The Generative Lexicon

The Generative Lexicon

The Generative Lexicon

James Pustejovsky

The MIT Press Cambridge, Massachusetts London, England

Second printing, 1996 © Massachusetts Institute of Technology All rights reserved. No part of this book may be reproduced in any form by any electronic or mechanical means (including photocopying, recording, or information storage and retreival) without permission in writing of the publisher. This book was set in Computer Modern by author and printed and bound in the United States of America. Library of Congress Cataloging-in-Publication Data Pustejovsky, J. (James) The Generative Lexicon / James Pustejovsky. p. cm. Based on the author's unpublished manuscript, "Towards a Generative Lexicon." Includes bibliographical references (p. ) and indexes. ISBN 0-262-16158-3 1. Semantics. 2. Generative grammar 3. Computational linguistics. I. Title P325.P85 1995 401'.43-dc20 95-32875 CIP

Contents

Preface

ix

Acknowledgements

xi

1

Introduction

1

2

The Nature of Lexical Knowledge

5

2.1 Semantic Classes and Categorial Alternation 2.2 Interlexical Relations

8 23

3 3.1 3.2 3.3 3.4

The Logical Problem of Polysemy Varieties of Sense Extension Contrasitive Ambiguity Complemetary Polysemy An Elemetary Lexical Semantic Theory

27 27 29 31 33

4 4.1 4.2 4.3 4.4 4.5 4.6 4.7

Limitations of Sense Enumerative Lexicons The Goals of Lexical Semantic Theory The Creative Use of Words Permeability of Word Senses Difference in Syntactic Forms Semantic Expressiveness Generative Lexical Models Strong vs. Weak Compositionality

39 40 42 46 50 55 57 59

5 5.1 5.2 5.3 5.4 5.5

The Semantic Type System Levels of Representation Argument Structure Extended Event Structure Qualia Structure The Interaction of Semantic Levels

61 61 62 67 76 81

vi

Contents

6 Qualia Structure 6.1 Modes of Explanation 6.2 The Qualia Structure of Nominals

85 85 90

7 7.1 7.2 7.3 7.4

Generative Mechanisms in Semantics Coertion and Type Shifting Co-composition Selective Binding Semantic Selection

105 106 122 127 131

8 8.1 8.2 8.3 8.4 8.5 8.6

The Sematics of Nominals Basic Issues Nominals with Unified Types Nominals with Complex Types Propositions and Event Descriptions Varieties of Nominalization Lexicalisation of LCPs

141 141 142 149 157 165 177

9 9.1 9.2 9.3 9.4 9.5 9.6

The Lexical Semantics of Causation How Language Encodes Causation Causation and Uncaussativity Aspectual Causatives and Coertion Experiencer Predicates Modal Causatives Conclusion

183 183 188 198 208 215 218

10 Consequences of Generative Lexicon 10.1 Co-composition and Verbal Semantics 10.2 Stage-Level Predication 10.3 Further Applications of Coertion

221 221 225 230

Contents

vii

10.4 Lexical Inference and Rhetoric 10.5 Summary

236 238

Notes

241

Bibliography

269

Author Index

289

Subject Index

293

Preface

This book is basedon a larger unpublished manuscript, Towardsa Generative Lexicon, written between 1988- 1993. Many of the basic ideas for this manuscript are first explored in a general way in an article writ ten in 1987 with Bob Ingria , entitled "Active Objects in Syntax and Semantics." Since the original publication of the article "Generative Lexicon" in the journal Computational Linguistics in 1991, what had started merely as a critique of theories of lexical senseenumeration has developed into a fairly specific proposal for how to perform lexical semantic analysis. Substantial new material has emergedfrom cooperative work with several colleagues. This includes the original work done with Bran Boguraev and discussionwith Ted Briscoe and Ann Copestakein the context of their Acquilex researchfunded by Esprit in Europe. The work on unaccusativity stems from joint research with Federica Busa of Brandeis. Extensions and elaborations of the coercion analysis for aspectual predicates in French has been done in closecollaboration with Pierrette Bouillon of ISSCO and the University of Paris. Several chapters have been omitted for clarity of presentation and in some casesfor clarity of the content. Hence, some topics that were expected to be included have been deleted entirely. For example, two important themes in generative lexical studies (the role played by Lexical Inheritance Theory and the theory of co-specification) are not e~amined in any depth in this monograph; both of these areas have become too large to make only passingreferenceto , and I felt the discussionpossible in this monograph would do no justice to these issues. Regarding lexical inheritance, recent work in computational lexicography and semantics, much of it done in the context of 'the Acquilex project (and reported in Briscoe et al., 1993) has pointed to new and exciting directions for how lexicons should be organized globally. The promise of realizing a projective inheritance model, as suggestedin my 1991 article, awaits further investigation, although it is currently one of the topics of research at Brandeis in conjunction with Bran Boguraev at Apple. The secondmajor omission in this monograph is a comprehensivediscussion of co-specification and processesof selection. This has proved to be a central concern in the applied computational researchat Brandeis as well as the recent work on lexical acquisition and induction from corpora. In the present study, however, I have chosento concentrate on the core mechanismsinvolved in semantic selection as they relate to

x

Preface

syntactic expression . Because co-specification treats the subtle use and variation in selection below the level of conventional semantic and syntactic types , it is impossible to do it justice without extensive discussion . This can be found in Pustejovsky (forthcoming ) and to a certain extent in Boguraev and Pustejovsky ( 1996) . Finally , I should point out that many questions relating to natural language semantics are not investigated in any great detail here . In par ticular , issues surrounding quantification and genericity are only touched on briefly , if at all . Furthermore , details of several of the mechanisms of composition are to be found not here but in other works , including Pustejovsky ( 1995b) and Pustejovsky and Johnston (forthcoming ) . My aim in the current work has been to outline what I feel is the necessary infrastructure for a truly generative , highly distributed , and lexically based semantic theory for language .

James Pustejovsky Brandeis University

Acknowledgments

The ideas contained in this work have been greatly influenced by conversations and interactions with numerous friends and colleagues. Those who have contributed to the specifics of this work include : Noam Chom sky, Bob Ingria , Bran Boguraev , Ted Briscoe , Federica Busa , Patrick Saint -Dizier , Nicholas Asher , Ray JackendofI , Geoffrey Nunberg , Michael Johnston , Pierrette Bouillon , Scott Waterman , Sabine Bergler , Peter Anick , Paul Buitelaar , Evelyne Viegas , and David McDonald . Early encouragement from Dave Waltz , Jane Grimshaw , Jim Higgin botham , Remko Scha, and Edwin Williams helped shape the direction of the work , and their influence is seen throughout the pages of this book . Critical interactions with many colleagues have been useful in sharpening my proposal and the specific analyses. In particular , George Miller , Jerry Hobbs , Manfred Bierwisch , Jiirgen Kunze , Ewald Lang , Beth Levin , Barbara Grosz , Sergei Nirenburg , Aravind Joshi , Chuck Fillmore

, Rick

Alterman

, Yorick

Wilks

, David

Waltz

, and

Peter

Cariani

have provided various degrees of commentary and criticism , making , I believe , the resulting work much clearer than it might have been. Large portions of the material reported on in this book have been presented in front , of audiences from linguistics and computer science conferences and departments , including Brandeis , UMIST , MIT , University of Pennsylvania , Princeton , McGill , the University of Texas , Stuttgart , Humboldt University in Berlin , the CUNY Sentence Processing Confer ence in New York , IBM , SUNY Buffalo , Toronto , Toulouse , Cambridge University , University of Copenhagen , and Charles University in Prague . Many of the comments and questions from these interactions have found their way into the work presented here, and I would like to thank the participants of these talks for their critical and helpful remarks . I would like to also thank the following for carefully reading the final manuscript : Bob Ingria , Federica Busa , Ann Copestake , Henk Verkuyl , Ted Briscoe , Nicholas Asher , Alex Lascarides , Anne -Marie Mineur , Michael Johnston , and Marc Verhagen . I would like to give particular thanks to MIT Press, for their encour agement and editorial support of this project . To Harry Stanton , for his patience and faith , and to Amy Pierce , for her enormous help with guiding the book through the Press on time , and further expediting the publication . And finally , to Deborah Cantor -Adams , for her wonderful editorial help and support with the manuscript .

.. XII

Acknow ledgments

For moral support, there are really too many peopleto thank . I would like to thank my mother, Frances, for her constant energy and encouragement; Charlie and Helga Marqusee, for their wonderful creature comforts and Charlie's personal mail and delivery service; John Broglio, for lending me his apartment in Northampton to rewrite the manuscript; and to Brent Cochran, Peter Cariani, Bob Ingria , and Fede Busa, for their undeservedfaith in me. And finally and most significantly, to my family, Cathie, Zachary, and Sophia for always being there, and enduring the tedious hours and my testy moodiness. This work was supported in part by grants from The Advanced ResearchProject Agency (MDA904-91-C-9328); the National ScienceFoundation (IRI -93-14955); Digital Equipment Corporation ; and the National Institutes for Health. I would like to dedicate this book to the memory of my father, George Harry Pustejovsky.

At this point , as it often happensin philosophy, we suddenly realize that the path of inquiry we hoped to open is already marked by the footprints of Aristotle . Zeno Vendler Linguistics in Philosophy

1 This

Introduction

book

deals

semantics

of

with

words

,

compositionality

.

words

of

a

have

largely

with

features

Under

guage

the

.

for

an

In

have

of

study

I

as

general

goal

of

of

a

the

goals

we

,

then

will

see

the

form

,

and

ative

of

of

I

current

cal

.

associated

of

natural

lan

issues

basis

of

-

:

composition

review

-

on

distinctions

, "

.

distinction

plementary

As

I

accidently

carries

issues

to

represent

models

.

the

recursively

as

If

.

way

in

in

The

traditional

which

,

to

in

ar

-

categories

.

means

language

,

on

passive

changes

a

antonymy

or

itself

be

the

meanings

,

functors

can

view

ultimately

denote

the

affect

we

synonymy

active

change

address

assign

language

either

in

lexical

Hence

,

if

reevaluate

order

to

what

This

satisfy

chapter

several

former

distinct

,

senses

3

,

( 1964

basic

and

,

both

incorporating

on

however

is

and

ambiguity

view

Weinreich

The

representation

lexical

word

in

. !

.

-

adequately

will

depends

semantics

distinguishes

between

ambiguity

directly

ways

such

we

we

composi

by

meaning

several

compositionality

lexical

basic

argue

,

,

.

how

computational

techniques

if

of

and

that

lexical

to

composition

some

view

senses

in

as

that

theory

meaning

argue

compositionality

of

semantic

semantic

word

of

phenomena

then

behave

,

nature

First

the

the

will

word

able

categories

words

I

theory

being

for

lexical

that

But

creative

theory

as

,

.

Our

semantic

comprehensively

very

item

a

metonymy

basic

studied

the

tagged

;

of

concerns

.

accounting

and

denote

the

,

information

important

on

interaction

of

theory

been

.

can

the

linguists

generally

two

of

what

senses

theories

models

these

problem

of

,

guments

no

the

semantic

,

has

the

to

compositionality

expressions

what

problem

and

word

been

contexts

semantic

the

semantic

formal

address

novel

of

and

have

,

to

in

lexical

design

polysemy

view

words

relate

for

the

,

.

how

the

theoretical

set

senses

little

examine

they

accounting

issue

static

i . e

of

particular

.

this

the

of

study

and

a

in

,

the

Nevertheless

done

and

combination

morphological

word

.

,

Computational

as

,

items

use

in

is

.

different

evaluation

tionality

to

,

lexical

ty

and

lexicon

syntactic

view

semantics

semantics

denote

creative

ali

alone

the

semantics

.

both

treated

distinct

language

Lexical

language

this

with

natural

,

)

theoreti

" sense

the

basis

such

an

calls

homonymy

unrelated

present

in

of

finite

,

makes

and

,

meanings

-

feature

approach

contrastive

-

enumer

where

a

,

com

lexical

whereas

-

2

Chapter 1

the latter refers to logically related word senses of the same lexical item . I then turn to some further problems with the enumeration method for lexical description illustrated in chapter 3. It will be shown that the representations assumed by current theories are inadequate to account for the richness of natural language semantics . As I show in chapters 2 and 3, most of the careful representation

work has been done on verb classes(e.g., Levin, 1993). In fact, the semantic weight in both lexical and compositional terms usually falls on the verb . This has obvious consequences for how lexical ambiguity has been treated . In chapter 4, I discuss several devices which simplify our semantic description , but which fall outside the conception of enumera tive lexical semantics . Looking at these devices closely, we notice that they point to a very different view of lexical semantics and how word meanings

are combined .

Given the discussion in these chapters , the following conception of lexical semantic systems emerges. Under such a theory , a core set of word senses, typically with greater internal structure than is assumed in previous theories , is used to generate a larger set of word senses when individual lexical items are combined with others in phrases and clauses. I will

refer to such an organization

as a generative

lexicon , and the op -

erations which generate these "extended senses" as generative devices, including operations such as type coercion and co- composition . I discuss how this view supports an explanatory view of semantic modeling . I then examine the goals of linguistic theory in general and lexical semantics in particular . I argue that our framework of knowledge for lexical items must be guided by a concern for semanticality in addition to grammati cality . The model of semantic interpretation we construct should reflect the particular properties and difficulties of natural language , and not simply be an application of a ready -to -wear logical formalism to a new body of data . I will view natural languages as positioned on a hierarchy of semantic descriptions , characterized in terms of their underlying poly morphic generative power . I argue that natural languages fall within the weakly polymorphic languages, more expressive than monomorphic , but well below the power of unrestricted polymorphic languages . This par ticular characterization is rich enough to capture the behavior of logical polysemy as well as effects of co-compositionality . Next , in chapter 5, I outline the type system for our semantics . A generative theory of the lexicon includes multiple levels of representation for

3

Introduction

the different types of lexical information

needed. Among such levels are

Argument Structure (for the representation of adicity information for functional elements), Event Structure (for the representation of information

related

to Aktionsarten

and event type , in the sense of Vendler ,

1967, and related work), Qualia Structure (for the representation of the defining attributes

of an object , such as its constituent parts , purpose

and function , mode of creation, etc.), and Inheritance Structure (for the representation of the relation between the lexical item and others in the lexicon ) . Chapter 6 presents in more detail the structure of qualia , and the role they play in distributing the functional behavior of words and phrases in composition . Chapter 7 presents the application of the mechanisms outlined in chapters 5 and 6 to the polymorphic behavior of language . A variety of poly morphic types is studied and I consider what operations are needed to adequately account for the syntactic expressiveness of semantic types . In particular , I examine the role of coercion in the grammar as well as the need for other generative devices, such as selective binding and co-composition . There is no single form of polymorphism ; rather , poly semy and type ambiguity are a result of several semantic phenomena in specific interaction . Chapter 8 examines briefly what the consequences of qualia structure are for the semantics of nomina Is. Nouns can be formally characteriza .ble in terms of three dimensions of analysis , involving argument structure , event type , and qualia structure . An analysis of nominal polysemy is presented , making use of the type system outlined in the previous chapters , and explaining in more detail the distinction between unified types and dot objects . In the next two chapters , I outline some areas of grammar that can be greatly simplified if we apply to them principles of generative lexical analysis through the use of the generative devices and the type system presented in chapter 5. In particular , I treat argument selection as driven by semantic types , modulated by constraints on coercion rules , selective binding , and co-composition operations in the grammar . This approach will permit us to explain the polymorphic nature of verbs taking multiple syntactic types . In chapter 9, I discuss the role that qualia and event structure have in describing the way causal relations are lexicalized in

language. Specifically, I look at the semantics of causativejinchoative verbs , aspectual predicates , experiencer predicates , and modal causatives

4

Chapter 1

such as risk. Finally , I discuss how this view of lexical organization relates to current theories of metaphor and pragmatically-induced metonymy. I argue, on methodological grounds, for a strong distinction between commonsenseknowledge and lexical structure, although the issue is clearly an empirical one. The types of creative polysemy examined in this work exhibit a regularity and systematicity across languagesthat is absent from patterns of pragmatic senseextension or modes of metaphor.

2

The Nature of Lexical Knowledge

Onlya fewyearsago,it wasconventional practice in boththeoretical andcomputational linguistics textbooks to coverall that needed to be saidregarding thelexiconin onequickchapter , beforegettingto the moreinteresting andsubstantive topicsof syntactic formandseman tic interpretation . Suchanimpoverished coverage todaywouldscarcely reflectthevibrancyof the fieldof lexicalresearch or the centralrole playedbylexicalknowledge in linguistic theoryandprocessing models . It is nowstandardly assumed bymostlinguistic frameworks (bothcom putational andtheoretical ) thatmuchofthestructural information of a sentence is bestencoded froma lexicalized perspective .! Themostpressing problems for lexicalsemantics , I believe , arethe following : (a) (b) (c) (d)

Explaining thepolymorphic natureof language ; Characterizing thesemanticality of naturallanguage utterances ; Capturing thecreative useof words in novelcontexts ; Developing a richer , co-compositional semantic representation .

I believe wehavereached aninteresting turningpointin research , where linguisticstudiescanbeinformed by computational toolsfor lexicol ogyaswellasanappreciation ofthecomputational complexity of large lexicaldatabases . Likewise , computational research canprofitfroman awareness ofthegrammatical andsyntactic distinctions oflexicalitems ; naturallanguage processing (NLP) systems mustaccount forthesedifferences intheirlexicons andgrammars . Thewedding ofthese disciplines issoimportant , in fact, thatI believe it willsoonbedifficultto carryout serious computational research in thefieldsoflinguistics andNLPwithoutthehelpof electronic dictionaries andcomputational lexicographic resources (cf. Zampolli andAtkins,1994 , Boguraev andBriscoe , 1988 ). Positioned at thecenterofthissynthesis is thestudyofwordmeaning , lexicalsemantics . Before addressing these questions , I wouldliketo discuss twoassump tionsthatwill figureprominently in mysuggestions fora lexicalseman ticsframework . Thefirstis that, withoutanappreciation of thesyntacticstructure of a language , thestudyof lexicalsemantics is bound to fail. Thereis nowayin whichmeaning canbecompletely divorced fromthestructure thatcarries it. Thisis animportant methodological

6

Chapter 2

point , since grammatical distinctions are a useful metric in evaluating competing semantic theories. The second point is that the meanings of words should somehowreflect the deeper conceptual structures in the cognitive system, and the domain it oper"ates in . This is tantamount to stating that the semantics of natural languageshould be the image of nonlinguistic conc--eptual organizi.ng principles, whatever their structure. Computational lexical semantics should be guided by the following principles. First , a clear notion of semantic well-formednesswill be necessary in order to characterize a theory of possible word meaning. This may entail abstracting the notion of lexical meaning away from other semantic influences. For instance, this might suggestthat discourseand pragmatic factors should be handled differently or separately from the semantic contributions of lexical items in composition.2 Although this is not a necessaryassumption and may in fact be wrong, it will help narrow our focus on what is important for lexical semantic descriptions. Secondly, lexical semantics must look for representations that are richer than thematic role descriptions (cf. Gruber, 1965, Fillmore , 1968). As argued in Levin and Rappaport (1986), named roles are useful at best for establishing fairly general mapping strategies to the syntactic structures in language. The distinctions possible with thematic roles are much too coarse-grained to provide a useful semantic interpretation of a sentence. What is needed, I will argue, is a principled method of lexical decomposition. This presupposes, if it is to work at all, (1) a rich, recursive theory of semantic composition, (2) the notion of semantic well-formednessmentioned above, and (3) an appeal to severallevels of interpretation in the semantics (cf. Scha, 1983). Thirdly , and related to the preceding point , lexical semantics must study all syntactic categories in order to characterize the semantics of natural language. That is, contrary to the recent trends in semantic representation, the lexicon must encodeinformation for categoriesother than verbs. Recent work has done much to clarify the nature of verb classesand the syntactic constructions that each allows (cf. Levin 1985, 1993). Yet it is not clear whether we are any closerto understanding the underlying nature of verb meaning, why the classesdevelop as they do, and what consequences these distinctions have for the rest of the lexicon and grammar. The curious thing is that there has been little attention paid to the other lexical categories (but cf. Miller and Johnson-Laird ,

7

The Nature of Lexical Knowledge

1976, Miller and Fellbaum, 1991). That is, we have little insight into the semantic nature of adjectival predication , and even less into the semantics of nominals . Not until all major categories have been studied can we hope to arrive at a balanced understanding of the lexicon and the methods

of composition

.

Stepping back from the lexicon for a moment , let me say briefly what I think the position of lexical research should be within the larger semantic picture . Ever since the earliest attempts at real text understanding , a major problem has been that of controlling the inferences associated with the interpretation process. In other words , how deep or shallow is the understanding of a text ? What is the unit of well -formedness when doing natural language understanding ; the sentence, utterance , paragraph , or discourse ? There the

sentence

researchers

, these would

is no easy answer to this question terms agree

are not

even

formalizable

because , except in a way

that

for

most

Oll .

It is my opinion that the representation of the context of an utter ance should be viewed as involving many different generative factors that account for the way that language users create and manipulate the

context

under

constraints

, in order

to

be understood

. Within

such

a theory, where many separate semantic levels (e.g., lexical semantics, compositional semantics, discoursestructure , temporal structure) have independent interpretations , the global meaning of a "discourse" is a highly flexible and malleable structure that has no single interpretation . The individual sources of semantic knowledge compute local inferences

with a high degree of certainty (cf. Hobbs et al., 1988, and Charniak and Goldman, 1988). When integrated together, these inferencesmust be globally coherent , a state which is accomplished by processes of cooperation among separate semantic modules . The basic result of such a view is that semantic interpretation proceeds in a principled fashion , always aware of what the source of a particular inference is, and what the certainty of its value is. Such an approach allows the reasoning process to be both tractable and computationally efficient . The representation of lexical in a richer

semantics , therefore , should be seen as just one of many levels characterization

of contextual

structure

.3

Given what I have said , let us examine the questions presented above in more detail . First , let us turn to the issue of methodology . In this chapter , I shall review the most common methods used for semantic classification of lexical items ,. and characterize the richness of the problem

8

Chapter2

of

representing

It

lexical

is

the

the

goal

lexical

and

semantic

the

way

it

facilitates

Within

of

a

word

' s

egorial

or

meaning

,

pIe

,

the

individuals

in

all

or

viduals

the

,

of

applying

standard

texts

.

.

1

.

A

recently

of

words

.

cf

.

1972

linguistic

.

For

)

as

Harris

is

of

1951

nouns

generally

)

and

to

Schubert

,

a

1989

.

by

For

and

example

,

due

to

literature

mass

)

of

collocation

classes

book

least

-

procedural

accomplished

basis

,

as

indi

subsets

selectional

at

or

set

of

are

this

For

the

sets

selectional

the

.

such

over

,

,

out

distinctions

on

1955

to

operators

example

in

1993

)

English

(

example

e

in

,

methodology

classes

begun

(

for

type

the

and

semantic

work

in

,

different

due

Pelletier

developed

Levin

1974

-

between

pick

Logical

cat

/

the

con

-

partition

count

distinction

.

Alternations

into

nations

while

classes

,

.

,

behaves

relations

would

Conventionally

,

,

.

such

view

word

refer

as

distinguishes

into

this

a

operations

Because

Chomsky

partition

Verbal

,

.

On

category

woman

analysis

animacy

participate

ample

classes

(

book

,

words

)

the

women

further

selectional

1

1975

basis

in

aspect

how

viewed

theoretic

,

categories

and

Verkuyl

-

distributional

different

cf

are

set

Montague

,

these

involving

into

.

Woods

tests

dog

cf

of

noun

who

as

.

only

be

the

semantics

cooccurrence

in

.

lexical

members

2

cf

(

merely

fundamental

type

would

whereas

world

not

individual

words

most

semantic

elements

hate

viewed

world

(

broad

the

be

the

instructions

nouns

,

the

not

the

and

an

between

,

its

what

world

might

in

perhaps

love

in

and

so

also

individuals

the

(

but

on

Alternation

determines

verbs

syntactic

should

,

Categorial

semantics

information

theory

their

.

formal

is

type

syntactically

exam

of

of

a

relations

and

tradition

such

classify

of

sentence

mapping

Classes

the

per

semantic

this

adequately

predictive

,

items

the

to

classes

Furthermore

lexical

capture

Semantic

theory

into

.

should

.

semantic

language

of

which

1

information

lexical

a

expression

,

.

any

of

meanings

Rather

2

of

items

map

a

semantic

. g

.

is

,

common

the

outlines

in

order

,

the

grouping

the

broad

Project

of

sink

verbs

,

roll

,

into

and

break

meanings

patterns

that

)

and

classification

classify

the

syntactic

alternations

Lexicon

to

verbs

study

grammatical

MIT

a

,

for

to

recently

verb

For

ex

-

codified

argument

alter

semantically

all

.

-

unique

have

both

tran

-

The Nature of Lexical Knowledge

9

sitive and intransitiveforms, wherethe lexicalsensesare relatedby the interpretivefeatureof causation . There are of course, numerousexampIes of intransitive verbs which have no zero-derived causative forms, e.g., arrive, die, fall (cf. Fillmore , 1968, Lakoff, 1970, Hale and Keyser, 1986, 1993, and Kunze, 1991): (1) a. The boat ~!!k in stormy weather. b. The plane ~ k the boat in stormy weather. (2) a. The ball ! Qllg4 down the hill .

b. Bill !:QIl~ the ball downthe hill. (3) a. The bottle .QrQ ~ suddenly .

b. MaryQ! Q~ the bottle suddenly . (4) a. The letter arrived on time . b. *The mailman arrived the letter on time . (5) a. My terminal gi g last night . b. *The storm .cti d my terminal last night . (6) a. The block tower full.. b. * Zachary full. the block tower. c. Zachary .fuil~ the block tower. While the sentencesin (4b)- (6b) are ungrammatical, they are certainly understandable. The lexical semantics should specify what it is that these two classesshare, such that they have grammatical intransitive forms, but equally important is the characterization of how they differ, such that the latter class permits no transitive form. Other useful alternation patterns include the conative, as illustrated below in (7)- (10):4 (7) a. Mary h.9-.t. the target . b. Mary ,!!.Qtat the target . (8) a. Mary scraped the window. b. Mary scraped at the window. (9) a. The cat touched my leg. b. *The cat touched at my leg.

10

Chapter 2

(10) a. Mary hQ .t. the arrow (at the target ). b. * Mary hQi at the arrow. The question is whether it is possible to identify the semantic discriminants leading to the distinct behavior of the transitive verbs above, while still explaining why (9b)- (10b) are ungrammatical.S Perhapseven more interesting is how the polysemy of those verbs taking multiple forms can be representedlexically. What the examples above clearly show is that participation in one grammatical alternation does not sufficiently determine the semantic class of the verb. In fact, even once a complete cataloguing of participa tion in alternation classesis achieved, we must ask ourselvesjust what we have accomplished. Descriptively, we may have achieveda great deal, in terms of how verbs behave according to semantically-labeled classes. But we must realize that explaining the behavior of a verb's semantic class can come only from acknowledgingthat the syntactic patterns in an alternation are not independent of the information carried by the arguments characterized in the very patterns themselves. In other words, the diversity of complement types that a verb or other category may take is in large part also determined by the semantics of the complements themselves. One of the methodological points I will argue is that alternation classifications do not constitute theory. Indeed, as Levin (1993) herself points out , the theoretical mechanismswhich give rise to the descriptive distribution of syntactic behavior are not transparent in the classesby themselves. Still another kind of syntactic diagnostic that seemsto have some theoretical utility is polyadicity more narrowly construed. As Bresnan (1982), Fillmore (1986), and Levin (1993) point out , there are not only argument changing alternations such as those discussedabove, but also argument dropping alternations as well. The rule of "indefinite NP deletion" is the term for the following alternation paradigm: ( 11 )

( 12 )

( 13 )

a . The

woman

~

her

b . The

woman

~

quickly

a . The

dog

devoured

meal

the

b . * The

dog

devoured

a . John

. dm ! ! k his beer

quickly

. cookie

.

. feverishly

.

.

11

The Nature of Lexical Knowledge

b. JohngI~nk feverishly . (14) a. John~~! ~~ his beerfeverishly . b . * John gulped feverishly .

(15) a. Mary hummed a song while she walked. b . Mary hummed while she walked .

(16) a. Mary performed a song while she ate her dinner. b . * Mary performed while she ate her dinner . In the examples here one might attribute the possibility of object -drop to an aspectual difference between the verbs being contrasted . That

is, while eat denotes an activity of unbounded duration (at least lexically) , devour, one might argue, denotes a transition . Although devour is generally considered a manner specification of the verb eat, it carries a completive implicature that is absent from eat. Similar remarks hold for the other two pairs above: while drink is an activity , gulp carries the implicature of completive aspect ; and finally , while hum is an activity , perform has a completive aspect lexically .6 If this were a complete account of the above data , we might expect it to explain the patterns of deletion for the other cases of complement dropping . This would seem difficult for many of the verbs entering into

this alternation . For example, as Fillmore (1986) points out , there are cases where near synonyms seem to behave differently with respect to

licensing of complement-drop (cf. (17) and (18)).

(17) ( 18

As

a. Mary k! ~ to start her car in the morning. b. Mary .tI!~ in the morning.

)

we

all

a

.

Mary

b

.

*

can

see

In

,

no

to

ditransitive

.

start

Pustejovsky

in

one

drop

addition

( cf

to

attempted

complement

mented

below

attempted

Mary

her

the

semantic

cases

car

the

morning

.

.

parameter

will

be

sufficient

to

explain

. 7

transitive

- intransitive

- transitive

,

in

morning

polyadicity

shifts

1992

)

such

:

(19) a. John~~ ~ a book to Mary.

,

as

those

there

shown

are

well

in

( 19

- docu

) -

-

( 24

)

12

Chapter 2

b. *John gave abook . (20) a. John ~~~ a lecture to the academy. b. John gave a lecture. (21) a. John mailed a book to his brother. b. * John mailed a book. (22) a. John mailed a letter to his brother.

b. Johnmaileda letter .

(23) a. Bill showeda book to Mary. b. *Bill showeda book. (24) a. Bill showeda movieto the audience . b. Bill showeda movie. Thus , in certain cases, the otherwise obligatory expression of the goal argument is dropped and the verb becomes a simple transitive . What allows the alternation , I will argue in later chapters , is the interaction of the verbal semantics with semantic information itself .

from the complement

Such grammatical alternations can be used throughout the grammar of a language to make semantic distinctions on the basis of syntactic behavior . Using category and selectional information as well as gram matical alternation data , words can be grouped into semantic classes following more or less predictable syntactic behaviors . Nevertheless , it is still necessary to explain why these and just these grammatical forms are part of a certain alternation class. This is addressed in chapter 9 below . Finally , let us consider briefly one of the oldest semantic classifications for verbs , that of aspectual class or Aktionsarten . The essential idea behind this classification is that verbs and verb phrases differ in the kinds of eventualities in the world they denote . It is normally assumed that there are at least three aspectual types : state, activity , and event, where the last class is itself sometimes broken down into accomplishment , and achievement events .8 For example , the verb walk in sentence (25) denotes an activity of unspecified duration . That is, the sentence itself does not convey information regarding the temporal extent of the activ ity , although deictically it is an event in the past which did terminate .9

13

The Nature of Lexical Knowledge

(25) a. Mary walked yesterday. b. Mary walked to her house yesterday. Such 1963

a ,

Vendler

Dowty and the

sentence ,

,

1979

drink

.

1967

) .

of

a

( 25b

her

logical

home

.

house

.

of

the

type

as

the

are

verbs

which

the

verbs

build

plishment

( 26

)

conveys however

seem

,

verbs

the

to

lexically

their

there

is

is

( 25a

when

over

to

1993

,

work

,

) ,

with

to

the

process

Mary

is

at

event

an

activity .

use

,

activity

accomplishment

transitive

logical

,

the

accomplishments

a

,

run

reference that

default

typical

Kenny

her

assert

an

.

1972 ,

as

explicit

does

denote

denote

, in

, sleep

terminated

activity

to

lexically

because

)

the said

Verkuyl

information

making

( 25b

,

( cf

are

Mary

not

activity

1978

same

that

,

seems

to

activity

,

is

an

of

whereby

destroy

denote

Mourelatos

activity

walk

and

a .

Maryhyilt Mary

) in

direct

the ( 26b

a

Creation mance

existence

of

) , the

,

For

example

, denote

culmination

to

.

there ,

accom

the

-

activity

or

As

permit

best

the

piano

the

sonata

in

in

( 27b

such ( 28

)

)

above

as

that

an in

both ,

a .

Mary

walked

b .

Mary

. Ql ! ilt

an

hour

to a

house

the

for is

i .e . ,

( 29

store in

in a

year

the

do

an .

)

accomplishments not

hour

.

.

testing

modification

so - called

lexical )

,

the

and

Perfor

-

accom

-

: 10

.

diagnostic

and

activities

act

is

) . minutes

classic

,

' s

table

. ( 27a structure

accomplishment

derived

while

usage

hours

a

accomplishments

complement

15

Mary

as

. of

the

in

, one

denotes

act

activity

( for

of

denotable

her

on

played

culmination

example

depending

E ! ~ ~ ~ ~

phrase

the

both

Mary

modification

)

the

play ) ,

is something of

Mary

verb

.

of

a .

adverbials

Notice

house

b .

or

poral

the

only

as ( 27b

illustrated

verb

table

consequence

are

such usage

)

. the

nonexistence

- verbs - verbs

plishment

house

destroyed

culmination

( 28

) ,

sentence

verb

b .

( 26a

such

,

to

.

while

( 27

said

1949

,

of

events

performed

is

Although

culmination

This

Just

,

constraint at

)

examples

duration

has

Ryle

Sentence

walking

In

,

( 25a

Other

additional

temporal

a

as

frame

whether by

tem

adverbials license

.

14

Chapter2

(29) a. * John gr2!!k in 20 minutes. b. *Mary worked in an hour. The

frame

adverbial

reference

to

and

( 29b

)

The

is

instantaneously

.

is

it

not

.

Of

)

( 31

,

John

gi

~

fuylli1

b

.

John

c

.

Mary

arrived

point

adverbial

a

.

She

b

.

The

.

J

the

specific

by

the

an

an

( 33

)

make

in

( 29a

)

~

!

state

thought

that

at

noon

)

,

of

,

( 30b

)

has

and

( 30c

an

)

- like

as

.

as

occurring

point

such

( cf

.

just

as

,

a

adverbials

achievement

at

3

Dowty

,

the

qual

-

pm

1979

is

) .

3

pm

.

.

is

not

restricted

verbs

at

10

3

am

show

at

seminar

pm

achievements

,

:

.

sonata

hour

: 00

to

below

: 00

the

his

4

of

achievement

.

channel

at

of

at

noon

2

.

: 30

pm

.

d

.

He

deijvereg

.

indicates

the

starting

time

of

an

event

of

some

.

lexical

could

in

as

not

( 32

in

)

,

properties

possibly

where

( 32a

)

specified

with

)

~

is

point

performed

apparently

lexically

( 32

phrase

that

change

( 30a

modification

~

a

sentences

accomplishment

~ ~

that

activity

verb

missing

something

an

wallet

at

- adverbial

activity

appear

his

lecture

sentences

the

in

but

pm

the

ames

are

factors

3

pianist

duration

What

or

is

change

by

at

swam

point

,

,

denotes

with

his

Here

verb

precondition

in

the

modification

sentence

.

c

a

results

where

one

a

examples

)

,

classification

example

that

,

the

the

state

that

but

gradual

a

course

as

event

,

For

a

Hence

suggestive

( 30

an

does

to

that

of

aspectual

achievement

ity

require

change

conventional

accomplishment

change

to

explicit

.

last

An

seems

an

we

to

an

of

be

see

a

either

a

"bare

( cf

.

a

.

Mary

~

cookies

b

.

Mary

~

a

a

.

Brown

and

cookie

Root

and

.

(

.

Inc

build

or

( 33b

activity

) )

of

( 32b

)

.

mass

term

,

. il1

thereby

assuming

)

the

runway

in

eat

above

)

QY

consider

Tehran

from

Similarly

mentioned

.

( accomplishment

.

in

affected

,

meaning

as

object

)

be

instance

the

accomplishment

plural

can

For

in

verb

( 33a

verb

.

shift

accomplishment

reading

the

lexical

.

,

can

15

The Nature of Lexical Knowledge

b .

The

Brown

and

presence

ically

of

telic

1986

,

,

subject

patterns

aspectual

)

)

a .

* John

b .

* The

a .

John

b .

The

Finally

, )

cates

individual

the

as of

as

the

( 36

)

of ( 37

( 37

,

sick

)

)

.

,

a .

John

b .

Watching

c .

Bill

these clearly

a .

* Bill * John

,

pluralization

and

finish of

.

Normally

these

,

verbs

subjects

,

as

suggests

himself the

wiped

,

on

typically

TV

permit

tall

can

be

,

Properties - permanent

predicates

. II

himself

overweight intelligent

in

brandy

the with

( 36

) .

hungry us

our

- level

years the

. coffee

predicates

. Great

is

.

John

serving

individual

over

It

construction

sentences

made

, -

with

. non

-

re

identified

predicates

cheap

predi

intelligent

individual

resultative

the

before

stative

an

- level

the

that

~

as

with

stage of in

with

~ ~

and

Carlson

of

that

- level

forms

Following

kinds such

called

k

.

two

.

himself read

!

states

identified

shown

counter

constructions

ate

individual usually

as

.

properties

are

commercial

the

dog

lifetime

in ,

.

Predicates as

been

predicate

his

are

appears

dog

of

of

have

his

.

.

its

and

illustrates

b .

typ

Bach

complementation

plural

distinguish

thought

clean

drank

on ~

- level

These

, and

on

behavior

throughout

typically

culminating

None as

which

a .

.

~ gi

can

stage

be

individuals

class

begin

with

flea

fleas

the ) , we

and

directly ,

from from

complements

arrive

to

( 1989

might

hungry

states this

~ ~ ~ ! ~ ~

examine

less

a

to

began

- level

individual

such

( cf

) .

comes

predicates

.

of

process

resulting

as

Asia

interpretation

details

such

~ ~ ~ ! ~ ~

began

or

for shift

as

began

Kratzer

more

the

Southwest

.

guests

us

1989

aspectual

same

in

unbounded

predicates

the

began

overweight

,

an

an

grammatical

but

guest

let

and

,

Krifka

distinction

( 1977

tains

) ,

runways

to

predicates

not

( 34

~

shifts

event

achievement

are in

( 35

and

aspectual

achievements illustrated

( 34

,

Q , yil

object

)

of

of

with

and

1993

.

plural

indication

the

an

bare

Inc

completive

Verkuyl

Another of

a

( or

Root

Books

.

.

,

16

Chapter2

One final characteristic for distinguishing

activities from accomplish -

ments, known as the "imperfective paradox" (cf. Bach, 1986, Dowty, 1979) , involves the possible entailments from the progressiveaspect. To illustrate the nature of this paradox, consider the sentencesin (38). (38) a. John is running . (Therefore, John has run.) b. John is building a house. (*Therefore, John has built a house.) What

this

difference

in entailment

indicates

is whether

an action

is ho -

mogeneousin nature or has a culmination of some sort. Sentence(38a) is an activity

and entails

the statement

John has run . That

is , John has

already engagedin some running. Sentence(38b), on the other hand, does not allow the entailment John has built a house because building is not a homogeneous process, but rather culminates in a changed state , i .e., it is an accomplishment . Thus , if x is
should

be able to account

for this behavior , and not just

use it

to classify propositions into aspectual types . Summarizing , we have considered the following categorization of aspectual types for verbs , verb phrases, and sentences: ACTIVITIES : walk, run , swim , drink ; ACCOMPLISHMENTS: build , destroy , break; ACHIEVEMENTS: die, find , arrive ; and STATES: sick, know, love, resemble, think , be. Membership in an aspectual class determines much of the semantic behavior of a lexical item , but it should be noted that the aspectual properties of a sentence may change as the result of other factors , such

as adverbial modification (both durative and frame), the structure of the NP in an argument position (e.g., definite vs . bare plural ) , or the presence of a prepositional phrase . Such non-lexical issues are problems in compositional semantics and are discussed in the context of "type -

shifting " phenomena in Bach (1986), Link (1983), Krifka (1989), and Verkuyl (1993). In the discussion that follows, I will restructure the above classification slightly , by making reference to subevents and to an event focusing mechanism called event headedness. 2 .1 .2

Nominal

Alternations

Nouns also have characteristic grammatical behaviors, depending on semantic category. For nouns as well, studying the behavior of grammatical alternations has certainly been the point of departure for the

The Nature of Lexical Knowledge

17

semantic classification of nominal types. Probably the most studied distinction for nominal semantics is that of count versus mass. This is a distinction which dates back to Aristotle , and more recently has played an integral role in the structuring of the semantic model for language (cf. Pelletier and Schubert, 1989, Link , 1983). How "stuff" is individ uated will determine how we talk about it ; hence, sand, although in fact composedof individual grains, is a mass noun and refers to undifferentiated stuff in our daily experienceof it .12 A house, on the other hand, is obviously perceivableas an individuated object and is classified as a count noun. As is well-documented, count nouns and mass nouns select for different quantifier types and allow very different patterns of predication. (39) a. MASSNOUNS : much ~ .lli1., more water; b. COUNTNOUNS : several houses, every ~hild . Not surprisingly, however, there are nouns that have both mass and count interpretations , and these will figure in our later investigations quite prominently. They include nouns such as beer, where we can talk about amounts of (40a) or quantities of (40b) the substance. Similarly, nouns such as e-mail refer either to the massof correspondencesI have or have sent, as in (41), or to the individual transaction or correspondence, seen in (42).

