Seeing clearly: Frame Semantic, Psycholinguistic, and Cross ...

Seeing clearly: Frame Semantic, Psycholinguistic, and Cross-linguisticApproaches to the Semantics of the English Verb See

by

Collin Freeman Baker

A. B. (Brown University) 1968M. A. (University of South Carolina) 1984

M. A. (University of California, Berkeley) 1993

A dissertation submitted in partial satisfaction of the

requirements for the degree of

Doctor of Philosophy

in

Linguistics

in the

GRADUATE DIVISION

of the

UNIVERSITY OF CALIFORNIA, BERKELEY

Committee in charge:Professor Charles J. Fillmore, Chair

Professor Andreas KatholProfessor Eleanor RoschProfessor Eve Sweetser

Fall 1999

The dissertation of Collin Freeman Baker is approved:

Chair Date

Date

Date

Date

University of California, Berkeley

Fall 1999

Seeing clearly: Frame Semantic, Psycholinguistic, and Cross-linguistic

Approaches to the Semantics of the English Verb See

Copyright 1999

by


1

Abstract

Seeing clearly: Frame Semantic, Psycholinguistic, and Cross-linguistic Approaches

to the Semantics of the English Verb See

by


Doctor of Philosophy in Linguistics

University of California, Berkeley

Professor Charles J. Fillmore, Chair

This dissertation studies lexically-speci�c (irregular) polysemy, using a case study of the

English verb see as the major example. Clearly, words such as see have di�erent meanings

in di�erent contexts, but how can we distinguish di�erent senses from mere di�erent uses

(modulations) of the same sense? What are the semantic and paradigmatic relations among

the senses? Answers to these questions were sought through a series of psycholinguistic

experiments, formal analysis in terms of Frame Semantics (Fillmore 1976; Fillmore 1982;

Fillmore & Atkins 1992) and other cognitive linguistic theories, and analysis of entries

in monolingual and bilingual dictionaries. The results show that speakers can reliably

distinguish many senses of see, that the English pattern of senses is partially shared across

languages, and that frame semantics is a good way of representing the relations among

senses.

First, the relation of semantics to world knowledge and categorization is discussed,

and Frame Semantics, homonymy, polysemy, and monosemy, traditional tests for polysemy,

and other types of linguistic evidence are de�ned.

Then, the semantics and syntax of see are outlined and detailed frame represen-

tations are given for 19 senses (e.g. recognize (saw that he left), ensure (see (to it) that

he leaves) and experience (saw combat)) and 6 purely compositional uses, e.g. tour and

hallucinate, including inheritance (complete or partial) among senses and from more gen-

eral frames. Representative collocations with see are discussed, along with other cognitive

linguistic representations, including mental spaces (Fauconnier 1985 [1994]).

2

Next three psycholinguistic experiments are described, involving (1) free sorting of

examples of uses of see, and (2) untimed, and (3) timed classifying of examples into a priori

categories. Results suggest that speakers can reliably access �nely di�erentiated senses like

those proposed above.

The sense divisions for see in entries from several English dictionaries are shown

to be problematic. Entries from bilingual dictionaries between English and Spanish, Chi-

nese, and Japanese demonstrate that there is partial overlap between the senses and sense

inheritance of English and those of other languages. The similarity of English to Spanish

is greater than to the non-Indoeuropean languages.

Finally, conclusions are drawn from all the approaches and future research is out-

lined. Includes index and experimental stimuli.

Professor Charles J. FillmoreDissertation Committee Chair

iii

Contents

Preface vii

Outline of the Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . viii

1 Introduction 11.1 The Research Question . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 The Domain of Linguistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

Semantics and World Knowledge . . . . . . . . . . . . . . . . . . . . . . . . 4Frame Semantics, Construction Grammar, and Other Formalisms . . . . . . 5Concepts, Categories, and De�nitions . . . . . . . . . . . . . . . . . . . . . 7Folk Models of Language and the Lexicon . . . . . . . . . . . . . . . . . . . 10

1.3 Structures of the Lexicon . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12De�nitions of Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12Tests for Lexical Ambiguity . . . . . . . . . . . . . . . . . . . . . . . . . . . 15Distinguishing Monosemy, Polysemy and Homonymy . . . . . . . . . . . . . 18Regular Polysemy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Irregular Polysemy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

1.4 An Example of Possible Polysemy: Crawl . . . . . . . . . . . . . . . . . . . 241.5 Other Types of Evidence for Lexical Semantics . . . . . . . . . . . . . . . . 27

Types of Linguistic Data and Collection Methods . . . . . . . . . . . . . . . 28Some Major Types of Evidence and Their Uses in Lexical Semantics . . . . 33

1.6 The Polysemy of See . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2 A Frame Semantic Analysis 39

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392.2 The Seeing Frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402.3 A Sense-Enumerative Approach . . . . . . . . . . . . . . . . . . . . . . . . . 44

Brief List of Senses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45Traditional Tests for Sense Di�erences . . . . . . . . . . . . . . . . . . . . . 48Co-occurrence of Complement Patterns with Senses . . . . . . . . . . . . . 55

2.4 Semantics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58The Event Structure of Seeing . . . . . . . . . . . . . . . . . . . . . . . . . . 59

2.5 Detailed Frames and Descriptions for Senses . . . . . . . . . . . . . . . . . . 60Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

iv

Basic Senses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63Semi-collocations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92Compositional Uses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

2.6 Collocations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104Examples of Collocations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

2.7 Some Recalcitrant Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1092.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

3 Other Cognitive Approaches 1113.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1113.2 A Mental Spaces Approach to the Semantics of the seen . . . . . . . . . . . 1123.3 The Semantics of Motion Expressions with See . . . . . . . . . . . . . . . . 1193.4 A Brief Discussion of Two Complex Idioms . . . . . . . . . . . . . . . . . . 121

Envisionment and Classi�cation as Reality . . . . . . . . . . . . . . . . . . . 121\Would rather see him hanged" and Related Patterns . . . . . . . . . . . . 123

3.5 Senses as a Function of the Semantic Types of Arguments . . . . . . . . . . 1243.6 Uni�cation and Re exives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1263.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

4 Psycholinguistic Experiments 1324.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

On-line vs. O�-line Methods . . . . . . . . . . . . . . . . . . . . . . . . . . 133Previous Work on Priming . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

4.2 Experiment 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140Statistical Measures of Agreement . . . . . . . . . . . . . . . . . . . . . . . 141Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

4.3 Experiment 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

4.4 Experiment 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

5 What the Dictionaries Say 1645.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1645.2 Monolingual English Dictionaries . . . . . . . . . . . . . . . . . . . . . . . . 166

Webster's Third New International|Text . . . . . . . . . . . . . . . . . . . 167Webster's Third New International|Discussion . . . . . . . . . . . . . . . . 172Small English Dictionaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

5.3 Bilingual Dictionaries|Introduction . . . . . . . . . . . . . . . . . . . . . . 179

v

5.4 English-Spanish Dictionaries . . . . . . . . . . . . . . . . . . . . . . . . . . . 180Collins English-Spanish Dictionary|Text . . . . . . . . . . . . . . . . . . . 180Collins English-Spanish Dictionary|Discussion . . . . . . . . . . . . . . . . 184A Pocket Spanish-English Dictionary . . . . . . . . . . . . . . . . . . . . . . 186

5.5 English-Mandarin and Mandarin-English Dictionaries . . . . . . . . . . . . 187Far East English-Chinese Dictionary|Text . . . . . . . . . . . . . . . . . . 187Far East English-Chinese Dictionary|Discussion . . . . . . . . . . . . . . . 190Everyday Chinese-English Dictionary|Text . . . . . . . . . . . . . . . . . . 192Everyday Chinese-English Dictionary|Discussion . . . . . . . . . . . . . . 196

5.6 English-Japanese Dictionaries . . . . . . . . . . . . . . . . . . . . . . . . . . 198Kenkyusha English-Japanese Dictionary|Text . . . . . . . . . . . . . . . . 198Kenkyusha English-Japanese Dictionary|Discussion . . . . . . . . . . . . . 203Small English-Japanese Dictionaries . . . . . . . . . . . . . . . . . . . . . . 207

5.7 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208English-English Comparisons of Entry Structure . . . . . . . . . . . . . . . 208Cross-linguistic Comparisons of Entry Structure . . . . . . . . . . . . . . . . 211Cross-linguistic Comparisons of Translations . . . . . . . . . . . . . . . . . . 220

5.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227Possible Alternative Organizations for Dictionary Entries . . . . . . . . . . 227

6 Future Research Directions and Conclusions 228

6.1 Future Research Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . 228Psycholinguistic Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 228Corpus Studies and Word Sense Disambiguation . . . . . . . . . . . . . . . 230Cross-linguistic Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232Other Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233

6.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

Bibliography 238

A Additional Corpus Examples of Senses 249A.1 Basic Senses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249A.2 Semi-collocations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251A.3 Compositional Uses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252

B Summary of Morphology and Syntax of Senses 254

C Experiment 1 256

C.1 Original nineteen senses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256C.2 Stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259

Brown Corpus Sentences (selected) . . . . . . . . . . . . . . . . . . . . . . . 259Constructed Sentences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261

D Stimuli for Experiments 2 and 3 263

Index 275

vi

Acknowledgements

Careful citations are the standard academic way to show the intellectual background of a

work such as this, but they can never tell the whole story. The ideas in this dissertation

have been developed with the bene�t of many careful readings and thoughtful comments

from all the members of my committee, and of extensive discussions with George Lako�,

Jane Edwards, Christopher R. Johnson, Susanne Gahl, Kevin Moore, and other faculty,

postdocs, and graduate students at UC Berkeley. I began by trying to acknowledge each

contribution by citing it as a personal communication, but soon found that there were too

many to count. I greatly appreciate all the suggestions that I have received and hope that

those who see their ideas used here will not be disappointed by the lack of speci�c reference.

I feel fortunate to have participated in the life of the Linguistics Department at

UC Berkeley, and to have received the intellectual stimulation of faculty such as Paul Kay,

John Ohala, Jim Matiso�, Larry Hyman, Leanne Hinton, Ian Maddieson, and Robin Lako�

and the encouragement of fellow linguistics students such as Sarah Taub, Joe Grady, Sara

Gesuato, Matt Juge, Kyoko Hirose Ohara, Orhan Orgun, Laurel Sutton, Marvin Kramer,

Jason Patent, Mary Bucholtz, Kathleen Hubbard, Anita Liang, and many others, including

a distinguished series of colloquium speakers and visiting scholars. It has also been my

good fortune during much of my stay at Berkeley to be associated with the International

Computer Science Institute (ICSI) and to have had the chance to discuss natural language

and cognitive science with researchers there such as Jerry Feldman, Dan Jurafsky, Terry

Regier, David Bailey, and Srini Narayanan.

I am grateful to the taxpayers of California for the Regents' Scholarship I received

during my �rst years at Berkeley and to U.S. taxpayers for the NSF funding of the work

of the FrameNet Project at ICSI (IRI 9618838, Charles Fillmore, PI), which has made it

possible for me to be paid for work related to my dissertation research.

Looking back at how I reached this point, I feel grateful to my parents for their

concern for my education (to the point of helping to found a new school) and to a long

succession of brilliant, dedicated teachers, too numerous to name. Above all, I want to

thank my wife and my son (now three years old) for their support and patience throughout

this long project.

vii

Preface

From an early age, children enjoy jokes and riddles that play on language: What

has eyes but cannot see? A potato. An air conditioner can run all day and still not go

anywhere. As in many other �elds, many of the simplest questions about language can lead

to profound insights. Why does the word tongue sometimes mean `language'? Why does

heart mean both ìnternal organ' and `courage or compassion'? Why do we say \I see what

you mean" when there's nothing to look at?

This thesis is about some of those \simple" questions. How can we tell how many

senses a word has? If a word has more than one sense, how can they be related to each

other? We will examine the general problem of polysemy (Greek for `many meanings') as

exempli�ed in a thorough case study of the semantics of the English verb see. A variety of

approaches will be used to study the question of what senses see can have and how they

are related to each other.

I will not deal with the historical evolution of the senses of see except insofar as they

are relevant to understanding the relations among the senses in the minds of contemporary

adult speakers of English. In general, if current speakers believe that one sense of a word

is older than another, then they may also tend to regard it as more central or basic. For

example, people who are aware that the English word republic derives from Latin r�es publica,

meaning `public a�airs', may have a di�erent concept of what it means to say that the United

States is a republic than people who are not aware of this fact. There is reason to believe

that the types of meaning change found through history are likely to be similar to those

involved in contemporary processes of meaning extension, such as metaphor and metonymy,

but this thesis does not attempt a full survey of these types of meaning shift.

Nor will I examine the childhood acquisition of the senses in detail, although I

will make occasional reference to the extensive research being done on questions of lexical

viii

acquisition. Aside from needing to limit the scope of the study, there is also no guarantee

that the patterns of childhood acquisition and the structure of adult lexical semantics are

very closely related; the nature of that relation would be a fascinating study in itself.

Outline of the Structure of the Dissertation

Chapter 1. Introduction

The �rst section outlines the research questions addressed in the thesis. These

questions, of course, are all applicable to any apparently polysemous word. I consider the

possibility that we might be able to treat all the uses of see as instances of a single, highly

abstract sense, along the lines of Ruhl (1989). I show that such a solution not only fails to

distinguish adequately among related words, but also contradicts psycholinguistic studies

of priming e�ects.

I discuss my pre-theoretical biases and assumptions about linguistics in general,

the types of data that it should seek to explain, what constitutes an adequate explanation

of a set of facts, etc. Brie y, I believe that linguistic theories should strive to cover as

broad as range of facts as possible; this calls for a very rich, expressive representation of

linguistic structures. The chapter also discusses a variety of sources of evidence that can

usefully bear on questions of lexical semantics, including traditional tests for ambiguity,

corpus data, psycholinguistic data, and cross-linguistic data.

Chapter 2. A Frame Semantic Analysis

In this chapter, I �rst discuss the syntax and semantics of see and try to apply

some of the traditional tests for sense di�erences. Then, I lay out a �ne-grained breakdown

of senses and uses according to a Frame Semantic approach, dividing them into basic senses,

compositional uses, collocations and semi-collocations.

Chapter 3. Other Cognitive Approaches

This chapter provides an account of some of the senses of see based on Fauconnier's

(1985 [1994]) mental spaces, which provide a satisfactory way to talk about the similarities

and di�erences in the conceptualization of what is seen, and thus to di�erentiate certain

senses; however, not all senses can be handled in this way. Also included are a discussion

ix

of \metaphorical motion" uses of see (e.g. I can see all the way to the bay), a typed-feature

structure treatment of the seer and seen, and a discussion of how to handle some problems

with re exives (e.g. He sees himself as a hero).

Chapter 4. Psycholinguistic Experiments

In this chapter I report the results of a series of experiments intended to shed

light on the psychological reality of separate senses. A variety of tasks were used: free

categorization by sorting cards containing example sentences with see into sense categories,

forced categorization using a set of categories devised by the experimenters, and two timed

tasks measuring reaction times for lexical decision and categorial judgement for the same

example sentences.

Chapter 5. What the Dictionaries Say

This chapter compares the sense divisions given for see in a large monolingual

dictionary and several small ones, and discusses what lexicographers have to say about how

they decide about sense divisions. I also examine bilingual dictionaries between English and

Spanish, Chinese, and Japanese, to see how well the sense divisions and the words used in

translating see match the English sense divisions based on other evidence.

Chapter 6. Conclusions and Future Research Directions

This chapter summarizes the results and suggests directions for future study. While

we should not assume that all the pieces will �t neatly together, if there appears to be little

or no coherence among the various types of evidence, that would bode ill for the prospect

of the establishment of an empirically-based theory of polysemy. I also discuss a number of

questions that remain to be answered and outline a program of research designed to answer

some of them.

1

Chapter 1

Introduction

Xu�e �er b�u s��, z�e w�ang; s�� er b�u xu�e, z�e da��.

Study without thought is wasted; thought without study is dangerous.

{Confucius, Analects. I.2.15

1.1 The Research Question

The English verb see, referring to one of the most basic of human faculties, is

correspondingly frequent in speech and writing. It is acquired relatively early by children, no

doubt through the association of the experience of seeing with the experience of hearing the

word in sentences like \Do you see the doggie?" One might suppose, then, that the syntax

and semantics of see would be very simple, but the opposite is true. Like most common

verbs, it occurs as the head of clauses displaying a broad range of syntactic patterns, with a

variety of senses. (The tendency of common words to have many senses is well documented;

one of the earliest clear discussions was by Zipf (1949[1965]:27-31)). This dissertation will

explore the relation between the syntax and semantics of the verb see and some of the

constructions in which it participates.

If asked to �nd out how many senses a word has, the average speaker of English

would most likely turn to \the" dictionary. The Random House Unabridged (1987) lists no

fewer than thirty-four senses of see, and even medium-sized dictionaries list more than two

dozen.1 Having perused the long entry, he might feel a sense of satisfaction that his language

1There are, of course, two homophonous words spelled see(s), the noun derived from Latin s�ed�es, relatingto the \seat" of a bishop, and the common verb, which is solidly Germanic. Linguists agree that it is ahistorical accident that the forms of these two words have come to partially overlap and, given the fact that

CHAPTER 1. INTRODUCTION 2

is so rich in senses for this common word. The linguist who reads the same dictionary entry

could not fail to ask herself whether there are really so many distinct senses, and, if so, how

native speakers can possibly learn to distinguish them. If the linguist looks at more than

one dictionary, she will also notice that even dictionaries of about the same size di�er about

how many senses the word has and how they are de�ned, tending to reinforce her doubts

about the reliability of the sense divisions shown.

One of the major topics of this chapter will be the types of evidence that should be

used in deciding questions in the �eld of lexical semantics. The great majority of argumen-

tation in linguistics in this area (as in most others) is based on introspective evidence, that

is, certain judgements as to grammaticality or acceptability, usually made by the author

and by the authors of references consulted. In the next part of this chapter, I will set out

some basic ideas about the proper domain of linguistics, in order to delimit the sorts of ques-

tions that linguistics tries to answer. Then, I will address some theoretical issues related to

lexical semantics in general and polysemy in particular. Next, I will discuss some di�cult

lexico-semantic problems, using traditional, largely introspective, linguistic evidence. The

following section will propose a program for gathering new types of evidence that may shed

light on the issues raised so far, and the next section will attempt to draw some conclusions.

1.2 The Domain of Linguistics

In undertaking any �eld of study, one of the fundamental questions is how to

delimit the �eld itself. One must decide what set of facts one is seeking to explain and what

one can safely ignore or at least treat in a summary fashion. For the purpose of designing

bridges for human beings to use, quantum e�ects can be ignored; in subatomic physics, the

weather outside the laboratory is irrelevant. In the humanities, such decisions tend to be

explicit, re ecting the scholar's personal and political choices; a committed Marxist is more

likely to be interested in labor relations than in Beowulf. In the sciences, it has traditionally

been assumed that the facts are independent of the experimenter (an assumption that is

increasingly being questioned); the researcher must at least be able to show that there are

no major factors that are not appropriately represented in the data on which the theories

are based.

they are di�erent parts of speech with quite di�erent semantic domains, it is unlikely that a native speakerwill ever confuse the two. The noun see will not be discussed further.


In linguistics, both of these viewpoints are found, and linguists are continually

rede�ning the boundaries of the �eld, often seeking to include phenomena which are of

interest to them personally on the grounds that it would be unscienti�c not to do so.

For example, the �eld has gone through at least one full swing of the pendulum from (1)

depending mainly on introspective evidence as to what expressions mean and what speakers

intend by them, to (2) depending almost entirely on corpus data, more or less ignoring

meaning and mental processes (at least in Bloom�eldian Structuralism), and back to (3) a

mentalist position relying largely on introspective data (at least in Generative Grammar).

If linguists are to understand each other and build on each other's work, it is important to

make explicit our decisions about what facts we seek to explain and what research methods

and types of argument we �nd convincing. Therefore, I will begin by outlining some of my

own predispositions (what Lako� (1990) called \commitments") regarding the proper study

of linguistics.

I believe that it is important for linguistics to pay attention to as broad a range

of facts about language use as possible. This takes precedence for me over the aesthetic

delight of producing the minimal and hence most elegant theory. This is not to say that

there is no room in linguistics for elegant, parsimonious theories, only that I consider the

coverage of a wide range of data to be more important. For example, theoretical linguistics

has made great progress by concentrating on linguistic competence and ignoring perfor-

mance errors by virtue of the competence/performance distinction introduced by Chomsky

(1965:4); nevertheless, I believe that the study of performance errors can contribute to lin-

guists' understanding of how speakers produce utterances. Finding adequate descriptions

for the full range of phenomena and making generalizations at the proper levels of abstrac-

tion should then lead us to satisfactory explanations of the phenomena, which motivate

the patterns we �nd. In some cases, we may even be able to predict what patterns will

occur in an area where we have not gathered data, but this is usually not possible, given

the complex interaction of di�erent types of organizing principles on di�erent levels.

I also believe that linguists must strive to make their theories consistent with

results in other relevant disciplines. In this dissertation, in addition to linguistics proper, I

will also be concerned with the �elds of psycholinguistics, lexicography, and natural language

processing. It should be clear that even these closely related �elds have somewhat di�erent

assumptions, methods, and goals, which will be discussed at appropriate points below.

While it is important not to equate linguistics with cognitive science or psychology,


I believe that linguists should be concerned with the psychological processes of language

production and understanding and their underlying neurological mechanisms, and that

linguistic theories should be consistent with what is known in those �elds. The �eld of psy-

cholinguistics undoubtedly has the most to contribute to linguistics proper; if nothing else,

it often demonstrates the di�culty of experimentally studying constructs which linguists

may assume as given, such as derivational complexity (Are passives really more complex

than actives?), or which linguists rarely consider in detail, such as lexical access (Exactly

what is retrieved from memory at what point in sentence processing?).

Finally, linguists should be aware of developments in computer science, particularly

in areas such as knowledge representation, connectionist networks, information retrieval,

and natural language processing. Although relatively few computer scientists have based

their work directly upon contemporary linguistics, many have adapted some elements of

linguistic theory for particular purposes, and it should be instructive for linguists to see

what works and what does not in such implemented systems.

Semantics and World Knowledge

One of the most di�cult questions in the �eld of semantics is where (if anywhere)

to draw the line between meanings conveyed by language and knowledge of the world in

general. At one extreme, some cognitive linguists deny the possibility of delimiting lin-

guistic semantics in any clearcut way: \The distinction between semantics and pragmatics

(or between linguistic and extra-linguistic knowledge) is largely artifactual, and the only

viable conception of linguistic semantics is one that avoids such false dichotomies and is

consequently encyclopedic in nature." (Langacker 1987:154) Langacker gives the example

of the noun compound buggy whip (p. 157), which cannot be explained compositionally

unless the semantics of buggy is taken to include the notions of a driver and a horse. At

the other extreme, we �nd statements such as \Crucially, all aspects of meaning that are

situational must be removed from the study of Grammar, since they overload the Grammar

with redundant material and prevent the formulation of important empirical generaliza-

tions."(Bouchard 1995:16).2

2Situational is a tricky word to de�ne. One cannot understand the concept of buggy whip withoutunderstanding what we will call the frame in which it is used, involving at least a horse, a buggy, and adriver. On the other hand, the meaning of buggy whip does not depend on whether or not one is riding ona buggy at the time of speaking the words.


Although Langacker and Bouchard agree to a surprising extent on the isomor-

phism of semantic structures and syntactic structures, Langacker postulates a rich \seman-

tic space" as the basis of all linguistic symbols, while Bouchard wants to �nd a minimal

semantics that will map onto the required syntax, and to relegate all other meaning conveyed

by language to an ill-de�ned pragmatic interpretation mechanism.

Bierwisch & Schreuder (1989) also take what is called a \two-level" approach, sep-

arating linguistic semantics from world knowledge and de�ning the pragmatic, interpretive

processes in somewhat more detail. But in the long run, such an approach must either deal

with the hard problems of compositionality, conversational implicature, etc. or not. If these

problems are dealt with, the mechanisms for doing so must ultimately be as complex as

those of cognitive linguistics in any case; if they are not fully dealt with, then we will have

just another example of sweeping the hard problems under a rug and waiting for someone

outside \linguistics proper" to take care of them.

If linguists really are concerned with the psychological reality of the constructs of

their theories, and are willing to accept psycholinguistic evidence as constraining possible

linguistic theories, then it seems that the burden of proof is on those who want to postulate a

separate set of mental mechanisms for lexical semantics, distinct frommore general cognitive

abilities. On the other hand, it is obvious that most linguistic analysis does not require

the full range of world knowledge at every step. For example, deciding that a particular

situation is going to be described in a sentence by a particular noun in a particular argument

relation with a particular verb requires a considerable degree of what would be called \world

knowledge" in some schools of linguistics. But having decided this, choosing an in ectional

form of the noun, in highly in ected languages such as Latin, requires only the information

as to what declension the noun belongs to, the case and number to be expressed, and some

very schematic representations of the in ectional a�xes for each declension. (This account

is adequate for \regular" nouns; additional complexity is involved for \irregular" nouns,

with various levels of subregularity.)

Frame Semantics, Construction Grammar, and Other Formalisms

In discussing semantics, I will assume, as is customary in cognitive linguistics,

that language does not directly re ect facts about an \objective" external world existing

independent of human observers, but rather re ects the conceptual structures which people


build as a result of shared experience of the (more or less common) external world and

of their culture. De�ning semantics as the relation between linguistic forms and concepts,

rather than between linguistic forms and an objective reality allows us to explain semantic

relations in terms of cognitive operations such as metaphorical and metonymic mappings.

The general approach to semantics adopted here is based on Frame Semantics

(Fillmore 1976; Fillmore 1982; Fillmore & Atkins 1992), which analyzes situations in terms

of frames containing participants and their roles in events, and then describing lexical items

in relation to such semantic frames. Frame Semantics attempts to make the linguistically

appropriate generalizations about words related to each other by virtue of being used to

describe the same type of predicate, without trying to encode the full range of a speaker's

knowledge of events in the world. As Fillmore (p.c.) expresses it, it must be part of our

knowledge of the semantics of the word carpenter that carpenters work with wood, but the

frame should probably not include information about whether or not they are unionized.

Although I may refer to other theories of grammar at various points, in this dis-

sertation I will represent syntax using the formalism of Construction Grammar (Fillmore

et al. 1988; Fillmore & Kay 1994; Goldberg 1995; Kay & Fillmore 1999), in a compromise

among the slightly di�erent notations which appear in Fillmore & Kay 1994, Fillmore &

Kay 1996 and Kay & Fillmore 1999. Construction grammar also posits that a single type

of unit, the construction, can be used to represent all the knowledge of a language which

a speaker possesses, both syntactic and semantic. Thus, the principal senses of see and

highly idiomatic phrases like see the light at the end of the tunnel are all represented as

constructions. These lexical entries are combined through uni�cation with other construc-

tions, such as the Subject-Verb Construction, to create all of the sentence patterns found

in the language.

In practice, because I will be dealing primarily with semantics, I will use mainly

attribute-value matrices to represent semantic frames and frame elements, in the style of

Frame Semantics, with the valence structure shown within them in the style of Construction

Grammar (see Section 2.5). Also, for clarity of exposition, I will deal �rst with those senses

of see that do not involve the presence of any other speci�c lexeme, and then brie y discuss

some of the many collocations of see with other words (Section 2.6).

Certain aspects of the semantics are also discussed in Section 3.2 in terms of Fau-

connier's (1985 [1994]) mental spaces. At several points I will also refer to the notion of

the interaction of the semantics of the verb with that of its NP complements, based in


part on the de�nition of qualia in Pustejovsky 1995 (Ch. 6) (without, however, subscrib-

ing to Pustejovsky's contention that all the information relevant for such interaction can

appropriately be �t into the same four categories for all lexical units).

Concepts, Categories, and De�nitions

For purposes of this dissertation, I will follow Rosch (1978) in de�ning categories as

sets of objects sharing certain properties (equivalent to \extensional meaning") and concepts

as the supposed mental constructs underlying the division of things into categories (roughly

equivalent to \intensional meaning"). As will soon be seen, this is an oversimpli�cation,

but it should be adequate for the moment. Psychologists today, while more willing than

behaviorists were to speak of the reality of mental objects and processes, are cautious about

extrapolating from experimental results to underlying causes, hence the phrase \supposed

mental construct".

The Western philosophical tradition, stretching back to Aristotle, has generally

assumed that a concept can best be de�ned in terms of necessary and su�cient conditions

for membership in the category. Although anyone who has tried it knows that the art of

writing a de�nition is far from easy, this tradition forms the basis not only of much of

lexicography (i.e. de�nitions in terms of genus and di�erentia) but also folk theories of

natural and cultural kinds, and hence, the extensions of words; e.g. \It can't be a skunk

because a skunk has a white stripe down its back."

Basic categories and Co-occurrence

It is a fundamental fact about all language that the basic pairings between form

and meaning in the lexicon are arbitrary; this had been known for centuries before Saussure

made it explicit and formal. The great contribution of cognitive linguistics has been to

remind us that we should not overstate the arbitrariness of language; that processes such

as semantic composition, metaphor, metonymy, and historical change are not only parts of

the everyday use of language, but also part of cognition in general, and therefore function

in accord with general principles that also apply to non-linguistic cognition, such as those

that govern learning and recall. With regard to lexical semantics, this implies that the

combination of meanings conveyed by a polysemous word will not be an arbitrary set;


rather, we expect to �nd connections between them based on general cognitive principles.

For example, we expect that people will often classify things and events on the basis of

perceived similarity, and that to a large extent, people will agree on those similarities and

those classi�cations. As Lako� (1987:Ch. 6) has shown, the basis of the classi�cation can

be quite complex, but it is not completely arbitrary. Wide-ranging psychological results

have also shown that certain types of categorization are more basic than others, and that

the tendency of certain events to co-occur is one of the bases of such categorization.

Category Structure and Gradation

Questions about the adequacy of Aristotelian de�nitions have also raised questions

about the assumptions on which they are based, such as the \Law of the Excluded Middle",

which says that for any set X, any individual must be either in or out of the set. If X

represents the set of things having the attribute A, then we can say that any individual

must be either A or not A; philosophers have spilled a lot of ink over what to do about

cases in which an attribute seems to be neither true nor false of an individual, but simply

inapplicable, e.g. is the number 3 blue or not?

A new spurt of interest in these questions began in the 1970s with the research

of Eleanor Rosch and her associates. They found that for many types of categories, it was

possible to measure not just whether or not an instance was a member of the category, but

whether or not the instance was a \good example" of the category; these judgements of

whether an instance is a better or worse example of a category have been called \represen-

tativeness", \exemplariness", \graded category structure", and \prototype e�ects"; I will

use the last term. Prototype e�ects have been found to be very robust and to be manifest

in a variety of measures (Rosch 1978; Rosch in press).

There was an initial burst of excitement after the �rst discovery of prototype e�ects

in which it was assumed that they were a direct re ection of the mental representation

of the concept. It was gradually discovered that this was incorrect, but this fallacious

interpretation of the data has plagued the �eld ever since. Lako� (1987:136 �.) claims that

the error arises from two incorrect interpretations of e�ects: (1) E�ects = Structure and

(2) Prototype = Representation. The former assumes that if there is variation in ratings of

\goodness of example" over a group of items that this means that the category in question


is not classical, but fuzzy3, and the items rated as better examples are members to a greater

degree.

Lako� (1987:77-90) gives an extended series of examples of how prototype e�ects

can arise with regard to even classical categories and exemplars that are indisputably mem-

bers of them. For example, one type of prototype is the \Ideal"; the Ideal husband is \a

good provider, faithful, strong, respected, attractive" (p. 87). Of a group of men all of

whom are legally married and live with their wives, some may be judged better examples

of husbands than others depending on their similarity to the Ideal husband, but all of them

are 100% members of the category \husband".

The other error (Prototype = Representation) involves the assumption that the

mental representation of the category is simply the best example, or prototype, i.e. that a

single exemplar is what de�nes the category. Degree of category membership would then

be determined by degree of similarity to the prototype. The Ideal husband example also

disproves this error; it would be perfectly possible to have an Aristotelian de�nition of the

category \husband" in terms of necessary and su�cient conditions (as for example is needed

in law to refer to the rights and responsibilities of a husband vis-�a-vis his wife), and still

to record prototype e�ects when people are asked to rate individuals as better or worse

examples (for more discussion see Medin 1989).

Thus prototype e�ects do not in themselves prove anything about the structure

of categories, as Rosch and other psychologists have been at pains to point out. Of course,

there are categories with fuzzy boundaries and graded membership, such as the set of tall

people, and they will inevitably show prototype e�ects. But the measurement of prototype

e�ects is not the royal road to understanding human categorization. Barsalou (1983) found

that even categories de�ned in such a way that the subject had never conceived of them as

a category before (e.g. things that might fall on your head, things to take on a picnic) also

display prototype e�ects, even though they are not pre-existing and do not have names.

One of the strongest experimental proofs of the E�ects = Structure error was the study by

Armstrong et al. (1983) which showed that the same subjects would claim to believe the

opinion that a particular category was crisply de�ned (the categories used in the experiment

were male, female, odd, and even), and still give better/worse example judgements on a

3Fuzzy is used here as a technical term, �rst used by Zadeh (1965), which describes a category whosemembership function can have values intermediate between true and false, i.e. that individuals can bemembers of a category to some degree. Classical categories, which do not allow partial membership, canalso be described as crisp, the opposite of fuzzy.


scale, which would correlate well with those of other such subjects! Armstrong et al. consider

that their results call into question the whole concept of de�ning categories by features, and

admit that \the sum of the features is not the whole concept". In fact, they conclude \. . .we

ourselves are not optimistic that a general theory of categorization, one that will answer to

the serious problems . . . is just around the corner. To the contrary, the continuing failure of

the search for such units leads us to doubt whether there is a general psychological domain

encompassing àll concepts' . . . ." (See Lako� 1987 for a criticism of their views and a

reinterpretation of their results).

The variety and reproducibility of the fundamental experiments in this area prove

that prototype e�ects are, in some sense, a real fact about human categorization. However,

prototype e�ects vary from subject to subject, from context to context, and even from time

to time for the same subject. As Barsalou (1987) concludes:

The graded structures within categories do not remain stable across situations.Instead a category's graded structure can shift substantially with changes incontext. This suggests that graded category structures do not re ect invariantproperties of categories but instead are highly dependent on constraints inherentin speci�c situations. (p. 107, quoted in Sch�utze 1996)

A number of linguists have argued that linguistic categories (notably the basic

syntactic categories, or parts of speech, such as noun, verb, and adjective) display prototype

e�ects; Lako� (1987:289-92) and Taylor (1995:Ch. 10) argue that both syntactic categories

and syntactic constructions display such e�ects.4

Folk Models of Language and the Lexicon

In Chapter 4 we will be discussing the results of psycholinguistic experiments in

which subjects are asked to make decisions about the senses involved in various examples

of uses of see, and in Chapter 5, we will be discussing how dictionaries divide examples

into senses. It is therefore useful to consider what theories speakers typically have about

lexical semantics and how this might in uence their thinking about the semantics of see

and (indirectly) the way lexicographers write entries for non-specialist dictionaries.

The usual folk theory of language use is based on the communication is object

exchange metaphor (Sweetser 1987, also called the Conduit Metaphor (Lako� & Johnson

4Langacker (1987:189-208) argues that although these syntactic categories display prototype e�ects theyare still crisp, because all members of each category \instantiate a single, abstract schema subject to rea-sonably explicit characterization."(p.189).


1980:10-12), based on Reddy 1979), which runs something like this: ideas are objects, and

language is a means of transmitting them from speaker to hearer. The forms of language

are containers (\packaging") for ideas; the speaker takes some of her ideas, wraps them up

in words and sends them to the hearer, who opens the package and \takes out" the ideas,

thus coming into possession of them. This implies that the container (language) and the

contents (ideas) are functionally distinct, and that (barring a failure of transmission), the

hearer, by a relatively simple process, \takes out" exactly the same ideas that the speaker

\put in".

There also exists a di�erent folk theory of lexical semantics (which may even con-

ict sometimes with the communication is object exchange metaphor), which runs

like this: words are linguistic objects connected with conceptual objects (ideas, meanings);

there are proper words to use for each meaning, possibly more than one word per meaning

(synonymy); producing language consists of putting together a string of words that corre-

sponds to to one's ideas, and understanding language consists of recovering the meanings

attached to the words one hears or reads. Like the communication is object exchange

metaphor, this metaphor implies that the hearer comes into possession of the identical ideas

that the speaker has, but the conventional connection between words and meanings is em-

phasized; there is some limited number of words to express each idea. The converse relation,

that a word might express more than one idea (i.e. homonymy and polysemy), is usually

not considered very much by most people, if only because the process of disambiguation

while understanding one's native language(s) is, under normal circumstances, unconscious.

On those occasions when the question arises of what ideas a word might properly

express, a dictionary may be consulted, which obliges with a list of appropriate de�nitions for

each word, sometimes with examples of use; this list is usually assumed to be unquestionably

accurate and complete. Two of the most likely reasons for consulting a dictionary de�nition

are to settle an argument about whether a particular word can be appropriately used with a

given sense, and to �nd the meaning of an unfamiliar word encountered (usually in reading).

For the �rst purpose, if one of the de�nitions listed corresponds with the idea in question,

then the word is appropriate; otherwise, it is not. For the second purpose, if one of the

de�nitions in the dictionary �ts in the context of the unknown word, then the user concludes

that that sense must be the one intended. In either case, each de�nition tends to be regarded

as a discrete, more or less independent entity, without much attention to the lexicographer's

attempts to express the relations between them using devices such as a hierarchical system


of numbers and letters for the senses, or words like hence, especially , usually , etc. or to

indicate limitations of usage, such as archaic, o�ensive, and chie y Brit. As the experiments

of Fellbaum et al. (1998) suggest, naive users are likely to start at the �rst de�nition and

read down only until they �nd one which seems to �t, ignoring the rest of the entry.

The two ways of using the dictionary mentioned above, starting from a mean-

ing and seeking an appropriate form and starting from a form and seeking an appropri-

ate meaning are more formally known as onomasiological and semasiological, or more

transparently, as encoding and decoding (Makkai 1966). A thesaurus is probably a better

resource for encoding, but most people have even less experience with and understanding

of thesauri than dictionaries.

1.3 Structures of the Lexicon

In this section, I try to de�ne a set of terms to be used in this dissertation. In

general, I will follow Cruse (1986:Ch. 3), but with certain modi�cations in the directions

of Frame Semantics and cognitive grammar. One of the most confusing terms in lexical

semantics is the word word itself; it will be avoided as much as possible.

De�nitions of Terms

A word form is a particular written or spoken form, such as running . A lexical

form is a set of in ectionally related word forms, e.g. frun, runs , ran, runningg; it can

also include idioms consisting of several word forms occurring in a construction, including

in ectional variants, e.g. the tail that wags/is wagging/wagged the dog . A lexical unit

(LU) is an association among (1) a lexical form, (2) a single meaning, and (3) a set of

grammatical properties, including part of speech.

Cruse (1986:77) de�nes the lexical unit as \the union of a lexical form and a

single sense", i.e. equivalent to Saussure's sign, but it is clear elsewhere in the book he

considers that some syntactic information is also attached to LUs. Most of the rest of this

chapter will be devoted to discussing what \a single meaning" might mean. A lexeme is a

set of lexical units. Figure 1.1 on page 13 shows graphically the relations among the units

described above.

Lexical ambiguity can basically be de�ned as the association of two distinct

senses with one word; intuitively, we feel that arriving at the meaning of an ambiguous


Lexeme 3MetaphoricalExtension

L. U. L.U. Lexical Unit

Lexical form Semantics Syntax

UsesWord Forms

context_token_context

context_token_context context_token_context

Syn. Patterns

Lexeme 1

Sem.Reg.Sem. Reg.L. U.

Lexeme

L. U. L.U. L.U.

2

Semantic Spectrum

Semantic Regularities

Groupings

Generalizations (Schematicization)

Metaphorical Leaps

Figure 1.1: Lexical Units and Lexemes

word in a given context context is not merely the result of the constraining of the meaning

by the context, but the activation of a sense already associated with the word. More

precisely, an entire lexical form may may be associated with more than one lexical unit (of

the same part of speech), as in bat/bats ànimal' vs. `baseball bat', or a single ambiguous

word form may be associated with more than one lexical unit as in saw `saber saw' vs.

`viewed'. (Both of these are examples of homonymy). Tests for ambiguity will be discussed

below in Section 1.3.

Generality refers to the phenomenon of a lexical unit having a range of uses

without the sort of quantal structure that would indicate ambiguity. In this case, we say

that the context of use modulates the sense of the unit, rather than selecting a sense

(and thus, a lexical unit) from among those associated with the form. In other words,

the variation across the uses of a general sense is continuous, and di�erences between uses

are not speci�ed as part of lexical semantic structure at any level; generality corresponds

roughly to what Langacker calls schematicity. A standard example is kick , which speci�es

that a foot is involved, but says nothing about which foot of the kicker is involved. For

human agents, this means that, rather than saying that there is ambiguity between two


senses `kick with the left foot' and `kick with the right foot', we want to say that kick is

general with regard to which foot is involved. This saves us a great deal of trouble if we

want to talk about a centipede kicking a leaf.

I will use the term fuzzy to refer to categories whose boundary is di�cult to de�ne.

A classic example is color terms, which do not have well-de�ned boundaries, even though

they form part of a well-de�ned and tightly interrelated semantic �eld. Other examples are

tall , young , and beautiful ; speakers will agree to a great extent on who is tall, young, or

beautiful and who is not, but there is no way to de�ne the boundary so that everyone will

agree on who is inside and who is outside. I will avoid using the word vague because it is

sometimes used to mean general and sometimes to mean fuzzy (as I am using them).

The basic de�nitions of the terms homonymy, polysemy, and monosemy are de-

ceptively simple; homonymy is the situation in which one form has two or more unrelated

meanings; polysemy, one form having two or more related meanings, andmonosemy, one

form having only one meaning.5 In other words, homonymy and polysemy are both types

of lexical ambiguity, and both can occur at the level of individual word forms or at the level

of lexical forms. As a �rst approximation, we can talk about a continuum from homonymy

to monosemy, with polysemy occupying a middle ground.

At the homonymous end, we have relatively clearcut cases like box (`container'

vs. à blow with the hand'), pool (`billiards' vs. `small lake'), and squash (`vegetable'

vs. `sport'). These examples have come about by chance intersections of unrelated roots

through historical change.

If the meanings of the LUs are su�ciently di�erent, speakers will rarely think of

the words as related, and they will persist as homonyms. This is especially true if the

meanings are of di�erent syntactic categories, as in pine (`long for' vs. èvergreen tree'),

rose (`went up' vs. ` ower'), and saw (`viewed' vs. `saber saw').

Note that in written languages, word forms and lexical forms can refer to either

spoken or written entities; homographs refers to two (or more) di�erent LUs that are asso-

ciated with a single written form, and homophones refers to two (or more) di�erent LUs

that are associated with a single spoken form. In languages with poor sound-writing corre-

5Note that the word homonymy comes from the Greek `same name', while the words polysemy andmonosemy come from Greek roots meaning `many meanings' and òne meaning'. This nomenclature implic-itly takes the position that there are two words that have the same form in homonymy but there is onlyone word that has several meaings in polysemy. I have tried to state my de�nitions without recourse to theslippery notion of word.


spondences, like English and French, non-homophonic homographs (e.g. tear `weep'/`rip')

and non-homographic homophones (e.g. main/mane/Maine) are not uncommon; even so,

most homographs are also homophones and conversely. (The complexity of the situation in

Chinese and Japanese is of a di�erent order.) I will use the term homonymy to cover both

homography and homophony.

Many of the best examples of monosemy are relatively technical terms, like eu-

calyptus , expropriate, intestine, monogamous , and stalactite, because they are precisely

de�ned and do not easily lend themselves to metaphorical uses. As relatively clear exam-

ples of polysemy we can cite late for an appointment vs. the late Mary Gonzalez and high

tea vs. herbal tea .

Tests for Lexical Ambiguity

Arguments based on parsimony require that we assume that a word is monosemous

unless we have evidence for more than one sense. An array of constructions to license various

types of combinations of lexical units (e.g. noun compounding, modi�cation, apposition)

will be needed in any construction-based grammar, so the general mechanisms whereby the

general meaning associated with one LU is constrained by the meanings of the other LUs

in the context must be worked out in any case.

To take a common example, cousin in English is general with regard to the sex

of the person referred to; in contexts such as My cousin helped himself to the cake or My

cousin got pregnant , the meaning is restricted to one gender, but I want to call this a case

of modulation6 of a general sense, not selection between senses. Modulation also occurs

with fuzzy categories such as tall ; varying the context (e.g. tall for a sixth-grader/for an

adult/for a professional basketball player) can move the fuzzy boundaries up and down the

scale of height, but I still want to call this modulation, not an indication of ambiguity.

We will use the following tests to distinguish ambiguity from generality; the dis-

cussion is based both on Geeraerts (1993, 1994a, 1994b) and on Cruse (1986).

The De�nitional Test

This test is founded on the notion of de�nitions in terms of necessary and su�cient

conditions; the basic idea is that a lexical form is monosemous so long as a single de�nition

6I am using the terminology of Cruse 1986; Alm-Arvius 1993 distinguishes between pragmatic exten-

sion and pragmatic restriction.


can cover all of its uses and not those of any other lexical form. If more than one de�nition

is required, then each de�nition is proof that the lexical form has a corresponding sense,

i.e. participates in that number of LUs.

This test actually dates back to Aristotle (cf. Geeraerts 1993:230); its most vig-

orous modern proponent is probably Wierzbicka (1996:242-244). Wierzbicka's program is

based on the assertion that it is possible to write de�nitions that clearly de�ne the meanings

of all LUs in all languages by building them up from a set of atomic semantic universals.

We will see how this works in speci�c cases in Section 1.4.

Yes/No Tests

These are originally based on Quine's (1960) discussion of polysemy in terms of

truth-theoretic semantics, but it is possible for proponents of other semantic theories to

make use of similar tests.

In their original form, these tests are basically assertions of the form \X is Y but

not Y"; if we can truthfully make such an assertion, the argument goes, then there must be

two senses of Y, by the Law of the Excluded Middle. Although this sounds very simple, it

is actually di�cult to �nd simple, acceptable sentences of this type.

(1) a. She's rich, but she's not really rich.

b. She's rich, but she's not rich.

(with lengthening and higher volume on the second rich),

In actual speech, we often �nd sentences like Ex. (1-a), containing the hedge really , meaning

that on a scale from poor to rich, she is closer to the rich end, but she is not near or at the

end, i.e. the second occurrence picks out a more extreme point. As (Kay 1983:135) points

out, \. . . a hedged sentence may contain a metalinguistic comment regarding the way in

which a word or phrase of the sentence is being used in the sentence." Thus, we need not

conclude that rich is polysemous on the basis of Ex. (1-a). The intonation contour on rich

in Ex. (1-b) (which is usually more dramatic than the contrastive stress that \Y but not

Y" would require in any case) probably serves a similar function, so that we cannot regard

either sentence as simply using rich in two di�erent senses.

On the other hand, a sentence like \I'm happy, but I'm not happy" can be inter-

preted as meaning that I feel two contradictory emotions or that di�erent parts of me feel


happy and not happy respectively; both of these situations seem to be perfectly plausible

given the complexity of human emotions, and sentences of this type are actually fairly com-

mon in everyday conversation.7 In this case, we would be wrong to assume that there are

two meanings of happy based on this evidence.

Cruse o�ers a number of clear examples of this test, such as:

(2) a. X: Has Charles changed his position?

b. Y: Yes, he's now sitting next to the chairman.

c. Y: No, he still supports capital punishment.

It is not hard to imagine a meeting to discuss some question of criminal law in which the

question in Ex. (2-a) could be appropriate answered either with Ex. (2-b) or Ex. (2-c); in

fact genuine ambiguity in conversation could easily arise.

Identity Tests

These tests are closely related to the yes/no tests, in that they involve judgements

of the acceptability of sentences given real world contexts. They were particularly popular

with Generative Grammarians, and the fullest account of them, Zwicky & Sadock 1973, is

closely related to a framework that has been largely superseded, even in the minds of its

former fervent advocates. We can, however, omit much of the theory-speci�c detail and still

keep the essentials of a useful test for ambiguity. Basically, the argument revolves around

sentences in which one predication is applied to two participants; the question is whether

the predication can only be appropriately applied to both participants in the same sense.

If it is easy to imagine di�erent extra-linguistic contexts in which the predication could

apply to each participant in a di�erent sense, such a reading for the sentence is usually not

acceptable, as shown in the examples in Ex. (3). All are considered bad on the \crossed

reading" in which Charles moved around the room and Nancy adopted a new philosophical

point of view, but good if both changes are physical or both are mental.

(3) a. \Conjunction reduction": Charles and Nancy both changed their positions.

b. \Pro-verb": Charles changed his position and so did Nancy.

7Not all words for emotions allow this pattern; �I'm enraged/furious/stunned, but I'm not stunned. Thosethat do allow it seem to be \topic-oriented" emotions: happy/excited/worried/delighted about A, but not aboutB.


c. \VP deletion": Charles has already changed his position and Nancy might.

d. \Gapping": Charles changed his position and Nancy hers.

The \bad" combinations are instances of what is called zeugma in rhetoric; it is

sometimes done deliberately for humorous e�ect, e.g. Her spirits rose with the temperature,

and by noon she was actually cheerful .

\Extra Information" Tests

The basic conception here is that the selection of one or the other of a pair of

alternative LUs immediately carries with it a lot of information that is not otherwise in-

ferable from the context. Modulation within a general sense, on the other hand, adds only

the information that results from the composition of the semantics of the context.

The test, therefore, consists in trying to �nd a synonym that will cover both cases

in question, and observing whether there is a considerable \loss" of information when it is

substituted. The e�ect is clearest (as usual) for homonyms:

(4) a. The object ew out from under the roof.

b. The object hit the ball.

c. The bat ew out from under the roof.

d. The bat hit the ball.

In Ex. (4), sentences a. and b. give us only the most general kind of image of the

processes, while c. and d. give us two quite distinct images. In the �rst place, we have to

go all the way up the taxonomy to a term as general as object to �nd a synonym, itself a

suggestion that ambiguity is involved. Having done so, we have little idea about the manner

of the ying motion or the hitting action. As soon as we go back to sentences c. and d.,

much more becomes clear, and the images of the bat in c. and d. are quite distinct. This

must be due to the ambiguity of bat .

Distinguishing Monosemy, Polysemy and Homonymy

Even though the di�erences between monosemy and polysemy on the one hand

and between polysemy and homonymy on the other seem intuitively clear, it has proven ex-

traordinarily di�cult to make each of these distinctions reliably and convincingly. Table 1.1


shows the distinctions to be made.

Ambiguity? No =) Monosemy+ YesRelated senses? No =) Homonymy+ Yes, PolysemyRule? Yes =) Regular, one lexeme (Cruse)+ NoTwo lexemes (Cruse)/ Idiosyncratic Polysemy (Cognitivists)

Table 1.1: Decision Tree for Structure of a Lexical Item

The �rst distinction is between ambiguity and generality. If we have ambiguity,

we must decide what the sources of ambiguity in a given sentence are. Ambiguity can be

pragmatic, e.g. Why don't you talk about it? can be a true question, a polite suggestion,

or command, depending on the extra-linguistic context; this would be of interest mainly to

sociolinguists, not to lexical semanticists. Distinguishing syntactic and lexical ambiguity is

sometimes easy; consider the following:

(5) a. He learned about the problem in the prison.

b. We all like squash.

c. They saw her duck.

In Ex. (5-a), the ambiguity is purely syntactic, having to do with where the PP

is attached, i.e. whether the problem is in the prison and the learning might have taken

place elsewhere, or the learning took place in the prison and the problem might have taken

place elsewhere. None of the ambiguity is due to the semantics of the words. Conversely, in

Ex. (5-b), there is no syntactic ambiguity, only semantic ambiguity associated with squash.

But very often lexical and syntactic ambiguity interact, so that one reading of a word �ts

with one syntactic pattern and another with another, as in the standard example Ex. (5-c),

where the syntax and semantics of both her and duck are involved. This kind of interaction

is the basis of many of standard tests for ambiguity.

The �rst question is whether a particular word form is ambiguous in a particular

context; if there is only one possible interpretation of a sentence, given all the possibilities

of each lexical form, then the lexical forms in it must be monosemous. This does not mean

that it must refer to a very restricted domain. In the sentence All vertebrates have central


nervous systems , the lexical form vertebrate refers to an enormous variety of mammals,

birds, and �shes and central nervous system to a corresponding variety of things inside

them, but both of them are monosemous. Nor are their senses particularly fuzzy; for the

vast majority of animals in the world, virtually every biologist will agree on their status as

vertebrate or invertebrate.

If a word form is ambiguous in a given context, the question as to whether we have a

case of polysemy or homonymy depends on whether the two senses are related. Traditionally,

this has been considered to depend on whether one of the two senses is historically derived

from the other (avoiding for the moment cases with more than two senses); this is in accord

with another popular cognitive model of the lexicon, which states that words have meanings

\in perpetuity". Thus, if one wants to determine, what a Latinate English word means, for

example, one should �nd out what the corresponding Latin word (if any!) or its component

morphemes would have meant to Cicero. Although the idea is seldom stated so baldly, it is

not uncommon to hear people saying things like \Sympathy really means the same thing as

compassion, they both mean `feeling with' somebody else", or \The root meaning of senator

is òld man' ". Of course this is contrary to the division of linguistics into diachronic and

synchronic; if homonymy is to be de�ned in terms of semantic unrelatedness, then we must

mean \in the minds of present-day speakers of the language".

Paradoxically, for those speakers who believe in the unchanging meanings of lin-

guistic forms down through the centuries, the more they learn about the etymology of forms

which had a common origin but have diverged greatly in semantics, the more likely they are

to think of them as related! There is also a constant tendency to construct folk etymologies

that will \explain" perceived relations among senses, whether they bear any relation to

historical fact or not.

This holds even for some parade-ground examples of homonymy. The two lexical

units bank , as in bank of a river and Bank of America, actually are related historically

(Geeraerts 1994a), and one might say that, to the degree that an individual remembers the

intervening steps in the etymology, the words have become synchronically related for that

person. Of course, only a tiny fraction of the speakers of a language will know about most

such etymological facts, and learning them will usually not a�ect peoples' understanding

of the meanings of the words in question, so this process will not have much e�ect on the

lexical semantics of the language as a whole.


Regular Polysemy

There are a number of types of polysemy in which regular relations exist among

LUs; since we will be concerned in this study about the irregular polysemy that is unique

to see, let us �rst look at some common types of regular polysemy. (This discussion will

be based in part on Cruse 1986.)

One type of regular polysemy can be described through various types of rules

or relations among lexical items. These have been analyzed in di�erent ways by di�erent

authors under di�erent names: regular polysemy (Apresjan 1974), lexical implication rules

(Ostler & Atkins 1991), semi-productive polysemy (Copestake & Briscoe 1995), and lexical

functions (Mel'cuk 1996). Examples of lexical relations are listed below.

Unit/Type alternation

The request in Ex. (6-a) might receive the reply in Ex. (6-b), treating that jacket as

an instance (unit) of the type, but the more probable reply, in Ex. (6-c), treats it as

expressing a type.

(6) a. A: I want to buy that jacket in the window.

b. B: Fine, I'll take it o� the mannequin.

c. B': Fine, what size do you wear?

Membership in di�erent contrast sets (a.k.a. Neutralization)

(7) a. dogs and cats vs. dogs and bitches (Cruse 1986, Sect. 11.5)

b. lions and tigers vs. lions and lionesses

c. alcohol and drugs vs. alcohol and glycerol

quality of opus - quality of author

Patterns of the form brilliant/witty/stupid book/play/poem :

brilliant/witty/stupid person, i.e. where the properties of the author are ascribed to

the work8 (Cruse 1986:78)

feeling-evoking

8This may be distantly related to the metonymy of author for work, e.g. Shakespeare is di�cult forstudents today.


She's sad/happy/angry/nostalgic about it. :

It was a sad/happy/angry/nostalgic day. (Ostler & Atkins 1991)

food item - mass

Here's an egg./He won't eat egg. (Ostler & Atkins 1991)

Another, more complex type of relation has to do with a continuum of senses

(which Cruse (1986:72�) calls a \sense spectrum") connected with the same lexical form,

e.g. the mouth (of a human | of a sea squirt | of a bottle | of a cave | of a river) (an

example apparently originally from Lyons (1977:550)). This is a series of meanings such

that each \adjacent pair" are near-synonyms, but the ends of the series are quite di�erent

from each other; although the relation among these senses can be considered metaphorical,

there is no sudden leap from one domain to another, but rather a series of small steps that

are hard to di�erentiate.

