has a couple of
disadvantages. First, a hyponymous relation between X and Y does not always
lead to entailment between F(X) and F(Y) (Basil became a Catholic does not
entail Basil became a Christian – he may have started out as a Baptist). Secondly,
entailment between F(X) and F(Y) does not guarantee hyponymy (The wasp stung
John on the knee entails The wasp stung John on the leg – knee is not a hyponym
of leg.) However, there is no reason in principle why these disadvantages cannot
be overcome by a more careful formulation of the test.
6.1.2 Hyponymy and context

A difficulty with the notion of hyponymy, which was noted in Cruse
2002b, is that naive subjects classify pairs like dog:pet as hyponyms, even though
It’s a dog does not entail It’s a pet. Furthermore, dogs and other pets, which
would appear to be diagnostic of hyponymy, is judged fully normal. In Cruse 1986
the relation between dog and pet was labeled ‘para-hyponymy.’ Cruse (2002b)
argued that a correct definition of hyponymy would be one that matched native
speaker intuitions, but he failed to provide such a definition. This difficulty does not
arise with the present approach. Since sense relations do not hold between words as
such, but between specific construals of words, there is no necessary inconsistency
between the normality of dogs and other pets, and the lack of entailment between
dog and pet. The frame Xs and other Ys induces construals of X and Y such that
the former is hyponymous to the latter. The context-dependence of sense relations
was strongly emphasized by Lyons (1968), but in fact it played a surprisingly small
role in his lexical semantics; here, it is a central feature of the approach. Strictly
speaking, the relation of hyponymy depends on only one aspect of construal,
namely, the setting of boundaries. Hyponymy could therefore be said to be a
relation between pre-meanings.
Contextual variation in boundary placement was illustrated in §5.4.2.1. Here we
concentrate on variation with particular relevance to hyponymy.
Relations of meaning between construals of different words in the same dis-
course are important because they are frequently necessary for both discourse
cohesion and the well-formedness of inference patterns. Take a simple example of
an expression pattern whose well-formedness depends on a relation of hyponymy,
such as X and other Ys. Consider the case of dogs and other pets, which most
speakers find perfectly normal. The claim here is that the construals of dog and pet
in this context are such that dog is a hyponym of pet, in that the class of dogs is a
proper subpart of the class of pets. This is in spite of the fact that most speakers
judge that the truth of This is a dog does not guarantee the truth of This is a pet.
How can these two apparently contrary positions be reconciled? They are recon-
cilable because two different construals of dog are involved. The construal in This
is a dog, therefore it is an animal involves some sort of default construal of dog
(and animal), either one that emerges in minimal contexts or one that is triggered
by the ‘logical’ domain evoked by the fairly unfamiliar sentence type; the construal
in dogs and other animals, on the other hand, is strongly constrained by the X and
other Ys format, which requires the construals of X and Y to be adjusted so that
hyponymy holds. The adjustment can affect either X or Y or both: what emerges
if there is no other context is the result of the easiest adjustment, that is, involving
the most easily accessible construals (there may also be an additional factor of the
contextualizability of the result). For instance, the construal of dog in dogs and
other pets is probably more specific than the default zero context construal, but
is easily accessible; in handbags and other weapons, however, the construal of
weapons is less specific than the default zero context construal.
Construability is not infinitely flexible. Sufficiently strong conventional con-
straints can prevent hyponymous construals from emerging. For instance, dogs
and other cats is virtually unconstruable in any imaginable context (at least it is
not literally construable). There are other constraints, too. For instance, dogs and
other dogs is not acceptable, even though there are readings of dog that would
satisfy the requirement of hyponymy. This is perhaps because of a constraint that
discourages the repetition of a form if a different construal is required.
As defined above, hyponymy is a transitive relation, based on containment,
which is also transitive. There are claimed examples of nontransitivity in the liter-
ature, which our conception of hyponymy does not permit. The apparent cases of
transitivity failure provide interesting examples of different construals of the same
word in different contexts. Two such examples are as follows:
(11) A car seat is a type of seat.

A seat is an item of furniture
*A car seat is an item of furniture.
(12) A hang glider is a type of glider.
A glider is a type of aeroplane.
*A hang glider is a type of aeroplane.
These examples are based on informant responses (see Hampton 1991).

Hampton’s solution to this problem, as has already been noted, is to say that
for X is a type of Y to be acceptable, it is sufficient for prototypical Xs to fall within
the category boundaries of Y. This assumes that X and Y are lexical items, and that
they denote fixed conceptual categories. Our account is different, as it assumes
that informants’ judgements are based on construals in context. We suggest that
hearers are predisposed to look for nonanomalous interpretations of input utter-
ances, and hence an utterance of X is a type of Y will be judged acceptable if there
are easily accessible construals of X and Y such that X is hyponymous to Y. For
instance, A car seat is a type of seat is judged normal because there is an accessible
construal of seat that includes car seats. The sentence A seat is a type of furniture is
judged normal because there is a construal of seat (probably the everyday default
construal) that excludes car seats, and is hyponymous to furniture. Hence, from
our point of view, these examples do not illustrate a breakdown in transitivity.
A similar account goes for (9). We are not aware of any convincing examples of a
genuine breakdown in the transitivity of hyponymy.
Sense relations have traditionally been viewed as relations between items po-
tentially occurring in a fairly strictly defined ‘paradigmatic slot.’ This viewpoint
has cognitive and communicative validity. The choice of any term out of a paradig-
matic set carries implicit information about aspects of its meaning that are shared
with other possible choices, information about meanings that are excluded, and it
opens up a particular range of more specific meanings, any or all of which may
be important to the message being transmitted. However, the traditional assump-
tion that words have inherent meanings has the consequence that paradigmatic
sets and the interrelations among members have been viewed as relatively stable
structures. Here, while the paradigmatic viewpoint is accepted as valid, the items
in the paradigm are not lexical items but contextual construals of lexical items, and
the relationships are relations between a particular construal of the item actually
chosen and potential construals of other items that might have been chosen in that
context.
For instance, consider the following exchange (the reader is invited to imagina-
tively construct a fuller context):
(13) A: What’s that noise?

B: It’s just a dog.
A full understanding of significance of dog in B’s reply needs at least (a) a construal
of a domain of potential noise-producing agents, given the context, (b) a construal
of what is excluded (the use of just implies that there are potential alternatives to
dog that would have had more serious consequences) and (c) a construal of the
range of possibilities opened up by dog in this context. Relations of hyponymy
and incompatibility hold between these various construals.
The paradigmatic viewpoint underlies the entailment approach to sense rela-
tions, under which relations are defined in terms of truth-conditional relations
between F(X) and F(Y). However, we need to carefully reconsider entailment

definitions of relations in the light of the present approach. On the traditional
approach, It’s a dog unilaterally entails It’s an animal, so dog is a hyponym of
animal; on the other hand, It’s a dog does not entail It’s a pet, therefore dog is
not a hyponym of pet. On the present approach, these results hold only within a
specific context. It may seem that the only context here is It’s a(n) – . But this
may not be correct: simply asking people to make a judgement of whether or not
there is entailment creates a particular mind-set that influences the way the lexical
items are construed. (It is worth noting that in the case of It’s a dog and It’s a pet,
students often have to be persuaded to adopt the ‘correct’ mind-set, otherwise they
see entailment where they are not supposed to.)
6.1.3 Relations between lexical items

We have consistently maintained that sense relations hold between con-
textual construals of lexical items and not between the lexical items as such.
However, it has been common practice for decades to speak of relations between
words. Are we simply to say that the traditional conception of sense relations was
misguided? This does not seem entirely satisfactory. The question therefore arises
as to whether there is a way of rescuing the traditional notion of hyponymy as a
relation between words.
Two possible ways of establishing hyponymy as a lexical relation suggest them-
selves. One is to say that there is a special relation between two words A and B
if, in any particular context, the construal accorded to A is always hyponymous
to the construal accorded to B. This would seem to point to an important relation
between words. Intuitively, one might assume that this is the case with dog and
animal. If so, it would mean that some relations between construals of certain
lexical items are not context dependent, even though the construals themselves are
context dependent. The question is whether such context-independent relations
exist. The indications are that, strictly speaking, they do not. Take Lyons’ example
of horse and mare: It’s a mare unilaterally entails It’s a horse, but in Our horse
has just given birth to a foal and Our mare has just given birth to a foal, mare
and horse are synonyms. In principle, it seems that all traditional hyponymous
pairs are capable of yielding this type of result in appropriate contexts. This line
of inquiry does not look promising.
Another possibility is to specify relations between words in terms of privileged
construals of some kind. Perhaps we can recruit the notion of a default construal
here, that is, a construal that appears in a zero context. We might say, for instance,
that the word koala stands in a particular relation to the word marsupial, on the
grounds that the default construal of koala is hyponymous to the default construal
of marsupial. This has some plausibility.
What result would this approach lead to in the case of dog and pet? At this
point we encounter a certain indeterminacy in the notion of a default construal.
It is arguable that the unmarked notion of ‘dog’ for the average western city-
dweller is the domestic variety, in which case dog will have to be considered a
bona fide hyponym of pet. But is this a truly context-free construal? And if not, is
there such a thing? There is another possible approach. The construal of dog that
includes wild dogs and is thus hyponymous to animal, but not to pet, has some
sort of scientific basis. The arguments for regarding this construal as the default
version do not seem particularly strong as it is probably of relatively infrequent
occurrence. But perhaps there is a need for what one might term a ‘core’ construal,
which emerges naturally in some way from purport. If we apply this notion to pet,
it seems clear that the core construal will exclude electronic pets. The fact that
the question Is an electronic pet a real pet? is interpreted restrictively is evidence
for this: real plausibly functions to direct our attention to the core. This, however,
leaves unanswered the question of the nature of the core.
6.1.4 Taxonymy
Lyons states that taxonomic lexical hierarchies are structured by the rela-
tions of hyponymy (class inclusion) and incompatibility (class exclusion). This is
true as far as it goes, but it is necessary to make a distinction between two relations
of inclusion. The first is the relation that is exemplified in An X is a Y (which
corresponds to ‘simple’ hyponymy); the second is the relation for which An X is
a kind/type of Y is diagnostic, which is more discriminating than hyponymy, and
which functions as the ‘vertical’ relation in a taxonomy. In Cruse 1986, the sec-
ond relation is called taxonymy. It is perhaps noteworthy that Wierzbicka (1996)
includes ‘a kind of’ as one of her semantic primitives, and does not subject it to
further analysis; here, an attempt is made to elucidate its nature. (The following
discussion draws heavily on Cruse 2002b.)
Taxonomizing is clearly not simply a matter of dividing a larger class into smaller
classes. Some logically impeccable subdivisions do not yield good taxonomies:
(14) ?A stallion/mare/foal is a kind/type of horse.

(A stallion is a horse.)
(15) ?A blonde/queen/actress is a kind of woman.
(An actress is a woman.)
(16) ?A red/green/blue hat is a kind of hat.
(A red hat is a hat.)
(17) A mustang is a kind of horse.
An ash-blonde is a kind of blonde.
(18) A Stetson is a kind of hat.
It is obvious that the expression is a kind/type of exerts some kind of selectional

pressure on pairs of items. It is, of course, possible that being a kind of something
else is an arbitrary fact about a word that has to be learned on an individual basis.
For instance, mustang could have as part of its semantic specification a feature
[KIND OF HORSE], whereas stallion would not possess this feature. A more
interesting possibility is that there is a principle of taxonomic subdivision that
allows the relationship of taxonymy to be predicted from the meanings of the
related items.
One approach is to think of the nature of the resultant subcategories and their
relations to one another, in the light of the purpose of taxonomization. Taxon-
omy exists to articulate a domain in the most efficient way. This requires ‘good’
categories that (a) are internally cohesive, (b) are externally distinctive, (c) are max-
imally informative (cf. the discussion of basic level categories in §4.3.3). In many
of the instances where a good hyponym is not a good taxonym of a hyperonym,
there is a straightforward definition of the hyponym in terms of the hyperonym
plus a single feature, as in:
(19) stallion = ‘male horse’

kitten = ‘young cat’
A significant number of good taxonyms, on the other hand, are not easily defin-
able in terms of their hyperonyms, and require encyclopedic characterization (for
instance, the difference between animal and horse cannot be captured in a single
feature). In Cruse 1986 it was suggested that single-feature definitions, in general,
are not good taxonyms because they do not create optimal categories. However,
there are problems with this analysis. It is not difficult to find cases where a sat-
isfactory taxonomy seems to be founded on a single-property division. Take the
case of spoons: these are taxonomized on the basis of what they are used in con-
nection with (teaspoon, coffee spoon, soup spoon etc.). It is significant that neither
large spoon, metal spoon, round spoon nor deep spoon is a satisfactory taxonym
of spoon:
(20) a. A teaspoon/soup spoon is a type of spoon.

b. ?A large/metal/round/deep spoon is a kind of spoon.
Perhaps the explanation lies rather in the nature of the features? Some features
are conceptually ‘simple’ (LARGE, ROUND etc.), others are more ‘complex’
(FUNCTION): perhaps complex features are better able to support taxonomy?
What, then, about the subdivision of blondes, which relies on shade of hair color
(An ash-blonde/strawberry blonde is a kind of blonde)? The fact is, there are
reasons to believe that the problem of taxonomy cannot be solved merely by
looking at the nature of the subcategories, for the following reasons.
Firstly, take the domain of BLONDES. Ash-blonde and strawberry blonde are
satisfactory taxonyms of blonde, but they are by no means the optimal subcate-
gories of BLONDE. We can get much ‘better’ subcategories than ASH BLONDE
and STRAWBERRY BLONDE by dividing the domain into BLONDE HOUSE-
WIVES, BLONDE DOCTORS, BLONDE TEACHERS, BLONDE LAWYERS
and so on, but these are not good taxonyms of blonde.
Secondly, suppose that a particular species of bird has a number of varieties
and a very marked difference between males and females. In such a case, it is
not inconceivable that a male/female division would yield the best categories, in
that the males of different varieties resembled one another more than the male and
female of a particular variety. However, even if that were the case, a sex-based
division would still be taxonomically ‘wrong.’
Thirdly, a given category may be a satisfactory subdivision of one superordinate,
but not of another. For instance, ?A lumberjack is a kind of man is not good, but
A lumberjack is a kind of manual worker is fine. If the crucial factor was the nature
of the resultant category, this would be hard to explain.
It is possible that the good category principle has a role to play in the charac-
terization of taxonymy, but it clearly cannot provide a full explanation.
It seems, then, that a satisfactory taxonym must engage in a particular way with a
particular aspect of the meaning of the hyperonym. In Cruse 1994, it was suggested
that taxonym and hyperonym must share the same perspective. The reason stallion
is not a good taxonym of horse, it was argued, is that it has a ‘sexual’ perspective,
while horse does not; the reason blonde is not a good taxonym of woman is that
it adopts a ‘hair-color’ perspective, while woman does not. Ash-blonde, on the
other hand, has the same ‘hair-color’ perspective as blonde, and that is why it is a
satisfactory taxonym. This notion of perspective needs to be made more precise;
it is not identical to the perspective as defined in §3.4, so to avoid confusion it will
be referred to henceforth as focal orientation.
Two possible lines of approach may be singled out. The first recruits Langacker’s
concept of profiling (see §2.2). The proposal is that what is profiled in a taxonym
must further specify what is profiled in the hyperonym. From this it will follow
that the reason stallion and foal are not good taxonyms of horse is that what they
profile, namely, ‘male’ and ‘young’, are not specifications of what is profiled in
horse; a similar explanation holds for blonde and woman. Likewise, lumberjack and
navvy are not good taxonyms of man because their profiles are not specifications
of what is profiled in man, but they are good taxonyms of manual worker, because
they specify further and profile what is profiled in manual worker, namely, type
of work. (It seems likely that there is also a constraint involving the relationship
between the domains evoked by taxonym and hyperonym, but it is not at present
clear what this is.)
A second suggestion for characterizing the above notion of focal orientation is

more tentative, but may be closer to the truth. It is that a good taxonym must have
as its core a specification of the core of the hyperonym. (For present purposes,
‘core’ is to be thought of intensionally.) Given that the core feature of a blonde
is the possession of fair hair, it follows that the taxonyms of blonde must specify
fair hair more restrictively; if it is the core feature of a manual worker to perform
manual work for money, then a taxonym must specify those activities further; if
the core feature of a spoon is to fulfill a particular function, then the taxonyms
of spoon must have as their essence a more specialized function. What, in that
case, is the core of horse? This is more difficult, but one possibility is its species:
intuitively, mustang would seem to have a more highly delimited version of the
same core.1
A question that is pertinent at this point is whether profiled features and core
features are the same thing. There are some indications that profile and core are
not necessarily the same, and that, when they are different, it is the core that
gives the correct prediction with regard to taxonymy. Take the case of woman. It
seems reasonable to assume that woman profiles [FEMALE], and that [HUMAN]
functions as base. On the assumption that profiling governs taxonomy, we would
predict that the taxonyms of woman would specify sexuality more narrowly. Hence,
we would expect lesbian to be a satisfactory taxonym. But this does not seem to
be the case. When asked to suggest types or kinds of women, people tend to offer
things like career woman, nest-builder, femme fatale and so on. Now these are not
subdivisions of sexuality (although some might have sexual consequences), but of
personality or character, and it is not implausible (it is even reassuring!) that these
come out as the core of womanhood. (It is harder to think of parallel sex-specific
categories for men, but machotype, new man and lady’s man are possibilities.)
The notion of core that we have appealed to as an explanation of taxonymy can at
present only be offered tentatively, pending a firmer characterization. However, one
further point needs to be considered. According to the approach to word meaning
adopted here, both the core of a category and the relation of taxonymy should
in principle be construals, subject to contextual constraints. Again, a definitive
statement cannot at this stage be made, but it does seem that both core and the
relation of taxonymy are much less subject to contextual variation than boundary
placement.
6.2 Lexical aspects of the part-whole relation
The lexical relation of meronymy (sometimes referred to as partonymy)

presents a number of problems that do not affect hyponymy. It is usually informally
1 In Cruse 2002, what is here called ‘core’ was referred to as ‘essence.’
described as the ‘part-whole’ relation. However, this is not strictly correct, and a
failure to separate the part-whole relation from meronymy leads to a great deal
of confusion. Meronymy is a relation between meanings, whereas the part-whole
relation links two individual entities. Of course there is an important connection
between the two, which will be examined in due course, but the distinction between
them must not be lost sight of. Meronymy is also bedeviled by the range of possible
construals of the notion of ‘part,’ some of which are relevant to meronymy and
some of which are not. We shall begin by examining the part-whole relation.
6.2.1 The part-whole relation

In this section we shall attempt to distinguish the lexically relevant
part-whole relation from neighboring relations, and examine some of its salient
characteristics.
6.2.1.1 The portion-whole relation

The part-whole relation may be considered to be a special subvariety of
a more general, more basic relation which here will be called the portion-whole
relation. This relation is illustrated in the following examples (the term ‘portion’ is
used here in a technical sense: it is not claimed that the word portion could appear
happily in all these sentences):
(21) A portion of the cake was given to each of the guests.

(22) Part of the garden was waterlogged.
(23) Parts of the letter were illegible.
(24) My portion of the omelette had bits of eggshell in it.
(25) One section of the house was out of bounds.
(26) I love this stretch of the river.
(27) It was a game of two halves.
The basic notion here is the containment of one region or regions (interpreted in
the broadest possible manner) within another region. The boundaries of a contained
region must be neither identical with, nor must they transgress, the boundary of
the containing region. The notion of containment is, of course, a very basic form of
construal, which we can assume corresponds to Lakoff’s container image schema.
For reasons of space, we shall concentrate on central instances and assume that
both contained and containing regions are bounded and continuous.
6.2.1.2 Parts and pieces

Consider the following two sentences:
(28) All the parts of the aeroplane were carefully packed into crates, ready for shipping.
(29) After the explosion, pieces of the aeroplane were scattered over a wide area.
The most accessible construals of parts and pieces in these two sentences are
distinct. Each unit referred to in (28) falls into one of a limited number of nonar-
bitrary categories which group together similar items from different wholes, and
which are defined by a characteristic set of properties such as shape, size and rela-
tionship to the whole aeroplane. We shall designate such items by the term part:
‘part’ is thus a hyponym of ‘portion.’
In contrast, the pieces referred to in (29) belong to a single category with only
one defining property, namely, that they once formed part of a whole aeroplane.
The label aeroplane pieces can be construed as uniting portions from different
aeroplanes into one category, but these pieces do not fall into distinct groups
with stable properties that unite similar pieces from different aeroplanes, and
they do not have definable relations, other than origin, with the whole entity.
(Pieces from a particular type of entity rarely have distinct lexical labels: the
example of shard comes to mind, but such cases are few and far between.) We
shall use piece as a technical term alongside part to designate portions that do
not qualify as parts: ‘piece,’ too, is thus a hyponym of ‘portion.’ Although most
pieces are not parts, and vice versa, they may be, accidentally, so piece and
part are not strictly incompatibles. (The technical term ‘piece’ covers a wider
category than the everyday lexical item piece: the waterlogged portions of the
garden and the illegible portions of the letter mentioned above are technically
pieces.)
The notion of recurrence (identifiability across different wholes) may need to
be further restricted. Suppose plates of a certain type always fracture in the same
way, yielding recurrent portions. Would they then qualify as parts? Arguably not.
Another characteristic of pieces that distinguishes them from parts is that they
are not contemporaneous with their canonically constituted wholes: there are no
pieces until the whole is destroyed, whereas true parts can be construed even in a
well-formed whole. The pieces of the plates that always fracture in the same way,
although they can be predicted, have no existence in the unbroken plate. (Even if
they were to be admitted into the PART category they would not be prototypical
parts, because they have no function relative to the undamaged whole.)
A further feature that distinguishes parts from pieces is what in Cruse 1986 was
called ‘autonomy.’ This is related to the notion of ‘spare part.’ Parts of artifacts
such as machines are often replaceable by functionally equivalent items: in such
cases, the replacement counts as, for example, a machine part, even though it has
never entered the constitution of a complete machine. (The same is perhaps soon
to be true of human body parts.) On the other hand, an exact replica of a piece of
a broken plate cannot properly be described as a piece of the plate. This property
of parts is presumably related to the fact that they prototypically have a species
of self-sufficiency, that is, cohesion plus distinctness, as ‘objects’ in their own
right. They are, in other words, subordinate wholes; pieces prototypically lack this
self-sufficiency.
6.2.1.3 Factors affecting the GOE of parts

The category PART, like any other natural category, has good and less
good exemplars. We may attempt to summarize the main factors that seem likely
to contribute to a high GOE for X as a part of Y as follows. (There is a problem
with testing for this, since the word part has a wide range of construals, only one
of which corresponds to PART. A form of expression that may select for PART is
The parts of a Y are A, B, C. . .):
(i) The boundary of X does not transgress the boundary of Y.

(ii) X shares all its substance with Y.
Marbles do not become part of a matchbox by being contained in it. Features

(i) and (ii) can be regarded as necessary.
(iii) The boundaries of X can in principle be demonstrated in a well-formed

whole Y.
(iv) The sharper (more salient) the discontinuity between X and not-X the better the
part.
(v) The greater the internal cohesion of X the better the part.
The discontinuity and cohesion may involve any or all of: shape, texture, color,
internal structure, makeup. Detachability and/or independent movability are strong
indicators of discontinuity.
(vi) X has a definable function relative to Y.
What is meant here is function, such as wings for flying or legs for perching
in a bird. There is another notion of function that is relative to human users. For
instance, the parts of an animal carved out by a butcher may not have distinctive
functions for the animal.
(vii) X is autonomous: exact replicas of X also count as parts.

(viii) There is type-consistency between X and Y.
The relevant notion of ‘type’ here is difficult to pin down. One aspect is what
is usually called ‘ontological type.’ There is no agreement on a basic ontology,
but the sort of thing referred to by Jackendoff (1983), namely, THING, STATE,
PROCESS, EVENT, PROPERTY, TIME, PLACE and so on seem relevant to parts.
That is, the parts of a period of time, for instance, should themselves be periods of
time; the parts of a thing should be things (rather than, for instance, substances);
the parts of an event or a process should be subevents or processes; the parts of
an abstract entity should be abstract entities; the parts of a place should be places,
and so on.
There is another notion of ‘type’ that is relevant. There are two broad types
of part, called segmental and systemic parts in Cruse 1986. Segmental parts
are clearly spatially delimited and are typically encountered sequentially as the
whole is traversed; they may well have heterogeneous internal consistency. Typical
examples are the externally visible parts of the body such as arms, legs, head,
trunk and so on. Systemic parts are typically spatially interpenetrating, but they
are functionally distinct and typically have a greater internal consistency. They are
less likely to be perceptually salient. Typical examples in the human body are the
nervous system, the vascular system, lymphatic system and the skeleton.
The two types can be observed in other entities besides the human body.
A house, for instance, can be divided into segmental parts (rooms) and systemic
parts (brickwork, plumbing system, wiring, etc.); a string quartet (to take a non-
spatial example) has movements as segmental parts, and first violin, second violin,
viola and cello as systemic parallels (notice that we speak of the second violin part
quite naturally); a similar case could could be made for the scenes and acts of a
play (segmental), and the various actors’ roles (systemic).
6.2.1.4 Part-whole chains

The part-whole relation generates chains of elements: A is a part of B,
B is a part of C, C is a part of D and so on. For instance:
(30) A finger-tip is a part of a finger.

(31) A finger is a part of a hand.
(32) A hand is a part of an arm.
(33) An arm is a part of a body.
We may say that a constituent element A of a part-whole chain W is an immediate

part of another element B, and B an immediate whole of A, if on a particular
construal of W there is no element X that is a part of B and of which A is a part.
Thus, a finger is an immediate part of a hand, and a hand is an immediate whole
of a set of fingers. An immediate whole is the normal scope of predication for its
part (§2.4).
6.2.1.5 Ultimate parts and ultimate wholes

A part-whole chain prototypically has a beginning and an end; that is,
there is a smallest part, which itself has no construed parts, and there is a largest
whole, which is not construed as part of a yet larger whole. A crucial problem is
therefore what determines the location of the ends of the chain. It is always possible
to think of some portion smaller than the ultimate parts of something: for instance,
the fingertips are composed of skin, nerve fibers, capillary blood vessels and so on,
and these are composed of various chemical substances that can be further broken
down into molecules, atoms, electrons and so on; at the other end of the chain, a
body can be part of a family or team, which is part of a population, which is part of
the terrestrial biomass and so on. Yet, intuitively, there is a self-contained system
beginning with fingertips and ending with the body.
Looking first at ultimate parts, it might well be the case in some instances that
no subpart is motivated – there is simply no subregion bounded by a sufficiently
salient discontinuity to justify the construal of a boundary. In other cases we
can appeal to the notion of type-consistency. One reason why we do not divide
fingertips into nerve fibers, capillaries and so on is that they are of the wrong type.
Ultimate wholes pose more puzzling problems. A crucial factor is undoubtedly
the existence of a major discontinuity with surroundings, coupled with internal
cohesion. However, there are a number of apparent inconsistencies. The following
examples arose in the course of class discussions as part of an undergraduate course
in lexical semantics. When student informants were questioned about pan and lid,
the majority said that the lid was not part of the pan, but some said they might
reconsider their judgement if one or more of the following applied:
(34) a. the lid was sold as a unit together with the pan
b. the lid was essential to the functioning of the pan
c. the lid was attached to the pan by means of a hinge.
When asked about teapot and lid, however, the majority said that the lid was
part of the teapot. If the lid was attached by a hinge, the judgment was unanimous.
When asked why the teapot lid was different from the pan lid, the most frequent
comment was that it was essential to the functioning of the teapot, and was sold
together with it. The screw caps of soft drink bottles and the cork and capsule
of a wine bottle were unanimously rejected as parts of the bottle, in spite of the
fact that these are arguably essential to function, and bottles are normally sold
complete with caps. A possible confounding factor is that they are normally made
of different materials from the bottle, which gives rise to a salient discontinuity.
One final type of example will be mentioned. Most informants reject battery
as a part of flashlight, whereas bulb is unanimously accepted as a part (there are
many similar cases). This is reinforced by the fact that one normally has to buy the
batteries separately when one buys a flashlight, but one does not expect to have to
buy a bulb separately. Furthermore, the description of a flashlight in a shop will
say battery included if it is included in the price, but will not say bulb included.
This is strange, because the battery is contained within the body of the flashlight,
and is essential to its functioning.
It appears that a judgement of where the boundaries of an ultimate whole are

situated is a result of the interaction of a number of factors, but the details of the
interaction are far from clear. It is possible that convention plays a role here.
6.2.1.6 Core parts

Consider the following sentences:
(35) There were scratches on the hand, but not on the arm itself.
(The hand is part of the arm.)
(36) The monitor is faulty, but the computer itself is OK.
(In the case of an iMac, the monitor is part of the computer.)
The expression the X itself selects some sort of core. In the case of the part-whole
relation, what seems to be selected is the smallest portion that can be construed as
‘a whole X’ – any smaller unit Y can only be construed as a part of X (although it
can of course be construed as a whole Y): certain (genuine) parts may be stripped
off without completely destroying wholeness. We shall call the smallest possible
portion of an X that can be construed as a whole X the core part. (The term
‘core’ is deliberately used to suggest a parallel between core parts and the core of
a category.) The factors that determine the existence and boundaries of a core part
in particular cases are not clear. In the case of the iMac, the fact that most PCs have
separate monitors may influence our construal of the iMac. In any case, the core
part may be presumed to need adequate motivation as a whole in its own right.
The notion of core part overlaps with, but is not identical with, the ‘main func-
tional part,’ which can function as active zone in a use of the name of the whole
(see Cruse 1986). For instance, a stainless steel X is often ambiguous between an
X that consists entirely of stainless steel, and one whose main functional part only
consists of stainless steel: consider a stainless steel screwdriver/hammer/knife, any
of which may have a handle made of some other material. This only applies to
certain privileged parts, however: a screwdriver with a tungsten head and a stain-
less steel handle would not qualify as a stainless steel screwdriver. My intuitions
are that the description ‘a totally new type of computer’ would be valid for an
iMac whose central processor had been redesigned, but not for one whose only
new component was the monitor portion: in this case, the main functional part and
the core part are the same. However, it is not clear that the part of a screwdriver
that engages with the screw can be called the screwdriver itself, so in this case the
main functional part and the core part are distinct.
6.2.1.7 Variable construal and the transitivity of the part-whole relation

In Cruse 1979 and 1986 a distinction was made between ‘integral parts’
and ‘attachments.’ These were distinguished as follows:
(i) If X is a part of Y and X is attached to Y are both normal, then X is an

attachment of Y:
(37) The hand is part of the arm.

(38) The hand is attached to the arm.
(ii) If X is a part of Y is normal, but X is attached to Y is not, then X is an integral

part of Y:
(39) The handle is a part of the spoon.

(40) ?The handle is attached to the spoon.
This distinction was then recruited as part of an explanation for one type of
so-called ‘transitivity failures’ in the part-whole relation, as in:
(41) a. An arm has a hand.

b. A hand has fingers.
c. ?An arm has fingers.
(42) a. Fingers are parts of a hand.
b. A hand is a part of an arm.
c. ?Fingers are part of an arm.
From the point of view of the current approach, both the distinction and the
transitivity failures are arguably illusory. Consider, first, the proposed distinction
between integral parts and attachments. Notice, first, that there is nothing odd in
principle about a part being attached to a sister part: we can say, for instance,
without oddness, that the arm is attached to the trunk at the shoulder. Second,
there is an easily accessible construal of arm that excludes the hand:
(43) There were burns on the victim’s hand, but none on the arm.
(44) He had a tattoo on his arm (contrasts with a tattoo on the hand).
(45) A broken arm.
Hence, the most obvious explanation of the normality of (38) is that arm receives
the construal that excludes hand. In (37), on the other hand, arm receives the
inclusive construal. Hence, (37) and (38) contain different construals of arm: the
hand is attached to arm1 , which is a sister part, but not to arm2 , the whole that
includes it. If one makes the effort to construe arm in (38) as arm2 , it becomes
as anomalous as (40). Turning now to (39) and (40), a slight twist is needed in
the explanation. Of course, the handle of the spoon is attached to the ‘bowl’ of
the spoon; the reason (40) is odd, however, is that there is no construal of spoon
that designates only that part. (This account of the oddness of [41c] and [42c]
differs from Langacker’s account as presented in §2.4. The two accounts are not
necessarily mutually exclusive.)
What about the apparent transitivity failures? It should perhaps first be acknowl-
edged that these are weird sentences: how many of us have encountered them in
naturalistic settings? They are also virtually uncontextualized, so one should hes-
itate before drawing general conclusions from them. The previous explanation
was that the relation cannot cross an attachment boundary, that is to say, the parts
of an attachment do not count as parts of the whole of which the attachment is
part. The evidence for this was extremely limited, not to say contrived. There is,
however, an alternative explanation if we assume that the exclusive construal of
arm is the default (i.e. arm1 ), then we can say that (41a) and (42b) exemplify the
inclusive construal of arm (i.e. arm2 ), but that (41c) and (42c) represent the ex-
clusive (default) construal. In support of this account of the ‘transitivity failures,’
we would claim that (41c) and (42c) improve considerably in normality if one
makes the effort to give arm the inclusive construal. The general conclusion is that
these examples give no grounds for impugning the transitivity of the part-whole
relation. However, an explanation is still required of why the inclusive construal
is not evoked in (41a) and (42c).
The account of the other transitivity failure given in Cruse 1986 is nearer the
mark, but can be more felicitously expressed using the current approach. It concerns
the following examples cited originally by Lyons as an unresolved puzzle:
(46) a. The jacket has sleeves.

b. The sleeves have cuffs.
c. The jacket has cuffs.
(47) a. The house has a door.
b. The door has a handle.
c. ?The house has a handle.
There are two issues here: the first concerns the logical validity of the conclusion
expressed in (46c) and (47c) based on the respective a and b sentences as premises;
the second issue is the normality of (46c) compared with (47c). Let us look first
at the logical question. The basic principle is simple: the conclusions are valid if
the construals of all terms are consistent between premises and conclusion. This
is the case in (46), so the conclusion (46c) is valid. In (47), on the other hand,
there is a discrepancy between the construal of handle in (47b) and (47c). The
word handle is typical of those exhibiting microsense behavior: when we hear
the word handle, in most contexts, unless a hyperonymic construal is explicitly
triggered, we look for a subtype of handle that is intended (default specificity),
and the resultant subtypes manifest the properties of microsenses. The contextual
frame The X has a handle elicits an appropriate microsense, that is, ‘an X handle.’
Hence, The door has a handle elicts the construal ‘door handle,’ while The house
has a handle elicits the microsense construal ‘house handle.’ This discrepancy
destroys the logical validity of the conclusion. (In [46], ‘sleeve cuffs’ are identical
with ‘jacket cuffs.’)
The situation with regard to normality is not entirely clear in these examples.
Out of context, there is probably a general tendency to a slight oddness if a part-
whole statement misses out an intervening term, especially if it is salient, hence
we might expect (46c) to be a little hard to contextualize. The reason it is not may
be due to another principle, namely that one can always relate parts at any level
to the overall whole: thus, for instance, The human body has fingers and fingers
are parts of the body are not so odd as The arm has fingers. The normality of
(46c) may therefore be due to the fact that jacket is construed as the overall whole.
The oddness of (47c) is more than we would expect simply from a level-skip:
as suggested in Cruse 1986, it is more plausibly explained as being due to the
fact that houses do not usually have handles, and it is difficult to imagine what
purpose they might serve (Cruse [1986] pointed out that substituting doll’s house
for house considerably reduces the oddness, but does nothing for the validity of
the conclusion.)
6.2.2 Meronymy
Like every other sense relation, meronymy is viewed here as a relation
between contextually construed meanings (or more precisely, by pre-meanings
created by boundary construal). However, the relationship is less straightforward
than hyponymy, and it is not easy to select the optimum way of expressing it.
The problem stems from the fact that the essential relation, the part-whole rela-
tion, does not hold between construed classes of elements, but between specific
individuals belonging to those classes. Also, the relation itself is subject to con-
strual, unlike the hyponymic relation between two classes. Given two classes, the
definition of hyponymy can decide whether hyponymy holds or not: there is no
need for a separate construal of the relation. In the case of meronymy, on the other
hand, a part-whole relation between two entities is itself a construal, subject to a
range of conventional and contextual constraints. So, let us examine the following
characterization of meronymy:
(48) If A is a meronym of B in a particular context, then any member a of the extension

of A maps onto a specific member b of the extension of B of which it is construed
as a part.
Hence, finger is a meronym of hand because for every entity properly describ-
able as a finger (in the default construal), there corresponds some entity properly
describable as a hand (also in the default construal), of which it is construed as a
part. This characterization seems to capture something essential about meronymy.
But it is too restrictive as it stands, because it does not allow for ‘spare parts,’ which
have never participated in any whole (of the relevant sort). We might therefore relax
the characterization to admit potential part-whole relations:

of A either maps onto a specific member b of the extension of B of which it is
construed as a part, or it potentially stands in the construed relation of part to
some actual or potential member of B.
(Notice that to take account of preserved body parts where the rest of the body
has been destroyed, we need to give a ‘timeless’ interpretation of ‘the extension
of B.’)
This now covers spare parts, but it arguably still misses something essential
to meronymy, namely, the construal of partness as an essential element of the
construal of the part. Take the case of lake and park. Many parks have lakes
within their boundaries, and in such cases the lake would be construed as ‘part
of the park.’ Furthermore, every lake is potentially a part of some park, so by the
second definition, lake would qualify as a meronym of park. There is, however,
a difference between the relation linking finger and hand and the relation linking
lake and park. In construing lake we are under no pressure to construe it as being
part of something: in cases where a lake is part of a park, the ‘partness’ is imposed
on the construal as it were from the outside. In the case of finger, ‘partness of hand’
is an essential component of the original construal. Let us call this difference one
between an intrinsic and an extrinsic construal of partness. Then we can say that
the relation of meronymy concerns only intrinsic construals of partness:

of A either maps onto a specific member b of the extension of B of which it is
construed as a part, or it potentially stands in an intrinsically construed relation
of part to some actual or potential member of B.
Notice that this characterization introduces an asymmetry between meronym

and holonym. A holonym is characteristically more independent, in that there is
no obligatory construal orienting it towards particular meronyms. Of course, the
information that says a human body consists of arms, legs and so on is represented in
the conceptual system, and is in principle accessible, but does not form a necessary
part of every specific construal of body. Parts are evoked only under particular types
of construal (e.g. under the ‘constitutive perspective’).
The above characterization has some plausibility when applied to construals of
part names. There are good reasons for retaining meronymy as a semantic relation
between construals. For instance, certain distinctions can only be made in terms
of classes rather than in terms of individual items. This is the case with what in
Cruse (1994) is called range congruence. Cruse (1986) distinguished the relation
between a ‘super-meronym’ and a ‘hypo-holonym’ from that between a ‘hypo-

meronym’ and a ‘super-holonym.’ The first is illustrated by the relation between
nail, on the one hand, and toe and finger on the other. Here we have a part name,
nail, which is related to two different wholes. There is a hyperonymic class of
NAILS, which is indifferent to the identity of the wholes:
(51) Nails had been torn from the victim’s fingers and toes.
However, the default use of nail denotes a microsense, one for each related
whole. Hence, B’s answer in (52) is acceptable as true:
(52) A: (examining B’s hands) Have you cut your nails this week?
B: (who cut his toe-nails the previous day, but had not cut his finger-nails for a
long time) No.
The second relation is illustrated by body in relation to penis and vagina. Here
it is the holonym that has the greater ‘range’: penis and vagina are parts, but in
relation to different subclasses of body. However, in contrast to the first case, there
is no evidence whatsoever that there are two microsenses of body corresponding to
‘man’s body’ and ‘woman’s body.’ This is interesting, and demands an explanation,
but the important thing for present purposes is that it is not a difference at the level
of the part-whole relation.
There are, however, difficulties that cast doubt on the utility of meronymy as a
lexical relation. Let us take the example of the word lid. The purport of this word
takes us into the conceptual area of containers, access points into containers and
means of controlling movement into and out of containers. There seems to be a very
strong conventional constraint forcing some sort of construal of incompleteness:
a lid is designed for use in conjunction with a container. However, this is not an
obligatory construal of partness, since not all lids are parts (cf. bottles, jars, teapots
etc.). There is undoubtedly a possibility of construing lid in a superordinate sense:
(53) In this box you’ll find a lot of things to put things in, and in the other box, a lot
of lids: your job is to sort out which lid goes with which container.
But most commonly, a much narrower construal is required:
(54) We’ve lost the lid.
Here, a particular type of container is referred to. Perhaps, then, we should recruit
the notion of microsense and say that lid has microsenses (‘teapot lid,’ ‘jamjar
lid,’ ‘pan lid’ etc.), some of which involve intrinsic construal of partness (cf. Cruse
1986). However, this will not work, because, as was mentioned earlier, whether,
for instance, a teapot lid is construed as a part or not depends on factors such as
the presence of a hinge, whether sold separately and so on. In other words, we can
only be sure of a construal of partness at the level of the individual item. Even the
local construal of ‘lid type’ is not necessarily specified as to partness. (There is no
evidence for ‘hinged lid’ and ‘free lid’ as distinct microsenses or perspectives.) The
same phenomenon can be observed with a wide range of words. The implication
seems to be that we cannot in general deal with the part-whole relation except
at the level of the individual referent. In (53), for instance, the relation between
container and lid is indeterminate. In other words, the whole notion of meronymy
as a lexical relation is dubious.
The picture of meronymy that seems to emerge from the above discussion is
the following. There is, first, a very indeterminate purport, then a series of pre-
meaning construals that take us nearer and nearer to the target construal, and may
involve a commitment to partness at some point before the final construal, but in
many cases the part-whole relation cannot be inferred until we reach the level of
individual referents.
This is a very awkward picture from the point of view of lexical semantics:
there seems little to be said at the lexical level. One conclusion from this might
be that meronymy, unlike, say, hyponymy or antonymy, is not a relation between
meanings. After all, strictly speaking, the parallel at the individual level of the
class-based relation of hyponymy is the part-whole relation, and the strict parallel
to incompatibility is a relation between sister parts: incompatibles denote sets with
no members in common; co-parts are, in general, parts of the same whole which
have no substance in common (provided they are of the same type – segmental parts
may share substance with systemic parts). Does this mean that we can dispense with
meronymy altogether? There are reasons why this is not an entirely satisfactory
proposal.
Perhaps the most cogent arguments in favor of retaining meronymy concern
its intuitive appeal, the fact that children learn the names of parts early, the fact
that all languages have names for parts of things, and, further, that cross-linguistic
generalizations can be made regarding the naming of parts, especially parts of the
human body (see Brown 2002 for a survey).
One such generalization concerns the equivalents in various languages for the
English hand. In many languages this denotes the region of the arm from the
fingertips to the elbow. (This is the case of the Modern Greek to xeri. However,
Greek speakers assure me that it can also be construed to refer to the region
from the fingertips to the wrist only, and that there is rarely any confusion.) The
generalization is that, where a language has an ‘extended’ hand, it is highly likely
also to have word for ‘foot’ that designates the region of the leg from the toes to
the knee (this is the case with Modern Greek to podi). Another generalization, or at
least strong tendency, is that extensions from hand to foot are common, extensions
from foot to hand are much rarer. Thus, the basic meaning of the French doigt is
‘finger,’ but we also have doigt de pied for ‘toe’ (alongside orteil). However, this
is not an absolute rule, as the heel of the hand is well established in English.
Finally, there is a hard core of word pairs, like finger and hand, that are extremely
difficult not to construe literally with a relation that satisfies our definition (but
they are remarkably few in number).
The situation with regard to parts and wholes and their linguistic expression can
therefore be summarized as follows:
(i) The part-whole relation properly applies to individual entities. It is a
construal that is subject to variation.
(ii) The recognition of a relation of meronymy between construals is justified
by the existence of a small number of generalizations and distinctions that
only apply to classes of parts.
(iii) Every language has a range of ways of referring to parts of things. Many
of these ways involve specialized lexical items, but apart from a very
restricted core set of strict lexical meronyms, the relations between these
and expressions for whole things are very various.
7
A dynamic construal approach to
sense relations II: antonymy
and complementarity
7.1 Oppositeness
7.1.1 Aspects of the construal of oppositeness

The notion of oppositeness is well established in everyday language:
(1) We were traveling in opposite directions.

(2) I found myself sitting opposite the new Minister for Moral Regeneration.
(3) John is tall and thin; Pete is just the opposite.
(4) Mary is extrovert and makes friends easily; Jane is just the opposite.
(5) He doesn’t seem to be interested in the opposite sex.
There are undoubtedly different construals of opposite involved here, but intu-
itively they belong to a family. None of them are metalinguistic. But even speakers
innocent of semantic theory have robust intuitions about lexical opposites, and
even quite young children rapidly catch on to the idea. Like all sense relations,
oppositeness is a matter of construal, and is subject to cognitive, conventional and
contextual constraints.
There seem to be two main components in a construal of oppositeness. The first
of these is binarity. Opposite meanings are construed as mutually exhausting some
domain: within the appropriate domain, there are only two possibilities. Some
domains are difficult to construe in any other way. For instance, there are only two
directions along a linear path, so up and down, and forwards and backwards are
natural opposites; likewise, there are only two extreme points on an axis, so top
and bottom, and front and back are also natural opposites. At a more abstract level,
there are only two directions of change between two states, that is, from A to B
and from B to A. This confers oppositeness onto dress:undress, tie:untie, and the
like.
These examples have a kind of built-in logical twoness. But a binary opposi-
tion does not have to be inherently logically watertight, only locally construed as
such. Take the case of town and country (urban:rural), which are often used as
164
Antonymy and complementarity 165
a binary opposition. Here a domain something like ‘areas of normal residence in

a temperate climate’ needs to be construed to justify the binarity. The binarity of
even an obvious pair of opposites like male:female rests on a restricted domain
that excludes hermaphrodites, beings with no sex organs (for whatever reason) and
so on.
While binarity is undoubtedly an essential feature of oppositeness, it is not, on its
own, sufficient. There are many situations where a domain is construed with only
two members, but oppositeness seems to be absent. Cruse (1986) mentions double-
decker:single-decker (as applied to buses); and monocotyledon:dicotyledon (a
botanical division of the domain of angiosperms – itself one of only two types
of seed plant). Another non-opposite two-way division is food:drink, in the do-
main of nutrition. Even tea:coffee is a binary choice (within the domain of hot
drinks) on an airplane.
What is missing from these examples? Cruse (1986) suggested that the contrast
expressed by a pair of true opposites has to be not merely binary, but ‘inherently
binary’; that is to say, the binarity has to be logically necessary, and not just a
contingent fact about the world. But then this, too, was found to be insufficient, as
there are logically binary notions that are not good opposites, the main example
quoted being Friday and Sunday, which are the only two days that are one day
removed from Saturday. (This is of course only one of many binary oppositions
that could be invented using the days of the week.) Why are they not opposites?
Cruse’s suggestion was that their meanings (i.e. Friday and Sunday) do not encode
a salient mutual orientation towards Saturday, in contrast to, say, yesterday and
tomorrow whose meanings are oriented towards today, and which exhibit a certain
degree of oppositeness. However, if the approach adopted here is correct, it ought
to be possible to devise a situation where such a mutual orientation is induced
by contextual constraints, leading to at least some sense of oppositeness. The
following is an attempt to do this:
Some important committee regularly meets on Saturdays. An influential caucus
holds a pre-meeting on Friday to prepare for the main committee meeting, and
a post-meeting on Sunday to discuss the events in the Saturday meeting. For the
members of this group, the Friday meeting could develop a (weak?) relation of
oppositeness to the Sunday meeting.
7.1.2 Main varieties of opposite

Several types of opposite can be recognized. (Some of the types described
in Cruse 1986 have been shown by informant testing to give rise to only a very weak
intuition of oppositeness.) The most salient types are complementaries, antonyms
and reversives, together with their morphological derivatives:
complementaries: dead:alive, open:shut, true:false (default construals). These

exhaustively bisect some domain into two subdomains.
antonyms: long:short etc. These are gradable adjectives or stative verbs and de-
note degrees of some property that diverge significantly from some reference
value.
reversives: rise:fall, dress:undress etc. These are verbs that denote changes in
opposite directions between two states.
Cruse (1986) also has a class of converses like buy:sell, parent:child and so
on. Many converses are good opposites, but the position taken here is that their
oppositeness is not a necessary consequence of their being converses, but arises
from other factors (for instance, the oppositeness of the direction of transfer of
goods and money in buy and sell). They will not be discussed as a class here.
7.1.3 Goodness-of-exemplar in opposites

Certain pairs of lexical items are readily judged to be better opposites
than others, presumably on the basis of default construals. The following would
seem to be some of the relevant factors:
(a) Intrinsic binarity

This has been discussed above.
(b) The ‘purity’ of the opposition

For instance, male:female are better opposites than, say, man:woman,
which in turn are better than aunt:uncle, whereas convent and monastery are hardly
felt to be opposites at all. The basic opposition underlying all these is MALE vs.
FEMALE, which appears in an ‘undiluted,’ or ‘pure’ form in male:female; in all
the other pairs mentioned, the basic opposition appears alongside other semantic
material that is, as it were, inert with respect to the opposition, and the more of
this there is, the less salient the opposition.
(c) Symmetry
For instance, large and tiny are not such good opposites as large and
small (although the sense of oppositeness is not completely absent). One possi-
ble reason for this is that large and tiny are not symmetrically disposed about
the reference point. The residual oppositeness of large and tiny may be due to
their counterdirectionality: when intensified, they move apart on the scale of
size.
(d) Matched non-propositional features

The default opposite of clean is not mucky but dirty, and this is because
mucky is restricted to certain registers, while clean has a more general applica-
bility – in this instance, we can say that they evoke different frames. Ideally,
non-propositional features such as register and expressivity should be the same for
both members of an opposite pair.
7.2 Complementarity
This chapter concentrates on adjectival opposites, of which the main types

are complementaries and antonyms. The discussion begins with complementaries.
7.2.1 Gradable vs. non-gradable construal of properties

There are two basic ways of looking at properties, both of which involve
basic image schemas. We can either think of properties as present or absent, or
we can presuppose presence and think in terms of more or less of them. In the
first case, a pair of complementaries will be the most natural linguistic expression;
for the second view, we will construe a continuous scale representing values of
the property, and linguistic expression will most naturally be by means of a pair
of antonyms. Which construal is the most natural, varies from domain to domain.
Take the property of ‘being married’ – you either have it or you don’t, and it is
odd (but not impossible) to think of one person as ‘more married’ than another.
Therefore the easiest construal of married:single is as a pair of complementaries.
Or take the domain of spatial extent. It is not very illuminating to think of the
presence or absence of something such as length – if something has no length,
it most likely does not exist – so ‘more or less’ is the obvious construal, and
opposites in this domain, such as long:short are naturally antonyms. However,
there are some properties that can plausibly be viewed in either way. Take the
domain of ‘clean/dirty.’ This can be construed either with a binary division or a
scale. The pair clean:dirty thus show a split personality, behaving at times like
complementaries (It’s not clean implies It’s dirty), and at times like antonyms
(This shirt is cleaner/dirtier than that one). Antonyms will be dealt with in detail
later: for the moment, let us concentrate on complementaries.
7.2.2 Profiling against domains

Complementaries are construed as both mutually exclusive and mutually
exhausting some domain. Hence, if X and Y are adjective complementaries, then if
some entity is X, then it is not Y, and if it is not X then it is Y. The notion of domain
is essential to the understanding of complementarity. Take the pair married:single.
We can say that if someone is not married, then they are single, and if they are
not single, then they are married. But this relationship is highly dependent on
the construed domain within which the two terms are operating. First of all, we
need to specify that we are talking about humans: the inference does not work for
angels or spiders (or chairs). Second, we must restrict the domain to ‘marriageable’
persons: the logical relation does not hold for babies (in Western societies at least)
or the Pope. Third, we must exclude noncanonical forms of cohabitation. Hence,
the logical properties only appear within an appropriately construed domain. In
some cases, the appropriate domain is subject to strong conventional constraints,
as in the case of the default construals of dead:alive: if someone says that John
is not dead, it will normally be inferred that John is still alive, that is to say, we
are strongly constrained to construe the domain in such a way that the relation of
complementarity holds.
The position of the boundary between a pair of complementaries X and Y is an
aspect of particular construals. In the case of dead:alive, the location of the bound-
ary is sometimes a matter of dispute, including legal dispute, and could well differ
in different discourse domains. In everyday language, John is dead can usually be
taken to mean that John’s state is well clear of the zone of uncertainty. It could
also, however, function argumentatively to indicate that John was on the wrong
side of the speaker’s construal of the boundary, that is, as part of a dispute (imagine
a discussion as to whether attempts to revive a patient should be abandoned). No-
tice that if one construes the notion ‘life-state’ to include zombification and/or the
vampiric state, and so on, as well as ‘ordinary’ death and life, the logical relation
of complementarity will not hold.
It is important to bear in mind that complementarity is a relation between con-
struals and not between lexical items: in many cases, properties can be construed
either in absolute terms or in gradable terms. In some cases, there is a species of
alternation, according to context, between a pair of complementaries and a pair of
antonyms; we shall postpone consideration of these until antonyms have been ex-
amined in detail. But there also appear to be cases where one term of an opposition
is construed as absolute and the other as gradable:
(6) A: Is John dead?
B: No, he’s very much alive.
Here, dead is construed absolutely and alive gradably. The two construals of
alive as gradable or absolute do not seem to amount to a difference between distinct
senses. There is, for instance, no zeugma in (7):
(7) A: Is John alive?
B: Very much so.
It appears to be possible for the absolute term to function as a zero point on the
construed scale for the other term. That is to say, in (6), dead is construed as ‘zero
aliveness.’ This does not appear to affect the logical relation between the terms;
that is to say, they are still complementaries by the ‘not-X entails Y’ criterion.
7.3 Antonymy
The following treatment of antonymy is based on Cruse and Togia (1995).

