BRIGITTP NERLICH

ZAZIE rOOD
VIMAlA HERMAN
DAVID D. ClARKE
(Editors)

Polysemy

Flexible Patterns of Meaning

in Mind and language
Polysemy
Trends in Linguistics
Studies and Monographs 142

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

Mouton de Gruyter
Berlin . New York
Polysemy
Flexible Patterns of Meaning
in Mind and Language

Edited by
Brigitte Nerlich
Zazie Todd
Vimala Herman
David D. Clarke

Mouton de Gruyter
Berlin . New York 2003
Mouton de Gruyter (formerly Mouton, The Hague)
is a Division of Waiter de Gruyter GmbH & Co. KG, Berlin.

@ Printed on acid-free paper which falls within the guidelines

of the ANSI to ensure permanence and durability.

ISBN 3-11-017616-5

Bibliographic information published by Die Deutsche Bibliothek

Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie;
detailed bibliographic data is available in the Internet at <http://dnb.ddb.de>.

© Copyright 2003 by Waiter de Gruyter GmbH & Co. KG, D-10785 Berlin
All rights reserved, including those of translation into foreign languages. No part of this
book may be reproduced or transmitted in any form or by any means, electronic or mechan-
ical, including photocopy. recording or any information storage and retrieval system, with-
out permission in writing from the publisher.
Printed in Germany.
This book is dedicated to the memory of
Andreas Blank (1961-2001), our friend
and colleague, who died far too young.
Acknowledgements

This book was written with the generous support of the Leverhulme Trust.
The book has been long in the making. The editors want to express their
gratitude to all contributors for their patience and forbearance, to Birgit Sievert
for her help and encouragement, and to Simon Cave and Sue Lightfoot for
helping the book over its last formatting and indexing hurdles.
Contents

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. vii
List of contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. xi

Setting the scene

Polysemy and flexibility: introduction and overview 3
Brigitte Nerlich and David D. Clarke
Cognitive models of polysemy 31
John R. Taylor
Polysemy: past and present .................................. 49
Brigitte Nerlich

Cognitive approaches
Polysemy and conceptual blending ............................ 79
Gilles Fauconnier and Mark Turner
Reconsidering prepositional polysemy networks: the case of over 95
Andrea Tyler and Vyvyan Evans
Polysemy as flexible meaning: experiments with English get
and Finnish pitiiii .......................................... 161
Jarno Raukko
Metonymic polysemy and its place in meaning extension .......... 195
Ken-ichi Seta

Synchrony/diachrony approaches
Polysemy in derivational affixes .............................. 217
Adrienne Lehrer
The role of links and/or qualia in modifier-head constructions ...... 233
Beatrice Warren
x Colltents

Polysemy and bleaching ..................................... 253

Jean Aitchison and Diana M. Lewis
Polysemy in the lexicon and in discourse ....................... 267
Andreas Blank

Psycho linguistic approaches

Irony in conversation: salience, role, and context effects ........... 297
Rachel Giora and Inbal Gur
Young children's and adults' use of figurative language: how
important are cultural and linguistic influences? .................. 317
Ann Dowker
Emerging patterns and evolving polysemies: the acquisition
of get between four and ten years ............................. 333
Brigitte Nerlich, Zazie Todd and David D. Clarke

Computational approaches
"I don't believe in word senses" .............................. 361
Adam Kilgarriff
Senses and texts ........................................... 393
Yorick Witks

Index .................................................... 411

List of contributors

Jean Aitchison, Professor of Dr. Diana Lewis, Faculty of English

Language and Communication, Language and Literature,
University of Oxford, UK. University of Oxford, UK.
Andreas Blank, Professor of Dr. Brigitte Nerlich, Institute for the
Romance Linguistics, University Study of Genetics, Biorisks and
of Marburg, Germany. Society, University of Nottingham,
UK.
David D. Clarke, Professor of
Psychology, University of Jarno Raukko, Department of
Nottingham, UK. English, University of Helsinki,
Finland.
Or. Ann Dowker, Department of
Experimental Psychology, Ken-ichi Seto, Professor, Faculty of
University of Oxford, UK. Literature and Human Sciences,
Osaka City University, Japan.
Dr. Vyvyan Evans, Department of
Linguistics and English Language, Dr. John R. Taylor, Linguistics
University of Sussex, UK. Section, University of Otago, New
Zealand.
Gilles Fauconniel', Professor,
Department of Cognitive Science, Dr. Zazie Todd, School of Psychology,
UC San Diego. University of Leeds, UK.
Rachel Giora, Professor, Department Mark Thrner, Professor, Department
of Linguistics, Tel Aviv University, of English and Doctoral Program in
Israel. Neuroscience and Cognitive
Science, The University of
Inbal Gur, Department of Linguistics,
Maryland, USA.
Tel Aviv University, Israel.
Dr. Andrea Tyler, Department of
Dr. Vimala Herman, English Studies,
Linguistics, Georgetown University,
University of Nottingham, UK.
USA.
Dr. Adam KilgarritT, Lexicography
Beatrice Warren, Professor,
MasterClass Ltd. and Information
Department of English, University
Technology Research Institute,
of Lund, Sweden.
University of Brighton, UK.
Yorick Wilks, Professor, Department
Adrienne Lehrer, Professor of
of Computer Science, University of
Linguistics, Emerita, Department
Sheffield, UK.
of Linguistics, University of
Arizona, USA.
Setting the scene
Polysemy and flexibility: introduction and overview

Brigitte Nerlich and David D. Clarke

This chapter has three parts. In the first part we provide a brief overview of
the varied fortunes of polysemy research since it was made popular by Michel
Breal in 1897 (Breal 1924 [1897]: Chapter 15; for more details see NerIich,
this volume), we summarize what has been done so far, what has proven
problematic in this type of research, and which gaps are still left to be
filled. We also link the recent upsurge of interest into polysemy with a
renewed interest into the cognitive bases of figurative language (metaphor
and metonymy). In the second part we explore various theories of polysemy,
proposing ourselves a graded theory of polysemy as flexible meaning, which
underlies discursive polysemy, lexicalized synchronic polysemy and dia-
chronic processes of polysemization, and we contrast this view with other
monosemous views of meaning. Throughout this introduction, we combine
the analysis of central issues with references to the topics discussed in the
articles that follow. In the last part of the introduction we shall provide an
overview of the chapters that follow and indicate the ways in which they
contribute to the overall theme of the book. In all the chapters a case will be
made for a theory of polysemy as flexible meaning.

1. Polysemy: problems and solutions

Fifty years ago the linguist and semanticist Stephen Ullmann wrote that
polysemy, the fact that some words have a network of multiple but related
meanings, is "the pivot of semantic analysis" (Ullmann 1957 [1951]: 117).
He was referring to traditional synchronic and diachronic lexical semantics
as it had developed after Breal. Fifty years after Ullmann and a century after
Breal polysemy has become central to modem cognitive semantics of the
synchronic and diachronic type as well as to computational semantics and
AI (see Brugman 1997). It has become clear that the study of polysemy is of
fundamental importance for any semantic study of language and cognition.
Polysemy can therefore be regarded as providing a privileged access to the
4 Brigitte Nerlich and David D. Clarke

network of interrelated theories of language, meaning and cognition proposed

throughout the 20th and into the 21st centuries and as pointing out directions
for future research into language and cognition.
And yet, ever since Michel Bn!al introduced the term polysemie into
linguistics, polysemy has caused problems in linguistic research. Some have
argued that it actually doesn't exist and is only an artefact of linguistic analysis
(see Victorri 1997; Kleiber 1999). It is certainly true that speakers are rarely
aware of the multiplicity of meanings unless prompted by the production or
comprehension of jokes or puns.
In lexical semantics, where the existence of polysemy is accepted, it has
always been difficult to distinguish polysemy from homonymy (see Heger
1963) - the existence of different but unrelated meanings for a single word
form - and thus to construct lexical entries coherently. To give a relatively
clear-cut example, the homonyms (river) bank and (financial) bank would
be accommodated in two entries, the meanings of the polysemous word nose
('facial organ', 'sense of smell' and 'attribute of a wine') would be accommo-
dated in one. But not all cases are so clear-cut.
Another problem arising from polysemy and homonymy is lexical ambi-
guity, and the precise relationship between polysemy, homonymy, ambiguity
and vagueness is still an unresolved issue in lexical semantics. Some solutions
are proposed in Andreas Blank's contribution to this volume (see also Geeraerts
1993; Tuggy 1993).
Most importantly, however, polysemy has proven almost intractable within
structural, especially feature-based and componential, theories of language
(Lyons 1977: 550-579; Wunderli 1995; for a critique see Deane 1988: 325),
especially those that are based on the axiom of language as an autonomous
system, cut off from cognition, emotion and bodily influences. In their efforts
to come to grips with polysemy, structural linguists fought to maintain the
theorem of "one form, one meaning" and transformational linguists focused
on invariant features to define meaning, leaving out issues of contextual and
co-textual modulation. From being an inspirational subject to historical lexical
semanticists, as envisaged by Breal and his followers, polysemy turned into
an obstacle to linguistic theory.
This all changed with the advent of cognitive linguistics in the 1980s,
when the obstacle became an opportunity, an opportunity to link language
back up with mind, meaning and society, a link already made by Breal (see
Nerlich, this volume). This change in perspective was facilitated by the
emergence of new theories in anthropology and psychology, most importantly
 Polysemy and flexibility: introduction and overview 5

new theories of how humans establish categories on the basis of prototypes

and family resemblances. In cognitive linguistics, the word itself with its
network of polysemous senses came to be regarded as a category in which
the senses of the word (i.e. the members of the category) are related to each
other by means of general cognitive principles such as metaphor, metonymy,
generalization, specialization, and image-schema transformations.
Unlike traditional research into polysemy inside historical and lexical
semantics, these analyses go both "beyond" words and "below" words,
studying poly se my at the level of morphological as well as syntactical con-
structions, and even viewing polysemy as a systemic phenomenon (see
Cuyckens and Zawada 2002; Lehrer, this volume).
Despite their inherently controversial nature, multiple (word) meanings
have been accounted for in dictionaries for at least two centuries. Polysemy
has also been the basis for research into semantic change, which can be
regarded as a result of the polysemous accretion and variation of meanings
clustering around a word form over time (see Nerlich and Clarke 1992;
Geeraerts 1997; Aitchison and Lewis, this volume; Blank, this volume). Using
the older insights into metaphor and metonymy as mechanisms of sense
extension, polysemy has become the focus of attention for modern cognitive
linguists and cognitive semanticists studying metaphors, metonymies and
conceptual blending (Fauconnier and Turner; Tyler and Evans; Seto, this
volume), prototypes, semantic fields, semantic frames, and semantic networks
(Wilks 1977; Brugman 1981; longen 1985; Norvig and Lakoff 1987; Deane"
1988; Schneider 1988; Schulze 1988, 1994; Lehrer 1990, this volume; Sweetser
1990; Fillmore and Atkins 1992; Taylor 1992; Dolling 1993; Cuyckens 1994;
Sinha and Kuteva 1994; Cruse 1995; Geeraerts 1997; Raukko 1999, this
volume; Cuyckens and Zawada 2002). In these contemporary fields of research
polysemy is no longer regarded as a problem, but rather as an inherent feature
of language, language use and cognition.
For the last two decades polysemy has also been at the centre of attention
of computational linguistics (see Wilks, this volume), where problems of word
senses and word sense disambiguation are vividly discussed (see Kilgarriff,
this volume; see also Ostler and Atkins 1991; Pustejovsky 1991, 1995; Asher
and Lascarides 1995; Kilgarriff and Gazdar 1995; Viegas and Raskin 1998;
Ravin and Leacock 2000).
Psycholinguists and students of AI have also studied polysemy, ambiguity
and sense disambiguation for over three decades and have developed various
models of semantic networks, first inside the framework of feature semantics,
6 Brigitte Nerlich and David D. Clarke

then prototype and frame semantics (see WiIliams 1992; Rueckl 1995).
Polysemy can in fact be regarded as the hub around which all the multiple
but related synchronic and diachronic theories of language and cognition
turn.
Developmental psychologists and linguists interested in language acqui-
sition have also begun to study polysemy (see Mason, Kniseley and Kendall
1979; Durkin, Crowther and Shire 1986; Johnson 1997,2001; Nerlich, Todd
and Clarke 1998; Israel, to appear). One contribution to this volume will
explore the acquisition of the different meanings of the polysemous verb
get from 4 to 10 years, and it will also investigate whether developmental
psychology can shed light on recent developments in cognitive linguistics
and vice versa (Nerlich, Todd and Clarke, this volume; see also Raukko, this
volume, for a parallel investigation of the "adult polysemy" of get).
There are still many questions to be answered in this field of research.
Here are just a few of them: do children learn the various subsenses of a word
in fixed collocational patterns and only later relate them to one core meaning?
Or do they first have a very vague global meaning for one word and then
tease out the subsenses? Do they learn the most frequently used subsenses
first? How do children start to enjoy and understand jokes and puns based
on the exploitation of polysemy/ambiguity? Why does this seem to happen
only after the age of 7? And why is this difficult for autistic children? As
conceptual metaphorical mappings are not primarily matters of language,
but part of our conceptual systems which allow us to use sensory-motor
concepts in the service of abstract reason, it is assumed by many cognitive
linguists that children acquire conceptual metaphorical mappings automatically
and unconsciously via their everyday functioning in the world (see Lakoff
and 10hnson 1999: Chapter 4). All this needs to be studied empirically.
Could it be that children put together knowledge elements into networks
that follow the same pathways as the accretion of polysemous meanings
by a word form over time? Do children's spontaneous creations of multiple
meanings match the patterns of adult polysemy, that is, follow metaphorical
and metonymical pathways? And if not, how are children's networks of
polysemous meanings adjusted to the adult ones? Another question one should
try to answer is: do children readily understand polysemous words? What
kind of polysemous words do they understand best? Does it depend on the
grammatical category? Does it depend on the frequency/familiarity or salience
(see Hughes 1989; Giora 1997)? Does it depend on the type of figurative
mapping underlying the mUltiple meanings of a word? Are metonymical
 Polysemy and flexibility: introduction and overview 7

extensions more readily understood than metaphorical ones or vice versa?

Does it depend on the degree of prototypicality? Or does it depend on a
particular combination of all these factors? In this book only a few of these
questions can be addressed (see Dowker, this volume; Nerlich, Todd and
Clarke, this volume), but they have at least to be stated.
Furthermore, it has become clear that polysemy cannot be studied inside
linguistics and psychology alone. It can only be understood if we also look
at human culture, as Brown and Witkowski have already stressed:

Polysemy is ubiquitous in language and its investigation has considerable

potential for illuminating human cognition. In addition, the regular patterns
of lexical change ... indicate that the lexicon is amenable to systematic
in vestigation as are other components of language. Most importantly, the study
of these regular lexical patterns can contribute significantly to knowledge of
the processes and capacities which underlie human language and culture.
(Brown and Witkowski 1983: 83, italics added)

Throughout the book we shall attempt to link linguistic, cognitive, and

cultural approaches to the study of polysemy, and also to link these to
diachronic and synchronic approaches (Fauconnier and Turner; Dowker;
Aitchison; Blank, this volume).
In recent years, research in cognitive semantics has shown that the lexi-
calized meanings of polysemous words can be explained in terms of basic
conceptual and cultural metaphors, that polysemy is motivated partially by
our metaphorical and metonymical structuring of experience, especially of
our bodily experience (Lakoff and 10hnson 1980; Bartsch 1984,2002; Dirven
1985; Gibbs 1994: 40-41; see Seto, this volume). Words do not accumulate
meanings at random, but follow certain cognitive pathways or patterns that
are natural to human cognition and that structure our acquisition of experi-
ence, knowledge and language. So, to understand the emergence and structure
of polysemy, we have to understand the nature of metaphor and metonymy,
an insight familiar to those studying semantic change a century ago (see
Nerlich and Clarke 1997; Blank, this volume; Nerlich, this volume).
More recently, it has become clear that rather than being a system for
representing concepts, language is a system of prompts for conceptual
integration. In this view linguistic expressions prompt for meanings rather
than represent meanings. Polysemy is a dynamic byproduct of this operation
of conceptual integration, it is not a static property of the words themselves.
Polysemy is therefore ubiquitous but also barely noticeable in most cases.
8 Brigitte Nerlich and David D. Clarke

We only become aware of its existence when the emergent meaning in a

blend, for example, seems remarkably distant from the domain of the input
from which the words came. When we notice this distance, we call it by one
of many names used in traditional semantics: extension, bleaching, analogy,
metaphor, and displacement (see Fauconnier and Turner, this volume; Aitchison
and Lewis, this volume). The problem of "graded polysemy" will be explored
further in the next section. Some of the problems that this view throws up
for computational approaches to polysemy are explored in Kilgarriff's contri-
bution to this volume.

2. Polysemy: theories and hypotheses

In the context of this book, we adopt as a working hypothesis the view that
almost every word is more or less polysemous, with senses linked to a
prototype by a set of relational semantic principles which incorporate a greater
or lesser amount of flexibility. We follow the now common practice in
poly se my research and regard polysemy as a graded phenomenon (see Cowie
1982; Lipka 1990; Gibbs 1994; Cruse 1995), ranging between what some
call contrastive polysemy and complementary polysemy (Weinreich 1966;
Pustejovsky 1995), where contrastive polysemy deals with homonyms such
as match (a small stick with a tip which ignites when scraped on a rough
surface) and match (contest in a game or sport), whereas complementary
polysemy deals with interrelated semantic aspects of a word, such as, in the
case of record, for example, the physical object and the music. Cases, such
as smart as in "a smart person" or "a smart dress", and/air as in "a fair trial"
and "fair hair" lie somewhere in between. Other polysemy researchers, such
as Cruse, have tried to deal with this semantic gradedness by invoking the
notion of sense-nodules or regions of higher semantic density within the
extreme variability of senses (Cruse 2000). The concept of gradedness also
applies to the difference between literal and metaphorical meaning, as recent
research seems to confirm (see Gibbs 1994; Giora, this volume).
Take the following examples:

(1) He gave me a book.

(2) He gave me a strange look.
(3) He gave me a hearing test.
(4) He gave me hell.
 Polysemy and flexibility: introduction and overview 9

In (1) give refers to a process in which ownership of an object changes

from person A to person B as the result of an action on the part of person A.
"Possession" is what prototype theorists would call the most prototypical
meaning of give. In the other cases this prototypical meaning of "possession"
is mapped onto various other human experiences. In (2) give refers to another
kind of action, in which no change of ownership is involved and the "object"
is of an abstract kind. In (3) give is synonymous with 'administer', and in
(4) the "object transferred" is of an entirely abstract, metaphorical, nature
and the expression in which give is used has become idiomatic (see Lee
1986: 1-2).
The way in which we understand the various meanings of the word give
depends on accommodation in context (which includes knowledge about
idioms, where accommodation is reduced to "zero" and understanding has
become automatic). As Langacker has pointed out:

It must be emphasized that syntagmatic combination involves more than the

simple addition of components. A composite structure is an integrated system
formed by coordinating its components in a specific, often elaborate manner.
In fact, it often has properties that go beyond what one might expect from its
components alone .... [O]ne component may need to be adjusted in certain
details when integrated to form a composite structure; I refer to this as
accommodation. (Langacker 1987: 76-77).

In the interpretation of polysemous words (in fact almost any word) there
is therefore always a process of accommodation (and assimilation) involved,
between what is given semantically, syntactically, and what we infer from
the surrounding pragmatic context of discourse -leading to what Fauconnier
and Turner call "conceptual integration". As Wittgenstein said in his Philo-
.mphical Grammar: "Well, 'Understanding' is not the name of a single process
accompanying reading or hearing, but of more or less interrelated processes
against a background, or in a context" (Wittgenstein 1974, §35: 74).
Wittgenstein, the "father" of the well-known concept of "family resemb-
lances", so central to prototype theory, also stressed the gradedness of the
multiple senses of a word in this passage from his Philosophical Grammar:

What a concept-word indicates is certainly a kinship between objects, but

this kinship need not be the sharing of a common property or a constituent. It
may connect the objects like the links of a chain, so that one is linked to
another by intermediary links. Two neighbouring members may have common
10 Brigitte Nerlich and David D. Clarke

features and be similar to each other, while distant ones belong to the same
family without any longer having anything in common. Indeed even if a
feature is common to all members of the family it need not be that feature
that defines the concept.
The relations between the members of a concept may be set up by the
sharing of features which show up in the family of the concept, crossing and
overlapping in very complicated ways.
Thus there is probably no simple characteristic which is common to all the
things we call games. But it can't be said either that "game" just has several
independent meanings (rather like the word "bank"). What we call "games"
are procedures interrelated in various ways with many different transitions
between one and another. (Wittgenstein 1974, §35: 75; emphasis added)

This polysemous gradedness is not just a synchronic, but also a diachronic

phenomenon, as conversational implicatures generating ad hoc ambiguities
and polysemies can become conventionalized or grammaticalized over time
(see Traugott 1994), and as the distant senses of a polysemous word can at
any time change status and generate new homonyms, that is, the semantic
link in the network of senses can become obscured. What was motivated
once becomes arbitrary. Polysemy always teeters on the edge of the semantic-
pragmatic interface and of the synchronic-diachronic interface - between
order and disorder.
This means that there is not so much a dichotomy between polysemy and
homonymy, but rather a synchronic and diachronic gradient based on complex
networks of meaning relations, of which the most basic ones are radial
networks surrounding one prototype, and family resemblances, based on
several linked prototypes or subnetworks. One can try to distinguish between
various types of poly se my along this gradient (see Norrick 1981; eruse 1995;
Blank, this volume), but this will always be fraught with difficulties, as
Geeraerts has pointed out (Geeraerts 1993, 1994; see also Nunberg 1979,
1992).
Geeraerts has written a number of important articles on cognitive sem-
antics, polysemy, prototype theory, and semantic change. In his plea for an
analysis of semantic change and polysemy in the framework of prototype
theory, Geeraerts tries to answer two questions:

(i) Why should diachronic structures reflect synchronic structures? And

(ii) Why should the human brain have a prototypical conceptual organiz-
ation at all? (Geeraerts 1985: 140)
 Polysemy and flexibility: introduction and overview 11

Geeraerts gives three reasons, all of which are linked to the functions of
the human mind, namely: storing information or accumulating world knowl-
edge and using it for cognitive purposes.

(i) Informational density. Categories should be as informative as possible,

which is achieved by prototypical clustering of subconcepts.
(ii) Structural stability. The categorical system can only work efficiently
if it does not change drastically, but at the same time it should not be
too rigid.
(iii) Flexible adaptability. Concepts should be able to adapt to changing
conditions in the world and to changing expressive needs on the part
of the users of the concepts; they should also be able to integrate
marginal or new members, without losing their stability and holistic
quality.

If it is true that the structure of categories, including semantic categories,

should be as dense as possible and as flexible as possible so as to be concep-
tually efficient, then one should expect in natural languages a tendency
towards the maximization of lexical polysemy. This tendency is evidently at
work in languages as any glance at a dictionary can demonstrate, especially
an English one.
However, there seems to be a certain point when the network of senses
becomes "saturated" and "strained" and begins to "shed" certain meanings
(see also Nerlich, Todd and Clarke, this volume). Either senses become so
distant that they are perceived as unrelated, such as French voler 'to fly' and
voler 'to steal' (the words become "homonyms"), or the word becomes part
and parcel of a free-standing semantic conglomerate. In the case of the highly
polysemous verb get, for example, standing expressions, such as Get a life
have developed, as well as phrasal verbs, such as to get into something, get
over something, get out of something. These phrasal verbs can again become
polysemous themselves, developing their own networks of metaphorical and
metonymical senses based on the primary locative ones, and so on. There is
therefore an ongoing cycle of emergent polysemy, conventionalized or lex i-
calized polysemy, and dead polysemy or homonymy.
As larno Raukko has pointed out in his contribution to this volume, it is
important "to see polysemy as patterns of flexibility in (lexical) meaning in
much the same way as it is accepted that situational (utterance or discourse
level) meaning is nonfixed, inexact and negotiable."
12 Brigitte Nerlich and David D. Clarke

This view is opposed to the so-called unitary, isomorphic or monosemous

view of meaning (on this issue see Bierwisch 1983; Ruhl 1989; Groefsema
1995; Geeraerts 1997; Blank, this volume). What is ideal for certain artificial
languages, that is, an unambiguous one-to-one correspondence between a
word and its meaning, is far from ideal for ordinary language. In ordinary
language we do not match words and meanings one by one, instead we have
learned to match (flexible and adaptable) patterns of meanings surrounding
a polysemous word (based on natural conceptual patterns) with certain con-
textual patterns in which the word is embedded - we match patterns. People
seem to be able to disambiguate polysemous words instantly and auto-
matically by recognizing and adjusting certain correlations between the
adaptable and changeable network of senses and the context of discourse in
which a word is used habitually; but they can also leave multiple meanings
"hanging in the air" if that serves their communicational purpose.
We hypothesize that the relation between words, meanings and context
naturally assigns appropriate meanings in the mind of the hearer/child whether
the word is polysemous or not. It is not the case that polysemy is a different
and harder "semantic nut" to crack for the hearer; rather, monosemous words
(e.g. aspidistra) have failed to exploit the full power of this mechanism. If
this hypothesis is true, the acquisition of polysemous words should not be
more difficult, but instead rather easier for children than the acquisition of
monosemous words.
Recent research by Pustejovsky is also based on the axiom that there is
no direct correspondence between words and meanings and that polysemy is
widespread. His theory of how we interpret polysemous words in context
has become a high-point of 1990s polysemy research and therefore needs to
be summarized briefly (see also Warren, this volume). Following Weinreich's
lead, Pustejovksy distinguishes between words which exhibit contrastive
polysemy, such as match (in sports) and match (in the match box) and words
which exhibit complementary polysemy, such as paper, as in making some-
thing out of paper, reading a paper, writing a paper, giving a paper, and so
on. He proposes to treat these two types of polysemous words in two different
ways.
In the case of contrastive polysemy the meanings correspond to so many
different words. In the case of complementary polysemy one word has
complementary meanings, what he calls a lexical conceptual paradigm.
Pustejovsky claims that a word is structured semantically by being embedded
in four fields: the argument structure, the event structure, qualia roles and
 Polysemy and flexibility: introduction and overview 13

the structure of lexical inheritance. According to Pustejovsky, nominals have

qualia structures which determine their meaning just as argument structures
determine verb meaning. He proposes the following qualia roles: constitutive
(constituent matter/parts), formal (size, shape, position etc.), telic (purpose)
and agentive (ontogeny). When a verb is combined with a noun its argument
structure must fit the qualia roles of that noun.
This may lead to the highlighting of particular roles. Take book as an
example. When highlighting the domain or qualia role of a physical object
(constitutive) we can say "I destroyed the book"; when highlighting its telic
role we can say "I read the book"; and so on. This highlighting is also
exploited in well-known jokes such as this: "I spent two months in the south
of France finishing my novel. I am a very slow reader", which plays with
two quaJia roles, the telic role (a novel "for reading") and the agentive role
("writing" a novel).
As this joke shows, speakers and hearers do not just "disambiguate"
polysemous words, they also have the option of exploiting polysemy and
ambiguity for conversational profit, and this despite the fact that keeping
mUltiple meanings in mind has some cognitive cost (Nerlich and Chamizo
Dominguez 1999; Giora, this volume). In ordinary language we often "ambig-
uate" purposefully so as to negotiate conversational turns and increase our
conversational prestige. Exploiting polysemy in this way leads to social
bonding (through shared laughter, for example), but it might also increase
the semantic, cognitive, and maybe even neuropsychological bonds between
meanings and concepts. We want to stress that (lexical) polysemy is not just
a phenomenon of the dictionary or of rather sterile disambiguation tasks in
psycholinguistics. Polysemy is an ordinary language and ordinary life phen-
omenon (see Nerlich and Clarke 2001).
However, the exploitation of polysemy for communicative and even
commercial purposes (as in advertising) has a price in terms of cognitive
processing. Experimental studies of the understanding of polysemous words
have shown that processing polysemous words results in the continued acti-
vation of inappropriate senses for quite some time after the word has been
encountered (WilJiams 1992; see also Gibbs 1994: 41). Other studies have
shown that interlocutors make use of what is available to them, regardless of
contextual information or speaker's intent (see Giora, this volume). They
access the most salient (frequent, conventionalized) meaning first, which is
sometimes neither the literal meaning nor the intended meaning (Giora 1997,
1999). Keeping several semantic options open has therefore cognitive costs,
14 Brigitte Nerlich and David D. Clarke

but it may also have communicational benefits, as salient meanings that have
not been deactivated may easily be reused by the discourse participants for
special purposes. This may then lead to strengthening of the semantic bonds
between the senses of a word. Becoming salient in conversation, they will
become more accessible and therefore more frequently used and usable.
However, the communicational costs of using polysemous words are also
obvious. There is always a danger of being misunderstood or of falling into
semantic traps: as when the topic of the long overdue renovation of the staff
toilets is discussed at a departmental meeting and someone says: "Any
movement on this issue would be welcome" ...
Any theory of polysemy has to achieve a very precarious balancing act
between the maximization of polysemy, where the words themselves carry
most of the polysemous workload and speakers just have to choose correctly
in context, and a minimization of polysemy which leaves most of the work
to the pragmatic component, that is to the interpretational work done by the
speakers. One way out of this dilemma is proposed by Anna Wierzbicka
(1996) who distinguishes between semantic primes (undefinable words),
words that can be defined or paraphrased reductively using these primes and
are therefore not polysemous, and words which need several paraphrases
and can therefore be regarded as polysemous.
Maximalist descriptions of polysemy differ from minimalist ones in their
psycho-communicational consequences. If most of the relevant meaning is
coded, it will be recoverable from memory and need not be constructed on
the spot. The consequence is reliable functioning of the communication but
lack of flexibility. If little of the relevant meaning is coded, the interpreter's
memory will not be overburdened, but his reasoning capacity will be occupied
in the situation concerned. The message will not become available indepen-
dently of the communication situation and will be less stereotyped. As a
consequence there will be less reliable functioning but more flexibility and
creativity (see Posner 1996: 236; on ways out of the minimalistlmaximalist
dilemma, see Warren, this volume; Tyler and Evans, this volume; Taylor,
this volume; Janssen, to appear).
Related to the minimalistlmaximalist difference in approaches to polysemy
is the abstractivistlcognitivist difference (see Behrens 1999). According to
the abstractivist view advocated for example by Caramazza and Grober
(1976), each polysemous word has an abstract overall (literal) sense and the
extended senses of the word can be derived contextually. This view has been
contested in more recent work in lexical semantics which suggests that the
 Polysemy and flexibility: introduction and ove",iew 15

meaning of many polysemous words can be explained in terms of basic

metaphors, metonymies, cognitive models and frames that motivate the sense
extensions (see Gibbs 1994).
As Fritz has recently pointed out, the maximalist and the minimalist
approach have their advantages and disadvantages, but the maximalist one is
better suited for both synchronic and diachronic semantics:

Semantic minimalism, advocated both by structuralists ... and, more recently,

by Griceans ... , is basically a healthy principle in that it forces researchers
to differentiate between what a word means and what is conversationally
implicated in a certain context. But it also fosters the tendency to explain
as implicatures what must be seen as established uses of expressions. The
latter tendency is definitely harmful for historical semantics. Authors
working in a cognitive-semantics framework accept polysemy as a funda-
mental semantic fact and explain the unity of meaning not by a minimalist
reduction to basic meanings, but by positing an internal structure within a set
of uses. This approach permits treating semantic developments as develop-
ments of sets of uses and their internal structures. (Fritz 1998b: 185; see also
Fritz 1998a)

A decade or so before Fritz, Lieberman had already pointed out that:

We know that the word table has a fuzzy floating set of references; but we
also know that there is some precise limit to the range of references. The
trouble comes in attempting to capture formally the precision and the
fuzziness. We can use the word table to refer to all manners of things that
have some property of tableness in a particular setting. The degree to which
we all ascribe the property of tableness to something in any setting will vary.
And the quality of tableness will change for each of us. What seems to be a
table at some time in some place may not be a table to us at another time in
another place.
Language is inherently ambiguous and uncertain. That is the problem and
the power of the system .... We must always creatively interpret an utterance:
the new interpretation always has the potential of achieving a new insight.
(Lieberman 1984: 82)

Although there is a danger that this might be taken the wrong (namely
modular) way, one could say that it is the function of sentential context to
mediate between semantics and pragmatics and, in a sense, between maxima-
list and minimalist polysemy:
16 Brigitte Nerlich and David D. Clarke

The rules of syntax have a functional purpose: they limit some of the semantic
referents of a word. In the sentence It's important to bank your money, the
syntax of English limits the semantic referents of the word bank to the range
of concepts relevant to its functioning as a verb. (Lieberman 1984: 85-86)

A tiny bit of syntagmatic bonding is sometimes sufficient to define the

sense of an expression successfully. Take for example rock quarry and rock
band (see Osgood 1980: 226). Where such syntagmatic or collocational
information is insufficient, as in rock idoL or hard rock (to give two very
simplified examples), our world knowledge, our knowledge of context, and
our knowledge of conceptual metaphors and metonymies, cognitive models,
and frames comes into its own.
Polysemy is a phenomenon that exposes the multiple relations and connec-
tions between syntax, semantics and pragmatics, and between language,
cognition and social interaction. It can only be studied if we try not to isolate
one from another.

3. Contributions to the volume

This introduction will be followed by two contributions which set the scene
for the following articles, a theoretical one by Taylor and a historiographical
one by Nerlich, both exploring the puzzles that polysemy has posed to linguists
and philosophers of language over time and is still posing to linguists now.
The articles by Fauconnierffurner, TylerlEvans, Raukko, and Seto explore
the wider theoretical and methodological debates connected with issues of
polysemy, blending and flexible meaning, and the status of metonymy inside
a cognitive theory of polysemy. Fauconnierffurner and TylerlEvans focus
on polysemy as part of an exploration of human conceptualization by studying
examples of blending and the various uses of over in (American) English.
Raukko's contribution provides detailed comparative analyses of the English
word get and the Finnish word pitiiii. It establishes a direct link to the
chapters by Warren and Lehrer, where polysemous English word forms
and polysemous constructions are investigated in detail. All three articles
widen the field of polysemy research from the lexicon to morphology and
syntactic constructions and make new contributions to methodology. The
articles by AitchisonlLewis and Blank study various levels of polysemy and
bleaching from the synchronic as well as diachronic point of view, not only
 Polysemy and flexibility: introduction and overview 17

in English, but also in the Romance languages. As Blank focuses on polysemy

in the lexicon and in discourse, his article is followed by GioralGur's contri-
bution, which discusses irony and polysemy from a psycholinguistic and
discursive point of view. The developmental dimension of polysemy (and
metaphor) is explored by Dowker, who provides an empirical analysis of
children's use of figurative language from a cross-cultural perspective, and
by Nerlichrrodd/Clarke, who examine the acquisition of get by English
children between age 4 and 10. This chapter links back to the empirical
investigation of adult uses of get by Raukko, with which it shares the experi-
mental approach. In the final part of the book Kilgarriff and Wilks debate
whether or not there are word senses and how to disentangle them by compu-
tational means.
Following this introduction, John Taylor explores the controversies
associated with the concept of polysemy, but does not intend to solve them.
Instead of refining the concept he examines the cognitive models that are
used to frame the conceptualizations of polysemy. He draws attention to
three mutually supporting cognitive models: the semiotic model of language,
the building block metaphor of syntagmatic combination, and the demarcation
of lexicon and syntax.
In her chapter Brigitte Nerlich traces the history of the concept of
"polysemy" from Antiquity to the first half of the 20th century. Breal's
treatment of polysemy is the pivot around which the article itself turns, as it
was Breal who invented the term "polysemy" a century ago and laid the first
theoretical foundations for a study of polysemy as a linguistic, historical and
cognitive phenomenon.
In their article "Polysemy and conceptual blending" Gilles Fauconnier
and Mark Thrner put polysemy research into the broader context of research
in cognitive linguistics and literary theory. They look at some aspects of
polysemy which derive from the power of meaning potential. More specifi-
cally, they focus on aspects linked to the operation of conceptual blending, a
major cognitive resource for creativity in many of its manifestations. They
argue that polysemy is pervasive in language and appears in many forms. It
is not just an accident of history or of synchrony, but rather an essential
manifestation of the flexibility, adaptability, and richness in meaning potential
that lie at the very heart of what a language is and what it is for. Fauconnierl
Turner review a number of classical cases of polysemy associated with
conceptual blending in order to give an idea of the overall importance of
blending in polysemy phenomena.
18 Brigitte Nerlich and David D. Clarke

The paper by Andrea Tyler and Vyvyan Evans (reprinted here with the
permission of Language) explores lexical poly se my through an in-depth
examination of the English preposition over. Working within a cognitive
linguistic framework, it illustrates the non-arbitrary quality of the mental
lexicon and the highly creative nature of the human conceptual system. The
analysis takes the following as basic: (i) human conceptualization is the
product of embodied experience, i.e. that the kinds of bodies and neural
architecture humans have, in conjunction with the nature of the spatio-
physical world humans inhabit, determines human conceptual structure; and
(ii) semantic structure derives from and reflects conceptual structure. As
humans interact with the world, they perceive recurring spatial configurations
which become represented in memory as abstract, imagistic conceptualiz-
ations. The authors posit that each preposition is represented by a primary
meaning, which they term a protoscene. The protoscene, in turn, interacts
with a highly constrained set of cognitive principles to derive a set of additional
distinct senses, forming a motivated semantic network. Tyler and Evans
conclude that previous accounts have failed to adequately develop criteria to
distinguish between coding in formal linguistic expression, and the nature of
conceptualization which integrates linguistic prompts in a way that is maxi-
mally coherent with and contingent upon sentential context and real-world
knowledge. To rectify this, they put forward a methodology for identifying
the protoscene and for distinguishing among distinct senses.
Like Seto in the following chapter, Jarno Raukko disputes some traditional
assumptions underlying polysemy research and puts forward a new experi-
mental approach focusing on flexibility. His chapter is therefore entitled:
"Polysemy as flexible meaning: experiments with English get and Finnish
pitiiii". Raukko claims that it is common to view polysemy as a collection or
network of several (interrelated) meanings that is fairly stable, fairly unprob-
lematic to segment and establish, and fairly much agreed upon by different
speakers. Yet, as he shows in his chapter, there are also opponents who wish
to reduce polysemy to patterns of contextual specifications or claim that
polysemy is irrelevant to the study of communication. One way of building
a bridge between these camps is to see polysemy as patterns of flexibility in
(lexical) meaning in much the same way as it is accepted that situational
(utterance or discourse level) meaning is nonfixed, imprecise, and negotiable.
The view about polysemy as flexible meaning does not imply that word-
specific descriptions of semantic variation would be unnecessary, but it seeks
to leave room for dynamicity and open-endedness in categorization as well
 Polysemy and flexibility: introduction and overview 19

as for intersubjective disagreements in views about the structure of the

polysemy of a given word. The article demonstrates some advantages of this
view with an analysis of results from polysemy experiments with nonlinguist
informants.
Blending research and research into metaphor as sources for polysemous
sense extensions are well represented in cognitive linguistics. More recently
we have seen a rising concern with metaphor's ugly sister, namely metonymy.
Recently published work on metonymy in the framework of cognitive ling-
uistics has shown that metonymy is not only as pervasive as metaphor but
also no less important in the daily use of language. However, in his chapter
"Metonymic polysemy and its place in meaning extension" Ken-ichi Seto
claims that the theory of metonymy presented in the mainstream cognitive
camp is flawed for a number of reasons. He argues specifically (i) that there
has been no satisfactory definition of metonymy yet, this being due to con-
fusions about the difference between entities and categories; (ii) that the
ultimate reason why those confusions so often occur resides in the (inevitable)
spatial representation of categorical relations; and (iii) that Langacker's
network model, which is supposed to deal with polysemy, does not quite
work because metonymy has no proper place in the model. Having criticized
the prevailing trend in polysemy and metonymy research, Seto then proposes
a new way of looking at polysemy: the cognitive triangle whose vertices are
metaphor, metonymy, and synecdoche.
The empirical exploration of polysemy in all types of linguistic phenomena
is continued in the chapter contributed by one of the earliest and best known
researchers in the field of polysemy, namely Adrienne Lehrer, working
within the framework of lexical field theory. In her chapter, "Polysemy in
derivational affixes", she pursues research into a domain of polysemy which
has often been overlooked. Whereas polysemy has been well investigated in
lexemes, much less research has been done on bound forms, especially on
derivational affixes. It has been established that the range of meanings
expressed by affixes (and function words) is limited, but in this chapter Lehrer
further explores the kinds of polysemy that can be found in derivations,
comparing them with the full range found in lexemes.
The chapter by Beatrice Warren, entitled 'The role of links and/or qualia
in modifier-head constructions", adopts one of the novel approaches to
polysemy developed by Pustejovsky and puts it to the empirical test. Like
the previous chapters by Fauconnierl1Urner and Raukko, Warren argues that
polysemy is evidence of lexical creativity. But she asks: is this creativity
20 Brigitte Nerlich and David D. Clarke

rule-governed as is believed to be the case with syntactic creativity? That is,

are there some finite means allowing infinite production of new senses? Her
position is that certain aspects of sense-formation are regular and predictable.
Building on work by Pustejovsky and herself, Warren traces the origin of a
certain kind of polysemy in adjectives, denominal verbs, compounds and
metonymies to a common cognitive basis.
With Jean Aitchison and Diane Lewis's chapter "Polysemy and bleaching"
we leave the detailed analysis of synchronic polysemies and their underlying
structures behind to enter the realm of diachronic poly se my and grammatical-
ization. We examine the reasons why words become polysemous and what
patterns of polysemy are established over time. AitchisonlLewis argue that
words which describe appalling events, such as disaster or catastrophe, tend
to "bleach" (fade in meaning). Yet their older unbleached meanings (such as
'a major catastrophic event') are often retained alongside their new, bleached
meanings (such as 'a minor mishap', 'a failure'). AitchisonlLewis study many
examples taken from the British National Corpus, a database of modern
written and spoken language, such as:

(5) At least 62 people were killed and 3,000 missing last night after an
underwater earthquake sent 50ft tidal waves crashing into the coast
of Nicaragua. More than 227 people were injured in the disaster.
(6) All other efforts to lose the fat from the offending areas proved to be
a disaster. If [lost weight below 54 kg my bust disappeared, yet nothing
went from my Legs or posterior!

This chapter explores some of the resulting polysemy. It asks two main
questions: first, how do hearers successfully interpret such fast-moving,
polysemous words? Second, what general processes can be identified in the
development of such polysemy? The diachronic theme explored in Aitchisonl
Lewis's chapter is continued in Blank's chapter.
In his chapter "Polysemy in the lexicon and in discourse" Andreas Blank
agrees initially with Breal who claimed that polysemy could be regarded as
the "synchronic side" of lexical semantic change. He argues that if we
conceive of semantic change as being based on associations between concepts
or concepts and linguistic signs, we can describe polysemy as the continuation
of these associative relations in synchrony. Blank warns us, however, not to
commit an easy fallacy: there is no complete isomorphism between diachronic
processes and synchronic states. His first aim is to explain the specific
 Polysemy and flexibility: introduction and overview 21

differences between diachrony and synchrony and to provide a typology of

the semantic relations that underlie polysemy, i.e. "types of polysemy", such
as "metaphoric polysemy", "metonymic polysemy", and so on (relations
which are also explored by Dowker, and Seto, this volume). He then addresses
another problem that concerns semantics: the distinction between contextual
variation ("vagueness"), polysemy, and homonymy ("ambiguity"). Here again,
the conception of semantic relations reveals its explanatory power, as it serves
as the main criterion used to distinguish between the three topics. Blank also
takes into account complex lexical developments, such as initial polysemy
that turns into homonymy and initial homonymy that is reinterpreted as
polysemy. However, the article not only deals with diachronic processes and
synchronic structures on which polysemy is based, but also reveals the
underlying discursive bases of polysemy, something often neglected in poly-
semy research, especially of the diachronic form. Blank therefore had to
modify Breal's view of polysemy as the direct lexicalized consequence of
semantic innovation. This is only one type in a set of possibilities.
The chapter by Rachel Giora and Inbal Gur, "Irony in conversation:
salience, role, and context effects", again challenges many presuppositions
underlying traditional polysemy research. Polysemy as flexible meaning is
directly studied in its discursive context, not from a historical point of
view, as in Blank's chapter, but from a psycholinguistics one. This chapter
focuses on one aspect of polysemy, namely irony, and it also contributes to
the debate about the flexible nature of literal and figurative meaning. Gioral
Gur report that previous research has demonstrated that familiar and less
familiar ironies are initially accessed literally. However, while familiar ironies
availed their ironic meaning in parallel with the literal meaning, less familiar
ironies facilitated ironically related concepts later. These findings support
the "graded salience hypothesis", and are inconsistent with the view that
context affects comprehension significantly. According to the graded salience
hypothesis, salient meanings should always be activated, even when they are
incompatible with the context. A meaning of a word or an expression is
salient if it is coded, i.e. retrievable from the mental lexicon. Factors contri-
buting to degrees of salience are conventionality, frequency, familiarity, and
prior context. Investigating irony reception in a spontaneous environment,
the authors show that more often than not, irony is responded to by resonating
with its salient literal interpretation. In friendly conversations, listeners very
often react to the literal meaning of the ironic utterance while at the same
time making it clear that they have also understood the implicated meaning.
22 Brigitte Nerlich and David D. Clarke

After having analysed some of the main synchronic, diachronic and dis-
cursive patterns of polysemy, we now come to patterns of polysemy which
emerge in and perhaps even structure the acquisition of certain lexical items.
Linking back to the articles by Seto, Blank, and Giora/Gur we put polysemy
back into the context of the study of figurative language, especially metaphor
and metonymy, but from a specifically psychological perspective. Ann Dowker
stresses in her chapter "Young children's and adults' use of figurative language:
how important are cultural and linguistic influences?" that there have been
numerous studies of the development and of the use and comprehension of
metaphors and other forms of figurative language. At least some types of
figurative language appear to develop very early: well before school age.
Figurative language seems to be a cross-cultural universal, but the precise
forms that it takes vary with language and culture. However, a recent study
by the author showed marked cross-cultural differences in the extent to which
4- to 6-year-old children used figurative language in their invented rhymes
and chants. One possible reason for cross-cultural variations in figurative
language may involve differences in the nature and extent of polysemy in
different languages. This chapter discusses (i) the broad course of develop-
ment of figurative language in early and middle childhood; (ii) cross-cultural
similarities and differences in its development and use; and (iii) the extent to
which cross-cultural variations in the development of figurative language
may reflect cross-linguistic variations concerning polysemy.
The developmental theme is continued in the chapter "Emerging patterns
and evolving polysemies: the acquisition of get between four and ten years",
in which Brigitte Nerlich, Zazie Todd and David D. Clarke study the
acquisition of the different meanings of the polysemous verb get by children
aged between 4 and 10 years. This chapter therefore complements Raukko's
chapter devoted to the flexible use of get by adults. Fifty-nine children took
part in an experiment that involved production and ranking tasks. The pro-
duction task showed that 4-year-olds only produced the main senses of 'have',
'fetch' and 'obtain'; by 10 years a much wider array of meanings was
produced; and at all ages syntactic frames were embedded in everyday
experiences. The ranking task involved selecting the best example of get from
'obtain', 'fetch', 'go', 'become', and 'understand'. Four-year-olds' knowledge
was structured around 'obtain' as the most prototypical meaning, and 10-
year-olds' responses were best explained by a prototype model of semantic
representation. It seems that in between, that is with the 7-year-olds, knowl-
edge of meaning was not yet organized into a prototypical scene. A theoretical
 PoLysemy and flexibility: introduction and overview 23

approach to the development of word meaning based on (synchronic and

diachronic) theories of polysemy and embodiment is proposed.
With the last two chapters by Kilgarriff and Wilks we enter the compu-
tational field of polysemy research and link it to issues in lexicography.
However, these papers not only address computational issues, but deal with
very fundamental theoretical, methodological and even philosophical issues
of polysemy research, such as: what are word senses and how can we disting-
uish between them? In the provocatively titled chapter "I don't believe in
word senses" Adam Kilgarriff argues that we often have strong intuitions
about words having multiple meanings, and that lexicography aims to capture
them, systematically and consistently. But how can lexicographers do this?
Can they find help in the literature dealing with the philosophy of language?
No, the philosophy literature does not provide a taxonomy of the processes
underpinning these intuitions, nor does it analyse the relations between the
word sense distinctions a dictionary makes and the primary data of naturally-
occurring language. This is a gap that this chapter aims to fill. Kilgarriff
shows that various attempts, such as Cottrell's connectionist approach (1989),
to provide the concept "word sense" with secure foundations have been
unsuccessful. He goes on to consider the lexicographers' understanding of
what they are doing when they make decisions about a word's senses, and
develops an altemati ve conception of the word sense, in which it corresponds
10 a cluster of citations for a word. Citations are clustered together where
Ihey exhibit similar patteming and meaning. The various possible relations
hetween a word's meaning potential and its dictionary senses are catalogued
and illustrated with corpus evidence, thus linking back to the method used
hy Aitchison!Lewis.
In "Senses and texts" Yorick Wilks addresses the question of whether it
is possible to sense-tag systematically, and on a large scale, and how we
should assess progress so far. That is to say, how to attach each occurrence
of a word in a text to one and only one sense in a dictionary - a particular
dictionary of course, and that is part of the problem. The paper does not
propose a solution to the question, although the author has reported empirical
findings elsewhere, and intends to continue and refine that work. The point
of this paper is to examine two well-known contributions critically: the first
(see Kilgarriff, this volume) is widely taken to show that the task, as defined,
l:annot be carried out systematically by humans whereas the second claims
strikingly good results at doing exactly that. This means that the aim of this
l:hapter is to attack two claims, both of which are widely believed, though
24 Brigitte Nerlich and David D. Clarke

not at once: that sense-tagging of corpora cannot be done, and that it has
been solved.
All the chapters in this book either propose novel approaches to polysemy,
or stringent criticisms of older approaches, or else are built on older approaches
to investigate novel problems. They all combine theoretical analysis with
empirical research and provide the reader with a multitude of examples and
a wealth of theoretical reflections. We hope that the field of poly se my research
will be enriched by them and that readers will come away with a deeper
knowledge, not only of polysemy, but of the functioning of language in mind
and in discourse, in synchrony and diachrony, in use and in acquisition.

References

Asher, Nicholas and Alex Lascarides

1995 Lexical disambiguation in a discourse context. Journal of Semantics
12(1): 69-\08.
Bartsch, Renate
1984 The structure of word meanings: polysemy, metaphor, and metonymy.
In: Fred Landman and Frank Veltman (eds.), Varieties of Formal
Semantics, 27-54. Dordrecht: Foris.
2002 Generating polysemy: metaphor and metonymy. In: Rene Dirven and
Ralf Porings (eds.), Metaphor and Metonymy in Comparison and
Contrast, 49-74. Cognitive Linguistics Research 2. Berlin and New
York: Mouton de Gruyter.
Behrens, Leila
1999 Aspects of polysemy. In: David A. Cruse, Franz Hundsnurscher,
Michael Job and Peter Rolf Lutzeier (eds.), Lexikologie - Lexicology
(HSK) Vo!. I. Berlin: WaIter de Gruyter.
Bierwisch, Manfred
1983 Semantische und konzeptionelle Reprasentation lexikalischer Einheiten.
In: Rudolf Ruzicka and Wolfgang Motsch (eds.), Untersuchungen
zur Semantik, 61-99. Berlin: Akademie-Verlag.
Breal, Michel
1924 [1897] Essai de semantique (Science des significations). Reprint of the 4th
Edition. Paris: Gerard Monfort.
Brown, Cecil H. and Stanley R. Witkowski
1983 Polysemy, lexical change, and cultural importance. Man 18: 72-89.
Brugman, Claudia
1981 Story of over. M.A. thesis. University of California, Berkeley.
1997 Polysemy. In: Jef Verschueren, Jan-Ola bstman, Jan Blommaert and
 Polysemy and flexibility: introduction and overview 25

Chris Bulcaen (eds.), Handbook of Pragmatics. Amsterdam and

Philadelphia: John Benjamins. Printed 1999.
Caramazza, Alfonso and Ellen Grober
1976 Polysemy and the structure of the subjective lexicon. In: Clea Rameh
(ed.), Semantics: Theory and Application, 181-206. Washington, DC:
Georgetown University Press.
Cottrell, Garrison W.
1989 A Connectionist Approach to Word Sense Disambiguation. London:
Pitman.
Cowie, H. P.
1982 Polysemy and the structure of lexical fields. Nottingham Linguistic
Circular 11(2): 51-65.
Cruse, David A.
1995 Polysemy and related phenomena from a cognitive linguistic point
of view. In: Patrick Saint-Dizier and EveIyne Viegas (eds.), Compu-
tational Lexical Semantics, 33-49. Cambridge: Cambridge University
Press.
2000 Meaning in Language. Oxford: Oxford University Press.
Cuyckens, Hubert
1994 Family resemblance in the Dutch spatial preposition op. In: Schwarz
(1994), 179-196.
Cuyckens, Hubert and Britta Zawada (eds.)
2002 Polysemy in Cognitive Linguistics. Amsterdam and Philadelphia: John
Benjamins.
Deane, Paul D.
1988 Polysemy and cognition. Lingua 75: 325-361.
Dirven, Rene
1985 Metaphor as a basic means for extending the lexicon. In: Wolfgang
Paprotte and Rene Dirven (eds.), The Ubiquity of Metaphor, 85-
119. Amsterdam: John Benjamins.
Dolling, Johannes
1993 Polyemy and sort coercion in semantic representations. In: Peter
Bosch and Peter Gerstl (eds.), Discourse and Lexical Meaning, 61-
78. Arbeitspapiere des SFB 340, Bericht Nr 30.
Durkin, Kevin, R. D. Crowther and B. Shire
1986 Children's processing of polysemous vocabulary in school. In: Kevin
Durkin (ed.), lAnguage Development in the School Years. London:
Croom Helm.
Fillmore, Charles and B. T. Sue Atkins
1992 Towards a frame-based lexicon: the semantics of risk and its
neighbors. In: Adrienne Lehrer and Eva F. Kittay (eds.), Frames,
Fields and Contrasts, 75-102. HiIIsdale, NJ: Lawrence Erlbaum
Associates.
26 Brigitte Nerlich and David D. Clarke

Fritz, Gerd
1998a Historische Semantik. Stuttgart and Weimar: Metzler.
1998b Review of Geeraerts (1997). Lexicology 4(1): 183-192.
Geerael1s, Dirk
1985 Cognitive restrictions on the structure of semantic change. In: Jacek
Fisiak (ed.), Historical Semantics. Historical Word-Formation, 126-
153. Berlin, New York, Amsterdam: Mouton de Gruyter.
1993 Vagueness's puzzles, polysemy's vagaries. Cognitive Linguistics 4(3):
223-272.
1994 Polysemy. In: R. E. Asher and J. M. Y. Simpson (eds.), Encyclopedia
of Language and Linguistics, 3227-3228. Oxford: Pergamon.
1997 Diachronic Prototype Semantics. A contribution to historicallexicology.
Oxford: Clarenden.
Gibbs, Raymond w., Jr.
1994 The Poetics of Mind: Figurative Thought, Language, and Understanding.
Cambridge: Cambridge University Press.
Giora, Rachel
1997 Understanding figurative and literal language: the graded salience
hypothesis. Cognitive Linguistics 7(1): 183-206.
1999 On the priority of salient meanings: studies of literal and figurative
language. Journal of Pragmatics 31: 919-929.
Groefsema, Marjolein
1995 Can, may, must and should - a relevance theoretic account. Journal
of Linguistics 31(1): 53-79.
Heger, Klaus
1963 Homographie, Homonymie und Polysemie. Zeitschriftfur Romanische
Philologie 79: 471-491.
Hughes, Alva T.
1989 Polysemy and Frequency in Word Knowledge Acquisition. Dissertation
Abstracts International, 1989 Mar, v49 (n9-B): 4046-4047.
Hiillen, Werner and Rainer Schulze (eds.)
1988. Understanding the Lexicon: Meaning, Sense and World Knowledge
in Lexical Semantics. Tiibingen: Niemeyer.
Israel, Michael
(to appear) Systematic idiomaticity in the acquisition of English get construc-
tions. In: Arie Vergagen and Jeroen van de Weijer (eds.), Agreement
in Form: Conventions on Different Linguistic Levels.
Janssen, Theo A. J. M.
(to appear) Monosemy versus polysemy. In: Hubert Cuyckens, Rem! Dirven and
John Taylor (eds.), Cognitive Linguistic Approaches to Lexical Semantics.
Cognitive Linguistics Research Series. Berlin: Mouton de Gruyter.
Johnson, Christopher Ronald
1997 Learnability in the acquisition of multiple senses: SOURCE recon-
 Polysemy and flexibility: introduction and overview 27

sidered. Proceedings of the 22nd Annual Meeting of the Berkeley

Linguistics Society, 469~80. Berkeley: Berkeley Linguistics Society.
2001 Metaphor vs. conflation in the acquisition of polysemy: the case of
see. In: Masako K. Hiraga, Chris Sinha, and Sherman Wilcox (eds.),
Cultural, Typological and Psychological Perspectives in Cognitive
Linguistics, 155-170. Amsterdam and Philadelphia: John Benjamins.
Jongen, Rene (ed.)
1985 La polysemie. Louvain-La-Neuve: Cabay.
Kilgarriff, Adam and Gerald Gazdar
1995 Polysemous relations. In: Frank R. Palmer (ed.), Grammar and Meaning:
Essays in Honour of Sir John Lyons, 1-25. Cambridge: Cambridge
University Press.
Kleiber, Georges
1999 Problemes de semantique: La polysemie en question. Villeneuve
d' Ascq: Septentrion.
Lakoff George and Mark Johnson
1980 Metaphors We Live By. Chicago: University of Chicago Press.
1999 Philosophy in the Flesh: The Embodied Mind and its Challenge to
Western Thought. New York: Basic Books.
Langacker, Ronald W.
1987 Foundations of Cognitive Grammar; Volume J: Theoretical Prerequisites.
Stanford, CA: Stanford University Press.
Lee, David
1986 Language, Children and Society: An Introduction to Linguistics and
Language Development. Brighton, UK: The Harvester Press.
Lehrer, Adrienne
1990 Polysemy, conventionality, and the structure of the lexicon. Cognitive
Linguistics 1(2): 207-246.
Lieberman, Philip
1984 The Biology and Evolution of Language. Cambridge, MA: Harvard
University Press.
Lipka, Leonhard
1990 An Outline of English Lexicology: Lexical Structure, Word Semantics,
and Word-Formation. Tilbingen: Max Niemeyer.
Lyons, John
1977 Semantics, Volume 11. Cambridge: Cambridge University Press.
Mason, J. M., E. Kniseley and 1. Kendall
1979 Effects of polysemous words on sentence comprehension. Reading
Research Quarterly 15(1): 49-65.
Nerlich, Brigitte and Pedro J. Chamizo Domfnguez
1999 C6mo hacer cos as con palabras polisemicas: El uso de la ambigiledad
en el lenguaje ordinario. Contrastes. Revista Interdisciplinar de
Filosoffa IV: 77-96.
28 Brigitte Nerlich and David D. Clarke

Nerlich, Brigitte and David D. Clarke

1992 Semantic change: case studies based on traditional and cognitive
semantics. Journal of Literary Semantics 21: 204-225.
1997 Polysemy: patterns in meaning and patterns in history. Historio-
graphia Linguistica 24(3): 359-385.
200 I Ambiguities we live by. Towards a pragmatics of polysemy. Journal
of Pragmatics 33: 1-20.
Nerlich, Brigitte, Zazie Todd and David D. Clarke
1998 The function of polysemous jokes and riddles in lexical development.
Cahiers de Psychologie Cognitive/Current Psychology of Cognition
17(2): 343-366.
Norrick, Neal R.
1981 Semiotic Principles in Semantic Theory. Amsterdam: John Benjamins.
Norvig, Peter and George Lakoff
1987 Taking: a study in lexical network theory. Proceedings of the 13th
Annual Meeting of the Berkeley Linguistics Society, 185-206.
Berkeley: Berkeley Linguistics Society.
Nunberg, Geoffrey
1979 The non-uniqueness of semantic solutions: polysemy. Linguistics and
Philosophy 3: 143-184.
1992 Systematic polysemy in lexicology and lexicography. In: Hannu
Tommola et al. (eds.), Euralex '92 Proceedings 1/,387-396. Tampere,
Finland.
(http://www.parc.xerox.com/istl/members/nunberglEuralex.html)
Osgood, Charles E.
1980 The cognitive dynamics of synesthesia and metaphor. In: Richard P.
Honeck and Robert R. Hoffman (eds.), Cognitive and Figurative
Language, 203-238. HiIlsdale, NJ: Lawrence Erlbaum Associates.
Ostler, Nicholas and B. T. Sue Atkins
1991 Predictable meaning shift: some linguistic properties of lexical
implication rules. In: lames Pustejovsky and Susan Bergler (eds.),
Lexical Semantics and Knowledge Representation: ACL SIGLEX
Workshop, 87-100. Berkeley, California.
Posner, Roland
1996 Pragmatics. In: Roland Posner, Karl Robering and Thomas A. Sebeok
(eds.), SemiotiklSemiotics. Ein Handbuch zu den zeichentheoretischen
Grundlagen van Natur undKultur/A Handbook on the Sign-Theoretic
Foundations of Nature and Culture. Volume 1, 219-246. Berlin and
New York: Waiter de Gruyter.
Pustejovsky, James
1991 The generative lexicon. Computational Linguistics 17(4): 409-441.
1995 The Generative Lexicon. Cambridge, MA and London, England: The
MIT Press.
 Polysemy and flexibility: introduction and overview 29

Raukko, Jamo
1999 An 'intersubjective' method for cognitive-semantic research on
polysemy: the case of get. In: Masaka Hiraga, Chris Sinha and
Sherman Wilcox (eds.), Cultural, Psychological, and Typological
Issues in Cognitive Linguistics, 87-105. Current Issues in Linguistic
Theory, 152. Amsterdam and Philadelphia: John Benjamins.
Ravin, Yael and Claudia Leacock (eds.)
2000 Polysemy and Ambiguity: Linguistic and Computational Approaches.
Oxford: Oxford University Press.
Rueckl, Jay G.
1995 Ambiguity and connectionist networks - still settling into a solution.
Journal of ExperimentaL PsychoLogy: Learning, Memory and Cognition
21(2): 501-508.
Ruhl, Charles
1989 On Monosemy. New York: SUNY.
Schneider, Edgar
1988 On polysemy in English, considering consider. In: Hiillen and
Schulze (1988), 157-169.
Schulze, Rainer
1988 A short story of down. In: Hiillen and Schulze (1988), 395-414.
1994 Image schemata and the semantics of off In: Schwarz (1994),197-
214.
Schwarz, Monika (ed.)
1994 Kognitive SemantikJCognitive Semantics. Ergebnisse, Probleme,
Pespektiven. Tiibingen: Gunter Narr Verlag.
Sinha, Chris and Tania Kuteva
1994 Spatial and non-spatial uses of prepositions: conceptual integrity
across semantic domains. In: Schwarz (1994), 215-238.
Sweetser, Eve
1990 From Etymology to Pragmatics: Metaphorical and CuLtural Aspects
of Semantic Structure. Cambridge: Cambridge University Press.
Taylor, John R.
1992 How many meanings does a word have? Stellenbosch Papers in
Linguistics 25: 133-168.
Traugott, Elizabeth Closs
1994 Grammaticalization and lexicalization. In: R. E. Asher and J. M. Y.
Simpson (eds.), The Encyclopedia of Language and Linguistics,
Volume 3, 1481-1486. Pergamon Press. Reprinted (1999) in: Keith
Brown and Jim Miller (eds.), Concise Encyclopedia of Grammatical
Categories, Oxford: EIsevier Science.
Tuggy, David
1993 Ambiguity, polysemy, and vagueness. Cognitive Linguistics 4(3):
273-290.
30 Brigitte Nerlich and David D. Clarke

Ullmann, Stephen
1951 The Principles of Semantics. Oxford: Basil Blackwell; Glasgow:
Jackson, Sons and Co.
Victorri, Bernard
1997 La polysemie: un artefact de la linguistique? Revue de Semantique
et Pragmatique 2: 41-62.
Viegas, Evelyne and Victor Raskin
1998 Computational Semantic Lexicon Acquisition - Methodology and
Guidelines. Technical Report MCCS-98-315, CRL, NMSU.
Weinreich, Uriel
1966 Explorations in semantic theory. In: ThomasA. Sebeok (ed.), Current
Trends in Linguistics rn, 395-477. The Hague: Mouton.
Wierzbicka, Anna
1996 Semantics: Primes and Universals. Oxford: Oxford University Press.
Wilks, Yorick
1977 Frames for machine translation. New Scientist 76(108): 802-803.
Williams, John N.
1992 Processing polysemous words in context: evidence for interrelated
meanings. Journal of Psycholinguistic Research 21(3): 193-218.
Wittgenstein, Ludwig
1974 Philosophical Grammar, edited by Rush Rhees; translated by Anthony
Kenny. Oxford: Basil Blackwell.
Wunderli, Peter
1995 Strukturelle Semantik, Polysemie und Architektur der Sprache. Zu
einigen Problemen der Bedeutungsanalyse. In: Ulrich Hoinkes (ed.),
Panorama der Lexikalischen Semantik. Thematische Festschrift aus
Anlaft des 60. Geburtstags von Horst Geckeler, 791-806. Tiibingen:
Narr.
Cognitive models of polysemy

John R. Taylor

l. Introduction

One of the firmest results to have come out of the Cognitive Linguistics
enterprise over the past couple of decades has been the realization that word
meanings need to be understood against broader knowledge configurations,
variously studied as "frames", "scenes", "domains", and "idealized cognitive
models" (Lakoff 1987; Langacker 1987; Croft 1993). Consider, as an illus-
tration, the word bachelor, and the fact that it would be odd to speak of the
Pope as a bachelor, even though the Pope clearly instantiates each of the
four defining features of bachelorhood, namely, "human", "male", "adult",
und "unmarried" (Fill more 1982; Lakoff 1987; Taylor 1995). The problem
arises not because the definition of bachelor in terms of these four features
is wrong in any material sense. Rather, it is because the concept "bachelor"
needs to be understood against an idealized cognitive model of bachelorhood,
und, more generally, of marriage practices. The model prescribes, for both
men and women, a "marriageable age", a stage in life at which a person is
expected to marry. Bachelor is used to designate males who have reached the
marriageable age but fail to marry, because they do not yet want the "commit-
ments" of marriage. The model is "idealized" in the sense that it offers a
simplified view of society and overlooks the many individuals and groups
who do not meet its background assumptions. For example, the model makes
!IO provision for celibate clergy. Given the presuppositions of the cognitive
model, it is quite legitimate to define a bachelor, quite simply, as a man who
husn't married. The application of the word becomes problematic with respect
10 the Pope, because the Pope lies outside the idealized situation covered by
the model.
In this article, I examine some of the cognitive models against which we
understand the concept of polysemy. Such an exercise may be useful, for a
number of reasons. First, it may wean us away from the idea that polysemy
IR a well-defined natural category - a "brute fact" (Searle 1969: 50) about
language, as it were, which, like brute facts of the material and biological
32 John R. Taylor

world, exists independently of our conceptualization of it, and whose essence

can be discovered by the application of the appropriate analytical tools. The
exercise may also suggest that at least some of the problems which we
encounter in the study of polysemy may be traceable to the cognitive models
that frame our conceptualization of the phenomenon. These models en-
capsulate idealizations of language, which might not always be applicable
to the linguistic data at hand. Just as the celibate clergy are not covered by
the cognitive model of bachelorhood, so certain linguistic data may not
fully conform with the cognitive models which frame our understanding of
polysemy.
Polysemy is commonly defined as the association of two or more related
meanings with a single phonological form (Taylor 1995: 99; Cuyckens and
Zawada 200 1: ix). The definition looks straightforward enough, and unlikely,
of itself, to arouse controversy. Nevertheless, problems are likely to be
encountered when we try to apply the definition to any particular set of
data (Taylor, to appear). Note, first of all, that the definition presupposes
that we have procedures in place for reliably identifying, characterizing, and
enumerating the meanings of linguistic units. In case we have reasons for
supposing that a given linguistic form is associated with more than one
meaning, we need to be able to assess whether and in what way the meanings
are related. To appreciate that these are very real issues in polysemy research
we need look no further than the now sizable literature on the lexical item
over (Brugman 1981; Lakoff 1987; Vandeloise 1990; Deane 1993; Dewell
1994; Kreitzer 1997; Queller 2001; Tyler and Evans 2001; amongst others).
Recurring (and to this day, unresolved) issues have involved determining
just how many different meanings this lexical item has, how the meanings
are to be characterized, and the manner in which they are related. The
definition of polysemy also presupposes that we can be confident that the
different (though related) meanings are indeed associated with a single lin-
guistic form. Do we want to recognize the preposition over (as in the lamp
over the tabLe) as the "same" linguistic form as the particle over (as in fall
over), predicative over (The party s over), adverbial over (Do it over again),
the prefixed morpheme in overeat and overjoyed, or even the noun over (as
in cricket) (see Tyler and Evans, this volume). The question also arises, what
are the linguistic forms which are candidates for a semantic analysis? Are
they word-sized units, such as over, multi-word expressions, such as over
here, constructional idioms, such as [V NP over again], or even fully specified
phrases, of the kind It ain't over till the fat lady sings?
 Cognitive models of polysemy 33

In this article, I draw attention to three mutually supporting cognitive

models of language and their relevance to our understanding of polysemy.
These are the semiotic model of language, the building block metaphor of
syntagmatic combination, and the demarcation of lexicon and syntax. The
models conspire to accord special significance to word-sized units. It is words
that are listed in the lexicon, and words are the building blocks which the
syntax combines. Words themselves are viewed as stable associations of
fixed, determinate "chunks" of semantic information with equally fixed and
determinate chunks of phonological material. According to the building block
metaphor, then, complex expressions are a compositional function of their
fixed and stable parts, at both the phonological and the semantic levels;
conversely, complex expressions can be exhaustively broken down into their
constituents. The need to postulate sometimes quite extensive polysemy arises
from the fact that a chunk of phonological material rarely contributes exactly
the same semantic chunk to the various expressions in which it occurs. In
order for compositionality to go through, it becomes necessary to associate
the phonological chunk with a range of distinct meanings, only one of which
is selected on any particular occasion of its use.
The flip-side of sense selection in the compositional process is sense
selection on the part of the hearer. Given that a phonological form may be
associated with different meanings, and that only one of these is involved in
the compositional process, how does a hearer select just that sense that is
involved? Suppose that word w is associated with n different meanings. Any
expression containing w will be in principle n-ways ambiguous. Since most
words may be supposed to be polysemous to some degree, the number of
possible readings of an expression will increase exponentially as the length
of the expression increases. Thus, an expression containing two words, w/
and w 2' which are associated with n/ and n 2 meanings, respectively, will in
principle have n l X n 2 possible readings.
While disambiguation is indeed a major issue in natural language pro-
cessing (Ravin and Leacock 2000: 24-5), for most language users, most of
the time, sense selection in the comprehension process is actually not a
problem at all. Most people, most of the time, are simply unaware of the
extent of the ambiguity generated by polysemy. This rather paradoxical
situation should give us pause for thought. At issue is not the definition of
polysemy as multiple meanings attaching to a single phonological form.
Neither should we rush to denounce the cognitive models presupposed by
the definition and which lead us to postulate extensive polysemy and the
34 John R. Taylor

ensuing ambiguities. Rather, it could be the case that the models are applicable
only to highly idealized data. Polysemy presents itself as a problematic
concept in those cases where the language data do not fit the idealized models.
In the following sections, I discuss the models in more detail, drawing
attention to some discrepancies between certain kinds of data and the ideal
situation presupposed by the models.

2. The semiotic model

According to the semiotic model, a language is a set of linguistic signs

(prototypically: words), each of which associates a phonological structure
with a semantic structure. The model was the foundation for Saussure's (1964
[1915]) theory, where the basic unit of language was the "linguistic sign",
whose essence resided in the association of a "concept", or signified, with
an "acoustic image", or signifier. As Saussure was careful to point out, neither
the concept nor the acoustic image were to be identified with the specific
details, whether conceptual or phonetic, of an actual utterance. Both sides of
the linguistic sign are mental entities which serve to categorize specific
instances that occur in an act of speech. The semiotic model, in its men-
talistic form proposed by Saussure, has been endorsed by many linguists,
including Sapir (1921), Chomsky (1988), Pinker (1994), as well as Lakoff
(1987) and Langacker (1987). Thus, the basic unit in Langacker's theory is
the "symbolic unit", which associates a phonological representation with a
semantic representation.
The semiotic model invites us to imagine an ideal semiotic system, in
which each signifier is paired off with a unique signified (Taylor 2003). The
ideal is captured by the well-known slogan "one form, one meaning". Degrees
of increasing deviation from the one form, one meaning situation can,
however, be identified:

(i) Both pronunciation and conceptual content are liable to vary according
to context of use. This kind of variation need not of itself be prob-
lematic for the semiotic ideal. As noted, both poles of the linguistic
sign, the "concept" and the "acoustic image", are taken to be mental
entities, representations which are schematic for the specific semantic
and phonological values manifest in an act of speech. Depending on
speech tempo, degree of stress, and perhaps other factors, upper may
 Cognitive models of polysemy 35

be articulated as [AP;:'], with a bilabial stop, or [AI}>;:,], with a bilabial

fricative. These different pronunciations can still be regarded as in-
stances of the schematic phonological form IAp;:,/. Likewise, whether
eat is used of eating a steak or eating an ice cream - two very different
activities, if we focus on their microstructure - we may still regard
the two activities as instances of a more schematically characterized
semantic unit [EAT].
(ii) As the degree of phonological and semantic variation increases, it
becomes increasingly difficult to bring the variants under a single
schematic representation. While eating a steak and eating an ice cream
arguably instantiate the same schematic process (having to do with
the ingestion of solid food), we may be less confident about bringing
these and other uses of eat (as when we say that the acid eats away
the metal, or that inflation eats up my savings) under a single
schematic representation. In such a situation, we may be inclined to
associate eat with more than one semantic value; that is to say, we
regard the item as polysemous. A comparable situation may obtain
with respect to variation in an item's phonological form. This phen-
omenon is sometimes discussed under the heading of "alternation"
(though "polyphony" might be a better label, one, moreover, which
brings out the parallels between variation at the semantic pole and
variation at the phonological pole). Thus, the indefinite article can
appear as [;:,], [;:,n], [re], [ren], or [El]; the morpheme (serene) has a
different phonological shape in the adjective serene and the noun
serenity, to cite just two examples.
(iii) The semantic values associated with a phonological form may be so
dissimilar that they are not perceived to be related at all, as with ball
"spherical object" or "social event"; this constitutes homonymy.
Alternatively, a semantic unit is associated with two or more unrelated
phonological forms, such as [8;:,uf;:,] and [kautS]; this constitutes
synonymy. These extreme deviations from the semiotic ideal are
perhaps best regarded as cases of two or more different linguistic
units happening to share the same phonological form, or, alternatively,
the same conceptual content.

The three degrees of deviation from the semiotic ideal constitute a con-
linuum, with borderline cases often difficult to classify as, say, polysemy vs.
context-conditioned variation, or, alternatively, as polysemy vs. homonymy.
36 John R. Taylor

Leaving aside such demarcation issues, we should expect that deviations

from the semiotic ideal as characterized under (ii) and (iii) will tend to
undermine the communicative efficiency of a language. Through the associ-
ation of more than one signified with a signifier, polysemy (and homonymy)
will generate ambiguity, which, if not properly resolved, may lead to com-
munication failure. To be sure, the "cost" of polysemy may be offset by
other factors, whose role may actually enhance the semiotic potential of a
language. The number of established phonological forms in a language is
going to be smaller by far than the number of conceptual categories that a
person may wish to designate. Moreover, in view of the changing environ-
ment and changing concerns of language users, conceptual categories, and
what count as members of these categories, are liable to undergo modification
over time. The possibility that new meanings can accrue to existing word
forms renders a polysemy-tolerant language more viable than a rigidly iso-
morphous system, in which each signifier is associated with a unique semantic
representation, and vice versa (Geeraerts 1985). It is worth noting that while
Langacker, on the opening page of his Foundations of Cognitive Grammar
(Langacker 1987: 11), is able to endorse the Saussurean conception of the
linguistic sign, he is also able to maintain (Langacker 1988: 50) that polysemy
constitutes the normal, expected state of affairs in lexical semantics. Indeed,
scholars working in the Cognitive Linguistics tradition have often been
censured for promoting a "rampant polysemy" (Cuyckens and Zawada 2001:
xv), that is, a readiness to associate linguistic forms with an ever-increasing
number of distinct senses.
Other linguists have tried to remain true to the semiotic ideal, seeking
to minimize the role of poly se my in natural languages (often, though, at the
expense of having to postulate additional processes, or levels of representation,
in order to account for the range of meaning variation encountered in actual
usage). Thus, for Coseriu (1977), polysemes are a matter of usage norms,
distinct from the unitary values that constitute the language system. For
Kirsner (1993), words are "invariant signals of invariant meaning" (p. 85);
polysemy effects arise through the use which speakers make of these linguistic
signs. Wunderlich (1993) claims that polysemy, so called, is merely an effect
of "conceptual" elaboration of unitary (i.e. non-polysemous) "semantic" rep-
resentations. Van der Leek (2000) takes a similar line, arguing that words
designate "Platonic" (that is, highly general and abstract) concepts, which
participate in, but do not uniquely determine, context-specific construals.
Concerning phonological representations, it will be recalled that one of the
 Cognitive models of polysemy 37

drivers of early generative phonology (Chomsky and Halle 1968) was the
desire to associate each morpheme of a language with a single, unique
"underlying form", from which all the surface altemants could be derived by
rule.
Since variation is not limited to the semantic pole, the possibility arises
that deviations from the one meaning ideal may be accompanied by deviations
from the one form ideal. Langacker (1987: 398) reports that in his speech,
route is associated with two distinct phonological shapes, [raut] and [ru:t],
both pronunciations being possible for the noun, but only the latter for the
verb. Orthography suggests that we are here dealing with a single linguistic
unit (though the matter is indeed moot). Sometimes, however, orthography
is also subject to variation. (Variation at the orthographic level might be
referred to as "polygraphy".) Speakers of British-based varieties of English
often distinguish, in writing, between "program" and "programme", and
between "disk" and "disc". To be sure, the written variants may not be
differentiated in pronunciation; they are, however, likely to be differentiated
semantically, "program" and "disk" being reserved for computer-related uses.
It is again a moot point whether we are here dealing with a single linguistic
sign, unified at the phonological pole, or with two different signs differen-
tiated both semantically and orthographically (Love, to appear).

3. The building block metaphor

The semiotic model provides us with a set of linguistic signs, each of which
associates a meaning with a phonological form. According to the building
block metaphor, complex expressions are formed by assembling these smaller
units; conversely, complex expressions can be exhaustively segmented into
their component parts. The metaphor requires that each constituent building
block have a fixed and determinate semantic and phonological content which
it contributes to the whole. Hence, the meaning of a complex expression will
be a function of the meanings of its constituent parts, just as the phonological
form of a complex expression will result from the alignment of the phonol-
ogical forms of its constituents.
The metaphor is pervasive in our deliberations on language and its
structure; indeed, as Langacker (1991: 186) observes, the metaphor may be
unavoidable "for expository purposes", as when we introduce beginning
linguistics students to the concept of the morpheme. Yet, paradoxically, it is
38 John R. Taylor

in the realm of morphology where the metaphor fails most conspicuously.

Take the word butcher. Butcher, like writer, appears to contain the agentive
suffix -er. The semantics fit, as well. Both butcher and writer characterize a
person in terms of what they do. Yet there is no base verb (to) butch to
which the -er morpheme attaches. Butcher cannot be exhaustively analysed
into its constituent building blocks, even though the word would appear to
be morphologically complex.
When it comes to analysing phrasal expressions, problems of segmentation
are less likely to arise. (But consider the case of opaque idioms, such as kick
the bucket, I as well as some examples to be discussed in the next section.)
Applying the building block metaphor to complex expressions, however, is
likely to result in an explosion of polysemes. Take, as an example, the
adjective old, and its use in expressions such as old man, old friend, and oLd
student (of mine). In these expressions, oLd does not modify the associated
noun in exactly the same way. An "old man" is a man who is advanced in
years; an "old friend" is a person who has been a friend (to someone) for a
long time (but who need not be advanced in years); an "old student (of mine)"
is a person who used to be a student of mine but who no longer is my student.
In order for the meanings of these Adj-N combinations to be built up comp-
ositionally, we need to postulate three distinct senses of old, only one of
which is selected for the compositional process. On the other hand, if oLd is
indeed polysemous in this way, oLdfriend ought to be three-ways ambiguous,
with the readings "aged friend", "friend of long standing", and "former
friend". This prediction seems not to be entirely correct. If I refer to someone
as "an old friend", only the second of these readings is likely to be appropriate.
The example illustrates the fact that a word of a given semantic type may
force a specific reading of a word with which it combines. Thus, old friend
is likely to be interpreted differently from old man because of the semantic
structure of friend (Taylor 1992). The matter was addressed by Pustejovsky
(1991), in his study of the ways in which words can mutually "coerce" their
readings. Begin a novel forces a particular reading of novel, namely, "novel-
as-text". Contemporaneously, this reading of noveL triggers an enriched
interpretation of begin, namely, "begin to read" (see Warren, this volume).
We can imagine other mutually coercing interpretations, as when talking of
a writer, a typesetter, a translator, a bookbinder, or a book-devouring insect
"beginning a novel". Given these coercion effects, we might hesitate to claim
that novel, or indeed begin, is polysemous. We might also begin to question
the assumption that word meanings are fixed and determinate entities. For
 Cognitive models of polysemy 39

example, how might we characterize the concept that gets coerced into the
specific readings "novel-as-text" and "novel-as-object"?
A more general failing of the building block metaphor is that it ignores
the possibility that the whole may be organized in ways which go beyond, or
which are even at variance with, the properties which are contributed by the
parts. Take Langacker's example of a football under the table (Langacker
1987: 279-280). The (literal) interpretation of football under the table goes
well beyond the compositionally derived meaning of the expression. For
example, we would probably picture the football as lying on the floor within
an area circumscribed by the table's legs. It is not only that the specifics of
this scene are not encoded in the expression; the football, if we consider
the matter carefully, is not actually "under" the table at all- the football, that
is, is not located at a place that is lower than the place occupied by the
table. For the football to be "under" the table, it would have to be under
the floor on which the table is standing! Conceptualization of the global
scene has forced an elaboration and readjustment of the compositionally
derived meaning (Taylor 2002). The matter is equally evident with respect to
phonological structure. Typically, did you would have the pronunciation
[ldI.d3U], rather than the "compositional" pronunciation [dld.ju:]. It is not
just that the boundary segments of the building blocks [dId] and [ju:] have
mutually influenced each other. Rather, the complex expression has been
organized in terms of syllable and foot structure. The component syllables of
[ldI.d3U] do not match up with the units contributed by the component
structures; moreover, the two syllables have been structured in terms of a
strong-weak relation within a trochaic foot.
The interplay of word meanings (that is, the semantic units which words,
in their status as building blocks, contribute to the whole) and the meanings
of larger expressions in which the words occur, has come to the fore in studies
of constructions. Consider Goldberg's (1995: 29) often-cited example of a
person "sneezing the napkin off the table". Sneeze does not plausibly belong
in the class of caused-motion verbs (put, push, throw, etc.). The caused-motion
sense of sneeze - "cause (a thing) to go (to a place) by sneezing on it" - is
contributed by the syntactic construction [V NP PP] in which it occurs. The
primacy of the construction over its parts is supported by the fact that in
different instantiations of the construction the meanings of the parts are
mapped in different ways onto the meanings of the complex expressions
(Mandelblit and Fauconnier 2000). Thus, in sneeze the napkin off the table,
the verb designates the causing activity, whereas in trot the horse into the
40 John R. Taylor

stabLe, the verb designates the caused activity. Appeal to constructions, then,
may again lessen the need to postulate extensive polysemy at the level of
words. On the other hand, polysemy is liable to re-emerge at the level of
constructions. As argued by Goldberg (1995), the caused-motion construction
itself has a number of distinct variants, each of which severely restricts the
range of items which are eligible to occur in it.

4. The lexicon and the syntax

Language description is typically couched in terms of a lexicon (which lists

the words and their fixed and conventionalized phonological and semantic
properties) and a syntax (which lists the rules for the combination of words).
Thus, when we study a foreign language, the essential tools are the dictionary
and the grammar book. Recent interest in idioms, however, has put in question
the validity of this modular view (Jackendoff 1997; Taylor 2002). At issue
are not just semantically opaque idioms of the kind kick the bucket, which
clearly have to be learned as such and which cannot be assembled in the
syntax, but the vast inventory of collocations, fixed expressions, formulaic
expressions, and constructional idioms.
Consider, for example, Queller's (2001) recent discussion of one use (or,
rather, set of uses) of over, concerning the expression all over. In water all
over the floor, over appears to have the covering sense exemplified in the
tablecloth over the table, while all appears to have its standard quantifying
sense. The compositional sense of water all over the floor, therefore, would
be that there is water covering all of the floor. The problem with this account
is that "total coverage" does not capture the distinctive semantic value of all
over. We should not say of a table cloth, for example, that it is "all over the
table", even though the cloth might be completely covering the table (or, at
least, the table top). The relevant sense of all over, according to Queller, is
not so much "total coverage", as "chaotic dispersal". It is for this reason that
one might say that there are blood stains all over the table cloth, but not that
there are red squares all over the cloth (the red squares being inherent to the
design of the table cloth, whereas the blood stains are not). All over turns out
to be an idiomatic expression, which needs to be separately listed as such in
a speaker's mental grammar. Although we can identify its parts (at least, on
the phonological level), and perhaps see their relevance to the composite
meaning, the expression is not a compositional function of its parts. There
 Cognitive models of polysemy 41

could be little point, therefore, in enquiring which of the various senses of

over is instantiated in all over.
Not all uses of all over exemplify the chaotic dispersal sense. Consider
the expression He had guilt written all over his face, in contrast to the odd-
ness of ?He had satisfaction written all over this face. At issue here is the
uncontrolled, or involuntary display of an emotion which interferes with a
person's desire to present a front of composure. We might want to say that
all over, in its status as an idiomatic phrase, is polysemous. But this may
not be the appropriate solution. Note that the "uncontrolled display" sense of
all over is associated with a specific syntactic frame, namely, [NP have
EMOTION written all over NP's face]. We are dealing here with a phrasal
idiom, some of whose components are lexically specified, the other "slots"
in the idiom being able to be filled by any items meeting the idiom's semantic
specifications. Again, it could be fruitless to enquire which of its various
senses over, or indeed all over, is contributing to the idiom. Similar remarks
apply to lexically fully specified expressions involving all over, as when we
tell a student that his essay is "all over the place", or say of an experiment
that the results are "all over the map".
A comparable situation in the phonological domain is discussed by Bybee
and Scheibman (1999). Don't may be subject to phonological reduction, even
emerging, in I don't know, as [f::l]. Although exemplifying very general
processes of reduction which are operative in English, the highly reduced
form of don't tends to be restricted to the fixed expression I don't know,
particularly when the expression is used, not to convey the subject referent's
ignorance of something (the "compositional" meaning of don't know), but to
soften a speaker's assertion, or to convey the speaker's polite disagreement.
Bybee (2001) emphasizes that the conventionalized pronunciation of I don't
know (along with its specialized meaning) cannot be regarded as a com-
positional function of the properties of its parts. It is the whole expression I
don't know, in both its phonological and semantic aspects, which needs to be
listed in a person's mental grammar.

5. Beyond the models

Although I have focused on some of their problematic aspects, there is a sense

in which each of the three models surveyed above is true (or at least, not
obviously false). It is entirely legitimate to regard a language as a semiotic
42 John R. Tay/or

system, that is, as a means for associating meanings (or conceptualizations)

with phonological (and graphological) forms, the task of the linguist being to
identify and to describe the conventionalized associations of form and mean-
ing which the language makes available to its speakers. The building block
metaphor also captures an important ingredient of language. It is uncontro-
versial that speakers do construct new expressions by assembling their smaller
parts in new and creative ways, and that hearers, on the whole, are able to
interpret these in the sense intended by the speakers. It is equally evident that
the creation and interpretation of new utterances is based on knowledge of
general patterns of syntagmatic combination which operate over smaller-sized
units. To this extent, also, the lexicon-syntax split is fully justified.
Taken together, the cognitive models entail the following idealization of
a language:

(i) A language can be analysed into a lexicon and a syntax. The lexicon
lists the words, while the syntax contains general rules for the com-
bination of words.
(ii) Each word associates a fixed and determinate semantic structure with
a fixed and determinate phonological structure.
(iii) The semantic and phonological properties of a complex expression,
as assembled by the syntax, is a compositional function of the semantic
and phonological properties of the component words.

Given the above idealizations, polysemy can be straightforwardly and

unproblematically defined in terms of a deviation from idealization (ii).
Polysemy, namely, consists in the association of a single word form with
two or more semantic representations. The need to postulate polysemy fol-
lows from idealization (iii). In order to be able to derive the compositional
meanings of the complex expressions in which it occurs, it may be necessary
for a word to be associated with a range of different meanings, only one of
which is selected for a particular expression.
As the discussion in this chapter will have demonstrated, the above idealiz-
ations may not be directly applicable to actual language data. The following
factors, in particular, distort the ideal situation:

(i) Complex expressions may indeed be segmented into smaller units.

The smaller units are observed to recur in different expressions, and
they may well be associated with fairly stable phonological and se m-
 Cognitive models of polysemy 43

antic values. However, these units need not correspond to words, as

traditionally defined. The units may comprise multi-word units of
various sizes: typical collocations, fixed turns of phrase, formulas,
cliches, and all manner of idioms. In the limiting case, the expression
itself has unit status.
(ii) Speakers of a language certainly extract patterns which sanction the
combination of smaller units. However, these patterns need not be
restricted to the very general phrase structure rules of traditional
syntax. They may comprise rather low-level generalizations, which
are able to sanction only a small number of instances, or they may
have the status of constructional schemas, whose slots can be filled
only by items meeting the semantic specifications of the construc-
tion. Sometimes, one or more of the slots may be lexically specified,
that is, the construction involves the occurrence of particular words.
Indeed, knowledge of a word may involve knowledge of the construc-
tions in which it can occur. In the limiting case, the construction may
be lexically fully specified.
(iii) Speakers draw on the resources of their language in order to give
expression to particular conceptualizations. Mostly, the available res-
ources allow only a partial and imperfect representation ofthe desired
semantic content. Consequently, the intended meaning of an expres-
sion may go beyond, or may even be at variance with, the meaning
that is contributed by the parts, in association with the relevant con-
structional schemas.

Given the extent and the ubiquity of these deviations from the idealiz-
at ions, it is not surprising that polysemy (understood as the association of
related meanings to a single linguistic form) should be such a problematic
concept, and one which has engendered so much discussion and controversy.
It was not my intention, in this article, to attempt to resolve these con-
troversies. The above discussion may, however, be useful, to the extent that
it may throw some light on why the seemingly so straightforward notion of
polysemy is so problematic. It also suggests that it may be useful to approach
the problematic aspects of polysemy, not so much by a refinement of the
concept itself, but by addressing the cognitive models which frame our
understanding of it. 2
44 John R. Tay/or

Notes

I. The point here is that. although kick the bucket can be segmented into words.
and can even be regarded as a perfectly regular verb phrase. the (idiomatic)
meaning of the expression cannot be distributed over the meanings of its parts.
2. Given my somewhat critical account of the cognitive models discussed in this
chapter. it may be asked whether there are other models of language which enable
us more accurately to conceptualize the object of our study. Here are two sugges-
tions. which I offer without further discussion. but which I think may be worthy
of further investigation:
(i) The corpus model. Instead of partitioning the facts of a language into a
lexicon and a syntax (a dictionary and a grammar book), it may be worth
exploring the metaphor of a language as a giant corpus, and to construe
knowledge of a language in terms of a mental corpus, comprising memory
traces of previously encountered utterances. Words. with their phonological
and semantic properties, would then emerge as the summed activation of
already encountered instances.
(ii) The cut-and-paste model. Instead of studying the production of complex
expressions in terms of the assembly of smaller units in accordance with the
rules of syntax, it may be more appropriate to view language production as
a process of sticking together various bits and pieces retrieved from the
mental corpus.
Interestingly, neither model has a place for polysemy, as traditionally understood.