In all �ve of the above examples of relations, the LUs involved are of the same

syntactic type and have the same valences. Many authors have also discussed other reg-

ularities which involve di�erent syntactic types and di�erent valences, as well as semantic

di�erences; I will call these alternations.

Cruse gives examples (p.80) of three alternations, Causative/Inchoative (J moved

the rock./The rock moved.), Count/Mass (Have some apple./Have an apple.), and Put them

in a can./Can them., which we could call \Cognate Container"; the last example shows that

LUs of di�erent syntactic categories can be part of the same lexeme, in Cruse's terminology.

These types of regular relations have been studied extensively by various scholars,

e.g. recently for English verbs by Levin (1993), who has produced an extensive classi�cation

of English verbs on the basis of such alternations. The work of Pustejovsky (1995) on the

qualia structure of nouns is also relevant, although he uses a particularly rigid system of

relating di�erent aspects of nouns together, based on an Aristotelian world view, and the

system seems to be more suitable for arti�cial kinds than natural ones. On a wider scale,

we can think of Sweetser's (1990) work on the evolution of verbs of knowing from verbs of

perception, particularly seeing. Although she is looking at it from a historical perspective,

there are quite real synchronic relations which result in systematic polysemy between these

semantic domains.

There are many cases in which a noun and a verb both belong to the same semantic

frame. In English, it is fairly common for some of the word forms to be identical; i.e. the


base form and the base \plus s" of the plural noun and the third singular present verb

are orthographically and phonologically identical, so that they are ambiguous as to part of

speech. In such cases, where both the lexical forms overlap and the senses are part of the

same semantic frame, di�ering only by syntactic category, I do want to call this a case of

polysemy, in accord with Cruse (1986) and Fillmore & Atkins (1992:100-101), and contrary

to most dictionaries, which always give separate main entries (or at least sub-entries) for

di�erent parts of speech, regardless of how closely related they are semantically.

Pattern Levin

1993

Examples

N one instanceof Ving

bump, charge, hit, jar, step, walk

N is a socialevent consist-ing promi-

nently of Ving

dance, lecture, talk

V install N(s)

in/on

light, roof, wire

V remove Nfrom

10.7 Pitverbs

dust (clean house), peel, shell, string(beans, celery), weed

V apply N to 9.9 Butterverbs

paint, stain, varnish, lacquer, caulk,spackle, grout, paper, dust, spray,salt, butter, oil

V expose to N steam, smoke, air, sun

V using N tool hammer, iron, plane, saw

V fasten by

means of N

22.4 Tapeverbs

cement, clip, glue, hook, nail, paste,pin, screw, staple, strap, tape

V place in acontainer or

storage loca-tion N

9.10 Pocketverbs

bag, bottle, box, can, case (mail inpost o�ce), crate, �le, rack (balls inpool), shelve

V split to forma con�gurationlike a N

branch, fork, fan (out)

Table 1.2: Zero-relations between English Verbs and Nouns

These kinds of regularities occur on all scales, such as that between the abstract

unit of currency and the physical cash, as in He pulled a dollar out of his pocket vs. It's not

worth a dollar . Table 1.2 lists a few everyday alternations, including references to Levin


1993 for those that she has included.

Irregular Polysemy

Beyond regular polysemy, however, we �nd a considerable amount of polysemy

that is speci�c to individual lexical forms. For example, in their study of risk , Fillmore &

Atkins (1992) �nd a unique pattern of relations between the noun and the verb, as shown

in Ex. (8).

(8) a. Take a great/huge/enormous risk in the stock market.

b. Risk *greatly/*hugely/*enormously in the stock market.

c. Risk everything in the stock market.

Cruse would presumably not want to group the LUs into a single lexeme in this case, since

he requires that all the LUs in a lexeme be related either by an more general alternation (i.e.

one shared by other lexemes) or by being members of a sense continuum (Cruse 1986:76-79).

There are many such lexically speci�c alternations; a few examples are given in Table 1.3.

V surround with N fenceV conduct through N to pipeV form an N around ringV ow out like an N streamV to follow the Ns of trackV chase up into a N treeV supply active ingredi-ent so that DO containsN

charge

Table 1.3: Some Lexically Speci�c Noun/Verb Alternations

1.4 An Example of Possible Polysemy: Crawl

Consider the following examples of the verb crawl :

(9) a. Ants were crawling all over the kitchen table.

b. The kitchen table was crawling with ants.

c. The ant crawled across the table.


d. The injured man crawled to the road.

e. The ants crawled up one side of the fence and down the other.

f. The baby quickly crawled over to the couch.

g. The tra�c was just crawling along.

h. If you run out of money, don't come crawling back to me!

i. ?The dog crawled quickly down the walk to greet his master.

It seems likely that we want to postulate separate LUs for Ex. (9-a) and Ex. (9-b),

which would be linked in a single lexeme by regular polysemy, demonstrated by the Swarm

subtype of the Locative Alternation (Levin 1993:2.3.4,47.5.1). Other verbs participating in

the alternation are bustle, swarm, and teem.

Ex. (9-c) is clearly an instance of the same LU as Ex. (9-a), although the plurality

and the words all over in the latter modulate the sense to give a rather di�erent image,

given our knowledge of the relative sizes of typical ants and tables. But ants normally crawl;

adult humans do not. We might want, therefore, to suppose that Ex. (9-d) is a di�erent

LU; the motor programs involved for ants and humans are quite di�erent.

What do our tests indicate? Starting with the de�nitional test, we face some

immediate di�culties. In trying to cover Ex. (9-a), Ex. (9-c), Ex. (9-d) and the obviously

related sense of Ex. (9-f) (i.e. babies, injured people, and ants on table tops), we might

come up with something like `to move with the body close to a horizontal surface, making

contact with it with more of the body than (the) two feet'. But in order to incorporate

Ex. (9-e), we have to omit horizontal from the de�nition. If we try to include Ex. (9-g),

we are left with nothing but `move close to a surface', which is clearly unsatisfactory, as

it is not su�ciently delimited from other verbs in the motion domain, such as glide and

run. (This is also the fundamental problem with Ruhl's (1989) venture in monosemy.) If

we try to add Ex. (9-h), where the motion may not be physical at all, we will be left with

no de�nition at all.

We could try to deal with the metaphorical senses (Ex. (9-h) and Ex. (9-g)) as

Wierzbicka apparently does, by appealing to \a general principle which allows us to use

words of di�erent kinds in the sense `like x ' (where `x ' stands for the word's literal meaning)."

(1996:244), but this seems like cheating. If we take the �rst attempt at de�nition given above

as the \literal" meaning and claim that the \like x" device can account for Ex. (9-e) by

picking out the multiple-footed contact and ignoring the orientation of the surface, what is


to prevent us from using \like x" to license Ex. (9-i) to describe a cocker spaniel bounding

down the walk to greet his master, by picking out the horizontality and multiple-foot-contact

and ignoring the proximity to the surface? We �nd ourselves asking what experience tells

us are the wrong kinds of questions|how many features do two things have to share to be

\similar"?

So we conclude that there must be polysemy, and we try to devise multiple de�ni-

tions; this is also problematic. In fact, we seem to have a good example of the kind of \family

resemblance" situation that Wierzbicka claims \never had any empirical basis" (1996:245).

That is, di�erent sets of the examples share di�erent sets of features (e.g. [low �], [slow �],

[multiple-feet �], [horizontal �]), and we have no justi�cation for preferring a de�nition

that groups them one way over one that groups them another way.

Trying the Identity tests, we produce sentences like the following:

(10) a. The ant was crawling on the sidewalk and so was the wounded man.

b. The ant was crawling on the sidewalk and so was the baby.

c. ?The ant was crawling on the road and so were the cars.

d. ??The ant was crawling on the sidewalk and so was the repentant spouse.

e. The oor of the restaurant was crawling with babies and the tables with ants.

f. The table was crawling with ants and the road with cars.

Ex. (10) d. is basically impossible with the `grovel' reading, but c. is better,

because at least the cars are physically moving. Even Ex. (10) a. and b., however, will

probably strike some speakers as zeugmatic. With regard to the other form of the alterna-

tion, both the gapped sentences (Ex. (10) e. and f.) sound zeugmatic, with e. marginally

better, probably due to the pragmatic plausibility of the scene described.

The Yes/No test gives us patterns like Ex. (11) and Ex. (12), which delimit the

literal and �gurative senses pretty well, and also Ex. (13), which suggests that we do not have

two separate senses for vertical and horizontal motion, even though prototypical crawling

is clearly horizontal.

(11) a. A: Was the baby crawling?

b. B: Yes, she was on all fours.

c. B': No, she was moving right along.


(12) a. Was her car crawling?

b. Yes, it was in �rst gear.

c. No, it was rolling on wheels.

(13) a. A: Were the ants crawling?

b. B: Yes, they were walking on the table.

c. B': ?No, they were walking up the wall.

Thus it seems that we have another instance of a �gurative extension (slow move-

ment of tra�c) from one of the literal senses (probably the injured man, although someone

who had never had a baby might wrongly suppose that they crawl slowly, also). `Grovel'

must come from the injured person (or possibly the baby), since the ants are in their normal

body posture. These cases of extension from a single sense of a lexeme to a metaphorical

sense are parallel to that mentioned by Taylor (1995) of John got round his mother as an

extension of one sense of spatial motion round , `get round an obstacle'.

1.5 Other Types of Evidence for Lexical Semantics

The divisions among academic disciplines have traditionally had to do not only

with di�erences in content but also di�erences in what counts as evidence and how it is to be

gathered; indeed di�erences in methodology may be more important in separating disciplines

than di�erences in content, as in the case of certain overlapping areas of anthropology and

sociology. Schools and movements within linguistics, likewise, have often dealt with the same

problems but reached quite di�erent conclusions because they appealed to di�erent types

of evidence and used di�erent types of argumentation. For example, one of the hallmarks

of the transition in America from Structuralism to Generative Grammar was the virtual

abandonment of the study of texts in favor of constructing examples and making intricate

arguments on the basis of their grammaticality, even though both schools were grappling

with many of the same problems about how to delimit morphological and syntactic units

and how to characterize their relations to each other.

The �rst section of this chapter appealed in general terms for the use of a wider

range of evidence and a greater variety of research method in linguistics; this section will

describe in more detail what is being suggested. I will �rst discuss the range of types of

data and data collection methods, and then focus on speci�c studies and what value they


have for the study of lexical semantics.

Types of Linguistic Data and Collection Methods

Since the methods of data collection and the nature of the data obtained are

inextricably linked, they must be discussed together. Let us consider �rst the features that

are common to all data collection, and then move on to those that are largely or exclusively

associated with linguistics.

Selection of Informants and Sample Size

The study of language is inextricably bound up with human behavior; as Lako�

(1987:157-218) has convincingly argued, the attempt to study semantics in terms of truth-

conditions relating language to an objective external world is fundamentally awed. There

thus can be no point in trying to study language without the involvement of human factors,

but it is essential to recognize the subjective elements in our methodology. A striking case

is the use of the researcher's own individual judgements as the basic data to be explained.

The initial hope in Generative Grammar was that the data for the new enterprise

would consist of sentences that were uncontroversially grammatical or ungrammatical and

that a rigorous theory that would account for uncontroversial judgements could itself be

used to decide whether or not certain marginal cases were grammatical (Chomsky 1957:14).

If this had always been the case, if the data for Generative Grammar had always consisted

only of truly uncontroversial sentences, then the question of who supplied the judgements

of acceptability would not matter very much. (For discussion of the di�erence between

grammaticality and acceptability, see Sch�utze (1996).)

In practice, however, grammatical theory has developed on the basis of judgements

about more and more di�cult cases; this has cast increasing doubt on two fundamental

assumptions: (1) that the individual acceptability judgements of the linguist herself are

adequate evidence of the grammatical system shared by an ideal speech community, and (2)

that the linguist's judgements of examples are not in uenced by the theory that she is trying

to construct. There is no external way to evaluate the truth of the latter assumption, but

it is obvious that data from someone without an interest in the results would be preferable.

The problem, of course, is not just that consulting oneself about sentences in a language

of which one is a speaker is the easiest way to get judgements, but that the use of truly


linguistically naive informants begs the question as to whether they can make judgements on

the basis of grammaticality at all. We may not agree with McCawley (1982) in saying \The

alleged ability of speakers of a language to distinguish `grammatical' and ùngrammatical'

strings of words is about as rare and as perverse as the ability to construct puns, an

ability to which I believe it is closely related." (p.78), but the problem of asking questions

which non-specialists can reasonably be expected to answer is a real one. An excellent

demonstration of precisely how to go about this in the case of acceptability judgements is

to be found in Cowart (1997). Sch�utze (1996) is a detailed discussion of the theoretical basis

of gathering such data; he says \While I most often talk about grammaticality judgements

in this book, I treat this as a cover term. . . It should be understood that wherever possible,

I intend the discussion to extend to other sorts of intuition, and I do not wish to imply

that grammaticality judgements in the narrow sense have any special status." (p.2) Some

methods for collecting data on lexical semantics will be discussed below.

The other assumption, about the uniformity of the speech community, is also

quite relevant for gathering data on lexical semantics. On the one hand, the thrust of

most of the work on color naming and basic-level categorization is that some aspects of

categorization are relatively independent of language and culture. On the other hand, there

can be no doubt that (1) much of any lexicon consists of words for cultural kinds, both

objects and actions, and (2) even natural, non-culture-speci�c objects and actions (sky,

ocean, rain, walking, eating, sleeping) are categorized and lexicalized quite di�erently in

di�erent languages, as seen in linguistic studies of epistemic expressions (Traugott 1989),

manner of motion, etc.

If we can demonstrate di�erences in patterns of lexical semantics between lan-

guages, we can hardly assume that they don't exist within languages; in fact, there are

obvious dialect di�erences, such as between British and American, or those re ected in

the lexical portions of dialect atlases. We must assume some variation between individu-

als, as well; as Cowart (1997) shows (at least for grammaticality judgements), statistically

sound conclusions can be drawn even from data containing a high level of variation between

subjects.

Worse yet, the multiple redundancy of the lexicon (Sadock 1984) and the creative

power of the interpretive process, involving both the ability to make metaphorical leaps and

the contribution of pragmatics, means that A may well understand what B means by a given

sentence in context, even if one of the important lexical units it contains is unknown to A.


The novel LU may be permanently added to A's mental lexicon immediately (i.e. \one-

shot learning" from context) or it may spring into existence only brie y and be forgotten

moments later, depending upon a host of poorly-understood factors.

For example, suppose you are standing in a harbor in a hurricane, watching a ship

tugging on a large chain that moors it to the dock. A sailor on the ship looks over at the

chain and yells to his mate, \That cable could break any minute now!" You will probably

immediately grasp the sense of cable that includes what are more commonly called chains ,

even if you had never heard it before; you might or might not remember this sense a month

later.

Although these issues are not often considered in \mainstream" linguistics, it

would seem, then, that the prudent course would be to adopt at least some of the methods

used by other social scientists in similar situations. This means, at least for the collection

of relatively intuitive types of data (Section 1.5), we must collect an adequate sample of

speakers by appropriate means of selection, and make generalizations about our �ndings

with the appropriate caution. Obviously, the linguist's intuition and practiced judgements

play a vital part in directing the inquiry and setting up the hypotheses; if, however, we want

to make statements about, e.g. \all speakers of American English", we must base them on

data drawn from a suitable sample of that population.

Methods of Data Collection

The methods listed here are intended to cover the gamut of ways of obtaining

linguistic information, including those not normally called \data collection"; the uses and

abuses of each will be brie y sketched.

Intuitive and casual elicitation is the dominant source of data for contempo-

rary linguistics. This is not in any way to suggest that linguistic analyses are not the product

of considerable re ection over long periods of time, or that no intellectual skill is involved.

On the contrary, doing good linguistics generally requires careful thought, a highly system-

atic consideration of alternate linguistic forms and alternative analyses, a lively imagination

to produce plausible contexts and convincing (counter-)examples, and an attention to very

subtle di�erences in meaning. At the same time, it requires a sort of \innocence" in making

acceptability judgements; as expressed above, it does not seem that the same person should

be expected to jump back and forth between these two roles.


Introspection, in the sense of conscious attention to and reporting of one's own

mental processes was important in the early period of psychology, e.g. Wundt, but is not

usually considered a credible source of linguistic data. A notable exception to this opinion

is Wierzbicka (1985), who is clearly within the philosophical tradition when she claims that

\. . . to understand the structure of the concept means to describe fully andaccurately the idea [of the prototype]. And to describe it fully and accurately wehave to discover the internal logic of the concept. This is best done not throughinterviews, not through laboratory experiments, and not through reports ofcasual, super�cial impressions or intuitions (either of ìnformants' or the analysthimself), but through methodical introspection and thinking. (p.19, emphasis inthe original)

Monolingual corpus data can be an invaluable counterpoise to the analyst's

own judgements, as eloquently expressed in Ruhl 1989:

Actual data without intuition hardly qualify as data at all. But intuition with-out su�cient supporting actual data is barren; both are necessary. The debateon relative merits has been at best short-sightedness, at worst a barrier to gen-uine semantic research. The greatest need presently in semantic research is forabundant actual data with which our intuitions can do their work well. (p.16)

In any serious work with a corpus, one invariably �nds discrepancies in both di-

rections, i.e. one expects things one does not �nd in the corpus and one �nds things one

did not expect. Finding the unexpected simply provides more linguistic data to explain.

Failing to �nd an expected pattern might mean either that one's hypothesis was wrong, or

(supposing that one has other attestations proving that the pattern is part of the language)

that the corpus is too small or wrongly selected for one's purposes9. For example, most

large corpora are compiled predominantly from written texts; failing to �nd (e.g.) certain

kinds of slang in such corpora is by no means proof that they are not part of the spoken

language.

An inherent problem in working with a corpus is the tradeo� between uniformity

of the dialect represented and the total size and coverage of the corpus; the larger the corpus

and the more varied its sources, the more likely one is to �nd examples that are unacceptable

in the dialect under investigation (usually by default the researcher's own) but acceptable

in some other dialect. Obviously, the better corpus searching tools at one's disposal and

the more fully (and accurately!) annotated the corpus, the more one can discover.

9Fillmore (1992) has a good description of how linguists should use corpora. For a well-reasoned discussionof the need to balance introspection and corpus studies, see also Ruhl (1989:13-16)


Multi-lingual corpora, speci�cally aligned parallel corpora are, in e�ect, a

source for translations. The automatic alignment programs themselves work by �nding

regular equivalences on a rough, statistical basis. The researcher can �nd many more sub-

tle kinds of equivalence. Parallel corpora can be particularly helpful in lexical semantic

research, because di�erent senses of a word will often be translated di�erently, in essence

automatically dividing the examples into senses. Of course, one should not assume this

simplistically; the existence of a translation di�erence in language B does not prove a sense

di�erence in language A. For example, English put on will be translated by di�erent verbs

in Japanese depending on whether the item in question is a hat, a coat, pants, belt or wrist-

watch; the motor programs involved in putting these items on are distinctive, but in the

context of English, wear should be considered general rather than ambiguous; for example,

there is no zeugma in He put on his necktie, coat, and hat . A simple de�nition that will

cover all the cases is a little tricky, however; AHD has \to clothe oneself with; don" (?she

donned her watch). Of course, put on does have other senses which can lead to zeugma, as

in ?He put on his overcoat and an air of indi�erence.

Cross-linguistic elicitation: Given the usefulness of translations, it is often

helpful to be able to actively elicit them rather than to merely look for them in a corpus. The

usual problems can arise when serving as one's own informant for purposes of translation,

but even when relying on others, one must be careful to elicit truly idiomatic, natural

expressions rather than word-for-word calques.

Surveys can be used to gather many types of lexico-semantic information quickly

from a large number of informants; it is common to elicit both onomasiological and sema-

siological data in the same survey, e.g. for a study of student slang, one would ask both

\What do you call a guy who is mainly interested in sports?" and for the de�nition of jock .

To ensure valid data, it is helpful to have a context that discourages frivolous answers,

although instructions may sometimes specify to give \the �rst answer you think of", if too

much re ection may actually interfere with a natural answer.

A good example of the use of a survey to gather semasiological information is

Coleman & Kay 1981.

Laboratory experiments allow the collection of many kinds of data simultane-

ously, including many that cannot be collected in any other way, such as response latency,

which can sometimes o�er insights into cognitive processes. The laboratory setting can

reduce extraneous variation and allow informants to concentrate on a task for a long time;


on the other hand, it is an extremely \unnatural" setting for language use, and often makes

people nervous, so that the normal use of language is very di�cult. Setting up a psycholin-

guistic experiment to vary only one factor of interest without varying other related factors

can be exceedingly di�cult.

Consultation, either in-person with experts or indirectly by the use of references

(including dictionaries and thesauri, both printed and automated) is of course indispensable.

Even if one is convinced that much of what is in a dictionary is misleading or even wrong

with regard to the semantic structure of a lexeme, it can still be invaluable for purposes

such as suggesting rare or archaic senses of a word that would otherwise not come to mind,

o�ering synonyms (and distinguishing among them), providing examples of use and giving

etymologies.

This method of data collection is one of the few that does not lend itself to statis-

tical analysis. In the �rst place, there are usually only a few possible sources of information

available, so that, e.g. one would not necessarily come to believe something more because

one more reference book listed it. In the second place, the value of this kind of evidence is

based on the reputation of its source; one may know, for example, what book is generally

acknowledged to be the \best" source for the meaning of Indo-European roots, and not

bother to consult any others.

Table 1.4 on the following page lists some of the more common types of data that

can be collected for semantic research, together with their classes, i.e. nominal, ordinal,

interval, or ratio.

Some Major Types of Evidence and Their Uses in Lexical Semantics

Time does not permit a full discussion of the following studies, but they are in-

dicative of what seems to be the increasing use of more varied sources of data in linguistics:

Corpus Evidence The FrameNet Project will be a major application of Frame Semantic

principles to corpus data.

Psycholinguistic Evidence Some examples include Coleman & Kay 1981, Sandra &

Rice 1995, Jorgensen 1990, Gibbs 1980 and 1990 (studies of idiom processing), and

(Williams 1992), a study which demonstrates the prototype e�ect between senses of

polysemous adjectives


Type Class

acceptability judgements (not\grammaticality judgements") N/O

similarity rating or ranking O

de�nition or context ! lexeme N

lexeme ! de�nition or example of use N

category ! giving examples N

examples ! naming category N

lexeme ! paradigmatically related lexeme (synonym, antonym, etc.) N

categorization of items (items ! grouping ) N

classi�cation (items + category names ! grouping) N

categorial judgement (items + category names ! judgement) N/O

representativeness rating/ranking O

zeugma judgement N/O

historical relatedness judgement N/O?

timed reading R

response latency priming and probing R

Table 1.4: Types of data collected

Studies of Language Acquisition Johnson (1996), for example,proposes the Con ation

Hypothesis: that children acquire at least some concepts in a manner opposite to Lan-

gacker's idea of acquisition of examples, i.e. that they �rst acquire an undi�erentiated

sense and later subdivide it.

Let us look in some detail at Fillenbaum & Rapoport (1971), one of the most

interesting uses of similarity judgements. The researchers elicited similarity judgements

for words in nine semantic domains, chosen mainly on the basis of previous work having

been done on them: color names, kinship terms, pronouns, emotion names, prepositions,

conjunctions, HAVE verbs, Verbs of Judging, and Good-Bad terms. They used two di�erent

techniques (one rather innovative, involving drawing labeled trees connecting the words) to

derive a ranking of all possible pairs of words in each domain, and then performed some

rather sophisticated statistical analysis on the data. It would take us too far a�eld to discuss

the analysis in detail, but results for two domains will be summarized. In the case of the

color terms, the data basically reproduced the color wheel, with somewhat di�erent results

for the two elicitation techniques.

In the case of the verbs of judging, Fillenbaum and Rapoport compare their �nd-

ings with two linguistic analyses, Osgood (1970) and Fillmore (1969); regarding the latter,


they conclude:

In some measure the present results would appear to be consistent with Fill-more's analysis, although one could hardly claim that they reveal in any clearway the subtleties exhibited by that analysis. This, of course, is hardly surpris-ing, given the necessarily coarse and crude means the subjects had to indicatetheir understanding of the similarities in meaning among the 30 terms judged,and also given that Fillmore's analysis. . . constituted a rather deep analysis ofthe semantic �eld, something probably not within the power of our subjects. (p.206)

The greatest weakness of the study is undoubtedly that, since subjects were asked only the

judge similarity, they had no way of indicating that, e.g. acquit and forgive are alike in

one way and acquit and convict in another. Furthermore, like many other psycholinguistic

studies, it dealt only with single words in isolation; sentential contexts might have helped

make meanings of many words more precise, but the single-word task already took most

subjects more than an hour to complete for the 30 verbs of judging.

Historical Evidence

The diachronic/synchronic distinction, like many idealizations used in linguistics,

is in some respects an oversimpli�cation. For example, an idiolect will inevitably change over

the course of an adult speaker's lifetime, mainly in response to changes in the surrounding

speech community; the most noticeable of these changes is undoubtedly the acquisition of

new vocabulary (and changes in the meanings of everyday words), but there is also likely to

be some atrophy, as words become less common or disappear from the living language. We

may be conscious, for example, of an older speaker's continued use of the word icebox , since

it has been substantially replaced by refrigerator , but not conscious of other words which

that same speaker once used but no longer does. New words spread across populations

irregularly, over varying periods of time, in a manner quite parallel to lexical di�usion of

sound change; this can be clearly seen in the case of slang and technological innovation, since

the spread is quite rapid, but similar changes occur at slower rates throughout the lexicon.

We assume that existing words similarly acquire and lose senses, becoming monosemous or

polysemous (or more polysemous or less polysemous).

Insofar as speakers are aware of the etymology of the expressions they use, records

of historical change may give us clues as to the notions speakers have about the relations


between expressions. Of course, speakers will vary greatly in their awareness of the true

etymology of words they use.

The other use of records of historical change is as a collection of documented

examples of meaning shifts over which generalizations can be made. We presume that the

kinds of meaning shifts which individual speakers creatively produce are likely to be similar

to those that have been common throughout history. Consider the processes that produce

some of the vocabulary associated with cultural innovations e.g. metaphorical extensions

such as drive a car, kill a (computer) process, click on an icon with a mouse. Each coining

or sense extension is a speci�c response to a speci�c set of circumstances, butmechanisms

such as metaphorical extension are unlikely to have changed over recorded history. This

presumes that such linguistic mechanisms are stable, probably by virtue of being special

cases of general cognitive processes which themselves change on an evolutionary time scale

rather than a merely cultural one.

There seem to be, however, cultural or linguistic preferences for one innovative

mechanism over another; for example, some languages, such as English, coin or borrow

words and expressions rather freely, while others, such as Chinese, tend to compose new

words from existing morphemes. We might also note that (apparently in all languages)

the use of novel extensions with no overt marking (metaphors) alternates with the use of

overt comparisons (e.g. similes) (He's a pig vs. He eats like a pig); one or the other may

be preferred depending upon the language, register, milieu, etc. Careful historical research

might �nd that an new meaning passed through a stage of being a common simile before

being acceptable as a metaphorical extension in one language, while in another language

the metaphorical extension came into being directly.

1.6 The Polysemy of See

It should by now not seem surprising that linguists disagree even about the idea

that the word see is polysemous. On one extreme, radical monosemists would claim that

the word has only one sense, and that the apparent di�erences found in di�erent sentences

are the result of interaction with elements of the linguistic and extra-linguistic context (as

in Ruhl's (1989) treatment of the verbs bear, hit, kick, and slap; see also Bouchard 1995).

At the other extreme, some recent approaches (e.g. Sinclair forthcoming) would attribute

meaning only to words used in a context and claim that it is impossible to discuss the


meaning of see as a single word.

In this dissertation, I will adopt a position between these extremes. One could

suppose that there are simply several verbs to see in English, i.e. that there is not only

homonymy between the noun see and the verb see but among the several verbs as well. But

that would be odd, since all the senses of see are related to the \same" lexeme, e.g. all of

them have the same irregular past and past participle forms.

There are cases of verbs which seem to be related in meaning but are in ected

di�erently, (e.g. They hung the picture vs. They hanged the rustler); that this does not

happen with see argues for fundamental unity on morphological grounds. On semantic

grounds, we �nd considerable overlap of meaning among the senses, but no simple set of

necessary and su�cient criteria for all of them.

It therefore seems logical to claim that the word see is polysemous (that it has

more than one sense), and that at least the majority of its senses are related in meaningful

ways that make it easier to learn the entire constellation of senses.

Throughout most of this dissertation, the vexed question of what is a separate

sense and what is merely a speci�c use of a more general sense will be dealt with only

in passing. Instead, the aim will be to look for the �nest distinctions which seem to be

justi�ed, and, near the end, to try to reach some conclusions about which deserve the

status of separate senses and which can be grouped together.

In other words, I will assume that \. . . di�erent constructions are typically, possibly

always, accompanied by slightly di�erent semantic interpretations. . . " (Goldberg 1995:8)

This principle is familiar from the diachronic point of view; whenever language changes so

that there are two competing forms in the same general semantic domain, speakers tend

to associate some (possibly novel) semantic distinction with the di�erence in form. This

accounts for the well-known fact that there are rarely or never any exact synonyms in a

language (Cruse 1986:265-270), but does not imply that there is one and only one sense per

grammatical form|the pairing actually seems to be many-to-many, as we will see below.

Methodological Note

Despite the admitted importance of the larger linguistic and non-linguistic context,

I will for the most part consider sentences in isolation in the manner of many traditional

linguistic analyses; if we believe that the meaning of a sentence is derivable from the mean-


ings of the words in it, and their relations to one another, then it should be possible to

learn something about the meanings of words by studying them at the level of the sentence.

Where relevant, however, pragmatic and discourse-level phenomena will be mentioned.

39

Chapter 2

A Frame Semantic Analysis

2.1 Introduction

In this chapter, I will introduce an analysis of the senses of see using Frame

Semantics, an approach to lexical semantics which describes lexical items in relation to

gestalts based on the speaker's experience of entire situations in which the items are used

(Fillmore 1976, 1982, and 1985, Gruber 1986, Fillmore & Atkins 1992, Fillmore 1994,

Petruck 1995, Lowe et al. 1997, Baker et al. 1998). I will discuss the especially varied

syntax of VPs headed by see, the di�culty of applying some traditional tests for sense

di�erences, and the event structure of seeing; I will also develop a set of senses, �rst in brief

outline and then in greater detail using a new frame notation.

Most of the examples given in this chapter have been constructed speci�cally to

exemplify clearly and simply the meanings and patterns under discussion; when one begins

to deal with real language use, things are rarely so simple or so short. To avoid unnecessary

multiplication of sentence patterns, we will consider mainly active forms, not passives,

although we will note certain senses that do not occur in the passive. Where possible,

corpus examples (showing the source) have also been given; some have been shortened by

omitting irrelevant material. The corpora used are the Brown corpus, a balanced million-

word corpus of American English compiled in the mid-1960s, and the British National

Corpus (BNC) a balanced 100-million word corpus compiled from 1991 to 1994.

CHAPTER 2. A FRAME SEMANTIC ANALYSIS 40

2.2 The Seeing Frame

Let us suppose that instead of depending upon the authority of \the" dictionary as

suggested in the introduction, we examine a few hundred sentences containing the word see

and determine the senses of the word from this data. In almost half the instances, we �nd

an NP representing a human being as the subject of see and an NP representing a physical

object as its direct object. As we look closely at the remainder, however, we �nd that see

occurs in a somewhat bewildering variety of sentence patterns. For example, although most

subjects of the verb are rather straightforward NPs, there is great syntactic variation in

the complements; Table 2.1 shows a fairly complete list of the patterns of complements in

clauses headed by see, except for the more speci�c patterns related to collocations, which

will be dealt with in Section 2.6. (Some of these patterns may be considered equivalent to

each other by many linguists, but I am simply listing the patterns at this point.)

Complementation Pattern Example Sentence

1 none He could barely see.

2 NP She saw the table.

3 NP AP/PP He saw her rich and in love.

4 NP NP He saw her a successful entrepreneur.

5 NP V-ing They saw him running down the street.

6 NP Vbrst We never saw her be kind to her mother.

7 bare clause Alice could see the dog was lost.

8 that clause Alice saw that the keys were missing.

9 if/whether clause I want to see if it's stopped snowing.

10 WH-headed IndirectQuestion

Let's see whether he wants to go with us.

11 PP[to] I unpacked the suitcase while he saw todinner.

12 to it that clause Please see to it that the house is cleanwhen you leave.

13 NP PP[to] The butler saw the doctor to the door.

14 (passive) V-to They had been seen to take bribes.

15 NP PP[as ] They see the contract as a liability.

16 PP[in] NP or NPPP[in]

He saw in her the beginning of a sense ofloyalty.

17 BE seeing NP He's seeing a girl in Des Moines everyweekend.

Table 2.1: The 17 Major Complement Syntactic Patterns (after Fillmore and Kay 1994:6.5)


Although we are considering mainly active sentences, not passives, we should note

that there is an interesting alternation between actives with bare verb complements and

passives with to-in�nitives (i.e. between patterns 6 and 14 in Table 2.1), as in Ex. (1). Fill-

more & Kay (1994:6.6) mention this alternation and Declerck (1983) gives a more detailed

discussion of it.

(1) a. Pat saw Jan write a letter.

b. *Pat saw Jan to write a letter.

c. Jan was seen to write a letter (by Pat).

d. *Jan was seen write a letter (by Pat).

With regard to clauses introduced by WH-phrases (including how), many of these are free

relatives, which can lead to some interesting ambiguities. In Ex. (2-a), what their parents eat

is a free relative, and could be paraphrased with whatever their parents eat or the food that

their parents eat . In Ex. (2-d), the WH-clause is an indirect question, the answer to which

is what the children want to �nd out (this is the sense of see which we will call determine

below). Ex. (2-b) and Ex. (2-c) are ambiguous between the two types of WH-clause, and

hence, between two di�erent senses of see; Ex. (2-c) could be paraphrased either as The

children want to see the food that their parents eat or The children want to �nd out the

answer to the question \What do my parents eat?". Therefore, in evaluating the sense of

see in sentences in which its complement is a WH-clause, we must take such ambiguities

into account; the presence of a word such as want may suggest that the sense of see which

is more volitional or intentional is to be preferred, or other sentence context may help to

resolve the question.

(2) a. The children eat what their parents eat.

b. The children see what their parents eat.

c. The children want to see what their parents eat.

d. The children want to see whether their parents eat.

The only thing that the 17 patterns in Table 2.1 (except the �rst one) have in

common is expressed in the minimal valence requirement for an experiencer and either a

stimulus for or a content of the experience. In fact, as we will see later in this chapter,

even these thematic roles are too restrictive for some uses; all that we can really say is that


the verb normally takes two arguments, which we can call a seer and a seen. Thus, the

minimal lexical entry for the verb see should contain only that information. In Fillmore &

Kay (1994:5.17), the minimal lexical entry for see was shown as in Fig. 2.1 (cf. Section 1.2).

syn

264cat v

lex +

375

sem []

CPL #1 [ #2

ADJ #1

ARG #2

�[seer],[seen]

�

lxm see

Figure 2.1: Minimal Lexical Entry for See (after Fillmore and Kay (1994))

In the �rst pattern in Table 2.1 (representing the sense which we will call faculty),

even the second argument is usually not expressed; one must still see some physical object(s)

but exactly what one sees is not relevant. In Construction Grammar terms, this is called

\inde�nite null instantiation" (INI).

If we ask the question \What types of NPs can be direct objects of see?", we are

thinking in terms of selectional restrictions (called \selectional rules" in Chomsky 1965);

even if we allow various transformations to generate di�erent \surface structures" from a

single \deep structure", this approach is too tied to syntactic con�gurations to capture the

appropriate semantic generalizations.

If we ask instead, what sorts of arguments (both subjects and complements) are

associated with clauses headed by the verb see, and how these are related to the semantic

frames for events like distant perception, coming to know, judging, etc., we are thinking in

terms of Frame Semantics. Leaving aside the human-SEE-physical-object cases, we �nd that

the seen can also be abstract entities, states and events of various types, and propositions.

In addition to humans, seers are animals (usually \sentient", higher animals), and even

inanimate objects, such as radar systems and ships.

There is also considerable variation in the \epistemic stance", i.e. implications


about the reality or unreality of the seen; in addition to conventional unreal objects, as

in see a unicorn or see pink elephants , we �nd patterns that clearly do not involve simple

visual perception, even though the objects perceived are in principle easy to see:

(3) a. I can just see her face when she opens the package.

b. Every time she closed her eyes she saw her father's face.

Ex. (3-a) removes the event from the present (via the clause which expresses futurity even

though the tense of both verbs is \simple present"!) and also adds the words can just ,

which are frequent collocates of this sense1. Ex. (3-b) contradicts a presupposed condition

for physical vision. These senses will be discussed in more detail in Sections 2.5 and 3.2.

Another area of variation that is not represented in the table above is the relative

frequency of di�erent forms of the verb. Some senses of see require or strongly prefer certain

tenses and aspects (e.g. pattern 17 in Table 2.1, used especially in the sense dating), some

lend themselves more readily to passivization or the formation of reduced relatives, etc.2

In the remainder of this chapter, we will follow the traditional linguistic practice

of noting what syntactic and semantic combinations are required, possible or impossible,

but we will not make further mention of frequencies; this will be discussed as a topic for

future research in Section 6.1.

Finally, we also note the presence of prepositional phrases expressing sources and

goals similar to those used with motion verbs, as in Ex. (4).

(4) a. From my o�ce, I can see the bay.

b. From my o�ce I can see to the bay.

c. From my o�ce I can see all the way to the bay.

Ex. (4-a) and Ex. (4-b) are similar in meaning3, and Ex. (4-c) seems to be merely a more

emphatic form of Ex. (4-b). These expressions and their origin will be discussed in detail

in Section 3.3.1cf. Ex. (41) for detailed discussion.2For some of the details, see Table B.1 on page 255.3For some speakers Ex. (4-b) implies that one can see things at intermediate distances as well, while

Ex. (4-a) does not.


2.3 A Sense-Enumerative Approach

One of the problems with using dictionary de�nitions as the basis of a discussion

such as this is that it is not at all certain that speakers of English will agree on or be able

to access consciously such a �ne categorization of meaning. Nevertheless, it seems useful to

create some list of senses as a starting point, which will be �ne enough to capture most of

the important distinctions but coarse enough that people can generally agree (after a bit of

training) about which sense occurs in a particular sentence. Therefore, the more than 1,300

sentences containing a form of the verb see were extracted from the Brown corpus, and a

list of senses of see was created on the basis of intuitive judgements of these examples.

Then, my colleague Chris Johnson and I, working independently, categorized the

senses occurring in a portion of the corpus according to the provisional list of senses. We

reviewed together sentences on which our answers di�ered, revised the list of senses, and

again independently classi�ed another batch of examples. We continued through several

repetitions of this process, on the assumption that if we had listed all the major senses of

see and the de�nitions were clearcut, we should be able to agree on the classi�cation of

previously unseen instances. The result was a list of senses on which we were able to reach

80 to 90% agreement on the classi�cation of new instances.

As far as possible, the sense di�erentiations have been made on on purely semantic

grounds, although we usually �nd corresponding di�erences in syntactic patterns among

senses; the complex relations between semantically de�ned categories and their syntactic

realizations will be displayed in Table 2.2 on page 57 and described in more detail beginning

on page 55 and again in Section 2.5. Further analysis, to be discussed in the rest of this

dissertation, has resulted in the list of senses given below in Section 2.3, which is intended

to be the �nest breakdown of senses of see that can reasonably be justi�ed.

The limitations of simply listing senses are well known; Kilgarri� (1997) criti-

cizes computational linguists for assuming the accuracy of dictionary lists of senses, and

Pustejovsky (1995:Ch. 4) devotes an entire chapter to the \Limitations of Sense Enumer-

ative Lexicons", focusing mainly on the fact that a list implies a number of senses stored

discretely. The only relation among senses which a list represents is a sequence with a

beginning and an end, which is certainly not the right sort; I do not intend to suggest

that this list is a good representation of how humans store their knowledge of the lexicon.

Nevertheless, listing a relatively �ne set of senses (or \uses") will be useful in the following


discussion, at least by way of de�ning a set of names.

Brief List of Senses

Let us consider brie y the senses that we would like to distinguish and examples

of each. The example sentences given in this section are constructed, for the sake of brevity

and clarity, but are similar to those found in real corpora.

Rather than refer to these senses by arbitrary numbers, we will use a name for

each sense, which should be easier to remember in later discussion. Unfortunately, choosing

common words as mnemonic labels for senses inevitably creates other problems; e.g., the

potential names themselves are often ambiguous (such as realize and state), and near-

synonyms for senses (such as regard and witness) may have patterns of valences which are

di�erent from those senses. (cf. Alm-Arvius (1993), who uses de�nitions like \see as a

near-synonym of experience").

The list is divided into three sections:

� Basic Senses

� Semi-Collocations

� Compositional Uses

The basic senses are those that are the least bound to any particular syntax or

lexical forms (aside from see itself). The \semi-collocations" are separate senses that tend

to co-occur with a small number of lexical forms or syntactic patterns, but are not as �xed

as real collocations; the range of words they require as part of their context is usually best

described intensionally rather than extensionally.

I have used the term \compositional uses" for patterns which may be partly con-

ventional, but whose semantics follows entirely from the regular composition of the meaning

of one of the basic senses of see with the meanings of the arguments. (I will discuss true

collocations in Section 2.6.)

Within each section senses are listed in alphabetical order, except that I list two

particularly important senses, eye and recognize �rst. (All of these senses will be dis-

cussed in much greater detail in Section 2.5.)


Basic Senses

eye: To perceive a physical object or motion with the eyes

She saw the cat on the mat.

He saw many armed soldiers in the streets.

He saw the uttering of butter ies in the garden.

recognize: To recognize a fact, understand a situation

They saw the di�culty of winning, but still insisted on bringing it to a vote.

She saw that nothing would change his mind about it.

accompany: To accompany someone on some part of their journey, usually to a departure

point; see out, see to [the door, the bus stop, etc.]

I'll see you to the door.

I o�ered to see him home, but he said he'd be �ne.

Jeeves, please see Mr. Johnson out and Miss Romano in.

condition: to recognize something/ someone as being in a condition

He saw her hungry and cold on the street corner.

I only want to see you happy again.

consult: To visit with an expert or authority in the hope of advice or action

You should see a doctor about that cough.

Going to see a divorce lawyer is a major decision.

dating: To visit with repeatedly for romantic reasons

They'd been seeing each other for a couple of years.

He had a girlfriend last year, but he's not seeing anybody now.

determine: To �nd out, determine

I just wanted to see if you were paying attention.

Let's see what happens to the reaction at 250 degrees.

ensure: To act to ensure that a situation obtains

We'll see to it that he never bothers anybody again.

She always saw that there was a fresh pot of co�ee ready for the meeting.


envision: To envision a counterfactual situation, often the future or past

Thank heavens they didn't try the north face; I could just see them on the cli� at

midnight, freezing and exhausted.

I don't see the Republicans nominating him this year.

faculty: The faculty of vision in general without any particular object.

He came home so drunk he couldn't see.

They operated on her eyes, and she can see much better now.

news: To learn about news by seeing a newspaper or TV newscast

I see that Congress �nally passed the budget.

I see that the Governor is running for President again.

process: To see something/someone perform a process

He saw her notice an item, pick it up, look at it, and then decide not to buy it.

She saw him gaining more political power month by month.

read: To read a short piece of text

Did you see Herb Caen's column this morning?

I saw the memo, but I couldn't believe they would really do it.

setting: (Of a time period or a place) To contain, be the site or occasion of

The 16th century saw an unprecedented owering of literary talent.

California has seen its share of natural disasters in the last few years.

visit: To meet with, visit (a person/persons)

Will you see Fritz while you're in Pocatello?

She's been to see her twice since her operation.

Semi-collocations

classify: To regard or interpret something as something else.

Do you see this o�er as the �rst step in a hostile takeover?

In her he saw the possibility of escape from the boredom of his daily life.

experience: To undergo, to witness and be a�ected by

He saw combat in Libya and Morocco.

I've seen lonely times when I could not �nd a friend.


gambling: To equal another player's bet in order to be allowed to see his or her cards.

I'll see your twenty and raise you ten.

He said he'd see me; his face fell when I turned up the kings.

Compositional uses

hallucinate: To have the sensation of seeing without the usual external physical stimulus.

The exhausted soldiers saw enemies behind every tree.

Hallucinogens can make people see the walls breathing and hear trees talking.

scan: A process analogous to eye or recognize, but performed by a machine.

The new satellite sees a larger area of North America.

When the monitor sees a sudden change in the heart rate, it alerts the nurses.

spectate: To witness the occurrence of an event (including performances)

He saw the accident and rushed over to help the injured people.

She saw the sack of the city by the Romans.

Have you ever seen a cricket match?

She went to New York to see Angels in America.

tour: To visit a place in order to learn about it or for pleasure

After years of dreaming about it, she �nally saw Italy last year.

If you're going to buy their products, you really should see the factory.

vide: To look at a speci�c location in printed matter in order to learn something.

See Appendix C for the mechanical speci�cations.

The megaliths were arranged in a nearly perfect circle (see Fig. 3, p. 48).

Traditional Tests for Sense Di�erences

Having gone through such a long list, the reader might be wondering what evidence

there is to justify such �ne distinctions, and which might be considered di�erent uses of

the same underlying sense. Lexical semanticists have used a variety of tests to determine

whether a word has more than one sense (see Cruse (1986:Ch. 3) for a more thorough

discussion). Let us consider four of them as evidence for the distinctness of some of the

senses shown above: synonymy, separate negation, zeugma, and identity of senses.


Substitution of Synonyms

The synonymy test is quite simple in principle; if we can �nd two words, each of

which serves as a synonym for one type of use but not for another type of use, we have reason

to suppose that the two uses represent di�erent senses. In clear-cut cases of homonymy,

this is easy to do. e.g. ball has two senses, à spherical object' and à formal, elegant dance';

we can treat sphere and dance as synonyms for the two senses respectively, and paraphrase

sentences to demonstrate the di�erence:

(5) a. The sound of the leather sphere/ball hitting the bat electri�ed the fans.

b. Cinderella was the most beautiful one at the dance/ball.

Even in the best case, however, true synonymy is extremely rare (Cruse 1986:265-70); in

Ex. (5-a), sphere is too precise and formal for this context (although perhaps allowed by a

special \literary license" enjoyed by sports reporters), while in Ex. (5-b), the connotation of

elegance is lost when we substitute dance for ball . Likewise, it is unusual to �nd a synonym

that will perfectly �t all contexts even for a single sense of see; that is why the senses listed

above were named with keywords, many of which were not intended to be synonyms for the

sense.

For example, visit is named with a word that is sometimes substitutable for see

in the contexts in which it has that sense, e.g.

(6) On my way to London, I want to stop in Boston to see/visit my old friend Richard.

But see by itself is relatively neutral about how the two parties come to be in contact with

each other, while visit implies that the visitor moves toward the the visited:

(7) I'll see/?visit my sister when she comes here for the convention next month.

Visit with may sound a little better than just visit in this context, suggesting a more equal

participation, but see can express a more casual occurrence than either visit or visit with.

The frequent use of the collocations come to see and go to see suggests that speakers feel a

need to clarify which party is moving in such situations. Note that in Ex. (7) see should be

considered as visit, rather than eye; if we try to negate some the entailments of visit, as

in Ex. (8), most people would �nd the result odd (cf. Separate Negation of Senses, below).


(8) I'll see/?visit my sister when I go there for the convention next month, but I won't

have time to talk with her.

In many cases, a likely synonym will �t one or several of the argument patterns of a sense,

but not others, e.g. the ensure sense can take either a PP with to or a clause introduced

by that ; the partial synonym attend to allows the former but not the latter:

(9) a. You warm up the car and I'll see to/attend to the luggage.

b. I'll see (to it) that the house is properly closed up for the winter.

c. *I'll attend (to it) that the house is properly closed up for the winter.

Likewise, for classify, there are two common argument patterns, one with NP1

as NP2 and one with NP2 in NP1, but not NP1 NP2; one or the other of the NPs must

be oblique. Other semantically similar verbs in the Categorization frame (p. 93) license

di�erent patterns of arguments, as shown in (10).

(10) a. She saw him as

b. She saw in him

c. *She saw him

d. She regarded him as

e. *She regarded in him

f. *She regarded him

g. She considered him as

h. *She considered in him

i. She considered him

9>>>>>>>>>>>>>>>>>>>>=>>>>>>>>>>>>>>>>>>>>;

an ally in her struggle.

These examples could be multiplied for nearly every sense. In short, the match

between partial synonyms and word senses is not good enough to make this a reliable way

to discriminate senses.

Separate Negation of Senses

In this test, we �nd sentences that seem paradoxical until we realize that the same

word is being used in di�erent senses:

(11) A: Did you see Dr. Ramos yesterday?


B: Yes, but only at the country club.

Here, B believes that A is asking about consulting with the doctor, and answers

with regard to both consulting and visiting; the reply could be paraphrased \I met

socially with Dr. Ramos, but I didn't consult him professionally."

The basic idea of this test is to coordinate two such clauses, one with negative

polarity and the other with positive polarity with the same verb. It is possible to get a

simple positive and a simple negative clause together, but usually hedges are added to signal

what is going on:

(12) I saw Alice at work today, but I didn't really see her, because I was so busy.

This is most likely to be interpreted with the �rst see in the sense eye and the second in

the sense visit or consult. This test can be useful in di�erentiating senses that allow the

same argument structures (such as eye, visit, and consult, all of which can take a simple

NP referring to a person as a direct object).

If the arguments are of di�erent semantic types for the two senses, we are dealing

with the third type of test, zeugma.

Zeugma

The basic principle of zeugma is that coordination of two arguments that require

di�erent senses of the same word produces a sentence that is somehow odd, sometimes

comically so, sometimes as a literary device. Consider an example from Cruse (1986:13),

(13) Arthur and his driving license expired last Thursday.

The oddness of this sentence is due to the simultaneous use of two distinct senses of expire

(roughly `come to the end of the valid period' and `die'); syntactically similar sentences that

involve only one of these senses are not odd: 4

(14) a. Arthur and his sister both expired last year.

b. Arthur's car insurance and his driving license both expired Thursday.

4Note that coordination of two NPs which have di�erent thematic relations to the verb can produce asimilar feeling of oddness, as shown in the coordinated subjects of The logger and his chainsaw cut down thetree. We must be careful not to confuse these two issues.


We can test for distinctness of senses by trying to invent sentences that produce

zeugma; I will not create the thousands of sentences necessary to test all possible combi-

nations of the 23 senses listed above, but will give a few representative examples (15)-(21),

varying the syntax where possible so that it will not interfere with the semantics. In each

case, the �rst (a) sentence of a pair uses the same sense of see twice, while the second (b)

sentence uses di�erent senses. The reader may �nd some of the (b) sentences stranger than

others, but none of them seem as natural as the corresponding (a) sentences that do not

combine di�erent senses.

(15) a. determine + determine:

Let's see if she's interested in a picnic and, if so, where she suggests we go.

b. visit + determine:

�Let's see Mom and if she's interested in a picnic.

(16) a. consult + consult:

Next week I have to see my doctor and my lawyer.

b. consult + visit:

�Next week I have to see my doctor and my mother.

(17) a. experience + experience:

Bill saw combat in Vietnam and hard times after he returned to America.

b. experience + eye:

�Jim saw combat in Kuwait and the Pyramids on his way home.

(18) a. envision + envision:

I see a tall, dark woman in your future and an important meeting at which

you will be presented with an award.

b. eye + envision:

�I see the red Miata across the street and a tall, dark woman in your future.

(19) a. recognize + recognize:

Mary saw the sincerity of his a�ection for her, but also that he would never be

the kind of husband she needed.

b. eye + recognize:

�Mary saw the pile of work on her desk and the impossibility of going home

on time.


(20) a. setting + setting:

Japan and California have both seen quite a few earthquakes recently.

b. setting + experience:

�California and its residents have both seen quite a few earthquakes recently.

(21) a. news + news:

I see that the President is going to be in town tomorrow and (that) they're

going to close Main Street for the afternoon.

b. news + recognize:

�I see that the President is going to be in town tomorrow and why you don't

want to go downtown tomorrow afternoon.

Identity of Senses (Crossed readings)

A fourth test for sense di�erences, closely related to zeugma, uses sentences in

which \VP deletion" produces gaps. Supposed we believe that there are two senses of

teacher , one meaning `male teacher' and the other `female teacher' (corresponding to those

morphologically marked in many languages with grammatical gender). Then suppose that

John has written a letter to his high school math teacher Bill Smith, and Joe has written

a letter to his math teacher Sally Jones; we still would �nd nothing odd in Ex. (22). This

indicates that our supposition of two separate senses is wrong, that teacher is general,

covering both male and female, rather than ambiguous between two distinct senses.

(22) John wrote to his teacher and Joe did, too.

Now let us apply this test to see. Ex. (23) can be considered to contain two instances of

seeing, one where the experiencer was Jan, and one where the experiencer was Chris. If see

has two senses, visit and eye, either of which could be occurring here, we might expect

there to be four acceptable readings of the sentence, one for each combination of senses.

There is certainly nothing wrong with Ex. (23) in a context in which Jan and Chris both

went to visit Pat, or in a context in which both of them caught eeting glimpses of Pat. But

we �nd Ex. (23) unacceptable in a situation in which Jan visited with Pat for ten minutes

and Chris only saw Pat through the window of a passing train; the single word saw \resists"

interpretation as visit in the �rst clause and eye in the second.


(23) Jan saw Pat and so did Chris.

Suppose a memo is written in a language that Isabel can understand but Alicia cannot.

Then Ex. (24) is odd, because we suppose that Isabel read and understood the memo

(sense read), while Alicia just saw incomprehensible letters on a page (sense eye).

(24) Isabel saw the memo and so did Alicia.

Summary

Unfortunately, as many authors have noted, conclusions about sense di�erentiation

based on one of these tests often di�er from those based on another, and many other factors

may in uence judgements.

For example, Ex. (25-a), which should be an instance of setting + setting may

be questionable because of general semantic constraints on coordination, regardless of the

zeugma problem. But there cannot simply be a rule against coordinating a place and a

time; other factors seem to be at work, also, e.g. Ex. (25-b) seems fairly acceptable, but

Ex. (25-c) seems worse.

(25) a. ?The �fteenth century and northern Italy saw a owering of the visual arts.

b. When we think of great painting, we think of the �fteenth century and Italy.

c. ?The �fteenth century and Italy were Jeri's primary research interests.

Furthermore, there is considerable disagreement among speakers in their judgements on

these tests, such as whether Ex. (26) is acceptable in a \crossed readings" situation (e.g.

when John had a physical examination and Mary had a date). Ex. (16-a) (above) also elicits

di�ering judgements.

(26) John saw Dr. Jones today and Mary did, too.

Thus, the apparently simple question of exactly how many senses there are and

how we can test for divisions between them probably has no simple answer. It may be more

productive to take the rather �ne breakdown given above as a starting point and look into

the semantic and syntactic relations among these putative senses.


Co-occurrence of Complement Patterns with Senses

As an introduction to considering the relation between the semantics of the senses

of see and the syntax of the various types of complements, we should discuss brie y the

broader question of the relation between syntax and semantics. Strictly speaking, predict-

ing all of syntax from semantics or all of semantics from syntax is impossible; each sub�eld

has its own raison d'etre. Many linguists, however, tend to regard one of these two sub�elds

as more important, basic, or interesting than the other, so they may have a tendency to

de�ne the domain of the \more important" sub�eld more inclusively, at the expense of the

\less important".

In the recent history of linguistics, Chomsky 1957 (which marks what Wasow

(1985) calls the �rst phase of the Chomskian revolution) represents a high-water mark of

syntax:

Grammar is best formulated as a self-contained study independent of semantics.I particular, the notion of grammaticalness cannot be identi�ed with meaning-fulness . . .. . . one result of the formal study of grammatical structure is that a syntacticframework is brought to light which can support semantic analysis. Descriptionof meaning can pro�tably refer to this underlying syntactic framework, althoughsystematic semantic considerations are apparently not helpful in determining itin the �rst place. (1957:106,108)

Since that time, there has been a steady movement in various schools of linguistics

toward increased emphasis on the lexicon, with more and more semantic marking in the

lexicon5, and the gradual ascendance of the view that the structure of a phrase is largely

the result of projection from its head lexeme. As Wasow (1985:203) puts it, \Much of what

was stipulated in grammars of earlier theories is taken to be a function of lexical semantics."

(For the history of the most radical movement toward semantics-based syntax, Generative

Semantics, see Lako� 1989 and Harris 1993.)