Antonyms have the following properties:
(i) They are adjectives or stative verbs
(ii) They denote properties construed as varying in degree
(iii) They are counterdirectional in that one term when intensified denotes a
higher value of the relevant property, while the other term when intensified
denotes a lower value.
Antonymy is treated here as a relation between construals, and involves the struc-
turing of content domains by means of one of a limited repertory of image schemas.
The principal image-schema in this account of antonymy is SCALE, which con-
strues a property in terms of more and less. It will be assumed here (a) that image-
schemas vary in schematicity, in that very general ones can be manifested as more
specific ones, and (b) that they can join together into complex image-schematic
structures. The general notion of opposite probably corresponds to a single
image-schema. The different types – complementaries, antonyms, reversives –
will correspond to more specific image-schemas. Looking at antonyms in more
detail, as we shall see, we need to think in terms of quite complex assemblages of
yet more elementary schemas.
The different types of antonymous relation will be presented using default read-
ings of words. However, it must be borne in mind that antonymy is a relation
between construals, and particular lexical items can typically be construed dif-
ferently in different contexts. It should also be borne in mind that much more
goes into a local construal of an adjective such as long than the selection of the
appropriate antonym type: there is also the orientation of the scale relative to the
content domain, the identification of an implicit reference point on the scale, and
the degree of divergence from the reference point.
7.3.1 A survey of antonym types

7.3.1.1 Monoscalar and biscalar systems
In this section, the basic antonym types are presented. A detailed justi-
fication of the types will not be given at this point; some of the justification will
become apparent below. The first division is between antonyms that involve a
single scale and those that involve two separate scales working in tandem.
A simplified version of a monoscalar system is given in Figure 7.1.
LENGTH
0 short long
Figure 7.1 A simplified monoscalar system
The scale denotes a single property, in this case length; the scale has an end
point denoting zero value of the property at one end and extends indefinitely in
the other direction. One term of the opposition is associated with a higher value
of the property and the other term with a lower value. The terms move in opposite
directions along the scale when intensified.
There are three basic types of biscalar system, depending on the relative dis-
position of the two scales. In the equipollent patterns, the properties of the two
scales are fully symmetrical. There are two possibilities here: either the scales are
arranged end-to-end, and are completely disjunct, or they are completely overlap-
ping. The disjunct type is exemplified by ‘hot:cold’ in Figure 7.2. Here we have
two independent scales which meet at their zero points and extend indefinitely in
opposite directions.
COLDNESS HOTNESS
cold 0 hot
Figure 7.2 A disjunct equipollent system
The parallel type appears to be rarer. Here the two scales run parallel to one
another over their whole length (Figure 7.3).
HARDNESS hard
0
SOFTNESS
soft 0
Figure 7.3 A parallel equipollent system
In the overlapping system, there is partial overlap between the two scales. At
the same time, the scales are not equal: there is a major scale (MERIT) and a minor
scale (BADNESS), as in Figure 7.4 on page 171.
MERIT good
0
BADNESS
bad 0
Figure 7.4 An overlapping system
7.3.1.2 Mapping of systems onto content domains

It will be noticed that different antonym pairs have been assigned to the
different scalar patterns. Let us assume for the moment that these assignments are
correct. The question then arises: is the association between a particular pair and
a particular pattern arbitrary? Do we have to learn it? Or is it motivated by general
(cognitive or other) principles? It is claimed here that these and other properties of
antonyms are strongly motivated. One pointer in this direction is the fact that there
is so much cross-linguistic convergence in antonymic properties. Most differences
in antonym behavior between nearest translation equivalents in different languages
can be attributed to conceptual differences, that is, they are not arbitrary. It seems
likely that what linguistic differences there are tend to occur in areas where the
conceptual motivation is less strong.
Let us now consider briefly how the different modes of construal show different
affinity for different content domains. We shall begin with the monoscalar pattern.
The factors favoring this construal include the following:
(i) the salience of properties
A property such that more of it is more salient is preferentially chosen as the

basis of a scale (see, however, §7.4.2.1).
(ii) the ease of construal of a determinate end point for the scale
If there is a definite end point, this is preferred as zero value.
(iii) the calibratability of the scaled property
A calibratable property (in terms of conventional units) is preferred to a non-

calibratable property as the basis for a scale.
Take the case of LENGTH: why do we have a scale of LENGTH rather than
a scale of ‘SHORTH’? Firstly, more length is generally more salient than more
shortness. Secondly there is an end point at the short end, but none at the long end.
Thirdly, length is much more easily calibratable than shortness. Hence, a scale
of LENGTH is well motivated. Consider, now, the case of SPEED. Firstly, fast
moving things are generally more salient than slow moving things. Secondly, there
is a clear end point at ‘stationary.’ Thirdly, both fastness and slowness are easily
calibratable. Again, a scale of SPEED is well motivated, although marginally less

so than that of LENGTH.
The scales of LENGTH and SPEED may be contrasted with that of DIFFI-
CULTY. Firstly, difficult things loom larger than easy things, generally speaking.
Secondly, there are two possible end points to the scale, both somewhat vague,
namely ‘impossibility’ and ‘no impediment at all.’ Thirdly, there are no readily
available conventional units for calibration. It can be seen that the motivation for
having a basic scale of DIFFICULTY rather than a scale of EASINESS is not over-
whelming. It is not surprising, therefore, that (a) English-speaking undergraduates
find it relatively hard to decide which way round the scale goes, and (b) some
languages (e.g. Modern Greek) actually do it the other way round (that is to say,
they have a scale of EASINESS).
Let us now consider why some antonyms operate over a monoscalar system and
others over a biscalar system. The answer is that the different image-schematic
complexes show different affinities for different content domains. This point can
be illustrated with reference to the disjunct equipollent type: why are there two
scales and why are they arranged end-to-end? Imagine putting one’s hand into (i)
a bowl of cold water, (ii) a bowl of tepid water and (iii) a bowl of hot water. You
will get a strong temperature sensation in (i), no temperature sensation at all in
(ii), and a strong, but different temperature sensation in (iii). In other words, there
is a natural zero-temperature sensation in the middle, and two distinct positive
temperature sensations – a natural (disjunct) equipollent situation. Given the four
patterns to choose from, there is little doubt which fits best. (The two temperature
scales are felt to articulate a single coherent domain because basic experience
unites them into a single scale: pouring boiling water into cold water makes it first,
less and less cold, then tepid, then gradually hotter. There is no such experiential
continuity between, for instance, anger and surprise.) Other natural equipollents
are: like:dislike, proud of:ashamed of, beneficial:harmful.
Turning now to the overlapping type, the argument for the appropriateness of
two scales is once again that we have two distinct notions to quantify: one is merit,
and the other is badness. Things can be construed as ‘not-good’ in two ways:
either they lack merit, or they possess a ‘positive’ badness. Whereas nothing gives
us simultaneously a sensation of hot and cold (hence those scales are disjunct),
something can exist, for instance, a 25% mark in an exam, that we can choose to
see as either lacking merit (i.e. being poor), or possessing positive badness (i.e.
being bad). There is therefore an area of overlap between the two scales. The fact
that ‘normal’ (that is to say, neither good nor poor) corresponds to zero badness,
is perhaps evidence of an entrenched optimistic view of life (cf. the ‘Polyanna
principle’): we are willing to believe of some bad things (say, the 25% mark) that
they nonetheless possess some degree of merit; but normality and upwards on the
scale of merit is free of badness.
7.3.2 Monoscalar systems: polar antonyms

A detailed examination of the various types of antonym will now be given,
beginning with polar antonyms. A full monoscalar system can be represented as in
Figure 7.5 (the simpler diagram presented in Figure 7.1 indicated only the absolute
scales).
0
X-ness
Y X
Figure 7.5 A full monoscalar system
There are two basic ways of construing a quantity of something: we can either
look at it in absolute terms (e.g. 25cm, 1.7 kg) or we can view it as more or
less than some reference value (a long pencil, a heavy suitcase). These two ways
correspond to what we shall call the absolute scale schema and the relative scale
schema respectively. In the diagram, the upper scale is the absolute scale, and the
heavy line indicates the reference value (or range) for the relative scale. (Even in
the case of a single gradable adjective we need to postulate an absolute and a relative
scale, since gradable adjectives are prototypically relative in their basic use.)
7.3.2.1 Subs, supras and the relative scale

The members of a pair of antonyms do not have the same relationship
to the absolute scale: one term, when intensified, denotes a higher value of the
scaled property, while its partner denotes a lower value. Following Cruse and
Togia (1995), these will be called the supra and sub terms, respectively.
If there is only one term, it will be a supra. Having only a sub term seems
intuitively perverse. Why should this be so? Probably because the sub term has a
cognitively more complex relationship to the absolute scale than the supra term has,
and cognitive complexity has to pay for its keep. This means that supras have the
most basic association with the scale: if a nominal denoting the associated scaled
property is derived from one of the opposed adjectives, it is always derived from
the supra term. Thus we have a scale of LENGTH, not a scale of SHORTNESS, a
scale of THICKNESS, not a scale of THINNESS, a scale of DIFFICULTY, not a
scale of EASINESS, and so on.
Supras can be recognized by their normality in, for instance, questions of the
form What is its Nom?, in structures such as twice as/half as X, and for many
speakers also in How X is it? questions. By these criteria, the default reading of
long is a supra term and that of short a sub term. The oddness of (8b), (8d) and
(8e) is presumably due to the difficulty of construing short as supra (outside of
specific contexts):
(8) a. What is its length?

b. ?What is its shortness?
c. twice as long/half as long
d. ?twice as short/?half as short
e. ?How short is it? (for some speakers)
Notice that, by the same criteria, the default readings of good, bad, hot and cold
are all supras: this is how we know that they are part of biscalar systems.
(9) a. twice as good/ half as good
b. twice as bad/ half as bad
c. How good was it?
d. How bad was it?
(The oddness of What is its goodness/badness is due to the lack of conventional

units of measurement for merit. See §7.4.2.2 below.)
Notice, also, that for some speakers, twice as short and half as short are syn-
onymous, which is a sign that short is being construed as a sub term. No speakers
interpret twice as long and half as long as synonymous.
7.3.2.2 The representation of subs and supras

Now let us consider the question of how to represent the relationship
between a supra term and its sub partner. The way they are portrayed in Figure
7.5 makes them look like equipollents. But this is not correct, because there is an
asymmetry between the two that is not found in equipollents. The problem is how
to represent this asymmetry. In Cruse and Togia 1995, the meanings of long and
short were glossed as follows (with a slight modification):
long = ‘noteworthy by virtue of a relative abundance of length’
short = ‘noteworthy by virtue of a relative paucity of length’
This is not satisfactory: abundance and paucity are nominals related to much and
little – in other words, we are explaining one pair of antonyms in terms of another
pair, and the essence of antonymy is slipping through the net. There is a way of
expressing the relationship without covert resort to antonymy in the definition,
in terms of the ordering of points on the scale of length (0 = zero length; R =
contextually determined reference point; L = length of object):
long: 0, R, L
short: 0, L, R
We can perhaps put it into words thus:

long = ‘noteworthy by virtue of being longer than some contextually determined
reference value’
short = ‘noteworthy by virtue of some contextually determined reference value
being longer’
This shows short to be parasitic on long, and avoids the risk of surreptitiously
incorporating the problem into the solution, but it does not bring out the negative
nature of short, nor its greater cognitive complexity.
Probably, the essence of the difference between long and short lies in the fact that
‘longness’ has a direct correlation with the scale of length, whereas ‘shortness’ has
an inverse correlation, which renders ‘short’ inherently more complex than ‘long’,
but it is still not clear how this might be encapsulated in a definition.
7.3.2.3 The iconicity problem

There is a curious iconicity paradox in connection with basic relative
adjectives such as long:short, wide:narrow, deep:shallow, high:low and so on.
All definitions of these adjectives in their relative uses presuppose an underly-
ing scale of length, width, depth, height and so on. In other words, the absolute
scale appears as conceptually more primitive. Yet the relative adjective is nearly
always morphologically simple, and when the name of the underlying property is
morphologically related to the adjective it is always morphologically complex.
This at first sight appears to run counter to the principle that morphological
complexity mirrors cognitive complexity. One would expect the adjective to be
derived from the noun, as in speed:speedy, weight:weighty, length:lengthy and so
on. Here the iconicity principle is followed, in that the resultant adjective encodes
a more complex and subtle notion than the corresponding ‘simple’ adjective (fast,
heavy, long). (An apparent exception to the rule is beauty:beautiful, but this prob-
ably has a historical explanation, in that beauty came into the language first. It is
worth noting that in French, beau:beauté follows the more usual pattern, with a
morphologically simple adjective and a complex nominal.)
One possible explanation is that, in applying the iconic pinciple, we should
distinguish between structural complexity (in terms of the number of elementary
components and their interconnections) and processing complexity (in terms of the
cognitive effort involved). Perhaps they are acquired first of all in an unanalyzed,
primitive, ‘Gestalt’ sense, which is basically relative. Maybe in order to develop the
full adult system, analysis and restructuring are necessary. Some of the results of
the analysis may well be conceptually simpler in some sense than the analysand, but
the extra effort that has gone into them is mirrored by the morphological complex-
ity. This has some connection with the fact that superordinate terms, which may
be schematic relative to their hyponyms, are nonetheless often morphologically
more complex, especially if the latter are basic-level items.
7.3.2.4 Impartiality and committedness

Impartiality can be illustrated in the first instance with comparatives.
Although sentence (10) is contradictory, sentence (11) is not, showing that there
is no clash between heavier and light:
(10) ?This box is quite light, but it is heavy.

(11) This box is quite light, but it is heavier than that one.
In our terms, heavier is impartial with respect to the weight of the referents
that form its arguments, in that its use is not constrained to those contexts where
heavy is appropriate. The same is true of lighter, although a slightly more complex
test frame is necessary to establish this, as a ‘heavy’ argument displays a greater
affinity for heavier than for lighter (this factor, labeled ‘pull’ in Allen 1986, also
requires explanation):
(12) This box is quite heavy, but it’s still lighter than the other one, nonetheless.
Committed uses of heavy and light are illustrated in (13) and (14):
(13) This box is less heavy than that one.

(14) This box is not as light as that one.
Other examples of impartiality (the [a] examples) and committedness (the [b]
examples) are:
(15) a. How long is it?

b. How short is it?
(16) a. What is its length?
b. ?What is its shortness?
(17) a. I was surprised at the length of the programme.
b. I was surprised at the shortness of the programme.
(18) a. How clean was the room when you moved in?
b. I was surprised at how clean the room was/at the cleanness of the room.
For languages other than English, another type of question shows impartiality
effects. Compare (19a) and (19b) in French and (20a) and (20b) in Modern Greek:
(19) a. Elle est longue, ta nouvelle jupe?

b. Elle est courte, ta nouvelle jupe?
(20) a. Ine makri to kenurgio su fustani?
‘Is your new skirt long?’
b. Ine konto to kenurgio su fustani?
‘Is your new skirt short?’
In each case the (a) question is in some sense ‘open-minded,’ compared to the
(b) question, even though the difference has no effect on the nature of the answers
to the questions – that is to say, the answer must reflect the truth value of the
implicit proposition in each case.
7.3.2.5 The ‘Exposure’ Principle

It is assumed that any use of an antonymous adjective is either to be
given an absolute construal or a relative construal. One broad generalization that
can be made is that all absolute construals of either term of a polar antonym
pair are impartial. However, as far as relative construals are concerned, sub terms
are committed, but supra terms may be impartial or committed. As a working
hypothesis, it will be assumed that whether an adjective receives a relative or an
absolute construal in a particular context is not haphazard.
In this spirit, we may now survey the various uses of polar antonyms, to see
to what extent their partiality can be motivated. One explanatory hypothesis is
what is here called the ‘Exposure Principle.’ First, we need to establish a ‘scale of
exposure.’ In the following, the exposure of the adjective increases from (i) to (v):
(i) cases of suppletion: speed, weight, temperature, worse, better
(ii) cases of morphological distortion: length, width, depth
(iii) undistorted incorporation in a derived word: thickness, hardness, diffi-
culty, longer, thicker
(iv) expressions where the bare adjective occurs in an idiomatic construction:
How long is it?
(v) bare adjectives in fully compositional expressions: It’s long, a long piece
of string
The relevance of this scale to the distribution of impartial and committed expres-
sions is captured in the following generalization: the more exposed an adjective the
greater its affinity for the relative scale, and the less exposed, the greater its affinity
for the absolute scale. So, suppletive forms are maximally likely to be impartial,
while bare adjectives in fully compositional expressions are maximally likely to
have a relative construal. Why should this be so? One suggestion is that this has
to do with the cognitive primitivity of the relative construal: cognitive primitivity
is iconically mirrored by constructional transparency.
This first principle only comes into play when there is a choice of forms. If there is
only one form, a second principle operates, which limits the application of the first
principle, and that is that wherever there is a choice between an impartial reading
and a committed reading, but only one available form (i.e. they are in competition),
the impartial reading has priority. We shall now illustrate the operation of these
principles.
(a) Comparatives
Both terms of every polar antonym, in all the languages for which data
are available, have an impartial comparative form. A committed comparative is
also possible in theory, but it would be cognitively more complex, as it involves

two reference points rather than just one.
Let us consider first a language such as French, which has only one way of
forming the comparative (plus long, etc.). This is maximally transparent, which
would bias it towards a relative construal on the Exposure Principle. However,
this is overruled by the Impartial Priority Principle, with the result that French
comparatives are impartial. English is a little more interesting in this respect. Some
English adjectives, for example intelligent, have only a periphrastic comparative,
and in such cases, the periphrastic comparative is impartial:
(21) Neither of them is very bright, but John is more intelligent than Bill.
But in some cases, we can observe a difference between the interpretation of an

inflectional and a periphrastic comparative:
(22) a. Process X is fast, but it’s nonetheless slower than process Y.

b. ?Process X is fast, but it’s nonetheless more slow than process Y.
Here we can see the effect of the exposure scale: slow in (22b) is more exposed
than in (22a), so (b) is committed and (a) impartial. In English, we can regard
the morphological comparative as the default variety, and the periphrastic form in
(22b) as somewhat forced. We might therefore expect the morphological variety to
attract the priority construal. Hence there are two competing explanations for (22)
(although they are not mutually exclusive). A clearer case is provided by Modern
Greek, where some adjectives have a more evenly balanced choice between a pe-
riphrastic comparative and a morphological comparative. The situation, however,
is similar to English: an adjective that has only a periphrastic form is impartial,
but where it is in competition with a morphological form, it is the latter that is
impartial, and the periphrastic form is committed.
(b) Quantified comparatives

(23) X is twice as/half as long as Y.
(24) ?X is twice as/half as short as Y.
These are quite different to ordinary comparatives. We may assume that any
mention of an explicit numeral or numerical quantity predisposes an absolute con-
strual of the adjective, at least partly because only the absolute scale can be cali-
brated. The oddness of short is presumably because there is no conventional way
of calibrating shortness. Quantified comparatives can be used with non-calibrated
scales (twice as good, only half as bad); presumably these are metaphorical ex-
tensions from the prototypical calibrated scales. Such extensions are limited as
to permissible quantifications: This stick is three and a half times as long as that
one is normal, but ?Jane is three and a half times as pretty as Sarah is distinctly
odd.
(c) Equalities and inequalities

The expressions as long as and not as long as are impartial, but as short
as and not as short as are definitely committed:
(25) a. X is as long as Y/ X is not as long as Y.

b. X is short, but it is as long as Y.
c. X is as short as Y.
d. ?X is long, but it is not as short as Y.
However, although (26) has an impartial reading, (27) does not:
(26) X is not as long as Y.

(27) X is less long than Y.
This is in accordance with the Exposure Principle, since as long as is an idiom,

whereas less long than is arguably compositional. However, the difference between
long and short needs further thought. Perhaps it is due to the fact that as long as
means ‘has the same length as,’ not ‘has the same longness as,’ and this only makes
sense on the absolute scale; as short as, on the other hand, means ‘has the same
shortness as,’ which can only be interpreted with reference to the relative scale,
and is therefore committed.
(d) What is its NOM? questions

(28) What is its length/weight/thickness?
(29) *What is its shortness/lightness/thinness?
(30) *What is its goodness/cleanness?
(31) What is its difficulty/hardness?
Notice that (28) cannot be used to ask Is it long or short?, whereas this is a
possible meaning of How long is it? What is its NOM? seems to demand an absolute
quantity as an answer, and since shortness and the like cannot be quantified in this
way, that is sufficient reason for the anomaly of (29). The same goes for (30),
although here the reason is not that we are dealing with sub terms, but that we are
dealing with noncalibrated quantities (see §7.4.2.2). But why is it that this form of
question demands a quantity, whereas How ADJ is it? does not? A likely answer
is that what? demands a specific individual entity to be identified. The only way
of individuating a quantity of something is by its magnitude. That is fine once
we know that we are dealing with, say, the absolute scale of length. But that still
leaves the question of why we are limited to the absolute scale of length. One
possibility is that it is only on the absolute scale that there is the possibility of
a sufficiently definite identification to satisfy the requirements of what? In other

words, the impressionistic relative scale does not allow anything so definite to be
pinned down.
(e) How X is it? questions

Questions of the form How X is it? have two possible interpretations. For
instance, How long is it? can mean either (i) or (ii):
(i) ‘What is its length?’ (*How short is it?)
(ii) ‘Is it long or short?’ (How short is it?)
Interpretation (i) has no equivalent with short, for the reasons discussed above.
Interpretation (ii) does have a short version, but it is committed. Notice first that the
fact that How long is it? has a relative construal, whereas What is its length? does
not, is in accord with the Exposure Principle. But we need to think about why How
long is it?, on the relative interpretation, is impartial, while How short is it? is com-
mitted (remembering that shorter is impartial). Assuming a general semiotic prin-
ciple that different forms are preferentially assigned different meanings, then all we
have to decide is why long is preferred for the impartial sense. The reason for this
is perhaps a combination of ‘simple before complex’ and ‘impartial before com-
mitted,’ yielding the combination ‘simple’ + ‘impartial’ as the preferred option.
In Turkish and Modern Greek, the How-questions are committed as far as cal-
ibrated scales are concerned; impartial questions require What-questions. This
follows the Exposure Principle, since in these languages, How-questions are max-
imally transparent. This difference with English How-questions fits with the fact
that in both Greek and Turkish the adjective is more exposed, in that the struc-
ture of such questions is not identical to the idiomatic How Adj is it? but is fully
compositional. For instance in the Turkish Ne kadar uzun? (‘How long is it?’), the
expression ne kadar simply means ‘how much,’ and would occur in the Turkish
translation of ‘How much money do you have?’. With noncalibratable quantities,
What-questions are ruled out, and How-questions assume the role of impartiality,
in accordance with the Impartiality Priority Principle.
(f) Is it X? questions
(32) French:
a. Elle est longue, ta nouvelle jupe?
b. Elle est courte?
(33) Modern Greek:
a. Ine makris?
‘Is it long?’
b. Ine kontos?
‘Is it short?’
(34) Turkish:
a. O uzun mu?
‘Is it long?’
b. O kısa mı?
‘Is it short?’
Questions of this form containing the supra term (which seem not to occur
significantly in English) function in a similar way to the relative construal of
How X is it?, that is, to ask whether something is long or short. They occur
particularly in languages that do not have How-questions (e.g. French) or where
How-questions are committed (e.g. Greek and Turkish). The sub term in each
case yields a committed question. In both cases, the relative scale is involved,
and the reasons for choosing the supra term to encode the impartial question are
presumably the same in both cases, namely, that it is conceptually the simplest.
7.3.3 Bi-scalar systems

Biscalar systems can be diagnosed by the fact that both terms of an
antonymic pair are supras. Essentially, biscalar systems incorporate a pair of
counterdirectional monoscalar systems. However, in most cases the individual
monoscalar components do not exhibit the full range of properties described
above. Typically, the supra-supra opposition is the most salient, and the supra-
sub oppositions are to a greater or lesser degree backgrounded, in that sub terms
are either nonexistent, or their properties are shared among a cluster of highly
context-sensitive items.
There are two types of biscalar system, the equipollent type and the overlapping
type. These are easily distinguished by their patterns of impartiality and commit-
tedness. In the discussion above of monoscalar systems, the notion of partiality was
related to the sub-supra distinction. This will henceforth be designated as scale-
partiality (also scale-impartiality, scale-committedness). This notion is still
relevant to some degree to biscalar systems, but the most salient notion is partiality
with regard to the supra-supra distinction, which we shall term system-partiality
(also system-impartiality, system-committedness). Equipollent systems are dis-
tinguished by the fact that the two terms of the opposition are symmetrical in all
their properties. For English, this means that there are no system-impartial uses of
either term. In overlapping systems, one term has system-impartial uses, whereas
the other term has none.
7.3.3.1 Equipollent antonyms

The outstanding characteristic feature of equipollent antonyms is that
the behavior of the two members of a pair is completely symmetrical. In English
they are system-committed in all uses, except where there is a morphologically
unrelated nominalization, such as temperature for hot:cold: hence, hotter, colder,

How hot/cold is it? are all system-committed, but What is its temperature? and
I was surprised by its temperature are both system-impartial.
There are relatively few equipollent pairs in English; the most obvious pair are
hot:cold, but happy:sad, sweet:sour, ashamed of:proud of, and beneficial:harmful
may be cited. The numbers of pairs in this category seem to vary from language
to language. For instance, Cruse has failed (by questioning native speakers) to
establish any convincing pairs in Turkish, whereas Togia’s observations (1996)
indicate that they are relatively common in Modern Greek. We are at present unable
to say whether this difference between Turkish and Modern Greek correlates with
any other semantic or other characteristic of the two languages.
In the case of hot:cold, which may be regarded as the prototypical disjunct
equipollents in English, there is evidence that cool and warm, besides their more
salient function as attenuatives of cold and hot respectively, double up as subs,
this time for, respectively, hot and cold. Thus, (35) and (36) are significantly less
anomalous than (37) and (38):
(35) This pan feels hot, but it feels cooler than that one/still feels cooler than that one,
nonetheless.
(36) This bottle feels cold, but it feels warmer than that one/still feels warmer than
that one, nonetheless.
(37) ?This pan feels hot, but it feels colder than that one/still feels colder than that one,
nonetheless.
(38) ?This bottle feels cold, but it feels hotter than that one/still feels hotter than that
one, nonetheless.
The sub term perhaps also occurs in a cool oven, a cool flame and a relatively
warm ice age. Cruse’s intuitions are that This flame is only half as cool as that one
is difficult to process, and resembles half as short. Notice, however, that reversing
the terms in (35) and (36) leads to anomaly:
(39) ?This pan feels cool, but it feels hotter than that one/still feels hotter than that
one, nonetheless.
(40) ?This bottle feels warm, but it feels colder than that one/still feels colder than that
one, nonetheless.
This suggests that the default readings of cool and warm are not subs of hot and
cold, and that the sub construal has to be coerced by contextual pressure. Wherever
the sub reading is appropriate, either inherently hot/cold items are involved (such
as ice-age, flame), or there is some other explicit indication that the domain in
question is one of hotness or coldness.
Parallel equipollent pairs are even less common than the disjunct type. They
seem to occur when (a) there is no property that can be construed as having a
zero value in the middle of the scale, (b) each direction of construal is equally
motivated and (c) there is no dominant viewpoint, no difference that motivates
a ‘positive/negative’ construal. Hayes (2001) argues that hard:soft are plausible
candidates, and dark:light as applied to colors.
7.3.3.2 Overlapping antonyms

Overlapping antonym oppositions have the following characteristics:
(i) The comparative form of one member is system-impartial (and, because of
the disposition of the scales, scale-impartial), while that of the other is system-
committed (although scale-impartial):
(41) John and Bill both got bad marks in the exam, although John’s marks were better
than Bill’s.
(42) ?John and Bill both got first class marks in the exam; John’s marks, however,
were worse than Bill’s.
(ii) Both members are normal in quantified comparatives; one member is system-
impartial ([43] and [44]) and the other system-committed ([45] and [46]):
(43) Both marks were admittedly bad, but Bill’s was twice as good as John’s.
(44) It’s true that Bill’s wasn’t the only bad mark, but it was only half as good as even
the other failures.
(An impartial reading of half as good is difficult to construe in [44], but it is
possible.)
(45) ?Both marks were excellent, but Bill’s was twice as bad as John’s.
(46) ?Both marks were excellent, but Bill’s was only half as bad as John’s.
(iii) Both members are normal in How-questions; one yields a system-impartial

question, the other a system-committed question:
(47) How good were the results this year?

(48) How bad were the results this year?
Typical members of this group are: good:bad, kind:cruel, polite:rude. Every

supra term potentially has a sub partner, but in content areas that lend themselves
to one of the biscalar antonymic patterns, the supra-sub opposition may be less
salient than the supra-supra opposition. This, however, is not universally the case,
and it is at present not clear what circumstances favor one opposition or the other
as the most salient.
In English it appears to be the case that supra-supra contrasts have precedence.
Take the case of the good:bad pair. Each of these has sub partners. The partner
of good is poor. Compare the normality of ?How poor is it? and How bad is
it?, and ?twice as poor and twice as bad. (In German, although schlecht is a sub
partner of gut, the gut:schlecht opposition is more salient than the supra-supra
gut:schlimm opposition.) Notice that although How bad is it?, worse and twice
as bad are system-committed, they are scale-impartial, since they presuppose no
particular value on the ‘bad’ scale. There is no single sub term for bad, but a
choice of contextually restricted possibilities, such as mild and slight: compare
?How mild is your gout? and ?How slight was your accident? with How bad was
your gout/accident?
Although absolute scale-schemas are necessary to explain the scale-impartiality
of better and worse, they are not calibrated, and this gives rise to differences with
polars. First, neither member yields a normal nominalization question:
(49) ?What was the goodness of the film?
(50) ?What was the cleanness of the room?
(51) ?What was John’s politeness?
(cf. What was the length of the rope?)
Second, How-questions have only one interpretation, namely, a relative one.

This explains why (52) is normal, but (53) is odd:
(52) A: The last one was quite short.
B: How long was it?
(Long, here, gets an absolute reading.)
(53) ?A: The results were rather poor/bad.
B: How good were they?
(B’s question only has a relative reading, i.e. ‘Were they good or bad/poor?’ But
A has already given that information, hence the oddness of the question.)
Third, both members are committed in (54)–(56):

(54) I was surprised by its X-ness.
(55) I was surprised by John’s politeness.
(56) I was surprised at the room’s cleanness.
Given that there are no calibrated quantities with the scales of overlapping
antonyms, a relative reading seems adequately motivated.
(iv) Inherentness is a property of overlapping antonyms, and of no other variety.

It can be illustrated with good:bad. Sentence (57) is quite normal:
(57) John, Bill and Tom are all pretty hopeless at tennis; John is a little bit better than
Bill, while Tom is slightly worse.
This is what we would expect, with better showing impartiality. However, while
(58) is normal, neither (59) nor (60) is:
(58) The drought last year was worse than this year.
(59) ?The drought this year is better than last year.
(60) ?How good was the drought last year?
It seems that certain items do not collocate with better or How good . . .?,
even though they collocate perfectly normally with worse. These items share the
property of being normally construed as ‘inherently bad’: accidents, illnesses,
famines, droughts, earthquakes and so on. The inherent badness (or whatever) of
some entity is, of course, a construal, and varies with contextual factors. Sentence
(61) is more likely in a conversation between two seismologists than between two
victims of the disaster in question:
(61) A: How was the earthquake?
B: Quite good – better than the last one.
Example (62) provides an example involving clean (clean:dirty are construed

in some contexts as overlapping):
(62) ?This smudge is cleaner than that one.
A possible explanation of the inherentness effect utilizes the fact that the BAD-
NESS scale has only half the extent of the MERIT scale. Perhaps there is a kind
of Gricean implicature that the scale on which we place something is the smallest
that encompasses all the possibilities for a particular construed domain. Hence, if
we say, for instance, How good was the film?, we implicate that we are prepared
for an answer anywhere on the scale of MERIT. The reason we say How bad was
the famine, rather than How good was the famine?, is because the BADNESS scale
encompasses all the expected possibilities for a famine. How good was the famine
would implicate that we construe the domain of famines as potentially containing
good examples.
7.4 Variable construal of antonyms and complementaries
In illustrating the different antonymic patterns available, we have relied

heavily on default construals in minimal contexts. In this section we examine the
extent of context variability in the construal of pairs of lexical items conventionally
regarded as antonymic opposites.
7.4.1 Absolute vs. relative construal

Some opposite pairs behave in some contexts like a pair of complemen-
taries, in other contexts like a pair of antonyms and in yet other contexts as a hybrid
opposition in which one term is gradable and the other absolute. A typical example
is clean:dirty. Example (63) shows them behaving like complementaries:
(63) I’ve put the clean shirts in the drawer and the dirty ones in this bag.
Here, a particular domain of shirts is construed as having only two subdivisions,

clean ones and dirty ones, and what does not fall under one heading necessarily
falls under the other. In (64), on the other hand, dirtiness is construed as a gradable
property, and in (65) cleanness is construed as gradable. Most speakers would
confidently infer the truth of (65) from the truth of (64), which means that clean
and dirty are behaving like antonyms:
(64) This shirt is dirtier than that one.

(65) That shirt is cleaner than this one.
(Notice that both [64] and [65] imply that the shirt in question is at least slightly
dirty.) In general, a gradable construal of a property is indicated by the use of
the morphological comparative or superlative, or intensifiers such as more, fairly,
quite, rather, extremely and so on.
In (66), we have a hybrid situation, with dirty construed as relative, but clean
construed as absolute:
(66) A: How are you getting on with that very dirty pan?
B: Well, it’s almost clean – give me another ten minutes and I’ll have it clean.
Hybrid construals are particularly common with expressions of completeness

and incompleteness, such as half (as in half clean), almost, nearly, practically,
completely, quite, utterly. Notice that It’s neither X nor Y prevents a dichotomous
construal, hence, It’s neither clean nor dirty is either odd or it coerces a hybrid
construal with dirty being given a relative construal and clean an absolute construal.
On this interpretation, not dirty receives an interpretation ‘not far enough along the
dirty-scale to deserve the unmodified label dirty,’ that is to say, ‘only slightly dirty.’
If this possibility is explicitly ruled out, the expression is virtually unconstruable:
*It’s neither clean nor even slightly dirty. It’s neither dead nor alive either requires
a wider construal of the domain (e.g. A piece of chalk is neither dead nor alive), or
the envisaging of odd states such as ghosts or vampires; It’s neither possible nor
impossible is very hard to construe at all. The final construal, as always, is a result
of interaction between constraints of different kinds and different strengths.
In a hybrid construal, the terms are normally not interchangeable. For instance,
It’s neither clean nor dirty cannot be construed with dirty given an absolute con-
strual and clean a relative construal. And (66) with the terms reversed not only
denotes an odd activity, but is actually not interpretable:
(67) A: How are you getting on with that clean pan?

B: Well, it’s almost dirty – give me another ten minutes and I’ll have it dirty.
The reason for this constraint, which appears to be cognitive, rather than con-
ventional, lies in the nature of the clean-dirty opposition. It represents what in
Cruse 1986 was called a privative opposition, that is to say, one in which one
term denotes the absence of a property and its partner denotes the presence of the
property. In the clean-dirty opposition, clean denotes the absence of dirt, rather
than the converse:
(68) When something is clean, there is no dirt present.

(69) ?When something is dirty, there is no cleanness present.
The notion of ‘zero dirtiness’ is of course a context-dependent construal: we

demand different degrees of absolute objective cleanness in different circumstances
as a qualification for the label clean. Think of a pair of shoes, a kitchen knife,
a surgeon’s scalpel. In the expression cleaner than clean, cleaner needs to be
interpreted with reference to an absolute scale of cleanness, whereas clean denotes
the current contextually appropriate reference point.
The reason clean has positive polarity (defined as the greater potential for im-
partial use; cf. How clean is it?, which is impartial, while How dirty is it? is
committed), even though it is an ‘absence’ term, that is, logically negative, is be-
cause there is a strong conventional constraint favoring a positive evaluation. This
can be overridden by contextual constraints: after a rugby match on a particularly
muddy pitch, a description of the players as gloriously dirty is construable; dirty
as applied to a film or book is frequently given a positive evaluation nowadays –
in a recent review, a novel was described as gloriously filthy. However, although
dirty can be construed as positive, where there are sufficiently strong contextual
constraints, what seems to be virtually impossible is to construe dirty as the zero
point of a scale of cleanness.
There is a small number of opposites that behave like clean and dirty: they display
an unusual freedom in respect of construal type. In other cases there is evidence of
conventional or cognitive constraints operating. Take the case of dead:alive. With
these a symmetrical absolute construal is readily accessible:
(70) A: Is it dead?
B: No, look, it’s breathing – it’s still alive.
Also, it is not difficult to construe alive as gradable:
(71) You look rather more alive than you did half an hour ago!
However, dead is considerably less comfortable in this context:
(72) ?You look rather more dead than you did half an hour ago!
Hence, dead seems more tied to an absolute construal than alive (or, indeed,
clean). A hybrid construal is fairly normal (notice that it is not dead that is being
graded here, but ‘not-dead-ness’):
(73) A: You look half-dead.

B: I feel three-quarters dead.
Figurative uses of dead may be less odd when construed as gradable, but this is
strictly irrelevant as far as literal construals are concerned:
(74) Every time I come back to this town it seems even more dead than the last time I
was here.
As a final example of absolute/relative alternation, consider married:single.

These are usually considered quintessential complementaries, and the fact that
there are well-defined legal criteria that can be appealed to lends support to this
construal:
(75) I need to ask you a few questions. Are you married or single?
However, as I have noted elsewhere, Iris Murdoch has (76) in one of her novels:
(76) Jane was very married.
Most speakers find this interpretable, but at the same time feel they are working
against palpable constraints in coming to the appropriate construal. Once one
accepts this construal, (77) presents no problems:
(77) Jane is more married than Mary.
But it is much harder to construe (78) as representing the same relation between
Jane and Mary as (77):
(78) Mary is more single than Jane.
That is, (78) cannot be construed as attributing different degrees of ‘married-ness’

to Jane and Mary. However, (78) is a possible expression of a situation where
Jane has a more-or-less steady boyfriend, whereas Mary has several men on the
go at any one time, none of whom last very long. In other words, we can construe
‘married-ness’ as a gradable property, but we cannot construe single as sub partner
to married: we can, however, with some cognitive effort, construe ‘married-ness’
and ‘single-ness’ as equipollent partners.
In contrast, a careful examination of clean:dirty in their manifestation as grad-
able antonyms shows them to belong preferentially to the overlapping type. One
indication of this is the element of inherentness, which has already been remarked
on:
(79) This smudge is dirtier than that one.

(80) ?That smudge is cleaner than that one.
While many lexical items denoting properties can be construed, according to

context, as either absolute or gradable, others seem not to have the two possibilities.
Particularly interesting are those that are gradable in their default construals. Why
is it that, for instance, cold can receive an absolute construal, as in (81), whereas
slow, light and cheap cannot under any circumstances be interpreted as ‘stationary’,
‘weightless’ or ‘free’, respectively, that is, as zero on the respective scales (as in
[82]–[84]):
(81) Your dinner’s almost cold – hurry up and come to table!
(Here, cold represents ‘zero hotness,’ i.e. room temperature.)
(82) ?The car was traveling completely slowly.
(83) ?In space one is completely light.
(84) ?I didn’t pay anything – they were completely cheap.
This is a difficult question. From available examples, it seems that only supra
terms can support an absolute construal. However, not all supras have absolute
construals: *almost long/heavy/fast. Also necessary is some sort of natural deter-
minate value, frequently also construed as a zero. The reason for the absence of
absolute construals for sub terms is not at present obvious. It may have something
to do with the fact that they are impressionistic, and do not have a linear relation
with the absolute scale. That is to say, to get equal increments of impressionistic
shortness, we must, let’s say, successively halve the absolute length. That way, zero
will be approached asymptotically, but will never be reached. Put another way, the
default construal of the SLOW scale has no end. There is a parallel with impres-
sionistic scales of loudness and brightness, which do not have a linear relation to
absolute energy level.
Another possibility is that, in those cases mentioned, either the zero point on
the scale is of no practical communicative importance (what use have we for the
notion of zero length?) or the zero point is preempted by another lexical item (as
with stationary, weightless, free). (Of course this could work the other way round:
we have a separate lexical item for the zero point because the sub term cannot
cover it. This has some intuitive plausibility. More work is needed on this point.)
7.4.2 Scale features

7.4.2.1 Direction of scale and number of scales
Graded properties are normally construed in such a way that an increase
in salience is equated with a higher degree of the property, hence we operate with
a scale of LENGTH rather than a scale of SHORTH, with long as supra and short
as sub. However, there are contexts where the roles of long and short appear to
be reversed. For instance, (85) and (86) appeared in a newspaper article about the
miniaturization of computer components:
(85) The new device is ten times smaller than anything seen previously.
(86) ‘The nanotube transistors are about 350 times smaller than a conventional silicon
transistor and faster.’ (Guardian On-line Supplement, October 10, 2002, p. 11)
And (87), in the same context, seems normal (although not actually attested):
(87) A. Every week we produce smaller and smaller chips.

B. How small do you think you’ll be able to get them?
A: Well, we are not sure, but our best efforts last year were only half as small as
this year’s.
It seems that in (85)–(87) the scale of linear extent is being construed in the
reverse direction from the default construal, as a scale of SMALLNESS rather than
a scale of BIGNESS/LARGENESS, with small as the supra term. This reversal
goes against quite strong cognitive constraints, and would not be possible unless
there were definite factors favoring it. In this case there are such factors, namely, the
fact that an increase in smallness is correlated with an increase in interestingness,
salience, desirability.
Many instances of scale reversal are correlated with a change of antonym type.
This is particularly so where the motivation for the reversal is related to evalua-
tiveness. Take the case, discussed earlier, of easy:difficult. It is not hard to con-
strue a scale of DIFFICULTY with difficult and easy as polars, difficult as supra,
and How difficult is it? as an impartial question. But an average class of under-
graduates finds it difficult to decide whether difficult or easy is the supra term,
with a significant number opting for easy, and an underlying scale of easiness.
This seems to be supported by the relative normality out of context of How easy
is it?
A significant determining factor seems to be whether ‘difficulty’ is judged neg-
atively or objectively. If ‘easy’ is positively evaluated, and ‘difficult’ negatively
evaluated, this encourages the movement of easy:difficult from the polar construal
to the overlapping construal, with both easy and difficult as supras, but easy as
the major term (parallel to good). A similar alternation in antonym type can be
observed with cheap:expensive (Hayes 2001). Where these are used simply to in-
dicate price, they behave more like polars, with cheap as sub; but where cheap has
connotations of ‘poor quality’ and expensive of ‘high quality,’ they behave more
like overlapping antonyms, with cheap being a (minor) supra term.
A different alternation is found with thin, in opposition to thick and fat. There is
no doubt that thick:thin are polars, with thin as sub term; in other words, there is no
scale of thinness, only one of thickness (in the default construal). The majority of
native-speaker informants, however, class fat:thin as equipollents, out of context.
This involves reclassifying thin as supra, that is, construing a scale of thinness. The
motivation seems to be that there is a societal norm for girth, which is positively
evaluated, and significant deviations in either direction are negatively evaluated:
the evaluative feature has the effect of encouraging the construal of a scale of
thinness (more of the property being more salient).
7.4.2.2 Effects of calibration

The availability or unavailability of conventional units for measuring the
degree of a property can make a difference to the way a gradable adjective behaves.
Take the case of strong. We may speak equally normally of a strong man and
a strong solution of a chemical, and in both cases it would be normal to ask How
strong is the solution? and How strong is that man? However, there is a clear
difference in normality between What is the strength of the solution? and ?What is
the strength of that man? The reason is that we are not familiar with conventional
units for measuring the strength of a man, whereas the strength of a solution can
be measured in conventional units. In this case, it might be objected that we are
dealing with two autonomous readings of strong, since (88) has a definite air of
punning:
(88) ?Mary likes her tea and her men to be very strong.
Perhaps a more convincing example is provided by strong tea and strong beer.
These two coordinate happily without zeugma:
(89) John likes his tea and his beer to be very strong.
Yet (90) is much more normal than (91), the reason being that the strength of
beer is commonly measured in terms of the percentage of alcohol it contains, but
there are as yet no units for measuring the strength of tea:
(90) What is the strength of this beer?
(91) ?What is the strength of this tea?
It would be difficult to argue that the properties referred to in (90) and (91) are
different. The situation is similar in the case of hard. There is a scientific scale
for expressing the hardness of minerals, so the question What is the hardness of
quartz? is perfectly well formed in a scientific context. However, although some
types of wood are harder than others, the question What is the hardness of this
wood? is, outside of a specially contrived context, odd. Many other examples could
be given.
7.5 Conclusion
The investigations outlined here have shown the complex and varied
behavior of antonyms to be remarkably non-arbitrary. In comparison with the other
sense relations studied, the role of cognitive constraints is especially prominent.
8
Metaphor
8.1 Figurative language
Prototypical figurative language will be characterized here as language

use where, from the speaker’s point of view, conventional constraints are delib-
erately infringed in the service of communication, and from the hearer’s point of
view, a satisfactory (i.e. relevant) interpretation can only be achieved if conven-
tional constraints on interpretation are overridden by contextual constraints.
What is the motivation for figurative uses of language? Here we need to distin-
guish the speaker’s motivation for using an expression figuratively, and the hearer’s
motivation for assigning a figurative construal to an expression. Briefly, a speaker
uses an expression figuratively when he/she feels that no literal use will produce
the same effect. The figurative use may simply be more attention-grabbing, or it
might conjure up a complex image not attainable any other way, or it may permit
the conveyance of new concepts. As far as the hearer is concerned, the most obvi-
ous reason for opting for a figurative construal is the fact that no equally accessible
and relevant literal construal is available.
The major types of figurative usage are metaphor and metonymy. Metaphor and
metonymy both involve a vehicle and a target. Metaphor involves an interaction
between two domains construed from two regions of purport, and the content of
the vehicle domain is an ingredient of the construed target through processes of
correspondence and blending. For instance, in (1) (from Patricia Cornwell’s Black
Notice) the speaker’s mental processes are presented as having simultaneously the
character of thoughts and small sinister creatures:
(1) A myriad of ugly, dark thoughts clung to my reason and dug in with their claws.
In metonymy, the vehicle’s function is merely to identify the target construal.

For instance, in (2) there is no combining of the features of cars and humans –
the use of you is simply an easy route to the intended referent, the car relevantly
associated with the addressee:
(2) Where are you parked?
193
Metaphor and metonymy have been a major preoccupation of cognitive lin-

guists generally. Metonymy is discussed briefly in §3.2.1; here, the focus is on
metaphor.
8.2 The conceptual theory of metaphor
8.2.1 Introduction
Cognitive linguists reject the so-called substitution theory of metaphor,
according to which a metaphorical expression replaces some literal expression that
has the same meaning. Metaphors (‘true’ metaphors), in general, are not literally
paraphrasable: they have a character that no literal expression has. At the same time,
although metaphorical meaning has a special character that distinguishes it from
any literal meaning, it has the same range of basic functions as literal meaning. Of
course, many metaphorical expressions have a heavy load of expressive meaning,
but so do many literal expressions. In other words, metaphorical meaning is not,
at least in basic functional respects, a special kind of meaning: it is rather the case
that metaphor is the result of a special process for arriving at, or construing, a
meaning.
One of the most influential books to emerge from the cognitive linguistic tra-
dition is Lakoff and Johnson’s Metaphors we live by (Lakoff and Johnson 1980;
see also Lakoff and Turner 1989; Lakoff 1987, 1993). Lakoff and his colleagues
use evidence from everyday conventional linguistic expressions to infer the ex-
istence of metaphorical relations or mappings between conceptual domains (in
the sense of chapter 2) in the human mind. Lakoff’s primary goal in developing
the conceptual theory of metaphor is to uncover these metaphorical mappings be-
tween domains and how they have guided human reasoning and behavior, as can
be seen by his subsequent application of metaphor theory to literature (Lakoff and
Turner 1989), philosophy (Johnson 1987; Lakoff and Johnson 1999), mathematics
(Lakoff and Núñez 2000) and even politics (Lakoff 1996).
Because of Lakoff’s aim to uncover deeply embedded conceptual relations in
the mind, for him the ideal metaphorical expressions to analyze are not the widely
discussed type of examples in (3), but rather those in (4):
(3) a. Juliet is the sun. (Shakespeare)

b. my wife . . . whose waist is an hourglass (from André Breton; Lakoff and
Turner 1989:90)
(4) a. I’ll see you at 2 o’clock.
b. He is in danger.
Metaphor 195
c. Her anger boiled over.

d. She’s had to contend with many obstacles in her life, but she has come a long
way since her days in the orphanage.
The expressions in (3) differ from those in (4) in two important respects. First,
the expressions in (3) are all of the form X is Y where X and Y are both nominal ex-
pressions, a quite uncommon type of metaphorical expression in ordinary speech.
The expressions in (4) are all common, everyday constructions in which metaphor-
ically used prepositions, verbs and other expressions (typically relational in na-
ture) combine with literal phrases (typically nominals functioning as arguments
of the metaphorical relational elements). Hence the metaphorical expressions in
(3) are grammatically and semantically quite different from those in (4). This dis-
tinction is pertinent because much psycholinguistic research on metaphor (e.g.
Gentner 1983, 1988), including research purported to test Lakoff and Johnson’s
theory (e.g. Glucksberg 2001), is based on the metaphor type illustrated in (3),
not (4).
Second, and more important, the metaphors in (3) are novel creations ([3a–b]
are both from literary works, for example) while the metaphors in (4) are con-
ventionalized linguistic expressions, another aspect of their common everyday
character. Lakoff and Turner distinguish novel metaphors from conventionalized
metaphors, calling the former ‘image metaphors’ (Lakoff and Turner 1989:99;
Lakoff 1993:229; see below). Of course, many metaphors do not become conven-
tionalized. But certain metaphors do get conventionalized, and more interesting,
the same metaphors tend to become conventionalized independently across lan-
guages. There is presumably some reason why certain metaphors are conventional-
ized again and again across languages, while others are not. Lakoff and colleagues
argue that their repeated conventionalization is due to their cognitive significance,
which in turn is grounded in human experience (hence the title Metaphors we
live by). Thus, the main focus of their theory of metaphor is of conventional
metaphors, not novel metaphors; we will return to this point at the end of this
section.
The central characteristic of Lakoff and Johnson’s theory of (conventional)
metaphor is that the metaphor is not a property of individual linguistic expressions
and their meanings, but of whole conceptual domains. In principle, any concept
from the source domain – the domain supporting the literal meaning of the ex-
pression – can be used to describe a concept in the target domain – the domain
the sentence is actually about.
For example, the literal meaning of at in (4a) is locative in nature, but it has
been metaphorically extended to apply also to time. Likewise, in in (4b) has a
basic locative meaning, and the use in (4) is a metaphorical extension of this:
here, a state (danger) is conceived as a container that one can be inside of or
outside of. But many other locative expressions can be used to describe time, as
in (5), and many container expressions can be used for a wide range of states, as
in (6):
(5) a. We have entered the 21st century.

b. I finished this in two hours.
c. They worked through the night.
(6) a. They’re in love.
b. How do we get out of this mess?
c. He fell into a deep depression.
Lakoff and Johnson use a formula target domain is source domain to

describe the metaphorical link between the domains. The metaphorical mappings
in (4a) and (5) are manifestations of the time is space metaphor, and those
in (4b) and 6 of the states are containers metaphor (Lakoff and John-
son 1980:31–32). Likewise, (4c) is a manifestation of the anger is heat of
a fluid metaphor, and (4d), the love is a journey metaphor. As Lakoff
puts it:
What constitutes the love is a journey metaphor is not any particular word
or expression. It is the ontological mapping across conceptual domains, from the
source domain of journeys to the target domain of love. The metaphor is not just
a matter of language, but of thought and reason. The language is secondary. The
mapping is primary, in that it sanctions the use of source domain language and
inference patterns for target domain concepts. The mapping is conventional; that
is, it is a fixed part of our conceptual system, one of our conventional ways of
conceptualising love relationships. (Lakoff 1993:208)
A (conventional) metaphor is therefore a conceptual mapping between two

domains. The mapping is asymmetrical, however: the metaphorical expres-
sion profiles a conceptual structure in the target domain, not the source
domain.
The mapping between source and target domains involves two sorts of cor-
respondences, epistemic and ontological. The ontological correspondences hold
between elements of one domain and elements of the other domain; epistemic cor-
respondences are correspondences between relations holding between elements
in one domain and relations between elements in the other domain (this includes,
for instance, encyclopedic knowledge about the domain). The phenomenon of
correspondence will be illustrated using the example of a n g e r i s h e at o f a
fluid (Lakoff 1987:387):
Metaphor 197
(7) Ontological correspondences

source: heat of fluid target: anger
container body
heat of fluid anger
heat scale anger scale
pressure in container experienced pressure
agitation of boiling fluid experienced agitation
limit of container’s resistance limit of person’s ability to suppress anger
explosion loss of control
(8) Epistemic correspondences

When fluid in a container is heated When anger increases beyond a certain
beyond a certain limit, pressure limit, ‘pressure’ increases to point at
increases to point at which which person loses control.
container explodes.
An explosion is damaging to Loss of control is damaging to person
container and dangerous to and dangerous to others.
bystanders.
Explosion can be prevented by Anger can be suppressed by force of will.
applying sufficient force and
counterpressure.
Controlled release of pressure may Anger can be released in a controlled
occur, which reduces danger of way, or vented harmlessly, thus reducing
explosion. level.
In general, metaphors are conceptual structures, and are not merely linguistic
in nature, although, of course, they are normally realized linguistically. The cor-
respondences between domains are represented in the conceptual system, and are
fully conventionalized among members of a speech community. An open-ended
range of linguistic expressions can tap into the same conceptual structure in both
conventional and unconventional ways, and be understood immediately: a concep-
tual metaphor cannot therefore be reduced to a finite set of linguistic expressions.
What Lakoff calls ‘elaborations’ involve more specific versions of a basic metaphor
whose characteristics in the source domain carry over to the target domain. For
instance, the difference in intensity between boil and simmer in reference to a
heated liquid carries over to indicate corresponding differences in degree of anger
in to boil with anger and to simmer with anger.
Another consequence of the conceptual nature of metaphor is that certain pat-
terns of reasoning may carry over from the source domain to the target domain.
Lakoff calls these ‘metaphorical entailments’ (it is not clear how metaphorical
entailments differ from epistemic correspondences). For instance, with reference
to the metaphor She demolished his argument, an example of the argument
is war metaphor: if you destroy all your enemy’s weapons, you win the war;
similarly, if you demolish all your opponent’s points in an argument, you win the
argument.
We may summarize Lakoff’s conceptual theory of metaphor as follows:
(i) It is a theory of recurrently conventionalized expressions in everyday lan-
guage in which literal and metaphorical elements are intimately combined
grammatically.
(ii) The conventional metaphorical expressions are not a purely linguistic
phenomenon, but the manifestation of a conceptual mapping between
two semantic domains; hence the mapping is general and productive
(and assumed to be characteristic of the human mind).
(iii) The metaphorical mapping is asymmetrical: the expression is about a
situation in one domain (the target domain) using concepts mapped over
from another domain (the source domain).
(iv) The metaphorical mapping can be used for metaphorical reasoning about
concepts in the target domain.
8.2.2 Issues in the conceptual theory of metaphor

The sketch of Lakoff’s conceptual theory of metaphor in the preceding
section presents only the major premises of the model. In this section, we will
examine some issues about the conceptual theory of metaphor that have led to
elaborations of the basic model.
The first issue we describe is the deceptively simple one of how best to describe
a particular metaphorical mapping. For example, the expressions in (9) illustrate a
metaphor described by Lakoff and Johnson as an argument/theory is a
bu i lding (Lakoff and Johnson 1980:46):
(9) a. We need to construct a strong argument for that.
b. The argument collapsed.
c. We need to buttress the theory with solid arguments.
However, other expressions making reference to buildings do not participate in

the metaphor (Clausner and Croft 1997:260; Grady 1997:270):
(10) a. *Is that the basement of your theory?
b. *That line of reasoning has no plumbing.
c. *This theory has French windows.
The examples in (10) suggest that the metaphor should be formulated more
concisely, that is, using less schematic source and target domains, in such a
way that the metaphorical mapping is valid for the concepts in the source and
target domains. Clausner and Croft propose the more specific formulation t h e
convincingness of an argument is the structural integrity
Metaphor 199
of a building . In some cases, the more schematic metaphorical mapping is

replaced by two (or more) distinct specific mappings. Clausner and Croft
reformulate Lakoff and Johnson’s love is a patient metaphor (p. 49; ex-
amples [11a–b] and [12c] below) as two more specific metaphors, a social
relationship is bodily health (11) and a social relationship
is life ([12]; Clausner and Croft 1997:261–62):
(11) a. This is a sick relationship.

b. They have a strong, healthy marriage.
c. *Their relationship went to the hospital.
(12) a. The marriage is dead – it can’t be revived.
b. Her selfishness killed the relationship.
c. His effort to understand her breathed new life into the marriage.
Clausner and Croft also argue that metaphors vary in productivity. Many
metaphors, such as the convincingness of an argument is the
structural integrity of a building , appear to be completely pro-
ductive, once formulated at the appropriate level of schematicity. Other metaphors
are only partially productive, in that some expressions are acceptable and others
are not. Examples of partially productive metaphors are the revelation idioms:
(13) spill the beans, let the cat out of the bag, blow the whistle, blow the lid off, loose
lips
The expressions in (13) all have as a target domain t o r e v e a l a s e c r e t. But

other similar expressions are not conventional in English:
(14) *spill the peas, *let the cat out of the house
One cannot formulate a single metaphorical mapping at any level of schematic-

ity that would include the expressions in (13) but not the expressions in (14). Yet
the expressions in (13) are understood metaphorically. Gibbs and O’Brien (1990)
found that the conventional expressions in (13) had coherent mental images asso-
ciated with them by subjects, while the unacceptable expressions in (14) did not.
The fact that the idioms were ‘imageable’ (Lakoff 1993:211) indicates that there
is a metaphorical mapping present; the fact that the nonconventional yet semanti-
cally similar expressions in (14) are not consistently imageable indicates that the
metaphor is not completely productive. Finally, truly opaque idiomatic expressions
such as kick the bucket and by and large are not interpreted in terms of mapping
from a source domain. Clausner and Croft argue that the usage-based model (see
chapter 11) can be used to model degree of productivity of metaphorical/idiomatic
expressions.
Lakoff and Johnson allow for metaphors to exist at different levels of schematic-
ity, that is, in a taxonomic hierarchy. For example, they posit a schematic metaphor
love is a journey , and illustrate three more specific instantiations of the

metaphor, love is a car trip/train trip/sea voyage , illustrated in
(15)–(17) respectively (Lakoff and Johnson 1980:44–45):
(15) a. This relationship is a dead-end street.

b. We’re just spinning our wheels.
(16) a. We’ve gone off the tracks.
(17) a. Our marriage is on the rocks.
b. Their relationship is foundering.
Lakoff (1993:222) further adds that love is a journey can be grouped with
a career is a journey under a more schematic metaphor a purposeful
life is a journey , which in turn is an instance of what he calls the event
structure metaphor (roughly, action is directed motion ), which includes
the mappings in (18):
(18) States are locations.