References

Brugman. Claudia
1981 Story of over. M.A. thesis. University of California, Berkeley.
Bybee, Joan
2001 Phonology and Language Use. Cambridge: Cambridge University
Press.
Bybee. Joan and Joanne Scheibman
1999 The effect of usage on degrees of constituency: the reduction of don't
in English. Linguistics 37: 575-596.
Chomsky, Noam
1988 Language and Problems of Knowledge. Cambridge, MA: Massachusetts
Institute of Technology Press.
Chomsky, Noam and Morris Halle
1968 The Sound Pattern of English. New York: Harper and Row.
 Cognitive models of polysemy 45

Coseriu, Eugenio
1977 L'Etude fonctionnelle du vocabulaire: precis de lexematique. Cahiers
de iexicoiogie 29: 5-23.
Croft, William
1993 The role of domains in the interpretation of metaphors and metonymies.
Cognitive Linguistics 4: 335-370.
Cuyckens, Hubert and B. Zawada (eds.)
2001 Polysemy in Cognitive Linguistics. Amsterdam: Benjamins.
Deane, Paul
1993 Multimodal spatial representations: on the semantic unity of "over"
and other polysemous prepositions. Duisburg: Linguistic Agency,
University of Duisburg.
Dewell, Robert B.
1994 Over again: image-schematic transformations in semantic analysis.
Cognitive Linguistics 5, 351-380.
Fillmore, Charles
1982 Frame semantics. In: Linguistic Society of Korea (ed.), Linguistics
in the Morning Calm, 111-138. Seoul: Hanshin.
Foolen, Ad and Frederike van der Leek (eds.)
2000 Constructions in Cognitive Grammar. Amsterdam: Benjamins.
Geeraerts, Dirk
1985 Cognitive restrictions on the structure of semantic change. In: Jacek
Fisiak (ed.), Historical Semantics, ]27-153. Berlin: Mouton de
Gruyter.
Goldberg, Adele
1995 Constructions: A Construction Grammar Approach to Argument
Structure. Chicago, IL: University of Chicago Press.
lackendoff, Ray
1997 The Architecture of the Language Faculty. Cambridge, MA:
Massachusetts Institute of Technology Press.
Kirsner, Robert S.
1993 From meaning to message in two theories: cognitive and Saussurean
views of the Modern Dutch demonstratives. In: Richard Geiger and
Brygida Rudzka-Ostyn (eds.), Conceptualizations and Mental
Processing in Language, 81-114. Berlin: Mouton de Gruyter.
Kreitzer, Anatol
1997 Multiple levels of schematization: a study in the conceptualization
of space. Cognitive Linguistics 8: 291-325.
Langacker, Ronald
1987 Foundations of Cognitive Grammar, Volume 1. Stanford, CA: Stanford
University Press.
1988 A view of linguistic semantics. In: Brygida Rudzka-Ostyn (ed.),
Topics in Cognitive Linguistics, 49-90. Amsterdam: Benjamins.
46 John R. Taylor

1991 Foundations of Cognitive Grammar, Volume 2. Stanford, CA: Stanford

University Press.
Lakoff, George
1987 Women. Fire. and Dangerous Things: What Categories Reveal about
the Mind. Chicago, IL: University of Chicago Press.
Love, Nigel
(to appear) Rethinking the fundamental assumption of linguistics. In: Hayley G.
Davis and Talbot J. Taylor (eds.), Rethinking Linguistics. Richmond:
Curzon Press.
Mandelblit, Nili and Gilles Fauconnier
2000 How I got myself arrested: underspecificity in grammatical blends
as a source for constructional ambiguity. In: Foolen and van der Leek
(2000), 167-189.
Pinker. Steven
1994 The Language Instinct. New York: William Morrow.
Pustejovsky, lames
1991 The generative lexicon. Computational Linguistics 17: 409-441.
Queller. Kurt
200 I A usage-based approach to modeling and teaching the phrasal lexicon.
In: Martin Piitz, Suzanne Niemeier and Rem~ Dirven (eds.), Applied
Cognitive Linguistics, Volume 2,55-83. Berlin: Mouton de Gruyter.
Ravin. Yael and Claudia Leacock (eds.)
2000 Polysemy: Theoretical and Computational Approaches. Oxford:
Oxford University Press.
Sapir. Edward
1921 Language. New York: Harcourt, Brace and World.
de Saussure, Ferdinand
1964 [1915] Cours de linguistique generale. Paris: Payot.
Searle, John R.
1969 Speech Acts: An Essay in the Philosophy of Language. Cambridge:
Cambridge University Press.
Taylor, John R.
1992 Old problems: adjectives in Cognitive Grammar. Cognitive Linguistics
3: 1-46.
1995 Linguistic Categorization. Oxford: Oxford University Press.
2002 Cognitive Grammar: An Introduction. Oxford: Oxford University
Press.
2003 Near synonyms as coextensive categories: "high" and "tall" revisited.
Language Sciences 25: 263-284.
(to appear) Polysemy's paradoxes. Language Sciences.
Tyler, Andrea and Vyvyan Evans
200 I Reconsidering prepositional polysemy networks: the case of over.
Language 77: 724-765. (Reprinted in this volume.)
 Cognitive models of polysemy 47

Vandeloise, Claude
1990 Representation, prototypes, and centrality. In: Savas Tsohatzidis (ed.),
Meanings and Prototypes, 403-437. London: Routledge.
Van der Leek, Frederike
2000 Caused-motion and the "bottom-up" role of grammar. In: Foolen and
van der Leek (2000), 301-331.
WunderIich, Dieter
1993 On German um: semantic and conceptual aspects. Linguistics 31:
II 1-133.
Polysemy: past and present

Brigitte Nerlich

1. Introduction

It is fairly well known that research into polysemy began with the work of
the French semanticist Michel Breal in the late 19th century.· It is less well
known that the "multiplicity of meaning", with or without the label "polysemy",
was quite well researched by students of literature and by lexicographers
well before Breal. It should come as no surprise that the major pathways of
polysemous sense extension, namely metaphor, metonymy and synecdoche
have also been on the philosophical and linguistic agenda for a long time.
To fully understand the import of the concept of polysemy for contem-
porary semantic thought, it is useful to reconstruct some of the contexts in
which it appeared for the first time. We hope that such a reconstruction may
lead to a new evaluation of contemporary concepts and theories. However,
one should not expect this work to be definitive, because the field will
undoubtedly change under the pressure of new historical discoveries: the
historical survey modifying the present status of polysemy research, this latter
in turn influencing the future of the former (see Eco 1996: 744).
The first part of this survey will trace the development of various research
traditions analysing the pathways of polysemous sense extension before the
term "polysemy" was introduced into general linguistics by Breal in 1887.
This means, effectively, looking at research done in the wide field of rhetoric,
in the philosophy of metaphor, in etymology and in lexicography, where it
was essential to find elegant and plausible ways to write lexical entries for
words with multiple meanings. The second part of this survey will be devoted
to Breal's more revolutionary treatment of polysemy and the research leading
up to it. In the last part we shall give a brief and necessarily incomplete
survey of post-Brealian research into polysemy.2
50 Brigitte Nerlich

2. Metaphor, metonymy, and synecdoche as pathways of polysemous

sense extension

Metaphor has had three relatively unrelated waves of fame in the 20th
century: the first brought about by Ivor A. Richards, Kenneth Burke, and
Max Black in their reflections on metaphor published between 1936 and
1962 (Richards 1936; Burke 1962; Black 1962); the second triggered by
Roman lakobson's papers "Two aspects of language and two types of aphasic
disturbances" and ''The metaphoric and metonymic poles". published in the
1950s (Jakobson 1983 [1956b], 1956a);3 the third, more recent one, was
unleashed by the George Lakoff and Mark 10hnson book Metaphors We Live
By in the 1980s (Lakoff and 10hnson 1980; see also Lakoff and Thrner 1989;
Lakoff and 10hnson 1999). Metaphors We Live By has become a standard
text for those interested in what has now become a new paradigm in ling-
uistics, namely cognitive semantics. One of the basic claims of this new
research paradigm is that "a lexical item is typically polysemic - comprising
a family of interrelated senses, forming a network centred on a prototypical
value. Although the precise array of senses conventionally associated with
the expression is not fully predictable, neither is it arbitrary - as the net-
work evolves from the prototype, each extension is MOTIVATED in some
cognitively natural fashion, and often in accordance with a general pattern
or principle" (Langacker 1988: 392). Some of the most important patterns or
principles are metaphor, metonymy and synecdoche.
Although well aware that modem reflections on metaphor have their roots
in Aristotle, and in Greek and Latin rhetoric, and that they have parallels in
early 20th-century developments, such as the work done by Burke or Black
(see Turner 1999), only a small number of cognitive linguists seem to ap-
preciate just how much research into metaphor, as well as metonymy and
synecdoche, was done during the 19th and early 20th centuries in the fields
of rhetoric, lexical semantics and philosophy (see Smith 1982; lakel 1999;
Nerlich 1998; Nerlich and Clarke 1997,1999, 2oo0a, 2oo0b, 2001). These
figures of speech were gradually discovered as not only having an aesthetic,
but also an essentially cognitive function.
The foundations for this fundamentally new research into metaphor and
other core figures of speech were laid in the 17th and 18th centuries with the
work of Giambattista Vico, John Locke, lohann Gottfried Leibniz and lohann
Heinrich Lambert amongst the philosophers, and Cesar Chesneau Du Marsais
amongst the grammarians.
 Polysemy: past and present 51

Like many of his philosophical colleagues, Locke had still been sceptical
about the value of metaphors, as they seemed to obscure the direct link
between words and ideas, but at the same time he saw that metaphors were
unavoidable and deeply rooted in our language and thought. He gave the
following examples for the metaphorical grounding of mental concepts:

It may also lead us a little towards the Original of all our Notions and
Knowledge, if we remark, how great a Dependence our Words have on
common sensible Ideas; and how those, which are made use of to stand for
Actions and Notions quite removed from sense, have their rise from thence,
and from obvious sensible Ideas are transferred to more abstruse Significations
... v.g. to Imagine, Apprehend, Comprehend, Adhere, Conceive, Instill, Disgust,
Disturbance, TranquiJity, andc. are all Words taken from the Operations of
sensible Things, and applied to certain modes of Thinking. (Locke 1975
[1689]: Ill, i: 5)

Locke's most quoted example was perhaps that the Latin word for "mind",
that is, spiritus, had its origin in the word for "breath".4
Leibniz, who took up this example, approached the problem not only
from a philosophical angle, but also from a decidedly etymological one
and came to the conclusion that language was in fact permeated by meta-
phors, a view that became immediately popular. The German linguist Johann
Christoph Adelung, for example, pointed out almost a century after Locke,
in his 1798 entry on "metaphor" for his grammatical critical dictionary: "In
point of fact most of our words are metaphors. The word spirit when desig-
nating a spiritual rational entity is a metaphor, as it actually means breath"
(Adelung 1798: 192).5
In his Nouveaux essais sur l'entendement humain published in 1765
(Leibniz 1981), a reply to Locke's Essay on human understanding (1975
11689]), Leibniz also stressed the importance of tropes, especially of metaphor,
metonymy and synecdoche for the changes in meanings that one can observe
everywhere (see also Leibniz 1882, 1956).
A contemporary of Leibniz, the French philosopher and etymologist Anne
Robert Jacques Turgot, continued the reflection on how figures of speech
can bring about changes in the meanings of words. But he also pointed out,
in a thoroughly modern way, that for these figures to work and these changes
la occur, more than just "rhetoric" is needed. The use and uptake of figures
of speech depends on our ways of conceptualizing the world, of seeing the
world in certain ways/' What one would nowadays call "polysemization" is
52 Brigitte Nerlich

for him the result of the way in which the principal idea of a word spills over
onto the secondary idea. Then the word is used in its new extension for other
ideas, based solely on the secondary meaning without regard for the primitive
one - as when we say un chevalferre d'argent (a horse "ironed" with silver).
New metaphors are coined with this new meaning in mind, and then one on
top if the other, until we reach a point when a meaning emerges which is
completely opposed to the original meaning of the word (Turgot 1756: 105b).
Du Marsais, like Turgot a contributor to Diderot's Encyclopedie, devel-
oped a rhetoric of ordinary language use in which he studied the multiplication
of meanings as well as the forgetting of the original meaning, two processes
that go hand in hand. As early as 1730 he claimed that catachresis, or the use
of a word in a context that differs from its proper application, was funda-
mental to the extension and multiplication of meanings, and he rejected the
widespread view that catachresis was nothing but the "incorrect" (mis- or
ab-) use of a word (see Nerlich 2000). He illustrates this point of view with
the same example that Turgot used some time later.

... it is the ordinary costume to nail iron horseshoes under the feet of horses,
something we call "ferrer un cheval": ... if it so happens that instead of iron
we use silver we say that "les chevaux sont ferre d' argent" instead of inventing
an entirely new word which nobody would understand .... And so "ferrer"
signifies by extension shoeing a horse with silver instead of iron. (Du Marsais
1977 [1730]: 45)

Another contemporary German philosopher, Lambert (1764), was even

more keenly aware than Leibniz and Turgot of the ubiquity of metaphors
and their importance for language and thought. And like them he was inter-
ested in the process of polysemization.

It has however been known for a long time that we compare the visible with
the invisible, the world of the body with the world of the spirit, sensations
with thoughts, and that we use the same words for either. Hence words
necessarily acquire a double and sometimes a multiple meaning. Having light
in the room and a light in one's thoughts are examples for this way of
speaking. (Lambert 1764: Aleth. I., §45)

Metaphors are central to Lambert's "semantic tectonics" (see Ungeheuer

1980). One quote from Lambert's Neues Organon (1764) has to suffice
here:
 Polysemy: past and present 53

... all the words of a language could be divided into three classes, the first of
which doesn't need any definition, as one points to the thing itself in its
totality, and as one can make an immediate connection between word, concept,
and thing. Another class uses the words of the first class in a metaphorical
way and uses a tertium comparationis instead of a definition. The third class
comprises those words which have to be defined, and this insofar as one uses
the words of the first two classes to do so. The words of the third class thus
defined can again be used for definitions. It is obvious that the words of the
third class can also be used metaphorically and that most of them are used in
this way already. (Lambert 1764: 11, ii: 45)

Lambert tried to provide a solution to a problem that still haunts modem

linguists, namely the question: where does meaning come from? For Lam-
bert, it arises from ostensive definitions,? on which metaphors then build the
semantic structure of language. Modern cognitive semanticists would rather
stress the importance of perception, "preconceptual structures" and image
schemata, as such a foundation. They have discovered the human body as
the main source of "meaning".
Embodiment has become one of the most central tenets of cognitive
semantics. Meaning is seen as grounded in the nature of our bodies and in
our interaction with the physical, social and cultural environment we live in.
This means that concepts are grounded in our bodily experience and then
elaborated by structures of imagination. such as metaphor and metonymy.
This unity of the inner and outer, the corporal and the mental, had already
been stressed by Johann Wolfgang von Goethe. According to him we always
spiritualize the corporal and anthropomorphize the mental, and this mainly
through metaphor. The philosopher of metaphor, Alfred Biese, quotes from
Goethe's poem Procemion (Goethe 1816):

So weit das Ohr, so weit das Auge reicht,

Du findest nur Bekanntes, das Ihm gleicht,
Und Deines Geistes Feuerflug
Hat schon am Gleichnis, hat am Bild genug.
(Quoted in Biese 1893: 120)

[As far as ear can reach, or eyesight dim,

Thou findest but the known resembling Him;
How high so'er thy fiery spirit hovers,
Its simile and type it straight discovers.]
(hup:/lwww.everypoet.com!archive!poetry/Goethe/goethe_procemion.htm)
54 Brigitte Nerlich

Goethe very much agreed with the 18th-century Neapolitan philosopher

Vico, who was one of the most eminent thinkers in the field of "metaphor".
Unlike Kant, who only reluctantly accorded imagination and language an
important place in his philosophy (but who corresponded with Lambert about
this problem), Vico had given language and imagination, not reason, a central
status in his conception of the mind. "In all languages," he wrote in his
Scienza Nuova, "the greater part of expressions referring to inanimate objects
are taken by transfer from the human body and its parts, from human senses
and human passions ... Ignorant man makes himself into the yardstick of the
universe" (quoted in Ullmann 1962: 214; see Vico 1948 [1725]).
According to Danesi (1990), Vico saw the relevance of imagination in
the fact that it allows us to understand reality - and more.

For when we wish to give utterance to our understanding of spiritual things,

we must seek aid from our imagination to explain them, and like painters,
form human images of them. (Vico quoted in Danesi 1990: 228)

A last example of the sustained reflection on metaphor and other figures

of speech in the 18th century has to be the famous passage from Du Marsais's
book Des Tropes. In this passage, which prefigures much of modern sem-
antics, Du Marsais makes it quite clear that the use of figures is as necessary
to language as breathing to life. For him there is nothing so natural, ordinary
and common as the use of figures of speech in daily discourse - even that of
fishwives (Du Marsais 1977 [1730])." As Lakoff and Turner say in their book
More Than Cool Reason:

Metaphor is a tool so ordinary that we use it unconsciously and automatically,

with so little effort that we hardly notice it. ... metaphor is an integral part of
our ordinary everyday thought and language .... [It] allows us to understand
ourselves and our world in ways that no other modes of thought can. (Lakoff
and Turner 1989: xi)

In the 19th century the exploration of metaphor and other figures of speech
began to be an albeit small part of the new science of historical linguistics,
which detached it somewhat from the anthropological and philosophical
explorations of the previous centuries. Metaphor, metonymy, generalization
and specialization became part and parcel of the instruments used in histori-
cal semantics used to study and classify types of semantic change and of
meaning extension. This early form of semantics was influenced by rhetoric,
association psychology, and aesthetics (see Nerlich 1992).
 Polysemy: past and present 55

Aristotle had distinguished between various forms of mental association:

by simultaneity, succession, similarity, and contrast (De memoria, 2). In 1739
the Scottish philosopher David Hume had tried to establish laws that govern
the mental world and which would be analogous to the law of gravity so
spectacularly discovered by Newton. He postulated that these laws were laws
of association: by similarity, by contiguity in space or time, and by cause
and effect. This was the beginning of a long-standing research programme in
association psychology, which was only challenged after the advent of Gestalt
psychology at the beginning of the 20th century. In 1829 James Mill sum-
marized the by then well-established laws of association as (i) contiguity in
time; (ii) contiguity in space; (iii) the relations of cause and effect, of means
and end, of part and whole; (iv) contrast or similarity; (v) equal or different
forces directed at an object; (vi) objects which by accident are called by
similarly sounding names. He then reduced these six laws to two: the law of
Simultaneity and the law of Resemblance or Affinity (see Ritter 1972: 550).
Linguists working with the "laws" of rhetoric, such as metaphor and met-
onymy, could easily see the affinity between these laws of rhetoric and the
more scientific "laws of association". They therefore developed a semantics
based on a theory of meaning where meanings were equated with mental
representations and the associations between mental representations, and
where changes in meaning were equated with changes in associations. The
first to formulate this new research programme for historical semantics was
Christian Karl Reisig, a classicist, who gave lectures on the syntax, etymology
and semantics of Latin in about 1829, and stated that:

The basis for the development of ideas in words is the association of thoughts
in the group of representations .... However, certain associations of ideas are
preferred in human representation, and rhetoric has given them certain names,
which however are to some extent also appropriate for semasiology, namely
synecdoche. metonymy. and metaphor. Insofar as these so-called figures aim
at something aesthetic. they certainly belong to rhetoric, even when individuals
make use of them; but insofar as a usage based on these figures of speech has
been established in a particular language, and this is particular to the nation,
these figures should be dealt with here. (Reisig 1890 [1839]: 2; emphasis
added)

Quite independently of these advances in historical linguistics and its new

branch "semantics" (or as Reisig called it "semasiology"), the reflection on
figures of speech also continued to be part of an ongoing philosophical
56 Brigitte Nerlich

examination of the relationship between language, thought and imagination,

triggered in part by Kant's neglect of language and in part by Kant's analysis
of aesthetic judgements (e.g. Herder, Goethe, Jean Paul, Gerber, Vischer,
Nietzsche, Biese, Mauthner) (see Schumacher 1997; Nerlich and Clarke
2001).
The analysis of metaphor also continued to be part of an ongoing study
of tropes and figures of speech in the more applied fields of rhetoric, poetics
and stylistics (e.g. Fontanier, Gross, Kleinpaul, Brinkmann, Wackemagel),
an approach which was only partially integrated into the linguistic current of
thought on metaphor and which was rejected as mere classificationism by
most of the philosophers of metaphor.
By the mid-19th century the linguistic and the philosophical approaches
to metaphor began to intermesh in the works of certain scholars (such as
Gerber, Wegener, Biese, Mauthner, Stiihlin, and Cassirer), who looked back
for inspiration to earlier philosophico-linguistic studies of metaphor by Vico
and Lambert for example. They all recognized the crucial role of metaphor
in language and in the structuring of thought and set the stage for such 20th-
century developments as Karl Biihler's declaration that metaphor is funda-
mental to all concept formation (Biihler 1990 [1934]).
Continuing the 18th-century tradition of stressing the ubiquity of meta-
phor in human thought and language, the linguist and philosopher Gustav
Gerber (1884) pointed out that there is no real difference between "literal"
language and "figurative" language, only between what Hermann Paul (1920
[1886]) later called usual and occasional or established and creative uses of
language. For Gerber all language is figurative or pictorial from its inception.
Those meanings which have acquired a certain stability through usage can
be called "literal" as opposed to those meanings whose special status is
highlighted through the use of tropes and figures of speech, which we then
can call "figurative". However, both types of meaning are only understood
insofar as those who know the language establish connections between the
"literal" or the "figurative" meaning and other words in a language (Gerber
1871-1874, II, I: 21, quoted by Knobloch 1988: 257). He also stressed that
the meaning of words and sentences can only be understood in the context
of their use and that the meaning of words is therefore never fixed but always
flexible and adaptable (Gerber 1884: 104). The meaning of words and sen-
tences always has to be "figured out" in text and context.
Based on work by Vico, Lambert, Goethe and Gerber, Alfred Biese then
elaborated, at the end of the 19th century, a philosophy of metaphor, which,
 Polysemy: past and present 57

well before Mark 10hnson (1987), proposed a close link between "the body
and the mind" as the basis for metaphor. He wrote: "metaphor is therefore
not a poetic trope, but an original form of intuition for thought". For Biese,
metaphor is the most essential inner schema of the human mind" (Biese
1893: VI). Biese was one of the most outspoken advocates of a philosophy
of the metaphoric according to which metaphor was neither just a "figure of
speech", nor just a poetic fiction or decoration, but was regarded as underlying
the structure and evolution of human thought and language.
A more empirical study of what Eve Sweetser calls the "Mind-as-Body
Metaphor" can be found in Hans Kurath's 1921 Chicago dissertation on the
semantic sources of the words for the emotions in Sanskrit, Greek, Latin,
and the Germanic languages:

Kurath (1921) notes that Indo-European words for the emotions are very
frequently derived from words referring to physical actions or sensations accom-
panying the relevant emotions, or to the bodily organs affected by those
physical reactions. (Sweetser 1990: 28)

Although the link between the historical development of emotion words and
the psychosomatic nature of the emotions is not as fundamental as Kurath
thinks, we can see here a first insight into the metaphorical nature of expres-
sions such as "my anger boiled over" which have so frequently been studied
in modern literature on emotions and metaphors (see Koevesces 2000).
While philosophers such as Biese explored the all-pervasive influence of
metaphor on language and thought, linguists continued to study the processes
and results of metaphorization in language, and psychologists began to ex-
plore what Gilles Fauconnier has recently called "backstage cognition", that
is, "how language works in concert with 'behind the scenes' understandings
and cognitive processes" (Fauconnier 1998: 277).
In the following we shall first investigate how philosophers and linguists
analysed the results of the use of figurative language, namely the increase in
the spectrum of meanings of a word, what Breal called "polysemy", then
how Breal revolutionized this study.
58 Brigitte Nerlich

3. Polysemy research before Breal

The modern term polysemy was popularized by Breal in 1887 (Breal 1991
[1887]). Most modern linguists dealing with the topic of polysemy refer to
this crucial date, but they rarely look further back into the past (but see
NerIich and Clarke 1997).
The roots of the concept of "polysemy" lie in Greek philosophy, that is,
the debate surrounding the problem of the naturalness or arbitrariness of
signs as debated in Plato's (429-347 B.C.) Cratylus. In his account of Plato's
contribution to linguistics, Fred Householder points out that

Democritus [460/457-mid-4th century B.C.] (as quoted in Proclus'

commentary on the CratyLus 16) offered four arguments (with four specially
coined names) in favour of arbitrariness: (a) "homonymy" or "polysemy",
i.e., the same sequence of phonemes may be associated with two or more
unrelated meanings; (b) "polyonymy" or "isorrophy", Le., the existence of
synonyms; (c) "metonymy", i.e., the fact that words and meanings change;
(d) "nonymy", i.e. the non-existence of single words for simple or familiar
ideas. (Householder 1995: 93)

Polysemy meant primarily what was later to be called "homonymy", referring

to the multiple, but unrelated meanings of a word. Breal still subsumed
homonymy under the heading of polysemy.
The term polyonymy was also used by the Stoics studying how one and
the same object can receive many different names, how it can become "many-
named" or polyonymous.
During the Middle Ages the interpretation of the Holy Scriptures came
up against the problem of polysemy, a polysemy that was acknowledged,
but one that had to be tamed (by the theory of the "four senses"). One of
those who marvelled at the polysemous profundity of the Scriptures was
Gilbert of Stanford, who compared them with

a rapid river that flows by producing new senses which, as it waves and
whirls, come one after the other in such a way that no single one annuls the
other; instead they accumulate and increasingly enrich this immense treasure
of divine wisdom; everyone, according to one's own intellectual ability, can
glean something from this inexhaustible storage of senses. (Eco 1996: 739)

The first to use the term polysemous in a relatively modern sense was
Dante, who wrote about the polysemous character of the poem: "Istius operis
 Polysemy: past and present 59

non est simplex sensus, immo dici potest polysemum, hoc est plurium
sensum" (This work doesn't have one simple meaning, on the contrary, I say
that it can be polysemous, that is can have many meanings) (quoted in OED
1989, Vol. XII: polysemous, p. 75, col. 1). As Eeo points out:

When presenting his poem to Cangrande della Scala. Dante makes

immediately clear that it has to be read as a "polisemos" ("polysemic")
message. One of the most celebrated examples of what Dante means by
polysemy is given in his analysis of some verses of Psalm 1l3. In exit Israel
de Aegypto. Following the mediaeval theory [of the four senses, BN], Dante
says concerning the first verse of the Psalm: If we look at the letter it means
the exodus of the sons of Israel from Egypt at the time of Moses; if we look
at the allegory. it means our redemption through Christ; if we look at the
moral sense it means the conversion of the soul from the misery of sin to the
state of grace; if we look at the mystical sense it means the departure of the
sanctified spirit from the servitude of this corruption to the freedom of the
eternal glory. (Eco 1996: 741)

The difference between the old reading of the bible and Dante's is that Dante
is actually "taking a way of reading the bible as an example of how to read
his own mundane poem!" (Eco 1996: 741.) The debate about the literal or
figurative, the single or multiple meaning of the bible continued throughout
the 16th and 17th centuries and is still with us today (see McFague 1983).
Thinking about meaning, language and its relation to the real and spiritual
world advanced enormously during the Renaissance (see Was wo 1987), but
real mundane research into the multiplicity of meaning only began in the
18th century, with the study of neologisms, synonyms and the figures of
speech. The emancipation of semantics or the study of meaning from religious
lhought was followed by the emancipation of semantics from etymology or
the quest for the true original meaning of a word.
In the early 19th century the literary scholar August Wilhelm Schlegel
spoke about the difficulties encountered when dealing with "les termes
polysemantiques" in dictionaries (Schlegel 1832: 42) and the etymologist
August Friedrich Polt pointed out in 1861:

that even without the creation of new words a language can be enriched
through the new conceptual use of old words (Polysemantics [Polysemantie])
alone. at least on its internal level, that is, semantically, linguistic structures
can gain an extraordinary depth and focus. (Pott 1861: 5)
60 Brigitte Nerlich

Historical linguists and lexicographers became increasingly interested in

the multiplicity of meaning from the point of view of etymology, historical
lexicography or historical semantics. As we have seen, figures of speech,
such as metaphor, metonymy and synecdoche provided lexicographers with
instant ways of charting the development of the multiple meanings of words.
However, lexicographers were no longer merely looking for the true, original,
first and etymological meaning of words, but came to examine how words
were used to make sense by those who used them in actual texts; they looked
at the various "readings" of a word (a theme taken up again in Schlieben-
Lange 1997). They were interested in finding the connections between the
meanings of polysemous words, in finding patterns in the evolution of
meaning and in putting order into the meanings of lexical entries - again a
problem that is still with us, especially in computational linguistics and natural
language processing.

4. Breal's study of polysemy

Bringing this emancipation of semantics from etymology to a head, Bn!al

felt compelled to create the term polysemie at the end of the 19th century
because he wanted to establish semantics as a new branch of generalling-
uistics, independent of etymology and lexicography. This is the reason why
this term and Bn!al's research into polysemy were the starting point for a
whole new tradition of studies into polysemy. From looking at polysemy in
disembodied lexical entries, Breal turned to polysemy as a phenomenon of
language use, language acquisition, language change and even neuroling-
uistics avant la lettre.
Breal knew that, diachronically, polysemy stems from the fact that the
new meanings or values that words acquire in use (through extension, restric-
tion, metaphor, etc.) do not automatically eliminate the old ones - polysemy
is therefore the result of semantic innovation. The new and the old meanings
exist in parallel (Breal 1924 [1897J: 143-144; English translation 1964
[1900]: 139-140) (see Blank, this volume, and Blank 1997).
And yet, synchronically, or in language use, poly se my does not really
exist (it is rather an artefact of lexicographers) - sense selection in the
comprehension process is actually not a problem at all. In the context of
discourse a word always has one meaning - except, one should point out, in
jokes and puns (see Tay tor. this volume). (Sense selection is, however, a
 Polysemy: past and present 61

problem for computers - see Kilgarriff, this volume; Wilks, this volume.)
The most important factor that brings about the multiplication of meanings
diachronically and that helps us to "reduce" the multiplicity of meanings
synchronically is the context of discourse, a fact already mentioned by
Leibniz. We understand polysemous words because the words are always
used in the context of a discourse and a situation, which eliminate all the
adjoining meanings in favour of only the one in question (Brea11991 [1887]:
156-157).
However, in the constant dialectical give and take between synchrony and
diachrony, and between meaning and understanding, or what Traugott and
Blank call in Gricean terms conversational and conventional implicature
(Traugott 1988; Blank 1997: 366-369), incremental changes in the meaning
of words occur, insofar as hearers, having understood a word in a certain
context in a slightly divergent way, become themselves speakers and might
use a word in the newly understood way in yet another context, which again
brings about a different type of understanding, and so on. In the long run
these slight variations in use and uptake might lead to major semantic changes
(see Turgot 1756).
More sudden shifts in meaning are brought about by the use of metaphor
and metonymy. Aristotle in The Rhetoric had already remarked that "strange
words simply puzzle us; ordinary words convey only what we know already;
it is from metaphor that we can best get hold of something fresh" (Smith
1982: 128). There are also shifts in meaning which have a more social than
poetic root, as when the word operation, studied by Breal, means something
different according to the social context in which it is used (by a mathema-
tician, a general, a surgeon, and so on). Analysing the multiplication of
meanings based on the speakers' and hearers' social and cognitive needs and
activities was central to Breal's semantics.
Breal was fascinated by the fact that when talking to each other we neither
get confused by the multiplicity of meanings that a word can have, some of
which are listed in dictionaries of usage, nor are we bothered with the
etymological ancestry of a word, traced by historical dictionaries.
Both the dictionary of usage and the historical dictionary classify the
meanings of polysemous words, which have been produced over time by a
nation. This is a social (abstract and decontextualized) classification, whereas
the classification of meanings in the head of a speaker and hearer is in each
case an individual (concrete and contextual) classification. Breal has in mind
an "isosynchronic competence", a half-conscious type of user knowledge, which
62 Brigitle Nerlich

only works inside concrete situations (see Desmet and Swiggers 1995: 283).
Modem polysemy research still debates whether it should predominantly deal
with the former type of polysemy or with the latter, and how to reconcile the
one with the other.
Breal observed that most of the time it is the latest, most modem meaning
of the word, yesterday's or today's meaning, with which we first become fam-
iliar (Breal 1991 [1884]: 149) - something recently rediscovered in England:

In 1985, the department of English at the University of Birmingham ran a

computer analysis of words as they are actually used in English and came up
with some surprising results. The primary dictionary meaning of words was
often far adrift from the sense in which they were actually used. Keep, for
instance, is usually defined as to retain, but in fact the word is much more
often employed in the sense of continuing, as in "keep cool" and "keep
smiling". See is only rarely required in the sense of utilizing one's eyes, but
much more often used to express the idea of knowing, as in "I see what you
mean". Give, even more interestingly, is most often used, to quote the
researcher, as "mere verbal padding", as in "give it a look" or "give a report".
(Bryson 1990: 143)

Language understanding and language acquisition follow the opposite route

of language change. In both cases, the last, not the first or primitive meaning
of a word is the basic meaning. In modem parlance one would say that the
most salient, not the most "literal" meaning is the one that we acquire first
and also use and understand first (see Giora 1997).
Breal therefore wondered at how children acquire word meanings. He
thought that: "In order to recognize how meanings are arranged in our heads,
the surest way is to see how they got in there in the first place" (Breal 1991
[1884]: (49).

The child who hears a word first pronounced retains it with the meaning with
which he has heard it used, and he associates it in his thought with the object
itself to which it was applied, and to no other. We could say that for him all
names are at first proper names. If a little later he hears the same word applied
to other similar objects, then he learns to generalize the meaning ....
Let us assume that the same word returns in another context, with another
meaning [valeur]. If the two meanings are not too far apart, the child senses
that it is the same word; he perceives the connexion between the two meanings
and, taking stock of the distance travelled, he extends his initial concept. ... If
on the other hand, the two meanings are too distant, he does not try to connect
 Polysemy: past and present 63

them, and he takes account of the new meaning as if dealing with a separate
word. (Bn!al 1991 [1884]: 149)

Breal was acutely aware of the fact that the semantic, cognitive, and
developmental side of language studies was not yet on a par with the advances
made in the study of phonetics, of the more physiological side of language.
In his article "How words are classified in our mind" (Breal 1991 [1884]),
Breal appeals to the future to supply us with insights into the cognitive aspects
of human language and writes:

When we compare intelligence to a file cabinet in which ideas are arranged

in order, or to a photographic plate on which images are deposited, or to an
instrument whose various strings vibrate in turn, it is clear that these are
merely analogies. In fact, the true nature of intellectual phenomena is unknown
to us. We perceive the effects, but the cause remains hidden. We are obliged
to transpose facts of a higher order into the language of the five senses. To be
precise, we should speak neither of ideas nor of words, for there are no such
things: there are only states, the habits of our brain and the movements of our
vocal apparatus. But if we spoke in this way no one would understand us.
Thus it is better to speak in an approximate way, while waiting for the
establishment of the science of human intelligence which one of our former
colleagues, one of the great figures of the Institute, Claude Bernard, liked to
call the science of the twentieth century. (Breal 1991 [1884]: 151)

With Breal semantics as a linguistic discipline made a first step into this
future, a future in which we are still participating and to which we are still
contributing beyond the end of the 20th century, the century of psycho-
linguistics, artificial intelligence, brain scanning and neuropsychology.

5. Polysemy research from Breal to the present

As we have seen, Breal had provided the conditions for new trends in
polysemy research, in general linguistics, in sociolinguistics, in psycho-
linguistics, and in the study of language acquisition. By the beginning of the
20th century a shift was noticeable from accommodating polysemy in his-
torical semantics towards observing and explaining polysemy in synchronic
semantics, as the study of la langue as well as the study of lexical fields. The
turn from the 19th to the 20th century also saw a steady influx of new types
64 Brigitte Nerlich

of psychology which started to compete with association psychology. There

were the act psychology of Franz Brentano, the physiological and "cultural"
psychology of Wilhelm Wundt, and the psychology of the unconscious of
Eduard Hartmann, Carl Gustav lung and Sigmund Freud, Gestalt psychology,
and Denk-psychology as developed by Oswald Kiilpe and Karl Buhler. As in
the days of association psychology, all these psychologies had an influence
on how linguists studied word meaning, polysemy, vagueness, and semantic
change.
Linguists, such as Karl Otto Erdmann and Hans Sperber, began to distin-
guish between the fundamental semantic value of a word and its emotionally
charged secondary meanings (Erdmann 1910 [1900]). According to Sperber
(1923), semantic change is driven by emotions. For Sperber semantic change
can only be studied as a change in an emotionally "charged" field of words.
The Swedish scholar Gustaf Stem tried to integrate all psychological and
linguistic approaches to meaning in his 1931 book on semantic change. As
Michael K. Smith pointed out in his review of Metaphors We Live By, "Stem's
masterful book has an excellent discussion of metaphor which includes a
much broader taxonomy and numerous examples of each type. In many places
Stern is theoretically close to the authors and even applies the term "gestalt"
to the relationship between words and concepts" (Smith 1982: 131-132).
Stern also developed an elaborate taxonomy of types of polysemy, as sum-
marized in Nerlich and Clarke (1997).
Despite these advances in a psychologically-inspired semantics, little was
known about the actual psychological processes underlying the creation and
understanding of metaphor, a major field of research in modern theories of
blending and conceptual integration. The linguist Philipp Wegener had
pointed the way towards the study of these processes. He stated as early as
1885 that sentence understanding in general and metaphor understanding in
particular are based on a filtering and blending process. The predicate (the
new) is seen through the lens or filter of the exposition (the given, which is
either the co-text of the utterance or the situational context), and the exposition
is filtered through the predicate. When a word is used in the co-text of a
sentence the collocation with other words filters out those mental represen-
tations which are congruent in this sentence. The others drop below the level
of consciousness (see Wegener 1991 [1885]: 50).
But a new psychology of language was needed to lead to real advances in
this area. This new psychology gradually emerged at the beginning of the
20th century. The old association psychology, with its notions of contiguity
 Polysemy: past and present 65

and similarity inherited from Hume and which had structured most of 19th-
century metaphor research up to and including Wilhelm Wundt, was replaced
by a psychology of consciousness and thinking, developed mostly in
Wiirzburg, and by a new Gestalt psychology developed mainly in Berlin.
Insights from both traditions were merged in empirical research into the
understanding of metaphors, especially in the works of Biihler and Wilhelm
Stahlin (1913), who both explored metaphor as a process of conceptual
integration or Sphiirenmischung.
These advances in the psychology of metaphor and polysemy were accom-
panied by advances in the sociology of poly se my. The polysemy of words
was explained by looking at how different social groups use one and the
same word for different purposes. An example (already found in Breal) for
the social differentiation underlying some types of polysemy was the word
operation, as used by a mathematician, a general or a surgeon, for example
(Meillet 1921 [1905]: 193). The end-point and summa of this tradition of
studying polysemy from a linguistic, conceptual, psychological and socio-
logical point of view was the work of the semanticist Stephen Ullmann
(1951).
There followed a period of polysemous latency, so to speak, after the
advent of transformational generative grammar with its focus on syntax and
later feature semantics. However, polysemy was not completely forgotten,
as illustrated by the research undertaken by Hans Blumenberg (1960), Uriel
Weinreich (1962, 1996), Harald Weinrich (1967), James McCawley (1968),
Jurij Apresjan (1974), Charles Fillmore (1975), Andrew Ortony (1975), and
Alfonso Caramazza and Ellen Grober (1976).
At the same time a new wave of polysemy research, under the heading of
fJoly-isotopie and polyphonie, developed in France in the framework of
general rhetoric and text analysis. This type of polysemy research harks back
to Dante's first use of the word "polysemous" as indicating the mUltiple
readings of a poetic text. It was directly influenced by the rediscovery of
work on the multiplicity of meanings, polyphony, and heteroglossia by the
Russian philosopher of language Mikhael Bakhtin (see Todorov 1977).
In the Anglo-American world, polysemy was rediscovered with the advent
of cognitive semantics in the 1980s (see Brugmann 1981; Lakoff 1987;
Langacker 1987; Geeraerts 1997). Cognitive linguists not only rediscovered
metaphor and metonymy as pathways and patterns which structure polysemous
sense extensions, they also began to reconnect synchronic and diachronic
research into meaning.
66 Brigitte Nerlich

Synchronic polysemy and historical change of meaning really supply the same
data in many ways. No historical shift of meaning can take place without an
intervening stage of polysemy ... But if an intervening stage of polysemy was
involved, then all the historical data, as evidence of past polysemy relations, is
an interesting source of information about the reflection of cognitive structures
in language. Even more crucially, the historical order in which senses are
added to polysemous words tells us something about the directional relation-
ship between senses; it affects our understanding of cognitive structure to
know that spatial vocabulary universally acquires temporal meanings rather
than the reverse. (Sweetser 1990: 9)

More recently polysemy has become central to a wide variety of researchers

in computational and "generative" semantics, especially in the works of Adam
Kilgarriff and James Pustejovsky.
In Germany, polysemy is explored nowadays in a new approach that tries
to integrate advances in cognitive semantics with the older tradition of
structural semantics, as well as with Gestalt theory and frame theory (see
Blank 1997; Koch 1998). These researchers try to steer a fine line between
two other approaches to polysemy proposed in Germany: the type of research
done in the framework of Manfred Bierwisch's "two-level-semantics" based
on generative linguistics (which makes a distinction between a layer of
meaning defined by grammar, and a level of interpretation based upon con-
ceptual knowledge) (see Bierwisch 1983; Bierwisch and Schreuder 1992;
Sweetser 1986; Schepping 1994; Schwarze and Schepping 1995; Schwarze
1995; Pause, Botz and Egg 1995), and the "simplest systematics" approach
recently proposed by Brigitte Schlieben-Lange (1997) and based on the one
hand on Gricean principles, on the other on a type of linguistics elaborated
by Eugenio Coseriu (see Koch 1998: 126). Bierwisch's approach has been
compared to the Langackerian approach favoured by cognitive linguists
working in the Lakoff/Johnson tradition by John Taylor in several articles
(see Taylor 1994, 1995,2(00).
In France and Belgium, polysemy is studied by Georges Kleiber (1990,
1999), Dirk Geeraerts (1992, 1997), by the cognitively and hermeneutically
inspired Fran~ois Rastier at the CNRS (Rastier 1987), as well as by the
semiotically inspired Pierre Cadiot (1992, 1993) at the University Paris VII,
by Catherine Fuchs and Bemard Victorri (see Fuchs 1991, 1996; Victorri
and Fuchs 1996; Fuchs and Goffic 2002) at the CNRS and many others.
Research into the multiplicity of meaning has certainly come a long way
 Palysemy: past and present 67

since the first efforts at systematizing the study of "la polysemie" were made
by Breal at the end of the 19th century.

Notes

I. Breal created the term in 1887 (Breal 1991 [1887]: 156-157) in his review of
two seminal books: Arsene Darmesteter's book La Vie des mats (1887) and the
second edition of Hermann Paul's book Prinzipien der Sprachgeschichte (1886
[1880)). He used it again in his famous Essai de semantique in 1897, and from
1900 onwards, when the Essai was translated into English, the term palysemy
made its way into mainstream linguistics.
2. Some parts of this chapter are based on previously published papers, such as
Nerlich and Clarke (1997, 2000a, 2(01) and Nerlich (1998).
3. lakobson's work launched two famous schools of thought, integrating research
on metaphor with structuralism, hermeneutics and speech act theory via the work
of Claude-Levi Strauss and Paul Ricoeur.
4. In his seminal article on the inescapable metaphoricity of language, H. WaIter
Schmitz quotes the Dutch philosopher lohann Clauberg (1622-1665) as the first
to use this example (see Schmitz 1985). Those interested in Locke's views of
metaphor and truth should read Oosthuizen Mouton (to appear).
5. All translations are mine, BN.
6. This type of argument can be found again in modern proponents of "cognitive
rhetoric", such as Dan Sperber (1975) and Mark Turner (1991).
7. "Explaining the meaning of a word by ostension, by pointing to something to
which the word applies, has been variously thought to constitute (i) a form of
explanation which provides language with a foundation, (ii) an explanation which,
in presupposing a general grasp of language, is only secondary, and (iii) a
procedure which does not qualify as a definition or explanation at all. While
ostension may serve to point the learner in the right general direction, there is
certainly a question as to how much eventual understanding may owe to any
such procedure, and how much it requires exposure to word usage over a period
of time." (http://www.xrefer.com/entry/55307l)
H. This view echoes Lorenzo Valla's 15th-century saying that "housewives some-
times have a better sense of the meaning of words than the greatest philosophers"
(quoted in Waswo 1987: 95).
68 Brigitte Nerlich

References

Adelung, 10hann Christoph

1798 Metapher. Grammatisch-kritisches Worterbuch der Hochdeutschen
Mundart. Dritter Theil, von M-Scr. 2. Vermehrte und verbesserte
Ausgabe. Leipzig: Breitkopf und Hiirtel.
Apresjan, Jurij D.
1974 Regular polysemy. Linguistics 142: 5-32.
Bierwisch, Manfred
1983 Semantische und konzeptionelle Repriisentation lexikalischer
Einheiten. In: Rudolf Ruzicka and Wolfgang Motsch (eds.),
Untersuchungen zur Semantik, 61-99. Berlin: Akademie-Verlag.
Bierwisch, Manfred and Robert Schreuder
1992 From concepts to lexical items. Cognition 4: 23-60.
Biese, Alfred
1893 Die Philosophie des Metaphorischen. In Grundlinien dargestellt.
Hamburg and Leipzig: Leopold Voss.
Black, Max
1962 Models and Metaphors: Studies in Language and Philosophy. Second
printing. Ithaca, New York: Cornell University Press, 1963.
Blank, Andreas
1997 Prinzipien des lexikalischen Bedeutungswandels am Beispiel der
romanischen Sprachen. (Beiheft der Zeitschrift fiir romanische
Philologie, Volume 285.) Ttibingen: Max Niemeyer.
Blumenberg, Hans
1960 Paradigmen zu einer Metaphorologie. Archiv fur Begriffsgeschichte
6: 7-142.
Bn!al, Michel
1924 [1897] Essai de semantique (Science des significations). Reprint of the 4th
edition. Paris: Gerard Monfort. First edition 1897.
1964 Semantics: Studies in the Science of Meaning. New York: Dover
Publications. Translation of Essai de semantique - unabridged and
unaltered republication of the work first published by Henry Holt
and Company in 1900.
1991 [1884]How words are organized in the mind (1884). In: Breal, The
Beginnings of Semantics. Essays, lectures and reviews, edited and
translated by George Wolf, 145-151. Oxford: Duckworth.
1991 [1887] The history of words (1887). In: Breal, The Beginnings of Semantics.
Essays. lectures and reviews, edited and translated by George Wolf.
152-175. Oxford: Duckworth.
Brugman, Claudia
1981 Story of over. M.A. thesis. University of California, Berkeley.
 Polysemy: past and present 69

Bryson, Bill
1990 Mother Tongue: The English Language. London and New York:
Penguin Books.
Buhler, Karl
1990 [1934] Theory of Language: The Representational Function of Language.
Amsterdam and Philadelphia: John Benjamins. English translation
by Donald Fraser Goodwin, edited with an introduction by Achim
Eschbach.
Burke, Kenneth
1962 A Grammar of Motives and a Rhetoric of Motives. Cleveland: The
World Publishing Company.
Cadiot, Pierre
1992 Extensions et glissements polysemiques ... d'une langue a l'autre.
In: Michele Lorgnet (ed.), Atth delta Fiera Internazionale della
Traduzione J, Riccione, (decembre 1990), ateneo, Forli: 31-57.
1993 Representations d'objets et semantique lexicale: qu'est ce qu'une
bofte? Journal of French Language Studies 4: 1-23.
Caramazza, Alfonso and Ellen Grober
1976 Polysemy and the structure of the subjective lexicon. In: Clea Rameh
(ed.), Semantics: Theory and Application, 181-206. Washington, DC:
Georgetown University Press.
Danesi, Marcel
1990 Thinking is seeing: visual metaphors and the nature of thought.
Semiotica 80(3/4): 221-237.
Darmesteter, Arsene
1887 La vie des mots etudiee dans leurs signijications. Paris: Delagrave.
Desmet, Piet and Pierre Swiggers
1995 De la grammaire comparee Cl. la semantique. Textes de Michel Breal
publies entre 1864 et 1898. Introduction, commentaires et bibliographie
par Piet Desmet et Pierre Swiggers. Leuven and Paris: Peeters.
Dirven, Rene and Johan Vanparys (eds.)
1995 Current Approaches to the Lexicon. Frankfurt: Peter Lang.
Du Marsais, Cesar Chesneau
1977 [1730] Traite des Tropes ou des differents sens dans lesquels on peut prendre
un meme mot dans une meme langue. Postface de Claude Mouchard,
suivi de Jean Paulhan, Traite des Figures ou La Rhetorique decryptee.
Paris: Le Nouveau Commerce.
Eeo, Umberto
1996 History and historiography of semiotics. In: Roland Posner, Kurt
Robering and Thomas A. Sebeok (eds.), SemiotiklSemiotics: Ein
Handbuch zu den zeichentheoretichen Grundlagen von Natur und
KulturlA Handbook on the sign-theoretic foundations of nature and
culture. Volume 1,730--746. Berlin and New York: Waiter de Gruyter.
70 Brigitte Nerlich

Erdmann, Karl Ouo

1910 [1900] Die Bedeutung des Wortes: Aufsatze aus dem Grenzgebiet der Sprach-
psychologie und Logik. 2. Aufl. Leipzig: Eduard Avenarius. First ed.
Leipzig: Haessel; 3rd ed. 1922; 4th ed. 1925.
Fauconnier, Gilles
1998 Mental spaces and conceptual integration. In: Michael Tomasello
(ed.), The New Psychology of Language: Cognitive And Functional
Approaches To Language Structure, 251-280. Mahwah, NI and
London: Lawrence Erlbaum.
Fillmore, Charles I.
1975 An alternative to checklist theories of meaning. Proceedings of the
Annual Meetings of the Berkeley Linguistics Society (BLS) 1: 123-
131.
Fuchs, Catherine
1991 Polysemie, interpretation et typicalite: I' exemple de pouvoir. In:
Daniele Dubois (ed.), Semantique et Cognition: Categories, Prototypes,
Typicalite, 161-170. Paris: Editions du CNRS (ColI. Sciences du
Langage).
1996 L'interpretation des polysemes grammaticaux en contexte. In: Georges
Kleiber and Martin Riegel (eds.), Les Formes du Sens: Etudes de
linguistique franraise, medievale et generale offertes a Robert
Martin, 127-133. Louvain: Duculot (Coli. Champs linguistiques).
Fuchs, Catherine and Pierre Le Goffic
2002 La polysemie de comme. In: Olivier Soutet (ed.), La Polysemie. Paris:
Presses de I'Universite Paris IV. (Actes du Colloque international
"La Polysemie", Paris, novembre 2000).
Geeraerts, Dirk
1992 Polysemy and prototypicality (Review article on Kleiber 1990).
Cognitive Linguistics 3(2): 219-231.
1997 Diachronic Prototype Semantics: A Contribution to Historical Lexi-
cology. Oxford: Clarendon.
Gerber, Gustav
1871-1874 Die Sprache als Kunst. 2 Bde. (Bd. 2 in zwei Teilen). Bromberg:
Heyfleder. Second ed. 1885; reprint Hildesheim: Olms 1961.
1884 Die Sprache und das Erkennen. Berlin: Gaertner.
Giora, Rachel
1997 Understanding figurative and literal language: the graded salience
hypothesis. Cognitive Linguistics 7( 1): 183-206.
Householder, Fred W.
1995 Plato and his predecessors. In: E. F. K. Koerner and Ronald E. Asher
(eds.), Concise History of the Language Sciences, 9-93. Oxford:
Pergamon.
 Polysemy: past and present 71

Jakel,Olaf
1999 Kant, Blumenberg, Weinrich: some forgotten contributions to the
cognitive theory of metaphor. In: Raymond Gibbs Jr. and Gerdard J.
Steen (eds.), Metaphor in Cognitive Linguistics. Amsterdam and
Philadelphia: John Benjamins.
Jakobson, Roman
1956a The metaphoric and metonymic poles. In: Roman lakobson and
Morris Halle (eds.), Fundamentals of Language, 76--82. Gravenhage:
Mouton and Co.
1956b Two aspects of language and two types of aphasic disturbances. In:
Krystyna Pomorska and Stephen Rudy (eds.), Language in Literature,
95-120. Cambridge, MA; London, England: The Belknap Press of
Harvard University Press 1983.
Johnson, Mark
1987 The Body in the Mind: The Bodily Basis of Meaning, Imagination
and Reason. Chicago: Chicago University Press.
Kleiber, Georges
1990 La semantique du prototype. Paris: Presses Universitaires de France.
1999 Problemes de semantique: La Polysemie en questions. Paris: Presses
Universitaires du Septentrion.
Knobloch, Clemens
1988 Geschichte der psychologischen Sprachauffassung in DeutschLand
von 1850 bis 1920. Ttibingen: Niemeyer.
Koch. Peter
1998 Saussures mouton und Hjelmslevs trre: zwei Schulbeispiele zwischen
Semstructur und Polysemie. In: Edeltraud Wemer, Ricarda Liver.
Yvonne Stork, and Martina Nicklaus (eds.), Et multum et multa.
Festschrift fur Peter Wunderli zum 60. Geburtstag, 113-136.
Ttibingen: Gunter Narr.
Koevecses, Zoltan
2000 Metaphor and Emotion: Language, Culture and Body in Human
Feeling. Cambridge: Cambridge University Press.
Lakoff. George
1987 Women, Fire and Dangerous Things. 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: The Cognitive Unconscious and the Embodied
Mind: How the Embodied Mind Creates Philosophy. New York: Basic
Books.
Lakoff, George and Mark Turner
1989 More Than Cool Reason. Chicago: University of Chicago Press.
72 Brigitte Nerlich

Lambert, Johann Heinrich

1764 Philosophische Schriften. Hrsg. v. Hans-Werner Arndt. Bd. 1/11: Neues
Organon oder Gedanken iiber die Erforschung und Bezeichnung des
Wahren und dessen Unterscheidung vom Irrthum und Schein. Repro-
graphischer Nachdruck der Ausgabe. Leipzig: bey Johann Wendler
1764. Hildesheim: Georg Olms, 1965.
Langacker, Ronald W.
1987 Foundations of Cognitive Grammar, Vol. I: TheoreticaL Prerequisites.
Stanford, CA: Stanford University Press.
1988 Review of George Lakoff: Women. Fire, and Dangerous Things,
Chicago: Chicago University Press. Language 64(2): 384-395.
Leibniz, Gottfried Wilhelm
1981 Nouveaux essais sur L'entendement humain. English: New essays on
human understanding. Translated and edited by Peter Remnant and
Jonathan Bennett. Cambridge: Cambridge University Press.
1882 Die philosophischen Schriften von Gottfried WilheLm Leibniz, Volume
V. Edited by C. I. Gerhardt. Berlin: Weidmann.
1956 PhiLosophicaL Papers And Letters. A Selection translated and edited,
with an introduction by Leroy E. Loemker. Volume I. Chicago: The
University of Chicago Press.
Locke,John
1975 [1689] Essay on Human Understanding, edited by Peter H. Nidditch. Oxford:
Oxford University Press. First ed. 1689; 5th ed. 1706.
McCawley, James D.
1968 The role of semantics in grammar. In: Emmon Bach and Robert T.
Harms (eds.), UniversaLs in Linguistic Theory. New York: Holt,
Rinehart and Winston, Inc.
McFague, SaIlie
1983 Metaphorical TheoLogy: ModeLs of God in Religious Language.
London: SCM Press Ltd.
MeiIlet, Antoine
1921 [1905] Linguistique historique et linguistique generate. [Vol. I.] Paris:
Champion.
Nerlich, Brigitte
1992 Semantic Theories in Europe 1830-1930. Amsterdam and Philadelphia:
John Benjamins.
1998 La metaphore et la metonymie: aux sources rhetoriques des theories
semantiques modernes. Semiotiques 14: 143-\70.
2000 La semantique et la polysemie: de la conceptualisation a la
designation de domaines et concepts linguistiques nouveaux. In
Gabriel Bergougnioux (ed.), Breal et Le sens de La Semantique, 183-
194. Orleans: University of Orleans Press.
 Polysemy: past and present 73
Nerlich, Brigitte and David D. Clarke
1997 Polysemy: patterns in meaning and patterns in history. Historiographia
Linguistica 24(3): 359-385.
1999 Synecdoche as a communicative strategy. In: Andreas Blank and Peter
Koch (eds.), Historical Semantics and Cognition, 197-213. Berlin
and New York: Mouton de Gruyter.
2000a Blending the past and the present. Reflections on conceptual and
linguistic integration. LOGOS. Journal of General Linguistics and
Language Theory 1: 3-18. Reprinted with some changes in Rene
Dirven and Ralf Porings (eds.), Metaphor and Metonymy in Com-
parison and Contrast, 555-594. Berlin and New York: Mouton de
Gruyter.
2000b Semantic fields and frames. Historical explorations of the interface
between language, action and cognition. Journal of Pragmatic 32(2):
125-150.
2001 Mind, meaning, and metaphor: 'the philosophy and psychology of
metaphor in nineteenth-century Germany. History of the Human
Sciences 14(2): 39-61.
OED
1989 Oxford English Dictionary. 2nd edition. Oxford: Clarenden.
Oosthuizen Mouton, Nicholas T.
(to appear) Metaphor and truth in the seventeenth century. In: Armin Burkhardt
and Brigitte Nerlich (eds.), Reflections on Tropical Truth: Studies on
the Epistemology of Metaphor. Berlin and New York: Mouton de
Gruyter.
Ortony, Andrew
1975 On putting apples into bottles: a problem of polysemy. Cognitive
Psychology 7: 167-180.
Paul, Hermann
1920 [1886] Prinzipien der Sprachgeschichte. Wiederabdruck der 9. Auflage.
(Konzepte der Sprach- und LiteraturwissenschaJt 6.) Tilbingen:
Niemeyer. Second edition 1886,
Pause, Peter E., Achim Botz and Markus Egg
1995 Partir c'est quitter un peu: a two-level approach to polysemy. In: Urs
Egli et al. (eds.), Lexical Knowledge in the Organization of Language,
245-283. (Current Issues in Linguistic Theory 114.) Amsterdam:
Benjamins.
Pott, August Friedrich
1861 Etymologische Forschungen auf dem Gebiete der Indo-Germanischen
Sprachen. unter Beriicksichtigung ihrer Hauptformen. Sanskrit; Zend-
Persisch; Griechisch-Lateinisch; Littauisch-Slawisch; Germanisch
und KeLtisch. 2nd ed. Zweiten Theiles erste Abtheilung: Wurzeln;
Einleitung. Lemgo-Detmold: Meyersche Hofbuchhandlung.
74 Brigitte Nerlich

Rastier Fran~ois
1987 Semantique interpretative. Paris: Presses Universitaires de France.
Reisig, Christian Karl
1890 [1839] Vorlesungen uber lateinische Sprachwissenschajt. Mit den Anmerkungen
von Friedrich Haase. Unter Benutzung der hinterlassenen Manuskripte
neu bearbeitet von Hermann Hagen. Zweiter Bd. 1890. Lateinische
Semasiologie oder Bedeutungslehre. Neu bearbeitet von Ferdinand
Heerdegen. Berlin: Calvary.
Richards, Ivor Armstrong
1936 The Philosophy of Rhetoric. Second printing 1950. New York: Oxford
University Press.
Ritter, Joachim (ed.)
1972 Historisches Worterbuch der Philosophie; herausgegeben von
Joachim Ritter; unter Mitwirkung von mehr als 700 Fachgelehrten;
in Verbindung mit Gunther Bien ... [et al.]. Base1: Schwabe 1971-
<1998>. From Volume 4 edited by Karlfried Grunder.
Schepping, Marie-Therese
1994 Un cas de polysemie dans le domaine verbal: arriver. In: Hiltraud
Dupuy-Engelhardt (ed.), Actes d'EUROSEM 94. Colloque du 20 au
23 septembre 1994, Reims.
SchlegeJ, August Wilhelm von
1832 Rejlexions sur l'etude des langues asiatiques adressees a Sir lames
Mackintosh, suivies d'une lettre a M. Horace Hayman Wilson. Bonn:
Weber; Paris: Maze.
Schlieben-Lange, Brigitte
1997 Uberrlegungen zu einer einfachen Systematik der Zuweisung von
(polysemen) Lesarten. In: Ulrich. Hoinkes and Wolfgang Dietrich
(eds.), Kaleidoskop der Lexikalischen Semantik, 239-247. Tiibingen:
Narr.
Schmitz, H. WaIter
1985 Die durchgangige Tropisierung der Sprache. Uber einen Aspekt von
"Zeichen im Wandel". In: Klaus D. Dutz and Peter Schmitter (eds.),
Historiographia Semioticae: Studien zur Rekonstruktion der Theorie
und Geschichte der Semiotik, 241-270. Miinster: MAkS Publikationen.
Schumacher, Rene
1997 "Metapher". Erfassen und Verstehen frischer Metaphern. (Basler
Studien zur deutschen Sprache und Literatur 75.) Tiibingen und Basel:
A. Francke.
Schwarze, Christoph
1995 Polysemie im Sprachvergleich. In: WaIter Dahmen et al. (eds.),
Konvergenz und Divergenz in den romanischen Sprachen, 204-219.
Romanistisches Kolloquium vm. Tiibingen: Gunter Narr Verlag.
 Polysemy: past and present 75

Schwarze, Christoph and Marie-Therese Schepping

1995 Po1ysemy in a two-level-semantics. In: Urs Egli et al. (eds.). Lexical
Knowledge in the Organization of Language, 283-300. Amsterdam:
Benjamins.
Smith, Michael K.
1982 Metaphor and mind. Review of George Lakoff and Mark Johnson,
Metaphors We Live By, Chicago: University of Chicago Press 1980.
American Speech 57: 128-133.
Sperber, Hans
1923 Einfuhrung in die Bedeutungslehre. Bonn-Leipzig: Schroeder.
Stahlin, Wilhelm
1913 Zur Psychologie und Statistik der Metaphern. Eine methodologische
Untersuchung. Archiv fur die gesamte Psychologie 31: 297-425.
Stern, Gustaf
1931 Meaning and Change of Meaning: with Special Reference to the
English Language. Goteborg: Elanders Boktryckeri Aktiebolag.
(Reprinted, Bloomington and London: Indiana University Press,
1968.)
Sweetser, Eve E.
1986 Polysemy vs. abstraction: mutually exclusive or complementary?
Berkeley Linguistics Society, Vo!. 12, Berke1ey, 528-563.
1990 From Etymology to Pragmatics: Metaphorical and Cultural Aspects
of Semantic Structure. Cambridge: Cambridge University Press.
Taylor, John R.
1994 The two-level approach to meaning. Linguistische Berichte 149:
3-26.
1995 Models of word meaning in comparison: the two-level model
(Manfred Bierwisch) and the network model (Ronald Langacker).
In: Dirven and Vanparys (1995), 3-26.
2000 Approaches to word meaning: the network model (Langacker) and
the two-level model (Bierwisch) in comparison. In: Bert Peeters (ed.),
The Lexicon-Encyclopedia Inteiface, 115-141 (Current Research in
the SemanticslPragmatics Interface, Vo!. 5). Amsterdam: Elsevier.
Todorov, Tzvetan
1977 Theories du Symbole. Paris: Editions du Seui!.
Traugott, Elizabeth Closs
1988 Pragmatic strengthening and grammaticalization. In: Proceedings of
the Annual Meeting of the Berkeley Linguistics Society 14: 406-416.
Turgot, Anne Robert Jacques, baron de I' Aulne
1756 Etymologie. In: Encyclopedie, ou Dictionnaire Raisonne des Sciences,
des Arts et des Metiers. Tome Sixieme. Nouvelle impression en
facsimile de la premiere edition 1751-1780, 98a-1I1 b. Stuttgart-
Bad Cannstatt: Friedrich Fromann Verlag (Giinther Holzboog) 1967.
76 Brigitte Nerlich

Turner, Mark
1999 Contribution to cogling list, 30 January 1999 on "history of cognitive
linguistics". X-UIDL: 284bda4409504c43ad4da2924844ab36
Ullmann, Stephen
1951 The Principles of Semantics. Oxford: Basil Blackwell; Glasgow:
Jackson, Sons and Co.
1962 Semantics. Oxford: BIackwell.
Ungeheuer, Gerold
1980 Lamberts semantische Tektonik des Wortschatzes als universales
Prinzip. In: Gunter Brettschneider and Christian Lehmann (eds.),
Wege der Universalienforschung. Sprachwissenschaftliche Beitriige
zum 60. Geburtstag von Hansjakob Seiler, 87-93. Tiibingen: Gunter
Narr.
Vico, Giambattista
1948 [1725] The New Science of Giambattista Vico (Revised edition; T. G. Bergin
and M. H. Fisch, editors and translators). Ithaca, NY: Corn ell
University Press, 1948.
Victorri, Bernard and Catherine Fuchs
1996 La Polysemie: Construction Dynamique Du Sens. Paris: Hermes.
Was wo, Richard
1987 Language and Meaning in the Renaissance. Princeton: Princeton
University Press.
Wegener, Philipp
1991 [1885] Untersuchungen iiber die Grundfragen des Sprachlebens. HaIIe a.d.S.:
Niemeyer. New edition by E. F. K. Koerner with an introduction by
Clemens Knobloch. Amsterdam and Philadelphia: Benjamins 1991.
Weinreich, Uriel
1962 Lexicographic definition in descriptive semantics. In: Fred W.
Householder and Sol Saporta (eds.), Problems of Lexicography, 25-
43. Bloomington, IN: Indiana University Press. Reprinted in:
Weinreich (1980), 295-314.
1966 Explorations in semantic theory. In: Thomas A. Sebeok (ed.), Current
Trends in Linguistics, Volume 3, 395-477. The Hague: Mouton.
1980 On Semantics. Edited by WilIiam Labov and Beatrice S. Weinreich.
Philadelphia: University of Pennsylvania Press.
Weinrich, Harald
1967 Semantik der Metaphor. Folia Linguistica I: 3-17.
Cognitive approaches
Polysemy and conceptual blending'"

Gilles Fauconnier and Mark Turner

1. Introduction

Science is an eternal battle against common sense. How can it be the earth
that moves, when we so clearly see the sun travelling majestically across the
sky? Since when do feathers fall as fast as stones? Where did imaginary
numbers ever come from?
In thinking about meaning, common sense is no less of an obstacle. What
could be more obvious than the platitude that words have meanings, that
"dog" means dog and "house" means house? This reasonable and simple
view serves us well in everyday life and is widely shared. And yet there is
considerable evidence that it is deeply wrong - not just wrong because it is
oversimplified and in need of refinement, but more deeply wrong and mis-
leading in the very notion of "meaning" that it takes for granted.
The clash between our common-sense, self-evident view of meaning
contained in words and the infinitely more complex and remarkable reality
of meaning construction has certainly not escaped the notice of thinkers
through the ages. But it is only recently that we have started to come to grips
directly with the dynamics of on-line meaning construction and the wealth
and variety of cognitive capacities that we bring to bear on the most ordinary,
mundane situations.
It has been useful, in approaching such issues, to forget notions like
"meaning of an expression", "semantic representation", "truth-function", and
the like, and to think instead of the "meaning potential" of a language form.
Meaning potential is the essentially unlimited number of ways in which an
expression can prompt dynamic cognitive processes, which include concep-
tual connections, mappings, blends, and simulations. Such processes are
inherently creative, and we recognize them as such when they are triggered
or produced by art and literature. In everyday life, the creativity is hidden by
the largely unconscious and extremely swift nature of the myriad cognitive
operations that enter into the simplest of our meaning constructions. It is
also hidden by the necessary folk-theory of our everyday behaviour which is
80 Gilles Fauconnier and Mark Turner

based quite naturally on our conscious experience rather than on the less
accessible components of our cognition.
In this article, we look at some aspects of polysemy which derive from
the power of meaning potential. More specifically, we focus on aspects linked
to the operation of conceptual blending, a major cognitive resource for
creativity in many of its manifestations.
Poly se my is pervasive in language and appears in many forms. It is not
just an accident of history or of synchrony, but rather an essential manifes-
tation of the flexibility, adaptability, and richness in meaning potential that
lie at the very heart of what a language is and what it is for. It is also a
symptom (rather than a primitive component) of the way in which various
cognitive operations allow for creativity at many levels.
The diversity and wide range of polysemy, and the wealth of theoretical
implications associated with it, are richly attested in the present volume. In
this article we review a number of cases of polysemy associated with concep-
tual blending. A majority of these cases have been discussed in other contexts
and for a variety of theoretical purposes. Rather than repeat the analyses in
extenso, we will frequently refer the interested reader to the appropriate source
for a more detailed treatment. In the sources, polysemy was seldom in itself
the major focus. By bringing together a large number of cases in the present
context, we hope to give an idea of the overall importance of blending in
polysemy phenomena.

2. An example of blending: the history of the world record in the mile

On 8 July 1999, The New York Times reported that Hicham el-Guerrouj had
broken the record for the mile, with a time of 3:43.13. The illustration that
accompanied the article, reproduced in Figure 1, shows a one-quarter mile
racetrack with six figures running on it, representing el-Guerrouj in a race
against the fastest milers from each decade since Roger Bannister broke the
4-minute barrier in 1954. EI-Guerrouj is crossing the finish line as Bannister,
trailing everyone, is still 120 yards back. This illustration prompts us to
construct a conceptual packet that blends together structure from six separate
input mental spaces, each with a one-mile race in which the record is broken
by a runner. The blend places all six of the runners on a single racetrack,
with a single beginning time.
This blend has the familiar features of conceptual integration networks.
 Polysemy and conceptual blending 81

Aug. 6. 1958 May 6. 1954

Herb E1I1oH RoC.r BannIster
Jun. 23. 1961 Australia Britain
Jlm Ryun .......
United States :
July 17, 1919
SebastIan Co. .. ....
Britain :

July 27. Ise

Steve Cl'am ....
Britain

Figure 1. A comparison of world record times for the mile.

There is a cross-space mapping connecting counterparts in each of the six

spaces: winners, racetracks, finish lines, the mile distance, and so on, There
is a generic space containing the structure and elements taken to apply to all
these spaces, which constitute the fairly rich frame of running the mile and
breaking the record, There is selective projection to the blend: from each of
the six input spaces, we project to the blend the entire frame of running the
mile, but not, for example, a specific location for the race, or any of the
runners except the winner. Some counterparts projected to the blend are fused,
such as the racetracks. Others, such as the record breakers, are not. There is
emergent dynamic structure in the blend, namely, structure that cannot be
found in any of the inputs: the blend is a simulation of a mythic race between
giants of the sport, most of whom in fact never raced against each other. In
this mythic race, el-Guerrouj "defeats" Bannister by 120 yards,
This blend is immediately intelligible and persuasive, but its construction
is remarkably complicated. Projecting to the blend el-Guerrouj, his location
at the finish line, and his winning time as he crosses the finish line does not
tell us how to locate the other runners behind him. Naturally, their historical
records do not indicate where they were at time 3:43.13. Their location on
the track at this time must be calculated separately. In this case, the calculation
is made by assuming that each runner ran his race at a uniform speed, despite
the fact that this never happens, We see therefore that the input mental spaces
to the blend, however useful, are fictions that do not correspond to the real
situations that occurred, With these fictions in place, it is easy to compute
the distance each runner has travelled at time 3:43.13 as the product of 1
mile and the ratio of el-Guerrouj's winning time to the runner's winning
82 Gilles Fauconnier and Mark Turner

time. Subtracting the distance travelled from 1 mile yields the distance by
which the runner trails el-Guerrouj. For example, Bannister trails el-Guerrouj
by [1760 yards] - [(3:43.13/3:59.4)(1760 yards)] = 120 yards, rounding to
the nearest yard.
To see further that there is nothing automatic or inevitable about this
blend as an instrument for highlighting competition and record-breaking, we
can compare it to the blend for the history of breaking the distance record
for a fixed time in bicycling. In the standard one-hour competition in bicycling,
the time of the performance is invariant, while the distance varies. So one
breaks the record by going farther in one hour than anyone else has ever
gone in one hour. Now, for this blend, we can project both the time and the
distance for each of the previous record holders without having to perform
any calculation. We simply place all the record-holders on the same track,
each at the distance he had achieved after one hour. In fact, the blend for the
milers looks like the blend for the bicyclists, but in the first case some
aggressive manipulations were required to achieve the blend. In the bicycle
competition, the contestants in the inputs and in the blend all do in fact stop
after an identical period of time has elapsed, namely one hour. In the mile
race, the contestants in the blend effectively stop competing to win the
moment the winner crosses the finish line, even though their counterparts in
the input spaces continue to compete to win, to finish the mile, and in fact to
break the record.
In the rest of this article, we will show how various kinds of polysemy
occur as a result of blending. We will argue that the following principles
guide the development of polysemy and furthermore that most polysemy is
invisible:

(i) Through selective projection, expressions applied to an input can be

projected to apply to counterparts in the blend. In this way, blends
harness existing words in order to express the new meanings that
arise in the blend.
(ii) Combinations of expressions from the inputs may be appropriate for
picking out structure in the blend even though those combinations
are inappropriate for the inputs. In consequence, grammatical but
meaningless phrases can become grammatical and meaningful for the
blend.
(iii) Terminology that naturally applies to the blended space ends up,
through connections in the integration network, picking out meaning
 Polysemy and conceptual blending 83

that it could not have been used to pick out if the blend had not been
built.
(iv) Blending provides a continuum for polysemy effects. Polysemy is an
inevitable and routine outcome of blending, but it is only rarely
noticed. The noticeability of polysemy is a function of the availability
of certain frames through defaults, contexts, or culture.

The most obvious case of harnessing an existing word to express new

meaning is category extension. Suppose we refer to a particular domestic
relationship between two members of the same sex as a same-sex marriage.
As we have analysed in Turner and Fauconnier (1995), this expression
prompts us to create a conceptual blend that modifies the category "marriage".
The expression "same-sex" comes from the mental space with the domestic
relationship between the two adults of the same sex, while the word "marriage"
comes from the mental space with the frame of conventional marriage. These
expressions, attached to the inputs, are now used to evoke the blend, so
"marriage" now picks out new meaning. Conceivably, this category modific-
ation could become so conventional for the entire linguistic community that
one could without risk of appearing uncooperative refer to a same-sex marriage
as simply a "marriage". By Principle (i), "marriage" applies to an element in
the blend that is quite different from its counterpart in the inputs. By Principle
(ii), "The brides married each other at noon" is a combination of expressions
from the inputs that is now appropriate for the blend but impossible for the
inputs.
The expression computer virus is a parallel example, where the new
meaning is produced by technological innovation rather than social change.
A conceptual blending network links the inputs of computer processes and
health and medicine and the blended space of computer viruses, vaccines,
disinfectants, and so on. By Principle (i), some vocabulary that applies to the
inputs is projected to counterparts in the blend and so ends up expressing
new meaning. In fact, in this case, the vocabulary has become conventional
for the entire linguistic community, so one can say "I got a virus at the office",
meaning a computer virus. This example, unlike same-sex marriage, is felt
to be metaphorical, but the general blending mechanisms for creating poly-
semy are the same for same-sex marriage and computer virus. In fact, we
point out below that on a continuum of blending networks, some are felt to
be completely literal, some absolutely metaphoric, and others at various stages
in between. In this way, the mechanism for extending meaning in the case of
84 Gilles Fauconnier and Mark Turner

computer virus is not metaphor per se, although metaphor is a collateral

feature of this particular blending network.
Complex numbers is a category extension in mathematics, analysed at
length in Fauconnier and Turner (1998). One input space has the Euclidean
plane and the other has real numbers. Before the invention of complex
numbers, there was already a historical connection between the geometric
line and the real numbers. Under blending, this connection was extended to
involve the entire plane. In the blend, a complex number is both a number
and a point. This point uniquely specifies a vector from the origin. By
Principle (i), vocabulary from the geometric input is applied to this number,
which is now said to have a magnitude and an angle (or argument); and
vocabulary from the number input is applied to the complex number in the
blend, so we can now speak of operations like "addition" and "multiplication".
Clearly, the meanings of "angle", "product", "number", and "sum" - in a
way, the meanings of all terms in number theory - have been extended and
deeply modified. Vocabulary that applies to the inputs has been projected to
pick out counterparts in the blend, and consequently applies to the new
meaning that has been developed in the blend. By Principle (ii), it becomes
mathematically correct to say that "the angle of the product of two numbers
is the sum of the angles of the two numbers". This combination of expressions
from the inputs becomes appropriate for the blend even though it does not
apply to the inputs. Additionally, by Principle (ii) we can also refer to "the
square root of negative one" for the blend, but not for the inputs.
In the case of complex numbers, the meaning extension is absolutely
precise and rigorous. It defines an extension of mathematics itself. We see
that it is not through linguistic or psychological properties of terms like
"number" and "sum" that polysemy occurs. It occurs as a byproduct of the
conceptual change brought about by the blending network. Given the connec-
tions in the network, by Principle (i), words like "number" or "angle" from
the inputs come to apply to counterparts in the blend, producing a sharply
different mathematical meaning. Mathematics loses none of its rigour by
having words like "number" be polysemous. In some contexts, "number"
refers to elements that do not have angles; in other contexts, "number" refers
to elements that do have angles. "Number" retains all of its old meanings
but acquires a new one to pick out elements in the complex number blend.
The examples we have considered are all traditionally considered to be
category extension. For these cases, it is intuitively tempting to think that
the category is extended by adding or deleting criterial features. But, as we
 Polysemy and conceptual blending 85

have just seen with the example of complex numbers, category extension
occurs by blending, which is not simply an operation of adding and deleting
features. We now turn to an example where it does not seem even intuitively
that the blend arises by adding or deleting features.
Coulson (1997) analyses caffeine headache as having two conventional
readings, one in which the headache is caused by caffeine, the other in which
the headache is caused by lack of caffeine. In this second case, we need an
integration network involving a general schema for a headache and its cause,
a present scenario in which the person with the headache has had no caffeine,
and a counterfactual scenario in which someone has had caffeine and so has
no headache. In the blend, the particular person with the headache has had
no caffeine, the lack of caffeine is the cause of the headache, and the term
"caffeine" has been projected from the all-important and desirable counter-
factual scenario in which there is no headache. Although this looks and is
intricate, it is an instance of a general pattern in which the integration network
contains a fully activated and highly important mental space that is counter-
factual to the blend, and the simple term for the blend is taken from the
counterfactual scenario. For example, a nicotine fit is a fit caused by lack of
nicotine, where the term "nicotine" is taken from the counterfactual scenario
in which the person does not have the fit. Easy error, used in tennis to describe
an error in making what should have been an easy shot, takes the modifier
for the error from the counterfactual space. Perhaps most conventionally, a
money problem activates a counterfactual space where there is money and
no problem. In the blend, there is a problem, a causal relation, and a cause,
namely, no money, but the term indicating the cause is taken from the
counterfactual space in which there is money and therefore no problem.
We see that terms like caffeine headache have more than one meaning,
being polysemous, because there is more than one blending possibility. The
striking possibility pointed out by Coulson depends on the general availability
of blending networks that have two highly active input spaces where one is
directly counterfactual to the other in some crucial respect. For examples
like money problem and caffeine headache that are licensed by such networks,
it is very clear even at the intuitive level that the polysemy cannot be a result
of adding and deleting semantic features attached to the two words. A caffeine
headache situation, on the counterfactual reading, has no features of caffeine.
In fact, its indispensable feature is a total absence of caffeine.
86 Gilles Fauconnier and Mark Turner

3. Gradients of blending

In this section, we focus on a single word, father, to show how gradients of

blending yield gradients of poly se my effects.
Consider an exceptionally simple limiting case in which a generic space
in a conceptual integration network has two people and no relations. Take
Input 1 in the network to be the father-child subframe of our more general
kinship frame. And take Input 2 to consist only of two people with no relation
between them, e.g. Paul and Sally. A simple cross-space mapping can link
father and Paul, connecting them to one of the people in the generic space,
and ego and Sally, connecting them to the other person in the generic space.
Projecting father and Paul to the blend and fusing them there and ego and
Sally to the blend and fusing them there yields a very simple network, in
which the structure in the blend is almost entirely obtained by composition
of the input structures. It is essentially equivalent to a Fregean composition,
expressed in logical notation by something like

(I) FATHER (Paul, Sally)

It is also equivalent to filling in slots in a frame (father, ego) with fillers

(Paul, Sally). In English, this blend would be triggered by sentences like
Paul is the father of Sally. Another way to think about the resulting blend is
to view it as instantiating the projection of the kinship frame in Input 1 onto
the situation in Input 2, consisting of Paul and Sally.
These are very simple networks, and if they were the only form of inte-
gration ever observed, there would be scant justification for setting up a
theory of conceptual blending. Simple framing (or its Fregean equivalent)
would suffice.
But in fact these simple networks are only the beginning of a long gradient
of increasing complexity. Crucially, the same word (father for example) can
operate in all the networks of the gradient. Superficially, the result is that the
word appears to have many different meanings. On the contrary, the word is
always playing the same role in inviting us to use our potential to construct
meanings through mechanisms like conceptual integration. To show this, we
will work through a series of cases along this continuum all using the word
father. Consider:
 Polysemy and conceptual blending 87

(2) Zeus was the father of Sarpedon. He watched from Mount Olympus
as his mortal son met his fated death.

This example points to the fact that there was more pattern completion
and projection from inputs in the Paul and Sally case than we had realized.
In Paul is the father of Sally, we quietly projected from Input 1 a range of
conventional knowledge, such as the mortality of the father and his normal
paternal limitations. But it is a technical fact of blending that we can project
equally from either input. So in the case of Sarpedon, the framing of the
father's powers and limitations comes not this time from Input 1, but from
Input 2.
Now consider:

(3) Zeus is the father of Athena. She was born out of his head. fully clad
in armour.

Now, from the kinship space, we bring in general schemas of human

progeneration, such as the offspring's coming out of the body of a parent,
but we bring in from our knowledge of divinity the possibility of unusual
birth. We explicitly build in the blend, on the warrant of the second sentence,
the particular kind of progeneration which involves neither a mother nor an
infant.
The divinity in Input 2 allows for many wonderful blends, each of which
contains a "father" and a creative method of progeneration. For instance,
Zeus is also the father of Aphrodite, this time in virtue of having castrated
Chronos and cast his genitalia into the ocean foam, whence Aphrodite is
born.
Of course, the Zeus cases cannot be attributed to figurative speech or
analogy. Zeus is still felt to be quite the father of Sarpedon, Athena, and
Aphrodite. Family structure is inferred along with sentiments and emotions
that come with it.
Now consider Joseph was the father of Jesus. In this case, in the blend,
we do not project the usual structure of the father's role in procreation or the
non-virginity of the mother. But we can project family structure and family
sentiments and emotions. Again, this use of "father" is not felt to be meta-
phoric or analogical.
Now consider a neighbour who takes care of Sally for the day while Paul
is away, carrying out fatherly duties like making her lunch, accompanying
88 Gilles Fauconnier and Mark Turner

her to school, reading bedtime stories. That neighbour can say to Sally: "I'm
your father for today". Like the Zeus and Joseph blends, some family structure
and genealogy is projected. As in the Joseph blend (but not the Zeus blends),
progeneration is not projected. Many of the typical aspects of the father-
offspring relationship are projected (routines, taking care, responsibility,
affection, protection, guidance, authority, and so forth). Compositionality is
no longer at all an option to account for this case. Too many properties felt
as central are missing. We have moved along the conceptual integration
network continuum from the pole of "Fregean" networks. But clearly, we
have not reached a point on the continuum that would be felt intuitively to
be metaphorical. Fatherhood is not a metaphor for what the neighbour is
doing. In fact, although some analogy has now contributed to the mapping,
the function of this blend is stronger than just analogy between the neigh-
bour's actions and a father's actions. The neighbour in this local context is
really filling in the role of the father in relevant respects, not just doing
something "similar" to what the father does. The flexibility of blending with
selective projection and contextual elaboration allows for this intermediate
kind of situation which doesn't fit a prototypical semantic or pragmatic
characterization.
In the Zeus and Joseph cases, there are obvious Principle (i) and Principle
(ii) polysemy effects. By Principle (i), "father" is projected to the blend from
the father-ego input, but now picks out new meaning in the blend. By
Principle (ii), we can now refer in general, across all contexts, to Zeus as
"the parent of Athena", whereas, by contrast, Paul cannot be referred to in
all contexts as "the parent of Sally". We can refer to birth as "leaving Zeus's
head" in the way we normally refer to birth as "leaving the womb". Many
similar expressions, each using words that already apply to the inputs, can
be fashioned that pick out meaning only in the blend. We can also refer to
Joseph as "Jesus's mortal father", giving "mortal father" a contrastive rather
than redundant meaning, which is likewise inappropriate for the father-ego
input.
Consider further examples linked to father:

(4) a. The Pope is the father of all Catholics.

b. The Pope is the father of the Catholic Church.
c. George Washington is the father of our country.

They are further along the continuum. The first example still has people
 Polysemy and conceptual blending 89

in both inputs. From the kinship input that provides the word "father", we
project not progeneration at all but instead authority, size of the family,
responsibility, leadership, social role. From the second input, we project
specific properties of catholicism.
The second example arguably projects the role of a child to a single social
entity (the Church). The blend reflects a type of socio-cultural model, in
which a social entity (church, nation, community) is the "child" of its leader.
The word "father" is now felt to have a different meaning, but not a particu-
larly metaphoric one.
With the George Washington sentence, we go a little bit further by high-
lighting the causality in time between the parent and child, and between the
founder and the nation. This abstraction increases the perceived difference
between the two inputs and their domains. The impression of metaphor is
undoubtedly stronger. And that subjective impression reaches a higher point
when the two domains are even more explicitly distinguished, as in Newton
is the father of physics. Physics, as opposed to church and country, does not
even stand in metonymic relation to people and groups of people. Yet Newton
and Washington as adult men have all the criterial biological features of
possible fathers plus some of the stereotypical social ones (authority, respon-
sibility ... ). The conceptual integration networks directly bring in frame
structure from both inputs.
Even more subjectively metaphorical are cases like Pound's Fear, father
of crueLty (Turner 1987), where the two domains (emotions/qualities and
people/kinship) have no literal overlap at all, and the projected shared schema
is correspondingly abstract (causality). And finally, Wordsworth's acrobatic
metaphor The Child is Father of the man comes around almost full circle by
using background knowledge (children grow into men) to create emergent
structure in the blend giving a rich instantiation to the abstract generic causal
structure which maps kinship to the human condition in an unorthodox way.
The oddness of its counterpart connections and the extensive drawing on the
frames of both inputs to create a new organizing frame for the blend help
make Wordsworth's line feel figurative. But the syntax and mapping scheme
of The Child is Father of the man are the same as the syntax and the mapping
scheme of Paul is the father of Sally.
The kinds of blend we have been talking about are often constructed using
language. The reason language can prompt for blends that result in the same
word's being used to pick out different meanings is that language does not
represent meaning directly; it instead prompts for the construction of meaning
90 Gilles Fauconnier and Mark Turner

in systematic fashion. All of the "father" examples are examples of the

familiar XYZ construction ("x is the y of z") whose purpose is to prompt for
blends in systematic fashion, in ways that naturally result in polysemy.
This gradient of polysemy for the word "father" falls out naturally as a
consequence of the facts that (1) "father" is in each case attached to one of
the inputs; (2) blending as a conceptual operation applies to those inputs;
and (3) by Principles (i) and (ii), "father" comes to pick out elements in the
blend and to participate in phrases that pick out structure in the blend but not
the inputs. Polysemy is in this view not a property of words but a byproduct
of the operation of conceptual integration and the fact that words are attached
to its inputs. The cognitive operation of conceptual blending, with its mecha-
nisms of selective projection and elaboration, is not restricted to linguistic
examples. But a mind that can do blending, and that also knows language,
will inevitably develop polysemy for words through blending. If words show
up in inputs, they can be projected like any other element of an input. This
will change their domain of application, unnoticeably in most cases, but
noticeably when the emergent meaning in the blend to which they apply
seems remarkably distant from the domain of the input from which they
came. When we notice this distance, we call it by one of many names:
extension, bleaching, analogy, metaphor, displacement. On our view, poly-
semy is a very common phenomenon, a standard byproduct of conceptual
blending, but noticed only in a fraction of cases.
One way of thinking of language is as a system of prompts for integration.
Since the conceptual structures to be integrated are many, each with ranges
of words attached to them, an expression that prompts for their appropriate
integration has to combine words, and language has to have forms to make
these combinations possible. Obvious examples are predication (this beach
is safe) and compounding (likely story, possible solution, eligible bachelor,
fake gun). Consider, for example, this beach is safe. A common way of des-
cribing the meaning of this sentence is to say that a particular property, SAFE,
is predicated of an object, BEACH, by means of the words "safe" and "beach".
On this view, this house is safe asks us to apply the same particular property,
SAFE, to a different object, HOUSE. SO, "safe" just has one meaning, SAFE. It
would be straightforward to say the beach is safe when we want to let a
child play there. And in that situation, it would be equally true that the child
is safe. But now we see the purported property SAFE attributed to the beach
in the beach is safe and to the child in the child is safe would have to be two
different properties, namely, on a first approximation, something like NOT
 Polysemy and conceptual blending 91

POTENTIALLY HARMFUL as opposed to NOT LIKELY TO BE HARMED. By the same

token, the word "safe" in the sentence the beach is safe would have to apply
many different properties on the readings that the beach is legally protected
from development, has a statistically low number of drownings, is not a site
of violent crime, is owned in such a way that its ownership cannot be taken
away from the owner, is a vacation spot that can be proposed without problem
to someone (as in a safe bet), and so on. In one sense, "safe" can mean many
different things, but at the same time, there is no subjective apprehension of
polysemy in these cases.
The details in these cases are actually surprising. They show that in order
to make sense of "safe", we need to construct a counterfactual situation in
which there is a victim, a location, instruments, possessions, and harm to the
victim. In the case of the beach that is legally protected, the beach is the
victim and the developers do harm to it. In the case of the beach with few
drownings, the swimmers are victims and the beach (meaning by metonymy
the water) does harm. In the case of the beach without crime, the vacationer
is the victim and the criminals do harm. Alternatively, the owner of the beach
can be the victim, or the person to whom we propose vacation spots can be
the victim. We see that the noun to which "safe" is applied can point to
many different roles in many different scenarios, not just the role of victim.
This leads, as shown elsewhere (Fauconnier and Turner 2002), to an
analysis showing that, in order to do justice to the meaning of "safe", we
must regard it not as applying a particular property but instead prompting
for a particular kind of blend. The blend takes into account the frame of
harm and the specifics of the situation referred to in the rest of the expression.
We are required to blend them to create a counterfactual scenario in which
there is specific harm, and to understand how the present situation is disanalo-
gous to the specific counterfactual scenario. In fact, the linguistic expression
singles out the disanalogous counterparts. For example, the beach is safe,
meaning that the child will not drown, singles out a counterfactual counterpart
beach with riptides, deadly waves, and so on, and asks us to understand that
the beach in the present situation is disanalogous to it. If "safe" does have a
meaning, it is something like "perform a conceptual integration, finding on
your own appropriate connections, given the other words in the expression,
and building a suitable counterfactual space on the basis of the harm scenario".
In fact, it is just like the caffeine example above that involved a counter-
factual space of not having a headache and having had the caffeine, and
understanding how that counterfactual space is disanalogous to the present
92 Gilles Fauconnier and Mark Turner

situation. Here, too, the linguistic expression selects "caffeine" and "head-
ache" in order to identify the disanalogous counterparts.
Now consider a word like "likely". Sweetser (1999) considers the case in
which likely candidate means not someone likely to become a candidate or
succeed as a candidate but, e.g., a candidate likely to grant an interview. As
she writes, "So long as we can think up a scenario relative to the candidate
in question, and evaluate that scenario for likelihood, likely candidate can
mean the candidate who figures in the scenario we have labeled as likely."
On her analysis, conceiving of such a scenario and evaluating it consists of
finding a blend of the frame for likelihood, conceived of as probability of
occurrence in a sequence, and the frame for candidate. Like "safe" above,
"likely" prompts for a blend. Sweetser's examples make the point clearly
that the scenarios necessary to do the appropriate blending mayor may not
be connected at all to the particular lexical items (e.g. "candidate"), as we
see, for example, in the case where possible textbook refers to a textbook
that might possibly be chosen as the one to be used in a college course.
Just as the different meanings of "safe" may go unnoticed, so the different
meanings of "possible" and "likely" may go unnoticed. But from a logical
standpoint, a possible textbook in the sense of one that may be adopted is not
the same as a possible textbook in the sense of one that might exist, or might
be written, or a trade book that could double as a textbook. As before, in the
cases of "likely" and "possible", blending opens the possibility of extensive
polysemy in the logical sense, but which may go unnoticed. This is no
accident, since the function of these linguistic forms, like the linguistic forms
"Adj-Noun" and "Noun-Noun", is to prompt for blending.

4. Conclusion: the purpose of linguistic form

Human beings are confronted by a fundamental problem: conceptual systems

are vast, rich and open-ended, while linguistic systems, impressive though
they be, are relatively quite thin. How can a linguistic system be used to
convey the products of conceptual systems, and how can these products find
expression in language, given the stark mismatch in their respective infinities?
If forms of language had to represent complete meanings, language could
communicate very little. The evolutionary solution to this problem is to have
systems of forms prompt for the construction of meanings that go far beyond
anything like the form itself. The "of' found in a range of examples like
 Polysemy and conceptual blending 93

Paul is the father of Sally, Father of cruelty, Father of the Catholic Church,
Vanity is the quicksand of reason, Wit is the salt of conversation, and so on
does not single out any particular blend or even any particular projection; it
only prompts for finding a way to construct a conceptual network that will
have a relevant meaning. What we have to do to construct that network is
nowhere represented in the linguistic structure. The single word "of' is thus
associated with an infinity of mappings. Of course, this infinity of mappings
is anything but arbitrary. It is constrained itself by the requirements on con-
ceptual integration networks. Different grammatical forms prompt different
infinities of conceptual mappings.
Because linguistic expressions prompt for meanings rather than represent
meanings, linguistic systems do not have to be, and in fact cannot be, ana-
logues of much richer conceptual systems. Prompting for meaning construc-
tion is a job they can do; representing meanings is not. As we have shown in
this article, a byproduct of constructing conceptual integration networks will
be massive, though often unrecognized, polysemy.

Notes

* © Gilles Fauconnier and Mark Turner.

References

Coulson, Seana
1997 Semantic leaps: the role of frame-shifting and conceptual blending
in meaning construction. Ph.D. dissertation, University of California,
San Diego.
Fauconnier, Gilles and Mark Turner
1998 Conceptual integration networks. Cognitive Science 22(2): 133-187.
2002 The Way We Think: Conceptual Blending and the Minds Hidden
Complexities. New York: Basic Books.
Sweetser, Eve
1999 Compositionality and blending: working towards a fuller understanding
of semantic composition in a cognitively realistic framework. In:
Theo Janssen and Gisela Redeker (eds.), Scope and Foundations of
Cognitive Linguistics, 129-162. The Hague: Mouton De Gruyter.
(Cognitive Linguistics Research Series.)
94 Gilles Fauconnier and Mark Turner

Turner, Mark
1987 Death is the Mother of Beauty: Mind. Metaphor; Criticism. Chicago:
University of Chicago Press.
Turner, Mark and Gilles Fauconnier
1995 Conceptual integration and formal expression. Journal of Metaphor
and Symbolic Activity 10(3): 183-204.
Reconsidering prepositional polysemy networks:
the case of over *

Andrea Tyler and vyvyan Evans

1. Introduction

We focus here on the issue of semantic polysemy, the phenomenon whereby

a single linguistic form is associated with a number of related but distinct
meanings or senses. In particular, we consider how the notorious polysemy
of the English preposition over might be accounted for in a principled,
systematic manner within a cognitive linguistic framework. At base, we argue
that the many senses of over constitute a motivated semantic network orga-
nized around an abstract, primary meaning component, termed aprotoscene.
The many distinct senses associated with over are accounted for by interaction
of the protoscene with a constrained set of cognitive principles. Accordingly,
our more general claim is that the lexicon is not an arbitrary repository of
unrelated lexemes. Rather, the lexicon constitutes an elaborate network of
form-meaning associations (Langacker 1987, 1991a, 1991b), in which each
form is paired with a semantic network or continuum (Brisard 1997). This
follows from two basic assumptions, widely demonstrated within the frame-
work of cognitive linguistics. First, semantic structure derives from and
mirrors conceptual structure (see, for example, Fauconnier 1994, 1997; Heine
1997; lackendoff 1983; Lakoff 1987). Second, the kinds of bodies and neural
architecture human beings have - how we experience - and the nature of the
spatio-physical world we happen to live in - what we experience - determine
the conceptual structure we have (Clark 1973; Evans 2000; Grady 1997;
Heine 1993, 1997; lohnson 1987; Lakoff and 10hnson 1980, 1999; Svorou
1993; Sweetser 1990; Talmy 1983, 1988, 1996, 2000; Turner 1991; Varela,
Thompson and Rosch 1991).
This model of the lexicon generally, and the model of polysemy proposed
here in particular, contrasts with traditional models in a number of ways.
The traditional view holds that all regularity and productivity are in the syn-
tax, with the lexicon serving as a repository of the arbitrary. Aronoff (1994)
points out that Bloomfield articulated this perspective as early as 1933. More
96 Andrea Tyler and \yvyan Evans

recently, Chomsky has reasserted this stance: "I understand the lexicon in a
rather traditional sense: as a list of 'exceptions', whatever does not follow
from general principles" (1995: 235). Models within this framework have
tended to represent different word senses as distinct lexical items (Croft 1998).
Polysemous forms are simply represented as an arbitrary list of discrete words
that happen to share the same phonological form.
Over the years, this stand has been criticized for failing to account for
systematic ways in which numerous forms are clearly related (Jackendoff
1997; Langacker 1991 a; Levin 1993; Pustejovsky 1998). Croft (1998) notes
that a number of linguists have argued for some type of derivation within the
lexicon that would represent distinct senses as arising from a primary sense
via a set of lexical operations. By and large, these analyses have focused on
polysemy involving changes in the argument structure of verbs or alterna-
tively in category changes, and have had little to say about the type of
polysemy demonstrated by English prepositions in which syntactic category
changes are often not involved.
In fact, most linguists (cognitive linguists excepted) have not paid much
attention to the phenomena of polysemy. Pustejovsky notes that "The major
part of semantic research ... has been on logical form and the mapping from
a sentence-level syntactic representation to a logical representation" (1998:
33). The lexicon has been represented as a static set of word senses, tagged
with features for syntactic, morphological and semantic information, ready
to be inserted into syntactic frames with appropriately matching features.
Within this tradition the lexicon has been viewed as "a finite set of [discrete]
memorized units of meaning" (Jackendoff 1997: 4).
Cognitive linguistics takes a significantly different perspective on the
nature of the mental lexicon. Of primary importance is the notion of embodied
meaning: the meanings associated with many individuallexemes are instanti-
ated in memory not in terms of features, nor as abstract propositions, but
rather as imagistic, schematic representations. Such image-schemas are held
to be embodied, in the sense that they arise from perceptual reanalysis of
recurring patterns in everyday physical experience (see Johnson 1987;
Mandler 1992, 1996 for a developmental perspective). I Perceptual analysis
creates a new, abstract level of information - information tied to the spatio-
physical world we inhabit but mediated by human perception and concep-
tualization. The central assumption of embodied meaning stands in stark
contrast to approaches to the mental lexicon that represent lexical items as
bundles of semantic. syntactic and morphological features.
 The case of over 97

A second distinguishing tenet of cognitive linguistics involves the repres-

entation of lexical items as natural categories involved in networks or con-
tinuums of meaning. Research into human categorization (Rosch 1975)
strongly suggests that speakers distinguish between prototypical and per-
ipheral members of a set, based not on criterial properties or features, but
rather on how predictable a member is, based on a prototype (Lakoff 1987).
Consequently, cognitive semantic accounts of polysemy (Brugman 1981;
Brugman and Lakoff 1988; Lakoff 1987) have argued that lexical items
constitute natural categories of related senses organized with respect to a
primary sense and thus form semantic or polysemy networks. Hence, such
accounts are strongly suggestive that the lexicon is much more motivated
and organized than has traditionally been assumed (Dirven 1993; Lakoff
1987; see also Langacker 1991a; the work in construction grammar argues
in a related vein, e.g. Fillmore, Kay and O'Connor 1988; Kay and Fillmore
1999; Goldberg 1995).
In the 1980s, Brugman conducted pioneering work in the polysemy of
the English preposition over (1981 [1988]). This research was followed by
Lakoff (1987), Brugman and Lakoff (1988), Dewell (1994) and Kreitzer
(1997). Brugman and Lakoff treated prepositions as denoting a spatial rel-
ation between an element in focus (the figure), and an element not in focus
(the groundV The BrugmanlLakoff framework took a highly fine-grained
approach to the semantics of prepositions. Accordingly, Lakoff (1987)
provides a network that contains at least 24 distinct senses. More recently,
work such as Evans (2000), Kreitzer (1997), Rice (1993), Ruhl (1989),3
Sandra (1998), Sandra and Rice (1995), Tyler and Evans (2003), and
Vandeloise (1990), has questioned whether such a fine-grained analysis is
warranted, arguing that the BrugmanlLakoff analysis is methodologically
unconstrained.
We will argue that a significant problem with previous approaches is that
they fail to distinguish between what is coded by a lexical expression and
the information that must be derived from context, background knowledge
of the world, and spatial relations in general. That is, previous analyses fail
to take account of meaning construction as a process which relies upon
conceptual integration of linguistic and nonlinguistic prompts, guided by
various global cognitive principles. Hence, we follow recent work in cognitive
linguistics (Fauconnier 1994, 1997; Fauconnier and Turner 1998; Turner
1991, 1996), which posits that formal linguistic expression underspecifies
for meaning. We will further argue that this failure stems in large part from
98 Andrea Tyler and lryvyan Evans

the fact that previous approaches have not developed well-motivated criteria
for (i) distinguishing between distinct senses within a network versus interpre-
tations produced on-line and (ii) determining the primary sense associated
with a preposition.
Our first objective in the present article is to outline what we term a
"principled polysemy framework". This will anchor the semantic network of
over to a foundational conceptual representation (our protoscene), deriving
directly from uniquely human perceptions of and experience with the spatio-
physical world. The protoscene we posit is a highly abstract representation
of a recurring spatial configuration between two (or more) objects. Hence,
details of the physical attributes of the objects involved in a particular spatial
scene will be shown not to involve distinct senses (contra Brugman/Lakoff).
We will argue that many of the distinct senses posited in previous approaches
are produced on-line, as a result of a highly constrained process of integrating
linguistic prompts at the conceptual level. Key to distinguishing our frame-
work from previous ones will be outlining clear, motivated methodology for
determining the protoscene associated with a preposition and distinguishing
between senses that are instantiated in memory versus interpretations pro-
duced on-line. Our second objective is to demonstrate the usefulness of the
framework by providing a complete account of the polysemy exhibited by
over.