(40) a. Texans drinka lot of beer . b. Patsyrelished everybeershedrank . (41) a. Moree-mailis arrivingeveryday. b. Is thereanye-mailformetoday ? (42 )

a . The last e-mail I sent you was yesterday . b . Every

A semantic vidual

e- mail I send gets bounced .

distinction

and group

related . to count

and mass is that

nouns , and this is also differentiated

For example , group

nouns

satisfy

semantic

plurality

between

indi -

by predicability requirements

selection , as shown in (43 ) below : (43 )

a. The committee ill ~ for lunch . b . The crowd dispersed after the police introduced

tear gas .

. on

18

Chapter 2

For purposes of anaphoric binding , group nouns do not parallel plural NPs completely , however . So far , all the noun classes we have discussed have been predicative in a fairly direct way. That is , both woman and water , when used in full NPs , refer independently to something out in the world . Relational nouns , on the other hand , are dependent on another referent in terms of how they themselves denote . For example , neighbor and brother denote individuals standing in relation to at least one other individual in specific ways . The grammatical consequences of this semantic distinction have been long recognized , and give rise to the following interesting distinction ; (44) a. The men arrived yesterday . b . ?The neighbor arrived yesterday . c. The neighbors arrived yesterday .

(45) a. *The brother came home. b. The brothers came home together. As Bierwisch (1983) and Eschenbach(1993) have pointed out , the two types of relational nouns can be distinguished with respect to pluralization behavior. Nouns such as neighborand sister denote "horizontal relations ," while father and daughter denote hierarchical relations. Within the latter class, the noun daughter is the dependent object in the relation , and behaves differently from father, which is the independent individual . Note that contextual saliencewill improve the acceptability of these NPs. (46) a. *The daughter is in the house. b. ?The daughters are gathering upstairs. c. The fathers are meeting tomorrow. Explicit mention of the independent variable in the relation , of course, results in fully acceptable sentences: (47) a. My daughter phoned me. b. John's brother is in town. c. My neighbor lent me a chainsaw. The distinctions between count/ mass, individual / group, and predicative / relational , are motivated by distinct grammatical behaviors as well

19

TheNatureof LexicalKnowledge

as the underlying semantic distinctions perhaps giving rise to these dif ferences

. A more

traditional

method

of nominal

classification

is based

on taxonomies of the speaker 's intuition or commonsense perspective of what the nouns denote in the world . For example , we might distin guish between "concrete referring " nouns , such as woman , boy, horse

(all count nouns) , as well as grass, water, and gold (mass nouns), and "abstract referring " nouns such as time , place, age, and shape. Such taxonomies of entity types are common in computational treatments of language phenomena , but are largely ignored or seen as irrelevant by the majority of theoretical linguists . The major exception to this is the

semantic taxonomic tradition as carried out by Wierzbicka (1988) and Dixon (1991) and their colleagues. This tradition cannot be so quickly dismissed as is so often the case in theoretical circles . Many of the generalizations they hope to capture are legitimate goals for linguistic theory and cognitive

science . More to the point , however , much of their

work

attempts to achieve these goals without always applying the proper tools of analysis . 13

The structuring of such taxonomic information for nouns (and other categories as well) in computational linguistics and AI is not simply an exercise in domain modeling ; it is necessary for driving the inferences that a language reasoning system must perform in order to understand a sentence. From primitives -based inferencing techniques such as Wilks

(1975,1978) to commonsensemetaphysics reasoning systems applied to language such as Hobbs et ale (1987) , the taxonomic classification of objects in the world through language can be a serious enterprise and not merely metaphysical play . These concerns have received renewed interest in computational approaches to language analysis both in computational mal semantics , and point

linguistics and for -

back to the work done on selection

restriction

from the 1960s and 1970s in the generative tradition . Where selectional features were seen as conditions on lexical insertion in previous theo ries , sortal specification is viewed in terms of type satisfaction within an interpreted model . I mention this trend because I believe there are im portant underlying motivations in both computational and theoretical linguistics communities for modeling the conceptual or epistemological ground assumptions for language research. Yet these motivations can differ dramatically and I hope to identify what the goals are for the diverse communities , and then outline what I think the common goals are

20

Chapter2

for linguistic research in the different approaches . 2 .1 .3

Adjective

Classes

Finally , let us turn briefly to the semantics of adjectives . By their very nature , adjectives are generally taken to denote states . Some of the previous discussion on types of stativity , e.g., the individual -level vs. stagelevel distinction , will be a useful device for distinguishing adjectives as well . This

is related

to but

not

identical

with

the

oldest

distinction

ap -

plied to adjectives, that of accidental vs. necessaryqualities (as used in Aristotle and the scholastics) . This distinction gives rise to the classification of properties such as hungry , dirty , and broken as accidental qual ities as distinct from necessary qualities such as the properties bipedal

(of a species), tall (of an individual ) , and hard (of a substance). We saw in the previous

discussion

that

there

are grammatical

distinctions

to be made on the basis of this typological difference . One diagnostic we did not consider concerns the progressive aspect , and the ability of most stage - level predicates

to enter into predicates

with the progressive ,

while individual -level predicates cannot , as illustrated

below .

(48) a. The horse is being ~~!~!~ with her rider . b . You 're being so angry again ! c. Stop being so impatient .

(49) a. * John is being .t.illl today. b . * Aren 't you being beautiful tonight ! c. * Stop being so intelligent . There are, of course, other ways to classify adjectives by virtue of syntactically distinct behaviors , including the basic distinction between

predicative and attributive position, illustrated in (50)- (51). (50) a. the ~ ~~~~~ criminal b . * This criminal is alleged .

(51) a. the trightened boy b . The boy is frightened . With respect to complementation patterns , there are structurally many similarities between adjectives and verbs . For examples , just as there are

21

The Nature of Lexical Knowledge

intransitive and transitive verbs, there are unary and binary predicative adjectives, which can be seen as intransitive and transitive forms. An adjective such as old, for example, takes no complement, as in (52a), while adjectives such as envious and jealous as in (52b) are inherently relational , and might be analyzed as transitive . (52) a. Sophia is not glQ. b. John is envious of Mary 's position. Another structural distinction differentiates adjectives allowing movement like behavior , such as certain , from non-alternating adjectives .

(53) a. Mary is certainto be the next President. b. It is ~ertain that Mary will be the next President. Perhaps the most celebrated example from this class involves the raising / control distinction seen with adjective pairs such as easy and eager.14 Adjectives such as eager, anxious , and unwilling are subject -control pred icates , and have no alternating construction , while tough- Movement adjectives such as easy, tough, and difficult , enter into the following alter nation .

(54) a. It is ~~ ~ to teach this class. b. This class is easy to teach. (55) a. It is ~angerousto drive on this road in the winter . b. This road is ~~ngerousto drive on in the winter . ( 56 )

a . It

is ! nteresting

b . Bill

These

are

as

(57)

a .

Jim

b .

We

(58) a. b.

interesting

meaning

( 57 ) and

Bill John

many

has

decided going

has had

to

Bill

.

imagine

from

our

.

perspective

adjectives

because

assume

to get

take

a

give a

dangerous

interesting

an

difficult

easy

exam

exam

road suggestion

. for

to .

the

get

of

in constructions

:

to

an

President

to

of these

( 58 ) below

' re

imagine

is interesting

particularly

underspecified such

to

President

final

here

.

.

the

22

Chapter2

Depending on the nominal , however , interpretation

of the "ellipsed " in -

finitival may depend on local context, as the sentencesin (59) illustrates . (59) a. John is teaching an ~~~ class this semester. b . Bill is taking an easy class this semester. What

is curious

about these sentences is that the " understood

predicate "

in the NP an easy class is determined by the governing predicate in the

VP ; that .is, in (59a) the class is easy to teach while in (5gb) it is easy to take. This is not the case, however, with the NP an easy/ difficult exam, where the ease or difficulty seems to refer in most cases to the taking of the exam .IS This seems to depend on the semantics of the noun being modified . I will return to these examples in chapter 10, where local semantic context is able to bring out the appropriate reading compositionally , and the appropriate semantic distinctions are presented for differentiating between nouns such as exam and class. Rather than first examining the grammatical behavior of adjectives , it might seem reasonable to look first at what concepts are expressible in a language adjectivally . This is the approach described in Dixon (1982) , where he takes a field -descriptive perspective on the taxonomic classification of adjectives . Dixon distinguishes adjectives according to the general semantic field associated with the term . For example , he arrives at the following classes for adjectives from cross-linguistic study : 1. DIMENSION: big , little , large , small , long , short 2. PHYSICAL PROPERTY: hard , soft , heavy, light 3 . COLOR : red , green , blue

4. HUMAN PROPENSITY: jealous , happy , kind , proud , cruel , gay 5 . AGE : new , old , young

6. VALUE: good , bad , excellent , fine , delicious 7. SPEED: fast , quick , slow 8 . DIFFICULTY

: difficult

, easy

9 . SIMILARITY

: alike , similar

10. QUALIFICATION: possible , probable , likely

The Nature of Lexical Knowledge

23

Such classescan be very useful for descriptive purposes, but reveal little about the functional or relational properties of the predicate. For example, the adjectives difficult and easyare tough-movement predicates and behave the same with respect to this construction. In the class QUALIFICATION, however, the adjective likely allows raising while possibleand probably do not. Similarly, .the other semantic classesdo not reflect a uniform syntactic behavior. What is needed, I believe, is a semantic classification that captures the intuitions listed by Dixon but based on the relational and logical behavior of the predicates and not on their folk-epistemology. I return to this discussionbelow in chapter 10.

2.2

lnterlexical

Relations

Besides grouping words into distinct semantic classes, lexical semantics is the study of how words are semantically related to one another . In this section , I will briefly examine five classes of lexical relations :

1. SYNONYMY 2.

ANTONYMY

3.

HYPONYMYAND LEXICAL INHERITANCE

4.

MERONYMY

5.

ENTAILMENTAND PRESUPPOSITION

Synonymy is generally taken to be a relation betweenwords rather than concepts. One fairly standard definition states that two expressionsare synonymous if substituting one for the other in all contexts does not change the truth value of the sentencewhere the substitution is made (cf. Lyons, 1977, Cruse, 1986).16 A somewhat weaker definition makes referenceto the substitution relative to a specific context. For examplein the context of carpentry, plank and boardmight be consideredsynonyms, but not necessarilyin other domains (cf. Miller et at., 1990) . It should be noted that if synonymy is defined by substitutability of expressions, then it is an intra -category relation , e.g., nouns for nouns, verbs for verbs, and so on. Antonymy is a relation characterized in terms of semantic opposition , and, like synonymy is properly defined over pairs of lexical items rather than concepts. Examples of antonymy are rise/ fall , heavy/ light,

24

Chapter 2

fast/ slow, long/ short (cf. Cruse, 1986, Miller , 1991). It is interesting to observe

that

co - occurrence

data

illustrate

that

synonyms

do not

neces -

sarily share the same antonyms . For example , rise and ascend as well

as fall and descendare similar in meaning, yet neither fall / ascendnor rise / descend are antonym pairs . For further details see Miller

et al.

(1990). By far the lexical relation most studied in the computational commu nity is hyponymy , essentially the taxonomic relation defined in inheri tance networks . For example , specifying car as a hyponym of vehicle is equivalent to saying that vehicle is a superconcept of the concept car, or that vehicle

the set car is a subset of those individuals

denoted

by the set

.

One of the most difficult lexical relations to define and treat formally is that

of meronymy

, the relation

of parts

to the whole . The relation

is familiar from knowledge representation languages with predicates or slot -names such as part - of and made- of (cf . Brachman and Schmolze ,

1985, Hobbs et al., 1987). Similarly in the domain of planning, the issue of meronymy arises when defining the necessary or optional subparts of a plan or event (cf . Kautz , 1987, and Cohen et ai., 1990) . For treatments

of this relation in lexical semantics, see Miller et at. (1990) and Cruse (1986) . Another important respect in which words can be related is through entailment and presupposition . Although there is no complete agreement on how to define these relations , one fairly established distinction is the following . An expression A semantically entails an expression B if and only if every situation that makes A true , makes B true . On the other hand , A semantically presupposes B if and only if both (a) in all

situations where A is true , B is true , and (b) in all situations where A is false, B is true (cf . Strawson , 1952, Keenan , 1972).17 To see how important these concepts are for determining lexical mean-

ings, observe how (60a) entails the proposition denoted by sentence (60b ) . (60) a. John kill g Bill . b . Bill ill ~ . c . Bill

is dead .

That is, if there is a killing event , then there is also a dying event . Capturing such entailment relations was one of the motivations for lexical

25

The Nature of Lexical Knowledge

decomposition search

in

That

kill

dying

entails

, becomes

dying

event

Notice

in

but

also

action

( 62 )

( 63 ) Thus

the

~ to

finish

finished

the

exam

a . Mary

didn

' t managed

b . Mary

didn

' t finish

lexical of the and

of

managing

Fodor

trade of

( 65 )

the

is selling owns

a . Mary

is trading owns

( 66 )

items

~ ~! ~ ~ ~ that didn

is , regardless

associated

the

a . John

locked

the with

interpretation

the

the

( 60a ) , where

complement

not

no

, however

attempts

or

with

.

event to

do

,

the

.

.

exam

.

.

like

manage

attempts , 1971 the or

must

to

presuppose

bring

, 1974

lexical

this

, Beuren

event , 1985

semantics

ownership

that

of

about ).

verbs

is presupposed

such by

the

piano

for

a computer

.

. what

example truth

he

, the of

and

John . It

verb

.

.

' t forget

of door

of

person

the

exam

, determining

. For

b . John

John

associated

.

with

her

, cf . Kiparsky

That

finish

also

piano

a piano

presuppose

verb

exam

, Karttunen

his

b . Mary

to

a factive

the

a verb

a piano

so straightforward

appears

re -

I

a . John

lexical

succeeds

possession

relation

b . John

some

it

the

arise

, where

entails the

exam

event

, 1963

event

negation

that

to

finish

semantics

much

.

the

to

motivates

of presupposition

manage

, whether

presuppositions

( 64 )

verb

an

behavior

a presupposition

attempted

still

).

the

the

manage

and

assertion

factive

the

, and

, 1979

examining is not

a . Mary

Similar

1960s

presupposes

b . Mary

agent

For

when

. This

complement

Mary

sell

the

, Dowty

than

( 61 ) - ( 63 ) that

, the

not

clear

occurs

( cf . Katz

as

in

, 1983 rather

carries

in

( 61 )

the

linguistics

( e .g . , Jackendoff

the

locked he

' s memory

would forget

associated

, 1971

the

with

door

locked

in

( hence

is

a fact

appear

the

, it

is

( 66b )

is

called

the

world

that

this

factivity

. door

however

are

( 66a ) and

).

the

, there

furthermore . Notice

presuppositions

forget

complement

Kiparsky

that

the

verb

, that

complement

.

in

that in .

( 67 ) there

is is no

26

Chapter 2

(67) John ~~!~~! to lock the door. In fact, in some ways it appear to be counterfactive, in that the process of forgetting prevents the event from even occurring. We will return to issuesof factivity in chapter 8.

3

The Logical Problem of Polysemy

In chapter2, I reviewed the basiccomponents of knowledge necessary for lexicaldescription . In this chapter , I turn to the problemof lexical ambiguity . A propertreatmentof the description of the semantics of lexicalitemsin the language shouldpermitus to adequately describe the behaviorof ambiguityaswellasthe lexicalselection process in the grammar . If d9necorrectly , this shouldsimplifyboththe grammatical andsemantic description of the language . I exploretwo dimensions of the problemof lexicalambiguityandthendiscuss the simplestlexical modelthat is ableto accountfor thesephenomena . 3.1

Varieties of Sense Extension

It is certainly true that many words in a language have more than one meaning, a property usually called polysemy. But the ways in which words carry multiple meaningscan vary. For example, Weinreich (1964) distinguishes two types of ambiguity, the first of which he calls contrastive ambiguity. This is seen where a lexical item accidently carries two distinct and unrelated meanings (i .e., homonymy) . Examples of this are shown in (1)- (5) below. (1) a. Mary walked along the .Q ~nk of the river . b. HarborBank is the richest .Q~ k in the city. (2) a. Drop me a ~ when you are in Boston . b . We built a fence along the property ~ . (3) a. First we leave the gate , then we ~ b . John saw the taxi down the street .

down the runway .

(4 )

a . The discussion turned on the feasibility b . The bull turned on the matador .

of the scheme .

( 5)

a . The judge asked the defendant to approach b . The defendant was in the pub at the hM .

the hM .

In the examples above , for whatever reason , the underlined items have more that one lexical sense . Whether these senses are historically re lated or accidents

of orthographic

and phonological

blending , is largely

28

Chapter 3

irrelevant for purposesof lexicon construction and the synchronic study of meaning. The other type of ambiguity Weinreich refers to , illustrated in (6)(10) below, invoJveslexical senseswhich are manifestations of the same basic meaning of the word as it occurs in different contexts. (6) a. The b..@ !!k raised its interest rates yesterday. b. The store is next to the newly constructed .b~ k . (7) a. John crawled through the window. b. The window is closed.

(8) a. Mary paintedthe door.. b. Mary walkedthroughthe door..

(9)

a. The furm will fail unless we receive the subsidy promised . b . To farm this land would be both foolish and without reward .

(10) a. If the store is ~E~ , checkthe price of coffee. b. Zac tried to open his mouth for the dentist. Following Weinreich's usage, I will refer to these sensedistinctions as complementary polysemies. Somehow, our model of lexical meaning must be able to account for how the word for bank can refer to both an institution and a building , how the word for window can refer to both an aperture and a physical object, and how stative predicates can also refer to causative acts. In the examples above, there are two types of sense complementarity: (a) category preserving, and (b) category changing. I will define logical polysemyas a complementary ambiguity where there is no change in lexical category, and the multiple sensesof the word have overlapping, dependent, or shared meanings. Hence, complementary polysemy is a slightly broader term than logical polysemy, since the former also describeshow cross-categorial sensesare related, for example with the use of hammer as both a noun and a verb. In the next section I discuss the nature of contrastive ambiguity in more depth, and examine what factors in the grammar and what types of knowledge seemto be at play in the disambiguation processfor this type of polysemy.

29

The Logical Problem of Polysemy

3.2

Contrastive

Ambiguity

Making use of the distinction mentioned above, it quickly becomes clear that most work to date on ambiguity has dealt with contrastive ambiguity , the essentially arbitrary association of multiple senses with a single word ; furthermore , if it has dealt with complementary polysemy at all , it has been cross-categorial ambiguity alone, usually treated as a subspecies of contrastive senses. In some sense this is not surprising , since given the current representational techniques and strategies for dif ferentiating

word

senses , there would

a logical distinction

appear

to be no reason to make

between these types of ambiguity . This strategy ,

which I will call senseenumeration lexicons (SELs) , appears at first to adequately handle the sense differentiation for both ambiguity types . Let us see what factors are at play in the disambiguation process for lexical items that have contrastive

senses . Consider

first , the ambiguities

in sentence(11), presentedin Waltz and Pollack (1987). (11) John .hQta few Qygk.. In this sentence both the verb shoot and the noun buck are contrastively ambiguous , and this sentence asserts either that John was successful on a hunting trip or that he spent some money gambling . This example illustrates what could be called pragmatically constrained disambigua tion , since comprehension of such an utterance is performed in a specific context

of who John is and what activity

he was involved

in . Notice that

lexical disambiguation does not occur independently for one lexical item , but

rather

, once

the

context

or domain

for one item

has been

chosen

or

identified , the ambiguity of the other items is also constrained . We will see that , while this is a property of contrastive ambiguity , it does not characterize sense narrowing in logical polysemy .

Consider next the sentencein (12) below, discussedin Hirst (1987, 1988) . (12) Nadia's !~~~ taxied to the terminal . Both the nouns plane and terminal are ambiguous .1 Here plane has at

least two senses, (1) as an aircraft and (2) as a tool used in carpentry . The noun terminal also has two senses , as (1) a computer terminal and (2) as a building at an airport , train station , or bus station . The

30

Chapter 3

computational the question

concern in the disambiguation of how to arrive

given sentence , given particular .

of such lexical items is

at the appropriate strategies

word

for contextual

sense within

a

and pragmatic

.

prImIng

.

Another example discussedin Hirst (1988) and similar examples discussed in Lascarides and Asher (1993) involve sentencessuch as (13) below, and (14) mentioned in the previous section: (13) Ross was escorted from the ~

to the gQ~ .

(14) a. The judge askedthe defendant to approach the ! :!M. b . The defendant

was in the pub at the ! : ! M.

For a sentencesuch as (14a), although it is possible that a judge could be at a drinking establishment and furthermore could refer to the indi vidual as a defendant at this location , this is unlikely , given the normal use of these terms . Hence, what is at play in these cases is an intuitive notion of priming and context setting that is providing for the disam biguation of the lexical items in the sentence by virtue of the discourse within

which

the sentence appears . From a theoretical

perspective

, the

major problems posed by contrastive ambiguity involve issues of discourse inferencing and the correct integration of contextual information into processing . Finally , there are some cases of contrastive ambiguity that do not require context and pragmatic information for disambiguation , so much as the disambiguation that comes by virtue of the predication relation in

the sentence. For example, in (15) below, the appropriate sensefor the noun club is arrived at by virtue of sortal knowledge of the NP appearing in the inverted subject position (cf . Hirst , 1988) .

(15) a. Nadia's favorite lYh is the five-iron . b . Nadia

' 8 favorite

club

is The

Carlton

.

Because of the way the appropriate sense is identified in this example , I will refer to this as a case of sortally constrained disambiguation . There

are , of course , many

finer

distinctions

to make

in the

nature

of

contrastive ambiguity , as well as the in the strategies and information sources that help disambiguate senses. My concern here, however , is to compare this type of ambiguity with complementary polysemy , and to explore what lexical representation is adequate for expressing such sense distinctions

.

31

The Logical Problemof Polysemy

3.3

Complementary

Polysemy

Unlike the cases of ambiguity discussed in the previous section , complementary polysemy seems to entail a very different type of relation between

senses . The sentences given in the first

section

above involv -

ing the nouns door and window , for example , are part of a larger set of alternations called Figure - Ground Reversals, which include a large class of nouns in the language , such as fireplace , pipe, room , gate, etc . The ambiguity in such nouns involves the two senses of 'aperture ' and 'physical object ' used to frame this aperture . This sense alternation is just one of many nominal alternations that can be described as logical polysemies , where the noun seems to have systematically related senses. These

include

:

(16) Count / Mass alternations ; lamb . a . The l ~mQ is running

in the field .

b . John ate l ~mh for breakfast .

(17) ContainerjContainee alternations; bottle. a. Mary broke the QQt.tk . b . The baby finished the QQ1tk.

(18) Figure/ Ground Reversals; door, window. a . The window

is rotting

.

b . Mary crawled through the window .

(19) Product / Producer alternation ; newspaper, Honda. a . The

newspaper

fired

its editor

.

b . John spilled coffee on the newspaper .

(20) Plant / Food alternations; fig , apple. a. Mary ate a fig for lunch . b . Mary watered the figs in the garden .

(21) Process/ Result alternation ; examination, merger. a. The company 's merger with Honda will begin next fall . b . The merger will prod lice cars . (22) Place / People alternation ; city , New York . a . John

traveled

b . New

York

to New

kicked

the

York mayor

. out

of office .

32

Chapter 3

Like

the

the

contextually

case

a

of

logical

noun is

cases

related

.

the

of

sense

is

available

of

a

logically

the

in

particular

they

are

modifying

a

~

car

a

good

meal

c .

a

good

knife

, the

nominal

head

it

however

, there

rather

a

jective

senses

is

Logical types

Verbs for

a

senses

chapters

verbs

7

be

select

the

to

in

other

.

-

noun

.

is , one

Both

-

senses

interpretation

of

'

of

for

purposes

exhibit

this

categories

as

meanings

10

below

also

be

for

, as

read

Mary

~ ~~ ~~

to

Mary

began

reading

c .

Mary

began

the

such

as

are

seen in

novel

polysemous of

syntactic

the

the

it

the

a

logical

well

.

For

, depending

positive

on

evaluation

nominal or

head

being

is

not

clear

effect

modified

.

what

the .

exact I

,

, but

Such

polysemies

judgment

the

above

focusing

complementary

positive

of

polysemies

alternation

as

a

a .

begin

irrele

) , complemen

effect

seem

merely

an

on

beyond and

can

number

( that

available

for

multiple

Unlike

to

, although

b .

multiple

is

classified

senses

polysemy that

( 24 )

better

these in

.

seem

dependency

contrastive between

question

not

seem

and

seems

nature

-

are

.

seen

have

good

modifying

does

functional

also

senses

priming , it

' focused

above

noun

good

adjective is

pairs the

the

con

polysemous in

seems

it .

the

.2

a .

sense

than

the

is as

which

a logically

not

within

around from

senses

relevant

sense

sense

contextual

is

with

context

shadowing

seem

of

such

b .

some

of

sense weaker

one

polysemy

, adjectives

( 23 )

All senses

in

while

of

correct

the

contradictory

other

, but

.

,

encountered

polysemy

manner

sense

much noun

the

Complementary

what

a

logical

that

are

every

context

between

example

In

if have

context

relation

a

of

contrastive

senses

to

the

is

, the

virtue

the

the

polysemous

noun

is

determining

only

seem

in

disambiguate

contrastive

senses

by

senses

we

section

only

difference

helps

issue

previous

discussed

biggest

is , while

tary

have

setting

to

the

the

we

The

discourse

That

in

disambiguation

identified

distinguishes

trastive

vant

club

polysemy

What

a

the

determined

ad rather

relation

return

to

this

. as

relating

the

sentences

the

novel

.

the

novel

.

multiple

below

complement

.

.

in and

that

they

semantic

must contexts

be

able , such

to as

select Verb

33

The Logical Problem of Polysemy

Phrase, Gerundive Phrase, or Noun Phrase. To a large extent, the verb itself retains the same meaning, varying slightly depending on the type of complement it selects. Hence, this would appear to be a legitimate example of logical polysemy. Other related senseswhich could possibly be viewed as polysemiestake us further into the area of verbal alternations more broadly defined, such as the inchoativejcausative alternation , seenbelow in (25) and (26).

(25) a. The bottle QIQ ~. b. JohnQIQ .k the bottle. (26) a. The window~.E~~~~ suddenly . b. Mary openedthe windowsuddenly . These differ from the contrastive ambiguity casespresented in the previous section in several respects. Not only are the sensesrelated in a well-defined way, but it is fairly uncontroversial that one sense(that in (25a) and (26a)) is actually entailed by the other sense. Thus, even such verbal alternations as these can be seenas logical polysemiesas well. These are but a few of the types of complementary polysemy that languages allow. The purpose of this discussion has been merely to introduce the distinctions in ambiguity types, and not to exhaustively study the nature of these polysemiesthemselves, something that is addressedin subsequentchapters. In the next section, I present the most elementary model for lexical semanticsthat would adequately describe the sensedistinctions just discussed. 3 .4

An Elementary

Lexical

Semantic

Theory

Given the preliminary discussionof polysemy from the previous section, I present the simplest model of lexical design possible, and one which is widely assumedin both computational and theoretical linguistics. As I mentioned earlier, the form that a lexicon takes influences the overall design and structure of the grammar. The major part of semantic researchuntil fairly recently has been on logical form and the mapping from a sentence-level syntactic representationto a logical representation language. Hence, it is not surprising that many assumptionsregarding lexical meaning are based on models that are 10- 20 years old.

34

Chapter3

Let

us

more

outline

clearly

semantics

is

string

in

forms

and

the

. one

the

described

in

.

the

This

is

as

a

correct hope

the

, and

is

allow

the a

has

the

grammar

Lexicon

Let

the to

possibly polysemies

have

multiple

meaning

encoding

or

sense

effect

us

on

define

any word

and

lexicon

) , and

to between

for

separate of

language

- defined

smallest .

( BEL

mapping well

model

natural

account

means

furthermore the

the a to

with

simplest

in

interpretation

way to

annotated

elementary

semantic

in

direct

the for

that

proceed

section

each

certainly

lexicon

L

every

Given

how

could

be a a

listed

lexical

variation

the

such

nature

of

a dictionary

characterize

it

the

=

directly

as

of

the

count

genus

fact

that

complicate

the

word

as

in of

term

( GENUS

of

the

a word

the of

structure

structure

if as -

expressing

interpretation of

and

entries

fashion data

taxonomic

a

as

and

.

used

above

( 28 )

category

type

) , which

dictionary

is

compositional sentence

bank ( 27 )

- form

below

,

( CAT

) ,

locates

the

.3

- noun

count _ noun = shore

financialjnstitution ]

]

selectional

features would

features ( cf .

or

the

only

81 , . . . , 8n

} .

, the

senses

lexical

CAT = GENUS [ bank2

of

in

straightforward

=

ambiguation the

a

lexical

' . . . , wsn

compromise

contrastive

specification

that set

two

the

{ WS1

if

senses

organization

combine

GENUS [ bank1

Assuming same

to

Lexicon multiple

, then as

sense

seem

in

CAT

( 28 )

word stored

standard

within

( 27 )

not

Enumeration

L , having

lexical

, the

fairly

basic

concept

in

words

example

using

Sense

that

of

does of

a w

are

view

ambiguous

For

with senses

this

process

is

word

so ~ iated these

entries

most

Enumeration

for

ing

. The

present problem

:

A

and

can

to core

would

forms

operations

Sense

follows

, we

,

form

semantic

the

previous

words

order the

assigning

language

in of

a lexical

the

of

process

listings

in that

semantic

deterministic

sense

problem

Assuming

of Hirst

requirements

( or

types

appear functor , 1987

) to

and ) .

For

as be

for the

genus

merely

the

arguments example

from , a

verbs terms

verb

are

process the such

defined

from

themselves of

correctly

available as

lend

the

, then

set might

dis

match of

-

lexical select

,

The Logical Problem of Polysemy

35

in one of its senses(for it will certainly have many sensesin an BEL), for financial

_institution

as subject , shown below :

(29) The h. @k willl ~oo the money to the customer. lend CAT

(30)

1 = verb

SEM= Ro(B1,B2,()3) ARGSTR=

[ARGl =np ~+finanCial ~nstitutiOn ]] ARG2 = np + money ] ARG3 = np + human ]

From the point of view of linguistic theory , this is a perfectly reasonable model for lexical design , since, as long as the structural and semantic requirements are satisfied , there is no reason to change or enrich the compositional mechanisms making use of this lexical knowledge .4 A similar approach applied to verbs would allow variation in comple ment selection to be represented as distinct senses, related through a sharing of the lexical sign itself . This is the strategy adopted in most current linguistic frameworks , in some fashion or other .5 Informally , such an approach assumes each lexical item to be uniquely selective for a particular syntactic environment , as illustrated below for the verb begin. The semantics of each form , shown below simply as a relation

Ri ((}l , O2) , can be related to each other by a lexical redundancy rule or meaning postulate . beginl CAT

(31)

= verb

SEM= Rl ((}l ,(}2) ARG2 = = np vp [+ lnf ARGSTR =[ARGI . ]] begin2 CAT

= verb ARGl

= np

(32 ) SEM =R2 ,()2) =vp[+prog ARGSTR =(()1 [ARG2 ]] .

begln3 CAT

= verb

(33 ) SEM =R3 (()1 ,()2) =npJ' [ARGl ARGSTR

=

ARG2

= np

Given this preliminary definition of sense enumeration lexicons , let us examine more carefully the way in which gELs are able to account for

36

Chapter3

lexical selection and ambiguity in the two classesof ambiguity discussed in the previous section. We return to the sentencein (12) , repeated below.

(34) Nadia'8El~~~ taxiedto theterminal. Assuming that the contrastive sensesof plane and terminal can be distinguished by appropriate features or sorts (as illustrated in (35) and (36) for plane), then this example is similar to the disambiguation of the noun club given above. plane ! CAT = count _noun GENUS = aircraft

[ [

(35)

plane2 CAT = count _noun GENUS = tool

(36) That

is

,

entiated

the

,

contrastive

propriate

discourse

sense

shown

in

type

senses

hence

( 37

.

)

selection

of

plane

context

Assuming

below

the

,

the

are

is

sortally

not

sortal

really

constrained

needed

restrictions

subject

is

on

therefore

or

to

select

the

differ

-

the

ap

predicate

-

taxi

disambiguated

by

strict

.

taxi

( 37

)

CAT SEM

= =

verb P ( Bl

ARGSTR

As

mentioned

,

rowing

of

Cottrell

,

noun

and

in

as

1965

us

above

turn

,

they

]

sense

has

disambiguation

1988

in

)

.

of

given

that

that

)

,

the

and

basic

two

is

predication

.

senses

a

nar

-

Small

for

( a

sense

in

the

( cf

termina12

appropriate

the

fixed

processing

Assuming

the

been

facilitates

subsequent

( computer

selection

the

representation

variations

lexical

and

]

,

the

building

accomplished

is

fixed

at

this

. 6

to

that

aircraft

contrastive

terminall

then

processing

enumerated

)

,

,

the

[ +

senses

,

are

straightforwardly

np

one

contrastive

)

quite

=

once

Tanenhaus

aircraft

Let

,

1

constrained

other

point

saw

ARG

pragmatically

terminal

an

[

above

sentence

for

)

=

appear

of

in

verb

senses

to

since

adequately

complementary

polysemy

complementation

the

have

Aspects

describe

-

Model

syntactic

been

( cf

.

distribution

.

We

encoded

Chomsky

,

.

I

The Logical Problem of Polysemy

37

will attempt to analyze the casesof nominal polysemy discussedabove in terms of SEL representations. Theseinvolved figure/ ground reversals, container/ containeealternations, and count/ massalternations, repeated below: (38) a. The l~!!!h is running in the field. b. John ate lamb for breakfast. (39) a. Mary broke the QQ~.:t.~ . b. The baby finished the QQt.:tlli. (40) a. The window is rotting . b. Mary crawled through the window. Traditionally these have been treated as simple cases of sense enumer ation , along the lines of contrastive ambiguity . Indeed , the representations below for the complementary senses of the noun lamb seem as well -motivated as the listings for plane given in (35) and (36) .

(41)

CAT = coun ~_noun GENUS = ammal [ lamb ! ]

(42)

CAT = mass _noun GENUS = meat [ lamb2 ]

The fact that these two sensesare logically related is not captured in the two representationsabove, but the sensesare distinguished by type , which is usually the most important consideration for compositionality . One possible modification to the SEL framework we could make, which would differentiate contrastive from complementary sensesfor a lexical item , would be to store complementary sensesin a single entry, distinguished by sense-identification number. lamb

(43)

= meat CAT = mass _noun ] SENSE ! = [ GENUS GENUS = animal = count -lloun ] SENSE2= [ CAT

Thus , we could restate the definition of a sense enumeration lexicon to account for this distinction in how senses are stored :

38

Chapter3

(44) A lexiconL is a SenseEnumeration Lexiconif and only if for everywordw in L, havingmultiplesenses 81, . . . ,Snassociated with that word, then: (i) if 81, . . . , Snarecontrastive senses , the lexicalentriesexpressing thesesenses arestoredasWS1 ,.' . ,Wsn ' (ii) if 81, 00. , 8narecomplementary senses , thelexicalentryexpress ing thesesenses is storedasW{Sl,...,Sn}o Everyambiguityis eitherrepresented by (i) or (ii) above . This is in fact the approach takenby manyresearchers within both theoreticalandcomputational traditions. The advantage of this model of lexicaldescriptionis that the lexiconremainsa separateandindependentcomponent or sourceof data, or a plug-in modulefrom the computational perspective . Hence , onecanstudypropertiesof syntax andsemantic interpretation , knowingat leastthat the lexiconis a fixed point of reference , interactingwith othercomponents of grammarin a predictableandwell-definedway. Nevertheless , in the next chapter ,I showhowthe sense enumerative lexiconmodeloutlinedaboveis inade quateforthepurpose of linguistictheory. I will thenoutlinewhatI think arethe necessary components for an adequate semanticdescription of the language , asviewedfromthe lexicon .

4 In

Limitations

this

of

chapter

method

I

for

be

.

is

view

the

antonymy

ity

will

language

.

ments

.

way

in

will

itself

us

As

reevaluate

that

of

CREATIVE

contexts

(2)

for

of

are

the

semantic

USE

OF

then

of

,

current

the

lexical

form

if

of

argu

change

Words

-

the

compositionality

actually

in

force

language

showing

of

the

theory

we

can

arguments

description

:

of

passive

composition

basic

WORDS

,

semantics

semantic

three

-

semantic

or

however

,

compositmnal

.

the

language

inade

-

.

assume

new

senses

in

novel

.

THE

PERMEABILITY

definitions

OF

but

word

(3)

,

view

nature

there

SELs

THE

very

denote

our

)

If

synonymy

categories

functors

.

assign

as

of

lexical

in

ways

recursively

view

active

( 1991a

can

,

to

-

will

several

such

basic

either

meaning

in

our

will

language

lexical

able

then

It

lexi

natural

phenomena

. ,

.

enumeration

of

being

the

Pustejovsky

Hence

the

show

quacies

(1)

.

theory

assumption

as

categories

change

will

standard

in

etc

what

behave

argued

which

to

I

on

The

as

,

sense

theory

for

metonymy

words

a

enumeration

chapter

of

semantic

accounting

ultimately

that

that

our

the

previous

description

theory

,

,

the

by

the

note

of

semantic

polysemy

denote

requires

a

with

in

allowed

to

of

depend

problems

for

structure

expressions

,

intrinsic

account

important

goal

to

Lexicons

described

representations

general

meanings

some

to

It

the

we

the

inadequate

semantics

affect

to

Enumerative

description

that

are

turn

lexical

shown

cons

Sense

WORD

overlap

SENSES

and

make

Word sensesare not atomic

:

reference

to

other

senses

of

the

.

THE

EXPRESSION

OF

MULTIPLE

SYNTACTIC

FORMS

:

A

single

word

sense canhavemultiplesyntacticrealization . Each

of

these

models

semy

.

gELs

Taken

will

sign

unable

poor

. ,

,

models

that

sense

to

complementary

regarding

express

together

argue

( i . e

points

adequately

are

I

is

considerations

to

of

,

would

address

natural

meaning

real

,

and

is

of

lexical

the

language

must

touch

and

for

to

contrastive

.

every

.

-

.

polysemy

on

poly

incorporating

approach

~ ent

of

enumerative

frameworks

semantics

suffic

compositionality

sense

knowledge

conventional

nature

we

inability

of

that

the

)

the

the

nature

seem

although

enumeration

polysemy

word

it

to

the

How

lexicon

de

ambiguity

To

adequately

treat

assumption

this

,

we

is

accomplished

have

-

it

40

Chapter 4

without proliferating word senses is not a simple task and requires rethinking the role played by typically non-functor elements in the phrase ; that is, in order to maintain compositionality we must enrich the semantics of the expressions in composition . In the last section of this chapter , I outline a model of semantics built on this principle , called a generative lexicon .

4 .1

The Goals of Lexical

Semantic

Theory

As mentioned above, I assume that the primary goal of a theory of lexical semantics , and with

it a compositional

semantics , is to both

describe

adequately the data and to be transparent regarding two points : the system must be learnable is an obvious way and the various phenomena of polymorphisms must be adequately addressed. I do not distinguish between the goals of theoretical linguistics and computational linguistics , but rather consider the use of computational tools and descriptions as an important part of the machinery for the analysis of linguistic theories . Furthermore , I believe the data set has necessarily expanded to include corpora

, but

we must

be cautious

with

what

inferences

we draw

from

corpus data. In this respect, the criticism of Chomsky (1955, 1957) is just as relevant today as it was in the 1950s. I will introduce a notion of semanticality , analogous to the view of

grammaticality (cf. Chomsky, 1964), but ranging over semantic expressions rather than syntactic structures . Semanticality refers to the semantic well - formedness

of expressions

in a grammar . Within

the standard

assumptions of logical languages with truth -functional interpretations ,

an expressionis either well-formed or not, and a sentence(i .e., an expression having a propositional type) is or is not a proposition. Hence, what would the definition of semanticality be if not simply a binary judgment on whether

an expression

is truth - functional

or not ?

Consider the way in which arguments are expressed in the language , and how this affects the acceptability of an utterance . (1) a. ?Mary kicked me with her foot . b . Mary kicked me with her left foot . (2) a. ?John buttered the toast with butter . b . John buttered .

conSlll

.

the toast with

an expensive

butter

from Wis -

41

Limitations of SenseEnumerative Lexicons

Although the (a)-sentencesare not ungrammatical in any strict sense, they are semantically less acceptable than the b-sentences. As argued in chapter 9, the licensed expression of certain types of verbal arguments is predictable from the semantics of the expression and semantic operations in the grammar . The degree to which these conditions are met will intuitively define what is semantically well -formed or not , that is, its degree of semanticality . These include , for example , the conditions on

closure of arguments in various positions (cf. chapter 9) . Other sentences such as (3b) and (3c) are semantically odd because of what we normally associate with the semantic possibilities of a noun such as dictionary

and rock .

(3) a. Mary began the book. b . ?John began the dictionary . c . ??Mary

began the rock .