Verb phrases present the most prominent and clear-cut examples of the projection

of the properties of the lexical head onto the larger syntactic structure, and linguists have

long noted that the syntax of argument structure seems to be related to the semantics of the

verbs involved. Levin (1993) is the most ambitious recent work on the argument structure

of English verbs and contains extensive references to earlier, more speci�c studies. Levin

5It is striking that Chomsky 1965 introduces features such as � animate and � human as part of thelexicon, but insists on calling them syntactic rather than semantic features.


uses patterns of alternation of argument structure to group verbs and then attempts to

describe these groups in semantic terms. She says,

This work is guided by the assumption that the behavior of a verb, particu-larly with respect to the expression and interpretation of its arguments, is toa large extent determined by its meaning. Thus verb behavior can be used ef-fectively to probe for linguistically relevant pertinent aspects of verb meaning.. . . [This book] should help pave the way toward the development of a theory oflexical knowledge. Ideally, such a theory must provide linguistically motivatedlexical entries for verbs which incorporate representation of verb meaning andwhich allow the meanings of verbs to be properly associated with the syntacticexpressions of their arguments. (p.1)

Let us then consider, for the verb see, which participates in so many of Levin's

(1993) alternations, what connections there might be between the semantics of its senses

and the syntax of its arguments. The relation between the senses listed in the \Brief List

of Senses" (p. 45�) and the complement patterns given in Table 2.1 (p. 40) is summarized

in Table 2.2 on the next page. Each dot indicates that a particular sense can occur with a

particular syntactic pattern.

As Table 2.2 suggests, the situation is complex, but not hopeless. In an ideally

simple world, there would be exactly one dot in each row and one dot in each column; that

is, each sense would have exactly one syntactic form of expression. In fact, by looking across

the rows, we can see that this is the case for all of the senses which require physical vision

(eye, faculty, news, read, visit, vide, and audience).6 The other senses of see have

more diverse forms of expression, largely because many of them refer to states and events,

which can be expressed in a variety of syntactic forms.

Looking down the columns of Table 2.2 instead, we �nd that most syntactic pat-

terns are ambiguous among two or three senses. Unfortunately, the most basic argument

structure, the simple transitive use of the verb with a single NP direct object, can express

more than a dozen meanings. In part this is a re ection of the semantic exibility of the

transitive sentence type, independent of the verb see. Another reason is the greater seman-

tic complexity of nominals (cf. Fillmore 1994); they can express not only objects of physical

perception and facts apprehended by cognition, but also facts to be investigated (deter-

mine, e.g. Let's see the temperature), people to be called upon (visit, consult), situations

6Many other senses allow physical vision as the source of evidence for a proposition, but do not requireit. For example, process does not require physical vision; in She saw him gaining political power month bymonth, the source of information can be what she heard just as well as what she eyeed. A similar argumentholds for determine, recognize, condition, etc.


none

NP

NPAP/PP

NPNP

NPV-ing

NPVbse

S�n

Swhether

WHIndirectQuestion

PP[to]

toitthatclause

(passive)V-to/Ving

(passive)tobeNP/AP

NPPP[as]

PP[in]NP/NPPP[in]

BEseeingNP

Basic Senses

eye �

recognize � �

accompany �

condition � � �

consult �

dating � �

determine � � �

ensure � � �

envision � � � �

faculty �

news �

process � � �

read �

setting � � � �

visit �

Semi-Collocations

classify � �

experience � �

gambling �

vide �

Compositional Uses

audience �

discourse � � �

hallucinate � � �

scan � � � � � � �

spectate � � �

tour �

Table 2.2: Senses by Complement Syntactic Patterns


to be imagined (envision), items to be read, performances to be watched, etc. Abstract

states and processes can also be expressed as nominals, with accompanying arguments (e.g.

see the di�culty of doing it perfectly, see her success as a stockbroker).

(27) a. determine: Let's see what he does.

b. recognize: She saw what he had done.

c. envision: I can just see what he must be doing now.

VPs that take WH-complements can express an unusual combination of senses as in Ex. (27)

but that is because they are often free relatives, usually functionally equivalent to NPs, as

in Ex. (27-b) and Ex. (27-c). True indirect questions, including clauses introduced by if or

whether work only for the sense determine.

I have included in classify the patterns see X as Y, see Y in X, and see in X

Y, as they are very similar semantically (cf. Ex. (10) on page 50). All three patterns are

speci�c constructions for this sense, although the latter two are of course identical in form

with a simple locative, e.g. to see his grandmother in his daughter vs. to see his grandmother

in his garden. The order of the NP and the PP[in] seems to be determined by \heavy NP

shift", and the PP[in] can also be located at the beginning of the clause.

2.4 Semantics

Merely presenting a list of senses like the one given in Section 2.3 above, while a

useful �rst step, does little to elucidate their interrelations. Yet there does seem to be some

structure to the senses; for example, understanding something about how the senses are

related to each other should help us understand what a learner of English (as either a �rst

or second language) must learn separately about each sense of the verb and what is more

general.7

There can be little doubt that the most common sense is eye; as noted above,

more than a third of the corpus examples fall into this category. recognize is another

frequent sense. Thus we would predict, for example, that if speakers of English were asked

to give examples of a typical sentence with see, examples of eye would be the most frequent,

recognize would be somewhat less frequent, and sentences with senses such as setting

7Later in this chapter, I will make some of these relationships much more explicit in terms of inheritanceamong frames; Chapter 4 will provide evidence about central vs. peripheral senses.


would be quite rare.

The most typical senses will also be those that are shared by the largest number of

speakers, while some of the less common senses may not even be familiar to all members of

a speech community. For example, speakers who rarely play card games may not be familiar

with the gambling sense, and some speakers with less formal education are unfamiliar with

the setting sense.

As Rhodes (n.d.) has noted, the less typical senses not only tend to have many

restrictions on their use but also carry with them more presuppositions. For example,

the gambling sense is normally used only in speaking, in the midst of a card game on

which money is wagered, the subject is normally the speaker, the direct object is a sum of

money, the tense is usually present or future, it is a performative, cannot be negated, etc.

Conversely, the most basic sense, eye, can be used in any tense or person, in the a�rmative

or negative, with or without modals, the object can be almost any physical object, etc. In

discussing the relations among the senses, we will start with the most typical ones and work

gradually away from them.

The Event Structure of Seeing

From our common experience as human beings, it seems reasonable to assume

that the prototypical use of see is to describe an event in which the following are true: a

human being sees a physical object with the eyes, the seer is aware of seeing, the image

or percept corresponds directly with a physical object present in the real world within the

seer's visual �eld, the eyes of the seer function normally, there is adequate light and no

obstruction of vision, the seer correctly categorizes (recognizes) the object seen, and comes

to hold some true beliefs about the world (e.g. that seen exists in a certain location and

has certain properties such as a color and a size) as a result of this process.

At the very least, we can consider this event to be composed of two sub-events,

perceiving and coming to know; the former is typically punctual, telic, and resultative,

while the latter is inchoative, a transition into the state of knowing or believing. Many

senses contain something of both of these sub-events, but senses will di�er with regard

to the relative importance of the two; in particular, when the latter is emphasized, as

in senses like recognize and condition, the direct object of the verb is more abstract

(e.g. a proposition or state) and more mental processing is implied, which requires the


experiencers (seers) for these senses to be humans, or at least sentient beings. Thus, the

\simple" recognition of objects and processes is �ne with subjects fairly low on the scale of

sentience (Ex. (28-a)) but not the recognition of more complex states of a�airs(Ex. (28-b)),

yet the latter is �ne with more \intelligent" subjects (Ex. (28-c)).

(28) a. The gold�sh saw the food oating on the surface of the water.

b. ?The gold�sh saw that the mesh was to �ne too swim through.

c. Fido saw that his master would die unless pulled from the ames.

(See Table B.1 on page 255 and Section 3.5 for a more complete treatment of these restric-

tions.) When the �rst sub-event (visual perception) is emphasized, there is no implication

about the resulting beliefs, as will be discussed in the following section.

2.5 Detailed Frames and Descriptions for Senses

Introduction

As we look more closely at the syntax and semantics of each sense, we �nd a wealth

of detail that will not �t into any chart or list of atomic features. In this section, I will show

a frame for each sense, followed by some example sentences for each sense, from corpora

where possible (some of the long corpus examples are abbreviated). These are followed by

a prose detailed description of the sense. The subdivisions and order of presentation are

the same as in Section 2.3.

Most of the frame notation should become clear from the examples, but it may be

helpful to discuss some of the concepts brie y here.

Frame Inheritance and Scenes (Sub-frames)

Two types of inheritance are shown in the frames, partial and complete. For partial

inheritance, the attribute \Uses" is given, followed by the name of the frame inherited from;

in this case, all the semantics of the ancestor frame is used, but not all of its roles. The

relation between the roles in one frame and those which it uses from another frame are made

explicit through role equations, usually with the name of the ancestor frame abbreviated.


Only those roles which appear in an equation in the daughter frame are used by it.8 For

complete inheritance, the attribute \Inherits" is given, followed by the name of the frame

inherited from; in this case, all the semantics and all the roles are inherited.

Within the entry for each sense of see, frames are listed in order of inheritance; this

means that the frame for the sense of see itself may be preceded by several more general

frames from which it inherits. The frames for senses of see all have names of the form

see eye, see recognize, etc. to distinguish them from the more general frames.

More general frames have been postulated not only where they are shared by two

senses of see, but also at levels where certain generalizations are best stated. In most cases,

there is a prose statement of the reasons for setting up each general frame, with a list of

some words that I believe will need that frame as part of their own description.

Where one event frame is must be described in terms of sub-events, these are listed

under the attribute \Scenes" in the frame. Scenes, like frames, have types and may use or

inherit from other frames, in addition to any inheritance of the frame as a whole.

Semantic Speci�cations

The semantics of frames and their participants are expressed in several ways,

formally through types, predications, and equations, and informally in prose description. It

is hoped that eventually all the information contained in the prose description can also be

expressed in the Frame Semantic formalism. Obviously, the choice of names for frames and

roles is also intended to be meaningful, although it is sometimes necessary to invent names

for them.

Frames themselves may be of two broad semantic types, event or state, indicated

by the \type" attribute. Roles also generally have semantic types, and these types are in-

herited along with the role. Frequently the daughter frame will further restrict the type of

an inherited role. For example, see eye uses perception distant which in turn uses per-

ception basic; perception basic has a role for the body part involved in the perception,

without restriction as to what body part is involved, perception distant requires that the

body part be eyes, ears, or nose (the distant senses), and see eye further restricts the body

8This mechanism might serve several functions. For example, it can represent Fillmore & Atkins's (1998)frame blending, as when one sense of argument inherits some structure from the frame for contentionand another sense inherits from the frame for reasoning, while both inherit from the talk frame. Partialinheritance may also allow representation of Turner & Fauconnier's (1995) blended spaces, but this hasnot been worked out in detail.


part to eyes. A tree showing part of the type hierarchy presupposed in these descriptions

is found in Fig. 3.8 on page 125.

Semantic restrictions are also expressed in the form of predications on roles, with

the attribute \Restr", in cases in which it seems inappropriate to call the restriction a type;

the following predicates are used in these frame descriptions:

a�ects(X,Y) Situation X a�ects Y.

authority(X) Person X is in a position of authority, possesses special knowledge, etc.

familiar(X,Y) Person X is familiar with Y.

partOf(X,Y) X is a part of Y|not necessarily a proper subpart.

temporalOverlap(X,Y) Events X and Y overlap in time.

Other Attributes

In some cases, valence descriptions of the standard type in Construction Grammar

link the roles to syntactic patterns in which they can be expressed. If no such valence

description appears, the assumption is that there are no syntactic restrictions speci�c to

the sense in question, and the syntax of the sentence will be built up by uni�cation with

more general frames and constructions. The frames for senses of see themselves also have

the attribute \Category" (Cat) with value \V", and the attribute \Lexeme" (Lexm) with

value see, i.e., they are lexically speci�c frames.

The standard construction grammar thematic roles stimulus and content (based

on thematic roles) are used in some places, even though the names are not entirely felicitous

in this context. Stimulus is used to refer to the physical objects or actions that are

perceived, while content refers to the mental interpretation resulting from the perception9.

Typographic Conventions

Frames are displayed as attribute-value matrices with the attributes down the left

side and the values down the right. Atomic values are shown in boldface; values which are

themselves AVMs are in large square brackets. Names of roles are capitalized, names of

9These correspond roughly to the base space and the belief space, respectively, in the mental spacesanalysis in Section 3.2.


frames and types are lowercase. Types are displayed to the bottom right of the object in

small italics. At the beginning of the discussion of each of the senses involving inheritance,

I have inserted a small diagram showing which frames inherit from (or \use") which. Where

a frame is used by a scene within another frame, the arrow showing this relationship goes

inside the boundary of the latter, directly to the inheriting scene.

Basic Senses

Sense EYE

perception basic

perception distant

see eye

Inheritance for see eye.

Frame perception basic

type event

roles:

266664Perceiversentient

Phenomenon

Body part

377775

The Body part role is appropriately called the \organ of perception" in the case

of see, hear , and smell , but the organ of perception is a little harder to de�ne in the case

of taste, and amounts to the entire surface of the body in the case of touch.


Frame perception distant

type event

Uses perception basic [pb]

roles:

2666666666664

Perceiver = pb.Perceiver

Phenomenon = pb.Phenomenon

Body parteyes _ ears _ nose = pb.Body part

Loc perceiver

Path

3777777777775

Frame see eye

type event

cat V

lexm see

Uses perception distant [pd]

roles:

266664Seer = pd.Perceiver

Seenphys obj _ phys motion = pd.Phenomenon

Body parteyes = pd.Body part

377775

val:8>>>>><>>>>>:

266664role Seen

syn nuc

pt NP

377775

9>>>>>=>>>>>;

The usual implicatures about physical vision described in Section 2.4 apply to

see eye. Ordinarily, if seen is a person, we would assume that a more speci�c sense can

be inferred, i.e. visit or consult, unless there is some indication to the contrary.


Sense RECOGNIZE

cognition inchoative

see recognize

Frame cognition inchoative

type event

roles:

264CognizersentientContentproposition

375

Words that use the cognition inchoative frame include: see recognize, learn,

�nd out, realize, recognize, ascertain, and discover.

Frame see recognize

type event

cat V

lexm see

Uses cognition inchoative [ci]

roles:

264Seerhuman = ci.Cognizer

Seen = ci.Content

375

val:8><>:

264role Seen

pt S�n _ NP

3759>=>;

(29) a. Brown: . . . , we see 14 major problems which fall into three broad groups|the

market place itself, marketing methods, and marketing management.

b. Brown: Despite a too long sustained declamatory ight, this �nal speech is

convincing, and we see why British audiences apparently were impressed by

\Roots".

c. BNC: I don't see there's anything to explain.

d. I don't believe there's anything to explain.


The seer is a conscious, sentient being, the seen is a state of a�airs, seer comes to know

seen. recognize implies that seen is true of the world. seen is typically an abstract

NP (Ex. (29-a)) or an embedded clause with or without a complementizer (Ex. (29-b)).

Sometimes there may be just a bare S; Ex. (29-c) illustrates such a situation. It is combined

with \negative raising", a phenomenon which is also common with other verbs of judging

and knowing, as in Ex. (29-d); the negative is logically part of the lower clause, so that these

two sentences could be paraphrased \I see/believe that there isn't anything to explain.")

When content is expressed as an NP, it must be of the sort capable of expressing

a proposition (cf. cognition inchoative). With see, this is usually a de-adjectival noun with

a possessive expressing the argument of the predicate as in Ex. (30-a); hence, this means

essentially the same thing as Ex. (30-b).

(30) a. She saw the futility of opposing them.

b. She saw (that) it was futile to oppose them.

Sense ACCOMPANY

directed motion

accompany

M1 M2

see accompany

Frame directed motion

aktionsart activity

type event

roles:

2666666664

Trajector

Source

Path

Direction

3777777775


The directed motion frame is very general, with a set of roles that are very

widely used by other frames.

Frame accompany

aktionsart activity

presupps 1

profiles Part1

type event

roles:

264Part1 = m1.Trajector

Part2 = m2.Trajector

375

scenes:

26666666664

264 event [m1]

Inherits directed motion [dm]

375

264 event [m2] 1

Inherits directed motion [dm]

375

37777777775

restr:

264partOf(m1.Path,m2.Path)

temporalOverlap(m1,m2)

375

As mentioned in the de�nition on page 62, the partOf relation need not denote a

proper subpart|one can accompany someone all the way from their starting point to their

destination, or just part of the way.


Frame see accompany

aktionsart accomplishment

type event

cat V

lexm see

Uses accompany [acc]

roles:

2666666664

Seerhuman = acc.Part1

Seenhuman = acc.Part2

Goal = acc.m1.Goal

acc.m1.Src = acc.m2.Src

3777777775

restr:

�protect(Seer,Seen)

�

(31) BNC: \It'll be no trouble at all to see you to your door," he had lied.

see accompany means to accompany someone who is leaving or arriving on some part

of their journey, often to some departure point. seer and seen are both human, goal is

the point of departure or arrival. The accompany sense implies a polite social situation,

and that the seer is \protecting" or \guiding" the seen. See out treats the entrance of

the building as the point of leave-taking. See [a person] to always speci�es the point of

leave-taking; see you home is essentially a more idiomatic way of saying see you to the door

of your house (cf. Ex. (31)).

These uses must inherit some of their meaning from the more general frame ac-

company (shown above) which also licenses VPs such as walk you home, walk you to your

car , show her up (i.e. to the parlor on the second oor). Take someone home and bring

someone home can also be included, provided that we are clear that home is treated here

strictly as a destination. All of the patterns for the sense accompany of see (and similar

senses of walk , drive, take, run and bring) are therefore distinct from those in Ex. (32);

in Ex. (32), the object of the PP-to is an expression for something that is metonymic for a

conventional activity; di�erent types of metonymy are used in each of three examples (cf.

Norvig & Lako� 1987 on the semantics of take).


(32) a. take someone to a movie/the movies

b. take someone to the doctor

c. take someone to church/school

Sense CONDITION

bare state

see condition

state

Frame bare state

type state

roles:

264PartStaterelational

375

val:8>>>>><>>>>>:

266664role Part

gf ext

pt NP[acc]

377775,264role State

pt AjP _ NP

375

9>>>>>=>>>>>;

The state is predicated of the Participant; note the semantic type of the state

must be relational or scalar, not intrinsic, what Lako� (1987:498-501) calls a \non-inherent,

nonpersistent predicate"10, cf. Ex. (33).

(33) a. him old/successful

b. *him Russian/good/brave/tall

c. him a father/king/a tyrant

d. *him a Japanese/?a coward

10There is a similar distinction in Spanish as to which which adjectives can be complements of ser andwhich of estar ; the former are inherent and stable, the latter temporary and \accidental".


Frame see condition

type event

cat V

lexm see

roles:

264SeerSeen = 1

375

scenes:

264264 state 1

Uses bare state [bs]

375375

(34) a. Brown: He was delighted to see them so happy.

b. He hopes to see her a member of Congress next year.

c. Brown: \. . .We see a nation that traditionally values sovereignty above all else

willing to give up its economy, placing this authority in Continental hands."

This sense is much less frequent than process. The types of states that can occur in this

sense are restricted by the inheritance from bare state. In this sense, the verb see can

be described as having two complements, NP XP, where XP is either AP or NP. The XP

represents a state predicated of the (�rst) NP, e.g. Ex. (34-a), Ex. (34-b).

Sense CONSULT

consultation with authority

see consult


Frame consultation with authority

type event

roles:

2666666664

Consulterhuman

Consultanthuman

Problem

Purpose

3777777775

restr:

264authority(Consultant)a�ects(Problem,Consulter)

375

(35) a. He went to the doctor for his back pain.

b. I think you should consult a specialist.

c. You should ask John about your transmission.

d. I want to check with someone at the consulate about renewing my visa.

e. You should see someone about your cough/tax problems/parking ticket.

f. The Parties agree to consult with each other before taking any action which

might a�ect the value of the Property.

Words that use this frame include see, consult , talk (with/to), ask , and check

(with). The Problem is typically realized as PP[about], sometimes PP[for] in a medical

context (Ex. (35-a)); it is often inferred from context (called \de�nite null instatiation"

(DNI) in Construction Grammar) as in (Ex. (35-b)) or via metonymy (Ex. (35-c)). The

Consultant includes not only doctors and lawyers, but also political leaders and anyone in

a gate-keeping position (Ex. (35-d)). The Consultant cannot be DNI, but can be expressed

by an inde�nite pronoun, relying on discourse context (Ex. (35-e)). Note that the verb

consult itself can also be used in situations of equality, such as Ex. (35-f), which is not

an example of this frame. In naming the roles, we are implying a situation of inequality,

calling the authority the consultant and the person seeking help the consulter.


Frame see consult

type event

cat V

lexm see

Uses consultation with authority [ca]

roles:

266664Seer = ca.Consulter

Seen = ca.Consultant

Problem = ca.Problem

377775

val:8><>:

264role Seen

gf obj

3759>=>;

(36) a. You should see a doctor about that cough.

b. You should have a doctor see about that cough.

c. ?You should get a doctor to see about that cough.

d. *Have a lawyer see [you] about writing your will.

The simplest and most common pattern is seen in Ex. (36-a), but seer and seen can some-

times be swapped by adding a causative without much change in meaning, as in Ex. (36-b)

or Ex. (36-c) (the latter is pragmatically odd because of the high social status of doctors).

But this causativization seems to work only for visits to the doctor; Ex. (36-d) is bad,

perhaps because a visual examination of the patient's body is an inherent part of a medical

consultation, but not of consultations with other experts. The situation in which a person

of higher social status sees one of lower social status is treated as a separate sense of see,

i.e. audience (q.v., p. 100). Furthermore, some common expressions like this, such as

Ex. (36-b) are ambiguous among several readings, consult, ensure, and perhaps even

determine, if one is mainly concerned with diagnosis rather than cure.

The uses with Problem appearing as PP[about] may be related to the sense de-

termine and frame ascertain. For example (37-a) can be interpreted as ensure (move it

before I get a parking ticket) or determine (�nd out why it keeps stalling at stoplights).

We would probably call (37-b) sense consult, but, as with consulting the doctor, the


purpose can still be either just getting a diagnosis of the problem or having the problem

solved.

(37) a. I need to see about my car.

b. I need to see a mechanic about my car.

Sense DATING

perception basic

recip event perception distant

social interaction see eye

romantic involvement see visit

see dating

State 1 State 2

Frame romantic involvement

type state

roles:

264Part1 = si.Part1

Part2 = si.Part1

375

scenes:

2666664

266664event

viewpt iterated

Uses social interaction [si]

377775

3777775

Obviously, romantic involvement is not merely social interaction, even if repeated.

For our present purposes, we will assume that the Western concept of romance is understood,

without attempting to de�ne it.


Frame see dating

denotes 1

entails 2

type event

cat V

lexm see

roles:

264Seer = sv.Seer = ri.Part1

Seen = sv.Seen = ri.Part2

375

scenes:

266666666666666664

264 state 1

Uses romantic involvement [ri]

375

2666666664

state 2

aktionsart activity

viewpt iterated

Uses see visit [sv]

3777777775

377777777777777775

restr:

�temporalOverlap( 1 , 2)

�

(38) He saw her on and o� for a year or so.

The iteration of the visiting allows it to be construed as a state, over the same time

span as the romantic involvement. The social interaction of visit is necessarily reciprocal

(often expressed by seeing each other); one could debate whether it is best to treat romantic

involvement as always reciprocal (as shown here) or to allow for the possibility of unrequited

love, but in any case, the inheritance from visit guarantees that dating will be reciprocal.

The fact that this sense has the activity aktionsart and the iterative viewpoint

makes it compatible with the progressive form of the verb. For some speakers the progressive

is required, while other speakers �nd sentences such as Ex. (39) acceptable, so long as the

durativity is expressed in some way.

(39) Federico and Olga saw each other for a couple of years but that ended when they

went o� to college.


Sense DETERMINE

ascertain

see determine

Frame ascertain

aktionsart accomplishment

type event

roles:

264CognizersentientFactssoa

375

val:8><>:

264role Facts

pt S�n _ NP _ IndQn

3759>=>;

Some words that use the ascertain frame are: ascertain, discover , learn. The

cognizer comes to know the facts of the situation.11

11The careful reader will have noticed a substantial overlap between the ascertain frame and the cogni-tion inchoative frame, not only in the formal notation, but also in the list of words that use the two frames.Both involve a cognizer and a content and inchoative cognition. Nevertheless, we need to distinguishascertain because it is volitional, involving the seer's intention to learn something. There is also a complexrelation between the possible syntactic realizations of the facts, factivity, and various readings of indirectquestions, but these also involve other verbs of cognition and communication, and would take us too fara�eld.


Frame see determine

type event

cat V

lexm see

Uses ascertain [asc]

roles:

264Seer = asc.Cognizer

Seen = asc.Facts

375

val:8><>:

264role Seen

pt IndQn _ DNI _ NP

3759>=>;

The seer intends to learn seen as the result of a purposive action. Although

this sense no doubt originated in situations where the seer �nds out something by eyeing

something, the visual modality is not required now, as in Let's see how the piano sounds

after it's tuned.

Sense ENSURE

intend cause

intentional causation

State 1 Event 3

see ensure

Frame cause

type event

roles:

264CauseE�ectsoa

375


Frame intend

type state

roles:

264IntendersentientIntention

375

Frame intentional causation

backgrounds 2

entails 2

type event

roles:

266664Intender

Action = 2

E�ect

377775

scenes:

26666666666666666666666666666666664

2666666664

state 1

Inherits intend [int]

roles:

264int.Intender = Intender

int.Intention = 2 and 3

375

3777777775

2664event 2

roles:

�Agent = Intender

�3775

2666666664

event 3

Inherits cause [c]

roles:

264c.Cause = 2

c.E�ect = E�ect

375

3777777775

37777777777777777777777777777777775


Frame see ensure

type event

cat V

lexm see

Uses intentional causation [ic]

roles:

264Seerhuman = ic.Intender

Seen = ic.E�ect

375

val:8>>>>><>>>>>:

266664role Seen

gf comp

pt S�n

377775

9>>>>>=>>>>>;

8><>:

264role Seen

pt PP[to]

3759>=>;

(40) a. BNC: Sarah will see to the hens and the milking.

b. I'll see that the house is properly closed up for the winter.

c. See that he gets home before dark.

d. *See that the sky is blue.

e. I want you to see clearly that the sky is blue.

f. ?I want you to see clearly that he gets home before dark.

The seer is a human being, the seen is a state of a�airs; seer makes sure that seen

obtains, either by doing what is necessary to bring it about or by causing others to do so.

Some uses of the form see to NP are good candidates for explanation in terms of

metonymy along the lines of Pustejovsky's (1995) qualia; e.g. in Ex. (40-a), our knowledge

of hens enables us to understand the hens as metonymic for the actions associated with

caring for hens, especially in the context of the milking . (However, it would be di�cult to

characterize the feeding, watering, etc. as either a Telic or Agentive quale of hens; the

concept of qualia works better where the NP refers to an artifact, in which case we can

felicitously speak of its Telic quale.)


The form see [that ] S with the ensure sense (as in Ex. (40-b)) is distinguishable

from recognize mainly on pragmatic/discourse level evidence. There are, however, some

syntactic di�erences, as well; ensure can be an imperative (Ex. (40-c)), while recognize

cannot (Ex. (40-d)), likewise, certain adverbs of manner are appropriate with recognize

(Ex. (40-e)) but not with ensure (Ex. (40-f)). All of these follow naturally from the

agentiveness of intentional causation.

The variety of syntactic realizations for see ensure might suggest that we will

need several constructions linked to this sense, but I believe that the single syntactically

underspeci�ed construction given here will cover most uses.12

(41) a. A: The cow needs milking.

b. B: I'll see to it.

c. I'll see to it that the cow gets milked.

d. I'll see to the milking of the cow.

e. I'll see to the cow.

Sense ENVISION

imagine

see envision

Frame imagine

type event

roles:

264Cognizerhuman

Contentthing _ soa

375

restr:

�irrealis(Content)

�

This frame includes both \forming a mental image of", related to the use of see,

hear , smell , taste and feel with irrealis objects, as in Ex. (42) (a-e), and more abstract

12We will have to list separately the collocation see to it S�n, which uses this sense, because of the uniquesyntax and the speci�city of the lexical items involved. Some interesting things happen when part of thispattern is omitted; in Ex. (41), given the two conversational turns in (a) and (b), speaker B could intendEx. (41-c), Ex. (41-d), or Ex. (41-e). In Ex. (41-e), the it functions like an \expletive" (It is important that. . . ); in Ex. (41-d), its antecedent is the VPing, and in Ex. (41-e), its antecedent is the cow , which in turnis metonymic for the actions needed to take care of the cow. All three can lead to the proper understandingof the reply in the given context.


kinds of cognition (Ex. (42)).

(42) a. I can just see my great-grandmother now, crossing the prairie in a wagon train.

b. I can hear Ronald Reagan now, saying, \There you go again."

c. I can smell her perfume now, just as if she were here.

d. I can taste my grandmother's biscuits now.

e. I can feel that �rst kiss even today.

f. Can you imagine a world in which the only armed forces would be those of a

world government?

Frame see envision

type event

cat V

lexm see

Uses imagine [img]

roles:

264Seer = img.Cognizer

Seen = img.Content

375

(43) a. Brown: Wagner replied, \Can't you just see the headline: `City Hooked for

$172,000'?"

b. BNC: \I can just see her dripping with tears of anxiety."

c. Whenever I read the poem, I can hear Dylan Thomas speaking.

d. Brown: He was again tingling with pleasure, seeing himself clearly in Slater's

shoes.

This sense often co-occurs with can/could just (Ex. (43-a), Ex. (43-b)), in [one's] mind's

eye, etc. The seen can be a state of a�airs or just a simple noun. The state of a�airs is

\envisioned" in an irrealis mental space.

There is a close parallel with some other verbs of vision, e.g. glimpse the future,

and there is even a related verb that speci�cally means \envision the future", foresee; there

is also a similar sense of hear (Ex. (43-c)). The other sense modalities are less commonly


used to come to know states of a�airs, but one can �nd a few examples such as get a taste

of the future. Ex. (43-d) is a collocation in its own right which uses this sense.

This sense is also similar to at least some uses of what Brugman (1996) calls the

depictive sense of have, cf. Ex. (44-a) vs. Ex. (44-b).

(44) a. Mary has herself running for Congress next year.

b. Mary sees herself running for Congress next year.

Sense FACULTY

perception basic

perception faculty

see faculty

Frame perception faculty

type event

Uses perception basic [pb]

roles:

266664Perceiver = pb.Perceiver

Phenomenon = pb.Phenomenon

Body part = pb.Body part

377775

val:8><>:

264role Phenomenon

pt INI

3759>=>;

The perception faculty frame denotes the ability to perceive by a given modal-

ity, rather than the perception of something in particular. Pragmatically, this frame is only

invoked when there is some question about or change in one's general ability to see (eye),

hear, smell, etc., but it is di�cult to represent this fact in terms of frames.


Frame see faculty

type event

cat V

lexm see

Uses perception faculty [pf]

roles:

266664Seer = pf.Perceiver

Seen = pf.Phenomenon

Body parteyes = pf.Body part

377775

Often occurring with can/could , this sense usually pro�les some sort of di�culty or

limitation of vision. In traditional syntax, faculty would be considered intransitive, since

no object is expressed in the sentence. Semantically, we can consider the seen to be àny-

thing' or `physical objects in general'; this is represented in the frame perception faculty

by the requirement that the seen be inde�nite null instantiated (INI).

This sense often occurs with adjunct PPs expressing either a path as an explanation

for a failure of vision (I couldn't see through the dense jungle), or a goal as a description of

the range of vision (From the dormer window, you can just see to the bay). It is not clear

whether that should be treated as a separate sense or a role goal should be added to this

sense to allow for the \metaphorical motion" uses (cf. Section 3.3).

Sense NEWS


see eye see recognize

see news

Event


Frame see news

profiles Seen

type event

cat V

lexm see

Uses see recognize [rec]

roles:

264Seer = eye.Seer = rec.Seer

Seen = rec.Seen = eye.Seen.Content

375

scenes:

26666664

2666664

event

Uses see eye [eye]

roles:

�Seenvisual medium

�

3777775

37777775

val:8><>:

264role eye.Seen

pt DNI _ PP

3759>=>;

(45) a. I see that the governor has �nally decided to support the university.

b. I hear that the President is in town.

c. I can hear (by/from your voice) that you have a cold.

Content is not a role in see eye, but because of the semantic restriction to visual medium,

we can be sure that it will have a content to which we can refer here13. When the medium

is expressed, it selects for speci�c prepositions:

(46) I saw in the Chronicle/on the ten o'clock news/?from the newspaper/?by the news-

paper that Congress has adjourned.

This sense can be considered an extension of the \ordinary" visual evidential use,

which we have called recognize, and Alm-Arvius (1993:122-128) treats both read and

news as \pragmatic expansions" of eye. However, because so much of our knowledge of

13Cf. Pustejovsky's (1995) treatment of the qualia for books.


the world depends upon the news media, this use seems to be so conventionalized that it

will be treated here as a separate sense. Unless one knows the full context, the distinction

is sometimes hard to make, e.g. Ex. (45-a) could be said on the basis of a tour of the

campus, noticing improvements in the buildings (recognize), or on the basis of reading a

newspaper (news).

There is a corresponding expression with hear which takes the same types of

complements (Ex. (45-b)). The uses with see are more or less limited to news acquired

from the media proper, while those with hear may refer ambiguously to news broadcasts,

gossip, story-telling, etc. Note also that, unless the context makes it clear, the pattern see

that S is ambiguous between news and recognize; such ambiguity is less common for hear

that S, but possible (Ex. (45-c)).

Both read and news incorporate both the idea of learning of a state of a�airs

and also a speci�cation of the sense modality.

Sense PROCESS

Frame bare event

type event

roles:

264Parteventevent

375

val:8>>>>><>>>>>:

266664role Part

gf ext

pt NP[acc]

377775,264role Event

pt VPing _ VPbrst

375

9>>>>>=>>>>>;

8><>:

264role Event

pt NP[deverbal?]

3759>=>;

In bare event, the Event is predicated of the Participant, without the de�nite

time of a �nite verb. Transitive events, with two participants, are still relatively simple

when the event is expressed with VPing or VPbrst, as the patient is in the usual position

(Ex. (47-a), Ex. (47-b)).

With NPs, there are more possibilities; we can have (1) the same order as with ac-


tive VPs, the agent genitive preceding the N and the patient following in PP[of] (Ex. (47-c)),

or (2) the patient preceding and the agent optionally following in a PP[by], on the model

of the passive VP (Ex. (47-d)). Ex. (47-e) is therefore ambiguous between active (army as

agent) and passive (army as patient) readings. When the event is expressed as an NP, the

Participant is not a constituent of the frame directly, but may be either a constituent of

the NP or null instantiated, along with other participants, if any (e.g. Ex. (47-f)). A full

treatment of the syntax and semantics of event NPs is beyond the scope of this discussion,

but see Fillmore 1994 for an outline of the complexities.

(47) a. me upsetting Mary

b. me upset Mary

c. the army's destruction of the city

d. the city's destruction (by the army)

e. the army's destruction

f. his gift of $50 to the Red Cross

Frame see process

type event

cat V

lexm see

roles:

264SeerSeen = 1

375

scenes:

264264 event 1

Uses bare event [be]

375375

In the condition sense, nearly all of the semantics is inherited from the more

general frame bare condition; in process, the same is true for the bare event frame.

The event is a process of which the participant is a patient or an action of which the

participant is the agent.

The VP can be headed by either a present participle or a bare verb stem (naked

in�nitive); there are semantic di�erences between these two patterns, but they do not seem

to represent a di�erent sense of see. Rather, these di�erences follow from the di�erence


in aspect expressed by the subordinate verb. This is probably best explained by Lan-

gacker (1991); in discussing the examples We saw the ship fsink/sinkingg, I heard him

fcall/callingg, and She felt the earth fshake/shakingg, he says:

. . .An episode of direct, physical perception has a limited duration that can bethought of as a temporal viewing frame; if based on perception alone, apprehen-sion is restricted to that portion of an event which temporally coincides withthe frame (cf. Vol. I, p 193).

It is generally recognized the the zero form (as in We saw the ship sink) indi-cates that the entire subordinate event is perceived, whereas -ing (We saw theship sinking) conveys that only part of it is. The notion of a viewing frameallows us to describe this contrast with reference to independently establishedvalues of -ing and zero. We can attribute to -ing precisely the same value thatit has in the progressive construction (Fig. 5.5) and in certain adverbial clauses(Fig. 10.1(b)): on the processes it imposes an immediate scope of predicationcomprising a representative series of internal states; the pro�le is necessarilycon�ned to these states, which it construes holistically and as being e�ectivelyhomogeneous. . . . the perceptual verb's viewing frame is identi�ed as beingresponsible for the restricted, \internal perspective" imposed by -ing on thesubordinate process.

The contrast between zero and -ing resides in the relationship between the im-mediate scope they impose and the overall pro�le of the subordinate verb: inthe case of -ing , the immediate scope falls within the boundaries of of the verbstem's processual pro�le, whereas with zero the immediate scope coincides withthose boundaries. (p. 442-3)

Semantically, the seer comes to know about the state or action of the entity repre-

sented by the NP. The syntax and semantics of this sense are very close to those of depictive

predication (Rothstein 1983; Brugman 1996), but here there is also a strong presumption

that the state holds in reality. The process sense is distinguished from recognize in

not completely encapsulating the event as a proposition, even though the content of the

knowing can be expressed as a proposition. For example, Ex. (48-a) foregrounds the propo-

sition, emphsizing that it it a fact in the speaker's opinion; Ex. (48-b), on the other hand,

presupposes the running event and gives us the point of view of the seer, foregrounding

Frances , even though the seer presumably does become aware of the running event as a

result of seeing.

(48) a. recognize: He saw that Frances was running up the hill.


b. process: He saw Frances running up the hill.

Sentences such as The only time I saw Monk play, he was stinking drunk and Seeing Hideo

Nomo pitch a no-hitter was the thrill of a lifetime, could be considered either as examples

of spectate if we think of the event perfectively as a single performance, or as condition

and process with the NP+bare-VP pattern.

Sense READ

see eye

see read

event

Frame see read

profiles Seen

type event

cat V

lexm see

roles:

264Seer = eye.Seer

Seen = eye.Seen.Content

375

scenes:

26666664

2666664

event

Uses see eye [eye]

roles:

�Seentext

�

3777775

37777775

restr:

�short(Seen)

�

The seen is text, short enough that it can be read in a few minutes; seer reads

seen.

There is an interesting interplay between see and read in regard to long and short

printed matter: I saw her latest novel typically means that I haven't read it (yet), but I saw

your column in the paper or I saw your memo usually means that I have read it. In the

case of the novel, see implies `not read' by scalar implicature (since the stronger term read


is available), but apparently for very short texts, seeing is tantamount to reading|they are

treated as being at the same point on the scale. Thus, in answer to the question \Did you

read the paper this morning?" one could say \No, I only saw the headlines", since reading

the headlines is not equivalent to reading the entire paper. One could not, however, intend

by this that one had looked at the headlines without reading them.

read is very similar to see news but we cannot assume that the content is a

proposition in this case. We may need to use some abstract frame for understanding to

express the result of the reading, but cognition inchoative is not suitable, for the same

reason.

Sense SETTING

container

see setting

Frame contain

type event

roles:

2666666666664

Container

Contents

Interior

Exterior

Boundary

3777777777775


Frame see setting

type state

cat V

lexm see

Uses contain [cnt]

roles:

264Seer = cnt.Container

Seenevent = cnt.Contents

375

(49) a. BNC: He believed that Paris was seeing a forward surge in the arts.

b. The 1940s saw the beginning of the Cold War.

c. Brown: The years 1812 and 1813 saw him in Germany and France again, but

on this visit to Berlin he did not seek out the philosophers as he had on his

�rst journey.

d. The Cambrian saw an explosion of speciation.

e. Brown: . . . the Italian painters, by universal consent, were the most brilliant

group of geniuses any art has seen.

The seer is a place or a period of time, and the seen is an event or combination of

events. The seen occurs within the seer, so that the seer contains the seen or is

the \setting" for it. Note that Ex. (49-b) and Ex. (49-c) are not simply metonymic for

\people living during the 1940s" or \people living in Europe in 1812", as shown by the

acceptability of Ex. (49-d). Such a metonymy may well have been the origin of this sense,

but it is not necessary synchronically. Ex. (49-e) seems to represent a variant, in which

the \container" is neither a place nor a time, but something that might be characterized

as a professional community; it may thus be closer to a true metonymy, with the name

of the profession standing for all the members, in this case, all artists of all times. Note

also that it is desirable to postulate several other roles in the contain frame, as these are

needed for other metaphorical containers, as in There isn't room in this town for both of

us. (interior) and He's on the verge of a nervous breakdown (boundary), but only the

container and contents are used by see setting.


Sentences with the sense setting can often be paraphrased with the word wit-

ness (The 1940s witnessed the beginning of the Cold War). Unfortunately, as previously

mentioned, the problem of polysemy is rampant|the word witness itself can have senses

similar to those we have labeled spectate, process, etc.

Sense VISIT

perception basic

recip event perception distant

social interaction see eye

see visit

Frame recip event

type event

roles:

264Part1 = e1.ProtoAgent = e2.ProtoPatient

Part2 = e2.ProtoAgent = e1.ProtoPatient

375

scenes:

2666666666666666664

2666664

event [e1]

roles:

264ProtoAgentProtoPatient

375

3777775

2666664

event [e2]

roles:

264ProtoAgentProtoPatient

375

3777775

3777777777777777775

restr:

264e1.type = e2.type

375


Frame social interaction

type event

Uses recip event [re]

roles:

264Part1human = re.Part1

Part2human = re.Part2

375

This frame represents a very basic concept, what people do when they see each

other and either know each other or want to communicate something. Typically involves

face-to-face conversation, but neither seeing each other nor use of language is necessary,

although one or the other must occur. Who is designated Part 1 and Part 2 is arbitrary,

since it's reciprocal; this does not mean that they are social equals or friends, only that an

unsuccessful attempt by one party does not constitute interaction.

Frame see visit

type event

cat V

lexm see

Uses see eye [eye]

Uses social interaction [si]

roles:

264Seer = eye.Seer = si.Part1

Seen = eye.Seen = si.Part2

375

restr:

�familiar(Seer,Seen)

�

Note that in this frame, we have multiple inheritance, from both see eye and

social interaction; this causes no problems, but only makes the semantics more speci�c.

seer and seen are both conscious beings physically present in same place and time, who see

(eye) each other; some minimal amount of social interaction takes place (not necessarily an

intentional formal visit), which is usually friendly, and usually involves some conversation.

A clear example of a common use is Nice to see you! visit includes eye, and implies that

the person is correctly recognized.

The social interaction is clearly primary, but eyeing seems to be required as well,


at least for participants with normal vision. The last constraint expresses the fact that one is

more or less required to use meet for the �rst social meeting with an unfamiliar social equal,

although this constraint may be defeasible for some speakers(Ex. (50-a)). Most apparent

counter-examples are probably consult (Ex. (50-b)) or audience.

(50) a. ?I'm going to see our new neighbor for the �rst time this afternoon.

b. I'm going to see my new dentist for the �rst time this afternoon.

Note that blind people use this sense quite comfortably; this seems to con ict with the

description of this sense as using eye, as shown in the frame above. What is intended is

that for normally sighted people, the sense visit requires that physical vision take place|

such people cannot use this sense, for example, in describing a telephone call, however long

or friendly it may be. It will be interesting to see what happens to this use if television

telephones become common.

Semi-collocations

Here we consider senses that are intermediate between those listed above a Basic

Senses, which do not require any particular lexical items to co-occur with them (although

they may have quite speci�c syntactic requirements), and true collocations, discussed in

Section 2.6, which do impose requirements on lexical items around them. What we here

call \semi-collocations" require some lexical item(s) to co-occur, often drawn from a small

set of semantically related words, usually in one of several syntactic patterns.

Sense CLASSIFY

categorization

see classify


Frame categorization

type event

roles:

266664Cognizerhuman

Item

Category

377775

Frame see classify

type event

Uses categorization [cat]

roles:

266664Seer = cat.Cognizer

Seen = cat.Item

Category = cat.Category

377775

val:8>>>>><>>>>>:

264role Seer

syn DA

375,

266664role Seen

syn nuc

pt NP

377775,264role Category

pt PP[as]

375

9>>>>>=>>>>>;

8><>:

264role Seer

syn subj

375,264role Category

syn obj

375,264role Seen

pt PP[in]

3759>=>;

This frame is still not complete, since we can have an AP after the as (see this

one as important).

(51) a. We see line A as shorter than B, even though they are really the same length.

b. recognize She saw that Sandy was brilliant.

c. classify She saw Sandy as brilliant.

d. I only saw her as an old woman, so it was hard for me to imagine her as a

young heart-breaker.

seer is a conscious being, seen a stimulus, and category is a state predicated of seen,


usually expressed as an NP or an AP. seer comes to regard seen as an instance of cate-

gory, a fact which is not immediately obvious.

classify can be considered as related to recognize and condition. The choice

of classify rather than either of these creates a scalar implicature that the seer's belief

is at least a matter of opinion and possibly not true, as in Ex. (51-c) vs. the much stronger

Ex. (51-b). This sense is semantically very similar to regard X as Y, or view X as Y, but

the syntactic patterns are di�erent for each word (cf. Ex. (10) on page 50). Some of the

same forms can also be used to express the e�ects of an optical illusion (Ex. (51-a)).

There are also sentences of the pattern see X as Y which mean something like àt

the time when he/she/it was/will be' (e.g. Ex. (51-d)). Distinguishing these cases, which

should probably be classi�ed as condition, from occurrences of classify will almost

certainly require inferencing based on knowledge of the world. For more discussion of the

valence alternations, see Levin (1993:181-2) on Appoint Verbs, including consider (which

hardly belongs there semantically), and Characterize Verbs, including regard and see.

Sense EXPERIENCE

service as

see experience


Frame service as

type event

roles:

266664Undergoerphys obj

Service

Role

377775

val:8>>>>><>>>>>:

lexm use _ service _ action _ combat _ duty264role Service

pt NP

375,264role Role

pt PP[as] _ DNI

375

9>>>>>=>>>>>;

8>>>>>>>><>>>>>>>>:

lexm use _ service _ action _ combat _ duty

264role Service

pt NP

375,

266664role Role

gf mod

pt AjP _ N

377775

9>>>>>>>>=>>>>>>>>;

This frame appears in \undergoer BE in service (as role)", \put undergoer

in service (as role)", and \undergoer's service (as role)", as well as experience.

There is clearly a further generalization be made to include the corresponding verbs (use

undergoer as role, and undergoer act/serve as role), but it is omitted here, as not

directly relevant to see.

DNI for role is very common, and adjuncts of place and time frequently provide

the context allowing the inference required for this omission, e.g. The aircraft was in action

in the Gulf War i.e., ying combat missions, as a combat plane. Each of the lexemes

for the service role is subtly di�erent. The word combat itself partially or completely

incorporates the role, and is equivalent to saying \service as a combatant". The second

valence set covers expressions such as military service, civilian use, and sea duty .


Frame see experience

type event

Uses service as [svc]

roles:

266664Seer = svc.Undergoer

Seen = svc.Service

Role = svc.Role

377775

val:8><>:

264role Seer

gf ext

375,264role Service

gf obj

3759>=>;

(52) a. Brown: Friend is o� to a great start with a 4-0 record but isn't likely to see

action here this week.

b. BNC: He had seen service in the Boer War and the Boxer Rebellion. . .

c. BNC: The FR.4 was too late to see service in World War Two. . .

d. BNC: In addition, there is a single HST Mk 3 saloon, which. . . has seen use by

both the Queen and the Prime Minister.

e. BNC: Charlotte, who has seen more than 20 prime ministers come and go in

Britain since she was born in 1877, received her 15th telegram from the Queen.

f. The company has seen pro�ts fall even while revenues rose.

seer undergoes seen and is a�ected by it.

Some of these examples are based on a sort of collocation with the small set of

nouns shown in the service as frame (service, combat , duty , use, action, and possibly

others), e.g. Ex. (52-a), Ex. (52-b). In this pattern, seer need not be a person, but still

\undergoes" the experience and is a�ected by it (Ex. (52-c), Ex. (52-d)); in this respect it

is distinct from setting, in which seer is a place or time which is not a�ected per se by

the events. There is also a more open pattern with events generally, as in Ex. (52-e) and

Ex. (52-f), which verges on process, except that the seer is a�ected by the event, not

merely a passive witness. Another construction closely related to experience is Have seen

better/worse times/days/years/etc., which has a restricted type of NP in the direct object


role.

Sense GAMBLING

gambling

gambling cards

see gambling

Frame gambling

type event

roles:

2666666664

gamblerhuman

opponenthuman

outcomeproposition

stakesmoney

3777777775


Frame gambling cards

type event

Uses gambling [gam]

roles:

266664gambler = gam.gambler

opponent = gam.opponent

outcome = 1 = gam.outcome

377775

scenes:

266666666666666666666666666666666664

�event draw cards

�2666664

event bet money1

roles:

264bettor = gambler

bet

375

3777775

2666664

event bet money2

roles:

264bettor = opponent

bet

375

3777775

2666664

event compare 1

roles:

264hand1playing cards

hand2playing cards

375

3777775

377777777777777777777777777777777775

Frame see gambling

type event

Uses gambling cards [gc]

roles:

264seer = gc.gambler

seen = gc.bet money2.bet

375

(53) a. I'll see your twenty and raise you ten.

b. He said he'd see me; his face fell when I turned up the kings.

c. BNC: \I'll see your `buggered' and raise you a `shagged out'."


d. SJM91: I'll see your billion and raise you two. What fun!

Most dictionaries list this sense, but unfortunately, perhaps because it is mainly

used in conversation, I have found only two corpus examples, one from the BNC Ex. (53-c)

and one from the LDC corpus of the San Jose, California Mercury newspaper Ex. (53-d).

Although these examples are very colorful, neither is very typical; more typical examples are

Ex. (53-a) and Ex. (53-b). We thus must depend on personal communication with several

speakers who have this sense in their idiolects. The direct object is usually the amount

of the bet being matched. Since the expression is apparently used mainly in �rst person

(except in reported speech), and the speaker usually accompanies the utterance with the

action of placing the matching bet on the table; this use is, in fact a performative, even

though the verb is usually in future tense, since the simultaneous speaking and placing of

a bet constitute the act in question. Given the limited semantic range of possible subjects

and objects, this use also verges on being a collocation, but the object is not limited to a

particular set of words.

Sense VIDE

see eye

see vide

Frame see vide

type event

Uses see eye [eye]

roles:

264seer = eye.seer = the reader

seenvisual medium = eye.seen

375

syn:

�mood imper

�

val:8>>>>><>>>>>:

266664role seer

gf ext

pt DNI

377775

9>>>>>=>>>>>;


This is in part simply compositional; it can largely be understood as inheriting

eye, and adding the restrictions that the seen is text or graphical material (hence not

inheriting read), and the verb must appear in the imperative. But there is also typically

a compressed, \telegraphic" style in the clause (e.g. See chart, p. 37.) and apparently a

rather limited set of nouns that can express the seen: page, section, chapter , appendix ,

chart , table, diagram, etc. These facts, taken together, suggest that it should be regarded

as a sense in its own right.

Compositional Uses

Sense AUDIENCE

(54) a. The doctor will see you now.

b. The King will see the Ambassador tomorrow.

c. Mrs. Riggs-Aston will see no-one today.

This is visit with the further restriction that the seer is of higher social rank

than the seen, meaning something like `grant an interview to'.14 The di�erence in social

rank may be con�ned to the particular situation in question; e.g. a millionaire or a state

governor may have to wait until the doctor is ready to see him or her. The syntax of seer

and seen are the inverse of those of consult, yet something more formal than a simple

social interaction (i.e., visit) is usually intended.

Stereotypically a set phrase used by receptionists (Ex. (54-a)), it also occurs in

more varied syntax and social settings (Ex. (54-b), (54-c)). In the last case, there is not

necessarily a great di�erence in social rank between the hostess and her guests, but the

situation is so formal, that the decision not to see anyone for a day is clearly the hostess'

prerogative.

14This is one of the places where the use of common words as names of senses may unfortunately cause someconfusion. The sense visit is by de�nition reciprocal, since it uses (indirectly) reciprocal event. Therefore,it is correct to say that audience is just a further restriction on the social ranks of the participants. Butthe ordinary verb visit has a speci�c directionality, which happens to run counter to the facts for the senseaudience. Mrs. Riggs-Aston stays at home and people come to visit her, she does not go to visit them, inthe ordinary sense of the word visit . But when they come to her house, they see (sense visit) each other,and because it is such an unequal relationship, we can say that she sees (sense audience) them; they donot see (sense audience) her.


Sense DISCOURSE


see discourse

Frame see discourse

type event

Uses cognition inchoative [ci]

roles:

266664seerhuman = ci.Cognizer

seen = ci.Content

discourse context

377775

val:8>>>>><>>>>>:

266664role seer

syn DA

person 2

377775

9>>>>>=>>>>>;

8>>>>><>>>>>:

266664role seer

syn DA

person 1

377775

9>>>>>=>>>>>;

(55) a. Brown: As we have seen, Methodism early took a stand against slavery.

b. Brown: Second, we will see how Sidney answered the charges, . . . .

c. As seen in Chapter 3, the war did not unfold as the General Sta� had planned.

d. You'll see that this is necessary when I come back to this point in a few minutes.

e. As the reader will see in the following section, . . .

The seer is �rst or second person, the seen is a situation, and there is a reference,

either direct or indirect to the (written or spoken) discourse context. This may be marked

simply on the verb, using past or perfect for discourse up to the present, or future for

following parts of the discourse, (Ex. (55-a), Ex. (55-b)) or it may be more explicit, as in

Ex. (55-c) and Ex. (55-d).

A wide range of registers is possible; in casual conversation, I see is a complete


turn, with both seen and seen de�nite null instantiated (DNI), i.e. completely recoverable

from context. In formal writing, the \editorial we" is common Ex. (55-b); in very formal

writing, the subject of see can sometimes be third person, although still standing for the

addressee: (Ex. (55-e)).

Sense HALLUCINATE

(56) a. BNC: Drinkers at a new pub may see more than pink elephants hovering over-

head if they have one too many this weekend.

b. Brown: He had been \seeing things".

c. He has been seeing things that upset him.

d. BNC: \Does that mean they'll be seeing green monkeys and blue snakes up

the rigging?"

e. BNC: He would be seeing shadows around every corner next.

f. The exhausted soldiers saw enemy troops everywhere.

This sense is unique in presupposing that an image is experienced, but implying that there

is no physical stimulus of the usual sort which could serve as the cause of the image. It is

syntactically very similar to eye, but is distinguished from it by virtue of the fact that the

seen is irrealis. Thus, we can use the same frame as for eye, except that the semantic type

of the seen will be irrealis, rather than phys obj _ phys motion, or add this restriction as

a predicate, so that the seen is an unreal physical object or motion.

As noted in the introduction to this chapter, some of the NPs that serve as seen are

conventional expressions for unreal entities, such as pink elephants (Ex. (56-a)). Likewise,

the collocation BE seeing things has the idiomatic meaning hallucinate, but only when

not further modi�ed (cf. Ex. (56-c)). In other cases, conversational implicatures must be

computed (Ex. (56-d), (56-e)), and there are real possibilities for ambiguity between eye

and hallucinate, as in Ex. (56-f).


Sense SCAN

Frame see scan

type event

roles:

�Seernonsentient

�

val:8><>:

264role Seer

syn subj

3759>=>;

This is a metaphorical extension from the process that animate beings perform

with their eyes to similar processes performed by machines, computer systems, etc. The

seen may be either the immediate physical stimulus or the content, i.e. scan may imply

some sort of \cognitive" processing as well. When the seen is a physical object, the con-

struction for scan will be identical to that for eye; when the seen is a state of a�airs,

the construction would be like that for recognize, except that the semantics of the seer

would not be restricted to sentient beings.

The idea is that this can unify with other senses, mainly eye, recognize, deter-

mine, and possibly ensure, and override their type \sentient" on the seer, which would

have to be made a default, rather than absolute. The alternative is to keep the sentient

type and de�ne a metaphorical mapping (a.k.a. \type coercion" in this context) that treats

non-sentient seers as though they were sentient.