2. Previous approaches

2.1. The full-specification approach

The full-specification approach (e.g. Lakoff 1987) characterizes the poly se my

network for over as subsuming distinct but related topographical structures
at a fine-grained level. Each sense is represented by a distinct image-schema;
each image-schema is related through various formal links and transfor-
mations. To see the level of granularity in this model, consider (1) and (2).

(1) The helicopter hovered over the ocean.

(2) The hummingbird hovered over the flower.

Following Langacker's cognitive grammar (Langacker 1987, 1991a, 1991b),

figure-ground relations denoted by prepositions were described in terms of
 The case of over 99

a trajector (TR) and a landmark (LM). Lakoff observed that in a sentence

such as (I) over describes a relation between a TR, the helicopter, and a LM
that is extended, the ocean, while in (2) the relationship is between a TR, the
hummingbird, and a LM that is not extended, the flower. Lakoff argued that
such differences in dimensionality of the LM should be represented as distinct
senses in the semantic network associated with over. He termed this approach
full specification (see Lakoff 1987 for full details and copious examples).
From this view it follows that for a word such as over, there would be a vast
number of distinct senses explicitly specified in the semantic network, in-
cluding many of the metric characteristics of the variety of TRs and LMs,
that can be mediated by the spatial relation designated by over.
While not in principle inconceivable,4 in practice, as Kreitzer observed,
the fine-grained distinctions between instances of over as in (I) and (2),
along with the proposed links and transformations, provide a semantic net-
work so unconstrained that "the model ... [allows] ... across, through and
above all to be related to the polysemy network of over" (1997: 292). Sandra
and Rice (1995), on the basis of their experimental findings, question whether
the actual polysemy networks of language users are as fine-grained as sugges-
ted by models of the sort proposed by Lakoff. This view is echoed forthrightly
in Vandeloise (1990).
Moreover, a Lakoff-type analysis fails to consider that detailed metric
properties of LMs and TRs are often not specified by the lexical forms used
by speakers in their utterances. For instance, the lexical form flower does
not specify whether the entity should be construed as [+ vertical], as a tulip
or calla lily might be, or [- vertical], as a lobelia or a water lily might be.
Thus, in a sentence such as (2), The hummingbird hovered over the flower, it
appears that verticality is not explicitly specified by the semantics of the
LM. This indicates that there must be a sense of over in which the TR is
higher than the nonextended LM and the verticality of the LM is not specified.
Thus, Lakoff's account results in the highly questionable consequence of
positing three senses of over in which the TR is located higher than a
nonextended LM - one which specifies for a vertical LM, one which specifies
for a non vertical LM, and one which does not specify for verticality and
hence subsumes the first two senses. Similarly, Lakoff's model would posit
three additional senses involving a LM which is extended, one which specifies
for verticality (e.g. a mountain range), one which specifies for non verticality
(e.g. an ocean), and one which does not specify for verticality (e.g. the area)
and hence subsumes the first two. s
100 Andrea Tyler and Vyvyan Evans

In essence, by building too much redundancy into the lexical repres-

entation, Lakoff's model vastly inflates the number of proposed distinct
meanings associated with a preposition such as over. An implicit consequence
of this representation is that real-world knowledge as weII as discourse and
sentential context, which are used in the conceptual processes of inferencing
and meaning construction, are reduced in importance, as much of the infor-
mation arising from inferencing and meaning construction is actuaIIy built
into the lexical representation.

2.2. The partial-specification approach

Kreitzer's approach (1997), which we term partial specification, offers a

notable refinement of the BrugmanlLakoff approach because Kreitzer is able
to constrain the number of senses within a polysemy network, in a consistent,
motivated way. Building on work by Talmy (1983), Kreitzer posits that there
are three distinct levels of schematization inherent in the conceptualization
of a spatial scene: the component level, the relational level, and the integrative
level. The component level constitutes conceptual primitives, notions such
as LM, TR, PATH, contact between TR and LM, lack of contact, whether the
LM is extended, vertical, and so on. These combine giving the relational
level. Crucially, for Kreitzer "the relational level schema is taken as the basic
level of 'granularity' representing a sense of a preposition" (1997: 295).
Whereas for Lakoff each additional topographical component constituted a
distinct sense, Kreitzer claims that these individual components apply com-
position ally at the relational level. As such, image-schema transformations
(which aIIow new components to be added to the image-schemas) are no
longer taken as providing a new sense. Rather, image-schema transformations
simply serve to widen the applicability of a particular sense. Examples (3)
and (4) illustrate this point.

(3) The boy climbed over the wall.

(4) The tennis ball flew over the wall.

In (3) there is contact between the TR, the boy, and the LM, the wall,
whereas in (4) there is not. For Lakoff, this distinction warranted two distinct
senses. Kreitzer, by claiming that the sense provided by an image-schema is
defined at the relational level (rather than at the component level), is able to
 The case of over 101

argue that both usages represent only one sense of over. His insight is that
the basic spatial relation between the TR and LM remains unchanged in (3)
and (4), even though the components of the spatial scene may vary contextu-
ally. For Kreitzer, topographical features, such as contact and extendedness
of the LM, are situated at the component level, and consequently do not
delineate distinct senses or image-schemas.
Consequently, Kreitzer argues that the plethora of separate image-schemas
posited by Lakoff can be represented by three image-schemas at the relational
level. The primary sense, which he terms overI, is static, over2 is dynamic,
and over3 is what Kreitzer terms the occluding sense. Examples of these are:

(5) The picture is over the sofa [overI, static sense].

(6) Sam walked over the hill [over2, dynamic sense].6
(7) The clouds are over the sun [overJ, occluding sense].

Although Kreitzer is successful in constraining Lakoff's analysis, his

account faces a significant problem because his three basic senses of over
are arbitrarily connected; they do not share a common TR-LM configuration.
As Lakoff's model with a system of links and transformations has been
abandoned, over now denotes three distinct relations, and it is difficult to see
how Kreitzer's occluding sense of overJ could be related to overI or over2.
In order to appreciate the difficulty, consider (7) in relation to (5) and (6). In
(7), over denotes a relationship in which the TR, the clouds, is beneath the
LM, the sun. In (6), over denotes a dynamic relationship in which the TR is
above the LM only at the midpoint of the TR, but in (5) the TR is stationed
above the LM. It would seem that his claim to polysemy is undermined by
three schemas so distinct as to have little in common. Moreover, he makes
no attempt to account for how overI could give rise to over2 and overJ
respectively.
Secondly, as with Lakoff's full-specification approach, Kreitzer's model
fails to fully address the issue of the contributions of sentential context and
background knowledge. Consider (8) for instance.

(8) The clouds moved over the city.

Kreitzer posits that (8) has two construals as a result of his assumption
that over has both a static and a dynamic relational schema. Construal 1
stipulates that the clouds moved above and across the city, such that they
102 Andrea Tyler and vyvyan Evans

originated in a position not above the city, moved over the city, and came to
be in a position beyond the city. Construal 2 stipulates that the clouds moved
from a position in which they were not over the city, to a position such that
they came to be directly over the city. These construals are diagrammed in
Figures I and 2.

o--..~o
.
D
Figure 1. The clouds moved over the city: construal 1 (after Kreitzer 1997: 305).

0---+0
D
Figure 2. The clouds moved over the city: construal 2 (after Kreitzer 1997: 305).

Kreitzer argues that construal I is the result of over2, while construal 2

represents an integration of move, which contains a path schema as one of
its components, and overl. On this view, the whole meaning of the sentence
depends on which image-schema for over is taken.
In addition to these two construals posited by Kreitzer, however, there is
a third construal in which the clouds move around but remain above the city.
This is represented in Figure 3.
Based on Kreitzer's account, we would expect construal 3 to result from
integration of move with over 1, as the TR is always "above" the city.
However, the problem for Kreitzer's account is that we have two construals,
2 and 3, which would thus not be distinguished image-schematically. How
do we obtain distinct construals without such being coded?
Kreitzer's account is problematic because he is assuming that distinct
construals either result from such being coded by a preposition at the
relational level or arise at the integrative level. But the integrative level simply
"conflates" the two linguistic codes. That is, the path schema of move is
 The case of over 103

0-.0

Figure 3. The clouds moved over the city: construal 3.

added to the static schema of overl, resulting in a dynamic construal. Since

we are able to distinguish construal 3 from construal 2, there must be a
further level of integration at which linguistic codes are elaborated, such that
linguistic underspecification is filled in, providing a variety of construals,
limited only by our perceptual abilities and what is possible in the world.
This is the level of integration we refer to as the conceptual level. Hence, a
fundamental problem with Kreitzer's account (as well as with Lakoff's) is
that it assumes that the rich understanding we obtain about spatial scenes is
derived entirely from what is coded by formal linguistic expression. This
represents a commitment to the view that conceptualization must always
derive from linguistic antecedents. We argue that the ambiguity (given that
there are three construals) arises precisely because move codes a path schema
whereas over does not, and because of what we know about cities and clouds
(cities, unlike walls, for instance, occupy an extended area). Accordingly,
the elements can be integrated in at least three different ways, as indicated
by the three construals. This is testimony to the highly elaborate and rich
process of conceptual integration. The linguistic prompts themselves do not
provide distinct interpretations; these come from our knowledge of what is
possible in the world and our ability to integrate minimal cues to construct a
complex and dynamic conceptualization of a spatial scene. Sentence (9)
illustrates this point.

(9) The clouds moved over the wall.

In (9) construals 1 (Fig. 1) and 3 (Fig. 3) are ruled out, not because over
has both a dynamic and a static sense, but because walls are not extended
landmarks (as noted in Lakoff's analysis), whereas cities are, and moved
codes a path schema. Thus, when the sentential elements are integrated, the
TR follows a path, as designated by moved, such that the TR occupies a
position relative to the LM, as specified by the mental representation for
104 Andrea Tyler and VYvyan Evans

over. The clouds move, neither away from the wall, nor in a vertical manner
without crossing the wall, but from a position prior to the wall to a position
beyond the wall. That this should be so follows from conceptual integration
of the cues prompted by the linguistic elements in the sentence. Accordingly,
we argue that a polysemy network needs to allow for the distributed contribu-
tion of meaning played by all sentential elements, as well as the constraints
imposed by our experience of the world and our ability to construct a rich
and highly dynamic conceptualization based on minimal linguistic cues.
Another problem with Kreitzer's account is that in attempting to constrain
Lakoff's analysis he has significantly understated the amount of polysemy
appropriately associated with over. For instance, many senses touched on by
Lakoff are simply ignored by Kreitzer. We will provide a detailed examination
of the semantic network for over in Section 4. Finally, neither Kreitzer nor
Lakoff attempts a serious account of how he determined which sense of over
should be considered the primary sense. We address this issue in detail in
Section 3.2.
The spirit of our model is coherent with a number of previous analyses
that have addressed the mUltiple meanings associated both with prepositions
(Herskovits 1986; Vandeloise 1991, 1994) and with other linguistic forms
(Cushing 1990, 1991). While these scholars differ from each other and from
us in several key assumptions (e.g. the nature of lexical representation), they
do entertain the possibility that the polysemy exhibited might be best modelled
in terms of a central (or ideal) sense. 7

3. Principled polysemy: the basic framework

3.1. Methodology for determining distinct senses

One of the problems with previous polysemy networks, as noted by Sandra

and Rice (1995), is that there appear to be as many different approaches to
how best to model a semantic network as there are semantic network theorists.
While we accept that all linguistic analysis is to some extent subjective, we
propose here to introduce methodology to minimize the subjective nature of
our analysis. We do so in the hope that other scholars can employ our metho-
dology and test the predictions made by our model. We aim to provide the
basis for replicability of findings, a prerequisite for any theoretically rigorous
study.
 The case of over 105

We suggest two criteria for determining whether a particular instance of

a preposition counts as a distinct sense. Firstly, accepting the standard assump-
tion that the primary sense coded for by prepositions is a particular spatial
relation between a TR and a LM (although we will nuance what "spatial"
means), for a sense to count as distinct, it must involve a meaning that is not
purely spatial in nature and/or in which the spatial configuration between the
TR and LM is changed vis-a.-vis the other senses associated with a particular
preposition. H Secondly, there must be instances of the sense that are context-
independent, instances in which the distinct sense could not be inferred from
another sense and the context in which it occurs. To see how this would work
let us reconsider the sentences in (1) and (2). In (l), over designates a spatial
relation in which the TR, coded by the helicopter, is located higher than the
LM. In (2), over also designates a spatial relationship in which the TR, the
hummingbird, is located higher than the LM, coded by the flower. Neither
instance of over constitutes a nonspatial interpretation, hence neither use
adds additional meaning with respect to the other. By virtue of our proposed
methodology, these instances of over cannot be treated as two distinct senses.
In contrast, examples (10) and (11) do appear to constitute a distinct sense.

( 10) loan nailed a board over the hole in the ceiling.

(11) loan nailed a board over the hole in the wall.

In these sentences the spatial configuration between the TR and LM

designated by over is not consistent with the "above" meaning designated in
examples (1) and (2). In addition, a nonspatial meaning appears to be part of
the interpretation. That is, the meaning associated with over appears to be
that of covering, such that the hole, the LM, is obscured from view by the
TR. Clearly, this notion of covering and obscuring represents an additional
meaning not apparent in examples such as (1) and (2). The fact that the
usage in (10) and (11) brings additional meaning meets the first assessment
criterion for whether this instance counts as a distinct sense.
In terms of the second criterion, we must establish whether the covering
or obscuring meaning can be derived from context. If it can be, then this
instance would fail the second assessment criterion and so could not, on the
basis of the present methodology, be deemed a distinct sense. Assuming that
the primary sense of over involves a spatial configuration between a TR and
LM and that this configuration involves some sense of the TR being higher
than the LM,9 we see no way in which the covering meaning component
106 Andrea Tyler and Vyvyan Evans

associated with over in (10) and (11) can be derived from context. To see
why this is so, contrast this instance with (12), in which the covering meaning
is derivable from context.

(12) The tablecloth is over the table.

The TR, the tablecloth, is higher than (and in contact with) the LM, the
table. As tablecloths are typically larger than tables, and the usual vantage
point from which such a spatial scene would be viewed is a point higher
than the table, the result would be that a substantial part of the table would
be covered and so obscured from view. The interpretation that the table is
covered/obscured could be inferred from the fact that the tablecloth is over
and hence higher than the table, in conjunction with our knowledge that
tablecloths are larger than tables and that we typically view tables from above
the top of the table. Such an inference is not possible in (10) as the spatial
relation holding between the TR and the LM is one that would normally be
coded by below (i.e. the board is below the hole in the ceiling), rather than
by over, given the typical vantage point. Similarly, in (11) the spatial con-
figuration between the TR and LM would normally be coded by something
like next to. In short, unless we already know that over has a covering/
obscuring meaning associated with it, there is no ready contextual means of
deri ving this meaning in sentences such as (10) and (11). From this, we
conclude that the covering/obscuring meaning associated with over in (10)
and (11) constitutes a distinct sense.
The two assessment criteria being proposed are rigorous and, in the light
of future empirical research, may be shown to exclude senses that are legiti-
mately instantiated in the language user's mental lexicon and hence would
have to be adjusted. Nonetheless, without prejudging future findings, we
suggest that this methodology predicts many findings that have already come
to light, and so represents a reasonable approximation for assessing where
we should draw the line between what counts as a distinct sense conventional-
ized in semantic memory, and a contextual inference produced on-line for
the purpose of local understanding. The appeal of such methodology is that
it provides a rigorous and relatively consistent way of making judgements
about whether a sense is distinct, and provides methodology that can be used
in an intersubjective way.
 The case of over 107

3.2. Methodology for determining the primary sense

An equally thorny problem is the question of what counts as the primary

sense associated with a polysemy network. In previous studies of semantic
networks, researchers have assumed that there is a single primary sense
associated with a preposition and that the other senses are derived from this
primary sense in a principled way. We share this assumption. Scholars,
however, have often disagreed about which sense should be taken as primary
(or central). Lakoff (1987) following Brugman (1981), argued that the primary
sense for over is "above and across", and included a path along which the
TR moves, as represented by sentences such as The plane flew over the city.
Kreitzer (1997) disagreed, suggesting that the primary sense (over]) is some-
thing akin to an "above" sense, as in The hummingbird hovered over the
flower. These decisions were primarily asserted rather than being argued for.
Because linguists have simply asserted what constitutes the primary sense
for a particular lexical category, appealing to intuitions and assumptions they
often fail to explicitly articulate, we are in the unfortunate position that Lakoff
(1987) and Kreitzer (1997) can offer equally plausible yet conflicting views
of what the primary sense of over should be.
Sandra and Rice (1995) observed that given the current state of theoretical
development, any analysis of a polysemy network, including what constitutes
its primary sense, is relatively arbitrary, reflecting each analyst's own pref-
erences (or indeed imagination). Langacker, however, has argued persuasively
that there are various kinds of evidence to help us discover and verify the
structure of a complex category (1987: 376). Building on his suggestions we
advance a set of criteria that we believe provides a more principled, inter-
subjective method of determining the appropriate primary sense for individual
prepositions. As with our criteria for determining distinct senses, we see these
criteria as the beginning of a plausible methodology leading to replicability
of findings. We hypothesize that some of these criteria may also be useful
for other classes of words. But because of the particular nature of prepos-
itions - that they code for spatial relations that may not have changed over
many thousands of years (that is, the way humans perceive space seems not
to have changed), and that they are a closed class - the nature of the primary
senses associated with lexical forms is likely to be at least somewhat distinct
from the primary senses associated with word classes such as nouns, adjec-
tives, and verbs.
We suggest that there are at least four types of linguistic evidence that can
108 Andrea Tyler and Vyvyan Evans

be used to narrow the arbitrariness of the selection of a primary sense. We

posit that no one piece of evidence is criterial but, taken together, they form
a substantial body of evidence pointing to one sense among the many distinct
senses being what Langacker (1987: 157) terms the sanctioning sense, from
which other senses may have been extended. The evidence includes (i) earliest
attested meaning; (ii) predominance in the semantic network; (iii) relations
to other prepositions; and (iv) grammatical predictions (Langacker 1987).
Given the very stable nature of the conceptualization of spatial relations
within a language, one likely candidate for the primary sense is the histori-
cally earliest sense. Having examined more than 15 English prepositions
(see Tyler and Evans 2003), we found that the historical evidence indicates
the earliest attested uses coded a spatial configuration holding between the
TR and the LM (as opposed to a nonspatial configuration as in The movie is
over [= complete]). Since English has historically drawn from severallang-
uages, not all prepositions entered the language at the same time and there
are instances of competing, near synonyms, for instance, beneath, below,
and under. In such cases, over a period of time the semantic territory has
been divided among such competing prepositions, but even so, they retain a
core meaning that directly involves the original TR-LM configuration. Unlike
words from many other word classes, the earliest attested sense for many
prepositions is still a major, active component of the synchronic semantic
network of each particle. Over is related to the Sanskrit upan 'higher' as
well as the Old Teutonic comparative form ufa 'above', that is, a spatial
configuration in which the TR is higher than the LM (OED).
Turning to the notion of predominance within a semantic network, by
this we mean that the sense most likely to be primary will be the one whose
meaning components are most frequent in other distinct senses. We have
identified 14 distinct senses associated with over. Of these, eight directly
involve the TR being located higher than the LM; four involve a TR located
on the other side of the LM vis-a-vis the vantage point; and three - covering,
reflexive, and repetition - involve multiple TR-LM configurations. Thus, the
criterion of predominance suggests that the primary sense for over involves
a TR being located higher than the LM.
Within the entire group of English prepositions, certain clusters of prep-
ositions appear to form compositional sets that divide up various spatial
dimensions. Above, over, under, and below appear to form a compositional
set that divides the vertical dimension into four related subspaces (see Tyler
and Evans 2003). Other compositional sets include in and out, on and off, up
 The case of over 109

and down. The linguistically coded division of space and spatial relations is
relativistic in nature, depending largely on construal of the particular scene
being prompted for (Langacker 1987; Talmy 1988,2000). To a large extent,
the label assigned to denote a particular TR-LM configuration is determined
in relation to other labels in the composite set. So, for instance, what we
label as up is partially determined by what we label as down. In this sense,
the meaning of a preposition that participates in a compositional set is partially
determined by how it contrasts with other members of the set. The particular
sense used in the formation of such a compositional set would thus seem to
be a likely candidate as a primary sense. For over, the sense that distinguishes
this preposition from above, under, and below involves the notion of a TR
being located higher than but potentially within reach of the LM. We expand
on this argument in the next section.
The choice of a primary sense gives rise to testable grammatical pre-
dictions. So, for instance, if we recognize that what are now distinct senses
were at one time derived from and related to a pre-existing sense and became
part of the semantic network through routinization and entrenchment of
meaning, we would predict that a number of the senses should be directly
derivable from the primary sense. This is consistent with Langacker's (1987)
discussion of a sanctioning sense giving rise to additional senses through
extension. Any senses not directly derivable from the primary sense itself
should be traceable to a sense that was derived from the primary sense. This
view of polysemy explicitly acknowledges that language is an evolving,
usage-based system. Grammatically, for any distinct sense that is represented
as directly related to the primary sense, we should be able to find sentences
whose context provides the implicature that gives rise to the additional
meaning associated with the distinct sense. We have already discussed this
notion briefly (Section 3.1) when we considered the additional meaning of
covering/obscuring associated with over in (10)-(12). We argued that the
use of over in (10) and (11) revealed additional meaning that could not be
derived from sentential context, while the additional meaning of covering/
obscuring could be derived from context in (12). By the criterion of gram-
matical prediction, (12) constitutes evidence that a likely candidate for the
primary sense associated with over involves the TR being located higher
than the LM, as the distinct covering/obscuring sense can be derived from
this primary sense and certain sentential contexts. Of course, the covering/
obscuring sense is only one of 14; all other senses would have to be tested
against this same criterion.
I 10 Andrea Tyler and lryvyan Evans

3.3. The protoscene

As we said earlier, we assume that English prepositions form polysemy

networks organized around a primary sense. At the conceptual level; the
primary sense is represented in terms of abstracting away from specific spatial
scenes, that is, real-world scenarios such as described by (l3a) and (l3b),
resulting in an idealized spatio-functional configuration.

(13) a. The picture is over the mantel.

b. The bee is hovering over the flower.

We call this abstracted mental representation of the primary sense the

protoscene. It consists of a schematic trajector (TR), which is the locand (the
element located, and in focus), and is typically smaller and movable; a
schematic landmark (LM), which is the locator (the element with respect to
which the TR is located, and in background), and is typically larger and
immovable, and a conceptual configurational-functional relation which
mediates the TR and the LM. In the case of over, the TR is conceptualized as
being proximate to the LM, so that under certain circumstances, the TR could
come into contact with the LM. The functional aspect resulting from this
particular spatial configuration is that the LM (or the TR) is conceptualized
as being within the sphere of influence of the TR (or the LM) (see Dewell
1994, and Vandeloise 1991, 1994 for a discussion of other prepositions).
In our label protoscene, the term proto captures the idealized aspect of
the conceptual relation, which lacks the rich detail apparent in individual
spatial scenes, while the use of scene emphasizes visual awareness of a spatial
scene, although the information included in the image can contain information
from other sense-perceptions. Because protoscenes are abstractions ultimately
arising from recurring real-world spatial scenarios, we will diagram them. to
In our diagrammatic representation of the protoscene posited for over (Fig. 4),
the TR is portrayed as a dark sphere, the LM as a bold line.
The dashed line signals a distinction between the part of the spatial scene
conceptualized as being proximal to the LM (i.e. within potential contact
with the LM) and that which is conceptualized as being distal. The vantage
point for construing the spatial scene is offstage, and external to the spatial
scene. Crucially, the linguistic form over prompts for the conceptual spatial
relation captured by the protoscene.
Two claims warrant more thorough investigation. The first is that the
 The case of over 111

•
Figure 4. The protoscene for over.

spatial configuration holding between the TR-LM is correctly expressed by

the description that over lexicalizes the protoscene depicted in Figure 4,
namely that the TR is above but within a region of potential contact with the
LM. This contrasts with the English preposition above, which we argue
prompts for a conceptual spatial relation in which the TR is higher than but
110t within reach of the LM. The second claim warranting further scrutiny is
that the TR and LM are within each other's sphere of influence.
Dealing with the first claim, using the criterion of relationship to other
prepositions which form a compositional set, consider the instances of over
and above in sentences such as (14).

(14) a. She walked over the bridge.

b. She walked above the bridge.

The sentences in (14) are characteristic of the distinction in English

between over and above. While in (14a) the conventional reading is one in
which the TR, she, is above but within reach (in this particular case, the TR
is in contact with the bridge), most native speakers of English would exclude
possible contact from their reading of (14b). The TR, she, might constitute a
ghostly presence capable of levitation, or the TR might be on a second bridge
higher than the first, but generally English speakers would not interpret the
bridge as the surface being walked upon. These examples strongly suggest
that we are right in positing that over does designate a spatial configuration
in which the TR is in potential contact with the LM.
We turn now to the functional aspect of the protoscene in Figure 4, namely
the claim that the TR and LM are within each other's sphere of influence. A
consequence of being within potential reach of the LM is that the TR can
112 Andrea Tyler and Vyvyan Evans

affect the LM in some way and vice versa. For instance, because of an
independently motivated experiential correlation (Grady 1997), we con-
ventionally understand power and control being associated with an entity
who is higher than the entity being controlled (we will discuss this in more
detail when we deal with the control sense for over). In physical terms we
can only control someone or something, and hence ensure compliance, if we
are physically proximal to the entity we seek to control. If, then, in recurring
human experience, control, and hence the ability to physically influence
someone or something, is dependent upon being higher than and physically
close to the entity we seek to control, we would expect that these notions
can be designated by over but not above. While both over and above desig-
nate spatial relations which are higher than, only over also designates the
functional relation of influence, precisely because part of its spatial con-
figuration involves the notion of potential contact between the TR and LM.
Consider (15).

(15) a. She has a strange power over me. (Lakoff 1987)

b. ?She has a strange power above me.

In terms of a control reading, while over in (15a) is perfectly acceptable,

above in (I5b) is decidedly odd. This suggests that the protoscene for over
does indeed have a functional element of influence between the TR and LM,
as a consequence of its spatial configuration designating potential contact
between the TR and LM (see Vandeloise 1994 for a discussion of the func-
tional nature of prepositions).
This relation places certain maximal constraints on what can count as
over: a spatial relation should be prompted for using the preposition over
only if the spatial relation ranges from a configuration in which there is TR-
LM contact to one in which there is no contact but the TR can be construed
as within potential reach of the LM. While there is strong evidence for
defining over in this way, a review of the many interpretations regularly
assigned to over by speakers of English shows that this representation alone
is inadequate. Hence, there is a need to posit a set of cognitive principles of
meaning construction and meaning extension that will account for the many
additional senses associated with over.
 The case of over 113

3.4. Cognitive Principles

3.4.1. Perceptual analysis and reconceptualization

Mandler (1988, 1992, 1996) argues that a basic aspect of human cognition is
the ability to submit salient (i.e. recurring) real-world scenarios and spatial
scenes to perceptual analysis that gives rise to a new level of conceptualized
information which is stored imagistically in the fonn of an abstract schematiz-
ation, termed an image-schema. 11 Once stored, the image-schema is available
for integration with other conceptualizations, further analysis, and reconcep-
tualization.
Earlier, we used the term conceptualization in a nontechnical way. In order
to distinguish our nontechnical usage from a more sharpened operationalization,
we here introduce the tenn complex conceptualization. A complex concep-
tualization is a constructed representation,12 typically (but not inevitably)
produced on-line. A complex conceptualization represents our projection of
reality (in the sense of lackendoff 1983), and can represent static and relatively
simple phenomena, e.g. The cloud is over the sun, or dynamic and relatively
complex phenomena, e.g. The cat ran over the hill and ended up several miles
away. Our claim is that the integration of linguistic forms with other cognitive
knowledge prompts for the construction of a complex conceptualization.
In our model, the image-schemas representing the spatial configurations
associated with prepositions are termed protoscenes. 13 The primary scene
(i.e. the protoscene) associated with a preposition can be used, in conjunction
with other linguistic prompts (i.e. within an utterance), to prompt for recurring
spatial scenes and real-world scenarios.
Figure 5 represents the complex conceptualization which would be con-
structed in the interpretation of the recurring scenario prompted by sentences
such as (16) and (17).

(16) The rabbit hopped over the fence.

( 17) The boy stepped over the pile of leaves.

At some point, such recurring complex conceptualizations become subject

to reanalysis and hence reconceptualization. 14 We posit that distinct senses
arise as a result of the reanalysis of a particular aspect of such a recurring
complex conceptualization. In other words, the recurring complex concep-
tualization from which a distinct sense originally arises is derivable from the
114 Andrea Tyler and \yvyan Evans

/
/ -- _.- -- -
B

I
/
\
\

•C
I

Figure 5. Schematization of sentences of the type The cat jumped over the wall.

protoscene and thus the distinct sense is related to the protoscene in a

principled manner.
On our analysis, while prepositions themselves do not prompt for dynam-
ism, prepositions do participate in prompting for complex conceptualizations,
which often are dynamic (in the sense that they include motion phenomena).
Minimally varying static spatial scenes can be integrated at the conceptual
level to provide a dynamic sequence. This is analogous to the way in which
movie stills (static images) are flashed onto a screen in sequence to create
the illusion of a moving image, a movie. Hence, we are arguing that prepos-
itions prompt for nondynamic conceptual spatial relations, while maintaining
that such relations can be integrated with other prompts, to create (dynamic)
complex conceptualizations. In sum, we hold that while human concep-
tualization of spatial scenes is rich and dynamic, the available linguistic
prompts underspecify such richness. Meaning is the result of integration of
linguistic prompts at the conceptual level. Thus, the protoscene for over is
integrated in the most felicitous way, given the sentential context, and given
what we know about what is possible in the world.

3.4.2. Ways of viewing spatial scenes

The notion of a vantage point mentioned in the discussion of the protoscene

suggests that how a particular spatial scene is viewed will in large part
determine the functional nature of a particular spatial scene, and thus in what
way it is meaningful. Four distinct issues affect the functional nature of a
particular spatial scene, based on the different ways in which such scenes
can be construed (i.e. "viewed").

(i) Every spatial scene is conceptualized from a particular vantage point.

 The case of over 115

The conceptualizer represents the default vantage point. Accordingly,

the same scene can be construed from many different vantage points
(Langacker [1987] divides this phenomenon into two aspects, per-
spective and vantage point).
(ii) Certain parts of the spatial scene can be profiled (Langacker 1987,
1992). Thus, in the sentence The cat is sitting in the middle of the
circle, the TR, the cat, is conceptualized as being surrounded by the
LM, described by the circle; here the LM is being conceptualized as
a container, and the space encompassed by the LM is being profiled.
In contrast, in the sentence Okay everybody, get in a circle, the outer
edge, or shape of the LM, is being profiled. IS
(iii) Related to (ii) is the fact that the same scene can be construed in a
different way. For instance, in a spatial scene in which a large cloth is
positioned in relation to a table such that the cloth covers the top of
the table, the scene can be construed by focusing on contact between
the cloth and the table. In this case, the scene is likely to be coded in
English by the sentence The tablecloth is on the table. Alternatively,
the relationship between the cloth and the table can be viewed as the
cloth occluding the table from the observer's vantage point. In this
case, the scene might be coded as The cloth is over the table. A less
typical, but perfectly acceptable construal would be to place the table
in focus, in which case the coding would be something like The table
is under the tablecloth.
(iv) The exact properties of the entities which are conceptualized as TR
and LM can vary. In the sentence The hot air balloon floated over
New York City, the LM is conceptualized as vertical and extended;
whereas in the sentence The plane flew over the ocean, the LM is
non vertical and extended.

3.4.3. Atemporality

In advancing the model of word meaning on which we will base our analysis
of over in Section 4, we note, following Langacker (1987, 1991a, 1991b,
1992; see also Talmy 1988, 2000) that prepositions profile (i.e. designate) a
spatio-functional relation that is scanned (i.e. apprehended) in summary
fashion. 16 That is, they do not profile a relation that evolves through time, as
is the case for example with verbs. Verbs profile processes that are scanned
116 Anarea Tyler and Vyvyan Evans

in serial fashion. For instance, in the sentence The boy runs home from school,
the process profiled by run constitutes a process that integrates all the points
occupied by the TR, The boy, which intervene between school and home,
hence the process evolves through time by integrating these sequential
components. The result is a sequential process. This contrasts with the relation
described by a preposition, which does not evolve through time. Prepositions
represent a conceptualized relation holding between two entities (a TR and a
LM), independent of sequentially evolving interdependencies. In this sense,
prepositions can be considered to profile atemporal relations.

3.4.4. Inferencing strategies

We have argued that not all meanings assigned to a preposition, which arise
from interpreting the particle within an utterance, are stored as distinct senses,
and that previous models have often failed to recognize the contribution of
encyclopaedic knowledge and inferencing involved in natural language proces-
sing. In deriving on-line interpretations we employ a number of inferencing
strategies. Because of space constraints we will mention just three of the
most important. In Section 4 we provide a detailed illustration of how these
strategies enable us to produce meaning on-line.

(i) Best fit. Only a tiny fraction of all PQssible spatial relations are coded
by discrete lexical items. In linguistic terms, prepositions represent a
closed class, that is, English speakers have a limited set of linguistic
choices to represent a virtually unlimited set of conceptual spatial
relations. Speakers choose the preposition that offers the best fit
between the conceptual spatial relation and the speaker's communic-
ative needs. The notion of best fit represents a crucial means for
allowing us to fill in information about a particular spatial scene. To
our knowledge, no other linguist has specifically discussed this notion,
but it seems to be a logical extension of the notion of relevance (Grice
1975; Sperber and Wilson 1986).
(ii) Knowledge of real-world force dynamics. Although a spatial scene is
conceptual in nature, in the creation and interpretation of an utterance
the speaker and hearer will assume that all elements in a spatial scene
are subject to real-world force dynamics. 17 For instance, in the inter-
pretation of a sentence such as The cat jumped over the wall, it is
 The case of over 117

assumed the interlocutors will apply their knowledge of the world,

which includes the information that entities cannot float in mid-air
unless they possess the means or ability to do so. General knowledge
of cats includes the information that they cannot hover above walls
and that they are subject to gravity. Hence, any responsible account
of the conceptual system and meaning extension must recognize the
large body of real-world knowledge we bring to bear (often uncon-
sciously) when constructing meaning. Vandeloise (1991) discusses this
in terms of a naIve theory of physics that applies to how humans
conceptualize spatial relations and use language to express those
conceptualizations.
(iii) Topological extension. This strategy involves the notion that the prin-
ciples of Euclidean geometry do not hold at the level of conceptual
structure (Tal my 1988, 2000). Conceptualized space and spatial rel-
ations are not held to be metric notions of fixed distance, amount,
size, contour, angle, and so on. Rather, conceptualized space and
spatial relations are topological in nature, that is they "involve relativ-
istic relationships rather than absolutely fixed quantities" (2000: 170).
Thus, a TR-LM configuration can be distorted conceptually, as long
as the relation denoted by the protoscene remains constant. In applying
this principle to prepositions, we argue that over denotes a relation in
which the TR is above but within reach of the LM. This functional
relationship has sometimes been referred to as the TRlLM being
conceptualized as in each other's sphere of influence (Dewell 1994).
The principle of topological extension allows us to account for ex-
amples in which, on first analysis at least, this relation does not appear
to hold, e.g. The plane flew over the city (the plane is a considerable
distance above the city, yet is being conceptualized as within potential
reach).

3.5. On-line meaning construction

How might on-line meaning construction apply to the protoscene (or indeed
any distinct sense) to produce a contextualized interpretation of a preposition?
To illustrate this process, we will consider the path sense posited by Lakoff
(1987) and Kreitzer (1997). Lakoff termed this the above-across sense, while
Kreitzer called it over2. Both Lakoff and Kreitzer sought to capture the
118 Andrea Tyler and \ryvyan Evans

intuition that over could be employed to designate a trajectory followed by a

TR in which it moves from a position on one side of a LM so that it comes
to be on the other side, as in (18).

( 18) The cat jumped over the wall.

Crucially, they suggested that over codes the trajectory or path as a distinct
sense instantiated in semantic memory. Following the methodology pre-
viously suggested for determining whether a sense is distinct or not, we posit
that in sentences such as (18) the interpretation that the TR follows a particular
trajectory described by "above and across" can be inferred from context.
Based on this methodology, over does not have a distinct above-across path
sense associated with it.
The case for attributing an above-across sense to over in examples such
as (18) relies on implied reasoning which runs as follows: (i) a spatial scene
is conceptualized in which a cat starts from a position on one side of the
wall and comes to be in a position on the other side; (ii) there is nothing in
the sentence, other than over, which indicates the trajectory followed by the
cat; (iii) therefore, over must prompt for an above-and-across trajectory. But
this conclusion is a non sequitur. Simply because a trajectory is not prompted
for by specific linguistic forms (formal expression) does not entail that such
information is absent. To reach this conclusion is to assume that the lack of
formal expression coding trajectory information implicates a lack of trajectory
information per se. On this view, all elements that are salient in the interpre-
tation of a scene must be coded linguistically.
We offer an alternative account that argues that the meaning assigned to
any utterance is radically underdetermined by the lexical items and the
grammatical structures in which they occur. That is, sentential interpretation
is largely the result of various cognitive/inferential processes and accessing
appropriate world knowledge. Consider the conceptualizations prompted for
by the sentence in (18) and contrast this with (19).

( 19) The tree branch extended over the wall.

Lakoff's full-specification account for over would argue that (18) and
(19) represent two different senses of over. For (19) he assumes that over
has a meaning that can be paraphrased as "above" while in (18) over has a
meaning, as already intimated, of "above and across", The implied reasoning
 The case of over 119

for adducing that over in (19) is associated with a static "higher than" sense
runs as follows: in the interpretation prompted for by (19), (i) no motion is
involved hence there is no trajectory; (ii) the branch is located above the
wall; and (iii) the only element that indicates the location of the branch in
relation to the wall is the word over; hence, (iv) over must have an above
sense.
We suggest that it is wrong to conclude that examples (18) and (19)
represent two distinct senses. Rather than representing prepositions as
carrying detailed information about each scene being described, we argue
that they prompt for schematic conceptualizations (a protoscene and other
distinct senses instantiated in semantic memory) that are interpreted within
the particular contexts in which they occur. Under our analysis, a path (or its
absence) is typically prompted for by the verb as it relates to other words in
the sentence. JK
In (18), the verb jumped does prompt for a conceptualization involving
motion, which entails a trajectory. Hence, the interpretation of the above-
across trajectory of the movement in (18) is not prompted for by over (i.e.
the concept of the TR in motion is not a semantic attribute of the protoscene),
nor for any of the other distinct senses associated with over, but rather arises
from the integration of linguistic prompts at the conceptual level. Most of
the information required to integrate the linguistic prompts and construct a
mental conceptualization of the spatial scene is filled in by inferencing and
real-world or encyclopaedic knowledge. In turn, this knowledge constrains
the possible interpretations that over can have in this particular sentence. In
the interpretation of (18), encyclopaedic knowledge (as adduced in part by
the inferencing strategy pertaining to real-world force dynamics) includes
(at the very least): (i) our understanding of the action of jumping, and in
particular our knowledge of the kind of jumping cats are likely to engage in
(that is, not straight up in the air as on a trampoline and not from a bungee
cord suspended from a tree branch extending above the wall); (ii) our
knowledge of cats (for instance, that they cannot physically hover in the air
the way a hummingbird can); (iii) our knowledge of the nature of walls (that
they provide vertical, impenetrable obstacles to forward motion along a path);
and (iv) our knowledge of force dynamics such as gravity (which tells us
that a cat cannot remain in mid-air indefinitely and that if the cat jumped
from the ground such that the trajectory of its path at point B matches the
relation described by over the wall, then it would have to come to rest beyond
the wall, providing an arc trajectory). Thus, we argue that the interpretation
t 20 Andrea Tyler and \ryvyan Evans

regarding the above-across interpretation of the trajectory in sentence (18) is

not prompted for by over, but rather arises from the integration of linguistic
prompts at the conceptual level, in a way that is maximally coherent with
and contingent on our real-world interactions.
We further suggest that part of the general understanding of this particular
sentence involves the interpretation of the wall as an obstacle which the cat
is attempting to overcome. There is an important conceptual connection
between the TR, the cat, and the LM, the wall, that is, the cat and the wall
are within each other's sphere of influence. Given this particular context and
the functional element we have assigned the protoscene, the salient point is
that the cat jumped high enough to overcome the obstacle. The exact metric
details of a spatial relation in a specific spatial scene are filled in by appli-
cation of inferencing strategies. These allow us to construct a likely interpre-
tation, based largely on knowledge gained from recurring daily interactions
with our environment. To make this point more concrete, reconsider Figure
5, which offers an approximate depiction of the complex conceptualization
constructed in the interpretation of (18).
In Figure 5, the various positions occupied by the TR, the cat, along its
trajectory are represented by the three spheres labelled A, B, and C. Notice
that only point B - the point at which the cat is higher than but in potential
reach of the wall - is explicitly mentioned in the sentence (i.e. this point in
the trajectory is explicitly prompted for by the occurrence of over. Points A
and C are inferred from what we know about jumping, cats and walls. The
verb jumped codes self-propelled motion using a solid surface to push off
from; thus, point A is implied as the initial point of the trajectory. The prompts
are integrated in such a way that the trajectory initiated by the verb jump
intersects with point B. Our knowledge of real-world force dynamics fills in
position C. Put another way, if a cat begins at point A and passes through
point B, then given our knowledge of gravity and the kind of jumping cats
are able to do, point C is entailed.
Many spatial relationships exist between the TR and the LM in the com-
plex conceptualization represented diagrammatically in Figure 5; thus, the
speaker has many choices of which relationship between the TR and LM to
mention. For instance, at both points A and C, the cat is beside the wall. The
cat could also be described as jumping near the wall. But, none of these choices
provides a sufficient cue for the construction of the relevant conceptualization
that the cat jumped such that at one point in its trajectory it was higher than,
but crucially within the sphere of influence of, the wall. Alternative prepo-
 The case of over 121

sitions fail to prompt for the key spatial configuration that prompts the listener
to construct the complex conceptualization represented in Figure 5. Given
the conceptualization the speaker wishes to convey, the speaker chooses from
the closed class of English prepositions the one that best fits the relevant
conceptual spatial relation between the TR and LM at one point in the cat's
trajectory, which will, in turn, prompt the appropriate entailments or inferen-
ces. This inferencing strategy is the notion of best fit. Accordingly, we
reiterate that a serious flaw in both the full- and partial-specification ap-
proaches is that neither fully distinguishes between formal expression in
language, which represents certain information, and patterns of concep-
tualization, which integrate information prompted for by other linguistic
elements of the sentence. Over does not itself prompt for an above-across
sense, that is, for a path. We hypothesize that all path or trajectory information
in the examples discussed results from conceptual integration of linguistic
and other prompts, following the notion of best fit, which determines that
the relation designated by the protoscene (and indeed other distinct senses)
will not precisely capture a dynamic real-world spatial relation, which is
constantly changing, but will provide a sufficient cue for conceptualization.
In order to illustrate the strategy of topological extension, we offer
example (20).

(20) There are afew stray marks just above the line.

Example (20) provides, on first inspection at least, a counterexample to

the spatial configuration we proposed for the protoscene associated with over
when it designates a spatial relation in which the TR is above but crucially
within potential contact with the LM. On this view then, we would expect
over, and not above to be employed in sentences such as (20), as this example
is describing a spatial scene in which the TR, afew stray marks, is physically
proximal to the LM, the line.
However, the inferencing strategy of topological extension places less
significance on the absolute metric distance between the TR and LM than on
the functional element associated with a particular sense. That is, the metric
distance between the TR and LM can be extended or contracted if the
functional element holds; in the case of over the TR and LM are understood
as being within each other's sphere of influence. Although the few stray
marks, the TR, are metrically proximal to the line, the LM, there is no contact
and no potential for contact between them. The stray marks are distinct from
122 Andrea Tyler and \t)!vyan Evans

the line and the LM is not within the sphere of influence of the TR. On the
basis of sentences such as She walked above the bridge, in which no contact
between the TR and LM is possible, we hypothesize that the functional
element of the protoscene for above places the focus on the notion of non-
bridgeable distance between the LM and TR. Thus, the relation in (14b) is
best designated by above. This analysis is supported if we attempt to use
over in place of above, as in There are a few stray marks over the line,
which presents the ambiguous interpretation that the marks are in contact
with the line and potentially obscuring parts of it. This interpretation arises
from the covering sense, which we will address later.
Grice (1975) noted with his maxim of manner that in everyday con-
versation speakers generally try to avoid ambiguity, unless there is a purpose
for the ambiguity. To avoid possible ambiguity, the inferencing strategy of
attempting best fit in the choice of lexical item suggests that the speaker will
choose the protoscene (or particular sense) that best facilitates conceptualiz-
ation of the scene he or she intends the listener to construct. In light of the
strategies of topological extension and best fit, we argue that above is the
most felicitous choice to prompt for the complex conceptualization that
involves a LM (a line), and a TR (stray marks) that is higher than and not in
contact with the LM, as attested by (20).19

3.6. Pragmatic strengthening

Earlier we presented a method for establishing when a sense is distinct and

hence putatively instantiated in semantic memory. Given our assumption that
the distinct senses associated with a particular preposition are related to one
another in a principled way, one of our purposes is to understand both how
and why new senses associated with a particular preposition came to be
derived. Since what are now conventionalized senses at one time did not
exist, we seek to explain how they are related to the protoscene. Our hypoth-
esis is that all the senses associated with the preposition over were at one
time derived from the protoscene or from a sense that can be traced back to
the protoscene for each individual preposition. 20
Grady (1997) has shown in detail that tight correlations in experience can
lead to conceptual associations between two quite distinct and otherwise
unrelated concepts. For instance, on a daily basis we experience recurring
correlations between quantity and vertical elevation. When a liquid is added
 The case of over 123

to a container or when more objects are added to a pile, an increase in quantity

correlates with an increase in height. Grady has suggested that correlations
of this kind result in lexical items relating to vertical elevation developing a
conventional reading in which they denote quantity, as in sentences such as
The prices have gone up, where gone up refers not literally to an increase in
vertical elevation, but rather to a quantificational increase.
A number of scholars who have investigated the meaning extension of lexi-
cal items have observed that inferences deriving from experience (analogous
to the situation just discussed) can, through continued usage, come to be
conventionally associated with the lexical form identified with the implicature
(see e.g. Bybee, Perkins and Pagliuca 1994; Evans 2000; Fleischman 1999;
Hopper and Traugott 1993; Svorou 1993; Traugott 1989). Following Traugott,
we term this process pragmatic strengthening, and it results in the associa-
tion of a new meaning component with a particular lexical form through the
continued use of the form in particular contexts in which the implicature
results. New senses derive from the conventionalization of implicatures
through routinization and the entrenchment of usage patterns.
Recurring implicatures that come to be conventionalized can result either
from independently motivated experiential correlations (as with quantity and
vertical elevation) or from construing a spatial scene in a certain way, that is,
from a new vantage point. Examples of each of these will be presented in
Section 4.
Prepositions can also be employed to express figure-ground relations
between nonphysical elements. In a sentence such as A feeling ofdread hung
over the crowd, the TR, dread, is an emotion rather than a physical entity.
We argue that this use is possible because over conveys a specific relationship
between an emotion, the TR, and the crowd, the LM; one in which the crowd
is being affected by, or within the sphere of influence of, the feeling of dread.
Being within the sphere of influence of a physical TR means the LM can
potentially be affected by the TR, as in Rain clouds hung over the city all
week. In A feeling of dread hung over the crowd, the TR is not physically
located higher than the LM, but because over has the functional notion of a
sphere of influence associated with it, over can be employed to designate
relations between nonphysical entities.
124 Andrea Tyler and Vyvyan Evans

3.7. The conceptual significance of syntax

Our model takes the view that formal aspects of language, such as syntactic
configurations, have conceptual significance. As syntax is meaningful, in
principle in the same way as lexical items, it follows that differences in
syntactic form reflect a distinction in meaning (Lakoff 1987; Langacker 1987,
1991 a, 1991 b; Sweetser 1990; Talmy 1988, 2000). We are using the generic
term "preposition" to describe the linguistic forms we are studying. But this
term subsumes a number of formal distinctions characterized by prepositions,
verb-particle constructions (or phrasal verbs), adpreps (which are adverbial
in nature, and do not overtly code a LM, e.g. the race is over; they are
discussed in Section 4), and particle prefixes (bound spatial particles as in
overflow, overhead, and so on).21

4. Beyond the protoscene: additional senses in the semantic network

Our methodology for determining distinct senses points to the conclusion

that in addition to the protoscene a number of senses must be instantiated in
semantic memory (contra Ruhl's 1989 monosemy framework).22 For instance,
we see no direct way of deriving the interpretation of completion normally
assigned to over in the sentence The movie is over (= finished), suggesting
that such an interpretation is due to a distinct completion sense associated
with over being stored in long-term memory. We now turn to a consideration
of the distinct senses, other than the protoscene conventionally associated
with the preposition over.
Figure 6 is a preview of the remainder of this paper; it represents our
proposed semantic network for over, subsuming a total of 14 distinct senses,
including the protoscene. Each distinct sense is shown as a dark sphere,
which represents a node in the network; the protoscene occupies a central
position indicating its status as the primary sense. In some instances our
representation of the semantic network depicts a distinct, conventionalized
sense arising from the conceptualization prompted for by another conven-
tionalized sense, rather than directly from the protoscene. For instance, in
the network represented in Figure 6, the "excess" sense is represented as
arising from the conceptualization associated with the "more" sense rather
than arising directly from a conceptualization in which the protoscene of
over occurs. Figure 6 represents the claim that reanalysis of conceptualiz-
 The case of over 125

2.B 2.C
Above-and- Completion
beyond
(excess I)
• 2.D 4.A

• 3. Focus-of-
attention

.-+----~D'_

2.A
On-the-
/
other-side- trajectory
of cluster

5. 6.
[ Reflexive

5.A
More
•
5.B Preference • 6.A
Control Repetition

• 5.A.l
Over-and-above
(excess IT)

Figure 6. The semantic network for over.

ations is potentially recursive and that a distinct sense can be the result of
multiple instances of reanalysis. Moreover, we believe that a complex concep-
tualization, such as the one represented in Figure 5, can be submitted to
mUltiple reanalyses and thus give rise to several distinct senses. When a
complex conceptualization gives rise to mUltiple senses, we term the set of
senses a "cluster of senses". A cluster of senses is denoted in our represen-
tation of a semantic network by an open circle. A single distinct sense is
represented by a dark sphere.
126 Andrea Tyler and VYvyan Evans

4.1. The A-B-C Trajectory Cluster

The four distinct senses in the A-B-C trajectory cluster (on-the-other-side-

of, above-and-beyond (excess I), completion, and transfer) all derive from
reanalyses of the complex conceptualization depicted in Figure 5, in which
the verb designates point A as a starting/push-off point. All involve TRs that
cannot hover and must return to ground; involve LMs construed as impedi-
ments to forward motion; and use over to designate the key spatial/functional
configuration (i.e. the TR being higher than the LM and both being within
each other's spheres of influence). This complex conceptualization, although
profiling a sequentially evolving process, is subject during reanalysis to
conceptualization in summary format. That is, although points Band C never
exist simultaneously in the world (because a TR such as a cat could not
occupy two such positions simultaneously), when such a spatial scene is
conceptualized in summary format, point C can be related to point B, and
hence the lexical form that prompts for point B can come, through entrench-
ment, to be employed to reference senses related to point C.

4.1.1. The on-the-other-side-of sense (2.A)

An unavoidable consequence of the unique trajectory prompted by sentences

analogous to (18) is that when the motion is complete the TR is located on
the other side of the LM relative to the starting point of the trajectory.
Although point C in Figure 5 and its relation to point A are not part of the
protoscene for over (and cannot be derived from the protoscene absent the
particular properties of the verb and TR discussed above), the on-the-other-
side-of sense has come to be associated with certain uses of over that are not
derivable from context. Consider (21).

(21) Arlington is over the Potomac River from Georgetown.

Notice in this sentence that the verb, is, fails to indicate any sense of
motion. In our model, the verb typically codes for motion and hence prompts
for a trajectory. Thus, the lack of motion coded by is, in turn, results in
failure to prompt for a trajectory. If there is no trajectory, there is no beginning
or endpoint, hence no principled way of deriving an on-the-other-side-of
sense from this sentential context. Native speakers nevertheless will normally
 The case of over 127

interpret this utterance such that Arlington is understood to be located on the

other side of the Potomac River from Georgetown. Consequently, over must
have a context-independent on-the-other-side-of sense associated with it.
Accordingly, the two criteria for establishing that a sense is distinct have
been met. The on-the-other-side-of sense adds meaning not apparent in the
protoscene and the use in (21) is context-independent.
We hypothesize that this distinct sense came to be instantiated in memory
as a result of reanalysis of the complex conceptualization represented in
Figure 5, specifically, the priviJeging of the consequence of the jump - that
the TR ends up on the other side of the LM. In addition, this conceptualization
involves a shift in vantage point from being offstage (Langacker 1992) to
being in the vicinity of point A. The default vantage point specified in the
protoscene for over, Figure 4, is offstage. Previously, we noted that spatial
scenes could be viewed from a number of possible vantage points, and these
different vantage points could give rise to different construals of the same
scene.
The on-the-other-side-of sense is illustrated in Figure 7. The eye icon on
the left represents the vantage point, the vertical line the impediment and the
dark sphere the TR.
Further evidence for this sense comes from examples like (22).

(22) Arlington is just over the river.

The sentence in (22) is felicitous only if the construer (the vantage point)
is located in the vicinity of point A (in Fig. 5) and Arlington is construed as
point C. Thus, the reanalysis of over which results in the on-the-other-side-
of sense involves two changes vis-a-vis the protoscene - the privileging of
point C and interpreting it as the point at which the TR is located, and a shift
in vantage point such that the construer is located in the vicinity of point A.
While the on-the-other-side-of component (point C in Fig. 5) is correlated in
experiential terms with arc-shaped trajectories and jumping over (i.e. higher
than) obstacles by TRs such as cats, without the shift in vantage point this

Figure 7. The on-the-other-side-of sense.

I •
128 Andrea Tyler and Vyvyan Evans

experiential correlation cannot be construed. We hypothesize that through

the use of over in contexts where on-the-other-side-of is implicated, this
meaning has come to be conventionally associated with over as a distinct
sense, a process we term pragmatic strengthening.
The on-the-other-side-of sense is highly productive in English, as attested
by the examples below. Notice that in neither of the following do we con-
ventionally obtain the reading in which the TR is physically higher than the
LM or that jumping or moving is involved. 23

(23) The old town lies over the bridge.

(24) John lives over the hill.

Moreover, examples such as (24), which have been described as having

endpoint focus, are reminiscent of the examples offered in Lakoff's (1987:
423) analysis for over, as evidence for an above-across sense. 24 We suggest
that misanalysis of the on-the-other-side-of sense contributed to a path above-
across sense being posited by earlier analyses. 25

4.1.2. The above-and-beyond (excess I) sense (2.B)

In (25) and (26) over is used as predicted by the protoscene but with the
additional implicatures that the LM represents an intended goal or target and
that the TR moved beyond the intended or desired point.

(25) The arrow flew over the target and landed in the woods.
(26) Lissa just tapped the golf ball, but it still rolled over the cup.

Given general knowledge of shooting arrows and targets, most speakers

would assume that whoever shot the arrow intended to hit the target but
aimed too high. The movement of the arrow, the TR, was above and beyond
the LM, or in excess of what the agent intended. Similarly, given general
knowledge of the game of golf and the goals of people who engage in the
game, most speakers would assume that the agent (Lissa) intended that the
movement of the ball (the TR), which she initiated with a tap, would result
in the ball going into the cup, the LM. Thus the movement of the ball was
above and beyond, or in excess of, what the agent intended.
The basic spatial configuration and trajectory followed by the TR is
 The case of over 129

identical to that associated with the protoscene in the context of a verb

depicting forward motion. But in sentences such as The cat jumped over the
wall, the TR's movement beyond the LM is presumed to be intentional, while
in sentences such as (25) and (26) the LM is construed as the target or goal
and the presumed intention is to have the TR come into contact with the
target. When the TR misses the target, it goes above and beyond the LM.
Going above and beyond the target is conceptualized as going too far or
involving too much. The implicatures of (i) the LM being construed as the
target/goal and (ii) the TR passing over the LM as going beyond the target/
goal have been reanalysed, resulting in a distinct sense being added to the
semantic network. Evidence for this sense being distinct comes from sen-
tences such as (27), in which the sense cannot be derived from context.

(27) Your article is over the page limit.

In this sentence, over cannot felicitously be interpreted as physically higher

than, or even on-the-other-side-of. Rather, the interpretation seems to be that
there is an established or "targeted" number of pages for the article and that
the actual number of pages "went beyond" that target.
Figure 8 diagrams the above-and-beyond (excess I) sense, representing
the LM as a bull's-eye target and highlighting the salient "beyond" portion
of the trajectory. (Our analysis provides for a second source of an excess
sense associated with over. This sense and its implication for the model are
discussed later.)
We emphasize that we are not claiming that the semantic network contains
criterial senses: that is, we are not suggesting that all uses of over will
absolutely reflect one sense or another. Often, specific uses of a preposition
will contain flavours of more than one sense, imbuing a particular reading
with complex nuances of meaning and providing both intra- and inter-hearer
differences in interpretation. Equally, we are not suggesting that application
of the model outlined in Section 3 will mechanistically provide a single,

- - - --~
" ... I
I'

Figure 8. The above-and-beyond (excess I) sense.

130 Andrea Tyler and Vyvyan Evans

unique derivation for each distinct sense, based ultimately on the protoscene.
We do not want to posit a simplicity rubric which claims that there is one
correct analysis and deny that there may be many means of instantiating a
distinct sense in memory. We find no strong evidence that human concep-
tualization and cognition is constrained by such a dictum (contra the wide-
spread view adopted in formalist approaches to meaning in the generative
tradition; for a critique of such views see Langacker 1991a: Chapter 10, and
the discussion of the generality fallacy in Croft 1998).
At this point we see no principled reason to rule out the possibility that
an excess interpretation might arise through an alternative route, as repres-
ented in the network by the over-and-above (excess IT) sense (5.A.1). We in
fact hypothesize that some speakers might derive an excess interpretation
through one route while others arrive at it through the other. Still others may
use both routes; the two resultant senses would then serve to inform each
other in various ways. We further argue that it is inappropriate to treat this
flexibility (or redundancy) as evidence that our model is flawed. Nor should
an alternative analysis of the derivation of a particular sense be taken to
constitute a counterexample to the overall model being posited. We see this
flexibility (and redundancy) as an appropriate reflection of the richness of
human cognition and the way in which experience is meaningful to us as
human beings.

4.1.3. The completion sense (2.C)

When over is integrated into a complex conceptualization, such as described

by Figure 5, the inferred shape of the trajectory has an endpoint C. The
endpoint of any trajectory (which represents the process of moving) is
commonly understood as representing the completion of the process.
We suggest that the completion sense associated with over has arisen as
a result of the implicature of completion being reanalysed as distinct from
the complex conceptualization represented in Figure 5. Once reanalysis has
taken place, the final location resulting from motion correlates with the
completion of motion, the distinct sense comes to be associated with the
form over in the semantic network via pragmatic strengthening.

(28) The cats jump is over [= finished/complete).

 The case of over 131

We suggest that the meaning component of completion results from re-

analysis of the spatial location of the TR as standing for an aspect of a process.
In (28), for example, the endpoint of the motion through space over an
impediment (i.e. the location at which the TR comes to rest) is interpreted as
the completion of the movement. In this instance the completion sense is not
describing a spatial relation but rather an aspect of a process. This is reflected
syntactically by the fact that the completion sense does not mediate a TR-
LM configuration in which the preposition is sequenced between the TR and
the LM, as illustrated by example (28). The completion sense, in formal
terms, is represented not by a preposition but rather by what we are terming
an "adprep" (Bolinger 1971; O'Dowd 1998).26
The completion sense differs crucially from the on-the-other-side-of sense
in that the latter focuses on the spatial location of the TR when the process
is completed (see Fig. 9) while the former focuses on interpreting point C as
the end of the motion or process. We tentatively hypothesize that an adprep
will always arise when the reanalysis involves interpreting the location of
the TR as an aspect of a process.
Figure 9 diagrams the completion sense. The dark sphere on the left
represents the location of the TR at the beginning of the process. The large
sphere on the right, which is in focus, represents the end point or completion.

••
Beg,"mng •
I
I
I
I
/
,-

----"1\
I End

Figure 9. The completion sense.

4.1.4. The transfer sense (2.0)

A consequence of the conceptualization represented in Figure 5 gives rise to

the transfer sense. Consider the following examples.

(29) SaLLy turned the keys to the office over to the janitor.
(30) The teLLer handed the money over to the investigating officer.

In these sentences, the conceptualization constructed is of a TR moving

132 Andrea Tyler and \ryvyan Evans

from one point to another. This follows from the conceptualization schema-
tized in Figure 5, in which an implicature of transfer arises, a consequence
of understanding the scene as one involving the transfer of a TR from one
location, point A, to a new location, point C (see Fig. 10). We suggest that
change in location of an entity is experientially correlated with transfer of
the entity; change in position often gives rise to the implicature that transfer
has taken place. Via pragmatic strengthening, this implicature is convention-
alized as a distinct meaning component and instantiated in the semantic
network associated with over as a distinct sense. As with the completion
sense, the transfer sense involves the reanalysis of the trajectory or process.
Again, in formal terms, over is represented not by a preposition but by an
adprep. In Figure 10, the TR has been transferred from the left side of the
impediment to the right side, as represented by the dark sphere, which is in
focusY

Figure la. The transfer sense.

4.2. The covering sense (3.)

In our basic definition of TR and LM we noted that the typical situation is

for the TR to be smaller than the LM, when the TR and LM are physical
entities (although as we have seen, it is not inevitable that such is the case).
All the senses and interpretations examined thus far have assumed that the
TR is smaller than the LM. This default ascription is also represented in the
protoscene we posited for over. However, there are instances in the real world
in which the object that is in focus (the TR) is larger or perceived to be
larger than the locating object (the LM). Such a situation is described by the
sentence in (31).

(31) Frank quickly put the tablecloth over the table.

Given our normal interactions with tables and tablecloths - we sit at tables
 The case of over 133

or walk past them such that both the table and the tablecloth are lower than
our line of vision - it follows that our typical vantage point is such that when
a tablecloth is over the table we perceive it as covering the table. This being
so, the vantage point is not that depicted in the default representation of the
protoscene, in which the viewer/construer is offstage. Rather the vantage
point has shifted so that the TR is between the LM and the cons truer or
viewer. The perceptual effect of having the TR physically intervene between
the viewer and the LM is that the TR will often appear to cover the LM or
some significant portion of it. 2H
In accordance with the position outlined previously - that spatial scenes
can be viewed from different vantage points - the covering interpretation
results from having a particular vantage point from which the situation is
construed. When a shift in vantage point occurs. the conceptualization con-
structed is likely to involve an additional implicature not part of the interpre-
tation when the default vantage of the protoscene is assumed. In sum, we
are arguing that the conceptualization constructed in the normal interpreta-
tion of (31) involves two changes from the default representation of the
protoscene: first, the TR is perceived as being larger than the LM and second.
the vantage point has shifted from offstage to higher than the TR. 29
The covering implicature has been reanalysed as distinct from the spatial
configuration designated by the protoscene (see Fig. 11). As noted with
examples (10) and (11), when over prompts for a covering sense, the TR
need not be construed as being located higher than the LM; hence. the
covering sense must exist independently in semantic memory.30

® Vantage point

.-- TR

Figure 11. The covering sense.

- LM
134 Andrea Tyler and \'yvyan Evans

4.3. Above and proximaL

4.3.1. The examining sense (4.)

As noted earlier, any spatial scene can be viewed from a variety of vantage
points. The construal that gives rise to the examining sense is the result of a
shift from the default (i.e. offstage) vantage point. In particular, we argue
that in the scene associated with the examining sense, the vantage point is
that of the TR, and further that the TR's line of vision is directed at the LM.
How might this construal arise? Consider the following sentence.

(32) PhyLLis is standing over the entrance to the underground chamber.

Here over is being used as designated in the protoscene and is mediating

a spatial relation between the TR, PhyLlis, and the LM, the entrance to the
underground chamber, in which the TR is higher than but proximal to the
LM. A consequence of Phyllis's being in this physical relation to the entrance
is that she is in a position to carefully observe the entrance. An important
way of experiencing and therefore understanding the act of examining is in
terms of the examiner being physically higher than but proximal to the object
being examined. Many recurring everyday examples of looking carefully at
objects involve the human eyes being higher than the object being scrutinized,
for example, examining tools, jewellery, a written text, or wounds on the
body. Further, if an object is not proximal to the viewer, it is generally not
possible to see the object clearly and therefore not possible to examine the
object thoroughly. The experiential correlation between proximity and poten-
tial thoroughness is reflected in sentences such as (33) and (34).

(33) I'LL give the document a close examination.

(34) I'LL give the manuscript a close read.

Two experiential correlates of examining are the viewer being located

above the LM and in proximity to the LM. Further, the functional aspect
associated with the protoscene is that there is a conceptual connectedness
between the TR and LM, i.e. the notion of sphere of influence. In this case,
the connection is construed as that between the examiner and the examined.
Because the protoscene for over contains these elements - a TR higher than
a LM, proximity between the TR and LM, and a conceptual connectedness
 The case of over 135

between the TR and LM - which match the physical correlates necessary for
examination, over is a likely candidate for developing an examining sense.
But this is not the entire story. Notice that the use of over in (32) does not
prompt for the interpretation that Phyllis is examining the entrance, only that
she is located such that she could examine it. For the examining sense to
arise, the scene must contextually imply examination. Put another way,
examination must be an implicature deriving from the particular linguistic
prompts in a given sentence. Consider (35).

(35) Mary looked over the manuscript quite carefully.

The normal interpretation of this sentence is something like "Mary

examined the manuscript". In this sentence, the TR, Mary, is physically higher
than and in proximity to the LM, the manuscript. Thus, the TR and the LM
are in the spatial configuration associated with the protoscene for over. In
addition, the TR is construed as directing attention toward the manuscript.
(This construal arises from our knowledge of the act of looking (it involves
looking at something) and our knowledge of humans (often when they are
looking, it is for some purpose).
This additional meaning element of directing attention towards the LM is
essential to the examining sense (see Fig. 12). Now consider sentence (36).

(36) The mechanic looked over the train s undercarriage.

The normal reading is that the mechanic examined the train's under-
carriage, but for such examination to occur, the mechanic, the TR, must be
physically underneath the train. In other words, in this conceptualization, the
TR is under the LM. Clearly, in this situation, there is no way of predicting
that over has associated with it an examination reading, given that the TR-
LM spatial configuration does not correspond with that normally associated

r------,
~ _____ .J

Figure 12. The examining sense.

136 Andrea Tyler and \-}'vyan Evans

with over, the very configuration that motivated the implicature of examin-
ation in the first place. This is good evidence, therefore, that the contextual
implicature of examination has been instantiated as a distinct sense in the
network via pragmatic strengthening. Hence, examination results from con-
struing a scene in a particular way. This being so, speakers are free to use
this examination-meaning component in the absence of the TR-LM con-
figuration which gave rise to the implicature of examination initially.

4.3.2. The focus-of-attention sense (4.A)

Sentences (37) and (38) illustrate what we call the focus-of-attention sense.
Notice that in (37) over can be paraphrased by about.

(37) The little boy cried over his broken toy.

(ef. The little boy cried about his broken toy.)
(38) The senator presided over the opening ceremonies.

In (37) and (38) the LM is the focus of attention. This sense is closely
related to the examining sense from which it derives. In the examining sense,
the vantage point is that of the TR, while the LM is physically below and
proximal to the TR. We further posited that the TR must be construed as
directing attention toward the LM. A natural consequence of the examining
sense is that the object being examined, the LM, is the focus of the TR's
attention. This natural consequence of examining has been privileged and
reanalysed as distinct from the spatial scene in which it originally occurred
(see Fig. 13), and via pragmatic strengthening, conventionalized as a distinct
sense. (Fig. 13 differs minimally from Fig. 12; here the LM is in focus.)
Once this sense has been instantiated in memory, nonphysical TRs and
LMs can be mediated by this sense.

-
Figure J3. The focus-of-attention sense.
 The case of over 137

(39) The committee agonized over the decision.

(40) The committee chair watched over the decision-making process.

4.4. The vertical elevation or "up" cluster (5.)

Four distinct senses fall under this cluster, as can be seen in Figure 6. Each
arises from construing a TR located physically higher than the LM as being
vertically elevated, or up, relative to the LM. Being up entails a particular
construal of the scene in which upward orientation is assigned to the TR (see
Fig. 14).

j.

Figure 14. The up cluster.

I
This construal arises frequently in real-world experiences associated with
the conceptual spatial relation over. For instance, in order to move over and
beyond many LMs, movement from a physically lower location to a physi-
cally higher location is often necessary, i.e. vertical elevation of the TR occurs.
Furthermore, an upward orientation is not typically construed in a neutral
way. As Clark (1973) and Lakoff and 10hnson (1980) have observed, an
upward orientation is meaningful in human experience. An element in a
vertically elevated position is often experienced as being positive or superior
to an element in a physically lower position. Notice that there is nothing in
the protoscene of over, i.e. of a TR being higher than the LM, that entails
this construal: in the scene described by The picture is over the mantel, the
picture is not construed as being in a better or superior position vis-a-vis the
mantel.
138 Andrea Tyler arui \ryvyan Evans

4.4.1. The more sense (S.A)

As noted in the discussion of experiential correlation, vertical elevation and

quantity are correlated in our experience. When there is an addition to the
original amount of a physical entity, the height or level of that entity often
rises. Because over can be construed as relating to a TR which is physically
up with respect to a LM, and vertical elevation correlates in experiential
terms with greater quantity, an implicature associated with having more of
some entity is associated with being over. This implicature is conventionalized
(via pragmatic strengthening), as attested by example (41).

(41) Jerome found over 40 kinds of shells on the beach.

The normal interpretation of over in this context is "more than". The LM,
40 kinds of shells, is interpreted as a kind of standard or measurement. The
TR is not actually mentioned; in interpreting the sentence, we infer that the
TR is shell types 41 and greater. If over were interpreted in terms of the
protoscene in this sentence, we would obtain a semantically anomalous
reading in which the additional shells would be understood as somehow being
physically higher than the 40 kinds actually mentioned in the sentence. Again,
we see no direct way in which this interpretation can be constructed from
the protoscene and the sentential context alone. Moreover, there is no direct
correlation between the concept of more types and vertical elevation. The
concept here is more variety not greater quantity of shells. We argue that
the "more" sense associated with over has arisen because of the indepen-
dently motivated experiential correlation between greater quantity and greater
elevation. Because of this experiential correlation, the implicature of greater
quantity comes to be conventionaIly associated with over (which in terms of
the designation prompted by the protoscene, has a greater height meaning,
and hence also implicates greater quantity).
The implicature of greater quantity or more comes to be reanalysed as
distinct from the conceptualization of the physical configuration that origi-
naIly gave rise to it (see Fig. 15). Once reanalysis has taken place, the distinct
sense comes to be associated with the form over, in the semantic network.
 The case of over 139

Figure 15. The more sense.

ii
4.4.2. The over-and-above (excess II) sense (S.A.l)

The over-and-above (excess II) sense is closely related to the more sense. It
adds an interpretation of "too much" to the "more" construal. We believe that
a likely origin for this sense is the reanalysis of scenes involving containment,
such as those described in (42) and (43).

(42) The heavy rains caused the river to flow over its banks.
(43) Lou kept pouring the cereal into the bowl until it spilled over and
onto the counter.

In these scenarios the LMs are containers and the TRs are understood as
entities held by the container. When the level of liquid or cereal (or whatever)
that has been placed in the container is higher than but within reach of the
top of the LM, then the amount constitutes more than the container can hold.
A consequence of the capacity of a container being exceeded is that more of
the TR becomes an excess of the TR, which results in spillage. In sum, more
of the TR, the water, equals a higher level of water. Too much more of the
TR results in a mess (see Fig. 16).
This node in the semantic network represents a second potential source
for the general notion of excess associated with certain uses of over. We see
subtle but distinguishable differences between the excess I sense, which seems
to us to be more closely tied to motion along a path and the interpretation of

Figure 16. The over-and-above (excess 11) sense.

140 AtuJrea Tyler and \yvyan Evans

going beyond a designated point, and the excess 11 sense, which seems to be
more closely related to exceeding the capacity of containers and exceeding
what is normal. For instance, in a compound such as overtired, it may be
that the conceptualization involved is not that an expected level of tiredness
is a goal that is missed, but rather, an expected or normal capacity for tiredness
has been exceeded. Consider (44).

(44) The child was overtired and thus had difficulty falling asleep.

In our interpretation of this sentence the child is conceptualized as having

a certain capacity for activity; the child is conceptualized as a container and
her or his activities are conceptualized as filling the container. When the activity
level reaches that capacity, the child is tired and the normal response to that
tiredness is to fall asleep. If the activity level exceeds the normal capacity,
the child becomes too tired, which results in irritability and difficulty going
to sleep.]' In this example we might construct a "more" conceptualization
for over, or we might construct an "excess" interpretation (which provides not
just a more meaning, but the additional too-much-more meaning) for over.

4.4.3. The control sense (5.B)

A third experiential correlate associated with vertical elevation is the phenom-

enon of control or power. This meaning component associated with over is
illustrated by (45) (from Lakoff 1987).

(45) She has a strange power over me.

Clearly, this sentence does not mean that the TR, she, is higher than but
within reach of me, the LM. Rather, the conventional interpretation derived
from such an example is that the TR exerts influence, or control over the
LM (as observed earlier). This meaning could not be derived from context,
and is therefore suggestive, given our methodology, that this constitutes a
distinct control sense instantiated in semantic memory. How then did the
control sense derive from the semantic network associated with over? We
suggest that this sense is due to an implicature becoming conventionally
associated with over, from an independently motivated experiential correlation
between control and vertical elevation.
 The case of over 141

For most of human history, when one person has been in physical control
of another person, control has been experienced as the controller being
physically higher. In physical combat, the victor, or controller, is often the
one who finishes standing, in the up position; the loser finishes on the ground,
physically lower than the controller. Hence an important element of how we
actually experience control (and presumably from where the concept itself is
derived) is that of being physically higher than that which is controlled.

(46) The fight ended with John standing over Mac. his fist raised.

Further, within the physical domain, the physically bigger, up, often
controls the physically smaller, down. Within the animal kingdom, a wide-
spread signal of the acknowledgment of power or status is for the submissive
animal to adopt a position in which its head is physically lower than the
head of the dominant animal. In experiential terms then, control and vertical
elevation are correlated. We suggest that because of an independently motiv-
ated experiential association between control and being vertically elevated,
there is an implicature of control associated with over.
Nonetheless, if control were understood only in terms of vertical elevation,
we would expect that the English preposition above should also implicate
control. But as (47) demonstrates, this is not the case.

(47) ?She has a strange power above me [control reading].

To exert control in order to affect the subject's actions and thus guarantee
compliance, one must be physically proximal to the subject. In experiential
terms, there are two elements associated with the concept control; the first is
up, and the second is physical proximity. As we have argued throughout this
article, while the protoscene for over designates a TR being physically higher
and proximal to the LM, there is good evidence for supposing that above
designates that the TR will be physically higher but precludes physical
proximity. In linguistic terms, we would expect over to develop a control
reading. The linguistic usage, then, accords with how we actually experience
(see Fig. 17: the spiral shape denotes that the TR [sphere] controls the LM
[vertical lineD.
As we have been arguing, distinct senses, once instantiated in semantic
memory, can be employed in situations that did not originally motivate them,
as a consequence of being instantiated as distinct within the semantic network.
142 Andrea Tyler and vyvyan Evans

A
(:~:~-=~~~
(- - - -- --
--- ----
-----
Figure 17. The control sense.

Accordingly, the control sense can be employed to mediate relations between

nonphysical TRs and LMs. In examples (48) and (49), either or both the TRI
LM are nonphysical entities.

(48) Cam ilia has authority over purchasing [= the act of deciding what
will be purchased].
(49) Personality has more influence over who we marry than physical
appearance.

4.4.4. The preference sense (5.C)

In the preference sense, that which is higher is conventionally understood as

being preferred to that which is lower.

(50) I would prefer tea over coffee.

(51) I like Beethoven over Mozart.

We suggest that the preference sense derives in the following way: being
physically up in experiential terms can implicate greater quantity, which
generally is preferred to a lesser quantity. In another experiential pattern being
s
physically up is associated with positive states such as happiness (He feeling
up today), while being physically down is associated with being unhappy
(I'mfeeting down today) (see Lakoff and 10hnson 1980). Given that happiness
is normally preferred to unhappiness, this experiential correlation results in
states associated with positions of vertical elevation being preferred to those
associated with a lower position. Hence, being over implicates a preferred
 The case %ver 143

state (see Fig. 18: the TR, which is higher, is to be preferred to the LM,
which is hence not in focus).
This implicature of preference is conventionalized, allowing a preference
interpretation (rather than a higher-than reading) in examples (50) and (51).

I I
I I
I I

Figure J8. The preference sense.

4.5. Reflexivity

4.5.l. The reflexive sense (6.)

Spatial reflexivity (first noted by Lindner 1981) is the phenomenon whereby

a single entity which occupies multiple positions is conceptualized such that
two salient positions occupied by the entity are integrated into a TR-LM
spatial configuration. A preposition such as over is then used to mediate a
spatial relation between the two positions, even though the same entity cannot
simultaneously occupy two distinct spatial positions in the world. The
dynamic character of experience is reanalysed as a static spatial configuration.
Langacker (1987) discusses this gestalt-like static conceptualization of a
dynamic process as summary scanning. Consider (52).

(52) The fence fell over.

In (52), the TR - the initial (upright) position of the fence - is distinguished

from the final position, in which the fence is lying horizontally on the ground.
We see the fence fall through a 90-degree arc and from this experience a
conceptual spatial relation is abstracted (via summary scanning), mediating
the two temporally situated locations into a single spatial configuration. In
the world, no such spatial configuration exists; after all, the same fence cannot
be in two locations at the same time, but by conceptualizing the fence
reflexively, the same entity can be both the TR and the LM (see Fig. 19).
Additional examples of the reflexive sense are given in (53) and (54).
144 Andrea Jyler and "Yvyan Evans

Figure 19. The reflexive sense.

(53) He turned the page over.

(54) The log rolled over.

This sense arises from reanalysis of a process. As noted previously, when

over is used to profile a process, it is coded as an adprep.

4.5.2. The repetition sense (6.A)

The repetition sense adds an iterative meaning component to the use of over,
a meaning component that could not be predicted from the protoscene alone
(or from any other sense considered so far). In examples (55) and (56), over
can be paraphrased by again or anew.

(55) After the false start, they started the race over.
(Cf. After the false start, they started the race again/anew.)
(56) This keeps happening over and over.

Many native speakers have informed us that sentences such as (56) prompt
for a conceptualization of a wheel or cycle, which seems to be evoked by the
notion of repetition. We hypothesize that the repetition-meaning component
associated with over may be the result of iterative application of the reflexive
sense (i.e. the 90-degree arc is repeated such that the TR passes through 360
degrees returning to its original starting point).
Such an analysis is consistent with the intuition that repetition is concep-
tualized as cyclical in nature (Fig. 20). An alternative derivation may be due
to an iterative application of the A-B-C trajectory, such that when the end-
point or completion of the trajectory is reached the process begins again.32
A third possibility may be that the notions of completion and reflexivity are
conceptually integrated forming a conceptual blend (in the sense of Fauconnier
and Turner 1994, 1998, 2002). We remain agnostic about which of these
 The case %ver 145

I
I

r---
IL ___
~/
~

I ,

---.,
'~
_ _ _ .JI
\
\

I I
I I I
I I
~

Figure 20. The repetition sense.

routes led to the instantiation of the repetition sense in the semantic network
for over.

5. Conclusion

Previous polysemy accounts of over offer analyses that are too fine-grained.
These accounts fail to distinguish between coding in formal expression and
a level of conceptualization that integrates linguistic prompts in a way
maximally coherent with sentential context and real-world knowledge. The
selection of a linguistic prompt is, we argued, motivated by a principle of
best fit. That is, given that prepositions represent a closed class they cannot
possibly code the infinite array of all conceptual spatial relations. The speaker
selects the preposition which, given the scene being described, is closest to
accurately describing the key spatial relation. Conceptual integration results
from such underspecified cues being used to construct a complex conceptual-
ization, which elaborates the relatively impoverished linguistic input. A
sentence such as The cat jumped over the wall results in a dynamic complex
conceptualization in which the cat moves above and across the wall, not
because this trajectory is coded for linguistically but because this is the most
coherent and reasonable conceptualization, given the particular prompts, and
given what we know about cats and walls.
In addition, we distinguish between constructed meanings and senses.
The former are constructed on-line in the course of constructing a concep-
tualization of a specific scene prompted by a particular utterance, whereas
senses are instantiated in memory, and can be recruited for the process of
conceptual integration. While complex conceptualizations result from the
process of conceptual integration taking account of motion and hence
146 Andrea 1Ijler and lYvyan Evans

temporal frames, it does not follow that prepositions themselves code dyna-
mism. Accordingly, we maintain the general assumption that prepositions
code atemporal relations.
Within the polysemy network for over set forth here, the primary sense is
termed the protoscene, and represents a highly idealized abstraction from
our rich recurring experience of spatial scenes. We set forth a set of explicit
criteria for determining the primary sense. Other distinct senses instantiated
in the polysemy network for over result from pragmatic strengthening, i.e.
reanalysis and encoding. We recognize a use as distinct only if its interpre-
tation involves a change in the spatial configuration between the TR and
LM and/or additional non spatial information is involved. The polysemy
network for over contains 14 distinct senses. Other interpretations derive
from conceptual integration constrained by the cognitive principles discussed
in Section 3.
The results of our study provide a means for distinguishing between
distinct senses and the process of on-line meaning construction, which is
primarily conceptual in nature. Clearly, a recognition of this distinction is
imperative for future research into the nature of semantic networks, and
provides additional insight into (i) the fundamentally non-arbitrary quality
of the mental lexicon; (ii) the highly creative nature of the human conceptual
system; and (iii) the fact that the way we experience renders spatio-physical
interactions meaningful. which in turn gives rise to emergent conceptual
structure.

Acknowledgements

Our thinking has benefited greatly from conversations with a number of colleagues.
We would particularly like to thank Joseph Grady, Elizabeth Lemon and the members
of the Georgetown Metaphor Group. We also thank Mark Aronoff, Steven Cushing,
and two anonymous Language referees for their detailed comments and suggestions.
We are grateful to Angela Evans for her assistance with the diagrams throughout,
and for supplying us with many of our linguistic examples. Andrea Tyler would like
to acknowledge the many insights and persistent questions raised by the students in
her classes on pedagogical grammar and applying cognitive linguistics. Vyvyan Evans
has benefited from detailed discussions with Craig Hamilton, and owes special thanks
to Mark Turner for his insight and encouragement.
 The case of over 147

Notes

* This article was first published in slightly different form in Language, Volume
77, Number 4 (2001), 724-765. Reprinted by kind permission.
1. Johnson's (1987) pioneering work argues that image-schemas are representations
of recurring aspects of bodily sensory-motor experience, such as verticality, con-
tainment, and so on, which are stored in long-term memory. Hence, they are not
"mental pictures", but rather abstractions from rich experience. See also Cienki
(1998) for an analysis of a single image-schema: straight.
2. The figure-ground notions were developed by the cognitive linguist Leonard
Talmy (e.g. 1978), and are derived from gestalt psychology.
3. Ruhl (1989) has elegantly argued against a polysemy position, championing
instead a monosemy framework. Monosemy holds that each lexical item is
associated with a single highly abstract sense. On this view, the sense is so abstract
that its precise meaning is filled in by context in conjunction with pragmatic
knowledge. We will demonstrate (Section 4) that some senses cannot be predicted
by context alone, a strong argument against a monosemy position.
4. Future empirical analysis might find that speakers make such fine-grained distinc-
tions, but the evidence to date does not bear this out. Although we cannot
definitively prove Lakoff's full-specification model is wrong, it does result in
questionable consequences, both in terms of its linguistic representations and in
terms of the little experimental evidence that is available.
5. The variations among just the two attributes of + / - or unspecified extended,
and + / - or unspecified vertical, result in nine distinct senses. Each time another
attribute is added to the model, the list of distinct senses multiplies accordingly -
consider Table 1. The predictions become even more questionable when one
considers that five of the nine senses involve attributes being unspecified.
Analogous arguments can be made for specification of the exact, metric
relationship between the TR and LM in terms of the presence or absence of
contact, as Kreitzer (1997) underscores with the example Sam went over the wall,
in which the precise manner of passing over the wall, either jumping or crawling,
is unspecified, therefore the presence or absence of contact is unspecified.

Table 1. Topographical features (after Lakoff 1987).

+ Vertical - Vertical Unspecified

+ Extended S S U
- Extended S S U
Unspecified U U U

S = specified; U = unspecified.
148 Andrea Tyler and \ryvyan Evans

6. In order to motivate the distinction between over] and over2, Kreitzer appeals to
Langacker's notion of summary scanning (Langacker 1987, 199Ia). Langacker
posits that summary scanning provides a means of integrating points occupied
by a TR along a path into a construal of motion along a path. The path is reified
at the conceptual level, even though it never actually exists in the world. Kreitzer
argues that the dynamic over2 describes a relation between a TR and a LM in
which it is the path that is the TR.
7. The term ideal meaning is from Herskovits 1986: Chapter 4.
8. It is important to note that some central (= most basic, to be explicated) senses
associated with prepositions will crucially involve a coordinate system along the
vertical or horizontal axes, while others will not. We will argue that the primary
sense associated with over does involve such a system in which the spatial relation
of the TR being located higher than the LM is essential. But this should not be
interpreted as a claim that all prepositions prompt for such a system. While the
English prepositions over and under regularly code respectively for the TR being
in a higher-than or lower-than position relative to the LM, the preposition out
appears to be insensitive to this dimension. Thus, we find sentences like The
rain poured out of the sky (in which the TR is lower than the LM) and The water
bubbled out of the hot springs (in which the TR is higher than the LM) which do
not affect the basic interpretation associated with out. Whether a particular
preposition is sensitive to the horizontal or vertical dimensions is part of its basic
lexical entry.
9. Although there has been disagreement about the appropriate representation of
the primary sense associated with over, all published analyses accept these two
basic assumptions. Synchronically, evidence that the basic spatial configuration
prompted for by over is something like a TR in a higher-than position relative to
the LM comes from sentences with clearly contrasting interpretations: Nicole
decided to walk over the bridge versus Nicole decided to walk under the bridge.
Having argued that the primary sense for over involves a spatial configuration in
which the TR is higher than the LM, we readily acknowledge that in many
instances this spatial configuration is not prompted for by over. Our analysis
attempts to model how these non-canonical spatial configurations have come to
be associated with the form over.
10. It should be noted that our diagrammatic representations of protoscenes are made
for ease of explication. They should not be interpreted as making any serious
claim about the neurological nature of imagistic representation.
11. An image-schema, as Mandler uses the term, constitutes a representation distinct
from purely perceptual information. As such, it constitutes a rudimentary "theory"
as to the nature of a particular object or relation between objects. The image-
schema relating to containment, for instance, is a concept as opposed to a
percei ved entity, insofar as it constitutes a means of understanding the functional
aspects of a particular spatial configuration.
12. This is akin to what Iackendoff (1983: 29) refers to as the projected world, and
 The case of over 149

is constructed at what Fauconnier (1997: 36) tenns the cognitive level or level C.
13. In terms of specifics our claim is as follows: a particular spatial scene is a rich
real-world scenario. mediating two objects (TR and LM) via a conceptual spatial
relation. Recurring spatial scenes perceived as resembling each other are stored
as an abstract protoscene. The aspect of the protoscene coded by a preposition is
the spatial relation mediating the TR and LM. and not the whole protoscene.
From this. it follows that a preposition presupposes a TR and a LM (as the
conceptual spatial relation holds by virtue of mediating a relation between a TR
and a LM). In minimal terms. a preposition prompts for a TR and LM. which
are typically supplied linguistically. e.g. The picture [TR] is over the mantel
[LM].
14. The reanalysis of an aspect of a particular complex conceptualization results in
privileging a different aspect or perspective on the complex conceptualization.
Yet. because the pertinent complex conceptualization is first prompted for by the
use of over. as in Figure 5. the derived sense is coded by the same linguistic
form. namely over.
IS. eruse (1986) discusses this in terms of modulation of a lexical item. For instance.
various parts of the car are highlighted in the following sentences: The car needs
to be washed (where car is interpreted as the exterior body of the car) versus The
car needs to be serviced (where car is interpreted as the engine) versus The car
needs vacuuming (where car is interpreted as the interior). This constitutes
modulation or highlighting different parts and backgrounding others.
16. Langacker ([ 992) discusses the atemporal nature of prepositions in terms of the
relationships they profile. "With before and after. time functions as the domain
in which the profiled relationship is manifested. Its role is consequently analogous
to that of space in the basic sense of in. on or near. A verb. on the other hand, is
said to be temporal in a very different way ... the profiled relationship is conceived
as evolving through time and is scanned sequential\y along this temporal axis. It
is by incorporating this further level of conceptual organization that precede and
follow differ from the prepositions before and after ... [Verbs] specifical\y track
[a process] through time ... A preposition can thus be characterized as profiling
an atemporal relation that incorporates a salient landmark" ([ 992: 292).
17. Unless the world being discussed is explicitly designated as science fiction.
18. In sentence ([ 9) the lack of motion is the result of integrating what is coded by
the verb extended with our knowledge of trees. In particular. the interpretation of
lack of motion depicted by (19) is the result of the interpretation of extended as
it relates to a tree branch. We understand trees to be slow-growing plants such that
humans do not perceive the growth of a branch as involving motion. Thus. we
interpret extended to depict a state. Notice that the stative interpretation of extended
is contingent upon the precise sentential context in which it occurs. Extended
can also be interpreted to convey motion as in He extended his arm towards the
door. Since there is no sense of motion prompted for by the verb in the sentential
context provided in (19). no path or trajectory is projected for the TR.
150 An.drea Tyler and \ryvyan Evans

19. We hasten to acknowledge that there are contexts in which two prepositions
appear to be interchangeable and virtually synonymous: Susan hung the picture
over the mantel versus Susan hung the picture above the mantel. We hypothesize
that such substitutability arises because the semantic networks associated with
each preposition represent continuums and at certain points the interpretations of
two continuums can overlap. In addition, for over and above we find a close
diachronic relationship, with over initially being used as the comparative form
of above. The diachronic link may surface in these overlapping uses.
20. In terms of synchronic polysemy networks, the empirical work by Sandra, Rice,
and their colleagues suggests that it may not be the case that a particular lexical
form has a single primary sense from which language users perceive all other
senses being derived. Their empirical work raises questions about the view that
we can define polysemy as a strictly synchronic phenomenon in which speakers
are consciously aware of a relationship holding between distinct senses of a
particular lexical form. This is an empirical question for which we do not yet
have sufficient evidence to determine the answer. If extensive experimental
evidence shows that language users systematically and consistently fail to perceive
some senses as being related, then we must question whether what we term
polysemy constitutes a phenomenon that is wholly synchronic in nature. While
we believe all the senses in a particular semantic network are diachronically
(and perhaps developmentally) related, in terms of the adult lexicon, there may
be differences in the perceived relatedness between distinct sets of senses, due to
routinization and entrenchment, obscuring the original motivation for the deriv-
ation of senses from pre-existing senses such as the protoscenes for language
users (see in particular Rice, Sandra and Vanrespaille 1999).
21. In formal terms, the particle in a verb-particle construction (VPC) is a more
grammaticized preposition in that the LM is linguistically covert, that is, it is contex-
tually understood without being linguistically coded (Lindner 1981; O'Dowd
1998). Such particles form part of a verb-particle construction with a verbal element,
and each unit (the particle and the verb) contributes to the meaning of the whole
unit (see Goldberg 1995 for a construction grammar approach, Morgan 1997 for
a study of verb-particle constructions). We introduce the term adprep to describe
a spatial particle which has adverbial meaning, that is, certain usages of the form
over are adverbial in nature, describing an aspect of a conceptual process, as in
The movie is over (= finished). Each formal component - preposition, particle (in
a VPC), particle prefix, or adprep - contributes different kinds of meaning.
22. Recall that we are using the term "sense" for distinct meanings instantiated in
memory (i.e. in the semantic network associated with each preposition).
23. It is worth pointing out that sentences such as (21 )-(25) offer strong evidence
against a monosemy theory of word meaning. Monosemy (see Ruhl 1989), as
noted previously, posits that all interpretations of a linguistic form, such as a
preposition, are contextually derivable from a highly abstract primary sense.
However, as can be seen from the on-the-other-side-of sense, neither of the
 The case of over 151

original aspects of the spatial configuration hold - the TR is not above the LM
and the TR is not proximal to the LM. The nature of a primary sense that would
derive both these senses simply from contextual cues would need to be extremely
abstract. We cannot see how a representation so abstract would also be constrained
enough to distinguish among many other English prepositions.
24. Lakoff (1987: 422-423) represents sentences such as Sam lives over the hill as
an example of schema I.VX.C.E. (above-across, with a vertical, extended LM,
contact between the TR and LM, and endpoint focus).
25. There is arguably a distinct sense which is derived from the on-the-other-side-of
sense. In examples such as

(57) The festival will take place over the weekend;

(58) The friendship remained strong over the years;
(59) Lets take a look at changes over time;

over mediates a temporal relation of concurrence between a process or activity

and the times during which the process or activity elapses. This sense is likely to
have developed from the on-the-other-side-of sense, when the physical LM is
extended, as, for example, in The boy walked over the hill, The cable runs over
the yard, and The bridge stretches over the river. In such situations the activity
is concurrent with the duration required for the activity. Because of pragmatic
strengthening, a duration sense may have become associated with over.
26. This is consistent with Langacker (e.g. 1987) who argues that grammatical
class is determined by virtue of what is profiled. For instance, the relationship
profiled by adverbs crucially differs from the relationship profiled by prepositions
in that an adverb takes a relationship as its TR and does not have a salient LM.
In contrast, a preposition takes an entity as its TR and elaborates a relational
LM.
27. Nonphysical entities can be identified as TRs or LMs, if they are construed as
focal and backgrounded respectively, and if a relation holds between them. As
over has a conventionalized transfer sense associated with it, the relation
between nonphysical TRs and LMs cannot be spatio-configurational, but as in
The government handed its power over to the newly elected officials, it can
involve the notion of transfer. This further illustrates that transfer must be a
distinct sense: it could not be derived from context in such sentences. There is
a conventional reading in which the members of the government transfer their
authority, i.e. their mandate to govern, to a new set of officials. In literal terms,
nothing is physically transferred, as the TR, power, is a nonphysical entity. None-
theless, to say that power is a nonphysical entity is not to say that the concept
"power" is without foundation in real-world experiences. In fact, the concept of
power derives from a variety of very real experiences: physical forces, socially
constructed relationships and hierarchies, and social interactions such as taking,
issuing and following orders, commands, edicts, and so on. In this sense, we
152 Andrea TyLer and Vyvyan Evans

each experience power in a real way, although the variety of experiences

subsumed by the concept of power does not have physical substance or spatial
dimensionality in the same way that a chair or a table has. Accordingly, it
makes sense that power can be transferred, thus licensing the use of the transfer
sense.
28. Again, following our argument that metric properties concerning the relationship
between the TR and LM are filled in on-line, over can be used to prompt for this
covering interpretation when there is contact between the LM and TR, as in (31),
or when there is no contact between the TR and the LM, as in The fibregLass
protector was put over the drained swimming pooL for the winter.
29. These two changes are closely intertwined in everyday experience. We are often
involved in real-world scenarios where the TR is physically larger than the LM
and we normally view the TR-LM from above, as in The cloth is over the tabLe.
In this real-world scene, if the TR were smaller than the LM, the preposition of
choice (best fit) would be likely to change.