Changes are movements.
Causes are forces.
Actions are self-propelled movements.
Purposes are destinations.
Means are paths to destinations.
Grady (1997, 1998) argues for a combination of decomposition of specific

metaphors and the subsumption of the parts into highly schematic metaphors
that combine with each other. For example, Grady takes the narrowed version
of an argument/theory is a building and analyzes it into the two
metaphors organization is physical structure and persisting is
remaining erect (Grady 1997:273). The two metaphor parts are formulated
in a schematic fashion in order to capture other metaphors, which Grady argues
are part of the same mapping, for example other argument metaphors as in (19)
(Grady 1997: 272), and other target domains for structures as in (20) (Grady 1997:
271):
(19) a. They tore the theory to shreds [fabric ]

b. Their theory of a masterpiece of logical construction [wo r k o f a r t]
(20) a. The Federal Reserve is the cornerstone of the nation’s banking system.
[financial system]
b. Recent land development has caused the near collapse of the Bay’s ecosystem.
[e c o s y s t e m ]
Grady (1998) presents a similar decomposition of the conduit metaphor (Reddy

1979[1993]).
Metaphor 201
Grady’s and Lakoff’s highly schematic analyses raise the question of which
metaphors are more basic to human understanding, the more specific or the more
schematic ones? To address this question, however, we must first address another
issue, namely what conceptual structures are mapped in the metaphor.
Lakoff proposes the Invariance Hypothesis as a constraint on metaphorical map-
ping (Lakoff 1990:54):
(21) Invariance Hypothesis: Metaphorical mappings preserve the cognitive topology

(that is, image-schematic structure) of the source domain.
Image schemas, as discussed in chapter 3, include much of the basic structuring

of experience, such as scales, causation, containment, motion and so on. Lakoff
argues in particular that reasoning in the target domain (metaphorical entailments)
is governed by the image-schematic structure of the source domain (for exam-
ple, consider the epistemic correspondences for anger is heat of a fluid
above).
Turner proposes an important constraint on the Invariance Hypothesis:
In metaphor, we are constrained not to violate the image-schematic structure of

the target; this entails that we are constrained not to violate whatever image-
schematic structure may be possessed by non-image components of the target.
(Turner 1990:252)
Lakoff calls these ‘target domain overrides’ (1993:216), and illustrates them with
give a kick and give an idea. When you give someone a kick, the person does not
‘have’ the kick afterward, and when you give someone an idea, you still ‘have’ the
idea. The target domain of transfer of energy or force does not allow that energy to
continue to exist after the transmission event, hence that metaphorical entailment
does not hold. Likewise, the target domain of knowledge does not imply that
knowledge transmitted is lost: that metaphorical entailment does not hold either.
The Invariance Hypothesis and the target domain override raise a fundamental
issue about conceptual metaphors: why do they exist in the first place? If the target
domain has image-schematic structure already, which can override the metaphor,
then why do we have metaphors? Likewise, if we can isolate image-schematic
structure, or construct highly schematic metaphors such as organization is
physical structure , is it not simply a highly schematic conceptual struc-
ture that is instantiated in both source and target domains? If so, then is it really
a metaphorical mapping, or simply an instantiation of the image-schematic con-
ceptual structure in two different cognitive domains? (This latter view has been
propounded by Glucksberg [2001] and Jackendoff and Aaron [1991:328–30].)
Lakoff and Johnson present two counterarguments against these criticisms. First,
they argue that, although target domains of metaphors are structured, they are not
fully so: ‘they are not clearly enough delineated in their own terms to satisfy
the purposes of our day-to-day functioning’(Lakoff and Johnson 1980:118). Thus,
target domains lack at least some (image-schematic) structure. Lakoff and Johnson
argue that the linking of otherwise independent conceptual domains by metaphor
in fact creates similarity:
. . . the ideas are food metaphor establishes similarities between ideas and
food. Both can be digested, swallowed, devoured, and warmed over, and both
can nourish you. These similarities do not exist independently of the metaphor.
The concept of swallowing food is independent of the metaphor, but the concept
of swallowing ideas arises only by virtue of the metaphor. (Lakoff and Johnson
1980:147–48)
The second counterargument that Lakoff and Johnson present for why concep-
tual metaphors exist is that there is an asymmetry between source domain and target
domain. For example, love is expressed in terms of journeys, but journeys are not
expressed in terms of love (Lakoff and Johnson 1980:108). If image-schematic
structure were simply a highly schematic concept subsuming the corresponding
concepts in the source and target domains, then one would expect metaphorical
mappings to go in either direction; but they do not. Even when it appears that
there is a bidirectional metaphorical mapping, Lakoff and Turner argue that the
two mappings are different:
We can have cases like people are machines , as in
At the violet hour, when the eyes and back

Turn upward from the desk, when the human engine waits
Like a taxi throbbing, waiting (Eliot, The Waste Land)
and also the different metaphor m ac h i n e s a r e p e o p l e, as in when we say,

‘The computer is punishing me by wiping out my buffer.’ But these are two
different metaphors, because the mappings go in opposite directions, and different
things get mapped. In machines are people , the will and desire of a person
are attributed to machines, but in the people are machines metaphor, there
is no mention of will and desire. What is mapped instead is that machines have
parts that function in certain ways, such as idling steadily or accelerating, that they
may break down and need to be fixed, and so on. (Lakoff and Turner 1989:132,
emphasis original)
These two counterarguments are persuasive; but they also imply that the Invariance
Hypothesis and the target domain override captures only part of the nature of
metaphorical mappings, and perhaps not the most important part. The fact that
both people and machines have parts that function (or malfunction) is part of why
human engine works as a metaphor. But Eliot is certainly conveying more than
that with the metaphor, including perhaps the mechanization of twentieth-century
Metaphor 203
life and its dehumanizing effects (not to mention further images created by the
simile on the next line).
It is likely that a far richer structure than simply compatible image-schemas is
brought into the target domain from the source domain. It also suggests that Lakoff
and Johnson’s first counterargument – the target domain lacks (image-schematic)
structure that is added by the metaphorical mapping from the source domain –
makes too sharp a distinction between target domain structure and mapped source
domain structure. It implies a minimum of interaction between the target domain
structure (already there) and the source domain structure (filling in for the ab-
sence of target domain structure). Instead, many metaphor theorists argue for a
more interactive relationship between source and target domain structure, involv-
ing something like a ‘fusion’ or ‘superimposition’ of structure from both domains
(Jackendoff and Aaron 1991:334; they also cite Black’s [1979] ‘interaction’ and
Ricoeur’s [1978] ‘reverberation’). Jackendoff and Aaron suggest that the source
domain concepts are transformed as well in being metaphorically applied to the
target domain (ibid.). It is this intuition that blending theory attempts to capture
(see §8.3.3). This interactive relationship of course strengthens the first counter-
argument: the metaphor brings much more than extra image-schematic structure
to the target domain.
The final issue we wish to raise is the relationship between the conventional
metaphors that Lakoff centers his attention on and novel metaphor creation. Lakoff
and Johnson (1980:52–53) argue that some novel metaphors are extensions of
existing conventional metaphors, such as the song lyric We’re driving in the fast
lane on the freeway of love for love is a journey (Lakoff 1993:210). Lakoff
and Johnson also allow for completely novel metaphors, using as an example l ov e
is a collaborative work of art (Lakoff and Johnson 1980: 139–43).
They argue that a novel metaphor of this type can be a systematic mapping between
two conceptual domains.
Lakoff and Turner argue that other novel metaphors are more restricted; these
are image metaphors. An example they give is my wife . . . whose waist is an
hourglass ([3b] above; Lakoff and Turner 1989:90). Lakoff and Turner argue that in
image metaphors, specific and richly specified mental images are mapped, whereas
in conventional image-schematic metaphors, ‘there is no rich imagistic detail’
(Lakoff and Turner 1989:91). They also argue that image metaphors do not involve
the mapping of rich knowledge and inferential structure of conventional image-
schematic metaphors (ibid.).
For conventional image-schematic metaphors themselves, Lakoff and Johnson
argue that they ultimately originate in human bodily and cultural experience. For
example, conscious is up/unconscious is down, exemplified by wake
up and fall asleep, are based on the fact that ‘humans and most other mammals
sleep lying down and stand up when they awaken’ (Lakoff and Johnson 1980:15).
A cultural example is l a b o r i s a ( m at e r i a l ) r e s o u r c e, in that it can be
measured, assigned a value, used and so on. This metaphor arises from the fact
that we use material resources for various purposes; by virtue of those purposes
the resources have value; and the use of those resources requires labor (Lakoff and
Johnson 1980: 65–66).
It seems likely that many conventional metaphors – the kind found in everyday
language – have a basis in everyday human experience. However, many novel
metaphors do not, and some conventional metaphors do not either, except in terms
of very general image schemas such as those described by Grady (1997, 1998).
Also, where Lakoff & Johnson discuss truly novel image-schematic metaphors
such as love is a collaborative work of art , and Lakoff and Turner
discuss novel image metaphors, they do not describe how a speaker comes up with
the new metaphor: they only describe what the structure of the metaphor is. But
we have suggested that even conventional metaphor involves a blending of richer
structure than just image-schematic structure between source and target domains.
It seems plausible that even conventional metaphors draw on the full richness of our
encyclopedic knowledge of our bodily and cultural experience, especially when
they are first coined. Nor does there appear to be a difference in kind between the
‘rich detail’ mapped in novel image metaphors and the ‘rich knowledge’ mapped
in conventional image-schematic metaphors. If so, then there is only a difference
in degree between conventional metaphors and novel metaphors.
8.3 Novel metaphor
8.3.1 The life history of a metaphor

The Lakoffians make a virtue of concentrating on fully established and
conventionalized metaphors. However, it can be argued that, if one wants to get
to the heart of metaphor as an interpretive mechanism, one must look at freshly
coined examples. These are the only ones all of whose properties are currently
available for study: conventionalized metaphors have irrecoverably lost at least
some of their original properties. Complex literary metaphors are unsuitable for
initial study for different reasons: one must understand the simple before tackling
the complex. Easily comprehended fresh metaphors are abundantly available in
popular literature, the daily press, on TV and so on.
Several stages can be recognized in the life history of a durable metaphor. When
it is first coined, the only way to interpret it is to employ one’s innate metaphorical
interpretive strategy, which is subject to a wide range of contextual and com-
municative constraints. Once a metaphor takes hold in a speech community and
Metaphor 205
gets repeated sufficiently often, its character changes. First, its meaning becomes
circumscribed relative to the freshly coined metaphor, becoming more determi-
nate; second, it begins to be laid down as an item in the mental lexicon, so that in
time, it can be retrieved in the same way as a literal expression; third, it begins a
process of semantic drift, which can weaken or obscure its metaphorical origins.
At the beginning of its life, even if it is being laid down as an item in the lexicon,
speakers are very conscious of its status as a metaphor, and they can recreate easily
the metaphorical path of its derivation. As time passes, however, the sense of the
expression’s metaphorical nature fades and eventually disappears (although it can
be brought to life by means of Lakoffian elaborations etc.). Once that happens,
the expression is no different from a literal expression, and only etymologists and
historians of the language can recreate the path of derivation. At some point along
this path of change, the expression acquires a capability to act as a literal basis for
further metaphorical extensions, which is not possible for a fresh metaphor.
How does this impinge on the Lakoff account of metaphor? Let us look at some
relevant examples. Take the metaphor mentioned by Kövecses (2002:8), social
organizations are plants . Kövecses gives the following expressions (not
in the original order) that are said to exemplify this metaphor:
(22) They had to prune the workforce.

(23) Employers reaped enormous benefits from cheap foreign labor.
(24) He works for the local branch of the bank.
(25) There is a flourishing black market in software there.
Each of these metaphors is well established in the language, but they are at dif-
ferent stages in their life history. The use of prune in (22) still strongly evokes the
source domain of arboriculture, together with the therapeutic function of pruning,
that is, to remove unnecessary growth and increase vigor. This is therefore still in
its youth as a metaphor. Arguably, reap in (23) is at a later stage of assimilation into
the language: the evocation of harvesting is relatively weak; the expression reap
benefits also shows a degree of frozenness – one cannot reap anything other than
benefits – and by some definitions this would count as an idiom. (The deadness of
reaped can be felt if it is contrasted with harvested in the same context.) When we
come to (24), a normal speaker will probably not activate the PLANT domain at
all, although if the connection is pointed out they will assent to the connection. The
word branch has become the normal literal term for a local office, shop or other
premises forming part of a larger organization. In this sense branch has developed
a completely independent set of syntagmatic and paradigmatic relations that have
nothing to do with the source domain. For instance, branch office contrasts with
head office, rather than, say, trunk office; one opens a new branch (one does not
grow one) and one closes it down (one does not cut it off); there is typically a
branch manager. In (25), although the historical origin of flourish is undoubtedly

the French verb florir, it is hard to believe that any contemporary speaker habitu-
ally makes the connection (according to the Compact Oxford English Dictionary,
flourish came into English ca. 1300 with both literal and metaphorical meanings).
Most people probably think of the literal meaning of flourish as having to do with
businesses, or perhaps families, and may even feel, for example, a flourishing
garden to be an extension from this. (The Oxford Advanced Learners Dictionary
gives ‘be successful, very active or widespread; prosper’ as the first definition
of flourish, with No business can flourish in the present economic climate as an
example.)
The point of these examples is that one cannot expect to learn much about the
mechanism of metaphor by studying cases such as branch, flourish or even reap,
although prune might be more rewarding. Lakoff argues that it is precisely because
metaphors such as argument is war have become fixed in the language (and
because they are so widespread) that they are so significant, and reflect fundamental
properties of the human mind. This is undoubtedly a valid point, but it does not
reveal the basic mechanism of metaphor. All these established metaphors must
have started life as novel ones. At the very least, there is a separate and worthwhile
study to be done on novel metaphors, and that is what is being attempted here.
8.3.2 How do we recognize metaphors?

Lakoff is very much against the view that an essential property of a
metaphorical expression is deviance. Basically, Lakoff asks how something so
widespread and natural as metaphor can possibly be described as deviant. However,
there are two different ways of interpreting ‘deviance.’ The argument of those who
claim that anomalousness is a necessary feature of a metaphorical expression is
not that the use of metaphor is an unnatural or deviant practice, it is that if a literal
interpretation of an utterance is anomalous, that is normally a signal that we need to
apply a different interpretive strategy. The metaphorical interpretation that we end
up with is not deviant at all. A valid attack on the ‘anomalist’ position would have
to show that there are perfectly normal metaphors whose literal interpretations are
not anomalous. No such examples that bear close scrutiny have yet been proposed.
A common line to take is to quote examples of metaphors whose literal inter-
pretations are true. A famous example is that of Max Black (1979), who first puts
forward the metaphor:
(26) Man is a wolf.
He admits that this is literally untrue, and thus seems to support the anomalist
position. However, he then goes on to point out that (27) is true:
(27) Man is not a wolf.
Metaphor 207
However, this, too, can bear a metaphorical reading, as in (28):

(28) Man is not a wolf, he is a lion.
Does this undermine the anomalist case? Not really, because although Man is not
a wolf may be true, it is nonetheless, in the sort of context in which a metaphorical
reading would be normal, pragmatically or conversationally anomalous. Black
himself effectively concedes this point. It is not obvious how a metaphor can
function if there is nothing perceptibly odd whatsoever about its literal construal
(Eco 1996 gives a concise account of this argument).
8.3.3 Blending Theory and novel metaphors

Lakoff’s model does not capture what is perhaps the most characteristic
feature of metaphor: a metaphor involves not only the activation of two domains,
not only correspondences, but also a species of blending of two domains. This
blending becomes weakened, eventually to disappear altogether, as a metaphor
becomes established, but is a vital feature of a novel metaphor, whether it is totally
fresh, or is a revitalization of a conceptual metaphor by using original linguistic
means. A model of metaphor that takes this on board is the one presented in
Grady et al. 1999, which builds on the notion of blending presented in Fauconnier
and Turner 1996. (Fauconnier and Turner say very little on the topic of metaphor
as such.) Grady et al. (1999) claim that the blending model (henceforward BT,
for ‘Blending Theory’) is not a rival to Lakoff’s model (henceforward CMT, for
‘Conceptual Metaphor Theory’) but presupposes it.
Whereas CMT operates with two domains and correspondences between them,
BT works with four mental spaces (see §2.6). However, while CMT domains
are permanent structures, BT’s spaces are partial and temporary representational
structures constructed at the point of utterance: they are thus at least partially
responsive to contextual factors. However, there is a strong element of conventional
fixedness: as Fauconnier and Turner put it, ‘Dynamically, input spaces and blends
under construction recruit structure from more stable, elaborate, and conventional
conceptual structures . . .’ (1996:115).
Two of BT’s spaces are like CMT domains: they parallel the target and source
domains in CMT, except that they are more partial. They add to these, first, a
generic space, which represents what the target and source domains have in com-
mon; second, and most importantly, there is the blended space, where selected
conceptual material from source and target spaces is combined to form a new
structure. This is expressed by Grady et al. as follows:
In a metaphoric blend, prominent counterparts from the input spaces project to

a single element in the blended space – they are ‘fused’. A single element in the
blend corresponds to an element in each of the input spaces. A ship in the blend
[as in the ship of state] is linked to a ship in the source space and a nation in the
target, a surgeon is linked to both a surgeon and a butcher, and so forth. Intuitively
speaking, the point of metaphors is precisely that one thing is depicted or equated
with another. (Grady et al. 1999:114)
The two input spaces have different roles: the material in the target space functions
as topic, whereas the material in the source space ‘provides a means of re-framing
the first [i.e. the material in the target space] for some conceptual or communicative
purpose . . .’ (Grady et al. 1999:117).
However, the blended space does not only contain a selection of properties
drawn from the two input domains: it also contains new conceptual material that
arises from an elaboration of the conceptual blend on the basis of encyclopedic
knowledge. As an illustration of the model, consider the following metaphor:
(29) This surgeon is a butcher.
Grady et al. point out that this expression carries a strong implication that the
surgeon is incompetent, although incompetence is a normal feature neither of
surgeons nor butchers. The contents of the four mental spaces are described in
(30) (adapted from Grady et al. 1999):
(30) Generic space: Agent

Undergoer
Sharp instrument
Work space
Procedure: cutting flesh
Input space I (Target): Role: Agent: Surgeon
(X) (i.e. some individual)
Role: Undergoer: Patient
(Y) (a different individual)
Instrument: Scalpel
Work space: Operating theatre
Goal: Healing
Means: Surgery
Input space II (Source): Role: Agent: Butcher
Role: Undergoer: Dead animal
Instrument: Butcher’s knife etc.
Work space: Butcher’s shop
Goal: Producing edible portions
Means: Cutting flesh
Blended space: Role: Agent: Butcher
(X)
Role: Undergoer: Patient
(Y)
Work space: Operating theatre
Goal: Healing
Means: Butchery
Metaphor 209
According to Grady et al., the inference of incompetence arises through an

elaboration of the basic elements of the blended space, that is, we imaginatively
reconstruct a scene in which a butcher is in charge of an operation, and uses his
normal butcher’s techniques on the patient: there is a basic incompatibility between
the goal and the means, which leads to the inference of X’s incompetence.
The informal description of the metaphorical process given by Grady et al.
and quoted above is acceptable as far as it goes, and is not particularly new (cf.
I. A. Richards’ ‘interanimation of words’ [Richards 1936]). But it is difficult to
see how the mental spaces format throws any light on the process of blending.
Phenomenologically, a metaphor is highly distinctive, and this distinctiveness is
due in no small measure to the phenomenon of blending. Stern (1999) tries valiantly
to get to grips with this notion. He refers to ‘seeing as’: in X is Y we are made
to ‘see X as Y.’ He illuminatingly invokes the type of visual conceptual blending
that one sees in cartoons.
It seems clear that any account of metaphor that ignores this phenomenon is
seriously flawed. In this sense, BT is an advance on the original Lakoff model.
However, a full explanation of it is still lacking. It is not clear how the process
envisaged in BT gives rise to a blend in the sense of ‘seeing one thing as another.’
A simple combination of conventional features from one domain with conventional
features from another gives no account of the experience of a novel metaphor. One
important factor that is missing from the BT account is what is referred to in
§8.4 as the openness of mapping between the source and target domains: the
correspondences simply cannot be enumerated. This point is well illustrated by
example (1), repeated in (31):
(31) A myriad of ugly, dark thoughts clung to my reason and dug in with their
claws.
Perhaps the ‘seeing as’ account of blending is the most illuminating suggested so
far, except that it is not really explicit enough.
8.3.4 Context sensitivity

A characteristic of novel metaphors is that they involve domains that are
construed in context. Of course, such construals are subject to conventional con-
straints, but context also has an important role. Furthermore, the correspondences
between domains in a novel metaphor are also subject to construal, and in a sense
are created by the metaphor, rather than being preexisting. This aspect of novel
metaphor is not recognized in either CMT or BT. There is an element of context
sensitivity in the BT model, in that the features that enter into the input spaces
are constructed on-line. However, no account of how the features are selected is
offered.
Stern (1999) makes the lack of context sensitivity the major plank of his critique
of Lakoff. He argues that the aspects of a source domain that are relevant to a
target domain are heavily dependent not only on the domains themselves, but on
the whole context of the utterance, even of the discourse, and that any model that
depends on fixed structures in the mind is doomed to failure. He points out that
every context structures domains that are invoked in it in a characteristic way,
in terms of what is salient and what is backgrounded, patterns of inference, and
expressive or attitudinal factors. He illustrates his point by considering possible
metaphorical interpretations of the sun (with literal reference to the center of our
solar system). He begins with the following example uttered by Romeo in Romeo
and Juliet 2.2:
(32) But soft, what light through yonder window breaks?

It is the East, and Juliet is the sun.
He offers the following characterization: ‘[S]he is exemplary and peerless, wor-

thy of worship and adoration, one without whose nourishing attention another
cannot live, one who awakens those in her presence from their slumbering, who
brings light to darkness’ (Stern 1999:9). Stern contrasts this with Achilles is the
sun, where it ‘expresses Achilles’ devastating anger or brute force’; then with
Before Moses’ sun had set, the sun of Joshua had risen (The Talmud), ‘where it
expresses the uninterrupted continuity of righteousness, which, according to the
Talmud . . . preserves the world’; and finally with The works of great masters
are suns which rise and set around us. The time will come for every great work
that is now in the descendent to rise again (Wittgenstein 1980), ‘in which “sun”
expresses the cyclicity and eternal recurrence of greatness, that things once great
will be great again . . .’ (ibid.).
8.3.5 Asymmetry of vehicle and target

An important aspect of a metaphor is the different roles of the participating
domains in the blend. In Juliet is the sun we are talking about Juliet rather than
about the sun, and in the Patricia Cornwell example, we are talking about thoughts
rather than about creatures. At the same time, as a result of the blending process
we create entities that are both, for instance, thoughts and creatures. The notion
of ‘seeing as,’ although it is suggestive, might seem to allow a complete takeover
of the target domain by the source domain, which, of course, does not happen.
The puzzle is how to reconcile these two aspects of metaphor. It is stipulated in
the version of BT presented by Grady et al. that the ‘target’ input domain forms the
‘topic’ and the ‘source’ domain provides ‘framing.’ However, this addresses only
Metaphor 211
the first aspect. A satisfactory account of novel metaphor will have to be more
explicit on this matter.
8.4 Metaphor and simile
8.4.1 Two types of simile

Most writers distinguish two sorts of expression of comparison: Glucks-
berg (1999) speaks of ‘literal similes’ and ‘metaphorical similes.’ Here, the two
types will be referred to as similes proper (henceforward, simply similes) and
statements of similarity. A simple diagnostic test is whether they transform read-
ily into metaphors or not. For instance (33) transforms easily into (34) without a
great change in meaning (it will be argued in a moment that there is a difference,
but here it is not great):
(33) John is like a lion.

(34) John is a lion.
On the other hand, (35a) and (36a) are statements of similarity, and do not
correspond to metaphors:
(35) a. My house is like yours.

b. *My house is yours.
(36) a. Nectarines are like peaches.
b. *Nectarines are peaches.
There is a different but related distinction between true similes and what might
be called speculations. Both may have the same form. Take the example of He
sounds like someone with a severe cold. Said of a person at the other end of a
telephone line, this could be a speculation that they indeed are suffering from a
severe cold. On the other hand, if it is said of someone known by speaker and
hearer to be perfectly healthy, it is a simile.
8.4.2 Theories of the relation between simile and metaphor

8.4.2.1 Metaphors are implicit similes
This is what Glucksberg (2001) calls the ‘classical’ view. To under-
stand a metaphor (which is prototypically false on a literal interpretation) we
first transform it into the corresponding simile, which, if the metaphor is a valid
one, will prototypically be true. On this view, the simile gives a more direct pic-
ture of the semantic structure of the expression; the metaphor is to be seen as
a kind of shorthand. Otherwise there is no substantive difference between the

two.
8.4.2.2 Similes are implicit metaphors

Currently, this is the favorite position, held, among others, by Stern (2000)
and Glucksberg (and attributed by him even to Aristotle). Glucksberg’s position
is as follows. We must first introduce Glucksberg’s notion of ‘dual reference.’ Put
simply, it is that all statements of the form X is Y are to be interpreted as class-
inclusion statements, as in obvious cases such as Dogs are animals. A metaphor,
on this view, is no different: it is a class-inclusion statement. John is a lion means
that John is a member of the class of lions. However, words are systematically
ambiguous, in that they can refer to a category (i.e. literally) or to a supercategory
for which the literal category is a prototypical example. Since the lion is conven-
tionally held to be representative, or emblematic, of the category of strong, coura-
geous beings, then John is a lion simply means that John is a strong courageous
being.
Every metaphor thus depends on our knowledge of what specific things are
good examples of, and this more abstract category, or rather its defining features,
tells us what we are to attribute to the subject. Every time we retrieve a word
from the mental lexicon, we have equal access to the narrow meaning and the
broad meaning (of course, an item may be an excellent example of more than one
category): choosing the metaphorical reading over the literal one is no different
in principle than choosing one reading of a (lexicographically) ambiguous word.
Let us designate the literal meaning of a term as X and its supercategory X . A
metaphor then has the general form A is X , while a simile has the general form
A is like X. A simile is interpreted, on this view, by translating it into a metaphor,
that is, interpreting X as X , reconstructing the supercategory, and applying its
defining features to A.
8.4.2.3 Similes and metaphors are distinct

The position to be adopted here is that metaphors and similes are pro-
totypically distinct, even though both involve two distinct domains. However, the
difference in meaning between a metaphor and its corresponding simile (in cases
where there is an equivalent) varies according to context and according to type of
both simile and metaphor.
The first and most obvious difference is in propositional structure. An expression
of the form A is like B asserts that there is a resemblance between A and B in some
respect. An expression of the form A is B, on the other hand, predicates certain
properties directly of A. In Langackerian terms, A is like B profiles the resemblance,
Metaphor 213
while A is B profiles the properties predicated. This would be a difference even if

there were no other. However, there are other differences.
There are two major differences between similes and metaphors. The first in-
volves the scope of the correspondences between the two domains. Most of the
discussion one encounters in the literature on the relation between metaphor and
simile centers around examples of simile that are not prototypical. In fact, exam-
ples of simile of the form X is like Y are comparatively rare: in the vast majority
of similes, there is a specification of the respect in which the resemblance holds,
without which a proper interpretation is not possible. This feature will be called
restricted mapping between the two domains. The following examples, taken
from Patricia Cornwell’s novel Black Notice (2000), will illustrate the point (the
restrictions are in italic):
(37) Marino was breathing hard like a wounded bear.

(38) [The victim had been] shot and dumped somewhere like garbage.
(39) Marino’s voice was soft and muted like bourbon on the rocks.
(40) And I know Anderson follows her around like a puppy.
(41) . . . strong gusts of wind pushed me like a hand.
In contrast, in prototypical metaphors, the correspondences between the domains

do not form a closed set, and cannot be exhaustively listed. This property will be
referred to as open mapping between the two domains.
(42) His sarcasm could have shred paper.

(43) . . . his words grabbed her by the collar.
(44) The mention of Lucy’s name squeezed my heart with a hard, cold hand.
(45) A myriad of ugly, dark thoughts clung to my reason and dug in with their claws.
(46) Bray’s self-confidence slipped just enough to unmask the evil coiling within.
The second major difference is that in a prototypical metaphor, what is presented

is a blend of two domains, whereas in a simile, the two domains are presented as
separate. In a simile, we are certainly invited to consider the two domains together,
but they are presented as distinct.
(47) Icicles bared long teeth from the eaves.

(cf. Icicles were like long teeth being bared.)
(48) The elevator has a mind of its own.
(cf. The elevator behaves like a being with a mind of its own.)
(49) Her eyes were dark holes.
(cf. Her eyes were like dark holes.)
(50) A headache began boxing with my brain.
(cf.?? My headache felt as if someone were boxing with my brain.)
However, in addition to prototypical metaphors and prototypical similes, there

are less prototypical examples of each. For instance, a characteristic of similes of
the form A is like B is that they display open mapping (at least where the restriction
is not latent). This makes them more ‘metaphor-like’ than the prototypical variety
(although it does not make them into metaphors). But even some similes of the
canonical form seem to have open mapping:
(51) She was gone in a flash of red, like a vengeful queen on her way to order armies
to march in on us.
(52) Cameras were already on her like a storm of hurled spears.
(53) It’s like I’m on the wrong planet.
(54) . . . now it’s like he’s a hunted animal with no place to go?
(55) I felt as if I were inside cut crystal.
(56) It was as if a wild animal had dragged her dying body off to its lair and mauled
it.
Similarly, there are apparent metaphors, where the mapping is highly restricted –
the motivation for the transfer can be reduced to one or a small number of features.
In such cases, one can say that they are more simile-like than are prototypical
metaphors. In fact, in certain cases, there appears to be no blending, either, which
would make them almost indistinguishable from similes.
(57) . . . her breath smoking out (= came out like smoke, because it was a cold day,
and it condensed)
(58) Grass was a thick, stiff carpet (because it was frozen) (change from metaphor to
simile makes very little difference)
(59) A computer mouse (any blending?)
(60) A splinter of light glinted in the dark. (appearance only?)
There are many cases where there is a striking difference in interpretation be-
tween a metaphor and its corresponding simile:
(61) . . . containers lined up at loading docks like animals feeding from troughs.
(cf. The containers were animals feeding from troughs.)
Notice how this opens up the mapping possibilities: there is no obvious way of
building in the mapping restrictions within the format of the metaphor. In (62), the
meaning is changed completely:
(62) Broken wooden packing cases littered the beach like the debris of a disordered
mind. (J. G. Ballard: Cocaine Nights)
(cf. Broken wooden packing cases littered the beach, the debris of a disordered
mind.)
An interesting (and so far unanswered) question concerns what these have in

common that distinguishes them from cases where the difference is slight.
Metaphor 215
In summary, we can recognize four types of expression:
(63) Mapping Blending?

Prototypical metaphors open yes
Simile metaphors restricted yes
Prototypical similes restricted no
Metaphorical similes open no
8.4.3 Metaphor-simile combinations

Metaphor and simile very frequently combine, and the combination seems
to have a unique function, that is, one cannot obtain the same effect in any other way.
There are two major modes: metaphor within simile and simile within metaphor.
(a) Metaphor within simile

What usually happens in these is that the simile serves to clarify the source
domain, often because the key word in the metaphorical vehicle is one with a wide
range of construals:
(64) Bizarre, angry thoughts flew through my mind like a thousand starlings.
Here, the phrase like a thousand starlings gives a precise picture of the sort of
flying that is to serve as the vehicle of the metaphor.
(65) She was standing there, her eyes fastened to me like steel rivets.
Here, like steel rivets specifies the quality of fastened: rivets provide a particularly
strong and rigid type of fastening.
(66) Grief tumbled out of her like a waterfall.

(67) This is really twisting my brain like a dishrag.
There are many different sorts of ‘twisting’ and ‘tumbling,’ but the similes function
to narrow them down.
(b) Simile within metaphor

These cases are in some ways more complex: in them, the second term
of the simile is itself a metaphor. There are two varieties of this. In the first, the
simile contains a complete metaphor, with topic and vehicle:
(68) He looked tired, as if life had pushed him too far.

(69) Rose looked dejected and somewhat embarrassed, as if afraid that her being so
upset had sent her spinning threads of truth into tapestries of conviction.
In the second type, the second term of the simile appears completely literal in
itself, but is only metaphorical when the comparison is made with the first term:
(70) Talley made love as if he were starving.

(71) Bray’s tone had the effect of a metal box slamming shut.
8.5 Metaphor and metonymy
Just as an examination of the relations between metaphor and simile

throws light on the nature of both, a consideration of the relations between metaphor
and metonymy is similarly illuminating. This is in two ways: first, a comparison
of the two highlights the special nature of each, and second, they can be shown
to interact in significant ways. However, since the focus of this chapter is on
metaphor, no attempt will be made to explore the further reaches of metonymy. (A
fuller account can be found in Kövecses and Radden 1998.)
8.5.1 Characterizing metonymy

The term ‘metonymy’ is sometimes interpreted very broadly, to include,
for instance, the relation between form and meaning within a sign, the relation
between a linguistic sign and its referent, the relation between, for instance, an
acronym and its full form, and the special relation between a prototype and the
category it represents. For present purposes, we shall construe metonymy more
narrowly. First of all, we shall say that metonymy involves the use of an expression
E with a default construal A to evoke a distinct construal B, where the connection
between B and A is inferable by general principles (i.e. is not a private prearranged
code between individuals). In novel uses, there is normally an intuitive violation
of conventional constraints.
This characterization covers metaphor as well as metonymy. Metonymy is dis-
tinguished by the fact that (i) A and B are associated in some domain or domain
matrix,1 (ii) any correspondences (in the Lakoffian sense) between A and B are
coincidental and not relevant to the message and (iii) there is no blending between
A and B. In chapter 5 we treated cases where A and B are unified separately,
as facets; however, whether facets are considered distinct from metonymy, or a
special variety of metonymy, has no serious theoretical consequences.
1 While identity of domain does seem to be a factor, we agree with Feyaerts (2000) and Riemer (2001)
that on its own it remains an unreliable criterion in the absence of independent means of delimiting
domains.
Metaphor 217
This narrower characterization still covers a wide range of types in terms of

the sorts of association that are operative. No attempt will be made to provide an
exhaustive list (it is not clear that this is possible, even in principle); the following
is to be taken merely as illustrative. We can divide the associations that support
metonymy roughly into ‘intrinsic associations,’ which are either inherent, or at
least relatively permanent, and ‘extrinsic’ associations, where A and B are asso-
ciated contingently and non-inherently. The following are examples of intrinsic
associations:
(a) Part-whole
Part for whole: I noticed several new faces tonight.
Whole for part: Do you need to use the bathroom?
(b) Individual-class
Individual for class: He’s no Heifetz.
Class for individual: Postman, this letter is covered in mud!
(c) Entity-attribute
Entity for attribute: Shares took a tumble yesterday.
Attribute for entity: He’s a size ten.
(d) Different values on same scale
Hyperbole: It’s practically absolute zero in here – shut the window!
Understatement: I’m feeling a bit peckish – I haven’t eaten for three days.
(e) Opposites
Irony: Now let’s move on to the small matter of the £30,000 you owe us.
Extrinsic associations in metonymy are exemplified by (72)–(76):

(72) Room 23 is not answering.
(73) I’m parked out back.
(74) The french fries in the corner is getting impatient.
(75) Sperber and Wilson is on the top shelf.
(76) England are all out for 156 runs.
8.5.2 Metaphor–metonymy relations

In this section we look at relations between metaphor and metonymy, in
particular, the question of whether they can be sharply distinguished from one an-
other. According to Radden (2000), prototypical cases of metaphor and metonymy
are situated at opposite ends of a continuous scale, with no clear dividing line be-
tween them. This would make the distinction analogous to the traditional one
between homonymy (no motivated relation between senses) and polysemy (mo-
tivated relation). Before considering cases that might support this position, we
consider cases where there is interaction between metonymy and metaphor, but
where their distinct identities are not compromised, and that arguably do not there-
fore support Radden’s position.
Although the Lakoffians distinguish between metaphor and metonymy, they

nonetheless emphasize that metonymy can play a vital role in the genesis of
metaphorical expressions (§8.2.2). Take the a n g e r i s h e at metaphor. At the
heart of this is a metonymy: an angry person subjectively feels hot, so one can
refer indirectly to anger by way of mentioning heat. At this level, one could say
that there are no correspondences, only the holistic one. The typical Lakoffian
correspondences arise only when the basic metonymy is elaborated, for instance,
in the idea of anger pictured as a liquid in a closed container undergoing a heating
process, or as a fire. Another, more basic, example is the more is up metaphor.
This, too, originates in a metonymy. If we take a pile of sand, say, and add more
sand, the top of the pile rises, or if we add more books to a pile of books. Thus
there is a real-life, literal correlation between ‘more’ and ‘greater height,’ which
justifies us in saying The pile is higher now, meaning that there are more books.
This metonymy can then be metaphorically extended to any case of ‘more,’ such
as higher prices, higher temperature and so on. This kind of cooperation between
metonymy and metaphor does not make them any less distinct.
A different type of case concerns what Goossens (1990) calls, somewhat in-
elegantly, metaphtonymy. This is when both metaphorical and metonymic pro-
cesses are recruited in the construal of an interpretation. Different types can be
distinguished. In one type, the elements that undergo metaphoric and metonymic
transfer, respectively, are different. One example of this is the following, from
Goossens:
(77) She caught the minister’s ear and persuaded him to accept her plan.
A plausible account of this is that we construe ear metonymically for ‘attention,’

which forces a metaphorical construal of caught; catch X’s attention is interpreted
as ‘make X attend.’ Another similar example (from Patricia Cornwell) is (68):
(78) He stopped on the sidewalk and looked into my eyes as people flowed around us
and light from shops unevenly shoved back the night.
In this case, night is first interpreted metonymically as ‘area of darkness,’ then

shoved back is metaphorically interpreted as ‘illuminated.’
In another variety of metaphtonymy, it is the same expression that undergoes
successive metaphorical and metonymic construal. Take the case of (69):
(79) My lips are sealed.
A literal interpretation of this can be metonymically understood to indicate that

the speaker is physically unable to speak. This metonymy can then be metaphori-
cally extended to a situation where the speaker is non-physically constrained. The
Metaphor 219
metonymic construal of the expression thus precedes a metaphorical construal of

the same expression. Another example is (80) (from Patricia Cornwell):
(80) Anger slipped out of hiding.
The context makes clear that this refers to someone who has been trying to conceal
her anger, but has lost control, and has allowed it to be perceptible. A possible
interpretation of this is that slipped out of hiding is first metonymically interpreted
as ‘become visible,’ which is then metaphorically extended to apply to someone’s
anger, that is, ‘become perceptible (not necessarily visually).’ In all these cases,
metaphor and metonymy, although both present, can be seen to make separate
contributions.
8.5.3 Types of indeterminacy

In the above cases, although both metaphor and metonymy can be shown
to be operative, we can nonetheless easily separate the effects of the two processes
and their distinctness is not compromised. But there are cases where there is
arguably a genuine indeterminacy between metaphor and metonymy.
The first type may be labeled ‘etymological indeterminacy.’ These cases arise
because a now conventionalized extended meaning could have been reached by
either route. Claimed examples of this phenomenon are head of the bed and back
of the chair. What is the motivation for these terms? Is it because a person’s head
normally rests near that part of the bed, or a person’s back rests on that part of a
chair? Or is it because of some resemblance between a bed and a supine person,
or between a chair and a standing or sitting person? This is a question of historical
fact, and is probably unresolvable. It is also possible that different speakers have
different conceptions of the relationship between, say, a human head and the head
of a bed. However, this is unlikely to lead to any observable difference in synchronic
usage.
A more immediate type of indeterminacy can be observed in certain expressions.
Consider, first, (81):
(81) The car stopped in front of the bakery.
Of course it is possible that the car had been left with the brakes off, and had rolled
down an incline on its own, coming to a stop in front of the bakery. This would be a
fully literal construal of (81). But in the normal course of events, we would interpret
this as referring to a car that was being driven, and that the agent of the stopping
action was the driver. This could then be taken as an example of metonymy – the
car is used to refer indirectly to ‘the driver of the car’ (whether this would be a
case of noun transfer or verb transfer is not relevant here). But there is a third
possible interpretation, which is that we are metaphorically attributing animacy

to the car. The question then arises of whether the metonymic and metaphorical
interpretations can really be separated.
The question is perhaps thrown into greater relief in (82):
(82) A yellow Porsche drew up in front of the bakery.
The action of drawing up is not something the car can do on its own, so any
interpretation must involve the driver. But (82) undoubtedly describes the motion
of the car and the driver may not have been visible to the speaker. In this case,
perhaps because of the specificity of the description of the car, it is easier to see
this as a humanization of the car. But a pure metaphor account would not give full
credit to the role of the driver.
A more revealing explication of (82) might be that a yellow Porsche denotes
a single entity that represents a kind of fusion of car and driver (notice that this
is different from the ‘unity’ observed with facets). If this were true, it would be
neither pure metonymy nor pure metaphor, nor would it be part metonymy, part
metaphor – it would be something intermediate. Another example is (83):
(83) Britain declares war on Iraq.
It is very hard to specify exactly what Britain refers to on a pure metonymic

construal (and on a facet analysis, it would be hard to pinpoint the facet involved).
It does not seem to be ‘the government,’ because, although (84) is possible (in
British English), (85) is not:
(84) The government have decided to restrict immigration.
(85) ?Britain declare war on Iraq. (cf. England win the World Cup.)
Once again, we seem to have a fusion – this time of country, government, final
decision-taker, monarch (perhaps) and so on, forming a single, semi-animate agent,
by a process that is neither pure metonymy nor pure metaphor. In one sense these
examples would seem to support Radden’s contention in the sense that expressions
can be placed on a scale of metaphoricity-metonymicity. In another sense, however,
the distinction between metaphor and metonymy as processes arguably remains
intact.2
8.6 Conclusion
Much work remains to be done on metaphor and its relationship with

metonymy and simile. Some facts about metaphor are well established. Metaphor
2 Riemer (2001) comes to a similar conclusion, but by a different route.

Metaphor 221
essentially involves the use of an expression to elicit a construal whose content is

the result of an interaction between two construed domains. One of these domains
is a construal on the basis of the expression’s conventionally associated purport; the
other domain is construed on the basis of an alien region of purport. The interaction
between the domains is a species of blending, whereby one domain, the target, is
modified under the influence of the other domain, the source. The result is a unique
semantic confection, unobtainable by any other means.
This summary conceals a multitude of mysteries: for instance, the role of contex-
tual and other constraints and the mechanism by which they produce their effects;
but most notably of all, the exact nature of the blending process, and the nature
of the resultant blend. Most metaphors studied by cognitive linguists have been
relatively simple. A further challenge is presented by more complex metaphors.
Even a small increase in complexity is daunting. The reader is invited to unravel the
interplay between metaphors and metonymies in one last example from Patricia
Cornwell:
(86) The temperature in my house slowly dropped, hours slipping deeper into the still
morning.
PA RT I I I
Cognitive approaches to
grammatical form
9
From idioms to construction grammar
9.1 Introduction
The cognitive linguistic approach to syntax goes under the name of con-
struction grammar. It is not an exaggeration to say that construction grammar
grew out of a concern to find a place for idiomatic expressions in the speaker’s
knowledge of a grammar of their language. The study of idioms led to calls for a
rethinking of syntactic representation for many years before construction grammar
emerged, and some of this work will be referred to in this chapter. At least partly
independently of construction grammar, a number of researchers have emphasized
the need to represent linguistic knowledge in a construction-like fashion. But in
cognitive linguistics, these concerns led to a grammatical framework in which all
grammatical knowledge is represented in essentially the same way. This chapter
presents the arguments for a construction grammar.
Construction grammar, like any other scientific theory, did not arise in a theoreti-
cal vacuum. Construction grammar arose as a response to the model of grammatical
knowledge proposed by the various versions of generative grammar over the period
from the 1960s to at least the 1980s, and other syntactic theories that emerged as
direct offshoots of generative grammar. (These models in turn represented exten-
sions of the organization of a traditional descriptive grammar of a language, albeit
with significant changes in terminology.)
In most theories of generative grammar, a speaker’s grammatical knowledge is
organized into components. Each component describes one dimension of the prop-
erties of a sentence. The phonological component, for example, consists of the rules
and constraints governing the sound structure of a sentence of the language. The
syntactic component consists of the rules and constraints governing the syntax –
the combinations of words – of a sentence. The semantic component consists of
rules and constraints governing the meaning of a sentence. In other words, each
component separates out the specific type of linguistic information that is contained
in a sentence: phonological, syntactic and semantic. In addition, all versions of
Chomskyan generative grammar have broken down the syntactic component fur-
ther, as levels or strata (such as ‘deep structure,’ later ‘D-structure,’ and ‘surface
225
226 Cognitive approaches to grammatical form
structure,’ later ‘S-structure’; Chomsky 1981) and modules or theories (such as

Case theory, Binding theory etc.; Chomsky 1981).
Further components have been proposed by other linguists. Some have argued
that morphology, the internal formal structure of words, should occupy its own
component (e.g. Aronoff 1993). Others have suggested that information structure,
that is, certain aspects of discourse or pragmatic knowledge, should have its own
component (Vallduvı́ 1992). However many components are proposed, the general
principle remains: each component governs linguistic properties of a single type:
sound, word structure, syntax, meaning, use.
From our point of view, the number of different components is not as crucial as
the fact that each type of linguistic knowledge is separated into its own component.
We may describe this as a ‘horizontal’ model of the organization of grammatical
knowledge, following its typical diagrammatic representation:
(1) phonological component
syntactic component
semantic component
In addition to these components, there is the lexicon, which characterizes the

basic units of syntactic combination. The lexicon differs from these components
in that the lexicon gives, for each word, its sound structure, its syntactic category
(which determines how it behaves with respect to the rules of the syntactic com-
ponent) and its meaning. Thus, a lexical item combines information from the three
components in (1) (and can include information from other components, such as its
morphological structure and its stylistic pragmatic value). It represents a ‘vertical’
component as against the ‘horizontal’ components:
(2) phonological component

lexicon
syntactic component
semantic component
The components are intended to be highly general rules that apply to all struc-
tures of the relevant type. Thus the rules of the phonological component apply
to all word forms and all phonological phrases (for prosodic and other phrasal
From idioms to construction grammar 227
phonology); the rules of the syntactic component apply to all sentences and
sentence types; and the same applies to rules for other components.
Of course, there must be some general way to map information from one compo-
nent onto another; for instance, there must be a way to map the syntactic structure
of a sentence onto the semantic structure of the meaning conveyed by the sentence.
These rules are called linking rules, and are also intended to be highly general,
applying to all sentences of the language. One might ask at this point, why are
the linking rules just a bunch of rules that link components, while the components
define the way that grammatical knowledge is divided up in the speaker’s mind?
As we will see, that is essentially the question that construction grammar asks.
The response of the generative grammarians is that the rules inside each compo-
nent are so highly intertwined and self-contained that they represent a cohesive
structure relative to the linking rules (and if they are not so highly intertwined, the
components are broken down further into levels, modules etc.).
In sum, the final model of the organization of grammatical knowledge in the sorts
of syntactic theories prevalent from the 1960s to the 1980s will look something
like the diagram in (3):
(3)
phonological component
linking rules
lexicon
syntactic component
linking rules
semantic component
One of the crucial characteristics of this model is that there are no idiosyncratic
properties of grammatical structures larger than a single word. Phrases and sen-
tences are governed by the highly general rules of the syntactic component and
their counterparts in the semantic and phonological components, and the equally
highly general linking rules. On the other hand, words represent an arbitrary and
idiosyncratic joining of form (phonological and syntactic) and meaning. The re-
striction of arbitrariness in grammar to the lexicon is a central principle of gener-
ative grammar, reiterated in recent versions of generative grammar (e.g. Chomsky
1993:3, 4).
One of the consequences of this model is the rejection of the concept of con-
struction in the traditional grammar sense of that word. In traditional grammar,
one describes a syntactic structure such as is found in the sentence in (4a) as ‘the
passive construction’:
(4) a. Janet was promoted by the company.

b. [Subject be Verb-PastParticiple by Oblique]
The passive construction would be described as necessarily possessing the com-

bination of syntactic elements given in (4b), including the subject noun phrase,
the passive auxiliary verb be in some form, a verb in the past participle form, and
(optionally in the case of English) a prepositional phrase with the preposition by. In
addition, a traditional characterization of the passive construction would indicate
that the agent of the action is expressed by the object of the prepositional phrase,
and the undergoer is expressed by the subject.
In the generative model, as many of these properties of the passive construction
as possible would be described by the general rules of the various components, and
any idiosyncratic properties would be placed in the lexicon. For example, the fact
that the subject precedes the verb is true of a large class of constructions in English
(see [5]); the fact that the auxiliary follows and is in a finite form in contrast to
the verb is also true of a large class of constructions (see [6]), and the fact that
the prepositional phrase follows the verb (and also a noun) it modifies, and the
preposition governs the object form of the noun phrase, is also true of a large class
of constructions (see [7]):
(5) a. Active: John ate.
b. Relative Clause: the tart that John ate . . .
c. Adverbial Clause: before John ate the tart . . .
d. Conditional: If John ate a tart, then I will have a tart as well.
e. Comparative: John ate a bigger tart than I did.
(6) a. Perfect: John has eaten the tart.
b. Progressive: John is eating a tart.
c. Future: John will eat [infinitive] a tart.
d. Modal: John might eat [infinitive] a tart.
(7) a. Oblique Adjunct Phrase: John ate the tart with a fork and spoon.
b. Prepositional Complement Phrase: John put the tart in the refrigerator.
c. Circumstantial Phrase: John ate the tart in the living room.
d. Nominal Prepositional Phrase Modifier: the tart on the table . . .
The logical conclusion of this process of analysis is the hypothesis that all
properties of syntactic constructions – that is, a grammatical structure larger than
just a single word – can be captured with the general rules of the grammatical
components and their interfaces, and thus there is no need for constructions in
grammatical analysis. Chomsky makes this claim explicit:
a central element in the work discussed here, as in recent work from which it
evolves, is the effort to decompose such processes as “passive”, “relativization”,
etc., into more fundamental “abstract features”. (Chomsky 1981:121)
UG provides a fixed system of principles and a finite array of finitely valued
parameters. The language-particular rules reduce to choice of values for these
parameters. The notion of grammatical construction is eliminated, and with it,
construction-particular rules. (Chomsky 1993:4)
Chomsky’s position on the generality of syntax and the irrelevance of construc-

tions to the analysis of grammar is the complement of his view that all arbitrary
and idiosyncratic aspects of grammar should be restricted to the lexicon.
The componential model of grammatical organization given above is charac-
teristic of the generative grammar and its offshoots. As we noted, in Chomskyan
theory the syntactic component is internally complex, beginning with levels in the
earliest versions and a further subdivision into modules in Government and Bind-
ing theory (Chomsky 1981). The most recent version, Minimalist theory (Chomsky
1993, 1995), apparently ends the internal organization of the syntactic component
and recasts the phonological component as an ‘articulatory-perceptual interface,’
which links the language faculty to the perceptual-motor system and the seman-
tic component as a ‘conceptual-intentional interface,’ which links the language
faculty to other human conceptual activity. Nevertheless, it appears that the broad
division into the three components in (3) remains, even if two components are
now thought of in terms of their embedding in the cognitive system as a whole,
and the third (syntactic) component is now the whole language system in between.
Likewise, the notion of the lexicon as the repository of idiosyncratic information
remains, and as such provides information linking the three components together
(Chomsky 1993:3; 1995:235–36).
Other syntactic theories that diverge from Chomskyan theory also retain the
organization into components. Earlier offshoots such as Relational Grammar
(Perlmutter 1983) and its offshoot Arc-Pair Grammar (Johnson and Postal 1980)
included multiple levels in the syntactic component but retained the separation
of components. Later offshoots, such as Lexical-Functional Grammar (Bresnan
1982), Generalized Phrase Structure Grammar (Gazdar, Klein, Pullum and Sag
1985; see also Pollard and Sag 1993) and Categorial Grammar (Wood 1993), re-
ject the concept of multiple levels but still retain the separation of components. In
more recent theories of non-Chomskyan syntax, there is a shift in emphasis from
separation of components to their interaction. But the development of construction
grammar marks a more direct break from the componential view of grammatical
organization.
9.2 The problem of idioms
Certain exceptions to the principle of the generality of rules governing

larger grammatical structures have been made in the history of generative grammar.
For example, the different syntactic structures required by different verbs, such
as those illustrated in examples (8)–(9), must be represented somewhere in the
grammatical model in (3):
(8) a. Tina slept.

b. *Tina slept bananas.
(9) a. David consumed the bananas.
b. *David consumed.
Until at least recently (e.g. Haegeman 1994:40–42), generative grammar has

accounted for this pattern of distribution by subcategorization frames associated
with lexical items in the lexicon, as in (10a–b):
(10) a. sleep: V, [ – ]
b. consume: V, [ – NP]
The effect of this device is to include phrasal syntactic information in the lexicon,
under the lexical entry for each verb. However, this way of handling the distribu-
tional patterns in (8)–(9) is consistent with the general principle that idiosyncratic
information is to be found in the lexicon.
There is another class of syntactic phenomena that poses a much greater problem
for the componential model of grammar and the principle that all grammar above
the word level can be explained by highly general rules. These are idioms. Idioms
are, by definition, grammatical units larger than a word which are idiosyncratic in
some respect. Some examples of idioms are given in (11):
(11) a. It takes one to know one.
b. pull a fast one
c. bring down the house
d. wide awake
e. sight unseen
f. all of a sudden
g. (X) blow X’s nose
h. Once upon a time . . .
It is difficult to give a precise definition of the category of idioms, for reasons

that will soon become clear. Nunberg, Sag and Wasow (1994:492–93) offer a
prototype definition of idioms with one necessary feature and a number of typical
features. The necessary feature is conventionality: ‘their meaning or use can’t
be predicted, or at least entirely predicted, on the basis of a knowledge of the
independent conventions that determine the use of their constituents when they
appear in isolation from one another’ (492). The other, typical, properties of idioms
they list are:
(12) a. Inflexibility: restricted syntax, as in shoot the breeze vs. *the breeze is hard to
shoot
b. Figuration: figurative meaning, as in take the bull by the horns, lend a hand
c. Proverbiality: description of social activity compared to a concrete activity, as
in climb the wall, chew the fat, spill the beans
d. Informality: typically associated with informal speech styles or registers

e. Affect: usually have an evaluation or affective stance towards what they de-
scribe
It is the necessary property of idioms that Nunberg et al. identify – their conven-
tionality – which is the relevant property of idioms with respect to the componential
model of grammar. If expressions such as those listed in (11a–h) are conventional,
then they must somehow be stored as such in a speaker’s mind. If so, then idioms
are part of a speaker’s grammatical knowledge. However, at least some aspects of
an idiom cannot be predicted by the general rules of the syntactic and semantic
components and their linking rules (we will leave out the phonological component
for the time being). Hence they pose a problem for the componential model. It
is possible to make certain sorts of stipulations to handle the conventionality of
idioms (cf. Nunberg et al. 1994:507). However, a more general treatment would
be preferable to such stipulations, if such a general treatment were available.
The linguists who ended up proposing the original construction grammar (in
Fillmore, Kay and O’Connor 1988) approached the problem of idioms from the
opposite direction. Instead of treating idioms as a problematic phenomenon from
the point of view of the componential model of grammar, they analyzed the wide
variety of idioms, and their analysis became the basis for a new model of grammat-
ical organization. The remainder of this chapter will take the reader from idioms
to construction grammar.1
Idioms can be characterized in many different ways. The description and clas-
sification that we will begin with is drawn from Fillmore et al. 1988, who used
their analysis to argue for a construction grammar. Fillmore et al. begin with three
features that can be used to classify idioms. The first feature they describe, drawn
from Makkai 1972, is the distinction between encoding and decoding idioms.
An encoding idiom is one that is interpretable by the standard rules for in-
terpreting sentences, but is arbitrary (i.e. conventional) for this expression with
this meaning. Examples given by Fillmore et al. are answer the door, wide awake
and bright red. These are all expressions that a hearer could figure out upon hear-
ing them. However, a speaker would not have guessed these expressions are the
natural-sounding English way to describe ‘open the door in response to some-
one knocking,’ ‘completely awake’ and ‘intense color.’ Another way of looking at
1 Fillmore et al. were not the first linguists to analyze idioms in a systematic way. There is a vast
literature on idioms, particularly in Europe where the study of idioms is called ‘phraseology.’ Another
important line of research on idiomatic expressions is the Firthian research on collocations. Nor were
Fillmore et al. the first to perceive the problem of idioms for a componential model of grammar, and
to propose an alternative; one such antecedent is Becker (1975). However, as in the case of many
scientific ideas, variants of the idea are proposed independently but commonly only one variant is
propagated through the scientific community (Hull 1988).
encoding idioms is from the point of view of someone learning a foreign language.
For instance, an English learner of Spanish would not be able to give the correct
way of asking ‘how old are you?’, if s/he did not know it already; but if s/he heard
a Spanish speaker say Cuantos años tiene? (lit. ‘how many years do you have?’),
s/he would very likely figure out what the speaker meant.
A decoding idiom is one that cannot be decoded by the hearer: a hearer will
not be able to figure out the meaning of the whole at all from the meaning of
its parts. Fillmore et al. give the examples of kick the bucket and pull a fast one.
Any decoding idiom is also an encoding idiom: if as a hearer you cannot figure
out what it means, then you are also not going to be able to guess that it is a
conventional way to express that meaning in the language. One of the reasons that
a decoding idiom is a decoding idiom is because there are not any correspondences
between the literal and idiomatic meaning of the parts of the decoding idiom. For
example, kick the bucket is a transitive verb phrase, but its idiomatic meaning is
the intransitive ‘die,’ and there is nothing that corresponds even metaphorically to
a bucket.
The encoding/decoding idiom distinction corresponds rather closely to Nun-
berg, Sag and Wasow’s distinction between idiomatically combining expressions
and idiomatic phrases respectively (Nunberg, Sag and Wasow 1994:496–97).
Idiomatically combining expressions are idioms where parts of the idiomatic
meaning can be put in correspondence with parts of the literal meaning. For in-
stance, in answer the door, answer can be analyzed as corresponding to the action
of opening, and of course the door denotes the door. In an idiomatically combining
expression such as spill the beans, meaning to divulge information, spill can be
analyzed as corresponding to ‘divulge’ and the beans to ‘information.’ In contrast,
no such correspondences can be established for kick and the bucket in kick the
bucket (‘die’). Nunberg et al. call the latter idiomatic phrases.
The encoding/decoding distinction is not the same as the idiomatically combin-
ing expression/idiomatic phrase distinction, however. Some idioms, such as spill
the beans, are encoding idioms even though they are idiomatically combining ex-
pressions. The encoding/decoding distinction is rather vaguely defined: it refers to
how clever (or lucky) the hearer is in decoding an expression of the language. For
this reason Nunberg et al.’s distinction is preferable.
The encoding/decoding distinction, as well as the idiomatically combining ex-
pression/idiomatic phrase distinction, characterizes idioms in contrast to regular
syntactic expressions with respect to the interpretation rules linking the syntac-
tic component to the semantic component. With idiomatic phrases such as kick
the bucket, the interpretation rules cannot apply because the parts of the syntactic
phrase do not correspond to parts of the semantic phrase at all. With idiomatically
combining expressions such as spill the beans, the parts of the syntactic phrase
correspond to semantic elements, but only in an interpretation that is unique to the

idiomatically combining expression (spill does not mean ‘divulge’ except in the
idiom spill the beans). Thus, the idiomatic meaning of idiomatically combining
expressions cannot be determined from the general rules of semantic interpretation
for the words or for the syntactic structure.
The second distinction that Fillmore et al. offer for defining idioms is between
grammatical and extragrammatical idioms. Grammatical idioms are parsable by
the general syntactic rules for the language, but are semantically irregular (i.e.
they are encoding or decoding idioms). Examples include those discussed so far,
such as kick the bucket and spill the beans, and also examples such as (X) blows
X’s nose. All of these idioms follow the general English syntactic rule that di-
rect objects follow the verb, and that a possessive modifier precedes the noun it
modifies.
Extragrammatical idioms are idioms that cannot be parsed by the general syn-
tactic rules for the language. Fillmore et al. give as examples of extragrammatical
idioms first off, sight unseen, all of a sudden, by and large and so far so good. One
might think that extragrammatical idioms are rare, but Nunberg et al. suggest that
they may not be particularly rare. Nunberg et al. provide a sampling of extragram-
matical idioms, which are given here (Nunberg et al. 1994:515; note there is only
one idiom overlapping with Fillmore et al.’s list):
(13) by and large; No can do; trip the light fantastic; kingdom come; battle royal;
Handsome is as handsome does; Would that it were . . .; every which way; Easy
does it; be that as it may; Believe you me; in short; happy go lucky; make believe;
do away with; make certain
The grammatical/extragrammatical distinction characterizes idioms in contrast

to regular syntactic expressions with respect to the rules of the syntactic component.
Grammatical idioms conform to the syntactic rules, but are idiomatic in some other
fashion. Extragrammatical idioms do not conform to the syntactic rules, and for
that reason alone are idiomatic.
Fillmore et al.’s third distinction is between substantive and formal idioms. A
substantive, or lexically filled, idiom is one in which all elements of the idiom are
fixed. For example, the idiom It takes one to know one is completely fixed; one
cannot even alter the tense (*It took one to know one). A formal, or lexically open,
idiom is one in which at least part of the idiom can be filled by the usual range
of expressions that are syntactically and semantically appropriate for the slot. For
example, with the idiom (X) blows X’s nose, the expressions I have described as
X can be filled by a noun phrase (and a corresponding coreferential possessive
pronoun) that refers to a person possessing a nose: I blew my nose, Kim blew her
nose, They all blew their noses and so on. Fillmore et al.’s use of the term ‘formal’
corresponds to Langacker’s term schematic to indicate a more general category