Sense SPECTATE

The seen is an event, typically a concrete event, including performances for en-

tertainment. This sense prototypically requires the spectator to be physically at the event,

seeing the participants or performers. With the advent of movies and television, spectate

no longer requires physical presence in the same place. spectate is distinguished from

process in that it pro�les the entire performance; I could say that I had seen Bill ironing

even if I had only stayed long enough for him to �nish the collar of one shirt, but to say that

I saw My Fair Lady, I have to have been watching for almost all of the show. Rather than

make a general speci�cation that spectate is always telic, we might be able to have this

follow from the Telic quale of artistic performances, that they are normally to be viewed in


their entirety. This is a special case of the more general fact that to see a bounded event

implies to see (sense process) it entirely, unless there is some evidence to the contrary,

either linguistically marked or derivable from the context.

classify might be considered to overlap with spectate in a special \theatrical"

sense (I only regret that I never saw Olivier as Prospero.) However, it is better to regard this

as a construction based on the word as alone, since this theatrical sense can also occur with

the other major sense modality hear (I heard Olga Borodina as Carmen) or without either

(Charles Laughton as Captain Bligh is unforgettable). Thus, the theatrical as construction

would have valence elements for a performer and a role, and its outer semantics would be

\performance", a subtype of \process" with perfective aspect.

Sense TOUR

Uses which involve visiting for the purpose of seeing the sights may be considered

to constitute a separate sense which we will call tour, but it is di�cult to decide whether

this is really lexicalized in English, or merely the result of the regular semantic composition

of eye with direct objects which are the sorts of places that people tend to take tours of.

(I will discuss this sense further when we look at bilingual dictionaries in Section 5.3).

2.6 Collocations

In this section, we will discuss collocations of other lexical items with see which

are lexicalized to varying degrees. More than one hundred such collocations are listed in

dictionaries of English idioms (e.g. the entries headed by see in Urdang & Abate 1983);

we will deal here only with a representative sample. All of these are \encoding idioms"

(Makkai 1966), in that, given a knowledge of the rest of the language, it is not predictable

that the particular meaning would be expressed in this particular way.

From a \decoding" point of view, many of the idioms with see are relatively easy

to understand on �rst hearing, as the sense of see involved is clearly one of those already

discussed, and the meaning of the constituents produces the meaning of the collocation by

normal rules of composition; examples are:

(57) a. can't see one's hand in front of one's face

b. see the sights


c. see at a glance

In others (which Fillmore (1996) calls decoding idioms), the collocation as a whole has

a meaning that cannot be produced by the composition of the parts, although the internal

syntax is quite ordinary, as in Ex. (58).

(58) a. see a man about a dog

b. see the color of someone's money

Many, probably most, collocations are intermediate cases, in which the semantics

of the collocation is only partly compositional, so that the remainder must be regarded as

attached to the collocation as a whole, as in Ex. (59).

(59) a. so angry (drunk, etc.) one can't see straight

b. can't see the forest for the trees

c. can't see beyond one's nose

Examples of Collocations

Let us examine a few more relatively common collocations in more detail; although

of course not exhaustive, the following list is intended to be at least representative of the

range of collocations using see. I will not attempt to give full frame de�nitions for each

of these, since they are for the most part elaborations of senses already discussed above,

with some of the arguments �xed as speci�c lexical items and some additional meaning

attributable to the construction as a whole. I will, however, give examples of each.

let's see

(60) a. Let's see, the jewelry will be unguarded for at least �ve minutes. . .

b. Let's see, there should be another box of staples here somewhere. . .

c. BNC: `Let's see, have I got this right?

d. Brown: \Let's see," Cousin Ada said.

This means `to consider carefully'. The lexical form is �xed, with no variation. This is a

\pure" pragmatic expression; no vision is involved, and no other person need be present; it

may be an extension of determine.


see one's way clear

(61) a. I hope you will see your way clear to give it to her.

To come to the conclusion that one can (or should) do something. This is a very �xed

metaphorical extension from the experience of walking on a path and seeing the path ahead.

see fit

(62) a. Pat had it for a week before seeing �t to tell me about it.

b. You can handle the situation however you see �t.

c. They never saw �t to visit us until we moved to Hawai'i.

d. You can sell it or hang on to it as you see �t.

e. Brown: \I should . . . be honored and grati�ed should the Democrats see �t to

nominate me."

The seer is a conscious being, seen a to-in�nitive. In the basic syntactic pattern (seer

sees �t to B), the subject of the in�nitive phrase is the subject of see, so we can regard this

sense as a verb of the type exempli�ed by endeavor , fail , refuse, and dare, which undergo

what has been called Subject-Equi-NP-deletion, but do not allow the alternative in which

the embedded verb has a di�erent subject marked with for , as shown below:

(63)

Ms. Ding hated [for Ms. Ding] to go out with Igor.for Natasha

saw �t [for Ms. Ding]�for Natasha

The seen must involve a process, not a pure stative (�They saw �t to live in Spain

for �ve years). There is another common syntactic pattern with a free relative, (They can

exercise their stock option whenever they see �t), which blocks the extraction.

see fit may be related to classify (i.e. see X as �tting) or condition (i.e. see

X �tting , where �tting functions as an AP). see fit is often used ironically, implying that

someone's actions are grudging and slowed by sel�shness. Heather Jones (p.c.) has noted

that there is a similar expression in Welsh, equivalent to `please', that means literally ìf

you see well'.


see off

(64) a. I have to go to the airport tomorrow to see my brother o�.

b. Sorry I won't be able to see you o� to France, but bon voyage, anyway.

c. We went to the airport to see them o� on their trip to Hawai'i.

d. The entire family came to the station to see o� Francesca and her sister.

e. Brown: After a dinner party for which she had come down to New York, Mrs.

Lewis and Casanova arrived to see them o�, . . .

To be present at the time of leave-taking on a journey in order to wish someone a good

trip. seer and seen are both human. This sense is often followed by adjuncts giving the

location of the leave-taking and/or the path of the journey.

see reason

(65) I hope they'll see reason before things get ugly.

This means to come to what the speaker regards as a reasonable position, i.e., his or her

own; variations include the (cool) light of reason, etc.

see through x

(66) a. She soon saw through his bluster and realized that he was actually afraid of

what she might know about him.

b. The very rich are targets for all sorts of con games and often become skilled

at seeing through them.

c. I didn't see through the con man in time.

The seer comes to recognize the reality underlying some false impression which has been

deliberately created. The seer is a human, the seen is almost always null instantiated,

and the object of the PPthrough is some kind of deception or, metonymically, a person who

deceives, as in Ex. (66-c). This use is only distantly related to see x through; there, the

metaphorical motion is through some process, extended in time. Here, the gaze itself is

what moves, from the seer through the deception, to the reality that lies \behind" it. The

entire sense is a metaphorical mapping from the physical domain (She couldn't see through


the trees) to an abstract, interpersonal domain (She couldn't see through the scheme).15

see x through

(67) a. We had enough wood to see us through the winter.

b. Do you really have the courage to see such an audacious plan through?

c. Friends will often see you through (hard times).

d. Our country will see the war through to victory.

e. Brown: Theresa had seen him through the right college, into the right frater-

nity, and . . . safely married to the right sort . . .

This collocation means to carry out a task until it is complete, or to help someone until the

end. It is based on the life is a journey (Lako� 1987:439) metaphor, which implies that

di�culties in life are obstacles on the path, and overcoming di�culties is passing through

the areas of obstacles. The second NP (the object of through) can be omitted in cases of

de�nite null instantiation. In one variant, the seer is either a consumable resource (often

money, but not time) or a person who provides such a resource, the seen(x) is a human,

and the NP object of through is some sort of di�culty, which may be DNI in appropriate

contexts (Ex. (67-c)). In another variant, the seer is a person, the seen(x) is a task,

and the NP is either an expression meaning `the end' or `success' (Ex. (67-d)) or a DNI

with the same meaning. Ex. (67-e) is an interesting example of conjoining senses without

creating the feeling of zeugma, or perhaps better said, the skillful literary use of zeugma;

we have see x through followed by what are probably best considered two examples of

condition.

y'see

(68) a. You see, there's only so much medicine can do in this case.

b. I've been there before, y'see, so I know my way around.

c. Brown: You see, she's on a diet.

d. Brown: . . . \You see, both of them, I mean the President and Je� Lawrence,

are romantics.

e. BNC: \You see," he said, \all bureaucracies are one bureaucracy."

15Cf. Section 3.3 for discussion of the notion of metaphorical motion of the gaze.


The seer must be a phonetically reduced form of you, the seen is a situation.

Also, the phrase y'see is intonationally marked as an interjection, usually with a relatively

high, level pitch if sentence initial; this is re ected in writing by setting it o� with commas,

although the reduced form of you is often not shown in writing, as in Ex. (68-d).

2.7 Some Recalcitrant Data

In the exposition of the senses and uses above, I have naturally tried to select clear

examples from the corpora where possible. But not all of the sentences in the corpus are

easy to classify, even given the extensive list of senses we have already discussed; here are

a few problematic cases.

(69) Brown: Not until the words had been spoken did Abel suddenly see the old house

and the insistent sea, and feel his contrition blotted out in one shameful moment

of covetousness.

This sentence is ambiguous between eye and envision, because it is out of context;

if we were to see even few sentences of context, the ambiguity could undoubtedly be resolved.

No other senses are plausible, so the situation is one of simple ambiguity.

(70) Brown: This last was probably not in Brumidi's palette, but was needed to take

the chill, bluish look o� the new work next to the old, where softening e�ects of

time were seen, even after thorough cleaning.

This example represents a more subtle problem; as in many other real-world examples,

the author has blended physical perception and inference based upon it. This is a natural

re ection of cognition; we are very often conscious only of our conclusions, not of the

evidence that leads up to them. Very likely the �rst impression of the people looking at the

paintings was simply that the new work and the old were clearly distinguishable by sight;

the only part that is directly perceptual is that the new work is bluer than the old. The

\chill" of the new work and the \softness" of the old work (and the idea that the latter is

due to the passage of time) are all the results of inference.

A striking example of our unconsciousness of what goes on in visual processing and

related inference is the recognition of faces, for which humans have a specially developed


area of the brain. We recognize friends almost instantly, with no conscious processing, yet

most people have di�culty in describing even very familiar faces accurately. We simply

don't know what cues we are using to recognize our friends.

2.8 Conclusions

I hope that the reader has been convinced of the original premises of this chapter,

that see is highly polysemous and that most of the various senses of see are related to

each other in ways that are motivated, if not wholly predictable. The Frame Semantic

representation allows us to express clearly the relation of the senses to each other and to

other lexical items, especially verbs of perception and cognition.

111

Chapter 3

Other Cognitive Approaches

3.1 Introduction

Although the Frame Semantic analysis of the uses of see given in Chapter 2 allows

us to make many useful generalizations about the relations among the participants in various

situations which we talk about using the word see, there are some points which are handled

in a somewhat unsatisfactory fashion. In particular, the use of the feature irrealis as part

of the semantics of envision and hallucinate doesn't really tell us everything we need to

know about these senses, and this feature appears only with regard to these two senses|we

are left with just the stipulation that all the other senses are realis by default. Another

way of explaining the relation between perception and cognition in the event structure of

the various senses is to represent the semantics of the seen in terms of Fauconnier's (1985

[1994]) mental spaces. This approach is complementary to Frame Semantics; as Lako� &

Sweetser (1985:x-xi) point out, Idealized Cognitive Models or frames can be used to express

conceptual structure (relations among roles), while mental spaces can be used to express

referential structure (relations between roles and their �llers).1

In this chapter, I will show how a mental spaces notation can express and clarify

the relations among some of the senses de�ned in the previous chapter (only those senses

where mental spaces seems to reveal something substantive will be discussed). I will also

1I am using the term \referential structure" in a very broad sense here. The �llers may themselvesbe frames, as in Weddings always make me cry, where the whole wedding event is the cause of the tears.The �llers may also be mappings, as in His description of his job history was less than truthful , whichdescribes a relationship of \inequality" between two mappings (descriptions) from actual events in his past,an idealized (truthful) one and the one \he" gave. Mental spaces can also be used to represent these sortsof \meta-referential" mappings, but I will not discuss them here.

CHAPTER 3. OTHER COGNITIVE APPROACHES 112

discuss brie y some uses that I have lumped under the faculty sense but which have

additional complements connected with paths; these situations can be understood in terms

of metaphorical motion. Next, I will show that many of the sense distinctions can be closely

related to the semantics of their arguments, using a semantic type hierarchy which is, for

the most part, needed for independent reasons. Finally, I will discuss what happens when

two roles are to be �lled with the same referent (She saw herself in the mirror) and what

needs to be added to the frames for re exives to work properly.

3.2 A Mental Spaces Approach to the Semantics of the seen

Fauconnier (1985 [1994]:xxxvii) introduces the idea of mental spaces as follows:

The simple idea behind the approach explored in this book is that when weengage in any form of thought, typically mediated by language . . . , domainsare set up, structured and connected. The process is local. A multitude ofsuch domains|mental spaces|are constructed for any stretch of thought, andlanguage (grammar and lexicon) is a powerful means (but not the only one) ofspecifying or retrieving key aspects of this cognitive construction. Reference,inference, and, more generally, structure projection of various sorts operate byusing the connections available to link the constructed mental spaces. Tech-nically, such connections are cross-domain functions that specify counterpartsand projected structure from one space to another. In simple cases, two spacesare connected by only one function, and intuitively this function seems to re ectsome form of identity of the connected counterparts.

The theory is thus very general, and applies to the relations between many types of domains

such as:

� objects in the world (\base space") and their counterparts in paintings and movies,

� a speaker's conception of a portion of the real world2 and a counterfactual mental space

containing corresponding objects, such as that set up by the introductory clause \If I

had been born in the Middle Ages. . . ", and

2I follow Fauconnier (1985 [1994]:15) in ignoring the philosophical questions about the relation betweenthe speaker's belief space and some \external", \absolute" reality. However, to avoid suggesting that thestarting point for a mental space mapping is \external reality", I will not follow Fauconnier (1985 [1994]) inusing the term \reality space", but will use the term \base space", as in some of the papers in Fauconnier &Sweetser 1996. This also has the advantage that mental spaces and their mappings can be embedded withineach other without changing the terminology.


� a past-time space set up by a phrase such as \Before I was married. . . " (which also

contrasts with the reference time).

Using the theory of mental spaces, we can say that a verb of perception sets up two

spaces in addition to the base space (B), a perceptual space (P) and a conceptual (belief/

interpretive/ judgemental) space (C). In the simplest sort of seeing (sense eye), all three

spaces are completely congruent; each object, property, and relation in base space has a

corresponding unique counterpart in each of the other spaces. Thus, the relation between

the entities in the base space, the percepts, and the concepts in the sentence Jan sees the

cat on the mat could be represented graphically as in Fig. 3.1:

C

C’

C’’

MM’’

M’

B

P

C

Figure 3.1: Mental spaces for Jan sees the cat on the mat.

The circles represent the three mental spaces, and the lines connecting entities in

them can represent various kinds of relations between them, as mentioned above. Where

a correspondence exists between two entities, they will not only be connected by lines,

but also be given the same name, using primes to distinguish among the members of the

correspondence. Thus, in Fig. 3.1, the cat, the mat and the on relation all exist in all three

spaces. In other words, besides the real cat really on the real mat, there is (in perceptual

space) a mental image of the cat (C') and the mat (M') and the former is \on" the latter,

and the seer correctly interprets the image, thus coming to believe that there are a cat and

a mat and that the relation on holds between them, i.e. she sets up corresponding entities

(C" and M") and relations in her conceptual space. The sentence says nothing per se about

the existence of the cat and the mat in the base space, but by asserting that they exist

in Jan's perceptual space, it triggers the transfer of the structure to both Jan's conceptual

space and also (normally) base space, by the \strategic principles" discussed in Fauconnier

(1985 [1994]:Ch. 3); i.e., the base space is the \parent" of the perceptual space and unless

there is some reason (from the context of the utterance) to do otherwise, the structure of


the child space will \ oat" up to the parent space.

This may seem to be too much \multiplication of entities" for such a simple case,

and, in fact, our usual experience is that these spaces all cohere completely, forming a much

simpler gestalt than the above analysis. But as we shall soon see, we need a fairly complex

representation to account for other cases, in which there is less coherence among the spaces.

Most of the senses that require physical vision will include at least this much structure and

also add some presuppositions, such as the social interaction of visit, and the news medium

of news. (In each diagram, I will draw all circles for three spaces, even if one or two are

empty, to emphasize when there are not corresponding entities across spaces.)

Now consider the sense hallucinate, (Sandy saw pink elephants while using that

drug). Pink elephants is a conventional name for some things that do not exist in the real

world, but which people in strange states sometimes think they see. Therefore, the usual

connections between the perceptual space and base space will presumably not be made by

the hearer of the sentence. As for the conceptual space, without more context, the sentence

does not indicate whether Sandy was su�ciently compos mentis to realize that there were

not really pink elephants present. If Sandy was aware at the time that the elephants were

merely an illusion caused by the drug, then the elephants are only in the perceptual space

(cf. Fig. 3.2).

B

P

C

elephants’

Figure 3.2: hallucinate (\Undeceived")

If Sandy believed that the room was �lled with pink elephants, we have the situ-

ation shown in Fig. 3.3, with links from the perception space to Sandy's conceptual space,

but not to base space, since only Sandy is deceived, not the speaker. This di�erence, while

clear, does not rise to the level of a sense di�erence, and does not a�ect the unreality of the

pink elephants from the speaker's point of view. (I ignore for the moment the space-building

implied by the use of the past tense.)


B

P

C

elephants’

elephants’’

Figure 3.3: hallucinate (\Deceived")

envision is related to hallucinate in that there is no transfer of the structure of

the conceptual space back to base space, but it can occur either when the seer really has a

mental image or when the seer has no such mental image. For example, in Sandy sees lots

of problems , the problems may or may not be the sorts of which a visual image can exist.

If they are of the sort can be pictured \in the mind's eye" (e.g. envisioning problems in a

garden, or with the design of a house), we would have a mental space diagram like Fig. 3.3;

if the problems are more abstract, we will have structures only in the conceptual space, as

shown in Fig. 3.4.

B

P

C

problems

Figure 3.4: envision: Sandy sees lots of problems.

Now consider the sense recognize as in She saw that nothing would change his

mind about it. Here the focus or foreground is the conceptual space. The seen is abstract,

and there is no implication that the concept is based upon any kind of visual perception.

There is nothing to block the transfer of entities and relations back to base space, but there

is some question about the existence of a anything in (visual) perceptual space. Since the

conclusion is presumably based on some sort of evidence, we can set up a more general

\evidence" space (represented by a dashed circle marked \E" in the diagram), with some


vaguely de�ned entity in it that triggers the conclusion|in any case, the reason for the

belief is not given in the sentence.

B C

himmind

mind’Ehim’

evidence

Figure 3.5: recognize: She saw that nothing would change his mind.

discourse, which does not involve physical perception, would be represented like

recognize, but without setting up the perceptual space at all.

The sense which I have labeled news is an extension of recognize, which carries

an additional presupposition that the seer has learned of the situation from a visual news

medium; example sentences are I see that Congress wants to pass a tax cut , Did you see

where Microsoft is suing IBM? , and He saw that the price of his stock had fallen again.

The general case can be represented schematically as in Fig. 3.6.

BC

medium

SOASOA’

image

P

content

Figure 3.6: news

There is no presumption as to the exact type of the news medium, but it is

presumed to be visual, because the word see is used in preference to hear . This sense is

triggered when the hearer supposes that the speaker is not likely to have direct knowledge

of the state of a�airs. It is di�cult to �nd words that will cover both the case of reading

a newspaper and the case of seeing a television news program, but we could describe the

process like this: perception of the image leads to setting up some notion of the content


(with luck, the same as the intended message) in conceptual space. Then, depending on the

degree to which the seer believes the content to be true, a state of a�airs (SOA') is set up

in conceptual space (the correspondence between the message and the believed SOA' is the

dashed line inside conceptual space). The usual default mapping into base space causes us

to set up the both physical news medium and the state of a�airs depicted by it in base space.

The conceptual space is foregrounded; the visual percept is backgrounded and consequently

less speci�c. (I have symbolized this by the dashed circle around the perceptual space.)

read is in some ways similar to news, except that there is a restriction that the

physical source of the image is a text, and the type of processing from percept to conceptual

space can be speci�ed as reading. Depending on the nature of the writing, there may or

may not be any other entities set up in conceptual space or base space. Here base space

is foregrounded, so that the object of see is the text. Consider the di�erences among the

following:

(1) a. ?I saw the Chronicle this morning. read

b. I see that there's going to be a parade today. news

c. I saw in the Chronicle this morning that there's going to be a parade today.

news

Ex. (1-a) presupposes that the seer read at least the usual portion of the newspaper, but

has nothing to say about whether he or she believed anything written there. Ex. (1-b) talks

only about the concept; it would be most appropriate when the source of the concept is

either known from context (e.g. the newspaper is in front of the speaker).

Ex. (1-c) foregrounds the concept (assumed to be true) and also gives the source,

perhaps by way of corroboration; a mental spaces diagram for this would add the Chronicle

to the base and perceptual spaces, with an in relation between the Chronicle and the article

in both spaces.

Note that Ex. (1-a) is marginal for some speakers, probably because of the di�culty

of knowing how much of the newspaper the speaker is claiming to have read. On the other

hand, the negative Ex. (2-a) and the question Ex. (2-b) are �ne, since the hearer doesn't

need to estimate the portion read. Because of this uncertainty, in reply to the question, a

\no" is acceptable, but a simple \yes" Ex. (2-c) is too little to satisfy the Maxim of Quantity

(Grice 1975; Grice 1978). A cooperative response must further specify the amount read, as


in Ex. (2-d) or Ex. (2-e).

(2) a. I didn't see the Chronicle this morning.

b. Did you see the Chronicle this morning?

c. Yes, I did.

d. Not really, I just glanced at the sports page.

e. Yes, but I only glanced at the sports page.

Finally, we note that, like envision, recognize can also occur in contexts which

do set up a perceptual space that contains some clue that leads to the concepts in the

conceptual space. Usually the hearer must also do some inferencing to understand the

connection. Consider the following example from an American folksong:

(3) I see by your out�t that you are a cowboy.

The out�t is the clothes that the addressee is wearing, the physical source of the perception,

but it is expressed by an oblique in the sentence; instead, the conclusion is foregrounded,

and forms the complement of see. The speaker has had to do some inferencing to connect

the boots, hat, chaps, etc. with the addressee's occupation, and the hearer must reverse the

process to imagine what the speaker saw. The situation can be represented as in Fig. 3.7.

BC

P

outfit

outfit’

outfit’’you’

you’’

cowboy’’

a

bc

cowboy

you

Figure 3.7: recognize: I see by your out�t that you are a cowboy.

The perceptual space contains the person (you) and the out�t; in the conceptual

space, we have the person, the out�t and the role cowboy . In both of these spaces, you and

out�t are linked by dashed line (a) representing the \wearing" connection; from the image

of the body partially covered by clothes, the seer knows that the addressee is wearing a

particular kind of out�t. But in the conceptual space, the out�t and the role \cowboy" are


also linked (dashed line (b)), representing the pre-existing knowledge of what kind of out�t

cowboys wear. From this the seer deduces that the addressee has the role cowboy (dashed

line (c) in the conceptual space). (Expressions of the form see by X that Y are will be

discussed further in the following section on page 121.) There is no reason to suppose that

the seer is wrong, in fact, the �rst person pronoun shows us that the seer is the speaker,

so we can con�dently copy everything in the conceptual space to the base space. Note that

the term \base space" is not equivalent to \reality space"; the whole process expressed by

the sentence could be embedded in a dream, with someone dreaming of seeing a person

in an particular out�t and concluding that he was a cowboy; this conclusion would still be

\true" within the dream, and therefore spread to the appropriate space.

3.3 The Semantics of Motion Expressions with See

Although it has been known to scientists for centuries that light travels from in-

candescent objects to the experiencer, either directly or by re ection, expressions of motion

connected with seeing do not always accord with this fact. On the one hand, many percep-

tion situations (both visual and other) are expressed in terms of motion from the stimulus

to the experiencer. In discussing this phenomenon, Lako� (1995) gives the sentences in

Ex. (4) as examples of the metaphor perception is reception.3

(4) a. A comet came into my sight.

b. The noise came through the walls.

c. The smell of the bay came through the fog.

On the other hand, there is also a metaphor involving motion in the opposite direction, also

applied both to vision and to other sense modalities (examples from Lako�).

(5) a. From my o�ce, I can see the bay. (=Ex. (4-b) on page 43)

b. From my o�ce, I can see all the way to the bay. (=Ex. (4-c) on page 43)4

c. From my o�ce, I can hear the trains.

d. From my o�ce, I can smell the bay.

e. From the mountain, I can pick up broadcasts from Moscow on my radio.

3I have omitted some of Lako�'s question-begging examples. Note also that it is di�cult to constructexamples with the verb see and this direction of \motion".

4In our sense divisions, this is faculty, since all the way to the bay is a path, not something seen.


f. Through the piles of leaves . . . , the dogs picked up the scent of the escaped

prisoner.

Thus we could say that there is a folk theory of vision which is \extramissive", in which

vision or the \gaze" travels from the eyes to the seen. Let us agree with Lako� that this

is an example of the more general metaphors perception is touch and perceptual

organs are limbs, and further that, since we have found that the motion can be con-

strued in either direction, that both directions can be subsumed under the more general

metaphor perception is contact between perceiver and perceived. We need not

agree exactly with Lako� that the prepositional phrases in Ex. (5) make the main verbs

metaphorical, but the examples in Ex. (4) and (5) do suggest that perception events can

be construed as metaphorical motion, and that this metaphorical motion can be in either

direction.

When we look at perception which is per se metaphorical, such as recognize, we

often �nd these combined with metaphorical motion as well, but the mappings are somewhat

di�erent. As Lako� (1995:141) points out, what is usually the content of perception as

in Ex. (5-a) also can be a metaphorical goal, as in Ex. (5-b).

Furthermore, the experiencer can be construed as a goal and the stimulus as

a source, both of the physical perception and of the knowing. This mapping works best

when the metaphorical motion is from stimulus to experiencer, as in Lako�'s example,

(6) John looks sick to me from the pictures.

There are many other ways to express judgements of this sort without the to and

from, and hence, the suggestion of motion, as in Ex. (7).

(7) a. When I saw the pictures, I was convinced John was sick.

b. Judging by the pictures, John must be sick.

c. Based on how he looks in the pictures, I'm sure John is sick.

When the main verb is see, the direction of motion is presumably from experiencer

to percept, yet we still get blends with a PPfrom; thus these can be either a physical starting

place for eye, or the abstract starting place for the mental event that leads to knowing;


evidence is a cause for belief, and reasoning is a journey from evidence to belief5 (I could

see from the calluses on his hands that he did a lot of manual labor).

With see as the main verb, there is a tendency for the content to map to goal; if we

want to use see as a causative, as in sense ensure, we cannot mark the goal of the causation

with the most common preposition for goals, to (Ex. (8-a)). But Ex. (8-b), without the

preposition, is ambiguous between cognition (recognize) and causation (ensure). To

make it clearly causal, we can use the \empty" pronoun it as the object of the goal-marking

proposition to and follow it by the clause that expresses the caused state or event.

(8) a. *She sees to (that) they're busy/them busy/their business.

b. She sees that they're busy.

c. She sees to it that they're busy.

We can also treat the evidence as a means to the goal of knowing, and mark it with by

(I could see by the calluses on his hands . . . ). Similar patterns exist with tell and know ,

which can also be verbs of cognition. (Cf. Lako� (1995) and Sweetser (1990:Ch. 2) for more

details of verbs of perception used in the cognition frame with metaphorical movement

expressions.)

Furthermore, see participates in other semantic and syntactic patterns that are

partially shared by other verbs (and nouns) of perception. (For discussion of these verbs

in English see Fillmore (1994), Atkins (1994) and Declerck (1982); for cross-linguistic com-

parisons, see Viberg (1983).) Many of these generalizations can probably best be captured

in a typed feature structure hierarchy, as I will discuss in Section 3.5.

3.4 A Brief Discussion of Two Complex Idioms

Envisionment and Classi�cation as Reality

(9) Brown: As far as he could see there was no hole to climb through it.

In the Brown corpus, we �nd the very problematic sentence shown in Ex. (9). We

assume that as far as he could see is intended literally, i.e. to refer to physical perception

(eye), as suggested by hole and climb. Without looking at a larger context, it is not clear

5Cf. Sweetser (1987:450) on the path inherent in deduce (< Lat. de+ducere `lead out/from')


whether \he" could see all of the area of \it" in which a hole would be useful, or whether

his range of vision was limited, and a climbable hole might actually exist just out of his

range of vision. The former interpretation is possible because the construction as far as NP

can/could see is frequently used with abstract objects (i.e. recognize), and there is also a

question of the scope of the negation.

But Ex. (9) is also part of a larger pattern. To anticipate Chapter 5 slightly, we also

�nd similar sentences in the Kenkyusha English-Japanese Dictionary , shown in Ex. (10).

(10) a. I see a great danger in that sort of thing. (401-h-iii)

b. I don't see any harm in what he is doing. (401-h-ii)

c. I cannot see myself submitting to it. (401-n-ii)

d. He won't see being made use of. (401-o-i)

There are also many similar sentences in the BNC; a search for see followed by no

turned up more than a thousand examples, a few of which are shown in Ex. (11).

(11) a. BNC: I see no reason why this could not apply to di�erent sizes of swimming

pool.

b. BNC: They see no possibilities of widening the issues beyond their congrega-

tions.

c. BNC: Dot could see no way out.

d. BNC: I can see no point in reacting to it.

e. BNC: [On hearing] of what the Tans had done Churchill said he could see no

harm in it.

The construction seems to be SEE + Quanti�er + fuse, need, point, reason,

cause, . . . g + PP/VPto, where the quanti�er is usually of negative polarity and the use of

preposition or VPto depends on the lexical item. This use may be based on a metaphorical

extension of what Lako� (1987:128) calls the \idealized cognitive model (ICM) of seeing":

A. You see things as they are.

B. You are aware of what you see.

C. You see what's in front of your eyes.


The metaphorical extension is to the categorization frame, which can be expressed with

the pattern \see X in Y". The X in Y pattern is based in turn on the Container metaphor,

according to which objects are containers and their qualities are their contents; thus, if Y

has the quality X, then X is in Y. Since part A of the ICM states that you see things as

they are, if you see Y and see X in Y, then X is in Y. Conversely, according to the ICM,

if you see Y and don't see X in it, then X is not in Y. Then by the Container metaphor,

Y does not have the quality X. Hence, if you see no harm in something, there is no harm

in it, and you can conclude that it is not harmful, and likewise for the other sentences in

Ex. (11).

\Would rather see him hanged" and Related Patterns

Many examples of these sorts of sentences are found in corpora such as the BNC;

a few are shown in Ex. (12).

(12) a. BNC: . . . I'm warning you, I'll see you dead �rst, before you take that girl.

b. BNC: I won't let him carry out his plan, whatever it is. I'll die, or I'll see him

dead �rst!

c. BNC: You can tell that old bag I'll see her stu�ed before I move an inch in her

direction.

d. BNC: I may be old-fashioned, but I'd rather see them married than indulging

in a sordid little a�air.

These can be analyzed as uni�cations of several constructions. In broad outline, the ele-

ments are as follows:

(I) Preference, expressed in any of several ways:

(13) a. will/would rather A than B

b. will/would A before will/would B

c. will/would A �rst

d. NP1 before NP2

A and B are XPs expressing states or actions, either VPbrst, VPs with will/would

or NPs expressing states. Pattern (13-d) allows only the last of these, as in Ex. (14-d).


(II) The use of \extreme values" of A to express strong unwillingness to do B.

Many of these are conventionalized, such as die, face a �ring squad , be damned .

(14) a. I'd rather die than go to the prom with Billy!

b. I'd go broke before I'd sell to your company!

c. I'd die �rst!

d. Death before dishonor!6

e. Over my dead body!

f. I suppose that's a compliment. . . though I had rather have had a plague of

boils, if I'd uv had my druthers. (Robert Heinlein, Citizen of the Galaxy 1957

Del Rey [1978] pp.57-8)

(III) Finally, one or both of the VPs in the comparison can use see + AdjP to

express a state, i.e., the condition sense, yielding the patterns we started with in Ex. (12)

All languages have ways of expressing preferences; the use of negative extreme

values (especially death) for strong unwillingness is found in many languages.

(15) a. Chinese

T�a n��ng s�� y�e b�u xi�ang.

He would rather die than surrender.

b. Japanese

Sonna koto ni kane wo tsukau gurai nara suteta hou ga mashi da.

It would be better to throw your money away than to spend it on that sort of

thing.

However, since the condition sense of see is not common across languages, the full com-

bination, as in Ex. (12) seems to be unique to English.

3.5 Senses as a Function of the Semantic Types of Arguments

If we were to try to devise an algorithm capable of di�erentiating the senses of see

as part of a natural language understanding system, it might be considerably easier to learn

6A closely related pattern, as in Give me liberty or give me death! , is included by Lako� (1971:144) inwhat she calls \rhetorical disjunctions" among other examples of asymmetric or .


NON-SENTIENT SENTIENT

HUMANTEXT

PHYS. OBJ

ENTITY PROPOSITIONEVENT

ABSTRACT

PLACE TIME

STATE PROCESS QNED PROPOSITION

T

Figure 3.8: Minimal Type Hierarchy

something about the semantics of the arguments �rst and then �nd out what this might

enable us to deduce about the sense of see. Even so, I will have to make a few simplifying

assumptions to keep the problem within reasonable bounds. First, let us set aside for the

moment sentences in which metaphorical motion is expressed, i.e. those discussed in the

preceding section. Second, let us assume that we already have ways of deriving something

about the semantics of the arguments; I assume that the system can recognize that in both

saw the destruction of the city by the army and saw the army destroy the city , the seen

is a process, which happens to be syntactically realized as an NP in the former and as an

NP+VP in the latter.

The output from such an analyzer could be a semantic type, organized within a

typed feature structure (Carpenter 1992; Koenig & Jurafsky 1994), for both the seer and

the seen. Most of the types are fairly commonly used within the �eld of lexical semantics

and should not be controversial. Fig. 3.8 shows the sort of type hierarchy that will be

needed; the terminology varies from author to author, such as whether events are included

in entities, whether one uses the term thing , etc. but the general idea should be clear from

the �gure. One unusually detailed type of physical object, text, is included in the tree, as

well as Questioned Proposition as a subtype of Proposition. We will also need to add at

least one other semantic/pragmatic feature besides the semantics of the arguments, realis

vs. irrealis (cf. Section 3.2).

Most of the senses of see can then be represented as the join of two types, one for

the seer and one for the seen, at the appropriate level of generality. The basic distinctions

between senses involving physical vision and those not necessarily involving it largely follow

from the type of the seen. Senses that foreground more complex and abstract contents of


perception, such as news and classify, require not merely sentient seers, but human ones.

Because so many relations need to be represented, the diagram becomes very di�cult to

read; I have therefore separated it, purely for expository purposes, into two parts. Fig. 3.9

on the next page shows such a representation for those senses involving human seers and

Fig. 3.10 on page 128 shows the senses involving other types of seers. These �gures can

be interpreted as follows:

The semantic type hierarchies of the seer and the seen are shown at the left and

right of the diagram, respectively, in the trees drawn with heavy lines. The light lines show

the joins of the types, with the names of the corresponding senses written near the angles.

In two cases (envision and classify), it has been necessary to show senses as having two

possible types of seen; in this case the two alternates are marked with a dashed arc. Thus,

for both envision and classify, the seer must be human and the seen can be either an

entity or an event; the di�erence between these two depends upon the fact that envision

is marked irrealis, while classify has two non-subject arguments, one an oblique, usually

marked with as .

Note that the sense scan has only the semantics of the seer speci�ed; it can take

a seen of any type. We could also directly represent the fact that all the senses of see

are related to the same set of irregular past and past participle forms; this is not shown,

however, as doing so would add another dimension to our �gure and make it much harder

to read.

3.6 Uni�cation and Re exives

The mechanism of uni�cation that underlies Construction Grammar (and Frame

Semantics) is completely general; anything can unify with anything so long as there are no

con icting speci�cations. This is how re exives are handled within these frameworks. For

example, in Ex. (16-a), there is a cooking frame, with Matilda as the cook, dinner as the

food, and her mother as the beneficiary; in Ex. (16-b), the same frame is involved, but

Matilda is both cook and beneficiary.

(16) a. Matilda has to cook dinner for her mother.

b. Matilda has to cook dinner for herself.

c. Matildahuman has to cook dinner for herselfsentient.

CHAPTER3.OTHERCOGNITIVEAPPROACHES

127

Semanticsof Seer

Semanticsof Seen

ENVISION

phys. obj.

setting

non-sentient

abstract

proposition

T

entity

irrealis

entity

event

READ

textphys. obj.

ENSURE

NEWS

DATING

CONSULT

ACCOMPANY

VISIT

SPECTATE

state

process

human

CONDITION

sentient

human

Qned proposition

PROCESS

CLASSIFY

Figure 3.9: Senses in Relation to Types of Arguments (seer = Human)

CHAPTER3.OTHERCOGNITIVEAPPROACHES

128

Semanticsof Seer

Semanticsof Seen

phys. obj.

non-sentient

abstract

DETERMINE

proposition

HALLUCINATE

T

entity

humanSCAN

(FACULTY)

irrealis

RECOGNIZE

entityphys. obj.

Qned proposition

text

event

sentient

human

state

process

EYE

CONDITION

PROCESS

place/time

SETTINGEXPERIENCE

Figure 3.10: Senses in Relation to Types of Arguments (seer 6= Human)


Since the cook must be of type human, and the beneficiary of type sentient (at

least), there is no con ict in having the same human �ll both roles; we can represent this

more fully by marking the types in the sentence, as in Ex. (16-c). If, for some reason, one

thing cannot �ll two roles in a frame even if they are of similar semantic types, this must

be stated, either in that frame or in some more general, inherited frame. For example, all

comparative constructions presuppose that there are two distinct individuals (or groups)

to be compared, even though they are often of the same semantic type (He's taller than

his brother), so this must be speci�ed at a fairly general level for all comparatives in all

languages.

Although sentences like those in Ex. (17) look like comparisons of very di�erent

things, they are not.

(17) a. He's taller than he used to be.

b. He's taller than I thought.

Both are really comparisons of two heights, of the same person at di�erent times

(17-a) and of a real person and of a mental image of that person (17-b). Mental space

diagrams might make this clearer by linking entities representing the same person in two

spaces (created by a tense di�erence in the �rst case and by imagining in the second case).

With regard to see, for many senses, the types of the roles will prevent the same

individual from �lling more than one role, e.g., recognize, determine, ensure, and

news all require a sentient (or human) seer and a proposition or state of a�airs as the

seen, so there is no possibility of an overlap between them. Other type con icts prevent

re exives with read, setting, experience, vide, discourse, hallucinate, spectate,

and tour.

But some re exive uses of see do occur, and in general, the types on the roles

proposed thus far will work as they should. (It is not clear why Ex. (18-d) is bad for some

speakers, unless the problem is ambiguity with envision. In Ex. (18-g), the category is

not just herself but herself only smarter .)

(18) a. eye Shesentient saw herselfphys obj in the mirror.

b. envision Hesentient sees himself �ve years from now getting rich running his

own company.SOA


c. process Hesentient saw himselfentity becoming stronger day by day.

d. condition �Isentient don't want to see myselfentity working until after mid-

night again.relational

e. scan+eye ?The video cameranonsentient can see itselfphys obj in the mirror.

f. classify Theysentient see themselves as advocates for their clients.

g. classify Shesentient sees her sister as herself, only smarter.

The problem arises in the case of senses where seer and seen are both human; as shown

in Ex. (19), most of these re exives are bad. Three of these cases can be suitably handled

by requiring that the the two participants in the reciprocal event frame be distinct; that

will take care of dating, visit, and audience, all of which use these roles from recip-

rocal event, and follows from the concept of reciprocity. In addition, we will need to

block the re exive consult (Ex. (19-c)) by specifying that, in its ancestor frame consul-

tation with authority, the consulter and consultant are distinct; this will also be

needed for words other than see which use that frame (cf. p. 70). A similar requirement will

be needed for one of the valence patterns with gambling (Ex. (19-g)); the names gambler

and opponent in the gambling frame imply as much. Finally, we need to ensure that the

two participants in the accompany frame are distinct, so that their counterparts in the

accompany sense of see will also be; unfortunately, we must also make an exception to

even this generalization for the idiom see oneself out (Ex. (19-b)). See oneself out means

something like \I can take care of myself here, I know the way out"; it may be using the

\protection" concept from accompany, but that's about all.

(19) a. *accompany She'llhuman see herselfhuman to the bus stop.

b. I'llhuman see myselfhuman out.

c. *consult Dr. Joneshuman decided to see himselfhuman about his sore throat.

d. *dating Billhuman has been seeing himselfhuman for six months now.

e. *visit Sarah Janehuman saw herselfhuman for three hours this afternoon.

f. *audience The Chairhuman will see herselfhuman Wednesday afternoon.

g. *gambling Mariohuman saw himselfhuman and raised himself �ve.


3.7 Conclusion

There is nothing contradictory in any of the approaches discussed in this chap-

ter, or between them and Frame Semantics; rather, they are complementary. As so often

happens in linguistics, we �nd that di�erent approaches including Frame Semantics, mental

spaces, analysis in terms of metaphor, and semantic type hierarchies handle di�erent pieces

of the puzzle well. I will continue to use frame semantics as the basis of most of my analysis,

but we must bear in mind that no one theory provides a really good account of all the facts.

132

Chapter 4

Psycholinguistic Experiments

4.1 Introduction

After seeing the arguments in Chapter 2 for an analysis of the semantics of see

which postulates many senses, connected with each other in complex ways, the reader might

naturally have questions as to whether the proposed senses have any psychological basis.

A psychologist might be tempted to say, \Such a complex lexical semantics is purely a

theoretical construct, invented by linguists for linguists, in uenced by what is printed in

dictionaries. It bears no relation to any mental representation of the word see that English

speakers might have. Speakers' mental representations of the word probably have just one or

two senses, with all the subtleties you have talked about arising from cognitive processing

òn the y'". This chapter details a series of experiments1 intended to respond to such

objections and to seek answers to questions such as:

� Given examples of various uses of see, what sort of sense divisions will speakers make

on their own, without any guidance?

� Given an a priori set of categories based on our linguistic analysis, how well will

speakers agree with each other on which example falls in which category?

� How will their level of agreement be a�ected by the number of categories they are

1This is joint work with Jane A. Edwards, who has taken part in the design, running, and analysis of theexperiments. My colleague Chris Johnson also participated in the initial establishment of the list of senses,and I have received innumerable suggestions from the members of my committee, other UCB graduatestudents and faculty, and audience members at presentations on this topic at the The Linguistic Society ofAmerica (Baker 1999) and the Berkeley Linguistics Society (Baker forthcoming).

CHAPTER 4. PSYCHOLINGUISTIC EXPERIMENTS 133

given?

� Will they make the same sorts of categorization decisions when forced to do so quickly?

� Will they make certain categorization decisions more quickly and accurately than

others? If so, can such di�erences in response time tell us anything about their own

mental representations or processes?2

On-line vs. O�-line Methods

Psycholinguistic experiments can be divided into those that ask the subjects to

make high-level, conscious decisions about linguistic questions and those that ask subjects

to make quick responses to relatively simple questions but measure reaction times, seeking

clues as to the moment-to-moment processing of sentences; we can refer to these as \o�-line"

and \on-line" methods respectively. The term \on-line" is fairly standard in this �eld; I am

using \o�-line" to refer to all experimental designs in which the speed of response is not

a factor.3 O�-line methods allow us to construct experiments relatively easily that seem

to answer some fundamental linguistic questions directly (\Is this sentence grammatical?",

\Are these words synonyms?" \What is the antecedent of this pronoun?"), but the data

they provide is often hard to interpret, because so much higher-level cognition may be

involved in the decision; e.g. in a grammaticality judgement, the subject may actually have

time to remember some rule that she was taught in the sixth grade, rather than relying

on a purely intuitive judgement. On-line tasks have a much better chance of discovering

something about the semantic representation used in actual sentence understanding, but

the experiments are much harder to construct and a great many factors need to be carefully

controlled to produce valid results.

O�-line experiments in which subjects estimate semantic relatedness of words may

produce a great many numbers which in some sense re ect the subjects' semantic maps of

the world. But the subjects have time to consider many dimensions on which words may be

similar, even though the response is a single number for each trial. Also, judging semantic

similarity on a scale is arguably an even more \unnatural" task than grammaticality judge-

ments; most people have at least some experience in school of proofreading, either their own

2Di�erent mental representations are very di�cult to distinguish from di�erent processing strategies, cf.Section 4.5.

3See Section 1.5 in Chapter 1 for a more general discussion of types of data and collection methods.


writing or that of others, looking for grammatical errors, but the task of rating similarity on

a scale seems to be con�ned to psychological experiments4. Nevertheless, such experiments

do produce some intriguing results; for example, in their second step, Durkin & Manning

(1989) asked their subjects to rate the semantic relatedness of pairs of uses of ambiguous

words (some homophonous, but mostly polysemous); they found that the similarity judge-

ments distinguished well between the homonyms and the polysemous words and provided

a good measure of the relative salience of the senses of the latter. (Durkin & Manning also

went on to do a priming task on related data.)

Free sorting experiments have better face validity than similarity judgments, but

the resulting categories are hard to compare. For example, Jorgensen (1990) gave subjects

cards containing low-polysemy nouns and high-polysemy nouns, as measured by the number

of dictionary senses. In the �rst task, they did completely free sorting; in the second, they

were told to divide the cards according to a set of dictionary de�nitions they had been

provided with. She found that subjects basically produced approximately 3 categories for

the low-polysemy words in both tasks, but created an average of 5.6 categories on the

free sorting and 9.1 on the dictionary-guided sorting. This was still less than the average

number of senses given in the dictionary (14.6), but the increase was signi�cant. Jorgensen

uses measures of the number of categories produced by her subjects and the amount of

agreement between them on the classi�cation of individual items, but has nothing to say

about the relation between the semantics of the categories produced by one subject and

those of other subjects or those of the dictionary.

Previous Work on Priming

Various experiments have demonstrated that priming e�ects between related words

can provide information as to the structure of semantic �elds (de Groot 1984, Meyer &

Schvanenveldt 1971). Other experimenters have used priming techniques to study the rela-

tions between the separate senses of homonyms. Swinney (1979) used cross-modal priming,

with an auditory stimulus consisting of two sentences with the ambiguous noun occurring

in the predicate of the second sentence, and a visual probe related to one of the senses

presented either (1) at the o�set of the ambiguous word or (2) three syllables later. For

4Subjects making grammaticality judgements apparently make the same grammaticality judgementswhether they are asked to apply their own standards or what they believe to be \academic" standards.(Cowart 1997:56-59)


example, one pair of sentences referred to a government building containing bugs , probed

with either spy or ant ; the context could either be carefully ambiguous or deliberately

biased (. . . spiders, roaches, and other bugs. . . ). Ambiguous sentences alternated with un-

ambiguous controls, e.g. ones in which bugs was replaced with insects . When the probe

was immediate, Swinney found that both senses of the ambiguous noun were primed even

if the context was strongly biased toward one or the other of the readings. When the probe

was presented three syllables later, however, only the contextually appropriate sense was

primed. These results suggest that the lexical access process is not guided by contextual

information, but retrieves all of the senses of ambiguous items. However, contextual infor-

mation is used very soon thereafter to select the appropriate sense from those that have

been retrieved.

Seidenberg et al. (1982) conducted a methodical series of �ve experiments on two

types of homonyms, both noun-noun (e.g. ball dance/baseball) and noun-verb (a tire/to

tire), using auditory primes. In each case the last word of the auditory prime was either a

homonym or a control word. A visual probe consisting of a single word was presented either

zero milliseconds or 200 milliseconds after the end of the auditory prime. For example given

the noun-noun homonym spade, the following four sentences were used as primes:

(1) a. You should have played the spade.

b. You should have played the part.

c. Go to the store and buy a spade.

d. Go to the store and buy a belt.

If the following probe is the word card , Ex. (1-a) is congruent with it, using spade. Ex. (1-b)

is congruent using a control word, Ex. (1-c), using the homonym, is incongruent with the

probe, and Ex. (1-d) is also incongruent, using a control word.

Over the course of the �ve experiments, Seidenberg et al. carefully varied the

extent to which the context selected for one of the senses, either syntactically, semantically,

or pragmatically. Consistent di�erences between the e�ects of probing at zero milliseconds

and at 200 milliseconds demonstrated that lexical access per se was not in uenced by

context but that all the senses of the ambiguous words were retrieved \instantaneously"

and then a rapid selection process based on the context occurred. Apparent counterexamples

seem to the result of lexical priming caused by earlier words within the prime (The auto


workers protected the plant , The football player fumbled the ball). Their results generally

lend support to the idea of a more or less autonomous lexical access process, followed by a

rapid process of selection of the appropriate meanings on the basis of context.

Williams's (1992) experiments also have important implications for those described

here. Williams notes the work of Swinney 1979, Seidenberg et al. 1982, and others on

homonyms, and points out that polysemous words raise somewhat di�erent questions. Since

the various senses of a polysemous word are, by de�nition, more closely related than the

senses of homonymous words, it is not clear whether or not the model based on homonyms,

i.e. rapid simultaneous activation of all the senses followed by rapid selection of a contex-

tually appropriate sense, will work. Williams also cites the work of Durkin & Manning

(1989) measuring the relatedness of the senses of polysemous words using a questionnaire

technique and notes that the authors are correct in saying that the results of such survey

techniques cannot be assumed to apply to the moment to moment processing of language.

Williams' experiments sought to combine information about the relative importance and

degree of relatedness of the senses of polysemous words (such as that found in Durkin and

Manning's work) with priming experiments like those of Seidenberg et al. (1982).

Williams' �rst experiment was based on eight polysemous adjectives. For each

of them, a more central sense and a less central sense were determined, based on the

entries in the Collins Cobuild English Language Dictionary, in which more frequent and/or

more concrete senses are listed �rst; a one-word synonym was found for each reading, e.g.,

awkward (clumsy (central) or embarrassing (non-central)), strong (mighty or intense).

All of Williams' stimuli were presented visually, with the prime disappearing and being

replaced (either immediately or after an interval) by the probe; timing of stimulus onset

asynchronies (SOAs) is from the onset of the display of the last part of the prime to the

onset of the probe. The task was a lexical decision task.

First, a group of subjects were shown just single words as primes followed by the

probes at an SOA of 250 ms. There was signi�cant priming of the probes when the prime

was related to the probe, and this priming e�ect was approximately equal, whether the

sense of the probe was central or non-central, e.g. a prime of the ambiguous word �rm

facilitates a probe of either solid or strict about equally.

Next, as in other experiments, sentences were constructed as primes, with the word

of interest as the last word in each sentence:


(2) a. The schoolteacher was criticized for not being �rm. solid

b. Nobody went to the pub because the music was so loud. solid

c. The couple wanted a bed that was �rm. strict

d. The orchestra hated the symphony because it was so long. strict

These represent the combinations [related prime + central probe], [unrelated prime + central

probe], [related prime + non-central probe], and [unrelated prime + non-central probe].

Another group of subjects were tested using these sentences and probes at SOAs of 250,

750, and 1100 ms. In this case, there was signi�cant priming (about 40 ms.) of central senses

even up to 1100 ms., but only slight priming (about 20 ms., not statistically signi�cant) of

non-central senses. Williams concludes, \It does not appear to be possible to suppress the

irrelevant meanings of a polysemous adjective in the same way as it is possible to suppress

the irrelevant meanings of a homonym." (Williams 1992:202)

Finally, a second experiment was conducted, using the same adjectives, but each

sentence ended with an adjective modifying a noun. In this case, the subjects were asked

to press a button to indicate whether or not the probe was related to the meaning of the

sentence as a whole. The expectation was that, aside from yes responses for probes which

match the meaning of the sentence, no responses should be slower for probes representing the

inappropriate reading of the ambiguous adjective than for irrelevant controls. Furthermore,

a larger percentage of yes responses were expected, even for the inappropriate reading, due

to the semantic relatedness of the senses.

The results of the second experiment were more mixed; the analysis involved drop-

ping the data from the subjects who took, on average, more than 1 second to respond. But

for the \fast" responders, there was a signi�cant increase in yes responses and a signi�cant

inhibition of no responses for irrelevant but related probes vs. neutral controls, despite

the intervening noun and a 250 ms. delay. This e�ect occurred for probes for the central

sense, but not for the non-central sense, i.e. contexts which bias for the non-central sense

apparently activate the central sense, but the converse is not true.

Unfortunately, as in so many other researchers' work, both Williams' sense distinc-

tions and his decision as to which constituted the central sense were based on a dictionary,

rather than empirical investigation. The Cobuild dictionary may be better than some in

terms of ordering senses on the basis of frequency and conceptual centrality, but it is not

immune to the commercial constraints and other linguistically irrelevant factors (discussed


in Section 5.1) which in uence the number and type of senses listed in dictionaries. Ex-

amining the stimulus materials (commendably listed in an appendix) shows that some of

the words used as probes for senses are not very good cues, or are themselves ambiguous,

e.g. for tight , taut vs. ?compact. The choice for deep is especially confusing, profound,

which is ambiguous in exactly the same way as deep, vs. low, which is highly polysemous:

physically not tall or high/ vulgar or depraved/ de�cient, lacking (low in intelligence, vita-

mins, saturated fats, low birth weight/ at the \bottom" end of various scales (low rate of

crime, taxes, low-pitched (= deep), low (radio) frequency, lower (military) rank , etc.).

On the basis of this prior work, we chose to combine free sorting with two other

methods, a forced classi�cation task and a priming experiment, in the hope that the

strengths of each method would o�set the weaknesses of the others. In the classi�cation

task, subjects were forced to classify uses into many predetermined categories, which were a

subset of those de�ned in Chapter 2. As stated there, this set of senses represents what we

believe to be the �nest breakdown which is logically defensible, considerably �ner than that

available to most people through introspection. If we �nd that a substantial proportion of

subjects agree on distinctions in the free-sorting task which cross-cut ours, then we have

failed in our e�ort. If, on the other hand, we have succeeded, we would expect to �nd that

any distinction which subjects reliably make should be representable as a combination of

the �ner ones. We would also expect such combinations to be more or less systematic, that

is, motivated by concepts which would be postulated on independent grounds.

Of course, carrying the grouping process to its logical conclusion, by collapsing

all the categories together would produce complete \agreement" of the unsatisfactory sort

posited in the strong version of monosemy (Ruhl 1989). As Cruse (1992) points out, claiming

that a single highly abstract, unde�nable \sense" accounts for all the uses of a highly

polysemous word is not only ipso facto unprovable but also fails to distinguish such words

from each other.

Predictions

On the basis of the study so far, we would predict the following:

� Since see seems to be highly polysemous, we would expect that non-linguist native

speakers of English will be able to distinguish many di�erent senses when tested on

tasks involving similarity judgements, categorization (using either prede�ned or their


own spontaneous categories), etc. There is no reason to suppose that all speakers

have exactly the same set of senses, but it is likely that there will be a great deal of

overlap, which is essential to communication in general.

� Since see is a highly polysemous, high-frequency word, we expect that our subjects

will produce more senses than Jorgensen's subjects, on average.

� Since the senses appear to have a complex structure, some senses being more cen-

tral than others, we would expect to �nd broad, although not necessarily perfect,

agreement among speakers as to which are the central senses.

� In a cross-modal priming experiment using a lexical decision task, in accord with

Williams's (1992) �ndings, we would predict that sentences which provide a context

for one sense of see would facilitate (prime) responses to a probe consisting of the

keyword for that sense of see, compared to a non-word probes; e.g., after the sentence

Maria saw the cat on the couch, we would expect subjects to respond to the keyword

eye more quickly than to a non-word.

� Likewise, on cross-modal priming using a categorial judgement task, we would predict

that sentences which provide a context for one sense of see would facilitate responses

to a probe consisting of the keyword for that sense of see, relative to probes consisting

of keywords for other senses.

� Also on the basis of Williams (1992), we would predict di�erential cross-sense priming

e�ects, so that sentences biased toward non-central senses of see would facilitate

probes for central senses more than conversely.

4.2 Experiment 1

As a �rst step toward testing these hypotheses, we conducted an experiment to

answer the following questions:

� If non-linguist native speakers of English are given a set of sentences containing see

randomly selected from a corpus, and asked to sort them into their own categories,

to what degree will their categorizations be similar to each other? Will there be

\lumpers" and \splitters", that is, some subjects who tend to make few distinctions

and other subjects who tend to make many �ne distinctions?


� If they are shown the same set of sentences again in a random order and are also

provided with an a priori set of categories,

{ to what extent will their categorizations be similar to each other?