These differ in semanticality in rather significant ways . The sentence in

(3a) admits of two strong interpretations , that of doing what one normally does to a book as reader , reading, and that of doing what one normally does to a book as a writer , writing . There are, furthermore ,

any number of ways of beginning books. While (3b) has any number of interpretations

regarding activities related to creating or constructing

this object, it fails to readily allow an interpretation available to (3a), that of reading the dictionary . Of course, there are instances of individ uals "beginning a dictionary ," such as Malcolm X in prison , but in this case the dictionary

is a form of narrative

. That

is , it is functioning

in

a different role , dictionary qua novel or narrative . There is no generally available interpretation for (3c) because of what we understand begin to require of its argument and our semantic knowledge or what rocks are and

what

you

can

an interpretation

do to them

, with

them

, etc . This

does

not

is unavailable , however . The interpretive

say that

process in

understanding is both stubborn and persistent, and as speaker/ hearers of a community we accommodate to context and pragmatic effects in order to interpret otherwise semantically ill -formed expressions . Another subtle example illustrating semanticality is the phenomenon

illustrated in the pairs below, discussedin Grimshaw and Vikner (1993). (4) a. ?The housewas built . b . The house was built by accomplished builders .

42

Chapter 4

(5) a. ?The cookies were baked . b . The cookies were baked in the oven.

The point of this discussion is that such distinctions in interpretation are real, systematic, and part of the languageitself. Thesesdistinctions constitute a level of representation in the semantics, which operates according to its own set of constraints. By looking at levels of semantic representation, eachwith its own statement of semantic well-formedness, semanticality can becomea metric reflecting the range of data and how semantic interpretation interacts with syntactic form. It should be pointed out that what is not meant by semanticality is a correlation with the direct statistical frequency of occurrence of an expression in the corpus. Although the corpus can be used for data mining and a general notion of empirical verifiability for patterns and cooccurrences, it cannot be used as the only sourceof semanticality judgments, for example, as on the basis of probabilities . Work in this area that ignores the systematicity accompanyinggrammatical and semantic rule systems, therefore, is throwing away most of the data from which they hope to perform grammar induction .

4.2

The Creative Use of Words

The most convincing argument for the inadequacyof a theoretical model of description is to demonstrate that this model is unable to sufficiently account for the data being investigated. Another argument would be that the model accounts for the data, but in a post hoc fashion, without making any predictions as to whether a particular datum should be possible or not . I will show that SELs fail on both these descriptive criteria . The first argument against the senseenumerative model concernsthe creative use of words; that is, how words can take on an infinite number of meanings in novel contexts. This is not an argument from an "infinite polysemy" position (cf. Section 4.3 below) but from regular sense alternations that are as systematic as transformational regularities in the syntax. Consider first the ambiguity of adjectives such as good, discussedin the previous chapter.

Limitations of SenseEnumerative Lexicons

(6) a. Mary finally boughta ~

43

umbrella.

b . After two weeks on the road , John was looking for a good meal .

c. John is a good teacher . Within an SEL , the only way to represent distinct senses for an adjective such as good would be by an explicit listing of sensesin the received usage of the word : goodl , good2 , . . . goodn . For the sentences in (10) , this would correspond to the three fixed senses listed below . go od ( 1) to function

well ;

good ( 2) to perform some act well ; good ( 3 ) tasty .

The conditions which make an umbrella "good for something ," however , are very different from those which make John a "good teacher " . As

Katz (1964) and Vendler (1967) P9int out , "goodness" is defined relative to a scale, and this scale may vary for each nominal the adjective applies to . One needn 't search far to find natural applications of the adjective good that are not covered by the enumeration above ; for example , good weather , a good movie , and good children . For each novel sense we encounter , the SEL approach must enter a new lexical item into the dic tionary , creating one entry for each new sense. That is, the cardinality of the senses of good will equal (at least ) the number of distinct types in the language to which the adjective applies . As an alternative , one might simply keep the meaning of good vague enough to cover all the cases mentioned above. Then , world knowl edge or pragmatic effects could further specify the manner in which something is good , by commonsense understanding of the phrase and the situation within which it is uttered . The problem with this strat egy, however , is that the particular chunks of commonsense knowledge needed to interpret how good modifies in a specific phrase are actually part of the meaning of the noun being modified , and not simply part of world knowledge . If that is so, then deploying commonsense inference seems unnecessarily and heavy-handed . On the other hand , if such a move is not taken , then it remains for the theory to explain how these chunks of lexical knowledge are put to use in composition .! Another example of the context -dependence of creative word use is illustrated very clearly with adjectives such as fast and slow, where the

44

Chapter 4

meaning of the predicate also varies depending on the head being mod ified . As mentioned

above , an SEL model

requires

an enumeration

of

different senses for such words , to account for the ambiguity illustrated

below (cf. Pustejovsky and Boguraev, 1993):2 (7) The island authorities sent out a ~

little governmentQQ~ , the

Culpeper , to welcome us: a boat driven quickly or a boat that is inherently fast

(8) a fast typist : a person who performs the act of typing quickly

(9) Rackets is a fast game: the motions involved in the game are rapid and swift ( 10) a fast book : one

that

can be read

in a short

time

(11) My friend is a fast drive~ and a constant worry to her cautious husband: one who drives quickly (12) You may decidethat a man will be able to make the ~

, difficult

decisions , a process which takes a short amount As with

the adjective

three distinct

good , the examples

of time in ( 7)- ( 12 ) involve

at least

word senses for the word fast :

fast

( 1) to move quickly ;

fast

( 2 ) to perform

fast

( 3 ) to do something

some act quickly ; that

takes little

time .

As argued in Pustejovsky and Boguraev ( 1993 ) , for an actual lexicon , word senses would be further annotated with selectional restrictions : for instance , fast of movable action

( 1) should be predicated

entities , and fast

"that

takes a little

by the object belonging

to a class

( 3 ) should , ideally , know how to relate the

time " - reading , in the case of ( 10 ) above - to

the object

being modified . Upon closer analysis , each occurrence

predicates

in a slightly

different

way . Again , any finite

of fast

enumeration

of

Limitations of SenseEnumerative Lexicons

45

word senseswill not accountfor creativeapplicationsof this adjective in the language .

Toillustratethis, consider thephrases thefastestmotorway andafast garageasusedin (13) below . (13) a. TheAutobahnis the fastestmotorway in Germany . b. I needa fastgaragefor my car, sinceweleaveon Saturday . The adjective fast in sentence (13a) refers to a new sense, fast4 , i .e., the ability of vehicles on the motorway to sustain high speed. Should this be a separate lexical sense for fast ? If it is not , then how is such an interpretation arrived at in the semantics ? Notice that even this novel

interpretation can be "blended" with other sensesof the adjective in a sentencesuch as (14) below. (14) The fastest road to school this time of day would be Lexington

Street. This incorporates the new sense, fast4 , with fast3 given above; that is, the reference to road in (14) is implicitly a reference to a route , resulting from the goal PP within the NP ; this subsequently allows a d urative interpretation for the adjective , fast , meaning quickly traversed . The other example given in (13b) , a fast garage involves yet another sense, related to both fast2 and fast3 , but the actual interval refers to the length of time needed for a repair by the garage , and not to the garage itself . This is similar to the sense in the phrase fast typist . With these novel uses of fast , we are clearly looking at new senses not covered by the enumeration given above. The inability of an SEL to completely enumerate the senses for a particular lexical item is not limited to the above examples or construc tions , by any means. For example , there are many ways to want , begin , or finish something : (15) a. Mary wants another cigarette . b . Bill wants a beer . c. Mary wants a job . (16) a. Harry ~~~~~ his class. b . John finished his article . c. We had better postpone our coffee until 11:00.

46

If

Chapter 4

the

goal

of

expression

we

and

must

,

then

is

in

drink

, "

(

only

way

:

to

want

to

smoke

to

want

to

drink

.

want3

:

to

want

to

have

here

item

SEL

is

words

,

to

fact

,

senses

necessary

then

there

for

how

would

given

to

be

an

adjective

that

an

as

open

to

is

Enumeration

have

,

by

is

assume

in

"

but

explicit

also

be

new

senses

sense

in

a

of

usage

grammar

senses

that

there

is

;

of

from

such

the

the

able

without

a

some

system

system

our

generators

this

semantically

single

there

creative

,

are

not

as

noun

.

,

and

Furthermore

generators

to

be

contexts

the

.

number

word

of

.

extensions

hence

.

that

must

different

new

;

to

contexts

meanings

semantics

in

unable

new

is

possible

lexical

sense

infinite

corpus

.

the

words

generate

many

want

,

" want

defeasible

putationallexicons

,

applies

an

)

all

context

must

mean

,

of

particular

16

verbs

com

of

must

enough

a

(

explain

there

want

means

" want

of

.

.

interpretations

generation

sensitive

to

in

these

Somehow

adequately

the

use

as

)

;

unrestricted

to

general

( 15

;

and

use

it

are

each

characterize

creative

such

)

then

in

verb

( I5b

a

relation

verbs

.

a

in

an

;

that

lexicon

the

capture

semantics

completely

rise

this

the

for

appropriate

In

for

in

for

be

the

cannot

account

allowing

for

senses

with

, "

,

sentences

interpretations

wanting

:

the

these

to

the

want

lexical

must

of

SEL

play

smoke

of

expression

the

assumes

any

of

hold

at

to

- formedness

that

interpret

presumably

want2

enumeration

of

it

.

difficulty

of

we

" want

,

an

list

giving

,

well

interpretation

.

sense

use

)

manner

}

the

variability

means

course

remarks

an

15c

determine

how

a

Similar

If

(

Of

the

exhaustively

to

the

b

c

The

it

within

to

)

)

to

for

contextual

in

.

( 17

provide

a

15a

and

interpretation

reference

is

account

there

that

the

theory

somehow

Clearly

such

to

semantic

generator

are

.

derivative

discussion

above

would

,

be

47

Limitations of SenseEnumerative Lexicons

4.3

Permeability of Word Senses

While

the

the

sense

the

problem

zation

first

of

of

word

the

in

assume

anism

fixed

word

guity

of

in

like

bake

with on

For

a .

John Mary

of

alternation

class

( 19 )

( 20 )

as

cook

and

meal

.

cooked

a the

a .

J ohn

fi ~ . d an

omelet

b .

J ohn

. fr ! ~

egg

( 19a

)

of

the

an

and

( 20a

cooking

activity

Even

if

we

descriptive how

were mech

to

select

systematic

at . , 1988 versus

) , which

( b ) sentences

-

the ambi

-

require

creation

- oJ - state

the

readings

,

respectively

) .

)

)

interesting

holds fry

carrots

. a

the

et

( change

, the

same

organi

between

obvious

consider

( a ) versus

.

a different

rigid as

- oj - state

here

cooked

too

Atkins

( creation

concerns

for

fact

the

, shown

other

is

that

verbs

below

the

in

this

one sense

.

.

. .

) , the and

object

frying

comes

into

,

respectively

for

the

edible

there

is

responsible

existence , yet

food

by

this

is

resulting

virtue

no in

of

different ( 19b

)

and

) .

The

problem

mantic

here

is

components

guarantee tions

of

correct alone

appropriate

duces

notion semantics

natural

different selection

problem

with

of an for

lexical ideal the

too

of

on this

; hence the

abstraction

.

These

overlap

basis

approach .

ambiguity meaning

word

much

readings

sense

level

issue of

that the

word

Another

the the

core

.

or

addressing

the

, such

:

illustrated

SELs

boundaries

always

change

.

The

Mary

both

( 20b

.

Mary

process

by

discussion other

not

potatoes cake

a .

For

is

( cf . the

.d a

are

of for

the

adequate

context

to

b .

the than

the

the

above

( discussed

. d the

that

were

given

context

. b~ ~

includes

fast

models failing

argument

a claim

) , it

any

QQ ~

purposes

sense

of

enumerative second

is , the

on

respect

the

b .

That

assumption

discrimination

( 18 )

.

based

enumeration

sense

verbs

depending

sense , the

analysis

sense

( a dubious

correct

is

the

that

against problem

senses

lexicon

senses

to

argument

incompleteness

of for

a

undergo

in , it of

Herskovits

lexical

" core

not

that

is ( 1987

item

se to

restric lacks , 1988

prepositions

semantic

"

possible

selectional

is

spatial

is

the

, which deviations

,

any

) , in intro

provides due

-

48

Chapter 4

to convention or pragmatic factors , supplying additional or overriding information to the existing selectional restrictions of the preposition . Thus , from the core meaning of in , convention will elicit related but distinct senses for the preposition as used in the two expressions the hole in the wall and the crack in the bowl. As these examples clearly demonstrate , partial overlaps of core and peripheral components of different word meanings make the traditional notion of word sense, as implemented in current dictionaries , inadequate ; see Atkins (1991) for a critique of the flat , linear enumeration -based organization of dictionary entries . The only feasible approach would be to employ considerably more refined distinctions in the semantic content of the complement than is conventionally provided by the mechanism of selectional restrictions .3 Another problem for sense enumeration models of lexical knowledge is the inability to adequately express the logical relation between senses in cases of logical polysemy . For example , sense alternations involving nouns such as window and door, discussed in chapters 2 and 3, were analyzed as listings of sense pairs , such 88 that given for window below :

(21)

windowl CAT = count _noun GENUS = aperture

(22)

windoW2 CAT = count - Doun GENUS= physical _obj

'J ]

The problem with this is that the logical relation that exists between the things in the world denoted by these expressions is not expressed, and these' senses are embodied in the use of the word as in (23) below . (23) John crawled through the broken window . In chapters 7 and 8, we return to the question of how a single lexical item is able to denote two senses. Another case of sense permeability involves adjectives which have complementary senses in well -defined contexts . These are cases of certain psychological predicates , discussed in Ostler and Atkins ( 1991, 1992) , which have the ability to apparently change type . For example , adjectives like sad and happy are able to predicate of both individuals (24a) and (24b ) , as well as event denoting nouns (24c) .

Limitations of SenseEnumerative Lexicons

( 24 )

( 25 )

a .

The

b .

a

woman

c .

a ~

a .

The

president

b .

* the

~ ~ ~ !~ l

sad

is

~

49

.

1 woman day

/

event

is

/

occasion

~ ~ ~ !~ l '

president

c. *an~~~!~1 day/ event/ occasion (26) a. Themanis frightened !. b. a frightened ! man c. * a frightened2 day / event / occasion We need to explain two things with such apparently polysemous adjectives . First , assuming these adjectives select for animate objects , what licenses the modification of a nonselected type such as a temporal inter val ?

(27)

[sadl ] [sad2 ] CAT= adjective ARG 1 = animate

(28)

_ind

CAT= adjective ARGl

=

interval

Secondly , what constraints explain the inability

of the adjectives in

(25) and (26) to operate in a similar fashion? Within standard approaches to lexical semantics , these data would suggest two separate senses for each of these adjectives , one typed as predicating of animate objects , and the other predicating of intervals . Yet even with this solution , we must somehow explain that "a sad day " is still interpreted relative to a human judging the events of that interval as sad, and most naturally , in a causative relation . Briefly , as explored in Pustejovsky

(1994) , the distinction here seemsto relate to the fact that the adjectives differ in their relational structure . That is , sad- adjectives do not take a prepositional object , except by adjunction , i .e., sad about that . The frighterlradjectives , being passive participles , are underlyingly relational , i .e., frightened of snakes, as are the afraid - adjectives , i .e., afraid of swimming . Apparently only the non-relational adjectives permit this shift in sense. Assuming that most of the interval or event denoting

50

Chapter 4

nominals

carry

adjectives

-

polysemy

a type

a sad

is similar

Finally

, another

fiers such location .

of causative

occasion to the related

as noisy , which

interpretation

is one

that

causes

inchoativejcausative type

of adjectival

predicates

when one

to

pairs polysemy

of an individual

modified be sad such

by sad then

the

as break .

involves

modi -

or of a particular

(29) a. a ~~~~~l car b . the noisy 1 dog .

c . a nOISY 2 room

d . a noisY2 cafeteria

Typical dictionary definitions will require at least two sensesfor the adjective noisy: (30) an object making noise; and (31) a location accompanied by noise.

(30) (31)

nolSYl CAT phys _obj [ARGI .==adjective ] nolSY2 CAT location [ARG .1==adjective ]

This representation, however, does not do justice to the meaning of this adjective. For these are not unrelated senses, since even with the location reading there is obviously a "noise-maker" present. Thus, we need some mechanismto strongly type an adjective like noisy, such that the first reading is somehowmade available through a type of indirect modification . We return to this issuebriefly below within the framework of a generative lexicon.4

4.4

Difference in Syntactic Forms

It is equally arbitrary

to create separate word senses for a lexical item

just because it can participate in distinct lexical realizations - and yet this has been the only approach open to computational lexicons which assume the ambiguity resolution framework outlined above. A striking example of this is provided by verbs such as believe and forget . The sentences in (32)- (36) show that the syntactic realization of the verb 's complement determines how the proposition is interpreted semantically .

51

.

I

Limitations of SenseEnumerative Lexicons

The tensed-S complement, for example, in (32) exhibits a property called "factivity " (cf. Kiparsky and Kiparsky, 1971), where the complement proposition is assumed to be a fact regardless of what modality the whole sentencecarries. Sentence(35) contains a "concealedquestion" complement (cf. Baker, 1969, Grimshaw, 1979, Heim, 1979, and Dor, 1992), so called becausethe phrase can be paraphrased as a question. These different interpretation are usually encoded as separate sensesof the verb, with distinct lexical entries. .I

(32 )of Madison Avenue is toforget most folks aren 'tmembers the leisure class .apt (factive ) that (

33

)

But

like

factor

(

34

)

As

for

(

35

)

36

)

the

rate

senses

distinguish

(37)

(38)

(39)

' s

not

a

forget

,

state

being

the

same

Caliphobia

choice

.

run

by

Ronald

(

,

non

liberal

Reagan

he

forgot

- factive

to

)

environmental

came

from

.

who

forget

the

password

or

never

got

it

?

)

forgets

non

:

where

friends

question

leaves

made

plans

)

about

ellipsed

have

his

being

let

concealed

He

who

into

question

What

(

Under

,

embedded

(

(

others

thing

California

loonies

(

many

one

his

- factive

umbrella

,

and

comes

back

to

get

it

.

)

forget 1 CAT = verb SEM R2 ( } l , ( } 2 [ FACTIVE ] ) ARG2 vp [ + mf ] ARGSTR = [ A = np RGl . forget2 CAT = verb SEM R2 ( } 1 , ( } 2 [ + FACTIVE ] ) ARG2 s [ + tns ] ARGSTR = [ A = np RGl forgetg CAT =verb SEM Rg ({}[lARG2 ,A ()2)=np RGI ] standard

SEL

below

the

,

readings

ARGSTR=

model

where

appropriate

.

,

these

would

features

correspond

have

to

been

introduced

the

sepa

-

to

52

Chapter 4

As discussedin Pustejovsky and Boguraev (1993) , sensitivity to factivity would affect , for . instance , the interpretation by a question -answering system : when asked Did Mary lock the door ?, depending on whether

the input was Mary forgot that she locked the door (factive), or Mary forgot to lock the door (non-factive), the answersshould be Yes and No respectively . Such a distinction could be easily accounted for by simply positing separate word senses for each syntactic type , but this misses the 0 bvious relatedness between the two instances of forget . It also misses

not only the parallel betweenthe question-like readings in (34) and (35), but also the similarity between the non-factive in (33) and the ellipsed non -factive in (36) . Moreover , the general "core" sense of the verb forget , which deontically relates a mental attitude with a proposition or event , is lost between the separate senses of the verb . That is, the complement

of forget is always factive in somerespect, either directly, as in (32) with the tensed S complement , or with an infinitival The proper

Vp .5

approach , in our view , is to have one definition

for for -

get which could , by suitable composition with the different complement

types, generate all the allowable readings shown above (cf. chapter 7 below). Another example illustrating the property of multiple subcategorizations being associated with a common under lying meaning is the verb

remember

.

(40) John probably won 't

remember , that he alreadyfed the dog.

(factive) (41) The neighbor remembered to feed the dog. ( Jactive)

(42) Mary can never remember where he leaveshis car keys. (embeddedquestion) (43) Edith finally rememberedher husband's name. (concealedquestion) (44) John couldn 't remember his lines for the play .

(concealedquestion)

(45) Mary rememberedthe keys before she left . (ellipsedfactive)

Limitations

of SenseEnumerative Lexicons

53

Other examplesof multiple subcategorizationinvolvingconcealedquestions and concealedexclamationsare givenbelow(cf. Grimshaw, 1979 and Elliott , 1974).6 (46) a. John knowsthe plane's arrival time. (= what time the planewill arrive) b. Bill figuredout the answer. (= what the answeris) (47) a. John shockedme with his bad behavior. (= how bad his behavioris) b. You'd be surprisedat the big carshe buys. (= how big the carshe buysare) Although the underlinedphrasessyntacticallyappearas NPs, their semantics is the sameas if the verbshad selectedan overt questionor exclamation. Similarly, the predicateregrettakes Sand NP complements, whereboth are interpretedfactively, asin (48) below. (48) a. Mary regrettedthat shehadpublishedthe articlein Illustrated Semantics . b. Mary regrettedthe article in IllustratedSemantics . c. John regrettedpublishingthe photosin the magazine . d. John regrettedthe photosin the magazine . For all of thesecases , gELswould simply list the alternativestructures along with their apparentlydistinct but related meanings . The fact that thesesensesareso related, however , suggests that this approachis missingthe underlyinggeneralizationbehindthesesyntacticforms. Another interestingset of examplesinvolvesthe range of subjects possiblewith causativeand experiencerverbs.7 Considerthe sentences in (49) and (50). (49) a. Driving a car in Boston frightens me. b . Driving frightens me. c . John 's driving frightens me. d . Cars frighten me. e . Listening to this music upsets me . f . This

music upsets me .

54

(

Chapter4

50

)

a

b

c

d

As

.

John

.

The

gun

k

.

The

war

.

.

John

these

examples

kli

' s

~

.

d

!

Mary

ll

kil

pulling

.

~

~

g

Mary

.

Mary

.

the

illustrate

trigger

,

klll

the

syntactic

~

Mary

.

argument

to

a

verb

is

not

always the same logical argument in the semantic relation . Although superficially similar to cases of general metonymy (cf . Lakoff and Johnson , 1982, Nunberg , 1978) , there is an interesting systematicity to such shifts in meaning that we will try to characterize below as logical metonymy . Finally , other verbal alternations involving reciprocal relations such as meet and debate are polysemous and allow reciprocal subject inter pretation (cf . (51c) and (52c)) ; as the sentences in (54) show , however , the constraint on reciprocal interpretation for motion verbs must require that both subject and object be moving (cf . Levin , 1993, and Dowty , 1991) .

(51) a. Johnm~ Mary. b. Johnm~ with Mary. c. John and Mary mtl . (52) a. John debatedMary. b.

c.

(53)

John debatedwith Mary. John and Mary debated.

a. A car ran into a truck . b. A car and a truck ffi !! into each other.

(54) a. A car ran into a tree . b.

* A car and a tree ran into each other .

These examples illustrate the inherent inability of SELs to capture the relatednessbetweensensesin the aboveexampleswithout the addition of more powerful mechanisms, such as meaning postulates. More generally, what emergesfrom these three independent arguments is the view that the senseenumerative model of lexical description is simply inadequate for describing the semanticsof natural languageutterances.

55

Limitations of SenseEnumerative Lexicons

4.5

Semantic Expressiveness

In order to help characterize the expressive power of natural languages in terms

of semantic

expressiveness , it is natural

to think

in terms

of

semantic systems with increasing functional power . Furthermore , a nat ural way of capturing this might be in terms of the type system which the grammar refers to for its interpretation . Let me begin the discussion on expressiveness by reviewing how this same issue was played out in the realm of syntactic frameworks in the 1950s. When principles of context -free and transformational grammars were

first

introduced

, one

of the

motivations

for

the

richness

of these

new systems was the apparent inadequacies of finite -state descriptions for the different natural language constructions that were being discov -

ered (Chomsky, 1955, 1957) . That is, whereverthe structuralists did not appreciate

that

a particular

construction

was not sufficiently

characteri -

zable by finite -state description , the power of a more expressive grammar became necessary. When Chomsky and others argued that natural lan guages appeared to be not even context -free, but in fact appeared to require mechanisms moving out of the family of context -free languages , the motivation was clearly data -directed and seemed warranted by the most prevalent interpretations of what CFGs were . An analogous situation holds in the semantics of natural language today . We have reached a point where we are discovering phenomena in language that are beyond the range of explanation , given the current makeup of our lexical semantic systems . If we think

of the view within

SEts in terms of its generative

capacity ,

we have what might be characterized as a monomorphic language (cf. Strachey , 1967) . In a monomorphic language , lexical items are given a single meaning , that is, one type and denotation . Lexical ambiguity is handled by having a multiple listing of the word . The polysemy discussed in previous chapters with words such as door has forced monomorphic approaches to represent these senses as independent and separate lexi cal items . As we discovered , within monomorphic approaches to lexical meaning , the connection is captured at best by meaning postulates or lexical redundancy rules of some sort . Similar examples exist with an entire range of nominal types in natural languages , all of which seem to be logically polysemous ; e.g., fireplace , pipe, room , etc . All of these have the meaning of both physical object and spatial enclosure (cf . Pustejovsky

56

Chapter 4

and

Anick

,

1988

Something

is

relationship

clearly

that

particular for

) .

way

this

.

class

of

views

by

,

every

) , a

but

the

a

of

words

Treating

Levin

meaning

literal

) t

.

as

Levin

and

of

postulates

are

lexical

more

that

a

monomor

.

defined

logi

of

-

word

Montague

) .

Dowty

still

-

treated

listing

( 1988

clearly

com these

and

include

is

-

all is

ambiguity

which

the

and In

ambiguity

adherents

senses

clear

items

denotation

Rappaport

word

.

:

enumerative

and

a

typing

individuals

as

Lexical

an

,

of seems

contrastive

view

and

, it

in

complex

properties

and

for

lexicon

view

far

where

type

both

a

set

following

meaning

,

to

logical

object

characterized

language

single

the

( 1985

so

, the

physical

us

simple

be

the

predominate

complex

a

can

A

has

the

as

polysemy

with

a

( 1979

)

monomorphic

than

in

,

Montague

's

.

Although such

we

as

that

meaning

will

the

be

properties

a

the

view

an

is

of

based form

.

the

and

a

a

restrictive the

Searle

( 1979 in

.

the

Briefly

of

for

literal

example

paper

,

than

the

.

such

language

context ,

theory role

) ,

this

polymorphic by

Languages

of

Searle

brief

lexical

.

No

No

a

( UPL

any

properties

)

inherent of

of

an

the

is

is

such

nothing

discussion

different

inherent in

in

context above

theory

Although

a

a

"

acts

type sub

explicitly

in

-

lexically

fairly

unrestricted as

the

pragmatically the

the

between

not

knowledge in

on

distinction

polymorphic

" background

expression

words

semantic

)

restriction

operational

.

( 1979 of

there of

our

is

by

more

assume

contribution meaning

meaning From

of denying

adopted

transformations

theory

,

held

lexicon

may

polymorphic

The

contexts

inverse

theory

unrestricted

language

item

, the

shift

a

have

determined

Polymorphic lexical

classes

, the

be

:

Unrestricted a

here

we

termed is

of are

it would

perspective

meaning

system

that

Such

the

might

since

discuss above

.

from

theory

not

described entirely

Viewed

want

,

a

word

been

more

model

the

types

; namely by

leads

than of

provided

.

has

( 1974

of

approach

defined

ambiguity

listing

polysemy

senses

has

rather

Languages are

multiple

cal

)

this

definition

,

lexical

( ML

phrases

in

aperture

discussion

of

Monomorphic

an

particular

our

language

missed

is

nominals

theory

plex

door

This

Summarizing standard phic

being

a

trigger

to

determined

language

that

constrains

. ,

I

which

believe

it

accounts

is

clear

for

that

the

what

polysemy

we

in

Limitations of SenseEnumerative Lexicons

57

natural language while not overgenerating to produce semantically ill formed expressions . furthermore , if our observations above are correct , then such a theory must be in part lexically determined and not entirely pragmatically defined . Such a theory would generate what I will term weakly-polymorphic languages ( WPLs ) . Several lines of research have pointed to capturing the flexibility of word meaning and semantic inter pretation , from early observations made by Katz (1964) , Wilks (1975) , and Nunberg (1979) to th ~ work reported in Klein and van Benthem (1987) . The properties of such a system would be at least the following : " Weakly Polymorphic Languages All lexical items are semantically active , and have a richer typed semantic representation than convention ally assumed; semantic operations of lexically -determined type changing (e.g., type coercions ) operate under well -defined constraints . Different subclasses of polymorphic operations are defined , each with independent properties and conditions on their application . As with the increasing generative capabilities of families of grammars , the range of sense extensions for a lexicon (and with them the subsequent semantic expressions ) increases as restrictions on generation are lifted . What natural language data seem to require is a semantic system falling outside of monomorphic languages (ML ) , but well below the language of unrestricted polymorphic languages (UPL ) , what we have called weakly polymorphic languages (WPL ) :

ML ~ WPL ~ UPL This is the view of semanticsthat I will present in the remainder of this book. I will describethe varieties of coercion types, and then outline an approach within generative lexicon theory, how best to constrain the application of coercion operations. The presentation of each phenomenon will be somewhat brief, in order to give a broader picture of how all categories participate in generative processesgiving rise to logical polysemy. 4 .6

Generative

Lexical

Models

Given what I have said , I will outline what I think are the basic require ments for a theory of computational semantics satisfying the require ments outlined in the previous chapter . I will present a conservative

58

Chapter 4

approach to decomposition , where lexical items are minimally

decom-

posed into structured forms (or templates) rather than sets of features. This will provide us with a generative framework for the composition of lexical meanings , thereby defining the well -formedness conditions for semantic expressions in a language . We can distinguish between two distinct approaches to the study of word meaning : primitive - based theories and relation -based theories . Those advocating primitives assume that word meaning can be exhaus-

tively defined in terms of a fixed set of primitive elements (e.g., Wilks , 1975a, Katz , 1972, Lakoff, 1971, Schank, 1975). Inferences are made through the primitives into which a word is decomposed . In contrast to this view , a relation -based theory of word meaning claims that there

is no need for decomposition into primitives if words (and their concepts) are associatedthrough a network of explicitly defined links (e.g., Quillian , 1968, Collins and Quillian , 1969, Fodor , 1975, Carnap , 1956,

Brachman, 1979). This view relies on logical rules of inference to establish the connectedness

between

lexical

meanings

and propositions

. In a

sense, linguistic data are just another application of a general , more pow erful

set of reasoning

devices needed for commonsense

inference , naive

physics , and micro -world modeling . What I would like to do is to propose a new way of viewing decompo sition , looking more at the generative or compositional aspects of lexical semantics , rather than decomposition into a specified number of prim itives . Briefly , a generative lexicon can be characterized as a system involving at least four levels of semantic representations . These include the notion of argument structure , which specifies the number and type of arguments

that

a lexical

item carries ; an event structure

of sufficient

richness to characterize not only the basic event type of a lexical item , but also internal , subeventual structure ; a qualia structure , representing the different modes of predication possible with a lexical item ; and , a lexical

inheritance

st ~ cture , which

identifies

how

a lexical

structure

is

related to other structures in the dictionary , however it is constructed . A set of generative devices connects these four levels , providing for the compositional

interpretation

of words in context . The most important

these devices for our discussion

is a semantic

transformation

called

of type

coercion which captures the semantic relatedness between syntactically distinct

expressions . As an operation

coercion can be seen as transforming

on types within

a A- calculus , type

a monomorphic language into one

Limitations

of Sense Enumerative

59

Lexicons

with polymorphic types (cf. Cardelli and Wegner, 1985, Klein and van Benthem, 1987). Argument , event, and qualia types must conform to the well -formedness conditions defined by the type system and the lexical inheritance structure when undergoing operations of semantic composi tion . Lexical items are strongly typed yet are provided with mechanisms for fitting to novel typed environments by means of type coercion over a richer

notion

of types .

If lexical items are to be thought of as carrying several parameters '

(or dimensions) of interpretation , then the question immediately arises as to how a particular interpretation is arrived at in a given context . This question is answered in part by the semantic operation of type coercion . Intuitively , the process works as follows . In the construction of a semantic interpretation for a phrase or sentence, a lexical item is able to coerce an argument to the appropriate type only if that word or phrase has available to it , an interpretation of the expected type . Assuming that objects are represented within a typed semantic formalism , we can define type coercion

as follows :

TYPE COERCION: a semantic operation that converts an argument to the type which is expected by a function , where it would

otherwise

result

in a type error .

As I shall show in chapter 6, each lexical item has available to it a set of type shifting

operators

which

are able to operate

over the expression ,

changing its type and denotation , following specific semantic constraints . In the next chapter , I turn to the general motivations of linguistic theory , and specifically , to the overriding principles motivating develop ments

in lexical

semantic

research

.

4 .7

Strong

vs . Weak Compositionality

Before looking at specific analyses of how ambiguity can be reduced , there are some important issues to address regarding compositionality in general .8 We can view the principle of compositionality as being satisfied

in at least two ways , which

I will

refer to as weak and strong

compositionality . Briefly , the distinction can be articulated as follows . A basic Fregean concern with composition is not sufficient enough for a cognitive or computational model of natural language semantics . One

must also be concerned with space (and possibly time ) considerations

60

Chapter 4

within the system (i.e., for the algorithms therein). This meansthat two parameters are important for characterizing semantic devices: (A) The degreeof composition within an expression(i.e., how much unilateral or b.ilateral function application occurs within a phrase); (B) How many explicitly defined sensesare necessaryto accomplish a

uniqueinterpretationof the phrase. The first point I think is fairly clear . This refers to how functionally the elements in the phrase are treated , relative to the resulting inter pretation . In most conventional approaches , only one element for each phrase is treated functionally . I have argued elsewhere that natural lan guage exhibits a high degree of cocomposition that must be accounted for formally . The second point refers to the linguistic and logical tradition of simply multiplying senses on demand for new contexts as needed to create new word senses. That is, as mentioned above, compositionality is achieved by enumeration of word senses. This gives rise to a system where the number of distinct lexical senses (i .e, lexical listings ) needed in the lexi con rises proportional to the number of interpretations in the language . We will call this "weak compositionality ." It may still be compositional , but it results in a system that captures generative expressiveness only by virtue of listing an infinite number of senses, an unfortunate state of affairs for a computational system . A more adequate theoretical stance , I believe , is that where the num ber of lexical senses remains roughly constant relative to the space of possible interpretations in the language . On top of a simple type system are generative mechanisms , which through composition , produce the actual "senses in context " non-lexically up the tree , as it were . This is what I refer to as "strong compositionality ." I believe this is a much more adequate model for cognitive concerns and computational tractability , while still preserving the principle of compositionality .9 In what follows , I will explore the formal characteristics of some of these generative devices, as seen from the concerns of lexical semantics .

5

The Semantic Type System

5.1

Levels of Representation

In this chapter I detail the organization of lexical information within a generative lexicon . As should be clear from discussion in previous chapters , our goal is to provide a formal statement of language that is both expressive and flexible enough to capture the generative nature of lexical creativity and sense extension phenomena . To this end , I will characterize a generative lexicon as a computational system involving at least the following four levels of representations , mentioned briefly in chapter 4: 1. ARGUMENT STRUCTURE: Specification of number and type of logical arguments 2 . EVENT

, and how they are realized

STRUCTURE : Definition

syntactically

.

of the event type of a lexical

item

and a phrase . Sorts include STATE, PROCESS , and TRANSITION, and events

may

have

subeventual

structure

.

3. QUALIA STRUCTURE: Modes of explanation , composed of FORMAL, CONSTITUTIVE 4 . LEXICAL

structure

, TELlC , and

INHERITANCE

is related

contribution

AGENTlVE

STRUCTURE

to other

roles . : Identification

structures

of how

in the type

a lexical

lattice , and its

to the global organization of a lexicon .

A set of generative devices connects these four levels, providing for the compositional interpretation of words in context . Included in these generative operations are the following semantic transformations , all involv ing well - formedness

conditions

on type combinations

:

. TYPE COERCION: where a lexical item or phrase is coerced to a semantic interpretation by a governing item in the phrase , without change of its syntactic type . . SELECTIVE BINDING : where a lexical item or phrase operates specifically on the substructure of a phrase , without changing the overall type in the composition

.

. CO- COMPOSITION: where multiple elements within a phrase behave as functors , generating

new non - lexicalized

senses for the words

in

62

Chapter 5

composition forms

These

also

expressions

semantic

Argument conditions

itance

structure

when

Lexical

items

are strongly

fitting

to

By of

novel

notion defining

the

representation

con

meanings

, this

into

a single

on context

Semantic

restructuring Anick

, and

theoretical

what

a possible semantic

Argument

Following a lexical four

the initial item

( 1) a = A

is the

type , Q provides ture , and I

structure

.

. for

over

a

7. at different

levels

of sentence

composition

mechanisms conflate

the

plays

operates

here

such

word

meaning

can

lexi -

in com -

word

behavior

senses depen -

size of the

lexicon

an important

. Following

claim

is that

involved

different

of word

lexical

discussion

role

Pustejovsky

conceptual

paradigms

a characterization

be , through

. in

the

con -

mechanism

.

above , I will

Q;' can be defined


assume

as a structure

( cf . Pustejovsky

that

the semantics

, consisting

of the

of

following

, 1995a ) :

, Q ,I > argument the

structure

binding

is an embedding

tice , determining

coercion

items

meta - entries

expressions

inher -

of the

items

composition

lexical

mechanisms

of type

reducing

for lexical

such

with

to the

composition

in chapter

regularities

, greatly

conform and

of

distinct

in the

us to

,

Structure

components

where

, encoding

as a result

discussion

a characterization

expressive

enable

must

system

means below

at

to our

of semantic

of lexical

arrive

more

will

of how

of well - formed

5 .2

of the

( 1988 ) , I call

( lcps ) . The

by

component

underspecification

and

strains

to

types

type

co - composition

syntactically

are provided

behavior

integral

meta - entry

this

yet

environments

approach

dent

by the

, to be pres ~ nted

and

. Because

position

qualia

operations

typed

we hope

as an active

between

semantic

.

are important

defined

functional

underspecified

specification

relatedness

undergoing

typed

of types

of

, such as manner

verb

, event , and

well - formedness

richer

light

transformations

the

.1

cases

enriched

, and

semantic

capture

includes

contextually

transcription

three to

This

becoming

feature

how

.

what

Argument

, e is the

of these

two

transformation

information structure

, placing

is inheritable is by

far

specification

parameters

the

Q within

from best

of the

in the qualia

the

event struc -

a type global

understood

lat -

lexical of these

63

The Semantic Type System

areas, and is the logical starting semantics

of words

point for our investigation

into the

.

What originally began as the simple listing of the parameters or arguments associated with a predicate has developed into a sophisticated view of the way arguments are mapped onto syntactic expressions . For

example, Chomsky's (1981) Theta-Criterion and Bresnan's (1982) functional completeness and coherence conditions require arguments to be expressed as syntactic constituents , and syntactic constituents to be bound

to the argument structure . Chomsky's Projection Principle (Chomsky, 1981) further requiresthat such conditions are satisfied at all levelsof lin guistic representation . One of the most important recent contributions to the theory of grammar has been the view that argument structure it -

self is highly structured independentof the syntax. Williams ' (1981) distinction

between

external

and internal

arguments

and Grimshaw

's pro -

posal for a hierarchically structured representation (cf . Grimshaw , 1990) provide us with the basic syntax for one aspect of a word 's meaning .2 The argument

ification

structure

for a word

can be seen as a minimal

of its lexical semantics . By itself , it is certainly

for capturing the semantic characterization a necessary

component

the assumption

that

spec -

inadequate

of a lexical item , but it is

. Indeed , much research has been conducted

argument

structure

is the strongest

determinant

on or

constraint on the acquisition of verb meaning in child language acqui -

sition (cf. Gleitman, 1990, Fisher, Gleitman, and Gleitman (1991) and Pinker (1989) for discussion). I will lexical

introduce items

, here

a distinction illustrated

between

1. TRUE ARGUMENTS: Syntactically item

four

types

of arguments

for

for verbs ;

realized parameters of the lexical

; e .g . ,

"JQhn arrived

late ."

2. DEFAULT ARGUMENTS: Parameters which participate in the logical expressions in the qualia , but which are not necessarily expressed syntactically ; e.g., " John

built

the

house

out

of bricks

" .

3. SHADOW ARGUMENTS: Parameters which are .semantically incorpo rated into the lexical item . They can be expressed only by operations

64

Chapter 5

of subtyping or discourse specification ; e.g., " Mary

4 . TRUE

buttered

ADJUNCTS

but

are

part

any

particular

adjunct

True

and

conditions

alternations

expression

ternations

alternations the

shown

(2)

(3)

the

sentences

a . The

window

b . John

broke

a . Mary

which

may

the

as the

are

necessary

be

such

left

result

in

from

those

al -

.

The

former

alternation / product

in

(2) ,

alternation

):

~ AQil

material

is optional

created

object

material for

are

.

.

.

/ product the

unexpressed

arguments

.

of . 1YQQQ.

into

wood

the

above

which

phrase

/ causative

, 1993

ex -

{}- criterion

in

logical the

ellipsed

pair well

surface are

, its

Such

status

as

optional I will

call

default

- formedness syntax

discussed

of

. The to

an

ar -

arguments

the

ar sen -

conditions

some

extent

arguments

A

in

9.