(see §3.2), and we will use the term ‘schematic’ here.
Fillmore et al. note that one potential confusion with respect to the substan-
tive/schematic distinction is that there may be a substantive idiom that fits the
pattern of a counterpart schematic idiom. For example, they posit a schematic
construction which they loosely describe as The X-er, the Y-er. Examples of this
schematic construction are given in (14a–c):
(14) a. The more you practice, the easier it will get.

b. The louder you shout, the sooner they will serve you.
c. The bigger the nail is, the more likely the board is to split.
There is also a substantive idiom that fits the pattern of the The X-er, the Y-er
schematic idiom:
(15) The bigger they come, the harder they fall.
The existence of the schematic idiom The X-er, the Y-er does not preclude the
existence of a substantive idiom like (15), just as the existence of a general syntactic
rule where the direct object follows the verb does not preclude the existence of an
idiom such as kick the bucket.
The substantive/schematic distinction characterizes idioms in contrast to regular
syntactic expressions on the one hand and the lexicon on the other. Both substantive
and schematic idioms have parts that are lexically completely specified, although
schematic idioms have parts that are specified in syntactic terms (that is, by a
syntactic category such as ‘noun phrase’ or ‘possessive pronoun’). In contrast,
syntactic rules make reference only to general syntactic categories such as V
(verb), NP (noun phrase) and so on, as in the phrase structure rules given in (16)
for simple active intransitive and transitive sentences:
(16) a. S → NP VP
b. VP → V
c. VP → V NP
The last distinction that Fillmore et al. give is for idioms with or without prag-
matic point (Fillmore et al. 1988:506). Idioms with pragmatic point are idioms
that, in addition to having a meaning in the usual sense of that term, also are
specifically used in certain pragmatic contexts. Obvious examples of idioms with
pragmatic point are idioms used for opening and closing conversations such as
Good morning or See you later, and for other specialized discourse contexts
such as telling a fairy tale (Once upon a time . . .; ibid.). Other idioms with
pragmatic point are those that have a certain conventional pragmatic content, as
with the schematic idiom illustrated by Him be a doctor?! On the other hand,
many other idioms such as all of a sudden do not have any specific pragmatic
point.
The with/without pragmatic point distinction characterizes idioms with respect
to the ‘information structure’ or ‘discourse’ component that some linguists have
argued for. They demonstrate that some idioms have conventional information-
structure or discourse-contextual properties associated with them, which again
cannot be predicted from general pragmatic or discourse-functional principles.
For example, it may be a general pragmatic principle that in taking leave, one
may make reference to a future meeting; but it cannot be predicted that the
specific phrase See you later is conventionally used in English for that pur-
pose, whereas in Spanish Hasta luego (lit. ‘until later’) is used for the same
function.
Fillmore et al.’s analysis demonstrates that idioms are quite varied in their
syntactic, semantic and pragmatic properties, ranging from completely fixed ex-
pressions to more general expressions, which may be semantically more or less
opaque and may not even correspond to the general syntactic rules of the language.
The distinctions discussed above are summarized in (17):
(17) a. encoding vs. decoding

b. idiomatically combining expressions vs. idiomatic phrases
c. grammatical vs. extragrammatical
d. substantive vs. schematic (formal)
e. with pragmatic point vs. without pragmatic point
Fillmore et al. use the features given above for a final, three-way categorization
of idioms. Their first category of idioms are unfamiliar pieces unfamiliarly ar-
ranged. The new aspect of this definition is the fact that certain words occur only
in a idiom. Examples of (substantive) idioms with unfamiliar pieces are kith and
kin ‘family and friends’ and with might and main ‘with a lot of strength.’ In other
words, such idioms are lexically irregular as well as syntactically and semantically
irregular. Unfamiliar words are by definition unfamiliarly arranged: if the words
do not exist outside the idiom, then they cannot be assigned to a syntactic category
in terms of a regular syntactic rule. Also, unfamiliar words unfamiliarly arranged
are by definition semantically irregular.
However, an idiom containing unfamiliar pieces unfamiliarly arranged does
not imply that it is an idiomatic phrase; such an idiom can be an idiomatically
combining expression. This point is made clearer by the schematic idiom of this
type given by Fillmore et al., the idiom The X-er, the Y-er illustrated in (14) above.
The unfamiliar pieces are the two occurrences of the, which are not definite articles
(in fact, they come from the Old English instrumental demonstrative þy). The
unfamiliar arrangement is the parallel syntactic structure, with a degree expression
followed by a clause with a gap corresponding to the degree expression. In (14a) for
example, the parallel gapped structures are more . . . you practice and easier . . .
it will get ). Nevertheless, this idiom is an idiomatically combining expression,
in that the parts of the construction can be made to correspond with the parts of
its meaning (roughly, ‘the degree to which you practice determines the degree to
which it gets easy’).
Fillmore et al.’s second category of idioms are familiar pieces unfamiliarly
arranged. These idioms do not contain unique words but are extragrammatical. In
other words, such idioms are lexically regular, but syntactically and semantically
irregular. Fillmore et al. give all of a sudden and in point of fact as examples of
substantive idioms in this category. They give as an example of a schematic idiom of
this category the phrase Nth cousin (M times removed): this is a syntactically unique
construction. Again, idioms made up of familiar pieces unfamiliarly arranged may
be idiomatically combining expressions.
Fillmore et al.’s third and last category of idioms are familiar pieces familiarly
arranged. Such idioms are lexically and syntactically regular but semantically
irregular. Again, such idioms may be substantive or schematic; Fillmore et al. give
examples of both types. The substantive idioms they list are in fact not entirely
fixed expressions; they include pull X’s leg (which can have any person-denoting
noun phrase as X) and tickle the ivories ‘play the piano’ (which can be inflected
for tense/mood). Fillmore et al.’s schematic idioms are even more schematic; they
include what they call ‘fate tempting expressions’ such as Watch me (drop it, slip
etc.).
The types of idioms, and their comparison to regular syntactic expressions, are
given in Table 9.1.
Table 9.1 Types of idioms compared to regular syntactic expressions
Lexically Syntactically Semantically
Unfamiliar pieces unfamiliarly arranged irregular irregular irregular

Familiar pieces unfamiliarly arranged regular irregular irregular
Familiar pieces familiarly arranged regular regular irregular
Regular syntactic expressions regular regular regular
9.3 Idioms as constructions
Having presented their analysis and classification of idioms, Fillmore et

al. argue that the proper way to represent speaker’s knowledge of idioms is as
constructions. For Fillmore et al., a construction is a schematic idiom. That is,
some elements of the construction are lexically open on the one hand, and so the
idioms fitting the description cannot simply be listed as ‘phrasal lexical items.’ In
this respect, schematic idioms differ from substantive idioms. Fully substantive
idioms, such as It takes one to know one or The bigger they come the harder they
fall, can simply be listed as lexical items. Listing substantive idioms would require
the allowance for multiword lexical items in the lexicon. But this concession to
the linguistic facts does not conflict greatly with the principle of componential
grammar that arbitrary and idiosyncratic linguistic knowledge is found in the
lexicon (cf. the discussion of subcategorization frames in §9.2). Hence substantive
idioms do not require any drastic departure from the componential model of the
organization of grammar.
Schematic idioms, on the other hand, cannot simply be listed in the lexicon.
And schematic idioms are idioms; that is, they are semantically and possibly
also syntactically and lexically irregular. Syntactic, semantic and in some cases
pragmatic properties of schematic idioms cannot be predicted from the general
rules of the syntactic and semantic components (and the pragmatic component)
or the general rules linking these components together. Instead, the syntactic,
semantic (and in some cases pragmatic) properties must be directly associated
with the construction. Such a representation would cut across the components in
the componential model of grammatical knowledge, and hence represents a direct
challenge to that model, at least for idioms.
Fillmore et al. make the case for constructions as units of syntactic represen-
tation by examining one construction in great detail, the construction containing
the conjunction let alone, and demonstrating that it has syntactic, semantic and
pragmatic properties that cannot be described by the general rules of the language,
but is rule-governed within the context of the let alone construction and certain re-
lated constructions. The following discussion will present some of the more salient
unique properties of the let alone construction.
The syntax of the let alone construction is complex. Let alone can be described
as a coordinating conjunction; like other conjunctions, it conjoins a variety of
like constituents (Fillmore et al. 1988:514; the emphasized elements represent
prosody):
(18) a. Max won’t eat shrimp, let alone squid.

b. We’ll need shrimp and squid.
(19) a. Max won’t t o u c h the shrimp, let alone c l e a n the squid.
b. I want you to cook the shrimp and clean the squid.
However, let alone fails in some syntactic contexts where and is fine, and vice
versa (Fillmore et al. 1988:515–16; Fillmore et al. also discuss WH-extraction and
It-clefts):
(20) a. Shrimp and squid Moishe won’t eat.

b. *Shrimp let alone squid Moishe won’t eat.
c. *Shrimp Moishe won’t eat and squid.
d. Shrimp Moishe won’t eat, let alone squid.
Example (20d) shows that let alone allows sentence fragments for the second
conjunct. In this respect let alone is like certain other conjunctions, including
comparative than (Fillmore et al. 1988:517, 516):
(21) a. John hardly speaks Russian let alone Bulgarian.

b. John speaks Russian, if not Bulgarian.
c. John speaks better Russian than Bulgarian.
However, unlike comparative than and ordinary conjunctions, let alone is impossi-
ble with VP ellipsis (deletion of the verb phrase excluding the auxiliary; Fillmore
et al. 1988:516):
(22) a. Max will eat shrimp more willingly than Minnie will.
b. Max won’t eat shrimp but Minnie will.
c. *Max won’t eat shrimp let alone Minnie will.
The let alone construction is a focus construction, like a number of other con-
structions of English (see Prince 1981b, discussed below), hence its characteristic
prosody. In fact, let alone is a paired focus construction, like those given in (23b–c)
(Fillmore et al. 1988:517):
(23) a. He doesn’t get up for lunch , let alone breakfast .

b. He doesn’t get up for lunch , much less b r e a k fa s t.
c. She didn’t eat a b i t e, never mind a w h o l e m e a l .
The let alone construction allows for multiple paired foci (see [19a]), and in such
sentences allows multiple let alones (Fillmore et al. 1988:520):
(24) a. You couldn’t get a poor man to wash your car for two dollars, let alone a rich
man to wax your truck for one dollar.
b. You couldn’t get a poor man, let alone a rich man, to wash, let alone wax, your
car, let alone your truck, for two dollars, let alone one dollar.
In this respect, let alone is similar to the construction not P but Q (illustrated
in [25]) and to the respectively construction (illustrated in [26]); but Fillmore et
al. argue that in other respects let alone differs from both of these constructions
(1988:521–22):
(25) Ivan sent not an album but a book, and not to Anna on her anniversary but to
Boris on his birthday.
(26) Fred and Louise hated their shrimp and squid respectively.
Let alone is a negative polarity item, not unlike any; it occurs in negative contexts
and certain other contexts (Fillmore et al. 1988:518):
(27) a. He didn’t reach Denver , let alone Chicago .

b. He didn’t reach any major city.
(28) a. I’m too tired to g e t u p, let alone go running with you.
b. I’m too tired to do any chores.
However, unlike these polarity items, let alone is allowed in certain contexts where
other negative polarity items are disallowed (Fillmore et al 1988:519; example
[29a] is attested):
(29) a. You’ve got enough material there for a whole s e m e s t e r, let alone a w e e k.
b. *You’ve got enough material for any semester.
The semantics as well as the syntax of let alone is complex and not entirely pre-
dictable from more general rules of semantic interpretation from syntactic struc-
ture. As mentioned above, the let alone construction has at least one paired focus
(e.g. the pair semester and week in [29a]). The interpretation of a let alone sen-
tence requires the following steps. First the interpreter must recognize or construct
a semantic proposition in the fragmentary second conjunct that is parallel to the
proposition in the full first conjunct. Second, the interpreter must recognize or
construct a semantic scale underlying the elements in the propositions. This is not
always easy. For instance, the scale for (18a) may have to do with the assumed de-
gree of distastefulness of shrimp versus squid, or it may have to do with the relative
cost of shrimp versus squid (and Fred’s stinginess; Fillmore et al. 1988:524–25).
More specifically, the interpreter must perform the following semantic opera-
tions. The interpreter must construct a scalar model, which ranks propositions
on a scale – for example, the scale of distastefulness of eating seafood or the cost
thereof. The propositions in the two conjuncts must be from the same scalar model –
in this case, ‘Fred not eat shrimp’ and ‘Fred not eat squid.’ The two propositions
are of the same polarity (in this case, negative). Finally, the initial, full conjunct
denotes the proposition that is stronger or more informative on the scale – Fred not
eating shrimp is more informative than Fred not eating squid, on the assumption
that people who would eat squid would also eat shrimp but not vice versa. This
semantic analysis can be generalized to multiple paired focus versions of let alone
(see Fillmore et al. 1988 for details). This whole semantic apparatus is required for
the interpretation of the let alone construction, and is not necessary (as a whole)
for other constructions.
Finally, there is a specific pragmatic context in which the utterance of a let alone
construction is felicitous (Fillmore et al. 1988:532). First, the discourse context
is one such that the weaker (less informative) proposition, that is, the underlying
proposition of the fragmentary second conjunct, is at issue – for example, the issue
of whether or not Fred eats squid. The weaker proposition accepts or rejects this
context – in this case, Fred doesn’t eat squid rejects it. But simply uttering the less
informative proposition is not cooperative since the speaker knows that the strong
proposition represented by the initial conjunct is true. So the speaker utters the let
alone sentence. Fillmore et al. note that let alone is similar pragmatically to other
conjunctions allowing sentence fragments, such as those illustrated in (21b–c)
above. However, some of these conjunctions present the stronger proposition in
the second, fragmentary conjunct, unlike let alone:
(30) a. He didn’t make colonel, let alone general.

b. He didn’t make general; in fact, he didn’t even make colonel.
The preceding discussion has presented some of the evidence that the let alone
construction has its own syntactic, semantic and pragmatic properties that cannot be
predicted from more general rules of syntax, semantics and pragmatics. A number
of other studies done in the emerging framework of construction grammar demon-
strate that other constructions also have unique syntactic, semantic and pragmatic
properties. A reading of these studies gives rise to two general observations.
First, the construction on which attention is focused by the researcher(s) turns
out to be just one of a family of related constructions. For example, the let alone
construction turns out to be just one of a family of coordinate constructions that
allow certain kinds of sentence fragments in the second conjunct, two of which
were illustrated in (21b–c). The let alone construction also turns out to be just
one of a family of paired focus constructions, two of which were illustrated in
(23b–c). Paired focus constructions are in turn related to a family of single focus
constructions. The phrase let alone is itself related to other negative polarity items,
and let alone is also related to a number of items that require a scalar model for
their semantic interpretation, such as even, almost, few and merely (Fillmore et al.
1988:530).
Likewise, Lakoff’s seminal study of the There-construction, as in There’s a
fox in the garden, uncovered a large family of related constructions with slightly
different syntactic and semantic properties, which are illustrated in examples (31)–
(32) (see Lakoff 1987, Appendix 3 for the analysis of the differences among There-
constructions):
(31) Deictic There-Constructions:

a. Central: There’s Harry with the red jacket on.
b. Perceptual: There goes the bell now!
c. Discourse: There’s a nice point to bring up in class.
d. Existence: There goes our last hope.
e. Activity Start: There goes Harry, meditating again.
f. Delivery: Here’s your pizza, piping hot!

g. Paragon: Now there was a real ballplayer!
h. Exasperation: There goes Harry again, making a fool of himself.
i. Narrative Focus: There I was in the middle of the jungle . . .
j. New Enterprise: Here I go, off to Africa.
k. Presentational: There on that hill will be built by the alumni of this university
a ping-pong facility second to none.
(32) Existential There-Constructions

a. Central: There’s a fox in the garden.
b. Strange [Event]: There’s a man been shot.
c. Ontological: There is a Santa Claus.
d. Presentational: Suddenly there burst into the room an SS officer holding a
machine gun.
Michaelis and Lambrecht’s (1996) study of Nominal Extraposition, illustrated in

(33a), reveals a family of related exclamative constructions, illustrated in (33b–e):
(33) a. It’s amazing the amount I spent!

b. I can’t believe the amount I spent!
c. The amount I spent!
d. I can’t b e l i e v e how much I spent!
e. It’s i n c r e d i b l e how much I spent!
The Nominal Extraposition construction in (33a) is characterized by extraposition

of the NP and a metonymic interpretation of the extraposed NP as referring to a
scale, unique to exclamatives (the exclamative character of the sentence is due to
the assertion of an excessive degree on the scale). The constructions in (33b–c)
share the metonymic interpretation of (33a), but the NP is not extraposed (33b)
or is an independent utterance on its own (33c). The exclamative constructions in
(33d–e) directly express the scale (how much), unlike (33a–c). The construction in
(33e) extraposes the degree expression, not unlike the extraposed (33a), while the
construction in (33d) does not. All five constructions in this family are distinguished
by the fact that they are a distinct speech act (expressed by the simple present tense),
and they assert an affective stance, namely contravention of expectation (this fact
restricts the main clause predicate to a gradable, contrary to expectation assertion).
The second observation upon surveying these studies of particular construc-
tions follows from the first. The number and variety of constructions uncovered
in these studies imply that speakers possess an extraordinary range of specialized
syntactic knowledge that goes beyond general rules of syntax and semantic in-
terpretation on the one hand, and a list of substantive idioms on the other. The
detailed analysis of such constructions is not the exclusive preserve of construc-
tion grammarians. Linguists working in a variety of approaches to syntax and
semantics have examined schematic idioms/constructions and demonstrated that
they represent rule-governed and productive linguistic behavior, albeit limited to

the family of constructions analyzed.
One of the linguistic schools that calls itself ‘functionalist,’ which we will
call ‘autonomous functionalist’ (cf. Croft 1995:496–99), identifies constructions
that possess a specific discourse-functional or information-structural value. For
instance, Prince (1978) argues that the constructions known as It-cleft (illustrated
in [34]) and WH-cleft (35) each have their own discourse-functional value (Prince
1978:885):
(34) It is against pardoning these that many protest (Philadelphia Inquirer, February
6, 1977)
(35) What you are saying is that the President was involved (Haldeman, Watergate
tapes)
Prince notes that WH-clefts and It-clefts differ syntactically, in that the former
allow clefted adverbs or prepositional phrases as well as clefted noun phrases,
and the latter commonly allow verb phrases or sentences as clefted items (Prince
1978:884). Discourse-functionally, WH-clefts can be used when the information
in the subordinate clause is in the hearer’s consciousness (Prince 1978:894). In
contrast, Prince identifies at least two distinct ‘sub-senses’ for It-clefts, illustrated
in (36)–(37) (1978:896, 898):
(36) So I learned to sew books. They’re really good books. It’s just t h e c ov e r s that
are rotten. (Bookbinder in S. Terkel, Working, 1974)
(37) It was just about 50 years ago that Henry Ford gave us the weekend. (Philadelphia
Bulletin, January 3, 1976)
Example (36) illustrates what Prince calls a stressed focus It-cleft. In the stressed
focus It-cleft, the subordinate that-clause is given but not assumed to be in the
hearer’s consciousness. The stressed focus It-cleft is interesting also in that it
has a phonological property associated with it: only the focused part (in small
capitals in [36]) has strong stress; the that-clause is weakly stressed. Example
(37) is an informative-presupposition It-cleft: the that-clause presents information
that is a general known fact, albeit not to the hearer and hence new to the hearer
(and therefore also not in the hearer’s consciousness). Informative-presupposition
It-clefts have a normal rather than weak stress on the that-clause. These examples
indicate that constructions may have phonological features associated with them
as well as syntactic, semantic and pragmatic/discourse features.
Birner and Ward (1998) analyze a wide range of preposing constructions, such
as Topicalization (illustrated in [38]; Birner and Ward 1998:51), postposing con-
structions, such as right-dislocation (as in [39]; Birner and Ward 1998:146) and
argument reversal constructions, such as inversion (as in [40]; Birner and Ward
1998:159):
(38) As members of a Gray Panthers committee, we went to Canada to learn, and learn
we did. (Philadelphia Inquirer, June 16, 1985)
(39) It’s very delicate, the lawn. You don’t want to over-water, really. (father in the
movie ‘Honey, I Shrunk the Kids’)
(40) Behind them, moving slowly and evenly, but keeping up, came Pa and Noah.
(J. Steinbeck, The Grapes of Wrath, 1939)
Birner and Ward argue that, although there are commonalities among the different
constructions with respect to the discourse status of the preposed element and
the status of the postposed elements for syntactically similar constructions, each
has its own unique discourse properties. In other words, the various preposing,
postposing and inversion structures they discuss must each be analyzed as distinct
grammatical constructions.
Wierzbicka has discussed the properties of several families of constructions in
various publications (see the papers collected in Wierzbicka 1980, 1988, as well as
the examples discussed here). For example, Wierzbicka argues that the schematic
idiom have a V and the related types give a V and take a V, illustrated in (41)–(43),
represent rule-governed constructions (Wierzbicka 1988:293, 338):
(41) a. have a drink
b. *have an eat
(42) a. give the rope a pull
b. *give the window an open
(43) a. take a look at
b. *take a look for
Wierzbicka argues that the item following the indefinite article is a verbal in-
finitive, not a noun, and hence differs from other have constructions that do take
a noun, or more generally a noun phrase. For example, the phrase have a cough
is nominal, in that one can also have a headache/have pneumonia and so on in
which the word is indubitably a noun (Wierzbicka 1988:295–96). In this respect,
the have a V construction is syntactically unique.
Semantically, Wierzbicka argues that have a V represents an action as limited
in time but not punctual, lacking an external goal, and repeatable, and is of benefit
to the agent/experiencer (1988:297–302). Wierzbicka argues that this semantic
characterization is still incomplete, since it provides necessary but not sufficient
conditions for the occurrence of verbs in this construction. Instead, she presents
ten subtypes of the have a V construction, just as Prince offers two subtypes of the
It-cleft. One of these types she describes as ‘aimless objectless action which could
cause one to feel good,’ exemplified by have a walk/swim/run/jog/lie-down and so
on. In this subclass, the verbs are intransitive but durative and atelic (Wierzbicka
1988:303), hence the unacceptability of (44b–c):
(44) a. He had a walk.

b. *He had a walk to the post office.
c. *He had a get-up.
The verb cannot describe a purposeful action (one with an external goal), other
than a recreational activity, hence the unacceptability of (45b–c):
(45) a. He had a swim.

b. *He had a work.
c. *He had a pray.
One indicator of the conventional character of the interpretations of schematic

idioms/constructions, particularly pragmatic ones, is the lack of translatability of
the idiom. For example, it has been argued that the tautological statement in (46)
is inferred to mean something like ‘That’s the kind of unruly behavior you would
expect from boys’ on general pragmatic principles (Levinson 1983:125):
(46) Boys will be boys.
Wierzbicka (1987) points out that in fact, the literal translation of (46) in various
languages (see examples [47]–[50]) does not have the same pragmatic meaning
as (46). Instead, different constructions are used to obtain approximately the same
pragmatic force (Wierzbicka 1987:96–97). In examples (47)–(50), the (a) sentence
is the closest literal translation to (46), the (b) sentence the one with the closest
pragmatic meaning to that of (46), and the (c) sentence is the literal translation of
the (b) sentence:
(47) French:
a. *Les garçons sont les (des?) garçons.
b. ?Les garçons seront toujours les (des) garçons. [still questionable]
c. ‘Boys will always be boys’
(48) German:
a. *Knaben werden Knaben sein.
b. ?Knaben bleiben (immer) Knaben.
c. ‘Boys remain (always) boys’
(49) Russian:
a. *Mal’čiki budut mal’čiki.
b. (Čego ty xočeš’?) oni že mal’čiki.
c. ‘(What do you expect?) They are boys’
(50) Polish:
a. *Chlopcy be da chlopcy.
b. (Jednak) co Paryż to Paryż.
c. ‘(However) what (is) Paris this (is) Paris’
These examples indicate that the pragmatic interpretation in (46) is in fact con-
ventionally associated with the equational tautology in which it occurs in English.
In fact, Wierzbicka argues for several different equational tautological con-
structions in English, which cannot be substituted for each other (Wierzbicka
1987:104):
(51) Nabstract is Nabstract .

a. War is war.
b. *Kid is kid.
(52) Nplural are Nplural .
a. Kids are kids.
b. *Wars are wars.
(53) Nplural will be Nplural .
a. Boys will be boys.
b. *Wars will be wars.
(54) An N is an N.
a. A party is a party.
b. *A war is a war.
(55) The N is the N.
a. The law is the law.
b. *The war is the war.
Wierzbicka argues that the semantic interpretations for the constructions exem-
plified in (51)–(55) can be characterized as follows: a ‘sober’ attitude toward com-
plex human activities (51); tolerance for human nature (52–53), the future subtype
indicating ‘the willful and uncontrollable spontaneity’ of the human type (1987:
107); obligation with respect to a human role, activity or institution ([54]–[55];
Wierzbicka argues that [54] has other readings as well). These semantic differences
cannot be inferred either from general rules of semantic interpretation in English
or general rules of the pragmatics of communication.
Even in the generative grammatical tradition, which is the theory most closely
identified with the componential model, there have been studies of schematic id-
ioms, in particular by Jackendoff (1990, 1997; see also Akmajian 1984 and Lam-
brecht’s 1990 reanalysis in construction grammar terms). For example, Jackendoff
(1997) analyzes the ‘time’-away construction, illustrated in (56):
(56) Bill slept the afternoon away.
Syntactically, the noun-phrase after the intransitive verb acts like a direct object
complement, and normally cannot occur with a transitive verb ([57a–b]). In some
cases the ‘normal’ direct object can appear in a with phrase, which it cannot do in
an ordinary active construction ([57c–d]; Jackendoff 1997:535):
(57) a. Fred drank the night away.

b. *Fred drank scotch the night away.
c. Fred drank the night away with a bottle of Jack Daniels.
d. *Fred drank with a bottle of Jack Daniels.
Semantically, the ‘time’-away construction appears to have the same interpre-

tation as the durative adverbial with for (Jackendoff 1997:536):
(58) Bill slept for the (whole) afternoon.
However, unlike the durative adverbial, the ‘time’-away construction requires a

volitional subject ([59a–b]) and an activity rather than a state ([60a–b]; Jackendoff
1997:537):
(59) a. The light flashed for two hours.

b. *The light flashed two hours away.
(60) a. Celia sat for two hours.
b. *Celia sat two hours away.
The particle away as an aspectual particle seems to have a meaning and behavior
similar to away in the ‘time’-away construction:
(61) Bill drank away.

(62) *Celia sat away.
But the particle away is atelic (unbounded), while the ‘time’-away construction is
telic, as indicated by the It take NPtime construction:
(63) a. *It took a month for Lois and Clark to finally get to dance away.
b. It took a month for Lois and Clark to finally get to dance two blissful hours
away.
These properties indicate the uniqueness of the ‘time’-away construction with

respect to general syntactic and semantic rules. Jackendoff further argues that the
‘time’-away construction’s properties cannot be predicted from the properties of
other semantically related constructions, such as the resultative construction (64)
and the way construction ([65]; see Jackendoff 1997):
(64) The river froze solid.

(65) Dora drank her way down the street.
Jackendoff weighs two analyses of the ‘time’-away construction, the

construction-based account of the construction grammarians, and an account in
which a lexical rule derives the relevant verb that governs this type of construc-
tion. Jackendoff concludes that the only substantive difference between the two
accounts is that if one wants to ‘preserve the assumption that the lexical verb’s argu-
ment structure always determines the argument structure of the VP’ (Jackendoff
1997:557), then one must commit oneself to the lexical rule analysis. Jackend-
off himself inclines to the constructional analysis for the ‘time’-away construc-
tion, since he believes that constructions are necessary in other contexts anyway
(ibid.).
Jackendoff’s inclination in his 1997 paper is another step away from the com-
ponential model of generative grammar toward a construction grammar model.
Jackendoff’s inclination is also a step toward the construction grammarian’s bolder
hypothesis. Since one must posit constructions in order to account for a substantial
part of a speaker’s grammatical knowledge, is it possible to generalize the concept
of construction to account for all of a speaker’s grammatical knowledge? The next
section presents construction grammar’s arguments for the bolder hypothesis.
9.4 From constructions to construction grammar
The preceding section presented a number of case studies that argue for
the need to posit constructions as a unit of syntactic representation. A construction
is a syntactic configuration, sometimes with one or more substantive items (e.g.
the words let alone, have a . . . and away) and sometimes not (as with the focus
constructions, the exclamative constructions and the resultative construction). A
construction also has its own semantic interpretation and sometimes its own prag-
matic meaning (as with the tautological constructions). Hence a construction as a
unit cuts across the componential model of grammatical knowledge. The existence
of constructions would require a revision to the componential model in (3) that we
may represent as in (66):
(66)
phonological component
constructions
linking rules
lexicon
syntactic component
linking rules
semantic component
Constructions, like the lexical items in the lexicon, are ‘vertical’ structures that
combine syntactic, semantic and even phonological information (for the specific
words in a construction, as well as any unique prosodic features that may be
associated with a construction). As more and more constructions are discovered
and analyzed, construction grammarians came to argue that, in fact, grammatical
organization is entirely ‘vertical’ (indeed, this approach is already suggested in

Fillmore et al. 1988).
We begin with the syntactic structure of constructions. In §9.3, constructional
analyses were proposed for schematic idioms. Schematic idioms were defined in
§9.2 as idioms in which some element or elements are lexically open, indicated by
a category label as in have a V. Schematic idioms were contrasted with substantive
idioms such as It takes one to know one, in which there are no lexically open ele-
ments. Hence, substantive idioms can be listed in the lexicon without substantially
altering the basic principles of the componential model.
But Fillmore et al. observe in a footnote that there is in fact a continuum from
substantive to schematic (1988:505, n. 3). Although we described all idioms with
any lexically open elements as schematic idioms in §9.2, in fact schematic id-
ioms vary considerably in their schematicity. Some schematic idioms such as the
verb-phrase idiom kick the bucket are fixed except for grammatical inflectional
categories:
(67) a. Jake kicked the bucket.

b. Jake’s gonna kick the bucket. [etc.]
Other schematic idioms have one or more open argument slots as well as inflec-
tional flexibility, such as give NP the lowdown ‘tell NP the news’:
(68) a. I gave/I’ll give him the lowdown.

b. He gave/He’ll give Janet the lowdown. [etc.]
Still other schematic idioms have open classes for all ‘content’ words, leaving
just a salient form such as the connective let alone as a substantive element:
(69) a. She gave me more candy than I could carry, let alone eat.
b. Only a linguist would buy that book, let alone read it.
Finally, a constructional analysis has been proposed for some schematic id-
ioms in which all elements are lexically open, such as the resultative construction
(Goldberg 1995:181; attested examples):
(70) a. This nice man probably just wanted Mother to . . . kiss him unconscious.
(D. Shields, Dead Tongues, 1989)
b. I had brushed my hair very smooth. (C. Brontë, Jane Eyre, 1847)
Yet the resultative construction has no lexically specific element. It can be de-
scribed only by a syntactic structure, in this case [NP Verb NP XP], with a unique
specialized semantic interpretation.
It is a very short step from analyzing the resultative construction as a construc-
tion to analyzing all the syntactic rules of a language as constructions (Fillmore
et al. 1988:501, 534; Langacker 1999:19). After all, a syntactic rule such as
VP → V NP describes a completely schematic construction [V NP], and the seman-
tic interpretation rule that maps the syntactic structure to its corresponding semantic
structure is unique to that schematic construction. Indeed, Goldberg suggests that
there is a transitive construction just as there are more specialized schematic syn-
tactic constructions such as the resultative construction (Goldberg 1995:116–19).
Reanalyzing general syntactic rules as the broadest, most schematic constructions
of a language is just the other end of the substantive- schematic continuum for
idioms/constructions.
Turning to semantic interpretation, one can also argue that constructions and
compositional semantic rules differ only in degree, not in kind. As we noted in
§9.2, Nunberg et al. (1994) argue that most idioms are idiomatically combining
expressions. In an idiomatically combining expression, the syntactic parts of the
idiom (e.g. spill and beans) can be identified with parts of the idiom’s semantic
interpretation (‘divulge’ and ‘information’). Nunberg et al. argue that idiomatically
combining expressions are not only semantically analyzable, but also semantically
compositional.
Nunberg et al. observe that idiomatically combining expressions are only the
extreme end of a continuum of conventionality in semantic composition. The other
end of the continuum is represented by selectional restrictions. Selectional re-
strictions are restrictions on possible combinations of words which are determined
only by the semantics of the concepts denoted by the word. For example, the re-
strictions on the use of mud and car in (71)–(72) follow from the fact that mud is
a viscous substance and a car is a machine:
(71) a. Mud oozed onto the driveway.

b. ?*The car oozed onto the driveway.
(72) a. The car started.
b. ?*Mud started.
The restrictions on mud and car are not dependent on the conventional form in
which the concepts are expressed. If one used the word goo instead of mud or
automobile instead of car, the judgements in (71)–(72) would remain the same.
The combinations in (71a) and (72a) are semantically compositional: the meaning
of the whole can be predicted from the meaning of the parts.
An intermediate point on this continuum involves what are called collocations.
Collocations are combinations of words that are preferred over other combinations
that otherwise appear to be semantically equivalent. For example, Matthews argues
that toasted and roasted describe essentially the same process, but are restricted
in their acceptable combinations (Matthews 1981:5):
(73) a. roasted meat

b. toasted bread
(74) a. ?*toasted meat
b. ?*roasted bread
Most linguists would analyze (73a–b) as semantically compositional as well. In

both cases, the meaning of the whole can be predicted from the meaning of the
parts. It is just that speakers of English conventionally use toasted with bread and
roasted with meat, but not the other way around. This convention does not affect
the semantic compositionality of the expressions in (73a–b).
Typically, collocations are expressions that can be interpreted more or less cor-
rectly out of context, but cannot be produced correctly if the conventional expres-
sion is not already known to the speech community (Nunberg, Sag and Wasow
1994:495). In other words, collocations are encoding idioms. For example, the
expressions in (75a) and (75b) are the American and British terms for the same
type of object; each is compositional to the speakers of that dialect, but a speaker
of the other dialect would not be able to know what conventional expression is
used to refer to that type of object:
(75) a. thumb tack (American English)

b. drawing pin (British English)
Nunberg et al. argue that exactly the same reasoning applies to idiomatically
combining expressions. Idiomatically combining expressions are largely fixed in
their words; any substitution leads to ungrammaticality, as in (76b–c) and (77b):
(76) a. Tom pulled strings to get the job.

b. *Tom pulled ropes to get the job.
c. *Tom grasped strings to get the job.
(77) a. She spilled the beans.
b. *She spilled the succotash.
However, given the meanings of the words in the idiomatically combining expres-
sion, the meaning of the whole expression is compositional:
By convention . . . strings [in pull strings] can be used metaphorically to refer to

personal connection when it is the object of pull, and pull can be used metaphori-
cally to refer to exploitation or exertion when its object is strings. (Nunberg et al.
1994:496)
When we hear spill the beans used to mean ‘divulge the information’, for ex-
ample, we can assume that spill denotes the relation of divulging and beans the
information that is divulged, even if we cannot say why beans should have been
used in this expression rather than succotash. This is not to say, of course, that
spill can have the meaning ‘divulge’ when it does not co-occur with the beans, or
that beans can have the meaning ‘information’ without spill. The availability of
these meanings for each constituent can be dependent on the presence of another
item without requiring that the meaning ‘divulge the information’ attach directly
to the entire VP. Rather it arises through a convention that assigns particular
meaning to its parts when they occur together. (Nunberg et al. 1994:497)
At first, Nunberg et al.’s analysis may look odd. To say that pull and strings
each have a meaning found only in pull strings, and that those meanings are
compositional in the idiomatically combining expression, seems ad hoc. The more
natural description is the traditional one, that the meaning of the idiomatically
combining expression is ‘noncompositional.’ In fact, it is sometimes said that
one of the strongest pieces of evidence for constructions as independent syntactic
objects is that there is some degree of ‘noncompositionality’ in the meaning of
the construction. But there is evidence that Nunberg et al.’s analysis is the right
one.
Some English words exist only in idiomatically combining expressions, such as
heed in pay heed. It makes sense to say that heed has a meaning, that is of course
found only in pay heed. It has been argued that heed is idiomatic, because it is
essentially synonymous with attention in pay attention, and yet does not behave
the same way (Radford 1988; see Nunberg et al. 1994:505):
(78) a. You can’t expect to have my attention/*heed all the time.

b. He’s a child who needs a lot of attention/*heed.
Nunberg et al. argue that heed does not in fact mean the same thing as attention
does, when attention is the object of pay (Nunberg et al. 1994:505):
(79) a. The children paid rapt attention/?*heed to the circus.

b. I pay close attention/?*heed to my clothes.
c. They paid attention/??*heed to my advice, but didn’t follow it.
The semantic differences are related to the difference between the verbs attend and
heed: ‘we clearly attend to much that we do not heed . . . one can take heed but not
attention, and . . . attention but not heed can wander’ (Nunberg et al. 1994:506).
In other words, heed in pay heed does have its own meaning even though it occurs
(as a noun) in only that combination. Hence, it is reasonable to assume that other
words have specialized meanings in idiomatically combining expressions, and that
those meanings are compositional.
Another important line of evidence for the compositionality of idiomatically
combining expressions is psycholinguistic. Speakers of English recognize the
meanings of words in idiomatically combining expressions, and recognize them
as figurative meanings, even though the figurative meanings are found only in the
idiomatically combining expressions (Gibbs 1990). These two pieces of evidence
point to Nunberg et al.’s conclusion that ‘The dependency among the parts of
idiomatically combining expressions is thus fundamentally semantic in nature’
(Nunberg et al. 1994:505).
From a construction grammar perspective, Nunberg et al.’s analysis of idiomat-
ically combining expressions looks more natural. An idiomatically combining ex-
pression such as spill the beans is a construction. As a construction, it has unique
syntax: the verb must be spill and its object must be the beans. It also has a se-
mantic interpretation, namely ‘divulge information.’ All Nunberg et al. are saying
is that this construction has its own semantic interpretation rules, mapping spill
onto ‘divulge’ and the beans onto ‘information.’ The constructional analysis is
presented in the diagram in (80), using lowercase to describe form and uppercase
to describe meaning, boxes to represent the construction and its parts, and dotted
lines to indicate the syntax-semantics mapping (see §10.1):
(80)
spill the beans
DIVULGE INFORMATION
What Nunberg et al. have done is to dissociate conventionality from noncompo-

sitionality. Idiomatically combining expressions are not noncompositional. There
exist truly noncompositional expressions; these are idiomatic phrases such as saw
logs and kick the bucket. Idiomatically combining expressions differ from collo-
cations and ordinary expressions only in that the conventional way of expressing
the parts of its meaning are conventional and also relatively opaque, compared to
collocations and ordinary expressions.
Earlier analysts have assumed that an idiomatically combining expression is
‘noncompositional’ because the meaning of the whole is not predictable from
the meaning of the parts when those parts appear in other expressions than the
idiom. More precisely, idiomatically combining expressions have been treated
as ‘noncompositional’ because their meanings do not conform to the semantic
interpretation rules of regular syntactic expressions such as [Ve r b O b j e c t]VP
in the case of spill the beans. But spill the beans is compositional in the sense
that the parts of the syntactic expression can be mapped onto components of the
meaning of the idiom, as in (80). The way that spill the beans differs from regular
syntactic expressions is that there are rules of semantic interpretation associated
with just that construction that are not derivable from the [Ve r b O b j e c t]VP
pattern of which spill the beans is an instance.
Thus, the common perception that a particular construction must be represented

as an independent syntactic unit because it is ‘noncompositional,’ is technically
incorrect. Constructions other than idiomatic phrases are compositional, that is,
the meanings of the parts of the construction are combined to form the meaning of
the whole construction. The reason that they must be represented as independent
constructions is that semantic interpretation rules associated with the construction
are unique to that construction, and not derived from another more general syntactic
pattern, as construction grammarians carefully note (see, e.g., Goldberg 1995:13
and Michaelis and Lambrecht 1996:219).
Indeed, one can think of the general ‘compositional’ rules of semantic inter-
pretation as being semantic rules associated with general (schematic) syntactic
structures, just as specialized rules of semantic interpretation are associated with
syntactically specialized extragrammatical idioms. Nunberg et al.’s analysis of id-
iomatically combining expressions can easily be extended to the general rules of
semantic interpretation that link syntactic and semantic structures. Consider, for
example, the English predicate adjective construction, illustrated in (81), and its
semantic interpretation:
(81) Hannah is smart.
The English predicate adjective construction has the form [NP be Adj]. It differs
from the ordinary verbal construction in requiring the copula verb be. One can ana-
lyze the semantics of the predicate adjective construction as follows. The members
of the Adjective category have a meaning that requires them to be combined with
the copula be in order to be interpreted as ascribing a property to a referent (unlike
verbs). The copula be has a meaning that requires combination with a member
of the Adjective category in order to be interpreted as doing the job of ascribing
(a property) to the subject NP. This analysis is in fact essentially the semantic
analysis that Langacker argues for (Langacker 1987:214–22; 1991a:204–5). In
Langacker’s terminology (see §3.5), Adjective symbolizes an atemporal relation,
and the copula be symbolizes a process that Adjective meanings must be combined
with in order to be predicated.
In like fashion, semantic interpretation rules can be provided for any schematic
construction describing the most general syntactic structures of the language. In
other words, all syntactic expressions, whatever their degree of schematicity, have
rules of semantic interpretation associated with them, although some substantive
idioms appear to inherit their semantic interpretation rules from more schematic
syntactic expressions such as [Verb Object] (see §10.2.1). Hence, the difference
between regular syntactic expressions and idiomatically combining expressions
is not that the former are ‘compositional’ and the latter are ‘noncompositional.’
Instead, the former’s rules of semantic composition are more general and the
latter’s rules of semantic composition are more specialized. In semantics as well

as syntax, the concept of a construction can be generalized to encompass the full
range of grammatical knowledge of a speaker.
If syntax and semantics as a whole can be represented as constructions, what
about morphology and the lexicon? Morphology, like syntax, represents complex
grammatical units, made up of morphemes. From a structural point of view, the
only difference between morphology and syntax is that morphemes are bound
within a word, while words are morphologically free within a phrase or sentence.
Interestingly, analogs to almost all of the peculiar phenomena of idioms can be
found in morphology.
There are unfamiliar morphemes that exist only in single combinations, such as
cran- in cranberry (cf. kith and kin, pay heed). Such morphemes caused problems
for American structuralist analysis, because one had to assign a meaning (if any) to
the unfamiliar morpheme only in that word. This is, of course exactly the analysis
advocated by Nunberg et al. for their syntactic analogs.
There is also ‘extragrammatical’ morphology, that is, morphological patterns
that do not obey the general morphological rules of the language. The general rule
for plural formation in English is suffixation of an allomorph of -s to the noun stem.
The ablaut plurals of English such as feet, geese and so on are outside the general
plural formation rule. Arguably, the plural of brother-in-law, either brothers-in-
law or brother-in-laws, is also outside the general rule. Such examples are common
across languages. For example, the general rule for the position of agreement affixes
in K’iche’ Mayan is as a prefix immediately following the aspect prefix: x-at-w-
il-oh ‘I saw you [familiar].’ However, the second person formal morpheme is a
free word following the verb form, and hence is an ‘extragrammatical’ morpheme:
x-w-il alaq ‘I saw you [formal]’ (Mondloch 1978:27).
Morphological expressions can be placed on a continuum of schematicity. A
maximally substantive morphological expression is fully specified, as in book-s.
Partially schematic morphological expressions include book-NUMBER and NOUN -s.
Fully schematic morphological expressions include NOUN -NUMBER .
Finally, many words are what one might call ‘idiomatically combining words,’
where the meaning of a morpheme is specific to the stem it combines with (or a
subclass of stems). For example, -en is the plural of brother only when it refers to
a member of a religious community, and brother refers to a member of a religious
community when it is combined with -en.2 The derivational suffix -er refers to
the agent of the event denoted by the verb stem when that verb stem is in a class
including write, run and so on, but refers to the instrument if the verb stem is clip,
staple and the like, or the patient if the verb stem is fry, broil and so on. All of
2 We ignore here the fact that the plural stem for the plural in -en is distinct from the singular stem.
these observations suggest that in fact morphology is very much like syntax, and
that a constructional representation is motivated for morphology as well.
Lastly, the lexicon differs only in degree from constructions. Words in the lexi-
con are pairings of syntactic form (and phonological form) and meaning, includ-
ing pragmatic meaning. Constructions are also pairings of syntactic form (and
phonological form, for the substantive elements) and meaning, including prag-
matic meaning. The only difference is that constructions are complex, made up of
words and phrases, while words are syntactically simple. Some words are morpho-
logically complex, of course. But we have just argued that construction grammar
would analyze morphologically complex words as constructions whose parts are
morphologically bound. Morphologically simple words are atomic, that is, they
cannot be further divided into meaningful parts. But a word is again just the limiting
case of a construction (Fillmore et al. 1988:501).
The end point of this argument is one of the fundamental hypotheses of construc-
tion grammar: there is a uniform representation of all grammatical knowledge
in the speaker’s mind, in the form of generalized constructions. Table 9.2 com-
pares the different types of grammatical entities found in the componential model
of grammar and their analysis as constructions in construction grammar.
Table 9.2 The syntax-lexicon continuum
Construction type Traditional name Examples
Complex and (mostly) syntax [S b j be- t n s Ve r b -en by O b l]

schematic
Complex, substantive verb subcategorization [S b j consume O b j]
frame
Complex and (mostly) idiom [kick-t n s the bucket]
substantive
Complex but bound morphology [Noun-s], [Ve r b - t n s]
Atomic and schematic syntactic category [D e m], [A d j]
Atomic and substantive word/lexicon [this], [green]
Syntactic rules (and the accompanying rules of semantic interpretation) are

schematic, complex constructions. The subcategorization frames required to han-
dle verbal syntactic behavior are schematic constructions with a substantive verb.
Idioms are complex and (at least partly) substantive constructions. Morphology
describes complex constructions, but constructions of bound morphemes. Words
in the lexicon are atomic substantive constructions, while syntactic categories
are schematic atomic constructions. In other words, grammatical knowledge rep-
resents a continuum on two dimensions, from the substantive to the schematic
and from the atomic to the complex. This continuum is widely referred to as the
syntax-lexicon continuum. Thus, construction grammar conforms to Langacker’s

content requirement for a grammar: the only grammatical entities that are posited
in the theory are grammatical units and schematizations of those units (Langacker
1987:53–54).
The notion of a construction in construction grammar is much more general than
the traditional notion of a construction. In construction grammar, a construction can
be atomic or complex; it can have parts that are morphologically bound as well as
free; and any or all of the parts may be substantive or schematic. All constructions
in construction grammar, though, are pairings of a syntactic and morphological
(and, where relevant, phonological) form with a meaning, including pragmatic
meaning.
The model of grammatical knowledge in construction grammar is represented
in (82):
(82)
phonological properties
construction1
construction3
construction2
[etcetera]
lexicon
syntactic properties
semantic properties
A construction grammar consists of a large number of constructions of all types,

from schematic syntactic constructions to substantive lexical items. All of the
constructions possess properties of form (syntactic and phonological) and meaning
(semantic and pragmatic). All of these constructions are organized in a particular
way in a speaker’s mind. The next chapter describes how these generalizations are
elaborated in various theories of construction grammar.
10
An overview of construction grammars
10.1 Essentials of construction grammar theories
In chapter 9, we presented the argument for representing grammatical

knowledge as constructions. In this chapter, we will examine the structure of
constructions and their organization in the grammatical knowledge of a speaker.
This section introduces fundamental concepts and descriptive terms for the anal-
ysis of the structure of a grammatical construction. The concepts in this section
form the basis of any syntactic model, although they are combined in different
ways in different syntactic theories. Any grammatical theory can be described as
offering models of representation of the structure of an utterance, and models of
organization of the relationship between utterance structures (presumably, in a
speaker’s mind). The latter are sometimes described in terms of levels of represen-
tation, linked by derivational rules. But construction grammar is a nonderivational
model (like, for example, Head-driven Phrase Structure Grammar), and so a more
general description of this aspect of grammatical theory is ‘organization.’
Different versions of construction grammar will be briefly outlined in §10.2.
We survey four variants of construction grammar found in cognitive linguistics –
Construction Grammar (in capital letters; Kay and Fillmore 1999; Kay et al. in
prep.), the construction grammar of Lakoff (1987) and Goldberg (1995), Cogni-
tive Grammar (Langacker 1987, 1991) and Radical Construction Grammar (Croft
2001) – and focus on the distinctive characteristics of each theory.
10.1.1 Grammatical representation: the anatomy of a construction

Grammatical constructions in construction grammar, like the lexicon in
other syntactic theories, consist of pairings of form and meaning that are at least
partially arbitrary (but see §10.2.1). Even the most general syntactic constructions
have corresponding general rules of semantic interpretation. Thus, constructions
are fundamentally symbolic units, as represented in Figure 10.1 (compare
Langacker 1987:60).
257
CONSTRUCTION
syntactic properties
morphological properties FORM

phonological properties
symbolic correspondence (link)

semantic properties
pragmatic properties (CONVENTIONAL)

MEANING
discourse-functional properties
Figure 10.1 The symbolic structure of a construction
The term ‘meaning’ is intended to represent all of the conventionalized aspects

of a construction’s function, which may include not only properties of the situa-
tion described by the utterance, but also properties of the discourse in which the
utterance is found (such as the use of the definite article to indicate that the object
referred to is known to both speaker and hearer) and of the pragmatic situation
of the interlocutors (e.g. the use of an exclamative construction such as What a
beautiful cat! to convey the speaker’s surprise). We will use the terms ‘meaning’
and ‘semantic’ to refer to any conventionalized function of a construction.
The central difference between componential syntactic theories and construction
grammar is that the symbolic link between form and conventional meaning is
internal to a construction in the latter, but is external to the syntactic and semantic
components in the former (i.e. as linking rules). Figures 10.2 and 10.3 compare
a componential syntactic theory and construction grammar on this parameter,
highlighting in boldface the essential difference in the two models.
NP VP Quant N syntactic component

grammar
lexicon
linking rules
Pred(arg) Qx: Pred(x) semantic component
Figure 10.2 The relation between form and function in a componential syntactic
theory
An overview of construction grammars 259
word NP VP Quant N
grammar
Pred Pred(arg) Qx: Pred(x)
constructions
Figure 10.3 The relation between form and function in construction grammar
In the componential model, the various syntactic structures are organized inde-
pendently of the corresponding semantic structures, as represented by the high-
lighted boxes in Figure 10.2. In construction grammar, the basic linguistic units are
symbolic, and are organized as symbolic units, as represented by the highlighted
boxes in Figure 10.3.1 As a consequence, the internal structure of the basic (sym-
bolic) units in construction grammar is more complex than that of basic units in
the componential model.
The internal structure of a construction is the morphosyntactic structure of sen-
tences that instantiate constructions. For example, a simple intransitive sentence
such as Heather sings is an instance of the Intransitive construction. If we compare
a simplified representation of Heather sings in generative grammar to a simpli-
fied representation of the same in construction grammar, we can see that they
are actually rather similar except that the construction grammar representation is
symbolic.
The box notation used in Figure 10.4b (on page 260) is simply a notational
variant of the bracket notation used in Figure 10.4a (Langacker 1987; Kay and
Fillmore 1999). Thus, we can see that both the generative grammar representation
and the construction grammar representation share the fundamental part-whole or
meronomic structure of grammatical units: the sentence Heather sings is made
up of two parts, the Subject Heather and the Predicate sings.
The brackets in Figure 10.4a are labeled with syntactic category labels, while the
corresponding boxes in the syntactic structure of Figure 10.4b are not labeled. This
does not mean that the boxed structures in Figure 10.4b are all of the same syntactic
type. Construction grammarians assume, of course, that syntactic units belong to
1 Other theories that share construction grammar’s basis in symbolic units are Head-driven Phrase
Structure Grammar (HPSG; Pollard and Sag 1987, 1993), and Semiotic Grammar (McGregor 1997).
However, these theories are not explicitly construction based, although HPSG and Fillmore and Kay’s
version of construction grammar have converged in many respects.
(a) Generative grammar: (b) Construction grammar:
[ [Heather] [sings] ] Heather sings syntactic

NP VP S structure
HEATHER SING semantic

structure
construction
Figure 10.4 Simplified generative grammar and construction grammar
representations of Heather sings
a variety of different syntactic categories. The boxes have been left unlabeled
because the nature of those categories is one issue on which different theories
of construction grammar diverge. That is, we may ask the following question of
different construction grammar theories:
(i) What is the status of the categories of the syntactic elements in construction
grammar, given the existence of constructions?
Beyond the meronomic structure of grammatical units, generative grammar

and construction grammar diverge. First, as we have already noted, construction
grammar treats grammatical units as fundamentally symbolic, that is, pairings of
grammatical form and the corresponding meaning or semantic structure. As a
consequence, the representation of a construction includes correspondence rela-
tions between the form and the meaning of the construction. We will call these
correspondence relations symbolic links.
It will be convenient to use different names for the parts of a syntactic structure
and the parts of a semantic structure. We will call the parts of the syntactic structure
elements and parts of the semantic structure components. Thus, a symbolic link
joins an element of the syntactic structure of a construction to a component of the
semantic structure of that construction. There is also a symbolic link joining the
whole syntactic structure to the whole semantic structure (the middle symbolic link
in Figure 10.4b). This symbolic link is the construction grammar representation
of the fact that the syntactic structure of the Intransitive construction symbolizes
a unary-valency predicate-argument semantic structure. Each element plus corre-
sponding component is a part of the whole construction (form + meaning) as well.
We will use the term unit to describe a symbolic part (element + component) of a
construction. That is, the construction as a symbolic whole is made up of symbolic
units as parts. The symbolic units of Heather sings are not indicated in Figure 10.4b
for clarity’s sake; but all three types of parts of constructions are illustrated in
Figure 10.5 (compare Langacker 1987:84, Fig. 2.8a; Figure 10.5 suppresses links
between parts of the construction for clarity).
(syntactic)
Heather sings element
(semantic)
HEATHER SING component
(symbolic) unit
Figure 10.5 Elements, components and units of a construction
Figure 10.4b has two other relations apart from the symbolic relation: one join-
ing the two syntactic elements and one joining the two semantic components. The
link joining the two semantic components describes a semantic relation that holds
between the two components, in this case some sort of event-participant relation.
Thus, the semantic structure of a construction is assumed to be (potentially) com-
plex, made up of semantic components among which certain semantic relations
hold.
The link joining the two syntactic elements in Figure 10.4b is a syntactic rela-
tion. The syntactic relation does not obviously correspond directly to anything in
the generative grammar representation in Figure 10.4a. This is because the repre-
sentation of syntactic relations in most syntactic theories is more complex than a
simple syntactic link. One layer is the syntactic relation itself, such as the subject-
verb relation holding between Heather and sings in the construction grammar
representation in Figure 10.4b. A second layer is the means of representing syn-
tactic relations. Different syntactic theories use different means for representing
abstract syntactic relations. For example, generative grammar uses constituency
to represent syntactic relations, while Word Grammar (Hudson 1984) uses depen-
dency. The third layer is the overt manifestation of syntactic relations, such as
word order, case marking and indexation (agreement). We strip away the latter
two layers in comparing construction grammar theories.
An important theoretical distinction is made regarding the internal structure
of constructions (Kay 1997). The analysis of syntactic structure is unfortunately
confounded by an ambiguity in much traditional syntactic terminology. We can
illustrate this with the example of the term ‘subject’ in the Intransitive Clause
construction in Figure 10.5, illustrated once again by the sentence Heather sings.
The term ‘subject’ can mean one of two things. It can describe the role of a
particular element of the construction, that is, a part-whole (meronomic) relation

between the element labeled ‘subject’ in the Intransitive construction and the
Intransitive construction as a whole. This is the sense in which one says that
Heather is the ‘subject of the Intransitive Clause’ Heather sings. This part-whole
relation is represented implicitly in (1) by the nesting of the box for Heather inside
the box for the whole construction Heather sings.
(1)
Intransitive
Heather sings construction
Subject Predicate
The subject role defines a grammatical category. But the term ‘subject’ can
also describe a syntactic relation between one element of the construction – the
subject – and another element of the construction – the Verb. This is the sense in
which one says that Heather is the ‘subject of the Verb’ sings. In other words, the
term ‘subject’ confounds two different types of relations in a construction: the role
of the part in the whole, and the relation of one part to another part. The difference
between the two is illustrated in (2):
(2)
Intransitive Construction
role
Subject relation Verb
Different construction grammar theories develop different models of the internal

relations between elements of constructions and components of constructions.
These differences can be elucidated by answering question (ii):
(ii) What sorts of syntactic relations are posited?
10.1.2 The organization of constructional knowledge

Constructions are not merely an unstructured list in construction gram-
mar. Constructions form a structured inventory of a speaker’s knowledge of the
conventions of their language (Langacker 1987:63–76). This structured inventory
is usually represented by construction grammarians in terms of a taxonomic net-
work of constructions. Each construction constitutes a node in the taxonomic
network of constructions.
A taxonomic relation describes a relationship of schematicity or generality be-
tween two constructions. For example, in §9.2 we noted the existence of a schematic
idiom The X-er, the Y-er, and also of a substantive idiom The bigger they come,
the harder they fall. We noted (following Fillmore et al. 1988) that the existence
of the schematic idiom was not incompatible with the existence of the substantive
idiom. A construction grammarian captures the fact that the substantive idiom is an
instance of the schematic idiom by representing this relationship with a taxonomic
link, as in (3):
(3) [The X-er, the Y-er]

|
[The bigger they come, the harder they fall]
Any construction with unique idiosyncratic morphological, syntactic, lexical,

semantic, pragmatic or discourse-functional properties must be represented as an
independent node in the constructional network in order to capture a speaker’s
knowledge of their language. That is, any quirk of a construction is sufficient to
represent that construction as an independent node. For example, the substantive
idiom [S b j kick the habit] must be represented as an independent node because it
is semantically idiosyncratic. The more schematic but verb-specific construction
[S b j kick O b j] must also be represented as an independent node in order to specify
the verb’s argument structure (or in older generative grammar terms, its subcatego-
rization frame). Finally, the wholly schematic construction [S b j Tr Ve r b O b j]
is represented as an independent node.
These constructions are independent but related in terms of schematicity. For
example, several levels of schematicity can be represented between the substantive
idiomatic phrase kick the habit and the most schematic representation of the verb
phrase in (4).
(4) [Ve r bPhrase ]

|
[Ve r b Ob j]
|
[kick Ob j]
|
[kick [the bucket]]
Taxonomic relations between constructions allow construction grammarians

to distinguish and yet relate the grammatical knowledge that is represented by
different formal devices in the componential model of grammar. In (4), the top
two levels in the taxonomy corresponds to the phrase structure rule VP → V
NP in a componential model; the third level corresponds to the subcategorization
frame kick: [ NP]; and the lowest level an idiomatically combining expression
kick the habit, which would be listed in the lexicon in the componential model.
Thus, taxonomic relations complement the uniform representation of grammat-
ical knowledge posited by construction grammar. Taxonomic relations allow a
construction grammarian to distinguish different kinds of grammatical knowledge

while acknowledging the existence of the syntax-lexicon continuum.
Of course, kick the habit has the same argument structure pattern as ordinary
transitive uses of kick, and ordinary transitive uses of kick follow the same argument
structure pattern as any transitive verb phrase. Each construction is simply an
instance of the more schematic construction(s) in the chain [kick the habit] –
[kick Obj] – [TrVerb Obj] . Thus, these constructions can be represented in a
taxonomic hierarchy, as in (5):
(5)
CLAUSE
SBJ INTRVERB SBJ TRVERB OBJ
SBJ sleep SBJ run SBJ kick OBJ SBJ kiss OBJ
SBJ kick the bucket SBJ kick the habit
However, grammatical constructions do not form a strict taxonomic hierarchy.