{ to what extent will their categorizations be similar to those of the experimenters

for the same sentences?

We chose actual sentences from a corpus as the best approximation to natural uses

of see which could conveniently be used as experimental stimuli.

Method

Subjects

The subjects were 9 undergraduates at the University of California at Berkeley,

who received credit toward their introductory psychology courses for their participating in

the experiments. All subjects were native speakers of American English with little or no

training in Linguistics.

Materials

The stimuli for Experiment 1 consisted of two blocks of 100 sentences selected

at random from the Brown corpus, together with 44 constructed example sentences. (All

the constructed sentences, along with a random sample of 50 corpus sentences are listed

in Appendix A.) For the sorting task, each block of 100 sentences was printed on 3x5 inch

cards, forming two sets, Set 1 and Set 2. A set of cards was also prepared for the 44

constructed example sentences.

For the classi�cation task, a set of Perl programs and web forms were written to

display the material. Nineteen senses were used in the classi�cation task; the de�nitions

and examples are given in the Appendix on page 256, along with the names we used for

them at that time. The same pool of corpus and constructed sentences were used in both

tasks.

Procedure

Each subject participated in two tasks, Sorting and Classi�cation. One hour was

allocated for each task.


Task 1: Sorting

In this task, subjects were given cards with one sentence containing see and asked

to read each sentence one at a time and place it in a pile with sentences containing the

same sense of see. They were instructed to ignore grammatical factors such as tense and

voice, and to use as many piles as they wanted. Subjects were also asked to write a brief

de�nition or characterization of each group.

Each subject sorted 100 corpus examples, followed by the 44 constructed examples,

followed (if time permitted) by the remaining 100 corpus examples. To control for the

in uence of the particular examples in the �rst part of the sorting, the �rst 100 sentences

given to each subject were either all from Set 1, all from Set 2, or half from each set.

Task 2: Classi�cation

Each subject sat at a computer on which the materials were displayed using HTML

and a web browser. First the subjects read the directions and saw the list of de�nitions and

examples. Then the sentences were presented one at a time at the top of the screen, with

the list of senses displayed below. Subjects chose the sense which best matched the use in

the sentence by clicking next to it and then clicked again to go on to the next sentence.

Both the response and the response latency were recorded, but the task was not paced, and

subjects could refer back to the de�nitions and examples whenever they wanted, either on

screen or on a printed handout. All subjects completed the same 99 sentences (1 block of

trials) within one hour.

Statistical Measures of Agreement

Three di�erent measures of agreement were used in analyzing the results of the

experiments reported here, simple agreement, kappa, and omega.

Simple proportion of agreement with a \gold standard" is obviously the easiest to

interpret, but it has serious limitations. It does not re ect agreements among raters that

do not match the standard.

For example, suppose in an experiment on perception of dialects, recordings of

three speakers, A, B, and C, are played to a total of 12 subjects (called \raters"), who must

decide which speakers are from New York, which from Chicago and which from Philadelphia.

Several possible outcomes of such an experiment are shown in the four parts of Table 4.1.


The rows represent the speakers, and the columns represent the frequencies of the subjects'

judgements (called \ratings"). Suppose further that Speaker A is really from New York, B

from Chicago, and C from Philadelphia; then the numbers on the diagonal represent the

number of agreements between the ratings and the actual dialects of the speakers.

In Table 4.1 (b), the simple agreement between the actual dialect and the ratings

is 2/3, or 0.67. But (c) will have exactly the same simple agreement, even though it is

clear that the raters are agreeing among themselves in ways that do not match the actual

dialects. In Table 4.1 (d), the simple agreement is 1/3 (0.33), even though the data do not

show any relation at all between the actual dialects and the ratings.

(a)

NY Chi Phil

A 10 0 2B 2 10 0C 0 2 10

simple= 0:83K= 0:55

(b)

NY Chi Phil

A 8 2 2B 2 8 2C 2 2 8

simple= 0:67K= 0:18

(c)

NY Chi Phil

A 8 0 4B 0 8 4C 4 0 8

simple= 0:67K= 0:24

(d)

NY Chi Phil

A 4 4 4B 4 4 4C 4 4 4

simple= 0:33K= �0:09

Table 4.1: Examples of Measures of Agreement

Because of the limitations of simple agreement, it is often preferable to use the

kappa statistic (Scott 1955, Cohen 1960), which is corrected for chance agreement. This is

the standard statistic for inter-rater reliability when the number of categories is �xed for

all raters. The basic formula is

K =P (A) � P (E)

1� P (E)

where P(A) is the proportion of actual agreement, and P(E) is the proportion of expected

agreement. Since the numerator is the di�erence between the actual and the expected,

results no better than chance will give kappas around 0. Returning to Table 4.1, (a) shows

that kappa drops o� quite sharply when even a little disagreement occurs, and (c) gives a

higher kappa (0.24) than (b) (0.18) because of the greater agreement of the raters among

themselves, even if they are incorrect. As expected, kappa is near 0 (actually slightly

negative) when there is no agreement among raters, as in (d). (For further discussion of the


logic of the statistic, see the excellent introduction in Siegel & Castellan (1988:284-91).)

The omega statistic (Morey & Agresti 1984) is based on whether or not two raters

classify each pair of stimuli in the same category or not, without regard to the classi�cation

of other pairs. Like kappa, it is corrected for chance agreement, so that it varies from 0

for chance agreement to 1.0 for perfect agreement. Omega is inherently less powerful than

kappa, since it considers each pair of stimuli in isolation, however, it has the great advantage

that it can be used in cases in which the number of categories di�ers from rater to rater.

We will therefore use omega to measure agreement among subjects on the Sorting task,

where di�erent subjects created di�erent numbers of categories.

Both kappa and omega are insensitive to the number of categories involved, or the

type of distribution of instances into categories. The variance of the sampling distribution

is known for both, so that the probability of a particular outcome can be calculated.

Results and Analysis

Task 1: Sorting

The number of categories per subject ranged from 6 to 21, with a mean of 11.

The proportion of examples in the categories varied greatly, from 33% for eye and 15% for

recognize to 0 for some categories. The omega coe�cient of agreement ranged from 0.09

to 0.49, with a median of 0.245.


All subjects �nished 99 sentences of the �rst set. Some subjects continued on to

other sets, but the order of the sets was randomized across subjects, so that there was little

overlap beyond the �rst set. The overall agreement among raters, measured by the kappa

statistic, was .385. This value is low, but understandable; not only were a large number of

senses listed, but also many of the sentences were ambiguous when presented out of context.

Discussion

In evaluating the results of Experiment 1, the strengths and weaknesses of using

corpus examples became apparent. On the one hand we had learned something about the

5This and all of the values of kappa reported in this study are statistically signi�cant at p < 0:05.


distribution of the senses in naturally occurring texts. On the other hand, many of the

corpus examples were ambiguous, which made the task more di�cult for the subjects and

added noise to our results. Also, the very uneven distribution of the senses meant that we

had more data than we needed on eye and too little data to draw conclusions about many

of the less common senses. Furthermore, the tasks used in Experiment 1 were both o�-line,

metalinguistic judgements; we and others to whom we reported our �ndings had doubts

as to whether the distinctions that subjects made in these tasks were relevant to on-line

sentence processing.

At the same time, as a result of continuing linguistic analysis, we had revised and

expanded the list of senses from 19 to 23 senses and 9 collocations (names and examples

are given in Sections 2.3 and 2.6). Some of what we had called senses in Experiment 1 were

reclassi�ed as collocations, since they involve the presence of other speci�c words; among

these are: see x through, see through x, and let's see. On the other hand, senses

such as accompany, although \idiomatic", place semantic restrictions on their comple-

ments, but do not require any speci�c syntax, (I'll see you as far as the bus stop, I'll see

you home, I'll see you to your door).

It was therefore decided to construct a new experiment which would contain only

examples of seven clearcut senses which should be relatively easy to distinguish from each

other. Two online tasks were also added to the sorting and classi�cation tasks, to help us

investigate possible di�erences between o�-line and on-line processing.

The names (keywords) for the 14 senses used in the remaining experiments are

shown (in alphabetical order) in Table 4.2 on the next page for convenience, as they will be

used hereafter without further explanation.

4.3 Experiment 2

Method

Subjects

Subjects were 21 UC Berkeley undergraduates with quali�cations as in Experi-

ment 1.


Keyword Example

accompany He saw her home.

condition She saw him happy at last.

consult You should see the doctor.

determine I'll see if I can open the jar.

ensure I'll see to it that he's on time.

envision I just can't see living in Podunk all my life.

experience The freighter saw use as a troop carrier in the war.

eye She saw him through the window of the train.

faculty She doesn't see as well as she used to.

hallucinate He saw stars for a few minutes after hitting his head.

process He saw her dancing with the football hero.

recognize She saw that he was working too much.

setting The 1960s saw a construction boom in Japan.

visit I haven't seen my sister since Christmas.

Table 4.2: Names of Senses used in Experiments 2 and 3

Materials

The seven senses chosen for this experiment were: eye, faculty, recognize,

determine, ensure, experience, and setting.

Example sentences were constructed for each of the seven senses, systematically

varying other factors such as tense and aspect, question vs. statement, negation, voice, and

domain of discourse. (The three broad domains of discourse were \academic", \personal",

and \entertainment".) In practice not all combinations of these factors produced reasonable

sentences, but as many as possible were created. Appendix D contains the entire set of

stimuli used in Experiments 2 and 3.

Task 1: Sorting

The instructions were the same as those in Experiment 1, except that subjects

were also asked to choose the sentence which best exempli�ed each sense and to place it on

the top of the pile at the end of the hour.


The method was the same as that used in Experiment 1, except that a thoroughly

randomized order of presentation was used both for stimuli and for the list of senses, to


eliminate any possible ordering e�ects6.

Timed Tasks

The stimuli in these tasks were presented by use of the PsyScope program (Co-

hen et al. 1993) on Macintosh computers. In each case the subjects saw one of the same

example sentences as in the previous tasks, displayed in large text near the center of the

screen, and then heard an auditory prime consisting of a single word (or sometimes, in the

Lexical Decision task, a single non-word). The subjects then pressed one of the keys on the

keyboard to respond; they were instructed to use the index �nger of their dominant hand

for positive responses and the middle �nger of their dominant hand for negative responses.

The sentences were displayed for up to 4 seconds, although subjects could end this and go

on to the probe by pressing the space bar. This was followed by the auditory probe which

lasted approximately 500 ms. Subjects had 1500 ms. from the beginning of the auditory

probe to respond; responses after this time period were not used in further analysis.

Blocks of 40 trials of each task (Lexical Decision and Categorial Judgement) were

administered randomly across subjects, so not all subjects did the same amount of each

task, but all subjects participated in both. Subjects worked for one hour altogether on the

two tasks, and were allowed to rest after each block.

Task 3: Lexical Decision

In Lexical Decision blocks, the probe was either a keyword for the primed sense,

a keyword for another sense, or a non-word. The task was a word/non-word judgement.

Task 4: Categorial Judgement

In Categorial Judgement blocks, the probe was either a keyword for the primed

sense or a keyword for another sense, and the task was to decide whether the probe was an

instance of the primed sense.


024681012

��

HHHHHHHHHHH

HH

��HHHHHH

3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

Freq.

Figure 4.1: Experiment 2. Number of Categories used by each subject


Task 1: Sorting

Figure 4.1 shows the number of categories created by each subject in the Sorting

task. The median is 6, with 5 and 7 being the next most common numbers of categories;

this is a fair approximation to the 7 categories intended by the experimenters. (One subject

produced 19 piles of cards; after reading the de�nitions he wrote and looking at the cards

contained in each pile, we can �nd no regular basis for the distinctions made.)

A calculation of the omega statistic between each pair of subjects showed that there

was substantial agreement among subjects even before any instructions as to categorization

were given; the mean = 0:57. There was considerable variation among subjects, but there

was no cluster of subjects who agreed with each other and disagreed with the experimenters'

initial categorization. This suggests that, although there may be agreement among subjects

and disagreement with the experimenters on individual pairs of senses, there is no other

well-de�ned \dialect" for the senses of see among our subjects.

As a test of the e�ectiveness of our manipulation of the experimental variables,

the omega statistic was used to compare the subjects' initial sortings with the values of

the manipulated variables, including the intended sense. Table 4.3 on the next page shows

the results for a representative group of nine subjects; the agreement for the irrelevant

manipulated factors is essentially zero (because of the correction for chance agreement, the

value of omega can sometimes be less than zero). The agreement with the intended sense

ranges from a low of .36 for subject number 30 to a high of .82 for subject number 33; this

variation in agreement seems to be due to individual di�erences. These results suggest that

the subjects were able to follow the instructions to pay attention only to the sense of see

occurring in each sentence and to ignore the other syntactic and semantic factors.

6Fellbaum et al. (1998) found such e�ects in a similar classi�cation task.


Factors 30 31 32 33 34 35 36 37 38

Tense/Asp. 0.03 0.04 0.02 0.02 0.03 0.05 0.03 0.02 0.02Qn/State. 0.03 0.01 0.00 0.01 0.01 0.02 0.02 0.02 0.01Negation 0.01 0.00 0.01 0.01 0.00 0.01 0.00 0.01 0.00Voice -0.03 -0.01 0.03 0.01 0.00 0.03 -0.02 0.03 0.01Domain 0.08 0.06 0.06 0.06 0.06 0.06 0.06 0.06 0.06

Sense 0.36 0.40 0.75 0.82 0.62 0.46 0.50 0.71 0.61

Table 4.3: Agreement (Omega) between Subjects' Sorting and Manipulated Variables


In addition to recording subject responses, the response latency on the Classi�ca-

tion task was also recorded; the distribution has a strong right skew, as is typical of such

measurements. The median latency was 19 seconds, with the �rst quartile at 13 seconds

and the third quartile at 26 seconds. Latencies longer than 80 seconds were considered

errors, since it seems unlikely the subjects were actually attending to the current item for

so long.

In terms of simple agreement, 84% of the items were classi�ed as intended by the

experimenters; the mean kappa for agreement among all subjects on the seven categories

was .74. This is much higher than Experiment 1, presumably due to the smaller number of

senses and clearer example sentences.

Responses !

Intended #

eye

faculty

determine

ensure

recognize

experience

setting

Total

eye 623 49 3 2 8 0 0 685faculty 65 381 1 0 3 0 0 450

determine 46 11 394 25 2 2 2 482ensure 19 7 33 444 13 3 1 520recognize 11 9 25 7 597 3 0 652

experience 10 3 4 5 4 388 131 545setting 11 1 1 6 1 73 335 428

Total 785 461 461 489 628 469 469 3762

Table 4.4: Experiment 2. Intended Senses vs. Responses


Table 4.4 on the preceding page shows the relation between intended senses and

responses for all of the items on the Classi�cation task. The senses have been arranged so

that those which frequently overlap are in adjacent rows and columns. Thus, while eye and

faculty were \correctly" classi�ed most of the time, 49 instances of intended eye were

classi�ed as faculty, and 65 instances of intended faculty were classi�ed as eye. The

asymmetry between the two \errors" may be due to the general bias toward the response

eye.

There is also some overlap among the three senses determine, ensure, and

recognize. We note that all three of these senses involve a relation between the seer and

a proposition; in the case of ensure, the seer brings the proposition about; in determine,

the seer �nds out if the proposition is true; in recognize, the seer becomes aware of the

truth of the proposition.

Finally also we note overlap between experience and setting, especially in the

direction from intended experience to response setting. It may seem surprising that

these two senses overlap, especially as the setting sense is unique with respect to the

semantics of its subject. The similarity between the senses is that both of them allow non-

animate subjects (The house saw use as a barracks during the Revolutionary War). In the

Sorting task, several subjects created a category for non-animate seer, and this may be

the reason for the relation found between these two senses.

Clustering of Classi�cation Responses

One way of thinking about the disagreements between the intended senses and the

responses is to consider what would happen if we collapsed certain pairs of senses. If we

collapse senses where there is a good deal of overlap, we should increase the kappa; if we

collapse senses which were well distinguished, we could actually decrease the kappa. Doing

this collapsing systematically produces a clustering algorithm, based on the kappa statistic.

The steps are as follows:

1. For each pair of categories, compute the kappa that would result if they were combined

into one.

2. Actually combine the pair which produces the greatest increase in agreement.

(This represents the distinction which was hardest for the subjects to agree upon.)


3. Repeat this procedure, until combining categories produces no more improvement.

Depending on the data, this may be before all categories are merged.

The order of combining can be represented as a tree, with the branchings at the

bottom of the tree representing the categories with the most overlap. The height of each

branching represents the new level of agreement produced by combining the categories

below. The clusters can be thought as re ecting the speakers' hierarchy of mental represen-

tations in this semantic space. That is, distinctions within clusters are more di�cult than

distinctions between them.

DETERMINE ENSURE

RECOGNIZE

0.8

0.9

1.0

0.7

EXPERIENCE SETTING

EYE FACULTY

0.85

0.80

.88

.90

.92

Initial kappa = 0.74

Figure 4.2: Experiment 2. Clustering of Senses Based on Increasing Agreement

Fig. 4.2 shows the results of clustering on the basis of overall agreement between

subjects (measured by kappa). Three clusters are noticeable (in order of decreasing overlap),

eye/ faculty, experience/ setting, and determine/ ensure/ recognize. While


these clusters were not foreseen by the experimenters, they can be motivated on the basis of

the semantics of the arguments of the verb. experience and setting are the only senses

in the experiment which allow non-sentient seers (which must be the subjects of the verb).

determine, ensure, and recognize are the only senses which allow propositions as the

seen. eye and faculty, of course, both require physical vision and a physical object as

the seen, while none of the other �ve senses do.

Timed Tasks (Lexical Decision and Categorial Judgement)

The data for the Lexical Decision task and the Categorial Judgement task are

still being analyzed, but some preliminary results can be stated here. First, regarding the

accuracy of responses, we have found quite a high percentage of \correct" responses in the

Categorial Judgement task, i.e. simple agreement between the subjects' response and the

category intended by the experimenters; the median is 92% agreement (Q1 = 87%, Q3 =

96%).

With regard to reaction time measurements, the priming e�ects expected on the

basis of Williams (1992) and similar experiments are very small (less than 50 milliseconds).

Our measurement of reaction times is subject to an error of approximately 16 ms., due to the

polling frequency of the Macintosh keyboard. This error will tend to be random, masking

the e�ects we are seeking, but accumulating data from a su�cient number of subjects should

overcome this problem.

All reaction time data was �rst standardized on the basis of all the scores of that

subject on that task, either Lexical Decision (LD) or Categorial Judgement (CJ), so that

the mean of the standardized scores would be 100 and the standard deviation 10. This

allows comparisons of scores adjusted the individual subjects' overall reaction speeds and

variability.

First, let us look at the standardized reaction times for \correct" responses to the

situation in which the prime and the probe match for each of the 2 tasks, which are shown

in Table 4.5 on the following page. The rows are sorted to put the standardized RTs on the

categorial judgement task in ascending order. Since there are clearly di�erences according

to sense, paired t-tests were performed to test the signi�cance of the di�erence between

each pair of senses, using two di�erent methods of handling missing cases, listwise deletion

and analysis-by-analysis. Looking �rst at the categorial judgement task, the values of t


Sense CJ LD

eye 93.14 98.05faculty 95.29 98.97ensure 97.18 107.41determine 98.07 95.19recognize 98.57 100.80setting 101.45 99.32experience 101.97 103.68

Table 4.5: Mean Standardized RT for \IN" Responses

are shown in Table 4.6, with signi�cant results marked with an asterisk and trends marked

with a T. (All statements of signi�cance are based on one-tailed t-tests at the p < 0:05 level

unless otherwise speci�ed.)

Probes!

Primes#

eye facl ensr detr recg expr setgmeanRT

Listwise deletion of missing cases

eye { <1 -1.22 -2.30* <1 -1.99* -1.91* 94.32faculty { { <1 -1.05 <1 -1.48 -1.06 96.51ensure { { { <1 <1 -2.02* <1 98.68determ { { { { <1 <1 <1 99.49recog { { { { { -1.59 <1 98.04exper { { { { { { <1 104.66setting { { { { { { { 101.28

Analysis-by-Analysis deletion of missing cases

eye { <1 -1.09 -1.74T -1.02 -1.76* -2,41* 93.14faculty { { <1 -1.42 <1 -1.70T -1.27 95.29ensure { { { <1 <1 -1.54 -1.46 97.18determ { { { { <1 <1 1.02 98.07recog { { { { { -1.31 <1 98.57exper { { { { { { <1 101.97setting { { { { { { { 101.45

Table 4.6: Paired t-tests for di�erences in RT on the Categorial Judgement Task

These tests are suggestive, but the pattern is not as clear as might be desired. It

appears that part of the problem is the low number of data points in each cell, which could

be remedied either by gathering more data or (more immediately) by combining senses into

groups. The obvious candidate grouping is the one based on the clustering by agreement dis-


cussed in the preceding section, i.e., (1) eye-faculty, (2) ensure-determine-recognize,

and (3)settingexperience. T-tests were then performed to test the signi�cance of di�er-

ences between these groups, using both methods of handling missing cases. Using listwise

deletion, signi�cant di�erences are found between groups 1 and 2 and between 2 and 3;

signi�cant di�erences are found between groups 1 and 3 using either deletion method. No

signi�cant di�erences are found between senses within the proposed clusters with either

deletion method.

Returning to Table 4.5 on the preceding page, the pattern of results from the

lexical decision task is not so clear, and grouping as before does not yield the same pattern

of signi�cant di�erences. We note that, with the exception of determine, the two senses

having to do with physical vision (eye and faculty) are faster than the others; this

makes sense from a cognitive point of view, with \concrete" senses accessed faster than

\metaphorical" ones. The reason for the unexpectedly speedy reactions to determine and

setting is still being sought. There may be additional noise in this task due to perceived

similarities between some of the keywords and some of the non-words, despite our e�orts

to create a set of uniform, unrelated, English-sounding non-words.

Discussion

In general, the subjects distinguished among the senses reliably on at least three

out of the four tasks. They were able to understand and follow the instructions and their

unrestricted categories (from the Sorting task) were generally in accord with those used by

the experimenters.

Since the results showed that the subjects were able to handle 7 senses well, we

decided carry out a new experiment, adding 7 more senses to the stimuli, to bring us closer

to the numbers of senses hypothesized on linguistic grounds in Chapter 2.

4.4 Experiment 3

Method

Subjects

Subjects were 39 UC Berkeley undergraduates with quali�cations as in experi-

ments 1 and 2.


024681012

HHHHHH ��

��

HHHHHH��

HHHHHH HH

3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

Freq.

Figure 4.3: Experiment 3. Number of Categories used by each subject

Materials

The seven new senses were visit, consult, process, condition, envision, hal-

lucinate, and accompany. In order to keep the total set of stimuli small enough, only the

5 clearest examples of the seven senses used in Experiment 2 were retained in Experiment 3.

As before, not all combinations of the manipulated factors with the senses produced good

sentences.

Procedure

The procedure was identical to that used in Experiment 2.


Task 1: Sorting

The number of categories created by each subject in the Sorting task is shown in

Figure 4.3. The median number of categories is 10, which is signi�cantly larger than the

median of 6 on Experiment 2 (using the median test, �2 = 26:09, p < 0:01). Although the

median is somewhat less than the 14 intended categories, a substantial number of subjects

also made 12 or more distinctions. (As in Experiment 2, there was one such outlier, with

18 categories in this case, some of which were apparently inconsistent.)

With regard to marking central senses, most of the subjects chose either a category

corresponding to [eye+faculty], or eye alone if they separated faculty from it. For

example, of the 35 subjects in Experiment 3 who marked one sense as central, 31 chose

eye or [eye+faculty]. Of the remaining four subjects, two chose faculty (although less

prototypical, faculty can be logically argued to be more basic, since the visual faculty is a

precondition for eyeing), one chose something like recognize and one mysteriously chose


accompany. Most of the de�nitions were short, but one subject wanted to leave no doubt

in our minds that he meant eye, writing this de�nition:

to actually visualize something. To have an impulse that travels to the brain viathe optic nerve so that the brain actively con�rms the existence of the objectyou visualize.


We found that the 10 sentences with the lowest level of agreement were causing a

disproportionate amount of mismatches between the response and the intended sense, and

most of these also had no more than about 50% of agreement among the subjects' responses,

so we eliminated them from further consideration, reducing the number of stimuli from 115

to 105. (There had been no comparable problems in the data for Experiment 2.)

Even a total of 14 senses, 75% of the responses agreed with the experimenters'

categorization. The mean kappa among all subjects fell only to .70; after the elimination of

the 10 weakest items, it rose to .75.

Table 4.7 on the following page shows the relationship between intended senses and

responses for Experiment 3, after the 10 weakest items have been eliminated as described

above. Once again, we �nd the general bias toward the response eye, and some of the

same overlaps as noted in Experiment 2. In addition, the newly added senses create new

combinations; the most striking result is that the majority of examples of intended process

receive the response eye. Although some of the subjects created a separate category in the

sorting task for what we call process, the predominance of eye responses for intended

process stimuli suggests that most subjects regard perceiving a person performing an

action as a simple physical perception (i.e. eye), notwithstanding the secondary predication

associated with it. The newly introduced sense condition also creates considerable overlap,

although the vast majority of cases are \correctly" recognized.


Responses!

Intended#

eye

process

faculty

visit

consult

condition

experience

setting

envision

hallucinate

recognize

determine

ensure

accompany

Total

eye

175

4

11

0

0

2

1

1

0

0

2

1

0

0

197

process

239

174

18

4

0

7

8

2

16

0

10

1

0

1

480

faculty

20

0

153

0

1

0

0

0

0

0

0

0

0

0

174

visit

35

0

4

382

0

0

0

0

1

0

0

2

0

0

424

consult

1

0

1

25

487

0

0

0

0

0

0

1

0

0

515

condition

30

32

16

2

0

279

14

0

3

0

5

0

1

0

382

experience

0

6

1

0

0

2

122

6

6

0

1

1

0

0

145

setting

0

0

0

0

0

3

33

108

6

0

0

0

1

0

151

envision

9

9

2

0

0

0

1

1

335

2

1

0

0

0

360

hallucinate

19

0

3

0

0

1

0

0

0

397

1

0

0

0

421

recognize

1

4

1

0

0

30

4

0

3

0

168

7

1

0

219

determine

3

2

0

5

9

4

0

0

9

0

3

210

15

2

262

ensure

0

0

0

0

0

1

0

0

1

0

3

2

222

2

231

accompany

6

0

1

5

0

0

0

0

0

0

0

0

2

416

430

Total

538

231

211

423

497

329

183

118

380

399

194

225

242

421

4391

Table4.7:Experiment3.IntendedSensesvs.Responses


sight, but altered or not real (maybe a subsection of the other sight category)

seeing stu� in your mind; things that don't actually exist

possible hallucinations

look at (imaginary)

to perceive imaginary objects, colors, etc

hallucinate

imagine a stimulus, involuntarily

seeing something not really there; inside your head or imagining an image

to hallucinate visual image. to visualize something that seems real only to youbecause it is in your mind

visual delusions

hallucination; create/picture an image not really there

internal physical perception, internal cognitive (distorted), (hallucination), (use ofimagination)

visual (unreal); hallucinations

Table 4.8: Subjects' de�nitions for hallucinate

The categorization in the rest of the table is remarkably clearcut, but let us discuss

the 19 cases of intended hallucinate which were classi�ed as eye. This should in principle

be a relatively clear-cut distinction, between seeing a physical object and seeing \something

that is not there". In fact, 13 of the 39 subjects spontaneously listed senses in the sorting

task that clearly refer to our sense hallucinate, as shown by their own de�nitions in

Table 4.8. On the other hand, seeing hallucinations can be thought of as a kind of physical

visual experience, in which the physical source (the stimulus) is inside the the body of the

perceiver. (To delve any further into the actual cause of hallucinations would lead us into

clearly non-linguistic questions about neurology and mental illness.) Thus, in Chapter 2,

we treated hallucinate as a composition of eye and an irrealis seen.

The 19 instances of intended hallucinate which were classi�ed as eye arise from

the 6 sentences shown in Table 4.9 on page 159. As the table indicates, all of these examples

were usually classi�ed as hallucinate, and the 19 instances of eye do not seem to be part

of a regular pattern. Furthermore, the highest proportion of eye responses is found in the

�rst sentence, which is also inherently problematic, since spots in front of your eyes can

sometimes refer to \ oaters" (loose blood cells in the vitreous humor), and sometimes result

from low blood pressure or low blood sugar, all of them physical causes with which people

are generally familiar.

As in Experiment 2, agreement among subjects on the Classi�cation task was


0.8

0.9

1.0

0.7

Initial kappa = .75

.78

.80

.83

.85

.87

.89

.91

.86

.92.93

.95.96

EYE PROCESS

FACULTY

STATE

EXPERIENCE SETTING VISIT

ENVISION

RECOGNIZE

DETERMINE

ENSURE

CONSULT

HALLUCINATE

ACCOMPANY

Figure 4.4: Experiment 3: Clustering of Senses Based on Increasing Agreement


eye fac hal Sentence

7 2 58 Don't you sometimes sees spots in front of youreyes if you stand up suddenly?

5 0 32 I didn't exactly see dragons ying in the livingroom, but I did notice that colors and smellsseemed more intense.

2 0 19 It took him a few minutes to recover after beingtackled so hard, but he wasn't seeing stars oranything.

2 0 62 If this drug is as dangerous as they say, should Istart to see the walls breathing sometime soon?

1 0 35 How long after you took the drug did you seethe unicorn in your backyard?

2 0 66 If he keeps drinking like that, he's gonna be see-ing pink elephants soon.

Table 4.9: Distribution of Responses For Problematic Instances of hallucinate

also calculated (after the elimination of the 10 unsatisfactory items) and the clustering

algorithm was applied to that data. Fig. 4.4 on the page before shows the resulting tree;

unfortunately, the clustering is not nearly as clear-cut as in Experiment 2. Except for the

pair experience/setting, each new sense is an added to a single growing cluster. The

overlap between eye and process is shown by the fact that these are the �rst to merge,

and the eye/faculty cluster seen in the preceding tree is re ected in the close proximity

of faculty. The newly introduced senses condition and visit also merge soon thereafter.

The three senses which formed a cluster of their own in the preceding experiment, i.e.

envision, recognize, and determine, merge one after the other, but do not form a

separate cluster in this tree. The position of accompany at the top of the tree is in accord

with its relatively di�erent meaning. It is, however, hard to explain why envision and

hallucinate are so far apart, and why consult should be so far from visit.

Timed Tasks

Because of the way the stimuli were randomized, there were many combinations

of prime and probe for which we have only a small number of responses or even none. This

is particularly true for Experiment 3, with its larger number of senses; we are �lling a 14 x

14 table rather than just a 7 x 7 table, and would need four times as much data to achieve

the same coverage. The large number of gaps in the data have made it impossible to do


the sort of statistical analysis which was done on the timed tasks of Experiment 2. We

can, however, calculate the simple agreement between the subjects' categorial judgements

and the intended senses: after eliminating one subject who pressed the \yes" key on all of

his responses, the median is 94% agreement (Q1 = 89%, Q3 = 97%), even with the larger

number of senses.

4.5 Conclusions

Experiment Senses Stimuli N Subjects Tasks

1 20 Corpus 9 1 & 2

2 7 Constructed 21 all

3 14 Constructed 39 all

Table 4.10: Summary of the Experiments

Table 4.10 shows a summary of the stimuli, tasks, and number of subjects in each

of the experiments.

We must recognize that not all the interesting questions about the polysemy of see

can be resolved using experiments of this type. For example, nothing in this experimental

setup will help to resolve the well-known problem of distinguishing retrieval of di�erent �xed

representations from di�erences in processing strategies, with which cognitive psychologists

have been so concerned. There are also the ever-present dangers of generalizing from an

unrepresentative sample; UC Berkeley undergraduates are certainly not a representative

sample of English speakers in general. On the other hand, we have no a priori reason

to suppose that their mental representations of the semantics of see are systematically

di�erent from those of other speakers, since it is such a common word, and most of its

senses are very common as well.

Also, as with most psychological research, there is no way to distinguish inter-

subject di�erences due to di�erent mental representations (or processing) from those caused

by subjects' di�ering interpretations of (or attitudes toward) their duties in the experiment.

(One exception is that data from subjects whose performance is obviously contrary to

the instructions can legitimately be omitted from further analysis, as we have done in a

few cases described below.) If, for example, some subjects tried to divide senses as they


imagined a dictionary would have and others did not, but simply tried to construct their

own categories in their own way, this might produce variation across subjects in the type of

categorization. It seems unlikely that enough subjects carried out one such an interpretation

of the instructions consistently enough to substantially a�ect our results.

As noted earlier in this chapter, we expect to �nd individual variation in internal-

ized semantic structures for see across speakers; these patterns would be expected to follow

the usual lines of geographical and social dialects, with educational level playing a major

role, but we were not able to determine anything about such distribution in this study. To

do so would require many more subjects from much more diverse backgrounds. Nor do these

experiments provide any evidence as to the relative importance (or temporal precedence)

of semantic vs. syntactic factors. As noted in Chapter 2, many senses have quite speci�c

restrictions (syntactic and/or semantic) on their arguments, such as accompany or deter-

mine. The subjects may be learning to distinguish at least some of senses from relatively

straightforward cues which are primarily syntactic (although the assumptions that syntactic

cues (a) can be easily separated from semantic cues and (b) are more straightforward than

semantic cues may be characteristic of some linguists rather than most other speakers).

Even given these caveats, there are still a number of signi�cant conclusions to be

drawn; let us review our predictions in the light of our results:

1. Since see is highly polysemous, we predicted that our subjects would produce more

senses on the sorting task than the subjects in Jorgensen 1990.

In Experiment 1, the mean number of categories per subject was 11.6, substantially

higher than the mean of 5.6 found by Jorgensen (1990), as we had predicted. For

Experiments 2 and 3, the median numbers of categories produced were 6 and 10,

which approximate the number of senses intended by the experimenters, i.e. 7 and

14 respectively. The di�erence between the two medians is signi�cant, meaning that

subjects recognized that more senses were present in Experiment 3 on the basis of the

stimuli alone.

2. We predicted broad agreement among our subjects as to what the central sense(s) of

see is.

Most of the subjects chose as a central sense either a category corresponding to

eye+faculty, or eye alone if they separated faculty from it.


3. We predicted that, although there would be individual variation, subjects would be

in basic agreement regarding the sense structure.

In Experiments 2 and 3, our subjects were able to distinguish 7 and 14 senses respec-

tively as shown by the results on the Classi�cation and Categorial Judgement tasks.

This suggests that the relatively low rate of agreement in Experiment 1 was due to

the ambiguous and unbalanced stimuli rather than to inherent di�culty of the tasks.

The level of agreement among subjects on both the sorting task and the classi�cation

task suggests that the categories which we used in the classi�cation task matched fairly

well with the categories which the subjects had at the beginning of the experiments.

The very high accuracy found on the Categorial Judgement task under timed con-

ditions might be interpreted as proving that the subjects actually use the categories

which they displayed on classi�cation task in understanding natural language sen-

tences. However, we cannot rule out the possibility that subjects have merely learned

the categories created by the experimenters extremely well by that point. In any case,

the high level of accuracy obtained after a short training period suggests that the a

priori categories used in the experiment were easy for the subjects to learn. There are

also relatively good correspondences between many of the subjects' own spontaneous

de�nitions and those of the experimenters. It is di�cult to quantify either of these sit-

uations, but, taken together, they suggest that the experimenters' a priori categories

and the subjects' own initial representations match reasonably well.

4. We made three relatively speci�c predictions regarding cross-modal priming experi-

ments. Our data have not provided convincing evidence to support those predictions,

primarily due to gaps in the data. We have found tendencies in the expected directions

for speci�c senses, and fairly strong evidence of priming in the predicted directions

when the data for individual senses are clustered in ways that are in accord with our

�ndings from inter-subject agreement.

In addition to the hypotheses we began with, we have also found that, at least in

Experiment 2, clustering of senses on the basis of inter-subject agreement produces group-

ings that, while not predicted in advance, can be motivated on the basis of independently

supported semantic types. It is an empirical question whether these results would also hold

true in an expanded version of Experiment 3, gathering more data. An agenda of research


designed to answer some of these remaining questions will be discussed in Chapter 6.

164

Chapter 5

What the Dictionaries Say

5.1 Introduction

It is important to study the semantics of see based directly on primary data,

whether in the form of introspection, corpus searching or experiments. But it would also

be foolish to ignore the e�orts of professional lexicographers to de�ne the word; we should

now be able to look at dictionaries and to compare them with the results of the study so

far. The �rst part of this chapter discusses what English lexicography has to say about

the senses of see and the sense divisions that occur in a number of major monolingual

English dictionaries; the second part gives some comparisons between the sense divisions

in monolingual English dictionaries and those which appear in several bilingual dictionaries

(sometimes called translating dictionaries). The relevant sections of several dictionaries are

included in this chapter and discussed in detail; most of the original typography has been

preserved, but line breaks have been introduced and (in many cases) numbering added so

that the reader can follow the discussion more easily. Since we are discussing only the sense

divisions, the etymologies, pronunciation, and word forms (saw/seen/seeing) are omitted

from the entries. For the bilingual dictionaries, we use regular orthography in the case of

Spanish, and the national standard romanizations for Chinese and Japanese.

We must remember that, even though almost every new dictionary is compiled with

reference to many existing dictionaries, there are severe constraints on the amount of time

and e�ort which professional lexicographers can spend on the writing of individual entries

for words; the writing of dictionaries (whether mono- or bilingual) is a commercial enterprise

in which time spent on lexicographic analysis is a cost and productivity is measured in terms

CHAPTER 5. WHAT THE DICTIONARIES SAY 165

of words per day. As Patrick Hanks puts it,

\. . .The enormous cost of compiling a new dictionary, the absurdly long peri-ods for which the investment is tied up before any return on it can be expected,the terrible fear that the risk may not pay o�, and the resulting pressure tokeep the book to a predetermined length (predetermined by marketing consid-erations, rather than by what needs to be said) are just some of the factors thatcan all too often combine to sti e initiative in lexicography.. . . " (1992:110)

Students of lexical semantics can spend a potentially unlimited amount of time on

precise distinctions, but most users of dictionaries will be satis�ed if they �nd a de�nition

which applies to the context in which they found the word, without too much concern for

omissions and overlaps among senses.

Since seeing is a central part of human experience, all languages have words to

express varieties of seeing. What is interesting linguistically is what similarities and di�er-

ences we �nd in the patterns of lexicalization across languages. In the domain of motion,

for example, it is well known that some languages tend to have motion verbs (or their equiv-

alent) that incorporate the direction of motion, while the manner of motion is expressed

by another word, whereas other languages have motion verbs that incorporate manner, and

the direction of motion is expressed by separate words (Talmy 1985).

Other studies have looked at the main verbs of perception across languages. For

example, Viberg 1983 examines a wide range of languages, but depends largely on bilingual

dictionaries for data. Some of the work of the DELIS project deals with verbs of perception

(Heid & Krueger 1994) and is based on a substantial body of corpus data, but is limited

to a few European languages, all Indo-European (although substantial variation in lexico-

semantic patterning with regard to the complements of verbs of perception is found among

those languages). While evidence across two or three languages (especially if they are his-

torically or culturally related) does not prove very much, evidence of lexicalization patterns

from a range of unrelated languages may o�er glimpses of semantic universals in this �eld.

Since perception and consequent knowledge are part of the shared experience of

all people, we would expect to �nd generally similar senses across languages. This does not

mean that each language will have a single verb of visual perception that will have senses

matching all of the senses of English see; we have already observed that this in not the case.

Rather, we expect to �nd the notion of inchoative knowing of both concrete and abstract

entities and events being conveyed by verbs of (primarily visual and auditory) perception,


with or without intentionality (i.e. corresponding roughly to English see, look , hear , and

listen).

Because the breakdown of senses given in Chapter 2 is rather detailed, I do not

expect to �nd many languages making many �ner sense distinctions along these same

lines.1 There will undoubtedly be many cross-cutting distinctions which English does not

have, such as honori�cs, di�erent tense/aspect divisions, etc., but it would be surprising

to �nd, for instance, a language which used di�erent verbs of perception for animate and

inanimate seen within the eye sense. On the other hand, it would not be surprising to

�nd that many languages do not allow their common verb for see to be as a translation

of the verb in I see why he's happy , since, on our analysis, eye and recognize represent

di�erent senses of see in English.

5.2 Monolingual English Dictionaries

Buyers of English dictionaries have certain expectations as to what a dictionary

should contain, and dictionary writers must inevitably follow these traditional expectations

in order to satisfy their customers. Part of this tradition involves the notion of substi-

tutability, i.e., that the de�nition should include a synonym or paraphrase which can be

directly substituted for the word to be de�ned in the sense in question with no change of

either syntax or meaning (Hanks 1992:103). There may well be, in fact, no good semantic

substitute with the right part of speech in the language, but this tradition is still strong.

For example, when elementary school children are asked to write de�nitions for words, and

they write Tennis is when two people take rackets and hit a ball over a net. . . , they may be

criticized and told that they must write instead Tennis is a game in which two people. . . ,

i.e., that the noun tennis should not be de�ned by a clause, but by another noun game

(although the di�erentia may be expressed by a clause.)2

Readers of dictionaries have also come to expect some sort of sense divisions for

high frequency words with many sorts of uses, i.e., there seems to be a folk concept of

polysemy. However, this concept is apparently extremely vague; readers write to publishers

with many sorts of complaints about dictionaries, but never to complain about the sense

1Of course, speakers of any language can make semantic distinctions as �ne as they like by use ofappropriate devices such as modi�ers, in ections, syntactic variation, etc. I am referring here to the set ofsenses shared by all speakers when they use the word.

2Cf. Fillmore & Atkins (1992:101) on the relation between substitutability and Frame Semantics.


divisions for polysemous words (B.T.S. Atkins, p.c.).

Some monolingual dictionaries simply list a series of senses separated by semicolons

with no indication of the relations among them. This may give the impression that the senses

are supposed to be completely distinct, although in fact, the lexicographers are perfectly

aware that one may include another, or be related to it in a variety of ways (Hanks 1992:102).

Other dictionaries try to specify the major and minor sense divisions and give examples

of each. In some cases the relations among senses may be indicated by connectives and

anaphoric expressions, such as entries that start with \hence", \especially", etc. In some

dictionaries, entries are in the form of hierarchical outlines, and the numbering and lettering

indicate subdivision of senses (Atkins 1992:19).

Webster's Third New International|Text

vt

1. (a) to perceive by the eye, apprehend through sight

opens his eyes to see the sunlight coming in through the window

(b) to perceive as if by sight

it was wonderful what that boy saw who was blind|Stuart Cloete

(c) to detect the presence of

the supersonic streamlining of this vehicle makes it di�cult to see by radar|

L.N.Ridenour

2. (a) to have experience of: UNDERGO

saw sea duty on a minesweeper|Current Biog.

if anyone keeps my word, he will never see death|Jn 8:51 (RSV)

openings for keen, practical, �nal year student to see dairy cattle and small-

animal practice|Veterinary Record

seen better days ; see life

(b) to learn or �nd by observation or experience: come to know: DISCOVER

a point of view which I have since seen cause to modify|John Buchan

(c) to �nd out by investigation: ASCERTAIN

see if the hat �ts ; see if the car needs oil

see who's at the door


(d) to give rise to: be marked by

the late glacial times saw the complete triumph of our ancestral stock|Jacquetta

& Christopher Hawkes

(e) to serve as the setting for : be the scene of : WITNESS

that house saw more worry and unhappiness|Virginia D. Dawson & Betty D.

Wilson

3. (a) to form a mental picture of: VISUALIZE

can still see her as she was twenty years ago

saw her in his dreams

(b) to perceive the meaning or importance of : COMPREHEND, UNDERSTAND

because the frontier gives shape and life to our national myth, we have preferred

to see its story in romantic outline|Dayton Kohler

(c) to be aware of: recognize

planning to �re you tomorrow, because you just can't see a good news story|

Sinclair Lewis

sees the folly of further resistance|T. B. Costain

sees only his faults

(d) to form a conception of: imagine as a possibility: SUPPOSE

can you see me knowing how to furnish a house|Edith Sitwell

was never whipped. . . she was so digni�ed and superior you just couldn't see her

across my mother's lap|Myron Brinig

(e) to have presented for observation or consideration: be made aware of

we saw, in the previous lecture, how the problem arose

(f) to look at from a particular point of view

see oursels as others see us|Robert Burns

(g) to look ahead to: FORESEE

can see the day when a college will not try to cover the whole �eld of liberal

arts|Time

4. (a) to direct one's attention to; put under observation : EXAMINE, SCRUTINIZE

want to see how he handles the problem


(b) i. to inspect or read understandingly (something written or printed) have you

seen the story of yesterday's game

let me see your pass, soldier

seen and allowed

ii. to read of

I saw your appointment in the newspapers

(c) to refer to

for further information, see the documents printed in the appendix

see the explanatory notes at the beginning of the book

(d) to attend or visit as an observer or spectator

see a parade, see a play, see the sights of the city

5. (a) to take care of: provide for

would like him to have enough to see him easily to the end of his days|T.B.Costain

(b) to take care or heed: make sure

see thou say nothing to any man|Mk 1:44 (AV)

see that your wet umbrella is not placed between your seat and the next|Agnes

M. Miall

will see that he is brought up properly

6. (a) to regard as: CONSIDER, JUDGE

the electorate did not see �t to ratify the new frame of government|B.W.Bond

did not see it right to ask for special favors

(b) to prefer to have: allow to happen: WELCOME

would probably see himself shot before he told a deliberate falsehood|J.G.Cozzens

I'll see you dead before I accept your terms

(c) to regard with approval or liking: �nd acceptable or attractive

still can't see the portholes but this is our only complaint in an otherwise clean

design|Walt Woron

hope you'll be able to make her see it|W. S. Maugham

can't understand what he sees in her

7. (a) i. to make a call upon: visit

stopped o� at the o�ce to see his former employer


ii. to call upon or meet with in order to obtain help or advice

see a doctor ; see a lawyer

(b) i. to be in the company of regularly or frequently esp. in courtship or dating

had been seeing each other for a year before they became engaged

ii. to grant an interview to or accept the visit of: meet with: RECEIVE

the president of the bank will see you in a few minutes

sees only a few old friends these days

iii. to meet with for the purpose of in uencing esp. by bribery or pressure

charged that the witness had been seen by the defense

8. (a) accompany, escort

young men would wait to see the young ladies home|Agnes S. Turnbull

(b) to wait upon: be present with

saw her onto the plane

saw him o� at the station

(c) to give continued attention, assistance, or guidance to|used with through

saw a new edition of his book through the press

the sympathy of his friends saw him through this period of grief

9. to meet (a bet) in poker or to equal the bet of (a player): CALL

vi

10. (a) to give or pay attention

see, the train is coming

(b) to look about

stood up and �red his pistol in the air, and the naked Indians came out on the

shore to see|Meridel Le Sueur

11. (a) to have the power of sight: have vision

whereas I was blind, now I see|Jn 9:25 (AV)

he sees poorly with his left eye

(b) to apprehend objects by sight

it was so foggy that he could hardly see


(c) to perceive objects as if by sight

the butter y lightness that was teaching his �ngers to see|Marcia Davenport

12. (a) to grasp something mentally: have insight: UNDERSTAND

this fundamental bias of all thinking. . . is what enables us to see, gives thought

its real use|H. J. Muller

(b) to take note

these aren't ordinary trout, you see|Corey Ford

(c) CONSIDER, THINK

when can I �nish this|let me see

13. (a) to make investigation or inquiry

you'll see about the rates, won't you|Agnes S. Turnbull

(b) to arrive at a conclusion through observation and experience

I can't give you an answer yet, but we shall see

[Discussion of relation of see to its near-synonyms behold, descry, espy, view, survey,

observe, notice, remark, note, perceive, and discern omitted.]

Collocations

14. see about

(a) to attend to

I'll see about parking if you buy the tickets

(b) to think over before deciding

we can't give you an answer now, but we'll see about it

15. see after to attend to or care for

see after the baggage; see after the baby

16. see daylight

(a) to get over the initial di�culties of a problem or undertaking

after �ve years of trying, he began to see daylight


(b) slang, of a bronco rider: to bounce high in the saddle so that daylight can be

seen between the rider and the saddle

17. see for dial chie y Eng. to look for

18. see one's way to �nd a course of action possible or reasonable

think I can see my way to lending you 10 dollars

19. see red to become enraged: lose control of oneself

has an insulting manner that makes others see red

20. see the elephant slang. to gain experience of the world

21. see through to see the true meaning, nature, or character of

pride themselves on seeing through the motives of politicians|Times Lit. Supp.

we have seen through the environment theory as we saw through the race theory|

A.J.Toynbee

22. see to to take care of : attend to

saw to the education of the children|Nancy Mitford

23. see to it to make certain by taking necessary or appropriate action

saw to it that the men in the armed services received higher pay|Current Biog.

Webster's Third New International|Discussion

We will �rst study Webster's Third New International Dictionary (W3NI, Gove

1993), as an example of a monolingual dictionary, because it is one of the most authoritative

dictionaries of English, and the entry for see is one of the longest and most detailed to be

found. We �nd that most of the senses and \sub-senses" listed in the entry correspond to

senses which we have described in Chapter 2, although the relationship is far from one-

to-one. The reasons for the imperfect correspondence between the W3NI and our earlier

analysis will become clear as we discuss the entry in detail. As noted in Section 2.6, see

participates in dozens of collocations (some of which were discussed in that chapter); even

this large unabridged dictionary includes only ten of them. We will discuss only those whose

de�nitions and/or examples are linguistically interesting.


The �rst sense (1a), corresponds to our eye, the most basic sense. The de�nition

of the next sub-sense begs the question of exactly what sort of perception is meant; it is

obviously not simple vision, but is it the recognition of physical objects with the hands or

a marvelous sensitivity to social relations? Neither the example sentence nor the de�nition

makes it clear. The next sub-sense, however, (1c), probably is lexicalized, our sense scan.

Although no de�nition is given for the top level sense which includes all three of these

sub-senses, we note that all involve physical vision, with eyes or other systems.

No such unity can be found among the senses listed under (2). (2a) corresponds

closely to our sense experience, and (2c) corresponds to our sense determine. (2b) seems

rather similar to (3c), corresponding to our sense recognize, except that the example which

is given for (2b) is itself idiomatic (re ected in the omission of the article before the count

noun cause), perhaps related to the family of uses/senses see no reason to, see no need to,

etc., discussed in Section 3.4 on page 121.

(2d) and (2e), at least to judge by the examples, are temporal and locative versions

respectively of our sense setting; the semantics of the temporal and the locative seem close

enough not to require separate senses. The example given in (2d) could be an example of \be

marked by", but it does not seem to �t the de�nition \give rise to", unless one conceives of

\glacial times" as actively fostering human evolution. There may well have been a stage in

the development of the setting sense in which \PLACE saw EVENT" was metonymic for

\PEOPLE in PLACE saw EVENT". But it now has an independent existence, as shown by

this example; by de�nition, there were no people outside of our \ancestral stock" to witness

its triumph. Also, note that in (2e), the synonym WITNESS is itself ambiguous between

a sense like eye/process and a sense like setting, in addition to its legal senses. Such a

range of meaning seems to be language speci�c; the nearest equivalents in Spanish, Chinese,

and Japanese do not have a similar range.

(3) can be considered mainly a mixture of examples of our senses envision and

recognize. (3a), (3d), and (3g) are all envision. Note that the irrealis semantics charac-

teristic of envision can be expressed in various ways, by can plus the negative, metaphor-

ically by the day when, use with the future tense (will not try), in his dreams . etc. The

choice of one out of several linguistic forms to express a given situation conveys additional

information by conversational implicature (Levinson 1983:132-147). In this case, the use

of can at all rather than nothing suggests that the ability to see something is in doubt.

Furthermore, one of the non-negated can's is clearly a rhetorical question (can you see


me knowing . . . ), probably to be answered in the negative; some related constructions are

discussed in Section 3.4 on page 121.

(3c) and (3e) are both instances of recognize; the latter is a formal use associated

with the rhetoric and pragmatics of formal discourse, but we need not consider it a separate

sense (but cf. (4c) below). (3b) is an example of classify. (3f) is di�cult to classify; it

might be considered an example of eye or recognize, depending on how literal we take

the poet to be; it could also be considered classify, although the clause introduced by as

is not equivalent to the phrase introduced by as in the construction see X as Y.

(4) also combines several senses. In (4a), the example would fall under determine,

and does not imply any action on the part of the seer. The other three sub-headings do

involve intentionality on the part of the seer, although they contain a mixture of senses.

The distinction between (4(b)i) and (4(b)ii) is an important one which we have distinguished

as read and news, since the complement of the former is the physical object which is read

or its topic, while the complement of the latter is the event or proposition expressed by the

written material. (4c) is a specialized use of read typical of formal writing, and invariably

in the imperative. (4d) we have treated as two separate senses, the �rst two examples as

spectate, and the third as tour.

(5b) is clearly our ensure, but (5a) is not covered by any of the senses we have

discussed so far, unless we treat it as a metaphorical extension of accompany. A similar

extension from the collocation see out is noted in the New Oxford Dictionary of English

(NODE, Pearsall & Hanks 1998) as meaning \(of an article) last longer than the remainder

of someone's life" in British English. This metaphorical extension may well be lexicalized

for some U.S. speakers, too.

It is di�cult to see the rationale for the grouping we �nd in (6). (6a) contains

two examples, the �rst an instance of the collocation see fit, which may be related to

classify, but is best treated as a construction in its own right. The second example did

not see it right to ask special favors is not attributed, and no examples of this pattern

are found in the BNC; given the syntactic di�erence from see �t , these two should not be

grouped together, and the latter may not really exist.

(6b) seems to be identical with the collocation listed in the NODE (see someone

damned �rst), except for the order of the two conditions. These are combinations of con-

structions, as discussed on page 123. The use of the word welcome is at best misleading

as to the meaning of the examples. (6c) is also dubious; the �rst two examples are highly


ambiguous, but may well represent envision and recognize respectively. The third is an

idiomatic use of classify.

The senses in (7) all have to do with meeting socially. (7(a)i) is our visit and

(7(a)ii) our consult; (7(b)i) is dating. The �rst example under (7(b)ii) is audience,

discussed on page 100 as a compositional use; several dictionaries use the example of re-

ceiving an ambassador for this use. The second example, however, along with the second

de�nition \meet with", is ambiguous between visit and audience. (7(b)iii) seems to be

an instance of visit or consult which has been over-interpreted by the lexicographer; the

word see in this sentence conveys only the \meet with" part of the de�nition. The rest of

the de�nition was no doubt clear in the larger context which the lexicographer could see,

but is mysterious to the readers who see only the phrase cited in the dictionary.

(8a) is a use of accompany, as is the �rst example in (8b). The following example

is the collocation see off, and (8c), the collocation see x through. (9) is a sense we have

called gambling; it is found in all the large dictionaries, but is extremely rare in actual

corpora, probably because it occurs primarily in dialogue during a card game.

Many of the senses listed separately under the heading \vi" in W3NI are identical

to those listed under \vt"; this is a result of a fundamental problem in methodology. While

syntactic factors must be given due weight in distinguishing senses, it is also important to

recognize that some of the senses listed here as transitive can also occur where the com-

plement is understood in context but unexpressed. In Construction Grammar terminology,

they represent either de�nite null instantiation (DNI) or inde�nite null instantiation (INI).

For example, in (10b), the \naked Indians" presumably did not come out on the shore

merely to \look about" but to determine what had happened, or what had caused the

noise; this is DNI, because the complement of see can be rather clearly inferred from the

context. (11b) is an instance of eye with INI, the complement understood as something

like \anything around him"; likewise, (13b) can be treated as determine with INI.

(10a) seems to be a reduction of the question \Do you see?", pronounced with rising

intonation; the DNI complement is either \that the train is coming" (i.e., recognize), \the

train coming" (process) or \the train" (eye). Note that in the de�nition, \to give or pay

attention", the object attended to is also null instantiated. (11a) is our sense faculty,

while the example in (11c) seems to be a metaphorical extension of eye, not a separate

sense.

(12a) is our sense recognize. The examples given in (12b) and (12c) are both


collocations in our treatment, y'see and let's see respectively. As discussed in Chap-

ter 2, both these senses inherit the basic semantics of recognize but have very di�erent

pragmatics. (13b) seems to be identical to (14b), and should be classi�ed as instances of

determine.