(60) ?The small handkerchief was spread out over the tabLe.
(61) The small handkerchief was spread out on the tabLe.

However, there are also many real-world scenarios in which the TR is actually
smaller than the LM but because of the construer's vantage point (the TR inter-
venes between the viewer and the LM), the TR appears larger than the LM. For
instance, in The dark clouds moved over the sun, the clouds are not physically
larger than the sun, but they appear larger to the earthbound viewer.
30. Lakoff (1987: 429) accounted for cases of the covering reading in which the TR
is not higher than the LM by positing a rotation transformation. The covering
schemas all have variants in which the TR need not be above (that is, higher
than) the LM. In all cases, however, there must be an understood viewpoint from
which the TR is blocking accessibility of vision to at least some part of the LM.
We will refer to these as rotated (RO) schemas, though with no suggestion that
there is actual mental rotation degree-by-degree involved. This is an extremely
powerful transformation, potentially affecting all prepositions whose primary
sense involves either a vertical or horizontal orientation. In a number of instances,
the protoscenes for over, wider, before, and after would be essentially indisting-
uishable. And this analysis offers no explanation for why TR-LM configurations
that do not match the protoscene would develop this reading.
A common consequence of the LM being covered by the TR is that the LM
is occluded from the construer's view. Typically the scene described in (31) is
that the tablecloth occludes the tabletop from the observer. As we see in examples
such as the following, occlusion is not an inevitable consequence of covering.

(62) The mask is over her face.

(63) She wore a transparent veil over her face.
 The case of over 153

(64) The dark, heavy clouds are over the sun.

(65) There are afew wispy clouds over the sun.

In sentences (62) and (64) a consequence of the LM being covered by the TR

is that the LM is no longer visible. In (63) and (65), however, covering does not
obscure the object. We have not been able to find any instances of occluding
which involve the use of over! that do not include a covering sense. Funher, in
the examples in which we can tease apan covering from occluding, the physical
attribute of transparency/opacity of the TR must he specified. If the TR is not
specified as transparent (65) the normal reading is that covering entails occlusion.
Thus, we have concluded that the occlusion interpretation is a contextual impli-
cature of the covering sense and real-world knowledge of the propenies of objects
such as tablecloths and blankets. Given the absence of contextually independent
examples of occlusion - linguistic examples of over in which occlusion is not an
implicature deriving from covering - our methodological procedure suggests that
an occluding reading is an on-line interpretation.
31. In some cases, we see no clear way to determine which source is more appropriate.
As we noted in our discussion of the excess I sense, specific uses of over (or
any preposition) seem to contain "flavours" of more than one sense, which
imbues a reading with complex nuances of meaning. For instance, consider the
following.

(66) Hey! Why are you bringing in so many cases of motor oil? There must be
a dozen cases here. That's well over the two cases I ordered.

In this example we might construct a "more" conceptualization for over, or

we might construct an "excess" interpretation (which provides not just a more
meaning, but the additional too-much-more meaning) for over. In this latter case,
the example could be derived from either the above-and-beyond (excess I) sense
or the over-and-above (excess 11) sense. On the one hand, two cases could be
conceptualized as the target the customer was aiming for, and bringing in ten
additional cases could be construed as going beyond the designated target. On
the other hand, two cases could be conceptualized as the expected amount or
level of goods, and the additional ten cases could be construed as going above
the expected amount or level.
Alternatively, the hearer may construct a complex conceptualization in which
all three senses are influencing the interpretation. This reflects our claim that
there is a semantic network linking distinct senses, and that conceptualizations
may be due to a semantic network constituting a meaning continuum, as discussed
earlier. Accordingly, our network should be thought of as a semantic continuum,
in which complex conceptualizations can draw on meanings from distinct nodes
as well as the range of points between nodes, which provide nuanced semantic
values. In addition, an imponant consequence of our claims
154 Andrea Tyler and VYvyan Evans

(i) that the principles of meaning construction in conjunction with a distinct

sense such as the protoscene (or any other distinct mental representation
or sense), can be used to construct a wide range of conceptualizations;
(ii) that anyone conceptualization is subject to mUltiple construals (through,
for instance, privileging a particular aspect of the scene or shifting the
vantage point from which the scene is viewed);
(iii) that distinct senses can be extended to include nonphysical entities when
such are perceived as focal (TRs) and backgrounded reference points
(LMs);
(iv) that semantic networks form an interrelated continuum of interpretations
(rather than just a series of absolutely discrete points of meaning);

is that the model predicts that a particular sense may arise from more than one source.
In forms such as overachieve, overkill, overdo, and overdress we do not see
a clear basis for arguing for the superiority of the above-and-beyond interpretation
versus the over-and-above interpretation. As noted earlier, we do not consider
this a flaw in our model; rather we see it as testimony to the richness and
complexity of conceptualization. We also hypothesize that native speakers are
likely to vary in their intuitions about these cases.
32. Lindstromberg (1997) offers a very similar explanation.

References

Aronoff, Mark
1994 Morphology By Itself: Stems and Inflectional Classes. Cambridge,
MA: MIT Press.
Bloom, Paul, Mary A. Peterson, Lynn Nadel and Merrill F. Garrett (eds.)
1996 Language and Space. Cambridge, MA: MIT Press.
Bloomfield, Leonard
1933 Language. New York: Holt, Rinehart and Winston.
Bolinger, Dwight
1971 The Phrasal Verb in English. Cambridge, MA: Harvard University
Press.
Brisard, Frank
1997 The English tense-system as an epistemic category: the case of
futurity. In: Marjolijn Verspoor, Kee Dong Lee and Eve Sweetser
(eds.), Lexical and Syntactical Constructions and the Construction
of Meaning, 271-286. Amsterdam: John Benjamins.
Brugman, Claudia
1981 [1988] The story of over. M.A. thesis. Linguistics Dept. University of
California. Berkeley. Published [1988] as The Story of Over:
 The case of over 155

Polysemy. Semantics and the Structure of the Lexicon. New York:

Garland Press.
Brugman, Claudia and George Lakoff
1988 Cognitive topology and lexical networks. In: Steven Small. Garrison
Cottrell and Michael Tanenhaus (eds.). Lexical Ambiguity Resolution,
477-507. Palo Alto. CA: Morgan Kaufmann.
Bybee. Joan, Revere Perkins and William Pagliuca
1994 The Evolution of Grammar: Tense, Aspect. and Modality in the
Languages of the World. Chicago. IL: University of Chicago Press.
Chomsky, Noam
1995 Categories and transformations. In: The Minimalist Program. 219-
394. Cambridge. MA: MIT Press.
Cienki, Alan
1998 Straight: an image schema and its metaphorical extensions. Cognitive
Linguistics. 9(2): 107-150.
Clark, Herbert H.
1973 Space, time, semantics, and the child. In: Terence Moore (ed.),
Cognitive Development and the Acquisition of Language, 27-64. New
York: Academic Press.
Croft, William
1998 Linguistic evidence and mental representations. Cognitive Linguistics
9: 151-174.
Cruse, David A.
1986 Lexical Semantics. Cambridge, England: Cambridge University Press.
Cushing, Steven
1990 Prototypical considerations on modal meanings. In: Tsohatzidis
(1990), 74-90.
1991 Explaining a missing modal meaning: ideology and paradigm as
pragmatic parameter. In: Jef Verschueren (ed.), Levels of Linguistic
Adaptation, 19-32. Amsterdam: John Benjamins.
Dewell, Robert
1994 Over again: image-schema transformations in semantic analysis.
Cognitive Linguistics 5(4): 351-380.
Dirven, Rem!
1993 Dividing up physical and mental space into conceptual categories by
means of English prepositions. In: Cornelia Zelinsky-Wibbelt (ed.),
The Semantics of Prepositions: From Mental Processing to Natural
Language Processing, 73-97. Berlin: Mouton de Gruyter.
Evans, Vyvyan
2000 The structure of time: language, meaning and temporal cognition.
Doctoral thesis. Linguistics Dept, Georgetown University.
Fauconnier, Gil\es
1994 Mental Spaces. Cambridge. England: Cambridge University Press.
156 Andrea Tyler and vyvyan Evans

1997 Mappings in Thought and Language. Cambridge, England: Cambridge

University Press.
Fauconnier, Gilles and Mark Turner
1994 Conceptual Projection and Middle Spaces. Dept of Cognitive Science:
University of California, San Diego. Cognitive Science Technical
Report No. 9401.
1998 Conceptual integration networks. Cognitive Science 22(2): 133-187.
2002 The Way We Think: Conceptual Blending and The Minds Hidden
Complexities. New York: Basic Books.
Fillmore, Charles, Paul Kay and Mary Catherine O'Connor
1988 Regularity and idiomaticity in grammatical constructions: the case
of let alone. Language 64(3): 501-538.
Fleischman, Suzanne
1999 Discourse markers across languages: implications of a case study
for historico-comparative and sociolinguistics. Paper presented at
Georgetown University, November 1999.
Goldberg, Adele
1995 Constructions: A Construction Grammar Approach to Argument
Structure. Chicago, IL: Chicago University Press.
Grady, loseph
1997 Foundations of meaning: primary metaphors and primary scenes.
Doctoral thesis. Linguistics Dept, University of California, Berkeley.
Grice, Paul
1975 Logic and conversation. In: Peter Cole and lerry L. Morgan (eds.),
Syntax and Semantics, Volume 3: Speech Acts, 41-58. New York:
Academic Press.
Heine, Bernd
1993 Auxiliaries: Cognitive Forces and Grammaticalization. New York:
Oxford University Press.
1997 Cognitive Foundations of Grammar. New York: Oxford University
Press.
Herskovits, Annette
1986 Language and Spatial Cognition: An Interdisciplinary Study of the
Prepositions in English. Cambridge, England: Cambridge University
Press.
Hopper, Paul and Elizabeth Closs Traugott
1993 Grammaticalization. Cambridge: Cambridge University Press.
lackendoff, Ray
1983 Semantics and Cognition. Cambridge, MA: MIT Press.
1997 The Architecture of the Language Faculty. Cambridge, MA: MIT
Press.
lohnson, Mark
1987 The Body in the Mind. Chicago, IL: Chicago University Press.
 The case of over 157

Kay, Paul and Charles Fillmore

1999 Grammatical constructions and linguistic generalizations: the what's
X doing Y? construction. Language 75(1): 1-34.
Kreitzer, Anatol
1997 Multiple levels of schematization: a study in the conceptualization
of space. Cognitive Linguistics 8(4): 291-325.
Lakoff, George
1987 Women, Fire and Dangerous Things: What Categories Reveal about
the Mind. Chicago, IL: Chicago University Press.
Lakoff, George and Mark Johnson
1980 Metaphors We Live By. Chicago, IL: Chicago University Press.
1999 Philosophy in the Flesh: The Embodied Mind and its Challenge to
Western Thought. New York: Basic Books.
Langacker, Ronald
1987 Foundations of Cognitive Grammar; Volume J. Stanford, CA: Stanford
University Press.
1991a Concept, Image and Symbol. Berlin: Mouton de Gruyter.
1991b Foundations of Cognitive Grammar; Volume 2. Stanford, CA: Stanford
University Press.
1992 Prepositions as grammatical(izing) elements. Leuvenese Bijdragen
81: 287-309.
Levin, Beth
1993 Towards a Lexical Organization of English Verbs. Chicago, IL:
University of Chicago Press.
Lindner, Susan
1981 A lexico-semantic analysis of English verb-particle constructions
with out and up. Doctoral thesis. Linguistics Dept, University of
California, San Diego. Distributed [1983] by the Indiana University
Linguistics Club.
Lindstromberg, Seth
1997 English Prepositions Explained. Amsterdam: John Benjamins.
Mandler, Jean
1988 How to build a baby: on the development of an accessible represen-
tational system. Cognitive Development 3: 113-136.
1992 How to build a baby: n. Conceptual primitives. Psychological Review
99(4): 587--604.
1996 Preverbal representation. In: Bloom, Peterson, Nadel and Garrett
(1996), 365-384.
Morgan, Pamela
1997 Figuring out figure out: metaphor and the semantics of the English
verb-particle construction. Cognitive Linguistics 8: 327-357.
158 Andrea Tyler and Vyvyan Evans

O'Dowd, Elizabeth
1998 Prepositions and Particles in English: A Discourse-Functional
Account. New York: Oxford University Press.
Pustejovsky, James
1998 The Generative Lexicon. Cambridge, MA: MIT Press.
Rice, Sally
1993 Far afield in lexical fields: the English prepositions. In: Michael
Bemstein (ed.), ESCOL '92: 206-217.
Rice, Sally, Dominiek Sandra and Mia Vanrespaille
1999 Prepositional semantics and the fragile link between space and time.
In: Masako Hiraga, Christopher Sin ha, and Sherman Wilcox (eds.),
CuLtural, PsychoLogical and TypoLogicaL Issues in Cognitive Ling-
uistics, 107-127. Amsterdam: John Benjamins.
Rosch, Eleanor
1975 Cognitive representations of semantic categories. Journal of Experi-
mental Psychology: General 104: 192-233.
Ruhl, Charles
1989 On Monosemy: a Study in Linguistic Semantics. Albany: State
University of New York Press.
Sand ra, Dominiek
1998 What linguists can and can't tell you about the human mind: a reply
to Croft. Cognitive Linguistics 9(4): 361-378.
Sandra, Dominiek and Sally Rice
1995 Network analyses of prepositional meaning: mirroring whose mind -
the linguist's or the language user's? Cognitive Linguistics 6(1):
89-130.
Sperber, Daniel and Deirdre Wilson
1986 Relevance: Communication and Cognition. Cambridge, MA: Harvard
University Press.
Svorou, Soteria
1993 The Grammar of Space. Amsterdam: John Benjamins.
Sweetser, Eve
1990 From Etymology to Pragmatics: Metaphorical and CulturaL Aspects
of Semantic Structure. Cambridge, England: Cambridge University
Press.
Talmy, Leonard
1978 Figure and ground in complex sentences. In: Joseph H. Greenberg
(ed.), Universals in Human Language, Volume 4, 625~49. Stanford,
CA: Stanford University Press.
1983 How language structures space. In: Herbert Pick and L. Acredelo
(eds.), SpatiaL Orientation: Theory, Research and Application, 225-
320. New York: Plenum.
 The case of over 159

1988 Force dynamics in language and cognition. Cognitive Science 12:

49-100.
1996 Fictive motion in language and "ception". In: Bloom, Peterson, Nadel
and Garrett (1996), 211-276.
2000 Toward a Cognitive Semantics, VoLumes I and I/. Cambridge, MA:
MIT Press.
Traugott, Elizabeth Closs
1989 On the rise of epistemic meanings in English: an example of
subjectification in semantic change. Language 65(1): 31-55.
Tsohatzidis, Savas L. (ed.)
1990 Meanings and Prototypes: Studies in Linguistic Categorization.
London: RoutIedge.
Turner, Mark
1991 Reading Minds. Princeton: Princeton University Press.
1996 The Literary Mind. New York: Oxford University Press.
Tyler, Andrea and Vyvyan Evans
2003 The Semantics of English Prepositions: SpatiaL Scenes, Cognition
and the ExperientiaL Basis of Meaning. New York and Cambridge:
Cambridge University Press.
Vandeloise, Claude
1990 Representation, prototypes and centrality. In: Tsohatzidis (1990),
403-437.
1991 SpatiaL Prepositions: A Case Study in French. Chicago, IL: The
University of Chicago Press.
1994 Methodology and analyses of the preposition in. Cognitive Linguistics
5(2): 157-184.
Varela, Francisco, Evan Thompson and Eleanor Rosch
1991 The Embodied Mind. Cambridge, MA: MIT Press.
Polysemy as flexible meaning: experiments with
English get and Finnish pitaa

Jarno Raukko

1. Introduction

It is common to view polysemy as a collection or network of several inter-

related meanings that is fairly stable, fairly unproblematic to segment and
establish, and fairly well agreed upon by different speakers. Yet there are
also opponents (of at least two kinds) who either wish to reduce polysemy to
depend only upon contextual specifications, or simply claim that polysemy
is irrelevant to the study of communication.' One way to build a bridge
between these camps is to see polysemy as patterns of flexibility in (lexical)
meaning in much the same way as it is accepted that situational (utterance or
discourse level) meaning is non fixed, inexact and negotiable. This does not
imply that word-specific descriptions of semantic variation would be unneces-
sary, but it leaves room for dynamicity and open-endedness in categorization
as well as for intersubjective disagreements about the structure of the poly-
se my of a given word. I want to demonstrate the advantages of this view
with an analysis of some of the results from polysemy experiments with
nonlinguist informants.
Even if semantics is nonfixed, flexible, imprecise and in a pessimistic
sense intangible, language users believe that much of what is communicated
is based on a significant degree of intersubjective agreement. They need to
believe in such an agreement in order to believe in the value of communi-
cation (see Raukko 1996). A truly intersubjective type of experiment to
investigate this phenomenon might involve a manufactured dialogical situ-
ation where the linguist could, e.g., observe meaning negotiations between
the participants over the meaning of a polysemous word in a specific context.
So far I have employed a more modest "mock-intersubjective" framework
where informants can creatively produce their own representations of the
polysemy of a word, and I will in my analysis perform the "intersubjective
shuttling" between the responses.
The two words whose polysemy provides the example cases in this article
162 lamo Raukko

are two verbs, one from English (get, meaning e.g. 'obtain', 'receive', 'arrive',
'take', 'become', 'make', 'have', 'understand', 'answer', 'must' and 'be
allowed to') and the other from Finnish (pitiiii, meaning e.g. 'like', 'must',
'hold', 'keep', 'consider', 'organize', 'wear' and 'not leak'). In most cases
these verbs are not translational counterparts; the reason for selecting them
lies mainly in the shared complex type of polysemy. The section on get
concentrates on the results from an experiment called the production test
and uses some secondary findings from the results of a so-called difference
evaluation test (also known as the similarity rating test). The section on
pitiiii focuses on the results of a third type of test, the sorting test.

2. Variability, flexibility, fuzziness and methodological needs

2.1. Flexibility and juzziness

A new dynamic view of meaning gained ground by the end of the 19th century
(Nerlich 1992: 100). Instead of a one-to-one relationship between word and
idea, meaning was already seen as flexible, elastic, adaptable and open to
change; "imperfect", so to speak. Such notions were at least presented in
Germany (Erdmann 1910; Paul), France (BreaI1868; Nyrop 1913; Paulhan),
England (Gardiner) and the United States (Whitney 1867).
It could be said that mainstream semantics in the 20th century lost track
of the notion of flexible meaning as the driving force of semantics in general
and polysemy in particular. Structuralist semantics paid more attention to
the difference between homonymy and polysemy2 than to the Ubiquity of
polysemy. Indeed, many semanticists, including so-called cognitive semanti-
cists (e.g. Geeraerts 1993; Tuggy 1993), have paid much attention to another
tripartite division, namely ambiguity, polysemy and vagueness. Still, much
of that literature rests on a notion of vagueness as a fairly static property of
lexical meaning.
One recent addition is offered by Zhang (1998), who draws attention to
the concept ofjuzziness in this four-partite framework:jitzziness, vagueness,
generality and ambiguity. (For Zhang, polysemy is closest to vagueness.)
Her framework is Gricean, and her view of polysemy does not seem to allow
for ambiguities that Bre left unresolved by the context. In her view only
fuzziness is not "resolvable with resort to context" and "is closely involved
with language users' judgments" (Zhang 1998: 13), After discussing syntactic
 Experiments with English get and Finnish pitaa 163

and semantic tests for fuzziness, Zhang introduces some pragmatic tests.
Whereas for Zhang (i) bank is ambiguous irrespective of people's judgements;
(ii) person is general in that it does not specify sex, height or nationality;
(iii) good is vague as to whether it refers to the goodness of a student, food
or legs; (iv) beauty, by contrast, is fuzzy owing to the fuzziness of the concept
it denotes, to the fuzzy nature of language users' perceptions of its referential
boundary (1998: 27). Zhang calls such factors psychological, but does not
explicate if her understanding of "semantic cognition" would also involve a
more socially based procedure of "wanting to believe that interlocutors agree
on meanings" in place of truth conditions and Grice's maxims.

2.2. Flexibility and variability

All words and morphemes display poly se my to a greater or lesser extent. In

addition, polysemy reflects semantic flexibility. All discourse deals with this
fuzzy state of affairs, and because virtually nothing is monosemous and fixed,
there is no solid way to explain polysemy away by processes of contextual
specification. Therefore, all investigations of linguistic meaning and ling-
uistic function have to deal with semantic flexibility. Polysemy (flexibility)
becomes the default, and it affects everything we say about syntax, semantics,
pragmatics, text and discourse.
Semantic flexibility, semantic variability, and polysemy are to me in some
sense three ways of talking about the same thing. However, in the following
I shall concentrate on polysemy as one type of semantic variability. I am partly
following Raukko (1997) where I contrasted a mainstream "traditional view"
with an "alternative view".
Variability is different from variation (see Ostman 1988). Variability has
to do with a process, with a potential, with an inherent quality. Variation is
rather a product, a materialized state of affairs, a descriptive quality. In
research, variation is more like a top-down phenomenon: linguists first decide
what language is, and what is a relevant linguistic feature, and then see how
language varies according to that feature. Variability is more bottom-up: it is
a starting point for a theory, and only through building a theory of the
phenomena in language does the linguist gradually build a notion of what
language is and what a relevant feature is.
What then is semantic variability? I would like to suggest at least six (over-
lapping) shades of the idea.
164 lama Raukko

(i) Nonfixedness. We cannot claim that language has a lexicon with ready-
made meanings and polysemous categories for words; nor does a
person's cognition contain a mental lexicon in the sense of a store-
house of ready-made meanings and categories.
(ii) Fuzziness, unclarity, non-pinpointability. We cannot aim at preciseness
in communication nor in the analysis of meaning.
(iii) Flexibility. Language is constituted by social conventions. These social
conventions allow for change and flexibility, and some disagreement
between speakers. Still, it is not a dead end to try to describe seman-
tics; we just have to allow for loose ends, and for the possibility of
change and flexibility.
(iv) Ambivalence. Polysemous situations are not about two competing
meanings at a time but (in principle) about an indefinitely large
"number" of meanings. Whereas ambiguity is usually tied to intention,
ambivalence is the idea that even speakers themselves do not (always)
know what they intend to communicate (Ostman 1988, 2000).
(v) Intersubjective differences (this can also be about variation). When
we describe semantics, it is wiser to be descriptive and open-ended
than prescriptive and definitive. Of course the social conventions
themselves are (governed by) norms, but it is a different thing for
linguists to decide what is the correct meaning - or set of meanings -
ofa word.
(vi) Polysemy (in "static semantics" this is about variation). The default is
that words do not have just one fixed meaning. Therefore many
practical things that we do, such as provide glosses for foreign lang-
uage items, or make lists of contrastive vocabularies in different
languages, are very problematic endeavours.

These six shades of variability cover and partly explain the meaning and
place of polysemy and flexible meaning in semantics. The four Subsections
2.3, 2.4, 3.1 and 3.2 will show how I have operationalized variability and in
particular flexibility in my overall methodology and in my experiments.

2.3. Methodological consequences

When we want to address semantic questions, such as "What kinds of mean-

ings can this word 'have'?" and "What is crucial in the distinctions and
 Experiments with English get and Finnish pitiiii 165

potentials of that polysemy?", and take into account the properties of seman-
tics that I have spelt out above, we have to choose some methodology, and
choose some concrete methods.
One method that has become something of a default in linguistics is the
corpus method. The linguist looks at a large and somewhat pre-processed
selection of text material and tries to find the relevant instances (instantiations,
specimens) of the item that slhe wants to study. However, problems arise,
because most semantic questions would require some knowledge about the
meaning-maker, the human beings producing and receiving and under-
standing the text and all its bits. When using a corpus, one is basically by
oneself when trying to determine what the bits mean. Of course one can use
one's linguistic knowledge - as a native or fluent speaker and as a linguistic
expert - in order to determine the meanings. However, it is not necessarily a
very interesting, reliable, nor "scientific" way to deal with semantic questions.
One reason is the ever-presence of semantic variability and flexibility. If
meanings are not out there as fixed entities, but instead emerge in language
use situations, we need to know more about the context and the participants
of the communicative situation in order to say what is being meant. Also, if
meanings are fuzzy, it is even more difficult to determine the meaning if we
are not ourselves part of the context of the communicative situation. More-
over, if people can have different opinions on what something means, why
would it be enough for a linguist just to look at a corpus and say what he or
she finds there?
Another way might be to turn to a different scientific ideal, something
closer to the natural sciences, for instance cognitive neuroscience. So, for
instance, one could forget about questions that semanticians, pragmaticians
and discourse analysts are interested in, and concentrate on things that you
can see in the curves and measures that show features in the electrical activity
of the brain. Or feed a computer large masses of text and see how it categ-
orizes words and texts according to distributions. The problem here is that
we are quite far from linguistically interesting questions and methods. And
unlike natural scientists, semanticists should not even dream of finding
definitive or objective answers.
Then there is a middle way: do something about the tacit assumptions
and hence the procedure of scientific discovery and linguistic analysis. This
is the road I am taking. I have chosen to try out what certain kinds of
experimental methods, experiments, experimentations can do for us.
One field of linguistics where it has been customary to use experiments
166 larno Raukko

is psycho linguistics. Although two out of three test types that I am using
have been employed by semantically-oriented psycholinguists, my experi-
mental methodology as a whole is not exactly of psycholinguistic nature. I
am trying out experiments that fall between pure armchair speculation and
hard-core experimental methods with an ideal of objectivity. Some of these
experiments leave room for the creativity of the informant who is participating
in the experiment (and hence bear similarities with elicitation and inter-
viewing). Although these experiments use artificial settings, an open-minded
and non-objectivist semantician can use the experiments to make discoveries
about semantic flexibility and variability, as there is openness in the questions
and room for the informant's personal effort.
Section 3 will introduce the three types of experiments that I have been
using in my research into polysemy. In Section 3.2 I will show how the six
shades of variability mentioned in Section 2.2 are operationalized in the
experiments.

2.4. Polysemy as dynamic mass

For me polysemy, flexibility and variability are broad cover tenns, and I do
not even mind using them as covering one another. Polysemy is flexible
meaning, and it covers referential polysemy, regular polysemy, lexical poly-
semy (see Deane 1988), vagueness, generality and fuzziness. I prefer to
investigate all variants of "one form - not just one fixed meaning" with the
same tool pack, rather than concentrate on a conceptual typology of these
variants.
Cognitive linguists such as Langacker (1988a) use a specific network
model to describe polysemy with nodes that are joined by semantically
plausible links (see Seto, this volume). I have tried to develop a more flexible,
loosely structured, and "contiguous" version ofthe model (Raukko 1994,1997)
where the nodes are only used as methodological tools and, in principle, a
polysemous semantic range is (ontologically) more like a flexible "mass"
than a "network".
Flexible meaning is not only about graded membership of categories,
fuzzy boundaries of categories and continuum effects - characteristics of (at
least ideal) prototype theory. It is also about the nonspecificity of meaning
and meaning type category membership. In Section 4.2 we will see how
within a given range of the polysemy of get some instances can be more
 Experiments with English get and Finnish pitlili 167

reliably allotted to a category than others. Some instances are vague or fuzzy
enough to simultaneously match two different categories, but we will find at
least two versions of this phenomenon. In some cases one can interpret the
same meaning in two methodological ways, and therefore the semantic
instance falls into a double membership of these (contradictory) categories.
In other cases one does not need to decide the proper category, although
related examples would fall into distinctive (neighbouring but still contra-
dictory) categories.

3. Experimental applications of flexible meaning

3.1. Experimental settings

I will here report on three types of tests in my experimental research on

polysemy: (i) production tests, (ii) difference evaluation tests, and (iii) sorting
tests. As this article concentrates on get and pitiiii, the reader can find in the
Appendices reproductions of (i) the whole of the questionnaire of the get
production test, (ii) the instructions part for the get difference evaluation test,
and (iii) the translation of the instructions sheet for the pitiiii sorting test.
I have devised the production test myself; I used the first pilot version in
Helsinki in 1991 (Raukko 1993). The mature version (used since 1994)
contains four questions on a questionnaire, of which an example (the case of
get) is reproduced in Appendix 1. The informant (or subject) is asked to
produce examples (sentences or phrases) each containing the lexeme get (in
any form) so that each example portrays the word get in a meaning different
from all the rest that the informant produces. This is the outline of question
No. 1 in the questionnaire, and question No. 4 provides the analyst with the
crucial help of paraphrases of the respective examples in which the informant
is asked to avoid using the word get.
The analysis is based on 329 informants' responses to the get production
test. Most of these 329 informants were high school students in two schools:
Putnam City North High School, Oklahoma City, Oklahoma, USA; and Rider
High School, Wichita Falls, Texas, USA. There were also 23 university
students of English at the Midwestern State University, Wichita Falls, Texas,
and a few teachers from each school. The experiment was carried out in
March and April 1994; I was guiding the experiment on location (see Raukko
1999 for further details).3
168 lama Raukko

The difference evaluation test is also called the similarity rating test, and
my version follows the traditional setup (Lehrer 1974; Caramazza and Grober
1976; Colombo and Flores d'Arcais 1984; Gibbs et al. 1994; Sandra and
Rice 1995). The idea is to go through pairs of sentences which each contain
two uses of get and evaluate the semantic distance between the two uses in
each case. The function of this test is more confirmative in nature, and I
consider its reliability far lower than that of the production test. The reason
is that multiple choice tests require far less commitment, creative effort and
explicit signs of successful understanding of the instructions from the infor-
mant than tests where one has to write down one's own sentences, comment
on them and paraphrase them. Another reason is that when the informant
faces pairs of two stimulus sentences one pair at a time, it is much easier not
to take the question of semantic distance seriously than in the case of the
sorting test (introduced next below), where one needs to look at the whole of
50 stimuli from the outset and work much harder on them. It is no surprise
that the sorting test requires much more time and expertise from the informant
than the difference evaluation test. However, Section 4 will make two short
references to the results of the get difference evaluation test. The locus of
the experiment was the same as in the production test, i.e. USA 1994, and I
had 79 informants (see Appendix 2).
The sorting test also follows a tradition, although I have devised my own
elaborate version for my research on the Finnish verb pitiiii. The idea of
sorting stimuli is familiar from psychology and psycholinguistics, and for
polysemy I have seen it used by at least Colombo and Flores d' Arcais (1984)
as well as Sandra and Rice (1995). I have performed the production test and
the difference evaluation test for pitiiii (in Helsinki, during 1995-1999) as
well, but the main emphasis in Section 5 will be on the results of the sorting
test (in Helsinki, during 1998). The design and usability of the sorting test
draws heavily on the production test, starting from the fact that the stimulus
sentences originate in the production test responses. As the two tests address
the same problem of "how to categorize meanings of pitiiii" from two different
perspectives, the analyses of the two types of experiments are necessarily
interlinked. The pitiiii sorting experiment was conducted in October 1998 in
Helsinki with 21 informants who were university students, either majoring
in general linguistics or taking an introductory course in general linguistics.
The English translation of the instructions part from the original Finnish
questionnaire can be found in Appendix 3.
 Experiments with English get and Finnish pitiiii 169

3.2. How the experiments and analyses operationalize flexibility

Let us see how the experiments, or tests, help us grasp the six shades of
variability (listed in Section 2.2), or help us operationalize the research
questions pertaining to flexible meaning. The following is an illustrative list
of examples, not an exhaustive analysis.

(i) Nonfixedness. In the production test, instead of giving the informants

a ready-made selection of instances that would display different mean-
ings of a word, I ask them to produce these instances. They create
their own personal view of polysemy by means of giving examples,
commenting on the centrality of and links between the instances, and
giving paraphrases for the examples. Although ready-made examples
are given to informants as stimuli in the sorting test, the quantitative
and qualitative properties of the divisions (cf. meaning types) are not
fixed in advance, but the informant can establish them by her- or
himself.
(ii) Fuzziness, unclarity. I am using three different test types because they
approach the same issue from different angles, and even if we never
quite get into the "meaning itself', we have three kinds of evidence
suggesting certain trends. When I analyse the results, I do not aim at
clear-cut boundaries or discrete categories that include all possible
data, and I allow for uncertainty even in the final results of the data
analysis.
(iii) Flexibility is present in many details of the experiments and my
analysis of the results. Firstly, I have not fixed the territory of the
word under study too strictly in advance. If in the test for the word
get there appear examples with forget, I accept them and they tell me
something about the way my informants see boundaries between
categories. Secondly, if in the tests for back and date (also performed
in the USA in 1994) all possible word-classes and even compounds
occur, such as back-up, then this is evidence that people feel that
word-class boundaries do not form boundaries for semantic analyses
of polysemous words, as many traditional schools of thought have
believed.
(iv) Ambivalence. Even if the setting of the production test makes the
informants use a metalanguage to "translate" linguistic examples that
they have produced on their own (in the form of the paraphrasing
170 iarno Raukko

task), this does not guarantee disambiguation and nonflexibility. On

the contrary, paraphrases can distort the "intentions" of the informants
or lead interpretations astray or contradict with examples. Of course
methodologically I have to build my analysis on the paraphrases, but
in the end, paraphrases are not an omnipotent metalanguage. They
are just as much subject to flexibility effects as the original examples.
(v) Intersubjective differences. One of the main concerns of an experiment
is the search for both similarities and differences in the responses. In
some sense similarities - that is, intersubjective agreement - are more
relevant for the presentation of the results, if the objective is to describe
the most important aspects of the polysemy of a word. These are the
most basic social conventions about the use of the word. However,
differences must also be taken into account. They can tell us not only
about differences of opinion, but also about different perspectives on
the same matter. Some examples of the use of a word can have double
membership on the "map" of the semantics of that word. Some ex-
amples are so fuzzy that it is no use trying to fix them into a certain
point on that same map. The most reliable way to study intersubjective
variability in practice is to set up an experiment with native informants.
(vi) Polysemy. It is clear that all my experiments try to give a richer
understanding of polysemy.

It is clear that semantic studies lose depth if variability is ignored. The

acknowledgement of variability has important methodological consequences,
and I have chosen to tryout experimental methods for this end. Although
they cannot be used any more objectively than any other method, they have
important qualities which directly address facets of semantic variability.

4. Flexibility effects in the polysemy of get

This section focuses on the "problems" in the analysis of the polysemy of

get. After the introductory overview on the polysemy as a whole, it takes
three excursions to specific parts of the polysemy. Even if the experiments
are designed to pave the way for a richer analysis of polysemy, it should be
kept in mind that the problems are intrinsic in nature and therefore partly
insoluble. There is no escaping from the fact that in the polysemy of get there
are several points where examples and instances do not fall into categories
 Experiments with English get and Finnish pitaa 171

neatly. Categories can be seen as continuums, but they also overlap and yield
insoluble categorization situations. And as pointed out in Section 2.4, flexible
meaning is not only about borderlines of categorization: in many cases the
semantics itself of the instances is so fuzzy that it is difficult to interpret in
the context of the polysemy of a word. We shall view these "problems" as
supporting the theory of flexible meaning instead of leaving us with an
unsatisfactory outcome.
The polysemy of get forms an exceptionally abundant and complex net-
work of meanings and ranges of flexible meaning. This abundant poly se my
of get has been discussed by e.g. Kimball (1973), Niedzielski (1976) and
Lindstromberg (1991), whose main goal has been to provide a neat reduction-
istic representation (or even matrix).4 Lindstromberg points out (1991: 289),
however, that while his analysis mainly aims at simplicity (i.e. the polysemy
of get can be explained via extensions from the meanings 'obtain' and 'seize' /
'take hold of'), his approach is also more prototype-based and cognitive-
semantic in nature than that of Kimball and Niedzielski. I find Lindstromberg's
analysis of fairly little use, because my aim is not explanatory simplicity but
descriptive "relevance": I aim at concentrating on "central" and "crucial"
features in the polysemy of get partly from the perspective of a speaker who
has no training in linguistics (i.e. a typical informant of mine). Needless to
say, I am looking at colloquial and nonstandard (American) English, and
because my informants are mostly American teenagers, I do not make any
claims with respect to regional and social coverage and generalizability (see
Raukko (999).5
To get a first taste of some delicious details in the polysemy of get, here
are some introductory examples that in themselves are flexible in meaning.

(0) a. I didn't get that. What did you say? (Lindstromberg 1991: 289)
b. Part of being a nation is getting the history wrong. 6

The instance of get in (Oa) can be analysed at least as (i) 'understanding'

in the sense of 'grasping the content'; (ii) 'understanding' in the sense of
'hearing the words'; and (iii) 'hearing the words' in the sense of 'catching'
them. Getting in (Ob) can refer to (i) 'understanding' the history wrongly;
(ii) 'interpreting' the history wrongly; (iii) 'reconstructing' the history in a
wrong form; and (iv) 'making' the history wrong, causing a change of state,
creating an image of history that is wrong.
In order to understand the place of the meaning types that will be in
172 lama Raukko

focus in Sections 4.1-4.3, a brief overview of the major categories in the

polysemy of get is necessary. This overview is the outcome of a long and
complex analysis of the results of the 1994 get production test experiment
(partly reported in Raukko 1999). The meaning types listed below are the
result of the method used, and they concentrate on the most salient categories
(and the most salient features of the categories), and are continually subject
to processual changes.
In a very general and abstract manner, the polysemy of get can initially
be divided into such large groups of meanings as (i) CHANGE OF POSSESSION;
(ii) CHANGE OF LOCATION; (iii) CHANGE OF STATE; and (iv) static meanings.?
However, when we look at the responses of the 329 American infonnants in
the get production test, we can abstract out a slightly different pattern of four
most salient high-level meaning types: (a) OBTAINING; (b) RECEIVING; (c)
MOTION; and (d) CHANGE OF STATE; where (a) and (b) correspond to (i), (c) to
(ii) and (d) to (iii). In the case of each type, at least 70% of the informants
produced at least one instance of the type, and thus differentiated it from all
other meaning types. When we go further down to the "basic-level" meaning
types that these productions most often represent, we find four corresponding
but specified types: (a') CONCRETE OBTAINING FOR ONESELF; (b') CONCRETE
RECEIVING; (c') CONCRETE (non-causative) MOTION; and (d') EXPERIENTIAL CHANGE
OF STATE. The following sentences are prototypical examples of these basic-
level meaning types, respectively from (a') to (d'). They are prototypical in
at least two senses: they have the same characteristics as a more abstract
outline of a prototype of the given meaning type would have, and they are
examples which frequently occurred either in identical form or in safely
comparable forms.

(1) She had enough money to get a car.

(2) What are you getting for your birthday?
(3) Get out of here.
(4) I'm getting tired.

The meaning types behind these examples are the top salient meaning
types in the polysemy of get. Some others, comparable to these but which
are not part of the salient higher-level groups (a) through (d) are: (e) UNDER-
STANDING; (f) STABLE POSSESSION; (g) OBLIGATION; and (h) ABILITY (or 'pennission').
Examples of these are given in (5) to (8) respectively.
 Experiments with English get and Finnish pitiiii 173

(5) I don't get it.

(6) I've got a headache.
(7) I gotta go.
(8) I get to go to Dallas on Saturday.

UNDERSTANDING and STABLE POSSESSION are related to the abstract type

CHANGE OF POSSESSION, although they are more or less static in meaning. The
modal uses OBLIGATION and ABILITY are also fairly static, and at least the latter
is related to cases of CHANGE OF STATE with a verbal complement, such as
STARTING (Lets get going) and REACHING A STAGE (Getting to know a foreign
language is hard work).
But why should OBTAINING and RECEIVING be different meanings in such a
salient way? Why should not STABLE POSSESSION be more closely linked to
CONCRETE RECEIVING, as we can have (colloquial) cases such as I got a car
where it may not be specified whether the speaker refers to a change or a
state of her/his possessions. The answer to these questions is: categories leak;
we cannot establish such meaning types too tightly. Section 4.1 focuses on
the question of OBTAINING vs. RECEIVING. Section 4.2 tackles a continuum
between two high-level abstract groups. In the group of CHANGE OF LOCATION,
there are basically two parts: non-causative MOTION and causative CARRYING.
Instances of the latter, however, can come very close to certain types of
OBTAINING and even some other meaning types, and this is where we see the
necessary continuum. Section 4.3 questions the independence of both ABILITY
and OBLIGATION vis-a-vis verbal types of CHANGE OF STATE.

4. J. Obtaining vs. receiving

According to my findings, meaning types in the polysemy of get that have to

do with 'possession' (and that it is plausible to distinguish) include OBTAINING,
RECEIVING, CATCHING ETC., UNDERSTANDING and STABLE POSSESSION. While the
last two are static, CATCHING ETC. is a special type of CHANGE OF POSSESSION:
such types of meanings have to do with, e.g., CONCRETE CATCHING (Get the
ball), ANSWERING (I will get the door) and HURTING/KILLING/PUNISHING (I'm
gonna get you punk). Thus, I argue that UNDERSTANDING and CATCHING ETC. are
specialized enough not to belong to the type METAPHORICAL RECEIVING, which
is probably their historical source.
As for OBTAINING and RECEIVING, the two main poles of CHANGE OF
174 lamo Raukko

POSSESSION,I have suggested in Raukko (1993, 1995, 1999) a variety of

subordinate meaning types. Table 1 presents these types in a matrix form,
which is not an experimental result as such.

Table J. Types of RECEIVING and OBTAINING, in the core of CHANGE OF POSSESSION. All
example sentences are taken from the production test responses.

Beneficiary = subject Beneficiary = (direct or

prepositional) object

Subject is non-agentive Subject is agentive

Concrete CONCRETE RECEIVING CONCRETE OBTAINING CONCRETE OBTAINING

(What did you get FOR ONESELF FOR SOMEONE ELSE
for Christmas?) (Lets go get some (Please get me a drink.)
bread.)
Meta- METAPHORICAL METAPHORICAL METAPHORICAL
phorical RECEIVING OBTAINING FOR OBTAINING FOR
(I never get my way ONESELF SOMEONE ELSEK
or I got an A on the (Get a life!) [Not produced by my
test.) informants]

The actual responses of the informants do not form such a neat picture as
Table 1 does. In particular, there were several informants who not only made
differentiations among these five types, but also differentiations within one
or more of these types. The clearest trend of all is that within METAPHORICAL
RECEIVING there are three sUbtypes which are sometimes distinguished from
one another rather systematically, namely 'receiving a grade', 'catching an
illness' and 'other types'. The strong representation of 'receiving a grade' is
of course explained by the composition of the informant population. Apart
from METAPHORICAL RECEIVING, we find type-internal differentiations in many
responses, but no clear patterns among informants.
The problem is more general: how do we find organization in the rather
chaotic-looking, varied outcome of the responses? When addressing the prob-
lem of motivating the distinction between OBTAINING and RECEIVING, it is safe
to base the hypothesis on the concrete basic-level types, CONCRETE OBTAINING
FOR ONESELF and CONCRETE RECEIVING, which are often the source of vague
interpretations.
In the data, as many as 199 informants (61 % of the total population)
 Experiments with English get and Finnish pitiil:i 175

produced at least one instance of CONCRETE OBTAINING FOR ONESELF. I need to

say that this is an initial count, because this meaning type is at the very
centre of the discussion of problematic categorization. As many as 47 infor-
mants produced more than one instance, so that the overall number of
instances in this initial counting was 266. The most frequently used paraphrase
was buy, which occurred 88 times. Other paraphrases that clearly indicate an
agentive and intentional role on the part of the subject were obtain (20
instances), purchase (19), pick up (17), retrieve (13), take (9) and grab (3).
Thus 169 instances (64% out of 266) would at least partially meet the criterion
of prototypical agentive obtaining. Yet even some of these examples could
well slide towards other meanings: pick up could be said to contain a possible
CARRYING component (see Section 4.2).
On the other hand, 162 informants (49% of the total population) produced
at least one instance of CONCRETE RECEIVING. Half of these (82 informants)
used receive as the paraphrase. It turns out that as many as 104 of these
informants (64% out of 162) produced both CONCRETE OBTAINING FOR ONESELF
and CONCRETE RECEIVING, that is, they distinguished these meaning types from
one another. We thus have 104 cases of a pair of sentences that can be seen
to manifest a clear difference between the meanings. Let us take a look at
two such pairs; (9a) and (9b) come from my get informant No. 19, and (lOa)
and (lOb) from informant No. 259; (9a'), (9b'), (lOa') and (lOb') are the
paraphrases written by the same informants, respectively.

(9) a. I got a car when I turned 16. 9

a'. I received a car when I turned 16.
b. I went to the store to get a CD.
b'. I went to the store to purchase a CD.
(10) a. I got my prom dress in August.
a'. 1 bought a prom dress in August.
b. I have gotten three letters from a friend.
b'. I have recieved [sic: her spelling] three letters from a friend.

Without the paraphrases it would be difficult to decide whether the examples

refer to OBTAINING or RECEIVING. In examples like (9a) one gets extra hints
from the context - when you turn 16 it is your birthday and it is customary
to get presents then - but at least (lOa) would be vague in this respect without
the paraphrase.
Not surprisingly, the material is filled with examples that are not easy to
176 lama Raukko

classify, some of them because they lack a paraphrase, others because the
object of getting, the paraphrase and the encyclopaedic knowledge about
the things talked about are necessarily unclear with respect to the com-
ponent of agentivity. Consider examples (11)-(13) (and the corresponding
paraphrases) which were the only instances of any type of (CONCRETE) CHANGE
OF POSSESSION that the particular informant (Nos. 124, 171 and 276 respec-
tively) produced.

(11) Let'sgetthis.
( 1I ') Let's acquire this.
(12) Igetacoke.
(12') I picked up [sic: past tense] a coke.
(13) I need to get a new car.
( 13 ') I need to acquire a new car.

Acquire as such is as polysemous as get in this respect. One could argue

that Let's more probably invokes an agentive reading than a non-agentive
one; that pick up may be an American colloquial synonym for buy; and that
need may bring into mind that the speaker is planning to do something
agentive about the problem. But still we are on soft ground with examples
like these: the distinction between OBTAINING and RECEIVING seems to start
losing its relevance.
As a large proportion of the instances contain elements that make them
unclear cases, the temptation to do away with the distinction OBTAINING vs.
RECEIVING grows. Yet at the same time we find a great number of clearer
cases, some of them providing us with neat pairs of prototypical examples of
the two types, distinguished by the informant as different meanings of get.
After all, distinguishing (examples of) meaning types was what the production
task was all about, so if an informant comes up with more than one example
of CHANGE OF POSSESSION, we always have to give at least some explanation
and description of the similarities and differences between the two or more
examples.
The field of OBTAINING and RECEIVING, as presented in Table I, is so compli-
cated that we have to deal with several parameters at once. Some informants
may show us an example where they want to distinguish CONCRETE OBTAINING
FOR SOMEONE ELSE from METAPHORICAL RECEIVING (for oneself), calling our
attention to three differences at the same time: (i) concrete vs. metaphorical;
(ii) agentive vs. non-agentive; and (iii) for someone else vs. for oneself. Yet
 Experiments with English get and Finnish pitiiii 177

we the analysts may be led to search only for one minimal difference in each
pair of sentences that the informants have produced.
The other test type, difference evaluation, gives us some additional evi-
dence for the fact that the distinction is a difficult one. The pair in (14),
which contained CONCReTE OBTAINING FOR ONESELF and CONCRETE RECEIVING,
was not only judged as relatively different in meaning (the tenth highest
"different" rating among the tested 31 pairs, with average rating of 2.13 on
a scale from 0 to 4), but the judgements also deviated quite remarkably: the
standard deviation for this pair was the fifth highest among the tested pairs. 10

( 14) a. I went to get a candy bar.

b. I got a car for my birthday.

The analysis of particular points in the polysemy of get requires sensitivity

for continuum effects, generality effects and other kinds of flexibility. Ulti-
mately we also need to ask ourselves why so many informants seem to
consider the distinction between (agentive) OBTAINING and (non-agentive)
RECEIVING so important in the otherwise abundant and complex polysemy of
get, even if the distinction is partly too flexible to establish. The methods I
have used so far cannot answer that question.

4.2. Obtaining vs. carrying

A nice example to demonstrate the intricate but tricky difference between

and CONCRETE CARRYING comes from Lindstrom-
OBTAINING FOR SOMEONE ELSE
berg (1991: 291) who does not comment very much on the difference in
meaning:

(15) I got the letter to her [the letter, to her].

(16) I got her the letter [the letter, for her].

Lindstromberg terms both of these instances 'change of possession'. (Cf.

Giv6n and Yang's (1994) somewhat different treatise.) At first glance, it would
seem that (15) has more to do with motion than (16), while in (16) change of
possession is highlighted. However, a more careful look reveals that both of
these sentences can easily mean both things. Both of them could be para-
phrased as "I took/brought/carried the letter to her" and "I gave/handed/(made
178 lamo Raukko

her have) the letter", but we usually tend to think that if one of them had
more to do with 'motion' and hence CARRYING, it would be (15), maybe
because of the partly locative preposition to.
Some of my informants produced examples and paraphrases that portray
a possible difference between CONCRETE OBTAINING FOR ONESELF and CONCRETE
CARRYING; the following come from my get informants Nos. 6 and 77. (17a)
and (18a), which are incidentally both paraphrased with take, are the ones
that I would like to classify as CARRYING.

(17) a. I am getting something out of the closet.

a'. I am taking something out of the closet.
b. I have to get a new car.
b'. I have to purchase a new car.
(18) a. Get your hands off me!
a'. Take your hands off me.
b. She will get the milk from the store.
b'. She will pick the milk from the store.

In my scheme, it is typical for OBTAINING not to have locative complements

(as in [17a] out of the closet), but this is by no means impossible. In (18a)
the locative phrase off me highlights the motion component clearly more
than from the store in (18b), while (18b) may refer to a commercial transaction
as well. In some sense, however, (17a) and (ISb) are similar instances, and
yet they have to be classified differently, because the contrasting instances
are so different. Thus (17b) and (I8a) are at the opposite ends of a scale,
while (l7a) and (l8b) fall in the middle.
The difference evaluation test again brings us one relevant example pair
whose meaning difference informants strongly disagreed upon.

(19) a. Could you come and get me after school?

b. I am getting something out of the closet. (= 17a)

The average and standard deviation figures are almost identical to the
ones for the evaluated pair OBTAINING vs. RECEIVING discussed in Section 4.1.
Thus, informants again differ greatly in their opinion about the similarity or
difference of the meanings of get in (l9a) and (l9b), which supports the
analysis of OBTAINING and CARRYING as being idealized abstractions forming
the ends of a continuum of flexible meaning where 'change of possession'
 Experiments with English get and Finnish pitaa 179

and 'motion' are present to varying degrees. Even the component 'change of
state' is sometimes present as well, but in cases like getting something into
shape we are already dealing with METAPHORICAL CARRYING instead of the
concrete one.
There are also situations where the informant has produced three related
instances, and while two represent OBTAINING and CARRYING, the third one can
be classified as CATCHING ETC. - e.g. Get the ball, or I will get you.
Yet there are many cases where the informant has produced only one
example relating to our present discussion, and it is relatively impossible to
try and firmly classify the meaning type as either OBTAINING, CARRYING or
CATCHING ETC.

(20) He went to get his money from the drawer.

(20') He went to acquire his money from the drawer.
(21) Get me that pen.
(21') Fetch me that pen.
(22) Will you get your plate?
(22') Will you move your plate?

All these examples have components of at least two abstract meaning

types at the same time. Thus, they support the point about the importance of
flexibility in the analysis of polysemy.

4.3. Ability vs. reaching a stage vs. obligation

There are three basically distinct meaning types of get that all take an
infinitival complement: ABILITY, REACHING A STAGE and OBLIGATION. The first is
a technical term for meanings like 'permission' and 'ability', which were not
usually distinguished by informants and therefore do not occur as separate
meaning types in the analysis. REACHING A STAGE belongs to the macro-type
CHANGE OF STATE, and hence its meaning is more dynamic than that of the two
modals (OBLIGATION and ABILITY). Besides having obvious differences as to
aspects of modality and semantics, the three types favour different verb forms
of the main verb get. While OBLIGATION is supposed to be bound to the form
got (which appears as have got to, has got to, 've got to, S got to, got to,
gotta, gots to or don't got to), it is prototypical for ABILITY to occur in the
simple present tense (or future), while REACHING A STAGE is the most flexible
180 lama Raukko

morphologically, and "syntactically more marked" instances like I'm getting

to ... tend to be interpreted as such non-modal uses. In the following the first
two examples and paraphrases (23a-b) show informant No. 246 making a
distinction between ABILITY and REACHING A STAGE, and (24) is the most frequent
instance of OBLIGATION in the get production test, with two frequently occuning
paraphrases.

(23) a. 1 get to be on Television [sic: the informant's capitalization].

al. I'm going to be allowed to be on television.
b. I'm glad 1 "got to know you" [sic: the informant's quotation
marks].
l
b I'm glad we became acquainted.
•

(24) I've got to go.

(241) 1 need to go. /1 have to go.

The relationship between ABILITY and REACHING A STAGE is problematic

and partially overlapping, so that there are again many cases that are difficult
to classify. In informant No. 246's case the difference seems fairly clear:
getting to be on television can be considered a rare occasion, something that
is possible only because circumstances make it so, while getting to know
someone is usually considered a more everyday process.·· This analysis is
supported by the paraphrases, which in general are the only basis for distin-
guishing between ABILITY and REACHING A STAGE. Be able to and be allowed to
are, by default, likely paraphrases of ABILITY. Be allowed to occurs in 18
responses and be able to in 13 responses (together in 31 responses, or 53%,
out of 58 instances of ABILITY in total). In contrast, the omission of get is a
fairly typical way to paraphrase REACHING A STAGE. (E.g. one informant para-
phrased her I'd like to get to know you as I'd like to know you better.) There
is also a strong tendency in my informants' responses with ABILITY to take up
a certain lexical complement, i.e. the verb go. Forms of get to go occurred in
41 responses out of 58 (71 %). This collocation seems to be very central in
teenage American English. The most difficult form of get in this context is
got, because it can occur with all of the three meaning types. A suitable
example is 1 got to go, where both intonation and the paraphrase might
"disambiguate" the meaning into one of the three.
OBLIGATION is to a large extent bound to the form got (for variants, see
above), although I claim that there is one exception in my production test
data: informant No. 155 paraphrased 1 get to go to school today as 1 had to
 Experiments with English get and Finnish pitliii 181

go to school today, and although there is some confusion with the tense
choice, "going to school" would at least be a likely candidate for a complement
of OBLIGATION. Moreover, although an example like Why do I always get to
do the dishes? did not occur in the production test data, it could be seen as
marking the path between REACHING A STAGE and OBLIGATION. I claim that these
examples suggest variability for OBLIGATION, although the standard view limits
this meaning type to the form got.
I hope to have shown that despite the significant semantic differences (on
the surface) between the three meaning types with an infinitival complement,
there is a continuum from ABILITY via REACHING A STAGE to OBLIGATION. Once
we have a continuum, it will be difficult to neatly classify instances of get
into meaning types. At least the distinction between ABILITY and REACHING A
STAGE is hard to establish solidly, even if we have prototypical instances of
both and see informants distinguishing the two meaning types. We have seen
again that the relationship between these two meaning types is best understood
in terms of flexible meaning, where there is no strict division into two isolated
meaning types.

5. Flexibility manifested in the polysemy of pitiiii

Pitiiii is a Finnish verb with extensive polysemy. Some of its meanings

correspond to the polysemy of the English verbs hold and keep, but two
other salient and fairly distinct uses can be translated as like / be fond of and
must / have to. The latter items are more separable semantic nodes, they are
more "monolithic", whereas the ones that partly intersect with the semantic
spaces covered by hold and keep form an intricate network of flexible meaning.
In the case of get, we saw results of (primarily) the production test and
(secondarily) the evaluation test. I have performed similar tests for pitiiii, but
the main emphasis will now be on a third type of experiment, namely the
sorting test, which was briefly introduced in Section 3.l.
I will first give a brief overview of the polysemy of pitiiii as a whole in
the form of a list of glossed examples, and then concentrate on the areas
where flexibility and fuzziness are prominent features. The broader overview
is useful both for the understanding of the place of the flexible area in the
polysemy as a whole and for the assessment of quantitative results in the
general framework. The following set of examples (which were produced by
informants in the production test and later used as stimuli in the sorting test)
182 lamo Raukko

portrays a concise list of the most crucial meaning types in the polysemy of
pitaa. By "crucial" I refer to such salience factors as frequency and earliness
of occurrence in the production test, while the indicators for distinguishing
meaning types now mainly derive from the results of the sorting test. 12 The
list starts with LIKING (in [25]) and OBLIGATION (in [26]) (one of the meanings
shared by get and pitaa) and ends with six types (30)-(35) that belong to the
problematic area of very flexible meaning.

(25) Ma pidin tosta leffasta.

I-NOM like-PAST-SGl that-ELAT movie-ELAT 13
'I liked that movie.'
[LIKING]

(26) Minun pitiiii liihteii toihin.

I-GEN must-sG3 leave-INF work-PL-ILLAT
'I must leave for work.'
[OBLIGATION]

(27) Minii pidan sinua hulluna.

I-NOM consider-sG 1 YOU-PART crazy-ESs
'I consider you crazy.' / 'I think you are crazy.'
[CONSIDERING]

(28) Suksi pitaa tiiniiiin hyvin.

ski-NOM grip-sG3 today well.
'The skis have a good grip today.'
[MAINTAINING CONTACT]

(29) Pidii minua kiidesta.

hold-SG2-IMPER me-PART hand-ELAT
'Hold my hand' or more literally 'Hold me by my hand.'
[MANUAL HOLDING]

(30) Mii pidiin avaimia taskussa.

I-NOM keep-sG 1 keY-PL-PART pocket-INEsS
'I keep my keys in my pocket.'
[KEEPING IN A PLACE]
 Experiments with English get and Finnish pitiiii 183

(31) Pida suusi kiinni!

keep-IMPER-SG2 mouth-Acc-your shut (ADVERB)
'Keep your mouth shut.'
[KEEPING IN A STATE]

(32) Se pitiiii outoja vaatteita.

s/he wear-sG3 odd-PL-PART cIoth-PL-PART
'S/he wears strange/weird/odd clothes.'
[WEARING]

(33) Ma pidan huomenna juhlat.

I-NOM organize-sG 1 tomorrow partY-PL
'I'm having/giving a party tomorrow.'
[ORGANIZING]

(34) Pidii kiirettii!

keep-IMPER-SG2 hurrY-PART
'Hurry up.'
[UPHOLDING A CONDITION]

(35) Mina pidin lupaukseni.

I-NOM keep-PAST-SGI promise-Acc-my
'I kept my promise.'
[METAPHORICAL RETAINING]

The presence of flexibility becomes apparent in the variation of informant

behaviour in the sorting test. With the meaning types LIKING and OBLIGATION,
informants are quite unanimous (i.e. most informants group the same stimuli
in these groupings and name the groups in similar ways), whereas variance
increases when we go down the list.
The relationship between sentences (31) and (34) is quite intriguing. Firstly,
these sentences were grouped together by as many as 62% of informants (13
out of 21). Secondly, there is a risk that some informants may have looked at
a combining criterion that has nothing to do with the meaning types in the
polysemy of pitiiii, namely the imperative form; this criterion was mentioned
by three informants out of the above 13. Thirdly, not only are these two
sentences often grouped together, but they both tend also to be grouped with
other, somewhat different, sentences, such as (36).
184 lamo Raukko

(36) Pidii hauskaa!

keep-IMPER-sa2 fun-PART
'Have fun!'

Sentence (36) was grouped with (34) by as many as 17 informants (81 %),
but with (31) only by nine informants (43%); the difference is statistically
significant. In contrast, there were ten stimuli (out of 50) that were sometimes
sorted with (31) but never with (34), and conversely, only two that were
sometimes sorted with (34) but never with (31). I think all this shows that
(31) is a much more difficult stimulus sentence than (34), even if on the
surface (31) would seem less idiomatic and more "basic" in meaning than
(34). Sentence (36), for its part, was surprisingly sorted together with (33)
by as few as two informants, although "having fun" and "having a party"
could have been easily thought of as attracting similarity judgements. One
possible reason is that there were two other stimuli that were thought of as
matching (33) so well that there was no demand for other group members.
Let us look at one more example, namely (37). The interesting finding is
that (37) was sorted together with each of (38) through (40) by eight to ten
informants, but (38)-(40) in turn were sorted with one another by only two
to five informants. The hypothesis is that (37) seems to be a combining link
between the other three instances (38), (39) and (40), while these three are
not so easily linked without the presence of (37).

(37) Minii pidiin sinut vaikka viikisin.

I-NOM keep-sa 1 YOU-ACC even by-force
'I will keep you by force if needed.'

(38) Puut pitiiviit lehtensii pitkiille syksyyn.

tree-PL-NOM keep-PL3 leave-Acc-their long-ALLAT autumn-ILLAT
'Trees keep/retain their leaves till late in the autumn.'

(39) Minii pidiin sinua lukitussa komerossa.

I-NOM keep-sa 1 you-PART 10cked-INESs closet-INESs
'I (will) keep you in a locked closet.'

(40) Pidiin asian salassa.

keep-sal matter-ACC secrecY-INESS = 'in secret'
'I (will) keep the matter secret.' or 'I will keep it as a secret.'
 Experiments with English get and Finnish pitiiii 185

Thus (37) can be seen as a flexible channel or path between the other
three. Moreover, (39) was sorted together with (30) by half of the infor-
mants - an expected result considering the seeming similarity between the
two sentences - but (40) was grouped with (31) by only one informant,
despite the seeming similarity between the two instances.
We are left with a very unpredictable and intricate network of inter-
relations within the area of flexible meaning which cannot be reduced to
established categories.

6. Conclusions

There are two ubiquitous phenomena in linguistic semantics. One is polysemy:

the phenomenon that one form can be used for more than one meaning. The
other has to do with the indeterminacy, nonfixedness, fuzziness, negotiability,
variability and flexibility of the semantic values of all linguistic units. My
point is that these are different sides of one and the same problem. However,
this insight is only rarely exploited by linguists, at least when they are per-
forming their actual practice of linguistic analysis and theorizing. And yet
within the cognitive linguistic framework it has become almost a truism to
say that all lexical items are polysemous (see e.g. Langacker 1988b: 50-51).
If it is accepted that practically all morphemes have not just one meaning,
we can rephrase this claim in the form that practically all morphemes have
flexible meaning. Thus we are moving from the first ubiquitous phenomenon
to the second one. The difference remains that, while polysemy is generally
thought of as a more permanent property of the semantics of a language - a
product, so to speak - semantic indeterminacy is usually seen as a processual
phenomenon. I have chosen to tryout experiments in search of a richer
understanding of polysemy as flexible meaning, and whichever experiment
one chooses, there will be flexibility effects in the results.

Appendix 1. The questionnaire of the get production test

April _, 1994
female/male
Senior/Junior/SophomorelFreshman
age_
186 lamo Raukko

Dear student of Putnam City North High School [Oklahoma City]. My name is
larno Raukko; I am a researcher of linguistics at the University of Helsinki, Finland,
and I am studying some aspects of everyday English. I very much appreciate your
participation in this test.

(I) The verb get (get, getting. got. gotten) is very common in spoken American
English. and it seems to be used in very many different ways. You might say that
get has different meanings. Could you please write down examples of the ways
in which you use the word get differently. Your examples can be full sentences,
but they do not need to be. You do not have to fill in every blank from A to K;
just produce as many examples as you can think of where the use of get still
differs from every other use you have mentioned.

[Additional instruction: if I had to answer a similar question concerning the

word bank, I might produce examples like the following:

A. He had enough money left in his bank account.

B. She followed the man along the river bank.]

Now please write down examples with get. Feel free to use casual and colloquial
style! These examples will not be corrected!

A. ____________________________________________
B.
C.
D. __________________________________________
E.
F.
G. __________________________________________
H. __________________________________________
I.
l.
K. ________________________________________________

Please turn over when you have finished answering question No. 1.

(2) In your examples, would you say that one or some of the uses of get is more
typical or central or important than the others? Is one of them the most central
one? If you think so, please write down the letter(s) of the example(s). Also
explain briefly why you think so.

(3) Do you think that the uses of get in your examples are somehow connected or
 Experiments with English get and Finnish pitiiii 187

linked to one another? Is there a common idea? If you think so, please explain or
specify briefly.

(4) Finally, would you please once again read each of your examples for question
No. I and try to express the same idea in another way. In other words, for each
example, write down an alternative way of saying the same thing so that you
don't use get any longer.

[If I had to do this for "She followed the man along the river bank" and I had
to avoid using the word bank again, I could write "She followed the man along
the side of the river".]

Thank you very much for your help.

Appendix 2. The questionnaire of the get difference evaluation test

April_, 1994

Dear student! My name is larno Raukko; I am doing my Ph.D. in linguistics at the

University of Helsinki, Finland. I am studying some linguistic aspects of English,
and I would need your help in making judgements about meaning difference and
meaning similarity. Below you find sentences that were produced by American high
school students in a test of a slightly different kind than this one. The sentences are
grouped in pairs, and you have to compare two sentences at a time. Each sentence
has the word get in it, and you are supposed to decide whether that word carries the
same meaning or two different meanings in the sentences. Additionally, you can
decide the degree of the difference (from 0 to 4). I give you an example of how I
might understand meaning difference (= MD).

In these sentences get has the same meaning (MD =0):

I am getting tired.
The dog got sick.

In these sentences get has a somewhat different meaning (MD = 2):

I am getting tired.
A guy got me pregnant.

In these sentences get has a very different meaning (MD =4):

I am getting tired.
Did you get his joke?
188 lamo Raukko

Now I would like you to estimate the meaning difference between the uses of get in
the following pairs of sentences:

MD
I got home late.
01234
I got a new car.

Did you get any mail today?

Did you get his joke today?
o I 234

[etc.)

Appendix 3. The questionnaire of the pitiUi sorting test. (Translation)

RESEARCH QUERY ABOUT THE MEANINGS OF THE VERB PITM

initials _ _ _ _ __
year of birth _ _ __
female/male
main subject _ _ __

\.(a) In the enclosed envelope you will find 50 cards, each with a sentence containing
the word pitiiii. Could you please sort these instances of the meanings of the
verb pitiiii into groups where each member shares a "similar enough" meaning.
Let the sorting be solely based on the meaning of the word pitiiii, not on the
meaning of the sentence as a whole - even if the meaning of pitiiii is understood
on the basis of the sentential context. For the purposes of this research, it is not
essential, either, e.g. which form the verb pitiiii occurs in.
(The sentences are numbered only in order for you to be able to refer to those
numbers in some of the questions to follow.)
Sort the sentence cards into piles. A pile can consist of any number of
sentences, and a pile can also consist of one sentence alone. When you have
finished piling, join the cards in each pile with a small paper clip (found in the
envelope), so that it will be easier for me to collect the piles in the end.

(b) If some sentences are especially difficult to sort, mark a "D" on such sentence
cards.

(c) Can you think of names for these piles - i.e. for these types or classes of
meanings? (For example, you can use some other Finnish verb, which roughly
 Experiments with English get and Finnish pitiia 189

means the same as the meaning of the pitiiii in question.) In that case, write that
name on the topmost card.

(d) When you were doing the pile sorting, did it occur to you that some meanings
were quite close to one another but not as close as the members of one pile?
Would you be able to combine some piles into classes at an upper level? Use
the bigger paper clips for combining.

2. The piles that you sorted can be said to correspond to meaning classes. In your
opinion, what are the three most important/central meaning classes? Refer to the
names you used or to the number of the topmost card.

_ _ _ _ _ _ and _ _ _ _ _ _ and _ _ _ __

(Can you specify, in case it is unclear, whether you refer to a bottom-level

category or a higher-level category?)

3. Continuation to task No. 2: please choose one sentence (i.e. one card) out of
each of the three most important classes which best represents the respective
meaning class, in your opinion. Refer to the numbers on the cards.

_ _ _ _ _ _ and _ _ _ _ _ _ and _ _ _ __

4. The term polysemy is usually used to mean that one word (or some other item)
has several meanings which are "related to one another" in a plausible way. The
meanings have in other words developed from one another in the course of time,
and even the present-day speaker can in principle think of the connection between
the meanings. Do you feel that all of the meanings of pittltl that you dealt with
are in this manner "related to one another", or is one of them / are some of them
very much apart from the rest?

Yes, all meanings seem A large proportion of the meanings

to be related to one another. are probably related to one another,
but the following is/are very separate:

(Can you specify, in case it is unclear,

whether you refer to a bottom-level
category or a higher-level category?)
190 lama Raukko

Could you now place the card piles into the envelope together with this questionnaire.

Thank you very much for your help!

Notes

I. Roy HaITis (personal communication) belongs to the latter type, who commented
on Raukko (1997) that polysemy is a pseudo-problem in integrationist linguistics,
because it is only a feature of langue. Although a possibly relevant point, the
consequent discussion would ultimately concern the extension of polysemy, and
I see my attempt to group together polysemy, flexibility, fuzziness, and situational
variability as an attempt to bring polysemy also very much into the realm of
parole.
2. Langacker for instance (1988a: 136) argues that the distinction between polysemy
and homonymy is a matter of degree and cannot be reduced to a simple dichotomy.
3. I wish to express my gratitude to all the 329 anonymous informants as well as
the people who helped me in organizing the 1994 experiments, especially WaIter
Bower, Harry Brown, Kit Johnson, Melanie McGouran and Jackye Plummer.
4. I would also like to point out interesting ongoing research on the development of
the polysemy of get in children, done by Simone Duxbury (Monash University,
Melbourne, Australia) and Brigitte Nerlich, Zazie Todd and David Clarke (see
Nerlich, Todd and Clarke, this volume).
5. The American flavour is represented for example in the distinction between have
got ('possess', 'must') and have gotten (for all other meanings) and in the
American spelling used in all of the examples, which authentically come from
my American informants.
6. This sentence was used as an article title by my Helsinki colleague Pekka Kuusisto,
and the interpretations below reflect the discussion we had over the title.
7. Henceforth I will use SMALL CAPITALS to indicate technical terms that function as
names of meaning types that my experiments and my analyses suggest. Less
technical descriptions are given in single quotes, such as 'permission' a few lines
below.
8. METAPHORICAL OBTAINING POR SOMEONE ELSE is a fairly unimportant meaning type,
if not virtually non-existent, and in this matrix it mainly serves analogy. None of
my informants (neither [1991] in Helsinki nor [1994] in the USA) produced
examples that I would like to interpret and classify as METAPHORICAL OBTAINING
FOR SOMEONE ELSE; yet in dictionaries there are some possible candidates, such as
He got her a job with the telephone company (Collins Cobuild English Language
Dictionary [1987]: Reading 8.4), whose metaphoricity, however, is not very strong.
9. I have to explain an additional feature of informant No. 19's response, because
it functions as an explicit proof that American high school students seem to have
 Experiments with English get and Finnish pitaa 191

a clue of what they are doing in my production test. Namely, she not only
produced examples, but dictionary-type definitions. She explicated her first
example (9a) as To get is to receive something and her example (9b) as To get
something is to go buy something or accomplish something.
10. The average ratings ranged from 0.33 to 3.13, so that 2.13 is a relatively high
difference rating. The standard deviation in the ratings for this pair was 1.35.
The standard deviations ranged from 0.94 to 1.46.
11. Two orthographicaIly identical cases of, e.g., I'm glad J got to know you can
mean either ABILITY or REACHING A STAGE depending on the intonation: ABILITY has
stress on got, REACHING A STAGE on know. (I thank Diana ben-Aaron for pointing
this out.) Needless to say, my method fails to account for prosody.
12. As in the case of the get production test, most pitiiii production test infonnants
provided me with paraphrases of these examples (e.g. where they had to replace
pitiiii with another word) so that I did not have to rely only on my native speaker
intuition in order to interpret the meanings of pitiiii from the sentential context,
but to save space I will not show these paraphrases and their glosses.
13. Explanations of the abbreviations for the suffixes referred to in examples:
NOM nominative INF infinitive
GEN genitive IMPER imperative
PART partitive PAST past tense
ACC accusative SG singular
ESS essive PL plural
INESS inessive first person
ELAT elative 2 second person
ILLAT iIlative 3 third person
ALLAT aIlative

References

Bn!al, Michel
1868 Les idees latentes du langage. Le~on jaite au College de France
pour la reouverture du cours de grammaire comparee,le 7 dec. 1868.
Paris: Hachette.
Caramazza, Alfonso and Ell en Grober
1976 Polysemy and the structure of the subjective lexicon. In: Clea Rameh
(ed.), Semantics: Theory and Application. Georgetown University
Round Table on Languages and Linguistics 1976, 181-206.
Washington, D.C.: Georgetown University Press.
Colombo, Lucia and Giovanni B. Flores d' Arcais
1984 The meaning of Dutch prepositions: a psycholinguistic study of poly-
semy. Linguistics 22: 51-98.
192 larno Raukko

Deane, Paul
1988 Polysemy and cognition. Lingua 75: 325-361.
Erdmann, Karl OUo
1910 Die Bedeutung des Wortes. Aufsatze aus dem Grenzgebiet der Sprach-
psychologie und Logik. 2. Aufl. Leipzig: Avenarius.
Geeraerts, Dirk
1993 Vagueness's puzzles, polysemy's vagaries. Cognitive Linguistics 4:
223-272.
Gibbs, Raymond w., Jr., Dinara A. Beitel, Michael Harrington and Paul E. Sanders
1994 Taking a stand on the meanings of stand: bodily ex.perience as moti-
vation for polysemy. lournal of Semantics 11: 231-251.
Givon, Talmy and Lee Yang
1994 The rise of the English GET-passive. In: Barbara Fox. & Paul J.
Hopper (eds.), Voice: Form and Function, 119-149. Amsterdam and
Philadelphia: John Benjamins.
Kimball, John
1973 Get. In: John Kimball (ed.), Syntax and Semantics 2,205-215. New
York: Seminar Press.
Langacker, Ronald
1988a A usage-based model. In: Rudzka-Ostyn (1988), 127-163.
1988b A view of linguistic semantics. In: Rudzka-Ostyn (1988), 49-90.
Lehrer, Adrienne
1974 Homonymy and polysemy: measuring similarity of meaning. Language
Sciences 32: 33-39.
Lindstromberg, Seth
1991 Get: not many meanings. International Review ofApplied Linguistics
in Language Teaching 29: 285-302.
Nerlich, Brigitte
1992 Semantic Theories in Europe 1830-1930: From Etymology to Con-
textuality. Amsterdam and Philadelphia: John Benjamins.
Niedzielski, Henry
1976 Semantic considerations of get and some of its Polish equivalents.
Papers and Studies in Contrastive Linguistics 5: 219-238.
Nyrop, Kristoffer
1913 Semantique. Vo\. IV of his Grammaire historique de la langue
franraise. Copenhagen: Gyldendalske Boghandel Nordisk Forlag.
6stman, Jan-Ola
1988 Adaptation, variability, and effect: comments on IPrA Working Docu-
ments I and 2. IPrA Working Document 3: 5-40. University of
Antwerp: International Pragmatics Association.
2000 Merkityksen ajankohtaiset haasteet yleiselle kielitieteelle [The current
challenges of meaning for general linguistics]. In: Anu Airola, Heikki
 Experiments with English get and Finnish piiliii 193

1. Koskinen and Veera Mustonen (eds.), Merkillinen merkitys [The

mystery of meaning], 69-84. Helsinki: Gaudeamus.
Raukko, Jarno
1993 Get examined. A cognitive semantic study of the polysemy of get as
a dynamic network of meanings. Unpublished M.A. thesis, Department
of English, University of Helsinki.
1994 Polysemia: merkitysten verkosto - merkityksen verkko [Polysemy:
the network of meanings - the net of meaning]. In: Pentti Leino and
Tiina Onikki (eds.), Kieli 8. Suomen kielen kognitiivista kielioppia
2. Niikokulmia polysemiaan [Cognitive grammar of Finnish 2: per-
specti ves to polysemy], 36-69. Helsinki: The Department of Finnish
at the University of Helsinki.
1995 What do we get for get in Finnish? The New Courant 3: 170-180.
[University of Helsinki, Department of English.]
1996 "No more polysemy", says the nationalist language police. The para-
doxical battle between semantic flexibility and normativism.
Pragmatics, Ideology, and Contacts Bulletin 3: 36-44. [University
of Helsinki, Department of English.]
1997 The status of polysemy in linguistics: from discrete meanings to de-
fault flexibility. In: SKY 1997: The 1997 Yearbook of the Linguistic
Association of Finland, 145-170.
1999 An "intersubjective" method for cognitive-semantic research on
polysemy: the case of get. In: Masako K. Hiraga, Chris Sinha and
Sherman Wi\cox (eds.), Cultural, Typological, and Psycho linguistic
Issues in Cognitive Linguistics. Selected papers of the bi-annual ICLA
meeting in Albuquerque, July 1995, 87-lO5. [Current Issues in Ling-
uistic Theory 152.] Amsterdam and Philadelphia: John Benjamins.
Rudzka-Ostyn, Brygida (ed.)
1988 Topics in Cognitive Linguistics. Amsterdam and Philadelphia: John
Benjamins.
Sandra, Dominiek and Sally Rice
1995 Network analyses of prepositional meaning: mirroring whose mind -
the linguist's or the language user's? Cognitive Linguistics 6: 89-130.
Tuggy, David
1993 Ambiguity, polysemy, and vagueness. Cognitive Linguistics 4:
273-290.
Whitney, WiIliam Dwight
1867 Language and the Study of Language: Twelve Lectures on the
Principles of Linguistic Science. New York: Scribner, Armstrong.
Zhang, Qiao
1998 Fuzziness - vagueness - generality - ambiguity. Journal of Pragmatics
29: 13-31.
Metonymic polysemy and its place in meaning
extension

Ken-ichi Seto

1. Introduction

The last quarter of a century has seen a rising concern with metonymy. The
literature, from Nunberg's (1978) pioneering exploration to Radden and
Koevecses' (1999) penetrating investigation, by way of Lakoff and Johnson's
(1980) influential work, has shown that metonymy is not only as pervasive
as metaphor but also no less important in the daily use of language. In spite
of the continuing expansion of the metonymic realm as is shown in Panther
and Radden (1999) and Barcelona (2000), however, the core notion of
metonymy is, so far as I can see, not yet established. I shall argue specifically
(i) that there has been no satisfactory definition of metonymy yet, owing
to confusions about the difference between entities and categories; (ii) that
the ultimate reason why those confusions so often occur resides in the
(inevitable) spatial representation of categorical relations; and (iii) that the
network model (the prototype-extension-schema triangle), which is supposed
to deal with polysemy, does not work because metonymy has no proper place
in the model. After these arguments I shall offer a new way of looking at
polysemy: a cognitive triangle whose vertices are metaphor, metonymy, and
synecdoche.

2. Why metonymy is still a vague notion

In classical rhetoric metonymy and synecdoche have been defined, respec-

tively, as: "the substitution of a word denoting an attribute or adjunct of a
thing for the word denoting the thing itself' and "a figure of speech in which
a more inclusive term is used for a less inclusive one or vice versa, as a
whole for a part or a part for a whole" (The New Shorter Oxford English
Dictionary). However, since 1956, when Jakobson wrote a classic paper on
metaphor and metonymy, the concept of contiguity has often been used to
196 Ken-ichi Seta

define metonymy. But what does "contiguity" involve? So far as I can see,
this is the most important question to ask when defining metonymy. Does it
mean the contiguity of entities alone? Or does it also mean the contiguity of
categories? If the inclusive definition were taken, what would the contiguity
of categories mean? In other words, is metonymy concerned only with a
referential transfer between entities, or also with a nonreferential transfer
between a more comprehensive category and a less comprehensive category?
On the other hand, if one chose not to use the tenn "contiguity", at least in
some cases, then what should be the defining characteristic of metonymy?
To answer these questions, I shall first offer a new definition of metonymy
which departs in some crucial respects from traditional views.

2.1. The definition of metonymy

A new definition of metonymy can be given as follows:

(l) Metonymy is a referential transfer phenomenon based on spatio-

temporal contiguity as conceived by the speaker between an entity
and another in the (real) world.

This is coupled with a new definition of synecdoche:

(2) Synecdoche is a categorical transfer phenomenon based on semantic

inclusion as conceived by the speaker between a more comprehensive
and a less comprehensive category.

These two definitions are a declaration of independence for metonymy

and synecdoche from the centuries-old, rhetorico-cognitive Western tradition.
The major differences are:

(i) While the traditional notion of metonymy has kept synecdoche as its
servant, under the new definitions (1) and (2), synecdoche is indep-
endent of metonymy.
(ii) Whereas traditionally metonymy is a mixed category, in that it covers
entities and categories alike, the definition of metonymy in (1) makes
it clear that metonymy is a coherent category that only comprises
entity-based referential transfers and excludes categorical transfers.
 Metonymic polysemy and its place in meaning extension 197

(iii) Synecdoche now comprises only category-based nonreferential trans-

fers, leaving entity-based whole-part transfers to metonymy. As a
result, synecdoche also becomes a coherent category.
(iv) Both definitions, (1) and (2), avoid theoretical confusions caused by
the ambiguity (see below) of "whole" and "part", which permits both
referential and categorical interpretations.
(v) The new demarcation between metonymy and synecdoche makes it
possible to allocate "partonomy" (the "part-of' relation) to metonymy
along with other entity-based relations, and "taxonomy" (the "kind-
of' relation) exclusively to synecdoche. l

This division of work is not just notational but also substantial, because it
is supported on lexico-semantic grounds. Metonymy is concerned with ref-
erence; synecdoche is associated with (extensional and intensional) sense;
"partonomy" denotes a referential relation which holds, for example, between
a windmi\l and its sails (the sails are a part of a windmill); and "taxonomy"
(hyponymy) signifies a categorical relation which holds, for example, between
a ticket and a traffic ticket (a traffic ticket is a kind of ticket). Thus under the
new definitions (3a) is a metonymy, and (3b) is a synecdoche (in a parking
situation).

(3) a. The windmill is turning.

b. I got another ticket.

This distinction is so clear that it may seem hardly necessary to emphasize

the point because it is not a semantic practice to treat partonomy and taxon-
omy under the same heading. However, the realm of metonymy appears to
be a different world where partonomy and taxonomy have sat side by side
for centuries. Even Radden and Koevecses (1999), who provide the most
penetrating analysis of metonymy, do not raise questions about this matter.
For them, metonymy is sometimes referential and sometimes categorical.
Neither Lakoff and 10hnson (1980) nor Lakoff and 10hnson (1999) show
interest in the problem, although one of their great concerns is to define
what a category actually is. What makes them insensitive to this matter?
198 Ken-ichi Seta

2.2. The return of Lakoff and lohnson

It is a curious fact that while partonomy and taxonomy are distinguished in

lexical semantics, in research into metonymy they are lumped together. In
this respect, Lakoff and lohnson (1999) is no improvement on Lakoff and
lohnson (1980). For them metonymy is as partonomical as it is taxonomical.
What, then, is the essence of metonymy? Is it a coherent category? Apparently
so. Because taxonomy often looks like partonomy. Take the following two
figures (Fig. I and Fig. 2).

B C

D
~ E

Figure 1.

Figure 2.

Which figure represents partonomy and taxonomy better? Actually, this

is a pseudo-question because there is no substantial difference in represen-
tability between the two figures. One figure can represent partonomy or
taxonomy just as well as the other. For example, Figure 3 is a rough sketch
of the taxonomy of things in the world (see Ch om sky 1965: 83).
However, Figure 1 can also represent the partonomy of a tree (Fig. 4).
 Metonymic polysemy and its place in meaning extension 199

thing

living thing non-living thing

human animal

Figure 3.

a tree

branches a trunk

twigs leaves

Figure 4.

Hence, Figure I is called a "tree" diagram. Note that once the taxonomy of
things is represented by Figure I, it would become "a tree". And a tree as an
entity, not tree as a category, is composed of different parts such as a trunk,
branches, roots, and so on. Likewise, each node below the top (thing) in
Figure 3 would be interpreted as one of these parts. Therefore, taxonomy
would become partonomy! But, of course, this is fallacious reasoning. Where
are we going wrong? A pitfall is contained in the representation of Figure 3
itself. The moment the framework of Figure 1 is adopted to represent a
taxonomical relation, it automatically forces us to reason along the above
line and to conclude that there is no substantial difference between taxonomy
and partonomy because each node is a part of a whole tree. Confusion of
this kind, which may be called the partonomy-taxonomy fallacy (or the PT
fallacy), is pervasive and it is caused by a specific spatial metaphor such as
a tree diagram, as in Figure 1.
200 Ken-ichi Seto

Perhaps this is one of the reasons why it appears that categories are better
represented by rings within a circle (Fig. 2). Consider Figure 5, but note
again that although human beings are a kind of (living) thing, the small ellipse
of human beings could also be said to be a part of the larger ellipse of (living)
things in so far as the represented forms (figures) are concerned. Here taxon-
omy once again would look like partonomy. Or, one might take a slightly
different line of reasoning: in Figures 2 and 5, rings and smaller ellipses are
rather contained in a circle and a large ellipse, respectively, so in I got another
ticket, ticket (container) stands for "traffic ticket", contained, just as in The
kettle is boiling, kettle (container) stands for "the water in the kettle", contained.
Container-for-contained is a regular pattern of metonymy. Consequently, one
might conclude that (3b) is an instance of metonymy. But this reasoning,
too, is fallacious, because whereas "traffic ticket" is a kind of ticket, "the
water in the kettle" is not a kind of kettle.

Figure 5.

Lakoff and Johnson (1999: 51) apparently succumb to the force of the PT
fallacy (see also Lakoff 1987, Chapter 5). They point out that there is a
conceptual metaphor A CATEGORY IS A CONTAINER. The primary experiential
basis of the metaphor is, according to them, that "things that go together
tend to be in the same bounded region (correlation between common location
and common properties, functions, or origins)." I do not deny that there is
such a conceptual metaphor because in fact there is one, not only in English
but also in Japanese and perhaps many other (possibly all) languages. Thus
we ask, for example, "Are tomatoes in the fruit or the vegetable category?"
However, it is this conceptual metaphor that is the ultimate cause of the PT
fallacy because it blurs the distinction between folk understanding and expert
knowledge. It is necessary in the analysis of metonymy (as a piece of expert
 Metonymic polysemy and its place in meaning extension 201

knowledge) as well as in lexical semantics to tell "things that go together"

referentially from "things that go together" categorically.2
This blurred distinction between folk understanding and expert knowledge
may also occur in other fields. For example, Reddy's classic paper on meta-
phor and thought (Reddy 1979) points out that the folk understanding of
communication is dominated by the conduit metaphor so that, when talking
about communication, we can hardly avoid using the metaphor. For instance,
get the idea across, put each concept into words very carefully, and extract
coherent ideas from that prose, are all expressions based on this metaphor.
Reddy's (1979) most valuable finding, however, is not that the entire meta-
lingual apparatus of the English language is largely (at least 70%) determined
by the conduit metaphor, though this is a major discovery, but that this
conceptual metaphor distorts the true picture of what is happening in
communication. For the model of communication based on the metaphor
"objectifies meaning in a misleading and dehumanizing fashion" (Reddy
1979: 308); we "have the mistaken, conduit-metaphor influenced view that
the more signals we can create, and the more signals we can preserve, the
more ideas we 'transfer' and 'store' ," (p. 310); and we "neglect the crucial
human ability to reconstruct thought patterns on the basis of signals and this
ability founders" (p. 310). The conduit metaphor, which belongs to folk
understanding, does not adequately capture the reality of communication.
Metaphors thus necessarily lead and mislead us. The conduit metaphor
has contributed to establishing a distorted picture of how communication
actually works. Whereas the picture given by the conduit metaphor is not
wrong by one standard (folk understanding) because it is a picture we live
by, it is wrong by another standard (expert knowledge) because it fails as a
model of communication. This line of argument also applies to the conceptual
metaphor A CATEGORY IS A CONTAINER. It is a metaphor we live by, a metaphor
deeply entrenched in our minds, but it should not be so as part of the theory
of metonymy, because it has a strong numbing effect and obscures the
difference between partonomy and taxonomy. We should be careful to keep
from putting both partonomy and taxonomy under the same heading of
metonymy because, as eruse (2000: 47) correctly distinguishes, "[s]eeing
something meronymically [i.e. partonomically], i.e. as a whole consisting
of parts (cf. Pustejovsky's 'constitutive role')" is one thing, and "[s]eeing
something taxonomically, i.e. as a kind, in contrast with other kinds (cf.
Pustejovsky's 'formal role')" is quite another. 3
One might, however, argue that the definition of synecdoche previously
202 Ken-ichi Seta

raised in (2) also makes use of the metaphor A CATEGORY IS A CONTAINER. This
is true, for "inclusion" and "comprehensive" are based on the A CATEGORY IS
A CONTAINER metaphor. The two words might therefore better have been
omitted in the definition. Compare a revised definition (4) with (2):

(4) Synecdoche is a categorical transfer phenomenon based on hyponymy

as conceived by the speaker between a genus and a species.

However. this improvement is only apparent because it has just hidden

"inclusion" and "comprehensive" behind technical terms such as "hyponymy",
"genus" and "species". Ask what a genus is, and we would have to answer
that it is a more "comprehensive" or "inclusive" category, or we would come
up with another technical term such as a "higher-order" category. The reason
why a real improvement is next to impossible is simply that there seem to
be no metalinguistic terms precise enough to describe the true nature of
categories, no terms that are not (even potentially) contaminated with the A
CATEGORY IS A CONTAINER metaphor or other closely related spatial metaphors.
Four points may be worth noting.

(i) Categories are in the mind (i.e. represented in the brain) but are very
hard to access by way of (meta)language. perhaps because the nature
of categories resists verbalization. Categorization (e.g. edible vs. non-
edible) is an essential function of living things from human beings
down to amoebas. There is no doubt that amoebas can categorize for
this feature, edible vs. non-edible, but from this no-one would infer
that they have the A CATEGORY IS A CONTAINER metaphor. Categorization
itself is possible without language. 4
(ii) The force of the A CATEGORY IS A CONTAINER metaphor is compelling.
More generally, the pressure of spatialization metaphors is so high
that the moment one tries to talk about categories, the A CATEGORY IS A
CONTAINER metaphor or any other closely related spatial metaphor is
ready on the tip of the tongue. s
(iii) Perhaps a rare linguistic tool that is appropriate for describing the
true nature of categories, and is relatively free from the bias of folk
understanding, is "kind of'. "Kind of' can be used to distinguish
between taxonomy and partonomy (e.g. a traffic ticket is a kind of
ticket, not a part of a ticket; a leg is a part of a table, not a kind of
table). Thus (4) may again be paraphrased: synecdoche is a categorical
 Metonymic polysemy and its place in meaning extension 203

transfer phenomenon based on a "kind of' relation as conceived by

the speaker between a genus and a species.
(iv) It is only a step from the PT fallacy to the entity-category fallacy (or
the EC fallacy, i.e. to interpret categories in terms of entities) because
partonomy is just a kind (not a part) of entity-based contiguous relation.
In short, metonymy is an E(ntity)-related transfer and synecdoche is
a C(ategory)-related transfer. E-relation and C-relation are the bases
of two different routes of thought.

2.3. Langacker and the network model

Langacker's concern with metonymy started with "active zone" (Langacker,

1984), a usage-level metonymic fluctuation phenomenon, rather than lexical
metonymy based on regular semantic shifts. This is why when Langacker
(1990) presented the network model, a model supposed to capture meaning
extension from a prototypical sense, there was no place for metonymy.
Figure 6 is a slightly modified version. ti

prototype extension metaphor

(similarity)

Figure 6.

Figure 6 shows that the extension from the prototype is exclusively

metaphoric and that the schema C is the extraction of the commonality of A
and B. Since a schema corresponds to a genus, a prototype to a species, and
a metaphor to another species (in a domain different from the domain to
which the prototype belongs), the schematic relation (i.e. the species-to-genus
relation) and the instantiating relation (i.e. the genus-to-species relation) mean
what we call the C-relation. If we interpret the schema as another extension,
it corresponds to synecdoche (the species-to-genus type) in our sense. Now
204 Ken-ichi Seta

it is clear that whereas metaphor and synecdoche hold their places in the
network model, there is no place for metonymy.
Langacker started to change his position later, gradually and tantalizingly,
until he finally decided to give metonymy a place in the network model
when he referred to "extension (generally metaphorical and metonymic)"
(Langacker 1995b: 111).7 A new version of the network model may be shown
in Figure 7.

schema

"he~~~=atk
A --------------~~~ B
extension metaphor
prototype
metonymy

Figure 7.

However, the inclusion of metonymy in the network model poses a serious

theoretical problem: B, the extension from the prototype A, is not a uniform
node because the extension processes of metaphor and metonymy are differ-
ent, so the status of schema has to change accordingly. While node C continues
to be able to maintain a schematic relation with the prototype and a metaphor,
such a relation is no longer possible with the prototype and a metonymy,
because the prototype is related to its metonymic extension on the basis of
contiguity (the E-relation) in the world. There is no similarity to be extracted
from the two. Therefore, so far as metonymy is concerned, the network model
stops working.
For example, see the notorious example of metonymy in (5) (Nunberg,
1978: 22):

(5) The ham sandwich is getting restless at table 20.

"The ham sandwich" refers to "the customer who ordered it" in a

restaurant situation. It is obvious that there is no sense relation (lexical or
non lexical) between "The ham sandwich" and "the customer"; they are only
referentially connected by sharing the same space-time. It is therefore not
 Metonymic polysemy and its place in meaning extension 205

possible to extract a schema (i.e. abstract semantic commonality) from "The

ham sandwich" and "the customer". The same applies to the next example,
a case of "active zone" or metonymic contextual fluctuation.

(6) She slipped her hand through his arm.

In a normal, non-bloody situation, "his arm" metonymically refers to "the

space between his arm and the side of the body". Again, there are no (mean-
ingful) common semantic features to be extracted from the two referents in
question. And although the metonymies in (5) and (6) are contextual in nature,
things do not change with lexicalized metonymies:

(7) He is always chasing skirts.

"Skirts" are not similar to "the girls who wear them", but only contiguous
with them. Of course, ultimately any two things can be said to be similar in
one respect or another, but that is not the point here.
One merit of the network model in Figure 7 is that three major extension
patterns, metaphor, metonymy, and synecdoche, are all present, provided that
synecdoche is substituted for schema. Metaphor is based on the similarity
relation (the S-relation), metonymy on the E-relation, and synecdoche on the
C-relation. They make up the three vertices of what is called the cognitive
triangle.

3. The cognitive triangle and its implications

A theory of polysemy should attach equal weight to metaphor, metonymy,

and synecdoche. These are the major routes of meaning extension both
synchronically and diachronically.8 They also influence and partly stipulate
the ways we think and act.

3.1. The cognitive triangle

Figure 8 shows a simplified model of how a prototype extends its meaning.

P, the prototype, extends its meaning in three major directions: metaphor
(the S-relation), metonymy (the E-relation), and synecdoche (the C-relation),
206 Ken-ichi Seta

t
~~
t/p~t
C E
~ecd~~OnY~
Figure 8.

which together make up the cognitive triangle. And each vertex extends a
newly acquired meaning again in the three directions. Metaphor is similarity-
based, metonymy is entity-based, and synecdoche is category-based. The
extension to C is of two kinds: one is the species-to-genus type as in (8a)
and the other is the genus-to-species type as in (8b).

(8) a. Will you ship the goods by rail?

b. I have a temperature.