One of the simplifications in the hierarchy of constructions in (5) is the exclusion of
tense-aspect-mood-negation marking, expressed by auxiliaries and verbal suffixes.
If those parts of an utterance are included, then any construction in the hierarchy
in (5) has multiple parents. For example, the sentence I didn’t sleep is an instanti-
ation of both the Intransitive Verb construction and the Negative construction, as
illustrated in (6):
(6)
SBJ INTRVERB SBJ AUX-n't VERB
I didn't sleep
The sentence I didn’t sleep thus has multiple parents in the taxonomy of con-
structions to which it belongs. This is a consequence of each construction being a
partial specification of the grammatical structure of its daughter construction(s).
For example, the Negative construction only specifies the structure associated with
the subject, verb and auxiliary; it does not specify anything about a verb’s object
(if it has one), and so there is no representation of the object in the Negative
construction in (6).
A construction typically provides only a partial specification of the structure of
an utterance. For example, the Ditransitive construction [Sbj DitrVerb Obj1
Obj2], as in He gave her a book, only specifies the predicate and its arguments.
It does not specify the order of elements, which can be different in different
constructions: compare the Simple Declarative example just given above with
the It-Cleft construction It was a book that he gave her. Nor does the Ditransitive
construction specify the presence or position of other elements in an utterance, such
as modal auxiliaries or negation, whether in a declarative sentence (where they are
preverbal; see [7a]) or an interrogative sentence (where the auxiliary precedes the
subject; see [7b]):
(7) a. He won’t give her the book.
b. Wouldn’t he give her the book?
Hence, any specific utterance’s structure is specified by a number of distinct

schematic constructions. Conversely, a schematic construction abstracts away from
the unspecified structural aspects of the class of utterances it describes.
All versions of construction grammar employ taxonomic relations between con-
structions in the organization of grammatical knowledge. Constructions may be
linked by relations other than taxonomic relations. A third question we may ask
of different construction grammar theories is:
(iii) What sorts of relations are found between constructions?
Finally, the taxonomic hierarchy appears to represent the same or similar infor-
mation at different levels of schematicity in the hierarchy. For example, the fact
that the habit is the direct object of kick in kick the habit is, or could be, repre-
sented in the idiom construction itself [kick the habit], or at any one or more of the
schematic levels above the hierarchy, all the way up to [TrVe r b O b j]. Different
theories of construction grammar have offered different answers to the question of
how information is to be represented in the taxonomic hierarchy of constructions:
(iv) How is information stored in the construction taxonomy?
In §10.2, the answers that various theories of construction grammar give to

questions (i)–(iv) are presented.
10.2 Some current theories of construction grammar
This section surveys current theories of construction grammar in cog-

nitive linguistics. All of the theories conform to the three essential principles of
construction grammar described in chapter 9: the independent existence of con-
structions as symbolic units, the uniform representation of grammatical structures,
and the taxonomic organization of constructions in a grammar. Of course, the exact
means by which constructions and grammatical information are described in each
theory, and the terminology used, varies. In each of the following subsections,
the basic terminology used for the essential construction grammar features, and
the approach to the four questions introduced above, will be presented for each
theory. The different answers to the four questions bring out some current issues of
debate in construction grammar. It should be noted that the different theories tend
to focus on different issues, representing their distinctive positions vis-à-vis the
other theories. For example, Construction Grammar explores syntactic relations
and inheritance in detail; the Lakoff/Goldberg model focuses more on catego-
rization relations between constructions; Cognitive Grammar focuses on semantic
categories and relations; and Radical Construction Grammar focuses on syntactic
categories and typological universals. Finally, the last three theories all endorse
the usage-based model, which is described in chapter 11.
10.2.1 Construction Grammar (Fillmore, Kay et al.)

Construction Grammar (in capitals) is the theory developed by Fillmore,
Kay and collaborators (Fillmore and Kay 1993; Kay and Fillmore 1999; Kay et al.,
in prep.). Construction Grammar is the variant of construction grammar (lower
case) that most closely resembles certain formalist theories, in particular Head-
driven Phrase Structure Grammar, which also calls itself a sign-based theory (i.e. a
theory whose fundamental units are symbolic; Pollard and Sag 1993:15). Neverthe-
less, Construction Grammar conforms to the essential principles of construction
grammar; Fillmore and Kay were among the first to articulate these principles
(Fillmore, Kay and O’Connor 1988). Construction Grammar’s distinguishing fea-
tures are its elaborate, and still evolving, descriptive language for the internal
structure of constructions, which can be only briefly sketched here (the version
described here is essentially that of Kay and Fillmore 1999).
In Construction Grammar, there is a uniform representation of all grammatical
properties, formal and functional, as features with values, such as [cat v] (syntactic
category is Verb) and [gf –subj] (grammatical function is not Subject; Kay and
Fillmore 1999). The value of a feature may itself be a list of features with their
own values. The overall set of features with their values (including features) are
more generally called feature structures. A simple example of a feature structure
is the Verb Phrase construction (Kay and Fillmore 1999:8, Fig. 2). The Verb
Phrase construction may be represented by brackets around the features and feature
structures, as in (8), or by an equivalent box notation, as in (9):
 
(8) [cat v]
 
 [role head] 
 
 [lex +] 
   
 [role filler] 
 
  [loc +] + 
[gf ¬subj]
(9)
cat v
role filler
role head loc + +
lex +
gf subj
The equivalent diagrams in (8)–(9) are read as follows. The two inner boxes
(feature structures) indicate the features of the verb and its complements (if any).
The first box specifies that the first constituent of the VP construction is its head
and that it must be lexical. For example, in found her bracelet the first constituent
is the head of the VP and it is a word, not a larger constituent. The feature-value
pair [cat v] above it is actually a simplification of a more complex feature structure
(Kay and Fillmore 1999:9, n. 13), which specifies that the syntactic category of
the head of the VP, in this case found, must be ‘verb.’ The second box specifies the
complements, if any, of the verb. The + sign following the second box (‘Kleene
plus’) indicates that there may be zero, one or more complements in the VP. In the
VP found her bracelet, her bracelet is the one and only complement. In the VP
construction, the complements are given the role value ‘filler’ (see below). The
feature [loc(al) +] indicates that the complement is not extracted out of the VP.
An example of an extracted, [loc −], complement of find would be the question
word What in the question What did he find?
The internal structure of a construction in Construction Grammar can be most
easily understood by working from the parts to the whole. Minimal parts are words
(or more precisely, morphemes; we will ignore this distinction for now). Each
part has syntactic features, grouped under the feature syn, and semantic features,
grouped under the feature sem. Construction Grammar separates the phonological
features under a feature phon if the construction is substantive. The syn and
sem features are themselves grouped under the feature ss (formerly synsem),
which represents the symbolic structure of that part of the construction. The basic
symbolic structure for Construction Grammar is given in (10):
(10)
syn ...
ss
sem ...
...
phon <...>
(i) What is the status of the categories of the syntactic elements in

construction grammar, given the existence of constructions?
The elements of constructions in Construction Grammar fall into a small
set of atomic category types, such as [cat v] and [gf sbj]. That is, constructions in
Construction Grammar can be described in terms of complex combinations of a
set of primitive atomic units. This is a reductionist model of syntactic structure:
the atomic units are primitive and the complex units are derived. Why are con-
structions not superfluous, then, in Construction Grammar? It is because specific
constructions as a whole will contain syntactic and semantic information that is
not found in the units of the construction that make up its parts. For example, the
What’s X doing Y? or WXDY construction (Kay and Fillmore 1999), illustrated
by What’s this cat doing in here?, possesses a number of syntactic and semantic
properties not derivable from other constructions or the words in the construction.
Its distinctive semantic property is the presupposition of incongruity of the event,
which Kay and Fillmore argue cannot be derived by conversational implicature
(1999:4). The WXDY construction is found only with the auxiliary be and the
main verb do in the progressive (yet the progressive form here can be used with
stative predicates), and excludes negation of do or be, all properties not predictable
from the words, related constructions, or the constructional meaning (1999:4–7).

The manner in which Construction Grammar assembles the parts of a
construction into a whole uses three different sets of features: role, val and rel.
The use of role, val and rel are illustrated in (11), based on the diagram contrasting
roles and relations in (2) and using Heather sings as the example:
(11)
Intransitive Construction
'Heather sings'
[role filler] [role head]
[rel [gf sbj] [theta agt]]

Subject Verb
ARGUMENT 'Heather' 'sings' PREDICATE
[val { [sem {A}] }]
The role feature is used to represent the role of the syntactic element in the
whole. The role feature is associated with each part of a complex construction
and defines syntactic roles such as mod(ifier), filler and head. For instance, the
subject-predicate (Intransitive) construction in (11), Heather sings, has the roles
head for sings and filler for Heather (Kay and Fillmore 1999:13).
In addition to roles, each part of a complex construction has a relation to some

other part of the construction in Construction Grammar. The relations between
parts of a construction are all cast in terms of predicate-argument relations. For
example, in Heather sings, Heather is the argument and sings is the predicate.
The predicate-argument relation is symbolic, that is, both syntactic and semantic.
Semantically a predicate is relational, that is, inherently relates to one or more
additional concepts. In Heather sings, singing inherently involves a singer. The
semantic arguments of a predicate are the concepts to which the predicate re-
lates, in this case, Heather. Syntactically, a predicate requires a certain number
of arguments in specific grammatical functions to it: sing requires an argument
in the subject grammatical function. And syntactically, arguments are related to
the predicate by a grammatical function: in this case, Heather is the subject of
sings.
The remaining two features used to describe meronomic relations in Construc-
tion Grammar, val and rel, are used on predicates and arguments respectively.
The val feature structure is used to indicate the relation of the predicate to its
argument(s), and the rel feature structure is used to indicate the relation of each
argument to its predicate. The val feature is found in the predicate’s representa-
tion. The value of the val feature will be a set, indicated by the set notation {};
the val feature will be a set consisting of more than one member for predicates
with multiple arguments. For the predicate sings in (11), the val set consists of
just one member, namely the singer argument. Construction Grammar indicates
the argument of a predicate by a cross-reference to the set of semantic arguments,
which is part of the sem feature structure. In this example, we simply indicate that
the singer argument corresponds to the argument A in the sem feature structure
for sings.
The rel feature structure in the representation of the argument phrase indicates
what grammatical function the argument is found in, and what semantic role it
should have with respect to the predicate. The rel feature structure takes a syntactic
feature, gf (for ‘grammatical function’), and a semantic feature θ (for ‘thematic
role,’ indicated as ‘theta’ in [11]).2 In (11), the argument Heather’s rel feature
structure has a grammatical function of ‘subject’ and a thematic role of ‘agent.’
Finally, predicates and their arguments in a construction are matched with each
other such that each argument’s rel is matched up with one of the elements in the
val list of its predicate. This is achieved through indexes on the relevant feature
structures in the construction (not indicated in [11]). Kay and Fillmore call this
matching principle the Valence Principle (Kay and Fillmore 1999:10).
2 The version of the feature geometry given in Kay and Fillmore 1999:9, n. 10 includes another syntactic
feature under rel, namely case.
Construction Grammar keeps distinct part-whole relations (role) and part-part

relations (val and rel). Predicate-argument relations are independent of the role
relations each predicate and argument has. For example, in both The book is red
and the red book, red is the predicate and (the) book is the argument. However,
in The book is red, be red is in the head role while in the red book, book is in
the head role. Furthermore, Construction Grammar keeps distinct the val feature
for predicates and the rel feature for arguments. The reason that val and rel are
kept separate is that a single element in a construction can be a predicate taking
arguments and at the same time be an argument for another predicate. For example,
in You should read this, the element read is a predicate taking the argument this,
but is itself an argument of the predicate should (Kay 1997).
The meronomic relations of a construction in Construction Grammar are ana-
lyzed in terms largely familiar from other syntactic theories (head, modifier, pred-
icate, argument), although they are defined somewhat differently. In Construction
Grammar, predicate-argument relations between elements are syntactic and se-
mantic, and they are clearly distinguished from syntactic roles held by elements
in the construction as a whole.

(iv) How is grammatical information stored in the construction
taxonomy?
We address both of these questions together because the answer to (iii)
in Construction Grammar is dependent on the answer to (iv).
Construction Grammar, like all construction grammars, allows taxonomic rela-
tions between constructions. In examining a construction taxonomy such as those
illustrated in (4)–(5), it can be noted that what is more or less the same information
is represented at multiple levels in the taxonomy. For example, the taxonomy in
(4) appears to represent the fact that the object follows the verb at each of the
lower three levels. Redundant representation of information need not be the case,
however. One can represent the fact that the object has the grammatical function
[gf obj] just once, at the highest possible level in the taxonomy – in (4), the [Ve r b
O b j] level. The constructions at the lower taxonomic levels will then inherit
this property by virtue of being an instance of (an instance of) the [Ve r b O b j]
construction. For example, the idiom kick the habit does not separately and redun-
dantly represent the fact that the habit bears the object grammatical function to
kick; it inherits this feature from the [Ve r b O b j] construction.
Following Goldberg (1995:73–74), we will describe a model in which infor-
mation is represented nonredundantly and is inherited as a complete inheritance
model. Construction Grammar is a complete inheritance model (Kay and Fillmore
1999:7–8, 30–31). That is, Construction Grammar represents information only
once in the construction taxonomy, at the highest (most schematic) level

possible.
A consequence of the complete inheritance model is that some constructions
may not be pairings of form and meaning, contrary to the general principle given
in §10.1 that constructions are symbolic (Fillmore 1999:121, n. 11). Consider, for
example, the following examples of constructions in English (Fillmore 1999:126,
122, 123, 121):
(12) Did you understand what I said? Polarity question (positive)

(13) Boy, was I stupid! Subject-auxiliary inversion exclamation
(14) Don’t you even touch that! Emphatic negative imperative
(15) May I live long enough to see the Blessings-Wishes-Curses
end of this job!
All of these constructions have in common the syntactic property that the
auxiliary verb precedes the subject argument. In a complete inheritance model,
one would posit a Subject-Auxiliary Inversion (SAI) construction possessing this
syntactic property, which would be inherited by the constructions in (12)–(15)
(Fillmore 1999). However, there is no common semantic property that is inherited
in all of these constructions. Hence the schematic SAI construction will lack any
semantic specification, and thus will be a purely syntactic construction. Neverthe-
less, this is only a limiting case in a model of grammar that is organized in terms
of symbolic units.
In a complete inheritance model, a construction can inherit the feature structures
of its parent constructions; this is the significance of the taxonomic relation between
constructions in this model. Complete inheritance is an all-or-none relation, and
so the categories defined by a construction taxonomy in Construction Grammar
are classical (see chapter 4).
Kay and Fillmore also allow parts of a construction to inherit feature structures
from another construction (Kay and Fillmore 1999:18; see also Fillmore 1999;
Kay 2002). Kay and Fillmore argue that the nonsubject WH-question construction,
instantiated in Why did she leave him?, is made up of a left-isolated (traditionally
called ‘fronted’) WH question word and an inverted clause. Thus, the nonsubject
WH-question construction as a whole inherits the feature structure of the schematic
left-isolation (LI) construction, while the non-left-isolated part of the construction
inherits the feature structures of the SAI construction:
(16) Nonsubject WH-question construction:

Inherit LI
wh + Inherit SAI
In other words, parts of constructions can be children of other constructions, whose

feature structures they inherit. Thus, Construction Grammar models meronomic
relations between constructions by taxonomic relations between a parent construc-
tion and the corresponding parts of other constructions.
10.2.2 Lakoff (1987) and Goldberg (1995)

Lakoff (1987) develops a variant of construction grammar in his impor-
tant study of the There-construction in English. Lakoff’s analysis emphasizes the
complex, nonclassical structure of the category of There-constructions, in keeping
with his interest in prototypicality and radial category structure. Lakoff’s student
Goldberg also adopts Lakoff’s emphasis on constructional relations in her analysis
of argument structure constructions (Goldberg 1995). Goldberg also addresses the
other issues raised above, either explicitly or implicitly, in the context of analyz-
ing argument structure constructions. But the chief distinguishing characteristic
of Lakoff’s and Goldberg’s version of construction grammar is the exploitation of
nonclassical categorization in the analysis of relations between constructions.

Space prevents us from examining Goldberg’s model of argument linking
in detail; we touch only on those topics relevant to the representation of construc-
tions. Goldberg argues that one should analyze participant roles in complex events
as derived from the event itself, following the principles of frame semantics (see
chapter 2). For example, the participant roles for rob/steal are ‘robber’ and ‘vic-
tim’ (Goldberg 1995:47–48). This analysis of participant roles is an example of
a nonreductionist representation: the complex event or situation is treated as the
primitive unit of semantic representation, and the definitions of the roles in the
events are derived from the situation as a whole.
In contrast, Goldberg’s analysis of syntactic roles and relations in argument
structure constructions is reductionist. As in Construction Grammar, Goldberg
employs a set of atomic primitive grammatical relations such as subject and object,
and primitive syntactic categories such as verb.

This question has not figured centrally in Lakoff’s and Goldberg’s the-
ories. In Lakoff’s study of There-constructions, he represents constructions with
the following parameters of form (Lakoff 1987:489), which allow for relations
between syntactic elements as well as relations between the elements and the
construction as a whole:
r Syntactic elements (e.g. clause, noun phrases, verb etc.)

r Lexical elements (e.g. here, there, come, go, be etc.)
r Syntactic conditions (e.g. linear order of elements, grammatical relations
such as subject and object, optionality of elements etc.)
r Phonological conditions (e.g. presence or absence of stress, vowel length
etc.)
Goldberg’s monograph analyzes argument structure constructions, focusing on

relations between constructions (see immediately below), the semantics of argu-
ment structure, and the linking to syntactic roles. Because of the ambiguity of terms
such as ‘subject’ between role and relation interpretations, Goldberg’s representa-
tion of the syntactic structure of argument structure constructions (e.g. Goldberg
1995:50–55) is compatible with either interpretation.

Schematic constructions in taxonomies represent a categorization of ut-
terances in accordance with certain of their grammatical and semantic properties.
As such, it might be expected that construction taxonomies would display some
of the same properties as conceptual categories, properties that we have discussed
in chapters 4, 5 and 8. Two of the most central properties of category structure are
polysemy and prototype-extension structure. Both of these properties are found in
construction taxonomies.
Lakoff and Goldberg discuss a variety of relationships (links) among con-
structions, including taxonomic relations (Lakoff 1987, Appendix 3; Goldberg
1995:74–81). One of the links Goldberg discusses, the instance link (1995:79–
81), corresponds to the taxonomic links described above. A second type of link,
the subpart link, corresponds to a meronomic link: ‘one construction is a proper
subpart of another construction and exists independently’ (1995:78). This formu-
lation appears to represent the meronomic relation as a distinct type of link, unlike
Construction Grammar where a proper subpart of one construction may be an
instance of another construction. Elsewhere, Goldberg describes all links as inher-
itance links (1995:74–75), but the direction of inheritance in her diagrams is the
opposite to that in Construction Grammar (see 1995:80).
Goldberg also proposes a third type of construction link, the polysemy link, for
subtypes of a construction that are identical in syntactic specification but different
in their semantics. For example, Goldberg argues that the ditransitive construction
[Sbj Verb Obj1 Obj2 ] has a general meaning involving a transfer of possession
of Obj2 to Obj1. However, there are semantic variations on this syntactically
unified construction (1995:38, Fig. 2.2):
(17) S BJ causes OBJ2 to receive OBJ1:

Joe gave Sally the ball.
(18) Conditions of satisfaction imply S BJ causes OBJ2 to receive OBJ1:
Joe promised Bob a car.
(19) S BJ enables OBJ2 to receive OBJ1:
Joe permitted Chris an apple.
(20) S BJ causes OBJ2 not to receive OBJ1:
Joe refused Bob a cookie.
(21) S BJ intends to cause OBJ2 to receive OBJ1:
Joe baked Bob a cake.
(22) S BJ acts to cause OBJ2 to receive OBJ1 at some future date:
Joe bequeathed Bob a fortune.
Goldberg treats the first sense (the one in [17]) as the central, prototypical sense
and the other senses as extensions from the prototype. The extensions from the
prototype inherit the syntactic construction schema from the prototype.
Goldberg does not explicitly argue for a schema subsuming all of the senses
of the ditransitive construction. However, inheritance is a characteristic of taxo-
nomic links, so Goldberg’s analysis suggests that there is a schematic syntactic
ditransitive construction, even if there is not a semantic schema (Goldberg does not
propose a schematic meaning of the ditransitive). In other words, there is a syntac-
tic construction schema that has as instantiations the six senses in (17)–(22), and
the actual successful causation meaning in (17) is the prototype for the ditransitive
construction.
In fact, however, the syntactic schemas for the six subsenses are all slightly
different, because each sense has a distinct subclass of verbs associated with it
(Croft 2003a). Thus, each subsense has associated with it a syntactic schema
specifying the verbs or verb classes that each subsense applies to. In general,
semantic differences in the grammatical constructions such as the ditransitive or
the perfect are likely to have syntactic consequences, and so distinct constructional
senses are likely to also have distinct syntactic schemas.
It is still possible to posit a superordinate ditransitive construction specifying
what is common to all of the subordinate constructions. Such a construction would
have a syntactic schema such as [Sbj DitrV Obj1 Obj2]. The D i t r V category
in the ditransitive construction would be a polysemous category, that is, there are
no necessary and sufficient conditions to describe all and only the ditransitive
verbs in the language. But as we saw in chapter 3, this is true of many linguistic
categories, and so it is not surprising that a syntactic category turns out to be the
same. Likewise, the semantics of the superordinate ditransitive construction would
also be polysemous: although there would be a necessary condition that some
modulated transfer of possession is involved (actual, intended, future etc.), this
condition is not a sufficient condition defining the ditransitive construction’s use.
The most important property of the polysemy analysis is that one construction
sense is central and another is an extension from it. A clear case of extension
from a central sense in constructions is a metaphorical extension, another type
of link proposed by Goldberg, following Lakoff (1987) in his analysis of There-
constructions.
Lakoff argues that many of the extensions of the central There-construction
involve metaphorical extension. For example, the Perceptual Deictic There-
construction, illustrated in (23), involves a number of metaphorical extensions
from the Central Deictic There-construction illustrated in (24) (Lakoff 1987:511,
509):
(23) a. Here comes the beep.
b. There’s the beep.
(24) There’s Harry.
The Perceptual Deictic describes the impending (23a) or just-realized (23b) acti-
vation of a nonvisual perceptual stimulus, for example an alarm clock that is about
to go off. To express this meaning, the Presentational Deictic uses the metaphor of
deictic motion of a physical entity in physical space. The extension of the Central
Deictic to the Perceptual Deictic requires the following metaphorical mappings
(Lakoff 1987:511):
(25) Perceptual Deictic domain Central Deictic domain
nonvisual perceptual space is physical space
percepts are entities
realized is distal
soon-to-be-realized is proximal
activation is motion
A metaphorical extension (or any other semantic extension, for that matter)
need not establish a schema of which the basic construction and the metaphorical
extension are both instantiations. Lakoff’s based-on link, like Goldberg’s polysemy
link, involves inheritance of both syntactic and semantic properties, and so is not
unlike a taxonomic link. Lakoff, however, does not posit a superordinate Deictic
There-construction schema. On the other hand, Goldberg argues that there is a
superordinate schema subsuming both a central construction and its metaphorical
extension (Goldberg 1995:81–89).

Goldberg and Lakoff differ from Construction Grammar in allowing
normal (or default) inheritance (Goldberg 1995:73, citing Flickinger, Pollard
and Wasow 1985). Normal inheritance is a method for accommodating the fact
that much of what we know about a category is not true of every instance of a
category. For example, we know that most birds fly, to the point that if we hear
reference to ‘a bird,’ we will assume that it can fly. Of course, if we are fur-
ther informed that the bird in question is an ostrich or a penguin, or that it has
a broken wing or it is dead, we would cancel that assumption. One model for
representing this information is to store the information FLIES with the category
BIRD, instead of with the many instances of bird species and individual birds
that can fly. The property FLIES is inherited in those cases, but inheritance can
be blocked if it conflicts with information in the more specific case, such as pen-
guins, ostriches, a bird with a broken wing, a dead bird and so on. This is normal
inheritance.3
Lakoff uses normal inheritance in his analysis of There-constructions. Nor-
mal inheritance is part of Lakoff’s based-on link between constructions (Lakoff
1987:508); so does Goldberg (1995:74). For example, Lakoff argues that the
Presentational Deictic There-construction in (26) is based on the Central Deictic
There-construction in (27) (Lakoff 1987:520, 482):
(26) There in the alley had gathered a large crowd of roughnecks.

(27) There’s Harry with the red jacket on.
One of the properties of the Central Deictic is that the verb must occur in
the simple present tense, because the semantics of the Central Deictic is to point
out a referent in the speech act situation (Lakoff 1987:490–91). The Presentational
Deictic is based on the Central Deictic but also specifies that the verb may appear in
a variety of tenses as expressed in auxiliaries (Lakoff 1987:521). This specification
blocks the inheritance of the simple present tense requirement from the Central
Deictic.
Goldberg also allows for the representation of information at all levels in the
taxonomic hierarchy of constructions. Goldberg describes such a model as a full-
entry model (Goldberg 1995:73–74). She gives an example of a situation that
virtually requires a full-entry representation, namely a conflict in multiple inher-
itance. If there are multiple parents, then there will be inheritance from multiple
‘parents’ in the taxonomic network. It may be that the multiple parent nodes have
conflicting specifications of some properties, and this conflict has to be resolved
for the specific instance. Normal inheritance cannot handle this problem. Normal
inheritance adjudicates a conflict in specification between parent and child nodes
in the taxonomy (the child always wins). In multiple inheritance the conflict is
between the two parent nodes, and there is no principled way to choose which
parent would win in a conflict.
3 Construction Grammar eschews default inheritance; instead, default values are left unspecified and
default constructions fill in unspecified values (Fillmore 1999:115, n. 3; Kay 2002:470).
Goldberg gives an example of a conflict in multiple inheritance with the resulta-

tive construction, illustrated in (28), and the verb-particle construction, illustrated
in (29) (Goldberg 1995:97–98):
(28) a. She hammered the metal flat.

b. The metal was flat.
(29) a. He cleaned the mess up.
b. He cleaned up the mess.
Goldberg notes (following Bolinger 1971) that some resultatives allow for word
order variation of the same type as is found in the verb-particle construction:
(30) a. Break the cask open.

b. Break open the cask.
c. The cask is open.
Goldberg proposes that the class of resultatives illustrated with break open in
(30) are instances of the verb-particle construction as well as of the resultative
construction. However, the two parent constructions have conflicting properties.
While the verb-particle construction allows word order variation (compare [29]),
the resultative construction does not:
(31) *She hammered flat the metal.
In this case, break open inherits the word order variation of the verb-particle con-
struction, not the fixed word order of the resultative. Conversely, while the resulta-
tive allows for a simple predication (compare [28b]), the verb-particle construction
does not:
(32) *The mess is up.
In this case, break open inherits the predicability of resultatives, not the ungram-
matical predication of the verb-particle construction.
The two parent constructions of (30) give conflicting specifications as to whether
the word order change is acceptable or not, and whether the predication of the result
phrase is acceptable or not. Goldberg suggests that, in this case, the information
about the specific construction types is provided in the specific construction, even if
it is redundant with the information contained in (one of) the parent constructions;
then the problem of how to resolve the conflict of multiple inheritance does not
arise. In other words, Goldberg argues that a full-entry model in this situation is
desirable.
Is full entry plausible when the information could be represented nonredun-
dantly by inheritance? It might appear to the reader that, a priori, the inheri-
tance model is to be preferred over the full-entry model for reasons of parsimony.
However, most cognitive linguists argue that a cognitively based grammar should
not be constructed in an a priori fashion, because grammatical knowledge is a psy-
chological phenomenon. Clearly, speakers do not store a representation of every
utterance they have ever used or heard. Speakers form schemas that generalize
over categories of utterances heard and used. But it does not necessarily follow
from this observation that speakers store every piece of grammatical knowledge
only once. It does not even necessarily follow that actual speakers form a more
schematic category for every linguistic generalization that clever linguists have
found (see Croft 1998c and §11.2.5).
The principle that information should not be stored redundantly is motivated by
the desire for parsimony in representation. But parsimony in representation simply
pushes complexity to the processes of language use. A complete inheritance model
maximizes storage parsimony, that is, it minimizes the redundant storage of infor-
mation. A complete inheritance model thus requires maximum on-line processing
in order to access and use the information in the production and comprehension of
utterances (see Goldberg 1995:74; Barsalou 1992b:180–81). A full-entry model
maximizes computing parsimony: as much information as possible is stored in
multiple places, so that on-line computation is minimized during production and
comprehension (Barsalou 1992b:180–81; see §§12.1, 12.2.5).
On the whole, the psychological evidence suggests that ‘concepts and properties
in human knowledge are organized with little concern for elegance and [storage]
parsimony’ (Barsalou 1992:180). This does not mean that full entry is to be pre-
ferred in all situations, however: such a model is just as a priori as the inheritance
model. Instead, Goldberg, following Langacker (1987, chapter 10) and other cog-
nitive linguists, advocates a usage-based model, in which patterns of language use
are taken as evidence for the independent representation of grammatical informa-
tion (see especially Goldberg 1995:133–39). In chapter 11, we will examine some
suggested criteria for positing schematic constructions and the degree to which
information is stored redundantly in the mind.
10.2.3 Cognitive Grammar as a construction grammar

Cognitive Grammar is a detailed, carefully worked out theory of syntax
and semantics (Langacker 1987, 1991a, 1991b, 1999, inter alia; see also Taylor
2002). Langacker’s seminal volume (Langacker 1987) gives an abstract exposition
of the framework, and although the word ‘construction’ rarely appears there, and
a completely different set of terms is used, Cognitive Grammar’s model of syn-
tactic representation is a construction grammar model. The distinguishing feature
of Cognitive Grammar as a construction grammar is its emphasis on symbolic
and semantic definitions of theoretical constructs traditionally analyzed as purely

syntactic.
Langacker defines a grammar as a structured inventory of conventional linguistic
units (1987:57). Most conventional linguistic units are symbolic units, with their
two halves, form and meaning.4 Like Construction Grammar, Cognitive Grammar
assumes the symbolic character of the linguistic sign (to use the Saussurean term).
Cognitive Grammar, also like Construction Grammar, nevertheless emphasizes a
uniform representation of constructional form and function. Langacker argues that
all semantic, pragmatic and discourse-functional properties are ultimately concep-
tual, a part of what he calls semantic space, which he describes as ‘the multifaceted
field of conceptual potential within which thought and conceptualization unfold’
(1987:76; see chapter 1).
In the Cognitive Grammar representation of a construction, the symbolic unit
itself must link the form (signifier) and meaning (signified) of the construction.
Langacker describes the link as a symbolic correspondence. Langacker describes
the functional structure (the signified) of the construction as the semantic pole of
a symbolic unit, and its formal structure (signifier) as the phonological pole. The
term ‘phonological pole’ may sound odd: syntax at least is not ‘phonological,’
particularly with respect to schematic constructions. However, Langacker argues
that a schema such as Noun in the description of a construction should be thought
of as phonologically as well as lexically schematic: the schema ranges over pos-
sible nouns, and those nouns are all phonologically contentful, even if their exact
phonological form cannot be specified schematically.5

Cognitive Grammar argues that fundamental syntactic categories such as
Noun, Verb, Subject and Object are abstract (schematic) semantic construals of
the conceptual content of their denotations. Thus, fundamental syntactic categories
4 Langacker also allows for independent phonological and semantic units, but not independent syntactic
units.
5 Cognitive Grammar and Construction Grammar, like Head-driven Phrase Structure Grammar (Pollard
and Sag 1993), eschews the use of phonologically ‘null’ or ‘empty’ elements (see also Kay 2002).
Construction Grammar replaces the concept of a null element with the concept of null instantiation,
that is, some constructions have a feature that indicates that there is a (semantic) argument that is
not formally instantiated (Fillmore 1986b; Fillmore and Kay 1993, chapter 7). Fillmore and Kay
distinguish three types of null instantiation: definite (equivalent to null anaphora), indefinite (as in
The dog ate) and free (corresponding to unspecified adjuncts). Fillmore and Kay argue that null
instantiation is associated with either constructions or individual words; Croft (2001:275–80) argues
that instantiation is associated with constructions only.
have an essentially semantic basis, but in terms of the construal of experience, not
in terms of semantic classes. As described in chapter 3, Langacker has developed
semantic construal analyses of a wide range of syntactic categories, including
parts of speech, grammatical roles (subject and object), the count/mass distinction,
various English tense/aspect inflections and auxiliaries, the English possessives -’s
and of, ergativity, English complementizers and complement types, Cora locatives
and the Yuman auxiliary (see Langacker 1987, 1991a, b, 1999).
One question that can be raised about the Cognitive Grammar analysis of gram-
matical categories is the relationship between the abstract semantic construal def-
initions and the variation in both formal distribution and semantic polysemy of
such categories across languages. It has been suggested that cross-linguistic vari-
ation in putatively universal semantic categories can be accommodated in terms
of conventionalized construal: the same semantic category is found everywhere,
but the construal of specific experiences as belonging to the semantic category is
language-specific:
When we use a particular construction or grammatical morpheme, we thereby

select a particular image to structure the conceived situation for communicative
purposes. Because languages differ in their grammatical structure, they differ
in the imagery that speakers employ when conforming to linguistic convention.
(Langacker 1991b:12)
For example, the English root sick is construed as an adjective or atemporal relation,
that is, summarily scanned (see §3.2), and requires a copula verb be for predica-
tion/sequential scanning; but the equivalent Russian root bol(e)- is construed as a
verb (sequentially scanned) and requires an adjectival derivational suffix to be con-
strued atemporally. But it is not clear that there is any difference between positing
a universal semantic category plus language-specific conventionalized construal
for specific cases on the one hand, and positing a polysemous category with a
semantic prototype and language-specific semantic extensions on the other.

Cognitive Grammar takes a more radical departure from the more familiar
analyses of relations among parts of a construction (Langacker 1987, chapter 8).
The Cognitive Grammar concept of valence, like that of Construction Grammar,
is symbolic. Unlike valence in Construction Grammar, however, valence in Cogni-
tive Grammar is gradient. We will begin by looking at a straightforward predicate-
argument relation, where the Cognitive Grammar and Construction Grammar no-
tions of valence coincide, and then examine the extension of valence in Cognitive
Grammar to other semantic relations.
In a sentence such as Heather sings, sings is a predicate because it is relational.

The relationality of sings is due to the fact that singing requires a singer. Hence,
the semantic structure for sings includes a schematic singer as a substructure. In
Heather sings, Heather is an argument: it is nonrelational and it fills the role of the
singer for sings. Heather is nonrelational because the concept of a person does not
presuppose another concept. Langacker’s term for an argument filling the role of a
predicate is that the argument elaborates the relevant substructure of the predicate.
The substructure that can be elaborated by an argument is an elaboration site or
e-site (Langacker 1987:304). These relations are illustrated in (33):
(33)
HEATHER e-site SINGS
elaboration
As we noted in §10.2.1, a unit in a construction may be simultaneously a pred-

icate and an argument, as is read in You should read this article. How is this
possible? It is because the event of reading elaborates a substructure of the modal-
ity expressed by should, and the thing read, this article, elaborates a substructure of
the event of reading. Hence, predicate and argument status – valence – is relative:
predicate and argument status depend on what two semantic structures are being
compared.
Not only is valence relative, it is gradient. In a sentence such as (34), I and what
I am reading are traditionally analyzed as complements of read while on the train
is an adjunct to read (we ignore the progressive be in this example):
(34) I was reading this on the train.
Complements are arguments of a predicate: reading inherently involves a reader

and a thing read. Adjuncts are predicates and their head is the argument: on the
train inherently involves a Figure (the event) whose location is described by the
spatial relation. Hence, read elaborates a substructure of on the train.
But this description is an oversimplification. Reading is a localizable activity:
reading takes place in a location, as well as involving a reader and a thing read. This
is not true of all predicates; one cannot say for instance that *John was widowed on
the train. Hence the location of the reading event is a substructure of the semantic
structure of read, and on the train also elaborates that substructure of read. The
solution to this apparent paradox is that the substructure of read elaborated by on
the train is much less salient in the characterization of the reading event than the
substructures of read elaborated by I and this. Conversely, the substructure of on
the train that is elaborated by read is highly salient in the characterization of the
spatial relation. On the train is more of an adjunct of read than a complement
because read elaborates a salient substructure of on the train, whereas on the train
elaborates a not very salient substructure of read. The relative strength of the two
relations is illustrated in (35):
(35)
FIGURE READER
TRAIN TEXT
ON LOCATION
dependent
READ
autonomous
Langacker use the terms autonomous and dependent to describe the gradient
reinterpretation of the predicate-argument distinction: ‘one structure, D, is depen-
dent on the other, A, to the extent that A constitutes an elaboration of a salient
substructure within D’ (1987:300). Conversely, A is autonomous relative to D to
the extent to which it does not elaborate a salient substructure of D. In (35), on the
train is dependent on read because read elaborates the highly salient figure role
of the locative relation on the train. Conversely, read is autonomous relative to on
the train because on the train elaborates only the not very salient substructure of
the location of the reading event.
The Cognitive Grammar analysis of concepts of ‘head,’ ‘modifier’ and so on are
both similar and different from the analysis in Construction Grammar. In Construc-
tion Grammar, the roles represent a relation between the parts of a construction
and the whole, and are defined syntactically. In Cognitive Grammar, the analo-
gous concepts also represent a relation between the parts of a construction and the
whole, but they are defined semantically and symbolically.
Cognitive Grammar defines a semantic relation between part and whole as the
profile determinant: the profile determinant is the part of the construction whose
semantic profile the whole construction ‘inherits’ (Langacker 1987:289). The pro-
file is the concept designated by the unit, against the background knowledge pre-
supposed by that concept (see chapter 2). Langacker combines the concepts of
profile determinacy and autonomy/dependence to define ‘head,’ ‘complement’
and ‘modifier’ in the intuitively expected way (1987:309). A head is a dependent
predication that is a profile determinant; a complement is an autonomous predi-
cation that is not a profile determinant; and a modifier is a dependent predication
that is not a profile determinant.

Langacker advocates what he calls a unified approach to categorization
(1987, chapter 10). A category has a nonclassical structure, in that there is typically
a prototypical member or set of members, and nonprototypical members are cate-

gorized by extension from the prototypical members. However, it is also possible
for there to exist a schema subsuming both prototype and extension, which has
a classical category structure, with necessary and sufficient conditions specifying
its instances. Langacker’s model of categorization is of course applied also to con-
structions. Hence, for Langacker, as for Lakoff and Goldberg, one may have both
construction schemas and also nonclassical relations between constructions, such
as prototype-extension relations, including metaphorical extensions.

Cognitive Grammar is a usage-based model, in which the establishment
of schematic constructions is the result of language use as described briefly in
§10.2.2 and in more detail in chapter 11.
10.2.4 Radical Construction Grammar

Radical Construction Grammar (Croft 2001) was developed to account for
typological variation in a construction grammar framework, and to address basic
questions of syntactic argumentation. Radical Construction Grammar adopts the
nonclassical category structure and the usage-based model of the Lakoff-Goldberg
theory and Cognitive Grammar. Radical Construction Grammar takes a thoroughly
nonreductionist approach to constructions, and rejects autonomous syntactic rela-
tions between elements in a construction. Radical Construction Grammar adopts
the usage-based model, and brings in the semantic map model and the notion
of a syntactic space from typological theory to provide organizing principles for
constructions.
(i) What is the status of the categories of the syntactic elements, given
the existence of constructions?
The standard analysis of meronomic relations between syntactic struc-
tures is reductionist (§11.2.1): a construction such as the intransitive or transitive
construction is made up of parts, and those parts are themselves defined inde-
pendently of the constructions in which they occur. For example, various clausal
constructions have verbs, which are analyzed as belonging to the same part of
speech no matter what construction they occur in. This analysis is motivated in
part because they have the same inflections (present in third person singular -s and
non-third person singular zero, past in -ed or other allomorphs):
(36) Present third singular:

a. Intransitive: Toni dances.
b. Transitive: Toni plays badminton.
(37) Present non-third singular:

a. Intransitive: We dance-Ø.
b. Transitive: We play-Ø badminton.
(38) Past:
a. Intransitive: We danced.
b. Transitive: We played badminton.
In other words, the same units occur as the parts of many different constructions.
Ultimately, the decomposition of a construction will lead to a set of basic or
primitive elements that cannot be analyzed further, and out of which constructions
are built. These atomic elements include syntactic categories such as Verb or Noun
and relations such as Subject or Object and so on. A model of grammatical structure
of this type is a reductionist model: more complex structures are treated as built up
out of primitive and ultimately atomic units. In the example given here, the atomic
units are the basic categories and relations.
The reductionist model has a significant shortcoming: it does not capture certain
empirical facts about the distribution of words. For example, while many English
verbs occur in either the transitive or intransitive constructions, many others do
not:
(39) a. Judith danced.

b. Judith danced a kopanica.
(40) a. Judith slept.
b. *Judith slept bed.
(41) a. *Judith found.
b. Judith found a 20 dollar bill.
One solution is to divide Verbs into Transitive Verbs and Intransitive Verbs.
If so, then a decision must be made about verbs such as dance, which occur
in both constructions: do they simultaneously belong to both subclasses? or do
they form a third distinct class? One effect of dividing Verbs into Transitive
Verbs and Intransitive Verbs is that one is essentially defining the categories in
terms of the construction they occur in, Transitive or Intransitive. These problems
are multiplied in cross-linguistic comparison, where the variation found is more
extreme (Croft 2001).
One can deal with such problems in the reductionist model by adding syntactic
features that prevent certain category members from occurring in the unacceptable
constructions, as in (40b) and (41a). Again, the effect is that one is introducing a
feature that specifies the category in terms of the construction it occurs in/does not
occur in (in this case, Transitive and/or Imperative and/or VP Conjunction).
Radical Construction Grammar takes a different approach to the relations of
constructions to their parts. It takes the constructions as the basic or primi-
tive elements of syntactic representation and defines categories in terms of the
constructions they occur in. For example, the elements of the Intransitive construc-
tion are defined as Intransitive Subject and Intransitive Verb, and the categories are
defined as those words or phrases that occur in the relevant role in the Intransitive
construction. In other words, Radical Construction Grammar rejects the existence
of atomic schematic units (see Table 9.2 in §9.4), because atomic schematic units
are defined independently of constructions. This differentiates Radical Construc-
tion Grammar from reductionist theories.
Radical Construction Grammar is a nonreductionist model because it takes the
whole complex structure as basic and defines the parts in terms of their occur-
rence in a role in the complex structure. In effect, Radical Construction Grammar
takes to its logical conclusion one of the strategies for handling these problems
in reductionist theories, namely the subdividing of classes and the employment of
syntactic features that essentially specify which constructions a particular word or
phrase occurs in (see Croft 2001, chapter 1).
Constructions are individuated like any other conceptual object, by categoriza-
tion. Constructions possess formal features, including word order, patterns of con-
tiguity and specific morphemes (or very small classes of morphemes) in particular
roles. Constructions are also symbolic units, and typically possess discrete mean-
ings. Radical Construction Grammar assumes a nonclassical category model, and
allows for prototypes and extensions of constructions, as well as the possibility of
gradience between construction types.

Radical Construction Grammar, like Construction Grammar and Cogni-
tive Grammar, represents the role of a part of a construction in the whole construc-
tion. Radical Construction Grammar differs from Construction Grammar in that it
defines relations between parts of a construction in purely semantic terms, that is,
there are no syntactic relations in Radical Construction Grammar.
One motivation for the Radical Construction Grammar analysis is that relations
between syntactic elements are not strictly necessary in a construction grammar
framework, from the point of view of language comprehension. Consider the phrase
the song, illustrated in (42) below with the semantic relation between [DEF] and
[SONG] now indicated by a link (labeled r):
(42)
the song
s s s
r SONG
DEF
If a hearer recognizes the phrase the song as an instance of the construction

[[DEF/the] [THING/Noun]] – that is, the hearer can retrieve the semantic struc-
ture of the whole construction, and identifies the elements of the construction (i.e.
the words the and song), and can identify the corresponding components of the
semantic pole (i.e. [DEF] and [THING]), then the hearer can identify the semantic
relation r by virtue of the semantic relation between [DEF] and [THING] in the
semantic pole of the construction. Hence the hearer need not rely on any syntactic
relation between the and song.
In Radical Construction Grammar, the various morphosyntactic properties that
are taken to express syntactic relations in other theories – case marking, agree-
ment, adpositions, word order, contiguity and so on – are interpreted as expressing
the symbolic links from the elements in the phonological pole of the construc-
tion to their corresponding components in the semantic pole of the construction.
The evidence for this analysis of symbolic relations would take us too far afield
(see Croft 2001, chapters 5–6). We mention here only two cross-linguistically
widespread phenomena arguing against syntactic relations. The relationship be-
tween putative syntactic relations and semantic relations is noniconic in many
cases (Croft 2001:206–20). This fact defeats attempts to construct general map-
ping relations between syntactic structures and semantic structures; if there are no
syntactic relations, this problem disappears. More seriously, the absence of ele-
ments allegedly related syntactically, such as the absence of controllers of agree-
ment, and the absence of the morphosyntactic expression of the alleged syntactic
relation (Croft 2001:226–33), makes it impossible to represent syntactic relations
without ad hoc devices. Again, if syntactic relations are abandoned, this problem
disappears.
The morphosyntactic properties that appear to indicate syntactic relations in fact
aid the hearer in identifying the role that construction elements fill in the mean-
ing of the construction as a whole (Croft 2001:233–36). Also, the combination of
morphosyntactic properties in an utterance taken as a whole aid the hearer in iden-
tifying a construction (Croft 2001:236–37). For example, the Gestalt combination
of auxiliary be, the past participle form of the verb, and the preposition by, in the
proper syntactic combination with the subject phrase, the verb and the oblique
phrase, uniquely identifies the passive construction, while the individual elements
identify the action (verb inflection and position after auxiliary), the agent (by plus
oblique phrase) and patient (subject position). In other words, the syntactic prop-
erties that seemingly encode syntactic relations in fact encode symbolic relations,
between individual elements and components of the construction and between the
constructional form as a whole and its meaning.
Thus, in Radical Construction Grammar, concepts such as ‘head,’ ‘argument’
and ‘adjunct’ that are syntactically defined in other theories must be given semantic
definitions, as in Cognitive Grammar (Croft 2001, chapter 7). Moreover, such

definitions must ideally be defined relative to constructions, rather than individual
components. For example, Croft argues that the Cognitive Grammar concept of
profile determinacy (§10.2.3), which is defined in Langacker 1987 as the element
whose profile determines the profile of the construction as a whole, is better defined
as profile equivalence, that is, the element whose profile most closely matches the
profile of the construction as a whole (Croft 1996:51–53; 2001:254–57). Profile
equivalence allows for the cases where a construction has no profile determinant
or more than one profile determinant. For example, a conjoined phrase such as
Bill and Tim profiles a pair, but none of its elements has a profile matching that
of the construction as a whole. Also, in the determiner-noun construction, the
construction as a whole profiles the referent, but so do both elements. In the
song, for example, song profiles the entity as an instance of a type or class, and
the schematically profiles the entity as an entity in the shared knowledge of the
interlocutors.

As a nonreductionist model, Radical Construction Grammar makes a
radical shift in the conception of grammatical structure. Radical Construction
Grammar does not posit a set of atomic primitive elements out of which construc-
tions are built. Instead, complex constructions are the basic units of grammatical
representation, and the categories defined by the parts of the construction are de-
rived. However, the effect of the nonreductionist hypothesis on the organization
of grammatical knowledge in the constructional network is minimal.
In Radical Construction Grammar, each part (unit) of a construction constitutes
a category whose members are defined solely by their occurrence in that role in
the construction. In order to differentiate categories, we append the name of the
construction to the labels for each unit in the construction. A representation of the
Intransitive and Transitive constructions is given in (43):
(43)
IntrSbj IntrV TrSbj TrV TrObj
The establishment of a category Verb is a linguistic generalization over the cat-

egories IntrV and TrV. This generalization is thus a taxonomic relationship, with
Verb superordinate to IntrV and TrV. However, any superordinate category, such as
Verb, must be linguistically motivated. The motivation for a superordinate category
such as Verb must be its occurrence as the category in some other construction.
For example, the standard motivation for positing a category Verb is the ability of
its members to be inflected with the tense/agreement suffixes. In a construction
grammar, this linguistic fact is essentially another construction, the morphological

construction [MVerb-TA]. We use the label MVerb to emphasize that this cate-
gory is defined by a morphological construction, namely its occurrence with the
tense/agreement suffixes (abbreviated TA):
(44)
MVerb -TA
IntrSbj IntrV TrSbj TrV TrObj
In other words, Radical Construction Grammar represents meronomic rela-

tions between constructions in a similar way to Construction Grammar (§10.2.1),
namely as instances of a schema.6 That is, the treatment of meronomic relations is
not a distinctive characteristic of nonreductionist models. The primary difference
between a nonreductionist model such as Radical Construction Grammar and a
reductionist model such as Construction Grammar is that the latter uses syntactic
features and values for roles that are defined independently of the constructions in
which the units occur. (In addition, Radical Construction Grammar differs from
Construction Grammar in allowing for nonclassical relations between construc-
tions, as in the Lakoff-Goldberg and Cognitive Grammar theories.)