With regard to the collocations at the end of the entry, there are no examples of

see after (15) in the BNC; it may be a blend of look after and see to occurring in speech

but not actually lexicalized. There are 290 examples of the collocation see for (17) in the

BNC, but only seven could possibly have the meaning indicated here, and all of those are

ambiguous with other known senses. There are no examples of see the elephant (20). Given

the fact that speakers must have knowledge of vast numbers of very low frequency lexical

items, it is di�cult to say in what sense these collocations exist as lexical units in the

language, shared by a speech community.

Clearly, see to (22) and see to it (23) are both instances of ensure, the latter

taking a that-clause (which can be null instantiated) as its complement, and the former

an NP, in circumstances in which the action to be performed is clear in context (e.g. in

(22), the equivalent that-clause would be something like see to it that the children receive a

good education, which might entail intermediate actions such as registering the children in

appropriate schools, reading to them, paying their tuition, helping them with homework,

training them in the family trade, etc.). But ensure can also be expressed without the

words to it before the that-clause (cf. 5b).

In summary, we �nd that the sense divisions and subdivisions in the W3NI seem to

have quite di�erent statuses. One of our senses may be scattered across several dictionary

senses or one dictionary sense may include several of ours. The de�nitions themselves are

often not very helpful; it would be daunting to distinguish among \to learn or �nd by

observation or experience" (2b), \to �nd out by investigation" (2c), and \to perceive the

meaning or importance of" (12a). Usually, the examples make clear what is intended by

the de�nition, rather than the other way around. The reader can learn something from

the dictionary, but it is a highly interactive process, depending heavily on what the reader

already knows. While lexicographers must obviously depend on reader's general world

knowledge in many ways, explicitly providing information about both the semantics and

the syntax of each sense is helpful for readers from very di�erent cultural backgrounds and

will be essential for dictionaries to be used in NLP.

We note that this process is very di�erent for entries de�ning less common words,


especially technical terms. If one looks up petasus and �nds the de�nition \a broad-brimmed

hat worn by ancient Greek travelers and hunters, often represented in art as a winged hat

worn by Hermes or Mercury", one has been informed very succinctly of exactly what one

needed to know. This is also in part due to the fact that nouns are generally easier than

verbs to de�ne (and when necessary to draw a picture of, as might be done for petasus).

Small English Dictionaries

We already noted that the lexicographer is inherently constrained to produce con-

cise entries, even for highly polysemous words such as see. The extreme case of this is the

pocket dictionary; when constrained to produce a very short list of senses, will the lexicog-

rapher be able to express the major distinctions? Let us examine, by way of examples, the

sense divisions in three small English dictionaries. The �rst is Merriam-Webster's Pocket

Dictionary (1995) in which we �nd the following entry:3

(151) Merriam-Webster's Pocket Dictionary

a. perceive by the eye

b. have experience of

c. understand

d. make sure

e. meet with or escort

The �rst two correspond to eye and experience. Understand presumably encompasses

our senses recognize, process, and condition. Make sure is ambiguous between ensure

and determine. Finally, the two social senses visit and accompany are combined in the

last item as a disjunction.

The entry from the second dictionary, Webster's New World Dictionary (1997), is

shown below:

(152) Webster's New World Dictionary

a. look at

b. understand

3In the remainder of this chapter, I will leave gaps in the numbering of examples to mark transitionsbetween languages and between large and small dictionaries.


c. �nd out

d. make sure

e. escort

f. meet; visit with

g. consult

h. have the power of sight

Even in this brief entry most of the major senses are listed. determine and ensure are

distinguished, as are accompany, visit, and consult, and faculty is added. Although

we may consider understand to be too broad, as in the dictionary above, this entry covers

the ground very nicely.

Finally consider the Concise Oxford Dictionary (Allen et al. 1990), shown below:

(153) Concise Oxford Dictionary.

a. perceive with the eyes.

b. deduce after re ection or from information.

c. regard in a speci�ed way.

d. meet (someone one knows) socially or by chance.

e. escort to a speci�ed place.

f. attend to.

g. ensure.

h. (in poker or brag) equal the bet of (an opponent) and require them to reveal

their cards to determine who has won the hand.

Like the Webster's New World, there are eight sense divisions,but they are much

less clear; (153-b) is vague and confusing; (153-c) may be intended to mean classify or

condition, but it is hard to tell which. In (153-d), one assumes that the intended sense

is visit; by the de�nition we're using, one can do this either by appointment or by chance;

therefore, the words socially or by chance do not seem to add any meaning. The distinction

between (153-f). and (153-g) is also unclear. It is surprising to �nd the gambling sense in

such a small dictionary.

All three dictionaries attempt to cover something of the broad semantic range of

see, and none has space for examples or discussion. Although the lengths of the entries


are comparable, the Webster's New World Dictionary does a much better job on the sense

divisions.

5.3 Bilingual Dictionaries|Introduction

Most bilingual dictionaries apparently try to make all and only the sense divisions

that will be relevant in translating to the target language. But they do not provide just

one translation or paraphrase for the sense divisions which they make; each sense division

in a bilingual dictionary typically corresponds to several possible translations in the target

language. For example, consider the use of see which occurs in (154).

(154) a. The last time I saw Paris. . .

b. You must go up there and see the factory before you decide.

We treat this as a use of physical vision (eye), and consider that the tour sense which

see has in the sentences is the result of the composition of the meaning of the verb with its

object, i.e., a place which is to be seen in its entirety, especially a tourist destination. One

can argue, however, that this should be treated as a separate sense, on the basis of separate

negation, as in Ex. (155).

(155) I've never really seen Paris, although I saw it from an airplane as I ew to Frankfurt

once.

In the New English-Chinese Dictionary (Editing Group 1988) for example, the fourth sense

division in the entry for see includes tour along with spectate. Three examples are given:

(156) a. see a play

b. go and see our new workshop

c. see the sights of Hangzhou

The �rst two are translated with phrases using k�an, (156-c) with y�oul�an. But the �rst is

translated by an idiom in Chinese, a verb-object compound k�anx�� (`look drama'|go to a

play or opera), and (156-b) is translated by k�ank�an, the reduplication indicating intention-

ality or purposiveness. Y�oul�an is a two-verb compound which can be glossed as `tour-view'.


Finally, a good bilingual dictionary will have idiomatic translations of its exam-

ple sentences, but a good translation of a sentence does not necessarily include a good

translation of each sense of each word in it. It is important for students to learn that

word-for-word translations are often clumsy and sometimes impossible, and the bilingual

dictionary is supposed to demonstrate that fact.

It may help to think of the problem in terms of Frame Semantics. In some cases,

there are very similar semantic frames across two cultures and the translation problem

is a matter of �nding equivalent linguistic forms for them. For example, the commercial

transaction frame and many more speci�c frames derived from it will be found in the

cultures of all developed countries; we just need to learn that English for rent is French �a

louer and Chinese ch�uz�u, and we will be able to recognize advertisements for apartments.

In other cases, the frames are quite di�erent across cultures, and the problem of translation

are much deeper; e.g. it is very unlikely that the American culture-speci�c �sh story can

be simply translated as `�sh' + `story' anywhere else in the world, and the culture-speci�c

Japanese term yagou `house/room-tag/number' has to be rendered by an explanatory phrase

in English, `Kabuki actor's stage name'.

5.4 English-Spanish Dictionaries

Collins English-Spanish Dictionary|Text

(201) |vt

a. (gen) ver

(i) see page 8 v�ease la p�agina 8

(ii) he's seen a lot of the world ha visto mucho mundo

(iii) he'll not see 40 again los 40 ya no los cumple

(iv) I can see to read veo bastante bien para poder leer

(v) to see sb do sthg ver a uno hacer algo

(vi) I saw him coming lo v�i venir

(vii) he was seen to fall se le vio caer

(viii) I saw it done in 1968 lo v�i hacer en 1968

(ix) I'll see him damned �rst antes le ver�e colgado

(x) there was not a house to be seen no se ve�ia ni una sola casa


(xi) this dress is not �t to be seen este vestido no se puede ver

(xii) he's not �t to be seen in public no se le puede presentar a los ojos

del p�ublico

b. (accompany) acompa~nar

(i) to see sb to the door acompa~nar a uno a la puerta

(ii) to see a girl home acompa~nar una chica a su casa

(iii) may I see you home? >puedo acompa~narte a casa?

(iv) he was so drunk we had to see him to bed estaba tan borracho que

tuvimos que llevarle a la cama

c. (understand) comprender, entender, ver

(i) I don't see why no veo por qu�e, no comprendo por qu�e

(ii) I fail to see how no comprendo c�omo

(iii) as far as I can see seg�un mi modo de entender las cosas; a mi ver

(iv) this is how I see it �este es mi modo de entenderlo. yo lo entiendo as��

(v) the Russians see it di�erently los rusos lo miran desde otro punto

de vista, el criterio de los rusos es distinto

(vi) I don't see it (�g) no creo que sea posible; no veo c�omo se podr��a

hacer

d. (look, learn, perceive) mirar; observar; percibir

(i) I saw only too clearly that . . . percib�� demasiado bien que . . .

(ii) I see in the paper that . . . veo en el peri�odico que . . .

(iii) Did you see that Queen Anne is dead? >has oido que ha muerto

la reina Ana?

(iv) I see nothing wrong in it no le encuentro nada indebido

(v) I don't know what she sees in him no s�e lo que ella le encuentra

e. (ensure)

(i) to see (to it) that. . . procurar que + subj; asegurar que + subj

(ii) see that he has all he needs cuida que tenga todo la que necesita

(iii) see that it does not happen again que no vuelva a ocurrir

(iv) see that you have it ready for Monday procura tenerlo listo para

el lunes

(v) to see that sth is done procurar que algo se haga

f. (visit. frequent) ver, visitar


(i) to see the doctor consultar al m�edico

(ii) I want to see you about my daughter quiero hablar con Vd. acerca

de mi hija

(iii) what did he want to you about? >qu�e asunto quer��a discutir con-

tigo?; >qu�e motivo tuvo su visita?

(iv) we don't see much of them nowadays ahora les vemos bastante

poco

(v) we shall be seeing them for dinner vamos a cenar con elias

(vi) to call (or go) and see sb ir a visitar a uno

(vii) the minister saw the Queen yesterday el ministro se entrevist�o con

la Reina ayer

(viii) I'm afraid I can't see you tomorrow lamento no poder verle man-

ana

(ix) see you soon!, see you later! <hasta pronto!

(x) see you on Sunday! <hasta el domingo!

g. (imagine)

(i) I don't see her as a minister no me imagino verla coma ministra

(ii) I can't see myself doing that no me imagino con capacidad para

hacer eso

(iii) I can't really see myself being elected en realidad no creo que me

vayan a elegir

h. (experience)

(i) he's seen it all est�a de vuelta de todo;

(ii) this hat has seen better days este sombrero ha conocido mejores d��as

(iii) I never thought I'd see the day when . . . parece imposible que

llegara el d��a en que . . .

(iv) she's certainly seeing life es seguro que est�a viendo muchas cosas

(v) we'll not see his like again no veremos otro como �el

(202) | vi

a. (gen) ver

(i) let me see d�ejame ver; (�g) vamos a ver, veamos; a ver

(ii) I'll go and see voy a ver


(iii) see for yourself <v�ealo Vd!

(iv) now just see here <mire!

(v) so I see lo veo, lo estoy viendo

(vi) as far as the eye can see hasta donde alcanza la vista

(vii) from here you can see for miles desde aqu�� se ve muy lejos; desde

aqu�� se domina un gran panorama

(viii) he was trying to see in se esforzaba par ver el interior

(ix) we shan't be able to see out no podremos ver el exterior

b. (understand) comprender

(i) I see lo veo

(ii) I see! <ya!, <ya caigo!, <ya comprendo!

(iii) it's all over, see? se acab�o, >comprendes?

(iv) he's dead, don't you see? est�a muerto, >me entiendes?

(v) as far as I can see a mi ver, seg�un mi modo de entender las cosas

(203) Collocations

a. see about vt

(i) (attend to) atender a, encargarse de

I'll see about it lo har�e, me encargo de eso

he came to see about our TV vino a ver nuestra televisi�on

(and repair) vino a reparar nuestra televisi�on

(ii) (consider)

I'll see about it lo ver�e, lo pensar�e

we'll see about that <es lo que hay que ver!; <y c�omo!

we must see about getting a new car tenemos que pensar en

comprar un nuevo coche

b. see in vt

(i) person hacer entrar, hacer pasar

(ii) to see the New Year in celebrar or festejar (esp LAm) el A~no Nuevo

c. see into vt investigar, examinar

d. see o� vt

(i) (say goodbye) despedirse de

(ii) we went to see him o� at the station fuimos a despedirnos de �el


en la estaci�on

(iii) (send away) the policeman saw them o� el polic��a los acompa~n�e a

la puerta; el polic��a les dijo que se fueran

(iv) (defeat) vencer, cascar; deshacerse de; acabar con the minister saw

the miners o� el ministro acab�o con los mineros

e. see out vt

(i) (person) acompa~nar a la puerta

(ii) to see a �lm out quedarse hasta el �n de una pel��cula, permanecer

sentado hasta que termine una pel��cula

(iii) we wondered if he would see the month out nos pregunt�abamos si

vivir��a hasta el �n del mes

f. see over vt visitar, hacer la visita de, recorrer

g. see through vt

(i) (deal) llevar a cabo don't worry, we'll see it through no te preocupes,

nosotros lo haremos todo

(ii) to see sb through ayudar a uno a salir de un apuro, ayudar a uno en

un trance di�c��l $100 should see you through tendras bastante con

100 libras, con 100 libras estar�as bien

h. to see through sb calar a uno, conocer el juego de uno to see through a

mystery penetrar un misterio

i. see to vt atender a; encargarse de

(i) (repair) reparar, componer

(ii) he sees to everything se encarga de todo, lo hace todo

(iii) the rats saw to that las ratas se encargaron de eso

(iv) to see to it that . . . procurar que + subj, asegurar que + subj

Collins English-Spanish Dictionary|Discussion

This large, standard, well-known translating dictionary (Alvarez Garcia 1998) does

a rather good job of separating word senses. Thus we �nd sense divisions for our senses

accompany (201-b), ensure (201-e), visit (201-f), and experience(201-h), all marked

with the same keywords which we have chosen for these senses, and, for the most part, the

examples within the sub-entries clearly fall within our de�nition of the sense. Our sense


consult is included within visit (201-f), which it inherits; our sense envision is found

with the label \imagine" (201-g).

The �rst sense given (201-a), includes, but goes beyond, our sense eye; it seems

to include several otherwise unrelated senses all of which can be translated by Spanish ver .

Since the purpose of the dictionary is translation, this may not be an unreasonable practice.

However, it might be confusing to many users to �nd very general uses such as (201-a-vi),

(201-a-vii), and (201-a-viii), all of them instances of our sense process mixed in with very

idiomatic uses such as (201-a-ix) and (201-a-iii). (201-a-ii) seems misplaced here; it belongs

in sense experience (201-h).

Like the monolingual English dictionaries, Collins distinguishes the transitive and

intransitive uses of the same sense in a way which we would not �nd necessary. The

examples in the �rst sense in the intransitive section (202-a), like those in (201-a), seem

to have in common only that most of them are translated by the Spanish word ver. In

particular,(202-a-i), (202-a-iv), and (202-a-v) do not typically involve physical vision. The

metaphorical motion uses appear here ((202-a-vi) and (202-a-vii)) because they do not have

direct objects.

The two senses labeled \understand", ((201-c) and(202-b)), include examples of

our senses recognize(201-c-i) and classify (201-c-v) and the\conversation" senses related

to recognize ((202-b-iii)(202-b-i)). As far as I can see mysteriously appears in both the

transitive and intransitive groups. The uses listed under (201-d) are our sense recognize

((201-d-i) and (201-d-iv)) and our sense news ((201-d-ii) and (201-d-iii)).

As noted above, a number of idiomatic uses appear in examples for the major

senses. The converse is also true; a number of the items listed as idioms belong with

the major senses such as see in (203-b-i) and see out (203-e-i) which belong under the

accompany sense and see to it which appears both as a separate collocation (203-i-iv) and

under the sense ensure (201-e) (where it belongs). At least one of the senses of see about

(203-a-i) belongs under ensure also, although the dictionary is probably right from the

translator's point of view to treat \consider" as a separate sense; note that the phrase I'll

see about it , without further context, is ambiguous between the two. The treatment of the

other senses of see out, as well as see o�, see through X and see X through is good.

In general then, the sense divisions of the Collins dictionary are reasonable, and

where they have chosen to combine uses that we have regarded as separate senses, they have

generally followed the relations represented by inheritance in our description. In many cases


it must be a di�cult decision for lexicographers whether to treat an idiom as an example of

a more general sense or to list it separately among the collocations at the end of the entry.

The most noticeable omission is the sense we have called determine, which does appear in

some other English-Spanish dictionaries, usually translated averiguar. Is also notable that

there is no treatment of uses such as see combat and see use as, which we have grouped

under sense experience.

A Pocket Spanish-English Dictionary

By way of comparison, let us consider a very small Spanish-English dictionary, the

University of Chicago Spanish-English Dictionary (Castillo & Bond 1972). The entire entry

is shown below:

(251) ver

(252) Collocations

a. see that you do it no deje Vd. de hacerlo; tenga Vd. cuidado de hacerlo

b. I'll see to it me encargar�e de ello

c. let me see a ver

d. to see a person home acompa~nar a una persona a casa

e. to see a person o� ir a la estaci�on para despedir a una persona

f. to see a person through a di�culty ayudar a una persona a salir de un

apuro

g. to see through a person adivinar lo que piensa una persona; darse cuenta

de sus intenciones

h. to see to one's a�airs atender a sus asuntos

i. to have seen military service haber servido in el ej�ercito

The treatment of collocations is fairly good, considering the limitations of space

involved, but it seems surprising that only the single Spanish word ver has been given in the

main entry. The collocations suggest the existence of some senses, such as ensure (in three

syntactic patterns: (252-a) (252-b), and (252-h)), accompany (252-d), and experience

(252-i). But there is nothing to suggest that see also has the senses recognize, visit, and

envision; note that ver will cover some but not all of the uses of visit, as shown by the


examples in the Collins English-Spanish Dictionary above. The addition of just the four

words comprender, entender, visitar, and imaginar after the �rst occurrence of ver would

have greatly improved the entry.

5.5 English-Mandarin and Mandarin-English Dictionaries

Far East English-Chinese Dictionary|Text

(301) { v.t.

a. k�an; ji�an; sh�� `look; see; view'

(All translated k�an ji�an except as noted)

(i) Do you see that tree?

(ii) I saw some people in the garden.

(iii) Have you seen today's paper? k�and�ao

(iv) I looked but saw nothing.

(v) I have never seen a man hanged. w�ei ji�an gu�o

(vi) I saw something move (or moving).

(vii) I saw that the box was empty. k�an ne�� h�ezi sh�� k�ongde

(viii) to see a tennis match. k�an y��ch�ang w�angqi�u b��s�ai

b. li�aoji�e; l��ng hu��. (=understand); ch�aju�e (= perceive).

(i) I see what you mean. m��ngb�ai

(ii) I don't see the use of doing that. k�anb�uch�u `can't see'

(iii) Don't you see how to do it? b�um��ngb�ai

c. f�axi�an; xu�e zh��. `discover; �nd out'

(i) I will see what needs to be done.

W�o hu�� ch�ak�an x�uy�ao zu�o sh�emma.

(ii) See if the postman has come.

Q�u k�ankan yo�uch�ai l�aigu�o m�ei yo�u.

d. j��ngy�an; yu�el�� èxperience'

(i) He has seen a lot in his life. yu�el�� h�en du�o

(ii) He has seen better days. yo�ugu�o d�ey�� sh��ho�u

(iii) I never saw such rudeness. w�ei j��nggu�o

(iv) That coat has seen hard wear. chu�ande h�en ji�ule `wear-DE very old-


PERF'

e. hu��mi�an; w�uji�an (=meet); f�angw�u (=call on); sh�angt�an (=talk with); ji�eji�an

(= receive a call from).

(i) It's a long time since I saw you last. ji�anmi�an

(ii) I used to see much of my neighbor. ji�and�ao

(iii) I'll see Thomas about it. zh�ao... t�ant�an

(iv) He wishes to see you alone. t�any��t�an

(v) I think you ought to see a doctor. zh�ao... k�ank�an

(vi) He is too ill to see anyone. ji�eji�an

f. h�us�ong; zh�aog�u èscort; care for'

Who's going to see Miss Green home? s�ong

g. c�anji�a; c�angu�an `participate in; take a tour of'

We saw the World's Fair. c�angu�an

h. r�ang; y�unx�u `let; allow'

You can't see your sister starve without trying to help her! r�ang

i. zh�uy��; f�uz�e `pay attention to; take responsibility for'

(i) See that the defect is made good. zh�uy��

(ii) See (to it) that the door is fastened before you go to bed. zh�uy��

(iii) I will see that you are paid. w�o b�� f�uz�e sh�� n�� n�a d�ao qi�an

j. q��d�ai; d�engd�ai d�ao zu�� ho�u `hope; wait until the very last'

He said that whatever happened, he would see the struggle through.

f�endo�u d�ao d�� `struggle to the end'

k. (p�aix��) ('Card games') xi�a t�ongy�ang du�ode d�uj��n `put down an equal bet'

(302) { v.i.

a. k�an; ji�an; sh�� `look; see'

(i) It's dark; I can't see. k�anbuji�an

(ii) On a clear day, we can see miles and miles from this hilltop. k�and�ao

h�en yu�an

(iii) Can you see to write? k�anded�ao

b. j�uyo�u sh��l�� `have the power of sight'

The blind do not see. k�anbuji�an

c. ch�ak�an òbserve'


Go and see for yourself if you don't believe me! ch�ak�an

d. k�aol�u `consider'

Now, then, let's see. r�ang w�o k�aolu k�aolu

e. li�aoji�e ùnderstand'

(i) I see. m��ngb�aile

(ii) See? d�ongbud�ong

f. zh�uy��; li�ux��n `pay attention; take care'

(303) Collocations

a. to see about

(i) zh�uy�� He promised to see about the matter.

(ii) k�aoch�a; ch�ax�un

b. to see after (=look after).

c. to see into

I must see into it.

d. to see one's way

e. to see out w�anch�eng; gu�anch�e `to complete; to adhere to (principles, etc.)'

f. to see over

We want to see over the house before we decide to rent it.

g. to see red

(i) sh�engq��; f�an�u `to be angry, enraged'

(ii) `desirous of damaging or attacking other people'

h. to see service

(i) `to become an expert through much experience'

(ii) `worn out due to long use'

i. to see somebody o�

We went to the station to see him o�.

j. to see the back of

I shall be glad to see the back of that fellow, he is a nuisance.

k. to see the last of

I shall be glad to see the last of this job.

l. to see things

m. to see through


(i) k�anto�u; zh�enzh�eng li�aoji�e

I see through your little game.

(ii) `to hold on until the very end'

n. to see to (=to take care of).

o. She will see to everything herself.

p. He would see to it that his sons took a livelier interest in politics.

q. to see visions

(304) { interj.

See, here he comes!

Far East English-Chinese Dictionary|Discussion

The partial text of the entry for see in the Far East English-Chinese Dictionary

(FEECD, Liang et al. 1975) is given beginning on page 187; rather than include a transliter-

ation of all the Chinese de�nitions, we have translated them back to English again, except

where it is necessary to discuss the choice of Chinese lexical item in detail. The sense

divisions are relatively straightforward:

(301-a) has to do with physical vision; besides eye, it includes condition, pro-

cess (301-a-v) and spectate (301-a-viii). (301-a-vii) is probably recognize, but with

emphasis on physical vision as the source of knowledge.

(301-b) is recognize, and the most common word for inchoative cognition,

m��ngb�ai , is used to translate two of the examples. Note that (301-b-ii) is translated by

k�an in a negative resultative compound, directly parallel to English \can't see the use".

(301-c) is determine; the word sh�emma `what' and and the pattern V m�ei

y�ou (Yes/No Question) can be used to form either direct or indirect questions. (301-d)

is almost exactly our experience, with both human and inanimate experiencers. Two of

the examples, (301-d-i) and (301-d-iii) have words that directly translate èxperience', but

the other two express the concept di�erently, with `have-GUO' (301-d-ii) and a resultative

complement `wear-old' (301-d-iv).

(301-e) is a group of social interaction senses, including our visit and consult

(301-e-v) and the use we noted previously, `to receive visitors' (301-e-vi). Several are simply

rendered `to talk with', sh�angt�an or (less formally) t�ant�an.


Many of the other sense divisions correspond nicely to ours: (301-f) is accom-

pany, (301-i) is ensure, (301-j) is the collocation see x through, and (301-k) is gam-

bling. (301-g) is the tour sense; the translation c�angu�an is a compound of `participate'

and `see', meaning `to take a tour of, inspect'.

(301-h) is an instance of condition, but translated with r�ang àllow', which

makes sense in this example, but would not work more generally (I just want to see you

happy again). As mentioned in Section 5.3, a good translation of a sentence containing see

does not always include a good translation of that sense of the verb see per se.

As with the monolingual dictionaries, many of the intransitive sense divisions are

exactly equivalent to transitive ones, except for the null instantiation of the complement.

Thus (302-a) = (301-a) eye with DNI (but includes the metaphorical motion discussed in

Section 3.3), (302-c) = (301-c) determine, and (302-e) = (301-b) recognize. (302-b) is

faculty and (302-d) is the collocation let's see. It is hard to tell what (302-f) means

without an example.

Most of the collocations are similar to those in monolingual dictionaries, including

some British idioms e.g. (303-f). (303-g-ii) is puzzling, and there seems to be some confu-

sion about (303-h); experience does not necessarily imply either expertise in persons or

wear and tear on objects, although both of these may result from long service. There also

seems to be a tendency to restrict both see something through and see something out to

`maintaining one's principles to the very end', not allowing for cases where, e.g. money is

su�cient for one's trip. Thus, we see that, in general, the sense divisions of the FEECD

are quite reasonable and are either equivalent to ours or to groups of our senses. The table

below summarizes the organization of the FEECD entry in terms of our sense names; the

sense divisions of all the bilingual dictionaries will be discussed in Section 5.7.


(305) {v.t.

a. eye, condition, process, spectate

b. recognize, classify

c. determine

d. experience

e. visit, consult, audience

f. accompany

g. tour

h. condition

i. ensure

j. see x through

k. gambling

(306) { v.i.

a. eye, faculty

b. faculty

c. determine

d. let's see

e. recognize, discourse

f. ??pay attention; take care

{ 16 collocations, including see x through,

see through x, see off, see about

(= determine) and see to (it that) (= ensure).

Summary of Sense Divisions in the Far East English-Chinese Dictionary

Everyday Chinese-English Dictionary|Text

k�an

(351) a. see; look at; watch

(i) k�an di�any��ng see a �lm; go to the movies


(ii) k�an di�anshi watch TV

(iii) k�an qi�us�ai watch a ball game

b. read

(i) k�an b�ao read a newspaper

(ii) k�an sh�u read (a book)

c. think; consider k�an q��ngx��ng sh�� make a correct appraisal of the situation

d. look upon; regard

e. treat (a patient or an illness)

f. look after k�ang�u take care of

g. call on; visit; see m��ngti�an q�u k�an t�a go and see him tomorrow

h. depend on k�an ti�anqi depend on the weather

(352) Compounds and collocations

a. k�anb��ng (of a doctor) see a patient /(of a patient) consult a doctor

b. k�anb�ugu�an cannot bear the sight of; frown upon

c. k�anbuq�� scorn; despise

d. k�anch�eng look upon as

e. k�anch�u make out; see

f. k�anchu�an see through

g. k�anf�a view [n. {CFB]

h. k�an f�eng sh�� du�o trim one's sails [lit. look wind use rudder {CFB]

i. k�anji�an catch sight of; see

j. k�anl�ai it seems; it appears; it looks as if

k. k�anq��ng underestimate, look down upon

l. k�ansh�ang take a fancy to (a girl, etc.); settle on

m. k�ant�ai bleachers; stand

n. k�anw�ang call on; visit; see

o. k�anzh�ong regard as important; value

p. k�anzu�o look upon as; regard as

ji�an

(353) a. see; catch sight of su�o ji�an su�o w�en what one sees and hears

b. meet with, be exposed to


c. show evidence of; appear to be ji�anzh�� y�u x��ngd�ong be translated into action

d. refer to; see

e. ji�ansh�ang see above

f. ji�anxi�a see below

g. meet; call on; see n�� ji�and�ao t�a m�eiyo�u did you meet him

h. view; opinion

(354) Compounds and Collocations

a. y�� w�ozh�� ji�an in my opinion

b. ji�an ch�� insight; judgment

c. ji�an du�o sh�� gu�ang experienced and knowledgeable

d. ji�an gu�ai mind; take o�ence

e. ji�an j�� as the opportunity arises; according to circumstances

f. ji�an j�� x��ng sh�� do as one sees �t

g. ji�anji�e view, opinion; understanding

h. ji�an mi�an meet; see

i. ji�an sh��mi�an see the world; enrich one's experience

j. ji�anshi widen one's knowledge; enrich one's experience/experience; knowl-

edge; sensibleness

k. zh�ang ji�anshi widen one's knowledge

l. ji�anw�en knowledge; information

m. jin�ax�� learn on the job

n. jianx�� j��sh�uyu�an technician on probation

o. ji�anx��sh�eng probationer

p. ji�anx�� y��sh�eng intern

q. ji�anxi�ao become e�ective; produce the desired result

r. ji�anzh�eng witness; testimony

s. ji�anzh�engr�en eyewitness, witness

k�an

(355) a. look after; take care of; tend

(i) k�an h�aizi look after children

(ii) k�an gu�a keep watch in the melon �elds


(iii) k�an ni�u tend cattle

(iv) k�an j��q�� mind a machine

b. keep under surveillance keep an eye on sb.

(i) k�angu�an look after, attend to/watch; guard (prisoners, etc.)

(ii) k�anh�u nurse (the sick)/hospital nurse

(iii) k�an ji�a mind the house

(iv) k�an m�en guard the entrance/look after the house

(v) k�ansho�u watch; guard (storehouse prisoners, etc.)/turnkey, warder

gu�an

(356) a. look at; watch; observe gu�an r��ch�u see the sunrise

b. sight; view w�aigu�an outward appearance

c. outlook; view; concept

(357) Compounds and Collocations

a. gu�anch�a observe; watch; survey

b. gu�anch�a d�ongj��ng watch what is going on

c. gu�anch�aji�a observer

d. gu�anch�asu�o observation post

e. gu�anch�ayu�an observer

f. gu�andi�an point of view, viewpoint; standpoint

g. gu�ang�an impressions

h. gu�angu�ang sightseeing; visit; tour

i. gu�angu�angtu�an visiting group

j. gu�angu�angzh�e sightseer

k. gu�ank�an watch; view

l. gu�anl�� attend a celebration or ceremony

m. gu�anl��t�ai reviewing stand

n. gu�anmo�u view and emulate

o. gu�anni�an sense; idea; concept

p. gu�anni�an x��ngt�ai ideology

q. gu�ansh�ang view and admire; enjoy the sight of

r. gu�anw�ang wait and see; look on


s. gu�anzh�ong spectator; viewer; audience

Everyday Chinese-English Dictionary|Discussion

In this section, we take a brief look at the translation process in reverse, looking

at the entries for four of the Chinese characters most commonly used to translate English

see and the range of meanings they show in a medium-sized Chinese-English dictionary

Everyday Chinese-English Dictionary (Ch�angy�ong H�an-Y��ng C��d��an) (1986). The entries

are for k�an `look' j��an `perceive', k�an `look after' and gu�an òbserve'.

One di�culty in working with most Chinese-English dictionaries is in �nding all

the words and phrases which include a particular character, but do not begin with it. Since

they are organized strictly by character, words and phrases which have a particular character

as their second, third, etc. member will not be easy to �nd. A similar di�culty occurs with

English dictionaries, e.g., if one wanted to �nd out the range of meanings of suit , it is unlikely

that one would �nd the collocation birthday suit `naked' unless one already knew it. The

problem is more acute in Chinese, both because of the analytic character of the language

(many more words are transparent compounds of two or more morphemes) and because of

the Chinese penchant for proverbs (ch�engy�u) consisting of three or four morphemes. Thus,

looking at the entries for k�an, ji�an, and gu�an will not lead us to phrases such as hu�o ji�an

gu�� `living see ghost|imagining things' and g�e �an gu�an hu�o òther bank.of.river observe

�re|look on others' problems with indi�erence'.

Since Chinese is more analytic than English, it is not surprising that the most usual

way to express `see' in Mandarin is with the compound k�anji�an (352-i), `look perceive'. The

many compounds that express aspects of the meaning of see will generally use either k�an

or ji�an (depending on which aspect of the event is pro�led, the volitional or the passive

experience). But the choice of k�an or ji�an also involves a di�erent event type; k�an is an

activity with no implied accomplishment, whereas ji�an is the telic, punctual reception of a

sense impression.

K�an is thus basically closer to English look than to see, as most of the examples in

(351) demonstrate. But k�an expresses some of the senses of see involving conscious mental

activity, such as classify (352-d) `look-become', (352-p) `look-make' as well as more speci�c

verbs of judging, such as (352-k) `look-light|underestimate', (352-o) `look-heavy|regard

as important' and even (352-c) `look NEG rise|despise'. The mental activities associated


with judging extend the use of k�an to a variety of words such as (352-f) `look-pierce|

see through x' and (352-g) `look-way|point of view, opinion'. K�an as mental \looking"

leads to (351-h) ìt depends on' (N.B. not I depend on).

visit is also often expressed by k�an, as is consult, both involving agents, rather

than just experiencers. The dictionary says that (352-a) `look-sick' can be used in either

direction, but at least some speakers �nd it more natural in the sense `consult a doctor'

than `see a patient'.

Ji�an emphasizes the experience of perception and the knowledge thus gained. In

(354-g) `see-understand' and (354-j) `see-be.acquainted', the perception and the knowledge

are represented by the �rst and second morphemes; (354-l) combines `see' and `hear' to mean

`knowledge', relation that is made more explicit in (353-a). (354-m) `see-study' combine to

mean àpprentice', ìntern' ((354-p) for medical interns).

The most common way of saying `meet with, encounter' is ji�an mi�an (354-h) which

is necessarily explicitly reciprocal; the participants are expressed as \A with B" or a plural

(usually dual) subject such as \you two". Why ji�an should be preferred to the more active

k�an to express the vide sense is not clear.

K�an is written with the same character as k�an, but pronounced with a di�erent

tone. There are an number of such pairs of words, sharing a written form, somewhat

semantically related, and di�ering only in tone. Although the process is no longer productive

in Mandarin, these are apparently the result of su�xation in Old Chinese of which only the

tone di�erences remain. As the examples in (355) make clear, k�an usually means `watch,

tend, guard' and forms a verb-object compound with the thing or place guarded.

Finally, gu�an is an older verb, which meant simply `look at' in Middle Chinese, but

now has a more formal sense òbserve, view'. It is rather tightly bound in these compounds

and does not readily simple resultatives as k�an and ji�an do as shown in Ex. (358)

(358) a. N� k�an/ji�an/*gu�an d�ao t�a me��you?

b. You see reach him YN-QN

c. Did you see him?


5.6 English-Japanese Dictionaries

Kenkyusha English-Japanese Dictionary|Text

(401) | vt.

a. miru, . . . ga mieru (perceive by the eye); (ki o tsukete) miru (look at):

(i) I see some people in the garden. niwa ni s�unin no hito ga mieru

(ii) What do [can] you see? Nani ga miemasu ka

(iii) See page 5 [ch. 10]. 5 peeji [10 sh�o] o miyo (cf. vide)

b. (Yume nado ni miru): I saw him in a dream.

(i) see things genkaku (hallucinations) o okosu

(ii) see snakes ☞ snake (n.)

(iii) see stars ☞ star (n.) (idiom)

c. (geki, eiga, meisho nado o) miru (view), kanran [kenbutsu] suru (be a specta-

tor of), mi ni iku (visit):

(i) go to see a show misemono o mi ni iku

(ii) He has gone to Italy to see Rome. R�oma kenbutsu ni Itaria e itta

(iii) Have you ever seen (=been to, visited) France? Fransu e itta

koto ga arimasu ka

(iv) see the sights meisho o kenbutsu suru

d. (hito ni) au, deau (meet:), kaiken suru, menkai suru (interview):

(i) I haven't seen you for ages. Zuibun nagaiaida oai shimasen deshita

ne (Hisashiburi desu ne)

(ii) I am too busy to see you now. Ima wa isogashikute omeni kakarenai

(iii) I'm very pleased [glad, happy] to see you. Oai shite taihen ure-

shii(y�okoso irasshai mashita)

(iv) I have seen nothing of you these days; let's see a great deal of

each other. Chikagoro sappari omikake shinai ga, kongo wa tabitabi

aimash�o yo

(v) see company kyaku o �otai suru.

e. (hito ni) ai ni iku, mimau, tazuneru (call on); (isha ni) mite morau (consult);

(i) I think he ought to (go and )see a doctor. Kare wa isha ni mite

morawanakereba naru mai


(ii) I shall be seeing him tomorrow. Asu kare o tazunete [mimatte]

miy�o.

f. (jiken nado o )miru, mokugeki suru; . . . ni s�og�usuru (undergo), keiken suru

(experience):

(i) That year saw many changes. Sono toshi wa kawatta koto ga �okatta

(ii) He couldn't live to see his son's marriage. Musuko no kekkon o

mizu ni shinda

(iii) He has seen a lot in his life [a lot of life]. Nakanaka yononaka

no keiken o tsunde iru

(iv) He will never see 50 again. Ano otoko mo goj�u no saka o koshita

yo.

g. shiru, satoru, rikai [ry�okai]suru (comprehend, understand):

(i) see the point of the argument ronten ga wakaru

(ii) Don't [Can't] you see what I mean? Watashi no iu koto ga wakaranai

ka

(iii) I don't see why he doesn't come. d�oshite konai ka wakaranai

(iv) see the use [good, fun] of . . . . . . no neuchi [yosa, okashisa] ga

wakaru

(v) see a joke j�odan ga wakaru

(vi) I see you. (Amer.) Kimi no iu koto wa wakaru yo

(vii) see it (colloq.) ry�okai suru, wakaru (understand, comprehend)

(viii) Do you see? [See?] wakarimashita ka

(ix) I see. Naruhodo wakarimashita.

h. mitsukeru (descry), miwakeru (discern), kidzuku, mitomeru (notice);

(i) He saw at once that he had made a mistake. Machigai o shita

koto ni sugu kidzuita.

(ii) I don't see any harm in what he is doing. Kare no shiteiru koto

ni nani mo warui koto o mitomenai.

(iii) I see a great danger in that sort of thing. S�oiu koto ni wa hij�ona

kiken ga aru.

i. mite miru, yoku shiraberu, kensa[kenbun] suru (examine):

(i) see a house before taking it kariru mae ni ie o miru


(ii) I have in a man to see the drains. Hito o yonde gesui o mite

moratte iru.

j. (torishirabe nado ni yotte) tashikameru (ascertain), shiru (learn, �nd out):

(i) Go and see who it is. Daredaka itte mite goran

(ii) I believe you, but I'd rather see it for myself. Kimi no kotoba o

shinjinai wake dewa nai ga, jibun de tashikamete mitai.

k. [tsurei that-clause matawa kakobunshi-kei no hogo o tomonatte ] (. . . suru

[sareru]y�o) ki o tsukeru, tehazu o suru, torihakarau (attend to, take care),

kitto . . . suru (make sure) (cf. vi 4):

(i) See (that) you don't catch your foot. Tsumazukanai y�oni ch�ui

shinasai

(ii) See that the work is done. Kitto shigoto ga owaru y�oni se yo.

(iii) see a thing done kantokushite yaraseru

(iv) see justice done koto no k�ohei o kisuru; fukush�u o togeru.

l. kangaeru (think, consider):

(i) as I see it watashi no miru tokoro de wa

(ii) I see things di�erently now. Watashi wa ima de wa mono no

kangaekata ga izen to wa chigau

(iii) I do not see it in that light. Watashi wa s�oiu f�u ni wa minai [kan-

gaenai]

(iv) Well, I'll see what I can do. D�o dekiru ka kangaete miy�o.

m. tsukisou, okuri todokeru (escort), miokuru

(i) I saw him home [to the door, as far as the station]. ie [genkan,

eki] made miokutta

(ii) see a person o� hito o miokuru

n. omoi ukaberu, s�oz�osuru (imagine):

(i) Poets see many things in the ordinary. Shijin wa heebon na mono

no naka ni iroirona mono o s�oz�osuru

(ii) I cannot see myself submitting to it. Watashi ga sore ni fukuj�usuru

nado to wa omoi mo yoranai.

o. damatte mite iru, mokunin suru (recognize as tolerable):

(i) He won't see being made use of Kare wa dashi ni tsukawarete

damatte wa imai.


p. [zoku] wairo no juju ni karande (hito ni) kaiken suru, . . . ni wairo o tsukau

(bribe), baish�usuru: see an inspector

q. (poker nado de, kake ni) �ojiru (meet); (aite to) d�ogaku no kake ni �ojiru.

(402) |-v.i.

a. miru (look); mieru, me ni suru, me ga kiku:

(i) as far as one can see miwatasu [mieru] kagiri (cf. 2)

(ii) I can't see to read. Kurakute [me ga kikanakute] yomenai

(iii) A puppy cannot see till the ninth day. Inu no ko wa kokonokame

made wa me ga mienai

(iv) Owls see best at night. Fukur�o wa yoru ga ichiban yoku me ga mieru

(v) He sees no further than his nose. Kare wa osaki makkura da, ikou

mesaki ga kikanai

(vi) Seeing is believing. ☞ seeing.

b. wakaru, rikai suru, etoku suru

(i) (discern, understand):

(ii) I see. wakarimashita, naruhodo

(iii) You see, gozonji no t�ori, owakaridesh�oga, nee, sora; ohanashi shi-

nakereba narimasen ga (I must explain)

(iv) You shall see ato de hanashimash�o

(v) See? owakari desh�o

(vi) as you see goran no y�o ni

(vii) as far as I can see watashi no miru tokoro de wa, watashi no kangae

dewa (cf. 1).

c. minuku, kanpa suru, sh�osatsu suru (have insight into, through).

d. ch�ui suru, i o kubaru (give attention or care) (cf. vt. 11):

(i) See to it that the work is done before dark. Kuraku naranai uchi

ni shigoto ga dekiagaru y�o ni ch�ui shinasai.

e. tashikameru, shiru (�nd out), shiraberu (make inquiry):

(i) The post has come; I'll go and see. Y�ubin ga kita, itte mite koy�o

(ii) Go and see for yourself, if you don't believe me. Watashi no iu

koto ga shiny�o dekinakereba jibun de itte shirabete [tashikamete] mina-

sai.


f. kangaeru (consider, think), jukuk�osuru (deliberate) (whether, if).

g. hito ni au, menkai o yurusu:

(i) Can you see for a few minutes? nisanpun ome ni kakarasete

itadakemasen ka. [This seems to be an error |CFB]

(403) Collocations

a. have seen (one's) better [best] days (Ima wa ochiburete iru ga) mukashi

wa yoi toki mo atta

b. She must have seen better days. Mukashi wa kitto shiawasena mibun de

atta ni chigainai

c. He was dressed in an old coat that had seen its best days. Mukashi

wa rippa datta furui fuku o kite ita.

d. let me see kangae sashite kure, s�osana, �eto, hatena,

e. see (a person) blowed [damned, hanged] (�rst [before]) Donna koto

ga atte mo sonna koto dake wa mappira da

f. see (a person) further �rst = see (a person) blowed.

g. see (a person) o� (hito o) miokuru (cf. send o�):

h. I saw my friend o� at the station.

i. see (a person) o� the premises (usan kusai mono nado o) yashikigai e

okuri dasu [oidasu]

j. see about . . . no koto o kangaeru (consider, think about); . . . ni tsuite nan-

toka shudan o k�ojiru; . . . no koto o shiraberu, ki o tsukeru: I [We] shall

see about it. Sore wa nantoka shimash�o [shirabete mimash�o] (often used as

�xed expression if one refuses to take action immediately ).

k. see after . . . ni ki o tsukeru (more often look after); . . . o sewa suru.

l. see �t [good] (to do) (. . . suru) h�o ga (tsug�o ga )yoi to omou: Do it if you

see �t.

m. see here (Amer.) moshimoshi (I say) oi, kore (look here) (said to scold

and to get someone's attention)

n. see the old year out and the new year in Ky�unen o okuri shinnen o

mukaeru.

o. see into . . . ni ch�osa suru; . . . o minuku (see through).

p. see out


(i) owari made miru; (sh�ukyoku o) mitodokeru, kansee suru: see out a long

play Nagai geki o owari made miru.

(ii) (zoku)makasu (defeat).

(iii) genkan made miokuru.

q. see over . . . o mimawaru, kenbun suru.

r. see through . . . no shins�o o minuku (penetrate), . . . o kanpa suru (detect)

s. see through a ladder miyasui, wakarikitta koto da

t. see through a brick wall ganshiki ga surudoi.

u. see. . . through . . . o saigo made mitodokeru [tasukete yaru]:

v. see a person through his troubles komatte iru hito o saigo made tasukete

yaru

w. see a thing through koto o yari t�osu (carry out).

x. see to . . . ni ch�ui suru, ki o tsukeru (take care of), o teire suru, junbi suru

(provide) (hand vi.4; cf. vt.11):

(i) Leave it to me; I'll see to it [I'll have it seen to] omakase nasai,

watashi ga torihakaraimasu [torihakarawasemasu]

(ii) see to one's business jibun no shigoto o ch�ui suru.

y. see well and good (zoku) yoi to omou, sashitsukaenai to omou (think �t).

Kenkyusha English-Japanese Dictionary|Discussion

Iwasaki et al. (1960) is a large, standard dictionary, one of several catering to the

lucrative market of Japanese-English translators, college students, and high-school students

in Japan facing rigorous college entrance examinations which include an EFL component.

It is densely packed with information and cross references; we have tried to preserve as

many of these as possible while rendering it more readable.

Before discussing the sub-entries, let us �rst introduce the verbs miru and mieru.

Miru basically means `to look at' and mieru is used to express `see'. Their syntax, however,

is di�erent. Miru is a regular transitive verb, with the direct object in the accusative and

the agent marked with either wa or ga (depending on several factors which are not relevant

here). Mieru, on the other hand, which can be regarded as a special potential form of miru,

marks the experiencer with wa or ni wa (which we we will consider topic markers, the latter

combined with a dative) and the stimulus with ga, as shown in Ex. (404) (cf. Makino &


Tsutsui (1986) for a comparison between mieru and the regular potential form mirareru).

A similar pattern occurs with kikoeru `hear', as shown in Ex. (405).

(404) a. Watashi wa mainichi yama o miru.

b. I TOP every.day mountain ACC look

c. I look at the mountain(s) every day.

d. Watashi (ni) wa mainichi yama ga mieru.

e. I (DAT) TOP every.day mountain GA see.POT

f. I (can) see the mountain(s) every day.

(405) a. Watashi (ni) wa tori no koe ga kikoeru.

b. I (DAT) TOP bird POS voice GA hear.POT

c. I (can) hear the bird.

We will �nd that senses of English see that are simple experiencer-stimulus situations will

usually be expressed with mieru, those that emphasize active looking will use miru, and

those that do not require physical vision will usually use neither.

Like other dictionaries, the entry for see is divided into three sections, transitive,

intransitive, and collocations, apparently for reasons of tradition. As in other dictionaries

discussed above, we �nd that the �rst sense under \v.i." (402-a) is our faculty, while

most of the others are distinguished from corresponding senses under \v.t." only by virtue

of the null instantiation of their direct objects. In the case of (402-d) the equivalence to

(401-k) is noted by the lexicographers, but (402-e) and (401-j) are equally close.

Most of our senses are represented in this entry, as we can see in Table 5.1. As

before, however, we also �nd a considerable variety within the sub-entries; I will discuss

them in order:

In (401-a), the �rst two examples are eye, but the last is vide, with the imperative

form of miru. (401-b) is our sense hallucinate, with cross-references to the entries for

conventional hallucinated NPs, \things", snakes and stars . (401-c) combines spectate and

tour, both basically eye composed with the semantics of the seen.

Sub-entry (401-d) covers our sense visit, with a variety of expressions. In fact,

Japanese has a large variety of expressions for types of social interaction; it is di�cult to

quantify, but there seem to be �ner distinctions than English customarily makes with regard

to the relative social status of the participants, the pre-arrangement or lack thereof, the


Sense Sub-entry

accompany (401-m)

audience (401-d)

classify (401-l)

consult (401-e)

determine (401-j)

ensure (401-k)

envision (401-n)

experience (401-f)

eye (401-a),(401-i)

faculty (402-a)

hallucinate (401-b)

discourse (402-b)

recognize (401-g),(401-h)

setting (401-f)

spectate (401-c)

tour (401-c)

vide (401-a-iii)

visit (401-d)

Table 5.1: Senses and Sub-entries in Kenkyusha English-Japanese Dictionary

purpose of the visit, etc. Thus, our senses visit, consult, and audience will correspond

to many Japanese expressions, some of which appear in (401-d) and (401-e). Without going

into too much detail, we can take note of the following: the simple verb au `meet' (401-d), a

polite form derived from it oai suru `HON.meet do' (401-d-i), and the humble verb omikake

suru `HON.eye.attach do' (401-d-iv), all of which would generally have to be translated

into English using the word see and expressing the social connotations periphrastically,

i.e., have the honor of seeing , have the pleasure of seeing , etc.4 In (401-e) we �nd mimau

`to visit someone who is in distress' (in a hospital, refugee camp, etc.), tazuneru `make a

social visit' and mite morau. The last of these is morphologically `see-TE receive', where

TE can indicate progressive aspect, sequential actions, or a number of other meanings. It

appears that see a doctor gets translated with mite morau, but not see my lawyer or see

the manager , because it is not necessary for them to physically examine the consulter (cf.

Chapter 2, senses consult on page 70 and audience on page 100).

4Some very polite Japanese may be very di�cult to express; attempts at \literal" translation are oftenludicrous, your honorable grandson, We are deeply obligated to you for the honor of your presence at ourpoor shack , etc.


Included in (401-f) are setting (401-f-i) and three examples of experience, all of

which are idiomatic in di�erent ways, and none of which involve the usual direct objects for

this sense, such as combat , use, service, etc. Although the �rst two translations given, miru

and mokugeki suru both involve physical vision, only one of the examples does (401-f-ii);

the last two translations will �t most examples of this sense better.

Most of the examples in (401-g) are of discourse or recognize. (401-g-vi) and

(401-g-vii) appear to be obsolete slang. The worst example of this is found in (401-p),

meaning `to bribe', which also appears in the OED entry for see, which in turn cites a

1911 edition of Webster's Dictionary! (401-g-iv) is an instance of classify; this sense is

scattered across the sub-entries, e.g. (401-h) contains one example of recognize and two of

classify, which participate in a larger pattern to be discussed below.

Sub-entry (401-i) provides the clearest cases of miru, where the pragmatics of the

situations requires active looking. (401-j) is our determine, (401-k) ensure, and (401-m)

accompany. In (401-l), the �rst three examples have to do with making judgements, and

can probably be considered classify even though they are not of the patterns see X as Y

or see Y in X . (401-l-iv) is wonderfully ambiguous and deserves its own discussion along

with the collocation see about it (403-j); both have many uses depending on pragmatics.

Example (401-n-i) is probably envision, but could be recognize if we believe

that poets see a reality that most people overlook.

Most of the \intransitive" sub-entries have already been discussed in passing, so

we will only note a few more points here. (402-b) consists mainly of examples of recog-

nize; the null instantiation of the seen means that many of them refer to the immediate

context, especially the discourse context, and so they can be further categorized as dis-

course. (402-c) is really a combination of see through and see into. (402-f) also seems to

be determine, like (402-e); (402-g) is apparently an error|at least it is not acceptable or

familiar to any of the American or British speakers I have consulted.

One might quibble about some of the more obscure collocations that have been

included, such as (403-f) and (403-s), but at least a fair selection has been made from the

hundreds which could have been chosen. See the old year out and the new year in (403-n)

is found here (and in some other bilingual dictionaries) but not in the much larger W3NI.

It is odd to �nd see here (403-m) agged as \American", since it seems at least as common

in British English as in American.

In summary, the Kenkyusha treatment of see is remarkably complete and, except


for following lexicographic custom by making the major division between transitive and

intransitive, predominantly organized according to the semantics of the English uses rather

than the syntax of the Japanese translation. Such an organization is probably the most

informative for students, even though it frequently results in the same Japanese term being

scattered across many sub-entries.

Small English-Japanese Dictionaries

Y�ohan English-Japanese Japanese-English Dictionary. (Kaneda 1997)

see v

(451) a. (look at) miru

b. (meet) au

c. (understand) wakaru

d. (consider) kangaeru

e. (take care) ki o tsukeru

f. (undergo) keiken suru

g. (�nd out) mitsukeru

h. (make sure) tashikameru

i. Collocations

(i) see o� miokuru

(ii) see through mi-yaburu; mi-nuku

(iii) I see wakarimashita

T�oei's English-Japanese Japanese-English Dictionary. (Fujikake 1984)

(452) a. (perceive with the eyes) miru, nagameru

b. (understand) rikai suru, wakaru

c. (imagine, think) . . . to omou, . . . to kangaeru

d. (view) kenbutsu suru

e. (visit) au

f. (ascertain) shirabe tashikameru

Both of these entries have


1. a basic sense (eye) translated miru

2. a social sense like visit, translated au,

3. a sense characterized as \understand" in English and wakaru in Japanese, which can

be used for our sense recognize, but is probably more frequent in the discourse

sense,

4. a \cognition" sense translated by kangaeru `think, consider', which is usually closer

to our sense recognize or classify than to imagine,

5. a sense translated by tashikameru. The T�oei dictionary uses the English guide word

ascertain, which is better because it is unambiguous, as opposed to the guide word

in the Y�ohan dictionary, make sure, which is ambiguous between determine and

ensure.

Although both dictionaries are very small, both cover most of the basic senses of

the English word see, and both give the most common Japanese equivalents. The Y�ohan

is preferable, because it adds the senses experience and ensure, in addition to giving

two English translations for wakaru, one of them extremely common in ordinary discourse.

The second translation for her sense eye in the T�oei is also misleading, meaning something

more like `stare at' than see. A language learner would be better o� most of the time

with the Y�ohan dictionary; only at a fairly advanced stage would one need to consult the

Kenkyusha, except that many of the collocations listed in the Kenkyusha will be encountered

even by intermediate students. As we have noted previously in regard to other languages,

neither small dictionary distinguishes transitive from intransitive uses, which is a blessing

in disguise. Otherwise, the sense divisions given in the Y�ohan agree rather well with those

in the Kenkyusha.

5.7 Analysis

English-English Comparisons of Entry Structure

Table 5.2 on page 210 shows a summary of sense divisions for the four monolingual

English dictionaries we have examined. The numbers and letters in the cells refer to the

sense divisions (and, in the case of the W3NI, subdivisions) used in printing the texts in


this chapter. There are only three senses distinguished by all four, eye, accompany, and

visit, but another three senses are distinguished by at least three of the four, recognize,

determine, and ensure. All of these six are among those I have called \basic senses",

although it could be argued that accompany is less basic, being much more specialized.

W3NI distinguishes sharply between transitive and intransitive uses, but none of

the smaller dictionaries have the space to do this. This prevents them from recognizing a

sense faculty, but saves them from splitting several of our senses, as W3NI does. None

of the dictionaries, not even the much larger W3NI, distinguishes among recognize, con-

dition and process. Nor does W3NI distinguish between envision and hallucinate,

despite discussing envision in three places. Overall, the agreement between the smaller

dictionaries and our sense divisions is reasonably good, but the organization of the entry in

the W3NI is quite di�erent.


Dictionary W3NI MWP WNW COD

Basic Senses

eye 1a a a a

recognize 3c c b

accompany 8a, 5a e e e

condition ? c? b? c?

consult 7a(ii) e? g

dating 7b(i) e?

determine 2c,4a,13a-b d c

ensure 5b, 14a, 15,22, 23

d d

envision 3a,d,g

faculty 11a,b h

news 4b

process ? c? b?

read 4b c? b?

setting 2d,e

visit 7a(i) e f d

Semi-collocations

classify 3b?,f?, 6a c

experience 2a b

gambling 9 h

vide 4c g?

Compositional Uses

audience 7b(ii)

discourse 3e

hallucinate

scan 1c

spectate 4d

Collocations

let's see 12c

see one's way clear 18

see fit =6a

see off 8b

see reason

see through x 21

see x through 8c

y'see 10,12b

Table 5.2: Summary of Sense Divisions in Four Monolingual Dictionaries


Cross-linguistic Comparisons of Entry Structure

Next, we examine the structure of the entries in the major bilingual dictionaries in

more detail, with a view to comparing across languages. For this purpose, for each language,

we show a table with rows representing our usual list of senses (in a di�erent order) and the

columns representing the sense divisions of the dictionary in question. The rows have been

ordered so that related senses are grouped together, and groups are separated by horizontal

lines. Then for each sense division, we have classi�ed the de�nition into one of our senses,

and counted the examples (if any) which fall into each of our senses. Each such table is

followed by another table listing the de�nitions for each of the the sense divisions; these are

taken from the full listings on previous pages, but are repeated here for convenience.