Intuitively and

of from

such

. They , but

the

expression different

alternations

chapter

B

wood

the

~ - ~ 2li .

guments

under

dol ! out

carved

the

the

.

the

carved

is

necessarily by

a verb

material

to

include

, as mentioned of

,

tied

.

window

c . Mary

Because

are

covered

optional

as the

not

; e .g . ,

distinguished

inchoative

such

b . Mary

gument

tence

broke

carved

an

( 3 ) ( cf . Levin

the

carved

d . ? Mary

in

as the

in

which

forms be

expression

are

. These

."

structure

of

logical

Tuesday

generally

argument

."

modification

on

should

cases

the

, and

spatial

York

expression

include

butter

representation

polysemous

such

latter

in

on

the

modify

parameters

arguments

involving

include while

New

domain

between of true

or

those

is the

expensive

interpretation

' s semantic

to

define . This

an

which

temporal

down

arguments

Verbal

item of

drove

surface

with

situational

expressions

other

the

the

lexical

at syntax

toast

: Parameters

of

" Mary

pressed

her

a

schematically

, for default in

an

expression

argument ( 4b ) .

such C , the

as

( 4a ) with

interpretation

true can

be

represented

and

The Semantic Type System

65

(4) a. A verb B with C b. verb'(A,B,C) If the defaultargument is unexpressed however , thenexistentialclosure on that argumentgivesthe representation in (5). (5) ~x[verb'(A,B,x)] Like defaultarguments , shadow arguments referto semantic content that is not necessarily expressed in syntax, aswith the incorporated semantic contentin the verbsbutterandkickin (6). (6) a. Marybutteredhertoast. b. Harrykickedthewall. The "hiddenargument " in (6a) is thematerialbeingspreadonthetoast, whilein (6b) it is the legwhichcomesinto contactwith the wall. Unlike defaultarguments , however , whichareoptionallyexpressed dueto conditions abovethelevelof the sentence (i.e., discourse andcontextual factors ), shadowarguments areexpressible only underspecificconditionswithin the sentence itself; namely , whenthe expressed arguments standsin a subtypingrelationto theshadow argument (cf. Wunderlich , 1987 ) .3 Observe the expressed shadowarguments in (7). (7) a. b. c. d.

Marybutteredhertoastwith margarine / *with butter. Harrykickedthewallwith hisgammyleg/ *with hisleg. MaryandJohndanceda waltz/ *a dance . Harryelbowed mewith hisarthriticelbow/ *with hiselbow.

Becausethe conditions under which these arguments can be expressed are so specific, I will distinguish them in logical type from the larger class of default arguments. The final class in our typology of argument types is defined more by complementarity than in terms of specific properties of the class; namely, adjuncts. I will have little to say about this type of argument here, except that they are associatedwith verb classesand not individual verbs. Hence, for example, the ability of the verb sleepto be modified by the temporal expressionon Tuesdayin (8a) is inherited by virtue of the verb's classification as an individuated event; similar remarks hold for the verb seeand locative modifiers such as in Boston in (8b).

66

Chapter5

(8) a. John slept late on Tuesday . b. Mary sawBill in Boston. The aboveclassificationis a first attempt at refining the distinction betweenargumentand adjunct phrases . The theoreticalconsequences of this moveare potentially significant; namely, it is not j list the lexical propertiesof a singleitem which will determinethe logicalstatus of a phraseasa certainargumenttype. Compositionaloperationsmaycreate an argumentor shadowan argumentat a phrasalprojection, by virtue of compositionalityin the phrase. In other cases , however , a true argument is defaultedby virtue of a complement 's semantics . For example,for the verb show,the true argumentexpressing the GOALargumentin (9a), can be defaultedby virtue of the semanticsof the complement , as in (9b), therebybecomingan optional argument. (9) a. Mary showedher paintingsto John. b. Mary showeda movie(to John). That is, the phraseshowa moviesomehowdemotesthe otherwisetrue GOALargumentto a default argument, giving it an optional status in the syntax. I return to this topic in chapter7. Another consequence of this view of argumenttypes is that default argumentscan be satisfiedby full phrasalexpression asa PP (cf. (lOa)), or as a phraseincorporatedinto a true argument, as in (lOb): (10) a. Mary built a housewith wood. b. Mary built a woodenhouse. In fact, whenthis occurs, the materialis expressibleas a shadowargument, as in (11) below. (11) Mary built a woodenhouseout of pine. In (lOb) the default argumenthaseffectivelybeensaturatedindirectly as a modifier in the direct object, while in (11) further specification by the phraseout of pine is licensedin the samemanneras in shadow arguments .4 From the discussionabove, I will assumethat the argumentsfor a lexical item, ARGl, . . .,ARGn , are representedin a list structure where argumenttypeis directly encodedin the argumentstructure, ARGSTR ,

67

The Semantic Type System

as

shown

is

a

in

(

shadow

12

)

below

,

argument

where

D

-

ARG

is

a

default

argument

,

and

S

-

ARG

.

a

(

12

)

ARGSTR

=

ARG2 D -

S ARGl

[

For

example

now

be

,

the

ARGl

in

(

13

)

-

(

.

.

.

.

. .

semantics

represented

illustrated

. =

= =

lexical

partially

=

ARGI

)

.

.

.

. ]

for

with

15

.

argument

the

verbs

structure

discussed

above

specifications

can

,

as

.

build

(13)

ARGSTR

=

butter

(14)

ARGSTR

=

kick

(15)

ARGSTR

.

.

=

.

I have yet to discussthe formal conditions under which these arguments are licensed or expressed, but what 'should be clear from this discussion is the usefulnessof the logical distinction in argument types, both descriptively in terms of coverageof construction types, and theoretically in terms of the formulation of principles of mapping from lexical semantic forms.

5.3

Extended Event Structure

It has become quite standard to acknowledgethe role of events in verbal semantics. Conventionally, the event variable for a verb within an event-based semanticsis listed as a single argument along with the logical parameter defined by a particular predicate or relation. For example, a lexical representation for the verb build merging aspectsof both Davidson's (1967) and Parsons' (1990) analysesmight be as follows:5

68

Chapter 5

AYAXAe [build (e, x , y) 1\ (}l (e, x) 1\ (}2(e, y)] This assumesan atomic view on event structure , where internal aspects of the event referred to by the single variable are inaccessible. Moens and Steedman (1988) and Pustejovsky (1991b) argue that finer-grained distinctions are necessaryfor event descriptions in order to capture some of the phenomena associatedwith aspect and Aktionsarten . Assuming this is the case, we need a means for both representing the subeventual structure associated with lexical items while expressing the necessary relation between events and the arguments of the verb. In Pustejovsky (1995a), a mechanismcalled Orthogonal Parameter Binding is outlined , which allows us to bind into an expression from independent parameter lists, i .e., argument structure and event structure. Given a listing of arguments and an event structure represented as a listing of event variables,

[ARGSTR :=: ARG }, ARG2 , ..., ARGn ] [EVENTSTR :=: EVENT }, EVENT2 , ..., EVENT n] we can view the semanticsof the verb as being centrally defined by the qualia, but constrained by type information from the two parameter lists. The predicates in the qualia refer directly to the parameters: [QUALIA = [ . . . [Qi = PRED(EVENTj ,ARGk)] . . .] I will return to the specificsof qualia structure in the next section. First , let us turn to the exact nature of the event argument list illustrated above, and what motivation exists for its structure . Proceeding from our earlier discussion in chapter 2, I assume that events can be subclassifiedinto at least three sorts: PROCESSES , STATES , and TRANSITIONs .6 F\1rthermore, I assumea subeventualstructure to these event sorts as well. This has the advantageof allowing principles of predicate-argument binding to refer to subeventsin the semantic representation, a move which has significant theoretical consequences(cf. Grimshaw, 1990, and Pustejovsky, 1991b). As shown in Pustejovsky and Busa (1995), however, evidencefrom unaccusativity and the varied nature of causative constructions showsthat this notion of event structure does not fully capture the underlying semantics of unaccusative constructions, with respect to how the subeventsproject to syntax.

69

The Semantic Type System

Within an event semantics defined not only by sorts but also by the internal configurational properties of the event, we need to represent the relation between an event and its proper subevents. Extending the constructions introduced in van Benthem (1983) and Kamp (1979), we interpret an "extended event structure" as a tuple , < E , j , < , , ~ , * >, where E is the set of events, :S is a partial order of part-of, < is a strict partial order, 0 is overlap, ~ is inclusion, and * designatesthe "head" of an event, to be defined below. An event structure with structured subeventssuch as that in Pustejovsky (1988, 1991), shown in (16) below,

(16) e < cx:

el

e2

can be constructed in such a model as follows. Let us define an event tree structure in terms of the relation of "exhaustive ordered part of," < cx..7 (17) a. [e3el < cx e2] = def < cx({ el , e2} , e3) b. Vel , e2,e3[< cx ({ el , e2} , e3) +-'t el :::5 e31\ e2 :::5 e31\ el < e21\ Ve[e ~ e3 - t e == e1 V e = e2]] This definition states that the event e3 is a complex event structure constituted of two subevents, el and e2, where el and e2 are temporally ordered such that the first precedesthe second, each is a logical part of e3, and there is no other event that is part of e3. Verbs included in this description are causativesas well as inchoatives, as argued in chapter 9. An event tree structure is not restricted to representing strictly sequential relations between subevents, but structures other orderings as well. For example, an event composedof two completely simultaneous subeventscan be lexicalized with a relation we will call "exhaustive overlap part of," cx ::. It can be defined as follows (cf. Kamp , 1979, and Allen , 1984): (18) a. [e3 el

cx : e2 ] = def

cx :( { el ?e2} , e3)

70

Chapter 5

b. Vel , e2, e3[ocx :({ el , e2} , e3) ~ el ~ e3.I\e2 ~ e31\ el ~ e21\ e2 ~ ell \ 3e[e ~ ell \ e ~ e21\ e = e3] /\ \ie[e :Se3 -1- e = el Ve = e2]]

This event structure is denoted by verbs such as accompany, which involve two subevents occurring simultaneously. This can be illustrated in terms of an event structure tree as follows:

(19) eooc

el

The

interesting

ence

to

both

telic

an

( 20 )

and

In

this multiple

will

accompany

you

similar

, it event

, the

interpretations

be

a.

part

of

the

event

:

<

cx: e2 ]

, e2 , e3 [ <

The

relation

and

e2 , where

( el ) V e =

<

<

to

the

while

on

store

I was

it

to

the

refer

-

assumes

context

. ( telic

verb

makes

, and

the

walking

that

= def

two

basically

, called

.

) . ( atelic

go , which

)

also

, where event

<

simultaneous

" exhaustive init

, and

is

admits

( e2 ) 1\ end

subevents

ordered

a function

end

,

overlap over

is a function

,"

events

returning

( el )

=5 e3 1\ e2

== end

=5 e3 / \ el

( e2 ) / \ Ve [e ::s' e3

0 e2 1\ - - + e ==

e2 ] ]

starts

defines before

an e2 .

,

cx: ( { el , e2 } , e3 )

cx: ( { el , e2 } , e3 ) f--+ el init

o ()( above el

other

follows

initial

init el

as

of

[e3 el

b . Vel

the

defined

the part

of

before

, because

underspecified

.

lexicalization

starts

is that

, depending

is aspectually

final

:

interpretations

sense

returning

below

a verb

aspectually

me

one

( 21 )

is

accompanied

cx , can

the

such

, it

b . Mary

Finally where

about event

atelic

a . John

of

<

thing implicit

e2

event The

containing event

structure

two

subevents tree

is

, el shown

71

The Semantic Type System

(22)

e < ooc

el

e2

We shall see that because of this partial causative

relation

exists

between

ordered relation , a type of

the subevents

, but

not

in the

same

way

as with the relation < cx : . Verbs such as walk will be analyzed as involving this

subeventual

structure

, where

two

motion

processes

are structured

in

an overlapping relation ; that is, the efficient motion of the legs bringing about the final motion of the body . I will argue that this relation is also present with control interpretations for aspectual predicates such as begin, with the associated entailments regarding causation . Given

what

I have

said

so far , there

are two

facets

of an event

tree

structure that need to be represented for a lexical structure : the specific events and their types ; and the ordering restriction over these events .

This is illustrated schematically in (23). a

(23 ) EVENTSTR =[~~: ::: ] RESTR

=

.. .

For example , the verb build is typically analyzed as involving a develop -

ment processand a resulting state (cf. Dowty, 1979, Moens and Steedman, 1988, Pustejovsky, 1991b), ordered by the relation "exhaustive ordered

part of ," < cx :. build

(24)

[El=process ]

EVENTSTR = E2 = state RESTR

. .. Unlike

build , however , which

=

< cx:

constrains

the types of its two subevents

to PROCESS and STATE , the verb accompany TRANSITIONS , or PROCESSES. The typing

permits

constraint

either

telic events ,

, however , is similar

to that for a coordinate structure in that they must be of like type : accompany

(25)

EVENTSTR

=

E2 =Ti Ti RESTR =cx :] [El =

72 .

Chapter 5

Of course, there are many more possible relations between subevents .

than are actually realized in lexicalized forms in natural languages ;8 one of the principal goals of a semantic theory must be to constrain the model to reflect these restrictions . The structural information discussed thus far for event structure , although necessary, is not sufficient to capture lexical distinctions that languages systematically make with respect to the relative prominence or importance of the subevents of a larger event . Talmy (1975,1976) and others have long noted that the event information conveyed by a verb can be much richer than the "sequence of events" structure encoded in the representations above . These grammatical observations , however , can be accounted for in terms of something I will call event headedness (cf . Pustejovsky , 1988) . Event headedness provides a way of indicating a type of foregrounding and backgrounding of event arguments . An event structure provides a configuration where events are not only ordered by temporal precedence, but also by relative prominence . One instance of prominence for an event , e, is provided by the HEAD marker , annotated as e* . The conventional role of a head in a syntactic representation is to indicate prominence and distinction . Rules of agreement , government , etc . militate in favor of marking structures in terms of heads of phrases. Within the interpretive domain of events, when viewed in a structural or configurational manner , the possibility of referring to heads becomes available . Informally , the head is defined as the most prominent subevent in the event structure of a predicate , which contributes to the "focus " of the interpretation . We can view * as a relation between events , * (ei , ej ) ( "ei is a head of ej " ) , where ei .-:::Sej :

(26) *(ei, ej) ==def [e. . . . e; . . .] .1

Headednessis a property of all event sorts, but acts to distinguish the set of transitions , specifying what part of the matrix event is being focusedby the lexical item in question. Adding the property of headedness to the event structure gives the following representation: a

(27)

EVENTSTR

=

El = . . E2 RESTR = .. HEAD Ei

73

The Semantic Type System

Assuming that events have at most a binary event structure , and that

there are three temporal ordering relations realized in language 0( , CO ( , and < CO ( ) there are six possible head configurations with two events , given a single head; there are twelve possibilities , if unheaded and

double

- headed

constructions

are

included

. These

are listed

below

along with an example of each type , where HEAD is indicated by an asterisk , and the event tree structure is given in a linear representation :

(28) a. b. c. d. e. f. g.

[eO ' el * < cx : e2] - build [eO ' el < cx : e2*] - arrive [eO ' el * < cx : e2*] - give [eO ' el < cx : e2] - UNDERSPECIFIED [eO ' el * 0(:;(: e2] - buy [eO ' el cx : e2*] - sell [eO ' el * 0(:;(: e2*] - marry

h . [eO ' el 0(:;(: e2] -

i. j. k. 1.

[eO ' el * < [eO ' el < [eCT el * < [eCT el <

UNDERSPECIFIED

<x e2] - walk <x e2*] - walk home (x e2*] - ?? (x e2] - UNDERSPECIFIED

Intuitively , structure (28a) representsaccomplishmentverbs, where the initial event is headed, focusing the action bringing about a state; (28b) represents achievement verbs , for which the persistence of the final state

is the focus of interpretation ; (28c) illustrates events involving a relational predicate on each subevent , and characterizes unilateral transi tions with

three arguments , i .e., a subclass of ditransitive

transfer

verbs

such as give and take. Predicate pairs such as buy and sell are character -

ized by (28e) and (28f) respectively, where there are two simultaneous events involved in the transaction , but only one is focused by the lexical item

.

Finally , ordered overlap gives rise to the structure in (28i) , where one event begins , and subsequently gives rise to another process which continues only while the first event continues to hold .

Notice that the structures in (28d), (28h), and (281) are unheaded and hence ill -formed without

further specification at surface structure

(cf . chapter 9) . The role of semantic underspecification will figure promi nently in the analysis of verbal polysemy . In terms of event structure , polysemy occurs when a lexical expression is unspecified with respect

74

Chapter 5

to headedness , i .e., headless. Headless event structures admit of two possible interpretations . More generally, a predicate should be as many ways ambiguous as there are potential heads. This representation provides us with a mechanismfor relating the logical sensesof polymorphic verbs such as: causative/ unaccusativepredicates such as breakand sink, associatedwith (28d); argument inversion predicates such as rent, associated with (28h); and raising/ control predicates such as beginand stop, associatedwith (281).9 There are several motivations for positing a head as part of an event structure , a matter that is discussedin Pustejovsky (1,988) as well as in Pustejovsky and Busa (1995). When adjoined to predicates denoting transitions , prepositional and adverbial phrasesnot only can modify the entire event, but can also take scope over the individual subevents. In particular , heads seemto license certain types of modification . Observe that the durative adverbials in (29) modify the designated head of the event rather than the entire event structure .

(29) a. John ran home for an hour . b. My terminal died for two days . c .

Mary left town for two week .

The event tree structure is illustrated in (30) below, where, until the formal structure of qualia is presentedin the next section, I refer informally to the expressionsassociatedwith each event in a tree structure .

(30)

It is generally assumedthat only processesand states license durative adverbials, yet modification by an adverbial in (29) is grammatical, even though the sentencesdenote telic events. Obviously, the available interpretation has the adverbial modifying the duration of the final state; in (29a) John spent an hour at home, in (2gb) the terminal was dead for two days, and in (29c) Mary was out of town for a period of two weeks.

75

The Semantic Type System

A similar

phenomenon

TIONS ) , when

modified

quietly

; namely

( 31 )

In

, modification

a . John

built

b . Mary

quietly

( 31a ) carelessly

into

existence

event

tree

tion

structure

constitutive

relation , gives

the

which

Evidence

from

of making ( but

This erence structure verbal generate

the

of mass

two

house

to the

e1 and

as an integral these

the

( 31 b ) .10 In the

approach

/ / / " ' ~" -"

terms

part

variables

rela -

resulting . The

of qualia

.

"""- e2

I 3y [draw

ture

.

e < cx

' ~' / ei

vor

in

process

below

between

brought

sentence

in the

, defined

( 32 )

) subevent

and

.

' s ( 1972 ) treatment

relationship

( e .g ., TRANSI as carelessly

( headed

a predicative

drawn

Burge

such

.

for

, I assume

, caNST

the

initial

act of building hold

being to

the

events

adverbs

carelessly

the

below object

left - headed

manner

a picture

remarks

in e2 , similar

structure

house drew

modifies

'the

with

is over

the

; similar

between

picture

occurs by certain

event

modification

reference

complete

our .

In

is expressive semantics semantic

while

3x [picture

is only

to a focusing

cf . Pustejovsky

to events

I

(m , y )]

one of several

mechanism

( 1988 ) for more initial

later

discussion chapters

enough also

expressions

to being .

of how

the

sufficiently

in fa -

an event

struc -

).

lexical

see how

(y , x ) ]

arguments

within

discussion

we will capture

( x ) A Const

items this

polymorphic constrained

make

view

behavior to

ref -

of event

not

of over -

76

5.4

Chapter 5

Qualia Structure

In this section, I outline the structured representation which gives the relational force of a lexical item, called the qualia structure. In some sense, a generative lexicon analyzes all lexical items as relational to a certain degree, but the manner in which this property is expressed functionally will of course differ from category to category, as well as between semantic classes. A more detailed discussionon the structure of qualia is presented in chapter 6. Briefly, Qualia Structure specifies four essential aspectsof a word's meaning (or qualia): . CONSTITUTIVE : the relation between an object and its constituent parts; . FORMAL : that which distinguishes it ,vithin a larger domain; . TELIC: its purpose and function; . AGENTIVE : factors involved in its origin or "bringing it about" . This view of word meaning, inspired in part by Moravcsik's (1973) interpretation of Aristotle 's modes of explanations (aitiae), emphasizes aspects of a word's meaning that have been largely ignored by formal treatments of lexical semantics. By looking at modes of explanation for a word, we permit a much richer description of meaning than either a simple decompositional view or a purely relational approach to word meaning would allow. These qualia are in fact structures very much like phrase structural descriptions for syntactic analysis, which admit of something similar to transformational operations in order to capture polymorphic behavior as well as sensecreation phenomena. The latter are the generative mechanismscoveredin chapter 6. There are two general points that should be made concerning qualia roles: (1) Every category expressesa qualia structure ; (2) Not all lexical i terns carry a value for each qualia role. The first point is important for how a generative lexicon provides a uniform semantic representation compositionally from all elements of a phrase. This view of qualia structure is a natural extension of the original applications where qualia for verbal representations were not

The Semantic Type System

77

discussed. The second point allows us to view qualia as applicable or specifiable relative to particular semantic classes. To begin our discussion, let us seehow qualia structure encodesthe meanings of nominals. We can think of qualia, in some sense, as that set of properties or events associatedwith a lexical item which best explain what that word means. For example, in order to understand what nouns such as co'okie and beermean, we recognizethat they are a foodstuff and a beverage, respectively. While cookieis a term that describes a particular kind of object in the world, the noun foodstuff denotes by making functional referenceto what we do with something, i .e., how we use it . In this case, the term is defined in part by the fact that food is something one eats, for a specific purpose, and so on. Similar remarks hold for the information related with the noun beer. The TELIC quale for the noun food encodesthis functional aspect of meaning, represented informally as [TELIC= eating]. Likewise, the distinction betweensemantically related nouns such as novel and dictionary stems from what we do with these objects that is different. That is, although both objects are books in a general sense, how we use them differs: while one readsa novel, dictionaries are for consulting. Hence, the respective qualia values encoding this functional information for novel and dictionary are [TELIC= reading ] and [TELIC= consulting ]. This distinction , of course, is not the only way these concepts differ; the structure of the text in a novel is characteristically a narrative or story, while a dictionary is by definition a listing of words. This distinction is captured by the CaNST (constitutive ) role, expressingthe internal structural differences. Finally , even with their overall similarity as expressedin identical FORMALroles, novels and dictionary also differ in how they come into being (cf. Aristotle , Metaphysics), expressedin the AGENTIVErole. That is, while a novel is generally written , a dictionary is compiled.! ! As we shall see, however, the qualia are not simply a listing of interesting facts about an object or lexical item, but provide the jumping off point for operations of semantic reconstruction and type change, which in turn contribute to our overall goal of characterizing a natural languageas polymorphic. As in our previous discussion, I will employ a genericfeature structure as the initial representation for the qualia structure . For a lexical item a , let us first characterize the roles as given in (33).

78

(33)

Chapter 5

a .. CONST = . . QUALIA =FORMAL = . TELIC = .. . AGENT

The listing above tells us nothing about how a particular lexical item denotes, however. For example, although a novel's purposeis the activity of reading and it comes about by someonewriting it , we do not want to claim that the common noun novel actually denotes such activities . Therefore, we cannot simply list the qualia values for an item, such as in (34), even though they are intuitively correct, without somehowbinding them appropriately. novel (34)

QUALIA=

CONST= narrative FORMAL= book TELIC= reading AGENT= writing

Not only doesthis prove problematic from the view of predication, but it also ignoresthe data from previous chaptersmotivating a more polymorphic treatment of semantics. That is, the noun novel should predicate in some fairly conventional way (cf. chapter 7), but in particular contexts, it should permit of interpretations that are licensed both by local syntactic and semantic context as well as reconstruction from the semantics of the word, as with sentencessuch as Mary enjoyed the novel. The solution to these concernsis to treat the qualia values as expressions with well-defined types and relational structures (cf. Pustejovsky, 1991, Copestakeand Briscoe, 1992). For example, the arguments to the relation read are explicitly given in (35), indicating the proper binding of the predicating term : novel

(35)

QUALIA =

FORMAL= book (x ) TELIC = read (y,x )

This is equivalent to the A-expression below , as a partial denotation of the noun semantics for novel:

79

The Semantic Type System

(36) Ax[novel(x) 1\ . . . TELlO Ay(read(y, x)] . . .]

Given

the

where

representation

contextual

in

context

"

effect

termining

)

these

.

Mary

.

John

finished

~

~

.

John

began

structure

,

to

associated

the

me

turn

qualia

cussion

of

with

modes

of

stative

predicate

. g

. ,

ignoring

with

the

to

event

the

details

the

.

De

7

NPs

-

.

will

.

containing

and

provide

verb

required

I

activities

the

information

finish

the

return

semantics

to

which

for

this

a

which

of

.

in

to

" contex

issue

5

. 3

I

will

in

,

the

defer

event

-

later

.

in

be

for

the

-

how

Intuitively

qualia

,

a

structure

to

how

represented

now

dis

begun

broadly

reference

might

played

detailed

discussion

structure

without

structure

role

a

distinguish

role

tall

argument

,

' s

exists

I

the

FORMAL

predicate

and

.

Continuing

verb

the

verbs

semantics

9

into

the

of

the

structure

affairs

,

sense

chapter

properties

turn

chapter

map

of

see

"

finish

of

consequently

in

or

until

example

and

subject

particular

This

corresponds

For

the

.

in

qualia

state

.

begin

the

to

derive

.

explanation

that

.

now

extended

possible

to

verbs

is

and

representation

verbal

the

about

briefly

least

order

.

,

begin

g

- based

derived

about

of

at

in

the

beer

NP

"

now

,

with

second

them

is

.

nouns

chapter

it

cigarette

his

sense

in

a

novel

the

the

the

Let

a

complement

discussion

by

~

information

with

tualizing

~

,

from

are

enables

encode

governs

~

)

below

readings

a

Qualia

them

( 35

comes

illustrated

b

c

e

,

how

( 37

in

information

,

it

as

came

in

( 38

)

,

:

tall

EVENTSTR

( 38

At

QUALIA

)

first

ply

cates

. 1

it

the

is

the

being

~ .~

El

el

might

appear

tall

a

into

not

one

- level

resulting

of

( el

' X

]

or

particular

captured

by

stage

event

- type

such

- level

,

but

and

i . e

. ,

move

the

slot

a

.

strategy

is

a

?

qualia

I

argue

the

.

-

in

predi

- based

to

quale

sim

semantic

level

reference

AGENTIVE

to

What

individual

rather

Involves

,

superfluous

named

predication

state

)

gratuitous

a

is

]

tall

between

stage

the

: state

=

generalization

where

of

=

RMAL

distinction

actually

,

[

[

predicate

or

that

tinction

into

=

glance

shift

motivation

10

=

bringing

-

dis

-

80

Chapter 5

Consider

next

icates

(

i

volving

. e

.

,

an

two

the

qualia

structure

TRANSITIONS

)

initial

phases

act

map

respectively

or

The

associated

These

are

followed

into

the

transitive

with

verbs

process

directly

.

.

by

of

the

verb

pred

analyzed

a

AGENTIVE

form

causative

typically

resulting

state

and

FORMAL

break

is

-

as

.

in

-

These

qualia

roles

illustrated

in

(

,

39

)

.

break

EVENTSTR

=

[ (

39

)

QUALIA

-

FORMAL

-

AGENTIVE .

Intuitively

,

ticular

this

Finally

,

sider

here

cesses

.

guish

active

verbs

: process

not

only

(40)

(41)

(

break

e2

_

events

'

Y

act

)

(

and

el

, x

, y

)

relational

corresponds

expressions

roughly

,

but

to

where

the

the

to

par

annotated

expressions

in

passive

motion

-

event

are

now

as

typed

.

and

It

,

qualia

with

,

verb

with

,

be

an

while

,

many

where

for

run

.

For

example

this

verbs

distinction

distinction

process

,

-

-

includes

unergative

active

-

distin

includes

that

/

con

pro

former

latter

out

the

will

FORMAL

the

unaccusative

to

I

languages

the

pointed

.

and

)

,

structure

the

1985

processes

move

to

The

1975

should

isomorphic

.

according

associated

AGENTIVE

(

of

run

snore

distinguished

is

Talmy

classes

such

and

are

between

out

and

process

predicate

distinction

pointed

sleep

a

the

the

below

structure

oc

broken

=

above

necessarily

qualia

=

explanation

language

lustrated

]

<

.

denoting

of

of

the

: state

=

This

predicates

As

as

a

el

.

mode

is

.

qualia

which

in

e2

=

presented

named

such

=

El

specific

roles

structures

by

.

relates

qualia

tree

E2

RESTR

the

class

verb

is

run

il

carries

run EVENTSTR =[El =el :process ] QUALIA = [~ ~ ENTIVE = run _ a ( e ct , X ) ] l sleep EVENTSTR =[El =el :process ] QUALIA =[~ .~RMAL =Sleep (e,X )]l structure

in

in

(

41

)

.

(

40

)

while

the

passive

process

verb

sleep

has

the

-

81

The Semantic Type System

The relevanceof this distinction in qualia binding is that processesare quantifiable in different ways, one of which is to specify explicitly the resulting state in the FORMALrole, by meansof co-composition. This is possible only with active processes , and is illustrated with constructions such as run to the store and run home (cf. Hinrichs, 1985, Verkuyl and Zwarts, 1988, and Jackendoff, 1990). Quantification of passiveprocesses generally does not allow the cognate construction, and is limited to modification by durative adverbials, as in sleepfor an hour, cough all night. In later discussion, I will show how processesassociated with more complex event structures, such as that proposed for the verb walk above, bind into both AGENTIVEand FORMALqualia roles. In this section, I have outlined informally how qualia structure contributes towards describing various aspectsof noun and verb meaning. I have limited my discussionto these two categoriesin order to illustrate the basic descriptive power of this level of representation. In chapter 10 below and in Pustejovsky (1995b) I outline the appropriate extensions to qualia structure to encompassall categories.

5.5

The Interaction

of Semantic Levels

In this section, I integrate the three levelsof argument, event, and qualia structure , in order to construct a uniform languagefor lexical semantic representations. I will assume, for discussion, a system based on typed feature structures such as Carpenter's (1992) system, and following its application to lexical semantics described in Copestake et at. (1993). The type system has two parts; the type hierarchy itself, and the constraint system operating over the types. I will not discussthe details of the constraints within the typing languageexcept as they relate to linguistic data and questionsof generativity. I assume, furthermore, along with Sanfilippo (1993), that semantic class information may be inherited by subtyping specifications, providing constraints on the types of arguments.12 To illustrate how the three levels described in previous sections come together to provide an integrated representation, consider again the semantics of the verb build. We have already provided structures for each level of representation independently, and now we need to unify them in a coherent fashion. Recall that there are three arguments associated

82

Chapter 5

with the verb: two TRUE ARGUMENTS and one DEFAULTARGUMENT . Furthermore, we analyzed the verb as a lexical accomplishment, containing two subevents, a processand a resulting state. These are bound to the qualia in the representation below .

ENTSTR

(42)

El = el::state prOCess E2 = e2 = HEAD RESTR
The processis identified as that AGENTIVEact involving both the deep syntactic subject, ARGl, and the default argument, D-ARGl, which is related to the logical object by the CONSTITUTIVE relation of ARG-2. The FORMALrole expressesthe resulting state of there being such an object ARG-2. Since this individual is defined as being made of the material of the default argument D-ARGl, and yet is logically distinct from it , as suggestedin 5.3 above, there are two possibilities for existential closure on such an event structure , as ill ustr ated in (43) and (44): (43) a. John is building a house. b. 3z[build_act(el ,j , z) /\ material (z) . . .] (44) a. John built a house. b. 3z3y[build_act(el ,j , z) 1\ material (z) 1\ exist (e2, Y) I\ house(y) 1\ el < e2] There are several consequencesof this representation which we will not explore in this book. One result is that the binding problem in the imperfective (44) is overcome in a fairly natural way. That is, there is no assertion that a houseexists in this atelic form of the event. Rather, this representation assertsno more than that the substancemaking up some (future ) house has been acted upon.

The Semantic Type System

83

The notion that concepts are associated with other concepts by some sort of explicit listing is, of course, not a very new or controversial notion . Indeed , much work in AI and computational approaches to word meaning has been concerned with providing "hooks" on word meanings , in order to facilitate inference . In some respects , qualia structure shares with these traditions the goal of capturing the meaning of a word as it relates to reasoning viewed in a much broader sense. But the principal motivation for qualia structure is not simply a listing of properties for a word . Rather , it can only be appreciated in the context of the generative mechanisms in the grammar which allow speakers to use the language creatively .

6

Qualia Structure

In this chapter I detail my assumptionsabout how lexical items encode semantic information in the qualia, and particularly , what the qualia structure conveys. In the previous chapter, I argued for associating functional semantic information with all categories, and how this move helps us towards achieving a more generative description of lexical sense derivation. The point of this chapter is to explore more fully the syntax of qualia structure and what its descriptive and explanatory role can be in a theory of word meaning and compositionality.l

6.1

Modes of Explanation

Aristotle 's notion of modes of explanation (or generative factors), as pointed out by Moravcsik (1975), can be viewed as a system of constructive understanding and inference. Thesefour factors drive our basic understanding of an object or a relation in the world . They furthermore contribute to (or, in fact, determine) our ability to name an object with a certain predication.2 Following our earlier discussion, assumethat a lexical semantic structure is defined by the four interpretive levels, , and that furthermore there are four basic roles that constitute the qualia structure Q for a lexical item. Now let us elaborate on what these roles are and why they are a necessarycomponent of the semantics. In 6.2 below, the qualia are given an informal interpretation , with possiblevaluesthat each role may assume.3 1

.

CONSTITUTIVE

or proper

: the

relation between an object and its constituents ,

parts .

i . Material

ii . Weight iii . Parts and component elements 2. FORMAL : That which distinguishes the object within .

mall

.

i . Orientation

ii . Magnitude iii . Shape iv . Dimensionality

a larger do-

86

Chapter6

v . Color vi . Position 3. TELIC : Purpose and function of the object . i . Purpose that an agent has in performing an act . ii . Built -in function or aim which specifies certain activities . 4. AGENTIVE : Factors involved in the origin or "bringing about " of an object . i . Creator ii . Artifact iii . Natural Kind iv . Causal Chain There are of course many ways of approaching the definition of a word , and qualia might seem at first to be a simple listing of case roles or named features associated with lexical items . This is not the case, however . What qualia structure tells us about a concept is the set of semantic constraints by which we understand a word when embedded within the language . The mode of explanation that characterizes a word as denoting a particular concept is potentially distinct from the manner in which that word is used in the language . That is, definition and word meaning need not have anything to do with grammaticalization or grammatical behavior . This is a fairly uninteresting claim about the relation between language and thought , and certainly unintuitive from a philosophical perspective . Yet , this is exactly what many models of semantics assume: namely , words have simple denotations , and the methods of composition are borrowed from general logical inference mechanisms . The view within a generative lexicon is different . The qualia provide the structural template over which semantic transformations may apply to alter the denotation of a lexical item or phrase . These transformations are the generative devices such as type coercion , selective binding , and co-composition , which formally map the expression to a new meaning . These operations apply only by virtue of lexical governance relations ; that is, their application is conditioned by the syntactic and semantic environment within which the phrase appears . For example , when we combine the qualia structure of an NP with that of a governing verb , we begin to see a richer notion of compositionality emerging , one which captures the creative use of words and the "sense in context " phenomena discussed in chapters 3 and 4.

87

Qualia Structure

A semantic representation is only useful if it facilitates a logical inferenceor interpretation . Furthermore, such information can be attributed to a specific lexical item only if it can be argued that the interpretive processis driven by language-specific representationsor constraints rather than logical inference alone. Qualia are interesting in this respect, since they not only structure our knowledge of words, but also "suggest" interpretations of words in context. Consider, for example, how the NPs, both in subject and complement position contribute towards further specifying the interpretation of the verb use in the sentencesin (1) and the NPs in (2) below.

(1) a. b. c

d

John

used

Mary

.

This

.

My

the

has

car

wife

new

used

uses

uses

knife

soft

unleaded

the

on the turkey.

contact

subway

since college.

lenses

gasoline

.

every

day

.

(2) a. the tonersusedin c~pyingmachines b. the yeastusedin beer In sentence(la ) , our knowledgeof knives as tools which can cut permits an economy of expression, whereby mention of the particular activity of cutting may be ellipsed. Similarly, in (1b), contact lensesare visual aids, and the use of them refers to the act of wearing them. The utility of a verb such as use is that it is semantically light (cf. Grimshaw and Rosen, 1990), what I will refer to as simply underspecified , with respect to the particular activity being performed. Sentence(1d), for example, is a near paraphrase of "My wife travels on the subway every day," an interpretation that is made possible by our knowledge of what the function of a subway is. Finally , the NPs in (2) show an interesting sense distinction , where the objects in (2a) are understood as standing in a functional part -of relation , while in (2b), the substanceis interpreted as being part of the material used for the processof making beer.4 The factors allowing us to determine which senseis appropriate for any of these casesare twofold: (1) the qualia structures for each phrase in the construction; and (2) a richer mode of composition, which is able to take advantage of this qualia information . As further evidenceof the way in which qualia structure elucidates an interpretation in context, consider the contextualized meanings for the

88

Chapter6

verb enjoy, discussedin chapters2 and 4 above, wherethe "ellipsed" predicateis suppliedby informationfrom the complement . (3) a. Mary enjoyedthe movielast night. (watching ) b. John quite ~!~~~~ his morningcoffee. (drinking) c. Bill enjoyedStevenKing's last book. (reading ) Although there arecertainlyanynumberof waysof enjoyingsomething , our understandingof thesesentences is facilitatedby defaultinterpretations of propertiesand activitiesassociatedwith objects. The qualia,of an object canbe seenasthe initial pointsfrom whichto constructinterpretationsthat would otherwisebe ill-formed. Hence, the TELICroles for movie, coffee , and booksomehowproject the activities of watching the movie, drinking his morning coffee , and readingStevenKing's last book,respectively , to the interpretationof the VP. Noticethat the contextualizationof a sensefor a verb neednot come from the semanticsof the complement , but can be influencedin obvious waysby the subjectas well. For example, in (4a) there are two factors contributing "" to the interpretationof the ellipsedpredicate; the qualia structures'associatedwith the subjectNP and the doubleobject NPs. The complements areidentifiedasnamesof airports, therebyallowinga default co-compositionalinterpretationof landing, takingoff, and soon. (4) a. Most commercialpilots ~~~~ Kennedyto Logan. b. Most commercial pilots prefer New York to Boston Although easily

this

interpretation

defeasible

, since

the sortal

typing

discussed

in chapter

bution likely

made

by

the

In eat

other

words

catfish

rather

, the

and

, it the

farmers

remarks

agentive

nominals is not

fish

for sentence

the

farmers

usually

for the

( 5 ) below

" suggested

are preferrin

prefer

must

of the sentence

hold in

(4a ) , it is more

complements

in the composition

is one that

sellers

Christmas

reading

10 .5 Similar

a . Midwestern b . Book

this

of airport

interpretation

(5 )

is also available

under

, where

.r2 catfish to

inherit

, a process

semantic

" by the

cookbooks

.

VP this

textbooks

contri the

-

most

semantics

.

year . around

.

is not book

the sellers

are raising

case are the

that

fish

farmers

preferring fish

and

to the

read

book

are

preferring

the sellers

books

to , but

are selling

89

Qualia Structure

the books. The TELlOroles from the agentivenominals in theseexamples seem to override any VP-internal interpretation available (cf. 10.2 for discussion) . Thus far, I have restricted my discussion to verbal and nominal semantics, yet the utility of qualia-based representations extends to the other categoriesas well. To demonstrate how, consider the phenomenon of adjectival submodification mentioned in chapters 2 and 4 above, and illustrated in (6) and (7). (6) a. b.

a

~

an

!

!

~

!

.

a

bulb

bulb

(7) a. a ~ b

~

opaque

ill

typist ~

~

typist

The adjectives bright and fast in these examplesare actually event predicates, modifying some aspect of the head noun. In fact, they each seem to make reference to a qualia-derived event associatedwith the noun. For a bulb, this is obviously referenceto what function the bulb has, namely, its TELlC role, which would be illumination . For an agentive nominal such as typist, the TELICmakesdirect referenceto the processdenoting verb from which the nominal is derived. The adjectives opaque and male, on the other hand, make reference to the FORMALrole of the head. It appears as though adjectives are able to subselect on the basis of typing information contained within the qualia structure of the phrases they modify, although it is not yet obvious how this is accomplished. It is possiblethat the adjectives do not select for specific qualia in the head noun, but rather select a particular type, which is or is not available within the value of a quale role. In the next section, I present a more detailed interpretation of the qualia structure as applied to nominal semantics. Issues relating to the grammatical consequencesof this model will not be discusseduntil chapter 7.

90

Chapter 6

6.2

The Qualia Structure

In this section

I explore

of No minals

in more detail

the typing

system

necessary

to

characterize the semantics of NPs , and , in particular , the logically polysemous behavior of nominals such as window and door. The approach taken here is to introduce the analysis of a category in terms of the four levels of representation . In terms of nominals , this means viewing the semantic

classification

from the perspective

of event structure

, argument

structure , and qualia structure . In this section I show how fine-grained distinctions are possible in the semantic behavior of nominal types based on the interaction of argument and qualia structures . As mentioned earlier , I assume a system based on typed feature struc -

tures, as applied to lexical structures in Copestake et at. (1993). The type system includes the type hierarchy and the constraint system oper ating over the types . The hierarchy in 6.1, for example , illustrates how features are organized in a lattice structure , where nomrqs refers to the least upper

bound

type for the types entity

, proposition

, and event

(cf . Copestake and Briscoe , 1992, Copestake , 1993, and Pustejovsky

and Boguraev, 1993). The details of the typing constraints will not be discussedhere (but ct. Briscoe et at., 1993); rather I will focus on the linguistic aspects of how a generative lexicon makes use of typed feature structures .

nomrqs

enti ty

event

abstract

. . .