Although Radical Construction Grammar is identical to Construction
Grammar in its handling of part-whole relations between constructions, it is like
Cognitive Grammar and the Lakoff-Goldberg model in that it allows for redundant
representation of grammatical information in accordance with the usage-based
model (see §10.2.2 and chapter 11).
One salient feature of Radical Construction Grammar’s organization of gram-
matical knowledge is derived from typological theory but conforms to the usage-
based model. This is the semantic map model. In the semantic map model, con-
structions are mapped onto a conceptual space according to their function, and
thus constructions can be related to one another by virtue of having overlapping
or neighboring functions in the conceptual space. The semantic map model is
described in relation to the usage-based model in §11.3.3.
6 More precisely, Radical Construction Grammar allows parts of constructions to be instances of a

part of another construction (as in [44]), as well as allowing them to be instances of another whole
construction. It does not appear that Construction Grammar allows the former possibility.
A second salient feature of Radical Construction Grammar, also inspired by

typological research, is the introduction of the notion of a syntactic space (Croft
2001, chapters 8–9). Constructions that are functionally similar or identical across
languages (and sometimes within a single language) vary considerably in their
grammatical properties, so that, for example, one cannot posit a universal passive
construction based on a fixed set of grammatical properties. The typical European
passive has the grammatical properties of patient as Subject, agent as Oblique (or
prohibited, in many languages), and a special passive verb form distinct from the
active:
(45) The children [patient = Subject] were taken [distinct verb form] to school by
their parents [agent = Oblique].
This construction is by no means universal (even setting aside the problems

of defining Subject etc. across languages): many structurally distinct voice con-
structions occur cross-linguistically, three of which are illustrated in (46)–(48)
(examples from Croft 2001:292–94):
(46) Upriver Halkomelem: patient = Subject, agent not Oblique, distinct verb form
tə́s -1 əm θ úλ́ ’à tə swı́yə qə
bump.into -ACCID -3SG.PASS she ART man
‘She was bumped into by the man’
(47) Bambara: patient = Subject, agent = Oblique, verb form not distinct:
o fo’ra dugutigi fè
3sg greet’compl.intr chief with
‘S/he was greeted by the chief.’
(48) Maasai: patient not Subject, agent prohibited, distinct verb form:
aa- dɔ l -i
1sg.obj- see -pass
‘I am seen.’
Instead of representing constructions as discrete universal formal types that oc-

cur across languages, Radical Construction Grammar represents constructions as
language-specific structures occupying positions in a syntactic space defined by
the structural properties that vary across languages (in this case, patient coding,
agent coding and verb form). The syntactic space allows one to formulate uni-
versals of the relationship between formal properties of constructions and their
function, of the sort discovered by typologists. For example, the more ‘passive-
like’ voice constructions are associated cross-linguistically with higher-animacy
and higher-topicality patients, where distributional restrictions or structurally con-
trasting voice constructions exist (Croft 2001, chapter 8).
10.3 Conclusion
This chapter has presented the essential features of a construction gram-

mar, and some of the different positions on basic representational issues found in
different theories of construction grammar. Unfortunately space prevents us from
describing analyses of more specific grammatical phenomena in the various the-
ories, such as argument structure, so-called movement phenomena, information
structure constructions, word order variation and so on; or specific grammatical
constructions (other than those illustrated in chapter 9). There are also a variety
of issues of representation and the processes that use that representation which
have not been fully addressed by the construction grammar models at the time of
writing. Nevertheless, there is a large and growing body of construction grammar
analyses of a wide range of grammatical constructions, and a lively debate on
basic issues of grammatical representation and process in a construction-based
approach.
11
The usage-based model
11.1 Grammatical representation and process
Grammatical knowledge is not merely a representational structure in the

mind of a speaker. In a recent survey of knowledge representation models in
psychology, Markman argues that there are four basic elements to a model of
knowledge representation (1999:5–10). One element is the representing world,
that is, the domain of the representations themselves. In the cognitive linguistic
approach to language, the representing world is of course the mind. A second
element is the represented world. For grammatical knowledge, the represented
world is utterances, that is, the form of utterances and their meaning in the discourse
context. The third and fourth elements in Markman’s analysis are some mechanism
to link the representing world to the represented world, and processes using the
representation. This last element is particularly important:
It makes no sense to talk about representations in the absence of processes . . . Only
when there is also a process that uses the representation does the system actually
represent, and the capabilities of a system are defined only when there is both a
representation and a process. (Markman 1999:8)
The primary processes in which grammatical knowledge is involved are com-

munication – the production and comprehension of utterances; the acquisition of
grammatical knowledge by children and by adults; and the changes in grammatical
knowledge of speakers over time. These processes link the representing world –
the grammatical knowledge – and the represented world – the world of utterances
and their meanings. These processes are the locus for the link between the rep-
resenting and represented worlds, Markman’s third element of a representation
model.
Many cognitive linguists propose a usage-based model for language use, lan-
guage acquisition and language change (§§11.2.2–11.2.4; see inter alia Langacker
1987, chapter 10; Barlow and Kemmer 2000; Bybee and Hopper 2001).1 The
1 Construction Grammar, on the other hand, is intended to be a competence model (Paul Kay, pers.
comm., 1999).
291
usage-based model contrasts with the traditional structuralist and generative mod-
els of grammatical representation. In the structuralist and generative models,
only the structure of the grammatical forms determines their representation in
a speaker’s mind. For example, the traditional models make a sharp distinction
between regular and irregular word forms. Regular inflected word forms, such
as the English plural form boy-s, are derived by a highly general rule forming the
plural from the singular, because the structural relationship between boy and boys,
namely addition of -s, allows this possibility. Irregular word forms, such as the
plural form feet, do not have a straightforward structural relationship linking the
singular and the plural. Since they cannot be derived by a general rule, irregular
plural word forms are therefore listed in the lexicon.
In the usage-based model, properties of the use of utterances in communica-
tion also determine the representation of grammatical units in a speaker’s mind. In
particular, two usage-based properties are assumed to affect grammatical represen-
tation: the frequency of occurrence of particular grammatical forms and structures,
and the meaning of the words and constructions in use. In §11.2, we discuss four
hypotheses of the usage-based model derived from this assumption, and the evi-
dence for those hypotheses in morphology, the area of grammar most intensively
studied in the usage-based model (Bybee 1985, 1995, 2001). In §11.3, we discuss
how the hypotheses of the usage-based model can be applied to the representation
of syntactic constructions, their acquisition and syntactic change.
11.2 The usage-based model in morphology
11.2.1 Entrenchment and representation of word forms

The primary factor determining the independent storage of word forms in
the usage-based model is the frequency of occurrence of the word form in language
use, that is, the token frequency of the word form. The hypothesis is that each
time a word (or construction) is used, it activates a node or pattern of nodes in the
mind, and frequency of activation affects the storage of that information, leading
to its ultimate storage as a conventional grammatical unit. A word form that occurs
frequently enough in use to be stored independently is described as entrenched
(Langacker 1987:59–60). Entrenchment comes in degrees, even beyond the mini-
mum threshold required for independent storage.2 In the usage-based model, the
entrenchment of word forms is possible even if the word form is predictable from
a more schematic grammatical representation. For example, the plural form boys
2 Bybee has used the terms ‘autonomy’ and ‘lexical strength’ to refer to degree of entrenchment in her
work (1985, 1995).
The usage-based model 293
may be entrenched even though it is a regular instance of the noun plural schema
[Noun-s] because it has a high token frequency. In contrast, it is less likely that
the plural form cornices is entrenched, because its low frequency of occurrence is
probably insufficient to lead to the storage of this form independently of the base
form cornice and the noun plural schema [Noun-s]. This first hypothesis of the
usage-based model is summarized in (1):
(1) Hypothesis 1: the storage of a word form, regular or irregular, is a function of its
token frequency.
It is clear that irregular forms must be independently stored in any model of

morphology, since they are not entirely predictable from a more general schema. In
the structuralist/generative models, irregular forms are simply listed in the lexicon.
The usage-based model predicts that irregular word forms will be found among
the more frequent words in the lexicon. The more frequent words are those more
likely to be entrenched, and hence irregularities can survive in such forms. If an
irregular form is not frequent enough, or declines in frequency of use, then it will be
regularized: its representation will not be sufficiently entrenched and reinforced
through use, and so the regular schema will take over in the production of the
relevant inflection. The structuralist/generative model, in which there is no effect
of frequency, predicts that there should be an even distribution of irregular word
forms across the lexicon.
In this case, the evidence clearly favors the usage-based model. Irregular in-
flectional forms are consistently found among the higher frequency words in the
lexicon. For instance, it is not surprising that the verb with the most irregular per-
son forms in English is the extremely frequent verb be (am, is, are; was and were –
the only past forms in English that distinguish person).
There is also evidence that low-frequency irregular inflected forms are regu-
larized, while high-frequency irregulars resist regularization to a greater degree.
For example, Bybee and Slobin examined the English verbs that form the past
tense by changing the final stem consonant from d to t, as in build/built (Bybee
and Slobin 1982:275). They compared the list of verbs in Jespersen’s historical
grammar of English (Jespersen 1942) to the same verbs listed in the American
Heritage Dictionary; the verbs with the American Heritage Dictionary past tense
forms are given in (2):
(2) past in t only: past in ed or t: past in t or ed:

bend blend rend
lend geld past in ed only:
send gird wend
spend not listed in AHD:
build shend ‘to shame’
Bybee and Slobin observe that the forms that retain the irregular past in t are all
frequent, while the others are rare (in fact, one has dropped out of the language).
Further dynamic evidence that irregularity is correlated with frequency is the
evidence that low-frequency irregular forms are more likely to be regularized in
production. For example, Bybee and Slobin discovered that there was a signifi-
cant rank order correlation between the likelihood of regularization of irregular
past tense forms by preschool children and the token frequency of the verb in the
adult caretaker’s speech, such that lower token frequency correlated with a high
regularization rate (Bybee and Slobin 1982:270). Bybee and Slobin found signif-
icant correlations between token frequency and regularization for some though
not all irregular verb classes in experimental production tasks given to third-grade
children and adults (1982:270–71).
A more indirect piece of evidence that regular forms are stored independently
under some circumstances is that a regularly inflected word form, or a regularly
derived word form, may diverge semantically from its parent word. For exam-
ple, ‘something can be dirty without involving real dirt at all . . . someone can
soil an item without being anywhere near real soil’ (Bybee 1985:88). Examples
of divergence of a former inflectional form are clothes, formerly the plural of
cloth (Bybee 1985:91) and shadow, formerly an Old English oblique case form
of shade (Croft 2000:36). Presumably, semantic divergence presupposes the in-
dependent representation of the inflected form, which then is free to diverge in
meaning.
This evidence regarding irregular word forms has led some generative linguists,
including Pinker and colleagues, to accept that frequency effects associated with
degree of entrenchment are found with irregulars. Hence, they accept that the usage-
based model is valid for irregularly inflected word forms. However, Pinker and
colleagues argue that regularly inflected word forms are not sensitive to frequency
effects. Instead, regularly inflected word forms are represented by a grammatical
rule based only on the structure of the word forms and not any properties of their
use (Pinker and Prince 1994; Marcus et al. 1992). This model is called the dual-
processing model of grammatical representation.3
Evidence from an experiment on the more regular third person present inflec-
tion in English indicates that low frequency regular forms are not stored in the
lexicon, because they do not exhibit gang effects (Stemberger and MacWhinney
1988:111–12). Gang effects are effects on stored word forms of phonologically
3 In fact, the traditional model, in which regularly inflected forms are generated by a structural rule
and irregular forms are listed in the lexicon, is also a dual-processing model. The only innovation
in the model proposed by Pinker and colleagues is that the irregulars conform to the usage-based
model’s predictions, instead of merely being listed in the lexicon.
similar stored word forms; if word forms are not stored, gang effects would not
occur. The absence of gang effects has been taken to imply that regularly inflected
forms are not stored (see, for example, Prasada and Pinker 1993). Other experi-
ments, however, do suggest that high frequency regular inflected forms are stored.
Stemberger and MacWhinney conducted an experiment in which subjects were
required to produce past tense forms of regular verbs at high speed, and errors
occurred significantly less often on high frequency regular past tense forms, im-
plying that the high frequency regular past tense forms are stored (Stemberger and
MacWhinney 1988:106). Bybee reports an experiment conducted by Losiewicz
(1992) which provides some evidence of a frequency effect for regular forms
(Bybee 1995:450–51). Losiewicz observed that the acoustic duration of word-
final /t/ or /d/ is shorter if it is part of the word than if it is the regular past tense
ending (e.g. rapt vs. rapped). If the difference in acoustic duration is due to storage
of the word form, then high-frequency regulars should have shorter final /t/ or /d/
than low-frequency regulars. In a sentence reading task, subjects had an average
7 msec difference in duration between high-frequency and low-frequency final /t/
or /d/ duration, which was highly significant.
The evidence reported in the last paragraph is compatible with a model in which
high-frequency regular word forms are stored but low frequency regular word
forms are not. The results suggest that frequency affects the storage of regular
word forms, and supports Hypothesis 1.4
11.2.2 Regularity, productivity and default status

In a structuralist/generative model, regularity is modeled by a rule. A rule
is generally analyzed as an operation over strings, such as the phonological string
of a word form or the syntactic strings of words and constituents of a construction.
Affixation, such as the suffixation of the English past tense suffix allomorphs
/t, /d/ or / d/, is a relatively simple operation. Other rule-governed operations
are more complex: they may involve internal changes to a word form, or other
changes.
In a usage-based model, a simple rule such as the addition of the past tense would
be represented by a schema of the sort used for the representation of constructions
in chapter 10. The schema for the English past tense would be represented as
something like [Ve r b-ed] (ignoring for now the three different allomorphs of
4 It is possible that the absence of a frequency effect for regulars is what one would expect even in a
usage-based model; activation network models trained on regular inputs do not display a gang effect
(Daugherty and Seidenberg 1994).
the past tense). The representation of a generalization as a schema conforms to

the principle that all grammatical knowledge is represented in a uniform fashion
(§9.4).
There are some differences in what can be represented as rules vs. schemas,
which will be discussed in §11.2.3. In this section, however, we will look at what
determines regularity, rather than how regularity is to be represented.
One property of regular inflections is the productivity of the inflection. As the
name implies, a productive inflection is one that is applied to almost any seman-
tically and phonologically appropriate word form, including new forms coined or
borrowed into the language. In the rule-based model, productivity is the open-ended
scope of application of the rule, within the phonological and semantic constraints
imposed by the inflection. As we will see, however, the interpretation of productiv-
ity in the usage-based model has led to the clarification of some hidden distinctions
in the definition of regularity as productivity.
In a usage-based model, a productive inflection is one in which the schematic
representation of the inflection is entrenched. The difference between the -en plural
of oxen and the -s plural is that the only entrenched form in the former is [oxen],
while in the latter case a schematic construction [N OUN -S ] is entrenched. But the
productivity of the regular plural form -s implies the existence of a general schema
that can easily be combined with a particular noun to form a plural. From a usage-
based point of view, then, the question is: what factors determine the entrenchment
of a schematic morphological construction?
Bybee argues that the productivity of a schema is a function of type frequency
(Bybee 1985:132–34; 1995):
(3) Hypothesis 2: The productivity of a schema is a function of the type frequency of

the instances of the schema.
Type frequency is the number of different word forms that are instances of a
particular schema. The type frequency of the English past tense schema [Ve r b-
ed] is thus the number of regular past tense verbs in English. The English past
tense suffix -ed is highly productive under this account. There is a vast number of
lower frequency verbs with the -ed past tense suffix which reinforce the past tense
schema [Ve r b -ed]. In fact, of course, the past tense schema has three allomorphs,
/t/, /d/ and / d/. But each of these phonologically defined schemas has a high type
frequency of low token frequency instances, so each allomorph is highly productive
for its phonologically defined class.
There is another important aspect of productivity: one must be able to form a
coherent schema (Bybee 1995:430). That is, there must be enough resemblance
between the types that contribute to the entrenchment of the schema that one
can form a schema in the first place. The closer the resemblance, the more en-
trenched is the schema. Each allomorph of the English productive past tense is a
coherent schema, that is, for the past tense suffix each allomorph defines a phono-
logically and semantically coherent category, namely [-t/PAST], [-d/PAST] and
[- Sd/PAST]. Moreover, the three allomorphs possess a phonological family re-
semblance, reinforced by their complementary phonological distribution (in terms
of the final segment of the verb stem) and identity of meaning. Because of the
family resemblance a speaker may form a superordinate category, notated here
[-ed/PAST] with the orthographic representation standing in for the phonological
schema.
Bybee argues that instances of a schema that have a high token frequency will
not contribute to the productivity of a schema (1985:132–34). Instances with a
high token frequency are strongly entrenched (1995:434). Only the entrenched
specific word form will be activated in language use and thus will not reinforce the
superordinate schema. On the other hand, word forms with a low token frequency
will not be as strongly entrenched (if they are entrenched at all; see §11.2.1).
Bybee argues that low frequency word forms will contribute to the entrenchment
of a schematic representation of the inflectional ending that applies across many
different word forms, including new forms. However, the examples that Bybee
gives in support of this hypothesis do not fully separate token frequency and type
frequency. Instances of a productive schema with a high token frequency, such as
the most common regular English past tense forms, are swamped by the number of
instances with a low token frequency. One would have to find a conjugation class
in a language where excluding the high token frequency instances results in too
low a type frequency to make the schema productive, but including them would
result in a high enough type frequency to make the schema productive. Hence,
it is not clear that high token frequency instances in fact do not contribute to the
productivity of a schema.
The usage-based definition of productivity is gradient, because type frequency
is gradient. Thus, the usage-based model predicts that productivity might vary in
degree. Forms for which there is a low type frequency may exhibit a minor degree
of productivity. Evidence for this is found among the irregular English past tense
forms. Most of the irregular English past tense forms involve an internal change to
the stem, usually a change to the stem vowel. One particular class has a relatively
high type frequency of relatively low token frequency verbs. This class is Bybee
and Slobin’s Class VI. Class VI verbs fall into two subclasses, those with a past
tense form with /æ/ and a past participle with / / (Class VIa in example [4]) and
those with a past tense form with / / (Class VIb; Bybee and Slobin 1982:288,
Appendix).
(4) Class VIa: Class VIb:

m swim/swam/swum n spin/spun
n begin/began/begun win/won
run/ran ng bring/brung* [dialectal]
ng ring/rang/rung* cling/clung
sing/sang/sung fling/flung*
spring/sprang/sprung hang/hung*
nk drink/drank/drunk sling/slung*
shrink/shrank/shrunk sting/stung*
sink/sank/sunk string/strung
swing/swung
wring/wrung
nk slink/slunk
k sneak/snuck* [dialectal]
stick/stuck*
strike/struck*
shake/shuck* [dialectal]
g dig/dug*
drag/drug*
Class VI irregulars are relatively coherent as a phonological class: most of them

have present tense vowel /I/ and end in a velar or nasal or nasal+velar. The past
tense form of Class VIb in particular can be described as having a prototype pattern
of present tense /Ciŋ / and past tense /Ciŋ /, with extensions to velar and/or nasal
final consonants and some variation in the stem vowel (cf. Bybee and Moder 1983
and §11.2.3). However, many of the Class VI verbs, particularly Class VIb verbs,
are relatively low in token frequency (Bybee and Slobin 1982:278). Hence Class
VI can be predicted to be mildly productive, although not nearly as productive as
the -ed past tense allomorphs.
The evidence for the productivity of this schema is historical and psycholinguis-
tic. The verbs asterisked in (4) were not part of this irregular class in Old English
(Bybee and Slobin 1982:288, citing Jespersen 1942). In fact, three of the aster-
isked Class VIb forms are not standard English forms, but are used in nonstandard
English dialects, suggesting a relatively recent shift of these verbs to Class VI. The
asterisked forms were either irregular verbs of other classes, or even regular verbs
to which the Class VIb vowel alternation was extended because of the phonolog-
ical resemblance of their verb stems to the stems of Old English Class VI verbs,
and the relative productivity of Class VI.
Bybee and Slobin also conducted experiments with children and adults: the
children were asked to provide past tense forms in sentence completion tasks, and
the adults were asked to provide past tense forms under extreme time pressure.
Both tasks elicited novel past tense forms. Of these, the data for both children and
adults indicated a mild degree of productivity of Class VI, such as streak/struck

and clink/clunk (Bybee and Slobin 1982:278).
Advocates of the dual-processing model of morphological representation have
argued that, while irregular verbs may display frequency-related patterns such
as those found by Bybee and Slobin for the Class VI past tense verbs, regular
inflections do not display any type frequency effects. Instead, regular inflections are
productive because they impose the least phonological specificity on the verb stem
(Prasada and Pinker 1993). For example, the allomorphs of the English regular
past tense impose the least phonological specificity on their verb stems: verb
stems are constrained only by certain phonological features of the final segment.
In schema terms, the regular inflection provides a maximally open schema. The
dual-processing hypothesis is given in (5):
(5) Hypothesis 2 : productivity of a rule is determined by its being a (relatively) open

schema.
English does not provide a good case to differentiate these two hypotheses for
productivity, type frequency vs. open schema. The regular English past tense has
both a much higher type frequency than any other past tense form and a very open
phonological schema. Thus, both hypotheses predict (correctly) that the regular
English past tense is highly productive. However, there are languages in which
the open schema for an inflection does not have a high type frequency. Examples
include the German plurals (Marcus et al. 1995) and Arabic plurals (McCarthy
and Prince 1990).5
The German and Arabic plural cases both dissociate type frequency from open
schema. Both German and Arabic have a range of plural formation processes,
none of which has an overwhelmingly greater type frequency such as is found
with the English plural -s. Both German and Arabic have an open schema plural,
the German -s and the Arabic sound plural (a suffix instead of an internal stem
change). Both German and Arabic open schema plurals are used as the plural
schema with items other than standard common nouns, such as proper names,
new borrowings and derived nouns and adjectives (Bybee 1995:440–42; Janda
1990:146–48). The open schema plurals are open schema precisely because they
must be applicable to noncanonical nouns; but they are also of low type frequency
for the same reason.
5 Clahsen and Rothweiler (1992) argue that the German past participle ending -t has a lower type
frequency than -en and yet is more productive. However, Clahsen and Rothweiler used only the first
1,000 verbs of a 4,314-verb frequency list of German verbs, thereby leaving out a very large number
of regular verb types, and they counted the verb stems multiple times if they occurred with multiple
productive prefixes, which again artificially increases the type frequency of -en (Bybee 1995:438).
Marcus et al. and Pinker and Prince argue that the applicability of the open
schemas to noncanonical nouns is evidence of their high productivity. Bybee, on
the other hand, argues that this is evidence only of their ‘emergency’ or default
status by virtue of their open schema, and that investigating the full range of
common nouns indicates that the default schema is no more productive than plural
schemas with a high type frequency. For Arabic, the iambic broken plural has
the highest type frequency, and is productive with any noun (including borrowed
nouns) that fit its canonical phonological shape criteria. The iambic broken plural
is overgeneralized by children as well as the default sound plural (Omar 1973,
cited in Bybee 1995:442), indicating that both the iambic broken plural and the
default sound plural are productive, the former especially so.
The German plural situation is more complicated. No single plural formation
pattern is highly productive. However, the evidence for productivity apart from
noncanonical nouns indicates some degree of productivity of several different
plural endings, including -s (see Bybee 1995:440–41 and references cited therein).
Children overgeneralize -en most frequently, and in nonce-probe tasks, different
endings were preferred for different noun genders, particularly if the nonce forms
were identified as common nouns. Even recent borrowings use -en and to a lesser
extent -e especially for masculine nouns; -s is used about half of the time, and
some loans have given up -s for -en upon integration into the language. Also,
nouns ending in vowels favored -s in a nonce-probe task (Köpcke 1988) and in
acquisition (Köpcke 1998:313–15). Among ordinary German common nouns -s
is associated with nouns ending in a full vowel (Janda 1990:145–46). These facts
suggest that -s is not truly a default schema.
A default schema does not require a high type frequency to arise. A default form
can arise if the non-default forms form relatively narrow and phonologically well-
defined classes, while the instances of the default schema are scattered across the
remaining phonological space, even if those instances have a low type frequency
(Hare, Elman and Daugherty 1995:626–27; they also simulate this effect in a
connectionist network). The German and Arabic plurals conform to this pattern
(compare Hare et al. 1995:608). If the phonological unity of the non-default classes
breaks down through phonological change, as happened with the Old English
past tenses, then the system becomes unstable, leading to a reorganization of
the irregular forms around new phonological classes or survival by high token
frequency (Hare and Elman 1995).
11.2.3 Product-oriented schemas

Up to this point, we have assumed that morphological generalizations
such as the relationship between the present and past tense forms of English verbs
can be equally well captured by rules deriving one form from another (or both
forms from a common underlying form). Instead, we have focused on the fact that
the usage-based model can capture generalizations about morphological patterns
that are based on token frequency and type frequency, whereas rules in the dual-
processing model do not imply the existence of any generalizations linked to
frequency, at least for regular inflections. However, Bybee and others have argued
that rules and schemas in fact make slightly different predictions about what sort
of generalizations can be made over related word forms.
Structuralist and generative morphological rules are what Bybee calls source-
oriented. Source-oriented rules specify the basic word form, such as the present
tense of a verb like wait, and describe a single operation with a single set of
conditions that produce the derived form, such as addition of / d/ to a stem ending
in t or d to form waited. In the usage-based model, a source-oriented schema is a
schema with a systematic structural relationship to another schema (the source in a
rule-based model). For instance the regular past tense schema [Ve r b- d] contains
the same stem as the present tense schema [Ve r b(- z)], and so the past and present
tense verb form schemas can be uniformly represented across all stems. The term
‘source-oriented’ is somewhat misleading in the usage-based model because there
is no derivation of one schema from another by a rule in the usage-based model. The
term ‘source-oriented’ simply indicates that the ‘source’ schema is as coherent,
phonologically and semantically, as the ‘product’ schema.
Bybee argues (following Zager 1980) that in addition to source-oriented
schemas, there also exist product-oriented schemas:
(6) Hypothesis 3: In addition to source-oriented morphological rules/schemas, there

also exist product-oriented schemas, which cannot be easily represented by deriva-
tional rules.
A product-oriented schema is a morphological schema that is coherent in terms

of the phonological form and meaning of the ‘derived’ inflected form of the word,
not in terms of a rule deriving the inflected form from a base form (Bybee
1995:443). In a morphological inflectional category with a product-oriented
schema, the ‘product’ schema is more coherent phonologically than (and at least
as coherent semantically as) its counterpart ‘source’ schema.
An example of a product-oriented schema is the schema for the Class VI irregular
verb class illustrated in (4) above. Because of the variety of phonological shapes of
the present tense forms of Class VI verbs, one cannot construct a single coherent
rule to derive the past tense form from the base form. Instead, the past tense forms
are more or less converging on the prototypical past tense shapes [. . . æŋ] (for
Class VIa past tense forms) and [. . . ŋ] (for Class VIb past tense forms, and the
Class VIa past participle forms).
Product-oriented schemas cannot be described in terms of a rule that converts

a base form into a derived form. At best, one would have to have a different rule
for each word (or small sets of words) deriving the product form from the source
form. The reason for this is that product-oriented schemas are more coherent and
unified categories than their so-called source schemas. Product-oriented schemas
can best be described in terms of a prototype schema towards which the ‘derived’
forms converge, by whatever phonological means necessary (Bybee and Moder
1983). Since a purely structural description of a product-oriented schema is not
possible (short of a ‘rule’ for each word), the structuralist/generative model predicts
that product-oriented schemas should not exist.
The term ‘product-oriented’ is misleading in the usage-based model in the same
way that ‘source-oriented’ is. In a usage-based model, ‘product-oriented’ schemas
arise because they represent a schematic (taxonomic) generalization across the
‘derived’ forms, which are represented as independent units. The primary factor
determining the existence of a schema, ‘source’- or ‘product-oriented’, is a (rel-
atively) high type frequency (§11.2.2). Hence, if a ‘product-oriented’ pattern in
word forms is at least partially productive, then it provides evidence in favor of
the usage-based model.
Such evidence is found with the English Class VI past tense schema. No single
rule ‘derives’ the past tense form from the present tense form. But the past tense
forms of Class VI can be described in terms of a family resemblance category,
whose prototype is [. . . ŋ /PAST]. As we saw in §11.2.2, the schema is pro-
ductive, demonstrated by performance by children and adults in psycholinguistic
experiments, and in the extension of the Class VI past tense schema to other English
verbs in the history of the language (see Bybee and Moder 1983 for a more detailed
argument in favor of this schema).
Bybee cites further evidence from plural formation in Hausa (Bybee 1995:443–
44). Haspelmath (1989) demonstrates that there is no set of general rules for
forming the plural in Hausa, but the plural forms can be described in a set of
product-oriented schemas which in turn can be subsumed under a more general
product-oriented plural schema. Lobben (1991, cited in Bybee 1995) conducted
nonce-probe experiments with Hausa speakers that indicate that the plural schemas
were indeed productive.
11.2.4 Network organization of word forms

In chapter 10, we observed that in construction grammar, constructions
are organized in a network in a taxonomy. However, in §10.1.2, we also observed
that constructions can, and usually do, have multiple parents in the taxonomy. The
description of the construction grammar of a language in any detail is going to lead
to an extremely tangled network of construction taxonomies. Is there any evidence
to determine which categories are more important in organizing multiple parent

taxonomies?
Bybee proposes a model in which the essential organizing feature of the network
of words is similarity. Similarities are connections or links between words. Bybee
further makes a number of proposals regarding which similarities between words
are more important. We may summarize Bybee’s proposals in (7):
(7) Hypothesis 4: strength of connection between word forms, and thus forces influ-
encing their phonological shape (among other things), is a function of similarity.
Similarity is measurable by comparing words to each other in both meaning and
form; similarity in meaning is much stronger than similarity in form.
Hypothesis 4 is part of the usage-based model of morphological representation in

that the meaning of word forms, which is manifested in their use in communication,
influences the organization of the knowledge of those word forms in a speaker’s
mind. In the structuralist/generative model, only structural properties determine
the organization of word forms in a speaker’s mind.
Words may be similar in form, meaning or both. Bybee describes similarity in
form as implying a phonological connection, similarity in meaning a semantic
connection, and similarity in both a ‘morphological connection’; for the latter
case, we will use the more general term symbolic connection.
Bybee argues that a solely phonological connection between words – in other
words, homophony – is the weakest connection of all (Bybee 1985:118). Homo-
phones such as the two most distinct senses of bank or crane have relatively little
psychological effect. Homophony does give rise to a minor yet robust priming
effect. In lexical decision and target naming tasks using homonyms, there is a
priming effect of both the contextually appropriate meaning and the homonymous
meaning within 0-200 msec of presentation of the stimulus; after 200 msec, only
the contextually appropriate meaning is primed (Swinney 1979, 1982; Seidenberg
et al. 1982). This priming effect indicates that there is a lexical connection based
on mere phonological similarity, but not a strong one.
A solely semantic connection between words is much stronger. An example
of a close semantic connection without a phonological connection is suppletive
paradigms, as in English go/went. Bybee observes that suppletion is subject to
regularization, as found in innovations in language use such as goed for went
and the replacement of suppletive forms over time in languages. The semantic
connection between GO+PRESENT and GO+PAST is so strong that one of the
forms is changed (went > goed) in production so as to make it more similar to the
other form with the similar meaning (go). In contrast, the existence of similarity in
form in homonyms or near homonyms does not lead to one of the words changing
meaning so as to make it semantically more similar to the other word with a similar
form.
The strongest connection is a symbolic connection: similarity in form and mean-

ing. Bybee argues that there are three factors that determine the strength of a sym-
bolic connection: degree of semantic similarity, degree of phonological similarity
and degree of entrenchment. Beginning with the last factor first: relative degree
of entrenchment largely determines the direction of analogical changes in word
paradigms. For example, one would predict that it is more likely that a speaker
would produce writed in place of wrote, by analogy with write(s), rather than
wrotes in place of writes. The past tense is less entrenched because it is less fre-
quently used than the present tense, and so is less resistant to being replaced by an
analogical formation that results from activation of the more entrenched form.
As with solely phonological and solely semantic connections, in symbolic con-
nections degree of semantic similarity is the more important factor. Bybee describes
degree of semantic similarity as relevance. The notion behind relevance can be
described by examining the inflectional categories that Bybee studied, those of the
verb: aspect, tense, mood and person.
An inflectional semantic category that is highly relevant to the verb is one that
makes the greatest change in the verb’s meaning. For example, a change in aspect
from, say, present meaning to habitual/generic meaning is a substantial change in
meaning. The present time reference in (8a) describes a current state of affairs,
true at the present moment but not long lasting. The habitual/generic meaning in
(8b) describes a series of eating events over a long period of time (habitual), or an
inherent property of mine that disposes me towards eating ice cream (or at least
does not prevent me from eating ice cream):
(8) a. I’m eating ice cream.

b. I eat ice cream.
In contrast, changing the person who is doing the eating does not make a very
substantial change in the nature of the eating event itself:
(9) a. I’m eating ice cream.

b. She’s eating ice cream.
Relevance is inversely related to strength of semantic connection. Two different

aspectual forms of a single verb are more weakly connected semantically, because
the event types they describe are more different. Two different person agreement
forms of a single verb are more strongly connected semantically, because the event
types they describe are more similar.
On the basis of semantic argumentation, Bybee proposes the following ranking
of verbal inflectional semantic categories, from most relevant to the verb/event to
least relevant (see Bybee 1985:20–23 for the semantic argumentation):
(10) valence changing < voice < aspect < tense < mood < person/number agreement
Bybee puts forward several types of typological and diachronic evidence for
the ranking in (10). For example, more relevant inflectional categories of the verb
occur closer to the verb stem. The reasoning behind this prediction is that the
greater the meaning change, the more intimately associated with the stem meaning
is the semantic category of the inflection. Bybee tested the hypothesis on the
four most common verbal inflectional categories, aspect, tense, mood and person
agreement, in a fifty-language sample (Bybee 1985:33–35). There were virtually
no counterexamples to the ordering of aspect, tense and person/mood with respect
to the other categories; mood and person agreement were more equivocal.
The notion of relevance is a further refinement of Hypothesis 4: that seman-
tic similarity to different degrees influences formal similarity of word forms to
different degrees. That is, greater semantic similarity will favor greater phono-
logical similarity (and thus increase symbolic similarity). Semantic distinctions
expressed lexically will be more phonologically different than semantic distinc-
tions expressed inflectionally, on the whole. It should be remembered that other
factors such as degree of entrenchment also affect phonological similarity: a higher
degree of entrenchment weakens the connection between word forms and thus may
lead to greater phonological differences.
Thus, another prediction from Hypothesis 4 is that, other things being equal,
a stronger semantic connection between words will imply a greater phonological
similarity of those words. Also, one would expect to find that phonological similar-
ity can be increased through analogical change of semantically strongly connected
words. There is considerable evidence for this prediction as well. Bybee reports
that data from her survey and from Rudes (1980) indicate that suppletion in verbal
paradigms is most likely along aspectual distinctions, then along tense, and least
likely along mood. There is also some suppletion along person distinctions, but
only in very high frequency forms; this can be explained by the principle given in
the preceding paragraph, that a high degree of entrenchment weakens connections
between words.
Finally, when paradigms are leveled analogically, they are most likely to be
leveled among semantically closely related forms, in particular different per-
son/number forms of the same tense-aspect-mood paradigm. Also, the direction of
leveling will be most likely towards the most frequent form (third person singular,
followed by first person singular), because forms with weak connections will give
way to analogical formations from stronger forms. Bybee presents a number of
examples of such leveling within person-number forms (Bybee 1985, chapter 3;
see also Bybee and Brewer 1980 for Spanish). For example, the Old Provençal
person-number forms for the preterite indicative in (11) were reformed analogi-
cally on the third person form including that form’s -t person/number suffix, in a
number of Modern Provençal dialects, such as the Charente dialect in (12) (Bybee
1985:55; Charente data from Meyer-Lübke 1923:352):
(11) Old Provençal preterite of am- ‘love’:

am-éi am-ém
am-ést am-étz
am-ét am-éren
(12) Charente preterites of cant- ‘sing’:
cantı́ cantét-em
cantét-ei cantét-ei
cantét-Ø cantét-en
In (12), the original third person singular stem cantét has become the base for
the analogical reformation of the other person-number forms. The only form to
resist the analogical change is the first singular form, which has the highest token
frequency of the person-number forms after third singular.
Bybee and Pardo (1981) used a nonce-probe task with Spanish speakers to com-
pare the effect of semantic connection on phonological production. Many Spanish
verbs have a vowel stem alternation in the present vs. preterite forms, for example
third singular present comienza ‘s/he begins’ with a diphthong vs. third singular
preterite comenzó ‘s/he began’ with a simple mid vowel. Bybee and Pardo pre-
sented a third singular present form of a nonce verb with a diphthong followed by
either the infinitive or the third singular preterite form with a mid vowel, and then
asked subjects to produce a first singular preterite form. Subjects produced more
mid vowel variants when presented with the semantically closer related third sin-
gular preterite than when presented with the more distant infinitive.
The evidence presented by Bybee and others (e.g. Andersen 1980) implies that
semantic similarity of grammatical units such as words plays an important role
in the organization of grammatical knowledge in a speaker’s mind. Given a set
of word forms, each of which can be subsumed under several more schematic
categories (indicative, present, third person, singular), one can postulate a ranking
of those schematic categories in terms of the network connections between words.
One could go so far as to restructure the multiple-parent representation into a
hierarchy, with the semantically most relevant (and hence most weakly connected)
distinctions at the top of the hierarchy, as in (13) (compare the display of the
Spanish verb paradigm in Bybee 1985:61, Table 1):
(13)
Present Past
Indicative Subjunctive Indicative Subjunctive
1sg 2sg 3sg ... 1sg 2sg 3sg ... 1sg 2sg 3sg ... 1sg 2sg 3sg ...
The hierarchy in (13) is still an oversimplification. The connections between

members at the lowest level are not displayed, nor are the lowest levels (person and
number) stratified. There remain semantic connections between the corresponding
mood values (present indicative and past indicative etc.), and the corresponding
person/number values (1sg present indicative, 1sg past indicative, 1sg present sub-
junctive, 1sg past subjunctive etc.). However, the hierarchy does indicate degree
of semantic connection: one can heuristically measure degree of semantic con-
nection by the number of connections that must be traversed in (13) in order to
reach related forms. For example, only four connections need to be traversed to
reach first singular present subjunctive from first singular present indicative, but
six connections need to be traversed to reach first singular past indicative from first
singular present indicative (for an alternative representation of degree of semantic
connection, see §11.3.4).
11.2.5 Conclusion
The empirical data that we have discussed in support of the details of
the usage-based model are drawn from the processes of language use (as tested in
psycholinguistic experiments) and language change. These data provide evidence
supporting four hypotheses about the effects of language use on grammatical
representation. The independent representation of an inflected word form is a
function of its token frequency in language use. The productivity of a rule/schema
is a function of a high type frequency of low token frequency instances, not of
the structural openness of the schema. Product-oriented schemas exist, that is,
schemas can be formed from members of an inflectional category that cannot
be described by rules deriving the members of the category from a source form.
Finally, the organization of inflected word forms is influenced by the degree of
semantic similarity between word forms.
The hypotheses of the usage-based model can be accounted for by an interactive
activation network for the representation of knowledge (Elman and McClelland
1984). The storage of word forms is determined in part by patterns of activation of
the network as a result of language use (§11.2.1). The phenomena described in Hy-
potheses 2–4 of the usage-based model are all analyzed in terms of the interaction
of activation patterns, such that a schema activates an instance and vice versa, and
a structure’s activation can result from the activation of formally and especially se-
mantically related structures. The result of the interactive activation is manifested
not only in the conventional production and comprehension of word forms, but
also in ‘errors’ in certain contexts, and innovations in language acquisition and
language change.
The usage-based model contrasts with the complete inheritance model in two
important respects. As we noted in §10.2.2, in the complete inheritance model
information is stored only at the most schematic level possible. In the usage-based
model information can be represented redundantly in less schematic construc-
tions, if activation levels lead to entrenchment. The second contrast follows from
the first. In the complete inheritance model, information flows down from the
most schematic constructions in the processing of an utterance. In the usage-based
model, processing involves activation of the entrenched construction(s) whose
structure(s) most closely matches those of the utterance. Since more specific con-
structions match utterances more closely than more schematic constructions, the
former are more activated than the latter. It is therefore possible that speakers will
not have the most schematic constructions represented in their minds, if they are
not activated sufficiently (Croft 1998c).
11.3 The usage-based model in syntax
In §11.2 we examined morphological representations of words, what they

represent and the processes they are involved with. A number of concrete hypothe-
ses and supporting evidence were put forward on the nature of word representations
and processes, including the role of token frequency in entrenchment, the role of
type frequency in productivity, the formation of schemas, phonological and seman-
tic similarity in connections between words, and the emergence of generalizations
in language acquisition. How many of these hypotheses might hold for syntax as
well as morphology? In §9.4, we argued that the same types of phenomena found
in the study of syntactic idioms are also found in morphology. In this section,
we will examine the applicability of hypotheses about morphological networks to
syntax.
11.3.1 Type/token frequency, productivity and entrenchment

In §§11.2.1–2, it was seen that type and token frequency play distinct
roles in the empirical predictions of the usage-based model for morphology. In
morphology, token frequency determines the degree of entrenchment of individual
substantive word forms, and also implies that strongly entrenched words will have
weak links to related forms (§11.2.4). Type frequency, and phonological coherence,
determine the degree of entrenchment of a schema such as [Ve r b-ed] for the
regular past tense.
However, all syntactic constructions, except for completely substantive idioms,
are schematic to some degree. Even an idiomatic phrase like [kick-tns the bucket]
is schematic in that it can be used in different tense-aspect-mood forms, including

auxiliaries:
(14) a. He kicked the bucket.

b. He’s gonna kick-Ø the bucket one of these days. [etc.]
In syntax, one would like to be able to differentiate between constructions at

different levels of schematicity, such as [kick the bucket], [kick O b j] and [TrVe r b
O b j]. If we want to apply the generalizations of usage-based morphology to syntax,
then we will have to find a more general description of frequency and the role it
can play in morphology.
In the usage-based model, token frequency determines degree of entrenchment
of a single word. A high token frequency for a word corresponds to a high number
of specific usage events with that word. The network pattern for low vs. high token
frequency is illustrated in (15) (boxes with rounded corners correspond to usage
events, and a dashed box, a lower degree of entrenchment):
(15)
prevaricate lie
prevaricate prevaricate lie lie lie lie lie lie lie
Low Token Frequency High Token Frequency
Type frequency determines the degree of entrenchment of a schema. A high

type frequency for a schema means that the schema is more deeply entrenched.
The network pattern for low vs. high type frequency is illustrated in (16):
(16) ew -ed
flew blew rapped stored typed pined sated griped

Low Type Frequency High Type Frequency
The network structure for low vs. high token frequency (entrenchment) and
low vs. high type frequency (productivity) is the same: the higher the number
of instances, the more entrenched the superordinate category is. We may then
formulate the following generalized definition of productivity:
(17) Generalized entrenchment/productivity: entrenchment (productivity) of a con-

struction is proportional to the number of instances of the construction at any
level of schematicity, and to the degree of formal and semantic coherence of the
instances of the construction.
Of course, two other factors have to be recognized in determining the degree

of entrenchment of the superordinate category (as were noted in §11.2.2). First,
the instances must be similar so that a category schema can be formed. Second,
instances that are themselves highly entrenched may not contribute as much to the
entrenchment of the superordinate category schema as instances that are not very
entrenched.
Proponents of the usage-based model have presented data supporting frequency
effects in syntax. Bybee and Thompson observe that the syntax of English auxil-
iaries is conservative, in that auxiliaries may invert with the subject in questions
and precede the negator not rather than follow it (Bybee and Thompson 1997):
(18) a. Have you eaten?
b. *Eat you?
(19) a. I have not eaten.
b. *I ate not.
In Middle English, all verbs had this ability, but it was lost in Modern English
(Bybee and Thompson 1997, from Mossé 1952):
(20) Gaf ye the chyld any thyng?
‘Did you give the child anything?
(21) a. My wife rose nott.
b. cry not so
We will explain the analysis with the interrogative constructions; the same ar-
gument applies to the negative constructions. The auxiliary verbs have a very high
token frequency, compared to other verbs. In constructional terms, this means that
a construction schema such as [Have S b j . . . ?] is much more entrenched than
was [Eat S b j . . . ?]. That is, there are far more instances of the have constructions
(and other auxiliary constructions) than of the eat constructions (and other lexical
verb constructions). In early Modern English, the more schematic construction
[Ve r b S b j . . . ?] declined, while [Do S b j Ve r b . . . ?] spread and became
entrenched (see the account in Denison 1993, chapter 15). The (relatively) low
entrenchment of [Eat S b j . . . ?] meant that [Eat S b j?] died out and was replaced
with [Do Sbj eat?]. However, [Have S b j . . . ?] survived because of its high degree
of entrenchment.6
6 This analysis is slightly different from Bybee and Thompson’s. Bybee and Thompson argue that
this is a case of high token frequency, which is true for the individual auxiliary forms but not for
the auxiliary construction schema. The syntax of the auxiliary must be analyzed in terms of the
constructions in which it occurs. However, Bybee and Thompson’s analysis is easily translatable into
the analysis presented here.
Readers may wonder why the construction schemas abstract away from arguments first, leaving
invariant the verbs and auxiliaries. Arguments in favor of this abstraction based on semantic relevance
and on language acquisition data are presented in §§11.3.3–4.
Bybee and Thompson also discuss a large scale corpus study of the French sub-
junctive by Poplack (1992, 1996). The French subjunctive form is disappearing
from the spoken language, but is still variably used. However, the French subjunc-
tive has survived largely in the complements of the highly frequent main clause
verb falloir ‘have to,’ and/or in the most highly frequent complement verbs, in-
cluding avoir ‘have,’ être ‘be,’ aller ‘go’ and faire ‘make, do,’ as predicted by the
usage-based model.
The example of English auxiliaries in questions and negative sentences il-
lustrates the maintenance of irregularity in more entrenched constructions that
have resisted changes in the more schematic constructions of the language.
Another respect in which more entrenched constructions are irregular is that
they undergo changes that less entrenched constructions do not undergo. These
changes commonly involve reduction, a typical concomitant of high frequency. In
schematic constructions, reduction applies of course to the substantive unit(s) of the
construction.
An example of reduction in a highly entrenched construction is the contraction
of not with the auxiliary in the [Sbj Aux -n’t . . .] construction. This contraction
is recognized in written English, and includes the fused form won’t. Of course,
such reduction originated in the spoken language, and one would expect to find
reduction of other negative-auxiliary contractions in the spoken language. Bybee
and Scheibman (1999) investigate the reduction of one specific negative auxiliary
form, don’t, in spoken American English conversation. They demonstrate that the
phonetic reduction of don’t is strongly correlated with the frequency of the verb
and of the subject with which don’t is combined in the [S b j don’t Ve r b . . .]
construction. The highest frequency subject is the first person singular I, and the
highest frequency verb in this construction is know; in fact, this reduction is so
salient that it is loosely represented orthographically as I dunno. But I dunno is
only the extreme end of a continuum of phonetic reduction that spans the full range
of verbs and subjects used in this construction. Bybee and Scheibman further note
that the reduction applies across the substantive units in the constructions as a
whole, regardless of their internal constituent structure.
In addition to syntactic irregularity as a consequence of high frequency, one
would expect to find degrees of syntactic productivity. Of course, maximal syntactic
productivity is the characteristic of the major, most schematic constructions of
the language, such as the transitive construction [S b j Ve r b O b j]. The high
productivity of this highly schematic construction is due to the very high frequency
of instances of this construction, due to the high number of transitive verbs, the
vast majority of which have a relatively low token frequency. However, even with
completely schematic syntactic constructions, one can find varying degrees of
productivity.
Bybee and Thompson (1997) cite an example of different degrees of produc-

tivity from the analysis of argument structure constructions by Goldberg (1995).
Goldberg compares two constructions, the Caused Motion construction and the
Ditransitive construction, illustrated in (22) and (23) respectively:
(22) He told the news to the woman.

(23) He told the woman the news.
The two constructions overlap in their distribution: some verbs allow both con-
structions, with approximately the same meaning, as in (22)–(23). The Ditransitive
construction is used with many fewer verbs than the Caused Motion construction,
and hence is much less productive than the Caused Motion construction (Goldberg
1995:124):
(24) Sally whispered some terrible news to him.
(25) *Sally whispered him some terrible news.
However, Goldberg notes that the prototypical ditransitive construction (‘S b j

cause Obj1 to receive Obj2’; see §10.2.2) has the highest type frequency of
any of the ditransitive subconstructions. And in fact the prototypical ditransitive
has some productivity compared to the other ditransitive subtypes. The protoypical
ditransitive is extended to new verbs, as in (26), but the subtype ‘S b j enable Obj1
to receive Obj2’ is restricted to a subset of enabling verbs and does not extend to
other verbs with that meaning (Goldberg 1995:129, 130):
(26) Chris e-mailed/radioed/arpanetted him a message.
(27) a. Sally permitted/allowed Bob one kiss.
b. *Sally let/enabled Bob one kiss.
Finally, one must also differentiate between productive syntactic schemas and
default syntactic schemas, used, for example, in borrowing. Many languages, par-
ticularly languages with complex verbal morphology, do not directly incorporate
borrowed verbs into the productive native syntactic constructions of the language.
Instead, they use a default construction, combining an invariant form of the bor-
rowed verb with an inflected native verb (often meaning ‘make, do’) in a construc-
tion [B o r r Ve r b ‘make’ . . .]. An example of this phenomenon can be found
in K’iche’ Mayan, which uses the verb ban ‘make’ with the infinitive form of
borrowed Spanish verbs, as in (28) (Mondloch 1978:117):
(28) x- Ø- im- ban engañar lē achi

pst- 3sg.abs - 1s g . e r g- make deceive the man
‘I deceived the man.’
The generalization of the notion of type/token frequency for constructions of
varying degrees of schematicity allows us to make predictions about syntactic
regularity and irregularity based on the usage-based model of morphology, and to

seek evidence for those predictions.
11.3.2 Product-oriented syntactic schemas

In the usage-based model, word forms are not the output of rules but in-
stances of schemas. In §11.2.3, it was noted that source-oriented schemas capture
the same relationships between word forms as rules in the rule-based model. Evi-
dence was presented there for product-oriented morphological schemas. Product-
oriented morphological schemas are generalizations over word forms that would
be analyzed as the output of a morphological rule in the rule-based model, but
cannot be so analyzed because each word form would require a different ‘rule’
deriving the output. However, the ‘outputs’ have a phonological coherence that is
greater than that of the ‘input.’ The example of a product-oriented morphological
schema given in §11.2.3 is the [. . . ŋ] schema for the Class VIb irregular past
tense verbs of English (Bybee and Slobin 1982). In this section, we will discuss
possible cases of product-oriented syntactic construction schemas.
The classic rule-based model of syntactic representation is transformational
generative grammar. In transformational grammar, a construction is describable
as the product of general rules. The effect of these rules is to insert, move or (in
earlier versions) delete syntactic elements from the source structure to yield a target
structure. Although current generative theories use a variety of formal devices, and
some theories eschew syntactic movement rules, a systematic relationship between
constructions such as the active and passive voice is still recognized as a rule-based
relationship.
In construction grammar, some construction schemas can be thought of as
source-oriented. For example, the English active transitive and passive construction
schemas can be described as in (29):
(29) a. Active: [S b j i Ve r b k -t n s O b j j ]
b. Passive: [S b j j be-t n s Ve r b k -pp by O b j i ]
In these constructions, systematic correspondences can be established between

the elements of the two constructions (indicated by the indices on the elements),
which can then be used to formulate a rule in the rule-based model that derives
the product construction (the passive) from the source construction (the active, or
some underlying structure from which both are derived).
Product-oriented syntactic construction schemas, on the other hand, would be
construction schemas that have a coherent syntactic structure but would require dif-
ferent ‘rules’ for each type of ‘input construction’ in transformational-generative
terms. Thus, a product-oriented construction schema is a schema subsuming
the outputs of different ‘rules’ and ‘input constructions.’ Of course, since the
syntactic constructions in question are themselves schematic, rules can be devised
for each construction type. But this is no different than having a ‘rule’ for each word
in morphology; it is the same phenomenon but at a higher level of schematicity.
The crucial point is that there is a higher degree of structural coherence defining
the product-oriented schema than its counterpart ‘source’ schema(s).
There are some strong candidates for product-oriented syntactic schemas in
English. One is the pair of interrogative and negative construction types discussed
in §11.3.1. A very general schema can be formed for each: [Au x S b j . . . ?]
and [Sbj Aux -n’t . . .]. However, there would have to be at least two rules
linking a source schema to the interrogative or negative product schema: a rule
inserting the auxiliary do for sources without an auxiliary, and a rule applying
to the (first) auxiliary for sources with an auxiliary. The different source-product
correspondences are illustrated for the question schema in (30)–(32):
(30) a. She found it. b. Did she find it?