Let us begin by looking at Table 5.3 on page 213, which shows the sense divi-

sions for the Collins Spanish dictionary. The column on the left shows our sense divisions.

The �rst group of senses are those that have to do with physical vision, eye, process,

spectate, tour, faculty, hallucinate, read, and vide. The next is the cognition

group, recognize, classify, envision, discourse, and news, followed by a small group

having to do with human complements, visit, consult, and audience. The fourth group

is a heterogeneous collection of fairly speci�c senses, accompany, determine, ensure,

experience, setting, and gambling. The last two rows represent collocations already

discussed in the cognitive chapter, and idioms not elsewhere discussed.

The columns represent the sense divisions in the dictionary; those numbered 201

are the transitive senses, and those numbered 202 are the intransitive senses. Above the

numbers is a classi�cation of the de�nition according to our sense divisions; thus, 201-a is

our sense eye, 201-b, our sense accompany, etc. The numbers within the Collins represent

classi�cation of the examples under each sub-sense. For example, looking at sense a. of the

Collins dictionary, on page 180, we note that the �rst example under the �rst de�nition

is see page 8 , which would be our sense vide, and this is represented by the number 1 to

the right of the word vide in the �rst column of Table 5.3. Of the 12 examples under this

heading, only one is our sense eye, 4 are our sense process, 1 faculty, 2 experience,

and 3 are idiomatic. But seven of the 12 are clearly \visual" senses, and to the senses

we have called \idiomatic" do involve physical vision (201-a-xi and 201-a-xii). Example

201-8-ix has nothing to do with physical vision, and the last two examples are only slightly

related.


All four examples in 201-b are clearly accompany, as intended; 11 of the 12 exam-

ples in 201-c and 201-d are in the cognition category, although their translations vary. The

rest in the transitive senses are also well focused; note that 201-f contains 10 \social" senses,

divided among visit, consult, and audience. As previously noted the intransitive senses

are related to the transitive senses through null instantiation, so the senses in 202-a and

202-b are spread over a large range. As with the monolingual dictionaries, several examples

of intransitive eye are our sense faculty, and discourse deixis is also well represented here.

In general, then, most of the examples under each sense fall in the same general category

as the sense intended. Most exceptions to this pattern fall in the categories of collocations

and idiomatic expressions, which a good bilingual dictionary will include as part of its job.


201-aeye

201-bacom

201-crec

201-drec

201-eens

201-fvisit

201-genvis

201-hexpr

202-aeye

202-brec

eye 1 2proc 4spectourfacl 1 2halureadvide 1

rec 2 1class 2 2envis 1 3disc 1 2news 2

visit 5cons 4audi 1

acom 4detr 2ens 5expr 2 3setggamb

coloc 1 2id 3 1 2 1 1

Table 5.3: Sense Divisions in Collins Spanish Dictionary


201-a: (gen) ver201-b: (accompany) acompa~nar201-c: (understand) comprender, entender, ver201-d: (look, learn, perceive) mirar; observar; percibir201-e: (ensure)201-f: (visit. frequent) ver, visitar201-g: (imagine)201-h: (experience)202-a: (gen) ver202-b: (understand) comprender

Table 5.4: De�nitions from Collins Spanish Dictionary


301-aeye

301-brec

301-cdetr

301-dexpr

301-evisit

301-facom

301-gtour

301-hid

301-iens

301-jid

301-kgamb

302-aeye

302-bfacl

302-cdetr

302-dcoloc

302-erec

302-fens

eye 5proc 1 1spec 1tour 1facl 3 1halureadvide

rec 1 1class 1envisdisc 1 1news

visit 2cons 2audi 2

acom 1detr 2 1ens 3expr 4setggamb

coloc 1 1id 1

Table 5.5: Sense Divisions in Far East Chinese Dictionary

The sense divisions in the Far East Chinese Dictionary are shown in Table 5.5, with

the de�nitions repeated in Table 5.6 on the following page. The correspondence between

the sense divisions indicated by the de�nitions and the examples if quite good; for the

\minor" senses, accompany (301-f), determine (301-c and 302-c), ensure (301-i), and

experience (301-d), all of the examples are of the corresponding sense. Most of the visual

senses are in 301-a (eye, process, spectate) and 302-a (intransitive, faculty). 301-e

gives two examples of each of the \social senses", visit, consult, and audience, and most

of the cognitive senses are found in 301-b and 302-e, both with clearly cognitive de�nitions.


301-a: k�an; ji�an; sh�� `look; see; view'( All translated k�an ji�an except as noted )301-b: li�aoji�e; l��ng hu��. (=understand); ch�aju�e (=perceive).301-c: f�axi�an; xu�e zh��. `discover; �nd out'301-d: j��ngy�an; yu�el�� èxperience'301-e: hu��mi�an; w�uji�an (=meet); f�angw�u (=call on); sh�angt�an (=talk with); ji�eji�an (=re-ceive a call from).301-f: h�us�ong; zh�aog�u èscort; care for'301-g: c�anji�a; c�angu�an `participate in; take a tour of'301-h: r�ang; y�unx�u `let; allow'301-i: zh�uy��; f�uz�e `pay attention to; take responsibility for'301-j: q��d�ai; d�engd�ai d�ao zu�� ho�u `hope; wait until the very last'301-k: (p�aix��) ('Card games') xi�a t�ongy�ang du�ode d�uj��n `put down an equal bet'302-a: k�an; ji�an; sh�� `look; see'302-b: j�uyo�u sh��l�� `have the power of sight'302-c: ch�ak�an òbserve'302-d: k�aol�u `consider'302-e: li�aoji�e ùnderstand'302-f: zh�uy��; li�ux��n `pay attention; take care'

Table 5.6: De�nitions from Far East Chinese Dictionary


Finally, let us turn to the Kenkyusha English-Japanese dictionary, with de�nitions

in Table 5.8 on page 219 and sense divisions in Table 5.7 on the following page. The

pattern here is much more complex than in the Spanish and Chinese dictionaries. A few

of the sense divisions, such as 401-j (determine), 401-k (ensure) and 402-a (faculty),

are represented by a set of examples that are clearly \on point", and others are nicely

within the larger categories, such as 401-a (eye/vide), 401-d (visit/audience), and 401-c

(spectate/tour). But many of the sense divisions made by the dictionary are unclear,

such as the distinction between 401-g and 401-h, which seems to be mainly aspectual,

between stative and inchoative cognition; both include examples of both recognize and

classify. The distinction between 401-a and 401-i may be re ected in their very di�erent

de�nitions in Japanese, yet the examples in the latter use the simple verb miru, `see', as in

401-a.


401-aeye

401-bhalu

401-cspec

401-dvisit

401-econs

401-fexpr

401-grec

401-hrec

401-ieye

401-jdetr

401-kens

401-lclass

401-macom

401-nenvis

401-oid

401-pid

401-qgamb

402-aeye

402-brec

402-b-iirec

402-crec

402-dens

402-edetr

402-fdetr

402-gconsult

eye

2

2

1

proc

spec

1

tour

3

facl

3

halu

read

vide

1

rec

2

1

class

1

2

envis

2

disc

4

3

news

visit

3

cons

2

audi

2

acom

1

detr

2

1

2

ens

4

1

expr

2

setg

1

gamb

coloc

1

2

id

3

1

2

3

1

1

2

1

1

Table5.7:SenseDivisionsinKenkyushaJapaneseDictionary


401-a: miru, . . . ga mieru (perceive by the eye); (ki o tsukete) miru (look at):401-b: (Yume nado ni miru): I saw him in a dream.401-c: (geki, eiga, meisho nado o) miru (view), kanran [kenbutsu] suru (be a spectator of),mi ni iku (visit)401-d: (hito ni) au, deau (meet:), kaiken suru, menkai suru (interview)401-e: (hito ni) ai ni iku, mimau, tazuneru (call on); (isha ni) mite morau (consult);401-f: (jiken nado o) miru, mokugeki suru; . . . ni s�og�usuru (undergo), keiken suru (experi-ence):401-g: shiru, satoru, rikai [ry�okai]suru (comprehend, understand)401-h: mitsukeru (descry), miwakeru (discern), kidzuku, mitomeru (notice);401-i: mite miru, yoku shiraberu, kensa[kenbun] suru (examine)401-j: (torishirabe nado ni yotte) tashikameru (ascertain), shiru (learn, �nd out)401-k: [tsurei that-clause matawa kakobunshi-kei no hogo o tomonatte ] (. . . suru [sareru]y�o)ki o tsukeru, tehazu o suru, torihakarau (attend to, take care), kitto . . . suru (make sure)(cf. vi 4)401-l: kangaeru (think, consider):401-m: tsukisou, okuri todokeru (escort), miokuru401-n: omoi ukaberu, s�oz�osuru (imagine)401-o: damatte mite iru, mokunin suru (recognize as tolerable)401-p: [zoku] wairo no juju ni karande (hito ni) kaiken suru, . . . ni wairo o tsukau (bribe),baish�usuru401-q: (poker nado de, kake ni) �ojiru (meet); (aite to) d�ogaku no kake ni �ojiru.402-a: miru (look); mieru, me ni suru, me ga kiku402-b: wakaru, rikai suru, etoku suru402-b-ii: (discern, understand):402-c: minuku, kanpa suru, sh�osatsu suru (have insight into, through).402-d: ch�ui suru, i o kubaru (give attention or care) (cf. vt. 11)402-e: tashikameru, shiru (�nd out), shiraberu (make inquiry)402-f: kangaeru (consider, think), jukuk�osuru (deliberate) (whether, if).402-g: hito ni au, menkai o yurusu

Table 5.8: De�nitions from Kenkyusha Japanese Dictionary


Cross-linguistic Comparisons of Translations

As I discussed in Section 1.2, certain groups of events tend to co-occur everywhere

in the world, and the resulting correlations in human experience form the basis of much

linguistic categorization; I further argued in Chapter 2 that such a natural tendency to

co-occur lies behind the connection between eye and recognize. Now, with translations

of separate senses in bilingual dictionaries, we can look at how the di�erent senses are

lexicalized in di�erent languages, and perhaps answer the question as to which relations

among senses are natural extensions based on natural co-occurrences of events which would

be true for all cultures, and which are language-speci�c facts.

Unfortunately, as we have seen, just as in the monolingual English dictionaries, the

example sentences under a sense heading in a bilingual dictionary do not necessarily belong

to that sense according to our de�nition. I have therefore analyzed the material from the

bilingual dictionaries in a second way, by categorizing each of the example sentences into

one of our senses, regardless of where it appears in the entry, and creating tables showing

the translations given for each example. Such a table for the Collins Spanish dictionary is

shown in Table 5.9 on page 222; the columns represent our sense divisions, and the rows,

the various translations given in the 64 examples in this dictionary. We have organized the

senses into four broad categories:

1. basic vision (faculty, process, vide)

2. visiting (audience, consult, visit)

3. ensure (ensure)

4. cognitive (classify, discourse,recognize)

The remaining senses are shown in the next division of the table, and example

sentences whose meaning depends on collocations we have previously discussed or other

idioms are shown in the last division. The vertical lines in the tables show these divisions,

with the double line in the middle of the table dividing the four meaningful divisions on the

left from the arbitrary ones on the right. Insofar as possible, we have tried to to make the

same divisions according to the meanings of the translation equivalents, which are shown

in the rows of table. Those example sentences which do not contain any word or phrase


which can be properly said to be a translation of the English word see appear in the row

entitled \No equivalent".

Now let us consider the extent to which the patterns of sense extension found in

the English verb see are found in Spanish. First we notice that all of the examples of basic

vision, including eye itself, are translated with ver , and the majority of the instances of ver

are in this category. There can, therefore, be no question as to the basic meaning of ver .

But we also �nd two instances of ver in the visit category, three in the cognition category,

and examples under our senses determine, experience, and news, a range of meaning

not unlike that of see in English. Two of the expressions for \point of view" also use ver

and the related noun, vista. (At least two more examples, the last two in the visit some

entry, could also be expressed informally with ver , as <nos vemos! or <nos veremos! )

Of course, there are di�erences between the uses of see and those of ver ; ver is not

used to express ensure, accompany, or envision. Also, a large majority of the example

sentences are translated with a word other than ver , suggesting that ver may not be as

commonly used for some senses in Spanish as see is in English. Somewhat surprisingly, in

translating the news sense, the verb oir `hear' is used, which is reserved in English for news

heard from another person. In summary, although there is some overlap with the patterns

in English, most of the senses of English see are not expressed with ver in Spanish most of

the time, although the range of meanings of ver is fairly similar to that of see.


eye

facl

proc

vide

audi

cons

visit

ens

class

disc

rec

acom

detr

envis

expr

news

coloc

idiom

Totals

mirar

1

1

ver

3

2

4

1

1

1

2

1

2

2

1

4

24

consultar

1

1

entrevisarsecon

1

1

visitar

1

1

cuidar

1

1

procurar

3

3

comprender

1

1

1

3

conocer

1

1

creer

1

1

encontrar

2

2

entender

1

1

1

3

percibir

1

1

acompa~nar

3

3

imaginar

2

2

llevar

1

1

oir

1

1

Noequiv.

1

2

3

1

1

2

1

3

14

Totals

3

3

4

1

1

4

5

5

4

3

3

4

2

4

5

2

3

8

64

Table5.9:FrequenciesofSpanishTranslationsbySenses.


Table 5.10 on the following page shows the distribution of translations for the

Far Eastern English Chinese dictionary. We run immediately into the question of which

morpheme we mean. The resultative compound k�anji�an (lit. `look-see') is found only in

the sense eye5, but many of the other senses have either k�an or ji�an. determine (301-

c), emphasizing agentiveness has k�an, while visit (301-e) has ji�an in various combinations

(note that various more formal words are given as de�nitions, but the �rst two examples

give more common words for this sense ji�anmi�an) `see face', and ji�and�ao `see reach'. The

consult example (301-e-v) uses the more agentive k�an, but the phrase zh�ao y��sh�eng k�ank�an

`seek doctor look look' is actually ambiguous as to who is the agent of k�an. faculty is

expressed by the combinations kandejian and kanbujian, the usual resultative compound

combined with de and bu for positive and negative ability.

Again we �nd the concept of `viewpoint' translated with morphemes related to

vision, either as a noun k�anf�a `look way' or gu�andi�an `view point' or as an idiom w�o k�an. . .

Ì look' ! às I see it,. . . '. Note also that tour gets a separate sub-entry; the compound

c�angu�an `participate view' is speci�c to this sense, although it contains the morpheme gu�an

`see' (cf. Section (357) on page 197). Even if we include all of the combinations containing

either of the morphemes ji�an and k�an as examples of verbs of perception, we still �nd that

only half of the example sentences are expressed by a verb of perception. We have only 38

examples here, but it appears that the distribution of the Chinese verbs ji�an and k�an is

somewhat more limited than that of English see.

5This is only true of four of the �ve examples translated with k�anji�an; one example under the �rst sensedivision, I saw that the box was empty is translated k�anji�an, but this is actually recognize, although thefact recognized and the percept that leads to the recognition are quite physical.


eye

facl

proc

spec

tour

audi

cons

visit

ens

class

disc

rec

acom

detr

expr

colloc

idiom

Totals

c�angu�an

1

1

ch�ak�an

1

1

ji�an

1

1

ji�and�ao

1

1

k�an

1

1

2

k�anb�uch�u

1

1

k�anbuji�an

2

2

k�and�ao

1

1

2

k�andeji�an

1

1

k�anji�an

4

1

5

k�ank�an

1

1

ji�eji�an

1

1

ji�anmi�an

1

1

f�uz�e

1

1

zh�uy��

2

2

d�ong

1

1

m��ngb�ai

2

1

3

j��ng

1

1

k�aol�u

1

1

r�ang

1

1

s�ong

1

1

y�ou

1

1

yu�el��

1

1

Noequiv.

1

2

1

1

5

Totals

5

4

2

1

1

2

2

2

3

1

2

2

1

3

4

2

1

38

Table5.10:FrequenciesofChineseTranslationsbySenses.


Finally let us consider our other non-Indo-European language, Japanese, shown in

Table 5.11 on the following page. In this case we have 66 examples to work from, so that

better generalizations may be possible. The relatively large number of equivalents used in

the Japanese translations is primarily due to variations in politeness, which may involve

the choice of a di�erent basic morpheme, di�erent in ectional or derivational morphology,

and/or the addition of honori�c forms. The situation is further complicated by the presence

of many English idioms and collocations among the examples (shown in the righthand

columns of Table 5.11). The general conclusion is that the basic Japanese equivalent of

see, miru, has a much more limited range than see. None of the senses in the Cognitive

division can be expressed by miru, nor can any of the senses in the Visit division, although

mite morau, like the Chinese equivalent, means `to be seen by a doctor', and ome ni kakaru

contains a morpheme meaning èye'.

The usual way of writing these words, using a combination of Chinese characters

and Japanese syllabics, raises other interesting questions. In all of the translations mit-

sukeru \descry", miwakeru \discern", minuku \have insight into", and mimau \call on",

the �rst syllable, mi , is written with the Sino-Japanese character pronounced ji�an in Man-

darin, meaning `see'; does this re ect accurately the morphology of these words? If so,

can we conclude that mitomeru \notice" does not contain the same morpheme, because

it is written with a di�erent character? The question of the extent to which the Chinese

characters adopted by the Japanese for their (fundamentally unrelated) language re ect the

underlying morphology of the Japanese language at that time is fascinating, but delving

into this would take us too far a�eld. For the moment, we simply observe that most of

the senses of see other than eye are expressed most of the time with words unrelated to

miru, but that there is also a signi�cant minority of translations using words which may

be related to miru.


eye

facl

spec

tour

vide

audi

cons

visit

ens

class

disc

rec

acom

detr

envis

expr

setg

coloc

id

Totals

gamieru

2

2

goran

1

1

kenbutsu

2

2

megamieru

2

2

miru

2

1

1

2

1

4

11

miwatasu

1

1

au

1

1

mitemorau

1

1

oaisuru

2

2

ohtaisuru

1

1

omenikakaru

1

1

2

tazuneru

1

1

chuisuru

1

1

kisuru

1

1

yaraseru

1

1

younichuisuru

1

1

younisuru

1

1

kidzuku

1

1

mitomeru

1

1

owakaridearu

1

1

ryoukaisuru

1

1

wakaru

1

5

2

1

9

miokuru

1

1

2

souzousuru

1

1

tashikametemiru

2

2

Noequiv.

1

1

1

1

1

1

1

1

1

7

16

Totals

5

3

1

3

1

2

2

3

5

3

7

3

1

5

2

2

1

3

14

66

Table5.11:FrequenciesofJapaneseTranslationsbySenses.


5.8 Conclusions

In summary, there seems to be a tendency across historically unrelated languages

for verbs meaning `see' to be used also to mean (a) `visit', (b) something like ùnderstand',

and (c) `�nd out'. This might be related to a natural tendency for events involving these

concepts to co-occur. On the other hand, the fact that the pattern of translations in

Spanish resembles English more closely than the patterns of Chinese and Japanese suggests

that these relations among senses may be stronger among Indo-European languages, and

therefore, inherited rather than universal. In any case, it is di�cult to generalize on the

basis of a small amount of evidence from just four languages.

Possible Alternative Organizations for Dictionary Entries

As we have noted with regard to several of the dictionaries discussed in this chapter,

many of the examples given for intransitive uses are in fact semantically identical to those

given for transitive uses; the di�erence has to do with the pragmatics of the situation and

whether the direct object can be null instantiated. Many dictionary entries could be made

much clearer by eliminating the transitive/intransitive distinction, provided that the readers

understand something about null instantiation.

We have also pointed out several cases where what are, properly speaking, collo-

cations appear as part of the regular entry, and vice versa.6 Maintaining the distinction

between these two might be easier if the listing of collocations indicated which of the regu-

lar senses was involved in each collocation. In fact, particularly in learners dictionaries, it

would be very helpful if lexicographers could indicate by means of some compact notation

which senses are extensions of, elaborations of, or otherwise derived from other senses. If

the senses are numbered, this need not occupy very much space. For example, to show

that sense 7 is derived (in some way) from sense 5, one could simply write (<5) at the

beginning of the entry for sense 7. Of course, including such information would require

the lexicographers to consider carefully the question of which sense is derived from which,

and which senses should be considered the most \central" or \basic". Such careful study

of semantic relations among the senses for every polysemous word might be prohibitively

expensive for dictionary publishers, although a similar amount of e�ort must be required

to write the etymologies that are included for most words in some dictionaries.

6(Louw 1995) found similar problems in a study of dictionary entries for the word run.

228

Chapter 6

Future Research Directions and

Conclusions

In the �rst part of this chapter, I will outline the research that needs to be un-

dertaken to answer some of the remaining questions Finally, I will attempt to synthesize

the contributions of the various approaches discussed in the last four chapters, to see what

generalizations can be made about the results.

6.1 Future Research Directions

Psycholinguistic Experiments

Further Analysis of Existing Data

There is certainly room for more analysis of the data already collected. My col-

league Jane Edwards and I are continuing to work on analyzing the data from the timed

tasks on both Experiments 2 and 3, eliminating outliers and �nding appropriate ways to

standardize the measurements. As noted in Section 4.4, however, a good analysis of the

Experiment 3 reaction time data will probably not be possible unless we gather more data

to reduce the number of missing values.

One of the questions which we wanted to answer was, \What is the nature of

individual variation in the semantics of see? For example, can speakers be categorized on

the basis of cognitive style (such as \lumpers" vs. \splitters")? (This might be a stable

personality trait, not limited to the particular range of senses in question.) Is there an

CHAPTER 6. FUTURE RESEARCH DIRECTIONS AND CONCLUSIONS 229

implicational hierarchy of levels of sense di�erentiation, so that any individual who makes

distinction A would also make distinction B?1 We are not ready to make any claims on this

point, but there are some indications that our subjects do vary on such a dimension.

The technique of clustering on the basis of agreement statistics, described in Sec-

tion 4.3 above, is useful in revealing certain aspects of the underlying structure of the senses.

It is, however, one-dimensional, and thus cannot reveal the complexity underlying the sense

divisions; in this respect, it has the same weakness as was noted above with regard to ex-

periments using similarity judgements. Several approaches for further, multidimensional

analysis are being considered.

One way of getting at the larger number of dimensions that are assumed to underly

the sense judgements would be to consider the responses on the classi�cation tasks as

(partially) independent dimensions, and to �nd a method to reduce their dimensionality.

The most conservative approach might be to consider each category as orthogonal to all

the others and look for responses to the same item in two or more categories that could

be reduced to a single dimension based on their relative frequencies. We are currently

considering the relative merits of discrete vs. continuous-valued representations, and of

various methods of reduction of dimensionality.

New Experiments

In accordance with Williams 1992, we expect that probes for senses closely related

to the prime will be facilitated more than probes for relatively distant senses. The distance

between senses might be measured along trees based on clustering by cross-rater agreement

(Section 4.3) or based on inheritance of frames (Section 2.5). However, we will probably need

to collect much more reaction time data (especially for the 14 senses used in Experiment 3)

before we can prove or disprove this hypothesis.

One question that naturally arises with regard to our data (especially from the last

two tasks) is whether, by the end of the experiment, the subjects have simply learned the

categories that we de�ned a priori and are responding on the basis of what they have learned

in the last hour, rather than on the basis of sense divisions which they had coming into

the experiment. It may be that the only way to resolve this question would be to create

1This might in some ways be similar to Kay & McDaniel's (1978) discovery of an implicational hierarchyfor distinctions of color terms across languages. In that case, however, the cross-linguistic di�erences haveto do with the existence of names for categories, which is much easier to prove than the discrimination ofcategories which do not have names, like those for the senses of see.


several sets of a priori categories, some of them \natural" and some of them including

what we really believe to be disparate examples. Presumably a set of categories which are

closer to those which subjects had initially will be easier to apply, and will produce higher

agreement on categorization of examples, both among subjects and with the experimenter's

categorizations. This would be a procedure similar to that used by Rosch (1973), in which

subjects were taught color names with two di�erent sets of exemplars, one containing focal

colors and one containing non-focal colors.

We would expect other verbs (and nouns) of perception to have similar semantic

structures, but most of them to be somewhat less elaborated, since see is one of the most

important. Thus, we expect to �nd at least a physical perception sense and a non-physical

cognition sense for verbs such as glimpse, look at , perceive, and hear and phrases such as

get a glimpse of , get/have a taste of , the smell of , etc. Many of these examples are easy

to construct or to �nd in corpora. It might be instructive to run similar experiments using

a variety of these perception words, to test what sense divisions speakers have for them,

whether this two-way distinction is reliable, and whether any other senses are similar to

the kinds of elaborations found with see, such as discourse deixis (e.g. I hear you, brother

Ì understand and sympathize with both your intended message and its implications'),

classi�cation (He perceives the proposal as just a delaying tactic), etc.

We would also predict other sorts of prototype e�ects in addition to those tested

so far, with more central senses more likely to be spontaneously produced and sentences

containing them to be better remembered. Standard experimental techniques could be used

for eliciting \best examples" and testing recall.

Corpus Studies and Word Sense Disambiguation

From Senses to Statistical Pro�les

The frame semantic description of the senses given in Section 2.5 is intended to be

precise and to contain all the appropriate information about the syntax and semantics of the

arguments of each sense at the appropriate level of generality. But all of this information

is discrete and categorical: a sense can occur with exactly the two arguments listed, the

phrase type of the seen can be either S�n or NP and nothing else, the semantic type of the

seer is human (not merely sentient), etc.

It seems clear from a variety of psycholinguistic experiments that speakers' knowl-


edge of language includes some knowledge of the relative frequencies of alternate forms for

the same meaning and alternate meanings for the same form. This need not be seen as

an additional \burden" on the language learner, a separate \coe�cient" to be stored along

with all the other syntactico-semantic knowledge. Rather, it should be a natural conse-

quence of a spreading-activation model of language learning and language processing. It

is probably premature to try to describe in much detail the relation between connectionist

mental models and the patterns of preferences displayed by speakers, but a realistic model

of the speaker's knowledge of senses should also include knowledge of the statistical pro�les

of the senses.

There have been a number of corpus-based studies on inducing the possible ar-

gument structures for various verbs from large corpora (e.g. Manning 1993), but most of

them have not had adequate sources of semantic knowledge about the arguments. Al-

though part-of-speech tagged corpora have been available for some time, we are just now

beginning to have parsers that are robust enough to allow us to record the syntax of the

arguments of a verb in adequately (i.e. beyond a bigram or trigram grammar based on

part-of speech tagging). At the same time, resources such as WordNet (Miller et al. 1990;

Fellbaum 1998) are enabling us to do at least limited semantic typing of the heads of the

argument NPs.2

We should therefore be in a good position to undertake a corpus study of the

argument structure of see, using the frame descriptions as a starting point and inducing

statistical pro�les of the syntax and semantics of the arguments of each sense. A similar

study is being undertaken by Dan Jurafsky and his associates for several thousand words,

many of them polysemous, in connection with the FrameNet project (Lowe et al. 1997;

Baker et al. 1998).

For the reasons just mentioned, we would expect that a careful study of corpus

examples of see will show that even though there is a considerable overlap in the syntax and

semantics of the arguments of di�erent senses, the relative frequencies of di�erent argument

patterns will be quite di�erent for di�erent senses (as shown in similar studies of other

2For the present, however, natural language processing systems must use primarily syntactic frames andrelatively \shallow" semantics, because \real" semantics not only involves more world knowledge than anycurrent NLP system incorporates, but also requires too much processing time. This is not necessarily a badthing; the time constraints on human language processing suggest that people also do a very broad, \shallow"sort of processing, (cf. Jurafsky 1992), rather than performing a \deep" logical sort of deduction such asthat envisioned by Hobbs et al. (1993). People do, however, integrate more types of evidence, especiallysemantic evidence, more rapidly than current AI systems.


words, e.g. Resnik 1993).

From Statistical Pro�les to Senses

A statistical pro�le of this sort is interesting in its own right from the linguistic

point of view, but for those working on NLP, it is only a means to the end of word sense

disambiguation, which is increasingly recognized as a key problem in building practical NLU

systems (witness the fact that an entire issue of Computational Linguistics was recently

devoted to the topic). The statistical study of FrameNet words mentioned above is typical

of the sorts of work now being done, in that it will use the hand-tagged examples of senses

of words from FrameNet as a training set for an algorithm that will then look through the

whole BNC for sentences containing those words and make a guess as to the sense of the

word and the roles of its arguments, based on the statistical similarity of the contexts to

those in the training set. (For an overview of current word sense disambiguation techniques,

see Ide & V�eronis 1998, and Light 1997.)

The real test of the statistical pro�les of the senses see developed in the �rst step

would therefore be how well they can be used to predict the sense from the syntax and

semantics of the arguments. Standard evaluation procedures such as cross-validation can

then be used to judge the accuracy of this method in comparison with other word-sense

disambiguation algorithms currently in use.

Cross-linguistic Survey

The purposes of a cross-linguistic survey would be (1) to test whether native

speakers of various languages actually use the terms listed in bilingual dictionaries to convey

the various meanings of see and (2) to �nd out the relative frequencies of the words used

in such cases. For this purpose, we could use a translation task; this is less than ideal from

a psycholinguistic point of view, since we are testing a very high-level skill (translation),

rather than a more direct meaning to form correspondence. However, it is much simpler to

gather cross-linguistic data in this way than to try to replicate a psychological experiment

in each of the countries, especially when dealing with meanings that are very di�cult to

depict, such as those of see.

The survey would include at least 40 English sentences containing see, covering

the range of senses. For each target language, at least 10 subjects would translate all of


the sentences into their native language, so that some idea could be obtained about variant

expressions for each sense.

As in the dictionary study, it is expected that some of the instances of see will be

uniformly (or nearly uniformly) translated by the basic verb for `see' in the target language.

Those uses could then be argued to constitute universal parts of human visual experience3 ,

and hence, plausibly grouped together as one general sense or at least a few related senses,

perhaps related by event metonymy, pro�ling of di�erent participants, etc. Other uses,

which are lexicalized in a variety of ways, would have a better claim to be treated as

separate, less related senses.

Other Methods

In addition to those mentioned above, other research methods would be required

to investigate predictions such as the following:

� We would expect that core senses of see such as eye and recognize will be older

historically and more likely than peripheral senses and collocations to serve as sources

for new extensions.

� We would expect that core senses would be acquired earlier and more completely than

than peripheral senses. (But note that Johnson's (1996) work on acquisition data

suggests that the earliest-acquired senses of see are more general, and the acquisition

process involves splitting larger semantic areas into �ner senses, rather than adding to

a collection of distinct senses.) We would also expect the central senses to be retained

better in vocabulary loss, whether due to disuse (as in language death), senility, or

external agents (such as brain damage or drugs).

By combining syntactic and semantic information in our frame de�nitions and

corpus studies, we may also in a sense be modeling the actual process of �rst language

acquisition. As Naigles et al. (1993:60) have suggested,

. . . the child confronts the problem of acquiring a verb lexicon from two im-perfect data bases. Both situations (extra-linguistic observation) and utterances(linguistic observation) provide only probabilistic evidence for the determinationof verb meanings. Yet we know that children acquire categorical (or close to

3Cf. Wierzbicka's (1996) language-universal semantic primitives, which now include see (eye).


categorical) knowledge of the verb lexicon. Our hypothesis is that they succeedby playing o� these two imperfect data bases against each other, seeking thesimplest �t between them.

Although I am certainly not proposing a serious attempt to model the richness of �rst

language acquisition, the complexity of linguistic phenomena suggests that many of these

problems can only be solved more or less as humans do, by making e�cient and concurrent

use of the full range of information available.

6.2 Conclusions

In some ways, we have barely begun to answer our initial research question, \How

many senses does see have, and how are they related to each other?" But the evidence

accumulated so far does allow us to reach certain conclusions. Although we have concen-

trated on a very small area of syntax and semantics, we have inevitably been involved in

many larger theoretical questions.

� There can be little doubt that see is in fact, highly polysemous, that it is unrealistic

to postulate only one (or even a few) more general senses, with all of its uses as

being created or understood \on the y" by means of processes such as metaphor,

type coercion, etc. The psycholinguistic evidence suggests that speakers have mental

representations (or processing strategies) that are at least partially separated for at

least a dozen senses, and several of the most distinctive senses were not even tested in

the experiment, to prevent the tasks from becoming too time-consuming. The variety

of translations for the various uses in bilingual dictionaries also leads to the same

conclusion.

More generally, although models of the lexicon which minimize the number of senses

have a certain theoretical appeal, the preponderance of evidence suggests that people

do store quite a number of relatively speci�c senses for highly polysemous words.

� On the other hand, the situation is very far from homonymy; most of the relations

among the senses of see are well motivated, even though particular senses are not

predictable. For example, even though the senses recognize, visit, and determine

cannot be reduced to mere uses of eye, the experiential basis of co-occurring events

relating them to eye helps explain both their likely historical development and the


acquisition of these senses by individuals. The partial sharing of the commonest

words for see in other languages suggests that this experiential basis is not limited to

English, although there are language-speci�c di�erences from English as well.

� Some of the less predictable senses such as accompany and experience must be

learned individually. Their connections with more central senses such as eye and

recognize are probably too distant to make them transparent in most contexts even

to native speakers. At least one sense, setting, being used only infrequently and

mainly in formal writing, may not even be learned by all speakers.

� Nothing in what we know about the mind or language requires that all words or all

concepts be learned in the same way. The varied relations we have found among

senses of see in this study demonstrate that abstraction over examples, metaphorical

mapping between domains, and analogical development of patterns of alternation, to

name just a few processes, can all play a part in the development of lexico-semantic

structures.

� There are numerous collocations which cover the entire range from frequent but com-

pletely compositional patterns such as I see what you mean to truly opaque, decoding

idioms (see the handwriting on the wall , see a man about a dog). An adequate theory

of lexical semantics must treat this as a continuum, rather than a dichotomy; most

collocations fall somewhere in the middle, with their semantics derived partially by

the composition of their parts and partially from the construction itself. Recognizing

and explaining the partial regularities is also helpful for second-language learners who

must learn (e.g.) to recognize variations on such constructions (We can see a very dim

light at the end of a very long tunnel.), and it will be essential for natural language

understanding systems. Any theory of linguistics which tries to separate language

into a highly regular syntax and a completely irregular lexicon will not be able to

properly represent what speakers know about these collocations.

� I have taken more or less for granted that there is great value in both introspective

and data-based approaches to these questions, and, indeed, the use of di�ering ap-

proaches in this study, including experimentation, has been fruitful. I believe that

this demonstrates that researchers in lexical semantics should use methods involving

more informants, corpus data, and, where appropriate, quantitative analysis together


with intuitive/introspective data. Other new types of evidence should be sought and

used when proven reliable; at the same time, linguists should avoid expectations that

one more bit of evidence from a related �eld will resolve all the major controversies

within linguistics.

We have not, however, really answered the question of how many senses of see we

should postulate, in part because di�erent criteria lead us to di�erent conclusions. If we join

the camp of the \splitters" and postulate separate senses for every form-meaning pair which

behaves di�erently syntactically, we will have more than the twenty-odd senses discussed

above, exclusive of collocations. If we side with the \lumpers" and make maximal use of

the interaction of the verb with its arguments and the surrounding context (i.e., Cruse's

(1986) modulation and contextual selection, Pustejovsky's (1995) type shifting, co-

ercion, co-composition, and semantic selection, or Alm-Arvius's (1993) Pragmatic

Expansion, Diversion, and Restriction), we might be able to get by with far fewer,

perhaps, like Alm-Arvius, with eight or nine senses.

There is also the question of possible di�erences between the receptive lexicon

and the productive lexicon; although we would like to think (for reasons of economy) that

they are one and the same, it is clear that speakers understand much more than they can

produce, both in their �rst and second languages. Consider, for example, what we called

\Compositional Uses" in Section 2.5. Some of these, such as tour, are encoding idioms

(Makkai 1966), i.e. they can be understood (decoded) immediately in context on the basis

of regular composition of the verb with its object. Speakers must learn, however, that this

is one of the ways that English encodes the idea of `touring' or `seeing the sights', which in

some other languages cannot be expressed by such a simple combination of the basic verb

for see and a place name. Thus, they need not be included in the receptive lexicon in order

to be understood, but they must be in the productive lexicon if they are to be used.

Ultimately, we may �nd that to ask the question \How many senses does see

have?" is to fall prey to what Langacker (1987:29) calls the Rule/List Fallacy:

. . . the assumption, on grounds of simplicity, that particular statements (i.e.lists) must be excised from the grammar of a language if general statements(i.e. rules) that subsume them can be established. . . . I have argued (Langacker1982a) that this is a specious kind of simplicity for anyone taking seriously thegoal of \psychological reality" in linguistic description. . . .We do not lose ageneralization by including both the rule and the speci�c plural forms in the


grammar of English, since the rule itself expresses the generalization. To claimon an a priori basis that the rule precludes the list, or conversely, is simply toembrace the exclusionary fallacy.

Later in the same book, Langacker gives the sentences in Ex. (1) as an example

of variants of teach which share a common base but di�er in their choice of primary land-

mark, and comments \In the usage-based model I propose, the variants in [Ex. (1)] are all

listed in the grammar, together with a schema representing their commonality (this schema

speci�es trajector status for the agent but is neutral in regard to the choice of primary

landmark)." (p. 270 fn.11)

(1) a. Sally teaches handicapped children.

b. Sally teaches mathematics.

c. Sally teaches third grade.

d. Sally teaches Sunday school.

(pp. 269-70, my numbering)

Langacker suggests that in most cases, we develop more abstract schemata for

linguistic objects, which are real, but secondary, evolving over time on the basis of a collec-

tion of like items.4 Sadock (1984) also gives some linguistic evidence that some idiomatic

expressions must be listed in the lexicon, even though they are completely compositional

and transparent; he gives examples such as cheeseburger and If you've seen one , you've

seen 'em all.

Certainly, some of the senses of see are more di�erent from each other than those

of teach in Ex. (1), so that the only generalization they share seems to be the lexical forms.

But for some of the agonizing questions about listing separate senses versus regarding them

as due to regular rules of composition and more abstract senses, the right answer may well

be \both". The mind/brain is \big" enough and exible enough to encompass multiple,

redundant representations.

4This is in accord with a well-known language acquisition phenomenon (Bloom 1994:32-3). Childrentypically (1) learn a pattern in a restricted context and begin using it correctly, then (2) begin to gener-alize and use the pattern incorrectly as a result of overgeneralization, and then (3) learn the appropriaterestrictions on the use of the (generalized) pattern, at which point the number of errors decreases again.The percentage of appropriate uses is high at �rst (within a narrow context), then low, then high again,giving a U-shaped learning curve.

238

Bibliography

Allen, R.E., H.W. Fowler, & F.G. Fowler (eds.) 1990. The Concise Oxford Diction-ary of Current English. New York: Oxford University Press, 8th edition.

Alm-Arvius, Christina. 1993. The English verb see. Gothenburg studies in English.G�oteborg, Sweden: Acta Universitatis Gothoburgensis.

Alvarez Garcia, Teresa. 1998. Collins Spanish-English, English-Spanish dictionary= Collins diccionario espa~nol-ingles, ingles-espa~nol . New York: HarperCollins, 3rdedition.

Apresjan, Ju. D. 1974. Regular polysemy. Linguistics 142.5{32.

Armstrong, Sharon L., Lila R. Gleitman, & Henry Gleitman. 1983. What someconcepts might not be. Cognition 13.263{308.

Asher, R.E. (ed.) 1994. The Encyclopedia of Language and Linguistics . New York:Pergamon Press, 1st edition.

Atkins, B.T.S. 1992. Theoretical lexicography and its relation to dictionary-making.Dictionaries 4{43.

|| 1994. Analysing the verbs of seeing: a frame semantics approach to corpus lexicography.In Proceedings of the Twentieth Annual Meeting of the Berkeley Linguistics Society , ed.by Susanne Gahl, Andy Dolbey, & Christopher Johnson. Berkeley Linguistics Society.

Baker, Collin F., 1999. Experimental mapping of sense structures of polysemous words.Presentation at the annual meeting of the Linguistics Society of America, Los Angeles.

|| forthcoming. Proving basic polysemy: Subjects reliably distinguish several senses ofSee. In Proceedings of the Twenty-�fth Annual Meeting of the Berkeley LinguisticsSociety, February, 1999 , volume 25. Berkeley Linguistics Society.

||, Charles J. Fillmore, & John B. Lowe. 1998. The Berkeley FrameNet Project.In COLING-ACL '98: Proceedings of the Conference.

Barsalou, Lawrence W. 1983. Ad-hoc categories. Memory and Cognition 11.211{227.

|| 1987. The instability of graded structure: Implications for the nature of concepts. InConcepts and Conceptual Development: Ecological and intellectual factors in catego-rization, ed. by Ulric Neisser, 101{40. Cambridge, U.K.: CUP.

BIBLIOGRAPHY 239

Bierwisch, Manfred, & R. Schreuder. 1989. From concepts to lexical items. Cognition42.23{60.

Bloom, Paul. 1994. Overview: Controversies in language acquisition. In LanguageAcquisition: Core Readings , ed. by Paul Bloom, 5{48. Cambridge, Mass.: MIT Press.

Bouchard, Denis. 1995. The semantics of syntax : a minimalist approach to grammar .Chicago, Ill.: University of Chicago Press.

Brugman, Claudia. 1996. Mental spaces, constructional meaning, and pragmatic ambi-guity. In (Fauconnier & Sweetser 1996), 29{56.

Carpenter, Bob. 1992. The logic of typed feature structures, with applications to uni-�cation grammars, logic programs, and constraint resolution. Cambridge tracts intheoretical computer science. New York: Cambridge University Press.

Castillo, Carlos, & Otto F. Bond (eds.) 1972. University of Chicago Spanish-EnglishEnglish-Spanish Dictionary . New York: Pocket Books, revised edition.

Chomsky, Noam. 1957. Syntactic structures . Janua linguarum nr. 4. 's-Gravenhage:Mouton.

||. 1965. Aspects of the theory of syntax . Cambridge: M.I.T. Press.

Cohen, J. 1960. A coe�cient of agreement for nominal scales. Educational and Psycho-logical Measurement 20.37{46.

Cohen, Jonathan D., Brian MacWhinney, Matthew Flatt, & Jefferson

Provost. 1993. Psyscope: An interactive graphic system for designing and control-ling experiments in the psychology laboratory using macintosh computers. BehaviorResearch Methods, Instruments & Computers 25.257{271.

Coleman, Linda, & Paul Kay. 1981. Prototype semantics: the english word Lie. Lan-guage 22.26{44.

Copestake, Ann, & Ted Briscoe. 1995. Semi-productive polysemy and sense extension.Journal of Semantics 12.15{67.

Cowart, Wayne. 1997. Experimental Syntax: Applying Objective Methods to SentenceJudgments . Thousand Oaks, California: Sage Publications.

Cruse, D. Alan. 1986. Lexical Semantics . New York: Cambridge University Press.

||. 1992. Monosemy vs. polysemy. Linguistics 3.577{599.

de Groot, A.M.B. 1984. Primed lexical decision: Combined e�ects of the proportionof related prime-target pairs and the stimulus-onset asynchrony of prime and target.Quarterly Journal of Experimental Psychology 36A.253{280.

Declerck, Renaat. 1982. The triple origin of participial perception verb complements.Linguistic Analysis 10.1{26.

BIBLIOGRAPHY 240

||. 1983. On the passive of in�nitival perception verb complements. Journal of EnglishLinguistics 16.27{46.

Durkin, Kevin, & Jocelyn Manning. 1989. Polysemy and the subjective lexicon:Semantic relatedness and the salience of intraword senses. JPLR 18.577{612.

Editing Group (ed.) 1988. A New English-Chinese dictionary . Seattle: University ofWashington Press, second revised and enlarged edition.

Fauconnier, Gilles. 1985 [1994]. Mental Spaces: Aspects of Meaning Construction inNatural Language. New York: MIT Press [reprinted by Cambridge University Press].

||, & Eve Sweetser. 1996. Spaces, worlds, and grammar . Chicago: University ofChicago Press.

Fellbaum, Christiane (ed.) 1998. WordNet: An electronic lexical database and some ofits applications . Cambridge, Mass: MIT Press.

||, Joachim Grabowski, Shari Landes, &Andrea Baumann. 1998. Matching wordsto senses in wordnet: Naive vs. Expert di�erentiation of senses. In (Fellbaum 1998).

Fillenbaum, Samuel, & Amnon Rapoport. 1971. Structures in the subjective lexicon.New York: Academic Press.

Fillmore, Charles J. 1969. Verbs of judging: An exercise in semantic description.Papers in Linguistics: International Journal of Human Communication, 1.91{117.

|| 1976. Frame semantics and the nature of language. In Conference on the Origin andDevelopment of Language and Speech, volume 280 of Annals of the New York Academyof Sciences , 20{32. New York Academy of Sciences.

|| 1982. Frame semantics. In Linguistics in the Morning Calm: Selected Papers fromSICOL-1981 , 111{137. Seoul, South Korea: Hanshin Publishing Co.

|| 1985. Frames and the semantics of understanding. Quaderni di Semantica: RivistaInternazionale di Semantica Teorica e Applicata 6.222{254.

|| 1992. \Corpus linguistics" vs. \Computer-aided armchair linguistics". In Directionsin Corpus Linguistics , 35{60. Mouton de Gruyter. Proceedings from a 1992 NobelSymposium on Corpus Linguistics, Stockholm.

|| 1994. The hard road from verbs to nouns. In In Honor of William S-Y. Wang:Interdisciplinary Studies on language and language change, ed. by Matthew Y. Chen& Ovid J.L. Tseng, 105{130. Taipei: Pyramid Press.

||, 1996. Notes on idiomaticity. Unpublished course materials for Linguisitcs 181 (LexicalSemantics).

||, & B.T.S. Atkins. 1992. Towards a frame-based lexicon: The semantics of risk andits neighbors. In Frames, Fields and Contrasts , ed. by Adrienne Lehrer & Eva FederKittay. Lawrence Erlbaum Associates.

BIBLIOGRAPHY 241

||, & B.T.S. Atkins. 1998. FrameNet and lexicographic relevance. In First InternationalConference on Language Resources and Evaluation, Proceedings , ed. by Antonio Rubio,Natividad Gallardo, Rosa Castro, & Antonio Tejada, 417{423, Granada, Spain.

||, & Paul Kay. 1994. Reading materials for Linguistics X20. Course reader at U.C.Berkeley.

||, & Paul Kay, 1996. Causative ABC constructions. Presentation at Cognitive ScienceSeminar at U.C. Berkeley, Jan. 26, 1996.

||, Paul Kay, & Mary Catherine O'Connor. 1988. Regularity and idiomaticity ingrammatical constructions: The case of Let Alone. Language 64.501{538.

Frawley, William. 1992. Linguistic semantics . Hillsdale, N.J.: L. Erlbaum Associates.

Fujikake, Eizo (ed.) 1984. T�oei's English-Japanese Japanese-English Dictionary . Tokyo:T�oei Shuppan.

Geeraerts, Dirk. 1993. Vagueness's puzzles, polysemy's vagaries. Cognitive Linguistics4.223{272.

||. 1994a. Homonymy. In (Asher 1994), 1595{6.

||. 1994b. Polysemy. In (Asher 1994), 3227{8.

Gibbs, Raymond. 1980. Spilling the beans on understanding and memory for idioms inconversation. Memory and Cognition 8.449{456.

||, & J. O'Brien. 1990. Idioms and mental imagery: The metaphorical motivation foridiomatic meaning. Cognition 36.35{68.

Goldberg, Adele. 1995. Constructions: A Construction Grammar Approach to Argu-ment Structure. Chicago: University of Chicago Press.

Gove, Philip Babcock. 1993. Webster's Third New International Dictionary of theEnglish Language, unabridged . Spring�eld, Mass.: Merriam-Webster.

Grice, H. Paul. 1975. Logic and conversation. In Speech Acts , ed. by Peter Cole & J.L.Morgan, volume 3 of Syntax and Semantics , 41{58. New York: Academic Press.

||. 1978. Further notes on logic and conversation. In Pragmatics , ed. by Peter Cole,volume 9 of Syntax and Semantics , 113{27. New York: Academic Press.

Gruber, Jeffery S. 1986. Frame information and lexically-based inference. Quaderni diSemantica 58{78.

Hanks, Patrick. 1992. Lexicography: Theory and practice. Dictionaries 97{112.

Harris, Randy Allen. 1993. The Linguistics Wars . Oxford University Press.

BIBLIOGRAPHY 242

Heid, Ulrich, &Katja Krueger. 1994. On the DELIS corpus evidence encoding schema.Deliverable D-III-0 of DELIS LRE 61.034, DELIS. EC Document.

Hobbs, Jerry, J.M. Stickel, P. Martin, & D. Edwards. 1993. Interpretation asabduction. Arti�cial Intelligence 63.69{142.

Ide, Nancy, & Jean V�eronis. 1998. Introduction to the special issue on word sensedisambiguation: The state of the art. Computational Linguistics 24.1{40.

Iwasaki, Tamihei, Jujiro Kawamura, & Sanki Ichikawa. 1960. Kenkyusha's newEnglish-Japanese dictionary on bilingual principles . Tokyo: Kenkyusha, new, revisedand enlarged edition.

Johnson, Christopher. 1996. Learnability in the acquisition of multiple senses: Sourcereconsidered. In Proceedings of the Twenty-SecondAnnual Meeting of the BerkeleyLinguistics Society , ed. by Matthew L. Judge Jan Johnson & Jeri L. Moxley, volume 22,469{480. Berkeley Linguistics Society.

Jorgensen, Julia C. 1990. The psychological reality of word senses. Journal of Psy-cholinguistic Research 19.167{190.

Jurafsky, Daniel. 1992. An on-line computational model of human sentence inter-pretation: A theory of the representation and use of linguistic knowledge. TechnicalReport 92/676, UC Berkeley Computer Science Department. Reprint of UCB Ph.D.Dissertation.

Kaneda, Fujihiko (ed.) 1997. Y�ohan English-Japanese Dictionary . Tokyo: Y�ohan Pub-lications.

Kay, Paul. 1983. Linguistic competence and folk theories of language: Two Englishhedges. In Proceedings of the Ninth Annual Meeting of the Berkeley Linguistics Society ,volume 9, 128{137, Berkeley, Calif. Berkeley Linguistics Society.

||, & Charles J. Fillmore. 1999. Grammatical constructions and linguistic general-izations: The What's X doing Y? construction. Language 75.1{33.

||, & Chad K. McDaniel. 1978. The linguistic signi�cance of the meanings of basiccolor terms. Language 54.610{46.

Kilgarriff, Adam. 1997. \I don't believe in word senses". Computers and the Humanities31.91{113.

Koenig, Jean-Pierre, & Daniel Jurafsky. 1994. Type underspeci�cation and on-line type construction in the lexicon. In Proceedings of the West Coast Conference onFormal Linguistics .

Lakoff, George. 1987. Women, Fire, and Dangerous Things . Chicago: University ofChicago Press.

BIBLIOGRAPHY 243

||. 1990. The invariance hypothesis: Is abstract reasoning based on image schemas?Cognitive Linguistics 1.

||. 1995. Re ections on metaphor and grammar. In Essays in Semantics and Pragmatics ,ed. by Masayoshi Shibatani & Sandra Thompson, number 32 (new series) in Pragmaticsand Beyond, 131{143. John Benjamins.

||, & Mark Johnson. 1980. Metaphors We Live By . Chicago: University of ChicagoPress.

||, & Eve Sweetser. 1985. Foreword. In (Fauconnier 1985 [1994]), ix{xvi.

Lakoff, Robin Tolmach. 1971. If's, and's and but's about conjunction. In Studies inLinguistic Semantics , ed. by Charles J. Fillmore & D. Terence Langendoen, 114{149.New York: Holt, Rinehart, and Winston.

||. 1989. The way we were; or, the real actual truth about generative semantics: Amemoir. Journal of Pragmatics 13.939{88.

Langacker, Ronald W. 1987. Foundations of Cognitive Grammar, Volume I: TheoreticalPrerequisites . Stanford, California: Stanford University Press.

|| 1991. Foundations of Cognitive Grammar, Volume II: Descriptive Application. Stan-ford, California: Stanford University Press.

Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary Investigation.Chicago: University of Chicago Press.

Levinson, Stephen C. 1983. Pragmatics . Cambridge Textbooks in Linguistics. NewYork: Cambridge University Press.

Liang, Shih-ch'iu, Jeffrey C.H. Tung, & C.H. Wang. 1975. Far East English-Chinesedictionary = Yu�and�ong Y�ing H�an D�a C��di�an. Taipei: Far East Book Co.

Light, Marc (ed.) 1997. Tagging Text with Lexical Semantics: Why, What and How? .Special Interest Group on the Lexicon, Association for Computational Linguistics. Pro-ceedings of the SIGLEX Workshop held April 4-5, in Washington, D.C., USA in con-junction with ANLP-97.

Louw, Johannes P. 1995. How many meanings to a word? In Cultures, Ideologies, andthe Dictionary: Studies in Honor of Ladislav Zgusta, ed. by Braj B. Kachru & HenryKahane, number 64 in Lexicographica: Series Maior, 357{65. T�ubingen: Niemeyer.

Lowe, John B., Collin F. Baker, & Charles J. Fillmore. 1997. A frame-semanticapproach to semantic annotation. In (Light 1997). Proceedings of the SIGLEX Work-shop held April 4-5, in Washington, D.C., USA in conjunction with ANLP-97.

Lyons, John. 1977. Semantics , volume I & II. New York: Cambridge University Press.

BIBLIOGRAPHY 244

Makino, Seiichi, & Michio Tsutsui. 1986. A Dictionary of Basic Japanese Grammar .Tokyo: The Japan Times.

Makkai, Adam, 1966. Idiom Structure in English. New Haven, Conn: Yale Universitydissertation. DAI 27.195A.

Manning, Christopher. 1993. Automatic acquisition of a large subcategorization diction-ary from corpora. In Proceedings of the 31st Annual Meeting of the Association forComputational Linguistics , 235{242.

McCawley, James D. 1982. How far can you trust a linguist? In Language, Mind, andBrain, ed. by Thomas W. Simon & Robert J. Scholes, 75{88. Hillsdale, N.J.: Erlbaum.

Medin, Douglas L. 1989. Concepts and conceptual structure. American Psychologist44.1469{1481.

Mel'cuk, Igor A. 1996. Lexical functions: A tool for the descrition of lexical relations in alexcion. In Lexical Functions in Lexicography and NLP , ed. by Leo Wanner, volume 10of Studies in Language Companion Series , 37{102. Amsterdam/Philadelphia: JohnBenjamins.

Meyer, D.E., & R.W. Schvanenveldt. 1971. Facilitation in recognizing pairs of words:Evidence of a dependence between retrieval operations. Journal of Experimental Psy-chology 90.227{234.

Miller, George A., Richard Beckwith, Christiane Fellbaum, Derek Gross, &Katherine J. Miller. 1990. Introduction to wordnet: An on-line lexical database.International Journal of Lexicography 3.235{312. Special issue, George Miller, ed.

Morey, Leslie C., & Alan Agresti. 1984. The measurement of classi�cation agreement:An adjustment to the rand statistic for chance agreement. Education and PsychologicalMeasurement 44.33{37.

Naigles, Letitia G., Henry Gleitman, & Lila R.Gleitman. 1993. Children acquireword meaning components from syntactic evidence. In Language and cognition: adevelopmental perspective, ed. by Esther Dromi, volume 5 of Human development .Norwood, N.J.: Ablex Pub. Corp.

no author given. 1986. An Everyday Chinese-English dictionary (= Changyong Han-Ying Cidian). New York, N.Y.: Hippocrene Books.

||. 1995. Merriam-Webster's Pocket Dictionary . Spring�eld, Mass.: Merriam-Webster,Inc.

||. 1997. Webster's New World Dictionary . New York: MacMillan.

Norvig, Peter, & George Lakoff. 1987. Taking: A study in lexical network theory.In Proceedings of the Thirteenth Annual Meeting of the Berkeley Linguistics Society ,volume 13, 195{206, Berkeley, California. Berkeley Linguistics Society.

BIBLIOGRAPHY 245

Osgood, C.E. 1970. Where do sentences come from? In Semantics: An interdisci-plinary reader in philosophy, linguistics and psychology , ed. by D.D. Steinberg & L.A.Jakobovits. Cambridge, U.K.: CUP.