"Ship" in (8a) extends its sense from "send something by ship" to "send
something by any means of transportation". ''Temperature'' in (8b) extends
its sense from "the temperature of the body" to "fever". In the latter example,
"extends" is used in an extended (Le. species-ta-genus, or schematic) sense.
Metonymy is of three types depending on the kinds of entities: spatial (e.g.
book), temporal (e.g. earthquake), and abstract (e.g. beauty) (for details and
examples, see Seto 1999).9
Another important point to make about the cognitive triangle is that
metonymy and synecdoche belong to different cognitive domains. to See
Figure 9.
Metonymy works in the E-domain, which is associated, typically, with
the real world where entities are arranged concentrically. We tend to see two
entities as closely connected if they are in a closer position than others. The
 Metonymic polysemy and its place in meaning extension 207

metaphor

Ithe C-domainl Ithe E-domain I

synecdoche metonymy

Figure 9.

whole-part relation in the sense of partonomy may be the closest relation of

all the relations based on contiguity in the world. On the other hand, the C-
domain is in the mind, where a lot of categories, some stable and some
unstable, are arranged so that they may reflect how we group the things in
the world. Synecdoche works in the C-domain. And metaphor, the third
category, it may be said, does not have its own proper domain which might
be called the S-domain. Rather, it straddles the E-domain and the C-domain,
sometimes serving as a bridge to connect the two different domains, some-
times connecting two faraway entities or -two distant categories in a single
domain. I shall have to leave the details to another paper.

3.2. The classical quartet vs. the modern trio

The cognitive triangle also explains semantic changes along the time axis.
When speaking of the classification of diachronic semantic changes, Geeraerts
(1994) refers to "the classical quartet of specialization, generalization,
metonymy, and metaphor." Since specialization and generalization largely
correspond to the two aspects of synecdoche used in our sense, that is, the
genus-to-species transfer and the species-to-genus transfer respectively, the
classical quartet can be collapsed into the modern trio of synecdoche
(specialization and generalization), metonymy and metaphor. The cognitive
triangle works both for diachronic semantic development and for synchronic
semantic extension.
208 Ken-ichi Seto

This tripartite theory of polysemic sense extension is obviously an

improvement on Ullmann (1962), a classic of semantics, because Ullmann
regards similarity and contiguity (thus metaphor and metonymy) as the two
basic axes of semantic change. Moreover, his definition of metonymy based
on contiguity is too broad and vague, and he seems to have no clear vision
separating synecdoche from metonymy. For him, metonymy is "founded on
some relation other than similarity" (Ullmann 1962: 78), "based not on
similarity but on some other relation between two terms" (p. 163), and "based
on relations other than similarity" (Ullmann 1964: 83). On other occasions,
however, Ullmann is more specific: contiguity is a relation "between two
references" (Ullmann 1951: 240). Thus, "Some metonymic transfers are based
on spatial relations", while "[a]nother group of metonymies are based on
temporal relations" (Ullmann 1962: 218-219). But there is a deep flaw in
UlImann's idea of contiguity, since in his writing contiguity sometimes means
"sense-contiguity" (Ullmann 1951: 222, 243): "contiguity between the senses"
(p. 231). He, too, commits the EC fallacy.
Thus as he discusses pejorative and ameliorative developments of senses,
Ullmann refers to "fortune", a semantically neutral word, which comes to
have an exclusively positive value "when used metonymically in the sense
of 'wealth' " (Ullmann 1962: 235). This is a sense-relation, a special type of
semantic specification, or narrowing, special in the sense that the change in
evaluation is definitely ameliorative. This is a case of synecdoche in our
sense, a categorical transfer, where the neutral "fortune" (genus) stands for
the narrower "fortune (wealth)" (species). It is clear that, in the world, the
neutral "fortune" is not contiguous with the narrower "fortune (wealth)";
they are just categorically related. Unfortunately, many later semantic theorists
and others share the EC fallacy or the PT fallacy with Ullmann in the
discussion of metonymy and categories in general. But some are more
cautious. Tversky (1990), for example, makes a clear distinction between
partonomy and taxonomy, and so do Lyons (1977), Cruse (1986), and Nerlich
(to appear). The cognitive triangle, standing on secure semantic ground,
promises a better way of looking not only at synchronic polysemy but at
diachronic semantic change as well, as it is able to explain patterns of meaning
extension and meaning change better than other models do.
 Metonymic polysemy and its place in meaning extension 209

4. Conclusion

The main points can be summarized as follows:

(i) Metonymy is an E-transfer, i.e. a referential transfer based on the

contiguity between an entity and another in the world, as conceived
by the speaker.
(ii) Synecdoche is a C-transfer, i.e. a categorical transfer based on a "kind
of" relation between a genus and a species, as conceived by the
speaker.
(iii) Metaphor is an S-transfer, i.e. a structural transfer based on the simi-
larity between an entity or category and another entity or category, as
conceived by the speaker.
(iv) The PT fallacy is to confuse partonomy (the entity-based "part of"
relation) and taxonomy (the category-based "kind of" relation) and to
interpret taxonomy in terms of partonomy.
(v) The EC fallacy is to confuse entities and categories and to interpret
categories in terms of entities. The PT fallacy is a kind of EC fallacy.
(vi) Folk understanding and expert knowledge should be distinguished in
scientific discussion.
(vii) Categorization is possible without language.
(viii) Langacker's network model is flawed because metonymy has no proper
place in it.
(ix) The cognitive triangle can be a model to describe not only synchronic
polysemy but also diachronic semantic change.
(x) In the cognitive triangle, metonymy belongs to the E-domain, synec-
doche to the C-domain, and metaphor straddles both domains.

Notes

I. For the distinction between partonomy and taxonomy in relation to metonymy

see Seto (1999); see also Cruse (1986) and Tversky (1990). As to the classification
of whole-part relations, see Tversky and Hemenway (1984), Cruse (1979) and
Winston, Chaffin, and Herrmann (1987). Note that partonomy is just one kind of
referential relation which metonymy may exploit; there are many others, such as
container-for-contained, cause-for-result, and so on (Seto 1999).
2. Lakoff (1987: 79) says: "[A] part (a subcategory or member or submodel) stands
for the whole category - in reasoning, recognition, etc. Within the theory of
210 Ken-ichi Seto

cognitive models, such cases are represented by metonymic models." For him,
things that go together referentially and things that go together categorically can
both be the basis of metonymy. Thus Lakoff (1987: 90) claims: "In short, a
cognitive model may function to allow a salient example to stand metonymically
for a whole category". Radden and Koevecses (1999), following this statement,
refer to specific "category" metonymies such as STEREOTYPICAL MEMBER FOR A
CATEGORY, CATEGORY FOR THE STEREOTYPICAL MEMBER, IDEAL MEMBER FOR THE
CATEGORY, and CATEGORY FOR THE IDEAL MEMBER. All these fit the definition of
synecdoche given in (2).
3. Cruse (2000: 50) notes further that "Pustejovsky does not really explain why he
opts for four qualia roles ... Croft (p.c.) sees no justification for four roles, or
any definite number; if there were any super-domains, he would opt for just
two, the taxonomic and the meronymic, the latter subsuming the constitutive,
telic and agentive roles." While I agree with Cruse (and Croft) in dividing
Pustejovsky's four qualia roles into two "super-domains" (the formal qualia and
the other three), it seems hardly justified to squeeze the constitutive, telic and
agentive roles into one and the same category of meronymy, at least under Crose's
definition of the term. Rather, these three roles may naturally be put into the
category of spatio-temporal contiguity which essentially characterizes metonymy,
with the constitutive qualia related to spatial contiguity, and the agentive and
telic qualias to temporal (i.e. causal) contiguity (see Seto 1999 for further dis-
cussion and examples). Accordingly, Pustejovsky's formal qualia role may be
associated with synecdoche in my sense, and the other three with metonymy.
Although the latter three qualia roles do not exhaust the metonymic resources,
the formal qualia does exhaust the synecdochic relations (genus-for-species, and
species-for-genus).
4. Since Rosch (1973, 1978), the term "prototype" has become very popular. In
fact there may be little doubt of the psychological reality of the prototype structure
of categories. But how are categories prototypically structured? Taylor (1995)
points out that "[t]here are two ways in which to understand the term 'prototype'.
We can apply the term to the central member, or perhaps to the cluster of central
members, of a category.... Alternatively, the prototype can be understood as a
schematic representation of the conceptual core of a category" (emphases added).
Note that the prototypical understanding of the term "prototype" is based on the
centre-periphery metaphor. This may have partly contributed to the genesis of
the A CATEGORY IS A CONTAINER metaphor even on the expert knowledge level.
5. Barcelona (personal communication) points out that another undesirable conse-
quence of the A CATEGORY IS A CONTAINER metaphor is that it may mislead one into
a belief that categories are discrete, since containers such as bowls and cups
have discrete contours.
6. Langacker's (1990: 271) original figure is shown in Figure 10. In this, "[t]he
process of extension occurs because a speaker perceives some similarity between
the basic value (i.e. the local or global prototype) and the extended value. This
 Metonymic polysemy and its place in meaning extension 211

SCHEMA

Figure 10.

similarity perception represents the commonality of the basic and extended values,
so it constitutes a schema having the two for instantiations" (p. 271, emphases
added). From this it is clear that Langacker's extension in the network model is
exclusively metaphoric in nature (at the time the paper was written) because it is
only in metaphoric extension that some schematic commonality can be extracted
from a prototype and its extension, such as a real mouse and a computer mouse;
on the other hand, it is virtually impossible, unless by force, to draw similarity
perception from one sense - whether it is a prototype or not - and its metonymic
extension, because a metonymic extension is associated with a prototype by way
of contiguity, not similarity. It would be absurd, for example, to see some
meaningful similarity between "long hair" per se and "a person with long hair"
or between "a kettle" and "water in the kettle".
7. Langacker's (1984) "active zone" clearly constitutes part of metonymy, but there
is no mention of metonymy there. The relationship between "active zone" and
metonymy seems to have grown slowly. The first reference to metonymy is
made in Langacker (1991: 456 footnote). Then Langacker (1993: 31) observes:
"Metonymy largely overlaps with what I have called active zone phenomenon."
The relationship between the two then develops rapidly into the total inclusion
of "active zone" in the metonymy category in Langacker (1995a: 27): "Specifically,
they [active-zone/profile discrepancies] represent a special case of METONYMY".
8. When dictionaries are compiled on the principle of clarifying the semantic
networks of lexical entries, instead of describing their historical sense develop-
ments or current sense frequencies, they should seriously take into account the
distinction between metaphor, metonymy, and synecdoche. In this regard, The
New Oxford Dictionary of English is worth mentioning because it identifies three
major extension patterns: "figurative extension of the core sense", "specialized
case of the core sense", and "other extension or shift in meaning, retaining one
or more elements of the core sense". It is obvious that "figurative extension" is
metaphoric and that "specialized case" corresponds to one of the two kinds of
synecdoche (unfortunately, no mention is made of generalization). However, the
third category is plainly miscellaneous, although, judging from the examples
cited there (p. x), there is no question that metonymy (along with some others)
comes into this third category.
9. An earthquake is a temporal entity, so we can say, for example, that "there was"
212 Ken-ichi Seto

an earthquake yesterday. An earthquake acquires the status of entity through an

ontological metaphor, a kind of spatial metaphor. So does an abstract entity. For
instance, when "beauty" is used in the sense of a beautiful woman, it is supposed
that "beauty" is a salient property of that woman. The woman has this property,
as it were, as a part of her body. To take another example, suppose that there is
a woman at the other end of the phone whose voice is soft and whose name you
don't know. You might refer to her in your mind as "Miss Softly". This is a
metonymy. And "softly" here becomes an abstract entity to characterize the softly-
voiced woman. Here again the property, as an abstract entity, is referentially
related to the woman.
10. For different notions of the term "domain", see, among others, Lakoff (1987),
Croft (1993) and Barcelona (2002).

References

Barcelona, Antonio
2000 (ed.) Metaphor and Metonymy at the Crossroads: A Cognitive Per-
spective. Berlin: Mouton de Gruyter.
2002 Clarifying and applying the notions of metaphor and metonymy
within cognitive linguistics: an update. In: Rene Dirven and RaJph
Poerings (eds.), Metaphor and Metonymy in Comparison and Con-
trast. Berlin: Mouton de Gruyter.
Browm, Lesley (ed.)
1993 The New Shorter Oxford English Dictionary. New York: Oxford
University Press.
Chomsky, Noam
1965 Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
Croft, William
1993 The role of domains in the interpretation of metaphors and meton-
ymies. Cognitive Linguistics 4: 335-370.
Cruse, David A.
1979 On the transitivity of the part-whole relation. Journal of Linguistics
1: 29-38.
1986 Lexical Semantics. Cambridge: Cambridge University Press.
2000 Aspects of the micro-structure of word meanings. In: Yael Ravin
and Claudia Leacock (eds.), Polysemy: Theoretical and Compu-
tational approaches, 30-51. New York: Oxford University Press.
Geeraerts, Dirk
1994 Historical semantics. In: Ron E. Asher (ed.), The Encyclopedia of
Language and Linguistics, 1567-1570. Oxford: Pergamon Press.
 Metonymic polysemy and its place in meaning extension 213
Jakobson, Roman
1971 [1956] Two aspects of language and two types of aphasic disturbances. In:
Roman Jakobson (ed.), Roman lakobson Selected Writings Volume
2: Word and lAnguage, 239-259. The Hague: Mouton.
Lakoff, George
1987 Women, Fire, and Dangerous Things: What Categories Reveal About
the Mind. Chicago, IL: The University of Chicago Press.
Lakoff, George and Mark Johnson
1980 Metaphors We Live By. Chicago, IL: The University of Chicago Press.
1999 Philosophy in the Flesh: The Embodied Mind and its Challenge to
Western Thought. New York: Basic Books.
Langacker, Ronald W.
1984 [1990] Active zone. In: Ronald W. Langacker, Concept, Image, and Symbol:
The Cognitive Basis of Grammar, 189-201. Berlin: Moulon de Gruyter.
1990 A usage-based model. In: Ronald W. Langacker, Grammar and
Conceptualization. Berlin: Mouton de Gruyter.
1991 Foundations of Cognitive Grammar. Volume 11: Descriptive Applic-
ation. Stanford, CA: Stanford University Press.
1993 Reference-point constructions. Cognitive Linguistics 4: 1-38.
1995a Raising and transparency. lAnguage 71: 1--62.
1995b Cognitive grammar. In: Jan-Ola OSlman, Jef Verschueren, Jan
Blommaert and Chris Bulcaen (eds.), Handbook of Pragmatics, 105-
Ill. Amsterdam: John Benjamins.
Lyons, John
1977 Semantics. 2 volumes. Cambridge: Cambridge University Press.
Nerlich, Brigiue
(to appear) Synecdoche: a trope, a whole trope, and nothing but a trope. In:
Armin Burkhardt and Brigiue Nerlich (eds.), Reflections on Tropes
and Truth. Berlin: Mouton de Gruyter.
The New Oxford Dictionary of English
1998 The New Oxford Dictionary of English, edited by Judy Pearsall. New
York: Oxford University Press.
Nunberg, Geoffrey D.
1978 The Pragmatics of Reference. Bloomington: The Indiana University
Linguistics Club.
Panther, Klause-Uwe and Gtinter Radden (eds.)
1999 Metonymy in lAnguage and Thought. Amsterdam: John Benjarnins.
Radden, Gtinter and Zoltan Koevecses
1999 Towards a theory of metonymy. In: Panther and Radden (1999),
17-59.
Reddy, Michael J.
1979 The conduit metaphor: a case of frame conflict in our language about
214 Ken-ichi Seto

language. In: Andrew Ortony (ed.), Metaphor and Thought. Second

edition 1993, 164-201. Cambridge: Cambridge University Press.
Rosch, Eleanor
1973 Natural categories. Cognitive Psychology 4: 328-350.
1978 Principles of categorization. In: Eleanor Rosch and Barbara B. Lloyd
(eds.), Cognition and Categorization, 27-48. Hillsdale: Lawrence
Erlbaum.
Seto, Ken-ichi
1999 Distinguishing metonymy from synecdoche. In: Panther and Radden
(1999),91-120.
Taylor, John R.
1995 Linguistic Categorization: Prototypes in Linguistic Theory. Second
edition. New York: Oxford University Press.
Tversky, Barbara
1990 Where partonomies and taxonomies meet. In: Savas L. Tsohatzidis
(ed.), Meanings and Prototypes: Studies in Linguistic Categorization,
334-344. London: Routledge.
Tversky, Barbara and Kathleen Hemenway
1984 Objects, parts, and categories. Journal of Experimental Psychology:
General 113: 169-191.
Ullmann, Stephen
1951 The Principles of Semantics. Glasgow: Jackson.
1962 Semantics: An Introduction to the Science of Meaning. Oxford:
Blackwell.
1964 Language and Style: Collected papers. Oxford: Blackwell.
Winston, Morton E., Roger Chaffin and Douglas Herrmann
1987 A taxonomy of part-whole relations. Cognitive Science 11: 417-444.
Synchrony/diachrony approaches
Polysemy in derivational affixes

Adrienne Lehrer

1. Introduction

The question of whether lexical elements and grammatical elements are

semantically similar has been debated. Some linguists, e.g. Beard (1988, 1990,
1992), argue that lexemes and affixes, especially derivational affixes, are
quite different, while others, such as Baker (1988), Lieber (1992), Lehrer
(1993, 1996, 1999), and Panther and Thomburg (2002) show they are similar
in that they are signs. Derivational affixes and lexemes share many semantic
relations, such as synonymy, antonymy, and polysemy (the topic of this
paper). Beard (1990) made an important observation about the semantics of
derivational affixes (and function words), namely that the meanings expressed
are limited to those concepts that also become grammaticalized: space, time,
agency, possibility, animacy, etc. I Since the range of semantic concepts as
well as most examples of metaphor and metonymy involve highly lexical
(as opposed to grammatical) concepts, the question arises as to whether and!
or to what extent the polysemy found in derivational affixes parallels that
found in lexemes. Yet even with limitations we expect to find the same types
of polysemy that have been identified in the traditional studies of meaning
change and in the works of many cognitive linguists interested in polysemy.
If the concepts and relationships represented by words change and expand,
then we should expect to find the same phenomenon in derivational mor-
phemes too. So even if we do not find the range of polysemy reported in
works like Apresjan (1974), Norrick (1981), Nerlich and Clarke (1992) and
others, we should expect the polysemy we find in affixes to be a proper
subset of regular polysemy.
A common structure of the relationship among senses is that of a central
sense (Bloomfield 1933), which gives rise to a variety of different derived
senses, where the derived senses are not necessarily related to each other
(though they may be). Radial structure is the contemporary term for this
relationship, and the suffix -ship (examined in Section 5) exhibits this structure,
as do over (Brugman 1981; Lakoff 1987), and -er (Panther and Thomburg
218 Adrienne Leo/er

2(02). In some cases, however, the relationship of the senses has a different
"shape", as Joos (1958) observed. His analysis of the word code had circular
shape, and my analysis of -ist in Section 4 exhibits a chain-like structure,
where sense A gives rise to sense B, which in turn gives rise to sense C.
Although this article deals only with English, it may serve as an example
for similar studies of other languages. My main source is the classic study
by Marchand (1966), supplemented by recent neologisms, which often reveal
aspects of productive senses of words and morphemes that are not necessarily
revealed in the conventional lexicon.

2. English prefixes

In the case of some English prefixes,2 we find mUltiple meanings similar to

those of full lexemes, namely the polysemy of space and time. Consider
Table 1.

Table 1. Spatial and temporal polysemy.

Prefix Spatial Temporal

ante- antechamber antebellum

fore- foreground, forehead, foreplay foreshadow
pre- pre-abdominal prenatal, prewar
post- postfrontal postwar, post-modern

Pre- and post- are predominantly temporal senses. The spatial senses are
restricted to anatomy and zoology (Marchand 1966: 134), and this sense
emerged in English later than the temporal one. The temporal sense of pre-
also has a range of meaning, influenced in part by the word-class of the
base. With nouns (prewar, precontact, pre-election) the time denoted is before
the event or period named. With verbal bases (preheat, preshrink, prepay),
the sense is "to perform some action before some pragmatically interpreted
action". To preheat the oven is "to turn it on to a certain temperature before
putting the food in". To prepay is "to pay [a bill] in advance of the due date
or the service performed". There has been a gradual drift in words like
preboard [an airplane] and preregister [for a class], where some individuals
act before other individuals do the same thing. To preboardlpreregister is
 Polysemy in derivational affixes 219

"to board or register, not before some other action, but before the regular
boarding or registration takes place".
Another domain in which we find a familiar instance of polysemy is in
prefixes that denote hierarchies. Here the meanings include those of generality
(versus specificity), quantity (more or less), and, in the case of super- and
sub-, spatial orientation. See Table 2.

Table 2. The poly se my of quantity and generality.

Prefix Generality Quantity Space

hyper- hypernym hyperactive

hypo- hyponym hypothermal
super- superset superactive supertitle3
sub- subset, subdivision subnormal subtitle

The sense involving quantity implies a contextual norm, and all four of
the prefixes in Table 2 denote a quantity falling above or below that norm.
The prefix arch- shares some of the senses of the four items in Table 2.
Its primary sense has to do with a social hierarchy, as in archbishop and
archduke, but it has developed a sense of quantity as well. Here we find
arch-heretic and arch-villain, where the meaning is "excessive". Marchand
(1966: 144) suggests that the use is pejorative, which may explain why he
has no examples like arch-hero and arch-saint.
Generalizing over the nine items above, we see a set of meanings ranging
from space and time, and metaphorical extensions of spatial notions like
over and under, to relationships of power and to quantity. Generality is a
special case of quantity, since "more general" = "more inclusive".

3. Diminutives

Another domain with well-known polysemy involves diminutives - affixes

denoting small size, with a cluster of senses related to small size, such as
young age and small quantity. In addition, there are extensions to meanings
of affection and pejoration. Dressier and Merlini Barbaresi (1994) present a
comprehensive analysis of Italian, German, and other European languages,
where the diminutive is carried by suffixes. As Dressier and Merlini Barbaresi
220 Adrienne Lehrer

show, the meaning of small easily shifts to endearment - the affection we

feel for small children and small animals, and also to pejoration, since small
can denote "lesser importance". English has a few suffixes too (see Table 3),
although their productivity is limited. Interacting with diminutives is at least
one feminine suffix, -ette, which also carries diminutive meanings. I have
included some questionable items, where the suffix carries the meaning listed,
but the base is either meaningless or requires some re-interpretation. For
example, although booklet can be glossed as "little book", anklet is not "a
little ankle". However, the suffix -let stilI connotes "small size". A hamlet is
a "small town", but the base, ham- (not the homonymous pork product) has
no independent identifiable sense.

Table 3. The polysemy of size and amount.

Prefix: Size-Amount-Age Other senses

-ette dinette, kitchenette [PLACE] launderette, luncheonette

[FEMININE] majorette

-kin(s) napkin [ENDEARMENT] babykins

-let booklet, hamlet, anklet [PLACE] anklet
starlet 4 [PEJORATION] kinglet
-ling duckling, sapling [ENDEARMENT] darling
[PEJORATION] weakling
-y baby, doggy, horsy [ENDEARMENT, BABYTALK]

The metonymy of PLACE meanings for some of the items above are
transfers of the whole word, not the suffix. -ling, with the exception of
darling, is affectively pejorative, as in weakling, trivling "trivial, worthless",
giftling "trivial gift", witling "one with small wit" (examples and glosses are
from Walker [Dawson] 1936: 170-171). The meaning of the suffix -y is
primarily "like", "characterized by", "full of' (Marchand 1966: 352); the
diminutive sense is secondary. 1\vo diminutive suffixes are widely found in
personal names for girls, although speakers probably do not interpret them
as diminutives: Annette and Nancy. However, the -y is productively used for
nicknames: Barby, Keithy, Tommy.
 Polysemy in derivational affixes 221

4. Suffixes expressing agent/subject arguments

Another set of suffixes exhibiting polysemy are those for agent/subject

arguments - suffixes that can be paraphrased as "one who is" or "one who
does" (Table 4). The suffixes examined are -an/ian, -ant/ent, -arian, -er, -ist,
and -ite. 5 Although each suffix has its own semantic structure, items tend to
share some common meanings: MEMBER OF A CLASS, INHABITANT OF A PLACE,
AGENT, INSTRUMENT, or FOLLOWER OF AN IDEOLOGY.

Table 4. Agentive suffixes.

Suffix Agent Member of Inhabitant Instrument Follower! Other

a group of Ideology

-anl-ian logician plebeian, Persian, Lutheran HAS CHAR-

reptilian Arabian ACTER OF

-antl-ent servant. solvent

defendant

-arian sexagen- Unitarian,

arian libertarian

-er teacher commis- New Yorker, toaster, CONNECfED

sioner Berliner lighter WITH

-ist pianist, Marxist, PREJUDICE

chemist Calvinist

-ite socialite Mennonite TECHNICAL

The suffix -er displays a traditional radial structure, with the AGENT sense
in the centre (see Fig. 1). (See Panther and Thomburg 2002 for a more detailed
analysis of the various senses of -er and their relationship.)
Of the agent suffixes other than -er, the agent sense is not necessarily
central. Beard (1990) has described the frequency of agentive and instru-
mental polysemy in a number of languages, and such polysemy generally
occurs by means of -er and -ant/ent. The technical senses of -ite are mostly
from chemistry.
The suffix -ist, which is attached to nouns or bound bases, has three
overlapping senses. The first is the agentive one, such as violinist or physicist.
The second, which is related to -ism, describes a proponent of an ideology,
222 Adrienne Lehrer

blender

slipper,
wrapper

clincher,
finisher

commissioner

Figure 1. The po\ysemy of -er.

such as federalist or capitalist. The third is like the second, but it additionally
adds a connotation of prejudice (Lehrer 1988, 1999). Historically, this sense
evolved from the word racist, "one who believes in innate difference among
races". But racist soon took on the sense "being unjustifiably prejudiced
against a group on the basis of race" and the negative connotation transferred
to the -ist, which expanded to sexist, ageist, speciesist, and other terms. This
shift illustrates Traugott's analysis, where pragmatic implicatures give rise
to conventional uses. "Semantic change ... involves specification, achieved
through inferencing" (1988: 413). More interesting, however, is her obser-
vation that (in the case of metonymy) generally the meaning shifts to "the
subjective belief state or attitude toward the situation" (p. 414). With -ist, the
earlier sense, as in federalist, has a neutral connotation, while the newer
sense, as in sexist, is negative and reveals the speaker's evaluation.
 Polysemy in derivational affixes 223

5. Neologisms: a data source for semantic change in progress

One difficulty in determining the senses of derivational affixes (and lexemes

for that matter) lies in the fact that although the productive meanings change
over time, the language retains many items with older senses. For example,
the current productive meaning of -ster is "an agent who does things that are
bad or at least shady" (Lehrer 1999). Yet words like spinster and youngster
obviously do not have this meaning.
One affix that I have been watching for several years is the prefix meta-
(see Fig. 2). Numerous words with this prefix have existed in English for a
long time, e.g. metaphor, metaphysics, and metamorphosis. There are also
many scientific words with meta-, even new coinages based on analogy with
existing scientific terms. The meanings glossed for these uses include
"beyond" or "transcendental", and "situated behind" for some technical terms
in anatomy and zoology, e.g. metathorax "hindpart of the thorax of an insect"
and metapodial "one of the bones of the hand". Almost all of Marchand's
examples of meta- are from this class, with subcategories based on the specific
sciences (Marchand 1966: 125).
However, there are three related, newer productive senses, all of which
were first recorded in the OED after 1925, and all have an umbrella sense of
"second order representation". The specific overlapping (and not completely

metamathematics

metaquestion.
metafiction

metagalru:y

Figure 2. The po\ysemy of meta-.

224 Adrienne Lehrer

discrete) senses are: (i) "the foundations of a science or discipline", (ii)

"aboutness", and (iii) "more general". (Some words incorporate two senses.)
The first meaning, "a foundational theory of X", is exemplified by meta-
history, metatheory, metapsychology, metamathematics,6 and metalinguistics.
Metasociology can be paraphrased as "the foundations of the study of sociol-
ogy"; metalinguistics is "the theoretical foundation of the study of language
or linguistics".
The second sense, which is the most productive, can be glossed as ABOUT,
and some words contain both meanings. For example, metamathematics is
about mathematics, and metalanguage is about language, but the ABOUT sense
of meta- contains a variable where the word in the base is repeated: "(an) X
about (an) X".

A metalanguage is a language about language(s).

A metamovie is a movie about a movie (movies).
Metacognition is cognition about cognition.
A metarule is a rule about rules.
A metamessage is a message about a message (messages).

So whereas metalanguage contains senses (i) and (ii), metahistory contains

only the first sense, since it is not "history about history". Similarly, meta-
mathematics is about mathematics, but is not "mathematics about mathematics".
Some cases require a more elaborate paraphrase. Metamind is not "a mind
about a mind". Instead we must go to the related adjective, metamental, and
define it as something like "a mental phenomenon (activities, processes,
judgements, etc.) about mental phenomena". Then we can define metamind
in terms of metamental.
Some semantic differences between meta-X and the paraphrase follow
from the syntactic differences. The syntax about NP requires a decision on
the count/mass/number features and definiteness of the nouns, where no
contrasts are possible with the meta- words. Hence, metalanguage could be
defined in any of the following ways:

A metalanguage is a language about a language.

A metalanguage is a language about languages.
Metalanguage is language about language.
Metalanguage is language about languages.
 Polysemy in derivational affixes 225

As in interpreting any utterance, pragmatic factors require fine-tuning based

on contextual factors.
Let us turn now to the semantic requirements of the base of the second
sense. One aspect of the meaning of these novel uses of meta- is the ABOUTNESS
requirement. One can understand what a metamovie is - a movie about a
movie - but it is perplexing to imagine what a metapotato might be, just as
it is perplexing to interpret "a potato about a potato".
A primary class of items having ABOUTNESS properties are linguistic products
of various sorts: speech acts, texts of various genres, concepts of linguistic
analysis, and means of realization. Examples I have observed include meta-
question, metasentence, metawriting, and metajoke. The American Heritage
Dictionary has an entry for metafiction, "fiction that deals, often playfully
and self-referentially with the writing of fiction or its conventions." In the
domain of visual and performing arts we can have metapicture, meta-
documentary, and metamovie. For example, Robert Altman's film The Player
is not only a movie about a movie, but even a movie about itself. This sense
of meta- can also be extended to bases denoting emotions, since emotions
can have ABOUTNESS properties. Wilbur (1977) coined the term metagrumbles
to describe "grumbling about [other people's] grumbling".
The third meaning of the prefixe meta- is "more general, more inclusive".
The earliest citation of this sense in the OED is metagalaxy, "cluster of
galaxies" (1930). However, if this sense is glossed simply as "of', the
overlap in meaning of ABOUT and OF is a natural development, as seen in
the partial overlap/synonymy of phrases like a book of birds and a book
about birds. This recent nuance is illustrated by words like metapopulation
and meta-invention. A metapopulation is not "a population about popu-
lations", but rather "a higher-order population - a population OF populations".
An article about microprocessors predicts that "it is not merely an invention
but a meta-invention, which enables us to create yet other inventions"
(Modern Maturity, Nov.-Dec. 1998: 65; emphasis added). In this example, a
meta-invention is not "an invention ABOUT inventions", but "an invention OF
[other] inventions". This use of meta- keeps the variable, but shifts the
meaning from ABOUT to OF.
Of these three related senses, although the sense ABOUT is not the first
recorded, it is synchronically central, lying between "foundation of' and
"more general". (Geeraerts [1985] demonstrates how prototypicality can shift
diachronically.)
The next three suffixes, -dom, -hood, and -ship, overlap in meaning in
226 Adrienne Lehrer

that they have a sense STATE or CONDITION, but in addition each has its own
range of meanings.
-dom (see Fig. 3) is in fact a reduced form of a formerly independent
word, with a central sense of "state" or "jurisdiction", as in kingdom or
Christendom. This sense includes both concrete and abstract meanings, so
we can have a phrase like kingdom of God, referring not to a physical place
but to an abstract realm. From this meaning it is a short step to the sense of
"territory", as in kingdom "place where the king rules", dukedom, and
sheikdom. Also from the general sense of "state, condition" it is a natural
shift to a special kind of condition illustrated by stardom, having the rank or
being in the class of stars.

dukedom
/TERRITORY

STATE!
CONDITION • STATUS, RANK earldom

~colTIVITY clerkdom

Figure 3. The polysemy of -dom.

Marchand (1966: 262) lists others classes of meanings, such as humorous

and pejorative senses, but these are not distinct senses. There are, rather,
independent semantic dimensions that intersect with the meanings, and they
can be found in any of the above senses: dogdom, gangsterdom, autodom.
-hood also has a general sense of "state" or "condition", but this sense
includes "state of affairs", as in falsehood and likelihood, which leads to a
sense of "status", e.g. wifehood, widowhood, which in turn leads to "time
period connected with that state" as in babyhood, childhood. The first sense
also develops into a collective body, as in brotherhood, sisterhood. See
Figure 4.
The third item, -ship, has a more complex set of meanings (Fig. 5). The
sense "state, condition" is found in items like kinship and ladyship, which
blends into "office, rank" (ambassadorship, editorship), which is very closely
related to "role, position" (leadership, dictatorship). Both of these latter two
senses lead to "respectful designation" (ladyship, lordship). The general sense
also yields senses of skill and art (penmanship, marksmanship) and also to
 Polysemy in derivational affixes 227

/ STATE OF AFFAIRS faLfehood

STATE!
CONDITION • STATE ~ STAGE OF LIFE
motherhood - - -... babyhood

~ COLLECTIVE BODY brotherhood

Figure 4. The po\ysemy of -hood.

"community, collective" (partnership, fellowship}, and finally, the sixth sense

has developed into "stipend, financial benefit", in scholarship, fellowship,
and TAship (teaching assistantship). However, this last may be limited to
specific nouns, and may not be a feature of the suffix.
The last affixes to be discussed are three negative prefixes: de-, dis-, and
un-, attached to verbs and participles. All three have a general meaning of
producing the opposite state, usually by reversing an action or returning to
an earlier state, but each has at least one other sense or nuance. In addition,
there are many subtleties and complexities of meaning that I will not discuss
here, since I am concerned only with polysemy.

. / 0lCE, POSITION editorship. ambassadorship

/"
.----r ROLE, POSITION leadership
STATE OF BEINGI t
RELATIONSHIP ~ •
kinship
~ RESPECTFUL DESIGNATION ladyship. lordship

SKILL penmanship. marksmanship

COMMUNITYICOLLECTIVITY partnership, fellowship

~
STIPEND fellowship, scholarship

Figure 5. The poJysemy of -ship.

228 Adrienne Lehrer

Examples of de- that show the sense of "reversing an action or state"

include decentralize, destalinize, decompress and decode. But one special
sense of de- is "getting rid of, remove something", as in delouse, dethrone,
and deforest. One publication from several years ago coined the term deNewt
to describe Newt Gingrich's resignation from the US House of Representatives.
When Gloria Estefan was left out of a list of divas in People Magazine, she
described herself as being dedivaed. The specialization of meaning "get rid
of, remove" is closely related to the general meaning of "reverse state" or
"undo action", since to remove something or someone involves a change of
state, usually to a previous state. But an area can be deforested that has
"always" been forest land (i.e. as long as anyone can remember).
The reversal sense of dis- is illustrated by disarm and disappear. In
addition, this prefix can connote "refuse to, fail to" as in disobey and disallow.
Some words require greater complications in their interpretation. For example,
disinherit is not simply a reversal or negation of inherit, since a paraphrase
of Mr. lones disinherited his children is not a negation of Mr. lones inherited
his children. The syntactic frames of each verb are not parallel.
The last negative item is un-, whose reversal of state sense can be seen in
untie, unwind, unlace, and unbuckle. (Marchand also lists unfreeze.) A closely
related, specific sense is found in unhorse, unsex, and unman. To unhorse
someone is to reverse a state, since that person is no longer on a horse.
However, to unsex someone is not to return to a previous state but to create
a different state. (Presumably this word is used only metaphorically.)
I have not been able to find any systematic contrasts between un- and
dis- with regard to "deprive of', since bases tend to take one or the other. An
earlier contrast in meaning between uninterested and disinterested (which
disambiguated the word interest) has been lost as disinterested has been
replacing uninterested.

6. Discussion and Conclusions

One traditional issue in work on polysemy is a controversy between mono-

semy and polysemy - whether to try to find one general sense (as pursued in
the work of Ruhl [1989]) or to posit a large number of highly specific senses,
almost one for every image schema (Brugman 1981; Lakoff 1987). In some
cases, it is quite clear which way to go, since the traditional tests for ambiguity
will give clear results (Geeraerts 1993; eruse 1986, 1995). However, in many
 Polysemy in derivational affixes 229

other cases, decisions are hard to make because the ambiguity tests are
indeterminate. Ambiguity tests often fail where a general sense includes a
specific sense, such as egg "reproductive body from various animals", "egg
from a bird, with a yellow yolk, surrounded by a hard shell" [and other
senses]. This type of polysemy has been termed autohyponymous polysemy
by eruse (1995).
In some of the affixes I have examined, neither solution - monosemy nor
polysemy - seems optimal. Selecting monosemy leaves too much to the
pragmatics and ignores the rather specific, contextually determined senses;
speakers do not need to figure out that specific meaning each time they
encounter a new use of the affix. Yet postulating many specific senses fails
to capture the fact that there is a unity that connects the various meanings. In
Figures 1 and 2, I have tried to capture both the general and specific senses.
The general meaning includes the specific ones, but not necessarily exhaus-
tively, and this in turn allows for the generation of nuances and innovations
lying outside the specific senses. Tuggy's model (1993) is the closest one I
have found to resolving the dilemma.
Tuggy, like Geeraerts (1993), shows that there is a continuum, not a sharp
break, between ambiguity and vagueness. Moreover, diachronic factors blur
the distinction even more. Tuggy shows how general schemas [general senses]
and their elaborations [specific senses] co-exist in languages. Schemas or
elaborations, or both, can become well-established through repeated usage
and thereby become salient (Thggy 1993: 279). When an elaborated sense
becomes salient, it is considered prototypical. If the general sense is not
salient but some of the elaborations [specific senses] are salient, the senses
strike us as a case of ambiguity. However, if the general sense is salient, but
the specific senses are not, we judge the item as exhibiting vagueness. When
the general sense and at least one of the elaborated senses are salient, we are
inclined to judge ambiguity. Tuggy's model of schema and elaborations does
not require us to choose vagueness or ambiguity - we get both, along with
an account of context and conditions that explain our judgements in clear
cases and borderline cases.
Although the concepts represented by English affixes are limited, they
show polysemy that is similar to that found in lexemes. There are meta-
phorical transfers of "space" and "time", "vertical position (superiority)",
"generality", and "rank", and there is some metonymy, as in "a group or
class of people" and "inhabitant of'. Also there is the metonymic shift from
"membership of a group" to "award for election to the group" to "financial
230 Adrienne Lehrer

support". Another similarity can be found in the structure of polysemy, which

can be represented by radial categories, as exemplified by Brugman (1981),
Lakoff (1987), and many other cognitive linguists. Figures 1 through 5 present
a few of the affixes which show this pattern.
However, one claim by cognitive linguists is challenged by these data. It
is assumed that the direction of change and innovation is from concrete to
abstract, and specifically from space to time. 7 However, the prefixes pre-
and post- primarily express temporal concepts, and the transfer to space is
synchronically secondary and diachronically later. In any case, these examples
challenge the assumption of necessary directionality. Yet given the fact that
concepts that are expressed in affixes are already rather abstract, it could be
argued that such examples are not serious counterexamples.

Notes

I. Mithun (1996) provides data from American Indian languages showing that the
range of meanings of derivational affixes is larger than Beard claims.
2. Marchand (1966) lists all the items I discuss as prefixes, and I will follow him in
this point. However, these morphemes differ as to their boundedness. Unfor-
tunately, English spelling is no guide. Some items are usually spelt with a hyphen,
some with a space before the base, and some are always part of the following
word. Moreover, there are often inconsistencies in a given item.
3. Some opera companies call the translations of the text appearing above the stage
supertitles; other call them surtitles. The prefix sur- is much less productive in
English than super-.
4. Starlet suggests quantity - "a little bit of stardom" and possibly mild derogation.
It also seems restricted to females, reinforcing an association of feminine and
diminutive senses.
5. There are other agent/subject suffixes not dealt with here, some of which are
discussed in Lehrer (1999).
6. The OED provides an 1853 citation of metamathematics as "beyond the scope
of', but a 1926 citation is glossed as "pertaining to".
7. There is a whole scale from very concrete concepts expressed in words like
table to extremely abstract ones, such as the meaning of metaphysics, with an
indefinite number or points in between. Moreover, it is not always clear how to
arrange concepts on this scale. Is time more abstract than space?
 Polysemy in derivational affixes 231

References

American Heritage Dictionary

1992 Boston: Houghton Mifflin.
Apresjan, Jurij D.
1974 Regular polysemy. Linguistics 142: 5-32.
Baker, Mark
1988 Morphology and syntactic objects. A review of A. M. DiScillo and
E. Williams, On the Definition of Word. Yearbook of Morphology 1:
259-284.
Beard, Robert
1988 On the separation of derivation from morphology: toward a lexeme-
morpheme based morphology. Quaderni di semantica 9: 3-59.
1990 The nature and origins of derivational polysemy. Lingua 81: 101-140.
1992 Lexeme-Morpheme Base Morphology: A General Theory ofInflection
and Word Formation. Albany, NY: State University of New York
Press.
Bloomfield, Leonard
1933 Language. New York: Holt, Rinehard and Winston.
Brugman, Claudia
1981 The Story of Over. M.A. thesis. Linguistics Dept, University of
California, Berkeley.
Cruse, David A.
1986 Lexical Semantics. Cambridge: Cambridge University Press.
1995 Polysemy and related phenomena from a cognitive linguistic viewpoint.
In: Patrick Saint-Dizier and Evelyne Viegas (eds.), Computational
Lexical Semantics, 33-49. Cambridge: Cambridge University Press.
Dressier, Wolfgang U. and Lavinia Merlini Barbaresi
1994 Morphopragmatics: Diminutives and Intensifiers in Italian. German.
and Other Languages. (Trends in linguistics. Studies and Monographs
76). Berlin and New York: Mouton de Gruyter.
Geeraerts, Dirk
1985 Cognitive restrictions on the structure of semantic change. In: Jacek
Fisiak (ed.), Historical Semantics. Historical Word Formation, 127-
154. Berlin: Mouton.
1993 Vagueness's puzzles, polysemy's vagaries. Cognitive Linguistics 4:
223-272.
loos, Martin
1958 Semology: a linguistic theory of meaning. Studies in Linguistics 13:
53-71.
Lakoff, George
1987 Women. Fire and Dangerous Things. Chicago: University of Chicago
Press.
232 Adrienne Lehrer

Lehrer. Adrienne
1988 A note on the semantics of -ist and -ism. American Speech 63: 181-185.
1993 Prefixes in English word formation. Folia Linguistica 2: 133-148.
1996 Why neologisms are important to study. Lexicology 2: 63-73.
1999 Are affixes signs?: The semantic relationships of English derivational
affixes. In: Wolfgang U. Dressier. Oskar E. Pfeiffer. Markus A.
Pochtrager and John R. Rennison (eds.), Morphologica 1996, 143-
154. (Papers from the Seventh International Morphology Conference,
Vienna, Austria, February 1996.) The Hague: Holland Academic
Graphics.
Lieber, Rochelle
1992 Deconstructing Morphology. Chicago: University of Chicago Press.
Marchand, Hans
1966 The Categories and Types of Present-Day English Word Formation.
Birmingham: University of Alabama Press.
Mithun. Marianne
1996 The meanings of roots and affixes. Papers from the Seventh Inter-
national Morphology Conference, Vienna, Austria, February 1996.
Nerlich. Brigitte and David D. Clarke
1992 Semantic change: case studies based on traditional and cognitive
semantics. Journal of Literary Semantics 21: 204-225.
Norrick, Neal R.
1981 Semiotic Principles in Semantic Theory. Amsterdam: John Benjarnins.
Oxford English Dictionary
1989 Oxford: Clarenden.
Panther, Klaus-Uwe and Linda L. Thornburg
2002 The roles of metaphor and metonymy in -er nominals. In: Rene
Dirven (ed.), The Metaphor-Metonymy Continuum: The Ongoing
Debate, 279-319. Berlin: Mouton de Gruyter.
Ruhl, Charles
1989 On Monosemy. New York: State University of New York Press.
Traugott, Elizabeth C.
1988 Pragmatic strengthening and grammaticalization. Berkeley Linguistics
Society 14: 40~16.
Tuggy. David
1993 Ambiguity, polysemy, and vagueness. Cognitive Linguistics 4:
273-290.
Walker, John
1936 Rhyming Dictionary, Revised and enlarged by Lawrence H. Dawson.
New York: Dutton.
Wilbur. Ken
1977 The Spectrum of Consciousness. Wheaton, fi..: Theosophical Publishing
House.
The role of links and/or quaUa in modifier-head
constructions

Beatrice Warren

1. The role of qualia in semantics

Whereas most publications on linguistic topics represent advances analogous

to millimetres or at best inches, now and then there appears an article or
monograph that represents a leap forward in terms of yards. Although it is
too early to include Pustejovsky (1996) in this illustrious league, judging by
the interest his model of generative semantics has raised, he may be con-
sidered a potential candidate.
The attraction of Pustejovsky's theory is that it suggests ways in which
the flexibility of word meaning can be predicted. It is possible, he claims, to
specify formally the generative devices "by which words can achieve a
potentially infinite number of senses in context, while limiting the number
of senses actually stored in the lexicon" (1996: 105).
As this quotation makes evident, he accepts, like so many semanticists
in the last few decades, that word meaning is negotiable in context. However,
unlike other proponents of flexible word meaning, he insists that something
we can call default polysemy is "rule-governed". He accepts, again like
many semanticists, that word meaning is componential, but rejects a simple
list of features of meaning. Instead he argues for a rich representation
involving argument structure (number and type of arguments associated
with the word), event structure (type of event in which the word is involved)
and qualia structure (a set of possible properties explaining the meaning of
the word). These form the material on which generative devices operate in
a well-defined manner to yield appropriate contextual interpretations of
words.
Thus we appear to have a lexical parallel to syntactic rule-governed
creativity, which would explain the allure of the theory. The introduction of
qualia roles, which "not only structure our knowledge of words, but also
suggest interpretation of words in context" (Pustejovsky 1996: 87) is a
particular innovation, and a very audacious and strong claim. The origin of
234 Beatrice Warren

these roles, Pustejovsky informs us, is Aristotle's notion of modes of expla-

nation. There are four roles:

(i) CONSTInJTIVE: the relation between an object and its constituents, or

proper parts.
(ii) FORMAL: that which distinguishes the object within a larger domain.
(iii) TELIC: purpose or function of the object.
(iv) AGENTIVE: factors involved in the origin or "bringing about" of an
object.

These four factors "drive our understanding of an object or a relation in

the world. They furthermore contribute to (or, in fact, determine) our ability
to name an object ... " (1996: 85). This brings to mind a theory by Clark and
Clark (1979: 789), which was inspired by Rosch and her colleagues' research
into the principles of human categorization, and which suggested that the
basis for categorization of objects includes (i) physical characteristics (cf.
the CONSTITUTIVE and FORMAL qualia), (ii) ontogeny (cf. the AGENTIVE quale)
and (iii) potential roles (cf. the TELIC quale). The qualia roles would in other
words represent the type of features which cause us to think of objects as a
kind, i.e. common constituent matter or parts, common gestalts (shape, size,
position, etc.), common origins, common functions. If we think of words as
category labels, I we would then have a rationale for the qualia roles - at
least for first-order entities2 - which finds support in research into categ-
orization. The question is: is there also empirical evidence that qualia structure
plays an important role in the description of meanings of words and in lexical
creativity giving rise to polysemy? The focus of the present contribution will
be that very question and we will turn to two different types of empirical
studies in search of an answer, first to a study dealing with defining com-
ponents of meaning, and then to a collection of studies concerned with
relational meanings.

2. Empirical evaluation

Some years ago Geeraerts presented ingenious studies of the particular features
that are salient in naming new objects. We get a detailed description of one
of these concerning a garment first used in Belgium in 1987, frequently
referred to by the English loan word leggings (1997: 32-47). The occurrence
 Links and qualia in modifier-head constructions 235

of this word was examined in fashion magazines and mail-order catalogues,

but only if a picture or drawing illustrated the referent. This made it possible
to describe the referents of leggings according to the following six dimensions:
the length of the garment (three possible "values": to the knee, below the
knee, to the ankles); its width (tight-fitting, somewhat tight-fitting, loose-
fitting); the presence of a crease (a clear crease, no crease); the material
(finely woven or knitted, coarsely woven or knitted, transparent material,
any other type); its function (upperwear or underwear); and finally the sex
of the person wearing the garment. The results of the study showed that in
1988 the prototypicallegging is long, tight-fitting, has no crease, is made of
an elastic material and is worn by women like a pair of trousers. There are
similar results for 1989, 1990, 1991 and 1992, but each successive year there
appears an increasing number of referents which deviate from the prototype.
These findings support the prototypical approach to semantics. That is to
say, categories cannot be defined by a single set of necessary and sufficient
attributes but exhibit a family resemblance structure, allowing definitions of
words to expand along lines like the following: ABCD ~ BCDE ~ ACDE
~ ABC ~ BCD etc.
Pustejovsky's model does not address the changeability of criterial attri-
butes. Nor does it account for the fact that defining features may be more or
less salient. However, the referential features themselves, which were chosen
by the investigators because they were felt to play a part in determining
category members, appear to match the quaJia fairly well. Length, width,
crease could be said to represent FORMAL qualia; material a CONSTITUTIVE
quale; the features "upperwear for women" arguably TELlC qualia.
Obviously a single example cannot support a theory in a decisive manner.
There is, however, a collection of studies which could give substantial
support. These studies are concerned with the semantics of modifier-head
combinations and are relevant since an important point of Pustejovsky's
theory is that it is when we combine words in utterances that the represen-
tational structures and the generative devices are activated to produce implicit,
nonlisted meanings. The collection comprises the following (in particular):
Ljung (1970), Brekle (1970), Aarts and Calbert (1979), Bartning (1980), Levi
(1978), Warren (1978,1984) and Leitzky (1989). They have in common the
hypothesis that the interpretation of modifier-head constructions involves
the retrieval of implicit links. That this would be so in the case of noun-
noun compounds had generally been assumed: silver spoon "spoon made of
silver", apple tree "tree producing apples", cart wheel "wheel belonging to
236 Beatrice Warren

cart", etc. Ljung's study, however, showed that adjectives also contain a link,
or, more precisely, that the definition of adjectives typically involves a
predicative verb and its complement, which is normally identical to the
nominal stem in the case of denominal adjectives: stony "abounding in
stones", beautiful "possessing beauty", noisy "producing noise". Aarts and
Calbert (1979) formalized this mode of defining adjectives and stipulated
that there are at least two types of semantic elements in adjectives: one with
predicative and one with referential force. 3 The predicate links the referential
content of the adjective with its head: angry man "man experiencing anger",
sad event "event causing sadness". There is, they also postulated, a limited
number of preferred links. Which of these is relevant depends on the nature
of the referential content and the head of the adjective. Warren (1988) later
illustrated that the variability of the link accounts for a certain type of
polysemy in the case of adjectives. Consider Table 1.

Table 1. Variability of links in adjectives.

Link Referential content Example

sad experiencing sadness sad girl

manifesting sad eyes
causing sad event
healthy having health healthy child
causing healthy food
smoky emitting smoke smoky fuel
containing smoky room
resembling smoky colour

The studies listed above did not only have in common the assumption
that the semantics of nominal modifiers could involve implicit links. As just
indicated, they also postulated that there exists a list of preferred relations
or, to use my terminology, default relations, which would naturally be invoked
unless there was strong contextual evidence to the contrary (Warren 1985:
378-380, 1988: 132-134). Compare Pustejovsky's admission that a particular
interpretation specified by the qualia structure may be overridden in certain
contexts. For instance, Bob finished the novels could have two possible regular
readings, i.e. "Bob finished reading the novels", in which case a feature in
the TELlC quale would be activated (novels are for reading), or "Bob finished
 Links and qualia in modifier-head constructions 237

writing the novels", in which case a feature in the AOENTlvE quale is activated
(novels come about by being written). However, the interpretation "Bob
finished wrapping up the novels" could be triggered in a case such as the
following: Bob, we know, works at a publisher's firm and one of his jobs is
to wrap and send various types of books - novels, textbooks, children's books,
etc. Linguists have in the past not been interested in default values, possibly
for ideological reasons, "all and only the grammatical constructions" being
the goal. If Pustejovsky succeeds in persuading the linguistic guild of the
validity of default values, this would amount to a minor breakthrough in my
view, in particular in semantics. 4
The lists of links presented in the studies cited are not identical, but
strikingly similar, and could be said to be variations on a theme. My own
version is shown in Table 2.
As we see from the above, the role combinations are reversible with the

Table 2. Implicit links.

Link Example Role combination

Compositional
consist of/constituted by electric power SOURCE-RESULT
constituting tragic event RESULT-SOURCE

Possessive
belonging to facial skin WHOLE-PART
having rational creature PART-WHOLE

Locative
occurring in/at/on polar climate PLACE-oBJ
medieval games TIME-oBJ
containing poisonous plant OBJ-PLACE
during which ... prevails happy times OBJ-llME

Causative
caused by nervous breakdown CA USER-RESULT
causing noisy children RESULT-CA USER

Purpose
be for culinary utensils GOAL-INSTRUMENTS

Resemblance
be like golden hair COMPARANT-COMPARED
be in accordance with normal behaviour NORM-ADHERENT
238 Beatrice Warren

exception of the resemblance and the purpose relations. The reason for this,
in the case of resemblance, is that although it is possible to compare the
shape of a tulip to that of a bell (bell tulip) or vice versa (tulip bell), the
referring item in this role combination automatically assumes the role COM-
PARED. In the case of the purpose compounds, I found that they could be
assigned two links and two role combinations: tablecloth: "cloth (OBlIINS)
which is for putting on tables (LOC/GOAL)"; card table: "table (LOc/INS) which
is for playing cards (OBI/GOAL) at"; night-dress: "dress (OBI/INS) which is for
wearing at night (TIME/GOAL)"; dinnertime: "time (TIME/INS) which is for having
dinner (OBI/GOAL) at"; ball bat: "bat (CAUSERlINS) which is for hitting balls
(OBI/GOAL) with"; football: "ball (OBI/INS) which is for hitting with foot
(CAUSERlGOAL)". As we see, the basic roles are reversible, but the superimposed
purpose role is not.
There is quite substantial empirical support for the existence of these links.
Ljung (1970), whose aim was to establish the meanings that denominal
adjectivalization could express, examined the definitions of 218 adjectives
ending in -al, -Jul, -ic, -ish, -ly, -ous or -y in Websters Seventh New Collegiate
Dictionary and Webster's Third New International Dictionary. He found it
possible to reduce the numerous definitions he had collected into a small
number of types (gross definitions), either because they were synonymous
or hyponymous. The present writer analysed 4,500 noun-noun compounds
in context (types) and 291 adjectives (types) in altogether 12,890 contexts
(Warren 1978, 1984) and found that, if there was a covert link,5 it would
with very few exceptions be one of those listed above.
An important discovery made by Ljung was that certain links ("meanings"
in his terminology) are incompatible with bona fide adjectives, i.e. adjectives
that could occur attributively (sad girl), predicatively (girl is sad), which
could be graded (very sad girl), and which could tolerate insertions (sad,
anxious girl). These links could, however, be found with modifiers which
had the morphological make-up of adjectives, but which in all other respects
were nouns: glandular pain, sulphuric acid, continental capital. These con-
structions he called N+aff N (noun+affix noun) compounds6 and (obviously)
he considered them types of compounds. He therefore suggested that certain
links were restricted to bona fide adjectives and others to compounds, whether
of the noun-noun or of N+aff N type. My own research into the semantics
of nominal modifiers did not completely confirm this surmise. Instead I found
that all the links that could occur in adjectives - also those compatible with
bona fide adjectives - could occur in noun-noun compounds. The restriction
 Links and qualia in modifier-head constructions 239

was that not all links found in compounds could occur with bona fide
adjectives, although they could be found in the so-called N+aff adjectives.
Consider Table 3.

TabLe 3. Links which do not occur in bona fide adjectives.

Link Example Paraphrase

constituted by silver spoon "spoon made of silver"

eLectric current "current consisting of electricity"
?smoky rings "rings consisting of smoke"
belonging to family estate "estate belonging to a family"
presidentiaL home "home belonging to the president"
?flowery leaves "leaves belonging to flowers"
occurring in/at/on city streets "streets in the city"
suburban areas "areas in the suburb"
?beachy path "path along the beach"
afternoon tea "tea served in the afternoon"
nocturnaL events "events occurring in the night"
?summery holidays "holidays in the summer"
caused by hay fever "fever caused by hay"
electric shock "shock caused by electricity"
?dusty allergy attack "allergy attack caused by dust"
being for emergency telephone "telephone for emergencies"
educationaL facilities "facilities for education"
?womanLy doctor "doctor for women"

I also found that the function of the modifier was a crucial factor as far as
stress, syntactic behaviour, morphology and the semantics of the modifier
are concerned. This was so in particular if the modifier was a classifier; in
which case it combines with the head to form a referring unit, or if it was
purely descriptive, in which case the modifier does not affect the meaning or
reference of the head but contributes additional information about it (i.e.
strictly speaking about the referent(s) of the head). Bolinger (1967) referred
to this distinction as reference-modifying and referent-modifying, respec-
tively. Referent-modifying modifiers, I found, are invariably bona fide
adjectives, which - as just indicated - could only occur with certain links,
viz. those in Table 4.
240 Beatrice Warren

Table 4. Links in bona fide adjectives.

Link Example Role combination

constituting problematic case RESULT-SOURCE

causing harmful drug RESULT-CAUSER

having (possessive) rational man PART-WHOLE

having on (Iocative) dusty furniture OBJ-PLACE

resembling golden orange COMPARANT-COMPARED

in accordance with alphabetic order NORM-ADHERENT7

To a certain extent modifiers could be said to have a basic job description:

nouns in the genitive are basically identifiers; uninflected noun modifiers
(i.e. the modifier in noun-noun compounds) are basically restrictive (i.e.
either identifiers or classifiers); non-derived adjectives and adjectives in -y,
-Jul, -ish and ?-ly are basically descriptors (bona fide adjectives); and adjec-
tives in classical suffixes such as -al, -an, -ar, -ary, -ic, and -ous are basically
classifiers or descriptors. 8 As we can see, certain functions can be connected
to more than one type of modifier; notably uninflected nouns and adjectives
in -al, -an, -ar, -ary, -ic, -ous etc. both have a basic potential classifying
function, and the descriptive function can in principle be connected with
both types of adjectives (i.e. the "native" and the "classical" types). From
this it follows that there should not be a difference as to function between
modifiers in noun-noun compounds and in combinations of -al, -an, and -ic
adjectives plus nouns, both being basically classifiers. The examples in
Table 5 appear to demonstrate this.
However, there should be a functional difference between the modifier in
noun-noun compounds (basically nondescriptive) and in combinations
involving -y, -ful and -ish adjectives (basically descriptive), which indeed
seems to be the case, as illustrated in Table 6.

Table 5. Synonymous noun-noun and adjective-noun combinations.

Link Noun-noun Adjective-noun

consisting of metal ball metallic ball

part of moon surface lunar surface
be in/at/on coast road coastal road
caused by hand milking manual milking
 Links and quaJia in modifier-head constructions 241

Table 6. Non-synonymous noun-noun and adjective-noun combinations.

Link Noun-noun Adjective-noun

having beauty spot beautiful spot

glamour girl glamorous girl
pepper sauce peppery sauce
constituting problem child problematic child
surprise party surprising party
resembling goldfish golden fish
silver paper silvery paper

As I think the examples just given show, the meaning of the whole phrase
is influenced by the function of the modifier. I will expand on this point
later.
Another consequence of the distinction between basically bona fide adjec-
tives (i.e. non-derived adjectives and those in -y, -Jul, -ish etc.) and the
potential N+aff adjectives (i.e. those in classical suffixes) is that the latter-
being compatible with a greater number of links - should be polysemous to
a greater extent than the former. This appears to be true, at least in principle.
Consider Table 7, which deals with musical.
However, it should be pointed out in this connection that not all the -ai,
-an, -ar, -ic etc. adjectives occur with all the links in principle available for
classifiers. Nor can all bona fide adjectives occur with all the links in principle

Table 7. The polysemy of an adjective with a classical suffix.

Link Example Role combination

constituting musical tones SOURCE-RESULT

constituted by musical composition RESULT-SOURCE
having (aptitude for) musical child PART-WHOLE
containing musical comedy OBJ-ABSTRACT PLACE
during which ... occurs musical soiree OBJ-TIME
belonging to ?musical pitch WHOLE-PART
occurring within (the sphere of) musical term ABSTRACT PLACE-DBJ
causing ?musical drumming RESULT-SOURCE
caused by musical shock SOURCE-RESULT
be for musical instrument GOAL-INSTRUMENT
resembling musical voice COMPARANT-COMPARED
242 Beatrice Warren

available for these adjectives. There are of course natural notional restrictions:
the referential content of dusty could hardly take on the role NORM, musical
or nervous are hardly compatible with the role TIME, etc., but besides such
natural limitations (which would be taken care of by qualia), there are numerous
restrictions which must be characterized as idiosyncratic. For instance, sad
is compatible with the PART role (sad child) and with the RESULT role (sad
news "causing sadness"), but hungry is compatible only with the PART role
(hungry child), not with the RESULT role (?hungry smells "causing hunger");
nervous can occur with the PART role (my nervous sister) and the CAUSER role
(nervous breakdown), but not with the RESULT role (?nervous news "causing
nervousness"); sensational is - contrary to expectation - incompatible with
the experience link (?sensational being "having sensations").

3. Idiosyncrasy, conventionalization and compositionality

I have elsewhere (Warren 1988: 131-132) suggested that idiosyncrasies of this

kind tend to develop to avoid cumbersome ambiguity and that a systematic
investigation of restrictions may reveal two main types: (i) of some notionally
possible links, only one will be favoured in connection with certain nouns;
sad, e.g., strongly favours the link experiencing when modifying animate
nouns, but may suggest inducing, giving rise to when combined with non-
animates; (ii) the adjective is restricted to one link only, which in turn may
give rise to "doublets" of, for instance, the kind listed in Table 8.
The fact that there are these idiosyncrasies has certain theoretical implic-
ations. For instance, we cannot plausibly assume that links of adjectives are
computed according to context, their referential content, and heads, every
time we come across them. Instead it seems that links in the case of adjectives
often become conventionalized parts of their meanings. We find here a

Table 8. Examples of "doublets".

Experiencing (manifesting)9 Causing

compassionate pathetic
envious enviable
hungry appetizing
contemptuous contemptible
furious infuriating
 Links and qualia in modifier-head constructions 243

difference between adjectives and modifying nouns. Links do not become

conventionalized parts of the meaning of the modifiers in noun-noun com-
pounds or in genitive constructions, blocking others from occurring in other
combinations. Links in these cases are part of the whole phrase, not of a
particular part of it, and they are typically not listed meanings, to use Pustej-
ovsky's terminology. What could be listed - and often is in the case of
compounds - is the whole construction.
One of the merits of Pustejovsky's model is that it highlights the existence
of semantic elements which are not likely to be memorized. Easy, for example,
by activating different quale roles would predictably suggest "read" when
applied to textbook, "accomplish" when applied to job, "solve" when applied
to problem, etc. None of these elements are likely to be stored. However, the
above suggests that a clear distinction between nonlistable and potentially
listable meanings would improve the model.
Pustejovsky seems to presuppose that meanings of phrases and sentences
are compositional, possibly in the belief that no other position would be
compatible with an approach the aim of which is to specify an infinite number
of possible senses in context. In my experience, in the case of modifier-head
constructions in which the modifier is a classifier, the meaning of the phrase
is very frequently not compositional. This is a well-established observation
in the case of compounds: blackboard does not refer to any board that is
black; frogman does not denote any man that happens to be like a frog.
Consider also expressions such as black hole, compact disc, right angles,
etc. It is quite obvious that retrieving a link - if that is indeed what we do -
does not necessarily amount to retrieving a definition of a modifier-head
construction functioning as a referring unit. The source of the definition lies
in the character of the referent. Working out the coiner's rationale for com-
bining, say, dirt and road is one task for the interpreter; working out the
denotatum of dirt road is another task. The accomplishment of the latter is
what matters. It is perhaps possible to make Pustejovsky's model perform
this two-stepped task. The qualia of dirt and road would produce the link;
the class of referents of dirt road would supply this compound with its
particular qualia. This would, however, mean allowing for the importance of
the function of lexical items and - above all - for the influence of contextual
referents on definitions. The above also demonstrates that semantic regularity
does not presuppose compositionality, as is often assumed (on this, see also
Warren 2001).
244 Beatrice Warren

4. Referential metonymy

We have so far considered adjective-noun and noun-noun combinations and,

in passing, genitive constructions. III There is, however, another construction
which in my view could be considered to be a type of modifier-head construc-
tion, viz. referential metonymy. In these constructions we have not only an
implicit link, but also an implicit head, as shown in Table 9.

Table 9. Metonymic patterns.

Implicit head Implicit link Explicit modifying Referent

noun

that which consists of silver (cutlery)

that which has beauty (beautiful lady/thing)
that which is part of eye (iris, as in blue eyes)
that which is in/at/on pocket (money, as in out a/pocket)
place where ... walkabout (road, Australian English)
occurs
that which causes ecstasy (drug)
that which is caused by hand (aid/applause)
that which is/represents Ann (picture of Ann)11

There is not as much empirical evidence that metonymies would adhere

to the above pattern as in the case of adjective-noun and noun-noun com-
binations. (I have at present around 100 randomly collected examples on
which I base the above [see also Warren 1992: 64-73].) Nevertheless there
are obviously very strong (and exciting) indications that whenever we have
to work out the relation between two nouns, there is a common array of role
combinations that we tend to choose from, although not all types of modifiers
make use of all types of roles. If this is indeed correct, we have revealed the
source of a fairly common type of polysemy. However, I bring up referential
metonymy not only to strengthen my hypothesis concerning the basic semantic
regularity in modifier-head constructions, but also because examples of
referential metonymy very clearly demonstrate that we should distinguish
between the "raw" semantic structure of modifier-head constructions and
their definitions. The nature of the definition ultimately depends on the nature
 Links and qualia in modifier-head constructions 245

of referents. Hand used metonymically, and involving a producing link, need

not mean "aid". It can also mean "applause", for example. The semantic
structure merely constrains reference.

s. Evaluation

Let us finally examine whether the studies of relational elements in modifier-

head constructions can be related to Pustejovsky's model and possibly support
it. We find then that compositional and possessive links correspond well to
CONSTITUTIVE qualia roles, that causative and purpose relations correspond
equally well to the AGENTIVE and the TELIC qualia respectively. Locative
relations would be a type of FORMAL qualia, which include according to
Pustejovsky (1996: 85-86): (i) orientation, (ii) magnitude, (iii) shape, (iv)
dimensionality, (v) colour, (vi) position. See Table 10.

Table 10. Matching links and qualia.

Links Qualia

compositional and possessive ~ CONSTlTIlnVE

locative ~ FORMAL
causative ~ AGENTIVE
purpose ~ TELlC

Admittedly, the FORMAL qualia seem poorly represented among the links
since only one of the features specified by Pustejovsky is included. It should
be remembered, however, that the other features would be involved in many
PARr-WHOLE combinations: tall building "building having great vertical length" ,
small house "house having little size", etc. The resemblance relation, however,
truly lacks a matching quale. A moment's reflection will establish that this is
natural. Resemblance arises because some feature or features of COMPARED -
one's intended referent - is reminiscent of some feature or features of COM-
PARANT - some other named entity: foothill "hill positioned like a foot",
frogman "man having shape and function (i.e. diving) of a frog", peanut
"nut growing (in pods) like peas", puppet government "government being
manipulatable like puppets". As we see, resemblance in these constructions
presupposes already-named entities which belong to distinct categories. It
246 Beatrice Warren

can therefore only be a relation, not a quale, which is what causes us to refer
some entities or phenomena to a single category. (It is possible to make use
of the resemblance relation to name something unnamed, in which case we
have a non literal metaphorical use of a word. That is to say, the COMPARED
(the intended referent) is implied and the COMPARANT is mentioned, assuming
- but only apparently - the role of a referring item.)
This brings us to the question of whether we are dealing with a set of
implicit links or latent qualia roles. Pustejovsky's position ought to be that
the qualia give rise to the links: combining silver with spoon will highlight
the FORMAL role of matter in silver and the AGENTlVE role of artefact in
spoon, producing "made of'; combining tea and spoon will highlight the
CONSTITUTIVE role of "liquid" in tea and the TELIC role of "for stirring" in
spoon, giving the link "for stirring"; combining apple with tree will
highlight the AGENTIVE role of "produced by tree" in apple, which then
naturally links apple and tree; combining seal with skin will highlight the
CONSTITUTIVE role of skin, i.e. part of animate, giving "skin of seal". Whether
qualia roles could always be appealed to in this manner is an open question.
What latent qualia in apple and cake are activated to give a locative have
relation in apple cake, and similarly, in victory and garden to give "for
promoting victory" in victory garden, or frog and man to produce "resembling"
in frogman?
Assuming that we accept the existence of both default links and qualia,
there seem to be the following possible hypotheses as to their relation: (i)
links and qualia just happen to coincide, but are in fact quite unrelated; (ii)
qualia are basic and give rise to links, which are not listed but completely
dependent on qualia to be activated; (iii) qualia give rise to implicit links,
but these, being frequently activated, become abstracted as grammatical types
of meanings, and also become memorized and no longer necessarily depen-
dent on qualia to be suggested; (iv) qualia and links are distinct because they
have distinct functions but have a common conceptual basis.
Of these, I find the last two equally plausible, the second somewhat less
plausible and the first implausible for the following reason: in view of the
fact that modifier-head constructions pick out a referent - something we
conceive of as a unit - just as single nouns do, the links and the qualia ought
to reveal what constellations of entities or phenomena we tend to recognize
as potentially unit-forming. Those which coincide in time and/or place (i.e.
the relations involving composition, possession and location) are natural
candidates. In the case of causal relations we seem to recognize that producer
 Links and qualia in modifier-head constructions 247

and product were once a unit. Consider moonlight, piano sound. We here
seem to have revealed the cognitive basis for perceiving distinct elements
which nevertheless are accepted as units. If this is indeed so, then links in
modifier-head constructions, which denote units, and qualia in single nouns,
which also denote units, ought to be of the same kind (with the natural
exception of resemblance links). The fact that they are strengthens this
surmise. It is also natural that we think of entities which are perceived as
units for the same reason as forming a category.
Links and qualia are similar in that they become activated when words -
in particular nouns - are combined. As we have seen, they are also strikingly
similar as to type. Indeed it is remarkable that a deductively constructed
model happens to agree so well with the results of empirical studies, the
existence of which the constructor of the model was probably quite unaware
of. Our conclusion must be that this agreement lends considerable support to
the theory of qualia.
It would, however, be a mistake to think of Pustejovsky's model as a
"finished product", as he himself points out. Above I have pointed out some
weaknesses, e.g. that it seems to be restricted to first-order entities and that
the fact that defining features may vary as to degree of salience is ignored.
The main omission of the model is, however, that it fails to allow for the
influence of contextual referents on word meaning. The fact is that the
interpreter does not expect contextual referents invariably to agree with stored
meanings and denotata. 12 There may be good enough matches, in which case
we have non-prototypical senses. It may be possible to include the contextual
referent in the conventional denotatum and yet other salient class-distinctive
features suggest themselves, in which case we have the non-compositional
meanings I have repeatedly exemplified above, i.e. gold fish, glamour girl,
beauty spot, black board, compact disc, etc. It may be impossible to include
the contextual referent in the conventional denotatum, in which case we have
different types of nonliteral meanings. New meanings arise not only when
words are combined, but above all when words are matched with referents.
As we have seen, non-prototypical meanings, literal non-compositional
meanings and most types of nonliteral meanings,l) i.e. meanings which are
truly polysemy-forming and invigorating factors in lexical semantics, are as
yet not accounted for in Pustejovsky's model. Nevertheless, since Pustejovsky's
model has opened our eyes to a possible basis for the semantic regularity we
find in modifier-head constructions, to the distinction between listable and
nonlistable polysemy, and to the difference between polysemy which arises
248 Beatrice Warren

when words are combined and the polysemy which arises when words are
matched with contextual referents, it may very well turn out to be a substantial
push forward within semantics.

Notes

\. I have elsewhere (Warren 1992: 18-20) argued that it is a mistake to think of

words mainly as category labels and as invariably deriving from the formation
of categories. Not all features of word meaning are criterial, i.e. serve to identify
referents. There are also components of meaning, the basis of which is probably
not category formation, and which serve to indicate speaker evaluation, which
they can do because their application is free. Although semanticists acknowledge
the difference in terms of denotation and connotation, they nevertheless tend to
ignore all but criterial features in forming theories of word meaning. Qualia, too,
must be seen as representing criterial features.
2. Although all (content) words have qualia structures in Pustejovsky's system
(1996: 76), qualia must, I believe, be seen as primarily connected to first-order
entities and only indirectly to second- and third-order entities. (For definitions of
first-, second- and third-order entities, see Lyons 1977: 438-452.)
3. They also included negative elements as in blind ("not having sight"), bare
("having no cover") and quantifying elements as in hot ("having much heat"),
brief ("having little duration"). Negative elements are obviously not part of all
adjectives, but if they are, they always are. Positive quantifying elements are
often not stable parts of meaning, but can appear or disappear as required by
context. Consider: the wall is tall (tall contains a quantifying element) and the
wall is half a yard tall (tall has no such element).
4. Downing (1977: 810-842) maintains that "the semantic relationships that hold
between the members of these compounds (i.e. noun-noun compounds) cannot
be characterised by a finite list of 'appropriate compounding relationships'"
(p. 810). She makes this claim in spite of the fact that - on her own admission -
only a small minority of the compounds she had collected and analysed did not
fit any of the established compounding relationships. Her justification for rejecting
a finite list is that virtually any relation is possible in the right context. This is
correct. It is probably possible to think of a context in which apple tree could
refer to the tree that people have thrown apples at. However, there is a price for
rejecting a set of established relations, viz. the counter-intuitive claim that out of
context, if encountered for the first time, apple tree would not suggest any
compounding relationship, at least not any particular one. The introduction of
default values saves the day. It allows us to recognize semantic regularity, even
if it is not absolute, and to account for the fact that certain implicit meanings
suggest themselves naturally, whereas others need contextual support. Personally,
 Links and quafia in modifier-head constructions 249

I see default values as the outcome of two tendencies in human language, striving
in opposite directions: constraints and rules on the one hand, which facilitate
interpretation, and flexibility on the other, which promotes expressiveness.
5. Far from all adjective-noun combinations contain a covert link. For instance,
there is no implicit predicate in nominalizations such as national leader ("some-
body who leads nations") or presidential appointees ("people whom the president
appoints"). In some adjectives functioning as adverbials, there is no link at all:
perfectfool ("somebody who is perfectly foolish") (see Paradis 2(00). (Pustejovsky
accounts for these by means of selective binding [1996: 127-131].) Also, there
are quite a number of adjectives which have conventionalized implied or fig-
urative meanings. Familiar, e.g., can no longer mean "consisting oflbelonging
t%ccurring in etc. family".
6. The term was originally coined by Lees (1960: 181).
7. That bona fide adjectives are restricted to these links is an intriguing discovery
for which I have no good explanation.
8. Basically is a necessary qualification in this context, since it is possible to classify
by describing (short story, black coffee, white people) or by identifying (writer's
cramp, clown s attire), or identify by describing (bring me the red box), etc.
9. Experiencing is a type of possessive link which requires that the referential content
of the modifier denotes some sensation and that the head denotes an animate.
Manifesting is dependent on the experience link and requires that the referent of
the head is thought of as a natural mediator of sensations: (sad) eyes, (angry)
face, (nervous) hands, (happy) smile, (furious) letter, (bitter) comments. If this
restriction is violated, we have what is sometimes referred to as transferred
epithets or the Wodehouse effect, e.g. a sad cigarette, a contemplative lump of
sugar.
An alternative account of combinations of this kind sometimes suggested is
to look upon them as metonymic: happy face "the possessor of the face experi-
ences happiness"; angry letter "the author of the letter experiences anger".
10. For reasons of space I have refrained from discussing the semantics of genitive
constructions. I will be content with the comment that the same type of default
relations appear to be involved. Consider loan s picture "picture that Joan owns"
(possessive link), "that Joan made" (causal link), "that represents Joan" (com-
positional link).
11. For Pustejovsky's account of referential metonymy, see Pustejovsky (1996:
90-95).
12. Compare NerIich and Clarke (200l), who point out that words keep recharging
their semantic batteries because there always is and always will be a discrepancy,
a fundamental incongruence, between the supply of words and our communicative
demands.
13. Compare Kilgarriff (to appear), who also found that Pustejovsky's model failed
to account for metaphors.
250 Beatrice Warren

References

Aarts. Ian and Ioseph Calbert

1979 Metaphor and Non-metaphor. The Semantics ofAdjective-Noun Com-
binations. Tiibingen: Max Niemeyer Verlag.
Bartning. Inge
1980 Remarques sur la syntaxe et la semantique de pseudo-adjectifs
denominaux en franfais. Stockholm: Almqvist and Wiksell
International.
Bolinger. Dwight
1967 Adjectives in English: attribution and predication. Lingua 18: 1-34.
Brekle. Herbert E.
1970 Generative Satzsemantik und transformationelle Syntax im System
der englischen Nominalkomposition. Miinchen: Wilhelm Pink.
Clark. Eve and Herbert Clark
1979 When nouns surface as verbs. Language 55(4): 767-811.
Downing, Pamela
1977 On the creation and use of English compound nouns. Language 53(4):
810--842.
Geeraerts. Dirk
1997 Diachronic Prototype Semantics. Oxford: Clarendon Press.
Kilgarriff, Adam
(to appear) Generative lexicon meets corpus data: the case of non-standard word
uses. Talk presented at the Utrecht Congress on Storage and Compu-
tation in linguistics, 19-21 October 1998.
Lees, Robert
1960 The Grammar of English Nominalizations. The Hague: Mouton.
Leitzky. Eva
1989 (De)nominale Adjektive im heutigen Englisch. Tiibingen: Max
Niemeyer Verlag.
Levi, Iudith
1978 The Syntax and Semantics of Complex Nominals. New York, San
Francisco, London: Academic Press.
Ljung, Magnus
1970 English Denominal Adjectives. Gothenburg Studies in English 21.
Lyons,John
1977 Semantics. Cambridge: Cambridge University Press.
Nerlich, Brigitte and David Clarke
200 1 Ambiguities we live by. Towards a pragmatics of polysemy. Journal
of Pragmatics 33: 1-20.
Paradis. Carita
2000 Reinforcing adjectives: a cognitive semantic perspective on gram-
 Links and qualia in modifier-head constructions 251

maticalization. In: Ricardo Bermudez-Otero, David Denison, Richard

M. Hogg and Chris B. McCully (eds.), Generative Theory and
Corpus Studies, 233-258. Berlin, New York: Mouton de Gruyter.
Pustejovsky, James
1996 The Generative Lexicon. Cambridge and London: The MIT Press.
Warren, Beatrice
1978 Semantic Patterns of Noun-Noun Compounds. Gothenburg Studies
in English 4l.
1984 Classifying Adjectives. Gothenburg Studies in English 56.
1985 Making sense of making senses. In: Sven Backman and Goran Kjellmer
(eds.), Papers on Language and Literature, 372-384. Gothenburg
Studies in English 60.
1988 Ambiguity and vagueness in adjectives. Studia Linguistica 42(2):
122-172.
1992 Sense Developments. Stockholm: Almqvist and Wiksell International.
200 I Accounting for compositionality. In: Karin Aijrner (ed.), A Wealth of
English, 103-114. Gothenburg Studies in English 81.
Polysemy and bleaching

Jean Aitchison and Diana M. Lewis

1. The problem

Fading meaning is a commonly mentioned type of semantic change, a process

known variously as bleaching, desemanticization, weakening, depletion,
distortion, even verbicide. The term "bleaching" is perhaps the most common,
and may have originated with the German neogrammarian Georg von der
Gabelentz, who noted that forms 'grow pale' (verblassen) and their colours
'bleach' (verbleichen) (Gabelentz 1891: 242). This paper argues that bleaching
is more usefully regarded as a type of polysemy, and explores the bleaching!
polysemy process.
In the past, many writers have been more concerned to censure bleaching
than to study it, perhaps subconsciously echoing a 19th-century viewpoint
that semantic change is undesirable:

This tendency of words to lose the sharp, rigidly defined outline of meaning
which they once possessed, to become of wide, vague, loose application
instead of fixed, definite, and precise, to mean almost anything, and so really
to mean nothing, is ... one of those tendencies, and among the most fatally
effectual, which are at work for the final ruin of a language, and, I do not fear
to add, for the demoralization of those that speak it (Trench 1856: 192).

According to Trench: "The causes which bring this mischief about are
not hard to trace": words which get into general use are caught up "by those
who understand imperfectly and thus incorrectly their true value". Con-
sequently, words "become weaker, shallower, more indefinite; till in the end,
as exponents of thought and feeling, they cease to be of any service at all"
(1856: 193).
This disapproval has continued. In a guide for radio published in Britain
in 1981 by the editor of the Oxford English Dictionary, for example, broad-
casters were advised in a section labelled "Inflation or modishness of diction"
to avoid the use ofthe word tragedy in sport by rephrasing (Burchfield 1981).
Similarly, the author of a book on the social history of English vocabulary
254 Jean Aitchison and Diana M. Lewis

suggests that "Much verbicide seems to be an upper-class affectation"

(Hughes 1988: 14).
These writers have apparently not taken into consideration the fact that
the old, strong meaning often still exists alongside the newer, "weakened"
meaning. We are therefore dealing with an important type of polysemy. From
this viewpoint, the changes are not so much weakening as expansion, the
development of multiple meanings: "meanings expand their range through
the development of various polysemies ... these polysemies may be regarded
as quite fine-grained. It is only collectively that they may seem like weakening
of meaning" (Hopper and Traugott 1993: 100). Bleaching therefore must be
regarded not only as a diachronic process, but also as a synchronic phenom-
enon, a type of polysemy which needs to be explored more fully.
Serious work on polysemy has recently escalated (e.g. Aitchison 2003;
Aitchison and Lewis 1996; Geeraerts 1993; Geeraerts, Grondalaers and
Bakema 1994; Gibbs 1994; Nerlich and Clarke 1997; Pustejovsky 1995;
Pustejovsky and Boguraev 1996; Taylor 1995; Tuggy 1993), yet the facets
investigated so far have been limited. In particular, the relationship between
diachrony and synchrony has been sparsely explored, though with some
notable exceptions (e.g. Sweetser 1988, 1990).
Bleaching is the semantic side of "grammaticalization" or "grammati-
cization", a term coined by Antoine Meillet, who defined it as "the attribution
of a grammatical character to a previously autonomous word" (Meillet 1948:
131). In cases of "full" grammaticalization, words are demoted both semanti-
cally and syntactically (e.g. Bybee, Perkins and Pagliuca 1994; Heine, Claudi
and Htinnemeyer 1991; Hopper and Traugott 1993; Traugott and Heine 1991).
Typical examples are the English modals can and will, which were once
main verbs meaning 'know' and 'want'. Full grammaticalization, then, is the
final stage of a much wider phenomenon, which may affect a lexical item
only partially, as when the meaning alone of a word is involved. The indepen-
dence of semantic and syntactic aspects of grammaticalization has been pointed
out by several linguists (e.g. Fischer 1997; Lightfoot 1979; Scott 1996).
Relatively little is known about the detailed stages of bleaching. Sweetser
(1988, 1990) concentrated more on the overall direction of semantic change:
she was particularly concerned with the move of modals from a root sense to
an epistemic sense, for example. This paper therefore attempts to shed light
on the "nitty-gritty" of the bleaching process.
 Polysemy and bleaching 255

2. Disaster nouns

Certain types of words are known to bleachlbecome polysemous fast, notably

words relating to dire events, such as catastrophe, disaster, tragedy: What a
disaster! My new hat got wet; pejorative adjectives, such as aWful, bad,
dreadful, terrible: It was a terrible evening - Dulde wouldn't stop talking;
and "violent" verbs such as devastate, murder: I could murder a sandwich.
English contains a number of nouns for a disastrous event: calamity,
catastrophe, disaster, tragedy. Dictionaries mostly agree that these words
describe serious misfortunes as their primary meaning, and provide over-
lapping definitions for them, as in the The New Oxford Dictionary of English
(Oxford University Press 1998):

calamity, noun, an event causing great and often sudden damage or distress;
a disaster.
catastrophe, noun, an event causing great and often sudden damage or
suffering; a disaster.
disaster, noun, a sudden event such as an accident or a natural catastrophe
that causes great damage or loss of life ...
tragedy, noun, an event causing great suffering, destruction, and distress,
such as a serious accident, crime or natural catastrophe.

Of these words, only the entry for disaster includes the information that
it may be used in less dire circumstances: "an event or fact that has unfor-
tunate consequences ... informal, a person, act or thing that is a failure".
Yet investigation of the British National Corpus (BNC), a database which
includes both spoken and written language, suggests that informal, humorous
or figurative uses of all these words are extensive, alongside their more
serious, older meanings. (The BNC contains approximately 100 million
"w-units", roughly, lexical items, of which approximately 90 million are
written, and around 10 million spoken.) Examples of (a) serious, and (b)
trivial usages are given below:

( I) a. This will make it much less likely that the entire human race will
be wiped out by a calamity such as a nuclear war.
b. But what if ... you're out on the floor at some gay club, and you
spot or (calamity) are spotted by some workmates ... what then?
(2) a. A large comet hitting the earth would mean catastrophe.
b. To fall in love with Alexander would be a catastrophe.
256 Jean Aitchison and Diana M. Lewis

(3) a. ... the Hillsborough football disaster which killed 95 people.

b. To get a panama hat wet is to court disaster. The hat becomes
limp and shapeless.
(4) a. 259 passengers and crew ... were killed by a bomb. This was
Britain s worst air tragedy.
b. The great tragedy of modern music is that ... the results are less
and less significant from a human point of view.

Of these similar-meaning words, disaster is by far the most frequent in

the BNC (Table 1). It can therefore be regarded as the "prime" disaster word
in English at the current time, and likely to shed useful light on the polysemy
process.

Table J. Disaster nouns: frequency per million w-units.

BNC Spoken BNCWritten

Calamity 0.6 1.5

Catastrophe I.1 5.0
Disaster 22.3 32.1
Tragedy 6.3 19.3

As noted above, the dictionary entry labels non-serious uses of disaster

as "informal", but does not give any indication of the relative quantities of
the formal and informal usages, nor does it further characterize the informal
ones. The polysemy question is therefore investigated in this paper by con-
sidering:

(i) The scale of the disaster: what is the proportion of serious "real"
disasters, to trivial inconveniences?
(ii) The interpretation problem: how do speakers/hearers distinguish
between the various meanings of the word?
(iii) The development of polysemy: to what extent can this process be
successfully characterized?
 Polysemy and bleaching 257

3. Scale of disaster

Disaster is widely used in both spoken and written language: the BNC
contains over 3,000 examples. Of these, all the spoken ones were analysed
for this paper, as well as 600 randomly selected written ones. Removal of
unclear occurrences left 185 spoken and 589 written examples.
The proportion of "real" disasters to trivial inconveniences was assessed
by dividing the BNC examples of disaster into serious (S), medium (M) and
trivial (T) events. As a rough rule of thumb, an event which caused multiple
deaths was classed as (S); one which involved one or two deaths, inflicted
severe suffering on a small number of people, or caused environmental
damage was listed as (M); one which caused social inconvenience was classed
as (T). For example:

(5) At least 62 people were killed and 3,000 missing last night after an
underwater earthquake sent 50ft tidal waves crashing into the coast
of Nicaragua. More than 227 people were injured in the disaster. (S)
(6) But if all 22 million gallons escape, the disaster will be twice as bad
as the 1988 Exxon Valdez spill off Alaska. (M)
(7) All other efforts to lose the fat from the offending areas proved to be
a disaster. If I lost weight below 54 kg my bust disappeared, yet nothing
went from my legs or posterior! (T)

Table 2 shows how instances of disaster fell into the (S), (M) and (T)
categories.

Table 2. Disaster categories.

Serious Medium Trivial

Spoken n= 185 92 (50%) 27 (14%) 66 (36%)
Written n=589 255 (43%) 122 (21 %) 212 (36%)

Superficially, Table 2 suggests that serious disasters outnumber trivial

ones in both the spoken and written corpus. But these figures partly flatten
out the data, in that an imbalance occurred between singulars and plurals:
the singular examples tended to be more trivial than the plural ones, especially
in the spoken corpus, where the ratio of trivial to serious disasters was 1.2: 1
258 Jean Aitchison and Diana M. Lewis

in the singular, but only 0.1: I in the plural. This imbalance was less noticeable
in the written corpus, where the ratio of trivial to serious disasters was 1: 1 in
the singular, and 0.4: 1 in the plural.
Categorization into (S), (M) and (T) was a useful step, but proved to be
oversimple: the (S), (M), (T) categorizations sometimes overlapped, in that
the same event could seriously affect a huge number of people in one area,
by definition (S), yet only a single person in some other area, by definition
(M). We therefore subdivided the disasters into types which partly cut across
the (S), (M), (T) categories.
The largest single group involved wars or serious accidents, which killed
a large number of people. These were categorized as (S), as in (3a) and (5).
Environmental disasters were mostly classed as (M), as in (6).
Social, political and economic disasters were split between the (M) and
(T) classifications:

(8) If the tests prove positive, the flock is slaughtered: "For many farmers
it spells financial disaster ... " the association s chairman said. (M)
(9) The Tories were heading for disaster if they continued to delude
themselves that only a little fine tuning of presentation was required
to secure a fourth consecutive general election victory. (T)

A subdivision of economic disasters involved the technical phrase "disaster

recovery", a growth industry whose practitioners solve computer problems
after a system breakdown. These were classed as (M):

(10) Comprehensive, tried and tested disaster recovery procedures can

make the difference between getting vital computer systems up and
running again within hours or days, and going out of business.

Personal disasters were (M) when one or few people died or were injured:

( 11) Then disaster struck. The first two men to exit from the following party
were killed when their parachutes failed to open.

But the majority of personal disasters were quite trivial (T), covering
household inconveniences, sports losses, or minor social difficulties, as in
(3b) and (7), and also (12), (13) and (14):
 Polysemy and bleaching 259

(12) The only reason I'm running up these debts is that now I've got so
little capital left, that I've got to keep the capital for sheer disasters
Like the boiler.
(13) The last wicket fell ... So it was another bLackwash, another disaster
for EngLand.
(14) There have been many disasters along the road, Yorkshire puddings
you couLd sole your shoes with ... and last Christmas a chocolate log
that disintegrated, the proud little Santa on top sinking without trace
in a sea of chocolate gunge.

The breakdown into the various categories is shown in Table 3.

Table 3. Disaster by (S), (M), (T) and by type.

Serious Medium Trivial Total

War/Accident 342 4 346 (45%)

Environmental 3 28 31 (4%)
SoclPollEcon 2 55 84 141 (18%)
Personal 62 194 256 (33%)
Total 347 (45%) 149 (19%) 278 (36%) 774

4. Interpretation problem

How, then, do speakers use these words appropriately, and how do hearers
interpret them in the way intended? Broadly speaking, the surrounding context
clarified the level of disaster. Partly, this was explicitly specified, and partly,
covert conventions operated, which are implicitly understood by English
speakers.
The naming of a geographical location was the major clue that a serious
incident involving multiple deaths had taken place, sometimes (though not
necessarily) accompanied by further explanation of the type of disaster, as
in: the BradfordfootbaLl disaster, the 1986 Chernobyl nuclear disaster, the
Clapham disaster, the Clapham rail disaster, the HilLsborough disaster, the
Hillsborough football crowd disaster, the Kegworth air disaster, the Lockerbie
disaster, the Siberian pipeline disaster, the Stalingrad disaster, the Zeebrugge
ferry disaster.
For lesser events, the type of problem was often appended, as: ecological
260 Jean Aitchison and Diana M. Lewis

disaster, economic disaster, electrical disaster, environmental disaster,financial

disaster, the Great Onion disaster, industrial disaster, the poll tax disaster,
political disaster.
For trivial incidents, the cause of the problem tended to be specified, as
m:

(15) The gravys a disaster. Its got too much fat in it.
(16) Newspapers love a good disaster, and a wedding where the bride-
grooms trousers fell down at the altar would stand afar better chance
of being reported than one that went without a hitch.

Amidst this plethora of trivia, several topics recurred, especially cookery,

as in (14), (15) and (17):

(17) My own efforts at making custard ... may have been forgotten by those
who subsequently had to eat the awful stuff, but I am not likely to
forget the watery culinary disaster.

Sports losses were another common trivial disaster, as in (13), (18) and
(19):

(18) His spell in Italian football was a disaster and he eagerly accepted
the chance to join Manchester United for a record fee.
(19) X must get through to him in the first six rounds or face disaster.

The topics of food or sport therefore flagged that the disaster was a trivial
event. The origin of food disasters is unclear, though use of disaster in sport
has been attributed to Rudyard Kipling's If One sports report explained this:

(20) Behind many afootball club dressing room door, you '11 find a copy of
Rudyard Kipling's If pinned up, the poem that talks of "meeting
triumph and disaster and treating those two imposters just the same".

Nevertheless, it would be unwise to place too much emphasis on this, as

other disaster words are also used to describe sport, presumably because of
the difficulty of providing sparkling copy for events that are fundamentally
repetitive:
 Polysemy and bleaching 261

(21) In a history of disasters stretching across 30 years, Scotland has been

plagued by calamity, lapses in concentration and self-induced tragedy.
The goalkeeper is always to blame and always will be.

Food problems, however, were normally restricted to the word disaster,

perhaps on the grounds that a food tragedy or calamity might be misunder-
stood as a case of food poisoning.
Collocational clues also helped to signal whether an event was a genuine
disaster or a social hiccup. In particular. intensifiers often diminished the
seriousness of the disaster: an absolute disaster mostly indicated a fairly
trivial event, and so did a total disaster.

(22) The next morning was an absolute disaster. Loretta's hopes of a

conciliatory chat with Bridget over breakfast were dashed.
(23) England must pull itself together if a bitterly disappointing tour is
not to become a total disaster.

Similarly, disaster strikes and disaster struck often related to a trivial

event:

(24) Even if disaster strikes, as it seemed to for one student of mine who
dropped her nearly completed head on the concrete floor and an ear
snapped off [the wooden rocking horse she was carving], don't let it
worry you unduly. We simply glued the broken ear back in place.
(25) Disaster struck again for the home side after 57 minutes.

5. The development of polysemy

A prerequisite for the development of polysemy may be that the word must
be widely used. At first sight, this is a puzzle for disaster, in that serious
disasters are relatively rare. Yet the word disaster was frequently used,
primarily because many disasters discussed were potential, rather than actual.
Only a proportion (about halt) had happened, and the remainder were
impending, avoided, or hypothetical, as in:

(26) The research station's scientists comment privately that only a big
fire disaster will make the government look harder at fire research.
262 Jean Aitchison and Diana M. Lewis

(27) A second jet disaster was narrowly averted in Bogota on Thursday.

(28) When you rescue the old Christmas tree lights from the loft for the
umpteenth time, remember that they could be the cause of an electrical
disaster.

The potential, rather than actual, nature of many disasters is shown by

the linguistic expressions used with them, as in: avert, avoid, court, expect,
face,foretell, head for, head off, predict, prevent, save from, warn of, imminent,
impending, near, potential, brink of, chance of, doomed to, expectation of,
fear of, recipe for; can be, could be, could have been, would be, would have
been.
The main characteristic of these hypothetical expressions was their diversity:
in the written corpus, only can, could + verb, as in (28), and will, would +
verb, as in (26), occurred more than ten times. The hypothetical nature of so
many disasters, then, promotes bleaching, in that the word is used often in
cases where no serious event has in fact occurred. Once a word is widely
used in this way, bleaching/polysemy will escalate.
The word disaster therefore provides a blueprint for the birth of polysemy,
and the characteristics noted here are likely to be found elsewhere.
First, words bleach at different rates in spoken speech and written language:
spoken speech provides more bleached examples, as with disaster, and is
normally ahead in the polysemy process.
Second, bleaching moves through a lexical item unevenly, in that different
forms of the same word do not necessarily run parallel. Disaster (singular)
was more bleached than disaster (plural).
Third, bleaching typically begins in particular topic areas. In the case of
disaster, bleaching is especially frequent in sport (in common with other
disaster words), and also in cookery Ca feature of the word disaster, but not
tragedy).
Fourth, collocation is an important factor: certain combinations tend to
distinguish an event as serious or trivial. With disasters, an intensifier, as in
absolute disaster, is likely to signify a non-serious event, though each disaster
word has its own preferred intensifiers: an ?absolute tragedy or an ?absolute
calamity is not impossible, it is just less idiomatic. Such "freezes" (i.e. frozen
collocations) may provide the basis for a novel lexical item.
At first, "the persistence of older forms and meanings alongside newer
forms and meanings ... leads to an effect that can be called 'layering' or
'variability' at anyone synchronic moment in time" (Hopper and Traugott
 Polysemy and bleaching 263

1993: 123-124). But eventually, layering leads to full-blown polysemy, the

splitting of an original lexical item into more than one. This decision is
typically made by lexicographers. In the case of disaster, at least one dictionary
(NODE) has opined that the phrase disaster area should be a headword,
with its own independent lexical entry, away from its "mother" disaster:

disaster area, noun, an area in which a major disaster has recently occurred:
the vicinity of the explosion was declared a disaster area. n [in sing] informal,
a place, situation, person or activity regarded as chaotic, ineffectual, or failing
in some fundamental respect: the room was a disaster area. stuff piled every-
where I she was a disaster area in fake leopard skin and stacked heels.