(31) a. She will come. b. Will she come?
(32) a. She could have eaten already. b. Could she have eaten already?
There is more structural coherence to the product constructions in (30b)–(32b) than

to the ‘source’ constructions in (30a)–(32a). And of course, the product-oriented
schema is productive.
Another relatively clear case of a product-oriented schema in English is the
most general declarative construction. In the simple declarative construction, there
must be a preverbal subject phrase; the schema must be something like [. . . S b j
(Aux) Verb . . .]. In this case, the ‘sources’ are varied, as well as the ‘rules’
that would produce the output constructions. In the canonical case, the subject
argument occurs in preverbal position, as in (33a). In the case of certain verbs
which are often analyzed as lacking a subject participant, such as weather verbs,
a neuter subject it is put in preverbal position, as in (33b). In the case of certain
constructions in which a subject argument is not in preverbal position, the neuter
pronoun is put in preverbal position, leading to the appearance of two ‘subject
phrases,’ as in (33c):
(33) a. It’s in the cupboard.

b. It’s raining.
c. It is amazing how often it rains in Manchester.
Other analyses have been proposed for the constructions in (33b–c): some have
argued that it in (33b) refers to the general ambience, and others have argued that
there is only one ‘surface’ subject phrase in (33c), the ‘underlying’ subject having
been extraposed to a different syntactic position. However, whatever rule-based
analysis is proposed to derive (33a–c), the result is a coherent product-oriented

construction schema for the constructions exemplified in (33a–c), in which there
is a subject phrase in preverbal position. And this construction has been gradually
spreading in the history of English, replacing constructions in which the subject
could appear in different positions.
Another class of syntactic constructions that are easily analyzable as product-
oriented schemas are the so-called extraction constructions, in which an extracted
element, such as the question word in information questions, the head of a relative
clause, or the clefted noun phrase in a focus cleft construction, is positioned at the
beginning of the construction:
(34) a. Who (*did) met Jill yesterday?
b. the man that/who/*Ø met Jill yesterday
c. It was Ed that/who/*Ø met Jill yesterday.
(35) a. Who did Jill meet yesterday?
b. the man (that) Jill met yesterday
c. It was Ed that Jill met yesterday.
(36) a. Who did Jill talk to yesterday?
b. the man (who) Jill talked to yesterday
c. It was Ed that Jill talked to yesterday.
(37) a. What did Jill open the box with?
b. the hammer (that) Jill opened the box with
c. It was a hammer that Jill opened the box with.
The analysis of extraction constructions in rule-based models allows for move-

ment (or the equivalent thereof) of the question word/relative clause head from any
position in the sentence to initial position.7 Syntacticians are accustomed to for-
mulating a WH-movement rule as ‘move from anywhere to a particular position’:
it is not generally observed that the movement rule is slightly different depending
on the position from which it is moved; what the extraction constructions all have
in common is the target structure, that is, the product of the different rules. In fact,
the rule for English subject questions, relatives and clefts must be slightly different
from the rule for the other types of extraction, since the subject question does not
have an auxiliary (34a), and the subject relative and cleft prohibit deletion of the
relative pronoun in most dialects (34b).
Moreover, other languages have substantially different ‘rules’ for the forma-
tion of information questions, relative clauses and focus (cleft) constructions for
7 We disregard here the existence of constraints on the ‘path of movement’ (as it would be described
in a rule-based transformational model). Although it is widely assumed that the constraints must
be formulated in syntactic terms (beginning with Ross 1967), there are many counterexamples,
and alternative accounts have been formulated in semantic/pragmatic terms (see Deane 1991 and
references cited therein).
different arguments. In K’iche’, for example, questions, relatives and focus con-
structions formed on the ergative (transitive subject) argument require the -Vn
focus antipassive verb form (see example [38]; Mondloch 1978:74); those formed
on the absolutive (intransitive subject or transitive object) argument retain the ac-
tive voice verb form ([39]; Larsen and Norman 1979:357); those formed on the
instrument require the -bej focus antipassive form ([40]; Norman 1978:462); and
those formed on locative/directional phrases retain the active voice form but leave
the demonstrative pronoun wih in the normal oblique position ([41]; Mondloch
1978:42):
(38) jachin x- Ø- cun -an lē yawab?
who pf- 3sg - cure -antipass the sick.one?
‘Who cured the sick one?’
(39) jachin x- Ø- u- ch’ay -Ø lē achi
who pf- 3s g . a b s- 3s g . e r g- hit -act the man
‘Who did the man hit?’
(40) jas x- Ø- u- rami -bej lē achih
what pf- 3s g . a b s - 3s g . e r g - cut -inst.pass the man
r- ē le chēʔ
3s g . p o s s - g e n the tree
‘What did the man use to cut the tree?’
(41) jawiʔ c- at- bē wi?
where impf- 2sg.abs - go wi
‘Where did you go?’
Nevertheless, in K’iche’ as in English, there is a coherent product-oriented schema
such that the questioned, relativized or focused, phrase is put in initial position,
whatever happens to the rest of the sentence. This generalization is also productive,
and is another example of a product-oriented schema.
Another example of a product-oriented schema is the Japanese passive. The
Japanese passive construction is characterized by: (i) a subject NP which may be
marked with ga (‘subject’), wa (‘topic’), or may be absent altogether if highly
topical; (ii) an oblique agent NP marked with ni, which is optional; and (iii) a
verb form in -(r)are. The subject of a Japanese passive need not be merely the
object of the corresponding active, as in (42) (Tsukiashi 1997:18; all examples are
attested):
(42) Dietrich ga hangyakuzai de jusatu sareru

Dietrich s b j treason for shoot.to.death do:pa s s
‘Dietrich is shot to death for treason.’
The subject of the passive construction may correspond to the indirect object
of the active verb ([43]; Tsukiashi 1997:25), the possessor of the direct object of
the active verb ([44]; 1997:30–31), the subject of the complement of the active
verb ([45]; 1997:36), another NP somewhere in the sentence ([46]; 1997:38), or a

referent that is not a participant in the event at all ([47]; 1997:39; note that hiku is
intransitive only):
(43) ‘neetyan beppin dana’ to iwareta
girl beautiful are c o m p say:pass:pst
‘I was told, “you are a beautiful girl”’
(44) watasi wa sitagi o torarete-simatta
I t o p underwear o b j steal:pass :p e r f
‘I have had my underwear stolen.’
(45) tamago wa zenmetu -ka to omowareta
egg sbj all.broken -ques comp think:pass :pst
‘The eggs were thought to be all broken.’
(46) ushiro-no seki de wakarebanasi o saretari suruto
behind table l o c break.up.story o b j do:pa s s if
‘if I have people talking about their breaking up at the table behind me’
(47) karako no sugata ni hikarete . . .
Chinese.doll g e n figure ni attract:pass:conj
‘I was attracted by the figure of the Chinese doll.’
As with the extraction constructions in (34)–(41), the product construction for
the Japanese passives in (43)–(45) has a coherent structural schema, namely [(NP
ga/wa) (NP ni) Ve r b-(r)are . . .]. A similar argument can be made for the Bantu
passive-applicative patterns, as found in Kinyarwanda (Kimenyi 1980; see also
Hawkinson and Hyman 1974). The various passive-applicative constructions in
Kinyarwanda resemble the various K’iche’ extraction constructions, in that a dif-
ferent applicative suffix must be added to the verb depending on the role of the
oblique argument that is ultimately passivized. But all of them have in common a
subject NP and a passive verb form.
All of these examples can be analyzed as product-oriented construction schemas
in a usage-based model of construction grammar, in terms of their ‘syntactic’ or
symbolic grammatical structure (see §11.2). There is also a plausible candidate
for a product-oriented schema in English based on the phonological structure of a
family of construction schemas. These are the so-called quasimodals, or recently
grammaticalized constructions that have developed tense-aspect-mood functions
but are syntactically distinct from the older auxiliary category of English. All of
the quasimodal forms end in an alveolar consonant followed by schwa, although
this ending represents the reduction of different source forms, as in (48):
(48) a. She coulda done it. [from could have; also shoulda, woulda]
b. She oughta do it. [from ought to; also gotta, hadda, hafta, useta]
c. She oughta done it. [from ought to have; also gotta, hadda]
d. She’s gonna do it. [from -ing to; also non-third singular wanna]
e. She betta do it. [from had better]
We may describe the overall schema as [Sbj QuasiModal . . . Calv ə VP]. This
pattern seems to be mildly productive, in that almost all of the grammaticalizing
quasimodals have reduced to forms ending in [. . . Calv ə ]; of course, there is a
relatively low type frequency to this construction.
It should be pointed out that in the past two and a half decades of generative
syntactic research, emphasis has shifted from the description of rules to the de-
scriptions of constraints (‘principles’) on the output of rules (see, for example,
Chomsky 1981:3–4; 1993:5). To the extent that the principles and constraints of
generative grammar describe the structure of the ‘product’ schema, then genera-
tive grammar constraints are handling essentially the same kind of phenomena as
product-oriented schemas in the usage-based model. However, current generative
syntactic models utilize abstract syntactic structures and derivational processes,
and generate a wide range of outputs, many of which are invalid (‘crash’; Chomsky
1993:5) and only a few of which are actually occurring linguistic expressions (‘con-
vergence’; ibid.). In contrast, the usage-based model represents schemas abstracted
inductively from actually occurring utterances; in syntax, as in morphology, the
usage-based model does not posit underlying structures or nonexistent structures
that are filtered out (see also footnote 7).
11.3.3 Relevance and the organization of construction networks

In the usage-based model of morphology, semantic connections are ar-
gued to be much stronger than phonological connections. Moreover, degree of
semantic similarity predicts aspects of morphological structure, in particular the
likelihood of suppletion and other morphophonological irregularities. The degree
of relevance of semantic relations allows one to impose a roughly hierarchical
structure on a taxonomic network (see the diagram in [13] above).
The notion of relevance (relative semantic similarity) allows us to construct
hypotheses about the organization of syntactic knowledge as well. In this section,
we will examine the hypothesis that the simple relevance hierarchy in (49) governs
some aspects of the organization of syntactic knowledge of sentences (see Clausner
1991 for discussion of a more detailed hypothesis):
(49) illocutionary force < predicate type < participant type
We first offer semantic arguments for the relevance ranking in (49). Relevance
of sentences pertains to the meaning of the utterance in context. The illocutionary
force of an utterance has the greatest semantic effect on the meaning of a sentence,
since it alters the speaker’s intention and the hearer’s response to a proposition if
it is presented as an assertion, question, command or other speech act. Predicate
type is definable at two levels. At a general level, predicate type distinguishes

predicating an action (verbal predication) vs. describing, classifying, locating or
identifying a referent (various types of nonverbal predication). At a more specific
level, predicate type distinguishes the different kinds of states of affairs described
by different predicates (e.g. run, talk, dance, sleep etc.). Differences in predi-
cate type effect the greatest semantic changes in a proposition since they alter
the state of affairs; this is a more dramatic change to the semantic representa-
tion than merely changing the identities of the participants for a given state of
affairs.
Evidence in favor of the relevance ranking in (49) would be a greater likelihood
of significant differences in the phonological pole of a construction for semantically
more distant – that is, less similar/less strongly connected – construction types.
In morphology, evidence for semantic distance has been drawn from changes in
the phonological substance of word forms. In syntax, we are dealing with largely
schematic constructions, so evidence for distance will be drawn from the symbolic
organization of grammatical form, that is, the syntactic elements of constructions,
their presence/absence, and their order. Among the most significant structural dif-
ferences between sentence types are changes of word order, insertion of additional
units, and units positioned in first or last position, the two most salient positions
of the word string (see Clausner 1991).
Differences in illocutionary force are associated with sentence types, such as the
traditional categories of declarative, interrogative and imperative. Cross-linguistic
surveys indicate that there are substantial syntactic differences in the order and pres-
ence/absence of syntactic units in different sentence types (Sadock and Zwicky
1985; Clausner 1991). Interrogatives typically involve the repositioning of the
questioned element, either to sentence-initial or preverbal position, and the addi-
tion of a question morpheme. Imperatives typically lack the subject (addressee)
element, and have a stripped-down verb form.
In contrast, differences in predicate type, that is, verbal vs. nonverbal predication,
rarely involve any change in word order, although they may involve the presence
of an additional element (the copula), and the reduction or absence of ‘verbal’
inflectional categories on the nonverbal element (Croft 1991, chapter 2; Stassen
1997). Finally, differences in the participants in events rarely involve a significant
change in the structure of the sentence. The most common differences are the
employment of special case markings for certain participant types found in many
languages, such as the dative case for experiencers of mental state verbs (I in I
like Mozart) and the instrumental case for inanimate forces (the wind in The wind
knocked down the tree). And even these changes are more accurately described
as dependent on the differences in the type of predicate (state of affairs) which
requires special case marking of experiencers or inanimate forces.
Another sort of evidence for semantic distance in syntactic organization is ana-

logical changes of syntactic constructions. One example from the history of English
suggests that illocutionary force is more relevant than predicate type (and also that
polarity is intermediate in relevance between illocutionary force and predicate
type). This conclusion can be drawn by examining the constructional paradigm
given in (50):
(50) Declarative Imperative Prohibitive
Verbal a. He jumped. c. Jump! e. Don’t jump!
Nonverbal b. He is brave. d. Be brave! f. Don’t be cruel!
<g. Be not cruel!
English forms prohibitive (negative imperative) sentences from verbal predicates
by preposing Don’t to the bare verb stem, as in (50e); compare the imperative form
in (50c). Among nonverbal prohibitives, the construction with Be not, as in (50g),
gave way in the latter parts of the early Modern English period to a construction
preposing Don’t to be, as in (50f) (Denison 1998:252).
The construction [Don’t be A d j] is odd in the broader perspective of English
syntax because a stative predicate normally does not take do, and do does not
combine with be in any other construction, either declarative or (positive) impera-
tive.8 Yet the emergence of [Don’t be Adj] realigns the constructional paradigm in
(50) so that the illocutionary force constructions are more distinct from each other
and a single illocutionary force type is more uniform. The result of the change
from (50g) to (50f) is that the new nonverbal prohibitive construction [Don’t be
Adj] is now structurally more different from nonverbal declarative and imperative
constructions than the old construction [Be not Adj]; compare (50f–g) to (50c).
However, [Don’t be A d j] is structurally more similar to the verbal prohibitive con-
struction; compare (50f–g) to (50e). The constructions in (50e–f) are describable
with a single coherent construction schema [Don’t P r e d ]. In other words, the
change has led to a greater similarity within a single illocutionary force type and
greater differences between sentences of the same predicate type. This is what is
predicted by the semantic distance hypothesis, since differences in predicate type
are less relevant than differences in illocutionary force.
Additional evidence supporting the greater semantic distance between predicate
types than between participant types is found in language acquisition. Children
acquiring English tend to be very conservative in using verbs in different argu-
ment structure constructions, but very liberal in substituting a range of nouns in
argument position in a given verbal construction (see §11.3.4 and references cited
therein). That is, a child learning the verb break will first learn to substitute dif-
ferent participants in an argument position, such as Mommy break, Daddy break
8 Do is sometimes found with the positive imperative as well (Denison 1998:252).

and so on; and only later will the child learn to use different arguments with the
verb, such as Break cup, Mommy break cup, Break with stick and so on. These
results imply that children generalize across different participant types quickly and
early, but only later generalize across predicate types and the argument structure
constructions that characterize them.
The evidence presented here, if it is borne out by further studies, would al-
low us to restructure the taxonomic organization of sentence-level constructions.
Although a taxonomic organization for constructions must allow for multiple par-
ents, we can use degree of semantic similarity to rank the syntactic distinctions and
thus form a hierarchy, as was done in example (13) (§11.2.4) for morphological
paradigms. Such an organization is illustrated in (51) for the Imperative half of the
basic Declarative-Imperative split:
(51) Imperative: (POL) PREDPHRASE
,
Positive: PREDPHRASE Negative: Don t PREDPHRASE
, ,
VERB be ADJ etc. Don t VERB Don t be ADJ etc.
, ,
Jump! etc. Be happy! etc. Don t jump! etc. Don t be cruel! etc.
The diagram in (51) is a visual means to represent strength of connections

or semantic distance between constructions. Another means to visually repre-
sent semantic distance between constructions is employed in typological theory
and adopted by Radical Construction Grammar (§10.2.4). This is the semantic
map model (see Croft 2001:92–98; 2003b:133–39; Haspelmath 2003, for refer-
ences and general explication). In the semantic map model, semantic distance is
represented in a multidimensional conceptual space, whose dimensions corre-
spond to semantic properties. In the example in (50), the semantic dimensions
are declarative-imperative, positive-negative and verbal-nonverbal (semantically
defined as actions-properties etc.). Such a mapping is reduced by convenience to
two dimensions. In this case, it corresponds roughly to the spatial arrangement of
a morphological or constructional paradigm such as that in (50). For illustration,
we will restrict ourselves to the imperative half of the paradigm in (50), that is,
(50c–f).
Constructions in a single language (or across languages, for cross-linguistic
comparison) are mapped onto the regions of the conceptual space according to their
use. For example, one may map the English Copula construction, the prohibitive
Don’t construction, and the imperative Subjectless Clause construction on the
conceptual space corresponding to the imperative paradigm in (50), as in (52):
(52)
Subjectless Clause
,
Don t
positive negative
action action
imperative imperative
positive negative
property property
imperative imperative
Copula
The lowercase labels ‘positive action imperative’ and so on designate points in

conceptual space that can be expressed by constructions in a language such as
English. The horizontal dimension of the space corresponds to positive-negative
polarity and the vertical dimension to the action-property lexical semantic classes.
The boxes map the occurrence of different English constructions in the conceptual
space (compare the examples in [50c–f]).
The basic principle guiding the structure of conceptual space and semantic maps
is formulated in the Semantic Map Connectivity Hypothesis: constructions must
map onto a continuous region of conceptual space (Croft 2001:96; 2003b:134).
This is one manifestation of the effect of semantic distance on formal structure:
constructional uses must be semantically connected, at least historically. There is
a dynamic hypothesis as well, namely that constructions are extended in use along
connected paths in conceptual space (Croft 2001:101–2). In the example above, for
instance, the Don’t imperative was historically extended from the negative action
imperative to the negative property imperative.
Typologists have applied this model cross-linguistically, so that the Semantic
Map Connectivity Hypothesis is a universal hypothesis. Patterns of semantic maps
across languages also suggest that the internal structure of grammatical categories,
that is, relations among exemplars, is universal, while boundaries are language-
specific (Croft 2001:103; compare chapter 4). Hence, the typological evidence
indicates that at least the broad structure of the conceptual space is universal and
therefore makes up part of human cognition, although boundaries are variable
and hence less constrained by the nature of human cognition (Croft 2001:105;
2003b:138–39).
The semantic map model represents semantic relationships between construc-
tions. Formal taxonomic relationships can be superimposed on the semantic map
model. Relative semantic distances implied by the relevance hypothesis can be
represented by relative distance in the conceptual space: (in [52], for example, the
negative functions are closer to each other than either is to their positive counter-
parts). These relative distances impose constraints on the taxonomic organization
of construction (in this case, requiring the positive and negative forms to be grouped
together first in the taxonomy; compare [52] to [51]). Further structures must be
imposed on conceptual space to allow for the formulation of constraints on the
grammatical expression of conceptual structures (see, e.g., Croft 2001:160–61,
163–64, 169–70; 2003b:140–43).
The exploration of semantic relations between constructions and their con-
straints on formal properties of constructions is still in its infancy. Presumably,
further research will allow construction grammarians to impose further structure
on the network organization of syntactic as well as morphological knowledge.
11.3.4 The acquisition of syntax and syntactic change

In §11.3.3, we referred to evidence from language acquisition on the
organization of the construction network. More generally, recent research in child
language development offers evidence for a usage-based, inductive model of the
acquisition of syntax.
Evidence from very detailed longitudinal studies of early language development
demonstrates that children are in fact extremely conservative language learners
(Braine 1976 is an early important study along these lines; for more recent stud-
ies, see Tomasello 2000, 2003 and references cited therein). Children’s earliest
multiword utterances demonstrate that children use verbs and other predicates in
only one construction at a time (Tomasello 1992; Lieven, Pine and Baldwin 1997;
Tomasello et al. 1997; Pine, Lieven and Rowland 1998).
In other words, children do not utilize schematic categories such as [Ve r b]
or schematic constructions such as the transitive construction [S b j Ve r b O b j]
in their early acquisition, whether these schematic structures are innate or not.
Instead, children begin with very low-level generalizations based around a single
predicate and a single construction in which that predicate occurs, and only later
in acquisition learn more schematic categories and constructions.
The main exception to this highly specific acquisition process is that, as noted
in §11.3.3, children do substitute different object names in a single participant
role in a construction from early on. Tomasello (1992) proposes the Verb Island
Hypothesis, namely that verbs and other predicates form ‘islands’ of a single verb
plus a single construction, before joining together the ‘islands’ into a construction
network such as that illustrated in (50) above.
Although children substitute object names or ‘nouns’ early in acquisition, it
does not appear that this implies that children acquire a schematic [Noun] or
[Determiner Noun ] noun phrase category early on. Pine and Lieven (1997)
found that at the earliest stage of learning nouns and determiners, children also
proceed in a piecemeal fashion. In their study, Pine and Lieven found that, although
children use a variety of nouns with both a and the, the nouns they use with a and the
nouns they use with the overlap very little at first. Instead, it appears that children
learned nouns with one determiner, or that the determiner use was associated
with larger structures in which the noun and determiner occur, such as [in the X]
or [That’s a X].
Pine, Lieven and Rowland (1998) studied the first six months of twelve chil-
dren’s multiword speech and found evidence that children begin with lexically
quite specific constructions, but that it was not always verbs that functioned as the
‘islands’ around which utterances were learned. For example, children produced
utterances with the auxiliaries can, do, be and have, constituting an average 90.3%
of all children’s utterances (Pine et al. 1998:818). However, there was very lit-
tle overlap between the verbs used with each auxiliary for each child (only one
child with one pair of auxiliaries had an overlap significantly different from zero;
1998:819). This suggests that the children are learning lexically specific auxiliary-
verb combinations and have not yet developed a productive [Au x] or [Ve r b ]
category in their utterances.
A still more fine-grained study confirms that early acquisition begins piecemeal
and indicates that acquisition is sensitive to token frequency in the input. Rubino
and Pine (1998) conducted a longitudinal study of a child learning Brazilian Por-
tuguese, and argued that the acquisition of sentence constructions with subject-verb
agreement began in a piecemeal fashion, with the acquisition of singular and plural
agreement beginning independently (in fact, in succession). Later, the child began
to overregularize the third singular agreement affixes at the time that the child
began to produce the third plural agreement affixes (Rubino and Pine 1998:51).
This developmental sequence suggests that an initial stage of rote learning was
followed by ‘joining the islands’ of singular and plural agreement to induce a
system of number agreement in the third person.
The overall average of errors produced by the child was quite low, which is
what one would expect in a model of conservative, inductive language learning. A
breakdown of error rates by person and number indicated that frequency of forms
in the input defined the course of acquisition of subject-verb agreement. The child
acquired correct subject-verb agreement for the most frequent agreement forms in
the input first, and the first correct productions of the less frequent subject-verb
agreement combinations appeared with high frequency verbs in the input (Rubino
and Pine 1998:53). However, frequency in the input does not appear to be the only
factor determining acquisition. Gathercole, Sebastián and Soto (1999) examined
the acquisition of Spanish verbal forms in two children, and observed the same
piecemeal, incremental acquisition process, and some correlation with frequency

of the form in the input. However, it appeared that morphological complexity of
the verbal form also played a role in the order of acquisition of verbal forms by
the children studied.
These and other language acquisition studies suggest that a careful, detailed
examination of the actual course of development of children’s language acquisi-
tion conforms to the predictions of the usage-based model. Children begin with
very narrow construction types, even specific to individual verbs and nouns, and
gradually build more schematic grammatical constructions over time. The rate
of learning and generalization is influenced by the relative frequency of the con-
structions in the caregivers’ input. The order of acquisition is also sensitive to the
semantic distance between constructions, as described in §11.2.4 and §11.3.3.
Similar results are found in the detailed examinations of the paths of syntactic
change. As many historical linguists have observed in detailed studies, the birth
and growth of a construction proceeds in an incremental fashion, not unlike the
expansion from ‘islands’ of highly specific constructions as in child language
acquisition.
One example of a syntactic change, cast in a cognitive linguistic framework,
is Israel’s analysis of the development of the way construction, illustrated in (53)
(Israel 1996:218):
(53) a. Rasselas dug his way out of the Happy Valley.

b. The wounded soldiers limped their way across the field.
c. ?Convulsed with laughter, she giggled her way up the stairs.
All of the way construction examples given in (53) use a possessed direct object way
and require a complement describing the path of motion. Example (53a) describes
a means of achieving the motion along the path; (53b) describes a manner of
motion along the path, and example (53c) describes an incidental activity of the
subject as she travels along the path. The way construction is also syntactically
and semantically idiosyncratic: the verbs in the way construction are normally
intransitive, and their meaning does not normally entail motion.
Using data from the Oxford English Dictionary and the Oxford University Press
corpus of contemporary English, Israel argues that the modern way construction
grew gradually from two different, more narrowly used way constructions, the
means and manner constructions (a third source, the acquisition or continued
possession of a path, shrank rather than expanded, although it remains in certain
common instances such as find one’s way; Israel 1996:221, n. 3). The manner
construction began as a special case of the Middle English [go one’s Pat h]
construction, and was originally found with only the most common general motion
verbs, no more than sixteen verbs before 1700 (Israel 1996:221). Verbs encoding
manner and path of motion began to be used with the manner way construction,
and in the nineteenth century, expanded particularly in the domain of laborious
motion (plod, scramble, fumble) and tortuous path (thread, worm, insinuate). At
the end of the nineteenth century, the manner way construction expanded to verbs
expressing noise accompanying motion (crunch, crash, toot).
The means way construction does not emerge until around 1650, and began
with verbs describing path clearing (cut, furrow out) and road building (pave,
smooth), as well as forcible motion (force out; Israel 1996:223). Expansion begins
with the cutting verbs and extends to fighting verbs starting around 1770. In the
nineteenth century, progressively more indirect means of reaching the goal, as in
He . . . smirked his way to a pedagogal desk (Israel 1996:224, from New Monthly
Magazine VII.386, 1823). At this point the means and manner way constructions
appear to merge, and in the late nineteenth century one finds the first examples of
incidental activity, which is quite distantly related to motion, as in He . . . whistled
his way to the main front door (Israel 1996:225, from Blackmore, Cradock Nowell
xvi, 1866).
At the same time that the class of verbs in the way construction is expanding,
the overall syntactic form of the construction becomes narrower, from allowing
other nouns than way and an optional path expression to obligatory way and path
expression (Israel 1996:221, 226). This (common) pattern in syntactic change
illustrates how a new construction emerges from an often highly specific instance
of an existing construction schema and then expands in its own direction. A usage-
based model can account for this pattern in that it allows for the entrenchment of
specific instances of construction schemas, which function as ‘islands’ from which
a new construction expands, establishing and generalizing a new construction
schema with its own syntactic and semantic peculiarities.
11.4 Conclusion
In this chapter, the construction grammar model of the representation of

grammatical knowledge was linked to the processes that use that knowledge, and
to the relationship between the representations and what is represented (utterances
in discourse). The relationship between representations and what is represented
is essentially one of categorization: categorization of the experience to be com-
municated and the utterance that is used as instances of the grammatical category
of known constructions, symbolizing experiences of the same category. The cat-
egorization relation between language use and grammatical knowledge is also
sensitive to frequency of use of grammatical constructions at different levels of
schematicity, that is, the process of language use influences the structure of the
representation. This model of grammatical representation and the processes that

use it is the usage-based model. The formal representation of the usage-based
model is as an activation network, in which activation corresponds to the process
of language use, and entrenchment (or decay) is the effect of the process on the
representation.
A number of general hypotheses about grammatical representation and process
have been proposed by researchers into the usage-based model (in addition to those
proposed in chapters 9–11). Productivity is hypothesized to emerge from a high
type frequency, which can be generalized as the hypothesis that entrenchment of
schematic constructions is proportional to the number of discrete instances of that
construction. Generalizations are defined as schemas rather than rules producing
an output structure from an input structure. In addition to source-oriented schemas
that can be modeled by rules, there is evidence for product-oriented schemas in
morphology and syntax, which cannot be easily modeled by rules. The organi-
zation of constructions is sensitive to their relative semantic distance from each
other (relevance/semantic connections), which imposes further structure on the
taxonomic network of constructions with multiple parents. This organization can
be represented as semantic maps of constructions on conceptual space, the struc-
ture of which appears to be universal in large part. Finally, recent research on
language acquisition indicates that both morphology and syntax are acquired in
a gradual, piecemeal, inductive fashion. Although many of these hypotheses are
recent and hence the evidence supporting them is fragmentary, the cognitive lin-
guistic model of grammatical knowledge is an important and expanding strand of
cognitive linguistics research.
12
Conclusion: cognitive linguistics
and beyond
The contemporary movement of cognitive linguistics began largely as an approach

to the analysis of linguistic meaning and grammatical form in response to truth-
conditional semantics and generative grammar. In this book, we have focused on the
analyses of syntax and semantics in cognitive linguistics, based on the three funda-
mental hypotheses presented in chapter 1: language is not an autonomous cognitive
faculty; grammar is conceptualization; and knowledge of language emerges from
language use. However, these basic hypotheses have consequences beyond the
narrow confines of linguistics, and also the narrow confines of the mental repre-
sentation of linguistic knowledge. A number of cognitive linguists have pushed
the boundaries of cognitive linguistics; conversely, a number of critics have chal-
lenged cognitive linguistics to go beyond its boundaries. We conclude our survey
by pointing out some of the ways cognitive linguistics has gone, and should go,
beyond its boundaries.
The hypothesis that language is not an autonomous cognitive faculty has im-
plied that conceptual structures and processes proposed for language should be
essentially the same as those found in nonlinguistic human cognition. At first, this
meant that cognitive linguists drew on the results of cognitive psychology (and, to
a lesser extent, philosophy) in developing cognitive linguistic analyses. It also led
to the analysis of conceptual structure using linguistic evidence, as described in
Part I of this book. More recently, the reverse has taken place: cognitive linguists
have applied the conceptual analyses of cognitive linguistics to other cognitive do-
mains. The cognitive linguistic models of metaphor have been applied to literary
analysis (Turner 1987; Lakoff and Turner 1989), philosophy and ethics (Johnson
1987, 1993; Lakoff and Johnson 1999; for a sympathetic but critical review, see
Neisser 2001), politics (Lakoff 1996) and mathematics (Lakoff and Núñez 2000).
Blending theory has been applied to a general theory of mind, human culture and
the evolution of humankind (Turner 2001; Fauconnier and Turner 2002). These
forays outside of linguistics are ambitious and highly controversial, but if the first
hypothesis is valid, then there should be some implications from the results of
cognitive linguistic analysis for nonlinguistic cognition.
328
Conclusion 329
The hypothesis that grammar is conceptualization, as well as the other two

basic hypotheses, implies a close relationship between cognitive linguistics and
cognitive psychology. Indeed, research on categorization was inspired by the work
of psychologists such as Rosch (chapter 4), and research on construal operations
was inspired by the work of the Gestalt psychologists (chapter 3). More recently,
cognitive psychologists have been influenced by cognitive linguistic research (e.g.
Gibbs 1994; see also Tomasello 1998, 2000, 2003), and construction grammar in
particular has dovetailed with research in language acquisition (see §11.3.4). Also,
there has been considerable mutual interest between cognitive linguists and those
advocating activation network and parallel distributed processing models of lan-
guage (see, for example, Elman et al. 1996). Unfortunately, space has prevented us
from discussing the many experimental studies in psychology whose results are rel-
evant to cognitive linguistics. Nevertheless, there is considerable scope for further
interaction between cognitive psychology and cognitive linguistics, in particular
for critical experimental testing of cognitive linguistic hypotheses, and a refine-
ment of the linguistic assumptions behind the experimental designs of cognitive
psychologists.
Finally, the hypothesis that knowledge of language emerges from language use
reveals both a weakness and an opportunity for cognitive linguistics to respond to its
own critics, who look at language from a different perspective than either cognitive
linguists or the generative grammarians and truth-conditional semanticists that
cognitive linguists were originally responding to. These critics attack all of the latter
approaches to language for being focused exclusively on the mind of the speaker
or hearer, and ignoring the central function of language as communication and the
role of social interaction (‘discourse’) and the social structures it presupposes in
understanding why language is the way it is.
It is true that cognitive linguistics, as its name implies, has focused its attention
on mental representations and cognitive processes, and has only recently begun
to respond to the discourse and functionalist approaches to language (see, for ex-
ample, Langacker 1999, chapters 4 and 12; 2002). The hypothesis that knowledge
of language emerges from language use provides an opportunity for cognitive lin-
guistics to engage with the social-interactional nature of language. Usage events
are of course acts of social interaction. Speakers construe their experience for the
purpose of communicating that experience to others, which in turn has broader
social-interactional purposes; and hearers likewise invoke a construal of the utter-
ance for those broader purposes. In sum, cognitive linguistics has the potential to
make a contribution to a theory of language that goes beyond cognition, as well
as a theory of cognition that goes beyond language.
References
Akmajian, Adrian. 1984. Sentence types and the form-function fit. Natural Language and
Linguistic Theory 2:1–23.
Allan, Keith. 1986. Interpreting English comparatives. Journal of Semantics 5:1–50.
Andersen, Henning. 1980. Russian conjugation: acquisition and evolutive change. Papers
from the 4th International Conference on Historical Linguistics, ed. Elizabeth
Closs Traugott et al., 285–302. Amsterdam: John Benjamins.
Ariel, Mira. 1990. Accessing noun phrase antecedents. New York: Routledge.
Armstrong, Sharon Lee, Lila R. Gleitman and Henry Gleitman. 1983. What some concepts
might not be. Cognition 13:263–308.
Aronoff, Mark. 1993. Morphology by itself: items and inflectional classes. Cambridge,
Mass.: MIT Press.
Atran, Scott. 1990. Cognitive foundations of natural history: towards an anthropology of
science. Cambridge: Cambridge University Press.
Barlow, Michael and Suzanne Kemmer (eds.). 2000. Usage-based models of language.
Stanford: Center for the Study of Language and Information.
Barsalou, Lawrence W. 1983. Ad hoc categories. Memory and Cognition 11: 211–27.
1987. The instability of graded structure: implications for the nature of concepts. Concepts
and conceptual development: ecological and intellectual factors in categorization,
ed. Ulric Neisser, 101–40. Cambridge: Cambridge University Press.
1992a. Frames, concepts, and conceptual fields. Frames, fields and contrasts: new essays
in semantic and lexical organization, ed. Adrienne Lehrer and Eva Feder Kittay,
21–74. Hillsdale, N.J.: Lawrence Erlbaum Associates.
1992b. Cognitive psychology: an overview for cognitive scientists. Hillsdale, N.J.:
Lawrence Erlbaum Associates.
Becker, Joseph. 1975. The phrasal lexicon. Bolt, Beranek and Newman Report No. 3081,
Artificial Intelligence Report No. 28.
Berlin, Brent and Paul Kay. 1969. Basic color terms: their universality and evolution.
Berkeley: University of California Press.
Berlin, Brent, Dennis E. Breedlove and Peter H. Raven. 1973. General principles
of classification and nomenclature in folk biology. American Anthropologist
75:214–42.
Birner, Betty J. and Gregory Ward. 1998. Information status and noncanonical word
order in English. (Studies in Language Companion Series, 40.) Amsterdam: John
Benjamins.
Black, Max. 1993[1979]. More about metaphor. Metaphor and thought (2nd ed.), ed. An-
drew Ortony, 19–41. Cambridge: Cambridge University Press.
330
References 331
Bolinger, Dwight. 1971. The phrasal verb in English. Cambridge, Mass.: Harvard University
Press.
1980. Language, the loaded weapon. London: Longmans.
1992. About furniture and birds. Cognitive Linguistics 3:111–18.
Braine, Martin D. S. 1976. Children’s first word combinations. (Monographs of the
Society for Research in Child Development 41, no. 1.) Chicago: University of
Chicago Press.
Bresnan, Joan (ed.). 1982. The mental representation of grammatical relations. Cambridge,
Mass.: MIT Press.
Brown, Cecil H. 2002. Paradigmatic relations of inclusion and identity I: Hyponymy.
Lexicology: an international handbook on the nature and structure of words and
vocabularies (Handbooks of Linguistics and Communication Science, 21), ed.
D. Alan Cruse, Franz Hundsnurscher, Michael Job and Peter Lutzeier, ch. 47. Berlin:
Walter de Gruyter.
Brown, Roger. 1958. How shall a thing be called? Psychological Review 65:14–21.
Bybee, Joan L. 1985. Morphology: a study into the relation between meaning and form.
Amsterdam: John Benjamins.
1995. Regular morphology and the lexicon. Language and Cognitive Processes
10:425–55.
2001. Phonology and language use. Cambridge: Cambridge University Press.
Bybee, Joan L. and Mary Alexandra Brewer. 1980. Explanation in morphophonemics:
changes in Provençal and Spanish preterite forms. Lingua 52:271–312.
Bybee, Joan and Paul Hopper. 2001. Frequency and the emergence of linguistic structure.
(Typological Studies in Language, 45.) Amsterdam: John Benjamins.
Bybee, Joan L. and Carol Lynn Moder. 1983. Morphological classes as natural categories.
Language 59:251–70.
Bybee, Joan L. and Elly Pardo. 1981. On lexical and morphological conditioning of alter-
nations: a nonce-probe experiment with Spanish verbs. Linguistics 19:937–68.
Bybee, Joan L. and Joanne Scheibman. 1999. The effect of usage on degrees of constituency:
the reduction of don’t in English. Linguistics 37:575–96.
Bybee, Joan L. and Dan I. Slobin. 1982. Rules and schemas in the development and use of
the English past tense. Language 58:265–89.
Bybee, Joan L. and Sandra A. Thompson. 1997. Three frequency effects in syntax. Proceed-
ings of the 23rd Annual Meeting of the Berkeley Linguistics Society, ed. Matthew
L. Juge and Jeri O. Moxley, 378–88. Berkeley: Berkeley Linguistics Society.
Chafe, Wallace. 1994. Discourse, consciousness and time: the flow and displacement of
conscious experience in speaking and writing. Chicago: University of Chicago Press.
Chafe, Wallace (ed.). 1980. The pear stories. New York: Ablex.
Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris.
1993. A minimalist program for linguistic theory. The view from Building 20, ed. Kenneth
Hale and Samuel Jay Keyser, 1–52. Cambridge, Mass.: MIT Press.
1995. The minimalist program. Cambridge, Mass.: MIT Press.
Clahsen, Harald and M. Rothweiler. 1992. Inflectional rules in children’s grammar: evidence
from the development of participles in German. Yearbook of Morphology 1992,
1–34.
Clark, Eve V. and Herbert H. Clark. 1979. When nouns surface as verbs. Language
55:767–811.
332 References
Clark, Herbert H. 1996. Using language. Cambridge: Cambridge University Press.

Clark, Herbert H. and Susan Haviland. 1977. Comprehension and the given-new contract.
Discourse production and comprehension, ed. R. Freedle, 1–40. Norwood, N.J.:
Ablex.
Clausner, Timothy C. 1991. Sentence types and salience. University of Michigan working
papers in linguistics, ed. William Croft, 1–11. Ann Arbor: Department of Linguis-
tics, University of Michigan.
Clausner, Timothy C. and William Croft. 1997. The productivity and schematicity of
metaphor. Cognitive Science 21:247–82.
1999. Domains and image-schemas. Cognitive Linguistics 10:1–31.
Coleman, Linda and Paul Kay. 1981. Prototype semantics. Language 57:26–44.
Collins, Allan M. and M. Ross Quillian. 1969. Retrieval time from semantic memory.
Journal of Verbal Learning and Verbal Behaviour 8:240–47.
Cornwell, Patricia. 2000. Black notice. London: Warner Books.
Coseriu, Eugenio. 1975. Vers une typologie des champs lexicaux. Cahiers de Lexicologie
27:30–51.
Coulson, Seana. 2000. Semantic leaps: frame-shifting and conceptual blending in meaning
construction. Cambridge: Cambridge University Press.
Croft, William. 1991. Syntactic categories and grammatical relations: the cognitive orga-
nization of information. Chicago: University of Chicago Press.
1993 [2002]. The role of domains in the interpretation of metaphors and metonymies.
Cognitive Linguistics 4:335–70. Revised version printed in Metaphor and metonymy
in comparison and contrast, ed. René Dirven and Ralf Pörings, 161–205. Berlin:
Mouton de Gruyter, 2002.
1995. Autonomy and functionalist linguistics. Language 71:490–532.
1998a. The structure of events and the structure of language. The new psychology of
language: cognitive and functional approaches to language structure, ed. Michael
Tomasello, 67–92. Mahwah, N. J.: Lawrence Erlbaum Associates.
1998b. Event structure in argument linking. The projection of arguments: lexical and
compositional factors, ed. Miriam Butt and Wilhelm Geuder, 1–43. Stanford: Center
for the Study of Language and Information.
1998c. Linguistic evidence and mental representations. Cognitive Linguistics 9:151–
73.
2000. Explaining language change: an evolutionary approach. London: Longman.
2001. Radical construction grammar: syntactic theory in typological perspective. Oxford:
Oxford University Press.
2003a. Lexical rules vs. constructions: a false dichotomy. Motivation in language: studies
in honour of Günter Radden, ed. Hubert Cuyckens, Thomas Berg, René Dirven and
Klaus-Uwe Panther, 49–68. Amsterdam: John Benjamins.
2003b. Typology and universals (2nd ed.). Cambridge: Cambridge University Press.
In preparation. Verbs: aspect and argument structure. Oxford: Oxford University Press.
Croft, William and Esther J. Wood. 2000. Construal operations in linguistics and artifi-
cial intelligence. Meaning and cognition: a multidisciplinary approach, ed. Liliana
Albertazzi, 51–78. Amsterdam: John Benjamins.
Cruse, D. Alan. 1977. The pragmatics of lexical specificity. Journal of Linguistics
13:153–64.
1979. On the transitivity of the part-whole relation. Journal of Linguistics 15:29–38.
1986. Lexical semantics. Cambridge: Cambridge University Press.
References 333
1990. Prototype theory and lexical semantics. Meanings and prototypes: stud-
ies in linguistic categorization, ed. Savas L. Tzohatzidis, 382–402. London:
Routledge.
1992a. Antonymy revisited: some thoughts on the relation between words and concepts.
Frames, fields and contrasts: new essays in semantic and lexical organization, ed.
Adrienne Lehrer and Eva Feder Kittay, 289–306. Hillsdale, N. J.: Lawrence Erlbaum
Associates.
1992b. Cognitive linguistics and word meaning: Taylor on linguistic categorization.
(Review article on John R. Taylor, Linguistic categorization: prototypes in linguistic
theory.) Journal of Linguistics 28:165–83.
1994. Prototype theory and lexical relations. Rivista di Linguistica 6:167–88.
2000a. Lexical “facets”: between monosemy and polysemy. Sprachspiel und Bedeu-
tung: Festschrift für Franz Hundsnurscher zum 60. Geburtstag, ed. S. Beckmann,
P. P. König and T. Wolf, 25–36. Tübingen: Max Niemeyer Verlag.
2000b. Aspects of the micro-structure of word meanings. Polysemy: theoretical and
computational approaches, ed. Yael Ravin and Claudia Leacock, 30–51. Oxford:
Oxford University Press.
2002a. Microsenses, default specificity and the semantics-pragmatics boundary.
Axiomathes 1:1–20.
2002b. Hyponymy and its varieties. The semantics of relationships: an interdisciplinary
perspective, ed. Rebecca Green, Carol A. Bean and Sung Hyon Myaeng, 3–21.
Dordrecht: Kluwer.
Cruse, D. Alan., Franz Hundsnurscher, Michael Job and Peter Lutzeier (eds.). 2002.
Lexicology: an international handbook on the nature and structure of words and
vocabularies. (Handbooks of Linguistics and Communication Science, 21.1.)
Berlin: Walter de Gruyter.
Cruse, D. Alan and Pagona Togia. 1995. Towards a cognitive model of antonymy. Lexicology
1:113–41.
Daugherty, K. and Mark Seidenberg. 1992. Rules or connections? The past tense revisited.
Proceedings of the 14th Annual Meeting of the Cognitive Science Society 259–64.
Hillsdale, N. J.: Erlbaum.
1994. Beyond rules and exceptions: a connectionist modeling approach to inflectional
morphology. The reality of linguistic rules, ed. Susan D. Lima, R. L. Corrigan and
Gregory K. Iverson, 353–88. Amsterdam: John Benjamins.
Deane, Paul D. 1988. Polysemy and cognition. Lingua 75:325–61.
1991. Limits to attention: a cognitive theory of island phenomena. Cognitive Linguistics
2:1–63.
DeLancey, Scott. 1981. An interpretation of split ergativity and related patterns. Language
57:626–57.
Denison, David. 1993. English historical syntax. London: Longman.
1998. Syntax. The Cambridge history of the English language, vol 4: 1776–1997, ed.
Suzanne Romaine, 92–329. Cambridge: Cambridge University Press.
Dreyfus, Hubert L. 1991. Being-in-the-world: a commentary of Heidegger’s Being and
Time, Division I. Cambridge, Mass.: MIT Press.
Eco, Umberto. 1996. Metaphor. Philosophy of language: an international handbook of
contemporary research (Handbooks of Linguistics and Communication Science,
7.2), ed. Marcelo Dascal, Dietfried Gerhardus, Kuno Lorenz and Georg Meggle,
1313–23. Berlin: Walter de Gruyter.
334 References
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.
Elman, Jeffrey L. and James L. McClelland. 1984. Speech perception as a cognitive process:
the interactive activation model. Speech and language, vol. 10, ed. Norman Lass,
337–74. New York: Academic Press.
Fauconnier, Gilles. 1985 [1994]. Mental Spaces (2nd ed.). Cambridge: Cambridge Univer-
sity Press.
1986. Quantification, roles, and domains. Versus 44/45:61–80.
1997. Mappings in thought and language. Cambridge: Cambridge University Press.
Fauconnier, Gilles and Eve Sweetser (ed.). 1996. Spaces, worlds and grammar. Chicago:
University of Chicago Press.
Fauconnier, Gilles and Mark Turner. 1996. Blending as a central process in grammar.
Conceptual structure, discourse and language, ed. Adele E. Goldberg, 113–30.
Stanford, Calif.: CSLI Publications.
1994. Conceptual projection and middle spaces. San Diego: University of Cali-
fornia, Department of Cognitive Science Technical Report 9401 (available at
http://www.blending.stanford.edu).
2002. The way we think. New York: Basic Books.
Feyaerts, Kurt. 2000. Refining the inheritance hypothesis: interaction between metaphoric
and metonymic hierarchies. Metaphor and metonymy at the crossroads: a cognitive
perspective, ed. Antonio Barcelona, 59–78. Berlin: Mouton de Gruyter.
Fillmore, Charles J. 1975. An alternative to checklist theories of meaning. Proceedings of
the First Annual Meeting of the Berkeley Linguistics Society, ed. Cathy Cogen et
al., 123–31. Berkeley: Berkeley Linguistics Society.
1977a. Scenes-and-frames semantics. Linguistic structures processing (Fundamental
Studies in Computer Science, 5), ed. Antonio Zampolli, 55–81. Amsterdam: North-
Holland.
1977b. The case for case reopened. Grammatical relations (Syntax and Semantics, 8),
ed. Peter Cole and Jerrold M. Sadock, 59–82. New York: Academic Press.
1982a. Frame semantics. Linguistics in the morning calm, ed. The Linguistic Society of
Korea, 111–37. Seoul: Hanshin.
1982b. Ideal readers and real readers. Analyzing discourse: text and talk, ed. Deborah
Tannen, 248–70. Washington, D. C.: Georgetown University Press.
1985. Frames and the semantics of understanding. Quaderni di semantica 6:222–54.
1986a. U-semantics, second round. Quaderni di semantica 7:49–58.
1986b. Pragmatically-controlled zero anaphora. Proceedings of the Twelfth Annual Meet-
ing of the Berkeley Linguistics Society, ed. Vassiliki Nikiforidou et al., 95–107.
Berkeley: Berkeley Linguistics Society.
1999. Inversion and constructional inheritance. Lexical and constructional aspects of
linguistic explanation, ed. Gert Webelhuth, Jean-Pierre Koenig and Andreas Kathol,
113–28. Stanford, Calif.: CSLI Publications.
Fillmore, Charles J. and Beryl T. Atkins. 1992. Toward a frame-based lexicon: the semantics
of RISK and its neighbors. Frames, fields and contrasts: new essays in semantic and
lexical organization, ed. Adrienne Lehrer and Eva Feder Kittay, 75–102. Hillsdale,
N. J.: Lawrence Erlbaum Associates.
Fillmore, Charles J. and Paul Kay. 1993. Construction grammar coursebook, chapters 1
thru 11 (reading materials for Ling. X20). Berkeley: University of California.
References 335
Fillmore, Charles J., Paul Kay and Mary Kay O’Connor. 1988. Regularity and idiomaticity
in grammatical constructions: the case of let alone. Language 64:501–38.
Fillmore, Charles J., Paul Kay, Laura A. Michaelis and Ivan A. Sag. In preparation.
Construction Grammar. Stanford, Calif.: CSLI Publications.
Flickinger, Daniel, Carl Pollard and Thomas Wasow. 1985. Structure-sharing in lexical rep-
resentation. Proceedings of the 23rd Annual Meeting of the Association for Compu-
tational Linguistics, 262–67. Chicago: Association for Computational Linguistics.
Gadamer, Hans-Georg. 1989. Truth and method, trans. William Glen-Doepel, ed. John
Cumming and Garrett Barden, translation rev. Joel Weinsheimer and Donald G.
Marshall. London: Sheed & Ward.
Gathercole, Virginia C. Mueller, Eugenia Sebastián and Pilar Soto. 1999. The early acqui-
sition of Spanish verbal morphology: across-the-board or piecemeal knowledge?
International Journal of Bilingualism 3:133–82.
Gazdar, Gerald, Ewan Klein, Geoffrey Pullum and Ivan Sag. 1985. Generalized Phrase
Structure Grammar. Oxford: Basil Blackwell.
Geckeler, Horst. 1971. Strukturelle Semantik und Wortfeldtheorie. Munich: Fink.
Geeraerts, Dirk. 1993. Vagueness’s puzzles, polysemy’s vagaries. Cognitive Linguistics
4:223–72.
Geertz, Clifford. 1973. The interpretation of cultures. New York: Basic Books.
Gentner, Dedre. 1983. Structure-mapping: a theoretical framework for analogy. Cognitive
Science 7:155–70.
1988. Metaphor as structure mapping: the relational shift. Child Development 59:47–59.
Gibbs, Raymond W., Jr. 1990. Psycholinguistic studies on the conceptual basis of idiomatic-
ity. Cognitive Linguistics 1:417–51.
1994. The poetics of mind: figurative thought, language and understanding. Cambridge:
Cambridge University Press.
Gibbs, Raymond W. and J. O’ Brien. 1990. Idioms and mental imagery: the metaphorical
motivation of idiomatic meaning. Cognition 36:35–68.
Glucksberg, Samuel. 2001. Understanding figurative language. Oxford: Oxford University
Press.
Goldberg, Adele E. 1995. Constructions: a construction grammar approach to argument
structure. Chicago: University of Chicago Press.
Goossens, Louis. 1990. Metaphtonymy: the interaction of metaphor and metonymy in
expressions of linguistic action. Cognitive Linguistics 1:323–40.
Grady, Joseph E. 1997. Theories are buildings revisited. Cognitive Linguistics
8:267–90.
1998. The “conduit metaphor” revisited: a reassessment of metaphors for communica-
tion. Bridging the gap: discourse and cognition, ed. Jean-Pierre Koenig, 205–18.
Stanford, Calif.: Center for the Study of Language and Information.
Grady, Joseph E., Todd Oakley and Seanna Coulson. 1999. Blending and metaphor.
Metaphor in cognitive linguistics, ed. Raymond W. Gibbs Jr. and Gerard J. Steen,
101–24. Amsterdam: John Benjamins.
Gundel, Jeannette K., Nancy Hedberg and Ron Zacharski. 1993. Cognitive status and the
form of referring expressions in discourse. Language 69:274–307.
Haegeman, Liliane. 1994. Introduction to government and binding theory (2nd ed.). Oxford:
Basil Blackwell.
Hahn, Ulrike and Nick Chater. 1997. Concepts and similarity. Knowledge, concepts and
categories, ed. Koen Lamberts and David Shanks, 43–92. Hove: Psychology Press.
336 References
Haiman, John. 1980. Dictionaries and encyclopedias. Lingua 50:329–57.

Hampton, James. 1991. The combination of prototype concepts. The psychology of word
meanings, ed. P. Schwanenflugel, 91–116. Hillsdale, N. J.: Lawrence Erlbaum
Associates.
1997. Psychological representation of concepts. Cognitive models of memory, ed. Martin
A. Conway, 81–107. Hove: Psychology Press.
Hare, Mary and Jeffrey L. Elman. 1995. Learning and morphological change. Cognition
56:61–98.
Hare, Mary, Jeffrey L. Elman and Kim G. Daugherty. 1995. Default generalization in
connectionist networks. Language and Cognitive Processes 10:601–30.
Haspelmath, Martin. 1989. Schemas in Hausa plural formation: product-orientation and
motivation vs. source-orientation and generation. Buffalo Working Papers in
Linguistics, 89–01, 32–74.
2003. The geometry of grammatical meaning: semantic maps and cross-linguistic com-
parison. The new psychology of language, vol. 2, ed. Michael Tomasello, 211–42.
Mahwah, N. J.: Lawrence Erlbaum Associates.
Hawkinson, Annie and Larry Hyman. 1974. Natural hierarchies of topic in Shona. Studies
in African Linguistics 5:147–70.
Hayes, Victoria. 2001. A comparative analysis of antonymy and hybrid anto-
complementarity in English and German within a cognitive framework. Ph.D. thesis,
University of Manchester.
Heidegger, Martin. 1927 [1962]. Being and time, trans. John Macquarrie and Edward
Robinson. New York: Harper and Row. (Sein und Zeit. Tübingen: Neomarius
Verlag.)
Heider, Eleanor. 1971. Focal color areas and the development of color names. Develop-
mental Psychology 4:447–55.
1972. Universals in color naming and memory. Journal of Experimental Psychology
93:10–20.
Herskovits, Annette. 1986. Language and spatial cognition. Cambridge: Cambridge
University Press.
Hudson, Richard. 1984. Word grammar. Oxford: Basil Blackwell.
Hull, David L. 1988. Science as a process: an evolutionary account of the social and
conceptual development of science. Chicago: University of Chicago Press.
Husserl, Edmund. 1948 [1973]. Experience and judgement, trans. James S. Churchill and
Karl Ameriks. Evanston, Ill.: Northwestern University Press. (Erfahrung und Urteil,
ed. Ludwig Landgrebe. Hamburg: Claassen & Goverts.)
Israel, Michael. 1996. The way constructions grow. Conceptual structure, discourse and
language, ed. Adele E. Goldberg, 217–30. Stanford: Center for the Study of
Language and Information.
Jackendoff, Ray. 1983. Semantics and cognition. Cambridge, Mass.: MIT Press.
1990. Semantic structures. Cambridge, Mass.: MIT Press.
1997. Twistin’ the night away. Language 73:534–59.
Jackendoff, Ray and David Aaron. 1991. Review of George Lakoff and Mark Turner, More
than cool reason: a field guide to poetic metaphor. Language 67:320–38.
Janda, Richard D. 1990. Frequency, markedness and morphological change: on predicting
the spread of noun-plural -s in Modern High German and West Germanic. Pro-
ceedings of the Eastern States Conference on Linguistics (ESCOL ’90), 136–53.
Columbus: The Ohio State University.
References 337
Jespersen, Otto. 1942. A modern English grammar on historical principles, vol. 6. London:
Allen & Unwin.
Johnson, David E. and Paul M. Postal. 1980. Arc pair grammar. Guildford, N. J.: Princeton
University Press.
Johnson, Mark. 1987. The body in the mind. Chicago: University of Chicago Press.
1993. Moral imagination: implications of cognitive science for ethics. Chicago: Univer-
sity of Chicago Press.
Jolicoeur, Pierre S., Martin A. Gluck and Stephen M. Kosslyn. 1984. Pictures and names:
making the connection. Cognitive Psychology 19:31–53.
Kant, Immanuel. 1790 [1952]. The critique of judgement, trans. J. C. Meredith. Oxford:
Clarendon Press. (Kritik der Urteilskraft. Berlin: Bey Lagarde und Friederich.)
Katz, Jerrold J. and Jerry A. Fodor. 1963. The structure of a semantic theory. Language
39:170–210. (Reprinted in The structure of language: readings in the philosophy of
language, ed. Jerry A. Fodor and Jerrold J. Katz, 479–518. Englewood Cliffs, N. J.:
Prentice-Hall, 1964.)
Kay, Paul. 1997. Construction grammar feature structures (revised). Available at
http://www.icsi.berkeley.edu/∼kay/bcg/FSrev.html.
2002. English subjectless tagged sentences. Language 78:453–81.
Kay, Paul and Charles J. Fillmore. 1999. Grammatical constructions and linguistic gener-
alizations: the What’s X doing Y? construction. Language 75:1–33.
Kimenyi, Alexandre. 1980. A relational grammar of Kinyarwanda. (University of California
Publications in Linguistics, 91.) Berkeley: University of California Press.
Kleiber, Georges. 1996. Cognition, sémantique et facettes: une “histoire” de livres et
de . . . romans. Les formes du sens, ed. Georges Kleiber and Martin Riegel, 319–31.
Louvain: Duculot.
Koffka, Kurt. 1935. Principles of Gestalt psychology. New York: Harcourt, Brace & World.
Köpcke, Klaus-Michael. 1988. Schemas in German plural formation. Lingua 74:303–35.
1998. The acquisition of plural marking in English and German revisited: schemata versus
rules. Journal of Child Language 25:293–319.
Kövecses, Zoltán. 2002. Metaphor: a practical introduction. Oxford: Oxford University
Press.
Kövecses, Zoltán and Günter Radden. 1998. Metonymy: developing a cognitive linguistic
view. Cognitive Linguistics 9:37–77.
Kuno, Susumu. 1987. Functional syntax: anaphora, discourse and empathy. Chicago: Uni-
versity of Chicago Press.
Kuno, Susumu and Etsuko Kaburaki. 1977. Empathy and syntax. Linguistic Inquiry
8:627–72.
Labov, William. 1973. The boundaries of words and their meanings. New ways of analyzing
variation in English, ed. Joshua Fishman, 340–73. Washington D. C.: Georgetown
University Press.
Lakoff, George. 1973. Hedges: a study in meaning criteria and the logic of fuzzy concepts.
Journal of Philosophical Logic 2:458–508.
1987. Women, fire and dangerous things: what categories reveal about the mind. Chicago:
1990. The Invariance Hypothesis: Is abstract reason based on image-schemas? Cognitive
Linguistics 1:39–74.
1993. The contemporary theory of metaphor. Metaphor and thought, ed. Andrew Ortony,
202–51. Cambridge: Cambridge University Press.
338 References
1996. Moral politics. Chicago: University of Chicago Press.