~

Figure

. . .

proposition

p

h b. ' .c . ys _o

J

lllJ .ormatlon

6 .1

Fragment of a Type Hierarchy ; cf. Copestake and Briscoe (1992) , Pustejovsky and Boguraev

( 1993 ) .

91

QualiaStructure

In chapters 3 and 4, I discussed the problem of logical polysemy and

how nominals such as window and door in sentences(8) and (9) carry two distinct interpretations , each logically related to the other :

(8) a. John crawled through the window. b . Mary broke 1b~ windQW.

(9) a. Mary painted th ~ door. b . Mary walked through the door . Each noun has two word senses: a physical object denotation and an

aperture denotation. Pustejovsky and Anick (1988) characterize the meaning of such "Double Figure -Ground " nominals as inherently rela tional , where both arguments are logically part of the meaning of the noun . In addition to the true arguments corresponding to the two senses, there is a default argument making reference to the plane within which the door or window is positioned .6 There are many such classes of nom inals , referring both to 2-dimensional objects such as window and door, as well as to 3-dimensional objects , such as room , fireplace , and pipe . They are interesting semantically , because they are logically ambiguous as well , referring to either the object or to the aperture , or to both . The ability of a lexical item to cluster multiple senses is what Pustejovsky

and Anick (1988) referred to as a Lexical Conceptual Paradigm (lcp) . The intuition herent

behind the notion of an lcp is that there is something in -

in the semantics

of a noun

such that

it is able to project

any

of three separate senses of the noun in distinct syntactic and semantic environments . That is , the listing of the nouns in these separate environments is similar to a paradigmatic behavior . The notion of lcp is

further extended in Pustejovsky and Boguraev (1993) in order to project syntactic behavior from the semantics of the noun characterized as an Icp .7 The lcp provides a means of characterizing a lexical item as a meta entry . This turns out to be very useful for capturing the systematic ambiguities which are so pervasive in language . As discussed above, nouns such as newspaper appear in many semantically distinct contexts , able to function sometimes as an organization , a physical object , or the information

(10)

a.

contained

in the

articles

within

the

newspaper

.

The newspapers attacked thePresident forraising taxes .

92

Chapter 6

b. Mary spilled coffee onthenewspaper . c. John gotangry atthenewspaper . What

the

senses

,

notion

but

for

newspaper

are

those

.

)

a

.

.

ContainerjContainee

.

Figure

e

.

Product

Plant

.

us

that

allows

us

classes

.

a

g

,

application

)

~

;

,

see

.

Gunter

G

,

g

stated

using

a

,

,

the

ffi

/ 3 :

:

0' 1

.

and

such

a

typing

way

exhibiting

.

,

,

New

the

expression

- entry

this

York

.

merger

.

.

within

a

type

system

such

structuring

of

typing

the

of

nominals

with

judgment

,

T

.

a

judgments

The

has

types

in

judgments

typing

type

Honda

.

lcps

A

in

.

appropriate

be

)

application

{ 3 ( o: . )

,

to

1992

,

indicates

bottle

behavior

define

expression

0 ' 1

examination

polysemous

we

analyze

window

apple

;

the

distinct

meta

alternations

newspaper

,

of

how

,

city

analysis

to

can

as

the

.

;

door

fig

not

to

.

;

diathesis

an

be Efj

lamb

;

;

assignment

can operator

we

below

;

diathesis

these

nominal

diathesis

the

an

namely

alternations

that

,

,

Reversals

Result

treat

aspects

repeated

Producer

explain

under

4 those

to

that

alternations

here

( cf

is

alternations

as

examine

to

that

12

People

Imagine

assignment

(

/

/

grammar

states

the

Food

Process

now

do

different

alternations

/

outlined

to

of

Ground

/

Place

us

chapter

Mass

/

.

f .

Let

/

c

a

to

in

Count

allows

the

such

b

d

lcp

expressions

Among

polysemy

( 11

an

logical

presented

logical

as

of

as

these

respect

consists

of

judgment

type

9

T

as

. 8

follows

~

The

a

rule

,

an

:

T

of

where

:

~

_- ~

0' 2

In additionto theapplication ruleabove , wewill needconventional rules of projectionandabstraction . Fornow, however , let usconsider howto constructnewtypesout of existingones . Obviously , something like a recursive type constructorasshownin (13) is needed , heregivenasa rule: (13)

0"1 0"2 0"1 - + 0"2

Thiswouldcorrespond to theruleofrecursive typesin Categorial Gram mar, for example (cf. Lambek , 1958 , Moortgat , 1988 , vanBenthem ,

93

Qualia Structure

1991) . In order to capture the behavior of the nouns above, however , a less conventional type construction is needed. Let us analyze the logical type relation between the polysemous senses of a lexical item such as door as a dotted type. The type constructor , lcp , creates this complex type for a term a which carries senses 0"1 and 0"2, giving the following rule :

(14)

a

: 0" 1

a

: 0" 2

lcp(a) : 0"1 . 0"2

This says effectively that two types are able to combine to become a complex type , treated here as a Cartesian type. The lcp itself is then represented by the type cluster consisting of the two base types and its dotted type , shown below: (15) lcp = { al . a2, a1, a2} This is the paradigm of types associatedwith a lexical item by virtue of the type constructor lcp. Now imagine that 0"1 and 0"2 are the sensesfor a noun such as door, representedas the types phys_obj and aperture respectively. Then the lcp resulting from the type constructor lcp over these types would be the type cluster in (16). (16) phys_obj . aperture _lcp == { phys_obj .aperture , phys_obj , aperture } This is the lcp assignedto the lexical item door. All three types are available for expressionby this noun, as we demonstrate in chapter 7. Lexical conceptual paradigms illustrate very clearly that syntactic information is inheritable between lexical items. To illustrate this point , consider the class of process/ result nominals such as merger, joint venture, consolidation, etc. These nominals are ambiguousbetween a process interpretation (the act of merging) versus the resulting entity or state (the merger which results). This is a property of the whole paradigrr indicating that the alternation can be captured by an lcp. The statement above suggeststhat there are actually three senses available to a lexical item associatedwith an lcp constructed from two base types. This is substantiated by data presented in chapter 8. For the example of merger, given above, this sensewould essentially allow referenceto the entire event of the merging, as well as either processand

94

Chapter 6

result reading ,justasthetensed sentence equivalent ofthisNPdoes (cf. 8.3fordiscussion ). Consider thethreesenses forthenounconstruction . (17) a. Thehouse 'scpnstruction wasfinished in twomonths . b. Theconstruction wasarduous andtedious . c. Ther-onstruction isstanding onthenextstreet . These threesentences make reference totheentire dotted type,theprocess , andtheresultoftheprocess , respectively . Thetypehierarchy in 6.2 below illustrates whattheassociation is between typesresulting fromtheconstructor lcp.

t2 I t5 t6 ~ts .t6 to

t3

t4

Figure 6.2 Type Hierarchywith Dotted Types

Unlikea featureunification operation , wheretwotypes0"1 and0"2 withina latticecanbecharacterized ashavinga unique greatest lower bound(glb), 0"1n 0"2 (asused , for example , in Carpenter , 1992 , and Copestake et at., 1993 ), thisis a different interpretation for howtypes arecombined . Theconcept referred to bya lexicalitemcorresponding to anlcp is not theglbof twotypes , but ratheris theclusterof the twoindividual typesalongwiththedottedtype. Furthermore , thisis notmerelyanoperation of intersection ontwotypes , but a muchmore structured andinformative operation (cf. 8.3 below ). With thisviewof types , wecandistinguish thesenses of newspaper andbook asin (18) below , andwiththis, explain thedifference in logical

95

Qualia Structure

polysemiesavailableto thesenouns(cf. 8.2).

(18)

Given the discussion above, I now examine the specific properties of the qualia and how one can differentiate the semantic classesof nominals on the basis of argument type and qualia structure. I will assumethat for two types a and {3, seen as restrictions on the arguments x and y respectively, the complex or dotted type constructed from them is a . {3. The argument associatedwith this type is notated x .y, and inherits the functional behavior of both types x and y. 6.2.1

The Interpretation

of the FORMALQuale

The most familiar equivalenceclass associatedwith nouns is its basic category description. The FORMALquale distinguishes an object from a larger set. There are two possiblestructures associatedwith the FORMAL quale: (19) a. Simple Typing: Value of FORMALrole is identical to sortal typing of the argument; b. Complex (Dotted) Typing: Value of FORMALrole defines the relation between the arguments of different types. Concerning the first case, the typing of an argument for a nominal may exhaustively define the information contributed by the FORMALquale. In fact, for nouns denoting simple types (and not dotted-types), the FORMALis itself the typing restriction on the argument structure (i.e., the one referential argument). In such a case, the qualia structure has the following schematic form, minimally :9

96

Chapter6

0: ( 20 )

ARGSTR

=

QUALIA

By

way

of

illustration

distinguished tion two

by

is inherent sorts

[ ARGI

=

, the

gender in

within

= X ]

nouns

man

, a binary

the

the

a CONSTITUTIVE

= x :r ]

L FORMAL

compositional

type

make

human

distinction

and

woman

characterizing

, and

- up

hence

, utilizing

the

are

sorts

predicate of

this

the

of

. The

separation

feature

predicate

human

of

-

these

is represented

male

,

distinc

as

, shown

below

.

man ( 21 )

ARGSTR

=

QUALIA

Now

let

discussed

us in

such

as destruction

The

polysemy

following

each

other

schematic

consider

examination

.

The

.

For

a noun Quale

type the

of

the

qualia

a

overall structure

dotted

relational

are

classic the

cases type

the

nominal for

the

I ,

in

dotted are

complex

nominals

As door

polysemy

object

( or

. as

nominals

of logical of

arguments is the

such

such

alternating

a complex

how

types

nouns

- result

into denoting

defines

-

(22)

denoting

as process

directly

FORMAL

for

(x ) ]

objects

and encoded

]

, implicitly

, as well

manner

form

complex chapters

, window

is

e .g . , 71 . 72 , the

= x : human

FORMAL = x [ CaNST = male

=

previous

book , newspaper

[ ARGI

. the

) type

related type

is given

.

, to

The

below

:

X:'TI Y:'T2

Conformingto the behaviorof dotted types, in additionto the FORMAL definingthe relation betweenarguments , there must be referenceto the dotted argumentx.y in the qualiastructurefor this representation to be well-formed, e.g., R(e,w,x.y) . A goodexampleillustrating this point is the noun bookandthe relatedverb read. The activity of readingrequires not just that the complementbe informationor propositionalin nature, but that it is "readable " ; that is, it must havesomephysicalmanifes tation in order to be interpretedassymbolic. Furthermore , havingonly the property of physicalmanifestationis not sufficientfor somethingto

97

Qualia Structure

be readable; it must be something one can "read;" that is, it must be information . An alternative interpretation of the dotted object is also possible.l0 One might build the relation betweenthe types specifiedin the FORMAL role directly into the dot object itself. For example, the resulting type lattice would have the following structure :

a , B '"Y ~l ' a ' / '"~ ,.'Y R2 .R 1/.a.,8

Figure 6.3 Hierarchy Dotted Type with Relations Built into the Dot Object Judging between these alternatives seemsdifficult at this point , since the exact semantics of the resulting types is not yet well understood. In either case, the relations possible in the FORMALquale which define exactly how the types are related must be constrained in a way that does not give rise to unattested complex types in language. We return to this question in 8.3. 6.2.2

The Interpretation

of the AGENTIVEQuale

Knowledge of the "coming into being," in Aristotle 's terms (Physics 11) , of an object is encoded in the AGENTIVEquale of the lexical item associated with it . The manner in which something is created is a mode of explanation that will distinguish natural kinds from artifacts . How something comesabout is an important mode of explanation for distinguishing objects and events in the world . If the lexical form is a noun, the AGENTIVEis representedas an event predicate, where the object being defined is typically bound to the second argument of the relation. For a simple typed nominal, the schematic qualia structure is as follows:

98

Chapter 6

a

ARGSTR = [ ARG 1 = x:r ]

(23)

QUALIA =

AGENTIVE [FORMAL ==XR (e,y ,x ) ]

The representation in (23) above corresponds to the semantics of arti facts . For example , viewed from the constraint OfAGENTIVE, qua coming into being , objects such as cookies, cakes, and bread are typically baked . The process of baking , modulo such objects , is a creative activity , while relative to objects such as potatoes , carrots , and other natural kinds , it is simply a change of state predicate . As we discussed in previous chapters , to classify the verb bake as having both senses lexically specified is to miss the semantic generalization the language is expressing ; namely , that certain objects come into being by virtue of an activity which can otherwise simply change the internal state of an object . The sense is the same, but manifests itself with overall effects differently , resulting in the "sense in context " phenomenon . The explanation of what brings something about is of course not restricted to nouns , and within nouns certainly not to artifacts and simple typed objects . With the machinery of dotted types , we can provide a unique value for the AGENTIVE quale of a lexical item . The methodolog ical goal in GL is to associate individual qualia roles as unique values , avoiding both conjunction and disjunction , except when motivated by the semantics

of particular

lexical

items . As discussed

above , for a dot

object nominal , notice that the AGENTIVE may refer to the dotted ar-

gument directly (but cr. 8.3).11 Q

(24)

ARGSTR = ARG 1=X:7]1 ARG2

=

Y : 72

71 . 72-ICP

QUALIA =

6 .2 .3

The Interpretation

FORMAL= P (y,x ) AGENTIVE= R (e,w ,x .y ) of the CONSTITUTIVE Quale

The CONSTITUTIVE (henceforth CONST ) quale refers not only to the parts or material of an object , but defines , for an object , what that object is logically part of , if such a relation exists . The relation part _of allows for

both

abstractions

:

99

Qualia Structure

(25) a. Ax3y[part _oj (y, X)] b. Ax3y(part_of {x , y)] The function (25a) definesthe more conventional part-of-relation , while (25b) defines the relation between, for example, a hand and its body. The example of the noun hand below shows how it encodes what it is logically a part of : hand

(26)

ARGSTR = [ ARGI= x:limb]

CONST = = part QUALIA -- [FORMAL x _0 (x ,y :body ) ]

That is, the relation in the caNST allows for reference to what something is constituted

of as well as what

it constitutes

, in part ; i .e., a hand

is

part of a body, and a body has a hand (cf. Vikner and Hansen's 1994 suggestion of this interpretation 6 .2 .4

The Interpretation

for the CaNST role ) . of the TELIC Quale

The TELIC quale defines what the purpose or function of a concept is, if there

is such a constraint

associated

with

it .12

Rather

than

view -

ing the semantic functions associatedwith a lexical item (or cuncept) in terms of traditional

thematic role descriptions , I have argued that

lexical knowledge encodes the modes of explanation associated with a word . For this reason , there is no simple one-to -one mapping between f)-roles and qualia . For example , observe how the TELIC quale captures aspects

of several

different

f)-roles , but

cannot

be associated

with

any

one of them exhaustively . Modes

of

Telic

:

(i) Direct Telic: something which one acts on directly . a

ARGSTR = [ ARGl= x:r J QUALIA =

TELIC = R [FORMAL =(xe,y,x ) )

(ii ) Purpose Telic: something which is used for facilitating a particular activity

;

100

Chapter 6

a: ARGSTR =[ARG 1=X:T] QUALIA =[FORMAL TELlO =R =(e X,x,y)] Conforming to the first type above is the TELIC quale for a noun such as beer, where reference to the activity of drinking incorporates a variable for the noun as object of the predicate drink . The object 's purpose is the activity given in the TELIC role . beer ARGSTR=

(27)

QUALIA =

An

example

found as

with

with

of

the

objects

tools

as

}

= drink [ TELIC FORMAL = X ( e,y ,x ) ]

second that

, such

[ ARG1 = x :liquid

type

are the

of

used noun

TELIC

in

the

knife

.

use

, that

performance

of of

purpose an

telic

activity

, is

, such

knife ARGSTR

QUALIA

( 28 )

It

is ,

that sharing

=

of allow

course for causative

, some

=

just

[ ARG

1

=

x : tool

[ TELIC FORMAL

=

cut = x ( e ,x ,y ) ]

these

sorts

the

alternations

of structure

of

]

relations seen

between in

agents

types and

of

qualia

instrumentals

.

(29) a. The hammerbrokethe glass. b. Mary brokethe glasswith the hammer. (30) a. The knife cut the bread. b. John cut the breadwith the knife. Returning to the example of book from above, we can express the TELIC of a dotted type by direct referenceto the dotted argument directly ; read (PJWJX . y) . With this representationof the TELICfor dotted types and the AGENTIVEof bookgiven above, we can represent the complete qualia structure below:

101

Qualia Structure

ARGl

=

x : information

ARG2 = y :phys _obj

(31)

information

.phys _obj _lcp

FORMAL= hold (y,x ) TELIC = read (e,w ,x .y ) AGENT = write (e' ,v ,x .y ) The predicates read and write , on this view , select for objects of just the dotted type specified by the restrictions , and no others . This point is taken up again in chapter 8 when the semantics of nominals is studied more closely . 6 .2 .5

Mapping

from

Qualia

In this section , I examine briefly what the consequences of qualia - based representations are for how arguments are mapped to syntax . Our strat egy for projecting semantic arguments to syntax from underlying forms is very general , and would overgenerate unless constraints are operative in the grammar which block the expression of certain predicate -argument mappings . Given that event-headedness acts to foreground or "focus" a single quale of the verbal semantic representation , we will say that the abstracted quale that results from headedness must be saturated at s-structure . If one normally thinks of projection as specifying the appropriate grammatical functions to the arguments of a lexical item , such

as (32):13 (32) V (~, Y) --+ X:SUBJ, y:OBJ then it is clear what the task for a qualia - based representation is; namely to project from multiple semantic expressions to the appropriate gram matical

functions

in syntax . Whereas

the PAS strategy

accounts

for

how (32) is realized in syntax , projection in GL must make reference to qualia , thereby potentially complicating the lexicon to syntax mapping .

Within the approach taken here, the mapping illustrated in (32) is replaced with (33), which indicates the licensed projections from a particular quale , Qi :14

(33) a. Qi : R(el , x , y) b . Qj : P (e2, y )

~X:SUBJ, y:OBJ ~y :SUBJ

102

Chapter6

Given the presenceof more than one qualia role, individual qualia compete for projection , and mechanismssuch as headednessact as a filter to constrain the set of project able qualia. The headedevent, e* projects the configuration (or template) associatedwith that event's predicate (i .e., its Quale value). For example, from the qualia in (33), there are two possible mappings: (34) a. Qi : R(ei , x , y) ~ X:SUBJ, y:OBJ b. Qj : P (e2, y) ~shadowed

I.

(35) a. Qi : R(el , x , y) ~ shadowed b. Qj : P (e2' Y) ~y:SUBJ

:!

.

To demonstrate this principle , considerthe lexical representation for the verb kill .

(36)

Headednesswill determine that the argument associatedwith the second subevent, e2, cannot be expressed, since the qualia for the headedevent expressesthe template 88sociatedwith that relation. This is illustrated in (37) below.

(37)

QualiaStructure

103

For an unergative verb such as run, the projection to subject follows from the qualia structure , shown in (38).

(38)

In

addition

must

be

structure ( Cf

.

to

the

" saturated must

Chomsky

mapping "

be , 1981

by

fully , 1986

the

itself syntax

interpreted

, the .

That in

qualia is , the

the

of

a

lexical

variables

resulting

expression in

syntactic

the

qualia

structure

) .

(39) QUALIA SATURATION : A qualia structure is saturated only if all arguments in the qualia are covered. (40) COVERING : An argument x is coveredonly if : (i) x is linked to a position in s-structure; or (ii ) x is logically dependent on a covered argument y; or (iii ) x is existentially closed by virtue of its type . For the caseof verbs such as build, discussedin chapter 9 where they are referred to as constitutive causatives, an additional constraint is at play. Becausethere is a default argument in the semanticsof build (the material z used by x to bring about y), the binding is different than for a default causative relation. The qualia structure is repeated below in (41) in an abbreviated form.

(41)

build ARG2 = 0 CON = 0 ARGST = FORM phys [ a rtifa ] D A = ~ FOR RG = mas 1 r : l m ate QUALIA =[FORM exist (e ,0()e,0,0 2)ct AGEN build _ a create -I= cp ]l

104

With

Chapter 6

the initial

event headed , we would

expect

the default

argument ,

viz . the material to surface as the direct object , but this is not the case. Rather , the object which is constituted of this material is expressed as the object .

(42) a. QA : R (ei , x , f (y ))

b. Qp : P (e2' y)

~ X:SUBJ, y :OBJ

~ shadowed

Formally , such a default argument can be viewed as a Skolem function of

the argument it is dependent on, namely f (y). Thus, given the calculus of relations

in the qualia

and the templates

associated

with

them , the

FORMAL argument in this case ends up bound to the object position in syntax . It should

be pointed

out that the passive acts to head an event struc -

ture to a right -headed event . For a lexically left -headed event such as kill , this has the effect of shadowing the agent , and allowing expression of this argument only by adjunction . We return to this issue briefly in chapter 9 in the context of what licenses adjunction , but for further details of this proposal , see Pustejovsky (forthcoming ) . As mentioned earlier , unergative and unaccusative verbs are distin guished by their qualia structures , but unergatives themselves are not a homogeneous class by any means. The lexical distinction between agentlike and non- agent-like processes was distinguished in chapter 5 in terms of the qualia being made reference to ; namely , agent-like processes such as run are associated

with

the AGENTIVE quale while

processes such as

sleep are lexically associated with the FORMAL role . This is important for determining the manner in which adj uncts may co-compose with the verb to derive non-lexical senses, such as run to the store vs. sleep a restful sleep.

7

Generative

Mechanisms

in Semantics

In this chapter , we finally turn to the consequences of adopting a generative lexical approach to semantics , and the subsequent effect this has on the grammar in general . In the previous two chapters , we outlined the overall structure of the typing system incorporating the four levels of representation , as well as a more detailed exa,mination of the semantics of qualia . Our goal in this chapter is to see what machinery is necessary in order to make use of the representations presented , i .e., event structure , qualia structure , and an argument structure with dotted types , so that the criteria of descriptive and explanatory adequacy can be satisfied for our semantic

theory . What

we hope to achieve is a model of meaning

in

language that captures the means by which words can assume a poten tially infinite number of senses in context , while limiting the number of senses actually stored in the lexicon . The mechanism responsible for this polymorphic behavior of language is a set of generative devices connect ing the different levels of lexical semantics , providing for the composi tional interpretation of words in context . As mentioned in 5.1, included in these generative operations are the semantic transformations of type coercion , selective binding , and co-composition . Among co-compositional operations are transformations making use of semantically underspeci fied forms , such as manner co-composition , feature transcription , and light verb specification . All of these can be viewed as well -formedness conditions

on type combinations

in the grammar

.

A major consequence of this approach is that the isomorphism between syntactic and semantic categories cannot be maintained for all levels of linguistic description , nor is it desirable . What this means is that a syntactic phrase cannot be interpreted outside of the syntactic and semantic context within which it appears . Rather , only by embedding the phrase can the appropriate denotation be determined . There are, however , canonical syntactic expressions for realizing semantic types , and these are presented in 7.4 below . I first review the role played by type shifting in the literature , and how this tradition relates to the view of type coercion argued for in

Pustejovsky (1991a, 1993) and elaborated here. Then, I outline the mechanisms

of coercion

and how

we can make

use of the

richer

semantic

representations presented here to explain polysemy phenomena . The operations of selective binding and co-composition are discussed in 7.2 and 7.3. Finally , the study of how syntactic categories relate to semantic

106

Chapter7

types is taken up in section 7.4 , where the notion of lexical conceptual paradigm is related to canonical syntactic forms.

7.1

Coercion and Type Shifting

Type shifting was first introduced as a way of allowing operators such as negation and conjunction to change type according to what they modified or took as arguments (cf. Geach, 1968, Strachey, 1967). Subsequent work in Rooth and Partee (1982), Partee and Rooth (1983), Partee (1985), Klein and Sag (1985), and Chierchia (1984) developed the mechanismto allow an NP, or any expression, in general, to change its type (and hence its denotation) depending on the context. Briefly, we can imagine an expressionbeing assigneda default typing , where the model defineswhat the well-defined and undefinedtype shirtings allowed for that expressionare. The types for an expressionare related by what Partee and Rooth (1985) call a type ladder. The utility of this proposal is that it allows us to maintain a compositional semanticswhile also accounting for the different manifestations of an expressionin a principled way. 7 .1.1

Parametric

Polymorphism

and Type Shifting

The example of conjunction studied in Partee and Rooth (1985) is perhaps the most obvious type shifting phenomenonin natural language, and was independently classified as a parametric polymorphism in Strachey (1967). LeiB (1991) and Shieber (1992) treat conjunction operators as operators taking any type as argument. For this reason, such lexical items are said to be parametric polymorphic. For example, in English, and is generally viewed as being able to conjoin almost any category at any bar level, Xi . Although there are very few types of lexical items which exhibit this behavior, it is helpful to see how this mechanism works. The classic example illustrating type shifting within this operation is given in (1) , where an individual , John, of type e is coordinated with a quantified NP. (1) JQhn and every woman arrived. The NP every woman is of type e, t >, t >, and according to standard typing restrictions on coordination within this approach, the conjuncts

Generative Mechanisms in Semantics

107

must be of like type , which in this casethey are not. Under this strategy, the sentencewould be judged ill -formed even though it is fully grammatical. The solution to this difficulty is to lift the interpretation of John to that of a generalizedquantifier, namely "\P [P (j )], as shown in (2). (2) a. [John]e and [every woman] e,t >,t > : b. every woman E e, t >, t >: APVx(woman(x) -t P (x)] c. John E e type-shifts to e, t >t >: AP[P (j )] Type shifting has been useful for bringing together two distinct approaches to the semantics of interrogatives, as proposed recently by Groenendijk and Stokhof (1989). Other applications have also been proposed, and are explored in Partee (1985) and Dowty (1988).1 In Pustejovsky (1993) a particular application of type-shifting is discussed, illustrated in (3) below. (3) John considersMary ~jQQ} . In (3), the type of the NP a fool is changedto that of a predicate, <e , t >. Thus, Mary (of type e) and the predicate can combine in the standard fashion. Following a suggestion in Partee (1985), we can represent the verb consider as selecting for an argument of a particular type, namely a predicative phrase.2 (4) a. John considersMary ~_fulli . b. a fool: AP3x (fool (x) 1\ P (x)], c. Mary : E e

e, t > ,t > ~

<e , t >

The type that consider selectsfor by default is <e , t >, as seenwith the infinitival complement in (5), and it is exactly this type of complement selection which illustrates type coercion, to which I return below.3 (5) John considers [MarY]e [to be a fool] <ett > 0 The attraction of type-shifting is that it will provide us with one of the tools neededfor expressingthe semantics-to-syntax mapping in the grammar; namely, it can capture the semantic relatednessbetween systematically ambiguouslexical items. For example, Klein and Sag (1985) show how the "raised" and "non-raised" forms with believe(cf. (6)), and Equi and non-Equi sentenceswith prefer (cf. (7)) can be accounted for (cf. also Gazdar, 1982). Their analysis in effect provides one solution to the polymorphic syntactic nature of these verbs.

108

Chapter7

(6) a. John believes Marytobehonest . b. John believes thatMaryishonest . (7) a. Maryprefers toprogram inCommonlisp . b. Maryprefers forherstudents toprogram inCommonlisp -. They suggest lexical entries for believe and prefer as shown in (8) and (9) , where IR and IE are type -shifting operators for Raising and Equi , respectively . (8) a. believe E <S J
(9)

>

a. prefer E <S,
Lexically associated with each of these operators is a meaning postulate , ensuring the proper predication :4

(10) VVVPl...PnD[!R(()(V)(Pl) . . . (Pn) .(-t ( (V(Pl)) . . . (Pn))

(11 )VVVP1 [!E(()(V)(Pl)...(Pn )~Pl{AX [((V(X*))(X*)(P2 ) ...(Pn )... ]}]PnD What these rules do is to ensure the appropriate predication of the lower predicate to the proper NP, the binding that is accomplishedfor example, by Binding Theory in Chomsky (1981, 1986). The operators in (8) and (9) permit the appropriate syntactic context to be licensed, but the resulting compositions require the meaning postulates in (10) and (11) for semantic well-formedness.5 In other words, the type-shifting operator performs a destructive operation which the meaning postulate essentially acts to correct or undo. A more conservative solution is suggestedin Dowty (1985), where each particular grammatical usage of a verb is recorded in a separate lexical entry, what we characterized as senseenumeration in chapter 3 above. On this view, the relatednessbetweenwords is also captured via meaning postulates, and there is no single, "deep" type for a verb.6 These casesinvolving the verb considerillustrate that the type shifting of a phrase is often "licensed" by a particular lexical item . In this case, the governing verb consider licensesthe shift in the second NP to be

Generative Mechanisms in Semantics

109

interpreted as a predicate. In just such cases, we can think of a verb coercingits argument to assumea certain type. In the next section, we will try to make this proposal more concretein the hope of systematically accounting for why certain verbs allow the logical polysemy that they do. As stated above, one of the most seriousproblems in lexical semantics is accounting for the systematic ambiguity of lexical items. It should be a general goal of semantic researchto , whenever possible, reduce the amount of lexical ambiguity that the grammar requires. With this in mind , there are severalconstructions that have been treated as ambiguous verbs, but in fact lend themselveseasily to a type shifting explanation . The first is a case discussedin McCawley (1979) and Dowty (1979) concerning the complement types of the verb want, as illustrated in (12): (12) a. John wants to have a car until next week. b. John wants a car until next week. As Dowty points out, the temporal adverbial until next weekmodifies a hidden or understood predicate in (12b), just as it modifies the overt predicate in (12a). To relate these word senses , Dowty suggeststhe null hypothesis, namely that the verb want is simply ambiguous, and has the following entries, related by meaning postulates. (13) a. want! E <S,
E
c. want3 E
Similarly , Dowty (1985) argues in favor of the senseenumeration approach to these cases, partly because there are subtle differences in the verb meaning and the interpretation of the ellipsed predicate in each case, which must be accounted for somewhere , and presumably in the lexicon .7 Consider for example , the context dependence of the ellipsed predicate in each example below .

(14) a. John wants ~ Qeer. (to drink ) b. Mary wants a bOQk. (to read) c. Harry wants another cigarette. (to smoke)

110

Chapter 7

No generaltype shifting operator w'ould give us the appropriately specific readings required for eachof theseexamples. Dowty 's solution is to have meaning postulates relate the major word sensesfor a verb such as want, and then allow pragmatic factors to supply the contextual information that embeds the verb sensein context. This is a reasonableapproach given the limitations within conventional Montague Grammar, where information from the complement can in no way contribute to the overall composition of the phrase. There is something missing in this solution, however, that is similar to the problems we encounteredin previous chapters with the contextualization of sensesfor verbs such as enjoy, begin, and finish . Namely, the senseenumeration necessaryto account for the behavior of these verbs is systematically associatedwith the complements as much as they are with constraints from the verb itself. The other problem with this solution is that the only way to establish the relation betweenverb sensesis by meaning postulates. Although descriptively adequate, the meaning postulate solution for relating the different sensesof want, for example, seemsarbitrarily powerful and unconstrained. Similar remarks hold for the verb type-shifting solution mentioned above. Namely, that although type shifting does allow various complement structures to be accepted by the same verb, what the operations are in fact doing is to create new subcategorization frames or semantic categories for each of the different complement types. There are two major problems with this approach. First , in most of the cases, it is not the meaning of the verb which is changing, but rather the selectional properties on the verb's complement. Once again, the association in meaning is done by a meaning postulate and not with any more restricted mechanism. The secondproblem is that it fails to capture the polysemous behavior of these complementswhen they appear in other syntactic environments. That is, the nature of these complementssuch as a beerand a bookis such that they exhibit systematic polysemiesthat are independent of the verb want. Given these remarks, we will adopt a different strategy in what follows. Consider again the full range of complementation for the verb want. (15) a. Mary wants John to leave. (8 [+ INF]) b. Mary wants to leave. (vp [+ INFJ) c. Mary wants a beer. (NP)

Generative Mechanisms in Semantics

111

Rather than type shifting the verb to accommodateeach syntactic environment , let us assumethat the type of the verb remains the same; that is, it is monomorphic with respect to semantic selection. What changes, however, is the syntactic type of the complement to the verb, which undergoesa type shifting operation by virtue of lexical governance from the verb. Such an operation I will call type coercion, becauseit is lexically governed type shifting . Following Cardelli and Wegner (1985) and Pustejovsky (1993), I will define coercion as follows: (16) TYPE COERCION : a semantic operation that converts an argument to the type which is expected by a function , where it would otherwise result in a type error. As mentioned above, Partee and Rooth (1982) suggestthat all expressions in the language can be assigned a base type, while also being associatedwith a specific type ladder. Pustejovsky (1993) extends this proposal and suggeststhat each expressiona may have available to it , a set of shifting operators, which we call ~ a, which may operate over an expression, changing its type and denotation. The rules of function application and composition may make reference to these operators directly, allowing us to treat the function as behaving polymorphically , while the argument is actually the shifting type. Hence, we can give an initial formulation of the rule of function application incorporating type coercion as follows:8 (17)

FUNCTIONApPLICATIONWITH COERCION(FAC) : If a is of type c, and (3 is of type , then, (i) if type c = a, then (3(0;') is of type b. (ii ) if there is a 0- E Ea such that 0-(0;') results in an expressionof type a, then (3(o-(a )) is of type b. (iii ) otherwise a type error is produced.

To illustrate informally the effects of this rule, let us return to the ambiguity of want shown in (15) and (14) above. There are two things to explain: first , the different syntactic environments that are available as the complementtypes to the verb want; and secondly, the different interpretations that arise for the NP complement examples, which seems to require an enumeration of sensesalong with pragmat~.c enrichment for these cases. Rather than proposing different semantic types for the

112

Chapter7

verb, i .e., separate and distinct lexical entries, let us propose that the verb's type remains constant, where the internal argument is uniformly typed as a proposition ; if the syntactic form appearing in complement position matches this type, then the resulting structure is well-formed. If , however, the appropriate type is not present in complement position, it is coercedby the verb to match the type required by the typing restrictions on the verb. The tree in (18) below illustrates the relation between the "deep semantic type" and the syntactic realizations.

(18)

S[+INF] The mapping in (77) illustrates the ability of one semantic type to express itself in many syntactically distinct ways. From the perspective of syntactic categories, what this says is that a syntactic expression does not denote a single semantic type. Rather, it assumesa particular semantic type by virtue of context.9 What I am suggestingis that the syntactic forms associatedwith the argument of want, which appears to be a propositional type of some sort , are not uniquely realizable as these three syntactic forms for want, but are expressible by virtue of the semantic type itself. The methodology here is a fairly classic one in generative linguistics, arguing for the underlying properties of an expressionon the basis of its distribu tional behavior. What we would expect, then is that something of like type would appear mutatis mutandi in different contexts with the same expressiveabilities . In the next section I formalize the operation of type coercion and show how it is an integral component to the grammar. It is but one part of the semantic system that allows us to capture the semantic creativity and senseextension phenomena discussedabove. Only together with a richer representation system as emb'odied in the levels of argument, event, and qualia structure , can coercion appropriately apply to give rise to creative senses. In later sections, I also discusstwo other generative

113

Generative Mechanisms in Semantics

devicesrequiredfor adequatelymodelingthe polymorphicbehaviorof language . In sum, rather than assigninga newlexicalentry to a verbeverytime a differentsyntacticenvironmentfor it is discovered or everytime a new interpretationis needfor a new context, we will "spreadthe semantic load" more evenlythroughout the elementsin composition . This is accomplished by the representations as well as the ability to makeuse of them through the generativeoperationsdiscussed below. 7.1.2

Subtype Coercion

To demonstratethe formalpropertiesof type coercion , it is usefulto first examinewhat is perhapsthe simplestcaseof coercion, one involving a very specifickind of semanticshifting, namely the caseof subtype coercion . The formal propertiesaccompanying types and subtypesare well known in the semanticsand knowledgerepresentationliterature. Consider, for example,the sentences in (19) below, whereboth subject andobjectNPsaresubtypesof the sortalspecifications to the arguments of the verb. (19) a. ~i ~~l drivesa Hondato work. b. Thm readthe Tractatuson holiday. Althoughthis is a trivial point fromthe viewpointof syntax, for a seman tics with types, onemust establisha relation betweenthe type denoted by the NP in eachof theseargumentpositionsand the type that is formally selectedfor by the verbsdriveand read. The conventionalrelation givenbetweenthesetypesis oneof subtyping. Hence,within our typing system, we needto ensurethat if a function selectsfor type 71 and the actually occurringform is 72, where72 is a subtypeof 71 (72 ~ 71), it too shouldbe acceptedby the function as a legitimateargument. For example, assumingthat the lexical representationfor the noun car is givenasin (20), car ARGSTR= ARG

(20)

QUALIA =

TELlO

AGENTIVE FORMAL

1

=

x

=

: vehicle

]

drive

(

= =

create X

e

, y

, x

)

(

e

,

z

, x

) ]

then by typing Honda as a subtype of car , we establish the following relation : Honda~ car ~ vehicle . As with other lexical inheritance mech-

114

Chapter7

anisms , the more specific below supersedes

value for the AGENTIVE role in the structure

the more general value associated

with

an artifact

in

car , while still inheriting the values for the other qualia (cf . Beierle et at ., 1992 and Copestake et at ., 1993 ) . Note that AGENTIVE and TELlC val ues are both inherited ; the type specificity however , is locally defined .

of the AGENTIVE for Honda ,

Honda ARGSTR =

ARGI = x :car ]

QUALIA =

TELIC = drive ( e,y ,x ) AGENTIVE FORMAL = =x create ( e,Honda - Co ,x ) ]

(21 )

Assuming that the internal type selected by the verb in sentence ( 19a ) is vehicle , as illustrated below in the lexical representation for drive , then the selectional subtyping

requirements

relation

type of the actual

can be satisfied just in case there exists the

mentioned object

above , which

to the lexically

acts to formally

specified

relate

the

type .

(22)

Recall with

from

chapter

6 that

respect to a grammar

a type , r , such that

we defined

a typing judgment , 9 ~ Q' : T ,

to be an assignment , g , an expression , Q' , and

under

assignment

g , the expression

Q' has type T .

We can define a sub typing coercion relation , 8 , for these judgments follows ( cf . Gunter , 1992 ) : (23 )

a : 0"1, 8 [0"1 :::; 0"2] : 0"1 --t- ~ ~ 8 [0"1 :::; 0"2] (a ) : 0"2

This says that , given an expression a2 , there

as

is a coercion

possible

type of a in this composition The typing relation between by the governing shown below :

a of type aI , which

between

is a subtype

0"1 and 0"2, which

of

changes the

, from 0"1 to 0' 2 (cf . Beierle et at., 1992 ) . the subtype Honda and the type selected

verb drive is respected

by the coercion

relation , 8 , as

Generative

Mechanisms

( 24 )

in

a . 8 [Honda b . 8 [car

Similarly tion

the type

the

relation

, let

sented

in the

typing

polymorphism

that

in

coercion

section

the

a single

follows

complement

7 .1 .3

True

into

( 25 )

( 26 )

( 27 )

to

multiple qualia

, true

another

type

specified

shift

is not by the

wants

a beer .

wants

a cigarette

a . Mary

enjoyed

the

b . Mary

enjoyed

watching

proper

movie

a book . reading

a book .

c . John

began

to read

a book .

describes structure

.

to

, be

Subtype

lattice

, while

types

lattices

,

of the

rule

relatedness complement

to ensure

that

cases , regardless application just

how the

such

involves

embeds

coercion

the

strict

by lexical

gover -

the

operation

existing .

Let

us

7 . 1 : 10

. the

began

a lexical

sub -

.

movie

semantic

, but

in section

began

similarity

coercion

arbitrary

a . John

of function

pre -

.

type , licensed

b . John

ous section For

to

a . Mary

rule

type

the

cases

item

different

through

b . Mary

in all these

quite a single

the

are similar

of a lexical

reference

mentioned

a coercion

in

.

with

, although

below

available

type

to invoke

that

requires

resulting

the

clear

a rela -

individual

coercion

coercions

are formally

paradigm

as the

The

become

defines

coercion

complement

available

the

to capture

of type

functionality

to the

( as well

:'::; text

Coercion

, the

In order

satisfied

variable

polymorphism

of one type

return

to the

complement

Complement

nance . Furthermore type

mechanism

. It will

the

coercion

use of information

parametric

:.::; book

read and the actual

inferences

making

shifting

Tractatus

form , they

the

vehicle

by the verb

general

and

permit

- t car

~

us return

previous

they

expressed

Unlike

] : car

selected

illustrated

subtyping

true

~ car ] : Honda

~ vehicle

, in ( 19b ) above , The

between

Having

in

115

Semantics

as that

the

.

of these

semantic

of syntactic

with

coercion

semantic associated

different

denotations

verb

) , we will

type

of the

forms need verb

is

form . ! ! ( FAC ) given

transformation with

the

in the previ comes

verb

about

begin , given

.

116

Chapter7

below in a somewhat incomplete form (cf. 9.3 for details on aspectual coercion and control), the typing on the second argument is explicitly given as an event.