6. Conclusion

This paper has pointed out that words signifying catastrophic events are
subject to bleaching, and consequently, the development of polysemy. It has
provided a case study of the word disaster, noting that polysemy develops
from layering, simultaneous different usages of the same word. Speech and
writing, different forms of the same lexeme, different topic areas, and habitual
collocations were explored and compared. These enabled the sources of the
bleaching of disaster to be narrowed down and partially pinpointed. Only a
fine-grained approach of this type can lead to a full understanding of the
process of polysemy.

References

Aitchison, Jean
2003 Words in the Mind: An Introduction to the Mental Lexicon. 3rd
Edition. Oxford: Blackwell.
Aitchison. Jean and Diana M. Lewis
1996 The mental word web: forging the links. In: Jan Svartvik (ed.), Wonts,
39-47. Stockholm: Swedish Academy.
Burchfield, Robert
1981 The Spoken Word: A BBC Guide. London: British Broadcasting
Corporation.
Bybee, Joan L., Revere D. Perkins and WiIliam Pagliuca
1994 The Evolution of Grammar: Tense. Aspect and Modality in the
Languages of the World. Chicago: University of Chicago Press.
264 Jean Aitchison and Diana M. Lewis

Fischer, Olga
1997 On the status of grammaticalisation and the diachronic dimension in
explanation. Transactions of the Philological Society 95(2): 149-187.
Gabelentz, Georg von der
1891 Die Sprachwissenschaft. Ihre Aufgaben, Methoden und bisherigen
Ergebnisse. Leipzig: Weigel.
Geeraerts, Dirk
1993 Vagueness's puzzles, polysemy's vagaries. Cognitive Linguistics 4(3):
223-272.
Geeraerts, Dirk, Stefan Grondalaers and Peter Bakema
1994 The Structure of Lexical Variation: Meaning, Naming and Context.
Berlin: Mouton de Gruyter.
Gibbs, Raymond W.
1994 The Poetics of Mind: Figurative Thought, Language, and Under-
standing. Cambridge: Cambridge University Press.
Heine, Bernd, Ulrike Claudi and Friederike Hiinnemeyer
1991 Grammaticalization: A Conceptual Framework. Chicago: University
of Chicago Press.
Hopper, Paul J. and Elizabeth Closs Traugott
1993 Grammaticalization. Cambridge: Cambridge University Press.
Hughes, Geoffrey
1988 Words in Time: A Social History of the English Vocabulary. Oxford:
Blackwell.
Lightfoot, David
1979 Principles of Diachronic Syntax. Cambridge: Cambridge University
Press.
Meillet, Antoine
1912 [1948] Linguistique historique et linguistique generale. Paris: Champion.
Nerlich, Brigitte and David D. Clarke
1997 Polysemy: patterns in meaning and patterns in history. Historiographia
Linguistica 24(3): 359-385.
Pustejovsky, James
1995 The Generative Lexicon. Cambridge, MA: MIT Press.
Pustejovsky, James and Bran Boguraev (eds.)
1996 Lexical Semantics: The Problem of Polysemy. Oxford: Clarendon.
Scott, Biljana
1996 Aspectogenesis and the categorisation of directionals in Chinese.
D.Phil. dissertation, University of Oxford.
Sweetser, Eve E.
1988 Grammaticalization and semantic bleaching. Proceedings of the 14th
Annual Meeting, Berkeley Linguistics Society: 389-404.
1990 From Etymology to Pragmatics: Metaphorical and Cultural Aspects
of Semantic Structure. Cambridge: Cambridge University Press.
 Polysemy and bleaching 265

Taylor, John R.
1995 Linguistic Categorization: Prototypes in Linguistics. 2nd Edition.
Oxford: Clarendon.
Traugou, Elizabeth Closs and Bernd Heine (eds.)
1991 Approaches to Grammaticalization. Volumes 1 and 2. Amsterdam:
John Benjamins,
Trench, Richard Chenevix
1856 English, Past and Present. 3rd Edition. London: John Parker and
Son.
Tuggy, David
1993 Ambiguity, polysemy, and vagueness. Cognitive Linguistics 4(3):
273-290.
Polysemy in the lexicon and in discourse

Andreas Blank

1. Introduction

A fundamental problem lexicologists - and indeed all speakers - have to

face is the fact that the number of words we know is limited but that human
imagination is virtually unlimited (Schlieben-Lange 1997: 242), and that the
number of things, beings, processes and ideas that can be referred to is
endless. There are two main strategies which allow us to cope with this
discrepancy: first, we match the concrete referent with a lexicalized meaning
of a word and actualize this meaning in the concrete context; second, if the
first strategy cannot be applied or risks failure because there is no lexicalized
meaning to cover the actual referent, we must create a lexical innovation,
e.g. a semantic innovation, a new word-formation, a new idiom, or we must
introduce a loan word (see Blank 200la).
In the first case, the contextual meaning is located inside the range of an
existing semantic invariant: a learned semantic rule is applied to a given
context. In this paper I will try to define the criteria for this type of contextual
variation (or "vagueness") and to distinguish it from different senses of one
word, i.e. polysemy, and from different words having an identical signifier,
i.e. homonymy (see Section 3). When we decide to use a word but transgress
the traditional range of its semantic invariants, we create a semantic inno-
vation. If this innovation is successful, it becomes, in turn, lexicalized as a
new invariant of the word in question. The types of polysemy resulting from
this diachronic process and the possible semantic relations linking them in
synchrony will be discussed in Section 2. Finally, Section 4 deals with very
regular types of polysemy which cannot be satisfactorily explained on the
basis of more traditional theories of polysemy. This will lead us to a new
view of polysemy as a complex multi-level phenomenon right at the cross-
roads where cognition, discourse, discourse rules and idiosyncratic lexical-
ization meet.
268 Andreas Blank

2. The synchrony of semantic change: types of polysemy

The term "polysemy" first appears in 1897 in Michel Breal's fundamental

Essai de sbnantique:

Le sens nouveau, quel qu'il soit, ne met pas fin a I'ancien. IIs existent tous
les deux I'un a cote de I'autre. Le meme terme peut s'employer tour a tour au
sens propre ou au sens metaphorique, au sens restreint ou au sens etendu, au
sens abstrait ou au sens concret ... A mesure qu'une signification nouvelle est
don ne au mot, il a I'air de se multiplier et de produire des exemplaires
nouveaux, semblables de forme, mais differents de valeur. Nous appelons ce
phenomene de multiplication la polysemie. (Breal 1897: 154-155)

It is important to note that according to Breal polysemy arises as a con-

sequence of semantic change; it is the "synchronic side" of lexical semantic
change. Breal distinguishes various kinds of polysemy, using the types of
semantic relation between the old sense and the new one as criteria (see also
Fritz 1998: 57-58). Usually, these issues are not given the importance they
merit (see, however, Sweetser 1990, but only with regard to metaphor) and
most handbooks still define polysemy as being based on the existence of a
semantic relation between lexicalized senses of a word without further explan-
ation of the nature of this relation. I
To understand the diachronic background of lexical polysemy and how a
semantic innovation can be related to a lexicalized sense, we need, first of
all, a typology of semantic change. Such a typology, which is entirely based
on associations between concepts, or concepts and linguistic signs, was
suggested in Blank (1997) and is displayed in the left column of Table 1.
This typology serves as a basis for a detailed description of types of polysemy
based on associative relations in synchrony, presented in the right column of
Table 1.
As one can see, there is no complete isomorphism between diachronic
processes and synchronic states. Let us now briefly discuss the associative
backgrounds of the types of semantic change and of their synchronic counter-
parts.
The best-known type of polysemy is metaphoric poly se my which derives
in most cases from metaphor as a diachronic process. Both are based on a
more or less salient similarity between two concepts that belong to different
or even distant conceptual domains.
Similarity inside one and the same conceptual domain or folk-taxonomy
 Polysemy in the lexicon and in discourse 269

is the basis of co-hyponymous transfer, giving rise to co-hyponymous poly-

se my. These transfers probably occur because the speaker's knowledge of
the referential limits of the concepts involved is temporarily or permanently
blurred. This type of polysemy is limited to the same dialect or register. It is
quite unstable and tends to fade away.2
Types 3 and 4 are also based on similarity of concepts within the same
domain, as in most cases one of the two concepts involved in the semantic
change was conceived as a prototypical instance of the whole category and
therefore as a cognitive reference-point. In synchrony, however, we tend to
focus on the concomitant taxonomic relation between the two senses, one
being the hyperonym of the other. In this case, the associative relation
underlying the semantic change shifts to the background, while the taxonomic
inclusion of referential classes becomes dominant. 3
Type 5, lexical ellipsis - or better, absorption - has two subtypes, depending
on which part of a given complex word absorbs the sense of this complex
word. Synchronically, however, absorption is identical either with the result
of semantic restriction, i.e. taxonomic polysemy, or with the synchronic effect
of metonymy. Thus, absorption as a diachronic process has no proper syn-
chronic counterpart.
This leads us to metonymy (including meronymy) and to its analogous
synchronic counterpart. Both are based on conceptual contiguity, i.e. the
typical and salient co-occurrence or succession of elements in frames or
scenarios or of these frames themselves. 4 The same synchronic result is
produced by semantic change through popular etymology (with very few
exceptions). Diachronically, however, popular etymology necessarily com-
bines conceptual contiguity with formal similarity.
Type 8 deals with the reciprocal connection of participants in a frame,
such as the HOST and the GUEST in the frame "RECEIVING GUESTS". When such
a converse relation develops within the same word, we call this auto-converse
change leading to auto-converse polysemy. Although this can be seen as a
classical instance of opposition (see Aristotle, Categories 10), this is rather a
special case of contiguity which one could also list under metonymy.
Opposite senses within one word develop in Types 9 and 10, too. Here,
however, the underlying association is contrast, either on the con notational
level (antiphrasis) or on the core content level, producing a kind of inner
antonymy. Both cases rarely become conventionalized and polysemy is often
"asymmetric" as it usually does not function in the same register.
An analogous type of semantic change is the imitation of an already
Table 1. !j
o
Types of lexical semantic change Synchronic relation when conventionalized
:t.
1. Metaphor A. Metaphoric polysemy 5.
~
E mouse 'small rodent' > 'computer device' E mouse 'small rodent', 'computer device' ~
It afferrare 'to grasp' > 'to understand' It afferrare 'to grasp', 'to understand' b:I
tr
L brevis 'short' (spatial) > 'short' (temporal) L brevis 'short' (spatial), 'short' (temporal) ;::
;.:-

2. Co-hyponymous transfer B. Co-hyponymous polysemy

?ran- 'rat' > F (reg.), It (reg.) 'mouse' F (reg.) rat, It (reg.) rat, ratta, ratto 'rat', 'mouse'
Pt aborrecer 'to annoy s.o.' > 'to bore s.o.' Pt aborrecer 'to annoy s.o.', 'to bore s.o.'

J. Semantic extension C. Taxonomic polysemy~

MF pigeon 'pigeon raised for eating' > 'any kind of pigeon' F pigeon 'pigeon raised for eating', 'any kind of pigeon'
Sp tener 'to hold' > 'to have' Sp tener 'to hold', 'to have'

4. Semantic restriction
VulgL homo 'human being' > 'man' F homme, It uomo, Sp hombre etc. 'human being', 'man'
F ble 'corn' > 'wheat' F ble 'corn', 'wheat'

s. Lexical ellipsis (absorption)

a) absorption into the determinatum
Sp coche 'coach' > 'car' « coche automOvil) Sp coche 'coach', 'car'
G Schirm 'shelter' > 'umbrella' « Regenschirm) G Schirm 'shelter', 'umbrella'

b) absorption into the determinans D. Metonymic polysemy

F diligence 'velocity' > 'stage-coach' « carosse de diligence) F diligence 'velocity', 'stage-coach'
G Weizen 'wheat' > 'beer made of wheat' « Weizenbier) G (der) Weizen 'wheat', (das) Weizen 'beer made of wheat'
6. Metonymy
L lingua 'tongue' > 'language' L lingua 'tongue', 'language'
L defendere 'to defend' > F defendre 'to forbid' F defendre 'to defend', 'to forbid'
G wiihrend 'while' (temp.) > 'whereas' (advers.) G wiihrend 'while' (temp.), 'whereas' (advers.)

7. Popular etymology
F forain 'non-resident' > 'belonging to the fair' «foire) F forain 'non-resident', 'belonging to the fair'
Lat somnium 'dream' > Sp 'sleep' « somnus) Sp sueflO 'dream', 'sleep'

8. auto-converse change E. auto-converse polysemy

It noleggiare 'to lend' > 'to borrow' It noleggiare 'to lend', 'to borrow'
L hospes 'host' > 'guest' F hate, It ospite, Cat hoste, Occ oste 'host', 'guest'
cl'
9. Antiphrasis F. Antiphrastic polysemy ~
~
F villa 'country house' > F (argot) 'prison' F villa 'country house', F (argot) 'prison' ~
It brava donna 'honourable lady' > It (gergo) 'prostitute' It brava donna 'honourable lady', It (gergo) 'prostitute' S·
S-
III
10. Auto-antonymy G. Auto-antonymic polysemy
E bad 'not good' > E (slang) 'excellent' E bad 'not good', (slang) 'excellent' W
Sard masetu 'gentle' > 'irascible' Sard masetu 'gentle', 'irascible' §'
;::

11. Analogous semantic change all relations possible, e.g.: lS·

metaphoric polysemy
~
F poUr 'to polish', 'to steal' ,fourbir 'to polish' > F fourbir 'to polish', 'to steal' ~.
<::I
'to steal', nettoyer 'to clean' > 'to steal' etc. 10:
~
metonymic polysemy III

L levare 'to lift up', 'to erect', Sp alzar, It alzare Sp alzar, It alzare 'to lift up', 'to erect'

-
IV
'to lift up' > 'to erect' ......:J
272 Andreas Blank

existing polysemy, be it metaphoric, taxonomic or metonymic, by another

word, whose older sense is synonymous (or more rarely co-hyponymous or
antonymic) with the word that serves as a model: thus, during the history of
Romance languages, ?altiare developed almost the same bundle of senses as
levare, both being synonymous from the beginning (see Klein 1997: 134-
137). As will be shown in Section 4, analogy may play an important role in
cases of polysemy that cannot really be regarded as consequences of semantic
change.
Types A-G in the right column of Table 1 represent synchronic relations
between two senses of a word. The labels "metaphoric polysemy" etc. are
therefore somewhat imprecise, as they do not describe the whole polysemy
of a word. It is obvious that each of a pair of two related senses can establish
semantic links to other senses of the word in question. When we say that a
word is polysemous, this does not mean that all senses of a word are inter-
related or have "something in common". Poly se my should rather be con-
ceived as a chain or a network of senses. An application to lexicography
leads inevitably to complex representations, but allows a precise charac-
terization of the relations between the senses of one word, as shown in
Figure 1 with E man and in Figure 2 with F parler: 6

METON (I) 'adult. male person' TAX

(2) 'husband·-....::.::=-=-----------::::::.,..----~;;.;;.;.- (7) 'human being'

METON

(8) 'human individual as

(3) 'male having representing the species;
typically manly mankind'
qualities or virtues'

(5) 'vassal' (6) 'male servant;

valet'
(9) 'one of the pieces
used in playing certain
games. such as chess
or checkers'

Figure J. E man n.
 Polysemy in the lexicon and in discourse 273

(I) 'to produce the (4) 'to communicate by any

sounds of a language' means, non-vocally'
NAG - Vb - (Adv) NAG - Vb - de N/~ N/Adv
(3) 'to communicate vocally,
to express oneself' ~METON
METON
NAG - Vb - pour Inf/Advl ~ (5) 'to reveal what
de N/~ N one has concealed'
(2) 'to speak a given NAG - Vb
language'
NAG - Vb - Adj/N (8) 'to manifest an
intention'
SbAG - Vb - de Inf

(6) 'to converse' (7) 'to deliver an

NAG - Vb - avec NCOAG address, discourse, etc. '~A
NAG - Vb - de N/~ N
(9) 'to express oneself in
written or printed words'
~ NAG-Vb-deN/~N

(10) 'to contain (of a

book or sim.)'
NAG - Vb - de N/~ N

Figure 2. F parter vb.

3. Contextual variation - polysemy - homonymy

Polysemy as a concept of semantic description is distinct from but related to

"contextual variation" (also called "vagueness") and "homonymy" (also
known as "ambiguity"). Generally, it is difficult, if not indeed impossible, to
make a clear-cut distinction between these three phenomena (see Geeraerts
1993: 263). Indeed, in real-life situations, words only occur in concrete
utterances and not in their more abstract and somewhat idealized dictionary
definition. Thus, one has to determine where contextual variation of one
sense ends and where the semantic range of another sense starts - this is the
distinction between vagueness and polysemy - and one also has to find ways
of distinguishing polysemous words, whose senses are by definition related,
from homonyms, i.e. two lexemes which are not semantically related.
274 Andreas Blank

3.1. Contextual variation and polysemy

In a paper entitled "Polysemy and cognition", Paul Deane (1988) adopts the
so-called "standard version" (Kleiber 1990) of prototype-theory - which was
abandoned by mainstream Cognitive Linguistics at about the same time. 7
According to Deane, the prototypical structure of categories allows speakers
to utter sentences like (1a-d):

(I) a. My arm hurts.

b. Look at the arm of the statue.
c. My mother was overdosed on LSD, so my arm is this little thing
on my stomach.
d. A robotic arm reached out and grabbed me.
(All examples: Deane 1988: 347)

In (I a) we have something like the prototypical instance of an arm, while

in the other examples typical, but not necessary, features of an arm are
lacking: (I d) shares the essential functions of a human arm, but is inanimate
like (l b); (l c) has a strange position and may not function as it should.
Nevertheless, all three peripheral instances can be classified as "human arms".
The corresponding core sense comprises a bundle of properties, some of
which can be suppressed and produce the variation found in (1). Deane calls
this phenomenon "allosemy" and distinguishes this "non-distinct variation
in the meaning of a word" (1988: 345) from polysemy by using a number of
tests (for more details, see Deane 1988: 347-350).
Other cases of what one could call "allosemy" or "contextual variation"
are:

(2) a. John washed the car.

b. The mechanic lubricated the car.
(Both examples: eruse 1995: 33)

There is, indeed, a kind of PART-WHOLE relation between (2a) and (2b), or
rather between a general conception of a car and (2a) and (2b). On the other
hand, there are no proper senses of E car corresponding to the exact readings
in (2). They are, according to Cruse, "not even clearly conceptualizable"
(1995: 33). In these cases the context allows us to focus on the relevant part
or sub-domain of a complex concept.
 Polysemy in the lexicon and in discourse 275

In my view, It IS important to make a clear distinction between the

referential or extensionallevel and the level of semantic description: from a
referential point of view, "vagueness" can only mean that a given referent is
classified as a peripheral instance of a category, but still as a member of this
category. If we want to deal with vagueness, or rather "contextual variation",
semantically, it can only mean that in a given utterance the contextual meaning
of a word is recognized as an actualization of a lexicalized sense of this
word, although not all defining features are applicable to this context, e.g.
[animate] or [human] to examples (I b) and (Id). This rather subjective
intuition can be put to the test of intersubjective relevance: if two contextual
meanings show a clear semantic overlap, e.g. the arms in (1), but nevertheless
cannot be related properly by one of the seven semantic relations A-G, as
defined in Section 2, then we are dealing with contextual variation of one
sense.
If, on the other hand, two referents have to be considered as instances of
two different extensional classes, we are beyond the limits of referential
vagueness. If these two contextual meanings are instances of two lexicalized
senses which are related by one of the seven synchronic semantic relations
stated above, then we are dealing with polysemy, as in (3):

(3) a. The arm of a coat.

b. The arm of a record player.
c. An arm of the sea.
(All examples: WEUD, s.v. "arml")

The arm in (3a) is an instance of a CONTAINER--cONTENT metonymy; (3b)

and (3c) are different types of metaphors, (3b) clearly sharing formal aspects
with the human arm, (3c) being a kind of "landscape extremity". In contrast
to (1), none of these referents is directly connected to the upper part of the
human body or of a body built according to the human model: they are not
arms but can be conceptualized as such." If we adopt this point of view, it is
clear that different extensional classes correspond to different concepts and
senses which can all be related to one signifier. Metaphors, metonymies etc.
do not extend an extensional or conceptual category, but relate a new
extensional category and its concept to a given wordY If this metaphor or
metonymy remains a singular contextual effect, i.e. an innovation, nothing
happens, but if it is picked up by other speakers and eventually becomes
conventionalized, a new sense is created within the lexical range of one word.
276 Andreas Blank

3.2. Polysemy and homonymy

U sing both the referential and the semantic criteria we can now distinguish
polysemy from homonymy. Consider (4):

(4) a. The arm on that statue looks better than yours. (Deane 1988: 347)
b. A special arm of the government is to investigate the matter.
(WEUD, s.v. "arm2")
c. His religious convictions kept himfrom bearing arms, but he served
as an ambulance driver with the Red Cross. (WEUD, s.v. "arm2")
d. Three lions passant gardant ... the Royal Arms of England. (Pomy,
Heraldry [1787J; see OED, s.v. "arm2, IV014")

Referentially, all four examples belong to different extensional classes.

Now, while (4b) is another lexicalized metaphor related to (4a), there is no
proper reading of (4c) that allows us to establish a semantic relation to (4a)
or (4b); here arm means 'weapon'. (4d) is metonymically linked to (4c), as
indeed a noble's family crest was painted on a shield, i.e. on an arm (consider
G Waffen - Wappen). We can thus conclude that (4a) and (4b) can be assigned
to one polysemous word, and (4c) and (4d) to another, while there is no link
between the first pair and the second: this is homonymy.
Polysemy results from the lexicalization of an associative process and
therefore is semantic in nature, whereas homonymy, in the vast majority of
cases, arises from phonetic clash, as was the case in our example E arm I
'upper limb of the human body' and the like vs. arm2 'weapon':

(5) a. OE earm 'upper limb of the body' > ModE arml

b. OF arme 'weapon' > ME arme > ModE arm2

Divergent etymology is an important hint for the lexicographer and helps

in understanding synchrony, but etymology should not be taken to describe
synchrony. This is not only a question of methodological integrity; a des-
cription based on this criterion would not even succeed in explaining the
following cases. III

(6) a. OF voler 'to fly' (itr.) [>METON> 'to hunt with falcons' (tr.)
>METON> 'to catch the prey' >META>J 'to steal' (hence ModF
volerl 'to fly' - voler2 'to steal')
 Polysemy in the lexicon and in discourse 277

b. MHO sloz 'lock' [>META> 'castle locking a valley or a pass'

>EXTENSION>] 'castle, palace' (hence ModO Schlossl 'lock' -
Schloss2 'castle')
(7) a. Lat somnus 'sleep' > Sp sueno - somnium 'dream' > Sp sueno
(hence> Sp sueno 'sleep' -MIITON- 'dream')
b. OE corn 'grain' > ModE corn - OF corn 'horny induration on
the foot' > ModE corn (hence ModE corn 'grain' -META- 'corn
on the foot')

In (6) we should, diachronically, have polysemy, but the senses that linked
the first and the last acceptation disappeared from usage (put into square
brackets), so that there is no semantic relation between the remaining senses:
this is "secondary homonymy". In (7), original homonymy is re-interpreted
as polysemy by speakers who feel a semantic relation between the two senses
in question: this is "secondary polysemy".

3.3. Referential class and semantic relation: a double test for polysemy

The last four examples have demonstrated that we need criteria that work
exclusively in synchrony. A typology based on the features of referential
class and semantic relation fulfils this need and helps us to separate homonymy
from polysemy and the latter from mere contextual variation. See Table 2.
Contextual variation describes the semantic range of one lexicalized sense;
thus, if there is a semantic difference, it is below the level of the semantic
relations introduced in Section 2. Polysemy is a property of the semantic
status of a word and describes a network of related senses of this word, as
illustrated above with E man and F parler. Homonymy means that two words

Table 2.

Word form Referential class Semantic relation A-G

Contextual variation Identical Identical No (below the level of

("vagueness") semantic relation)
Polysemy Identical Different Yes
Homonymy Identical Different No (beyond the level of
("ambiguity") semantic relation)
278 Andreas Blank

are only phonetically identical; we are beyond the level of semantic relations. 11
The decision as to whether there is a semantic link or not is by no means
fortuitous - as it sometimes may appear - but can, in most cases, be found
by using the typology of semantic relations given above. This typology may
be based on diachronic processes, but it is essentially a description of syn-
chronic relations.
Considering finally the terms "vagueness" and "ambiguity", I would like
to stress that it is not the senses of a word which are vague, but that sometimes
contexts are vague when they allow different readings, and that often referents
are vague insofar as they are difficult to attribute to an extensional category.
But once we decide to attribute the "poor little thing on one's stomach" to
the class of human anns, the word becomes fully linked to the corresponding
concept and linguistic content. PrototypicaIity is thus a property of ex tensional
classes and not of concepts or senses.

4. Discourse rules, idiosyncrasy and lexicalization: levels of polysemy

4. J. Typical recurrent polysemies

Up to now we have stood by Breal's dogma that polysemy is the synchronic

side of semantic change. A great number of lexicalized semantic changes
support this view. It is, however, challenged by some types of polysemy
which can be found in all languages, e.g. E school which, according to the
context in which it is used, means 'building' (8a), 'a period of education'
(8b), 'the body of pupils and ofteachers' (8c), 'a course or a couple of courses'
(8d) and 'department of a university' (8e):

(8) a. The children are now at the school.

b. School starts at the age of six.
c. The entire school rose when the headmaster entered the auditorium.
d. After school the children rush home.
e. John now teaches at Harvard MedicaL School.

Even more universal are the CONTAINER--cONTENT and the CAUSE-RESULT

metonymies in (9) and (l0):

(9) a. I just bought Chomsky's latest book [= CONTAINER].

 PoLysemy in the Lexicon and in discourse 279

b. Chomskys latest book is awful [= CONTENT].

(to) a. Mary is sad [= STATE (AS RESULTING FROM SOMETHING)].
b. Mary brings sad news [= CAUSE].

Polysemy of this kind is quite common: the CONTAINER~ONTENT metonymy

can be applied to all sorts of containers (see the semantic description in
Copestake and Briscoe 1996: 30-31), the CAUSE-RESULT relation is generally
found with emotional adjectives, and the different metonymies in (8) are
common to a number of words referring to institutions with members in a
building (e.g. parliament, police station, church, country club, etc.), and they
are therefore highly predictable (see Bierwisch 1983: 82). To explain this
kind of "regular" relation, we can choose between two approaches: monosemy
(see Section 4.2) or a special type of polysemy which I will call "rule-based
polysemy" (see Section 4.3).

4.2. Two-level-semantics and polysemy

Let us first investigate the best-known monosemous approach to (8) and (9).
In his "two-level-semantics", Bierwisch treats examples such as E school as
conceptual shifts away from an abstract semantic representation, as a kind of
world-knowledge guided semantic extension of a core sense in specific
contexts (Bierwisch 1983: 85-88).12 Bierwisch's monosemous interpretation
leads to very complex or very abstract lexical entries (see his formalized
notation). This is a problem at the level of metalanguage, but does not
contradict monosemy as such. The "hard" problem with monosemy is that if
we really had a core sense that is extended by conceptual information, the
extension should work in all similar cases in one language and it should
even work universally. This is not the case, as e.g. (8e) would not be possible
in German, and (8d) would not allow insertion of E police station meaning
'duty hours':

(11) a. G ?Hans lehrt jetzt an der Medizinischen Schule der Harvard-

Universitiit.
b. ?After police station the policemen rushed home.

A case where universality breaks down completely is transitivization: 13

280 Andreas Blank

(12) a. E John sleeps in this hotel. - This hotel sleeps 100 guests.
b. F Jean dort dans cet hOtel. - ?Cet hOtel dort 100 clients.
c. G Hans schliift in diesem Hotel. - ?Dieses Hotel schliift 100
Giiste.
(13) a. F Marie sort de la maison. - Marie sort un pistolet de son sac.
b. E Mary comes out of the house. - ?Mary comes a pistol out of
her handbag.
c. G Maria kommt aus dem Haus. - ?Maria kommt eine Pistole
aus der Handtasche.

(12) and (13) show that although transitivization is a general conceptual

process, it is nevertheless highly idiosyncratic: e.g., in French, intransitive
verbs for DIRECTION OF MOVEMENT (sortir, entrer, monter, descendre, etc.) have
generally developed a causative sense (less commonly used with entrer,
however), which is not the case in English and German. And even within
one and the same language, a rule is not always applicable and idiosyncrasies
occur, as demonstrated by the following examples taken from Pustejovsky
and Boguraev (1996: 3--4).14

(14) a. Sam enjoyed the lamb.

b. The lamb is running out in the field.
(15) a. ?We ordered cow for dinner.
b. ?The frog here is excellent.

While E lamb admits the polysemy of 'animal' and 'meat prepared for
cooking/eating', this is not allowed with cow/beef, and in (15b) a reference
to the whole animal is excluded, since usually only hindlegs of frogs are
eaten. If all these words had, on the intralinguistic level, a rather abstract
core sense which allowed semantic extension based on encyclopaedic, extra-
linguistic knowledge, the different idiosyncrasies would be very hard, if not
impossible, to explain. Monosemy thus raises serious methodological prob-
lems and should be rejected as a semantic model.
We can conclude that the types of polysemy described in (8)-(14) result
from profiling against a cognitive background, a process which can be used
in similarly construed domains. Nevertheless these culture-specific rules are
only actualized on the level of one language and thus have to be learned one
by one (Schwarze and Schepping 1995). We have to learn their language-
specific restrictions and their individual "areas of non-application".
 Polysemy in the lexicon and in discourse 281

4.3. Discourse rules and polysemy

We can now understand why polysemy in Breal's sense does not provide a
satisfying explanation for this kind of polysemy, as it is indeed difficult to
explain the apparent regularities by pointing to a number of parallel semantic
changes. Here again, an interesting distinction can be found in Deane (1988):
Deane calls the nonpredictable, idiosyncratic polysemy (the "Breal-type")
"lexical polysemy", while polysemy as based on typical semantic relations,
as discussed throughout the present section, is called "regular polysemy"
(1988: 349-350). Deane's distinction in itself is ingenious; his terminology,
however, is problematic as all polysemies treated here are "lexical" by nature.
Despite this imprecision, Deane was pointing in the right direction, as
what he calls "lexical polysemy" is, as defined above in Section 2, a syn-
chronic consequence of "lexicalized semantic change", while "regular poly-
semy" is obviously different. The first type can be illustrated with examples
such as E mouse 'small rodent', 'computer mouse' or G Schirm 'shelter',
'umbrella' and will henceforth be called "idiosyncratic polysemy". Examples
of the second type include E book 'printed work', 'contents of this work',
and this type of polysemy will be called "rule-based polysemy", as the
polysemy of book arises from the rule that metonymic transfers from the
CONTAINER to the CONTENT are widespread in discourse. Using the term "con-
ceptual metaphor" introduced two decades ago by Lakoff and 10hnson (1980),
we can call the CONTAINER-CONTENT relation a "conceptual metonymy" or a
"contiguity schema" (see Blank 1999b). In fact, all examples cited in this
section are instances of different conceptual metonymies, e.g. BUILDING-
FUNCTION/GOAL (8b), BUILDING-AFFECTED PERSONS (8c), CAUSE-RESULT (10), CON-
TAINER-CONTENT (9, 12a), OBJECT-ACTOR (l3a), ANIMAL-MEAT OF THIS ANIMAL (14).
We can now evaluate our typology of poly se my in the greater context of
cognitive semantics. It is clear that not only conceptual metonymies but
also conceptual metaphors lead to rule-based polysemy, as shown in the
"extension" of the BUILDING-THEORY metaphor in (l6b).

(16) a. Is that the foundation of your theory?

b. His theory has thousands of little rooms and long, winding corridors.
(Examples: Lakoff and lohnson 1980: 46, 53)

The central tenet of Metaphors We Live by, viz. that "our ordinary con-
ceptual system, in terms of which we both think and act, is fundamentally
282 Andreas Blank

metaphorical in nature" (Lakoff and Johnson 1980: 3), can easily be extended
to metonymy. Conceptual metaphors are one way of structuring the world
via language; conceptual metonymies represent another, maybe even more
important way (see Seto, this volume). Less often, it seems, do we use
taxonomic relations deriving from the relation between a prototypical instance
of a category and the category itself. IS It is obvious that we use all major
strategies, metaphor, metonymy and hyponymy (what Seto, this volume, calls
synecdoche) when we think and talk about the world.
This means that on the level of discourse we use conventional metaphors,
metonymies (and taxonomies) and introduce innovations based on con-
ventional conceptual metaphors and metonymies and on prototype-<ategory
relations. It is also rather uncontroversial that metonymies rely on funda-
mental contiguities - often anchored in frames and scenarios - and that
metaphors result from perceptual or functional similarities between concepts
or domains that are not directly related (see Croft 1993; Koch 1995; Blank
L997).
Two questions remain, however. On which Level of knowledge are these
conceptual metaphors and metonymies stored in our mind? And, why are
there language-specific restrictions for the use of some words, while others
do not have these restrictions, despite the fact that they are embedded in the
same conceptual background (see the examples cited above)? It is obviously
not enough to say that the latter are based on encyclopaedic knowledge and
that the former is somehow idiosyncratic. At this point, it is necessary to add
two further levels of linguistically relevant rules, i.e. the level of rules
governing so-called "discourse types" or "discourse traditions" and the level
of language rules that are specific to a particular language.
Discourse traditions are sets of rules for the correct production of a specific
discourse, e.g. the set of metaphors and cliches of Renaissance poetry or the
typical phrases and strategies used when buying a used car or presenting a
paper at a conference (for details see Koch 1997). Although realized, of
course, in a concrete language, a discourse rule is not language-specific: it
characterizes a type of discourse and is common to speakers of all languages
who use a discourse type in a specific way (e.g. the typical Petrarchian
metaphors were known all over Europe during the 16th century, but only
within a small group of connoisseurs). Language rules, on the other hand,
are idiosyncratic and only govern the language use of a certain speech
community or a geographically or socially defined group.
These two levels interact with encyclopaedic or conceptual knowledge,
 Polysemy in the lexicon and in discourse 283

on the one hand, and with the actual discourse, on the other, as represented
in Figure 3.
According to this model, innovations are either based on prominent
encyclopaedic knowledge and have their psychological foundation in salient
associations (e.g. salient similarity, contiguity or contrast), or speakers model
them analogically on already-existing conceptual metaphors, metonymies or
taxonomies which, of course, themselves have cognitive foundations.
Innovations can be lexicalized directly as a specific language rule - an
idiosyncratic metaphor, metonymy etc. (represented by the broken arrow) -
or, and this seems to be more common, as a rule of the specific discourse
type in which the innovation was first used. This innovation may later become
lexicalized as a proper language rule.
In the case of metonymy, for instance, several conceptual contiguities are
permanently highlighted and enable us to construe analogous metonymies
for words that access the same, or a similarly construed, frame or the same
discourse type. As we have probably not learned all these potential meton-
ymies before, we must have learned the rules that underlie them, and their
specific restrictions. Waltereit (1998: 14-19, 26-28) has demonstrated that
these rules are mainly "discourse rules": every discourse type activates a set
of conceptual metonymies that can be filled with concrete lexical metonymies.
This explains why, e.g., (9) and (l0) are very widespread types of contiguity
corresponding to a rule in a large number of discourse types. The range of a
conceptual metonymy is thus limited by the range of its discourse type(s), as
shown in (17).

(17) a. The ham sandwich is waiting for his check [ORDERED DISH-CUSTOMER
in waiters ' discourse].
b. G Ich werde verliingert [EMPLOYEE-CONTRACT in discourses con-
cerning work].

We can now understand why (lOb) was odd: the contiguity schema PLACE
OF WORK-TIME OF WORK is not a usual discourse rule, although it is fully
understandable. The conceptual metonymy can be used creatively to produce
analogous metonymies, but only when based on an activated discourse type.
The same holds true for typical conceptual metaphors: once a similarity
scheme is anchored in the mind as a rule of discourse types it can be produc-
ti vely filled with analogous metaphors, as exemplified in (18).16
284 Andreas Blank

lexicalized as
Language-specific rules
I........ . _ - - - - - - - - - - Discourse rules
"-,..-------'

, ,lexicalized
in , as

,~----

i
reolired "

Encyclopaedic knowledge

Figure 3. Interaction of conceptual rules, language-specific rules, discourse rules

and discourse.

(18) a. E The trigger word opens a mental space [CONCEPTS-CONTAINERS

in linguistics].
b. F le paragonne cl ta jeune beaulte, Qui toujours dure en son
printemps nouvelle, Ce moys d'Avril, qui sesfleurs renouvelle ...
(Ronsard, Les amours cm [1552])
c. Thou art thy mother's glass and she in thee Calls back the lovely
April of her prime. (Shakespeare, Sonnets Ill)

The use of F avril, E April in the sense of 'youth' is a metaphor restricted

to Renaissance poetry, but here again, analogous transfers were possible:

(19) For since mad March great promise made of me; If now the May of
my years much decline, What can be hoped my harvest time will be?
(Sidney, Astrophel and Stella [-1580]. Cited after: Shakespeare, Sonnets,
edited by C. Knox. London: Methuen 1918: 6 fn.)

Restrictions based on a specific discourse type can also be found for

typical taxonomies, but here the situation remains less marked than in the
case of metaphor or metonymy and needs further investigation:
 Polysemy in the lexicon and in discourse 285

(20) I apprehended a vehicle on the premises. [BASIC LEVEL CONCEPT-

SUPERORDlNATE LEVEL CONCEPT in "police-speak"; see Nerlich and Clarke
1999].

Concrete metaphors, metonyrnies or taxonornies depend on rules of specific

discourses and exactly the same holds true for the conceptual metaphors,
metonymies and taxonomies that underlie them. They can be narrowly
restricted to a "small" discourse type or may function as a rule in a larger set
of discourse types. The specific character of the discourse type in which a
conceptual metaphor or metonymy is embedded also determines to what
extent we can apply this conceptual metaphor or metonymy to other words:
one cannot extend a conceptual metaphor or metonymy to a word that would
not be used in discourses where this rule is operative. The range of these
metonymies and metaphors is limited by cultural frames and the discourse
rules themselves, and can therefore vary from society to society.
Furthermore, it is possible for a metonymy, metaphor or a new hypon-
ymous or hyperonymous sense to become lexicalized as a specific language
rule in a specific language. 17 At this point, we have an idiosyncratic lex i-
calization of a discourse rule (itself originally profiled against a cognitive
background). Parallel lexicalized metaphors and metonymies deriving from
such rules can then be interpreted as instances of "analogous semantic change"
(see Section 2).

4.4. Four levels of polysemy

In order to summarize our findings, we can now revisit examples (8)-(19)

and determine their specific character, using the model of poly se my we have
developed. This leads to a fourfold typology of different levels of polysemy:

(i) Rule-based, nonlexicalized polysemy.

Ham sandwich 'customer who has ordered a ham sandwich'
follows a discourse rule which is restricted to a very specific discourse
type and which can only be applied to a limited number of contexts
(Fauconnier 1984 speaks of a "closed connector"): name the customer
by the meal he or she is eating. Within the framework of this discourse
type this can be widely used. There seems to be no further lexicaliz-
alion of this discourse rule or of one single instance of this metonymy
286 Andreas Blank

on the level of a specific language. It seems to be a discourse rule in

the discourse of waiters and waitresses, probably all over the world.
Roughly the same holds true for (17b), (l8a--c) and (19).
(ii) Rule-based and lexicalized polysemy with no or few idiosyncratic
restrictions.
The polysemies of E book and sad derive from an encyclopaedic
(maybe universal) cognitive background and are instances of rather
unspecified discourse rules which allow us to refer to the content via
the container, or to the cause via the effect, and so on. The examples
are also institutionalized as language rules in their specific languages
but, in contrast with the following types, they show almost no idio-
syncratic restrictions of use and of transferability to analogous
concepts.
(iii) Rule-based and lexicalized polysemy with idiosyncratic restrictions.
a. The different polysemies of E school, to sleep, lamb, to operate,
as well as F sortir, all derive from a universal cognitive background
and, as with the examples in (ii), they are instances of very wide
discourse rules and are lexicalized as rules of the English lexicon.
However, they are constrained by idiosyncratic restrictions of
varying strength. This inhibits full transfer to analogous cases,
which native speakers would regard as a violation of the language
norm, despite the fact that they are completely systematic and
understandable.
b. E mental space, vehicle and maybe also school in (8e) belong to
the same type as described in (iiia), but show furthermore a restric-
tion to a specific stylistic or socio-linguistic level.
(iv) Idiosyncratic lexicalized polysemy.
This last type covers cases of polysemy which have developed
without the overt application of a conventional pattern, i.e. a con-
ceptual metaphor, metonymy or taxonomy. Instances of this type cited
in Section I are It rat 'mouse', 'rat'; Sp coche 'coach', 'car'; Sp
sueiio 'dream', 'sleep'; F hate 'host', 'guest'; or Sard masetu 'gentle',
'irascible'. An intermediate stage of conventionalization as a discourse
rule can not, however, be excluded. Idiosyncratic polysemy of this
type seems to occur especially as a consequence of absorption, semantic
extension, popular etymology, auto-converse change and contrast-
based semantic change, i.e. types of semantic change which are based
on general linguistic rules (e.g. expressivity or efficiency; see Blank
 Polysemy in the lexicon and in discourse 287

1999c), but not on conceptual patterns. They differ therefore from

metaphor, metonymy and - less markedly - semantic restriction.

These different levels of conventionalization and idiosyncratic restrictions

correspond, to a certain extent, to a speaker's awareness of a polysemy: type
(i) appears to be completely normal in its context, but can easily be used to
amuse or scandalize people when applied in an inappropriate discourse
situation. 'K Type (iii b) can be used in many discourse types but should not
be used in the wrong register. Type (iii a) is usually "inconspicuous", but can
give rise to astonishment or reproof when transferred productively to non-
lexicalized items. This gives innovative speakers a chance to produce a
communicative effect. It is less easy to achieve a communicative effect in
the case of type (ii) which is open to transfers. There is also evidence that
with type (ii) speakers actually seem to have difficulties in distinguishing
polysemous senses. 19 Type (iv) is too heterogeneous for generalizations: here
everything depends on the individual instance of polysemy.

s. Conclusion

In the previous section of this article I have tried to show that not all types
of polysemy can be derived from types of semantic change. I had to modify
Bn!al's view in this respect. Polysemy as the direct lexicalized consequence
of semantic innovation is only one type in a set of possibilities. Indeed,
speakers seem to prefer to apply conceptual patterns anchored in discourse
traditions and to create semantic innovations based on these patterns. Inno-
vations can become conventional on the level of a discourse type and can
eventually become lexicalized as a language rule with or without specific
idiosyncratic restrictions of both use and transferability.
While even lay people can decide whether there is a semantic relation or
not, polysemy has puzzled specialists for a long time. In this paper I have
tried to develop a referential and semantic test that helps us to distinguish
polysemy from contextual variation and homonymy. I have also discovered
seven types of semantic relations between lexicalized senses of a word. This
allows us to define lexical polysemy positively as a theoretical concept in
semantics, and it can now be used with much more precision as a tool of
semantic and lexicographic description. It should be noted, though, that the
seven types of semantic relations and the distinction between polysemy,
288 Andreas Blank

vagueness and ambiguity can obviously not be used to explain all concrete
cases and to deliver explanations on which all those working in the field of
semantics could agree. This is, in fact, due to the "interpretative nature of
linguistic semantics" (Geeraerts 1993: 263).

Notes

1. See e.g. Cruse (1986: 80); Saeed (1997: 64); Taylor (1995: 99).
2. This type of relation can often be found across different dialects of one language,
e.g. EurSp tigre 'tiger' vs. AmerSp tigre 'jaguar', or EurSp le6n 'lion' vs. AmSp
'puma'. These are, however, marginal cases of polysemy because they have no
reality for speakers in ordinary life situations.
3. For a controversial discussion of this problem see Blank (1997: 191; 1999a:
Section 3; to appear); Gevaudan (1997); Koch (2001).
4. See Blank (l999b); Koch (1999). For a different view of metonymy, see Seto
(this volume).
5. Synecdochical polysemy according to the narrow definition of synecdoche in
Nerlich and Clarke (1999).
6. Example taken from WEUD and PR. Abbreviations: meta = metaphor; meton =
metonymy; tax = taxonomic relation; ag = agent; coag = counter-agent.
7. Polysemy is a core concept in newer studies in Cognitive Linguistics. According
to Lakoff (1987), the different senses of a word correspond to the members of
one single extensional cognitive category which, instead of having one proto-
typical member, shows a certain number of "prototype effects". This rather naiVe
equation of "one word-one concept" not only turns out to be a typical mono-
semous view of semantics, but leads to the puzzling interpretation of cases of
clear homonymy, such as F louerl 'to rent' - louer2 'to praise', as a kind of
monosemously conceived polysemy (e.g. in Geeraerts 1993: 234)! Cognitive
semantics thus simply misses the main task of lexical semantics, i.e. to describe
the senses of a word and the relations between them. For more details see Blank
(2001b: Section 11).
8. This is probably the difference between (3b) and the robotic arm in (Id), which
in fact is the arm of the robot, as it functions (and maybe looks) like a human
arm, while the arm of the record player is not an arm but an artefact that can be
perceived as an arm.
9. It is important not to equate word, concept and extensional category as is done
in some directions of Cognitive Linguistics (especially Lakoff 1987; see the
critique in Brown 1990: 23; Kleiber 1990: 147; Koch 1995: 37).
10. For further examples see Blank (1997: 430-433); Fritz (1998: 58); Ullmann
(1962: 164-179); Widlak (1992).
11. The same set of criteria can be used to define other lexical relations such as
 PoLysemy in the Lexicon and in discourse 289

paronymy (word fonn: similar; referential class: different; semantic relation: no),
synonymy (word form: different; referential class: identical; semantic relation:
no, but senses differ only on the connotational level), hyponymy (word fonn:
different; referential class: indifferent; semantic relation: yes).
12. For a more detailed discussion of (newer) approaches to monosemy see Koch
(1998: 126-130). Other possibilities used in structural approaches are maximizing
homonymy, which would override the semantic relations between the sense, or
the shift of polysemy from the level of the "system" to the level of the "norm",
as defined by Coseriu (1967). This direction is chosen by Dietrich (1997). In this
case, however, the semantic system of the language degrades from a network of
lexicalized word meanings to a mere "semantic potential". This systematic
semantic potential would not really contain language-specific semantic knowledge
but encyclopaedic knowledge, conceptual frames, rules of presuppositions etc.
which pennit deployment of this semantic potential on the level of lexicalized
meanings. Such a "minimalist program of semantics" is suggested by Schlieben-
Lange (1997: 244-247) (see the critique in Koch 1998). For a general critique of
the monosemous view of word meaning in structural semantics as developed by
Eugenio Coseriu and his disciples see Taylor (1999).
13. For further examples see Schwarze and Schepping (1995).
14. Examples of this kind not only challenge monosemous approaches to word
meaning but also the decidedly polysemous approach advocated by Pustejovsky
(1995): he has developed a complex semantic theory centred around four types
of semantic features called qualia, which are modelled on the four types of causes
as defined by Aristotle (see Organon IV, 2: Ch. 11). These qualia explain regular
metonymic and taxonomic polysemy very well, but they do not explain exceptions
as in example (13), and they seem to provide no explanation of metaphoric
polysemy, as this type would require qualia or argument-structure mapping. Indeed,
as Pustejovsky and Boguraev (1996: 4) admit: "the polysemy [in the case of cow
or frog; AB] is the result of lexical rules rather than of alternations within the
qualia of a single lexical item, such as with door [oo.]." An interpretation of these
cases as "semi-productive polysemy" is given by Copestake and Briscoe (1996).
15. See Nerlich and Clarke (1999), who describe typical discourse types where
taxonomic strategies are exploited.
16. There is, however, a greater number of very open conceptual metaphors (e.g.
"orientational metaphors"); see Lakoff and 10hnson (1980: 14).
17. According to Schwarze and Schepping (1995) this is even a necessary development.
18. Imagine someone talking about a restaurant where he/she has been for dinner
saying: "And then this ham sandwich sitting next to us went off without paying."
19. This was the finding of the empirical study by Soares da Silva (1992): his
Portuguese subjects had to classify the semantic relation of marked words in
pairs of sentences from "0" (no relation) to "4" (identical). The highest scores in
his test were actually given to Pg livro 'book' (CONTAINER-CONTENT) (3.92) and
triste 'sad' (RESULT-CAUSE) (3.62), which means that many test subjects were not
290 Andreas Blank

fully aware of the polysemy, i.e. of the semantic difference. Instances of type
(iii) were usually better classified as being polysemous (score between 1.5 and
3.5). See also Blank (1997: 418-419).

References

Aristotle
1995 Categories; Organon IV. Both in: Philosophische Schriften. Volume I.
Darmstadt: Wissenschaftliche Buchgesellschaft.
Bierwisch, Manfred
1983 Semantische und konzeptuelle Reprasentation lexikalischer Einheiten.
In: Rudolf Ruzicka and Wolfgang Motsch (eds.), Untersuchungen
zur Semantik, 61-99. Berlin: Akademie.
Blank, Andreas
1997 Prinzipien des lexikalischen Bedeutungswandels am Beispiel der
romanischen Sprachen. Tiibingen: Niemeyer.
1999a Les principes d'association et leur importance pour la structure du
lexique. Studi italiani di linguistica teorica ed applicata XXVID.
1999b Co-presence and succession: a cognitive typology of metonymy. In:
Giinter Radden and Klaus-Uwe Panther (eds.), Metonymy in Language
and Thought, 169-192. Amsterdam: Benjamins.
1999c Why do new meanings occur? A typology of the motivations for
lexical semantic change. In: Andreas Blank and Peter Koch (eds.),
Historical Semantics and Cognition, 61-89. Berlin, New York: Mouton
de Gruyter.
2oo1a Pathways of lexicalization. In: Martin Haspelmath, Ekkehard Ktinig,
Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology
and Universals: An International Handbook, 11, 1596-1608. Berlin,
New York: De Gruyter.
2oo1b Semantik 11. Neuere Entwicklungen in der lexikalischen Semantik.
In: Giinter Holtus, Michael Metzeltin, and Christian Schmitt (eds.),
Lexikon der Romanistischen Linguistik. I, 1,918-939.
(to appear) Fondamenti e tipologia del cambio semantico. In: Giampaolo Salvi
(ed.), Semantica e lessicologia storica. Rome: Bulzoni.
Breal, Michel
1897 Essai de semantique: Science des significations. Paris: Hachette.
Brown, Cecil H.
1990 A survey of category types in natural language. In: Savas Tsohazidis
(ed.), Meanings and Prototypes: Studies in Linguistic Categorization,
17-47. London: Routledge.
 Polysemy in the lexicon and in discourse 291

Copestake, Ann and Ted Briscoe

1996 Semi-productive polysemy and sense extension. In: James Pustejovsky
and Branimir Boguraev (eds.), Lexical Semantics. The Problem of
Polysemy, 15-67. Oxford: Clarendon.
Coseriu, Eugenio
1967 Sistema, norma y habla. In: Eugenio Coseriu, Teor{a dellenguaje y
lingu{stica general, 11-113. Madrid: Gredos.
Croft, William
1993 The role of domains in the interpretation of metaphors and metonymies.
Cognitive Linguistics 4: 335-370.
Cruse, David A.
1986 Lexical Semantics. Cambridge: Cambridge University Press.
1995 Polysemy and related phenomena from a cognitive linguistic view-
point. In: Patrick Saint-Dizier and Evelyne Viegas (eds.), Compu-
tational Lexical Semantics, 3~9. Cambridge: Cambridge University
Press.
Deane, Paul D.
1988 Polysemy and cognition. Lingua 75: 325-361.
Dietrich, Wolf
1997 Polysemie als 'volle Wortbedeutung' - gegen die 'Mehrdeutigkeit'
der Zeichen. In: U1rich Hoinkes and Wolfgang Dietrich (eds.),
Kaleidoskop der Lexikalischen Semantik, 227-237. Tiibingen:
Niemeyer.
Geeraerts, Dirk
1993 Vagueness's puzzles, polysemy's vagaries. Cognitive Linguistics 4:
223-272.
Gevaudan, Paul
1997 La polysemie verticale: hypotheses, analyses, interpretations. In: Phin
2 (www.fu-berlin.de/phin).
Fritz, Gerd
1998 Historische Semantik. Stuttgart: Metzler.
Kleiber, Georges
1990 La semantique du prototype. Paris: Presses Universitaires de France.
Klein, Franz-Josef
1997 Bedeutungswandel und SprachendifJerenzierung. Die Entstehung
der romanischen Sprachen aus wortsemantischer Sicht. Tiibingen:
Niemeyer.
Koch, Peter
1995 Der Beitrag der Prototypentheorie zur Historischen Semantik: eine
kritische Bestandsaufnahme. Romanistisches lahrbuch 46: 27-46.
1997 Diskurstraditionen: zu ihrem sprachtheoretischen Status und ihrer
Dynamik. In: Barbara Frank, Thomas Haye, and Doris Tophinke
292 Andreas Blank

(eds.), Gattungen mittelalterlicher Schrijtlichkeit, 43-79. Tiibingen:

Narr.
1998 Saussures mouton und Hjelmslevs trte: zwei Schulbeispiele zwischen
Semstruktur und Polysemie. In: Edeltraud Werner, Ricarda Liver,
Yvonne Stork and Martina Nicklaus (eds.), Et multum et multa.
Festchrijt Peter Wunderli, 113-136. Tiibingen: Narr.
1999 Frame and contiguity: on the cognitive basis of metonymy and certain
types of word-formation. In: Gtinter Radden and Klaus-Uwe Panther
(eds.), Metonymy in Thought and Language, 139-168. Amsterdam:
Benjamins.
200 1 Lexical typology from a cognitive and linguistic point of view. In:
Martin Haspelmath, Ekkehard Konig, Wulf Oesterreicher, and Wolfgang
Raible (eds.), Language Typology and Universals: An International
Handbook, IT, 1142-1178. Berlin, New York: De Gruyter.
Lakoff, George
1987 Women, Fire and Dangerous Things: What Categories Reveal about
the Mind. Chicago: University of Chicago Press.
Lakoff, George and Mark lohnson
1980 Metaphors We Live By. Chicago: University of Chicago Press.
Nerlich, Brigitte and David D. Clarke
1999 Synecdoche as a communicative strategy. In: Andreas Blank and Peter
Koch (eds.), Historical Semantics and Cognition, 197-213. Berlin,
New York: Mouton de Gruyter.
OED
1989 Oxford English Dictionary. 2nd Edition. Oxford: Clarendon.
PR
1977 Dictiollnaire Alphabetique et Analogique de la Langue Franraise
par Paul Robert. Paris: Parmentier.
Pustejovsky, lames
1995 The Generative Lexicon. Cambridge, MA: MIT Press.
Pustejovsky, lames and Branimir Boguraev (eds.)
1996 Lexical Semantics: The Problem of Polysemy. Oxford: Clarendon.
Saeed, lohn I.
1997 Semantics. Cambridge, MA: BIackwell.
Schlieben-Lange, Brigitte
1997 Uberlegungen zu einer einfachen Systematik der Zuweisung von
(polysemen) Lesarten. In: Ulrich Hoinkes and Wolfgang Dietrich
(eds.), Kaleidoskop der Lexikalischen Semantik, 239-247. Ttibingen:
Narr.
Schwarze, Christoph and Marie-Therese Schepping
1995 Polysemy in a two-level-semantics. In: Urs Egli et al. (eds.), Lexical
Knowledge in the Organization of Language, 283-300. Amsterdam,
Philadelphia: John Benjamins.
 Polysemy in the lexicon and in discourse 293

Soares da Silva, Augusto

1992 Homonimfa e polissemia: amilise semica e teoria do campo lexico.
In: R. Lorenzo (ed.), Actas do XIX Congreso Internacional de
Lingii{stica e Filoloxfa Romanicas. Volume 2,257-287. La Coruna:
Fundaci6n Pedro Bame de la Maza.
Sweetser, Eve
1990 From Etymology to Pragmatics. Cambridge: Cambridge University
Press.
Taylor, John R.
1995 Linguistic Categorization: Prototypes in Linguistic Theory. Oxford:
Clarendon.
1999 Structural semantics and cognitive semantics. In: Andreas Blank and
Peter Koch (eds.), Historical Semantics and Cognition. Berlin, New
York: Mouton de Gruyter.
UIlmann, Stephen
1962 Semantics: An Introduction to the Science of Meaning. Oxford:
BlackweIl.
Waltereit, Richard
1998 Metonymie und Grammatik: Kontiguitiitsphiinomene in der
franzosischen Satzsemantik. Tlibingen: Niemeyer.
WEUD
1994 Websters Encyclopedic Unabridged Dictionary ofthe English Language.
New York: Gramercy.
Widlak, Stanislaw
1992 Fra lessicologia e stilistica. Problemi di lessicologia e stilistica
deU'italiano e di altre lingue romanze. Cracow: Universitas Cracovia.
Psycholinguistic approaches
Irony in conversation: salience, role, and context
effects

Rachel Giora and lnbal Gur

1. Introduction

Much of what is going on in discourse comprehension and production

depends on the very first moments of comprehension. Consider the following
example:

Iddo and Omri (aged 7 years and 8 months, native speakers of Hebrew)
are playing together. Iddo fetches himself a glass of juice out of the
refrigerator.

(1) Omri: I want to drink too.

Mira (Iddo's mother): Iddo, totd 10 et ha-mitz (,take the juice out [of
the refrigerator] for him').
Iddo (laughingly): ha ... ha ... le-hotci 10 et ha-mitz ('to take/squeeze
the juice out of him' - a Hebrew idiom meaning 'drive one crazy
by imposing all kinds of hardships on herlhim').

What this example illustrates is that speakers and comprehenders make

use of what is available to them, regardless of contextual information or
speaker's intent (see also Horton and Keysar 1996; Keysar et at. 1998; Keysar,
Barr and Horton 1998). Though the idiomatic meaning of Mira's utterance
in (I) was not the intended meaning, nor was it compatible with context, it
was not ignored; contextual information did not inhibit its activation.
Why should contextually incompatible meanings be activated and infiltrate
the ongoing discourse? Why doesn't context block unintended meanings?
According to the graded salience hypothesis (Giora 1997,2003), words' and
expressions' salient meanings, i.e. meanings coded in the mental lexicon
(whose relative salience is further affected by, e.g., frequency, familiarity,
conventionality, or prototypicality) cannot be bypassed. Though prior context
may be instrumental in enhancing a word's or an expression's meaning (e.g.
298 Rachel Giora and lnbal Gur

Forster 1989), it can hardly inhibit activation of salient meanings (cf. Tabossi
1988; Titone 1998; but see Martin et al. 1999; and Vu, Kellas and Paul 1998
for a different view). Salient meanings should always be accessed and always
initially, regardless of contextual bias.
The question may need rephrasing, then. If contextual information is less
able to affect preselection of the appropriate meaning when there are other
equally or more salient competitors, why doesn't it suppress irrelevant mean-
ings that surface only because of their availability? This may not be the right
question either, because contextually inappropriate information does get
suppressed (e.g. Gemsbacher 1990; Onifer and Swinney 1981; Swinney
1979). However, Gemsbacher and Robertson (1995) also showed that skilled
and less skilled readers differ in their suppression abilities. While skilled
readers did not outperform less skilled readers at the initial, access stage,
they did at the later, suppression stage: Less skilled comprehenders were
less capable of suppressing contextually irrelevant information (which must,
therefore, interfere with their comprehension). But this will not explain the
use Iddo made of the irrelevant meaning (in [1]). His laughter suggests that
he computed both meanings (noticing their incongruity), deliberately choosing
to use the contextually inappropriate meaning for a special purpose: to crack
a joke.
Here, then, is an instance of an idiomatic meaning of an utterance, which
though inconsistent with contextual information, was neither inhibited nor
suppressed. Such behaviour is not explainable by an interactive approach to
discourse comprehension, which allows contextual information to affect
processing very early on so that only the contextually appropriate meaning
of words and sentences becomes available (e.g. the Relevance Theoretic
account proposed by Sperber and Wilson 1986/1995; Carston 1999; but see
Giora 1998a for an alternative interpretation of this account). Moreover,
activating inappropriate meanings after the appropriate meaning has been
captured is not motivated by such models. Such behaviour is not explainable
by the standard pragmatic model (Grice 1975; Searle 1979) either. According
to this model, language comprehension always begins with the literal interpre-
tation. If it is compatible with contextual information it is not revisited, and
search is terminated. Upon this view, then, literally intended language is
processed only literally. Since Mira's utterance was intended literally, this
model cannot account for the involvement of the contextually irrelevant
nonliteral meaning in Iddo's comprehension process (see also Gibbs 1980).
According to the graded salience hypothesis (Giora 1997, 2003), however,
 Irony in conversation: salience. role. and context effects 299

so-called irrelevant meanings are activated because they are salient. According
to the retention hypothesis (Giora 1995, 2003; Giora and Fein 1999a), such
meanings would, indeed, be attenuated or fade if they have no role in con-
structing the intended meaning. However, if they can be allocated some
function in the construction of the discourse interpretation, they would be
retained. Such is the case of the literal meaning of ironic utterances (e.g.
Giora 1995; Giora, Fein and Schwartz 1998). According to the graded salience
hypothesis, the contextually incompatible literal meaning of (the critical word
of nonconventional) irony is activated because it is salient, i.e., stored in the
mental lexicon. It is not dispensed with by contextual information, because
it has a role in constructing the ironic meaning: it allows the comprehender
to compute the difference between what is expected and what is (Giora 1995).
For instance, when I say What a lovely day for a picnic on a stormy day, the
literal meaning of the critical word lovely would be retrieved directly from
the mental lexicon on account of its salience. Despite its contextual misfit, it
would not be suppressed, because it is instrumental in deriving the ironic
meaning ('lousy' or 'far from being lovely'). Indeed, in Giora, Fein and
Schwartz (1998) we showed that the literal meaning of irony is made available
immediately, and remains active even 2000 msec after offset of the target
(ironic) sentence. However, after such a long delay, the same literal meaning
is no longer active in literally biasing contexts. Having been accessed and
integrated with contextual information, the literal meaning (of lovely) in the
literal context (e.g. when What a lovely day for a picnic is said on a sunny
day) has no further functions. It, therefore, begins to fade.
Similar findings are reported by Giora and Fein (1999a) regarding contex-
tually inappropriate meanings of salient/conventional metaphors. In Giora
and Fein's studies, the salient, literal meaning of utterances embedded in
metaphorically biasing contexts was retained in spite of contextual misfit.
However, the reverse did not hold: the salient metaphoric meaning of the
same utterance embedded in a literally biasing context showed deactivation.
For instance, concepts (e.g. 'rise') related to the literal meaning of conven-
tional (Hebrew) metaphors (e.g. Only now did they wake up. meaning 'only
now did they start doing something') were not suppressed in the metaphori-
cally biasing context, even though they were contextually incompatible in
that context. In contrast, the salient metaphoric meaning ('do') of the same
conventional metaphor was less active in the literally biasing context, where
it had no role in constructing the literal meaning of the utterance.
Such asymmetry has been shown to hold for balanced polysemous words
300 Rachel Giora and lnbal Gur

as well (words having related meanings that are similarly salient). Where a
contextually inappropriate meaning of a polysemous word was instrumental
in sustaining the contextually appropriate meaning, it was retained. Where it
was not, it was deactivated. In Williams (1992), (salient) central meanings
(e.g. 'solid') of polysemous words (e.g. firm) were activated immediately
and retained even after a long delay (of 859 msec) despite contextual misfit
(e.g. The school teacher was criticized for not being firm). Such meanings
are indeed conducive to the interpretation of less central interpretations.
(Salient) less central meanings (e.g. 'strict'), however, were not found to be as
context-resistant. Having been activated immediately, they were not retained
after a long delay, when they were not compatible with contextual information
(e.g. The couple wanted a bed that was firm). Though 'strict' may be related
tofirm, it is not conducive to the construction of its central meaning ('solid').
Therefore, it was not retained after it had been activated. However, 'solid' is
conducive to the 'strict' interpretation of firm, and therefore it was not
suppressed after being activated. This finding is consistent with the hypothesis
that contextually incompatible meanings that are instrumental in the interpre-
tation of the intended meaning are not suppressed (Giora 2003; Giora and
Fein 1999a).
Salient meanings that have not been deactivated because they have some
role in constructing the meanings currently being built, may be easily reused
by the discourse participants on account of their availability (e.g. the idiomatic
meaning in [1] above). To testify to the availability of the salient, though
contextually incompatible, literal meaning of ironic discourse, we examine
here the kind of response irony elicits in naturally occurring conversation. If
irony is responded to literally, this would be consistent with both the graded
salience hypothesis (Giora 1997) and the indirect negation view of irony
(Giora 1995; Giora and Fein 1999b, 1999c) which assume that irony compre-
hension involves activation and retention of its salient, though contextually
inappropriate, literal meaning. Such behaviour, however, is less compatible
with the direct access view (e.g. Gibbs 1986) which assumes that in a rich
context, irony is processed more or less directly, without having to go though
a contextually incompatible interpretive phase. Such behaviour is also less
compatible with the standard pragmatic model (Grice 1975; Searle 1979)
which posits that the contextually inappropriate literal meaning activated
initially should be rejected and replaced by the intended ironic meaning. I
This suppression hypothesis (see also Gernsbacher, Keysar and Robertson
2001) predicts that immediately after the utterance has been comprehended,
 Irony in conversation: salience. role. and context effects 301

there should be no residue of the contextually inappropriate literal meaning.

We have chosen to focus on spontaneous speech because such discourse
is less amenable to control and revision than written discourse. The assumption,
then, that information available to interlocutors may play a major role in
discourse comprehension and production must be particularly applicable to
such discourse. It is not the case, of course, that spontaneous talk is not
monitored (see, e.g., Zaidel 1987), or does not involve error correction.
However, given the time constraints imposed on face-to-face interaction,
spontaneous speech must be more "error" prone than written discourse, which
is self-paced and can be revisited any time. If one wants to examine the
extent to which contextually incompatible meanings are activated and mani-
pulated in the course of discourse construction, spontaneous talk seems the
natural environment to explore.

2. Irony and spontaneous discourse

So far, most of the research into the processes involved in comprehension

was conducted in the laboratory. Findings showed that lexical access is hardly
affected by contextual information (but see Martin et al. 1999; Vu, Kellas
and Paul 1998). Ample evidence has been adduced supporting the view that
lexical access is modular: lexical processes are autonomous and impervious
to context effects (see Fodor 1983; Forster 1979). Upon one version of the
modular view, lexical access is exhaustive; all the word's coded meanings
are accessed initially upon its processing, regardless of contextual bias (e.g.
Cairns 1984; Conrad 1974; Lucas 1987; Onifer and Swinney 1981; Picoult
and Johnson 1992; Seidenberg, Tanenhaus, Leiman and Bienkowski 1982;
Swinney 1979; Tanenhaus, Carlson and Seidenberg 1985; Tanenhaus, Leiman
and Seidenberg 1979; Till, Mross and Kintsch 1988; West and Stanovich 1988
and references therein). Upon another (ordered) version, lexical access is
exhaustive but frequency-sensitive: the more frequent meaning is accessed
first, and a search for the intended meaning continues only in case the more
frequent meaning is incompatible with the context (Bradley and Forster 1987;
Duffy, Morris and Rayner 1988; Rayner and Frazier 1989; Rayner and Morris
1991; Sereno, Pacht and Rayner 1992; Simpson 1981; Simpson and Burgess
1985; Simpson and Foster 1986; Simpson and Krueger 1991; Swinney and
Prather 1989; Tabossi 1988; for a review, see Gorfein 1989; Rayner, Pacht
and Duffy 1994; Simpson 1994; Small, Cottrell and Tanenhaus 1988). Con-
302 Rachel Giora and Inbal Gur

textual information has only post-access effects, suppressing contextually

inappropriate meanings and selecting the appropriate meanings (Swinney
1979).
The modular view of lexical access has been challenged by an interactive
direct access hypothesis, according to which lexical access is selective.
Contextual information directs access completely, so that only the appropriate
meaning is made available for comprehension (e.g. Glucksberg, Kreuz and
Rho 1986, Jones 1991; Martin et at. 1999; Schvaneveldt, Meyer and Becker
1976; Simpson 1981, Vu, Kellas and Paul 1998; but see Giora 2003 for a
critique).
Evidence from research into figurative language comprehension accumu-
lated in the laboratory seems consistent with the direct access view. Literal
and figurative utterances have been shown to involve equivalent processes
when embedded in supportive contexts (see Gibbs 1994 for a review). They
were shown to be processed equally automatically (e.g. Gildea and Glucksberg
1983; Glucksberg, Gildea and Bookin 1982; Keysar 1989), to involve the
same categorization procedures (see Glucksberg and Keysar 1990; Shen
1997), and to take equally long to read (lnhoff, Lima and Carroll 1984;
Kemper 1981; Ortony et at. 1978). Irony was no exception: it was shown to
take no longer to read than non-ironic discourse (see Gibbs 1986, 1994).
More recent findings, however, question the direct access hypothesis (and
see Giora 1997, 1998b, 2003, for a critical review). For instance, utterances
embedded in ironically biasing contexts took longer to read than when
embedded in literally biasing contexts (Dews and Winner 1997; Giora, Fein
and Schwartz 1998; Pexman, Ferretti and Katz 2000; Schwoebel, Dews,
Winner and Srinivas 2000). They were also found to be processed literally
first, and ironically later (Giora and Fein 1999c; Giora, Fein and Schwartz
1998). Even conventional ironies were found to be processed literally initially,
in parallel to the ironic meaning (Giora and Fein 1999c). Further, reanalysis
of Gibbs' findings (Giora 1995) is consistent with the view that irony compre-
hension involves a contextually incompatible (literal) phase (but see Gibbs
2002 for a different view). Would spontaneous speech support a modular
based view which allows meanings to be activated regardless of context?
To test this we examined naturally occurring conversations. In particular,
we looked at how irony affects text progression. Would it avail both the
literal and ironic meaning for further discussion and elaboration, as predicted
by the graded salience and retention hypotheses (Giora 1995, 1997, 2(03)1
Would it make available only the appropriate ironic meaning as predicted by
 Irony in conversation: salience. role. and context effects 303

the direct access view, the modular view, and standard pragmatic model?
To illustrate the way the contextually incompatible literal meaning of irony
may avail itself for further elaboration, consider the following example (taken
from Drew 1987) cited in Clark (1996: 374), in which the ironic/teasing turn
is responded to literally (both in bold). In this example, Gerald has just bought
a brand-new sports car, and is late for a meeting. He could respond to Lee's
ironic reprimand by addressing the ironic meaning (the reprimand), saying,
e.g., "I am sorry" or coming up with a real explanation. Instead, he proceeds
along the lines proposed by Lee, thereby elaborating on the tease or irony (in
bold), i.e., on the literal meaning of the ironic utterance (in bold):

(2) Gerald: Hi how are you.

Martha: Well, you are late as usual.
Gerald: Eheh eheh eheh eheh.
Lee: What's the matter couldn't you get your car started?
Gerald: Hehh. That's right. I had to get it pushed, eheh eheh
eheh eheh.

In our data, we looked for such responses to irony. Responses to the literal
meaning of irony are predicted by the graded salience hypothesis (Giora
1997, 2003) and the retention hypothesis underlying the indirect negation
view (Giora 1995). On the assumption that the literal meaning of
(nonconventional) irony is the only one coded in the mental lexicon, it should
be accessed once the irony is encountered. It should not be suppressed even
when the ironic interpretation is derived, because it facilitates the computation
of the difference between the expected and the derided situation.

2.1. Method

Participants: The participants were five Israeli friends (two women, three
men) who spent a Friday evening together. They were 29-33 years old, native
speakers of Hebrew, living in TeI Aviv.
Materials: Our data come from one-hour tape-recorded conversations
among the participants. The conversations took place in Tel Aviv, in October
1997. They comprise 9,380 words. For illustration, consider an English
translation of a Hebrew extract of the conversation (example [3] below, ironic
utterances underlined), containing about 300 words, which partly revolves
304 Rachel Giora and lnbal Gur

around Sara and Benjamin (Bibi) Netanyahu (the then Israeli Prime Minister
and his wife), who, in a newspaper interview, complained about the press
harassing them. The literal responses are in bold. 2

(3) 1. A: You don't understand one thing. You think they initiate these
things? This is maybe the first article they [wrote].
2. B: [They] stay home
and after them the paparazzi come ... and they are simply
[miserable].
3. A: [(They) ruin] their lives, what do you want?
4. B: I want to cite the last sentence of the article, yes, out of the
potency's (the word used in Hebrew is heroism) mouth
(meaning 'out of God's mouth', equivalent to the English
idiom 'out of the horse's mouth') [as they say].
5. A: [Out of the baby's mouth].
6. B: Out of the hero's mouth. In fact, it's not in this article, it's
in the, it's in the article that appeared in Ma'ariv (an Israeli
daily), but x Sara says that maybe following the tragedy of
Princess Diana they will begin to understand ... she and Diana
on the same level!
7. C: So, the last photograph, they chose the picture of Bibi and
Sara and their two kids on the beach in Caesaria and <xx>
ya'eni (a marker for irony) a spontaneous picture, but it's
obvious that this picture is carefully arranged: the big boy
with the father - the small one with the mother, and all this.
8. A: (0.1 second later) It's as if I take a picture of you (C) now,
say, you are sitting next to D (C's wife), because ... really,
come on!
9. C: One thing is certain, then ... th~ paparazzi photQ~raph~rs will
not catch them at the sp~ed Qf 160 kph,
10. Everybody: @ (2 seconds).
11. D: So she cannot compare herself to Diana.
12. B: <xx> She is really miserable because th~y do her injJ,!stice.
13. A: Diana, my ass, this entire story, believe me, I feel like retching.
14. D: Yes.
IS. A: Bi~ deal! [<xx>]
16. C: [1 dQD't f~~l] lik~ 1ll1!ObiD~ ill all. 1f~~llik~ libariD~
lbe profits.
 Irony in conversation: salience, role, and context effects 305

17. D: [@@@@]
18. C: [I feel like] sharine the profits. Throw me some bone
(equivalent to the English idiom 'throw me a scrap').
19. D: Open a florist shop.

2.2. A sample analysis

Speaker A (in [3.1]) sympathizes with the Netanyahus, while speaker B (in
[3.2]) is critical of them. Speaker B describes them ironically as "miserable"
- the literal sense - echoing their complaint and indicating that they are far
from being miserable, and that they must be happy - the ironic meaning -
about being so popular in the press. However, B (in [3.3]) elaborates on their
"misery" - the literal sense of the irony - by retorting that their life is
"ruined". Whether this response can be viewed as resonating with the literal
meaning is dubious: it could be a repetition of that person's belief rather
than a response to the previous utterance.
In (3.4), the reference to Bibi as God ("potency"/"heroism") is ironic.
Both the following responses referring to him as either "baby" (sort of
opposite) or "hero" in (3.6) are echoes of the literal meaning of the irony.
The irony at the end of (3.6) remains uncommented on at this stage. D is
going to refer to it later (in [3.11]), but it is hard to tell whether this is a
response to the literal or ironic meaning of the irony here.
The ironical meaning 'spontaneous' in (3.7) is responded to by A in (3.8).
A disagrees with C, i.e. with the ironical meaning, and attempts to defend
the genuine spontaneity of the photograph of the Netanyahus (appearing in
the press to affect "spontaneity" in order to support their claim that they are
haunted by the press).
The irony in (3.9), which suggests the couple will never really run away
from the paparazzo photographers, was responded to by a 2-second laughter.
The irony in (3.12) is a repetition of the topic of this conversation. It is a
repetition of the literal meaning of the previous ironies, particularly those
generated by the same speaker himself.
The irony in (3.16) is responded to by laughter in (3.17).
In (3.18), C echoes the literal meaning of his own irony, and in (3.19) the
literal meaning of his irony is elaborated on by D.
306 Rachel Giora and Inbal Gur

2.3. Results and discussion

Fifty-six ironic utterances were selected, on which there was 100% agree-
ment (as to their ironiness, reached at times after a discussion) between two
native speakers of Hebrew who listened to the recording. One judge was a
participant in the conversations. Of these 56 ironic utterances, 42 (75%) were
responded to by reference to their literal meaning. The responses were judged
as literal by the two judges as above. Only those on which there was 100%
agreement that they were indeed responses to the literal meaning of the irony
were counted.
These results suggest that the literal meaning of irony is accessed and
retained by both speakers and addressees. The occurrence of irony in the
conversations made its literal meaning available for further discussion and
elaboration, as predicted by the graded salience hypothesis (Giora 1997, 2003)
and indirect negation view (Giora 1995). However, they are partly inconsistent
with the standard pragmatic model (Grice 1975; Searle 1979) and modular
view (Onifer and Swinney 1981; Picoult and 10hnson 1992; Seidenberg et
al. 1982; Swinney 1979), attesting that the literal meaning was not suppressed
as irrelevant. Further, they are incompatible with the direct access view,
demonstrating that irony did not avail the ironic meaning exclusively.
It could be argued, of course, that tapping processes involved in under-
standing and producing naturally occurring conversations is not comparable
to testing comprehension on-line. While the direct access view may be chal-
lenged on the basis of evidence about processes, evidence accumulated from
conversational discourse may be telling only about products. While the
challenge is valid, it is still important to note that these products are better
explained by a graded salience view of comprehension rather than by a direct
access view, which does not allow for any contextually incompatible
interpretation to be activated initially. Recall, further, that these findings do
not stand in isolation, but are consistent with previous findings attesting to
processes attained by on-line measures such as reading times (Giora, Fein
and Schwartz 1998; Pexman, Ferretti and Katz 2000; Schwoebel, Dews,
Winner and Srinivas 2000) and lexical decision tasks (Giora and Fein 1999c;
Giora, Fein and Schwartz 1998), showing that irony is not accessed ironically
first.
Our findings can also be viewed as an instantiation of a more general
phenomenon of "dialogic syntax" (Du Bois 1998). Dialogic syntax occurs
when a speaker constructs an utterance based on an immediately co-present
 Irony in conversation: salience, role, and context effects 307

utterance. Du Bois discloses the ubiquity of dialogic syntax, showing that a

vast array of linguistic elements such as syntactic, semantic, pragmatic,
lexical, and even phonetic patterns in one speaker's discourse can be traced
back to an immediately co-present utterance. This suggests that activation of
any linguistic element makes it available for the same or next speaker to
elaborate on, the literal meaning of irony included. Our findings indeed
confirm that ironies avail their literal meaning, thereby allowing recurrence
of the salient/literal meaning in the next discourse segment. Evidence of
similar effects of a given utterance on adjacent ones (cf. Du Bois 1998)
suggests that salience of meanings is a major factor in discourse compre-
hension and production.

3. Summary

According to the graded salience hypothesis (Giora 1997, 2003), salient

meanings should always be activated initially, even when they are incom-
patible with contextual information. A meaning of a word or an expression is
salient if it is coded in the mental lexicon (e.g. the literal meaning of less
familiar irony but not its intended, nonliteral meaning made available by
context). Factors contributing to degrees of salience are, e.g., conventionality,
frequency, familiarity and prototypicality. Prior context may be instrumental
as well, though its role is limited. It may be predictive and facilitate activation
of a word's meaning(s), but it is less efficient in inhibiting activation of
salient meanings. In this respect, the graded salience hypothesis is consistent
with the modular model of lexical access (e.g. Rayner, Pacht and Duffy 1994;
Swinney 1979), particularly with an ordered-access version of it that is
frequency-sensitive (e.g. Hogaboam and Perfetti 1975; Simpson and Burgess
1985; Swinney and Prather 1989). It predicts that even rich and supportive
contexts biased in favour of less salient meanings should not inhibit activation
of salient meanings.
Indeed, previous research by Giora and her colleagues has demonstrated
that non salient ironies activate their salient literal meaning initially (Giora
1999,2003; Giora and Fein 1999c; Giora, Fein and Schwartz 1998): ironic
utterances were shown to facilitate literally related concepts 150 msec after
their offset, regardless of contextual bias. Similarly, salient ironies availed
their salient ironic meaning initially, in parallel to their salient literal meaning.
In contrast, nonsalient ironies facilitated ironically related concepts later -
308 Rachel Giora and lnbal Gur

l00{}-2000 msec after their offset. These findings support the graded salience
hypothesis, but are inconsistent with the view that context affects comprehen-
sion significantly (e.g. Glucksberg, Kreuz and Rho 1986; Sperber and Wilson
1986/1995, but see Burgess, Tanenhaus and Seidenberg 1989 for a critique).
Previous research (Giora and Fein 1999c; Giora, Fein, and Schwartz 1998)
has also demonstrated that, contra the standard pragmatic model (Grice 1975;
see also Searle 1979) and the modular view (Swinney 1979), the contex-
tually incompatible meaning of irony is not suppressed by contextually biased
infonnation. Both salient and nonsalient ironies retained their contextually
incompatible literal meaning in spite of the availability of the ironic meaning
(that emerged at a different temporal stage for the two types of irony). These
findings support the indirect negation view (Giora 1995) which maintains
that the contextually incompatible literal meaning of irony should be retained
because it has a role in irony interpretation - it provides a reference point
relative to which the ironicized situation is evaluated. Making the expected
explicit allows for the computation of the difference between what is and
what is looked for.
Spontaneous face-to-face talk can lend support to the retention hypothesis
(Giora 1995, 2003; Giora and Fein 1999a) if it is found to abound in ironic
utterances that get responded to literally. Indeed, having investigated irony
reception in a spontaneous environment, we found that more often than not,
irony is responded to by resonating with its salient, literal interpretation.
These findings corroborate those of Kotthoff (1998), who shows that in
friendly conversations, listeners very often respond to the literal meaning of
the ironic utterance while at the same time making it clear that they have
also understood the implicated meaning. Responding to the literal meaning
demonstrates that this contextually incompatible meaning has neither been
inhibited nor suppressed by contextual infonnation.
Empirical research supporting the graded salience hypothesis (e.g. Giora
1998a, 1998b, 1999, 2003; Giora and Fein 1999a, 1999b, 1999c; Giora, Fein
and Schwartz 1998; Pexman, Ferretti and Katz 2000; Schwoebel, Dews,
Winner and Srinivas 2000) and indirect negation view (Giora 1995; Giora,
Fein and Schwartz 1998) has so far focused on the processes involved in
comprehension of written, often contrived discourses tested in the laboratory.
In this study, we provide evidence in favour of the claim that salient meanings
are involved in spontaneous discourse. In particular, we demonstrate that
salient meanings are involved in text comprehension and production even
when they are incompatible with the context or the intended meaning.
 Irony in conversation: salience. role. and context effects 309

Acknowledgements

Support for this research was provided by grants from The Israel Science Foundation
and Lion Foundation to the first author. Thanks are also extended to Ray Gibbs,
John Du Bois and Brigitte Nerlich for their very helpful comments.

Notes

I. Even though Grice (1975) is not explicit about it, the processing model that
follows from his assumptions is taken to be a replacement or substitution account
(see, e.g .• Levinson 1983: 157).
2. Legend (following Du Bois et al. 1993):
half a second break
a shorter break
[] overlap
x unclear word
<xx> unclear utterance
@ laughter
() for words not appearing in the Hebrew text
underlinin~ ironic utterances
bold responses to the literal meaning

References

Bradley, Dianne C. and Kenneth I. Forster

1987 A reader's view of listening. Cognition 25(1-2): 103-134.
Burgess, Curt, Michael. K. Tanenhaus and Mark S. Seidenberg
1989 Context and lexical access: implication of nonword interference for
lexical ambiguity resolution. Journal of Experimental Psychology:
Learning. Memory and Cognition 15: 620--632.
Cairns, Helen Smith
1984 Current issues in language comprehension. In: Rita C. Naremore
(ed.), Recent Advances in Language Science. San Diego: College
Hill Press.
Carston, Robyn
1999 (June) Linguistic meaning and literal meaning. International Pragmatics
Conference on Pragmatics and Negotiation. Tel Aviv University and
The Hebrew University of Jerusalem.
310 Rachel Giora and Inbal Gur

Clark, Herbert, H.
1996 Using Language. Cambridge: Cambridge University Press.
Conrad, Carol
1974 Context effects in sentence comprehension: a study of the subjective
lexicon. Memory and Cognition 2: 130-138.
Dews, Shelly and Ellen Winner
1997 Attributing meaning to deliberately false sentences: the case of irony.
In: Charlotte MandelI and Alyssa McCabe (eds.), The Problem of
Meaning: Behavioral and Cognitive Perspectives, 377-414. Amsterdam:
Elsevier.
Drew, Paul.
1987 Po-faced receipts of teases. Linguistics 25: 219-253.
Du Bois, John W.
1998 Towards a dialogic syntax. Ms. UC Santa Barbara.
Du Bois, John W., Stephan Schuetze-Coburn, Susanna Cumming and Danae Paolino
1993 Outline of discourse transcription. In: Jane A. Edwards and Martin
D. Lampert (eds.), Talking Data: Transcription and Coding in
Discourse Research, 45-87. HilIsdale, NJ: Lawrence Erlbaum
Associates.
Duffy, Susan A., Robin K. Morris and Keith Rayner
1988 Lexical ambiguity and fixations times in reading. Journal of Memory
and Language 27: 429-446.
Fodor, Jerry
1983 The Modularity of Mind. Cambridge: MIT Press.
Forster, Kenneth I.
1979 Levels of processing and the structure of the language processor. In:
William E. Cooper and E. C. T Walker (eds.), Sentence Processing:
Psycho linguistic Studies Presented to Merrill Garrett, 27-65.
HilIsdale, NJ: Erlbaum.
1989 Basic issues in lexical processing. In: William Marslen-Wilson (ed.),
Lexical Representation and Process, 75-107. Cambridge: MIT Press.
Gernsbacher, A. Morton
1990 Language Comprehension as Structure Building. Hillsdale NJ:
Erlbaum.
Gernsbacher, A. Morton, Boaz Keysar, Rachel W. Robertson and Necia K. Werner
200 I The role of suppression and enhancement in understanding metaphors.
Journal of Memory and Language 45: 433-450.
Gernsbacher, A. Morton and Rachel W. Robertson
1995 Reading skill and suppression revisited. Psychological Science 6:
165-169.
Gibbs, Raymond w., Jr.
1980 Spilling the beans on understanding and memory for idioms in
conversation. Memory and Cognition 8: 449-456.
 Irony in conversation: salience. role. and context effects 311

1986 On the psycholinguistics of sarcasm. Journal of Experimental

Psychology: General 115: 3-15.
1994 The Poetics of Mind. Cambridge: Cambridge University Press.
2002 A new look at literal meaning in understanding what is said and
implicated. Journal of Pragmatics 34: 457-486.
Gildea, Patricia and Sam Glucksberg
1983 On understanding metaphor: the role of context. Journal of Verbal
Learning and Verbal Behavior 22: 577-590.
Giora, Rachel
1995 On irony and negation. Discourse Processes 19: 239-264.
1997 Understanding figurative and literal language: the graded salience
hypothesis. Cognitive Linguistics 7: 183-206.
1998a When is relevance? On the role of salience in utterance interpretation.
Revista Alicantina de Estudios Ingleses 11: 85-94.
1998b Irony. In: lan-Ola Ostman, lefVerschueren, lan Blommaert and Chris
Bulcaen (eds.), Handbook of Pragmatics, 1-21. Amsterdam: lohn
Benjamins.
1999 On the priority of salient meanings: studies of literal and figurative
language. Journal of Pragmatics 31: 919-929.
2003 On our Mind: Salience. Context and Figurative Language. New York:
Oxford University Press.
Giora, Rachel and Ofer Fein
1999a Understanding familiar and less familiar figurative language: the
graded salience hypothesis. Journal of Pragmatics 31: 1601-1618.
1999b Irony interpretation: the graded salience hypothesis. Humor 12:
425-436.
I 999c Irony: context and salience. Metaphor and Symbol 14: 241-257.
Giora. Rachel, Ofer Fein and Tamir Schwartz
1998 Irony: graded salience and indirect negation. Metaphor and Symbol
13: 83-101.
Glucksberg. Sam. Patricia Gildea and Howard Bookin
1982 On understanding non literal speech: can people ignore metaphors?
Journal of Verbal Learning and Verbal Behavior 21: 85-98.
Glucksberg, Sam and Boaz Keysar
1990 Understanding metaphorical comparisons: beyond similarity.
Psychological Review 97: 3-18.
Glucksberg, Sam. Roger Kreuz and Susan H. Rho
1986 Context can constrain lexical access: implications for models of
language comprehension. Journal of Experimental Psychology:
Learning. Memory. and Cognition 12: 323-335.
Gorfein, S. David (ed.)
1989 Resolving Semantic Ambiguity. New York: Springer-Verlag.
312 Rachel Giora and Inbal Gur

Grice, H. Paul
1975 Logic and conversation. In: Peter Cole and Jerry Morgan (eds.),
Speech Acts. Syntax and Semantics Volume 3, 41-58. New York:
Academic Press.
Hogaboam, Thomas W. and Charles A. Perfetti
1975 Lexical ambiguity and sentence comprehension. Journal of Verbal
Learning and Verbal Behavior 14: 265-274.
Horton, WiIliam S. and Boaz Keysar
1996 When do speakers take into account common ground? Cognition 59:
91-117.
Inhoff, Alber Werner, Susan D. Lima and Patrick J. CarroIl
1984 Contextual effects on metaphor comprehension in reading. Memory
and Cognition 12: 558-567.
Jones, Janet Lee
1991 Early integration of context during lexical access of homonym
meanings. Current Psychology: Research and Review 10: 163-181.
Kemper, Susan
1981 Comprehension and interpretation of proverbs. Journal of Psycho-
linguistic Research 10: 179-183.
Keysar, Boaz
1989 On the functional equivalence of literal and metaphorical interpre-
tations in discourse. Journal of Memory and Language 28: 375-385.
Keysar, Boaz, Dale J. Barr, Jennifer A. Balin and Timothy S. Paek
1998 Definite reference and mutual knowledge: process models of common
ground in comprehension. Journal of Memory and Language 39:
1-20.
Keysar, Boaz, Dale J. Barr and William S. Horton
1998 The egocentric basis of language use: insights from a processing
approach. Current Directions in Psychological Sciences 7: 46--50.
Kotthoff, Helga
1998 Irony, quotation, and other forms of staged intertextuality: double or
contrastive perspectivation in conversation. Interaction and Linguistic
Structures 5: 1-27.
Levinson, Stephen
1983 Pragmatics. Cambridge, UK: Cambridge University Press.
Lucas, Margery M.
1987 Context effects on the processing of ambiguous words in sentence
context. Language and Speech 30: 25-46.
Martin, Charles, Hoang Vu, George KeIlas and Kimberly Metcalf
1999 Strength of discourse context as a determinant of the subordinate
bias effect. The Quarterly Journal of Experimental Psychology 52A:
813-839.
 Irony in conversation: salience, role, and context effects 313
Onifer, William and David A. Swinney
1981 Accessing lexical ambiguities during sentence comprehension: effects
of frequency of meaning and contextual bias. Memory and Cognition
9: 225-236.
Ortony. Andrew, Diane L. Schallert, Ralph E. Reynolds and Stephen J. Antos
1978 Interpreting metaphors and idioms: some effects of context on
comprehension. Journal of Verbal Learning and Verbal Behavior 17:
465-477.
Pexman Penny, Todd Ferretti and Albert Katz
2000 Discourse factors that influence irony detection during on-line reading.
Discourse Processes 29: 201-222.
Picoult, Jonathan and Marcia K. Johnson
1992 Controlling for homophone polarity and prime target relatedness in
the cross-modal lexical decision task. Bulletin of the Psychonomic
Society 30: 15-18.
Rayner, Keith and Lyn Frazier
1989 Selection mechanisms in reading lexically ambiguous words. Journal
of Experimental Psychology: Learning, Memory, and Cognition 15:
779-790.
Rayner, Keith and Robin K. Morris
1991 Comprehension processes in reading ambiguous sentences: reflections
from eye movements. In: Greg B. Simpson (ed.), Understanding
Word and Sentence, 175-198. Amsterdam: North Holland.
Rayner, Keith, Jeremy M. Pacht and Susan A. Duffy
1994 Effects of prior encounter and global discourse bias on the processing
of lexically ambiguous words: evidence from eye fixations. Journal
of Memory and Language 33: 527-544.
Schvaneveldt, Roger W., David E. Meyer and Curt is A. Becker
1976 Lexical ambiguity, semantic context, and visual word recognition.
Journal of Experimental Psychology: Human Perception and Perfor-
mance 2: 243-256.
Schwoebel, John, Shelly Dews, Ellen Winner and Kavitha Srinivas
2000 Obligatory processing of the literal meaning of ironic utterances:
further evidence. Metaphor and Symbol 15: 47-61.
Searle, John
1979 Expression and Meaning. Cambridge: Cambridge University Press.
Seidenberg, Mark S., Michael K. Tanenhaus, James M. Leiman and Marie Bienkowski
1982 Automatic access of the meaning of ambiguous words in context:
some limitations of knowledge based processing. Cognitive Psychology
14: 489-537.
Sereno C. Sara, Jeremy M. Pacht and Keith Rayner
1992 The effect of meaning frequency on processing lexically ambiguous
314 Rachel Giora and Inbal Gur

words: evidence from eye fixations. Psychological Science 3:

269-300.
Shen, Yeshayahu
1997 Metaphors and global conceptual structures. Poetics 25: 1-17.
Simpson, Greg B.
1981 Meaning dominance and semantic context in the processing of lexical
ambiguity. Journal of Verbal Learning and Verbal Behavior 20:
120--136.
1994 Context and the processing of ambiguous words. In: A. Morton
Gernsbacher (ed.), Handbook of Psycholinguistics, 359-374. San
Diego, CA: Academic Press.
Simpson, Greg B. and Curt Burgess
1985 Activation and selection processes in the recognition of ambiguous
words. Journal of Experimental Psychology: Humo.n Perception and
Performance 11: 28-39.
Simpson, Greg B. and Mollie R. Foster
1986 Lexical ambiguity and children's word recognition. Developmental
Psychology 22: 147-154.
Simpson, Greg B. and Merilee A. Krueger
1991 Selective access of homograph in sentence context. Journal of Memory
and Language 30: 627-643.
Small, Steven I., Garrison W. Cottrell and Michael K. Tanenhaus
1988 Lexical Ambiguity Resolution: Perspectives from Psycho linguistics,
Neuropsychology, and Artificial Intelligence. San Mateo, CA: Morgan
Kaufmann.
Sperber, Dan and Deirdre Wilson
1986/1995 Relevance: Communication and Cognition. Oxford: Blackwell.
Swinney, David A.
1979 Lexical access during sentence comprehension: (re)consideration of
context effects. Journal of Verbal Learning and Verbal Behavior 18:
645-659.
Swinney, David A. and Penny Prather
1989 On the comprehension of lexical ambiguity by young children: inves-
tigations into development of mental modularity. In: David S. Gorfein
(ed.), Resolving Semantic Ambiguity, 225-238. New York: Springer-
Verlag.
Tabossi, Patrizia
1988 Accessing lexical ambiguity in different types of sentential contexts.
Journal of Memory and Language 27: 324-340.
Tanenhaus, Michael K., Greg N. Carlson and Mark S. Seidenberg
1985 Do listeners compute linguistic representations? In: David Dowty,
Lauri Kartunnen and Arnold M. Zwicky (eds.), Natural Language
 Irony in conversation: salience, role, and context effects 315

Parsing: Psychological, Theoretical and Computational Perspectives.

New York: Cambridge University Press.
Tanenhaus, Michael K., James M. Leiman and Mark S. Seidenberg
1979 Evidence for multiple stages in the processing of ambiguous words
in syntactic contexts. Journal of Verbal Learning and Verbal Behavior
18: 161-167.
Till, Robert E., Ernest F. Mross and Waiter Kintsch
1988 Time course of priming for associate and inference words in a dis-
course context. Journal of Verbal Learning and Verbal Behavior 16:
283-298.
Titone, Debra
1998 Hemispheric differences in context sensitivity during lexical ambiguity
resolution. Brain and Language 65: 361-394.
Vu, Hoang, George Kellas and Stephen T. Paul
1998 Sources of sentence constraint in lexical ambiguity resolution. Memory
and Cognition 26: 979-1001.
West, Richard F. and Keith E. Stanovich
1988 How much of sentence priming is word priming? Bulletin of the
Psychonomic Society 26: 1-4.
Williams, John N.
1992 Processing polysemous words in context. Evidence from interrelated
meanings. Journal of Psycholinguistic Research 21: 193-218.
Zaidel, Eran
1987 Hemispheric monitoring. In: David Ottoson (ed.), Duality and Unity
of the Brain, 247-281. Hampshire: Macmillan.
Young children's and adults' use of figurative
language: how important are cultural and
linguistic influences?