Lakoff, George and Mark Johnson. 1980. Metaphors we live by. Chicago: University of
Chicago Press.
1999. Philosophy in the flesh. New York: Basic Books.
Lakoff, George and Rafael Núñez. 2000. Where mathematics comes from. New York: Basic
Books.
Lakoff, George and Eve Sweetser. 1994. Foreword to Gilles Fauconnier, Mental Spaces,
ix–xlvi. Cambridge: Cambridge University Press.
Lakoff, George and Mark Turner. 1989. More than cool reason: a field guide to poetic
metaphor. Chicago: University of Chicago Press.
Lambrecht, Knud. 1990. “What, me worry?” – ‘Mad Magazine’ sentences revisited. Pro-
ceedings of the Sixteenth Annual Meeting of the Berkeley Linguistics Society, ed.
Kira Hall, Jean-Pierre Koenig, Michael Meacham, Sondra Reinman and Laurel A.
Sutton, 215–28. Berkeley: Berkeley Linguistics Society.
1994. Information structure and sentence form: topic, focus and the mental representa-
tions of discourse referents. Cambridge: Cambridge University Press.
Langacker, Ronald W. 1976. Semantic representations and the linguistic relativity hypoth-
esis. Foundations of Language 14:307–57.
1987. Foundations of cognitive grammar, vol. 1: theoretical prerequisites. Stanford,
Calif.: Stanford University Press.
1988. An overview of cognitive grammar. Topics in cognitive linguistics, ed. Brygida
Rudzka-Ostyn, 3–48. Amsterdam: John Benjamins.
1991a. Foundations of cognitive grammar, vol. 2: descriptive application. Stanford,
Calif.: Stanford University Press.
1991b. Concept, image, and symbol: the cognitive basis of grammar. Berlin: Mouton de
Gruyter.
1998. On subjectification and grammaticization. Bridging the gap: discourse and cog-
nition, ed. Jean-Pierre Koenig, 71–89. Stanford, Calif.: Center for the Study of
Language and Information.
1999. Grammar and conceptualization. Berlin: Mouton de Gruyter.
2002. Discourse and cognitive grammar. Cognitive Linguistics 12:143–88.
Larsen, Thomas W. and William M. Norman. 1979. Correlates of ergativity in Mayan
grammar. Ergativity, ed. Frans Plank, 347–69. New York: Academic Press.
Levinson, Stephen C. 1983. Pragmatics. Cambridge: Cambridge University Press.
Lewis, David. 1969. Convention. Cambridge, Mass.: MIT Press.
Lieven, Elena V. M., Julian M. Pine and Gillian Baldwin. 1997. Lexically-based learn-
ing and early grammatical development. Journal of Child Language 24:187–
219.
Lobben, M. 1991. Pluralization of Hausa nouns, viewed from psycholinguistic experiments
and child language data. M. Phil. dissertation, University of Oslo, Norway.
Losiewicz, B. L. 1992. The effect of frequency on linguistic morphology. Ph.D. dissertation,
University of Texas, Austin.
Lyons, John. 1963. Structural semantics. Cambridge: Cambridge University Press.
1968. Introduction to theoretical linguistics. Cambridge: Cambridge University Press.
Makkai, Adam. 1972. Idiom structure in English. The Hague: Mouton.
Marcus, G., U. Brinkmann, H. Clahsen, R. Wiese, A. Woest and S. Pinker. 1995. German
inflection: the exception that proves the rule. Cognitive Psychology 29:189–256.
References 339
Markman, Arthur B. 1999. Knowledge representation. Mahwah, N. J.: Lawrence Erlbaum

Associates.
Matthews, Peter H. 1981. Syntax. Cambridge: Cambridge University Press.
McCarthy, John and Alan Prince. 1990. Foot and word in prosodic morphology: the Arabic
broken plural. Natural Language and Linguistic Theory 8:209–83.
McCawley, James D. 1981 [1993]. Everything that linguists have always wanted to know
about logic (but were ashamed to ask) (2nd ed.). Chicago: University of Chicago
Press.
McGregor, William B. 1997. Semiotic grammar. Oxford: Clarendon.
Meyer-Lübke, W. 1923. Grammaire des langues romanes. New York: Stechert.
Michaelis, Laura A. and Knud Lambrecht. 1996. Toward a construction-based theory of
language functions: the case of nominal extraposition. Language 72:215–47.
Mondloch, James L. 1978. Basic Quiché grammar. (Institute for Mesoamerican Studies,
Publication 2.) Albany: Institute for Mesoamerican Studies.
Moore, Terence and Christine Carling. 1982. Understanding language: towards a post-
Chomskyan linguistics. London: Macmillan.
Mossé, Fernand. 1952. A handbook of Middle English, transl. James Albert Walker. Balti-
more: The Johns Hopkins University Press.
Murphy, Gregory L. and Mary E. Lassaline. 1997. Hierarchical structure in concepts and
the basic level of categorization. Knowledge, concepts and categories, ed. Koen
Lamberts and David Shanks, 93–132. Hove: Psychology Press.
Murphy, Gregory L. and Douglas L. Medin. 1985. The role of theories in conceptual
coherence. Psychological Review 92:289–316.
Neisser, Joseph Ulric. 2001. Review of George Lakoff and Mark Johnson, Philosophy in
the flesh. Language 77:166–68.
Norman, William M. 1978. Advancement rules and syntactic change: the loss of in-
strumental voice in Mayan. Proceedings of the Fourth Annual Meeting of the
Berkeley Linguistics Society, ed. Jeri J. Jaeger, Anthony C. Woodbury, Farrell
Ackerman, Christine Chiarello, Orin D. Gensler, John Kingston, Eve. E. Sweetser,
Henry Thompson and Kenneth W. Whistler, 458–76. Berkeley: Berkeley Linguistics
Society.
Nunberg, Geoffrey. 1995. Transfers of meaning. Journal of Semantics 12:109–32.
Nunberg, Geoffrey, Ivan A. Sag and Thomas Wasow. 1994. Idioms. Language 70:491–538.
Omar, M. 1973. The acquisition of Egyptian Arabic as a native language. The Hague:
Mouton.
Perlmutter, David M. (ed.). 1983. Studies in relational grammar 1. Chicago: University of
Chicago Press.
Pine, Julian and Elena V. M. Lieven. 1997. Slot and frame patterns and the development of
the determiner category. Journal of Child Language 18:123–38.
Pine, Julian, Elena V. M. Lieven and Caroline F. Rowland. 1998. Comparing different
models of the development of the English verb category. Linguistics 36:4–40.
Pinker, Stephen and Alan Prince. 1994. Regular and irregular morphology and the psycho-
logical status of rules of grammar. The reality of linguistic rules, ed. Susan D. Lima,
R. L. Corrigan and Gregory K. Iverson, 353–88. Amsterdam: John Benjamins.
Pollard, Carl and Ivan A. Sag. 1993. Head-driven Phrase Structure Grammar. Chicago:
University of Chicago Press and Stanford: the Center for the Study of Language and
Information.
340 References
Poplack, Shana. 1992. The inherent variability of the French subjunctive. Theoretical anal-
yses in Romance linguistics, ed. Christiane Laeufer and Terrell A. Morgan, 235–63.
Amsterdam: John Benjamins.
1996. The sociolinguistic dynamics of apparent convergence. Towards a social science
of language: papers in honor of William Labov, vol. 1: variation and change in
language and society, ed. Gregory Guy, John Baugh and Deborah Schiffrin, 295–
309. Amsterdam: John Benjamins.
Prasada, Sandeep and Steven Pinker. 1993. Generalization of regular and irregular morpho-
logical patterns. Language and Cognitive Processes 8:1–56.
Prince, Ellen F. 1978. A comparison of WH-clefts and it-clefts in discourse. Language
54:883–906.
1981a. Toward a taxonomy of given-new information. Radical pragmatics, ed. Peter
Cole, 223–56. New York: Academic Press.
1981b. Topicalization, Focus-movement and Yiddish-movement: a pragmatic differentia-
tion. Proceedings of the Seventh Annual Meeting of the Berkeley Linguistics Society,
ed. Danny K. Alford et al., 249–64. Berkeley: Berkeley Linguistics Society.
Pulman, Steven G. 1983. Word meaning and belief. London: Croom Helm.
Pustejovsky, James. 1995. The generative lexicon. Cambridge, Mass.: MIT Press.
Quine, Willard van Orman. 1951 [1961]. Two dogmas of empiricism. From a logical point
of view (2nd ed.), 20–46. New York: Harper. (Originally published in Philosophical
Review 60:20–43, 1951.)
Radden, Günter. 2000. How metonymic are metaphors? Metaphor and metonymy at the
crossroads: a cognitive perspective, ed. Antonio Barcelona, 93–108. Berlin: Mouton
de Gruyter.
Reddy, Michael J. 1979 [1993]. The conduit metaphor – a case of frame conflict in our
language about language. Metaphor and thought (2nd ed.), ed. Andrew Ortony,
Reinhart, Tanya. 1984. Principles of gestalt perception in the temporal organization of
narrative texts. Linguistics 22:779–809.
Richards, I. A. 1936. The philosophy of rhetoric. London: Oxford University Press.
Ricoeur, Paul. 1978. The metaphorical process as cognition, imagination and feeling. On
metaphor, ed. Sheldon Sacks, 141–57. Chicago: University of Chicago Press.
Riemer, Nick. 2001. Remetonymizing metaphor: hypercategories in semantic extension.
Cognitive Linguistics 12:379–401.
Rosch, Eleanor H. 1973. Natural categories. Cognitive Psychology 4:328–50.
1978. Principles of categorization. Cognition and categorization, ed. Eleanor Rosch and
Barbara Lloyd, 27–48. Hillsdale, N. J.: Lawrence Erlbaum Associates.
Rosch, Eleanor H. and Carolyn B. Mervis. 1975. Family resemblances: studies in the internal
structure of categories. Cognitive Psychology 7:573–605.
Rosch, Eleanor H., Carolyn B. Mervis, Wayne Gray, David Johnson and Penny
Boyes-Braem. 1976. Basic objects in natural categories. Cognitive Psychology
8:382–439.
Ross, John R. 1967. Constraints on variables in syntax. Ph.D. dissertation, Massachusetts
Institute of Technology.
Rubino, Rejane B. and Julian M. Pine. 1998. Subject-verb agreement in Brazilian
Portuguese: what low error rates hide. Journal of Child Language 25:35–59.
Rudes, Blair A. 1980. On the nature of verbal suppletion. Linguistics 18:655–76.
References 341
Sadock, Jerrold M. and Arnold Zwicky. 1985. Speech act distinctions in syntax. Language
typology and syntactic description, vol. 1: clause structure, ed. Timothy Shopen,
Schank, Roger C. and Robert P. Abelson. 1977. Scripts, plans, goals and understanding.
Hillsdale, N. J.: Lawrence Erlbaum Associates.
Searle, John. 1979. Literal meaning. Expression and meaning, ed. John Searle, 117–36.
Cambridge: Cambridge University Press.
Seidenberg, Mark S., Michael K. Tanenhaus, James M. Leman and Marie Bienkowski. 1982.
Automatic access of meanings of ambiguous words in context: some limitations of
knowledge-based processing. Cognitive Psychology 14:489–532.
Slobin, Dan I. 1991. Learning to think for speaking: native language, cognition, and rhetor-
ical style. Pragmatics 1:7–26.
Smith, Edward E., Edward J. Shoben and Lance J. Rips. 1974. Structure and process in
semantic memory: a featural model for semantic decisions. Psychological Review
81:214–41.
Smith, Linda B. and Larissa K. Samuelson. 1997. Perceiving and remembering: category
stability, variability and development. Knowledge, concepts and categories, ed. Koen
Lamberts and David Shanks, 161–95. Hove: Psychology Press.
Sperber, Dan and Deirdre Wilson. 1986. Relevance: communication and cognition. Oxford:
Basil Blackwell.
Stassen, Leon. 1997. Intransitive predication. Oxford: Oxford University Press.
Stemberger, Joseph P. and Brian MacWhinney. 1988. Are inflected forms stored in the
lexicon? Theoretical morphology, ed. Michael Hammond and Michael Noonan,
101–16. San Diego, Calif.: Academic Press.
Stern, Joseph. 2000. Metaphor in context. Cambridge, Mass.: MIT Press.
Sweetser, Eve. 1990. From etymology to pragmatics: metaphorical and cultural aspects of
semantic structure. Cambridge: Cambridge University Press.
Swinney, David A. 1979. Lexical access during sentence comprehension: (re)con-
sideration of context effects. Journal of Verbal Learning and Verbal Behavior
18:645–69.
1982. The structure and time-course of information interaction during speech compre-
hension: lexical segmentation, access and interpretation. Perspectives on mental
representation: experimental and theoretical studies of cognitive processes and ca-
pacities, ed. Jacques Mehler, Edward C. T. Walker and Merrill Garrett, 151–67.
Hillsdale, N. J.: Lawrence Erlbaum Associates.
Sylestine, Cora, Heather K. Hardy and Timothy Montler. 1993. Dictionary of the Alabama
language. Austin: University of Texas Press.
Talmy, Leonard. 1972. Semantic structures in English and Atsugewi. Ph.D. dissertation,
Department of Linguistics, University of California, Berkeley.
1976. Semantic causative types. The grammar of causative constructions (Syntax and
Semantics, 6), ed. Masayoshi Shibatani, 43–116. New York: Academic Press.
1977. Rubber sheet cognition in language. Papers from the Thirteenth Regional Meeting,
Chicago Linguistic Society, ed. Woodford A. Beach et al., 612–28. Chicago: Chicago
Linguistic Society.
1978. The relation of grammar to cognition: a synopsis. Proceedings of TINLAP-2:
theoretical issues in natural language processing, ed. David Waltz, 14–24. Urbana:
University of Illinois Coordinated Science Laboratory.
342 References
1983. How language structures space. Spatial orientation: theory, research and appli-
cation, ed. Herbert L. Pick, Jr. and Linda P. Acredolo, 225–82. New York: Plenum
Press.
1988a. The relation of grammar to cognition. Topics in cognitive linguistics, ed. Brygida
Rudzka-Ostyn, 165–205. Amsterdam: John Benjamins.
1988b. Force dynamics in language and cognition. Cognitive Science 12.49–100.
2000. Toward a cognitive semantics, vol. 1: concept structuring systems. Cambridge,
Mass.: MIT Press.
Taylor, John R. 1989 [1997]. Linguistic categorization: prototypes in linguistic theory (2nd
ed.). Oxford: Oxford University Press.
2002. Cognitive grammar. Oxford: Oxford University Press.
Togia, Pagona. 1996. Antonyms in English and Modern Greek: a cognitive approach. Un-
published Ph.D. thesis, University of Manchester.
Tomasello, Michael. 1992. First verbs: a case study of early grammatical development.
Cambridge: Cambridge University Press.
2000. Do young children have adult syntactic competence? Cognition 74:209–53.
2003. Constructing a language: a usage-based theory of language acquisition.
Cambridge, Mass.: Harvard University Press.
Tomasello, Michael, Nameera Akhtar, Kelly Dodson and Laura Rekau. 1997. Differential
productivity in young children’s use of nouns and verbs. Journal of Child Language
24:373–87.
Tsukiashi, Ayumi. 1997. A usage-based analysis of the Japanese passive construction. M.A.
dissertation, University of Manchester.
Tuggy, David. 1993. Ambiguity, polysemy and vagueness. Cognitive Linguistics 4:273–90.
Turner, Mark. 1987. Death is the mother of beauty: mind, metaphor, criticism. Chicago:
1990. Aspects of the invariance hypothesis. Cognitive Linguistics 1:247–55.
2001. Cognitive dimensions of social science. Oxford: Oxford University Press.
Ungerer, Friederich and Hans-Jürgen Schmid. 1996. An introduction to cognitive linguistics.
London: Longman.
Vallduvı́, Enric. 1992. The informational component. New York: Garland.
Wertheimer, Max. 1923 [1950]. Laws of organization in perceptual forms (abridged and
translated by Willis D. Ellis). A source book of Gestalt psychology, ed. Willis D.
Ellis, 71–88. New York: Humanities Press. (Untersuchungen zur Lehre von der
Gestalt. Psychologische Forschung 4:301–50.)
Whittlesea, Bruce W. A. 1997. The representation of general and particular knowl-
edge. Knowledge, concepts and categories, ed. Koen Lamberts and David Shanks,
335–70. Hove: Psychology Press.
Wierzbicka, Anna. 1980. Lingua mentalis: the semantics of natural language. New York:
Academic Press.
1985. Oats and wheat: the fallacy of arbitrariness. Iconicity in syntax, ed. John Haiman,
311–42. Amsterdam: John Benjamins.
1987. Boys will be boys. Language 63:95–114.
1988. The semantics of grammar. Amsterdam: John Benjamins.
1996. Semantics: primes and universals. Oxford: Oxford University Press.
Wilson, Deirdre. 1975. Presuppositions and non-truth-conditional semantics. New York:
Academic Press.
References 343
Wittgenstein, Ludwig. 1953. Philosophical investigations. New York: Macmillan.

1980. Culture and value, trans. P. Winch. Chicago: University of Chicago Press.
Wood, Mary McGee. 1993. Categorial grammars. London: Routledge.
Zager, David. 1980. A real-time process model of morphological change. Ph.D. dissertation,
State University of New York at Buffalo, New York.
Author index
Aaron, David 201, 203 Clausner, Timothy C. 22–23, 44–45, 68, 198–99,
Abelson, Robert P. 7, 17 318–19
Akhtar, Nameera 323 Coleman, Linda 31–32
Akmajian, Adrian 245 Collins, Allan M. 76
Allan, Keith 176 Cornwell, Patricia 193, 210, 213, 218, 219, 221
Andersen, Henning 306 Coseriu, Eugenio 141
Ariel, Mira 51 Coulson, Seana 38, 207–9, 210
Armstrong, Sharon Lee 88 Croft, William 4, 22–23, 26, 44–45, 57–58, 59,
Aronoff, Mark 226 64, 66, 68, 70–71, 73, 97, 102, 198–99, 242,
Atkins, Beryl T. 11–12 257, 274, 278, 279, 283, 285–87, 289, 294,
Atran, Scott 86 319, 321–23
Cruse, D. Alan 4, 79, 80, 83, 84, 109, 116, 117,
Baldwin, Gillian 323 120, 130, 136, 141, 142, 143, 147–50, 152,
Barlow, Michael 291 154, 156, 158, 159, 160–61, 165, 173–74,
Barsalou, Lawrence W. 87, 92, 96, 278 182, 187
Bates, Elizabeth A. 329
Becker, Joseph 231 Daugherty, K. 295, 300
Berlin, Brent 81, 82, 83 Deane, Paul D. 109, 315
Bienkowski, Marie 303 DeLancey, Scott 62
Birner, Betty J. 242–43 Denison, David 310, 320
Black, Max 203, 206 Dodson, Kelly 323
Bolinger, Dwight 85, 277 Dreyfus, Hubert L. 59
Boyes-Braem, Penny 82, 84
Braine, Martin D. S. 323 Eco, Umberto 207
Breedlove, Dennis E. 82, 83 Elman, Jeffrey L. 300, 307, 329
Bresnan, Joan 229
Brewer, Mary Alexandra 305 Fauconnier, Gilles 33–39, 71, 123, 207, 328
Brinkmann, U. 294, 299–300 Feyaerts, Kurt 216
Brown, Cecil H. 86, 87, 162 Fillmore, Charles J. 7, 8–14, 15, 16, 17–20, 28,
Brown, Roger 77 30, 60, 76, 95, 231–40, 248–49, 255, 257,
Bybee, Joan L. 291, 292–94, 296–306, 310–12, 259, 263, 266–71, 276, 279
313 Flickinger, Daniel 275
Fodor, Jerry A. 76
Carling, Christine 97
Chafe, Wallace 46, 50 Gadamer, Hans-Georg 21
Chater, Nick 76 Gathercole, Virginia C. Mueller 324
Chomsky, Noam 226, 227–29, 318 Gazdar, Gerald 229
Clahsen, Harald 299 Geckeler, Horst 141
Clark, Eve V. 47 Geeraerts, Dirk 109
Clark, Herbert H. 13, 18, 47, 60, 102–3 Geertz, Clifford 21
344
Author index 345
Gentner, Dedre 195 Kemmer, Suzanne 291

Gibbs, Raymond W. 199, 251, 329 Kimenyi, Alexandre 317
Gleitman, Henry 88 Kleiber, Georges 123
Gleitman, Lila R. 88 Klein, Ewan 229
Gluck, Martin A. 96 Koffka, Kurt 56, 63
Glucksberg, Samuel 195, 201, 211, 212 Köpcke, Klaus-Michael 300
Goldberg, Adele E. 248, 249, 253, 257, 266, Kosslyn, Stephen M. 96
270, 271–78, 283, 288, 312 Kövecses, Zoltán 205, 216
Goossens, Louis 218 Kuno, Susumu 61
Grady, Joseph E. 198, 200, 204, 207–9, 210
Gray, Wayne 82, 84 Labov, William 87
Gundel, Jeannette K. 51 Lakoff, George 3, 15, 16, 24, 28, 31, 32,
44–45, 55, 63, 72, 79, 80–81, 89, 95, 97, 142,
Haegeman, Liliane 230 151, 194–204, 205, 209–10, 218, 240–41,
Hahn, Ulrike 76 257, 266, 271–76, 283, 288,
Haiman, John 30 328
Hampton, James 81–82, 88, 89, 91, 144 Lambrecht, Knud 61, 241, 245, 253
Hardy, Heather K. 20 Langacker, Ronald W. 7, 14–15, 16, 18, 22,
Hare, Mary 300 23–24, 25–26, 29, 40, 43–44, 48–50, 51, 52,
Haspelmath, Martin 302 53, 54–55, 58, 59, 60, 62–63, 64, 67–68, 69,
Haviland, Susan 13 72–73, 89, 131, 133, 149, 157, 249, 253, 256,
Hawkinson, Annie 317 257, 258, 259, 261, 262, 278–83, 291, 292,
Hayes, Victoria 183, 190 329
Hedberg, Nancy 51 Larsen, Thomas W. 316
Heidegger, Martin 58–59 Lassaline, Mary E. 82, 84, 86
Heider, Eleanor 81 Leman, James M. 303
see also Rosch, Eleanor H. Levinson, Stephen C. 59
Herskovits, Annette 64–65, 69, 70 Lewis, David 102
Hopper, Paul 291 Lieven, Elena V. M. 323, 324
Hudson, Richard 261 Lobben, M. 302
Hull, David L. 231 Losiewicz, B. L. 295
Husserl, Edmund 54, 63 Lyons, John 88, 141, 143, 158
Hyman, Larry 317
MacWhinney, Brian 294–95
Israel, Michael 325–26 Makkai, Adam 231
Marcus, G. 294, 299–300
Jackendoff, Ray 153, 201, 203, 245–47 Markman, Arthur B. 291
Janda, Richard D. 299, 300 Matthews, Peter H. 134, 249–50
Jespersen, Otto 293, 298 McCarthy, John 299
Johnson, David 82, 84 McCawley, James D. 37
Johnson, David E. 229 McClelland, James L. 307, 329
Johnson, Mark 24, 44–45, 55, 63, 69, 194–96, McGregor, William 259
198–204, 328 Medin, Douglas L. 17
Johnson, Mark H. 329 Mervis, Carolyn B. 77, 82, 84
Jolicoeur, Pierre S. 96 Meyer-Lübke, W. 305
Michaelis, Laura A. 241, 253, 257, 266
Kaburaki, Etsuko 61 Moder, Carol Lynn 298, 302
Kant, Immanuel 54 Mondloch, James L. 254, 312, 316
Karmiloff-Smith, Annette 329 Montler, Timothy 20
Katz, Jerrold J. 76 Moore, Terence 97
Kay, Paul 31–32, 81, 231–40, 248–49, 255, 257, Mossé, Fernand 310
259, 261, 263, 266–71, 276, 279, 291 Murphy, Gregory L. 17, 82, 84, 86
346 Author index
Neisser, Joseph Ulric 328 Scheibman, Joanne 311

Norman, William M. 316 Schmid, Hans-Jürgen 79, 86
Nunberg, Geoffrey 48–50, 70, 230–31, 232, 233, Searle, John 29
249–54 Sebastián, Eugenia 324
Núñez, Rafael 194, 328 Seidenberg, Mark S. 295, 303
Shoben, Edward J. 88, 89
O’Brien, J. 199 Slobin, Dan I. 73, 293–94, 296–99, 313
O’Connor, Mary Kay 231–40, 248–49, 255, 263 Smith, Edward E. 88, 89
Oakley, Todd 207–9, 210 Smith, Linda B. 92–93
Omar, M. 300 Soto, Pilar 324
Sperber, Dan 100
Pardo, Elly 306 Stassen, Leon 319
Parisi, Domenico 329 Steen, Gerard J. 199
Perlmutter, David M. 229 Stemberger, Joseph P. 294–95
Pine, Julian 323, 324 Stern, Joseph 209–10, 212
Pinker, Stephen 294, 295, 299–300 Sweetser, Eve 33, 67, 97
Plunkett, Kim 329 Swinney, David A. 303
Pollard, Carl 259, 275, 279 Sylestine, Cora 20
Poplack, Shana 311
Postal, Paul M. 229 Talmy, Leonard 34, 44, 51, 53, 56–57, 58, 59,
Prasada, Sandeep 295 63, 66, 67 69–70
Prince, Alan 294, 299–300 Tanenhaus, Michael K. 303
Prince, Ellen F. 13, 238, 242 Taylor, John R. 3, 15–16, 79
Pullum, Geoffrey 229 Thompson, Sandra A. 310–12
Pulman, Steven G. 79 Togia, Pagona 173–74, 180
Pustejovsky, James 137 Tomasello, Michael 323, 329
Tsukiashi, Ayumi 316–17
Quillian, M. Ross 76 Tuggy, David 131
Quine, Willard van Orman 30 Turner, Mark 38–39, 44–45, 63, 194–95, 202–3,
207, 328
Radden, Günter 216, 217, 220
Raven, Peter H. 82, 83 Ungerer, Friederich 79, 86
Reddy, Michael J. 200
Reinhart, Tanya 59 Vallduvı́, Enric 226
Rekau, Laura 323 Vamling, Karina 20
Richards, I. A. 209
Ricoeur, Paul 203 Ward, Gregory 242–43
Riemer, Nick 216, 220 Wasow, Thomas 230–31, 232, 233, 249–54, 275
Rips, Lance J. 88, 89 Wertheimer, Max 63
Rosch, Eleanor H. 77, 82, 84 Whittlesea, Bruce W. A. 92
Ross, John R. 315 Wierzbicka, Anna 147, 243–45
Rothweiler, M. 299 Wiese, R. 294, 299–300
Rowland, Caroline F. 323, 324 Wilson, Deirdre 13, 100
Rubino, Rejane B. 324 Wittgenstein, Ludwig 77, 98, 210
Rudes, Blair A. 305 Woest, A. 294, 299–300
Wood, Esther J. 45
Sadock, Jerrold M. 319 Wood, Mary McGee 229
Sag, Ivan A. 229, 230–31, 232, 233, 249–54,
257, 259, 266, 279 Zacharski, Ron 51
Samuelson, Larissa K. 92–93 Zager, David 301
Schank, Roger C. 7, 17 Zwicky, Arnold 319
Subject index
Bold page numbers indicate definitions or major discussions. Category values (‘future’) are found
under category entries (‘tense’).
ablaut 254 antonymy 104, 141, 162, 165–66, 167, 168,

abstraction 4, 44, 52 169–92
see also scalar adjustment, qualitative applicative constructions 317
access node 30 Arabic 299–300
Access Principle 35 Arc-Pair Grammar 229
accessibility 46, 50 argument (grammatical) 49, 53, 70, 137, 195,
acquisition 1, 84, 291, 292, 294, 300, 307, 308, 264, 280–82, 310, 314, 320–21
320–21, 323–25, 327, 329 argument structure constructions 66, 263, 264,
activation 46, 292, 295, 304, 307–8, 327, 329 269–70, 273, 312, 320–21
see also interactive activation model argument reversal constructions 242
active zones 48–49, 137, 138, 140, 156 argumentation, syntactic 283
see also metonymy articles 11
adjective 53, 67–68, 118, 121, 142, 167–92, 253, definite 13–14, 258
299 indefinite 243
adjunct 228, 281 artifacts 17, 152
adposition 286 artificial intelligence 8, 17, 28
adverb 71, 242 aspect 64, 71, 254, 280, 304–5, 317
adverbial clause 228 atelic 243, 246
affectedness 66 habitual/generic 71, 304
agreement 11, 254, 261, 286, 287–88, 304–5, progressive 41, 52, 64, 228, 268
324 punctual 243
Alabama 20 simple (nonprogressive) 41, 52, 241, 276
ambiguity 118, 138, 140 telic 246
analogical change 304, 305–6, 320 assertion 61
antagonism 112–13, 115, 116, 121 atemporal relations 253, 280
anthropological linguistics 86 attachment 156–58
antonyms 3, 165–66, 172–92 attention 3, 43, 46–54, 62, 68, 100, 101, 112
committed 176, 177–81, 182, 183, 184, 187 autonomous (grammatical structure) 282
(im)partial 175–76, 177–81, 182, 183, 184, autonomy of cognitive faculty 1, 2, 3, 328
187, 190 autonomy of senses 109, 112–14, 116–20,
polar 172–81, 190, 191 121–22, 126, 127, 128, 131, 138, 140
see also monoscalar antonym attentional autonomy 112–13, 119–20
see also biscalar antonym/system; construal, see also antagonism
absolute; construal, committed; compositional autonomy 114, 118, 138
construal, hybrid; inherentness (in relational autonomy 113–14, 117, 126, 128,
antonyms); monoscalar antonym/system; 137
sub; supra truth-conditional autonomy 128–29, 133, 134
347
348 Subject index
auxiliary 228, 238, 264, 265, 268, 271, 276, 280, cognitive linguistics 1, 40, 42, 45, 105, 225, 291,
286, 310, 311, 315, 317, 324 325, 327, 328–29
cognitive psychology 3, 7, 17, 28, 30, 45, 46, 54,
background 59 75, 86, 328, 329
background assumptions 29–30 collocations 12, 18, 249–50, 252
Bambara 289 common ground 60–61, 102–3
Bantu 317 communication 19, 74, 99, 103, 193, 291, 326,
base 7, 15–16, 19, 25, 132 329
based-on link 275 comparative 177–79, 183, 186, 228
Being-in-the-world 58–59 quantified comparative 178–79, 183
binarity 164–65, 166 comparison 44, 54–58, 68
biological kinds 17, 86 complement 267, 281, 325
biscalar antonym/system 170, 172, 174, 181–85 complement constructions 280, 316
disjunct 170, 172, 182 complementaries 165–66, 167–69, 185,
equipollent 170, 172, 174, 181–83, 188, 191 188
overlapping 170, 172, 183–85, 188, 190 complementizer 280
parallel 170, 182–83 componential model (of a grammar) 225–29,
blending 38–39, 193, 213–15, 216, 221 231, 232, 237, 245, 247, 248, 255,
see also space, mental 258–59, 263
Blending Theory (BT) 38–39, 203, 207–9, 210, component, semantic (of a construction)
328 138–40, 260, 286
‘Blessings-Wishes-Curses’ construction 271 compositionality, semantic 105, 120, 177, 179,
borrowing 296, 299, 300, 312 249–54
boundary (of category/sense) 75, 76, 89–91, compound 31
93–95, 97, 102, 104, 105, 109–15, 122, comprehension 99, 100, 278, 285, 307
143, 146, 151, 153, 155–56, 159, 168, concepts 7, 14–15, 24–27, 30, 37, 47, 48, 88,
322 92–93
see also fuzziness (of boundaries) generic concepts 74–75
bounded/unbounded 64, 70, 71 individual concepts 74–75
bridging 13 Conceptual Metaphor Theory (CMT) 194–204,
207, 209
calibration 171–72, 178, 179, 180, 184, 191–92 conceptual space 109, 288, 321–23, 327
case (marking) 261, 286, 319 conceptual structure 2, 3, 15, 30, 34, 39, 46, 197,
Categorial Grammar 229 328
categorical 61 conceptualization see construal
categorization/category 3, 17, 46, 53, 54–55, conditional 228
74–106, 282–83, 285, 326 conjunction 237–38
ad hoc category 92 connection 303–4
see also boundary (of category); classical see also phonological connection; semantic
model; dynamic construal model; frame, connection; symbolic connection
semantic; levels (categories); prototype consciousness 46, 75
model constituency (syntactic) 261, 311
Caused Motion construction 312 constitution see Gestalt
change see language change constraints (conceptual) 100–1, 101–3, 109
circumstantial phrase 228 constraints (syntactic) 225, 315, 318
classical model (categorization) 76–77 construal 1, 19, 28, 40–69, 75, 79, 80, 93–98,
see also categorization 103–4, 109–10, 122, 127, 138, 140, 144,
clause 273 145, 150, 151, 153, 155, 158, 160, 161,
coercion 43 164–65, 167–68, 169, 182, 185–92, 216,
cognitive abilities/capacities 2, 3, 45 218, 221, 279–80, 328, 329
Cognitive Grammar 72, 257, 266, 278–83, 285, absolute construal (antonyms) 177, 178,
287, 288 179–81, 185–89
Subject index 349
default construal 71–72, 83, 102, 103–4, 126, declarative sentence construction 264, 265, 314,
140, 144, 146–47, 158, 159, 166, 168, 319–21
185, 189, 190, 191, 216 default construal see construal, default
hybrid construal (antonyms) 186–87 default schema see schema, default
relative construal (antonyms) 175, 177, default specificity 127, 129–33, 134, 135, 158
179–81, 184, 185–89 degree of membership (DOM) 79–81
construal operations 40–73 deictic center 60
interaction of 69–70 deixis 10–11, 44, 46, 58, 59–62, 63
construction grammar 4, 76, 225, 227, 229, 231, epistemic deixis 46, 60–61, 63
240, 245, 247–48, 252, 255–56, 257–90, demonstrative 51, 235
302, 313, 317, 326, 329 dependency 261
Construction Grammar (Flllmore, Kay et al.) dependent (grammatical structure) 282
257, 266–72, 279, 280, 282, 285, 288, derivation, grammatical 40, 41–42
291 determiner 324
Construction Grammar (Lakoff, Goldberg) 257, diachrony 111, 305
266, 272–78, 283, 288 dictionary meaning 30
constructions, grammatical 4, 8, 14, 34, 41–42, dimension 25, 69
53, 73, 177, 227–29, 236–49, 251–56, direct object see object
257–90, 295, 302, 308, 313, 319, discourse function see information structure
321–27 disjunction 34
container/containment image schema 80, 89, ditransitive construction 264–65, 273–74, 312
104, 142, 151, 201 domain, semantic 15–16, 17–32, 39, 44, 47, 65,
contextual constraints/pressure 102–3, 109, 68–69, 70, 79, 131–32, 164–66, 167–69,
122–23, 127–28, 130, 131, 134, 135, 172, 194–216, 221
136, 150, 159, 164, 165, 182, 193, 204, abstract domain 24
221 basic domain 24, 25, 26
contextual modulation 128, 129, 130, 135, image-schematic domains 68–69
140 source domain 55, 195–204, 207, 210, 215,
contextual pressure see contextual constraints 221
contiguity (syntactic) 286 target domain 55, 195–204, 207, 210, 221
controller (of agreement) 286 domain, social
convention, conventionality 31, 43, 72–73, 156, domain matrix 25, 27, 31, 47, 69, 122, 132, 216
195–98, 199, 203–4, 230, 231, 249–52, domain structure 26
258, 279, 280 dominion 46, 51–52
conventional constraints 102, 103–4, 109, dual-processing model 294, 299
111–12, 114, 117, 131, 135, 139, 144, durative adverbial 246
159, 161, 164, 166, 193, 209, 216 dynamic construal model (categorization) 4, 75,
conventional imagery 72 92–104, 141
conventional universalist position 73 see also categorization
conversational implicature see implicature
conversion 43 economy 74
conversive 166 elaboration 281–82
coordinate construction 240 elaboration site (e-site) 281
copula 253, 319, 321–22 element (syntactic) 260, 264, 285, 286, 287
Cora 280 embodiment 44
core (intensional) 150 empathy 46, 61, 62, 63
correspondence see metaphorical emphatic negative imperative construction 271
correspondence; symbolic encyclopedic knowledge/meaning 30, 86, 148,
correspondence 196, 204, 208
countability (count/mass) 71, 280 English (Modern) 14,21, 24, 41, 42, 64, 72,
see also noun 90–91, 110, 178, 181, 228, 235, 253,
counterfactual 36, 38, 39 280, 293, 310, 314, 316, 320–22
350 Subject index
English (Modern) (cont.) type frequency 296–300, 301, 307, 308, 309,
Middle English 310 312, 327
Old English 235, 294, 300 full-entry model 276–78
entailment 13, 104, 143, 145–46 functionalism (linguistics) 242, 329
entity/interconnection 44, 67–68 fuzziness (of boundaries) 77, 91, 94, 95
entrenchment 111–12, 131–33, 135, 136, 139,
292–95, 296, 297, 304, 305, 308–11, 327 games 33
epistemic correspondences 196–97, 201 gang effects 294–95
epistemic deixis see deixis gender (grammatical) 300
equality/inequality constructions 179–80 Generalized Phrase Structure Grammar 229
ergativity 280, 316 generative grammar 1, 2, 225–29, 259–60, 261,
split ergativity 62 263, 292, 293, 294, 302, 313, 317, 328,
e-site see elaboration site 329
evaluative terms 18–19 geometric structure 63, 64–65, 70
event 57–58, 70 German 10, 20, 21, 53, 183–84, 299, 300
evolution of humankind 328 Gestalt 46, 63–69, 75, 100, 101, 115, 116, 175,
exclamative constructions 241, 247, 258 286
experience 19, 24, 28, 44, 45, 54, 63, 68, 69, Gestalt psychology 56, 63, 329
71–73, 74, 101, 172, 195, 201, 203–4, goodness of exemplar (GOE) 77–79, 80–81, 92,
326 153, 166
expert systems 86 Government and Binding theory 229
explicature 100 gradability 71
exposure 177–81 graded centrality 3, 32, 75, 77–81, 88, 91
extraction constructions 237, 315–16, 317 grammar/grammatical knowledge 1, 3, 12,
106, 225–27, 229, 231, 247, 254,
facets 47–48, 101, 116–26, 131, 132, 137, 138, 255–56, 257, 263–65, 271, 278, 279,
140, 216, 220 287, 288, 291, 292, 296, 306, 326–27,
familiarity 78 328, 329
family resemblance 78, 82, 85 see also Arc-Pair Grammar; Categorial
feature Grammar; componential model (of a
grammatical 266–70, 284, 285 grammar); construction grammar;
semantic 7, 8–10, 76, 78, 87, 88, 91, 100, 148, Construction Grammar (Fillmore, Kay et
150 al.), Construction Grammar (Lakoff,
feature structure 266–67, 269, 271 Goldberg); generative grammar;
fictional situations 33 Generalized Phrase Structure Grammar;
fictive motion 46, 53, 54 Government and Binding theory;
figurative meaning 193, 230, 251 Head-driven Phrase Structure Grammar;
figure/ground 44, 46, 56–58, 59, 62, 71, 101, 281 Lexical-Functional Grammar;
filler 268 Minimalist theory; organization,
focal adjustments 43–44, 47, 58, 59 grammatical; Radical Construction
focal orientation 149–50 Grammar; Relational Grammar;
focus antipassive 316 representation, grammatical; rules;
focus constructions 238, 240, 247, 315–16 schemas; Semiotic Grammar;
folk classification 86 usage-based model; Word Grammar
force dynamics 43, 46, 66–67, 69 grammaticalization 63, 317–18
foreground 59 granularity 52
frame, semantic 8–22, 34, 37, 39, 46, 47, 53, 55, Greek, Modern 162, 172, 178, 180–81, 182
87, 91–92, 95–96, 167, 272 ground see figure/ground
French 20, 72, 90–91, 136, 178, 180–81, 311
frequency 78, 80, 133, 292, 304, 305, 309 have a X constructions 243–44, 247, 248
token frequency 292–95, 301, 306, 307, 308, head 267, 268, 270, 282, 282
309–10, 324–25 hearer 100, 286, 318, 329
Subject index 351
Head-driven Phrase Structure Grammar 259, inflections, grammatical 40, 41–42, 293–307,
266, 279 324–25
Hebrew, Modern 72 information structure 61, 226, 235, 242
historical linguistics 1 inherentness (in antonyms) 184–85, 189
see also change, language; diachrony inheritance 76, 270–72, 273, 274, 275–78
holonym 160 complete inheritance 270–71, 278, 308
homonymy 100, 111, 217, 303 default inheritance see inheritance, normal
homophony 303 multiple inheritance 264, 276–77
hyperonym 84, 114, 117, 120, 121, 127, 128, 129, normal inheritance 275–76
130, 131–32, 133, 134, 135, 148, 149, see also full-entry model
158 innate capacity for language 2–3
hyponym 14, 117, 120, 121, 127, 128, 132, 143, instance (of a construction) 259
144, 146, 147, 148, 152, 175 instrument 316
hyponymy 3, 7, 104, 141–50, 159, 162 integration (of components of a meaning) 125–26
intention 88
iconicity 175, 286 interactive activation network 307
ideal 80 interconnection see entity/interconnection
Idealized Cognitive Model (ICM) 28–32, 92, 95 interface (grammatical) 228, 229
cluster ICM 31, 92, 95 interpretation 98–100, 101, 109–10
identity 36 interrogative constructions 265, 310, 314, 319
idiom 199, 205, 225, 2300–37, 270 see also questions
decoding idiom 231, 232, 235 intransitive construction 234, 259, 260–62, 264,
encoding idiom 231–32, 235, 250 268, 283–85, 287
extragrammatical idiom 233, 235, 236 Invariance Hypothesis 201–2
formal idiom 233, 235 inversion constructions 242–43
see also idiom, schematic irregularity see regularity
grammatical idiom 232, 235 It-cleft construction 237, 242, 265, 315
schematic idiom 234, 235, 236–37, 241, 243, iteration 70
244, 247, 262–63
substantive idiom 233, 234, 235, 236, 237, Japanese 316–17
247, 253, 262–63 Javanese 21
see also pragmatic point judgement see comparison
idiomatic phrase 232, 235, 252, 253
idiomatically combining expression 232–33, K’iche’ Mayan 254, 312, 316, 317
235, 236, 249–53, 263 Kinyarwanda 317
illocutionary force 318–20 knowledge of language 1, 2, 3–4, 225, 328, 329
image 44
image nouns 33 landmark 58
see also picture nouns language 2, 71–73, 99, 328–29
image schemas 44–46, 62, 64–65, 68, 80, 104, language change 291, 292, 293–94, 298, 307,
167, 169, 172, 201–4 325–26
imaging systems 43 language use 1, 2, 3–4, 278, 307, 326, 328, 329
immediate scope, see scope of predication latency 134–37
imperative constructions 319–22 learning 74, 78, 325
implication 8 see also acquisition
see also entailment left isolation (LI) construction 271
implicature 10, 185, 268 let alone construction 237–40, 247, 248
incompatibility 104, 117, 126–27, 133, 141, 145, levels (categories) 75, 82–87, 96–97, 130
147, 152, 162 basic 82, 83–84, 96, 97, 130, 148, 175
individuation 64, 70 generic see level, basic
induction 4, 323–25, 327 subordinate 82, 84, 85–86, 96, 130
infinitive 247 superordinate 82, 84–85, 96, 97, 130, 175
352 Subject index
levels (syntax) 225–26 metaphtonymy 218–19

lexical decision task 78–79, 303 metonymy 46, 48–49, 70, 126, 193–94, 216–20
see also priming see also active zone
lexical field theory 10–11 microsenses 101, 116, 126–37, 138, 140, 158,
Lexical-Functional Grammar 229 161
lexical rules 246–47 microsense complexes 129, 131–33, 136
lexical semantics see semantics, lexical mind 2, 24, 72, 291, 292
lexicon 97, 205, 212, 226, 227, 229, 230, 234, Minimalist theory 229
237, 247, 255, 257, 263, 293, 294 modal 67, 228
linking rules 227, 231, 258 modifier 53, 54, 268, 282
links (between constructions) 273–75 modulation, contextual see contextual
list 292, 294 modulation
literary analysis 194, 328 module 1, 226
locative 50–51, 280 see also autonomy
monoscalar antonym/system 170, 171–81
Maasai 289 see also antonym, polar
mapping (between domains) 194, 198–99, 201–4 mood 304–5, 306–7, 317
open mapping 213–15 morphology 1, 4, 226, 254–55, 292–307, 313,
restricted mapping 213–15 318, 319, 321, 327
see also epistemic correspondences; motor activity 2
metaphorical entailments; ontological movement rule 313–15
correspondences
mathematics 194, 328 negation 13, 33, 36, 139, 239, 265, 269, 311
maximal scope see domain structure negative constructions 264, 310, 311, 314,
meaning 30, 97–98 321–23
word meaning 4 negative polarity item 239, 240
memory 3, 101 network 30
mental spaces see space, mental neural network 46
meronomic structure (of a construction) 259, node 262, 263, 292
260, 262, 269–70, 273, 283, 288 Nominal Extraposition 241
meronomy 3 nonreductionist model 272, 283, 287, 288
meronym 114–26 nonrelational see relationality
meronymy 7, 104, 141, 150–51, 159–63 nonsubject WH-question construction 271
metalinguistic 133–34, 164 nonverbal predication 319–22
metaphor 3, 39, 44, 46, 55–56, 70, 83, 119, 133, noun 12, 41, 42, 53–54, 67–68, 137, 141, 243,
193–221 254, 279, 284, 299, 323–24
conventional metaphors 194–204 count noun 24, 41, 42, 43, 64, 85
deviance in metaphors 206–7 mass noun 24, 41, 42, 43, 48, 64, 85
image metaphors 195, 203 noun phrase 233, 242, 243, 273, 324
novel metaphors 204–11 null instantiation 279
see also blending; Blending Theory; number 206–7
Conceptual Metaphor Theory; domain, plural 48, 254, 292–93, 294, 296, 299–300,
semantic; epistemological 302, 324
correspondences; Invariance Hypothesis; singular 48, 254, 292, 324
mapping; ontological correspondences; numeral 178
substitution theory of metaphor; target;
target domain override; vehicle object (grammatical) 42, 62, 66, 134, 228, 270,
metaphor within simile 215 273, 279, 280, 284, 316, 325
see also simile within metaphor objectivity see subjectivity/objectivity
metaphorical correspondence 193, 213, 218 oblique 42, 66, 228, 286, 289, 294, 316
metaphorical entailments 197, 201 odd number paradox 88
Subject index 353
ontological correspondences 196–97 portion 151, 152

ontological type 122, 126, 133, 153 Portuguese, Brazilian 324
open schema see schema, open possessive constructions 23–24, 51, 62, 73, 280
oppositeness 164–67, 170, 185–87 possible worlds 33
organization, grammatical 257 postposing constructions 242–43
orientation 46, 59, 63, 65 pragmatic point (of idiom) 234
pragmatics 12–13, 34, 50, 61, 127, 132, 136,
paired focus construction see focus constructions 226, 239–40, 244–45, 255
paradigm (grammatical) 304–5, 321–22 pre-meaning 103, 104, 105, 110, 112, 114, 117,
paragon 80 138, 140, 143, 159, 162
parallel distributing processing 329 predicate 259, 262, 264, 269–70, 280–82
parsimony 277–78 predicate adjective construction 253
computing parsimony 278 predicate transfer 49–50, 219
storage parsimony 278 predicate type 318–21
part 83, 120, 151–59, 160–63 prefix 254
core part 156 preposing constructions 242–43
extrinsic construal of part 160 preposition 41, 52, 286
integral part 156–57 prepositional phrase 228, 242
intrinsic construal of part 160 presentational sentence 41, 61
segmental part 154, 162 presupposition 13, 24, 27, 37, 61, 139, 268, 281
spare part 152, 160 presupposition float 38
systemic part 154, 162 priming 78–79, 93, 303
ultimate part 155 process (cognitive) 291
partial specification 264 process (semantic) 53, 54, 253
partiality production 2, 278, 294, 307
scale-partiality 181, 183, 184 productivity 199, 296–300, 302, 308–9, 311–12,
system-partiality 181, 183, 184 314
see also antonym, committed; antonym, profile 15–16, 17–23, 46, 47–50, 53, 92, 132,
impartial 149, 150, 196, 212–13
participant type 318–21 configurational profile 22–23
partonymy see meronymy locational profile 22–23
parts of speech 40, 41, 280 profile shift 47–50
passive construction see voice, passive profile determinant 282, 287
perception 2, 3, 88 profile equivalence 287
person (grammatical) 11, 254, 293, 304–7, 324 prohibitive constructions 320–22
perspective 30, 43, 44, 46, 58–63, 68 pronoun 36, 51, 59–60, 62, 233, 316
phenomenology 45, 58, 63 proper name 299
philosophy 28, 54, 194, 328 prototype 3, 77–92, 120, 152, 216, 272, 273–74,
phonological connection 303, 304, 318 283, 285, 302
phonological pole 279, 286, 319 prototype effects 32
phonology 1, 4, 225, 226–27, 242, 255, 294, prototype model 77–92
296–97, 298, 301–2, 319 see also categorization
phrase structure rule 263 Provençal 305–6
‘picture’ nouns 33, 62 psycholinguistics 251, 298–99, 302, 307
see also image nouns psychology see cognitive psychology
piece 151–53 purport 1–2, 100–1, 104, 109–10, 127, 147, 161,
planning 74 162, 193, 221
pluralia tanta 64
polarity question construction 271 qualia roles 137
politics 194, 328 see also ways-of-seeing
polysemy 3, 19, 97, 109–40, 217, 273–75, 280 quantification 34
354 Subject index
quantifier 48 scale 22–23, 46, 65, 69, 104, 166, 167, 169–92,
quasimodals 317–18 201
questions 311, 314, 315–16 scale-committedness see partiality,
How X is it? 173–74, 176, 179, 180, 183, 184 scale-partiality
Is it X? 181–82 scale-impartiality see partiality, scale-partiality
What is its NOM? 173, 176, 179–82, 184 scale-schema 173
see also interrogative constructions absolute scale-schema 173, 175
relative scale-schema 173
radial category structure 272 see also construal, absolute; construal, hybrid;
Radical Construction Grammar 257, 266, construal, relative
283–89, 321 scanning 44, 46, 53–54, 67, 280
range congruence 160 sequential scanning 46, 53–54, 67, 72, 280
reality 101–2 summary scanning 46, 53–54, 67, 280
reductionism 1 schema 295–302, 207, 308–11, 314, 317, 327
reductionist model 268, 272, 284–85 default schema 300, 312
reference 35–36, 54 open schema 299–300
reference point 51 product-oriented schema 301–2, 307, 313–18,
referential opacity 35–36 327
referential/attributive 36 source-oriented schema 301, 313, 327
regular syntactic expressions 232, 233, 234, 236, schematic relation 24, 26–27
252 schematic systems 43
regularity/irregularity (grammatical) 292–300, schematicity 175, 198–201, 202, 253, 254, 263,
303, 311, 313, 318 265, 309, 312
relation see semantic relation; symbolic relation; schematization 4, 44, 52–53
syntactic relation scope of attention 46
Relational Grammar 229 scope of predication 23–24, 46, 154
relationality 46, 48, 58, 67–68, 195, 269, 281 script 8, 17
see also entity/interconnection search domains 50–51
relative clause 315–16 selection 44, 46, 47–50, 65
relativity (semantic/linguistic) 72–73 selectional restrictions 249
relevance 304, 318–19, 322, 327 semantic connection 303, 304, 307, 318, 321,
see also similarity 327
Relevance Theory 100 semantic interpretation rules 232–33, 237, 245,
representation, grammatical 2, 257 249, 252, 253, 254, 257
representation, knowledge 291, 326 Semantic Map Connectivity Hypothesis 322
resultative construction 246, 247, 248, 249, 277 semantic map model 283, 288, 321–22, 327
reversive 165–66, 169 semantic pole 279, 286
right-dislocation construction 242–43 semantic relation 261, 286
roles (in mental spaces) 34–36 semantics 1, 2, 4, 12–13, 34, 40, 226, 297, 301,
roles, syntactic see syntactic roles 328
rules 225, 229, 237, 246, 292, 294, 295–96, frame semantics 8–32
300–2, 313–18, 327 lexical semantics 3
Russian 42–43, 280 semantics of understanding 4, 8, 13, 99
structural semantics 7, 76
salience 47 truth-conditional semantics 1, 2, 7, 8, 12, 33,
see also attention 38, 40–43, 64, 328, 329
sanction 55 see also lexical field theory
scalar adjustment 46, 51–53, 64, 65 semelfactive 43
qualitative 46, 52–53, 64 Semiotic Grammar 259
quantitative 46, 51–52, 64 sense unit 109
scalar model 239, 240 full sense unit 112, 115
Subject index 355
sign, linguistic 266, 279 symbolic connection 303, 304

see also symbolic unit symbolic correspondence/link 260, 279
similarity 82, 303–6, 308 symbolic relation 286
see also relevance symbolic unit 2, 258, 259, 266, 271, 279, 285
simile 211–16, 220 synonym 146
simile within metaphor 215–16 synonymy 8
see also metaphor within simile syntactic category 259–60, 268, 272, 274,
situatedness see perspective 279–80, 283–85
situational use (adpositions) syntactic relation 262, 263, 268–70, 272–73,
social interaction 329 280–82, 285–87
sociology 17 syntactic role 261–62, 268–70, 285
sounds 2 syntactic space 283, 289
space, mental 32–39, 71, 207–9 syntax 1, 2, 4, 225, 226, 229, 240, 241, 254, 255,
base space 33, 34, 35, 36 278, 279, 308, 319, 327, 328
blended space 39, 207–9 syntax-lexicon continuum 255–56, 264
generic space 39, 207–8 system-committedness see partiality,
input space 39, 207–8, 209 system-partiality
space builders 33 system-impartiality see partiality,
see also blending; Blending Theory system-partiality
Spanish 232, 235, 324–25
spatial relations 56, 58 target 193
specificity, default see default specificity see also domain, target
speculation 211 target domain overrides 201
speech act 241, 318 target naming task 303
speech act situation 10–11, 60, 276 tautological constructions 244–45, 247
speech community 17–18, 77, 103, 197, 204 taxonomic network 262, 302, 318, 327
state 53, 64 taxonomic relation 3, 24, 26–27, 262–64, 265,
state of affairs 319 270–72, 273, 275, 322
stereotypicality 80–81 taxonomy/taxonomic hierarchy 117, 128,
structural schematization 43, 46, 63–65, 69 147–50, 199–200, 264, 265, 271, 273,
structuralism 254, 292, 293, 302 302, 306, 321–23
structured inventory 262, 279 taxonym 148–49
see also taxonomic network; taxonomy taxonymy 147
sub 173, 174, 177, 179, 181, 182, 183, 188, 189, temporal expressions 33
190 tense 276, 287–88, 304–5, 317
subcategories 75, 81 future 228
subcategorization frame 230, 237, 255, 263 past 10, 283, 293–99, 304
subject 62, 66, 228, 259, 261, 262, 264, 265, perfect 228
269, 271, 273, 279, 280, 284, 286, 289, present 241, 276, 283, 304
314, 316, 319, 325 preterite 306
subject-auxiliary inversion (SAI) construction text 13–14
271 text coherence 14
subject-auxiliary inversion exclamation thematic role 269
construction 271 theory theory 18
subjectivity/objectivity 44, 46, 62–63 There-constructions 240–41, 272, 275, 276
subjunctive 311 thetic 61
subordination 57–58, 59 thought 71–73
substitution theory of metaphor 194 ‘time’-away construction 245–46, 247
suppletion 177, 303, 305, 318 token frequency see frequency, token
supra 173, 174, 177, 181, 183, 189, 190, 191 topic 316–17
Swedish 20 Topicalization construction 242–43
356 Subject index
topic-comment 61 values (grammatical) 266

topological structure 63, 64–65 vantage point 46, 59, 63, 69
‘Tough-movement’ 49 vehicle 193, 215
trajector 58 see also domain, source
transformational grammar see generative verb 12, 53, 66, 67–68, 137, 142, 169, 228, 254,
grammar 261, 264, 267, 268, 270, 273, 274, 279,
transitive construction 234, 249, 283, 284, 287, 283, 284, 286, 287, 289, 293–99, 304,
323 310, 312, 314, 323, 324, 325–26
transitivity (category) 232, 243, 316 Verb Island Hypothesis 323
transitivity (relation) 144–45 verb phrase 238, 264, 266–67
translation 19–21 verb-particle construction 277
truth-conditional semantics see semantics, Vietnamese 42
truth-conditional viewpoint 46, 59
type frequency see frequency, type voice 66
typicality 80 active 41, 61, 313, 316
typology 283, 288, 89, 305, 321, 322 passive 41, 61, 227–28, 286, 289, 313, 316–17
unbounded see bounded way construction 247, 325–26

unification of senses 113, 115, 116–22, 131, ways-of-seeing (WOS) 137–38
132 WH-cleft construction 242
Upriver Halkomelem 289 whole 151, 160
usage-based model 4, 199, 278, 283, 288, ultimate whole 155–56
291–327 word 8, 14, 34, 53, 98, 109, 251, 254–55, 267,
use see language use 268, 279, 292–307
utterances 2, 12, 33, 60, 62, 258, 265, 278, 291, Word Grammar 261
318, 326 word order 261, 264, 277, 286, 319
valence 280–82 Yuman 280

valence-changing (morphology)
Valence Principle 269 zeugma 49, 113, 118, 119–20, 121, 134, 136,
values (in mental spaces) 34–36 139, 168, 191

Cognitive Linguistics Croft Cruse (2004)

Uploaded by

Copyright:

Available Formats

Cognitive Linguistics Croft Cruse (2004)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cognitive Linguistics Croft Cruse (2004)

Uploaded by

Copyright:

Available Formats

This page intentionally left blank

william croft is Professor of Linguistics at the University of Manchester,

d. alan cruse was Senior Lecturer in Linguistics at the University of

Cambridge University Press

This publication is in copyright. Subject to statutory exception and to the provision of

First published in print format 2004

- ---- eBook (NetLibrary)

- ---- hardback

- ---- paperback

List of ﬁgures page xii

1 Introduction: what is cognitive linguistics? 1

Part I: A conceptual approach to linguistic analysis

2 Frames, domains, spaces: the organization of conceptual

3 Conceptualization and construal operations 40

4 Categories, concepts and meanings 74

Part II: Cognitive approaches to lexical semantics

5 Polysemy: the construal of sense boundaries 109

5.3.5 Contextual modulation 140

6 A dynamic construal approach to sense relations I:

7 A dynamic construal approach to sense relations II:

8.4 Metaphor and simile 211

Part III: Cognitive approaches to grammatical form

9 From idioms to construction grammar 225

10 An overview of construction grammars 257

11 The usage-based model 291

11.3.3 Relevance and the organization of construction

12 Conclusion: cognitive linguistics and beyond 328

2.1 RADIUS and CIRCLE page 15

3.1 Linguistic construal operations as instances of general

This book provides an overview of the basic principles and methods of

Cognitive linguistics is taken here to refer to the approach to the study of

Generative grammar and truth-conditional semantics are of course still vigorous

2.1 Arguments for frame semantics

What is it that words denote, or symbolize as cognitive linguists usually

(1) [MALE] [FEMALE]

(2) a. Those are live lobsters.

Moreover, one cannot define the features in terms of applicability to a semantic

(6) a. A dog was barking.

(8) My dad wasted most of the morning on the bus.

the nature of text coherence. A large class of presuppositional phenomena appear

(10) John regretted signing the letter.

(11) John didn’t regret signing the letter.

2.2 Concepts: profile-frame organization

In the preceding section, we described the frame as a coherent region

2.3 Some consequences of the profile-frame/domain

The terms frame (Fillmore), base (Langacker) and domain (Fillmore,

scale: the evaluation of STINGY-GENEROUS is the inverse of that for THRIFTY-

By distinguishing profile from frame/domain, we may give a more straightfor-

and drinking duty-free alcohol). A similar English example would be genuflect.

Basically, understanding the meaning of rasa presupposes understanding large

2.4 Extensions of the basic profile-frame/domain distinction

The distinction between profile and domain/frame is a fundamental one

2.4.1 Locational and configurational profiles

profiles. Antonymic adjectives such as TALL/SHORT (see chapter 7) profile a

2.4.2 Scope of predication

Figure 2.2 NIECE

(18) a. A body has two arms.

Nested scopes of predications can be generalized to nesting of frames/domains

2.4.3 Relationships between domains

TIME FORCE human beings