(28)

This states that the complement to beginis actually an event of some sort (discussedin 9.3 and Pustejovsky and Bouillon , 1995). Regardless of the surface syntactic form of the complement, the semantic typing environment is the same, namely an event . Where that type is not directly satisfied, as in (27c) or (27b), a coercion applies to reconstruct the semanticsof the complement. The coercion is, of course, successfulonly if the NP has availableto it an alias of the appropriate type (Pustejovsky and Boguraev, 1993). An alias can be thought of as an alternative type that is available to a element, be it lexical or phrasal. For a sentence such as (27a), the event type is forced on the complement a book, and comes about by reconstructing an event reading from the qualia of the NP. Recall that the lexical structure for book is a dot object , making referenceto two types, info and physobj .12

(29)

ARG2 = y :physobj [ ARGI = x :info ] info .p hyso b j _lcp FORMAL= hold (y ,x ) TELIC = read (e,w ,x .y ) AGENT = write (e' ,v ,x .y )

Becausethe NP a bookdoesnot satisfy the type required by the predicate begin, the verb coerces the NP into an event denotation, one which is available from the NP's qualia structure through qualia projection (see below 7.4). There are two event readings associatedwith this NP, namely the values of the AGENTIVEand TELIC qualia roles. How an

Generative Mechanisms in Semantics

117

event is actually reconstructed to satisfy the typing environment is a result of the coercion operation itself. Consider the tree representation below, where semantic types are notated in brackets. According to the lexical structure in (28) , the verb beginhas two arguments, [human] and [ event ] , and can be viewed as occupying a minimal syntactic projection for the verb, independent of any particular syntactic expression. That is, imagine for now the verb projecting semantic selectional information alone (cf. Chomsky, 1986).

(30)

S / / / / / / ~" "" "

"

[ human ]

VP / / / / / """ ~" " V

a

' " [ event

begin

]

(3

The statement of coercion on the complement from the governing verb begin can be seen as requiring it to recover an event-denoting expression, and then embed the NP semanticswithin this expression. This is illustrated schematically in (31) below.

(31)

S / / / / / "--'-" ""

'"

[ human ]

I a

VP

/ /"""""'~""-""""'" V

[ event ]

AXAe[read(e,x ,a_book)]

begin

abbok

Ignoring issueof subject control and binding for now, what this demonstrates is that the semantic typing of an environment can be respected without changing the syntax of the expression. It should be pointed out

118

Chapter 7

immediately that this does not operate without constraints, since the expression being coerced itself projects the appropriate semantic type required by the application rule. To see more clearly how such coercion interpretations are formally derived, consider the complement variations possible with the propositional attitude verb believe, given in (32) . (32) a. b. c. d.

Mary Mary Mary Mary

believesthat he left . (s [+ TNS]) believeshim to have left . (s [+ INF]) believesthe book. (NP) believesJohn. (NP)

Assuming that the verb believeselects for a propositional type, prop , as its complement, the tensed sentential complement in (32a) can be viewed as direct satisfaction of this selectedtype. Similar remarks hold for (32b), ignoring for this discussion, the exceptional casemarking on the complement subject him (cf. Chomsky, 1981). Now notice the interpretation of the complement in (32c). Although a definite NP such as the bookis formally a generalizedquantifier, with ~ype e Jt >, t , the interpretation it assumesin this position is clearly that of a proposition. We saw in (34) above that there are at least two possible event interpretations associatedwith a noun such as book(and subsequentlywith the NP containing it ), but in fact there is neither an event reading for (32c) nor is there a "reconstructed proposition" from the qualia structure of the NP. How, then, doesthis reading comeabout? Recall from our discussionin chapter 6 and the qualia structure in (34) above that the semantics of the bookis more complex than standardly assumedfor a nominal term phrase. Namely, the noun bookis a complex (or dot) object, and its type is info .physobj . Recall that this is the result of the type constructor lcp repeated here applying to the two types info and physobj : (33)

a : info

a : physobj

lcp(a ) : info . physobj

The dot object is the logical pairing of the sensesdenoted by the individ ual types in the complex type. That is, following the treatment taken for relational nominals outlined in Pustejovsky (1989) and in Pustejovsky (1994), we assumethat nominals such as bookare a sort of container ,

Generative Mechanisms in Semantics

119

which are further specified as a relation between a physical object and the

textual

information

contained

within

it .

Along with the complex type , we need to define the operations over this object which effectively foreground a particular sense. The coercion operations projecting one type from the complex type are a special case

of type pumping (or projection), and can be simply defined as El and ~ 2 below . These two operations , together with the dot object itself will form the definition of the type cluster we called an lcp in the previous chapter .

(34) lcp = { 0"1 . 0"2, E1[0"1 . 0"2] : 0' 1, E2[0' 1 . 0'2] : 0' 2} Hence , the lcp for book provides for the following aliases, licensing the polysemous behavior of the lexical item demonstrated previously . Assuming

the operators

El

and E2 to be defined

as above , then the full

type cluster , first mentioned in 6.2, can be given as in (35c) , with the appropriate derivation .

(35) a. ~ 1[info 'physobj ] :info b. E2[info .physobj ]:physobj c. info .physobj _lcp = { info .physobj , info , physobj } Intuitively , then , a book always denotes , in part , textual information . Because of the subtyping relations between these two types and the type

proposition

(prop ):

book ~ text

~ prop

the NP the book is able to "stand in " for a full propositional expression . This is illustrated in the tree structure below , once again showing semantic

selection

:

(36)

the

120

Chapter 7

Given this discussion, we return to sentence (32c) and the coercion involving the NP, the book. The type derivation below shows both the application of the type pumping operator, E, and the subtyping relation , 8 . The lcp interpretation of book, and there exists a subtyping relation O :' ~ S' . 37 (

)

The

El

(

info

associated

(

38

.

physobj

)

semantic

)

a

b

c

d

.

Mary

:

info

,

6

[ info

interpretation

believes

the

.

believe

(

"' 8

.

believe

'

(

A8

.

believe

'

(

" ' the

(

E1

(

(

book

the

the

-

-

-

is

book

:

given

5

prop

]

below

:

info

in

-

(

38

)

t

prop

.

.

book

)

book

:

:

)

)

(

info

prop

Mary

)

)

(

Mary

)

)

(

Mary

=

)

>

= > -

)

What the composition in (38) illustrates is how the deep typing on the complement is satisfied by enriching the semanticsof the object description as well as adding functionality to the manner in which phrases compose (cf. Pustejovsky, 1994). Finally , consider how the typing requirement for a verb such as sell, selecting for a complement of type ind , is satisfied by the lcp of book(cf. (39)). (39) Mary sold the book to John .

In sentence (39), the type physobj is available by application of the coercion operator E2, and likewise, the type ind is available through subtyping with the coercion operator 8 : i.e., physobj :,0::; ind .13 (40) a. Mary sold the book to John. b. sell (John )(e (~ 2(the -book )))(Mary ) =} c. sell (John ) (e (the - book : physobj )) (Mary ) =} d. sell (John )(the -book : ind ) (Mary ) Now let us return to the more difficult caseof coercion involving the complement of believe, namely (32d) above, repeated below. (41) Mary believes John . (NP)

121

Generative Mechanisms in Semantics

Unlike the case of an NP such as the book, where the type required by selection was part of the underlying semanticsof the expression, this is not the casein (41). What is similar , however, is the interpretation of the complement; namely, it refers to a proposition. Hence, under this view, there is coercion of an individual , John, to a propositional interpretation . In terms of typing , the shift simply involvestaking the type of an NP like John (i .e., a generalized quantifier, ind , prop > , prop and coercing it to a proposition (prop ).

(42) Type-shift:

ind , prop>,prop>

=>

prop

By itself, of course , this is both uninteresting and unintuitive . It is onething to simplyprovidethe appropriate type shiftingmechanisms to satisfythe typerequirements of a complement , suchasin (32c) and (32d). Thetask, however , is to explainwhysucha typeis available as an interpretationto beginwith. That is, why shouldJohnbe ableto metonymically projectto a propositional denotation , andindeed , what wouldthismean ? Intuitively,weunderstand thissentence to meanthat, whatJohnutters, typicallyor onthis oneoccasion , is believed by Mary. Thus, thereis an implicitrelationbetween Johnandthe propositionc/J expressed by him, conveyed by somemeansof communication R, andit is this proposition whichis beingjudged. Therelationhereis illustrated with an individualtype, indo (43) john ~ 33R[c/> 1\ [R(john,<jJ)])] The propositionct>is not simplyany proposition , but is construedas that whichis communicated by John. Our knowledge of humanagents as possiblespeakers or writersallowsus to infer (semantically ) that sucha relationasR in (43) exists, andthat it is sortallyrestrictedto communicative acts. Furtherdetails , however , on whatthis relationis seems out of the domainof linguisticknowledge , andproperlypart of defaultreasoning andabduction (cf. Lascarides andAsher,1993 , Hobbs , 1987 , and Hobbset al., 1988 ). Because the interpretationfrom the compositional semantics is conservative - by not completely specifying whatthe relationbetween thetwotypespropandind is- I will referto this particularapplication of typechange asminimalcoercion . That is, the typingenvironment is satisfied , but the semantics is not completely determined .14 Theassociated semantic derivationis below .15

122

Chapter7

( 44 )

As

a .

Mary believe

( Ap2 ( John

believe

( A3cjJ3R

[ / \

[ R ( John

, rjJ ) ] ) ] ) ( Mary

d .

believe

( A3 <j >3R

[ <j > A

[ R ( John

, cjJ) ] ] ) ( Mary

before

nicative

relation

information

,

but

say

radio

limits to

in

of

chapter

Co

In

section

we

begin

by

sense

( 45 )

( 47 )

( 48 )

Safire

is

.

For

most

example

likely

to not

lexical

inferences

,

an

-

, be -

you

is

how

with

commu semantic

that

illustrates also

of

, lexical

factors

example

only

type

something

, but

a

, the

verb

creation

bake

sense

. Q~ ~

. d the

potato

. d the

cake

verbs

in

heard believe

only

the

semantics issue

we

is return

.

Normally entries

are

Mary

~ !E ~i

the

table

b .

Mary

~ ! . E~ i

the

table

on have

function

verbs

a

application

to

meanings

dry

John

hammered

the

metal

John

hammered

the

metal

illustrate

, both

a

. this

change

of

) .

a .

Mary

.

between

a

presence

of

enter , 1995

.

. flat

the to

.

a .

car

one

al . , 1988

ambiguous

Rappaport

b .

the

two et

would and

a .

waxed

baking

, depending

Levin

involving describes

.

, lexicons ( cf .

of has

polysemy

, co - composition

.

( 46 ) - ( 49 )

reading

ballogical

than

( Atkins

Q~ ~

transition

ver

, more

polysemy

John

a

of

) . Briefly

, superficially the

below

cases , 1991a

studying

and

, the

( 46 )

those

John

lexical

other

William

a .

adjectival rate

possible

what

, however

believe

this

b .

and

not

commonsense

discuss

allows

( 45 )

Similarly ing

by to

knowledge

( Pustejovsky which

. In

supplied

What

linguistic

is

)

is specify

=?-

- composition

structure

state

.

subsequent

co - composition

point

This

)

10 .

7 .2

We

further

as .

believing

wrote

=?-

interpretation to

probably

, while

license

this

is is

he

)

complete

defined

rather

that

natural

is

Schorr

on

something

a

) ) ( Mary

operating R

Daniel

able

,

inferences

lieving

.

c .

mentioned

him

John

b .

subsequent

to

believes

.

both ) .

process a

forms

read

-

sepa

-

resultative as

123

Generative Mechanisms in Semantics

b. Mary ~~ ~ the car clean.

(49) a. Mary ~ . b. Mary ~ herselfsick. In order to capture the logical polysemy in all these casesand obviate the need for multiple listings of words, Pustejovsky (1991a) proposed that the complements carry information which acts on the governing verb, essentially taking the verb as argument and shifting its event type. Here we will make this proposal more explicit and describe what mechanism makes such an operation available. Let us assumethat the lexical structure for a verb such as bakeis that given below.

(50)

.I

I wish to claim that there is only one sensefor bake, and that any other readings are derived through generativemechanismsin composition with its arguments. What needsto be explained, however, is why nouns such as cake, bread and cookie "shift" the meaning of the verb bake, while other nouns (such as potato and garlic) do not. Intuitively , we would like to capture the fact that the former objects are prototypically brought about by the activity they are in composition with , something that the qualia structure should be able to express. Assume that the qualia for cakemakes referenceto an AGENTIVEvalue of ;\y;\e3x[bake(e, y, x)]. cake

ARGSTR ( 51

=

)

[

D - ARGl1 ARG

QUALIA

=

TELIC AGENTIVE

y : mass _ ind x= : food

=

CaNST FORMAL

= = =

]

y x

eat

( e2 =

bake

,z ,x

)

_ act

( el

,w

,y )

Notice that the AGENTIVEmakes referenceto the very process within which it is embeddedin this phrase, a relation which Pustejovsky (1991a) called cospecification.

124

Chapter 7

The

semantics

First

for

ment

its

The

governing

(B )

The

complement

The

7 and

the

verb

VP

The

derived

The

conditions

below

also

Keenan

by

then

a

we

, we

is

plements by

can

structure

for

the

:

in

a

derived

sense

AGENTIVE

becomes

the

, 1985

roles

FORMAL

of

matcha - ~- - - -

role

for

the

word

is

polysemous

not part verb

the

by

The

the

of

{ 3 , of

[ Qi

==

"Y ] ]

.

in

of

and

( 52 )

.

QSaC

Rather of

polysemous )

associated

Q Sa

/ 3) =

the

a

to with

by

the

Q S /3,

/ 3 . 16

putting , in

virtue

this

such of of

certain basic

"Y

qualia

sense

by

because add

=

and

that

,

[ Qi

QSanQS

bake

-

value

two

creation

cake

struc

. . .

suggests ,

two

Quale

these

like

view

meaning

verb

of

For

qualia a

[ QS {3

of

verbs

. This

:

is

bound

b with

senses

the

unification stated

a , with there

unification

type

appears

representation

type , if

lower

NP

are

UNIFICATION

, then

qualia

, a ( , a ) is

qualia

) :

greatest

co - specifying .

:

the

unique

call apply

QUALIA

. . .

on

The

( i . e . , those

[ QSa

weight

in .

I will can

< a , b > , and

both

itself

operation

WITH

define

derive

artifact

structure

verb

;

results

/ 3 , respectively ( 3,

the

semantic

co - composition

lowing

the

;

operation

Faltz

Further

contributed an

and can

.

verb

an

type

QS

/ 3 , as

structures

bake

-

unification in

include

complement

complement quale

APPLICATION

QSa

QSanQS

being

verb

argu

) .

qualia

These

.

the

feature

cake

a .

into

AGENTIVE

( the in

its

of

for

QA

structures

this

and

, a , of

, the

the

from

and

cases

=

to

and

which

expressions

the

applies

qualia

tures

of

)

operations

object

type

constituents

verbal

results under

shared

Thus

, a values

results

FORMAL

FUNCTION

]] ,

( bake

several

the

.

sense

( cf .

( 52 )

of the

from

binds

qualia

co - specifies

, where

results

Secondly

both

bake

complement

entire

. of

is , QA

of

composition verb

cake

co - composition

aspects

(A )

a

bake

identity

That

of

reflects

the

bake

application

verb

the .

operation

that

some

the

by

argument

The VP

of

, licensed

and

VP

function

structure

occurs

(C )

the

, conventional

it

com

-

meaning has

the

fol

-

Generative Mechanisms in Semantics

125

(53)

The result of co-composition is a semantic representation at the VP level that is identical in structure to the lexical form for a creation verb such as build. What this claims is that the creation senseof bakeembedsthe change-of-state reading within it by systematic rules of composition.17 The sensearises not by lexical enumeration, but generativelyin the semantics itself. As another brief example of how co-composition operates to give rise to "derived" senses, consider the lexical conflation examples first discussed in Talmy (1985), and characterized as lexical subordination in Levin and Rapoport (1988); namely, the systematic polysemy of the verb float and related verbs in the following construction: (54) a. The bottle is floating in the river. b. The bottle floated under the bridge. There is a systematic polysemy exhibited by these examples, where a processinterpretation for the verb float, as in (54a), is shifted to a tran sitional reading in sentence(54b). A partial lexical representation for the verb is given in (55).

(55)

In (54b), both manner and motion aspects of meaning are conflated into a new senseof float. In Pustejovsky (1991a) , it was suggestedthat

Generative Mechanisms in Semantics

127

the lexicon. Once a mechanism is in place which admits of making use of information from complements and other non-functor elements in a phrase, it is interesting to imagine how the notion of co-composition can be extended to handle other phenomena. Three applications are ill ustrative to this point ; namely, manner co-composition, feature transcription , and light verb specification. I will present an analysis of one phenomenon related to the semanticsof light verbs in chapter 10, and discussseveral further applications of this mechanism involving the underspecification of semantic forms. The other phenomenamentioned here are discussed in Pustejovsky (forthcoming) .18

7.3

Selective Binding

Now we return to the problem of adjectival polysemy presentedearlier. In chanters 3 and 4,. I discussedthree types of polysemy with adjectival ~ modification , repeated below :

(59) a. We will need a ~ boat to get back in time . b. John is a ~ typist . c. Th .t drivers will be caught and ticketed. (60) a. John put on a ~ album during dinner. b. I 'd like a really bright bulb for my desk. c. Mary dumped the pasta into the boiling pot . (61) a. The man is ~ . b. John is a ~ man c. That was truly a ~

day (event, occasion).

These are interesting from our point of view becauseof the apparent noncompositional nature of the modification as well as the productivity of these constructions. Consider first the examplesin (59). Recall that the standard view on selectionwithin an SEL for these types of adjectives is to enumerate the senses. This requires that every finely-nuanced sense of fast must be typed specifically for the noun class or type it modifies. We observed, however, that the meaning of fast is determined largely by the semantics of the head it is in construction with . Following our discussionof qualia structure in 6.2, we can now give more substanceto this observation.

128

Chapter 7

In predicativeposition, adjectivessuchasfast are ambiguousaswell, beingableto modify both NPs (cf. (62)) aswell asVPs (cf. (63) below). (62) a. That was M ! You're back already? b. Your dog is ~ . (63) a. Emanuel Ax 's hand moved so ~ were a blur . b . Mary

Even

in

the

pretation

was

individual

of

the

interpretation while

dog . The

There are

modifier

two

are

itself

Consider

issues

how

of

fast

to

be

able

the

( 63 ) are

of

the

( 62 ) , however an

event

.

moving

car .

, the In

inter

being

gone

is predicated

standard

-

( 62a ) , the

of someone

when

both

do

we

such

at

Assume

as

in

, 1985 in

modifier

arrive

the

here

event

, of

predicative

context

or

such

events

depends

on

as fast

; and

the

( b ) the

semantics

of

). ( 5gb ) above would

the

: ( a ) adjectives

individuals

. Treating

give

of

qualia

is as follows

the

us Ax [typist

interpretation

that

typist

with

modify

sentence

intersective

?"

dealt

to

adjective

the

an

typing

nominal

event

in

( cf . Bartsch

again

as simply

at

to

of the

being

control

in

reference

property

, being of

head

But

readings

duration

sentences

two

interpretation

fast

the

maintain

.

polysemous

the

fu ! ; . to

requires

to

( 62b ) , the

interpretations

too

predicative

refers

in

the

driving

during the scherzo they

" John

structure

for

adjective

' ( x ) Afast is

a typist

a deverbal

fast ' ( x ) J. who

is

agentive

:

typist ( 64 )

If

fast

ARGSTR

=

QUALIA

=

is

an

position

that

namely

, that

( 65 )

The event

adjective

such as

predicate

is

an

allow

( e ,x ) ]

, then the

]

there

desired

is

no

standard

interpretation

mode

for

of

sentence

com

=

able

Ae [type

to

make

contained

in

interpretation

selective

binding

' (e, X)

fast

available the possible

.

1\

( e ) ] . . .]

a selective

qualia is

for

the

a generative

interpretation head

noun

of . 19

mechanism

-

( 5gb ) ,

( 65 ) .

expression

to

= x : human

[ TELIC FORMAL = type = X

would in

AX [ . . . Telic

makes refer

event

[ ARGI

an

What I will

129

Generative Mechanisms in Semantics

(66) SELECTIVEBINDING: If a is of type , ~ is of type b , and the qualia structure of (3, QS {3, has quale , q of type a, then a ~ is of type b , where

[a{3] = (3 n a (q{3). The semantic device giving this interpretation

can be seen as treating

the adjective as a function and applying it to a particular quale within the N ' that it is in composition with . This same interpretive mechanism will now allow us to account

for the contextualized

senses for evaluative

adjectives such as good, discussed in chapters 2 and 3, and as used in

(67) below.2o (67) a good knife: a knife that cuts well Recall from chapter 6 that the qualia structure for knife is that given in (68) .

knife ARGSTR = [ ARG1 = x:tool ]

(68)

QUALIA= [ TELIC = cut FORMAL = x (e,x ,y) ] Sincegoodfunctions as an event predicate , the event description jn the TELIC quale interpretation given in ( 67 ) . Finally , consider the selective modification (69) below.

it is able to selectively of the

noun , resulting

within

the

NP

modify in the

in sentence

(69) a. John boughta long record. b. a long record: a recordwhoseplayingtime is long. Assuming

that

that

the

the

result

playing

adjective

the

qualia

structure

for

record

lon .g has an interpretation

is a selective

," as illustrated

interpretation

is that

as an event

over the TELIC

in ( 71 ) .

record = [ARGl =Y x:physobj (70) ARGSTR ARCl= .lnfo ] QUALIA = FORMAL = R(x.y) TELIC = play-Icp (e,x.y) ] [ info .physobj

given

event

in ( 70 ) and

predicate of " the

, then record

130

Chapter 7

(71)Ax [. . . Telic= [play(x)(e) Ae A long(e)] . . .J

Noticethat, because theoverall typeoftheNPdoesnotchangeas a resultoftheselective binding operation, sentences suchas (69)arenot

typingviolations.That is, whilethe verb buyselectsfor an individual for its internalargument,the adjectivelongselectsfor an event. Since

thesetypesarebothsatisfied, butat different levels inthecomposition,

the sentence is well-formed.

Whattheseadjectives demonstrate is not a violation or puzzlefor

coercionand selection;rather, they serveto illustratethe subselective

bindingproperties ofdifferent classesof adjectives, as modifying differentfacetsor qualiaofthehead,byvirtueoftheirtype.Modification by an adjective suchas long,fast,or bright,canbeseenaseventpredica-

tion, selectivelybindingthe appropriatequaleof the head. 2 As wesaw

above, theadjectives inthesecasesmodify a distinguished eventpredicate(i.e.,the TELIC quale)associated withthehead,readforbook, and

illuminate forbulb.Thus,a longbook isinterpreted asonetakinga long

timeto read,whilea brightbulbis a bulbwhichshinesbrightlywhen

illuminated.Theseadjectivescan be comparedto modifierssuchas ex-

pensive andopaque intheNPsin(72),bothofwhichreferto thephysical objectratherthanan activityor stateassociated withtheobject. (72) a. an expensive book b. an opaque bulb

Theseadjectives canbeseenasmodifying theFORMAL roleofthequalia

structure

for these nouns.

Thereis anotherinteresting phenomenon thatmightbeexplained by qualiaandselective binding mechanisms; namely, polysemous adjectives suchas oldandnewas usedin an oldfriendanda newneighbor, with thenon-intersective interpretation. Representationally, theadjective selectively modifies ina waysimilartofastandlonginouranalysis above. (73) a. an Qj j friend:(a friendfora longtime) b. {Formal [ AxBy = friend(x,y)A[Telic =

)teS{friend_state(eS,x,y) A long(eS)}} ...J Theadjective oldin thisexample is ambiguous between predication of theindividual andthatofthefriendship itself.Thedurative reading faciliated byselective bindingin (73a)doesnotseemto be available,

Generative Mechanisms in Semantics

131

however , with non-relational nouns such as movie and house, as illus -

trated in (74): (74) a. * an Qld movie : (one that I have had for a long time ) b . * an Qld house: (one that I have had for a long time ) It is not clear whether this is a semantic or pragmatic distinction . While friendship is something that continues , and one can have any number of friendships , possession of a house is typically something that persists uniquely , in successive stages. Hence, the phrase in (74b) does not mean " a house that I 've had for a long time ," but rather assumes the sense of previously lived in or owned. Such an analysis , however , cannot explain why (74a) is not a possible interpretation for a noun such as movie .22 In either case, it is the prepositional phrase of mine or the genitive which brings out the internal (i .e., selective ) reading which is possible , namely , the sense of former . In this section , I have tried to illustrate some further enrichments to the mechanisms of semantic composition , in order to capture the creative use of adjectives in adjective -noun constructions . The rule of selective binding is, in fact , an instance of a more general mechanism facilitating the selection of substructures through a path of features in the semantic description of a phrase . This is developed more fully in Pustejovsky and Johnston (forthcoming ) .

7 .4

Semantic Selection

As stated in above chapter 4, the goal of lexical semantic theory is to provide the foundation on which projection to syntax can be accomplished. One of the avenuesexplored in this researchis to determine to what extent syntactic behavior can be seen as following from semantic selection, and what is ,due to constraints on syntactic form. It is not our aim to completely reduce syntactic selection to underlying semantic types, a rather misguided goal in itself, since the semantic types would simply increasein specialization to reflect distinct syntactic patterning . Rather, the goal is to see how the grammar is affected by a specific approach to modeling lexical semantic knowledge. Although many of the problems in mapping semantic forms to syntax are left unanswered here, it is necessaryto identify the scopeof the changesthat result from adopting a generative lexical approach to semantics.

132

Chapter 7

There are two ways in which a generative lexicon as outlined above affects the mapping from lexical semanticsto syntax: (A ) There is no one-to-one mapping from underlying semantic types to syntactic representations; rather, a syntactic phrase is only fully interpretable within the specific semantic context within which it is embedded; (B) Because the representation of semantic information in the qualia structure , argument structure , and extended event structure is richer than what conventional models associatewith a word, a more complex model of filtering and checking is necessaryfor restricting the output to actual syntactic form. 7 .4.1

Canonical Syntactic Forms

Given the previous discussionof the mechanicsand operation of coercion and other generative devices, what appearsobvious is that the semantic type associated with an argument for a verb allows for a multiplicity of syntactic expressions; this forces us to addressthe issue of what role semantic selection is playing in generalin the grammar. The view taken here is that the grammar associatesa canonical syntactic form with a semantic type, but there are in fact many possiblerealizations for a type due to the different possibilities available from generative devices such as coercion and co-composition. Our methodology here, however, is to first look for semantic distinguishability betweendistinct sets of syntactic realizations. If no empirical evidence presents itself suggesting that two lexical items are truly distinguished in their semantic types, then the residual difference between these items is attributable to syntactic distinctions rather than semantic typing . For an expression of a particular type, there is a unique syntactic realization, which is able to adequately expressthe semanticsof the expression, and which can be seen as the most direct realization of the type , what I will call the canonical syntactic form (csf). The alternative syntactic forms which are available to an expressionare licensed by virtue of principles of semantic "recoverability," such that the information conveyedby the expressionwhosecanonical form is 7ris also able to be recovered by generative devices in another syntactic realization, 1[' . The proposal put forth here can be seenas an elaboration and extension of the ideas discussedin Chomsky (1986), where the notion of canonical

:ve Mechanisms in Semantics Generati

133

syntactic realization is introduced. In previous chapters, I discussedhow semantic types can be expressed in the lexicalization process. The lexical conceptualparadigm (lcp) can be seenas the lexicalization of a number of distinct semantic types into one lexical form. The semantic type which results is a dot object, and the lcp itself is the type cluster of the individual types together with the dot object . 1~he syntactic realizations for a semantic type are in part determined by virtue of the semantic type. Hence the form will exhibit the behavior of each type it is composedof in addition to the unique behavior of the dot object. The set of phrasal alternations associated with a semantic type is similar in nature to an lcp, in that they are determined by virtue of the semantic type. Unlike the lcp, however, which can be seen as taking a number of types and fusing them into a lexical item , the mapping from a single semantic type to syntactic forms is a one-to-many relation. What determinesthis mapping is the manner in which the generative devices are allowed to transform one semantic type into ano,ther, under a set of specified constraints. Factors that are outside of the realm of the syntax-to-semantic mapping might restrict an otherwise legitimate set of mappings or syntactic possibilities. As we shall see in chapter 8, an lcp may not necessarily exist as a unique lexicalization, but may be distributed across several lexical entries. The lcp itself as a conceptual relation may exist, but logical polysemy in a single lexical item would not occur for that concept in this language. For example, while the noun newspaperis logically polysemousbetweenthe organization and the printed information-containing object , the noun book refers only to the latter , while the noun author makes referenceto the "producer" of the book. In what follows, I will discusstwo examples of how the semantics of a lexical item is able to determine the resulting syntactic forms possible for complements to that item . The casesI wish to examine include selectional distinctions betweenthe following two minimal pairs: (75) a. the verbs like and enjoy; b. the interrogative selecting verbs ask and wonder. What emergesfrom this discussionis that semantic selection can in fact be a good indicator of the syntactic behavior of a lexical item , but only when viewed together with the generative operations that connect the distinct syntactic forms for a particular semantic type.

134

Chapter 7

Let us assume that , for any semantic

type , 7 , there is a unique

canon -

ical syntactic form {csf} that expressesthis type as a syntactic object, Xi . Expressed as a function , csf can be viewed as that unique mapping from semantic types to syntactic forms , cst 7 - t Xi , such that :

(76) For every type 7 in the set of semantic types, there is a function, canonical syntactic form {cst} , such that CSf (7) = Xi , except for a . when

7 = T , or

b . when

7 = -1,

in which case csf is undefined .

A syntactic expression, yj , of type a (where csf( a) = yj ), is substitutable for the csf of a type 7 only if this type is fully recoverable from licensed

semantic

operations

on a .

For example , assume a verb is typed for selecting an argument of

type 7, where cst( 7) = xi . The direct realization of this type as Xi is well -formed , assuming surface constraints are satisfied in the syntax ;

that is, [V Xi . . .] is a legitimate structure. Now consider the same verb appearing together with a syntactic phrase yj , of type a2; i .e.,

[V yj . . .]. From what we have said above, this structure is well-formed only if 7 is fully recoverable from a2. This is illustrated in (77) below .

(77)

Canonical

Form X ?

The resulting set of syntactic forms associatedwith a particular semantic type will be called a phrasal paradigm.23 The operations ensuring recoverability on the semantic type in this case are exactly the generative devices introduced earlier in this chapter; namely, type coercion, selective binding , and co-composition. It should be pointed out that the overall enterprise of establishing a lexicon operating along the principles outlined in this work are to some extent independent of the successor failure of semantic selection determining syntactic behavior.

135

Generative Mechanisms in Semantics

Let us turn now to specific examples illustrating

this mechanism , be-

ginning with the classic sub categorization differences between the verbs

enjoy and like. Observe in (78) and (79) how these verbs differ in their complement selection behavior :

(78) a. Mary likes to watch movies. b. Mary likes watching movies . c. Mary likes movies .

d. Mary likes (for) John to watch movies with her. e. Mary

likes that

John watches

movies with

her .

f . Mary likes it that John watches movies with her .

(79) a. Mary enjoys watching movies. b. Mary enjoys movies . The syntactic differences between these and other verbs are typically used as counterexamples to strong views of the semantic selection hy-

potheses(as held by Lakoff, 1971, Dixon , 1984, Wierzbicka, 1988, Chomsky, 1986). The discussionfrom the early literature on these differences were suggestive , but not conclusive of a real semantic type distinction . Thjs was due, in part , to the lack of a sufficiently rich type system that was both linguistically motivated as well as formally characterized . Viewed within the current enterprise , however , there are diagnostics in dicating that the semantics of these verbs are in fact distinguished by the type of complement each selects. Notice the distinction between these two verbs that emerges in the following discourse . Imagine that two roommates are discussing a chair in their apartment ; speaker A utters

sentence(80) to B , who in turn respondswith sentence(81). (80) I want to get rid of this chair. (81) But I ~

that chair!

The statement made by B in (81) is difficult for A to question, since the questioner is not privy to the actions . By uttering (81) , B towards the object , although now an alternate response to

attitudes of the hearer , only to his or her does not express any articulated attitude more explicit attitudes do exist . Consider (80) , that given in (82) :

(82) But I ~!~~~ that chair!

136

Chapter 7

Sentence (82) expressesa very different relation with respect to the complement. In this case, the proposition is readily verifiable, since it says of an object, that the speaker stands in a relation of performing some activity with the object; as such, (82) can be doubted, questioned, or denied, sincethe judgment relies on the person's observablebehaviors, entailed by the semanticsof the selectionalproperties of the verb. Hence, (83) is a legitimate responsefor A to make for (82) but not for (81). (83) But you never use it ! The distinction here is not just a pragmatic one, but is due to the underlyingly distinct types selectedfor by theseverbs. Namely, the verb enjoy selects an event function , while like selects for an attitude towards any type , which I represent simply as the top most type available to interpretation , i .e., T . This would include the semantic types event , prop , property , ind , and facti ve. These are shown with their respective canonical syntactic forms in (84).

(84)

VP[+ PRG ] NP S[+ INF] S[+ TNS] Becausethe verb like selectsfor T , it has no unique cst In fact, a unique csfexists only if there is a unique least upper bound in the type lattice . Notice from the figure that there is still only one canonical syntactic form given for the proposition even though full sentential infinitival and PRO-form infinitivals are often considered in free variation , and constrained due to binding constraints. What this states is that PRO interpretation might be usefully associatedwith a coerced form of the complement. In (85), the two verbs are compared, where their syntactic behaviors are associatedwith distinct semantic typings:

Generative Mechanisms in Semantics

137

(85)

[ind]

What

this

straints

illustrates to

thing

give

to

a

forms

the

us

turn

the

semantic

of

lection

is of

properties

( 87 )

)

in

relation questions

John

asked

me

what

John

asked

me

the

wondered

a .

John * John

a .

John

b .

* John

didn

of

concealed of

ing

semantic

and

care

that

differences

case

care

verbs

The

( 87 ) - ( 88 )

the

by

between

strategy in

distinction

is these

the

employed

syntactic in

in

the

verbs

. 24

.

wa

care

to

~.

verbs

take

this

the

NP

by

Pesetsky

rather

than

complements

expressibility

selectional

.

.

attributed

two

in

the

.

was

temperature

and

se -

and

.

temperature

wonder

from

semantic

.

temperature

the

of

below

was

nature

interrogative

difference

temperature

temperature

interpretation assigned

distinction .

the

and

~ ~ at

care

of

complements

Observe

the

.

discussion

idea

.

the

't

the

question

absence

't

didn

) , the

exists

the

a

-

those

there

type

problem

interrogative

the

which

continuing

temperature

only

illustrating

( 86 )

what

wondered

for

con

important

and

semantic

( 1982

to

a .

inability

semantic

Pesetsky

all

verb

, the ) ,

under

. The

that

pair

, namely

between

a

base

( 1986

and

b .

( 88 )

a

1981

concealed

b .

their

Chomsky

illustrated

( 86 )

The

,

is

minimal

relation

In

discussed

problem

from

, operates

polymorphism

to

second

selection

( 1979

verbal

complements

our

example

discussion

coercion to

.

, for

of

as

now

complementation Grimshaw

coercion

previous

appear path

Let

how

appearance

from

legitimate

of

the

note

syntactic

is

S[+ INF] S[+ TNS]

NP

VP[+ PRG]

work should

of ,

ask

however be

( 1982 to

an

is

of

the

) to

the

underly

versus ,

first

form

-

wonder to

attributed

assume to

138

The

Chapter7

difference

would

appear

to

be that

ask selects

for

a true

inter -

rogative (interpreted as a set of propositions), while wonder selects for an attitude

towards

a set of propositions

. Even with

this discussion , it

might be argued that such a finely -articulated distinction between ask and wonder is not something that should be encoded in the semantics typing of the complements , since such a difference could not possibly be learned by the child in linguistic experience . Interestingly , such semantic

distinctions

seem to be more difficult

for children

to acquire , and

the failure to make these subtle semantic differences is accompanied by a systematic misuse of the complement patterns associated with these verbs (cf . Pinker , 1984, 1989) . If that is possible to maintain , then it is interesting to speculate that the semantic differences are motivated and furthermore obviate the case theory analysis for this example . The syntactic consequence of such a semantic split might be reflected in distinct tendencies towards transitivity , but that too would be as a result of the semantic What

typing . would

the semantic

types

of these two verbs

order to give rise to both the appropriately mentioned

above , as well as the syntactic

need to be , in

fine-tuned interpretations

forms possible

for their

com -

plement position ? Exactly that distinction made above; namely , wonder takes as its complement an attitude towards a question , while ask selects a question directly . Some useful work towards characterizing this

distinction has already been made by Groenendijk and Stokhof (1989), who distinguish

between know and wonder as extensional and inten -

sional complement interrogatives , respectively .25 Assume that semantic selection

assigns the internal

argument

of ask the type quest ion , which

is a shorthand for the type <s , t >, where an interrogative is seen as a function from worlds to answers. Following the observations made in Groenendijk and Stokhof (1989) and others , let the type selected for by wonder be , simply the intension of the type assigned to the complement of ask. This distinction turns out to have consequences for the syntactic complementation pattern for each verb , as we see below . The tree in (89) gives the two semantic types associated with the interrogative argument of the verbs ask and wonder along with their canonical

and non -canonical

syntactic

forms .

139

Generative Mechanisms in Semantics

(89)

S[+ WH]

NP

It is possiblethat the subtleties of such distinctions are not really appreciated or recognizableuntil larger segments(or units) of analysis in the language are investigated, such as the way these verbs are embeddedin a particular discourse and context. This does not mean, however, that the information is not lexically encodedinto distinct semantic types for the complements. The NP complementation pattern for verbs like ask illustrates two aspect of the approach outlined here: A . the ability of the verb to coerceits complement; B . the ability of the complement to metonymically reconstruct the required coercing type from the semantic structure within the complement. Without detailing the mechanismsbehind these shifts in other classes, it is worth briefly looking at what the range of complementation patterns is for a small number of verbs, where the phrasal paradigm associated with each verb below is illustrative of a distinct semantic type

(90) LIKE-class: (like, hate, prefer) a. John would h~ Bill to leave. b. John h~ ~ (it ) that Bill left . c. John .h~ ~g to lose the game. d. John h~ .E?-dlosing the game. e. John .h~ ~ that he lost the game. (91) WAGER-class: ( wager, bet) a. * John wagered Bill to have left .

140

Chapter7

b . John wagered that Bill left . c. * John wagered to have left .

-class: (92) MAINTAIN a. * John maintainedBill to haveleft. b. John maintained(for a year) that Bill left. c. * John maintainedto haveleft.

(93) TRY-class: (try, attempt) a. * John Y !~ Bill to read the book. b. * John trig.Qthat Bill read the book. c. John k ! 4 to read the book. d. John t! !~ the book.

(94) REMEMBER -class: (remember, forget) a. John rememberedto lock the door. b. John rememberedthat he locked the door. c. John remembereglocking the door. d. John rememberegwhere he put the keys. e. John remember~Q his phone number. (95) PERSUADE -class: (PERSUADE , CONVINCE ) a. John convinced Mary to build a house. b. John convinced Mary that she should build a house. c. John convinced Mary that she had built a house. From our previous discussion, it follows that the broader the selectional possibilities for a verb, the more general is the semantic type associated with the complement. We saw above that the verb like selects for the type T , for which there is no specific canonical syntactic form (csJ) , but it is associatedwith the csfsfor its immediate subtypes; hence, indirectly , it behavesas though it has several csJs, along with the derived syntactic expressionswithin the phrasal paradigm for each type. The verbs .fo~ qet - ~ and rememberselect very generally for a factive inte~pretation of any type .

8

The Semantics of Nominals

8.1

Basic Issues

In this chapter , I give a general overview of theory can contribute towards a classification types . In studying the semantics of nominals major areas of concern from the perspective of

how a generative lexical of the different nominal , we can distinguish four a lexical semantic theory :

(1) The distinction in complement-taking behavior between nouns and verbs ;

(2) How nominalizations and event-denoting nominals are distinguished from their corresponding verbal representations and the events they denote : that is, what is the difference between an event represented as a sentence, and an event represented as an NP ? And likewise , for facts and propositions , represented as sentences and NPs .

(3) The representation of logical polysemy in nominals, such as window, record, book, and how these implicitly relational

nominals

relational nominals differ from

;

(4) How the semantics of nominals facilitates the richer compositional in terpretation required for characterizing natural language semantics as polymorphic ; that is, what allows for co-compositional interpreta tion in natural

language ?