Ann Dowker

For many years, it has been known that children - and adults - do not use
language only for communication, or for referring to objects and events in
the environment. Children's playful and nonliteral uses of language have
been the subject of study at least since the end of the 19th century (e.g.
Lukens 1894; Chamberlain and Chamberlain 1904; Trettian 1904; Jespersen
1922; Chukovsky 1925 [1968]; Weir 1962; Aimard 1977; Garvey 1977;
Iwamura 1980; Sutton-Smith 1980; Schwartz 1981; Kuczaj 1982, 1983; Fox
1983,1993; Schieffelin 1983; Nelson 1987; Dowker 1989,1991,1998; Joffe
and Shapiro 1991; lohnson 1991; Crystal 1998). These non literal uses of
language are very diverse, but one important feature that they share is their
metalinguistic nature: they involve attention to and manipulation of aspects
of language, rather than treating language as a means to the end of trans-
mitting information. Children manipulate linguistic patterns and relationships,
including phonological relationships (e.g. rhyme and alliteration); syntactic
relationships (e.g. "modified repetition" as defined by Dowker 1991) and
semantic relationships (e.g. simile and metaphor).
Metaphor and other forms of figurative language have been the object of
considerable interest for researchers: partly because of their importance in
the poetry and other literature - oral and written - of seemingly all cultures
(Finnegan 1977); partly because of their relevance to the ways in which
people form conceptual categories.
Metaphor subsumes a variety of devices. Simile and metaphor are disting-
uished for several literary and psychological purposes (Kennedy and Chiappe
1999); and some researchers (Happe 1991; Nerlich, Todd and Clarke 1998)
suggest that true metaphor emerges later and makes greater demands on
theory of mind than does simile; but simile will here be treated as just one
form of metaphor.
There have been numerous studies of the development and early use of
metaphor (Gardner et al. 1978; Billow 1981; Fourment, Emmenecker and
318 Ann Dowker

Pantz 1987; Gentner, Falkenhainer and Skorstad 1988; Winner 1988; Caramelli
and Montanari 1995).
Most of these studies suggest that some forms of metaphor begin very
early. Billow (1981) found that it occurred frequently in the spontaneous
play and conversation of children from 2112 to 6; and that it did not increase
in frequency with age within that age range.
Gardner et al. (1978) and Winner (1988) also report extensive use of
metaphor during the preschool period; though unlike Billow they found
significant age differences. Metaphor increased during the preschool period
and then declined with the onset of a "literal stage" at the age of about 6
or 7. It was suggested that during this period, children become more aware
of linguistic rules and category boundaries and are reluctant to violate them.
By adolescence, these rules and boundaries are more firmly defined, and the
children are in complete control of them, and thus may be more ready to
violate them for particular purposes. This may parallel the U-shaped rel-
ationships between expertise and strategy variability that Dowker et al. (1996)
have proposed for other domains such as mathematics.
Later studies have mostly supported the view that young children are
capable of comprehending and using metaphors, though there are some age
differences in their frequency and predominant types. Numerous studies
(Gardner et al. 1978; Dowker 1986; Gentner, Falkenhainer and Skorstad
1988) indicate that young children tend to use predominantly perceptual meta-
phors, e.g. They [children walking around the school] sound like horses. Older
children and adults use a relatively larger number of cross-sensory metaphors,
e.g. Her dress is so loud that it shouts, and psychologicaVphysical metaphors,
e.g. fear like a seep of water (from a poem by "Emma", quoted by Dowker,
Hermelin and Pring 1996), and I grasp your meaning. For example, Caramelli
and Montanari (1995) studied Italian children's use of animal terms as meta-
phors and found that they moved from basing metaphors predominantly on
visual resemblance at age 6 to basing them predominantly on moral judge-
ment at age 12.
There also appear to be age differences in the contexts in which people
use metaphor, though this has not been studied as systematically. Children
frequently use metaphor in the context of pretend play (Stross 1975; Billow
1981); while adults associate it strongly with poetry: an association that is
possibly less strong for young children (Dowker 1986; Dowker et al. 1998).
When researchers attempt to distinguish between literal and figurative,
including metaphorical, language, the distinction is often difficult to make.
 Young children's and adults' use offigurative language 319

Fass (1997: 2) points out that "it has proved extremely hard to develop precise
criteria for distinguishing literal from non literal language". Lakoff and Johnson
(1980) and Gibbs (1994) put forward the view that, in both children and
adults, the distinction between figurative and literal language is less sharp
than has sometimes been assumed. Gibbs (1994: 435-436) stated that "similar
cognitive mechanisms drive our understanding of both literal and figurative
speech"; that "people need not recognize figurative language as violating
communicative norms and maxims in order to understand what those expres-
sions figuratively mean"; and that "a great deal of our thinking is constituted
by metaphorical mappings from dissimilar source and target domains".
Examples of the latter include thinking of understanding as seeing (That's
very clear; I've got the picture; etc.) or anger as heated fluid in a container
(He got all steamed up; I was boiling over; She hit the ceiling).
Cross-cultural differences in metaphor use have only been studied to a
limited extent from a psychological point of view, though they have also
received attention from some other perspectives, e.g. the difficulties that they
pose for translators. The ways in which metaphor is used are certainly
influenced by pragmatics and cultural convention: for example, a British
person would use the term fox to mean 'sly and cunning person', while a
Canadian might use the same term to mean 'attractive woman'.
There have been some studies of the ways in which differences in meta-
phoric usage may reflect cultural differences in attitudes to and concepts of
certain emotional, social and cognitive domains, such as anger (Gibbs 1994;
Koevecses 20ooa), emotions generally (Palmer, Bennett and Stacey 1999;
Koevecses 2000b), and time (Zhou and Huang 2000; Moore 2001). Such
studies have revealed considerable commonalities, revealing universal cog-
nitive structures; for example both English and Wolof speakers have a
tendency to map spatial vocabulary onto temporal concepts (Moore 2001).
However, cultural differences are also found, and Moore concludes that
"While there is a substantial amount of metaphor structure that is shared
cross-linguistically, a full understanding of conceptual metaphor depends on
properties of particular languages, communities of speakers, or individuals".
Sometimes, even when a given type of metaphorical construction is found
generally across cultures, it is given more emphasis in some cultures than
others. For example, Yu (2001) points out that terms for the face are used
metaphorically in both English and Chinese to refer to broader aspects of
physical appearance (she's not just a pretty face), to emotion and character
(we mustface up to this), to interactions and relationships (we had aface-to-
320 Ann Dowker

face discussion), and to prestige and dignity (he lost face). However, such
metaphors were used more richly in Chinese than English, reflecting cultural
differences in the values attributed to the relevant concepts.
Far fewer studies have been carried out on the ways in which language or
culture may affect the actual frequency of figurative language.
Some years ago, I carried out an analysis of spontaneous poems produced
by children between the ages of 2 and 6 (Dowker 1986). The sources included
Timothy Rogers' book Those First Affections, his extensive unpublished
collection of poems by very young children, other published sources (e.g.
Trettien 1894; Dixon 1930; Britton 1970; Griffin 1981; Heath 1983; etc.),
and numerous personal communications from friends and acquaintances. In
all, there were nearly 400 poems by English children: 398 to be precise,
including 26 by 2-year-olds, 57 by 3-year-olds, 73 by 4-year-olds, 99 by 5-
year-olds, 99 by 6-year-olds, and 44 by nursery-school children of unspecified
age.
The English poems were compared with poems by French, Hebrew and
Russian-speaking children of similar age-range. The 37 French poems were
taken from various published sources (e.g. Aimard 1977; Chevalley 1982).
Most of the 66 Russian poems come from Chukovsky (1970) with a few
from other published sources. Some of the 19 Hebrew poems were unpub-
lished and sent to me by friends and acquaintances; the rest come from
Goldberg (n.d.) and Rivkai (1937).
The findings with regard to metaphor were as follows. One-sixth of the
English poems contained metaphor. There was some increase with age from
less than 10% of poems by 2- and 3-year-olds to 22% - just over a fifth - of
poems by 6-year-olds. In all, 78 metaphors were produced. Most of the
metaphors were similes; and about two-thirds of them were perceptually
based; e.g. When we walk around we look like shadows; The moon is just
like a little white bird/Except that it hasn't got a little face; This man was so
tall as the clouds; The sun is like a stove/Boiling water; Hands as soft as
seal. By contrast, only 7% were functionally or behaviourally based.
The poems by French children were more likely than those of the English
children to contain metaphor. One third of their poems did so. Over half of
the French metaphors were perceptually based (e.g. Lune d'argent ... Etoiles
d'or [Moon of Silver ... Stars of gold]), though a somewhat higher proportion
of the French metaphors did come into other categories, e.g. Les oiseaux
sont mignons/MalheureuxlComme les pauvres gens [The birds are sweetJ
UnluckylLike poor people].
 Young children's and adults' use o/figurative language 321

If metaphors were commoner in the French than the English poems, they
were much less common in the Russian and Hebrew poems. Only 8% of the
Russian poems and none of the 19 Hebrew poems contained any metaphor.
The study of children's spontaneous poems is important in showing what
they produce when not being actively directed by adults. However, there are
some disadvantages in confining oneself to the study of spontaneous poems.
They had to be noted down by adults, which may lead to selection bias:
adults may choose to record poems which they consider to be particularly
good or particularly funny or just particularly interesting. Moreover, these
poems were produced over a very wide timescale, from the 1890s to the 1980s,
and it is possible that children's poems could have changed somewhat over
this period, perhaps in response to changes in children's literature. Therefore,
a somewhat more controlled study was carried out, which involved eliciting
poems from children.
Dowker et al. (1998) carried out a cross-linguistic study of poem pro-
duction by young children. The participants were 122 English children, 59
French children, 148 Italian children, 118 Polish children and 118 Brazilian
(Portuguese-speaking) children between the ages of 4 and 6 years.
The basic task involved the successive presentation of three pictures. After
each picture was presented, the child was asked: "Could you tell me a story
about this picture?" and hislher response was recorded on a tape-recorder
and then played back to the child. The child was then told: "Now I'm going
to tell you a poem, which is a bit like a story but not quite. And I'd like you
to make up something like that." One of three poems dealing with the picture
(a rhyming poem, an alliterative poem and a simile poem) was then presented
as a stimulus, and the child was asked: "Can you make up something like
that?" Once again, the child's response was recorded and then played back
to himlher on the tape-recorder. The same procedure was followed with the
next two pictures. (The instructions were, of course, translated into the
relevant languages for the non-English children.) During this task, each child
heard one rhyming poem, one alliterative poem and one simile poem.
Differences between the procedure as applied to the different groups were
(i) that of necessity the stimulus poems were different, as the children of
each group heard stimulus poems in their own language; (ii) the pictures
sometimes differed between languages, as they had to permit the production
of stimulus poems with the appropriate devices in a particular language.
There were some differences in the use of phonological devices. The
Italian children made extensive use of both rhyme and alliteration; while
322 Ann Dowker

the Polish children made considerable use of alliteration but less use of
rhyme, and the English children used much rhyme but little alliteration. The
French and Brazilian groups were the only groups in the study to use phono-
logical devices in less than half of their poems. However, by far the most
marked differences were in the use of simile and metaphor. Metaphor, almost
always consisting of simile, was in general less common than the phonological
devices; but there were extremely striking group differences. Only a tiny
proportion of English, Italian or Polish children produced poems containing
similes, whereas similes were much more common in the French and Brazilian
poems.
It was inappropriate to investigate age differences in the use of simile in
the English, Italian or Polish group, because of the very small numbers
involved. They were nonsignificant in the Brazilian group <X2 = 0.45; df = 2;
p n.s.). In the French group, there was a significant increase with age <X2 =
8.82; df = 2; p < 0.05).

The following are some examples of poems containing similes:

(I) English poem by a girl of 4 years 6 months:

It was quiet as the night.

The rabbits were running across the street faster than the wind.
They were hopping around so fast.
We saw them run so fast
That the badgers watched us.
The trees were as big as a girl.
And they could fall down
Across the stream
Into the woods of a chipmunk.

(2) French poem by a 6-year-old boy:

Le cochon est gros comme une armoire.

[The pig is as fat as a wardrobe.]

(3) Brazilian poem by a girl of 5 years 5 months:

o gatinho e branquinho como nuvem.

o gatinho e branquinho como 0 papel.
 Young children's and adults' use a/figurative language 323

o gatinho gost de jogar bolinha e come

Leite feito urn bebe.

[The kitten is white as clouds.

The kitten is white as paper.
The kitten likes to play with a ball, and drinks
Milk like a baby.]

Thus, contrary to what might have been predicted on the grounds that
there are much greater phonological than semantic differences between lang-
uages, group differences were greater with regard to the use of metaphor and
simile than with regard to the use of phonological devices. Whereas all the
groups did make significant use of phonological devices in their poems, three
of the five groups (English, Polish and Italian) used almost no metaphor in
their poems.
The findings that metaphor was used more often, and phonological devices
less often, by French than English children are in agreement with the results
of the study of spontaneous poems. The spontaneous poems were not elicited
by means of examples, making it unlikely that the cross-cultural differences
in the studies were the artifacts of a particular methodology.
Both studies suggested a possible trade-off between the use of phono-
logical devices and of semantic devices (e.g. simile and metaphor), in that
the groups (French and Brazilian) that made the least use of phonological
devices made the greatest use of metaphor. Dowker et al. (1998) also found
that French and Brazilian children's poems outnumbered their stories, whereas
the reverse was true of the other groups. This may indicate differences in the
nature and sharpness of the division between the genres of stories and poems.
Perhaps to French and Brazilian children, poems are "stories with rhythm",
featuring semantic devices. For the other groups, poems may be more sharply
distinguished from stories through their emphasis on phonological as opposed
to semantic devices.
Do the group differences in poem production reflect differences in actual
capacities for understanding or producing metaphor? Though it is difficult to
answer this question with certainty, it appears unlikely that this is the case.
Cross-cultural and cross-linguistic differences in metaphoric development
have not been much studied; but much of the research on metaphoric develop-
ment has been carried out with English-speaking children (e.g. Gardner et
al. 1978; Billow 1981), and there is ample evidence that such children are
capable of using metaphor and simile from an early age.
324 Ann Dowker

Thus, there is no evidence that the differences in metaphoric production are

caused by or related to an inability to use or detect metaphor in certain groups.
Studies of the nature and extent of cultural differences in metaphor use
bring up several issues. Metaphor may be influenced by nonverbal conceptual
knowledge, by attitudes and emotions and by pragmatic context, and all of
these may be influenced by both age and culture. In addition, however,
metaphor may be influenced by factors within language itself.
As regards relationships between metaphor use and language development,
there has been considerable discussion of the relationship between early
metaphor use, and early literally-based overextensions of words. Numerous
researchers (e.g. Vosniadou 1987) have warned against excessive readiness
to label young children's utterances as metaphors, on the grounds that some
"metaphors" are overextensions of meaning due to a lack of precise knowl-
edge of word meanings, similar to the behaviour of an 18-month-old child
who has just learned the word "doggie" and applies it to all four-legged
animals (for discussion of this problem, see Vosniadou 1987). Gardner et al.
(1978), Billow (1981) and others have attempted to guard against this possi-
bility by only including renamings as "metaphors" if the children showed
evidence of knowledge of the correct name for the renamed object, and of
correct use of the term used in renaming.
Attempts to distinguish metaphor from overextension rest on the assump-
tion that at least some overextension is due to a genuine belief that the term
applies to the objects to which it is extended (or, at least, to the lack of a
clear understanding that it is). However, overextensions in production need
not be due to overextensions in comprehension (Fremgen and Fay 1980).
Young children have a limited vocabulary. Despite realizing that a certain
word is inappropriate to a certain referent, whose name (s)he does not know,
a child may use the word because it is the most appropriate term that (s)he
has available. Nelson et al. (1978) argue that overextensions based on percep-
tual similarities are really statements of analogy. The 18-month-old child,
who is not yet capable of making a statement such as "This grapefruit looks
like the moon" or even "This thing - I don't know it's name -looks like the
moon" may point to a grapefruit and say: "Moon". Such utterances, according
to Nelson et aI., are a part ofthe child's emerging fascination with similarities
and differences, which plays an important part in the early formation of
categories.
If a child uses an inappropriate term, not because of overextension in
comprehension, but as an imperfect substitute for the more appropriate term
 Young children's and adults' use o/figurative language 325

that (s)he lacks, then (s)he is doing one of the things that adults do when
they use metaphor: expressing "the otherwise inexpressible". The adult who
describes a state of depression as "a fog surrounding me", or who says "My
love is like a red red rose" is presumably using the figurative expressions
because (s)he cannot convey the meaning equally well by means of literal
description. Indeed, Carabine (1991) has pointed out that both children and
adults have "fuzzy boundaries" to the extensions of common word meanings,
and that there is far more similarity between children and adults in their
word extensions than is often recognized.
When word meaning extensions are or become standardized within a
language, rather than idiosyncratic to a particular individual or occasion, they
may be described as polysemy.
Polysemy has been defined (e.g. Ullman 1951) as the property of a word
that has multiple but related meanings (e.g. head can mean 'upper part of the
body', 'chief person in an organization', etc.). It is sometimes difficult to
distinguish it from homonymy (the property of a word that has multiple unrel-
ated meanings; e.g. bank can mean 'edge of a river' or 'financial institution ').
Some linguists (Ullman 1951; Kinberg 1991) have suggested a continuum
from figurative uses through polysemy to homonymy. Taking into account
the above-mentioned findings concerning children's early semantic develop-
ment, the following continuum is here proposed: overextension - figurative
uses - polysemy - homonymy.
Such a view implies a close relationship between polysemy and figur-
ative language; and indeed polysemy may be both a consequence and cause
of figurative language use. Many of the extra, polysemous meanings that a
word develops begin as metaphorical extensions of a primary meaning of
that word: e.g. head in the sense of 'chief person' presumably originated as
a metaphorical extension of head in its primary sense of 'upper part of the
body'. This is not to say that all polysemous words have one clearly definable
primary meaning; this is obviously not the case (Gibbs 1994). Other forms
of figurative language such as metonymy, a form of indirect reference in
which one word is replaced by a word referring to an entity closely associated
with it, are also the sources of polysemous meaning (e.g. 'face' for 'person').
Polysemy may also influence non literal language use: especially in puns
and riddles (Nerlich, Todd and Clarke 1998), but also metaphors and similes.
Since there appear to be cross-linguistic differences in metaphor use, and
since polysemy and metaphor seem to be closely related, are there also
cross-linguistic differences in polysemy? Do some languages have a greater
326 Ann Dowker

incidence of polysemy; a greater number of meanings for the same word

than do others? Sinha and Kuteva (1995) suggest that spatiallocatives, such
as on and in are polysemous in most or all languages, but are more so in
some languages than in others. Hunt and Agnoli (1991) propose that English
is more polysemous than Italian, and that Chinese and Japanese are more
polysemous than either. This conclusion was, however, based on investigation
of a relatively small number of words, e.g. 30 in English and 30 in Italian.
If there are cross-linguistic differences in polysemy, could this lead to
cross-linguistic differences in the use of metaphor? It is difficult to answer
this question, partly because of ambiguities in interpreting the direction of
causation, and partly because such effects could be in either direction. If a
culture has a strong tendency to use metaphoric expressions, then some of
these could become completely or partially standardized, increasing the
amount of polysemy in the language. On the other hand, explicit use of
metaphor could serve as an alternative to polysemy.
According to Deane (1988: 325) "Human thought displays two com-
plementary characteristics. While it displays flexible responses to novel situ-
ations, it is also highly structured, incorporating detailed information about
the world" and (p. 358), " ... viewed in cognitive terms, polysemy is a natural
consequence of the interplay of cognitive flexibility and structure".
Figurative linguistic extension, literal linguistic overextension and poly-
se my overlap considerably with one another, and all appear to depend on the
combination of flexibility and structure. In these respects, they do indeed
resemble cognitive processes in numerous other domains. Arithmetic, for
example, is a highly structured domain; but it also involves considerable
flexibility and within-individual and between-individual variability of strategy
use (Krutetskii 1976; Siegler and Jenkins 1989; Dowker et al. 1996; Siegler
1996; Baroody and Dowker 2003). A similar combination of flexibility and
patterned structure characterizes children's drawing (All and 1983; Karmiloff-
Smith 1990).
Nerlich, Todd and Clarke (1998: 361) propose that "the ability to use
polysemy in both metaphor and jokes facilitates the ability to play different
roles in life and see things from different viewpoints and vice versa". They
point out that both metaphor comprehension and the ability to form meta-
representations are impaired in autism; it might be added that autistic indi-
viduals tend to be inflexible in their strategy use in many domains, and in
language in particular.
The whole area of the relationship, especially in development, between
 Young children's and adults' use offigurative language 327

figurative language and polysemy is still little understood. Fragments have

been studied in isolation, but much research is still needed to bring the
fragments together. We may conclude that:

(i) Understanding of polysemous meanings of words improves with age.

(ii) Metaphoric production and comprehension begin very early in child-
hood, and become more sophisticated in adolescence and adulthood,
possibly after a temporary dip in middle childhood (Gardner et al.
J978; Winner 1988).
(iii) Polysemy, whether in the language development of an individual child
or the historical development of a language, is in part based on meta-
phoric extension (Nerlich and Clarke 1997; Nerlich, Todd and Clarke
1998).
(iv) The existence of polysemy facilitates the use of certain forms of
figurative language.
(v) Most crucially, there are no clear boundaries between polysemy, over-
extension and figurative language (Nelson et al. 1978; Gibbs 1994).

Clear-cut examples of metaphor, overextension, or literal polysemy without

metaphoric overtones are probably rare; the categories share many character-
istics and typically show significant overlap. The apparent divisions between
them are due in part to the fact that they have frequently been studied
separately, by researchers in different disciplines and subspecialities. Polysemy
has tended to be the province of linguists and artificial intelligence researchers;
overextension the province of developmental psychologists studying early
language development; and metaphor the province of developmental psychol-
ogists studying somewhat older children, of cognitive psychologists, and of
students of literature. Moreover, metaphor and other forms of figurative
language are frequently regarded as specialized poetic devices; though there
is considerable evidence not only that everyday language is often figurative
(Gibbs 1994) but that metaphor is less strongly associated with poetry in the
early stages of its development than later on. Young children produce both
poems and metaphors, but not necessarily in conjunction (Dowker et al. 1998).
However, one crucial gap in our knowledge is the question of the existence
and extent of cross-linguistic differences in the frequency of use of either
polysemy or metaphor. Even as regards age differences, although, as discussed
above, some research has been done on developmental changes in metaphoric
comprehension and production, very little research has been carried out on
328 Ann Dowker

the development and use of polysemy. Only when both the developmental
and linguistic influences on both polysemy and metaphor are better under-
stood can we begin to fully understand the nature of these phenomena. and
the undoubtedly close relationships between them.

References

Aimard, Paul
1977 Les Jeux de Mots de l'Enfant. Villeurbanne: Simep-Editions.
Alland, Alexander
1983 Playing with Form: Children Draw in Six Cultures. New York: Columbia
University Press.
Baroody, Arthur and Ann Dowker (eds.)
2003 The Development of Arithmetical Concepts and Skills. Mahwah, NI:
Erlbaum.
Billow, Richard
198 I Observing spontaneous metaphor in children. Journal of Experimental
Child Psychology 3 I: 430-445.
Britton, lames
1970 Language and Learning. London: Alien Lane.
Carabine, B.
1991 Fuzzy boundaries and the extension of object-words. Journal of Child
Language 18: 355-372.
Caramelli, Nicoletta and A. Montanari
1995 Animal Terms in Children's Metaphors. Lund University, Sweden:
Cognitive Science Research.
Chamberlain, Alexander and Isabel Chamberlain
1904 Studies of a child, I and n. Pedagogical Seminary 9: 264-291 and
452-483.
Chevalley, Etienne
1982 Miracles de l'Enfance. Paris: Guilde du Livre.
Chukovsky, Kornei
1925 [1968] From 'TWo to Five. Berkeley: University of California Press. 1968
translation by Miriam Morton of 1959 edition. First Russian edition
published in 1925.
1970 [1925] Ot Dvukk do Pyati [From Two to Five]. Moscow: NarodnayaAsveta.
Crystal, David
1998 Language Play. Harmondsworth: Penguin.
Deane, Paul
1988 Polysemy and cognition. Lingua 75: 325-361.
 Young children's and adults' use offigurative language 329

Dixon, Catherine
1930 Children Are Like That. New York: John Day.
Dowker, Ann
1986 Language play in young children. Ph.D. thesis. University of London
Institute of Education.
1989 Rhyme and alliteration in poems elicited from young children.
Journal of Child Language 16: 181-202.
1991 Modified repetition in poems elicited from young children. Journal
of Child Language 18: 625~39.
Dowker, Ann (ed.)
1998 Language play in children. Special Issue of Cahiers de Psychologie
Cognitive/Current Psychology of Cognition 17.
Dowker, Ann, Amanda Flood, Helen Griffiths, Louise Harriss and Lisa Hook
1996 Estimation strategies of four groups. Mathematical Cognition 2:
113-135.
Dowker, Ann, Beate Herrnelin and Linda Pring
1996 A savant poet. Psychological Medicine 26: 913-924.
Dowker, Ann, Grazyna Krasowicz, Giuliana Pinto, Antonio Roazzi and Anita Smith
1998 Phonological and semantic devices in very young children: a cross-
cultural study. Cahiers de Psychologie Cognitive/Current Psychology
of Cognition 17: 389-416.
Fass, Dan
1997 Processing Metonymy and Metaphor. London: Ablex.
Finnegan, Ruth
1977 Oral Poetry. Cambridge: Cambridge University Press.
Fourment, Marie-Claude, Nelly Emmenecker, and Valerie Pantz
1987 Etudes de la production de metaphores chez les enfants de 3 a 7 ans.
Annee Psychologique 87: 535-551.
Fox, Carol
1983 Talking like a book: young children's oral monologues. In: Margaret
Meek (ed.), Opening Moves: Work in Progress in the Study of
Children's Language Development. University of London Institute
of Education: Bedford Way Papers.
1993 At the Very Edge of the Forest: The Influence of Literature on Story-
telling by Children. London: Cas sell.
Fremgen, Amy and David Fay
1980 Overextensions in production and comprehension: a methodological
clarification. Journal of Child Language 7: 205-211.
Gardner, Howard, Ellen Winner, Robin Bechhoffer and Dennie Wolf
1978 The development of figurative language. In: Katherine Nelson (ed.),
Children's Language, Vo!. 1. Cambridge, MA: Gardner Press Inc.
330 Ann Dowker

Garvey, Catherine
1977 Play with language and speech. In: Susan Ervin-Tripp and Claudia
Mitchell-Kernan (eds.), Child Discourse. Academic Press.
Gentner, Dedre, Brian Falkenhainer and Janice Skorstad
1988 Viewing metaphor as analogy. In: David Helman (ed.), Analogical
Reasoning. London: Kluwer.
Gibbs, Raymond w., Jr.
1994 The Poetics of Mind. Cambridge: Cambridge University Press.
Goldberg, Lea
(not dated) Between a Children s Writer and his Reader. Tel Aviv: Sifriyat Hapoalim
(in Hebrew).
Griffin, Elinor
1981 Island of Childhood: Education in the Special World of Nursery
School. New York: Teachers' College Press.
Happe, Francesca
1991 The autobiographical writings of three Asperger syndrome adults:
problems of interpretation and implications for theory. In: Uta Frith
(ed.), Autism and Asperger Syndrome. Cambridge: Cambridge University
Press.
Heath, Shirley
1983 Ways with Words. Cambridge: Cambridge University Press.
Hunt, Earl and Franca Agnoli
1991 The Whortian hypothesis: a cognitive psychology perspective. Psycho-
logical Review 98: 377-389.
Iwamura, Susan
1980 The Verbal Games of Preschool Children. London: Croom Helm.
Jespersen, Duo
1922 Language: Its Nature. Development and Origin. London: Alien and
Unwin.
Joffe, Victoria and G. Shapiro
1991 Sound-based language play in four language-impaired subjects. The
South African Journal of Communication Disorders 38: 119-128.
Johnson, Janice
1991 Developmental versus language-based factors in metaphor interpre-
tation. Journal of Educational PsychoLogy 83: 470-483.
Karmiloff-Smith, Anneue
1990 Constraints on representational change: evidence from children's
drawings. Cognition 34: 57-83.
Kennedy, John and Dan Chiappe
1999 What makes a metaphor stronger than a simile? Metaphor and Symbol
14: 63-69.
 Young children's and adults' use offigurative language 331

Kinberg, Naphthali
1991 Figurative uses, polysemy and homonymy in systems of tense, mood
and aspect. Lingua 83: 319-338.
Koevecses, Zoltan
2000a The concept of anger: universal or culture-specific? Psychopathology
33: 159-170.
2000b Metaphor and Emotion: Language, Culture and Body in Human
Feeling. Cambridge: Cambridge University Press.
Krutetskii, Vadim
1976 The Psychology of Mathematical Abilities in Schoolchildren. London:
University of Chicago Press.
Kuczaj, Stan
1982 Language play and language acquisition. In: Hayne Reese (ed.),
Advances in Child Development and Child Behaviour 17. London:
Academic Press.
1983 Crib Speech and Language Play. Berlin: Springer-Verlag.
Lakoff, George and Mark 10hnson
1980 Metaphors We Live By. Chicago: University of Chicago Press.
Lukens, Herman
1894 Report on the learning of language. Pedagogical Seminary 3: 419-
460.
Moore, Kevin
2001 Spatial experience and temporal metaphors in Wolof: point of view,
conceptual mapping and linguistic structure. Dissertation Abstracts
International, Section A: Humanities and Social Sciences 61: 2685.
Nelson, Katherine (ed.)
1987 Narratives From the Crib. Cambridge, MA: Harvard University
Press.
Nelson, Katherine, Leslie Rescorla, lanice Gruendel and Helen Benedict
1978 Early lexicons: what do they mean? Child Development 49: 960-968.
Nerlich, Brigitte and David Clarke
1997 Polysemy: patterns of meaning and patterns of history. Historiographia
Linguistica 25: 349-385.
Nerlich, Brigitte, Zazie Todd and David Clarke
1998 The function of polysemous jokes and riddles in semantic development.
Cahiers de Psychologie Cognitive/Current Psychology of Cognition
17: 343-365.
Palmer, Gary, Heather Bennett and Les Stacey
1999 Bursting with grief, erupting with shame: a conceptual and gram-
matical analysis of emotion-tropes in Tagalog. In: Gary Palmer and
Debra Occhi (eds.), Languages of Sentiment: Cultural Constructions
of Emotional Substrates, 171-200. Amsterdam: Benjamins.
332 Ann Dowker

Rivkai, Yehuda
1937 Letters to Parents. Jerusalem: Achiasaf (in Hebrew).
Schieffelin, Bambi
1983 Talking like birds: sound play in a cultural perspective. In: Elinor
Ochs and Bambi Schieffelin (eds.), Acquiring Conversational Com-
petence. London: RoutJedge and Kegan Paul.
Schwarz, Judith
1981 Children's experiments with language. Young Children 36: 16-26.
Siegler, Robert
1996 Emerging Minds: The Process of Change in Children's Thinking.
Oxford: Oxford University Press.
Siegler, Robert and Eric Jenkins
1989 How Children Discover New Strategies. Hillsdale, NJ: Erlbaum.
Sinha, CJrris and Tania Kuteva
1995 Distributed spatial semantics. Nordic Journal of Linguistics 18:
167-199.
Stross, Brian
1975 Metaphor in the speech play of Tzeltal children. Anthropological
Linguistics 17: 305-323.
Sutton-Smith, Brian
1980 The Folkstories of Children. Philadelphia: University of Pennsylvania
Press.
Trettien, August
1894 Psychology of the language interest of children. Pedagogical Seminary
3: 419-466.
1904 Psychology of the language interest of children. Pedagogical Seminary
23: 113-177.
UlImann, Stephen
1951 The Principles of Semantics. Oxford: Blackwell.
Vosniadou, Stella
1987 Children and metaphors. Child Development 58: 870-885.
Weir, Ruth
1962 Language in the Crib. The Hague: Mouton.
Winner, Ellen
1988 The Point of Words: Children's Understanding of Metaphor and Irony.
Cambridge, MA: Harvard University Press.
Yu, Ning
2001 What does our face mean to us? Pragmatics and Cognition 9: 1-36.
Zhou, Rong and Xiteng Huang
2000 A cross-cultural study of the metaphorical representation of time.
Psychological Science - China 23: 141-145.
Emerging patterns and evolving polysemies: the
acquisition of get between four and ten years

Brigitte Nerlich, Zazie Todd and David D. Clarke

1. Introduction

The study of the acquisition of verbs has become central to developmental

psycholinguistics (Tomasello 1992; TomaseIlo and Merriman 1995). This
paper considers how the array of senses conventionally associated with the
polysemic word get is acquired by children between the ages of 4 and 10
years (see Tomasello 1992 for the acquisition of get between the ages of 17
and 24 months). In the process it will investigate whether developmental
psychology can shed light on recent developments in cognitive linguistics
and vice versa.
Polysemic verbs are important, since it is a basic claim of cognitive
linguistics:

that a lexical item is typically polysemic - comprising a family of interrelated

senses, forming a network centred on a prototypical value. Although the
precise array of senses conventionally associated with the expression is not
fully predictable, neither is it arbitrary - as the network evolves from the
prototype, each extension is MOTIVATED in some cognitively natural
fashion, and often in accordance with a general pattern or principle.
(Langacker 1988: 392)

Although the developmental sequence of the acquisition of the senses of

get is not fully predictable, neither can it be entirely arbitrary. Our central
hypothesis is that the first senses to be acquired will be the most prototypical
values of the word get, and that the developmental route followed from then
on will be motivated by some cognitively natural pattern.
There are several other reasons why the verb get was chosen for this
study. It is in the top five of the most common verbs in the English language,
with nearly 125,000 citations (out of toO million) in the British National
Corpus (New Oxford Dictionary of English 1998: 770). It is also one of the
334 Brigitte Nerlich, Zazie Todd and David D. Clarke

first and most commonly used verbs in children's speech, along with other
"general-purpose" or "light" verbs such as put, make, go, and do (Clark 1978).
Get is also one of the many highly polysemic verbs found in English.
The New Oxford Dictionary of English (1998), a dictionary of modern
usage, lists the following main senses and subsenses of get in current English
usage:

(i) come to have or hold (something); receive (experience, suffer, or be

afflicted with something, receive as a punishment, contract a disease);
(ii) succeed in attaining, achieving, or experiencing; obtain (move in order
to pick up or bring [something]; fetch, travel by or catch a bus, train,
obtain as a result of calculation, respond to a ring);
(iii) enter or reach a specified state or condition; become (used with past
participle to form the passive mood); cause or be treated in a specified
way, induce or prevail upon [someone] to do something, have the
opportunity to do, begin to be or do something;
(iv) come, go or make progress eventually or with some difficulty (move
or come into a specified position or state), succeed in making [someone
or something] come, go or make progress;
(v) have got; have;
(vi) catch or apprehend (someone) (strike or wound [someone] with a
blow, punish, injure or kill, be punished, injured or killed), annoy or
amuse, baffle;
(vii) understand.

U sing the historical Oxford English Dictionary (1888-1924), it is possible

to chart the development of the core senses of get over time. It seems that
the main subcategories of senses were already well-established by the BOOs.
The diachronic evolution of the sense of get can be roughly summarized as
follows:

1200: obtain possession

1250: receive
1300: seek out and take, procure (fetch)
1350: succeed in coming or going
1380: to receive knowledge
1500: make oneself, become
1897: to understand
 The acquisition of get between four and ten years 335

Some questions that might be asked are: which senses are first learned by
children and in which sequence? Does the developmental sequence of
acquisition map onto the most prototypical synchronic spread of senses and/
or the diachronic spreading of senses?
In this paper, we use recent developments in cognitive linguistics to
explore these questions. Some central tenets of cognitive linguistics and their
implications for the developmental study of get are outlined below. This will
provide the theoretical background for the study, but it will also allow us to
spell out more clearly the various hypothetical scenarios which could be
used to explain the acquisition of polysemous words by children.

2. Theoretical background

2.1. Embodiment

In recent years it has been proposed that meaning is grounded in the nature
of our bodies and in our interaction with the physical, social and cultural
environment we live in (see Lakoff and 10hnson 1999), It is claimed that
concepts are grounded in our bodily experience and then elaborated by struc-
tures of imagination, i.e. metaphor and metonymy. This means that metaphor
and metonymy motivate sense extension in some cognitively natural fashion,
as for example in expressions like 1 got it meaning 'I have understood it'.
The prototypical cognitive linguistic investigation focuses on the per-
ceptual, visual bases of the emergence and change of word meanings.
Meanings are seen as perceptually grounded. More recently researchers have
also begun to study hearing and smell (see Ibarretxe-Antuiiano 1999) as the
source of our embodied conceptual structures which are again the basis for
our construction of meaning. But the body's input into the construction of
meaning is not only limited to our senses of vision, hearing and smell; we
also interact with the world, and this much more actively, through our hands,
through touch, grasping, gathering, giving and receiving. The prototypical
and primordial experiential scene in which the word get is embedded can be
said to be a situation in which somebody comes to have or hold something.
From this underlying matrix get's polysemous sense extensions seem to flow.
In the case of get the scene, as illustrated in Figure I, corresponds to the
experiential Gestalt of a basic causal event in which an agent receives an
object from another agent, which results in grasping or holding something:
336 Brigitte Nerlich, Zazie Todd and David D. Clarke

GET

terminal motion,
'arrive'
initial motion

motion forward, backward, 'find'

change of state
to, from, up, down
(Joco)motion
search, result of search, 'fetch'
success, failure

'become'
...
'cause' 'capture' 'obtain'
'induce'

acquisitjon
'receive'
'eating'
(preparation)

CONCRETE or ABSTRACT

Figure 1. The prototypical scene associated with get.

someone causes something to be in somebody else's possession (see Tyler

and Evans, this volume, for a discussion of protoscene).

2.2. PoLysemy

Another central tenet of cognitive linguistics has been the poly se mic structure
of semantic categories. It has become clear that the study of the network of
senses surrounding a word form is of fundamental importance for any study
 The acquisition of get between four and ten years 337

of language and cognition. Psycholinguists studying the mental lexicon have

begun to ask questions like: are the meanings of a polysemous word "related
to each other in some way, perhaps via a core concept ... ? ... Are the various
meanings of words compartmentalized into many discrete and functionally
independent units, or is the mental lexicon organized more economically
with fewer but more highly structured lexical entries?" (Williams 1992: 194)
Some cognitive semanticists have investigated the polysemy verbs, such
as take or get, in an intuitive way, relying on their own knowledge of a
language (see Norvig and Lakoff 1987; Lindstromberg 1991). Others have
studied the network of meanings surrounding a word fonn, such as buy,
through the use of corpora (see Fillmore and Atkins 1992) and computerized
semantic networks. Some have started to integrate research in cognitive
linguistics with research in language acquisition and have begun to study
polysemous verbs, such as see (see 10hnson 1996, 1999).
Other cognitive psycho linguists have employed empirical methods to study
the cognitive-semantic aspects of polysemy in adults. Gibbs, for example,
analysed how subjects perceive the role of different image-schemas in the
meaning of stand (Gibbs et al. 1994), and Sandra and Rice have examined
the network of senses associated with certain prepositions using methods
from experimental psycholinguistics (Sandra and Rice 1995; Rice 1996).
More recently still, research has begun in the study of the acquisition of
get using the CHILDES database and applying Grady and 10hnson's concept
of "primary scene" (see Duxbury 1999 ms.; Grady and 10hnson 1997). Also
using CHILDES, Michael Israel has begun to chart the acquisition of get in
the framework of an argument/construction grammar approach. He argues:

that ultimately what children learn in acquiring these uses is a full set of
constructions and not just a few idiomatic senses for a single verb. The study
traces the emergence of these uses from a few verb-based formulae to a fully
productive family of constructions. The evidence shows that these uses are
not acquired in a random order, but rather are systematically linked to one
another, with more complex constructions learned as extensions from simpler
ones. (Israel, to appear)

In our study we use a lexically-based approach. Ideally, both approaches,

the lexical-semantic and the grammatical-constructionist one should be
combined.
Yet others are studying polysemy from an empirical point of view by
using adult infonnants and applying production tests and similarity rating
338 Brigitte Nerlich. Zazie Todd and David D. Clarke

tests (Raukko 1999; see Raukko 1997; see also Lehrer 1974). Others still are
investigating the acquisition of the get-passive using standard psychological
experiments (Kerstin Meints, personal communication; see also Giv6n and
Yang 1994).
Raukko (1999; see also Raukko, this volume) analysed the polysemous
structure of the verb get by asking adult informants to fill in questionnaires
where they were asked to produce sentence examples that manifest different
meanings of get. His aim was to build up a picture of the intersubjective
perception of the polysemy of get. He established that for young adults the
senses of get cluster approximately around the following base senses:
Obtaining, Receiving, Change of State, Motion, Understanding, Catching,
Obligation, Stable Possession, Carrying, Ability. This conforms grosso modo
to the main dictionary meanings of get, as well as to the diachronic evolution
of get (except for get used in the sense of understanding, which had quite a
high frequency in the examples produced by American high-school students;
on the difficulties in categorizing the senses of get, see Raukko, this volume).
However, one should be aware of the fact (already noted by the father of
the term polysemy, Michel Breal, in 1884 - see Breal 1991 [1884]):

that the real meanings are often far more complex than the simple dictionary
definitions would lead us to suppose .... The primary dictionary meaning of
words was often far adrift from the sense in which they were actually used.
Keep, for instance, is usually defined as to retain, but in fact the word is
much more often employed in the sense of continuing, as in keep cool and
keep smiling . ... Give, even more interestingly, is most often used ... as 'mere
verbal padding', as in give it a look or give a report. (Bryson 1990: 143)

Children learning a language might therefore be exposed to quite a dif-

ferent distribution of meanings than the dictionary one, and also quite different
from what adults believe the distribution to be in theory, if not in practice (as
captured by Raukko 1999, to some extent). In the case of get children would
certainly be exposed to a high proportion of uses where get is mere verbal
padding, is highly idiomatic, and where get, as part of a prefabricated
semantic whole, has almost lost its own semantic value. Although this needs
undoubtedly much further investigtion, we shall see whether this parental
influence is reflected in children's spontaneous production of sentences
containing get or whether this production task elicits utterances based on an
underlying knowledge of the prototypical core senses of get as found in
dictionaries of common usage.
 The acquisition of get between four and ten years 339

2.3. Polysemic networks and pre-patterned chunks

In the diachronic evolution of polysemic words there seems to be a certain

point when the network of senses becomes saturated and begins to shed
certain meanings. Either senses become so distant that they are perceived as
unrelated (they become homonyms, such as French voler meaning 'to fly'
and 'to steal'), or the word becomes part and parcel of a free-standing
semantic conglomerate. Standing expressions such Get a life, Get on with it,
and Lets get out of here, as well as phrasal verbs (a combination of a verb
with an adverb or preposition that has a single meaning) develop, such as to
get into something, get over something, get out of something. This corresponds
to a general tendency for groups of words to settle into pre-packaged chunks
(Deignan 1999: 34).
Most of these fossilized phrasal collocations of get are based on meta-
phorical or metonymical mappings, or mappings of more concrete locative
meanings onto more abstract mental meanings (see Brugman 1981; Lakoff
1987; Langacker 1987; Goddard 1998: 79; for a summary see Ungerer and
Schmid 1996: Ch. 4). Examples of concrete uses are: He got safely across
the bridge; Driving carefully, she finally managed to get around the obstacle;
I found the key and managed to get into the house; and so on. Built on these
are abstract uses such as: I really couldn't get my idea across; The news is
getting around quickly; To help get around this problem, some tanks are now
equipped with radar; I finally got a remark in [the conversation]; and so on
(some of these examples are taken from Cobuild online).
Whereas research into the fossilization of metaphorically based col-
locations helps us to understand some aspects of verbal polysemy, new
research in construction grammar helps us to understand others. One of the
foremost researchers in this field is Adele Goldberg. She discusses the
following examples:

(1) She sneezed a terrible sneeze.

(2) She sneezed her nose red.
(3) She sneezed her way to the emergency room.
(4) She sneezed the foam off the cappuccino.

Are we dealing here with four different senses of the verb sneeze? Goldberg
(1998; see also Goldberg 1995) thinks not:
340 Brigitte Nerlich, Zazie Todd and David D. Clarke

Instead of positing a new verb sense whenever a new syntactic frame is

available, it makes sense to associate some aspects of meaning directly to the
formal pattern itself. This allows us to account for the full semantic interpre-
tation without positing implausible and ad hoc verb senses ... (Goldberg
1998: 205)

In her view "each of these formal patterns and its associated meaning(s)
form a construction of the language" (Goldberg 1998: 205) and (argument
structure) constructions encode certain archetypal forms of human experience.
This has important implications for the study of language acquisition and
the study of get.
Prefabricated chunks help us to build sentences up more quickly; argument
structure construction provides us with ready-made blueprints for constructing
sentences. Both should be important in language acquisition (see Hopper
1998: 166).
Goldberg (1998) has argued that light verbs such as put, go, make, do and
get can be seen as "path breaking verbs", that is, as laying the foundations
for the acquisition of more and more complex grammatical constructions.
Some of these verbs are listed in Table 1.

Table J. Light verbs and constructional meanings they correspond to (reproduced

from Goldberg 1998: 207).

Verb Constructional meaning Construction

put X causes Y to move Z Caused motion

make X causes Y to become Z Resultative
go X moves Y Intransitive motion
do X acts on Y Transitive
get X acquires/possesses Y Possessive

Goldberg claims that the initial meaning of a construction is a basic

experiential Gestalt. This means that a basic pattern of experience is encoded
in a basic pattern of the language (Goldberg 1998: 208). The scene for get
was depicted in Figure 1.
As Slobin (1985) observed, children's first use of certain grammatical
marking is applied to "prototypical scenes": "In Basic Child Grammar, the
first Scenes to receive grammatical marking are 'prototypical', in that they
regularly occur as part of frequent and salient activities and perceptions, and
thereby become organizing points for later elaboration" (Slobin 1985: 1175).
 The acquisition of get between four and ten years 341

We shall explore whether the same applies to the acquisition of verbs like
get.

2.4. Levels of polysemy

In order to study the acquisition of the polysemous verb get, one has to be
aware of three levels of polysemy:

(i) A primary level of polysemy (based to a large extent on sense exten-

sions which follow the path of the prototypical frame or scene in
which get is embedded, as well as on metaphorical and metonymical
extensions); here get can mean 'obtain', 'receive', 'fetch', 'reach',
'cause', 'induce', 'understand', and so on. Some of these senses are
construction-based, such as the ditransitive to get somebody something.
(ii) A secondary level of polysemy, where certain semantic conglomerates
have formed, such as phrasal verbs and idioms.
(iii) A tertiary level of polysemy, based again on metaphorical and
metonymical extensions; here phrasal verbs derived from get have
themselves become polysemous, as in 1 got there in the end (to a
place) and 1 got there in the end (a task), 1 got into the house and 1 got
into trouble, and so on.

3. The acquisition of get

3.1. Hypotheses

(i) The first senses to be acquired will be the most prototypical values of
the word get related to the primary scenario illustrated in Figure 1; then
children will gradually work their way outwards to the more meta-
phorically and metonymically motivated meanings. This would be in
contrast to a pattern in which random chunks are only later integrated
into a conceptual whole roughly structured like the prototypical scene.
(ii) The first senses to be acquired will stay on the primary level of
polysemy.
342 Brigitte Nerlich. Zazie Todd and David D. Clarke

(iii) The synchronic spread of polysemes which cover the semantic area
outlined by the word get reflects to a certain extent the diachronic
spread of polysemes over time, and the developmental sequence in
which get is acquired by children follows the same path.
(iv) The conceptual frames associated with the various meanings of get
are closely linked to children's primary situational knowledge in which
the process of "getting" is integrated, of which the frame "obtaining
a present" is the most central. This will be the starting point for a
gradual exploration of the semantic and syntactic (argument structure)
framework surrounding get.

This means that children should learn the major senses of get in the
following sequence: 'obtain/receive', 'have', 'fetch', 'become', 'go/arrive',
'induce', 'having permission', 'understand'. We predict that the beginning
and the endpoint in the sequence will be stable, whereas there will be quite
a range of variation in the acquistion of the "intermediate" senses. We also
predict that the children will produce only a small number of get instances
from polysemy levels (ii) and (iii).

3.2. Method

This study uses elicitation and ranking tasks in the hope of tapping com-
petencies which do not normally show up in naturalistic tasks. We would
like to stress, however, that analysing the use of get by children in natural
settings is as important in the study of the acquisition of polysemy as using
experimental data (see Duxbury 1999 ms.).

3.2.1. Participants

Participants were recruited from local infant and junior schools. A letter was
sent to all parents of children in the selected age groups, and all children
who were given permission to take part in the study were used as participants.
There were eleven 4-year-olds, twenty 7-year-olds, sixteen 8-year-olds, and
twelve lO-year-olds.
 The acquisition of get between four and ten years 343

3.2.2. Materials

One Panasonic portable cassette recorder, cassette tapes, and some cards of
postcard size on which the experimental sentences were typed.

3.2.3. Design

The experimental stimuli consisted of a set of sentences using the word get
that were chosen to represent the following core senses of the word: OBTAIN,
FETCH, BECOME, Go, UNDERSTAND. I In order to keep the procedure simple
enough for the 4-year-olds to understand, only one sentence was used to
illustrate each meaning. The sentences were tested in a pilot study with ten
children, as a result of which one of the sentences (OBTAIN) was changed to
remove a possible desirability bias. The initial sentence involved getting a
bicycle as a birthday present; many of the children commented that this was
very boring and the sentence was replaced with a more neutral sentence
about getting a cake. The letter to parents explained that the experimental
sessions would be tape recorded and that all data would be kept anonymous
and confidential. All children saw the same set of experimental sentences.
The order of presentation of the sentences was randomized to prevent any
order effects. The independent variable was age and the dependent variables
were the sentences produced in the elicitation task and the rankings given
for the experimental stimuli in the ranking task.

3.2.4. Procedure

The experiment was conducted over a period of several weeks on the school
site. Children were taken from the classroom one at a time to participate in
the experiment. Children were told that the purpose of the experiment was to
study the meaning of words and that there were no right or wrong answers.
They were asked if they minded the cassette recorder being used and the
machine was then turned on. Children were given an initial warm-up task to
reduce shyness and then the experiment proper began. The production test
method used by Raukko (1999, this volume) was adapted for use with young
children and supplemented by a ranking task. Children were told that we
were interested in the meaning of the word get, and asked if they could think
344 Brigitte Nerlich, Zazie Todd and David D. Clarke

of any sentences that used the word. (Raukko had asked participants to
produce sentences showing as many different senses of get as possible; this
was felt to be too difficult for young children and instead they were just
asked to produce five different sentences.)
After the child had either produced five sentences or run out of ideas, the
ranking task began. Children were told that they were going to hear five
sentences using the word get and they would have to choose the one which
was the best example of what the word meant. After the sentences were read
out in a random order, the children were given the cards (in the same random
order) to choose from. Once they had selected the best example, they were
then asked to choose the best one from the remaining sentences, and so on
until all five sentences had been ranked. Once all the sentences had been
ranked, children were asked to give a reason for their best and worst choices.
Experimental responses were recorded both on paper and on cassette. Once
the experiment was over, children were thanked for their time, complimented
on their answers, and returned to the classroom.

3.3. Results

The results fall into two sections - the qualitative production data and the
quantitative rankings data.

3.3.1. Productions

To even out the numbers in each age group, data from five children were
selected at random from the 7-year-old group, and discarded. (N.B. This
applies to the production data only.) The productions were then categorized
according to the kind of get produced and the syntactic frames used.
The typical frames in which utterances were produced are shown in
Table 2. The numbers of utterances falling into each of the main senses is
given in brackets. Not surprisingly, the 4-year-olds produced fewer utterances
and also used a smaller range of types of get. This was also the only age
group in which 'obtain/receive' was not the most popular get production.
The IO-year-olds were the most creative in their production of sentences
containing get. They produced nine different senses of get, including get as
meaning 'induce somebody to do something', embedded in a wide variety of
 The acquisition of get between four and ten years 345

speech act types and argument structure frameworks. They are closely fol-
lowed, however, by the 7-year-olds who produced eight different senses of
get, including get as meaning 'become' and 'find'. The 8-year-olds produced
seven senses of get and used less syntactic variety. However, they produced
the only instances of get as idiomatic expressions, e.g. Get a life! and Get
lost. This shows a gradual progression towards a mature competence in the
use of get in its core senses, which accelerates at age 6-7. The use of get as
"verbal padding", in phrasal collocations as well as idioms was, however,
extremely rare.
As one can see from the prototypical objects and events listed in Table 2,
the frames and scenes in which the utterances were implicitly embedded
come mainly from the children's experiences with obtaining (more and more
sophisticated) presents on the occasion of birthdays and at Christmas, of
receiving pocket money and sweets, and of having, being in possession of,
certain special toys or pets. This also ties in with the cognitive linguistic
hypotheses about the embodiment of meaning, as the first senses children
learn are related to their pleasure in having and holding something which is
precious to them in some way.

3.3.2. Rankings

The rankings gi ven to the different sentences by the children were collated.
Since there were different numbers of children in the groups, the data is
expressed as percentages in the graphs. The 20% level is equivalent to chance,
since there were five choices.
Initially, the data were collated to see which sentences were ranked first,
that is, as the best example of what the word get means. For the 4-year-olds,
we can see (Fig. 2) that 'obtain/receive' was overwhelmingly selected as the
first choice (by 45% of children in this age group). This is significantly
different from chance (p < 0.05). This is not the case with the other age
groups, where there is no sentence emerging as a clear first choice. We can
see that for the 7-year-olds there is no clear leader. For the 8-year-olds,
'become' was never selected as first choice, but the others are ranked pretty
much equal. For the lO-year-olds, again there is no overwhelming first choice.
The 4-year-olds' overwhelming choice of get 'obtain' as first choice
becomes even clearer when we look at all the rankings that this sentence
received. We can see (Fig. 3) that it was never ranked lower than in the top
 w
Table 2. Typical frames of productions by children in the different age groups. ~
C\

4-5 6-7 7-8 9-10 .,t:tl

0;.
Senses have, fetch, obtain/receive, fetch, buy, obtain/receive, have, obtain/receive, fetch, S·
~
obtain/receive become, reach/go, have, fetch, become, buy, buy, have, reach,
retrieve, put on put on, being allowed retrieve, induce, find, ~
:::t.
;::;.
to do something being allowed to do .;:r-
something
~
N
~.

Prototypical [I] get X. 1 get X. I'll get X. 1 got/get X. 1 got/get X.

~
syntactic Get X! 1 can get X. 1 went to get X. 1 am going to get X. ~
frames Get me (some) X! 1 am going to get X. 1 am going to get X. 1 went to get X. I::l
1 (have) got X. 1 had to go and get X. 1 want to get X. ~
Let's get X. My mummy got X. Y let me get X. ~~
Let's go and get X. I got X. 1 have got X. Y needs to get X. s:
Let's go to Y to get X. Get X! Get me X! Let's get X. Get X! ~
1 want to get (some) X. Can I get X? Get X for Y! ()
I would like to get X. Could you go and get X? 5"
Can 1 get X?
Can I/we get (some) X?
Get (me) X!
Y got X for Z.
1 got to do X.
Can you get X?
Y got me X.
*~

Can you get me X? Y got some X from Z.

Can you go and get me X? Y got X for Z. 1 got X.
My mum told me to get X. 1 have got X.
I get X for you. Y got more X than Z.
Where shall 1 get X? Y got X to do Z.
Prototypical statement, order statement, order, question statement, order, question statement, order, question
speech acts

Prototypical sweets, car, cat sweets, toy, pocket money, sweets, pocket money, sweets, pocket money,
objects/events, play station, drink, shoes, pet, toy, ice-cream, shoes, video, TV, drink,
prototypical cat bike, cat, drink, Christmas, birthday
conceptual Christmas, birthday
frame
;l
Others Get up! get away from, get weighed, Get lost! Get a life! get told off, Got you! "'
I:l
get ready get ready, I got a long .fll::
1:;0
way to my birthdayo ::0,0
go
~
~
0-

~
i
l::
....
I:l
5.
~
1 <:J

~
348 Brigitte Nerlich, Zazie Todd and David D. Clarke

50

40

Q,)
bO
E
Q,)
30

~
20

10

Obtain Fetch Become Go Understand

Figure 2. Four-year-olds' first choice of meaning for get.

50

40

Q,)
bJl
«I
C
Q,)
30

~
20

10

First Second Third Fourth Fifth

Figure 3. Four-year-olds' rankings for 'obtain'.

 The acquisition of get between four and ten years 349

three. This is quite unusual when compared to all the other rankings given
across all age groups. The probability of this occurring by chance as first
choice this many times is less than 0.005. Four-year-olds also tended to place
'understand' in the last two at a level significantly different from chance
(p < 0.05).
For the 7-year-olds, the distributions are much flatter; there is no clear
order to their rankings at all (see Fig. 4). In fact, the 7-year-olds gave the
most undifferentiated responses out of the age groups surveyed, and none of
the results are significantly different from chance. This could mean that
7-year-olds are simply guessing. It is, however, also possible that this result
simply reflects the fact that they are aware of all the different meanings of
get, and have no overall preferences about which is most central. In other
words, they have learnt that 'obtain' is not necessarily a "better" example of
get, but they have a pluralistic sense of the meaning of get and have learnt to
embrace its alternative meanings without establishing particular preferences.
This might also show that the process the acquisition of get is relatively
slow but seems to accelerate around age 6, an age when children start to
immerse themselves in word play, jokes, and metaphors, which for the most

30

10

Obtain Fetch Become Go Understand

Figure 4. Seven-year-olds' first choice of meaning for get.

350 Brigitte Nerlich, Zazie Todd and David D. Clarke

part are based on the humorous exploitation of multiple meanings (see

Nerlich, Todd and Clarke 1998).
The 8-year-olds also have fairly flat distributions, with no overall first
choice, but a pattern is starting to emerge (see Fig. 5). 'Become' is ranked in
the last two positions significantly more often than would be expected by
chance (p < 0.05). 'Go' is ranked in the last two positions significantly less
often than would be expected by chance (p < 0.05), and there is a tendency
for it to be ranked third which just misses significance (p = 0.055). 'Under-
stand' is significantly likely to be ranked last (p < 0.01). One could argue
that they have achieved full competence and can produce all the various
core senses of get and that they have reached the age at which, as some
argue, metaphorical language goes underground (Gardner et at. 1978), a
tendency that becomes even clearer with the 10-year-olds.
The picture becomes more interesting with the 10-year-olds (see Fig. 6).
Here, a pattern is starting to emerge of the order in which sentences were
chosen. Individual differences are also apparent, with children making a
choice at the beginning and then following through in a pattern which is
consistent with that choice. 'Obtain' is ranked second significantly more often
than chance (p < 0.05). There is a trend for 'become' to be ranked third which
just misses significance (p = 0.053), but it is ranked in the middle three
significantly (p < 0.05). 'Understand' is overwhelmingly ranked last (p < 0.001).
Thus, there are signs emerging of the order in which sentences were chosen,
with 'understand' a clear last. The individual differences show that although
lO-year-olds may make different initial choices, their later choices are in a
pattern consistent with that.
To summarize the data from the rankings, the only completely clear first
choice was made by the 4-year-olds, who overwhelmingly favoured 'obtain',
and were less keen on those meanings which depend on metaphorical exten-
sion. The data for the 7-year-olds showed a fairly undifferentiated response,
in which there were no clear preferences and no clear patterns in the data.
By 8 years, signs of a pattern were starting to emerge. By 10 years, however,
there is evidence not just of a clear pattern in the data, but also of individual
differences in initial choice, which are then followed through consistently. It
would be desirable to have data for older children and adults to see how this
pattern develops further.
 The acquisition of get between four and ten years 351

30

10

Obtain Fetch Become Go Understand

Figure 5. Eight-year-olds' first choice of meaning for get.

30

10

Obtain Fetch Become Go Understand

Figure 6. Ten-year-olds' first choice of meaning for get.

352 Brigitte Nerlich, Zazie Todd and David D. Clarke

3.4. Discussion

Our initial hypotheses seem to be confirmed. Children learn the senses of

get from the most prototypical sense outwards and they seem, on the whole,
to follow the route of the diachronic semantic development of the verb get
over time and to mirror the distribution of senses in adult speech. In produc-
tion tasks they seem to stay mainly on the level of what we called "primary
polysemy". The prototypical scene they explore most is that of getting a gift
or a present.
The line of semantic development starts with the core sense of 'obtain'
and moves onto the most distant sense of get as 'understand' in various
overlapping stages. One can hypothesize that at each stage, at ages 4, 7, 8
and 10, the children's understanding of the meaning of get reflects a different
and changing model of semantic representation. At age 4, the children's
knowledge of the meaning of get could be described by using the model of
polysemy that posits general and abstract core meanings which are presumed
to underlie all uses of a polysemous word (see Caramazza and Grober 1976),
whereas the endpoint of the developmental process could be better described
by a prototype model of semantic representation, inspired by Eleanor Rosch's
research into human categorization (Rosch 1977; Geeraerts 1989). At the
midpoint between the two poles lies the semantic knowledge of the 7-year-
aids. Their responses suggest that they do not have a prototypical sense of
the meaning of get, since nothing was picked out as "best" overall and all
the meanings seemed to be equal. The prototypical structure of sub- and
super-ordinate levels is not found.
However, more research is needed to confirm these results. It would be
helpful to have data for older children and adults to see how these emerging
patterns develop further. One should also compare our experimental data
with naturalistic data. We have thrown light on children's growing passive
knowledge of the polysemy of get, not on their active knowledge.
The big question still is: why should children overwhelmingly produce
and choose the core senses of get, as found in elicitation tasks with adults, as
found in synchronic and diachronic dictionaries and so on? Why don't they
produce instances of get, which many of them must hear every morning
before they go to school, such as Get up, Get ready, Get your jumper on, Get
on with it, We are getting late, and so on? To answer this question, one
would have to look more closely at adult and peer group input at every stage
in lexical development between age 4, 10 and beyond.
 The acquisition of get between four and ten years 353

This study has also shown the usefulness of an interdisciplinary approach

which uses both psychology and cognitive linguistics. Developmental work
in this field is important because it can show the gradual growth and struc-
turing of the network of meanings surrounding a word form in language
acquisition, and the relevance of the various frameworks of thinking and
acting in which children learn certain words. We have seen how a pattern of
semantic networks emerges at different stages of development, which seems
to indicate that at different ages these patterns appear to map onto different
theories of word representation. This article has also shown that the structure
and organization of meanings around a polysemous word appear to change
with age in a pattern roughly consistent with the diachronic change in word
meanings and the spread of meanings in society.
Further research is needed to investigate whether the results found here
would also apply to other polysemous verbs, and to explore the extent to
which embodiment can be seen as motivation for word meanings. Finally,
the theoretical approach explored here provides a useful framework for future
research.

Notes

1. Sentences illustrating different meanings of get:

OBTAIN Peter got a cake from the bakers.
FETCH Can you get the milk out of the fridge?
BECOME Francesca got really wet today.
Go Let's get out of here.
UNDERSTAND The teacher spoke so fast I didn't get what she said.

References

Bn!al, Michel
1991 [1884] How words are organized in the mind. In: M. Bn!al, The Beginnings
of Semantics. Essays, Lectures and Reviews, 145-151. Edited and
translated by George Wolf. London: Duckworth.
Brugman, Claudia M.
1981 The Story of over. Trier: LAUT.
Bryson, Bill
1990 Mother Tongue. The English Language. London and New York:
Penguin Books.
354 Brigitte Nerlich, Zazie Todd and David D. Clarke

Caramazza, Alfonso and Ellen Grober

1976 Polysemy and the structure of the subjective lexicon. In: Clea
Rameh (ed.), Semantics: Theory and Application, 181-206.
Georgetown University Round Table on Languages and Linguistics
1976. Washington: University Press.
Cl ark, Eve V.
1978 Discovering what words can do. Papers from the Parasession on the
Lexicon, Chicago Linguistic Society, 34-57.
Deignan, Alice
1999 Linguistic metaphors and collocation in nonliterary corpus data.
Metaphor and Symbol 15(1): 19-36.
Duxbury, Simone
1999 (ms.) Getting the hang of multiple meanings - new perspectives on
polysemy. Paper for the Theme session on language acquisition at
the International Cognitive Linguistics Conference, Stockholm, July
14, 1999.
FilImore, Charles J. and Beryl Atkins
1992 Towards a frame-based lexicon: the semantics of risk and its
neighbors. In: Adrienne Lehrer and Eva Feder Kittay (eds.), Frames,
Fields, and Contrasts. New Essays in Semantic and Lexical
Organization, 75-102. Hillsdale, NJ: Lawrence Erlbaum Associates.
Gardner, Howard, Ellen Winner, R. Becchoffer and D. Wolf
1978 The development of figurative language. In: Katherine Nelson (ed.),
Children's Language, Vo!. 1. Cambridge, MA: Gardner Press Inc.
Geeraerts, Dirk
1989 Prospects and problems of prototype theory. Linguistics 27: 587-612.
Gibbs, Raymond w., Jr., Dinara A. Beitel, M. Harrington and P. E. Sanders
1994 Taking a stand on the meanings of stand: bodily experience as
motivation for polysemy. Journal of Semantics 11: 321-251.
Givon, Talmy and L. Yang
1994 The rise of the English GET-passive. In: Paul J. Hoppe and Barbara
Fox (eds.), Voice: Form and Function, 119-149. Amsterdam and
Philadelphia: John Benjamins.
Goddard, Cliff
1998 Semantic Analysis. A Practical Introduction. Oxford: Oxford
University Press.
Goldberg, Adele E.
1995 Constructions: A Construction Grammar Approach to Argument
Structure. Chicago and London: University of Chicago Press.
1998 Patterns of experience in patterns of language. In: Tomasello (1998),
203-220.
Grady, Joseph and Chris Johnson
1997 Converging evidence for the notions of subscene and primary scene.
 The acquisition of get between four and ten years 355

Proceedings of the 23rd Annual Meeting of the Berkeley Linguistics

Society, 123-136. Berkeley, CA: Berkeley Linguistics Society.
Hopper, P. J.
1998 Emergent grammar. In: Tomasello (1998), 155-176.
Ibarretxe-Antufiano, Iraide
1999 Metaphorical mappings in the sense of smell. In: Gerard Steen and
Raymond W. Gibbs Jr. (eds.), Metaphor in Cognitive Linguistics,
29-45. Amsterdam and Philadelphia: John Benjamins.
Israel, Michael
(to appear) Systematic idiomaticity in the acquisition of English 'get' con-
structions.
Johnson, Christopher R.
1996 Learnability in the acquisition of multiple senses: SOURCE recon-
sidered. In: J. Johnson, M. L. Luge and 1. Moxley (eds.), Proceedings
of the 22ndAnnual Meeting of the Berkeley Linguistics Society, 469-
480. Berkeley, CA: Berkeley Linguistics Society.
1999 Metaphor vs. conflation in the acquisition of polysemy: the case of
see. In: Masako Hiraga, Chris Sinha and Sherman WiJcox (eds.),
Cultural, Typological and Psychological Perspectives in Cognitive
Linguistics, 155-169. Amsterdam: John Benjamins.
Lakoff, George
1987 Women, Fire, and Dangerous Things. Chicago: Chicago University
Press.
Lakoff, George and Mark Johnson
1999 Philosophy in the Flesh: The Embodied Mind and its Challenge to
Western Thought. New York: Basic Books.
Langacker, Ronald W.
1987 Foundations of Cognitive Grammar, Vol. I. Theoretical Prerequisites.
Stanford, CA: Stanford University Press.
1988 Review of George Lakoff: Women, Fire, and Dangerous Things,
Chicago: Chicago University Press. Language 64(2): 384--395.
Lehrer, Adrienne
1974 Homonymy and polysemy: measuring similarity of meaning.
Language Sciences 32: 33-39.
Lindstromberg, Seth
1991 Get: not so many meanings. International Review of Applied
Linguistics (lRAL) 29: 285-302.
Nerlich, Brigitte, Zazie Todd and David D. Clarke
1998 The function of polysemous jokes and riddles in lexical development.
Cahiers de Psychologie Cognitive/Current Psychology of Cognition
17(2): 343-366.
356 Brigitte Nerlich, Zazie Todd and David D. Clarke

New Oxford Dictionary of English

1998 Edited by Judy Pearsall; chief editor, current English dictionaries,
Patrick Hanks. Oxford: Oxford University Press.
Norvig, Peter and George Lakoff
1987 Taking: a study in lexical network theory. Proceedings of the 13th
Annual Meeting of the Berkeley Linguistics Society, 195-206. Berkeley,
CA: Berkeley Linguistics Society.
Oxford English Dictionary
1888-1924 A new English dictionary on historical principles: founded mainly
on the materials coIlected by the Philological Society / edited by
James A. H. Murray ... with the assistance of many scholars and
men of science. Oxford: The Clarendon Press. Introduction,
supplement and bibliography published in 1933.
Raukko, Jamo
1997 The status of polysemy in linguistics: from discrete meanings to de-
fault flexibility. In: SKY 1997: The 1997 Yearbook of the Linguistic
Association of Finland, 145-170.
1999 An 'intersubjective' method for cognitive-semantic research of
polysemy: the case of get. In: Masko K. Hiraga, Chris Sinha and
Sherman WiIcox (eds.), Cultural, Typological and Psychological
Perspectives in Cognitive Linguistics, 87-107. Amsterdam: John
Benjamins.
Rice, Sally
1996 Prepositional prototypes. In: Rene Dirven and M. Plitz (eds.), The
Construal of Space in Language and Thought, 135-165. Berlin and
New York: Mouton de Gruyter.
Rosch, Eleanor
1977 Human categorization. In: Neil Warren (ed.), Studies in Cross-
Cultural Psychology 1, 3~9. London: Academic Press.
Sandra, Dominiek and SaIly Rice
1995 Network analyses of prepositional meaning: mirroring whose mind
- the linguist's or the language user's? Cognitive Linguistics 6(1):
89-130.
Slobin, Dan Isaac
1985 Crosslinguistic evidence for the language-making capacity. In: Dan
Slobin (ed.), A Crosslinguistic Study of Language Acquisition, Vol. 2
Theoretical1ssues, 1157-1256. HiIlsdale, NJ: Lawrence Erlbaum.
TomaseIlo, Michael
1992 First Verbs. A Case Study of Early Grammatical Development.
Cambridge: Cambridge University Press.
Tomasello, Michael (ed.)
1998 The New Psychology of Language. Cognitive and Functional
 The acquisition of get between four and ten years 357
Approaches to Language Structure. Mahwah, NJ and London:
Lawrence Erlbaum.
TomaseUo, Michael and W. E. Merriman (eds.)
1993 Beyond Names for Things: Young Children's Acquisition of Verbs.
Hillsdale, NJ: Lawrence Erlbaum.
Ungerer, Friedrich and Hans-Jorg Schmid
1996 An Introduction to Cognitive Linguistics. London and New York:
Longman.
Williams, John N.
1992 Processing polysemous words in context: evidence for interrelated
meanings. Journal of Psycholinguistic Research 21(3): 193-218.
Computational approaches
"I don't believe in word senses" *

Adam Kilgarriff

1. Introduction

There is now a substantial literature on the problem of word sense dis-

ambiguation (WSD).I The goal of WSD research is generally taken to be
disambiguation between the senses given in a dictionary, thesaurus or similar.
The idea is simple enough and could be stated as follows:

Many words have more than one meaning. When a person understands a
sentence with an ambiguous word in it, that understanding is built on the
basis of just one of the meanings. So, as some part of the human language
understanding process, the appropriate meaning has been chosen from the
range of possibilities.

Stated in this way, it would seem that WSD might be a well-defined task,
undertaken by a particular module within the human language processor.
This module could then be modelled computation ally in a WSD program,
and this program, performing, as it did, one of the essential functions of the
human language processor, would stand alongside a parser as a crucial
component of a broad range of Natural Language Processing (NLP) appli-
cations. This point of view is clearly represented in Cottrell (1989):

[Lexical ambiguity] is perhaps the most important problem facing an NLU

[Natural Language Understanding] system. Given that the goal of NLU is
understanding, correctly determining the meanings of the words used is
fundamental. The tack taken here is that it is important to understand how
people resolve the ambiguity problem, since whatever their approach, it
appears to work rather well. (Cottrell 1989: 1)

Word meaning is of course a venerable philosophical topic, and questions

of the relation between the signifier and the signified will never be far
from the theme of the paper. However, philosophical discussions have not
addressed the fact of lexicography and the theoretical issues raised by sense
362 Adam Kilgarriff

distinctions as marked in dictionaries. We often have strong intuitions about

words having multiple meanings, and lexicography aims to capture them,
systematically and consistently. The philosophy literature does not provide a
taxonomy of the processes underpinning the intuition, nor does it analyse
the relations between the word sense distinctions a dictionary makes and the
primary data of naturally occurring language. This is a gap that this paper
aims to fill.
I show, first, that Cottrell's construal of word senses is at odds with
theoretical work on the lexicon (Section 2); then, that the various attempts to
provide the concept "word sense" with secure foundations over the last 30
years have all been unsuccessful (Section 3). I then consider the lexicog-
raphers' understanding of what they are doing when they make decisions
about a word's senses, and develop an alternative conception of the word
sense, in which it corresponds to a cluster of citations for a word (Section 4).
Citations are clustered together where they exhibit similar patterning and
meaning. The various possible relations between a word's meaning potential
and its dictionary senses are catalogued and illustrated with corpus evidence.
The implication for WSD is that there is no reason to expect a single set
of word senses to be appropriate for different NLP applications. Different
corpora, and different purposes, will lead to different senses. In particular,
the sets of word senses presented in different dictionaries and thesauri have
been prepared, for various purposes, for various human users: there is no
reason to believe those sets are appropriate for any NLP application.

2. Thesis and antithesis: practical WSD and theoretical lexicology

2.1. Thesis

NLP has stumbled into word sense ambiguity.

Within the overall shape of a natural language understanding system -
morphological analysis, parsing, semantic and pragmatic interpretation - word
sense ambiguity first features as an irritation. It does not appear as a matter
of particular linguistic interest, and can be avoided altogether simply by
treating all words as having just one meaning. Rather, it is a snag: if you
have both river bank and money bank in your lexicon, when you see the
word bank in an input text you are at risk of selecting the wrong one. There
is a practical problem to be solved, and since Margaret Masterman's group
 "[ don't believe in word senses" 363

started examining it in the 1950s (see, e.g., Sparck Jones 1986), people have
been writing programs to solve it.
NLP has not found it easy to give a very principled answer to the question,
"what goes in the lexicon". Before the mid-1980s, many systems made no
claims to wide coverage and contained only as many words in the lexicon as
were needed for the "toy" texts that were going to be analysed. A word was
only made ambiguous - that is, given multiple lexical entries - if it was one
that the researchers had chosen as a subject for the disambiguation study.
This was clearly not an approach that was sustainable for wide coverage
systems, and interest developed in dictionaries, as relatively principled, wide-
coverage sources of lexical information.
As machine-readable versions of dictionaries started to become available,
so it became possible to write experimental WSD programs on the basis of
the dictionary's verdict as to what a word's senses were (Lesk 1986; Jensen
and Binot 1987; Slator 1988; Guthrie et al. 1990; Veronis and Ide 1990; Guthrie
et at. 1991; Dolan 1994). Looked at the other way round, WSD was one of
the interesting things you might be able to do with these exciting new resources.
Since then, with the advent of language corpora and the rapid growth of
statistical work in NLP, the number of possibilities for how you might go
about WSD has mushroomed, as has the quantity of work on the subject
(Brown et at. 1991; Hearst 1991; Gale, Church and Yarowsky 1992, 1993;
McRoy 1992; Yarowsky 1992). Clear (1994), Schiitze and Pederson (1995)
and Yarowsky (1995) are of particular interest because of their approach to
the issue of the set of word senses to be disambiguated between. Schiitze
and Pederson devised high-dimensionality vectors to describe the context of
each occurrence of their target word, and then clustered these vectors. They
claim that the better-defined of these clusters correspond to word senses, so
a new occurrence of the word can be disambiguated by representing its
context as a vector and identifying which cluster centroid the vector is closest
to. This system has the characteristic that a context may be close to more
than one cluster centroid, so at times it may be appropriate to classify it as
more than one sense.
Both Clear (1994) and Yarowsky (1995) provide a mechanism for the
user to input the senses between which they would like the system to disam-
biguate. They ask the user to classify a small number of statistically selected
collocates which the algorithm can use to "seed" the process of learning the
patterns which each sense is associated with. The user determines what the
relevant set of senses is when deciding on the senses he or she will assign
364 Adam Kilgarriff

seed collocates to.2 Clear then finds all the words which tend to co-occur
with the nodeword in a large corpus, and quantifies, for a very large number
of words, the evidence that it occurs with each of the seeds, and thus indirectly,
with each sense of the nodeword. Disambiguation then proceeds by summing
the evidence for each sense provided by each context word.
Yarowsky's method is iterative: first, those corpus lines for the nodeword
which contain one of the seed collocates are classified. Then the set of corpus
lines so classified is examined for further indicators of one or other of the
senses of the word. These indicators are sorted, according to the strength of
evidence they provide for a sense. It will now be possible to classify a larger
set of corpus lines, so producing more indicators for each sense, and the
process can be continued until all, or an above-threshold proportion, of the
corpus lines for the word are classified. The ordered list of sense-indicators
will then serve as a disambiguator for new corpus lines.
In the Semantic Concordance project at Princeton a lexicographic team
has been assigning a WordNet (Miller 1990) sense to each noun, verb, adjec-
tive and adverb in a number of texts, thus providing a "gold standard"
disambiguated corpus which can be used for training and evaluating WSD
programs (Landes, Leacock and Tengi 1996).
In 1994--1995, there was an extended discussion of whether WSD should
be one of the tasks in the MUC (Message Understanding Conference) program. 3
This would have provided for competitive evaluation of different NLP groups'
success at the WSD task, as measured against a "benchmark" corpus, in
which each word had been manually tagged with the appropriate WordNet
sense number (as in the Semantic Concordance). Some trials took place, but
the decision was not to proceed with the WSD task as part of the 1996
MUC-6 evaluation, as there was insufficient time to debate and define detailed
policies. The theme has recently been taken up by the Lexicons Special
Interest Group of the Association for Computational Linguistics (ACL), and
a pilot evaluation exercise is taking place in 1998: a milestone on the road
from research to technology.

2.2. Antithesis

Since the publication of Metaphors We Live By (Lakoff and 10hnson 1980)

and Women. Fire and Dangerous Things (Lakoff 1987), there has been one
approach to linguistics - cognitive linguistics - for which metaphor has been
 "[ don't believe in word senses" 365

a central phenomenon. Metaphor is, amongst other things, a process whereby

words spawn additional meanings, and cognitive linguists are correspondingly
interested in polysemy. Lakoff's analysis of the polysemy of mother is hugely
cited. Word sense ambiguity can often be seen as a trace of the fundamental
processes underlying language understanding (Sweetser 1990). The structures
underlying the distinct meanings of words are at the heart of the cognitive
linguistics enterprise (Taylor 1989; Geeraerts 1990).
Working in this framework, Cruse (1995) gives a detailed typology of
polysemy. He distinguishes polysemy, defined according to distinctness of
meaning, from polylexy, which is where, in addition to distinctness of meaning,
distinct lexical entries are required. A word is polysemous but not polylexic
where its non-base meanings are predictable, so they can be generated as
required and need not be stored. He also addresses where readings are
antagonistic and where they are not, and the characteristics of the different
semantic properties, or "facets", of a sense. He uses ambiguity tests to tease
out a number of issues, and a full Cruse lexical entry would contain: a
specification of polysemous senses; their lexical relations including their
relations to each other; whether they were antagonistic or not; the facets,
shared or otherwise, of each, and the extent to which distinct facets of
meaning could operate autonomously, so approach the status of senses on
their own. He considers several varieties of "semi-distinct" readings.
Lexical ambiguity has also moved centre-stage within theoretical and
computational linguistics. Both AAA! (the American Association for Artificial
Intelligence) and ACL have recently devoted workshops to the topic. 4 When
Pustejovsky and others discuss the generative lexicon (Briscoe, Copestake
and Boguraev 1990; Pustejovsky 1991), the generative processes they have
in mind are, again, ones whereby words spawn additional meanings (or, at
least, additional uses). Regular polysemy (Apresjan 1974) has recently been
discussed, and computational mechanisms for addressing it proposed, by
Ostler and Atkins (1991), and Copestake and Briscoe (1995), inter alia. Levin
and colleagues have also been finding systematicity in lexical ambiguity, in
relation to verb classes, their patterns of subcategorization, and their patterns
of alternation (Levin and Rappoport Hovav 1991; Levin 1993; Levin, Song
and Atkins 1997).
This combination of circumstances leads to an odd situation. Much WSD
work proceeds on the basis of there being a computationally relevant, or
useful, or interesting, set of word senses in the language, approximating to
those stated in a dictionary. To the WSD community, word senses are, more
366 Adam Kilgarriff

or less, as the dictionary says. Sometimes not all the sense distinctions
recognized in the dictionary are viewed as salient to the program. WSD
researchers tend to be "lumpers", not "splitters"S (Dol an 1994). (This is not,
of course, to say that WSD authors have not noted the theoretical problems
associated with dictionarys' word senses.) WSD research has gone a long
way on this basis: it is now common for papers to present quantitative
comparisons between the performance of different systems. Meanwhile, the
theoreticians provide various kinds of reason to believe there is no such set
of senses. To get beyond this impasse, the question "what is a word sense?"
needs to be considered more closely.

3. What is a word sense?

"No entity without identity" (Quine 1969).

Or, to know what something is, is to know when something is it. To know
what a word sense SI is, is to know which uses of the word are part of s I and
which are not, probably because they are part of Si where i :;t. 1. If we are to
know what word senses are, we need operational criteria for distinguishing
them.

3.1. Selection and moduLation

A good starting point is Cruse's textbook on LexicaL Semantics (Cruse 1986).

"Lexical units" are the object of his enquiry, and he devotes two substantial
chapters to specifying what they are. He states the heart of the problem thus:
One of the basic problems of lexical semantics is the apparent multiplicity
of semantic uses of a single word form (without grammatical difference). He
addresses in some detail the difference between those cases where the context
selects a distinct unit of sense, from those where it modulates the meaning.
In the pair

(1) Have you put the money in the bank?

(2) The rabbit climbed up the bank.

the two sentences select different meanings of bank, whereas in

 "[ don't believe in word senses" 367

(3) He doesn't often oil his bike.

(4) Madeleine dried off her bike.
(5) Boris's bike goes like the wind.

different aspects of the bicycle - its mechanical parts; its frame, saddle and
other large surfaces; its (and its rider's) motion - are highlighted in each
case. The meaning of bike is modulated differently by each context. 6

3.2. Ambiguity tests

The selection/modulation distinction is closely related to the distinction

between ambiguity and generality, also referred to as "vagueness", "indeter-
minacy" and "lack of specification".7 Where a word is ambiguous, a sense is
selected. Where a word-meaning is general between two readings, any par-
ticular context mayor may not modulate the word-meaning to specify one
or other of the readings. Thus, hand is unspecified between right hands and
left hands; some sentences modulate the meaning to specify a right or left
hand, as in When saluting, the hand should just touch the forehead, while
others do not. H
Clearly, bank is ambiguous between the readings demonstrated above;
bike is not. But for many reading-pairs, the answer is not c1ear: 9

(6) I planted out three rows of beans yesterday.

(7) Cook the beans in salted water.

(8) The cottage was charming.

(9) Our hosts were charming.

(10) Bother! I was about to talk to John, but now he's disappeared! (NOT-
HERE)
(11) I can'tfind it anywhere, it seems to have disappeared. (CAN'T-FIND)

A number of tests have been proposed for determining whether a word is

ambiguous or general between two meanings. They are catalogued in Zwicky
and Sadock (1975), eruse (1986), ten Hacken (1990) and Geeraerts (1993).
Here, I shall describe only one of the more successful tests, the "crossed
readings" one.
368 Adam Kilgarriff

(12) Mary arrived with a pike and so did Agnes.

could mean that each arrived with a carnivorous fish, or that each arrived
bearing a long-handled medieval weapon, but not that the one arrived with
the fish and the other with the weapon. On the other hand, in

(13) Tom raised his hand and so did Dick.

each might have raised a right hand, each might have raised a left, or one
might have raised his right, and the other, his left. The question now is, in

(14) Ellen bought some beans, and so did Harry.

is it possible that Ellen bought plants and Harry, food? If so, then the con-
clusion to be drawn from the test is that bean is ambiguous between the
readings, and if not, then it is not. IO

3.2.1. Criticisms of the tests

The tests are generally presented with the aid of an un problematic example
of ambiguity and an unproblematic example of vagueness. This is done in
order to demonstrate what the test is and what the two contrasting outcomes
are. However, this is not to use the tests in anger. What we want of a test is
that it is consistent with our intuitions, where our intuitions are clear, and
that it resolves the question, where our intuitions are unclear. The cross-
reading test fares tolerably well in meeting the consistency condition (though
see Geeraerts 1993 for a contrary view). But do the tests help where intuitions
are unclear? There is little if any evidence that they do. Here I discuss three
classes of problems.
Firstly, it must be possible to construct a plausible test sentence. The
word in its two uses must be able to occur with the same syntax and the
same lexico-grammatical environment. Consider the transitive and intransitive
uses of eat, as in John ate the apple and John ate. Is this a case of ambiguity
or vagueness?

(15) ?Mary ate, and John, the apple.

 "/ don't believe in word senses" 369

is unacceptable, but the reason is that elided constituents must have the same
syntax and subcategorization in both their expressed and elided occurrences.
It might be desirable to treat all words with alternative subcategorization
possibilities as ambiguous. But whether or not that is done, the test still fails
to elucidate on the topic of a word's meaning, where the word has different
syntax in different uses. The test can only be posed where the two uses are
syntactically similar.
The disappear example displays a different variant of this problem. The
CAN'T-FIND and NOT-HERE readings have different aspectual characteristics:
CAN'T-FIND is stative while NOT-HERE is a punctual "achievement" verb.

(16) Martha disappeared and so did Maud.

does not permit a crossed reading, but that is because we cannot construct a
viable aspectual interpretation for the conjoined sentence. Compare

(17) ?I evicted and knew her.11

It is not evident whether there is a conclusion to be drawn regarding polysemy.

In general, one can apply more or less effort into trying to find a test
sentence (and associated context) in which the crossed reading is plausible.
A test is clearly flawed, if, the more ingenuity the investigator displays, the
more of one particular outcome they will get. (The crossed reading test is the
test which suffers least from this flaw, but it is nonetheless in evidence.)
The second point is more general and theoretical. A certain amount of
interpretation of an utterance must have been undertaken before an accep-
tability judgement can be made. Three parts of the interpretation process are
lexical access; parsing; and "pragmatic interpretation", the final stage of incor-
porating the new information into the discourse model. The premise behind
acceptability judgements is that a subject can report on the outcome of the
first two stages, irrespective of what goes on in the third. For a wide range of
syntactic questions, the methodology is widely used and has proved its worth.
Nunberg's (1978) arguments illustrate the hazards of the premise for
questions in lexical semantics. Consider the sentence

(18) ?The newspaper costs 25p and sacked all its staff.

It is anomalous. We cannot place the origin of the anomaly in the lexicon

370 Adam Kilgarriff

unless we grant newspaper two lexical entries, one for a copy of the newspaper
and one for the owner or corporate entity. Then the size of our lexicon will
start to expand, as we list more and more of the possible kinds of referent for
the word, and still it will never be complete. So the origin of the anomaly
must be the interpretation process. But the anomaly seems similar to the
anomaly that occurs with bank. In a case lying between newspaper and bank,
how would we know whether the source of the anomaly was the lexicon or
the interpretation process? In the general case the point at which the lexical
process becomes a general-purpose interpretative one cannot be identified.
There is no accessible intermediate representation in which lexical ambi-
guities are resolved (for acceptable sentences) but in which the contents of
the sentence has not been incorporated into the hearer's interpretation of the
discourse. Geeraerts (1993) presents an extensive critique of the tests along
these lines, presenting evidence that the different tests give contradictory
results, and that even if we constrain ourselves to looking at just one of the
tests, they can all be made to give contradictory results by manipulating the
context in which the item under scrutiny is set.
The third problem is simply the lack of evidence that the tests give stable
results. It will sometimes happen that, for the same reading-pair, an informant
will deem crossed readings possible for some test sentences and not for others.
Or different informants will have conflicting opinions. There are, remarkably,
no careful discussions of these issues in the literature. The merit of the method
of acceptability judgements for syntax rests on the relative stability of their
outcomes: they work (to the extent they do) because linguists agree where
the question marks belong. Preliminary investigations into the stability of
outcomes in lexical semantics suggest that it is severely lacking.

3.3. Psycholinguistics and "semantic priming"

There is a set of findings in psycholinguistics which might allow us to base

an account of "word sense" directly on the mental lexicon. The experimental
paradigm is called "semantic priming". It is well-established that, if I have
just heard the word doctor (the "prime"), and then a sequence of letters (the
"target") is flashed upon a screen and I am asked to identify whether it is a
word or not, I respond faster if it is a word and it is nurse than if it is a word
but unrelated to doctor. 12 If an ambiguous prime such as bank is given, it
turns out that both river and money are primed for.
 "/ don't believe in word senses" 371

If bank is presented in isolation, priming for both river and money is

found for another second or two. In a context which serves to make only one
of these appropriate, after something between 50 and 200 milliseconds a
choice is made and after that only the appropriate target is primed for.
So, for ambiguous words, priming behaviour has a distinct "signature".
Perhaps it is possible to identify whether a word is vague or ambiguous by
seeing whether it exhibits this signature.
The hypothesis is explored by Williams (1992). He looked at adjectives,
for example firm, for which the two readings were represented by solid and
strict. After confirming that the prime, firm, in isolation, primed equally for
solid and strict, he tested to see if solid was primed for when firm occurred
in a strict context, and vice versa, after delays of 250, 500 and 850 ms.
His results were asymmetrical. He identified central meanings (solid) and
non-central ones (strict). Where the context favoured the central reading, the
non-central-sense targets were not primed for. But when the context favoured
the non-central reading, central targets were. The experiments provide evi-
dence that the various meanings of polysemous words are not functionally
independent in language comprehension, and that not all senses are equal, in
their representation in the mental lexicon. Williams discusses the asym-
metrical results in terms of hierarchical meaning structures.
Priming experiments do show potential for providing a theoretical grounding
for distinguishing ambiguity and generality, but more work needs to be done,
and the outcome would not be a simple, two-way, ambiguous/general dis-
tinction. Also, the method would never be practical for determining the
numbers of senses for a substantial number of words. The results of the
experiments are just not sufficiently stable: as Williams says, the priming
task "suffers from a large degree of item and subject variability" (p. 202).

4. Lexicographers, dictionaries and authority

What set of procedures do lexicographers have available to them to pin down

those protean entities, 'meanings'? Faced with the almost unimaginable diversity
of the language they are trying to describe, with the knowledge that what for
the sake of convenience we are pleased to call a language is in many ways a
synthesis of shifting patterns that change from year to year, from locality to
locality, from idiolect to idiolect, how do they arrive at those masterpieces of
consensus, dictionaries? How do they decide what, for the purposes of a
372 Adam Kilgarriff

dictionary, constitutes the meaning of a word, and where, in the case of

po\ysemous words, one meaning ends and the next begins? (Ay to 1983: 89)

In the middle of this debate stand the lexicographers. The word senses
that most WSD researchers aim to discriminate are the product of their
intellectual labours. But this is far from the purpose for which the dictionary
was written.
Firstly, any working lexicographer is well aware that, every day, they are
making decisions on whether to "lump" or "split" senses that are inevitably
sUbjective: frequently, the alternative decision would have been equally valid.
In fact, most dictionaries encode a variety of relations in the grey area between
"same sense" and "different sense": see Kilgarriff (1993) for a description of
the seven methods used in LDOCE (1987).
Secondly, any particular dictionary is written with a particular target
audience in mind, and with a particular editorial philosophy in relation to
debates such as "lumping vs. splitting", so the notion of specifying a set of
word senses for a language in isolation from any particular user group will
be alien to them.
Thirdly, many are aware of the issues raised by Lakoff, Levin, Pustejovsky
and others, with several lexicographers bringing valuable experience of the
difficulties of sense-division to that literature (see below).
Fourthly, the weight of history: publishers expect to publish, bookshops
expect to sell, and buyers expect to buy and use dictionaries which, for each
word, provide a (possibly nested) list of possible meanings or uses. Large
sums of money are invested in lexicographic projects, on the basis that a
dictionary has the potential to sell hundreds of thousands of copies. Investors
will not lightly adopt policies which make their product radically different to
the one known to sell. However inappropriate the nested list might be as a
representation of the facts about a word, for all but the most adventurous
lexicographic projects, nothing else is possible. J3
The division of a word's meaning into senses is forced onto lexicographers
by the economic and cultural setting within which they work. Lexicographers
are obliged to describe words as if all words had a discrete, non-overlapping
set of senses. It does not follow that they do, or that lexicographers believe
that they do.
 "[ don't believe in word senses" 373

4.1. Lexicographicalliterature

Lexicographers write dictionaries rather than writing about writing dic-

tionaries. Little has been written that answers the challenge posed by Ay to in
the quotation above. Zgusta's influential Manual (1971), while stating that
the specification of word meaning is the central task for the lexicographer
(p. 23) and the division of a word's meanings into senses is a central part of
that, gives little guidance beyond admonishments to avoid making too many,
or too few, distinctions (pp. 66-67).
Ay to's own offering in the (1983) paper is the "classical" or "analytic"
definition, comprising genus and differentiae. In choosing the genus term,
the lexicographer must take care to neither select one that is too general -
entity would not do as a genus term for tiger - nor too specific, if the specific
genus term is likely to be unknown by the dictionary users. Where two
meanings of a word have different genus terms, they need treating as different
senses. The next task is to identify the differentiae required to separate out
senses falling under the same genus term. He discusses cup, and argues that
there are three senses, one for the "trophy" sense, one for the varieties
standardly made of china or earthenware, and one for the prototypically
plastic or paper varieties. But his consideration of the arguments for treating
the second and third of these as distinct ends in a welter of open questions.
Stock (1983) is a response to Ay to's piece, and finds it wanting, firstly, in
the circularity involved in using different genus terms to identify distinct
senses - the lexicographer will only look for distinct genus terms after
determining there are distinct senses - and secondly, in that the model cannot
be applied to many words. She looks closely at culture, noting how different
dictionaries have divided the territory that the word covers in quite different
ways, and observes,

It is precisely the lack of clarity in our use of the word culture which makes
it such a handy word to have at one's disposal. It offers, as it were, semantic
extras just because in most uses its possible meanings are not clearly disam-
biguated. What can the dictionary maker do to reflect this state of affairs?
They do not, cannot by their very structure, show that there is slippage
between some of the senses that they give but not between others. (p. 139)

Hanks (1994), looking at climb, and Fillmore and Atkins (1992), studying
the semantic field centred on risk, make similar comments about the
inadequacies of dictionary conventions, and appeal to prototype theory and
374 Adam Kilgarriff

frame semantics for richer frameworks to describe the relationships between

the different ways a word (or word-family) is used.
Stock, Hanks, and Atkins were all involved in the early stages of the
COBUILD project, which, in the early 1980s, broke new ground in lexicog-
raphy through its use of very large computerized language corpora (Sinclair
1987). Good lexicographic practice had long used huge citation indexes, but
being able to see hundreds of instances of a word in context, ordinary and
extraordinary examples thrown together, was a radical development. It has
changed how lexicographers think about meaning. Where Ay to's paper offers
semantic analysis, Stock presents corpus evidence. The lexicographer's primary
source of evidence for how a word behaves switches from subjective to
objective; from introspection to looking at contexts.

4.2. A corpus-based model of word senses

This suggests a quite different answer to the question "what is a word sense?"
Corpus lexicography proceeds approximately as follows. For each word, the
lexicographer

(i) Calls up a concordance l4 for the word;

(ii) Divides the concordance lines into clusters, so that, as far as possible,
all members of each cluster have much in common with each other,
and little in common with members of other clusters;
(iii) For each cluster, works out what it is that makes its members belong
together, re-organising clusters as necessary;
(iv) Takes these conclusions and codes them in the highly constrained
language of a dictionary definition.