Ostler, Nicholas, & B.T.S. Atkins. 1991. Predictable meaning shift: Some linguisticproperties of lexical implication rules. In Lexical Semantics and Knowledge Representa-tion, ed. by James Pustejovsky & Sabine Bergler, 76{87. Association for ComputationalLinguistics. Proceedings of a Workshop Sponsored by the Special Interest Group onthe Lexicon.

Pearsall, Judy, & Patrick Hanks. 1998. The New Oxford Dictionary of English.Oxford: Clarendon Press.

Petruck, Miriam R.L. 1995. Frame semantics and the lexicon: Nouns and verbs inthe body frame. In Essays in semantics and pragmatics: in honor of Charles J. Fill-more, ed. by Masayoshi Shibatani & Sandra A. Thompson, number 32 (new series) inPragmatics & beyond, 279{97. Philadelphia: John Benjamins Pub. Co.

Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, Mass.: MIT Press.

Quine, Willard V.O. 1960. Word and Object . Cambridge, Mass.: MIT Press.

Reddy, Michael. 1979. The conduit metaphor. In Metaphor and Thought , ed. byA. Ortony, 284{324. Cambridge, UK: Cambridge University Press.

Resnik, Philip, 1993. Selection and Information: A Class-Based Approach to Lexicalrelationships . University of Pennsylvania dissertation.

Rhodes, Richard Alan, n.d. Some comments on the polysemy of see. Unpublished ms.UC Berkeley Linguistics Dept.

Rosch, Eleanor. 1973. Natural categories. Cognitive Psychology 4.328{50.

||. 1978. Principles of categorization. In Cognition and Categorization, ed. by EleanorRosch & Barbara B. Lloyd, 27{47. Hillsdale, N.J.: Lawrence Erlbaum Associates.

||. in press. Reclaiming concepts. In Reclaiming Cognition: The primacy of action,intention and emotion, ed. by R. Nu~nez & W.J. Freeman. Thorverton, U.K.: ImprintAcademic.

Rothstein, S., 1983. The Syntactic Forms of Predication. Cambridge, Mass.: MITdissertation. Distributed by the Indiana University Linguistics Club, 1984.

Ruhl, Charles. 1989. On monosemy: a study in linguistic semantics . Albany, N.Y.:State University of New York Press.

Sadock, Jerrold M. 1984. The poly-redundant lexicon. In Papers from the Parasessionon Lexical Semantics , ed. by David Testen, Veena Mishra, & Joseph Drogo, 250{269,Chicago. Chicago Linguistic Society.

BIBLIOGRAPHY 246

Sandra, Dominiek, & Sally Rice. 1995. Network analyses of prepositional meaning:Mirroring whose mind|the linguist's or the language user's? Cognitive Linguistics6.89{130.

Sch�utze, Carson T. 1996. The Empirical Base of Linguistics: Grammaticality judgmentsand Linguistic Methodology . University of Chicago Press.

Scott, J. 1955. Reliability of content analysis: the case of nominal scale coding. PublicOpinion Quarterly 19.321{325.

Seidenberg, M.S., M.K. Tannenhaus, J.M. Leiman, & M. Bienkowski. 1982. Au-tomatic access of the meanings of ambiguous words in context: Some limitations onknowlege-based processing. Cognitve Psychology 14.489{532.

Siegel, Sidney, & N. John Castellan, Jr. 1988. Nonparametric Statistics for theBehavioral Sciences . New York: McGraw-Hill, 2nd edition.

Sinclair, John, forthcoming. The search for units of meaning.

Sweetser, Eve. 1987. Metaphorical models of thought and speech: A comparison ofhistorical directions and metaphorical mappings in the two domains. In Proceedings ofthe Thirteenth Annual Meeting of the Berkeley Linguistics Society , ed. by Jon Aske,Natasha Beery, Laura Michaelis, & Hana Filip, volume 13, 446{459, Berkeley, Califor-nia. Berkeley Linguistics Society.

||. 1990. From etymology to pragmatics: metaphorical and cultural aspects of semanticstructure. New York: Cambridge University Press.

Swinney, David A. 1979. Lexical access during sentence comprehension: (re)considerationof context e�ects. Journal of Verbal Learning and Verbal Behavior 18.645{59.

Talmy, Leonard. 1985. Lexicalization patterns: Semantic structure in lexical forms. InGrammatical categories and the Lexicon, ed. by Timothy Shopen, volume 3 of Languagetypology and semantic description, 57{149. Cambridge, U.K: Cambridge UniversityPress.

Taylor, John R. 1995. Linguistic Categorization: Prototypes in Linguistic Theory .Oxford University Press, 2nd edition.

Traugott, Elizabeth. 1989. On the rise of epistemic meanings in English: An exampleof subjecti�cation in semantic change. Language 65.31{55.

Turner, Mark, & Gilles Fauconnier. 1995. Conceptual integration and formal ex-pression. Metaphor and Symbolic Activity 10.183{204.

Urdang, Laurence, & Frank Abate. 1983. Idioms and phrases index . Detroit, Mich.:Gale Research Co., 1st edition.

Viberg, �Ake. 1983. The verbs of perception: a typological study. Linguistics 21.123{162.

BIBLIOGRAPHY 247

Wasow, Thomas. 1985. Postsript. In Lectures on Contemporary Syntactic Theories ,number 2 in CSLI Lecture Notes. Stanford, California: CSLI.

Wierzbicka, Anna. 1985. Lexicography and Conceptual Analysis . Ann Arbor, Michigan:Karoma Publishers, Inc.

||. 1996. Semantics: Primes and Universals . Oxford University Press.

Williams, John N. 1992. Processing polysemous words in context: Evidence for interre-lated meanings. Journal of Psycholinguistic Research 21.193{218.

Zadeh, Lotfi. 1965. Fuzzy sets. Information and Control 8.338{353.

Zipf, George Kingsley. 1949[1965]. Human behavior and the principle of least e�ort:an introduction to human ecology . New York: Hafner Pub. Co.

Zwicky, Arnold M., & Jerrold M. Sadock. 1973. Ambiguity tests and how to failthem. Working Papers in Linguistics (Columbus, Ohio) 16.1{34.

BIBLIOGRAPHY 248

249

Appendix A

Additional Corpus Examples of

Senses

A.1 Basic Senses

(1) eye

a. BNC: I had seen her face before, but only in silhouette. . .b. BNC: May I see the room now?c. BNC: I could not see the heap; it was hidden by brambles nowadays.d. BNC: I saw him in the forest a couple of days ago.

(2) recognize

a. Brown: . . . the low-wage textile-producing countries in Asia and Europe will seethat \dumping" practices cause friction all around and may result in importquotas.

b. BNC: If there is to be a conspiracy, at least everyone will see that it exists andwill know who is involved.

(3) accompany

a. BNC: \No need to see me to my room, Alec," she said, reaching for the candle-stick.

b. BNC: \I'll see you home again on Friday, love," she said to Susan.c. BNC: I'm sure your sister can see us to the gate.

(4) condition

a. BNC: The daughter of Samuel Roberts, quite simply, could not be seen to beinvolved.

b. BNC: \Can't you understand that he's the last person on the sta� I want to seeupset?"

(5) consult

APPENDIX A. ADDITIONAL CORPUS EXAMPLES OF SENSES 250

a. Brown: After his pains got worse, Tom decided to see a real doctor, from whomhe learned he was su�ering from cancer of the lung.

b. BNC: . . .Ruud Gullit, who is seeing a doctor on Monday about his cartilageinjury.

c. BNC: You've been seeing your social worker regularly, have you, Vernon?d. BNC: . . . she had a child . . . and a husband called Bernard who was seeing a

fertility specialist.

(6) dating

a. Brown: Against my folks' wishes, we'd been seeing each other for short rides inthe truck.

b. BNC: . . . he broke it o� and confessed to Sarah that he had been seeing someoneelse.

c. BNC: My boyfriend and I are both 24 and have been seeing each other for threeyears.

d. BNC: . . . \Your son has been seeing my sister clandestinely."e. BNC: He had been seeing Molly on and o� for about six months.f. BNC: \She is seeing other men."g. BNC: Meanwhile Mick is reported to have been seeing glamorous 21-year-old

Aussie model Peta Wilson in Los Angeles.

(7) determine

a. Brown: \. . . I used to follow Williams every day in the box score, just to seewhether he got a hit or not".

b. BNC: They make parachute drops \to see what it's like".c. BNC: Let's see how you rate in that area.d. BNC: There were two engines, \Carlisle" and \No 1", and we would look eagerly

to see which was heading the train today.e. BNC: Try your local stamp dealer and see if he has any on his lists.f. BNC: She peeped at him to see the e�ect of her words.

(8) ensure

a. Brown: \I shall see about getting you a tent," he said.b. Brown: In order to see that this hindering situation remained e�ective, Wash-

ington detached several bodies of his troops to the periphery of the Philadelphiaarea.

c. Brown: You should see to it that the trap, the dirt-catcher in front of the �lter,is always clean.

d. BNC: We won't be disturbed for an hour and a half at least|Kate's out, andthen she'll be seeing to the meal.

e. BNC: `You see to the drinks.

(9) envision

a. Brown: In Arthur Clarke's Childhood's End [1953], . . . , we can see the bright


vision of science �ction clearly de�ned.

(10) faculty

a. Brown: \How can you tell an insane man to reason or a blind man to see"?b. Brown: He could see in almost total darkness.

(11) news

(12) process

a. BNC: The meeting was opened by the chairman, who welcomed those presentand was pleased to see so many attend.

b. BNC: He was layering a hedge in one of his top �elds when he saw the partyriding along the road towards the forest.

c. Brown: Despite his yearning, the colonel would not go down to see the mencome through the lines.

d. BNC: He's seen me talking to the great Walter Schellenberg.e. BNC: . . . \Is that why you'd been drinking, the night before you saw the rob-

bery on Handley Plain?"f. BNC: Through half-closed eyes he saw Alice rise from the bench and come

towards him.g. BNC: I mean, you've never actually seen anyone going in or out of them, have

you?"

(13) read

a. Brown: Andy did not see the newspapers the next day.

(14) setting

(15) visit

a. Brown: She hadn't seen him since.b. BNC: The Chirwas say they came to see a sick relative.c. BNC: \It's wonderful to see you."d. BNC: It's best I don't see Anna.e. BNC: \I'll see you about six this evening."f. BNC: \I'm always pleased to see old friends, Doreen."

A.2 Semi-collocations

(16) classify

a. Brown: Pope John sees the renewal and puri�cation of the Church as an ab-solutely necessary step toward Christian unity.

b. Brown: The indignant crusader sees in the nude or semi-nude human form athreat and danger. . .


c. BNC: . . . he does not see himself as a romantic orphan.d. BNC: He may have seen her death as a judgement.e. BNC: She looked at them, and saw dupes.

(17) experience

(18) gambling

(19) vide

a. Brown: The range is from 14 to 25 inches; . . . [see chapter on Laying, Brooding,Hatching, and Birth], the latter on a \normal" newly born individual.

b. BNC: (See use of ap on take-o� on pages 93 and 94.)c. BNC: . . . (see, for example, Hatim, 1984, 1987, 1988, 1989; Hatim and Mason,

1990), . . .d. BNC: For a detailed discussion of information structure, see section 5.1.2. be-

low.

A.3 Compositional Uses

(20) audience

a. BNC: \Nurse Dungarvan, why did you not explain at once that Mr Leland wasseeing a patient in here?"

b. BNC: The servants say the doctor's been three times but been forced to leaveagain without seeing him.

c. BNC: . . . more teeth were being extracted because dentists were seeing morepatients.

(21) discourse

a. BNC: Can you see the point I am trying to make?b. BNC: A �rm basis for the study of Oriental art came more slowly, and as

we shall see, some of the di�erences of approach between East and West stillrequire wider recognition.

c. BNC: \I see what you're getting at," he said.d. BNC: He does, however, modify this position slightly in his later writings as

we shall see shortly.

(22) hallucinate

(23) scan

a. Brown: SAMOS will be hard put to see through clouds|and to see in thedark.

b. Brown: The total picture is only seen by the camera which integrates the manysector scans over the entire 90-degree rotation period.

c. BNC: Where the camera sees white it sends to the output monitor the picture


seen by Camera A, where it sees black it sends out Camera B's picture.d. BNC: . . . the right hemisphere may recognise genuine words when it sees them,

. . .which suggests that . . . the right half of the brain is not inferior to the left.e. BNC: Each shape it sees is compared to a set of stylised representations of the

alphabet, . . .f. BNC: Again, the view�nder image will alert you to the problem by showing

you how the camcorder is seeing the scene.

(24) spectate

a. Brown: Later, a bus will carry members to the Chicago Stadium to see JackKramer's professional tennis matches at 8 p.m.

b. Brown: So I went to see \La Dolce Vita".c. BNC: \You should see his impersonation of Ray Bradbury," Cecil added.

(25) tour

254

Appendix B

Summary of Morphology and

Syntax of Senses

Some of the syntactic and semantic restrictions on senses discussed in Chapter 2are summarized in Table B.1 on the next page.

The second column indicates for each sense whether physical vision is presupposed,impossible, or unspeci�ed. We should note that blind people use senses very similar to manyof those listed here to mean `perceive by touching', `visit', `consult', etc. without the usualentailment of physical vision, so the items in this column may only apply when the seerpossesses normal vision. This may constitute a dialect di�erence for blind speakers, but atleast one such use is common to virtually all speakers, Let me see it , accompanied by holdingout the hand to receive a small object. The universal children's trick of responding to thisrequest by holding up the object but not handing it over and the resulting frustration of thespeaker, suggest that what is intended by the request is not only a closer visual inspection,but also a tactile one.

The next column shows whether a passive form exists for each sense, while thelast column tells whether or not the sense can occur with progressive aspect.

The items marked with (�) in the column for Progressive usually cannot occur inthe progressive, but can be \coerced" to take the progressive in the right context (Ex. (1-a)),just like many other verbs which normally resist the progressive (Ex. (1-b)) (also cf. Frawley(1992:147) on converting statives to actives by making them progressive).

(1) a. As the mist clears, he's gradually seeing more and more of the trees on the ridge.

b. As she matures, she's liking candy and ice cream less and co�ee more.

It is unclear how the facts represented by this table can be motivated from the analysispresented so far.

APPENDIX B. SUMMARY OF MORPHOLOGY AND SYNTAX OF SENSES 255

Sense Vision Subject Passive Tense Prog.

eye + A + any (�)a

recognize � A + any (�)a

process u A � any (�)a

visit + H � any any

classify u H + any (�)a

determine u A � any any

read + H + any any

spectate + H + any any

condition +/u A � any (�)a

envision � H � any (�)a

ensure u A? + any any

consult + H + any any

setting � place/time � ? any

see �t � H � any (�)a

esperience � b � any any

dating + H ? any (+)c

let's see � 's d � pres �

news + H � any (�)a

see through x � H ? any any

see x through � e � any any

see o� ? A + any any

faculty + A � any any

gambling ? H + pres? �

hallucinate �? A + any any

y'see � H � pres �

scan �? inan. + any any

a Usually not progressive, but see the text for exceptions.

b Must be a physical object; may be human.

c Usually progressive but there are exceptions in the right context.

d The colloquial form must be contracted; let us see is di�erent, i.e., sense discourse .

e See the discussion in the text for details of the semantics of the subject of this sense.

Table B.1: Syntactic Restrictions on Senses

256

Appendix C

Experiment 1

C.1 Original nineteen senses

(With de�nitions and examples seen by the subjects)

eye

Perceive an object or a physical action with the eyes

She saw the cat on the mat.

He could see something moving in the bushes.

visit

Meet, go to visit (a person/persons)

Will you see Fritz while you're in Pocatello?

She's been to see her twice since her operation.

determine

Find out, determine

I just wanted to see if you were paying attention.

Let's see what happens to the reaction at 250 degrees.

see as

Regard something as something else

Do you see this o�er as the �rst step in a hostile takeover?

In her he saw the possibility of escape from the boredom of his daily life.

recognize

APPENDIX C. EXPERIMENT 1 257

Recognize a fact, understand a situation

They saw the di�culty of winning, but still insisted on bringing it to a vote.

She saw that nothing would change his mind about it.

see state

To recognize something/ someone as being in a condition

He saw her hungry and cold on the street corner.

I only want to see you happy again.

see process

To see something/ someone perform a process

He saw her remember an item, pick it up, look at it, and then decide not to buy it.

She saw him gaining more political power month by month.

envision

Envision a counterfactual situation/ the future

Thank heavens they didn't try the north face; I could just see them on the cli� at midnight,freezing and exhausted.

I don't see the Republicans nominating him this year.

read

Read, consult a written reference

Did you see Herb Caen's column this morning?

See page 37.

spectate

Be a spectator at/of a performance/event

Have you ever seen a cricket match?

She went to New York to see \Angels in America".

service

To undergo, to witness and be a�ected by

He saw combat in Libya and Morocco.

I've seen lonely times when I could not �nd a friend.


contain

(Of a time period, a place, etc.) to contain, be the site or occasion of

The 16th century saw an unprecedented owering of literary talent.

California has seen its share of natural disasters in the last few years. (lexical Construc-tion) let's see (consider carefully)

Let's see, the jewelry will be unguarded for at least �ve minutes...

Let's see, there should be another box of staples here somewhere...

see to

(lexical Construction) see to / see to it that/ see that{to take charge of:

We'll see to it that he never bothers anybody again.

She always saw that there was a fresh pot of co�ee ready for the meeting.

you see

you see, I see (discourse markers)

As we will see in the next chapter, various solutions have been proposed.

You see, there's only so much medicine can do in this case.

see �t

to see �t (to V)

Pat had it for a week before seeing �t to tell me about it.

You can handle the situation however you see �t.

consult

To visit with an expert or authority, asking questions

You should see a doctor about that cough.

Going to see a divorce lawyer is a major decision.

dating

To visit with repeatedly for romantic reasons, BE seeing

They'd been seeing each other for a couple of years.

He had a girlfriend last year, but he's not seeing anybody now.

see news


to learn about news by seeing a newspaper or TV newscast

I see that Congress �nally passed the budget.

I see that the Governor is running for President again.

C.2 Stimuli

Brown Corpus Sentences (selected)

(1) (50 sentences randomly selected out of 200)

De�nition of the thighs at the uppermost part is quite commonly SEEN in mostchampionship Olympic lifters which is easily understandable.In the afternoon Miss Hosaka and her mother invited me to go with them andyoung Mrs. Kodama to SEE the famous Spring dances of the Geisha dancers.I SAW that I would soon run out of buildings at this rate, so I decided to takeanother measure|the whole state of Pennsylvania.Here he couldn't be SEEN by Blue Throat and his gang.It must have hurt her even to walk, for the sole was completely o� her left footand Morgan SAW that it was bruised and bleeding.Hereby, the external object viewed by the eyes remains the thing that is SEEN,not the retinal image, the purpose of which would be to achieve perceptivecooperation by stirring sympathetic impulses in the other sensory centers, motortensions, associated word symbols, and consciousness.Now as you step inside, instead of SEEING particles orbiting around like planets,you SEE waves and ripples very much like the ripples that you get on the surfaceof a pond when you drop a stone into it.I would try to memorize landmarks and SAW in a half-hour that it was hopeless.Now turn around so I can SEE your face.The next time the police SAW her she was dead.He went down the hall to Eugene's bathroom, to turn on the hot-water heater,and on the side of the tub he SAW a pair of blue wool swimming trunks.I wish you could have SEEN the crests fall on these two sparring coxcombs whenI told them that obviously the pasture belonged to their wives jointly.Evidence of this trend can best be SEEN in the recent activities of such leadingcompanies in the �eld as Advance Neon Sign Co., Los Angeles, Calif.Do you SEE that pretty girl standing next to the car with slacks on?On the other side of the ledger is the fact that he did SEE his niece and thewoman with whom she was staying.Third, the process of calci�cation is SEEN to begin later and to continue muchlonger for these boys than for the girls, a fact which con�rms data for othergroups of children.You'll have to go to town to SEE the doc.You don't SEE me stretched out on the deck, do you?


There was a tap at the door and Oliver entered with the word that Heiser wishedto SEE the Captain.Lying awake at night, he could SEE them, laid out on the oor of his mind.Apparently he was not a participant in the college or university theatricals,which he once attacked as utterly unworthy performances [SEE Apology, 3:300];but even in that famous passage, Milton was aiming not at the theatricals assuch but at their performance by `persons either enter'd, or presently to enterinto the ministry'.In contrast to the nuclear changes described above, another change in musclenuclei was SEEN, usually occurring in �bers that were somewhat smaller thannormal but that showed distinct cross-striations and myo�brillae.He lived in the dawn; he could only SEE the light coming over the horizon.At �rst I did not know what she meant; I thought she must be SEEING me assome one who had just come from SEEING her grandmother, in their distanthome-city.Chandler, looking to right and left to SEE how his men were faring, suddenlySAW another �gure bounding up the hill, hurling grenades and hollering thebattle cry as he ran.But in order to keep Letch in the public eye and out of trouble, I wrote in apart especially for him|that of a dashing ru�an who \SEES the light" and issaved by the inspiring example of Mother Cabrini.I didn't SEE her till several days later at the wedding, and her face looked likeit had never had a blemish on it.Not until the words had been spoken did Abel suddenly SEE the old house andthe insistent sea, and feel his contrition blotted out in one shameful moment ofcovetousness.I SAW holes in planes at the airport and in cars in the streets.The manager sat behind the group so he could SEE and count the hands thatwent up, and the director wrote the numbers on the blackboard.I never SAW him.Somehow our contemporary Moloch must be induced to SEE reason.He could not SEE objects as uni�ed, self-contained, and organized �gures, as aperson does with normal vision.Havana was �lled with an excitement which you could SEE in the brightness ofmen's eyes and hear in the pitch of their voices.He provoked outraged editorials when, after a post-Inaugural inspection of theWhite House with Mrs. Kennedy, he remarked to reporters, \We just cased thejoint to SEE what was there".\Well"|I didn't|I didn't ever want to SEE that woman again.The ledger was full of most precise information: date of laying, length of incu-bation period, number of chick reaching the �rst week, second week, �fth week,weight of hen, size of rooster's wattles and so on, all scrawled out in a handthat looked more Chinese than English, the most jagged and sprawling Alexhad ever SEEN.


At the same time he watched carefully to SEE how one attached pegboards tostone walls, but Mr. Blatz was usually standing in his line of vision and it allseemed so simple that he didn't like to disclose his ignorance.The meaning of this, as we shall SEE, is that he had no fund of visual memory-images of objects as objects; and, therefore, he could not recognize even long-familiar things upon SEEING them again.But there's one thing I never SEEN or heard of, one thing I just don't thinkthere is, and that's a sportin' way o' killin' a man!\Would you like to SEE my work?" Helva asked, politely.The Chicago contingent of modern critics follow Aristotle so far in this directionthat it is hard to SEE how they can compare one poem with another for thepurpose of evaluation.This would mean, it can readily be SEEN, that, again, for each new visual ex-perience, the tracing motions would have to be repeated because of the absenceof visual imagery.He SAW no life, but still stood there for a time peering at the unlovely hills, hisgaze continually returning to Papa-san.That is, it is literally a picture window: you don't SEE into the viscera; youSEE a picture|trees, or owers.Charity as she knew it was complex and reciprocal, and almost every roof sheSAW signi�ed charity.At last they SAW Calcutta, largest city of Bengal and the Caravan's destination.Some have SEEN revivalism and the search for Christian perfection as thefountain-head of the American hope.With distaste I SAW him assume a pompous air.\I never SAW him so anxious before", I said, lighting my pipe and o�ering hera cigarette.

Constructed Sentences

(2) She's been seeing him for years, but it doesn't look like anything will come ofit.She saw her working in the garden almost every day.I can see why they put him under house arrest.I've seen better days, but I'm not dead yet.Did you see Seinfeld last night?Modern ri es saw use on the battle�eld beginning with the Crimean War.W.W. II saw the use of Navajo as a secret code.Let's see, what time can you be at the station?I can see it now, in few more years there will be warning labels on ice cubesabout breaking a �lling.If you have these kinds of feelings often, perhaps you should see a psychiatristand talk about them.You see, the production of ivory has been outlawed for many years.You should see a lawyer before you sign anything.


I see that Macy's is having a sale on sheets.Did you see that they're going to raise bus fares again?They saw the snow swirling through the trees falling silently and gently into thelake.He often drank a great deal, but I never saw him drunk.The school sees to it that every child gets a hot lunch every day.The year 1723 saw Bach in Leipzig, in his new post at St. Thomas' Church.I haven't seen her for more than a year.Have you seen the new comet?You'll be needing some help to see to the plowing in the spring, won't you?He saw the keys hanging on the wall, just out of reach.He saw the sign that said he had to walk �ve more miles to reach the beach andgroaned.He said hello to her everyday, but never really SAW her.See what you can �nd out about the chances of getting a job there.She hasn't been seeing anyone since she moved to New York.You can �re him if you see �t; I'll back you up.I saw the headlines at the train station, but I haven't gotten the whole story.He wants to see her a success in her own right, not always depending on others.When we get their response, we'll see whether they're really willing to negotiateand make some progress or they're just stalling.The justices saw �t to make their ruling extremely narrow.Let's see, I'm about �ve foot ten, and she's a little taller than me...Can't you just see his face when he opens this package and tries to �gure outwhat to do with them?Despite the obstacles, his con�dence and character will see him through.I only saw Monk once, and he was so drunk he could hardly play.You can see mold is coming through the paint again.When you see an old friend for the �rst time in years, you usually spend a littletime catching up on the news of each other's lives.She saw him as an ally in her struggle for justice.I see much happiness in your future.He feigned innocence, but she saw right through him.You see what you're doing to yourself?Seeing that she's corporate vice president, it's not surprising that her o�ce isso big.I see it as a golden opportunity for you, and I wish you'd do it.I saw her in the ower of her youth, beautiful, generous, and con�dent of thefuture.

263

Appendix D

Stimuli for Experiments 2 and 3

(1) Stimuli for Experiment 2.

061 modl stmt + determine acad actvYou should see if you can �nish all your required courses by next spring.

055 past stmt + determine acad actvWe wanted to see whether the use of a catalyst would allow the reaction to takeplace at lower temperatures.

092 will ques - determine acad actvWon't you please see if Prof. Williams is in?

090 none stmt - determine acad passThe committee doesn't always see whether the credentials are valid before reach-ing its decision.

042 modl ques - determine entr actvShouldn't we see whether the full orchestra will �t on that little stage �rst?

057 modl stmt - determine entr actvThe people in the top rows won't be able to see who has the ball.

150 past ques - determine entr actvWhy didn't they see if they could get a �eld goal at that point?

014 past stmt - determine entr actvI assumed it was sold out, so I didn't even see if there were any tickets left.

075 modl ques + determine pers actvCould you see if the oven is hot yet?

040 none stmt + determine pers actvHe always sees if she needs anything from the store before he comes home.

091 will stmt + determine pers actvLet's call Mary and see if she's free for lunch.

095 past ques + ensure acad actvWho saw to the posting of the �nal grades for the course?

106 modl ques + ensure entr actvDoes the coach have to see that all the players are on the bus on time?

088 none ques - ensure entr actvDoesn't the coach see that the players get a good night's sleep before the game?

APPENDIX D. STIMULI FOR EXPERIMENTS 2 AND 3 264

008 none stmt + ensure entr actvThey see that the car is in perfect condition before every race.

013 will ques - ensure entr actvWhile I slice the cake, won't you see that everyone has something to drink?

124 will stmt - ensure entr actvTheir manager is useless{he won't even see to it that the band arrives on time.

074 modl ques - ensure pers actvCan't the baby sitter see to it that the kids have their baths on time?

012 modl stmt + ensure pers actvParents should see that their children learn basic safety rules at an early age.

127 modl stmt - ensure pers actvWithout a job, he could no longer see to it that his children had enough to eat.

122 past stmt - ensure pers actvNo one saw to the livestock for several days, so they were in bad shape.

064 will ques + ensure pers actvWho will see that the lawn gets mowed during your vacation?

097 past ques - ensure pers passWhy wasn't his injury seen to right away?

112 modl ques - experience acad actvShouldn't this beautiful old building see some new use rather than being torndown?

079 none ques + experience acad actvDoes this room see use for both humanities and science lectures?

101 none stmt + experience acad actvMany new Ph.D.s see several years of part-time teaching before landing a per-manent job.

006 none stmt - experience acad actvAcademic robes don't usually see use except at graduations.

069 past stmt + experience acad actvProf. Ramirez saw service as Dean of Humanities before becoming Chancellor.

089 will stmt + experience acad actvI'm sure you'll see better days once you've gotten your degree.

067 modl ques + experience entr actvIn fact, the �eld might see a lot of use this summer as a baseball camp.

027 none ques - experience entr actvDon't most players see a few years in the minor leagues before they get their�rst chance at the majors?

109 past ques + experience entr actvWhy did the women's swim team see its best season in years, while the men'steam did so poorly?

041 will ques - experience entr actvWon't the old stadium see use again for track and �eld events?

100 modl stmt + experience pers actvHe would've seen combat if the war hadn't ended suddenly.


139 past ques - experience pers actvDidn't you see duty in Italy during the War?

071 past stmt - experience pers actvHe was in Vietnam for a year, but he didn't see combat there.

068 modl ques - eye acad actvCan't you see the book the professor is holding?

020 modl stmt + eye acad actvYou might see a cheat sheet on a chair in the exam room.

060 modl stmt - eye acad actvAfter taking biochem, I couldn't see a green leaf without thinking of photosyn-thesis.

117 none stmt + eye acad actvI always see a lot of delivery trucks on campus in the morning.

065 will stmt - eye acad actvShe won't see the new computer until she comes into the o�ce.

062 modl ques + eye acad passCan the people at the front of the procession be seen by those at the rear?

056 none ques + eye acad passIs a di�erent color seen as a result of these contaminants?

099 will ques - eye acad passWon't the new lab equipment be seen by anyone who walks in?

126 none ques - eye entr actvDon't they see the same oats in the parade every year?

148 none stmt - eye entr actvI don't see anything in his sleeves, but he keeps pulling birds out of somewhere.

132 past stmt - eye entr actvThe other team didn't see the ball until it was too late.

050 will ques + eye entr actvWill the audience see both their faces all during the �ght scene?

053 past ques - eye entr passWeren't a lot of red tee-shirts seen in the streets just before the Big Game?

070 past ques + eye pers actvDid you see that ashy red coat they had in the window last week?

043 past stmt + eye pers passTwo suspicious-looking men were seen near the back door of the bank shortlybefore the robbery.

138 will stmt + eye pers passYou'll be seen by the neighbors if you sneak over the fence.

098 modl ques + faculty acad actvIf he could see better, would he be doing better academically?

145 modl stmt + faculty acad actvAs long as she can see well, she can continue doing library research.

141 modl stmt - faculty acad actvIf I can't see better tomorrow, I'll go to the health center.


017 none ques + faculty acad actvDoes Prof. Willams still see well enough to use a computer?

093 none stmt - faculty acad actvHe doesn't see very well any more, but he somehow keeps up with publicationsin his �eld.

007 past stmt + faculty acad actvHe saw well enough to drive to campus, but only during the day.

123 past stmt - faculty acad actvShe didn't see very well during the last few years of her life, but she continuedto teach courses.

114 none ques - faculty pers actvDoesn't your son see better now that he wears glasses?

130 none stmt + faculty pers actvSince her operation, she sees much better.

086 will ques - faculty pers actvWon't she see much better with her new glasses?

019 modl ques - recognize acad actvDon't they see the library should be open longer hours?

024 modl stmt + recognize acad actvSurely she must see that this student should have gotten a better grade.

044 past stmt - recognize acad actvAt �rst I didn't see that the result of the experiment could be interpreted intwo ways.

009 will ques + recognize acad actvWhen will they see that we need to o�er at least two semesters of physicalchemistry?

084 will ques - recognize acad actvWon't they see that the committee is opposed to the idea?

108 none stmt - recognize entr actvHe doesn't see that she really cares about him.

015 past ques + recognize entr actvAt what point did the manager see that he just had to change pitchers?

066 modl ques + recognize pers actvCould they see the cat was mad at them?

031 modl stmt - recognize pers actvShe might not see that he's having a bad e�ect on her, but all her friends do.

051 none stmt + recognize pers actvI see that he loves his job, but I don't understand why.

134 past ques - recognize pers actvDidn't you see that you were bound to get into trouble hanging around withpeople like that?

021 past stmt + recognize pers actvHe saw that she was serious about setting up her own company.

128 will stmt + recognize pers actv


Eventually he'll see that this isn't what he wants to do for the rest of his life.023 will stmt - recognize pers actv

She'll never see what a great guy he is unless she spends time with him outsidethe o�ce.

144 modl ques + setting acad actvShould the campus see a larger entering class next year?

005 past ques - setting acad actvWhy didn't the nineteenth century see as many wars as the twentieth?

036 past stmt + setting acad actv1985 saw the establishment of Cognitive Science programs at four major uni-versities.

046 will ques + setting acad actvWill next year see another drop in minority enrollment?

085 will stmt - setting acad actvThe campus won't see that many demonstrations again unless there's anotherwar.

034 none ques + setting entr actvDoes o�-Broadway see as many talented writers now as it did in the 70s?

001 none stmt + setting entr actvThis weekend sees �ve new major movies opening across the country.

058 none stmt - setting entr actvPittsburgh doesn't see musicals of this caliber very often.

063 will ques - setting pers actvWon't Ohio see more oods this spring after all that snow?


(2) Stimuli for Experiment 3.

050 past stmt + determine acad actvWe wanted to see whether the use of a catalyst would allow the re-action to take place at lower temperatures.

052 will ques - determine acad actvWon't you please see if Prof. Williams is in?

048 modl ques - determine entr actvShouldn't we see whether the full orchestra will �t on that little stage�rst?

053 past ques - determine entr actvWhy didn't they see if they could get a �eld goal at that point?

051 will stmt + determine pers actvLet's call Mary and see if she's free for lunch.

099 will stmt - ensure entr actvTheir manager is useless{he won't even see to it that the band arriveson time.

098 modl ques - ensure pers actvCan't the baby sitter see to it that the kids have their baths on time?

103 modl stmt + ensure pers actvParents should see that their children learn basic safety rules at anearly age.

107 modl stmt - ensure pers actvWithout a job, he could no longer see to it that his children hadenough to eat.

120 will ques + ensure pers actvWho will see that the lawn gets mowed during your vacation?

080 none stmt + experience acad actvMany new Ph.D.s see several years of part-time teaching before land-ing a permanent job.

074 will stmt + experience acad actvI'm sure you'll see better days once you've gotten your degree.

073 none ques - experience entr actvDon't most players see a few years in the minor leagues before theyget their �rst chance at the majors?

078 modl stmt + experience pers actvHe would've seen combat if the war hadn't ended suddenly.

071 past ques - experience pers actvDidn't you see duty in Italy during the War?

042 modl stmt - eye acad actvAfter taking biochem, I couldn't see a green leaf without thinking ofphotosynthesis.

043 none ques - eye acad actvDon't you see the book the professor is holding?

041 none ques + eye acad pass


Is a di�erent color seen as a result of these contaminants?037 past stmt + eye pers pass

Two suspicious-looking men were seen near the back door of the bankshortly before the robbery.

039 none stmt + eye acad actvI always see a lot of delivery trucks on campus in the morning.

121 modl ques + faculty acad actvIf he could see better, would he be doing better academically?

127 past stmt - faculty acad actvShe didn't see very well during the last few years of her life, but shecontinued to teach courses.

123 none ques - faculty pers actvDoesn't your son see better now that he wears glasses?

133 none stmt + faculty pers actvSince her operation, she sees much better.

126 will ques - faculty pers actvWon't she see much better with her new glasses?

054 modl ques - recognize acad actvDon't they see the library should be open longer hours?

057 will ques - recognize acad actvWon't they see that the committee is opposed to the idea?

059 none stmt - recognize entr actvHe doesn't see that she really cares about him.

070 past ques - recognize pers actvDidn't you see that you were bound to get into trouble hangingaround with people like that?

069 past stmt + recognize pers actvHe saw that she was serious about setting up her own company.

086 past ques - setting acad actvWhy didn't the nineteenth century see as many wars as the twentieth?

094 past stmt + setting acad actv1985 saw the establishment of Cognitive Science programs at fourmajor universities.

097 will stmt - setting acad actvThe campus won't see that many demonstrations again unless there'sanother war.

085 none stmt + setting entr actvThis weekend sees �ve new major movies opening across the country.

096 will ques - setting pers actvWon't Ohio see more oods this spring after all that snow?|The preceding were retained from Exp. 2.|The following are new in Exp. 3.

117 past stmt + accompany acad actvAfter the lecture, they had dinner with the speaker and saw him back


to his hotel.131 past stmt - accompany acad actv

If you work late in the lab, you should phone for an escort to see youto your car.

116 will stmt + accompany entr actvI promise I will see her to her apartment after the movie.

061 modl ques + accompany pers actvMay I see you to your car?

093 modl ques - accompany pers actvWhy couldn't your mother see Mrs. Latrobe to the door?

146 modl stmt - accompany pers actvYou should not see her home unless she asks you to.

067 none ques - accompany pers actvDoesn't he usually see visitors to the door when they leave?

035 past stmt + accompany pers actvA crowd of well-wishers saw the newly-weds to the door of their newhouse.

104 will ques - accompany pers actvAren't you at least going to see him to the bus stop?

134 none ques + condition acad actvHow often does one see an important question de�nitively answeredby a single experiment?

063 none stmt - condition acad actvI never see Prof. Liu stumped by students' questions.

088 past ques + condition acad actvDid you ever see him in a good mood?

143 will stmt - condition acad actvYou're not going to see her relaxed and enjoying life again until aftershe graduates.

065 will stmt + condition entr actvI hope we'll see him fully recovered in time for the next game.

136 none stmt + condition entr passIn the �nal minutes of the �lm, they are seen happy and satis�edwith their relationship.

044 modl ques + condition pers actvDo you think we might see Madeline rich and famous within the nextfew years?

075 modl stmt + condition pers actvIf you can really rest for a few days, we should see you in good shapeagain by Monday or Tuesday.

110 modl stmt - condition pers actvWe might not see her cheerful again for a long time.

036 none ques - condition pers actvHow come I never see you angry or upset?


122 past stmt + condition pers passHe was often seen drunk in one of the little cafes around closing time.

125 modl stmt - consult acad actvYou should see Prof. Johansen if you want suggestions about whatto read on this topic.

111 none ques + consult acad actvDid you see the dean to talk about hiring more teaching assistantsfor that new course?

101 none ques - consult acad actvDon't you see your advisor before you register for the next term?

135 past ques - consult acad actvWhy didn't you see someone in the Registrar's o�ce about your prob-lem with registration?

076 past stmt + consult acad actvShe saw a lawyer to �nd out if she could sue the university for sexualharassment.

068 modl ques + consult entr actvShould he see a lawyer before he signs the contract to play for them?

105 none stmt + consult pers actvWhenever he's sick for more than a day, he sees his doctor and followshis instructions.

130 none stmt - consult pers actvShe never sees a doctor until she's almost too sick to get out of bed.

087 past ques + consult pers actvDid she see a lawyer before she rewrote her will?

049 past stmt - consult pers actvI think she didn't see the minister to talk about her problem becauseshe was afraid of what he would tell her.

109 will ques - consult pers actvAren't you going to see a doctor about that cough?

077 will stmt - consult pers actvI'm not going to see my academic advisor until after the end of theterm.

070 modl ques - envision acad actvCan't you just see him now, getting his Ph.D. from Bill Gates Uni-versity in the year 2025?

092 none stmt + envision acad actvWithin the next �ve years, I see the University o�ering at least halfof its courses on the Internet.

066 past ques - envision acad actvBefore they hired her, didn't the department see her teaching bothundergraduate and graduate courses?

033 modl ques + envision entr actvCan you see Robin Williams playing George Washington in a comedy


about the U.S. Revolution?032 none ques + envision entr actv

Do you see any Indian �lms winning at the Cannes Film Festival nextyear?

137 none ques - envision entr actvYou don't see him making the team anytime soon, do you?

060 past ques + envision entr actvWhen the band �rst started, did you see yourselves winning a grammywithin two years?

031 past stmt - envision entr actvWhen he closed his eyes, he did not see his father's angry scowl, buthis mother's welcoming smile.

072 modl stmt + envision pers actvI can see myself now, sipping champagne and watching the sunsetfrom my yacht.

034 none stmt - envision pers actvSomehow, I don't see the Green Party winning the presidency in theyear 2000.

089 past stmt + envision pers actvI originally saw myself graduating in four years, but I still haven't.

055 modl ques - hallucinate entr actvIf she'd gotten a concussion, wouldn't she be seeing things and hearingvoices?

083 past stmt - hallucinate entr actvIt took him a few minutes to recover after being tackled so hard, buthe wasn't seeing stars or anything.

091 modl ques + hallucinate pers actvIf this drug is as dangerous as they say, should I start to see the wallsbreathing sometime soon?

030 modl stmt + hallucinate pers actvIf his fever rises again, he might see people who aren't there or failto recognize even you.

028 none ques - hallucinate pers actvDon't you sometimes see spots in front of your eyes if you stand upsuddenly?

040 none stmt + hallucinate pers actvAlcoholics who suddenly stop drinking often see snakes or bugs crawl-ing around them.

100 past ques + hallucinate pers actvHow long after you took the drug did you see the unicorn in yourbackyard?

029 past stmt + hallucinate pers actvI didn't exactly see dragons ying in the living room, but I did noticethat colors and smells seemed more intense.


141 will stmt + hallucinate pers actvIf he keeps drinking like that, he's gonna be seeing pink elephantssoon.

045 will stmt - hallucinate pers actvIf he'll stay on his medication, he won't see monsters hiding in theshadows everywhere.

142 modl ques - process acad actvIf he were really interested in this kind of research, wouldn't we seehim attending conferences on the topic?

138 modl stmt + process acad actvThe teacher can see some students falling behind schedule early inthe semester.

114 will stmt - process acad actvWe're not going to see her working in the lab much, now that she's�nished that experiment.

128 will ques + process acad passDo you think it's likely that we'll see him lecturing again next semester?

112 none ques - process entr actvDoesn't the audience see the soldiers warming around the camp�reas soon as the curtain opens?

047 none stmt - process entr actvYou almost never see a quarterback kick a �eld goal.

058 past ques + process entr actvHow many times did Americans see Rock Hudson kiss Doris Day onscreen?

064 none stmt + process entr passOut�elders are often seen jumping for a ball just before it goes overthe fence.

081 modl ques + process pers actvHow soon might we see the �rst cars rolling o� the production line?

124 modl stmt - process pers actvHe might not see himself getting older, but his friends can certainlytell the di�erence.

147 none ques + process pers actvDo you see her jogging around the lake every morning?

084 past ques - process pers actvDidn't you see her getting into a car with him shortly before mid-night?

062 past stmt + process pers actvHe saw them laughing together and asked what the joke was.

038 past stmt - process pers actvI haven't seen him walking the dog lately{I hope he's OK.

132 will ques - process pers actvWon't the boss see me walking across the parking lot?


113 will stmt + process pers passHe'll probably be seen raising money for good causes for many moreyears.

082 modl stmt + visit acad actvIf I went to my high school reunion, I could see all the people whonever left my home town.

129 modl stmt - visit acad actvIf she goes to France for her sabbatical, she might not see her studentsfor a whole year.

108 none stmt - visit acad actvWe don't usually see our colleagues very much when they're on sab-batical.

095 past ques - visit acad actvDidn't you see some of your former students while you were at theannual conference?

056 past stmt + visit acad actvI saw an old classmate of mine when I visited New York last summer.

102 past stmt - visit acad actvShe's been holed up writing her thesis and hasn't seen any of herfriends for weeks.

118 will ques - visit acad actvAren't you going to see all your friends at the college while you're intown?

090 none ques + visit entr actvDo you think movie stars see their parents during the holidays, or arethey always o� skiing in Europe?

079 modl ques + visit pers actvCould I see you and your husband next time I'm in the neighborhood?

115 modl ques - visit pers actvShouldn't you see your sister if you're going to be in L.A. for a week?

119 will stmt - visit pers actvI won't have time to see Alice in Denver unless I miss my plane.

275

Index

activation, 13, 136, 137, 231Allen et al. 1990, 178Alm-Arvius 1993, 15, 45, 83, 236Alvarez 1998, 184Apresjan 1974, 21Armstrong et al. 1983, 9, 10Atkins 1992, 167Atkins 1994, 121

Baker et al. 1998, 39, 231Baker, C. F. 1999a, 132Baker, C. F. 1999b, 132Barsalou 1983, 9Barsalou 1987, 10Bierwisch & Schreuder 1989, 5Bloom 1994, 237Bouchard 1995, 4, 36Brugman 1996, 81, 86

Carpenter 1992, 125Castillo & Bond 1972, 186Chinese, 15, 36, 164, 173, 179, 180, 187{

197, 215{217, 223{225, 227Chomsky 1957, 28, 55Chomsky 1965, 3, 42, 55Cohen 1960, 142Cohen et al. 1993, 146Coleman & Kay 1981, 32, 33collocations with see

let's see, 105, 144, 176, 191, 210see �t , 106, 174, 210see o� , 106, 175, 192, 210see one's way clear , 105, 210see reason, 107, 210see through x , 107, 144, 192, 197, 210see x through, 107, 108, 144, 175, 191,

192, 210

y'see, 108, 176, 210audience, 100

compositional uses of seeaudience, 56, 57, 72, 92, 130, 175,

205, 210{212, 215, 217, 220discourse, 57, 116, 129, 205, 206,

208, 210, 211, 220hallucinate, 48, 57, 102, 111, 114,

115, 129, 154, 157, 159, 204, 205,209{211

scan, 48, 57, 103, 126, 173, 210spectate, 48, 57, 87, 90, 103, 104,

129, 174, 179, 190, 204, 205, 210,211, 215, 217

tour, 48, 57, 104, 129, 174, 179, 191,204, 205, 211, 217, 223, 236

Copestake & Briscoe 1995, 21corpus, 3, 31{33, 39, 44, 45, 58, 60, 99,

109, 121, 123, 124, 139{141, 143,144, 164, 165, 175, 230, 231, 233,235

Cowart 1997, 29, 134Cruse 1986, 12, 15, 21{24, 37, 49, 51, 236Cruse 1992, 138

Declerck 1982, 121Declerck 1983, 41deGroot 1984, 134Durkin & Manning 1989, 134, 136

elaboration, 105, 227, 230Everyday Chinese-English dictionary 1986,

196

Fauconnier 1985, 6, 111, 112Fauconnier & Sweetser 1996, 112Fellbaum 1998, 231Fellbaum et al. 1998, 12, 146

INDEX 276

Fillenbaum & Rapoport 1971, 34Fillmore 1969, 34Fillmore 1976, 6, 39Fillmore 1982a, 6, 39Fillmore 1985, 39Fillmore 1992, 31Fillmore 1994, 39, 56, 85, 121Fillmore 1996, 105Fillmore & Atkins 1992, 6, 23, 24, 39, 166Fillmore & Atkins 1998, 61Fillmore & Kay 1994, 6, 41, 42Fillmore & Kay 1996, 6Fillmore et al. 1988, 6frame semantics, 230

de�nition, 5{6re exives in, 126{130

FrameNet Project, vi, 33, 231, 232frames

accompany, 68, 130description, 67

ascertain, 72, 75description, 75

bare condition, 85bare event, 84, 85description, 84

bare state, 70description, 69

categorization, 123description, 92

cause

description, 76cognition inchoative, 75description, 65

consultation with authority, 130description, 70

contain, 89description, 88

contention, 61directed motion, 67description, 66

gambling cards

description, 97gambling

description, 97

imagine

description, 79intend

description, 76intentional causation

description, 77perception basic, 61description, 63

perception distant, 61description, 63

perception faculty, 81, 82description, 81

reasoning, 61reciprocal event, 100n, 130description, 90

romantic involvement

description, 73see accompany, 68description, 67

see classifydescription, 93

see conditiondescription, 69

see consultdescription, 71

see datingdescription, 73

see determinedescription, 75

see discourse

description, 101see ensure, 79description, 77

see envision

description, 80see experience

description, 95see eye, 61, 63, 64, 83, 91description, 64

see facultydescription, 81

see gamblingdescription, 98

see news

INDEX 277

description, 82see processdescription, 85

see readdescription, 87

see recognize, 61description, 65

see scan

description, 103see setting, 89description, 88

see vide

description, 99see visit

description, 91service as, 96description, 94

social interaction, 91description, 90

talk, 61Fujikake 1984, 207

Geeraerts 1993, 15, 16Geeraerts 1994a, 15, 20Geeraerts 1994b, 15Gibbs 1980, 33Gibbs & O'Brien 1990, 33Goldberg 1995, 6, 37Gove 1993, 172Grice 1975, 117Grice 1978, 117Gruber 1986, 39

Hanks 1992, 165{167Harris 1993, 55Heid & Krueger 1994, 165historical linguistics, vii, 35{36Hobbs et al. 1993, 231homonymy, 11, 13{15, 18, 20, 37, 49, 134{

137, 234

Ide & Veronis 1998, 232Iwasaki et al. 1960, 203

Japanese, 15, 32, 122, 164, 173, 180, 197{208, 217{219, 225{227

Johnson 1996, 34, 233Jorgensen 1990, 33, 134, 161Jurafsky 1992, 231

Kaneda 1997, 207Kay 1983, 16Kay & Fillmore 1999, 6Kay & McDaniel 1978, 229Kilgarri� 1997, 44Koenig & Jurafsky 1994, 125

Lako� & Johnson 1980, 10Lako�, G. 1987, 9, 10, 28, 69, 108, 122Lako�, G. 1990, 3Lako�, G. 1995, 119{121Lako�, G. & Sweetser 1985, 111Lako�, R. 1971, 124Lako�, R. 1989, 55Langacker 1987, 4, 10, 236Langacker 1991a, 86language acquisition, vii, 233, 234Levin 1993, 22, 23, 25, 55, 56, 94Levinson 1983, 173Liang & Tung 1983, 190Light 1997, 232Louw 1995, 227Lowe et al. 1997, 39, 231Lyons 1977, 22

Makino & Tsutsui 1986, 203Makkai 1966, 12, 104, 236Manning 1993, 231McCawley 1982, 29measures of agreement, 141{143

kappa, 141{143, 148{150, 155omega, 141, 143, 147

Medin 1989, 9Mel'cuk 1996, 21mental spaces, 6, 62, 111{113, 117, 131

de�nition, 112Merriam-Webster's Pocket Dictionary 1995,

177metaphor, 6, 7, 10, 11, 15, 22, 25, 27,

29, 82, 89, 106{108, 112, 119{123,125, 131, 153, 173{175, 185, 191,234, 235

INDEX 278

metaphorical extension, 36, 103, 106,122, 174, 175

Meyer & Schvaneveldt 1971, 134Miller et al. 1990, 231monosemy, 14, 15, 18{20, 25, 35, 138Morey & Agresti 1984, 143

Naigles et al. 1993, 233natural language processing (NLP), 3, 176,

231, 232New English-Chinese dictionary 1987, 179Norvig & Lako� 1987, 68

Osgood 1970, 34Ostler & Atkins 1991, 21, 22

Pearsall & Hanks 1998, 174Petruck 1995, 39Pustejovsky 1995, 7, 22, 78, 83, 236

Quine 1960, 16

Reddy 1979, 11Resnik 1993, 231Rhodes ms, 59roles

beneficiary, 126, 129boundary, 89category, 93, 94, 129cognizer, 75consultant, 130consulter, 130container, 89contents, 89content, 75n, 120cook, 126, 129event, 85experiencer, 120facts, 75food, 126gambler, 130goal, 68, 82, 120, 121interior, 89opponent, 130participant, 85role, 95

seen, 42, 43, 59, 64, 66, 68, 72, 76,78, 80, 82, 87, 89, 91, 93, 94, 96,100{103, 106{109, 111, 112, 115,120, 125, 126, 129, 130, 151, 157,166, 204, 206, 230

mental spaces account of, 112-119seer, 42, 59, 60, 66, 68, 72, 75n, 76,

78, 86, 87, 89, 91, 93, 94, 96, 100,101, 103, 106{109, 113, 115{119,125, 126{129, 130, 149, 151, 174,230

service, 95source, 120stimulus, 120undergoer, 95

Rosch (in press), 8Rosch 1973, 230Rosch 1978, 7, 8Rothstein 1983, 86Ruhl 1989, 25, 31, 36, 138

Sadock 1984, 29, 237Sandra & Rice 1995, 33schema, 5, 10, 237Schuetze 1996, 10, 28, 29Scott 1955, 142Seidenberg et al. 1982, 135, 136senses of see

accompany, 46, 57, 68, 130, 144, 154,155, 159, 161, 174, 175, 177, 178,184{186, 191, 205, 206, 209{211,215, 221, 235

classify, 47, 50, 57, 58, 94, 104, 106,126, 174, 175, 178, 185, 196, 205,206, 208, 210, 211, 217, 220

condition, 46, 56, 57, 59, 85, 87, 94,106, 108, 124, 154, 155, 159, 177,178, 190, 191, 209, 210

consult, 46, 51, 56, 57, 64, 72, 92,100, 130, 154, 159, 175, 178, 184,190, 197, 205, 210{212, 215, 220,223

dating, 43, 46, 57, 74, 130, 175, 210determine, 41, 46, 56{58, 72, 103,

105, 129, 145, 149{153, 159, 161,

INDEX 279

173{178, 186, 190{192, 205, 206,208{211, 215, 217, 221, 223, 234

ensure, 46, 50, 57, 72, 78, 79, 103,121, 129, 145, 149{153, 174, 176{178, 184{186, 191, 192, 205, 206,208{211, 215, 217, 220, 221

envision, 46, 56, 57, 109, 111, 114,115, 118, 126, 129, 154, 159, 173,174, 185, 186, 205, 206, 209{211,221

experience, 47, 57, 95, 96, 129, 145,149{153, 159, 173, 177, 184{186,190, 191, 205, 206, 208, 210, 211,215, 221, 235

eye, 45, 46, 48, 49, 51, 53, 54, 56{59,76, 81, 83, 91, 92, 100, 102{104,109, 113, 120, 121, 139, 143{145,149{155, 157, 159, 161, 166, 172{175, 177, 179, 185, 190, 191, 204,205, 207{212, 215, 217, 220, 221,223, 225, 233{235

faculty, 42, 47, 56, 57, 82, 112, 119,145, 149{154, 159, 161, 175, 178,191, 204, 205, 209{212, 215, 217,220, 223

gambling, 47, 57, 59, 130, 175, 178,191, 210, 211

news, 47, 56, 57, 83, 84, 114, 116,117, 126, 129, 174, 185, 210, 211,221

process, 47, 56, 57, 70, 85{87, 90, 96,103, 104, 154, 155, 159, 173, 175,177, 185, 190, 209{211, 215, 220

read, 47, 54, 56, 57, 83, 84, 88, 100,117, 129, 174, 210, 211

recognize, 45, 46, 48, 56{59, 66, 79,83, 84, 86, 94, 103, 115, 116, 118,120{122, 129, 143, 145, 149{154,159, 166, 168, 173{177, 185, 186,190, 191, 205, 206, 208{211, 217,220, 223, 233{235

setting, 47, 54, 57{59, 89, 96, 129,145, 149{153, 159, 173, 205, 210,211, 235

vide, 48, 56, 57, 129, 197, 204, 205,210, 211, 217, 220

visit, 47, 49, 51, 53, 56, 57, 64, 74, 91,92, 100, 114, 130, 154, 159, 169,175, 177, 178, 184{186, 190, 197,204, 205, 208{212, 215, 217, 220,221, 223, 234

Siegel & Castellan 1988, 143Sinclair forthcoming, 36Spanish, 69, 164, 173, 180{187, 211, 213,

214, 217, 220{222, 227Sweetser 1987, 10, 121Sweetser 1990, 22Swinney 1979, 134, 136synonymy, 11, 18, 22, 33, 34, 37, 45, 48{

50, 133, 136, 166, 171, 173

Talmy 1985, 165Taylor 1995, 27Traugott 1989, 29Turner & Fauconnier 1995, 61

Urdang & Abate 1983, 104

Viberg 1983, 121, 165

Wasow 1985, 55Webster's NewWorld Dictionary 1997, 177Wierzbicka 1985, 31Wierzbicka 1996, 16, 25, 26, 233Williams 1992, 33, 136, 137, 139, 151, 229

Zadeh 1965, 9zeugma, 18, 26, 32, 34, 48, 51{54, 108Zipf 1949, 1Zwicky & Sadock 1973, 17

Seeing clearly: Frame Semantic, Psycholinguistic, and Cross ...

Documents