The first two issues are of course related , and I will argue that they cannot be addressed independently of each other . Furthermore , a formal distinction

in how nominals

and verbs refer to event descriptions

is nec -

essary to show how the grammaticalization of events differs i 1} these two syntactic domains . This is presented in 8.3 and 8.4. The third point was addressed in part in chapter 6, but many questions remain regarding the descriptive power and the exact nature of dot objects and how argument and event p~rameters interact in the qualia expressions . These issues are discussed below in 8.1 and 8.2, where I distinguish between unified types and dotted types . The former

are a more formal

interpretation

of the or -

thogonal types of Pustejovsky and Boguraev (1993), incorporating the structures employed in Copestake (1992) and Copestake et at. (1993). The last point above was touched on briefly in chapter 7, and will be the subject of later discussion . This concerns the exact manner in which

142

Chapter 8

coercion and co-composition rules makes use of qualia-based information . The variation in the expression of complements between nouns and verbs has long been a major concern to linguists, and has motivated many of the shifts in the theory of grammar. We will not be able to address this issue until we have explored in somewhat more detail the semantics of nominals from a GL perspective. As mentioned in chapter 6, for any category we can potentially distinguish three distinct dimensionsalong which the elementsof that category can be analyzed semantically. With respect to nouns, the interpretation can vary according to the three dimensions below: (A ) ARGUMENTSTRUCTURE : How many arguments the nominal takes; what they are typed as; whether they are simple, unified, or complex types. (B) EVENT STRUCTURE : What events the nominal refers to , both explicitly and implicitly . (C) QUALIA STRUCTURE : What the basic predicative force of the nominal is, and what relational information is associatedwith the nominal, both explicitly and implicitly . The manner in which these three representational levels can help us classify the semantics and associatedbehavior of the different nominal types will becomeclear in the subsequentsections.

8.2

N ominals with Unified Types

For purposes of lexical representation, it is often necessaryto allow a lexical item , making referenceto a type in a lattice , to be able to inherit from multiple parents. For some conceptual structures and their associatedlexicalizations, we will also needto employ a schemeallowing multiple inheritance. This has been explored in a generative lexicon to some extent in Pustejovsky and Boguraev (1993) and more generally in Copestake (1992) and Copestake et ale (1993). Previous attempts at structuring conceptual hierarchies (whether explicitly language related or not) have made heavy use of multiple inheritance, as systems have to grapple with accounting for the fact that , according to particular lexical-conceptual projections, biased by a variety of context factors,

TheSemantics of Nominals

143

different aspects of objects become more or less prominent as context varies . Thus , as illustrated below , a "book " is _a "information ," as well as a "physical _object ; a "dictionary " is _a "physical _object ," as well as "reference" ; a "car " is _a both "vehicle " and an "artifact ," and so forth . The conventional view on these sorts of inheritance relations is shown in the figure below .

information

.._-----~ -~ -artifact -

reference

Figure 8.1 Conventional Representation of Inheritance Relations

Still , as descriptive as such relations may appear, models like these suffer from a very limited notion of lexical structure; one particular consequence of this is the ambiguity of class membership (or, in our termi nology, 'hidden' lexical ambiguity). Thus, even though elaborate mechanisms have been proposed to control and limit the flow of information along the generalization/ specialization links, there has been no theory to either ( a) explain how to assign structure to lexical items, or (b) specify lexical relations between lexical items in terms of links between only certain aspectsof their respective lexical structures. The approach presented here, with its several distinct levels of semantic description, and in particular the qualia structure , are relevant to just this issue. On this view, a lexical item inherits information according to the qualia structure it carries. In this way, the different sensesfor words can be rooted into suitable, but orthogonal lattices. To illustrate this point , consider the two is _a relations below, and the differencesin what relations the objects enter into .

144

Chapter 8

play is_a book read

dictionary

ok

is_a book

no

buy

ok

ok

consult

no

ok

begin

ok

no

Figure 8.2 Table of is - a relations

This table illustrates a serious problem with most current inheritance systems for lexical knowledge. Namely, although it might seemreasonable to think of both plays and dictionaries as "books," they behavevery differently in terms of how they are selectedby different relations. This suggeststhat a single lattice for inheritance is inadequate for capturing the different dimensionsof meaning for lexical items. In Pustejovsky and Boguraev (1993), a proposal is made in regards to the structure of Lexical Inheritance Theory, and the need for typed inheritance for lexical information . The proposal, although not fully formalized there, was to posit a separate lattice per role in the qualia structure . Briefly, inheritance through qualia amounts to the following relations for this example: (1) a. book is _formal phys-object b . book is _telic information c. d. e. f. g. h.

book is _agent information dictionary is _formal book dictionary is _telic reference dictionary is _agent compiled -material play is _agent literature play is _telic book

The different inheritance structures just mentioned can be illustrated by the diagram below .

145

The Semantics of N ominals

reference

F

T compiled_matter

A A

The qualia roles differentiate the lattice structures resulting in a typed inheritance , excluding the unwanted inferences listed above, and allow -

ing only the desired ones.l Let us now integrate

this perspective

into the typing

system outlined

in previous chapters. In Carpenter (1992), type unification is defined as the unique greatest lower bound , glb , for two types al and 0"2, i .e., 0"1n0"2. This will allow us to define those objects which are non -polysemous , yet do involve the logical conjunction of their types . For this reason , I will refer to these objects as unified types. To illustrate

the formal

properties

of nominals

with

unified

types ,

consider the following examples . A type lattice provides the essential characteristics of a concept as a type in a hierarchy . Orthogonal inher itance arises not from the multiple assignment of essential typing of an object but from constraints on functional information about the object

146

Chapter 8

.I

suppliedby the qualia. The type hierarchyshouldbe so constrained as to allow only orthogonaltypes to be unifiable (i.e., distinct qualia types); in other words, differentqualiamayunify to form a unifiedtype, if the qualia unificationis well-formed. For example, giventwo orthogonaltypes a and T with qualia structures (2a) and (2b), (2) a. u = [0- . . . [QT = a ]] b. T = [r . . . [QA = {3 ]] they may unify, anT , to form the unifiedtype a_T, with the resulting qualiastructuregivenin (3):

TELIC =~={3 (3) [o QUALIA = AGENTIVE -_r [ r\I ] ]

The creation of a concept (i.e., type) that refers both to a physical object and a proposition, however, is not a possible unified type, since the FORMALqualia values for the two concepts are not unifiable. The ability to construct types that are otherwise outside of the set of unifiable types is something enabled by the construction of dot objects with the lcp type constructor, mentioned in 6.2, and discussedbelow. Let us look at a few examplesof unified types to illustrate what role the qualia are playing in supplying orthogonal dimensionsof an object's denotation. Consider first the nominal food. As mentioned in chapter 6, the TELIC role specifies for a given concept its use or function ; for food this value is the activity of eating. The construction of the type associatedwith food arises from a qualia-based constraint on the type physobj that it be edible. food

(4)

ARGSTR= U LIA = Q A

ARGI = x :physobj ] TELIC = eat FORMAL = X(eP,y,x ) ]

Similarly , the semantics for an artifactual object states minimally that it is something that was created , made , manufactured , or brought about by some human activity . The value of the AGENTIVE role for the type physobj indicates just this relation to the object , as shown in (5) , where top is the top -level type in the lattice , as defined above in Section 7.4.1.

147

The Semantics of N ominals

(5)

artifact ARGSTR= QUALIA =

D-ARGl y:human ] [ ARG 1== x:top = make [ TELIC FORMAL = X (eT,y,x) ]

This representation says nothing about the nature of the object except that it is human derived. That is, as an artifact , it could be a social construct, a verbal act, a physical object, and so on. Given the underspecified representation in (5) regarding what an artifact is, as opposed to how it came about, we can further restrict the type which partici pates in the qualia relation, effectively unifying a subtype with the qualia structure for the concept artifact . For example, to represent artifacts that are physical objects, we would like a type which combinesboth the FORMALvalue of a physobj and the AGENTIVEvalue of an artifact . Becausewe are treating the argument and the qualia relation both as types to be unified, it is possible to view the resulting lexical representation as a unified type, effectively arriving at a type structure similar to Copestake's (1992) treatment of greatest lower bound types. The resulting qualia structure is shown in (6) below.

(6)

y:human :physobj ] :~make (eT ,y,x)]

Another example of how the types instantiated as qualia structures can be unified to form new unified types is illustrated with the concepts of tool and artifactual tool . As discussedin chapter 6, a tool is simply defined as an object which has a TELICvalue indicating a purpose telic rather than a direct telic. That is, the predicative argument is the first argument in the TE~IC relation . Hence, for a word a , defining a TELIC value essentially classifiesit as something that has a use;

(7)

a ARGSTR =[ARG 1=x:top ] QUALIA =[FORMAL TELlO =R =(e x,x,y)]

Here again, this sayssimply that the object is usablein a certain capacity (i .e., TELIC), and saysnothing about how it cameabout (i.e, AGENTIVE ),

148

or

Chapter8

what

a

its

basic

naturally

device to

used

be

type

is , i . e . , its

occurring as

both

a tool

an

artifact

object of

persuasion

artifact

_ tool

FORMAL

being

and

, which

is

. a

role

used

as

.

tool

Restricting

tool

,

represented

the

however

as

Hence

a

in

,

, it ,

or

could to

nature

creates

a of

a

refer

to

rhetorical this

unified

object type

,

(8 ) .

a:

ARGSTR (

=

) QUALIA

x : top = y : human

TELIC

=

R ( e ,x ,y )

[ AGENTIVE FORMAL

Nouns

such

type

unification

be

as

knife

, discussed

very

specified

= =x

clearly

locally

for

in .

knife

make

6 .3

since

( eT

above

Observe ,

]

that lexical

,y ,x ) ]

, illustrate the

this

AGENTIVE

inheritance

operation value through

need the

.

not

= 1

=

8

of

[ ARGI D - ARG

FORMAL type

value artifact

will

bind

_ tool

.

the

AGENTIVE

to

the

value

inherited

from

the

(9)

ARGSTR

=

QUALIA

=

.

knife

D-ARGl y :physobj [ ARG 1 = = x :artifact _tool ] = cut [ TELIC FORMAL = X(e,x ,y ) ]

A look at the type lattice which results from such constructions illus trates the process of type unification , where unlabeled edges denote the FORMAL Quale. (10)

.

.

.

~

T entity , / /" 2 : " 1 Q ~ abstrac ~phys arti ~ tool _ o bj ~ . / ~ QAnQT .

.

.

ph ys _ai~t_ rt t1 oo

149

The Semantics of N ominals

8.3

Nominals with Complex Types

Having discussedthe propertiesof unified types, we turn now to the more complexstructure associatedwith dot objects, first discussedin chapters6 and 7. With the help of dotted types and qualia structure, we cananalyzethe inherentlyrelationalaspectof certainnominalswithout having to treat them as overtly relationalin denotation. The dot object is a type whichnecessarily incorporatesthe meanings(i.e., types) of its simpletypes into the complexobject. The examplesI haveused throughoutthis work haveincludednounssuchas book , door, and novel. The mannerin which suchnounsarerelationalis expressed in the FOR MALquale, asoutlinedin 6.2. For the noun book , for example,a relation specifyinga particularcontainmentrelationbetweenan individualphysical object andinformationis explicitly part of the makeup of the noun'8 FORMAL role, as repeatedbelow: a (11)

ARGSTR = ARG 1 = X:Tl ARG2 Y:T2] QUALIA =

FORMAL = R(y,x) T1. T2-1CP ]

The predicativeproperty of book , as discussed in chapter6, is givenby its type as a dot object. Hence, we canview the FORMAL Qualehereas the heador predicativequaliarole in this structure, givingthe equivalent representationin (12): (12) [FORMAL = x.y : R(x, y)] This is, in turn, translatableasthe expressionin (13) below: (13) Ax.y3R[book(x.y) /\ FORMAL = R(x, y) . . .] Notice that we have constructed a type x .y in (13) above without concern for how it is related by subtyping in the lattice . Because of the qualia -based relation between the types in the FORMAL role of the object , there is no way of exhaustively defining the behavior or characteristics of the dot type in terms of the type lattice alone . We can , however , provide for a partial characterization of the type for purposes of type selection

150

Chapter 8

within

the grammar

object

a . {3 to be selected

.

Namely , we need a mechanism in an environment

of its dot elements , a or {3. I will call this particular Object Sub typing , e ., and define it below . ( 14)

for allowing

less specific

than

an

either

coercion

rule Dot

0' 1 . 0' 2, which

is a dot

DOT OBJECT SUBTYPING : , : 0" 1 . 0"2,

e ' [0"1 ~ 7 , 0"2 ~ 7 ] : 0' 1 . 0' 2 - t 7

8 -[0"1 ~ 7 , 0' 2 ~ 7 ] (, ) : 7 This

states that , given an expression

, of type

object , there is a subtyping relation possible between the dot object and a type 7 , just in case 7 is the least upper bound of both of the dot elements , 0"1 and 0"2; coercion furthermore in an environment normally typed for 7 . To illustrate

the utility

lectional

properties

towards

a dot object

what

specific

allows the dot object

of this operation

of the verb

to pass

, we need only recall the se-

like . By expressing

a specific

such as a book , we do not commit

attitude

ourselves

aspect of the book causes me to like it . It might

be its " bookness ," independent

of what

information

it contains

to

simply ( a dot

element ) , what it looks like ( another dot element ) , what I can do to it (a true coercion ) , and so on . In fact , the generic nature of the statement in ( 15a ) suggests that process of reading terpretation . ( 15)

( 16 )

it cannot

or writing

, since that

a . Mary

likes the book .

b . Mary

liked the book .

a . ?Mary b . Mary

be associated would

why a generic interpretation telic

an iterative

book , she enjoyed

event

reading

it ) . As argued

The application

of the dot object

of book is illustrated material

a generic

in -

paradigm

with

Mary

read that

discussion , enjoy seman -

and coerces its complement subtyping

in ( 18 ) below . Assume

," associated

for ( 16a ) is only possible

(i .e., every time

in previous

tically selects for an event function appropriate type is not present .

conceptual

not entail

single

enjoys the book .

This also explains

" printed

any specific

enjoyed the book .

with

lexical

with

, if the

rule above to the type that

book is a species of

the type print

-D1atter . Then , the

for this dot object

is given in ( 17 ) below .

151

The Semantics of N ominals

(17) printJIlatter _lcp = {physobj.info , physobj, info } (18) a. e [physobj~ T] :physobj~ T b. e [info ~ T] :info ~ T c. e .[physobj~ T ,info ~ T] :physobj.info ~ T Hence , the type restriction for the complement position of like is satisfied by direct subtyping , as illustrated in (19) . ( 19) S / / / / " " " '" [human ] VP I a

/ V

-/

/ /"-.." " ~"" [T] e ' :physobj . info ~ T

like

the Jook

Now let us turn to how nominals denoting dot objects are different from relational nominals , such as father and sister . In chapter 6, I discussed simple and complex typed nominals , where the form of the qualia structure determined the general class of nominal a lexical item denotes . Hence , along the FORMAL dimension alone , we can distinguish between simple natural kind terms such as rock and man , as well as simple typed artifacts , such as knife , and complex typed nominals such as book and record. In spite of this difference in typing , however , all these noun classes are predicative nominals . The manner in which predicative nouns differ from nominals that are overtly relational in nature , such as brother , sister , and neighbor , must still be addressed. As discussed above in chapter 2, a relational noun denotes a set of individuals standing in relation to at least one other individual in a specific way.

(20) a. John's neighbor is visiting . b . Mary 's brothers came home together . In (2Gb) , Mary stands in a particular relation to the head brother , which is given directly in the semantics of the noun 's FORMAL role :

152

Chapter 8

(21)

The conditionsunder which the argumenty is existentiallyclosedor bound to a constantare discussedin Eschenbach (1993) and elsewhere (cf. Bierwisch,1983). The important point to noticeabout the represen tation above, however , is how the FORMAL Qualediffersfrom the forms we haveexaminedthus far.2 That is, unlikea nounsuchas cakeor rock, with simpletyping and a FORMAL valueno differentfrom the typing on the argument, relationalnounshavea "relationalrepresentation " in the FORMAL Quale . However , they must be distinguishedfrom the complex typed nominalssuchas book , whichalsocarry a relationalFORMAL value (cf. (3)):

(22)

ARGI= X:/ l ARGSTR = ARG2= Y:/ 2 a: ] ' 1. ' 2_lcp QUALIA = FORMAL = P{y,x)

]

The distinction betweenrelationalnominalsand nounssuchas bookis due to the latter being typed as dot objects. That is, although the complexobject denotedby bookis partially definedby a relation in the FORMALquale, its actual type is quite distinct from a simple typed relationalnominalsuchas brother. The ability of sucha nounto assume onecomplexpredicativeinterpretation, or oneof two simplepredicative interpretationsis what the lcp accomplishes (with the help of coercion operators). I return to this point in chapters9 and 10, whendiscussing constraintson coercion. It is perhapsworthwhile at this point to recall what the linguistic motivation is for the existenceof dot objects. There are, I believe, two major reasonsfor admitting suchobjectsinto our semantics : (i) SEMANTIC MOTIVATIONS : The knowledgewe haveof the concepts associatedwith doors, windows, books, computerprograms, etc. is not characterizable asthe conjunctionof simpletypes (or properties)

153

The Semantics of N ominals

in a conventional type hierarchy . The predicates and relations for the lexical

item associated

with

such a concept

are characteristic

of that

concept alone . For example , the concepts of "reading " and "writing " are not conceivable

activity

without

the existence

of the concept to which the

is applied , i .e., the dot object of printed material .

(ii ) LEXICAL MOTIVATIONS : The dot object captures a specific type of logical polysemy , one that is not necessarily associated with true complement coercion . There is strong cross-linguistic evidence suggesting that the way such concepts are lexicalized is systematic and predictable . So far we have focussed our attention on what the dot object is not : it is not a unified

type , created

from the meet n on types ; nor is it a

standard generalization (or join ) on types, in the conventional senseof this operation (cf . Morrill , 1994) . I have characterized it as a Cartesian type product of n types , with a particularly retricted interpretation . The product 71 x 72, of types 71 and 72, each denoting sets, is the ordered pair

< t1 , t2 > , where

t1 E 71, t2 E 7"'2. But

the pairing

alone does

not adequately determine the semantics of the dot object ; rather , the relation , R , which

of the definition

structures

the component

types must be seen as part

of the semantics for the lcp type constructor

itself .

That is, in order for the dot object 71 . 72 to be well -formed , there must exist

a relation

R , that

" structures

"

the

elements

of 71 and

72 ; i .e . ,

R (t1, t2)' For nouns such as bookand record, the relation R is a subtype of "containment ," while for partially event-denoting nouns such as lunch

and sonata, the relation is more complex (cf. 8.5 below). For now, we might view the set of relations, { Ri } , as specializedtype product operators , where the specific relation is built into the construc tor

itself :

(23) { Ri } == oR! , R2, oo. , Rn Then , for a dot object nominal such as lunch , which is polysemous between "event " and " food " interpretations (cf . 8.5) , the appropriate dot

object is that shown in (24). (24 ) lunch : event , lunch : food lCPR!(lunch) : event . food The common noun lunch would then have the following interpretation :

154

Chapter 8

(25 ) Ax .y3Rl [lunch (x : event .y : food ) 1\ FORMAL = Rl ( X , y ) . . .J With

abstraction

eralize

over tuples

the semantics

phically

such as that

of determiners

to any tuple

type . This is discussed

( 1995b ) , and in some respects , is similar of parametric polymorphisms . Given turn

a mechanism

for creating

to distinguishing

related

lexical

different

- matter

_lcp

to apply

more fully

to Morrill

types such as dot objects , let us now the nouns

- matter

for semantically

book and newspaper , which

. This type has the lcp referred

= { physobj

.info

, physobj

, info

tion

that

the physical

one generally

disbelieves

in ( 27 ) , where New York ( 27 )

object

that

to

}

As one would expect with this type , there is a logical polysemy paper between

polymor -

in Pustejovsky

's ( 1994 ) treatment

classes of polysemies

items . Consider

both refer to the type print above , repeated here ; ( 26 ) print

in ( 25 ) , we will need to gen -

and quantifiers

for news -

one purchases , and the informa -

or believes . Consider

Times is a subtype

of newspaper

the sentences .

a. Eno the cat is sitting on yesterday 's New York Time ~. b . Yesterday 's New York Times really got me upset .

In addition

to this

complex

of senses , represented

by the dot object

shown above , the noun newspaper refers logically also to the organization which publishes it , a sense which is unavailable to the noun book . ( 28 )

a. The newspaper

has just

b . The newspaper is filing for wire tapping . ( 29 )

suit

against

the federal

government

a . * The book has raised the price of paperbacks . b . * The author

While

fired its sports editor .

is suing the book for breach of contract .

newspaper carries a sense corresponding

publishes

the paper , the noun

to the organization

which

book is unable to denote the publisher

of

the book in these contexts . A concept such as that denoted by the noun newspaper is really a construction from types , one of which is itself a dot object .

The Semantics of N ominals

155

(30) newspaper_lcp = {print - matter .organization , print - matter , organization } As discussedin 8.6 below, whether a languageactually lexicalizesthe dot object representedabove in (30) by a single lexical item or not, the concept exists potentially with this structure in the semantic representation for the language. We discusssplit lexicalization below in 8.6. From these considerations, the observation here is that dot objects are constructed in a pair-wise recursive fashion, as illustrated in (31) and (32) below. There may, in fact, turn out to be instances of dot objects that are constructed from three dot elements or more, but this is question open to further investigation. (31) (32)

a : 0"1 a : 0"2 lcP1(a ) : 0"1 . 0"2 /3 : 0"3

/3 : 0"1 . 0"2

Icp2({3) : 0"3 . (0"1 . 0"2)

The lattice structure

relating the type associated with newspaper to

physob info organiza ~t m ~ atte '~ ".newsp . /(.)book

that for print - matter and book is shown in (33). (33)

The way in which a noun such as newspaperdenotes, however, is quite different from that of book, due to the type of lcp it belongs to . That is, the qualia structure for newspaperis a specialization of the lcp known as product .producer , where the AGENTIVEquale makes referenceto the dot element denoting the producer, and the FORMALquale refers to the product ; the dot object itself does not appear in the qualia except to define the type itself:

156

Chapter 8

ARGI ARG2

(34) -

Hence, although newspaper is logically polysemous , it can not denote the complete dot object , as can book. Rather , one sense or the other is available

for interpretation

, but not both .3

Given this type of representational strategy , one might ask what the appropriate

distinction

is between nominals

such as book and novel . Al -

though book may refer to novel in many contexts , the latter is obviously more informative and less extensive in its selectional distribution . They are both dot objects , and in fact , novel is arguably a subtype of the dot object book. But this subtyping alone does not inform us of what distinguishes these concepts ; namely , while a book is possibly an informa tion holder of potentially any type of information , a novel is restricted to a specific structure and form of the information , e.g., a narrative . The

relation

between

these

nominals

can

therefore

be characterized

as

the specialization of one dot element in the dot object . That is, for book denoting type physobj .info , the dot object associated with novel is physobj .narrati ve , where narrati ve ~ info . This is illustrated in

(35).

(35)

TheS_ emantics of Nominals

157

In this section , I have tried to better characterize the behavior and properties of the complex types known as dot objects . There are surely many questions that remain unanswered , but it is clear that the nature of the denotation relation for dot objects is distinct from the unified types presented in 8.2.

8.4

Propositions

and Event Descriptions

In order to understand the position taken here with regard to event descriptions and norninalizations, it is necessaryto digress briefly about our assumptions regarding the treatment of tense, and the distinction between propositions and events in semantics. I will make some very particular assumptionsabout the relation between events, propositions, and facts. An untensed utterance such as Mary build a housewill denote a set of events, and becomesa proposition only by virtue of being "tensed" (cf. Verkuyl , 1993). A fact is a proposition generally judged to be known as true . An unsaturated proposition, typically denoting a propositional function , will be replaced by an event function , to be defined below. The way in which a sentencedenotes events is distinct from the way NPs are event-denoting. To make this clearer, let us compare the referenceto the event of Mary 's arriving , expressedas first a sentencein (36a) and as an NP in (36b). (36) a. Mary arrived at 3 pm. b. Mary 's arrival at 3 pm. Although both (36a) and (36b) ostensibly may refer to the same event, they do not expressthe same content about this event, since (36a) is an assertion of an event having occurred while (36b) denotesan event with out an assertion that it has occurred. In order to make this distinction clearer, we will distinguish the way in which eventsare quantified as NPs or sentences. In (36a), the event is tensed and thereby interpreted as a proposition. I will argue that only through tensing of an event-selecting predicate can the event in (36b) contribute propositional information ; (37) Mary 's arrival was at 3 pm. Our assumptions regarding the connection between propositions and events is essentially the following: both untensed sentencesand NPs

158

Chapter 8

may denote event descriptions , but the manner with which they denote is quite different , distinguished by their types . A proposition is seen as the result of applying tense to an event description . To this end , I assume that tense acts as a generalized quantifier over event descriptions , and behaves in many respects similarly to SPEC in an N P structure . The proper name of an individual object is typically construed as denoting a type e directly . Let the simplest cases of event reference also be an individual , eO ' . Both are types that refer to individuated

objects ,

either spatial (e) or spatia-temporal (eO ' ). Consider the proper named event The Vietnam War or Vietnam as in (38) below. (38) We will never forget Vietnam . Just as the proper name Nixon is typed e, and yet is understood to refer to a particular

;'"1dividual , there is no reason for us to treat

Vietnam

any

differently , in terms of presupposition or existential closure due to the name alone . Consider the following two sentences.

(39) a. I read a book about ~ !~Q!! . b . I read

a book

about

The

Vietnam

War .

Since books can be about anything , the preposition about obviously selects

for a type

" general

that

individual

is illustrated

covers both

e and eO".

Let

us call this

type

a

," g , where e ~ g , eO' ~ g . The way about predicates

in (40) below .

(40) '\g'\x [about(x , g)] (V ietnameO ") The

type

for a quantifier

such as every or a is now represented

as

shown in (41), where 9 is the general individual type: (41)

g ,t >,

g , <eU, t

, <eU, t

For example , the quantifier a will have the following denotation , where : F is a variable

of type
.

(42) [a] == AP;\: F..\e::3x[P (x) 1\ F (e, x)] The interpretation of NPs such as a woman or a war as generalized quantifiers is differentiated by a sortal specification on the lamb da ex-

pression as well as the "predicate" variable, F , as illustrated in (43)

159

The Semantics of N ominals

and (44) below, where : Fedenotes a sortally specified <e, <eO ' ,t -type variable for individuals , and : FeCT is a sortally specified<eO ' , <eO ' ,t >>-type variable for events. ( 43 )

AFeAe3x

( 44 )

AFeuAe23el

The

variable

tally

as

Let the

type

: Fe

us

type

description

Now

[V P ( e , x ) ]

' \ PP

( ' \ xAe

claim

about

of

type

[ Tnsa

)

=

, it

is

this

( 45a

higher

this

,

view

)

is

PI

, and

the

the

type

stan

-

subject

functional

) , where

mentioned

plays

more

only

in

P

is

above

is

the

.

Since

the

the

.

of

the

not

the

tense

as

to opposed

sec -

refer

that be

the

this

will

understand

, since

/ aspect

in , I

-

Given

tense

examples

meant

proposi

tense .

marker

, but

utilize

than

presence

is

a

secondary

relevant

, and

creating

abstract tense

discussion

languages

is

embeds

]

of

on

where

the

and

tense

treated

< eO" , t > , and

( 1993

-

ex -

to

this

is

a substantive to

aspect

( cf .

) .

tense

anch

Verkuyl

other

, 1980

that

which

purposes

how

is

proposition

rest

expository

, 1976

relation

( 46 )

the

is

predicate

tense

primary

a language

, in

sor

, < eO' , t

in

, i .e . , V P ' ( N

that

relation is

the

English

Assume are

, the

of

, with

- denoting

[V P ( e , x ) ] )

role

aspect

anchors

through for

which

fact ,

system

in

meant

the

verb

quantifier

AxAe

are

Comrie

subject

generalized

=

that

"

its

a

) expresses

=

/ aspect

" tense

to

a

expression of

) ; ( 45b

[V P ]

In

The

description

examine

P

event

eO' , < eO' , t

of .

[ V P ]

.

and

interpretation

a .

us

remain

and

.

projects

applies

languages

marker

, < eO' , t

P / ( V PI

redefined

reading

tion

it

will

- denoting

b .

some

tense

the

))

quantifier

individual

functional

the

let

tional in

what which

, which for

( 45 )

generalized for

it , i . e . , N

variable

( e2 , el

: F eO" , respectively

VP

to

) 1\ Feu

e , < eO' , t

and

analyze

lower

applies

a

variables

resulting

dard

( el

notating

of

given

( x ) 1\ : Fe ( e , x ) ]

[ war

specified

pressions

in

(woman

an and

AE3i3e

event

Kamp

as is

a

function

itself

of

within and

[a ( i , n ) A

over

type an

event

descriptions

eO" , t > , t > . The interval

Reyle

( 1993

anch

( i , e ) 1\

structure )

( cf .

(e)]

also

, as En

,

anchoring explored

<;. , 1983

) .

,

160

Chapter 8

To

illustrate

descriptions

in

ference

how

these

NPs

and

between

the

changes

bring

those

at

out

the

the

distinction

sentence

interpretations

of

level

the

between

, let

sentences

us

in

event

look

( 47

at

a )

the

and

dif

-

( 4 7b

) .

(47) a. A man arrived yesterday. b. A man's arrival occurred yesterday. Let

as

us

assume

given

and

in

( 48

Kamp

( 48

)

that

a .

[ PAST [ PUT

c .

[ P

RES

in

structure

the

3i3e

A

) ,

the

basic

operator

n

::; [n

i

1\

1\

tenses

as

~

:

the

in

used

( i , e )

i

1\

( i , e )

anch

1\

&

1\

in

English

are

Kamp

( 1979

)

( i , e )

event

1\

( e ) ]

for

structure ,

i .e . ,

argument

( e ) J

( e ) ]

representation

event

single

anch

anch

lexical

- headed a

achieved

~

[n

3i3e

where

right

now

[i

3i3e

=

for

) .

underlying

specifies

cation

is

A A

]

( 49

telic

= =

the

given

specified

n ( 1993

] ]

that

denotations

where

Reyle

b .

Assume

that

) ,

and

the

the

shows

an

it

to

achievement

and

a

verb

;

default

be

arrive

a

the

is

lexically argument

argument

of

prominent

,

the

lo

-

. .

arrIve

ARGSTR

=

[

D - ARGI1 ARG El

( 49

)

EVENTSTR

=

=

= y : locatlon x : ind

=

el

E2 = RESTR

e2

HEAD

QUALIA

Because

of

projection to

reduces

to

in < ind

[

,

the

in

lexical

case forms

quantification

)

a .

AxAe3y

b .

APP

semantic

( 47a

)

type

is

, < event

an

[ arrive

( AXAe3y

eventual , prop

of

right

the

( el

( 50

.

below

making

Hence ,

, where

and

the

reference

this

expression 3y

: loc

is

.

, y ) ]

[ arrive

( e , x

, y ) ] )

VP

,

)

]

qualia

role

in

)

is

those

locations

below

,x

subevent

FORMAL

unsaturated

function shown

_ act ,x ,y )

through

( e , x

: loc

the

oc

shown

over

: loc

the

mediated

this

]

e2

= arrive at ( e2

=

,

is

: state = < =

AGENTIVE FORMAL

headedness

syntax

subevent

restricted

The

event to

that

( 50

=

.

: process

< e , .

> ,

associated

which

is

with

equivalent

the

to

sentence

the

type

a

.

.

!UION }O S~1~UUUIdSdq~

-

[

SIUU

III1II

-

. [ (x "d)9: V (X)UVW ]XE'd'\(9.:{'\( (Z9) (9)3 V (9-'~)VJU 'DV U :5>~]dExE3Y = [pd] (tg)

162

Chapter 8

NP . That

is , both untensed

sentences and common

event nouns denote

event descriptions . As mentioned earlier , Tense functions of S , just as Det is the specifier of N P .

as the specifier

Now let us see how NPs refer to events in the simplest complicated

modes of reference will be considered

section . Consider a simple the sentence in ( 56 ) . ( 56 ) A war occurred

event denoting

case . More

below and in the next

nominal

such as war , as in

last year .

I will assume that an NP may denote an event in one of two ways ; either it denotes an individual proper name event , such as Thanksgiving or Mardi Gras , of type eO " ; or it denotes a quantified expression of an event description , such as every war , which shown in (57 ) below .

is of type

eO" , t > , <eO" , t

, as

(57 ) [ every war ] == A: FeUAe13e2 [war ( e2) /\ : Feu(el , e2) ] The qualia

structure

for war is shown in (58 ) , where I assume that

logical agent of the activity Krifka , 1989 ) .

is the join of the two parties

the

( cf . Link , 1983 ,

war ARGSTR = [ D- ARG2 ARGI = yx :ind ] EVENTSTR = [ HEAD el El = el=:process ]

( 58 )

QUALIA = [ AGENTIVE = war _act (el ,XEBY) ] The basic predicative individuals

force of the noun is to denote an event between two

, both of which are represented

as default

ar ~uments , and are

optionally expressed in the syntax . They must be logically represented , however , because of the interpretation necessary for NPs such as those in ( 59 ) .4 ( 59 )

a. the war between the U .S. and Vietnam b . France 's war with Russia c . the American

There

are , furthermore

war with , other

Vietnam

uses of nouns such as war where there is

an extend .ed sense of attack or assault , such as the war on drugs and the war on poverty .

The Semantics of Nominals

163

by Aver b such as occur, selecting an event ar gument, is represented the following expression :

(60) APeCTPeCT (Ae2Ael[OCCUr (el, e2)]) The

semantic

tree

given

descri

selection in

ption

on

( 61 ) ,

the

where

selected

by

subject

we the

have

verb

for

this

simplified

verb the

is

similar

typing

for

to the

the event

.

( 61 ) [ prop

/

/

/

]

/ - " --"

", , -

Tense

S

I [

event

[ < event

, prop

> , prop

>] /

/

[ event

, prop -' ~

> ]

" " '~

,

]

VP I

a

V I occur

The of

event

description

tensing

, which

however

,

is

is

the

proposition

( 62 )

in

the

a .

A

~ i ~ e2 [i

:::::; n

First

,

( e , el

a

is

by

only

verbal

sentential

anchored

by

predication

event

virtue

. The

description

result

,

, namely

a

1\ anch

::3i ::3e2 [ i

:::::; n A anch

3i

~

::3e2 [ i

A

n

section

man

however with

the

section

( i , e2 ) 1\ & ( e2 ) ] ( Ae3el

[ war

( el

) A

)]])

c .

structed next

position

available

with

b .

below

( 63 )

only as

subject

.

next

peated

itself

same

[ occur

In

the

1\ anch

, we

will

( i , e ) A Ae3 ( i , e2 ) A

return

::3el ::3el

to

[ 'war

[ war

the

( el ( el

) 1\ [ occur

) 1\

example

[ occur

in

( e3 , el ( e2 , el

( 4 7b

)

) ] ] ( e2 ) ] ) ]]]

above

,

re -

:

's

arrival

,

we

the

occurred

will lcp

.

need type

yesterday

to

.

examine

constructor

how into

complex dot

objects

events , the

are subject

con

of

164

Chapter 8

Finally , let us revisit the semantics of true complement coercion in volving event selection , in order to examine the role played by the mechanisms introduced here for tense binding into the complement . Recall

from 7.4, that sentencessuch as (64) require coercion on the complement to reconstruct the type selected for by the verb , as well as to provide an appropriate interpretation of the complement .

(64) John enjoyed War and Peace. The complement of enjoy is an event description , <e , E>. (65) S

/ / '/ [ human ]

'-'--"-"" ' "", -", VP

I

/ / / --'-"""" "",

Q

V

[ <e , E >]

AxAe[read(e,x,war_and_peace)] .

enJoy

[

I

War

e , E > , <e ,

ana

Peace

]

The formal derivationassociatedwith this coercionis shownbelow. ( 66 )

a . John

enjoyed [enjoy

' ( QT

c . AeT

[enjoy

' ( AX , e [read

e . John f . AeT

this

propositions

) ) ( John

) ( eT ) ] =>-

( WP

) ( x ) ( e ) ] ) ( John

{ AX , eT [enjoY

' ( Ae [read

[enjoY

section are

, I

' ( 3e [read

have

related

event

descriptions

as the

relation

briefly

.

to of

( WP

these

analysis to

and

) ( x * ) ( e ) ] ) ( x * ) ( eT ) ] } =>) ( e ) ] ) ( John

a

view of

have propositions

of tense Pustejovsky

) ( eT ) ] =>-

) ( x ) ( e ) ] ( x * ) ) ( x * ) ( eT ) ] } =>-

treatment

questions

events

issues

( WP

) ( John

the

Many

the

( WP

presented

through

between

consequences discussion

.

' ( AX , e [read

over

further

( WP

Peace

{ AX , eT [enjoy

such

I defer

and

b . AeT

d . John

In

War

) ( eT ) ]

of

how

tense

been

events as

left to

facts

as a generalized ( forthcoming

a

and

function

unanswered , as well quantifier ).

, as .

165

The Semantics of N ominals

8.5

Varieties of Nominalization

In Davidson (1967) and Parsons (1981), the connection between nominalizations and events was explored , and the motivation for adding an event variable to verbs as well as to nouns such as burning and arrival was supported by inference patterns that mirror the sentential struc tures associated with such nouns . For example , consider the pair of

sentencesin (67), where in (67a), two sentencesare connectedby a temporal connective, while in (67b) a nominalization is predicated in a way that results in a reading similar to (67a) . (67) a. When wood burns, it requires oxygen. b . The burning of wood requires oxygen . This pair illustrates similar information

how ing - nominalizations and sentences contribute for purposes of inferences . Davidson 's point was to

elevate the event to a first -order object in the logic , such that quantifi cation over such objects would be possible with event nominals such as the ing- phrase in (67b) above just as it is with sentences as in (67a) .

Higginbotham (1983, 1985), Verkuyl (1990, 1993) and Grimshaw (1990) extend this notion to the role that event quantification has on the syntax and examine

a broad

function

event quantification . One of the remaining difficult

of nominalization

types with

respect

to

questions in event semantics is to for -

mally characterize the linguistic devices which facilitate an interpreta tion of these sentences as making similar if not identical causal state -

ments. To illustrate this point , consider the sentencesin (68). (68) a. Charcoal requires oxygen to burn . b . The burning

of charcoal

: equir ~~ oxygen .

c. When charcoal burns, it Eequire~ oxygen. d . Every

burning

of charcoal

requires

oxygen .

There are several things to point out about these sentences. First , the

quantified expressionin (68d) is simply a more explicit wording of what is implicitly stated in (68b) and (68c). Secondly, the statement in (68a) brings out the functional dependency between the subject phrase char coal and the rationale clause to burn ; namely , that only together with the rationale clause does the subject satisfy the event selecting property of the verb require .

166

Chapter 8

Let

me

require

is

subject

illustrate polymorphic

( cf .

infinitival

) ) ,

( cf .

a .

Making

b .

The

c .

For

the

and

dry

point in

( 69a

VP

(6 9 )

this

or

( 69b

that

it

selects

a

split

allows

dive

further either

data for

.

an

construction

requires

require divers

-

some

Observe

event

with

how

directly

as

control

of

an

) ) .

this divers

with

to

suits

having

both make

both

wet

and

this

dive

wet

dry

and

suits

to

requires

dry

suits

make

.

this

( having

)

dive

both

.

wet

.

d . Making this dive requires the divers to have both wet and dry suits . Notice

how

the

constructions

( 70

)

( 71

)

While

to

a

.

*

b

.

The

synonym

.

*

.

Animals

need

c

.

*

can

states

of

temporal

overlap

need

)

affairs

g

! ! ~ ~

both

,

however

,

seen

does

not

allow

both

,

as

resulting

discussed

~

more

reference

in

a

in

~ . Q

like

.

and

to

dry

suits

make

this

.

dive

.

.

animals

event

wet

suits

.

breath

patterning

making

dry

oxygen

to

another

as

both

and

! ! ~ ~

of

is

to

having

wet

oxygen

breathing

event

be

~ gg

breathing

! ! ~

Every

verb

dive

n

Animals

an

sitions

this

divers

a

the

( 72

Making

b

relate

or

near

.

a

The

oxygen

want

verb

to

a

meaning

chapter

VeVxVl: loc [burn (e, x) /\ at (e, x , l ) oxygen(y)]

- verb

in

, 5

these

require

the

seems

generic

propo

quantification

like

5

.

over

following

,

-

events

where

0

is

:

- t 3e

' [ o

( e

, e

' )

1\

at

( e

' ,

y

,

1 )

1 \

I amnot concerned in this section, however , with justifying the relevance of events - which we havediscussed previouslyin any case-- as muchas establishingthe way that linguistic items makereferenceto them. To fully understandthe contributionmadeby nominalsto causalinference , we must first explainthe distinctionsin event-denotingtypes. Wesawfrom Section8.5 that simpleevent-denotingnominalssuchas war havea fairly direct representation in a qualia-basedstructure. What would the qualia structureand event-denotingpropertiesfor a nominal suchas burningbe, suchthat the readingsin (68) could be correlated ?

167

The Semantics of N ominals

Notice

that

this

event

variable

is not

in both

the

( 73 ) VeVx [burning Regardless simply it

of

is telling

The

the

the

, that

verbal

expressions

of

semantics

reference

statement verb

appears

( x , y ) ]]

, this

require

of the

- like

, as in ( 73 ) .

( y ) 1\ require

this

of the

an individual

a Davidsonian

move

and

sentences

what

in

in subject

( 69 )

position

that

is required

by the

verb

is to

of ing - nominals

of

event the

description

such

as burning

should

make

argument

as the

, as the

verb , while

ungrammatical

still

ref -

requiring

forms

in

( 74 )

.6

( 74 )

a . * The

burning

b . * Burning The

event

tice

that

cated

placing

.

same

expression

suggest

underlying

underlying

representation the

and

behavior

even when

description

to

merely

expressiveness

polymorphic

us about

( as in ( 69b ) ) , the

erence

nominal

overall

the

above . Namely

by

( e , x ) 1\ wood ( x ) - 4 jy [oxygen

the

ignores

an event

answered

description the

requires

requires

associated

AGENTIVE

by the

oxygen

oxygen with

argument

argument

structure

.