Putting the concordance lines into clusters is data-driven rather than

theory-driven. The lexicographer mayor may not be explicitly aware of the
criteria according to which he or she is clustering. The interactions between
the lexicographers' clusters and the automatic clusters produced for Infor-
mation Retrieval purposes, and the potential for automating some of the
clustering that the lexicographer performs, are subjects of current research.
(It is a requirement for corpus lexicography software that it supports manual
clustering [Atkins 1993; CorpusBench 1993; Schulze and Christ 1994].)
Stage (iii) is just a fallible post hoc attempt to make the criteria explicit. The
 "[ don't believe in word senses" 375

senses that eventually appear in the dictionary are the result, at several
removes, of the basic clustering process.
Ambiguity tests failed to provide us with an account of what it meant for
two uses of a word to belong to the same word sense. Once we operationalize
"word sense" as "dictionary word sense", we now have a test that meets the
challenge. The identity test for a word sense in a particular dictionary is that
two usages of the word belong to it if and only if the lexicographer would
have put them in the same cluster. 15
We can now present a different perspective on the ambiguity/generality
debate. Where a word's uses fall into two entirely distinct clusters, it is
ambiguous, but where the clusters are less well-defined and distinct, "vague"
or "unspecified" may be a more appropriate description. There is no reason
to expect to find any clear distinction between the two types of cases.

5. Use, frequency, predictability, and the word sense 16

"Clustering" is a metaphor. It regards corpus lines as points in space with

measurable distances between them. To give the account substance, more must
be said about the ways in which corpus lines may be "close". In this section,
I classify the types of relationships that hold between a word's patterns of
usage, and consider how these considerations relate to lexicography.
There are five knowledge sources which come into play for understanding
how a word contributes to the meaning or communicative intent of the
utterance or discourse it occurs in. If a word in context is interpretable by a
language user, it will be by virtue of these knowledge sources.
Whether a dictionary provides a word sense that matches an instance of
use of the word, is dictated by considerations of frequency and predictability:
if the instance exemplifies a pattern of use which is sufficiently frequent,
and is insufficiently predictable from other meanings or uses of the word,
then the pattern qualifies for treatment as a dictionary sense. A use is pre-
dictable, to the extent that a person reading or hearing it for the first time can
understand it (in all its connotations). Clearly, different dictionaries have
different thresholds of frequency and predictability.
To illustrate the various processes whereby new types of usage may be
added to the repertoire for a word, let us consider the simple single-sense
word, handbag: "a small bag, used by women to carry money and personal
things (British; American English translation: purse)" (LDOCE 1995).
376 Adam Kilgarriff

As the 715 examples in the British National Corpus (BNC)17 make plain,
typical uses involve things being put into, or taken out of, or looked for in
handbags, or handbags being lost, found, stolen, manufactured, admired,
bought or sold. But a couple of dozen examples stretch the limits of the
definition or fall outside it altogether.
First, a proper name, and a reference to a unique object:

(19) the "Drowning Handbag ", an up-market eatery in the best part of
town
(20) an inimitable rendering of the handbag speech in The Importance of
Being Earnest

Next, metonymy, visual metaphor, simile:

(21) She moved from handbags through gifts to the flower shop
(22) "How about you? Did the bouncing handbag find yoU?"IH
(23) a weird, menacing building with bats hanging in the trees like
handbags
(24) Skin generally starting to age like old handbag or bodywork of car

Next, Mrs Thatcher:

(25) from Edward Heath s hip-pocket to Margaret Thatchers handbag and

on to Mr Majors glass of warm beer
(26) Thousands will be disgusted at the way she [Thatcher] is lining her
handbag
(27) send out Mrs Thatcher with a fully-loaded handbag
(28) "If you want to define the Thatcher-and-after era in a single phrase, "
he muses, "'accountants with plenary powers' says it." Well now-/
would have gone for something a little snappier: HA mad cow with a
handbag, " comes to mind as a first attempt.
(29) She [Thatcher] cannot see an institution without hitting it with her
handbag.

The last of these is cited in another citation as the launching-point of verbal

handbag. Of the three verbal citations, all were species of hitting and in two
of them, Mrs. Thatcher was the perpetrator. This sense still survives in 1999
in the following passage from an article in The Times (21 July, 1999):
 "[ don't believe in word senses" 377

(30) Baroness Thatcher chose a Buckingham Palace garden party to

demonstrate to William Hague her unswerving support for Michael
AshcroJt, the embattled Tory treasurer. It was a classic Thatcher hand-
bagging, according to witnesses who observed the encounters ...

Next, and closely related to Mrs. Thatcher, "handbag-as-weapon":

(31) Meg swung her handbag.

(32) determined women armed with heavy handbags
(33) it was time to race the old ladies back to the village for the tea and
scones of Beck Hall. I beat them, but only just - those handbags are
lethal.
(34) old ladies continue to brandish their handbags and umbrellas at the
likes of Giant Haystacks
(35) the blue rinse brigade will be able to turn out in force without having
to travel and give poor Louis Gerstner the handbagging of his life.
(36) Peterborough manager Chris Turner added: "Evidently one of their
players caught one of our players and it was handbags at 10 paces
and then someone threw a punch. "

The final, quite distinct group relates to discos, and the lexical unit dance
round your handbag, a pejorative phrase for the behaviour of certain exclu-
sively female groups at discotheques and dances where - prototypically -
they dance in a circle with their handbags on the floor in the middle. The
conversational speech subcorpus of the BNC provides two instances of the
full form while in the written corpus, the two related corpus lines, both from
music journalism. make only fleeting references to the collocation. and
strikingly indicate a process of lexicalization:

(37) The shoot was supposed to be a secret, but word got out and Hitman
regulars travelled down to Manchester. Two thousand couldn't get
into the club, and tension mounted between trendy regulars (locked
out of their own club) and the Hitman s handbag brigade (shut out of
their programme).
(38) New Yawk drawling rap over Kraftwerks "The Model" just does not
work, no way, no how. Handbag Dis will love it.

All these uses can be traced back to the standard sense: the potential for
378 Adam Kilgarrijf

using the word in the nonstandard way, is (in varying degrees) predictable
from

(i) Its standard meaning and use;

(ii) General linguistic knowledge (e.g. of processes of metonymy, regular
polysemy, and ellipsis, etc., and, in this case, the relation between
words for goods and words for shops or departments of shops where
those goods are sold);
Ciii) General world knowledge (e.g. regarding Mrs. Thatcher, or juvenile
female behaviour at discotheques) and
(iv) Knowledge of related collocations (e.g. "lining their pockets", "WEAPON
at NUMBER paces");
Cv) Taxonomic knowledge.

These five knowledge sources define the conceptual space within which
lexical creativity and productivity, and the idea of a "word sense", are located. 19
Needless to say, they frequently interact in complex ways. In handbags
at ten paces, the speakero assumes the addressee's awareness of handbag-
as-weapon. Note that ?briefcases at ten paces and ?shoulder-bags at ten paces
do not carry the same meaning. Although briefcases and shoulder-bags are
just as viable weapons as handbags, the words briefcase and slwulder-bag do
not carry the "weapon" connotations which make the citation immediately
understandable. Handbag-as-weapon is a feature of the word, over and above
the extent to which it is a feature of the denotation.
In the citation's context, there is no overt reason for a reference to hand-
bag; the people involved are men, not women, so not prototypical handbag-
users, and there is no other reference to femininity. It would appear that the
speaker is aiming to both distance himself from and minimize the significance
of the incident by treating it as a joke. The "duel" metaphor is itself a joke,
and the oddity of handbag in the context of either football or duel, along
with its associations with femininity and Mrs. Thatcher, contributes to the
effect. Moreover, there is a sexist implication that the men were behaving
like women and thereby the matter is laughable.
Interpreting handbags at ten paces requires lexical knowledge of handbag-
as-weapon, collocational knowledge of both form and meaning of "WEAPON
at NUMBER paces", and (arguably) knowledge of the association between
handbags and models of masculinity and femininity.
The "music journalism" use displays some further features. Handbag was
 "[ don't believe in word senses" 379

lexicalized in the clubbing world in circa 1990 as a music genre: the genre
that, in the 1970s and 1980s, certain classes of young women would have
danced round their handbags to.2 1 The coinage emanates from the gay and
transvestite club scene and is redolent with implications, from the appropri-
ation of the handbag as a symbol of gay pride, to changes in the social
situation of women over the last 20 years (and its expression in fashion
accessories), to transvestite fantasies of being naIve 17-year-old girls in a
more innocent age.
To restrict ourselves to more narrowly linguistic matters: the license for
the coinage is via the dance round your handbag collocation, not directly
from handbags. As shown by the spoken corpus evidence, the regular, non-
ironic use of the collocation co-exists with the music-genre use. It is of much
wider currency: all but two of a range of informants knew the collocation,
whereas only two had any recollection of the music-genre use. Also, "handbag"
music (or at least the use of that label) was a 1990-1991 fashion, and the
term is no longer current: 1996 uses of it will probably refer back to 1990-
1991 (as well as back to the 1970s and 1980s). Syntactically, the most
information-rich word of the collocation has been used as a nominal pre-
modifier for other nouns: in the music-genre sense, it is used as other music-
genre words, as an uncountable singular noun, usually premodifying but
potentially occurring on its own: "Do you like jazz/house/handbag?"

5.1. Frequency

These arguments make clear that there is a prima facie case for including
handbag-as-weapon and handbag-as-music-genre as dictionaries senses, and
dance round your handbag as an only partially compositional collocation.
Each exhibits lexical meaning which is not predictable from the base sense.
So why do the dictionaries not list them? The short answer is frequency.
Around 97% of handbag citations in the BNC are straightforward base-sense
uses. The music-genre sense is certainly rare, possibly already obsolete, and
confined to a subculture. The collocation is partially compositional and occurs
just twice in the corpus: for any single-volume dictionary, there will not be
space for vast numbers of partially compositional collocations. Not only is a
lex.icographer "a lex.icologist with a deadline" (Fill more 1988) but also a
lexicologist with a page limit.
380 Adam Kilgarriff

5.2. Analytic definitions and entailments

The handbag-as-weapon sense is rather more common, and a further con-

sideration comes into play. The denotations of base-sense handbag and
handbag-as-(potential)-weapon are the same. Correspondingly, the lexical fact
that there is a use of handbag in which it is conceptualized as a weapon does
not render the mOCE definition untrue. A lexicographer operating according
to the classical approach whose goal was simply to provide necessary and
sufficient conditions for identifying each word's denotation would say that
the "weapon" aspect of meaning was irrelevant to his or her task. A more
pragmatic lexicographer might also follow this line, particularly since space
is always at a premium.
The situation is a variant on autohyponymy (Cruse 1986: 63-65), the
phenomenon of one sense being the genus of another sense of the same
word. The prototypical example is dog (canine vs. male canine). Dog is a
case where there clearly are distinct senses. For knife (weaponlcutlerylbladed-
object), Cruse (1995: 39-40) argues for "an intermediate status" between
monosemy and polysemy, since, on the one hand, "bladed-object" is a
coherent category which covers the denotation, but on the other, in a scenario
where there was a penknife but no cutlery knife at a table setting, one might
reasonably say "I haven't got a knife". COBUILD (1995) distinguishes
"weapon" and "cutlery" senses, while LDOCE (1995) provides a single,
analytically adequate, "bladed-object" sense.
In a discussion of the polysemy of sanction, Kjellmer (1993) makes a
related observation. His goal is to examine how language breakdown is
avoided when a word has antagonistic readings. Nominal sanction is such a
word: in sanctions imposed on Iraq the meaning is akin to "punishment"
("PUN") whereas in the proposal was given official sanction it is related to
"endorsement" ("END"). A first response is that the context disambiguates -
punishment, not support, is the sort of thing you "impose", whereas "give"
implies, by default, a positively-evaluated thing given. Syntax is also a clue:
the plural use is always PUN, whereas determinerless singular uses suggests
END. Kjellmer then finds the following instances:

The process of social control is operative insofar as sanction plays a part in

the individual's behaviour, as well as in the group's behaviour. By means of
this social control, deviance is either eliminated or somehow made compatible
with the function of the social group .... Historically, religion has also functioned
 "[ don't believe in word senses" 381

as a tremendous engine of vindication, enforcement, sanction, and perpetuation

of various other institutions. (KjeUmer 1993: 119)

Here the context does not particularly favour either reading against the
other. In the second case, the co-ordination with both an END word (vindi-
cation) and a PUN one (enforcement) supports both readings simultaneously.
How is this possible, given their antagonism? How come these uses do not
result in ambiguity and the potential for misinterpretation? The answer seems
to be that

we may operate, as readers or listeners, at a general, abstract level and take

the word to mean 'control, authority' until the context specifies for us which
type of control is intended, if indeed specification is intended. In other words,
faced with the dual semantic potentiality of the word, we normally stay at a
higher level of abstraction, where the danger of ambiguity does not exist,
until clearly invited to step down into specificity. (Kjellmer 1993: 120)22

Citations where sanction is unspecified for either PUN or END are rare, and
there is no case for including the unspecified "control" sense in a dictionary.
The example demonstrates a relationship between a lexicographer's
analytic defining strategy and the interpretation process. There are occasions
where a "lowest common denominator" of the usually distinct standard uses
of a word will be the appropriate reading, in a process analogous to the way
an analytically-inclined lexicographer might write a definition for a word
like charming or knife, which would cover the word's uses in two or more
distinct corpus clusters. Some dictionaries use nested entries as a means of
representing meanings related in this way.

6. Implications for WSD

The argument so far exposes a lack of foundations to the concept of "word

sense". But, a WSD researcher might say, "so what?" What are the impli-
cations for practical work in disambiguation?
The primary implication is that a task-independent set of word senses for
a language is not a coherent concept. Word senses are simply undefined
unless there is some underlying rationale for clustering, some context which
classifies some distinctions as worth making and others as not worth making.
For people, homonyms like pike are a limiting case: in almost any situation
382 Adam Kilgarriff

where a person considers it worth their while attending to a sentence con-

taining pike, it is also worth their while making the fish/weapon distinction.
Lexicographers are aware of this: the senses they list are selected accor-
ding to the editorial policy and anticipated users and uses of the particular
dictionary they are writing. WSD researchers have generally proceeded as if
this was not the case: as if a single program would be relevant to a wide
range of NLP applications.
The argument so far shows that there is no reason to expect the same set
of word senses to be relevant for different tasks.
The handbag data show how various the nonstandard uses of handbag
are. These uses are sufficiently predictable or insufficiently frequent to be
dictionary senses (in a dictionary such as WOCE). They are licensed by a
combination of linguistic principles, knowledge of collocations, lexico-
syntactic contexts and world knowledge. Only in a single case, the department
store metonym, is there a plausible linguistic principle for extending the base
meaning to render the nonstandard use interpretable. The data suggest that
little coverage will be gained by an NLP system exploiting generative prin-
ciples which dictate meaning potential. The nonstandard uses of words tend
to have their own particular history, with one nonstandard use often built on
another, the connections being highly specific to a word or lexical field.
The handbag data also indicate how the corpus dictates the word senses.
The BNC is designed to cover a wide range of standard English, so is
consonant with a general-purpose dictionary. The common uses in the one
should be the senses in the other. But, were we to move to a music journalism
corpus, the music-genre sense would be prominent. A 1990s music-journalism
dictionary would include it. The practical method to extend the coverage of
NLP systems to nonstandard uses is not to compute new meanings, but to
list them. Verbal handbag can, if sufficiently frequent, be added to the lexicon
as a synonym for beat; "WEAPON at NUMBER paces" as one for "have an
argument". Given the constraints of the sublanguage of a given NLP application,
and the usually much narrower confines of the knowledge representation
(which defines the meaning distinctions the system can provide an interpre-
tation for) the proliferation of senses is not a problem. For the medium term
future, the appropriate language-engineering response to a use of a word or
phrase, for which there is a valid interpretation in the knowledge represen-
tation but where the system is currently getting the wrong interpretation
because use of the word or phrase does not match that in the lexicon, is to
add another lexical entry.23
 "[ don't believe in word senses" 383

The implications of the account for different varieties of NLP application

are addressed in Kilgarriff (1997a, 1997b).

7. Conclusion

Following a description of the conflict between WSD and lexicological

research, I examined the concept "word sense". It was not found to be
sufficiently well-defined to be a workable basic unit of meaning.
I then presented an account of word meaning in which "word sense" or
"lexical unit" is not a basic unit. Rather, the basic units are occurrences of
the word in context (operationalized as corpus citations). In the simplest
case, corpus citations fall into one or more distinct clusters and each of these
clusters, if large enough and distinct enough from other clusters, forms a
distinct word sense. But many or most cases are not simple, and even for an
apparently straightforward common noun with physical objects as denotation,
handbag, there are a significant number of citations in which it does not
straightforwardly mean what it standardly means. The interactions between
a word's uses and its senses were explored in some detail. The analysis also
charted the potential for lexical creativity.
The implication for WSD is that word senses are only ever defined relative
to a set of interests. The set of senses defined by a dictionary mayor may
not match the set that is relevant for an NLP application.
The scientific study of language should not include word senses as objects
in its ontology. Where "word senses" have a role to play in a scientific
vocabulary, they are to be construed as abstractions over clusters of word
usages. The nontechnical term for ontological commitment is "belief in", as
in "I (don't) believe in ghosts/God/antimatter". One leading lexicographer
doesn't believe in word senses. I don't believe in word senses, either.

Acknowledgements

This research was supported by the EPSRC Grant K 18931, SEAL. I would also like
to thank Sue Atkins, Roger Evans, Christiane Fellbaum, Gerald Gazdar, Bob Krovetz,
Michael Rundell, Yorick Wilks and the anonymous reviewers for their valuable
comments.
384 Adam Kilgarriff

Notes

* This article was first published in slightly different form in Computers and the
Humanities 1997; 31(2): 91-113. Reprinted by permission. © 1997 Kluwer
Academic Publishers.
The title of this article is quoted from Sue Atkins (past President, European
Association for Lexicography; General Editor, Collins-Robert English/French
Dictionary) responding to a discussion which assumed discrete and disjoint word
senses, at The Future of the Dictionary Workshop, Uriage-Les-Bains, October
1994.
1. Since the paper was written, seven years ago, the exercise, christened
SENSEVAL, has now taken place not once but twice, SENSEVAL-l in 1998
(Kilgarriff and Palmer 2000) and SENSEVAL-2 in 2001 (Cotton et al. 2001;
Edmonds and Kilgarriff 2002).
2. In Yarowsky's work, this is just one of the options for providing seeds for the
process.
3. MUC is a series of US Government-funded, competitive, quantitatively-evaluated
exercises in information extraction (MUC-5, 1994).
4. The AAAI Spring Symposium on Representation and Acquisition of Lexical
Information, Stanford, April 1995 and the ACL SIGLEX Workshop on The
Breadth and Depth of Semantic Lexicons, Santa Cruz, June 1996.
5. "Lumping" is considering two slightly different patterns of usage as a single
meaning. "Splitting" is the converse: dividing or separating them into different
meanings.
6. Cruse identifies two major varieties of modulation, of which highlighting is one.
7. See Zwicky and Sadock (1975) for a fuller discussion of the terms and their
sources.
8. Also related to this distinction is the polysemy/homonymy distinction: when do
we have two distinct words, and when, one word with two meanings? Most
commentators agree that there is a gradation between the two, with the distinction
being of limited theoretical interest. For some purposes, the distinction may be
more useful than the vagueness/ambiguity one (Krovetz 1996). In practice, similar
difficulties arise in distinguishing homonymy from polysemy, as in distinguishing
vagueness from ambiguity.
9. Examples (6)-(11) are taken by comparing four state-of-the-art English learners'
dictionaries (c/DE 1995; COBUILD 1995; WOCE 1995; OAWCE 1995) and
finding words where the lexicographers in one team made one decision regarding
what the distinct word senses were, whereas those in another made another. This
immediately has the effect of introducing various factors which have not been
considered in earlier theoretical discussions.
10. For many putatively ambiguous reading-pairs, there are intermediate cases. A
sprouting bean, or one bought for planting, is intermediate between food and
plant. But the possibility of intermediate cases does not preclude ambiguity: whether
 "/ don't believe in word senses" 385

two readings of a word are completely disjoint, permitting no intermediate cases,

is a different question to whether a word is ambiguous. This imposes a further
constraint on ambiguity tests. A speaker might say: "Ellen and Harry must have
bought the same kind of bean, unless, say, Ellen bought plants and Harry bought
beans sold at the supermarket but which he was intending to plant". We should
not infer that bean is vague. Rather, we must insist that both of the crossed
readings are prototypical. (There are of course further difficulties in making this
constraint precise.)
11. Eight out of ten informants found the related sentence, "I loved and married
her", odd. The two who found it acceptable were reading "and" as an indicator
of temporal sequence.
12. This is the "lexical decision" task in a mixed, visual and auditory procedure. It is
one of a variety of versions of semantic priming experiments. The basic effect is
robust across a number of experimental strategies.
13. The format of the dictionary has remained fairly stable since Dr. Iohnson's day.
The reasons for the format, and the reasons it has proved so resistant to change
and innovation, are explored at length in Nunberg (1994). In short, the develop-
ment of printed discourse, particularly the new periodicals, in England in the
early part of the 18th century brought about a re-evaluation of the nature of
meaning. No longer could it be assumed that a disagreement or confusion about
a word's meaning could be settled face-to-face, and it seemed at the time that the
new discourse would only be secure if there was some mutually acceptable
authority on what words meant. The resolution to the crisis came in the form of
10hnson's Dictionary. Thus, from its inception, the modern dictionary has had a
crucial symbolic role: it represents a methodology for resolving questions of
meaning. Hence "the dictionary", with its implications of unique reference and
authority (cf. "the Bible") (Leech 1981). Further evidence for this position is to
be found in McArthur (1987), for whom the "religious or quasi-religious tinge"
(p. 38) to reference materials is an enduring theme in their history; Summers
(1988), whose research into dictionary use found that "settl[ing] family argu-
ments" was one of its major uses (p. 114, cited in Bejoint [1994: 151]); and
Moon (1989) who catalogues the use of the UAD (Unidentified Authorising
Dictionary) from newspapers' letters pages to restaurant advertising materials
(pp. 60-64).
14. By "concordance" I mean a display which presents a line of context for each
occurrence of the word under scrutiny in the corpus, with all occurrences of the
key word aligned. Fuller details are, of course, system specific, but it has rapidly
become evident that this kind of display is the basic requirement for any corpus
lexicography system.
15. A psycholinguistic investigation along these lines is presented in lorgensen
(1990).
16. In a paper on this topic, it is critical to distinguish between uses of a word that
are candidates for treating as distinct senses. and ones that are distinct senses in
386 Adam Kilgarriff

an actual or idealized dictionary. Here and throughout the paper we use "use" for
the former, and "sense" for the latter.
17. For the BNC see http://info.ox.ac.uklbnc. Counts were: handbag 609, handbags
103, handbagging I, handbagged 2.
18. This turns out to be a (sexist and homophobic) in-group joke, as well as a case
of both metonymy and of a distinct idiomatic use of the word. Interestingly, in
the text, "the bouncing handbag" succeeds in referring, even though the idiom is
not known to the addressee, as is made explicit in the text.
19. In my thesis, in the context of an analysis of polysemy, I call the first four
knowledge types homonymy, alternation, analogy and collocation. (Taxonomy
is addressed separately.)
20. This is presented as a quotation of a football manager's spoken comment; quite
whether it is verbatim, or the Daily Telegraph journalist's paraphrase, we shall
never know.
21. Thanks to Simon Shurville for sharing his expertise.
22. Kjellmer implies that the further specification is a temporal process, there being
a time in the interpretation process when the lexical meaning of the word is
accessed but specified for "control" but not for either PUN or END. I see no grounds
for inferring the temporal process from the logical structure.
23. A well-organized, hierarchical lexicon will mean that this need not introduce
redundancy into the lexicon.

References

Apresjan, Juri D.
[974 Regular polysemy. Linguistics 142: 5-32.
Atkins, Sue
1993 Tools for computer-aided corpus lexicography: the Hector project.
Acta Linguistica Hungarica 41: 5-72.
Ay to, John R.
[983 On specifying meaning. In: Reinhard R. K. Hartmann (ed.), Lexicog-
raphy: Principles and Practice, 89-98. Sidcup, UK: Academic Press.
Bejoint, Henri
[994 Tradition and Innovation in Modern English Dictionaries. Oxford:
Oxford University Press.
Briscoe, Edward J., Ann A. Copestake and Branimir K. Boguraev
1990 Enjoy the paper: lexical semantics via lexicology. In: COLlNG 90(2):
42--47, Helsinki.
Brown, Peter, Stephen Della Pietra, Vincent 1. Della Pietra and Robert L. Mercer
199 [ Word sense disambiguation using statistical methods. In: Proceedings,
29th ACL, Berkeley, California.
 "I don't believe in word senses" 387

ClDE
1995 Cambridge International Dictionary of English. Cambridge: Cambridge
University Press.
Clear, Jeremy
1994 I can't see the sense in a large corpus. In: Ferenc Kiefer, Glibor Kiss
and Julia Pajzs (eds.), Papers in Computational Lexicography:
COMPLEX '94, 33-48, Budapest.
COBUlLD
1995 The Collins COBUlW English Language Dictionary. 2nd Edition.
Edited by John McH. Sinclair et al. London.
Copestake, Ann A. and Edward J. Briscoe
1995 Semi-productive polysemy and sense extension. In: James Pustejovsky
and Branimir Boguraev (eds.), Lexical Semantics: The Problem of
Polysemy, 15-67. Oxford: Clarendon.
CorpusBench
1993 CorpusBench Manual. Textware AlS, Copenhagen, Denmark.
Cotton, Scott, Philip Edmonds, Adam Kilgarriff and Martha Palmer (eds.)
200 1 Proceedings of SENSEVAL-2, Second International Workshop on
Evaluating Word Sense Disambiguation Systems. ACL, Toulouse,
France.
Cottrell, Garrison W.
1989 A Connectionist Approach to Word Sense Disambiguation. London:
Pitman.
Cruse, David A.
1986 Lexical Semantics. Cambridge: Cambridge University Press.
1995 Polysemy and related phenomena from a cognitive linguistic view-
point. In: Patrick Saint-Dizier and Evelyne Viegas (eds.), Compu-
tational Lexical Semantics, 33-49. Cambridge: Cambridge University
Press.
Dolan, WiIIiam B.
1994 Word sense disambiguation: clustering related senses. In: COLlNG
94, Tokyo.
Edmonds, Philip and Adam Kilgarriff (eds.)
2002 Evaluating word sense disambiguation systems. Journal of Natural
Language Engineering 8(4): 279-292.
FiIImore, Charles
1988 Keynote lecture, British Association of Applied Linguistics conference,
Exeter, September 1988.
Fillmore, Charles J. and Beryl T. S. Atkins
1992 Towards a frame-based lexicon: the semantics of risk and its neighbors.
In: Adrienne Lehrer and Eva Kittay (eds.), Frames, Fields and
Contrasts, 75-102. New Jersey: Lawrence Erlbaum.
388 Adam Kilgarriff

Gale, WiIIiam, Kenneth Church and David Yarowsky

1992 Estimating upper and lower bounds on the performance of word-
sense disambiguation programs. In: Proceedings, 30th ACL, 249-
156.
1993 A method for disambiguating word senses in a large corpus. Computers
and the Humanities 26(1-2): 415-539.
Geeraerts, Dirk
1990 The lexicographical treatment of prototypical polysemy. In: Savas
L. Tsohatzidis (ed.), Meanings and Prototypes: Studies in Linguistic
Classification, 195-210. London: Routledge.
1993 Vagueness's puzzles, polysemy's vagueness. Cognitive Linguistics
4(3): 223-272.
Guthrie, Joe A, Louise Guthrie, Yorick Wilks and Homa Aidinejad
1991 Subject-dependent co-occurrence and word sense disambiguation. In:
Proceedings, 29th ACL, Berkeley, California.
Guthrie, Louise, Brian M. Slator, Yorick Wilks and Rebecca Bruce
1990 Is there content in empty heads? In: COL/NG 90(3), 138-143, Helsinki.
Hanks, Patrick
1994 Linguistic norms and pragmatic exploitations or, why lexicographers
need prototype theory, and vice versa. In: Ferenc Kiefer, Gabor Kiss
and Julia Pajzs (eds.), Papers in Computational Lexicography:
COMPLEX '94, 89-113, Budapest.
Hearst, Marti A.
1991 Noun homograph disambiguation using local context in large text
corpora. In: Using Corpora: Proc. Seventh Ann. Con! of the UW
Centre for the New OED, 1-22, Waterloo, Canada.
Jensen, Karen and Jean-Louis Binot
1987 Disambiguating prepositional phrase attachment by using on-line
dictionary definitions. Computational Linguistics 13: 251-260.
Jorgensen, Julia C.
1990 The psychological reality of word senses. Journal of Psycholinguistic
Research 19(3): 167-190.
Kilgarriff. Adam
1992 Polysemy. Ph.D. thesis. University of Sussex, CSRP 261, School of
Cognitive and Computing Sciences.
1993 Dictionary word sense distinctions: an enquiry into their nature.
Computers and the Humanities 26(1-2): 365-387.
1997a Evaluating word sense disambiguation programs: progress report. In:
Robert John Gaizauskas (ed.), Proceedings, SALT Workshop on
Evaluation in Speech and Language Technology, 114-120, Sheffield,
UK, June.
1997b Foreground and background lexicons and word sense disambiguation
 "I don't believe in word senses" 389

for information extraction. In: Proceedings, Workshop on Lexicon

Driven Information Extraction, 51-62. Frascati, Italy, July.
Kilgarriff, Adam and Martha Palmer (eds.)
2000 Special issue on SENSEVAL: evaluating word sense disambiguation
programs. Computers and the Humanities 34(1-2). Dordrecht: Kluwer.
Kjellmer, Goran
1993 Multiple meaning and interpretation: the case of sanction. Zeitschrift
fUr Anglistik und Amerikanistik 41(2): 115-123.
Krovetz, Robert
1996 Surprises under the hood: an investigation of word meanings and
information retrieval. To appear.
Lakoff, George
1987 Women, Fire and Dangerous Things. Chicago: University of Chicago
Press.
Lakoff, George and Mark Johnson
1980 Metaphors We Live By. Chicago: University of Chicago Press.
Landes, Shari, Claudia Leacock and Randee Tengi
1996 Building semantic concordances. In: Christiane Fellbaum (ed.),
WordNet: An ELectronic LexicaL Database. Cambridge, MA: MIT
Press.
LDOCE
1987 Longman Dictionary of Contemporary English, New Edition. Edited
by Della Summers. Harlow.
1995 Longman Dictionary of Contemporary English, 3rd Edition. Edited
by Della Summers. Harlow.
Leech, Geoffrey
1981 Semantics. Cambridge: Cambridge University Press.
Lesk, Michael E.
1986 Automatic sense disambiguation using machine readable dictionaries:
how to tell a pine cone from an ice cream cone. In: Proceedings,
/986 SIGDOC Conference, Toronto, Canada.
Levin, Beth
1993 English Verb Classes and ALternations. Chicago: University of Chicago
Press.
Levin, Beth and Malka Rappoport Hovav
1991 Wiping the slate clean: a lexical semantic exploration. Cognition 41:
123-151.
Levin, Beth, Grace Song, and Beryl T. S. Atkins
1997 Making sense of corpus data: a case study of verbs of sound. Inter-
nationaL JournaL of Corpus Linguistics 2(1): 23-64.
McArthur, Tom
1987 Worlds of Reference. Cambridge: Cambridge University Press.
390 Adam Kilgarriff

McRoy. Susan W.
1992 Using multiple knowledge sources for word sense discrimination.
Computational Linguistics 18(1): 1-30.
Miller. George
1990 Wordnet: An on-line lexical database. International Journal of
Lexicography (special issue) 3(4): 235-312.
Moon. Rosamund
1989 Objective or objectionable? Ideological aspects of dictionaries. English
Language Research, 3: Language and Ideology. 59-94.
MUC-5
1994 Proceedings, Message Understanding Conference. DARPA.
Nunberg, Geoffrey
1978 The Pragmatics of Reference. Bloomington. IN: University ofIndiana
Linguistics Club.
1994 The once and future dictionary. Presentation at The Future of the
Dictionary Workshop. Uriage-Ies-Bains. France, October 1994.
OAWCE
1995 Oxford Advanced Learner's Dictionary of Current English. Fifth
Edition. Oxford: Oxford University Press.
Ostler, Nicholas and Beryl T. S. Atkins
1991 Predictable meaning shift: some linguistic properties of lexical impli-
cation rules. In: lames Pustejovsky and Susan Bergler (eds.), Lexical
Semantics and Knowledge Representation: ACL SIGLEX Workshop,
Berkeley, California.
Pustejovsky, lames
1991 The generative lexicon. Computational Linguistics 17(4): 409-441.
Quine, Willard van O.
1969 Speaking of objects. In: Quine, Willard van 0 .• Ontological Relativity,
1-25. New York: Columbia University Press.
Schulze, Bruno and Oliver Christ
1994 The IMS Corpus Workbench. Institut fur maschinelle Sprachverarbeitung,
Universitat Stuttgart.
Schiitze, Hinrich and lan O. Pederson
1995 Information retrieval based on word senses. In: Proceedings, ACM
Special Interest Group on Information Retrieval.
Sinc1air, John M. (ed.)
1987 Looking up: An Account of the COBUILD Project in Lexical Com-
puting. London: Collins.
Slator, Brian M.
1988 Lexical semantics and a preference semantics pars er. Technical Report
MCCS-88-16. Computing Research Laboratory, New Mexico State
University, New Mexico.
 "[ don't believe in word senses" 391

Sparck Jones. Karen

1986 Synonymy and Semantic Classification. Edinburgh: Edinburgh Uni-
versity Press.
Stock, Pen elope F.
1983 Polysemy. In: Proceedings, Exeter Lexicography Conference, 131-140.
Summers, Della
1988 The role of dictionaries in language learning. In: Ronald A. Carter
and Michael McCarthy (eds.), Vocabulary and Language Teaching,
111-125. London: Longman.
Sweetser. Eve
1990 From Etymology to Pragmatics: Metaphorical and Cultural Aspects
of Semantic Structure. Cambridge: Cambridge University Press.
Taylor, John
1989 Linguistic Categorization: Prototypes in Linguistic Theory. Oxford:
Oxford University Press.
ten Hacken, Pius
1990 Reading distinction in machine translation. In: COLlNG 90, Helsinki.
Veronis, Jean and Nancy M. Ide
1990 Word sense disambiguation with very large neural networks extracted
from machine readable dictionaries. In: COLlNG 90(2), 389-394,
Helsinki.
Williams, John N.
1992 Processing polysemous words in context: evidence for interrelated
meanings. Journal of Psycholinguistic Research 21: 193-218.
Yarowsky, David
1992 Word-sense disambiguation using statistical models of Roget's categ-
ories trained on large corpora. In: COLlNG 92, Nantes.
1995 Unsupervised word sense disambiguation rivalling supervised methods.
In: ACL 95, 189-196, MIT.
Zgusta, Ladislav
1971 Manual of Lexicography. The Hague: Mouton.
Zwicky, Arnold M. and J. M. Sadock
1975 Ambiguity tests and how to fail them. Syntax and Semantics 4:
1-36.
Senses and texts

Yorick Wilks

1. Introduction

Empirical, corpus-based, computational linguistics (CL) has reached by now

into almost every crevice of the subject, and perhaps pragmatics will soon
succumb. Semantics, if we may assume the sense-tagging task is semantic,
has shown striking progress in the last five years and, in Yarowsky's most
recent work (Yarowsky 1995), has produced very high levels of success -
over 90% - well above the key bench-mark figure of 62% correct sense
assignment achieved at an informal experiment in New Mexico in about
1990, in which each word was assigned its first sense listed in the Longman
Dictionary of Contemporary English (WOCE).
A crucial question in this paper will be whether recent work in sense-
tagging has in fact given us the breakthrough in scale that is now obvious
with, say, part-of-speech tagging. Our conclusion will be that it has not, and
that the experiments so far, however high their success rates, are not yet of
a scale different from those of the previous generation of linguistic, symbolic-
Al or connectionist approaches to the very same problem.
A historian of our field might glance back at this point to Small, Cottrell
and Tanenhaus (1988), who surveyed the AI-symbolic and connectionist
traditions of sense-tagging at just the moment when corpus-driven empirical
methods began to revive, but had not been published. All the key issues still
unsettled are discussed there and that collection showed no naivety about
the problem of sense resolution with respect only to existing lexicons of
senses. It was realized that that task was only meaningful against an assump-
tion of some method for capturing new (new to the chosen lexicon, that is)
senses and, most importantly, that although existing lexicons differed, they
did not differ arbitrarily much. The book further demonstrated that there was
also strong psychological backing for the reality of word senses and for
empirical methods of locating them from corpora without any prior assump-
tions about their number or distribution (e.g. in early versions of Plate's work,
published later in Wilks et aI. 1990; see also ]orgensen 1990).
394 Yorick Wilks

My purpose in this paper will be to argue that Kilgariff's negative claims

are wrong, and his errors must be combated, while Yarowsky is largely right
although I have some queries about the details and the interpretation of his
claims. Both authors, however, agree that this is a traditional and important
task: one often cited as being a foundational lacuna in, say, the history of
Machine Translation (MT), because of the inability of early Natural Language
Processing (NLP) systems to carry it out. It was assumed by many, in that
distant period, that if only word-sense ambiguity could be solved by the
process we are calling sense-tagging, then MT of high quality would be
relatively straightforward. Like many linguistic tasks, it then became an end
in itself, like syntactic parsing; and now that it is, we would claim, firmly in
sight (despite Kilgarriff) it is far less clear that its solution will automatically
solve a range of traditional problems like MT. But clearly it would be a
generally good tool to have available in NLP and a triumph if this long-
resistant task of CL were to yield.

2. The very possibility of sense-tagging

Kilgarriff's (1993) paper is important because it has been widely cited as

showing that the senses of a word, as distinguished in a dictionary such as
LDOCE, do not cover the senses actually carried by most occurrences of the
word as they appear in a corpus. If he can show that, it would be very
significant indeed, because that would imply that sense-tagging word occur-
rences in a corpus by means of any lexical data based on, or related to, a
machine-readable dictionary or thesaurus is misguided. I want to show here
that the paper does not demonstrate any such thing. Moreover, it proceeds
by means of a straw man it may be worth bringing back to life!
That straw man, Kilgarriff's starting point, is the "bank model" (BM) of
lexical ambiguity resolution, which he establishes by assertion rather than
quotation, though it is attributed to Small, Hirst and Cottrell as well as the
present author. In the BM, words have discrete meanings, and the human
reader (like the ideal computer program) knows immediately which meaning
of the word applies (Kilgarriff 1993: 367), "given that a word occurrence
always refers to one or the other, but not both" of the main meanings that a
word like bank is reputed to have. In the BM, the set of senses available for
a word does not depend on which particular dictionary you start with, but is
somehow abstractly fixed. The main argument of Kilgarriff's paper is to
 Senses and texts 395

distinguish a number of relationships between maCE senses that are not

discrete in that way, and then to go on to an experiment with senses in a
corpus. But first we should breathe a little life back into the BM straw man:
those named above can look after themselves, but here is a passage from
Wilks (1972: 12):

... it is very difficult to assign word occurrences to sense classes in any manner
that is both general and determinate. In the sentences "I have a stake in this
country" and "My stake on the last race was a pound" is "stake" being used
in the same sense or not? If "stake" can be interpreted to mean something as
vague as "stake as any kind of investment in any enterprise" then the answer
is yes. So, if a semantic dictionary contained only two senses for "stake":
that vague sense together with "Stake as a post", then one would expect to
assign the vague sense for both the sentences above. But if, on the other
hand, the dictionary distinguished "Stake as an investment" and "Stake as an
initial payment in a game or race" then the answer would be expected to be
different. So, then, word sense disambiguation is relative to the dictionary of
sense choices available and can have no absolute quality about it.

QED, one might say, since the last sentences seem to show very much
the awareness (a quarter of a century ago, but in the context of a computer
program for sense-tagging) that sense choice may not be exclusive if defined,
as it must be, with respect to a particular dictionary. Hence, in my view, the
BM is no more than a straw man because writers of the dark ages of CL
were as aware as Kilgarriff of the real problems of dictionary senses versus
text occurrences.
In general, it is probably wise to believe, even if it is not always true, that
authors in the past were no more naive than those now working, and were
probably writing programs, however primitive and ineffective, to carry out
the very same tasks as now (e.g. sense-tagging of corpus words). More
importantly, Wilks (1972), which created an approach called preference
semantics, was essentially a study of the divergence of corpus usage from
lexical norms (or preferences), and developed in the 1970s into a set of
processes for accommodating divergent/nonstandard/metaphorical usage to
existing lexical norms, notions that Kilgarriff seems to believe only developed
in a much later and smarter group of people around 1990, which includes
himself, but also, for example, Fass, whose work was a direct continuation
of that quoted above. Indeed, in Wilks (1972) procedures were programmed
(and run over a set of newspaper editorials) to accommodate such "divergent"
396 Yorick Wilks

corpus usage of one word to that of an established sense of a different word

in the same text, while in Wilks (1978) programmed procedures were specified
to accommodate such usage by constructing completely new sense entries
for the word itself.
A much more significant omission, one that bears directly on Kilgarriff's
main claim and is not merely an issue of historical correctness, is the lack of
reference to work in New Mexico and elsewhere (e.g. Cowie, Guthrie and
Guthrie 1992) on the large-scale sense-tagging of corpora against a machine-
readable dictionary (MRD)-derived lexical database. These were larger-scale
experiments whose results directly contradict the result he is believed to have
proved. I shall return to this point in a moment. The best part of Kilgarriff's
paper is his attempt to give an intuitive account of developmental relations
between the senses of a word. He distinguishes "generalizing metaphors" (a
move from a specific case to a more general one) from "must-be-theres"
(the applicability of one sense requires the applicability of another, as when
an act of matricide requires there to be a mother) from "domain shift" (as
when a sense in one domain, like mellow [of wine], is far enough from the
domain of mellow [of a personality] to constitute a sense shift).
It is not always easy to distinguish the first two types, since both rest on
an implication relationship between two or more senses. Again, the details
do not matter: what he has shown convincingly is that, as in the Wilks (1972)
quotation, the choice between senses of a given word is often not easy to
make because it depends on their relationship, the nature of the definitions
and how specific they are. I suspect no-one has ever held a simple-minded
version of the BM, except possibly Fodor and Katz, who, whatever their
virtues, had no interest at all in lexicography.
The general problem with Kilgarriff's analysis of sense types is that he
conflates:

(i) Text usage different from that shown in a whole list of stored senses
for a given word e.g. in a dictionary (which is what his later experi-
ment will be about), with
(ii) Text usage divergent from some "core" sense in the lexicon.

Only the second is properly in the area of metaphor/metonymy or "grinding"

(Copestake and Briscoe 1991) work of the group in which he places himself,
and it is this phenomenon to which his classification of sense distinctions
summarized above properly belongs. This notion requires some idea of sense
 Senses and texts 397

development, of senses of a word extending in time in a non-random manner,

and is a linguistic tradition of analysis going back to Givon (1967). However,
the straw man BM and the experiment Kilgarriff then does on hand-tagging
of senses in text, all attach to the first, unrelated, notion which does not
normally imply the presence of metonymy or metaphor at all, but simply an
inadequate sense list. Of course, the two types may be historically related, in
that some of the (i) list may have been derived by metaphoricaVmetonymic
processes from a (ii) word, but this is not so in general. This confusion of
targets is a weakness in the paper, since it makes it difficult to be sure what
he wants us to conclude from the experiment. However, since I shall show
his results are not valid, this distinction may not matter too much.
One might add here that Kilgarriff's pessimism has gone hand in hand
with some very interesting surveys he has conducted over the internet on the
real need for word-sense disambiguation by NLP research and development.
And one should note that there are others (e.g. Ide and Veronis 1994) who
have questioned the practical usefulness of data derived at many sites from
MRDs. My case here, of course, is that it has been useful, both in our own
work on sense-tagging (Stevenson and Wilks 2001; Cowie, Guthrie and
Guthrie 1992) and in that of Yarowsky, using Roget and discussed below.
Kilgarriff's experiment, which has been widely taken to be the main
message of his paper, is not described in much detail. He does not give the
reader the statistics on which his result was based even though the text quite
clearly contains a claim (p. 378) that 87% of (non-monosemous) words in
his text sample have at least one text occurrence that cannot be associated
with one and only one LDOCE sense. Hence, he claims, the poor old BM is
refuted, yet again. One must note here that this is a claim about word types:
so if 87% of them have at least one "defective" token, this could mean that
a very small number of text words (i.e. tokens) are "defective".
So, Kilgarriff's claim (about word types) is wholly consistent with, for
example, 99% of text usage (of word tokens) being associated with one and
only one dictionary sense! Thus the actual claim in the paper is not at all
what it has been taken to show, and is highly misleading.
But much empirical evidence tells also against the claim Kilgarriff is
believed to have made. Informal analyses by Georgia Green (1989) suggested
that some 20% of text usage (i.e. of word tokens) could not be associated
with a unique dictionary sense. Consistent with that, too, is the use of simu-
lated annealing techniques by Cowie, Guthrie and Guthrie (1992) at the
Computing Research Laboratory (CRL) New Mexico to assign LDOCE
398 Yorick Wilks

senses to a corpus. In that work, it was shown that about 75%-80% of word
usage could be correctly associated with LDOCE senses, as compared with
hand-tagged control text. That figure was subsequently raised substantially
by additional filtering techniques (particularly in Stevenson and Wilks 2001).
The two considerations above show, from quite different sources and
techniques, the dubious nature of Kilgarriff's claim. Antal (1963) argued
long ago (and some believe Wierzbicka 1989 maintains some version of such
a position) that words have only core senses. On such a view, dictionariesl
lexicons should express that single sense and leave all further sense refine-
ment to some other process, such as real world knowledge manipulations,
AI if you wish, but not a process that uses the lexicon.
Since the CRL result suggested that the automatic sense-resolution pro-
cedures worked very well (near 80%) at the homograph level (i.e. the sense-
cluster level, such as the LDOCE dictionary uses) rather than the ordinary
sense level (the latter being where Kilgarriff's examples all lie), one possible
way forward for NLP would be to go some of the way with Antal's views
and restrict lexical sense distinctions to the homograph level. Then sense-
tagging could perhaps be done at the success level of part-of-speech tagging.
Such a move could be seen as changing the data to suit what you can
accomplish, or as reinstating AI and pragmatics within NLP for the kind of
endless, context-driven, inferences we need in real situations.
This suggestion is rather different from Kilgarriff's conclusion. He pro-
poses that the real basis of sense distinction be established by usage clustering
techniques applied to corpora. This is an excellent idea and recent work at
IBM (Brown et al. 1991) has produced striking non-seeded clusters of corpus
usages, many of them displaying a similarity close to an intuitive notion of
sense (or you can now explore Google's own version of the very same
methodology at labs.google.com!sets).
But there are serious problems in moving any kind of lexicography,
traditional or computational, onto any such basis. Hanks (p.c.) has claimed
that a dictionary could be written that consisted entirely of usages, and has
investigated how those might be clustered for purely lexicographic purposes,
yet it remains unclear what kind of volume could result from such a project
or who would buy it and how they could use it. One way to think of such a
product would be the reduction of monolingual dictionaries to thesauri, so
that to look up a word becomes to look up which row or rows of context-
bound semi-synonyms it appears in. Thesauri have a real function both for
native and non-native speakers of a language, but they rely on the reader
 Senses and texts 399

knowing what some or all of the words in a row or class mean because they
give no explanations. To reduce word sense separation to synonym classes,
without explanations attached would limit a dictionary's use in a striking
way.
If we then think not of dictionaries for human use but NLP lexicons, the
situation might seem more welcoming for Kilgarriff's suggestion, since he
could be seen as suggesting, say, a new version of WordNet (Miller 1985)
with its synsets established not a priori but by statistical corpus clustering.
This is indeed a notion that has been kicked around in NLP for a while and
is probably worth a try. There are still difficulties: firstly, that any such
clustering process produces not only the clean, neat, classes like mM's
(Hindullew/ChristianlBuddhist) example but inevitable monsters, produced
by some quirk of a particular corpus. Those could, of course, be hand weeded
but that is not an automatic process.
Secondly, as is also well known, what classes you get, or rather, the
generality of the classes you get, depends on parameter settings in the clustering
algorithm: those obtained at different settings mayor may not correspond
nicely to, say, different levels of a standard lexical hierarchy. They probably
will not, since hierarchies are discrete in terms of levels and the parameters
used are continuous but, even when they do, there will be none of the
hierarchical terms attached, of the sort available in WordNet (e.g. ANIMAL or
DOMESTIC ANIMAL). And this is only a special case of the general problem of
clustering algorithms, well known in information retrieval, that the clusters
so found do not come with names or features attached.
Thirdly, and this may be the most significant point for Kilgarriff's pro-
posal, there will always be some match of such empirical clusters to any
new text occurrence of a word and, to that degree, sense-tagging in text is
bound to succeed by such a methodology, given the origin of the clusters
and the fact that a closest match to one of a set of clusters can always be
found. The problem is how you interpret that result because, in this
methodology, no hand-tagged text will be available as a control since it is
not clear what task the human controls could be asked to carry out. Subjects
may find traditional sense-tagging (against e.g. mOCE senses) hard but it
is a comprehensible task, because of the role dictionaries and their associated
senses have in our cultural world. But the new task (attach one and only one
of the classes in which the word appears to its use at this point) is rather less
well defined. But again, a range of original and ingenious suggestions may
make this task much more tractable, and senses so tagged (against WordNet
400 Yorick Wilks

style classes, though empirically derived) could certainly assist real tasks
like MT even if they did not turn out wholly original dictionaries for the
book-buying public.
There is, of course, no contradiction between, on the one hand, my sugges-
tion for a compaction of lexicons towards core or homograph senses, done
to optimize the sense-tagging process and, on the other, Kilgarriff's suggestion
for an empirical basis for the establishment of synsets, or clusters that
constitute senses. Given that there are problems with wholly empirically-
based sense clusters of the sort mentioned above, the natural move would be
to suggest some form of hybrid derivation from corpus statistics, taken
together with some machine-readable source of synsets: WordNet itself,
standard thesauri, and even bilingual dictionaries which are also convenient
reductions of a language to word sets grouped by sense (normally by reference
to a word in another language, of course). As many have now realized, both
the pure corpus methods and the large-scale hand-crafted sources have their
virtues, and their own particular systematic errors, and the hope has to be
that clever procedures can cause those to cancel, rather than reinforce, each
other. But all that is future work, and beyond the scope of a critical note.
In conclusion, it may be worth noting that the BM, in some form, is
probably inescapable, at least in the form of what Pustejovsky (1995) calls a
"sense enumerative lexicon", and against which he inveighs for some 20
pages before going on to use one for his illustrations, as we all do, including
all lexicographers. This is not hypocrisy but a confusion close to that between
(i) and (ii) above: we, as language users and computational modellers, must
be able, now or later, to capture a usage that differs from some established
sense (problem [iiJ above), but that is only loosely connected to problem (i),
where senses, if they are real, seem always to come in lists and it is with
them we must sense-tag if the task is to be possible at all.

3. Recent experiments in sense-tagging

We now turn to the claims (Gale et al. 1992; see also Yarowsky 1992, 1993,
1995) that:

(i) Word tokens in a single text tend to occur with a smaller number of
senses than often supposed; and, most specifically,
(ii) In a single discourse a word will appear in one and only one sense,
 Senses and texts 401

even if several are listed for it in a lexicon, at a level of about 94%

likelihood for non-monosemous words (a figure that naturally becomes
higher if the monosemous text words are added in).

These are most important claims if true for they would, at a stroke, remove
a major excuse for the bad progress of MT, make redundant a whole sub-
industry of NLP, namely sense resolution, and greatly simplify the currently
fashionable NLP task of sense-tagging texts by any method whatever (e.g.
Cowie, Guthrie and Guthrie 1992; Bruce and Wiebe 1994).
Gale et al.'s claim would not make sense-tagging of text irrelevant, of
course, but it would allow one to assume that resolving any single token of
a word (by any method at all) in a text would then serve for all occurrences
in the text, at a high level of probability.
Gale et al.'s claims are not directly related to those of Kilgarriff, who
aimed to show only that it was difficult to assign text tokens to any lexical
sense at all. Indeed, Kilgarriff and Gale et al. use quite different procedures:
Kilgarriff's is one of assigning a word token in context to one of a set of
lexical sense descriptions, while Gale et al.'s is one of assessing whether or
not two tokens in context are the same sense or not. The procedures are
incommensurable and no outcome on one would be predictive for the other:
Gale et al.'s procedures do not use standard lexicons and are in terms of
closeness-of-fit, which means that, unlike Kilgarriff's, they can never fail to
match a text token to a sense, defined in the way they do (see below).
However, Gale et al.'s claims are incompatible with Kilgarriff's in spirit
in that Kilgarriff assumes there is a lot of polysemy about and that resolving
it is tricky, whereas Gale et al. assume the opposite.
Both Kilgarriff and Gale et a1. have given rise to potent myths about
word-sense-tagging in text that I believe are wrong, or at best unproven.
Kilgarriff's paper, as we saw earlier, has some subtle analysis but one
crucial statistical flaw. Gale et al. 's is quite different: it is a mush of hard-to-
interpret claims and procedures, but ones that may still, nonetheless, be
basically true.
Gale et al.'s methodology is essentially impressionistic: the texts they
chose are, of course, those available, which turn out to be Grolier's Encyclo-
pedia. There is no dispute about one-sense-per-discourse (their name for claim
[ii] above) for certain classes of texts: the more technical a text the more
anyone, whatever their other prejudices about language, would expect the
claim to be true. Announcing that the claim had been shown true for
402 Yorick Witks

mathematical or chemical texts would surprise no-one; encyclopaedias are

also technical texts.
Their key fact in support of claim (i) above, based on a sense-tagging of
97 selected word types in the whole Encyclopedia, and sense-tagged by the
statistical method described below, was that 7,569 of the tokens associated
with those types are monosemous in the corpus, while 6,725 are of words
with more than two senses. Curiously, they claim this shows "most words
(both by token and by type) have only one sense" (Gale et al. 1992: 63). I
have no idea whether to be surprised by this figure or not but it certainly
does nothing to show that "Perhaps word sense disambiguation is not as
difficult as we might have thought" (p. 67). It shows me that, even in fairly
technical prose like that of an encyclopaedia, nearly half the words occur in
more than one sense.
And that fact, of course, has no relation at all to mono- or polysemousness
in whatever base lexicon we happen to be using in an NLP system. Given a
large lexicon, based on say the Oxford English Dictionary (OED), one could
safely assume that virtually all words are polysemous. As will be often the
case, Gale et al.'s claim at this point is true of exactly the domain they are
dealing with, and their (non-stated) assumption is that any lexicon is created
for the domain text they are dealing with and with no relation to any other
lexicon for any other text. One claim per discourse, one might say.
This last point is fundamental because we know that distinctions of sense
are lexicon- or procedure-dependent. Kilgarriff faced this explicitly, and took
LDOCE as an admittedly arbitrary starting point. Gale et al. never discuss
the issue, which makes all their claims about numbers of senses totally, but
inexplicitly, dependent on the procedures they have adopted in their experi-
ments to give a canonical sense-tagging against which to test their claims.
This is a real problem for them. They admit right away that few or no
extensive hand-tagged sense-resolved corpora exist for control purposes, So,
they must adopt a sense-discrimination procedure to provide their data that
is unsupervised. This is where the ingenuity of the paper comes in, but also
its fragility. They have two methods for providing sense-tagged data against
which to test their one-sense-per-discourse claim (ii).
The first rests on a criterion of sense distinction provided by corres-
pondence to differing non-English words in a parallel corpus, in their case
the French-English Canadian Hansard because, as always, it is there. So, the
correspondence of duty to an aligned sentence containing either devoir or impot
(i.e. 'obligation' or 'tax') is taken as an effective method of distinguishing
 Senses and texts 403

the obligation/tax senses of the English word, which was indeed the criterion
for sense argued for in Dagan and Itai (1994). It has well-known drawbacks:
most obviously that whatever we mean by sense distinction in English, it is
unlikely to be criterially revealed by what the French happen to do in their
language.
More relevantly to the particular case, Gale et al. found it very hard to
find plausible pairs for test, which must not of course share ambiguities across
the FrenchlEnglish boundaries (as interest/interet do). In the end they were
reduced to a test based on the six (!) pairs they found in the Hansard corpus
that met their criteria for sense separation and occurrence more than 150
times in two or more senses. In Gale et al.'s defence one could argue that,
since they do not expect much poJysemy in texts, examples of this sort would,
of course, be hard to find. Taking this bilingual method of sense-tagging
for the six-word set as criterial they then run their basic word sense dis-
crimination method over the English Hansard data. This consists, very
roughly, of a training method over lOO-word surrounding contexts for 60
instances of each member of a pair of senses (hand selected), i.e. for each
pair 2 x 60 x 100 = 12,000 words. Notice that this eyeballing method is not
inconsistent with anything in Kilgarriff's argument: Gale et al. selected 120
contexts in Hansard for each word that did correspond intuitively to one of
the (French) selected senses. It says nothing about any tokens that may have
been hard to classify in this way. The figures claimed for the discrimination
method against the criteria} data vary between 82% and 100% (for different
word pairs) of the data for that sense correctly discriminated.
They then move on to a monolingual method that provides sense-tagged
data in an unsupervised way. It rests on previous work by Yarowsky (1992)
and uses the assignment of a single Roget category (from the 1,042) as a
sense-discrimination. Yarowsky sense-tagged some of the Grolier corpus in
the following way: lOO-word contexts for words like crane (ambiguous
between 'bird' and 'machinery') are taken and those words are scored by
(very roughly, and given interpolation for local context) which of the 1,042
Roget categories they appear under as tokens. The sense of a given token of
crane is determined by which Roget category wins out: e.g. category 348
(TOOLS/MACHINERY) for the machinery contexts, and category 414 (ANIMALS/
INSECTS) for the bird contexts. Yarowsky (1992) claimed 93% correctness for
this procedure over a sample of 12 selected words, presumably checked
against earlier hand-tagged data.
The interpolation for local effects is in fact very sophisticated and involves
404 Yorick Wilks

training with the lOO-word contexts in Grolier of all the words that appear
under a given candidate Roget head, a method that they acknowledge intro-
duces some noise, since it adds into the training material Grolier contexts
that involve senses of a category 348 word, say, that is not its MACHINERY
sense (e.g. crane as a bird). However, this method, they note, does not have
the problems that come with the Hansard training method, in particular the
notion, unacceptable to many, that sense distinctions in a source languge are
defined by those in the target language (advocated in e.g. Dagon and Itai
1994), in that, for any word pair where two French and English words had
identical senses, there could be no sense distinction at all.
In a broad sense, this is an old method, perhaps the oldest in lexical
computation, and was used by Masterman (reported in Wilks 1972) in what
was probably the first clear algorithm ever implemented for usage dis-
crimination against Roget categories as sense-criterial. In the very limited
computations of those days the hypothesis was deemed conclusively falsified;
i.e. the hypothesis that any method overlapping the Roget categories for a
word with the Roget categories of neighbouring words would determine an
appropriate Roget category for that word in context.
This remains, I suspect, an open question: it may well be that Yarowsky's
local interpolation statistics have made the general method viable, and that
the lOO-word window of context used is far more effective than a sentence.
It may be the 12 words that confirm the disambiguation hypothesis at 93%
would not be confirmed by 12 more words chosen at random (the early
Cambridge work did at least try to resolve by means of Roget all the words
in a sentence). But we can pass over that for now, and head on, to discuss
Gale et al.'s main claim (ii) given the two types of data gathered.
Two very strange things happen at this point as the Gale et al. paper
approaches its conclusion: namely, the proof of claim (ii) or one-sense-per-
discourse. Firstly, the two types of sense-tagged data just gathered, especially
the Roget-tagged data, should now be sufficient to test the claim, if a 93%
level is deemed adequate for a preliminary test. Strangely, the data derived
in the first part of the paper are never used or cited and the reader is not told
whether Yarowsky's Roget data confirm or disconfirm (ii).
Secondly, the testing of (ii) is done purely by human judgement: a "blind"
team of the three authors and two colleagues who are confronted by the
Oxford Advanced Learners Dictionary (OALD) main senses for one of nine
test words, and who then make judgements of pairs of contexts for one of
the nine words drawn from a single Grolier article. The subjects are shown
 Senses and texts 405

to have pretty consistent judgements and, of 54 pairs of contexts from the

same article, 51 shared the same sense and three did not.
Notice here that the display of the DAW senses is pointless, since the
subjects are not asked to decide which if any DAW sense the words appear
in, and so no Kilgarriff-style problems can arise. The test is simply to assign
SAME or NOTSAME, and there are some control pairs added to force discrimi-
nation in some cases.
What can one say of this ingenious mini-experiment? Lexicographers
traditionally distinguish "Iumpers" and "splitters" among colleagues: those
who tend to break up senses further and those who go for large, homonymic,
senses, of which Antal would be the extreme case. Five Gale et al. colleagues
(one had to be dropped to get consistency among the team) from a "lumper"
team decided that 51 out of 54 contexts for a word in a single encyclopaedia
article (repeated for eight other words) are in the same sense. Is this signifi-
cant? I suspect not very, and nothing at all follows to support the myth of
discovery that has grown round the paper: the team and data are tiny and not
disinterested. The Grolier articles are mini-texts where the hypothesis would,
if true, surprise one least. Much more testing is needed before a universal
hypothesis about text polysemy enters our beliefs. Of course, they may in
the end be right, and all the dogma of the field so far be wrong.
More recently, Yarowsky (1993, 1995) has extended this methodology in
two ways: firstly, he has established a separate claim he calls "one sense per
collocation", which is quite independent of local discourse context (which
was the separate "one-sense-per-discourse" claim) and could be expressed
crudely by saying that it is highly unlikely that the following two sentences
(with the "same" collocations for plants) can both be attested in a corpus:

(l) Plastic plants can fool you if really well made [= organic]
(2) Plastic plants can contaminate whole regions [= factory]

One's first reaction may be to counter-cite examples like Un golpe bajo

which can mean either a low blow in boxing, or a score one below par, in
golf, although golpe could plausibly be said to have the same collocates in
both cases. One can dismiss such examples (due to Jim Cowie in this case)
by claiming both readings are idioms, but that should only focus our mind
more on what Yarowsky does mean by collocation.
That work, although statistically impressive, gives no procedure for large-
scale sense-tagging taken alone, since one has no immediate access to what
406 Yorick Wilks

cue words would, in general, constitute a collocation sufficient for disambigu-

ation independent of discourse context. An interesting aspect of Yarowsky's
paper is that he sought to show that on many definitions of sense and on
many definitions of collocation (e.g. noun to the right, next verb to the left
etc.) the hypothesis was still true at an interesting level, although better for
some definitions of collocation than for others.
In his most recent work, Yarowsky (1995) has combined this approach
with an assumption that the earlier claim ([ii]: one-sense-per-discourse) is
true, so as to set up an iterative bootstrapping algorithm that both extends
disambiguating collocational keys (Yarowsky 1993) and retrains against a
corpus, while at the same time filtering the result iteratively by assuming
(ii): i.e. that tokens from the same discourse will have the same sense. The
result, on selected pairs (as always) of bisemous words is between 93% and
97% (for different word pairs again) correct against hand-coded samples,
which is somewhat better than he obtained with his Roget method (93% in
1991) and better than figures from Schiitze and Pederson (1995) who produce
unsupervised clusterings from a corpus that have to be related by hand to
intelligible, established, senses.
However, although this work has shown increasing sophistication, and
has the great advantage, as he puts it, of not requiring costly hand-tagged
training sets but instead "thrives on raw, unannotated, monolingual corpora
- the more the merrier" (Yarowsky 1995: 173), it has the defect at present
that it requires an extensive iterative computation for each identified bisemous
word, so as to cluster its text tokens into two exclusive classes that cover
almost all the identified tokens. In that sense it is still some way from a
general sense-tagging procedure for full text corpora, especially one that tags
with respect to some generally acceptable taxonomy of senses for a word.
Paradoxically, Yarowsky was much closer to that last criterion with his (1991)
work using Roget that did produce a sense-tagging for selected word pairs
that had some "objectivity" predating the experiment.
Although Yarowsky compares his work favourably with that of Schiitze
and Pederson in terms of percentages (96.7 to 92.2) of tokens correctly tagged,
it is not clear that their lack of grounding for the classes in an established
lexicon is that different from Yarowsky's, since his sense distinctions in
his experiments (e.g. plant as organic or factory) are intuitively fine but
pretty ad hoc to the experiment in question and have no real grounding in
dictionaries.
 Senses and texts 407

4. Conclusion

It will probably be clear to the reader by now that a crucial problem in

assessing this area of work is the fluctuation of the notion of word sense in
it, and that is a real problem outside the scope of this paper. For example,
sense as between binary oppositions of words is probably not the same as
what the Roget categories discriminate, or words in French and English in
aligned Hansard sentences have in common.
Another question arises here about the future development of large-
scale sense-tagging: Yarowsky contrasts his work with that of efforts like
Cowie, Guthrie and Guthrie (1992) that were dictionary-based, as opposed
to (unannotated) corpus-based like his own. But a difference he does not
bring out is that the Cowie, Guthrie and Guthrie work, when optimized with
simulated annealing, did go through substantial sentences, mini-texts if you
will, and sense-tag all the words in them against WOCE at about the 80%
level. It is not clear that doing that is less useful than procedures like Yarowsky's
that achieve higher levels of sense-tagging but only for carefully selected
pairs of words, whose sense-distinctions are not clearly dictionary-based,
and which would require enormous prior computations to set up ad hoc sense
oppositions for a useful number of words.
These are still early days, and the techniques now in play have probably
not yet been combined or otherwise optimized to give the best results. It
may not be necessary yet to oppose, as one now standardly does in MT, large-
scale, less accurate, methods, though useful, with other higher-performance
methods that cannot be used for practical applications. That the field of sense-
tagging is still open to further development follows if one accepts the aim of
this paper which is to attack two claims, both of which are widely believed,
though not at once: that sense-tagging of corpora cannot be done, and that
it has been solved. As many will remember, MT lived with both these,
ultimately misleading, claims for many years.

Acknowledgements

Work referred to was supported by the NSF under grant #IRI 9101232 and the
ECRAN project (LE-2110) funded by the European Commission's Language
Engineering Division. The paper is also indebted to comments and criticisms from
Adam Kilgarriff, David Yarowsky, Karen Sparck Jones, Rebecca Bruce and members
408 Yorick Witks

of the CRL-New Mexico and University of Sheffield NLP groups. The mistakes are
all my own, as always.

References

Antal, Laszlo
1963 Questions of Meaning. The Hague: Mouton.
Brown, Peter E, Stephen A. Della Pietra, Vincent J. Della Pietra and Robert L. Mercer
1991 Word sense disambiguation using statistical methods. In: Proceedings.
ACL 91, 142-146.
Bruce, Rebecca and Janyce Wiebe
1994 Word-sense disambiguation using decomposable models. In: Pro-
ceedings. ACL 94, 251-256.
Copes take, Ann and Ted Briscoe
1991 Lexical operations in a unification-based framework. In: Proceedings.
ACL Siglex Workshop, 61-68, Berke1ey, CA.
Cowie, Jim, Joe Guthrie and Louise Guthrie
1992 Lexical disambiguation using simulated annealing. In: Proceedings,
COLlNG 92, 903-909.
Dagan, Ido and Alon Itai
1994 Word sense disambiguation using a second language monolingual
corpus. Computational Linguistics 20(4): 563-596.
Gale, William A., K. Church, Kenneth Ward and David Yarowsky
1992 One sense per discourse. In: Proceedings, Fourth DARPA Speech
and Natural Language Workshop, 61-69, Monterey, CA.
Givon, Talmy
1967 Transformations of Ellipsis, Sense Development and Rules of Lexical
Derivation. SP-2896, Systems Development Corp., Santa Monica,
CA.
Green, Georgia M.
1989 Pragmatics and Natural Language Understanding. Hillsdale, NJ:
Erlbaum.
Ide, Nancy and Jean Veronis
1994 Have we wasted our time? In: Proceedings, International Workshop
on the Future of the Dictionary, 103-107, Grenoble.
Jorgensen, Julia C.
1990 The psychological reality of word senses. Journal of Psycho linguistic
Research 19(3): 167-190.
Kilgarriff, Adam
1993 Dictionary word-sense distinctions: an enquiry into their nature.
Computers and the Humanities 26(1-2): 365-387.
 Senses and texts 409

Miller, George
1985 WordNet: a dictionary browser, In: Proc. First Internat. Cont on
Information in Data. Waterloo OED Centre, Canada.
Pustejovsky, James
1995 The Generative Lexicon. Cambridge, MA: MIT Press.
Schutze, Hinrich and Jan O. Pederson
1995 Information retrieval based on word sense. In: Proceedings, Fourth
Annual Symposium on Document Analysis and Information Retrieval,
86-94, Las Vegas, NV.
Small, Steven I., Garrison W. Cottrell and Michael K. Tanenhaus
1988 Lexical Ambiguity Resolution: Perspectives from Psycholinguistics,
Neuropsychology, and Artificial Intelligence. San Mateo, CA: Morgan
Kaufmann.
Stevenson, Mark and Yorick Wilks
2001 The interaction of knowledge sources for all-word word-sense
disambiguation. Computational Linguistics 27(3): 44-62.
Wierzbicka, Anna
1989 Semantics, Culture and Cognition. Oxford: Oxford University Press.
Wilks, Yorick
1972 Grammar; Meaning and the Machine Analysis of Language. London:
Routledge.
1978 Making preferences more active. Artificial Intelligence 11: 55-75.
Wilks, Yorick, Dan C. Fass, Cheng Ming Guo, lames E. McDonald, Tony Plate
and Brian M. Slator
1990 Providing machine-tractable dictionary tools. Journal of Machine
Translation 5: 183-199.
Wilks, Yorick, Brian M. Slator and Louise M. Guthrie
1996 Electric Words. Cambridge, MA: MIT Press.
Yarowsky, David
1992 Word-sense disambiguation using statistical models of Roget's categ-
ories, trained on very large corpora. In: Proceedings, COLlNG 92,
801-809.
1993 One sense per collocation. In: Proceedings, ARPA Human Language
Technology Workshop, 112-117, Princeton, NI.
1995 Unsupervised word-sense disambiguation rivalling supervised methods.
In: Proceedings, ACL 95, 189-196, MIT.
Index

AAAI (American Association for association 55

Artificial Intelligence) 365 Association for Computational
Aarts, Jan 235, 236 Linguistics (ACL) 364, 365
absorption 269, 2701 Atkins, Beryl T. S. 373-374
abstractivistlcognitivist difference 14- auto-antonymic polysemy 269, 271t
15 auto-converse po\ysemy 269,271t
accommodation 9 autohyponymous polysemy 229, 380
ACL see Association for Ay to, John R. 371-372,373
Computational Linguistics
Adelung, Johann Christoph 51
adjectives 236, 236t, 237-243, 2371, Bakhtin, Michael 65
239t, 240t, 2411, 248 n3, Barcelona, Antonio 195,210 n5
371 Bartning, Inge 235
adpreps 150 n21, 150 n26 Beard, Robert 217, 221
affixes see derivational affixes Bierwisch, Manfred 66, 279
Agnoli, Franca 326 Biese, Alfred 53, 56-57
all ove r 40-41 Billow, Richard 318, 324
ambiguity Black, Max 50
discourse context 60--61 Blank, Andreas 61,268
and fuzziness 162-163, 164, 325 bleaching
maxim of manner 122 censure of 253-254
purposeful ambiguity 13 development of polysemy 254,
and vagueness 229, 278, 368, 261-263
371, 375 disaster nouns 255-256,2561
see also ambiguity tests; grammaticalization 254
homonymy; word sense interpretation problem 259-261
disambiguation (WSD) scale of disaster 257-259,2571,
ambiguity tests 228-229, 367-370, 2591
375, 384-385 nn9-10 blending 64, 80
ambivalence 164,169-170 category extension 83-85
American Association for Artificial example 80-82, 8lf
Intelligence (AAAI) 365 gradients 86-92
Antal, Laszlo 398, 405 linguistic form 92-93
antiphrastic polysemy 269,27It metaphor 83-84
Apresjan, Jurij D. 65,217 principles 82-83
argument structure 12,340 Bloomfield, Leonard 95
Aristotle 55, 61, 234 Blumenberg, Hans 65
Aronoff, Mark 95 BNC see British National Corpus
412 Index

Boguraev, Branimir 280,289 n14 see also get: acquisition

Bolinger, Dwight 239 Chomsky, Noam 96, 199/
Breal, Michel 4,49,58,60-63,67 nI, Chukovsky, Kornei 320
268, 338 Cl ark, Eve 234
Brekle, Herbert E. 235 Cl ark, Herbert H. 137, 234, 303
Briscoe, Ted 289 n 14 Clarke, David 217, 249 n12, 326
British National Corpus (BNC) 255- Clear, Jeremy 363-364
256, 257, 376, 377, 379, 382 clustering 374, 375, 398, 399
Brown, Cecil H. 7 co-hyponymous polysemy 268-269,
Brugman, Claudia 97, 107,230 270t
Bryson, Bill 62, 338 COBUlW see Collins COBUlW
Biihler, Karl 56, 64, 65 English Language Dictionary
building block metaphor 33, 37-40 cognitive linguistics 4-5, 65-66, 96-
Burchfield, Robert 253 97, 274, 288 n7, 333
Burke, Kenneth 50 cognitive models 31-34, 41-43
Bybee, Joan 41 building block metaphor 33, 37-
40
corpus model 44 n2, 165
Cadiot, Pierre 66 cut-and-paste model 44 n2
Ca1bert, Joseph 235, 236 idioms 40-41
Carabine, B. 325 lexicon and syntax 40-41
Caramazza, Alfonso 65 semiotic model 34-37
Caramelli, Nicoletta 318 cognitive semantics 50, 53, 65, 281
catachresis 52 cognitive triangle 205-207, 206/
categorization 97,200-201,202-203, classical quartet vs. modern trio
210 nn4-5, 234,248 nI, 324 207-208
CHILDES database 337 domains 206-207,207/
children entities 206
development of metaphor 317- in lexicography 211 n8
318,323-324,327 Collins COBUIW English Language
drawing 326 Dictionary 374, 380
figurative language use 317, 319, Colombo, Lucia 168
320-321, 324-325 common sense 79
metaphor in poetry 320-321, communication 13-14
322-323, 327 complementary polysemy 8, 12
overextension 324, 325, 326, complex expressions 37, 38,42-43
327 comprehension 298-300,302, 326,
phonological devices in poetry 327 see also irony
321-322, 323 conceptual integration 7-8,9-11,90,
polysemy 6-7, 12, 325, 327, 338 93, 103
semantic development 62-63, conceptual knowledge 282-283, 284/
325, 340, 353 constructions 39-40, 339-340 see also
story-writing 323 modifier-head constructions
 Index 413

context effects Dagan, Ido 403

idioms 297-298 Danesi, Marcel 54
irony 299, 300, 301-302, 308 Dante Alighieri 58-59
metaphor 299,318 de- 227, 228
polysemous words 299-300 Deane,Paul D. 274,276,281,326
reading comprehension 298, 302 Democritus 58
salience and 297-299, 308 derivational affixes
contextual variation 267, 273 ambiguity tests 228-229
de- 227,228
and poly se my 274-275
diminutives 219-220, 220t
polysemy and homonymy 276-
dis- 227, 228
277
-dom 225-226, 226/
referential class and semantic
English prefixes 218-219,218t,
relation 277-278,277t 219t
see also vagueness -hood 225-226, 227/
contrastive polysemy 8, 12 meta- 223-225, 223/
conversation see comprehension; monosemy or polysemy 228-230
context effects; irony; neologisms 223-228
salience polysemy 217-218, 230 n I
Copestake, Ann 289 n 14 -ship 225-226, 227, 227/
corpus model suffixes (agent/subject arguments)
cognitive model 44 n2, 165 221-222, 221t, 222/
lexicography 165,374-375, 382, un- 227, 228
398 Dewell, Robert 117
see also sense-tagging dialogic syntax 306-307
Coseriu, Eugenio 36, 66, 289 n12 dictionaries see lexicography
Cottrell, Garrison W. 361,362 Dietrich, Wolf 289 n12
Coulson, Seana 85 diminutives 219-220, 220t
dis- 227, 228
Cowie, Jim 397-398, 405, 407
disaster
Croft, William 96, 210 n3
development of polysemy 261-
cross-linguistic differences 325-326,
263
327
interpretation problem 259-261
Cruse, David A. 8, 149 n15, 201, 208, nouns 255-256, 256t
209 nI, 2lO n3, 229, 274, scale of 257-259, 257t, 259t
365, 366, 367, 380 discourse rules 282-285, 284/
culture Dolan, William B. 366
frequency of figurative language -dom 225-226, 226/
use 320-321 doublets 242t
metaphor and 285, 319-321, Dowker, Ann 318,320,321
322-324, 326 Downing, Pamela 248 n4
and polysemy 7,325-326 Dressier, Wolfgang U. 219
cut-and-paste model 44 n2 Drew, Paul 303
414 Index

Du Bois, John W. 306-307 methodological consequences

Du Marsais, Cesar Chesnau 50, 52, 164-166
54 nonfixedness 164. 169
polysemy 164. 166-167, 170
production test 167. 168. 169.
EC (entity--category) fallacy 203, 208, 185-187
209 sorting test 168. 188-190
Eco, Umberto 58, 59 variability 163-164. 169-170
embodiment 96, 335 see also get: flexible meaning;
Erdmann, Karl Duo 64 pitiiii; qualia
etymology Flores d' Arcais. Giovanni B. 168
divergent 276 Fritz. Gerd 15
popular 269,271t Fuchs. Catherine 66
event structure 12 fuzziness 162-163. 164,325

Fass. Dan 319 Gabelentz, Georg von der 253

father 86-90 Gale. William A. 400-405
Fauconnier, Gilles 57,83.84. 149 Gardner. Howard 318. 324
n12, 285 Geeraerts. Dirk 10-11.66.207.229.
Fein, Ofer 299 234.288.288 n7, 367. 370
figurative language Gerber, Gustav 56
children's use of 317,319,320- Gernsbacher. A. Morton 298
321, 324-325 get: acquisition 333-335
culture and 320-321 argument/construction approach
development of metaphor 317- 337
318,323-324,327 levels of polysemy 341
and polysemy 325, 326-327 lexically-based approach 337
vs. literal language 318--319 hypotheses 341-342
see also metaphor study method 342-344
Fillmore, Charles J. 65, 373-374. 379 results
flexibility 162. 163, 164, 169, 183- productions 344-345. 346-
185 347t,352
flexible meaning 11-12, 161-162, 185 rankings 345, 348f, 349-
ambivalence 164. 169-170 350. 349f, 351f
analyses 169-170 semantic development 338, 352-
difference evaluation test 168, 353
187-188 get: construction
flexibility 162. 163. 164. 169. grammar 339-340
183-185 diachronic evolution 334
fuzziness 162-163. 169. 325 expressions 11
intersubjective differences 164. phrasal verbs 11. 339
170 polysemy 334, 336-338. 341
 Index 415

prefabricated chunks 340 Green, Georgia M. 397

protoscene 335-336. 336/ Grice. Paul 122, 309 n 1
see also get: acquisition; get: Grober, Ellen 65
flexible meaning
get: flexible meaning 170-171
ability 172-173. 179-181. 191 handbag 375-379. 382
nIl Hanks, Patrick 373-374. 398
basic-level meaning types 172 Harris, Roy 190 n 1
carrying 177-179 Hemenway, Kathleen 209 n I
change of location 172 history of poly se my 49-67
change of possession 172. 173- metaphor 50-57
174. 174t before Breal 49, 58-60
change of state 172 Breal's study 60-63
difference evaluation test 168, Breal to the present 63-67
177, 178, 187-188 homonymy 4, Il. 35, 58, 190 n2, 273,
major categories 172 276-277, 384 n8 see also
motion 172 ambiguity
obligation 172-173, 179-181 -hood 225-256, 227/
obtaining 172, 173-177, 174t. Hopper, Paul J. 254,262
177-179 Householder, Fred W. 58
production test 167, 174t, 178, Hughes, Geoffrey 254
185-187, 190-191 n9 Hume, David 55
reaching a stage 173, 179-181, Hunt, Earl 326
191 nIl hyponymy 289 nIl see also taxonomy
receiving 172, 173-177, 174t
stable possession 172-173
starting 173 I don't know 41
static meanings 172 idioms 38,40-41,297-298
understanding 172-173 image-schemas 96, 113, 147 nI, 148
Gibbs, Raymond w., Jr. 302,319, nll
337 irony
Giora, Rachel 297,299,302 comprehension 297-299, 300-
give 8-9 301,302
Giv6n, Talmy 397 context effects 299, 300, 301-
Goethe, Johann Wolfgang von 53-54 302, 308
Goldberg. Adele 39-40, 339-340, direct access hypothesis 302, 306
340t lexical access 301-302
Goldberg, Lea 320 research method 303-305
graded polysemy 8-11 results and discussion 306-307
graded salience hypothesis 297-299, sample analysis 305
300. 306. 307, 308 and spontaneous discourse 299,
Grady, Joseph 122. 337 300-308
grammaticalization 254 and teasing 303
416 Index

Israel, Michael 337 151 n26, 166, 190 n2, 203-

Itai, Alon 403 204, 210--211 nn6-7,211f,
333
language acquisition see children; get:
1ackendoff, Ray 96, 113, 148 n12 acquisition
1akobson, Roman 50, 195 language rules 282, 284f, 285
10hnson, Chris 337 LDOCE see Longman Dictionary of
10hnson, Mark 50, 57, 137, 142, 147 Contemporary English
ni, 195, 197, 198,200,281- Leech, Geoffrey 385 n 13
282, 319, 364 Lees, Robert 249 n6
10hnson, Samuel 385 n 13 leggings 234-5
Joos, Martin 218 Lehrer, Adrienne 223
Leibniz, Gottfried Wilhelm 50, 51
Leitzky, Eva 235
Kant, Immanuel 54 Levi,ludith 235
Kilgarriff, Adam 66,372,394-400, lexemes 217
401,402 lexical access 301-302, 307
Kimball,10hn 171 lexical conceptual paradigms 12
Kipling, Rudyard 260 lexical ellipsis 269,270t
Kirsner, Robert S. 36 lexical polysemy 281,286
Kjellmer, Goran 380--381 lexicography
Kleiber, Georges 66, 274 analytic definitions and
Koch, Peter 289 n 12 entailments 380--381
Koevecses, Zoltan 195, 197,210 n2 citations 362
Kotthoff, Helga 308 classification 61
Kreitzer, Anatol 97, 99, 100--104, 107, clustering 374, 375. 398, 399
117-118,147 n5 cognitive triangle in 211 n8
Kurath, Hans 57 corpus-based model 165, 374-
Kuteva, Tania 326 375, 382, 398
Kuusisto, Pekka 190 n6 dictionary format 385 n13
differentiae 373
frequency 379
Lakoff, George 50, 54, 97, 99-100, genus 373
107, 117-118, 128, 137, 140, implications for WSD 381-382
142, 151 n24, 152 n30, 195, literature 373-374
197, 198, 200, 209-210 n2, "lumping" vs. "splitting" 366,
230,281-282,288 n7, 319, 372, 384 n5, 405
364, 365 predictability 378
Lambert, 10hann Heinrich 50, 52-53, role of dictionary 385 n 13
54 sense division 59-60, 371-372,
Langacker, Ronald W. 9,34,36,37, 373,375.385 n16, 395
39, 50,98, 107, 108, 109, thesauri 398
115, 143. 148 n6. 149 n16, use 375-379,385 n16
 Index 417

see also sense-tagging; word language and 89-90

sense disambiguation meaning potential 79-80
(WSD) mental representations 55
lexicology, theoretical 362-366 os tension 53, 67 n7
lexicon role of dictionary 385 n 13
categorization 97 salience 62, 297-300, 307-308
definition 40, 95-96 synchronic/diachronic 65-66
embodied meaning 96 see also flexible meaning
generative 365 ~eillet, Antoine 254
innovation 267 ~erlini Barbaresi, Lavinia 219-220
Lieberman, Philip 15, 16 meronymy 210 n3, 269
likely 92 meta- 222-225, 223f
Lindner, Susan 143 metaphor 209
Lindstromberg, Seth 154 n32, 171, "a category is a container" 200-
177 201,202,210 n5
Ljung, ~agnus 235, 236, 238 cognitive triangle 205--208
LM (landmark) 99, 151-152 n27 comprehension 326, 327
Locke, John 50, 51 in conceptual blending 83-84
Longman Dictionary of Contemporary conceptual metaphor 281-282,
English (LDOCE) 372, 375, 283, 284, 285
380,393,394,397,398,407 conduit metaphor 201
Love, Nigel 37 context and 299,318
Lyons, John 208 cross-cultural differences 285,
319-321, 322-324, 326
development of 317-318,323-
~achine Translation (~T) 394, 400, 324, 327
401,407 domains 207, 207f
man 272f influences on 324
~andler, Jean 113 in poetry 320--321,322-323,327
~archand, Hans 218,219,220,223, polysemy 268, 270t, 271t
226, 228, 230 n2 research history 50--57, 61, 64,
~asterman, ~argaret 362-363, 404 65
maxim of manner 122 simile 317,320,322-323
maximalistlminimalist dilemma 14-15 see also building block metaphor
~cArthur, Tom 385 n 13 metaphoric polysemy 268, 270t, 2711
~cCawley, James D. 65 metonymic polysemy 270t, 271 t
meaning metonymy 55,209, 271t
acquisition (see children) cognitive triangle 205-208
classification 61 conceptual metonymy 281-282,
common sense 79 283, 285
construction 89-90, 92-93, 97-98, contiguity 195-196,208,269,
1l4f, 117-122, 149 n18 281, 282, 283
embodied meaning 96 definitions 195, 196-197,208
418 Index

domains 206-207,207f Niedzielski, Henry 171

entities 206, 211-212 n9 NLP see Natural Language Processing
entity--category (EC) fallacy 203, NLU (Natural Language
209 Understanding) 361
folk understanding vs. expert Norrick, Neal R. 217
knowledge 200--201 Nunberg, Geoffrey 195. 204. 369. 385
network model 203-205, 203f, n13
204f, 210-211 n6, 211f
partonomy-taxonomy (PT) fallacy
198-203, 198f, 199f, OAW see Oxford Advanced Learners
200f, 209 Dictionary
polysemy 270t,271t OED see Oxford English Dictionary
referential metonymy 244-245 orthography 37
Mill, James 55 Ortony, Andrew 65
Mithun, Marianne 230 n I ostension 53. 67 n7
Modern Maturity 225 over 97-98, 145-146
modifier-head constructions 235-242, A-B-C Trajectory Cluster 114f,
237t, 239t, 240t, 241t, 248 117-121. 126-132
nn3-4, 249 nn7-9 above-and-beyond (excess I)
monosemy 147 n3, 150-151 n23, 279, sense 128-130.
288 n7, 289 nl2 129f
Montanari, A. 318 completion sense 130-131,
Moon, Rosamund 385 nl3 13 If
Moore, Kevin 319 on-the-other-side-of sense
MT see Machine Translation 126-128, 127f, 151
MUC (Message Understanding n25
Conference) program 364, transfer sense 131-132.
384 n3 132f, 151-152 n27
above and proximal
examining sense 134-136.
Natural Language Processing (NLP) 135f
361, 362, 382, 394, 398, 399, focus-of-attention sense
401 136-137. 136f
Natural Language Understanding adpreps 131. 132
(NLU) 361 cluster of senses 125
Nelson, Katherine 324 covering sense 132-133. 133f,
neologisms 223-228 152 nn28-30
Nerlich, Brigitte 208,217,249 n12, distinct senses 105-106
326 full-specification approach 98-
New Oxford Dictionary of English. The 100. 147 n5. 147 nn4-5.
195.211 n8, 255, 263, 333, 147t
334 partial-specification approach
New York Times. The 80 100--104. 1O:lf, 103f
 Index 419

primary sense 107, 108, 109, 148 liking 182, 183

n9 maintaining contact 182
protoscene Ill, lIlf, 112 manual holding 182
reflexivity metaphorical retaining 183
reflexive sense 143-144, obligation 182, 183
144f organizing 183
repetition sense 144-145, production test 168, 191 n12
145f sorting test 168, 183, 188-90
semantic network 124-125, 125f, upholding a condition 183
153-154 n31 wearing 183
vertical elevation ("up" cluster) Plato 58
137, 137f polylexy 365
control sense 112, 140-142, polyonymy 58
142f polyphony 65
more sense 138, 139f polysemy
over-and-above (excess 11) definitions 32, 42, 325
sense 139-140, 139f as dynamic mass 166-167
preference sense 142-143, exploitation of 13
143f gradedness 8-11
see aLso principled polysemy levels 285-287, 341
framework problems and solutions 3-8
overextension 324, 325, 326, 327 terminology 49,58,67 nI, 268
Oxford Advanced Learners Dictionary theories and hypotheses 8-16
(OALD) 404-405 typology 8, 12, 229, 268-272,
Oxford English Dictionary (OED) 270--271 t, 281, 285-287
223,225,276,334,402 see also history of polysemy
possible 92
Pou, August Friedrich 59
Panther, Klause-Uwe 195 prefixes, English 218-219, 218t, 219t
Paradis, Carita 249 nS prepositional networks
par/er 273f adpreps 150 n21, 150 n26
paronymy 289 nil compositional sets 108-109
partonomy 197, 198-203, 198f, 200f, figure-ground relations 97, 98-
209,209 nl 99, 123
Paul, Hermann 56 landmark (LM) 99, 151-152 n27
Pederson, Jan O. 363, 406 trajector (TR) 99, 151-152 n27
perceptual analysis 113 see also over; principled
phrasal verbs 11, 339 polysemy framework
pitCiCi 181-182 principled polysemy framework 104
considering 182 atemporality 115-116, 149 n16
flexibility 183-185 cluster of senses 125
keeping in a place 182 complex conceptualization 113-
keeping in a state 183 114, 114f
420 Index

conceptual significance of syntax. roles 12-13,210 n3, 233-234,

124, 150 n21 248 nnl-2
coordinate system 148 n8 evaluation 245-248, 245t
distinct senses 105-106, 124, 150 Que\ler, Kurt 40
n22 Quine, Willard van o. 366
inferencing strategies ll6-117
on-line meaning construction
114f, 117-122, 149 n18 Radden, Giinter 195, 197,210 n2
perceptual analysis 113-114 radial structure 217
pragmatic strengthening 122- Rastier, Fran~ois 66
123, 150 n20 Raukko, larno ll, 338, 343
primary sense 107-109, 110 reading comprehension 298, 302
protoscene 110-112, 113, 149 Reddy, Michael l. 201
nl3 referential class and semantic relation
reconceptualization 113 277-278, 277t
spatial scenes 114-115 regular polysemy 281
vantage points 114-115 Reisig, Christian Karl 55
see also over representations 55
protoscenes 110-112,113,149 nB, retention hypothesis 299, 308
335-356, 336f, 340 Rice, Sally 97, 99, 104, 107, 150 n20,
prototype theory 274, 278, 288 n7 168, 337
psychology 55, 64-65 Richards, Ivor A. 50
PT (partonomy-tax.onomy) fallacy Rivkai, Yehuda 320
198-203, 198f, 199f, 200f, Robertson, Rachel W. 298
209 Rosch, Eleanor 97,210 n4, 343
Pustejovsky, lames 12-13,38,66,96, Ruhl, Charles 97, 147 n3
201,210 n3, 233-234, 235, rule-based polysemy 281, 285-286
236-237, 243, 245-248, 249
n5, 280, 289 n14, 365,400
Sadock, J. M. 367
safe 90-91
qualia salience 62, 297-300, 306, 307-308
components of meaning 234-235 Sandra, Dominiek 97, 99, 104, 107,
compositionality 243 150 n20, 168, 337
con ventionalization 242-243 Saussure, Ferdinand de 34
doublets 242t Scheibman, Joanne 41
idiosyncrasy 242-243 Schelegel, August Wilhelm von 59
modifier-head constructions 235- Schepping, Marie-Therese 289 n17
242, 237t, 239t, 240t, Schlieben-Lange, Brigitte 66,267,
241 t, 248 nn3~, 249 289 n12
nn7-9 school 278, 279
referential metonymy 244-245, Schiitze, Hinrich 363, 406
244t Schwarze, Christoph 289 n 17
 Index 421

Searle, John R. 31 Smith, Michael K. 64

semantic change Soares da Silva, Augusto 289-290
analagous 269, 271 t, 272, 285 n19
context 60-61 sociology of polysemy 65
emotions and 64 Sperber, Hans 64
referential class and semantic Stiihlin. Wilhelm 65
relation 277-278, 277t Stern, Gustaf 64
specification 222 Stock, Penelope F. 373, 374
synchrony 268-273, 270t, 272, suffixes (agent/subject arguments)
277-278, 277t 221-222, 221t, 222/
two-level-semantics 66, 279-280 summary scanning 148 n6
typical recurrent polysemies 278- Summers, Della 385 n 13
279 Sweetser, Eve E. 57, 66, 92, 254
typology 268, 270-271t, 281, synchronic polysemy networks 150
285-287 n20
Semantic Concordance project, synecdoche 55, 209, 210 n3
Princeton 364 cognitive triangle 205-208
semantic primes 14 definitions 195, 196, 197,201-
semantic priming 370-371,385 n12 203,210 n2
semantics 54, 55, 60, 63, 366-367 domains 207.207/
semiotic model 34-37 entity-category fallacy 203, 208
degrees of deviation 34-36 synonymy 35, 289 nIl
"one form, one meaning" 34 syntax 40,95, 124, 150 n21, 306--307
sense-tagging 393-394, 407
bilingual tagging 402-403
Gale et al.'s claim 400-405 Talmy. Leonard 97, 100, 117
and Kilgarriff's (1993) paper taxonomic polysemy 269, 270t
394-400, 401 taxonomy 197, 198-203, 198f, 199f,
against MRD-derived database 200/,209,209 nI, 284-285
396 Taylor, John R. 210 n4, 289 n12
one sense per collocation 405- temporality 115-116, 149 n16
406 ten Hacken, Pius 367
one-sense-per-discourse 401-402, thesauri 398
406 Times, The 376--377
Yarowsky's claims 405-406,407 Todd, Zazie 326
Seto, Ken-ichi 209 nl TR (trajector) 99, 151-152 n27
-ship 225-226,227,227/ transitivization 279-280
simile 317,320,322-323 see also Traugou, Elizabeth Closs 61, 123,
metaphor 222, 254, 262
"simplest systematics" 66 Trench. Richard Chenevix 253
Sinha, Chris 326 Tuggy,David 229
Slobin. Dan Isaac 340 Turgot, Anne Robert Jacques 51-52
Small. Steven I. 393, 394 Turner, Mark 54. 83. 84, 89
422 Index

Tversky, Barbara 208, 209 nl Wilbur, Ken 225

Wilks. Yorick 395, 396
WiIIiams. John N. 300, 337, 371
UAD (Unidentified Authorising Winner, Ellen 318
Dictionary) 385 n13 Winston, Morton E. 209 n 1
Ullmann, Stephen 3, 65, 208 Witkowski, Stanley R. 7
un- 227. 228 Wittgenstein, Ludwig 9-10
word sense disambiguation (WSD)
361-362
vagueness 162-163, 229. 278. 368, ambiguity tests 367-370, 375,
371. 375 see also ambiguity; 384-385 nn9-1O
contextual variation "bank model" (BM) 366, 394-
Van der Leek, Frederike 36 395,396,397,400
Vandeloise. Claude 97.99, 117 discourse context 60-61
variability 163-164. 169-170 polylexy 365
variation 163, 164 see also contextual selection and modulation 366-
variation 367
verbs semantic priming 370-371,385
acquisition of 333 n12
light verbs 334, 340, 340t sense 375, 381-382, 383, 385
phrasal verbs 11, 339 n16, 395
polysemic verbs 333, 337 and theoretical lexicology 362-
Vico. Giambattista 50, 53 366
Victorri. Bernard 66 use 375,385 n16
see also lexicography
WordNet 364, 399, 400
Walker, John 220 Words worth, William 89
Waltereit. Richard 283 Wunderlich, Dieter 36
Warren, Beatrice 235, 236 Wundt, Wilhelm 64, 65
Websters Encyclopedic Unabridged
Dictionary of the English
Language (WEUD) 275,276 Yarowsky, David 363-364. 393, 394,
Wegener, Philipp 64 403,404,405-406,407
Weinreich. Uriel 65 Vu, Ning 319-320
Weinrich. Harald 65
WEUD see Webster's Encyclopedic
Unabridged Dictionary of the Zgusta, Ladislav 373
English Language Zhang, Qiao 162-163
Wierzbicka, Anna 14. 398 Zwicky, Arnold M. 367