. burning

is illustrated

, x , is necessarily assignment

in ( 75 ) . No -

expressed

, as indi -

.

burning ARGSTR = ( 75 )

EVENTSTR

emphasis

the

initial

out

by

Asher

aspect

of the

a. b.

suffix form

event In

= burn

that

being

referred

this

is due to the

verbal

system

in .

above

seems to be on

to , an observation

appears

for left - headed

- ing , resulting

in the

_act ( el ,x ) ]

of the sentences

fact , this

, at least

( 1993 ) claims

nominalization

( 76 )

]

( 1993 ) .

in general

progressive

]

in the interpretation

Asher

nominals

=

= x :phYSobj

HEAD el [ El = el= :process

= [ AGENTIVE

QUALIA The

[ ARGI

to

be

events , as we shall aspectualizing

an interpretation

pointed

a property

of

ing -

see below

nature similar

of the to

the

.

168

Chapter8

Extending

to

Asher

agentive

such

as

baker

subevents

R

,

( el

of

' s

,

e2

)

,

el

and

e2

.

are

coercing

the

Space

preparatory

.

way

all

-

,

an

aspectualizing

.

much

Thus

as

event

,

the

the

' nominal

-

verbs

and

veT

nominalizers

,

by

where

a

kind

.

The

is

simply

of

can

complete

,

,

and

such

cut

are

also

over

completed

the

not

as

quit

operations

the

)

to

you

a

similar

denote

( 76b

aspectualizer

stop

perform

and

that

verbs

as

and

discussed

reference

fact

the

-

launching

denotes

abort

not

the

com

party

)

forcing

associated

to

as

( 76a

already

b

to

and

,

the

,

both

,

such

seems

R

of

headedness

such

denoted

The

seem

,

pair

by

the

examination

in

event

causatives

nominal

,

derived

events

NP

.

a

denotes

as

event

form

as

,

ing

other

the

the

suffixes

as

to

nominal

such

this

,

predicates

events

-

simple

that

well

ing

launching

argument

as

be

similar

interpretation

the

- or

seen

.

both

the

nominalizer

relation

is

the

that

-

and

eo

temporal

nominals

suggest

the

,

5

that

changing

of

event

this

ing

- er

would

,

chapter

to

the

the

an

fact

aspectualized

this

on

for

a

In

argue

might

suffix

'

be

might

,

terminate

,

nominal

)

event

to

at

is

identical

one

Following

ing

in

phase

)

, e2

might

predicates

( 76b

short

a

,

That

through

( ~

polysemous

Shuttle

the

the

in

as

Indeed

say

event

R

compare

with

process

. ,

view

well

below

i . e

might

suffixation

related

encountered

event

as

.

,

we

by

agentive

,

alternative

plete

event

initial

as

phase

of

the

.

As

plausible

scrutiny

as

from

.

Right

.

as

There

the

ing

that

be

relations

( ei

an

less

-

process

,

e2

,

,

where

operation

which

given

of

*

than

are

-

appear

)

and

to

withstand

arrive

of

the

the

have

die

double

- headed

one

to

;

of

suffix

are

unheaded

-

.

an

event

would

be

. ,

,

event

the

or

i . e

Hence

- headed

,

of

ill

is

.

accept

,

.

,

)

structure

application

left

result

said

and

less

transitions

destruction

nominalization

event

a

the

as

headedness

in

much

- headed

as

such

we

as

are

left

nominals

what

indicates

the

already

not

achievements

ing

result

resulting

acceptable

does

,

( such

the

subevent

) )

it

processes

nominalizations

point

since

,

. ,

are

of

ion

first

,

the

( R

than

transitions

nominals

head

( i .e

nominals

with

- headed

-

seems

:

interpretation

is

Regarding

ing

-

no

there

analysis

observations

transitions

ing

is

as

right

this

two

- headed

able

B

advisor

the

we

An

A

and

then

operation

,

created

externalization

war

observation

nominals

expect

- formed

to

ing

( R

we

might

as

effectively

( el

,

e2

structure

to

would

event

.

) )

expect

- denoting

This

seems

= =

The Semantics of N ominals

169

to be supportedby the contrastingbehaviorof arriving and arrival, as illustrated in (77)- (79) below. For somereason, the ing-form is grammatical only whensomereferenceis alsomadeto the culminatingstate of the event, as in (77b).7 (77) a. *The arriving of Johnwasgreetedwith mixedreactions. b. ?The arriving of John late wasmet with mixed reactions. (78) a. The arrival of Johnwasgreetedwith mixedreactions. - --- arrival -~~b. The of John late wasmet with mixedreactions.

(79) a. John 's arrivalwasgraceful . b. Thearrivalof thetrainis expected for3:00pm. Regarding not

the

actually

most

- ion

( 80 )

second

point

polysemous nominalizations

a . * The b . The

made between such

destroying destruction

( of was

above

, notice

that

and

result

process as

the

destruction

city

are

readings

are

, as

.

) was

widespread

ing - nominals

widespread

.

.

(81) a. *Theconstructing (ofthehouse ) hasadequate stability . b. Theconstruction hasadequate stability . Thewayin whichthe lexicalstructurereflectsthis singlesense , making reference onlyto the initial event,is illustratedin the representation in (82) of the semantics for the ing-nominalformconstructing . ing

(82)

El = el:process D-El = e2:state RESTR =
construct [ AGENTIVE FORMAL ==exist (e2,0 ~) ct(el' 0 , 0 ) ]

170

Chapter 8

In order to fully understand the inability of these nominals to carry the resulting event reading, we must look more closely at nominalizations which do exhibit this polysemy, such as examination, destruction, and arrival . Recall from chapter 6, that there are in fact three senses associatedwith these nominals, as demonstrated in (83). (83) a. The house's construction was finished in two months. b. The construction was interrupted during the rains. c. The construction is standing on the next street. I will argue that these process-result nominals are best treated as denoting dot objects, where both the dot elements are typed as events. This results in a class of event .event nominalizations, which are logically polysemous between process and result interpretations (cf. (17b) and (17c), respectively), but also admit of a third interpretation , that given by the dot object itself (cf. (17a)). The nominalized form for the accomplishment verbs derived by this processwill be logically polysemousin exactly the way that the aspectual system operates over their respective verbal forms. That is, (i ) the processreading in the nominal is the imperfective construction for the verbal; (ii ) the result reading in the nominal is the perfect construction for the verbal; and (iii ) the dot object reading in the nominal is the simple construction for the verbal form. To illustrate this parallel, consider the three sentencepairs in (84)- (86) below. Sentence(84a) and (84b) demonstrate how the force of the simple past is captured by the dot object nominalization sense; the sentencesin (85) show the past imperfective sensebeing coveredby the processnominalization sense; and finally , sentence (86b) illustrates how the result nominalization sense covers the present perfect sensein (86a) . (84) a. John constructed the roof frame for the houseyesterday. b. John's construction of the roof frame for the house was done yesterday.

(85) a. John was constructing the roof frame, when he fell from the ladder. b. John fell from the ladder during the construction of the roof frame.

171

The Semantics of N ominals

(86) a. Now that John has constructed the roof frame, he can start shingling. b. With the construction of the roof frame complete, John can start shingling. Now let us examin~ what type structure for these nominalizations would allow these three interpretations . I will argue that the polysemy exhibited by these nominalizations is formally the same as that seen with the nominals book, newspaper , and so on. Namely, two types are constructed to form a dot object with the lcp type constructor. In the caseof this classof event nominalizations, the types are both events, such that the dot object is itself an event. For an event el of type process and another event e2 of type state , the dot object is a complex type, e1 . e2, as shown below. (87) process -result _lcp = {process -result , process , result } To illustrate how this type is integrated into the representation of the nominal's semantics, consider the noun examination.

(88)

-

In some sense, the only thing different about this dot object is the typing on the dot elements . That is, whereas a book is a dot object composed of information and physobj , related by the FORMAL relation of con-

tainment (cf. 6.4 and 8.2), an examination is a dot object composedof process and state , related by the RESTRrelation in the event structure of precedence . It should be pointed out that the operation of headedness on the verbal representation is similar to the coercion operation selecting a dot element within the dot object of a complex event . We can , in fact , for purposes of discussion , assume that they are the same operation applying over different expressions .

172

Chapter8

As another example, consider the representation for the nominalization arrival shown in (90) , accounting for the polysemy exhibited in (89) below. (89) a. The party will begin after John's arrival . b. John's arrival was flamboyant.

(90)

Unlike with the ing- nominal arriving , the form above denotes a dot object , and carries the polysemy inherent with that typing . The type cluster associated with arrival is the same as that given above for the other process-result nominals, and a specific single dot element senseis arrived at by the headednessoperation mentioned earlier in chapter 8. For nominalizations such as construction and development, which are derived from verbs of creation, the result interpretation correspondseither to the individual which is created as a result of the initial process, or to the state itself. This is due to the fact that the FORMALrole of these verbs makes referenceboth to a state and to the object created. Verbs of destruction such as destroy, however, do not allow this sense, becausethe object is not predicated within the FORMALrole. This effect is demonstrated below with the verb developand the nominalization development. The full lexical semantic structure for a verb like developis given below in (91).

173

The Semantics of N ominals

-

(91) -

D- ARGl

The effect of the nominalizing

suffix -ment is to create a dot object ,

process .state , where in this case, the FORMAL quale predicates of the individual as well as of the state . Now consider some other typing possibilities available with the use of dot objects . Grimshaw (1990) and others have made much of the grammatical distinctions between examination , mentioned above, and nominals such as exam. What is it that gives rise to the differences between exam and examination in (92)- (94) below ? (92) a. The examination was long . b . * The examination was on the table .

(93) a. The exam was long. b. The exam was on the table. (94) a. The examination of the students. b. * The exam of the students. Grimshaw's analysis is partially correct, yet fails to observethat both these nouns are actually logically polysemous, in different ways, however. Notice that exam is polysemousin a manner similar to that of book. An exam contains an information object, such as that denoted by book, but specifically, of the type of question . This information object is part of the type of exam, along with the processof taking the exam. Thus, the exam is both the set of questionsand the event of the asking of the questions. This is the polysemy in the nominal exam, represented in (95) below.

174

Chapter 8

exam

ARGSTR

(

95

)

=

[

EVENTSTR

ARGI

=

QUALI

.o\

[

=

representation

the

set

the

event

have

questions

.

The

,

polysemy

of

In

,

of

that

they

are

while

the

dot

an

event

(

and

96

)

a

b

The

an

.

Mary

.

We

qualia

nology

is

get

structure

)

,

for

such

as

.

,

to

,

an

,

oral

can

exam

of

of

is

or

also

)

the

.

The

inherent

question

,

no

and

as

objects

(

can

event

denotation

reference

in

the

using

(

96

)

given

. 8

.

sonat

' s

as

both

sonatas

Bach

Dowty

be

.

to

Bach

the

that

artifact

object

after

-

is

an

make

for

until

-

is

shown

looking

event

acknowl

examination

to

,

the

to

physical

music

symphony

like

important

Exam

appears

event

and

refer

examination

behaves

it

.

with

concert

these

sonata

the

and

ways

of

the

. g

exam

,

Square

to

e

exam

event

Harvard

' t

can

objects

(

noun

type

in

won

exam

combination

Therefore

sonata

information

an

object

different

an

]

of

not

between

with

)

event

the

the

.

purely

)

, x

information

,

associated

x

.

difference

is

object

,

' Y lcp

_

information

need

of

very

e2

that

from

type

in

, z

(

any

but

structure

polysemous

el

the

symphony

major

(

like

itself

and

examination

The

the

qualia

sonata

ask

make . process

arises

examination

]

polysemy

,

,

in

the

terms

nominals

edge

=

compose

therefore

]

: process

=

questions

possible

event

el

the

manifestation

polysemy

: question

FORMAL

which

itself

physical

the

=

expresses

of

logical

El

x

AGENT question

[

This

=

,

~

1979

follows

.

termi

-

:

sonata

ARGSTR

=

EVENTSTR

(

97

[

ARGl

=

)

=

[

El

music

QUALIA

=

FORMAL

first

,

is

That

ment

performing

thing

as

to

well

that

as

which

is

,

notice

an

is

as

relation

,

.

the

the

event

and

the

(

e

(

'

,

the

types

book

are

el

'

, el

e

lexical

of

music

z

(

the

in

lcp

perform

type

types

-

compose

that

relates

two

]

hsten

=

information

specifically

just

]

: process

~

=

AGENT

The

: music

el

. process

TELIC

event

x

=

w

, x

"

, y

, x

item

music

. 9

directly

denotes

FORMAL

up

by

related

)

The

making

defined

are

)

)

the

a

by

dot

specific

the

an

relation

object

.

contain

relation

-

of

The Semantics of Nominals

Now are in

in

consider fact

nouns

such

logically sonata

dot

other

nominals

objects as

which

making

lecture

polysemous

175

appear

reference

, class

, and

behavior

as

to

lunch that

to

be

simple

a simple

. These

have

exhibited

events

events

. This

the

above

same

with

, but is seen type

exam

of and

.

( 98 )

a . Hurry b . We

( 99 )

( 100 )

up , Zac can

, your

talk

about

writing

a . John

is still

b . John

' s lecture

a . John

is late

b . John

is throwing

lunch it

l .ill ! ffi

tomorrow

lasted

for

is getting

during

his

over

' s lecture three

11 :00

cold

class

a barbecue

!

.

hours

. .

. brisket

party

for

his

class

.

Next consider the dot object created by an event type and a propo -

. 81 tion . This is the representation for lexical items such as belief and regret. belief

ARGSTR = [ARGl =x:prop ] D - ARGl

(101 )

= y : human

EVENTSTR = [ El = el:stateJ QUALIA

=

[state .prop _lcp ] FORMAL

= x

FORMAL= believe ( el ,y,x )

That is, the nominal denotes both the complement proposition and the

state of having this attitude , as illustrated in (102), where the lcp type constructor

has created

the new type .

. (102 ) 0: : state , 0: : prop lcp(o:) : state . prop

Notice that case assignment through the preposition in appears to be the only option available to a coerced NP complement with the nominalized form of the coercing

verb , but not to all nominalizations

(103) a. Mary believesthat John is sick. b . Mary believes the story . c. Mary :believes John .

(104) a. Mary 's b.illM that John is sick.

.

176

Chapter 8

b. Mary's ~ !M" in the story. c. Mary's .Q ~!! min John.

The important thing torealize with this analysis isthat the nominal belief is a dot object , and therefore denotes both the state of believing something as well as the proposition ofbelief itself . Now letus turn briefly tothe semantics ofpurported factive nominal izations .Asher (1993 )claims that allthe NPs in(105 )below are factive . The corresponding sentential structures are given in(106 ). (105 ) a.The collapse ofCarthage isafact . b.John informed Sue ofMary 'sdeparture . c. The collapse ofthe stock market isareal possibility . (106 ) a . That Carthage has collapsed isafact . b.John informed Sue that Mary has departed . c. That the stock market will collapse isareal possibility . Although it is true that these NPs carry factive readings , I would argue that the NPs themselves are not factive outside of these selectional environments , but simply event descriptions admitting a single inter pretation as result event nominals . The factive interpretation exhibited above in ( 105) is due to the coercive property of factive predicates such as inform and be a fact rather than the inherent semantics of the event nominal itself . There are two types of data suggesting this is correct . First , these event nominals appear in other contexts without factive in terpretations , as demonstrated in ( 107) . Secondly, factivity is an inter pretation that many NPs assume when placed within a factive coercive environment such as the complement of inform and as subject of be a possibility , as illustrated in (108) . That these NPs are not necessarily factive seems to correlate with the lack of tense/ aspect anchoring , behaving , in effect , as irrealis expressions . ( 107) a. The collapse of the stock market was prevented by government intervention . b . The thief 's departure was stopped by the police .

(108) a. Cathie informed me of your blister . b. A thunderstorm is a real possibility.

c. John informed hisparents ofhisscholarshiE .

177

The Semantics of N orninals

The

semantic

selection

predicate

(

which

109

structure

coerces

its

below

shows

complement

to

how

a

inform

factive

is

a

factive

interpretation

.

)

S

/

[

"

human

'

/

- - - - - "

"

"

"

"

"

]

"

VP

/

a

/

- /

"

"

"

~ "

"

V

[

[

inform

Following

Asher

tion

for

the

coerced

NP

is

110

,

1993

)

/

I

as

well

8

.

1

. 3

(

i

e

.

)

realize

)

departure

>

,

reveals

]

the

resulting

that

a

interpreta

the

manner

The

!

have

\

,

factive

to

sense

-

reading

similar

factive

1993

)

tried

.

in

true

is

comple

context

-

for

this

the

(

f

)

!

,

but

one

\

f

is

=

depart

that

(

this

arising

e

,

is

mary

)

not

through

a

]

reading

coercion

(

cf

.

10

to

the

fact

however

,

,

as

,

semantics

captures

,

109

' s

factives

in

anch

to

' s

,

lexicon

processes

\

Zucchi

section

(

]

,

.

n

NP

and

this

mously

in

in

110

the

of

description

point

in

Mary

treatment

above

: : ;

important

erative

give

a

general

polysemy

perspective

inherent

inability

of

other

on

in

many

nominals

how

a

gen

-

nominalization

to

behave

polyse

-

.

.

In

6

Lexicalization

this

and

section

,

icalization

I

theory

semantics

.

explored

,

sions

the

(

[ i

inherent

In

)

event

in

3i3e3f

Asher

1993

discussed

given

)

The

8

the

coercion

NP

(

shown

onto

ment

(

' s

of

fact

<

and

syntax

.

the

the

the

consequences

notions

characterization

is

expressions

what

given

The

as

explore

LCPs

of

of

correspondence

with

how

between

similar

structures

are

lcp

and

meaning

maps

lexicalized

,

generally

qualia

derived

for

structure

to

semantic

compositionally

lex

in

lexical

form

-

the

is

expres

-

in

178

Chapter 8

From our discussion above, we can think of logical polysemy as resulting from a complex object, which is constructed from other obje.cts with a type constructor, which we called lcp. The type cluster associated with a lexical paradigm (lcp) allows the realization of a number of senses, but is itself a functionally defined structure. Given such a system, the process of conceptual lexicalization is affected, such that two distinct situations may arise within a given language. For any possible lcp, IcPi, constructed in the type system, the languagemay expressthis paradigm through the unique expression of a single lexical item , or it may split this lcp into distinct lexical forms. The former case is illustrated schematically in (111), where the type cluster { tl , t2, tl . t2} is associatedwith the word Wi.

(111)

The second possibility is illustrated in ( 112) , where the dot object does not actually result in a logical polysemy in the lexical system for the language .

(112)

This distinction allows us to capture a number of interesting lexicalization phenomenain language, particularly with cross-linguistic lexical comparisons (cf. Pustejovsky and Busa, forthcoming). For now, let us

179

The Semantics of N ominals

restrict our discussion to English . Compare the nouns sale, purchase , and transaction , and how they relate to the dot object . The result of the lcp type constructor shown below represents the type of the nominalization transaction , independent of the financial aspect associated with selling and buying .

(113) ?_: give a : take lcp(a ) : give . take Notice

that

this nominal

is unable to refer to the individual

types , each

representing a unilateral event of transfer , but only to the dot object it self . Thus , transaction appears to be a case of split lexicalization within the lcp of this type cluster , since it cannot make reference to the in dividual events . The other lexical items in this split lexicalization are the nominals purchase and sale, which do refer to the individual events , while

still requiring

reference to the entire type cluster . Because of this

split , there is no logical polysemy associated with these nouns , in terms of the process-result ambiguity . This is not to say that these nouns cannot enter into logical polysemies . Just as the noun exam is not pol ysemous with respect to one type polysemy which examination is (i .e., process-result ) , it does exhibit the process-object polysemy , as discussed above . Interestingly , the noun purchase seems to be logically polysemous between the event and the object involved in the transaction , while the noun

sale is not . Hence , the former

behaves in some respects

like the

dot objects mentioned in the previous section , while sale seems to denote only the event .

(114) a. The government prohibits the purchas~ of illegal drugs. b . The purchases I made today are in the back of the truck .

(115) a. The ~ b . * The Boston

of the house this month seemsquite likely now. ~~

I made today are already distributed

throughout

.

This may be related to the distinction we saw between the physical 0 bject sense associated with the result reading of creation -verb nominaliza tions such as building , construction , and creation , which is absent from the nominalizations for verbs of destruction . The lexical representation

for purchaseis given in (116) below.

180

(116 )

Chapter 8

purchase ARGI = x:human ARGSTR = ARG2 = y:physobj D-ARGI = z:human El = el:process EVENTSTR = E2 =e2 RESTR =:state
In closing our discussion of lexicalization , it is interesting to compare how the semantic field around a particular lexical item is structured relative to other words in the same language . For example , while nouns such as apple, lemon , grape are polysemous between the product and the producer of the product senses as in (117) , there are many words that have only one sense or the other , as in ( 118) .11 (117) a. We have planted apples in the field behind the house. b . There are apples in that basket .

(118) a. The !~~~ are growing tall . b. *We have collected E!~~~ for the wreath. (pine cones) The tree and the fruit contained on this tree are both logical senses for the word apple, while pine is simply unable to refer to pine cone. It is the structure behind these concepts which gives rise to the polysemy . That is, an lcp relating product and producer for food items simply does not apply to pines , given our current dietary habits . If pine cones were to be edible or eaten , then the semantic

conceptualization

between the tree

and the cone would change as well , possibly permitting the polysemy as a result . It is possible that factors relating to suppletion phenomena may also be at play with sense assignment in such cases.

8.6.1

Referential

Transparency

To close this section , I would like to briefly explore the role that qualia structure might play in the modes of reference for an object . In the classification

of nominal

types , there are several standard

distinctions

made

in the way that NPs refer . For example , the distinction between pro nouns , proper names, and descriptions is important both for anaphoric

181

The Semantics of Nominals

binding

phenomena

these

do

,

a

( cf

however

follows

to

,

refer

at

the

)

the

expressed

this

for

1981

obvious

qualia

differences

structure

value

of

or

the

referential

in

permits

transparency

the

as

interpretive

What

relative

in

property

.

noun

us

opacity

' s

of

qualia

transparency

,

to

how

structure

and

.

define

it

as

:

( 119

)

The

REFERENTIAL

of

the

( i )

If

( ii

)

an

If

of

NP

it

is

is

an

it

TRANSPARENCY

specificity

then

only

is

of

"

in

the

reality

,

the

of

course

,

are

semantically

are

( i . e

such

quale

what

what

At

bread

,

these

that

which

carry

specific

distinct

.

,

modes

the

hearer

with

of

.

predicates

)

a

,

.

Ihe

such

lecturer

is

Namely

,

at

enjoy

pm

as

.

.

are

of

and

explanation

structure

my

for

.

understanding

is

,

baker

for

,

the

example

,

way

in

is

qualia

therefore

concepts

has

in

more

seen

)

clearly

in

below

composition

.

and

which

the

objects

these

.

lecturer

by

specified

( 120

not

which

,

transparent

in

~

In

information

as

when

,

types

himself

qualia

to

are

not

irltu

,

,

mode

these

clearly

: 00

The

referentially

words

3

no

.

referring

most

began

is

she

knowing

of

,

the

these

-

the

course

established

are

and

as

such

have

these

of

in

elements

individual

of

words

illustrated

,

nominal

require

that

virtue

they

is

relational

nouns

con

captures

the

specific

not

of

by

the

This

such

does

we

of

begin

then

strongly

and

there

are

knowledge

as

,

,

fall

pronominal

requirements

for

is

qualia

qualia

or

does

denotes

role

explanation

This

.

of

he

typing

.

its

predicates

term

of

devoid

use

its

weakly

Pronouns

spectrum

our

of

of

transparency

that

only

The

and

qualia

the

,term

AGENTIVE

the

are

the

of

in

of

the

consequences

the

,

understand

satisfy

they

more

( 120

we

objects

of

item

case

this

lexically

or

end

characterization

values

values

of

.

NPs

reference

TELlO

other

but

the

general

how

the

the

the

in

the

hierarchy

,

range

in

suggesting

are

Knowing

a

opaque

)

pronominals

of

is

,

the

informativeness

except

weak

referentially

. ,

type

the

,

an

where

sense

there

categories

on

global

between

absolute

by

for

,

technical

distinction

by

a

.

.

qualia

course

is

NP

.

constrained

the

of

NP

that

constrained

strongly

,

an

for

transparent

structure

" reference

itive

weakly

referentially

depends

lattice

values

opaque

whether

strained

of

qualia

referentially

NP

is

Determining

it

as

,

look

as

to

well

Chomsky

is

refers

will

.

,

noun

I

as

types

that

to

182

Chapter 8

b. Maryenjoyed the cake . Although certainly a defeasible inference, the referential transparency of the subject in (12Ga) suggeststhe activity that is begun, while a similar transparency of the object in (12Gb) suggeststhe manner in which are

the

cake

is enjoyed

. Both

pronouns

in

respect

, as demonstrated

uninformative

this

and

proper

names by

the

, however sentences

, in

( 121 ) .

( 121 )

a . ~ ~~ ~~ began b . Mary

As

we

will

determine

see the

constructed determine course

in

way

3 : 00

10 , the

interpretation a coercive that

pm

.

lie

chapter

default within

the .

at

enjoyed

individuals

degree of

environment are

of

transparency

how

a phrase .

described

This

of

an

NP

is semantically parameter

and

referred

re -

will to

will

also in

dis -

9

The Lexical Semantics of Causation

In this chapter, I explore how extended event structure and qualia structure interact in the semantics of verbs to account for the polysemy associated with certain verbal alternations. In particular , I will show how the theory of event headednessoutlined in chapter 5 gives us a constrained mechanism for lexically underspecifying the semanticsof verbs exhibiting polysemy in alternation classesinvolving causation. The representation responsible for the polymorphic behavior of these verbs is a fairly basic statement of event causation, called the Default Causative Paradigm. In section 9.1, I demonstrate how both causative and unaccusative forms for verbs such as sink and breaksharethe sameunderlying semantic representation, but project to distinct syntactic environments becauseof event headednesson this default causative expression. Analogously, aspectual verbs such as beginand start, which have both raising and control readings are analyzed in section 9.2 as logically polysemous in a similar fashion. In many respects, these alternations are formally similar to the behavior of dot object nominals such as door and book, which was discussedin previous chapters. Section 9.3 briefly examines the event structures associatedwith non-lexical (synthetic) causatives for comparison to the lexicalization phenomena discussedin previous sections. In Section 9.4, I show how the semantic representation of experiencer predicates is based on an underlying causal relation , similar to those discussedin 9.1 and 9.2, which is responsible for the syntactic binding violations associatedwith these constructions. Finally , in 9.5, I explore a limited class of verbs I will call modal causative, such as risk and gamble, and their associatedsemantic and grammatical properties.

9.1

How Language Encodes Causation

Our languageof causalrelationsis only asrich asthe descriptivemechanismswe employfor encodinghow objectsand eventsinteract in the world. Much of the groundworktowardssucha descriptiveformalism for causalreasoninghas beenelucidatedin the AI and computational linguisticscommunity. For example,Wilks (1978) developsa taxonomy of causalrelationsasmotivatedby the reasoningnecessary for language understandingsystems . Allen (1984)', Eberle (1988), and Kowalskiand Sergot(1986) eachexplorewhat possibletemporalconnections canexist

184

Chapter 9

between

propositions

taken

as individual

events , in order

to structure

an event-based logic for planning or reasoning. Kamp (1979) and van Benthem (1983) are both seriousexplorations addressingthe sameissues from the perspective of philosophical logic. Finally , Hobbs et ala (1987) presents an extremely rich language exploring not only possible causal connections , but also constraints on actually realized temporal binding between Most

events . work

from

formal

semantics

has

tended

to

take

a more

con -

servative approach to describing how events , as described in linguistic expressions , are combined

to facilitate

causal inference . Recent work in

Discourse Representation Theory (DRT ) (cf . Kamp and Reyle , 1993,

and Asher, 1993) has concentrated on event quantification in discourse, which

makes

no

use of internal

subeventual

structure

.

Hence , lexical

causation has not been explored in great detail in these frameworks (cf . however , Kamp and RoBdeutscher , 1992) . Parsons (1990) explicitly mentions how certain verbs encode causation , but limits his inquiry to a small class of causative predicates , without examining the issue more generally throughout the language . The approach taken in this work has been to build explicitly and extensively from the work mentioned above, in order to examine whether there is a general methodology for determining what information should be encoded in the lexical item itself , and furthermore , why . In other words , the goal is to study not only what causal relations are possible , but which relations are justifiably represented in the semantics of natural languages . In some sense, the entire structure of a generative lexicon is organized around this strategy ; namely , to provide an account for why certain causal relations and explanatory modes are lexicalized in the ways they are, as well as how they project to the particular environments they govern .

syntactic

One point that deserves mention is that there are identifiable anQ formally characterizable constraints on the representations which encode causal relations . Given that

we are concerned

causation , and not its general

form , there

with

the lexicalization

are several conditions

of that

must be met within a lexical expression . Capturing the necessary and sufficient conditions on causation in the physical world are, of course, beyond the scope of this work , and , fortunately , irrelevant for much of

this investigation. Both Bach (1986) and Chomsky (1994), echoing remarks made by Reichenbach (1978), argue that the conceptualization

185

The Lexical Semantics of Causation

of

the

world

tics

of

through

for

guage

lexicalizes

remarkably

will

satisfying

1

)

a

b

c

d

For

least

.

conditionalness

.

one

.

invariability

.

uniqueness

.

productivity

e

2

)

sided

the

to

construe

(

el

it

(

nor

(

,

2

)

the

la

)

.

pletely

must

the

this

same

property

lb

,

causal

this

statement

approach

event

best

are

e2

:

is

always

pro

-

in

individual

,

argument

and

of

view

,

be

.

in

order

Furthermore

into

,

the

semantics

productivity

these

constraints

seem

causing

role

to

and

i

overlap

some

way

at

,

least

coherence

5

. e

causal

.

,

be

the

consecutive

.

the

oc

,

be

must

cx :

and

cx :

of

event

)

.

a

can

Secondly

item

events

-

a

event

,

quale

lexical

re

exist

resulting

<

AGENTIVE

resulting

that

two

<

there

explanation

the

the

such

,

can

,

and

chapter

the

(

First

event

in

event

properties

.

in

any

over

as

relation

directly

uniqueness

the

precede

participants

must

.

causing

or

semantic

other

encoded

described

play

the

each

causal

point

the

cohere

of

to

linguistic

would

,

then

coherent

)

between

is

only

mention

a

,

relations

precede

events

(

following

two

That

of

relation

and

at

three

first

item

no

metaphysics

the

the

makes

as

relation

Given

addressing

a

illustrative

and

their

indirectly

precedence

is

then

invariability

naive

as

traditions

adopt

:

below

expression

items

stated

)

,

what

this

a

Analytic

typically

.

in

From

1

happens

by

represented

from

seem

;

statement

event

lexical

principles

-

.

conditions

of

lan

;

events

while

how

conceptualizations

properties

dependence

statement

in

following

-

of

;

duced

The

.

,

on

embody

causal

nature

nevertheless

seman

view

;

,

If

the

-

example

(

,

the

constraints

or

our

of

accepted

the

reflect

,

model

currently

,

not

anything

in

causation

at

(

If

a

the

example

Aristotelian

of

reflect

For

causation

issue

to

.

.

within

not

concepts

mechanics

be

of

need

those

quantum

the

structures

language

reality

to

linguistic

natural

is

.

We

,

.

only

lexical

com

,

predicated

will

refer

-

the

186

Chapter 9

(3)

ARGUMENTCOHERENCE : The relation expressed by the causing event and that expressedby the resulting event must make referenceto at least one parameter in common. This reference can be direct or indirect :

The

lexicalization

in

the

of

event

dence

of

tree the

accounts

causing

for

move

,

and

headed

a so

5

variety

build

,

is

of

,

as

( see in

reference

event

on

events

chapter

direct

structure

where

to

the

resulting

of

as

below

more

as

in

, die

semantics

bake

.

,

and ,

prece

present

as

arrive

of

Pustejovsky

shown and

both

such

,

causation

the

fully

are

verbs

such

schematically coherence

event

Constitutive

with

is

argument

state

verbs

) .

connection

discussed

( 4 ) ,

change

well

causation

in

kill

briefly verbs

( forthcoming

drop

,

right

discussed

creation

,

other

-

This

in

such

as

) .1

( 4 ) e

Cl /

/

/

x

/

---""

"

" ,'

e

2

, R

As

ing

argued

in

the

and

( 5 )

5

which for

the

This

says

two

individuals

that

[P

a

7

lexical

( e2

"

that ( the

P

above

item

[a

, Y ) ]

:

el

Q'

is

AGENTIVE

qualia

to

the

can

sort

be

ally

seen

as

defined

structur

-

arguments

.

AGENTIVE

1\

some

( . . . , y , . . .)

, the

reference

predicate

such about

and

. 3R =

, y , . . .)

make

AyAxAelAe23P FORMAL

" bringing

chapters

relations

subevents

( x

r

<

a

= cx:

e2

[R

relation

, x

, y ) ]

between

relation role

( el

1\

]

exists )

of

the

two

between resulting

subevents

x state

and

and

y of

in y

the ( the

187

The Lexical Semantics of Causation

FORMAL

role

tree

)

,

where

structure

derspecified

,

The

this

in

( 4

)

however

,

representation

( DCP

)

is

Our

here

a

is

to

of

the

tations

such

from

as

binding

.

is

between

)

a

in

( 5

)

,

the

subevents

,

( 6

of

as

various

well

,

,

.

is

.

of

and

one

to

or

more

-

-

constrained

only

map

also

explicit

in

the

to

the

elucidate

statement

temporal

restrictions

syn

represen

should

specific

-

also

admissible

not

but

in

but

respect

not

A

both

,

semantic

should

sortal

-

type

item

filtered

the

their

simplest

with

richer

are

forms

illustrating

as

,

which

verb

un

Paradigm

that

verb

type

are

structure

arguments

effectively

a

,

the

6

be

event

)

. 2

the

chapter

the

lexical

expression

semantic

)

the

a

semantic

to

these

in

embodies

between

need

in

The

( 5

Causative

forms

expressions

shown

,

.

in

head

with

in

before

Default

syntactic

mentioned

between

DCP

of

syntactic

event

associated

at

hold

structure

the

( 1995

be

not

no

accompanying

connection

the

did

behavior

As

possibilities

the

)

as

connection

Furthermore

syntactic

y

is

to

arrive

that

to

grammar

there

grammatical

.

'

qualia

Busa

the

realization

( e2

referred

number

regarding

predictive

,

can

in

aim

tactic

of

that

realizable

formative

that

)

P

equivalent

and

relation

that

( 5

Pustejovsky

causative

,

the

in

in

in

state

and

restriction

.

a

EVENTSTR

=

El

=

el

E2 RESTR

=

e2

: process

HEAD

ARGSTR ( 6

=

)

[

QUALIA

=

As

I

will

in

least

( 7

)

=

12

=

~

)

a

,

natural

three

.

-

ments

b.

RIGHT

.

as

kill

die

sink

arrive

break

,

L

~

,

e

.g

,

:

e

. g

event

. ,

etc

0

)

for

associated

murder

Direct

]

most

lexical

headedness

with

causative

forms

,

this

of

there

are

structure

;

accomplish

-

.

. ,

Direct

causative

achievements

,

.

EVENTS

,

~ '

classes

:

,

( el - lcp

of

EVENTS

,

( e2 _ act

Because

EVENTS

such

as

a

accounts

semantic

HEADLESS

as

paradigm

- HEADED

such

c.

]

aJesult

this

HEADED

,

cx :

]

= - causative

language

lexical

LEFT

=

AGENTIVE default

demonstrate

causation

( at

ARG2 ARGl

FORMAL

[

: state = < =

:

burn

.

e

.g

. ,

causativejunaccusative

verbs

,

such

188

Chapter 9

I argue in 9.2 and 9.3 that it is the semantic underspecification

occa-

sioned by a lexically headless event which gives rise to the polysemy

exhibited by the predicates in the causative/ unaccusative alternation and in the raising/ control alternation . 9.2

Causation

and U naccusativity

Let us now see how the underspecification mentioned above gives rise to the verbal polysemy associated with unaccusativity . The purely syntactic characterization of split intransitivity in language fails to capture the

semantic

relatedness

between

the

causative

and

unaccusative

forms

for the verbs in any systematic way. In fact , these analyses can be

seen as employing senseenumerative lexicons (as described in chapters 3 and 4), with all the shortcomings of that approach. The work of Van Valin (1990) , Zaenen (1993), and others has illustrated the problem in failing to link "unaccusativity " to the lexical semantics of the elements

in the sentence more intimately (cf. also Abraham, 1986); yet Levin and Rappaport (1989, 1995) have explained the syntactic behavior of these verbs in terms of fixed -form , lexically determined verb classes. Pustejovsky and Busa ( 1995) argue that a description of the behavior of unaccusatives between

the

in terms of fixed classes does not capture constructions

involved

in the

diathesis

the relatedness

alternation

or in the

unaccusativejunergative alternation of the same predicate. As a result, the sense enumerative lexical approach is not able to explain how unaccusativity can also arise non-lexically , through composition with other elements

in the sentence . Within

the framework

presented

here , how -

ever , where the number of senses remains fairly constant with respect to the possible space of interpretations , a set of generative devices allows us to determine semantic selection of complements by the rules of composition . These rules operate over underspecified lexical representations , such that

the configurational representation

the behavior

of a predicate

can be predicted

from

properties of different parameters in the semantic

.

First we examine briefly the behavior of unaccusatives in Italian ,

where the data suggest that the causativejinchoative alternation can be analyzed as a systematic form of logical polysemy . We will see that

the alternation shown below in (8) with the Italian verb affondare (sink)

189

The Lexical Semantics of Causation

requires neither a multiple listing of the entries (as in Levin and Rappaport , 1992, 1995, nor lexical rules as in Copestake, 1993), and Sanfilippo, 1993). (8) a. I nemici hanno affondato la nave. " The

enemy

b . La nave " The

sank

e affondata

boat

the

boat ."

.

sank ."

The characteristic behavior of verbs such as affondare is that , when occurring as intransitives , they assume the auxiliary selectional properties

of unaccusatives such as arrivare (arrive) , and are distinguished from unergative verbs such as camminare (walk). Chierchia (1989) suggests that the lexical representation for unaccusatives is in fact an underlying causative .3 Pustejovsky and Busa ( 1995) extend this notion by combining it with the concept of under specified event structures and argue that those unaccusatives which also have causative counterparts are logically polysemous because of the headless nature of the event structure

representation of the predicate, e.g., as with affondare in (12). As I will demonstrate

, whether

the

verb

surfaces

as an unaccusative

or

a tran -

sitive causative will be determined by which subevent in the semantic

e < o c el " / " " " e 2 [s_ a,(xink ,y)]ct [sI(y)]unk

representation is headed . For affondare , the unheaded event tree struc ture

is shown

(9)

in the

tree

below

.

Compare this structure to a lexically right -headed predicates such as arrivare (arrive) and morire (die), which have only unaccusative realizations, as demonstrated in the sentencesin (10) and (11). (10) a. II pacchetto e arrivato . "The packagearrived." b. *II postino ha arrivato il pacchetto.

190

Chapter9

" The

c.

(11) a.

* Il

mailman postino

ha

" The

mailman

Gianni

e

" Gianni

b.

* La * " The

arrived fatto

made

il

."

pacchetto

arrive

the

. package

."

. ."

pallattala

ha

bullet

package

arrivare

has

marta died

the

marta

died

Gianni

Gianni

."

."

.

I

Recall from chapter 8, that the semantics for the English verb arrive includes a specification that the resulting state in the transition , e2, is the head. This is true of the Italian verb arrivare as well. In fact, all lexical unaccusativeverbs will be analyzed as right -headedbinary event structures, meeting the following constraints on event structure : Q

(12)

EVENTSTR=

El

= el : prOCess

E2

=

RESTR HEAD

,

e2 : state

= < DC = e2

For arrivare in particular , the lexical structure is given in (13) (cf. chapters 6 and 7) .4 .

arrlvare

(13)

ARGSTR

=

QUALIA

=

D-ARGl = y:location [ ARGI = x:ind El = el:prOCess EVENTSTR = E2 = e2=:state RESTR X HEAD= e2

AGENTIVE [FORMAL =

The status of the location argument, y, as a default argument, obviates its obligatory expression in the syntax, while still allowing such constructions as that in (14). (14)

II pacchetto e arrivato a casa. "The packagearrived home."

According to the mapping principles outlined in c.hapter 6, the realization of such a lexical structure is dictated by the effect that headedness

TheLexicalSemantics of Causation

has

on

argument

associated structure owed

. ,

the

The

in

arguments

below

.

( 15 )

( 16 )

:

R ( ei

:

P ( e2 , y )

Qi

:

R ( el

Qj

:

P ( e2 ' y )

a lexically

structure

.

with with

6 .2 .5 ) .

only

arguments

expressed their

at

arguments

,

quantificational

This

is

surface

are

shad

closure

repeated

in

( 15 )

-

over

and

( 16 )

, y : OBJ

~ shadowed ~ y : SUBJ

unaccusative

there

is

" deep This

along

~ x : SUBJ

that

obligatorily

~ shadowed

, x , y )

the

, recall

are ,

in

, x , y )

, and of

particular

interpretation below

- determined

unambiguous

tive

an

Qj

In

event events

( cf .

Qi

expression

.

headed

headless

resulting

these

For

expression

with

191

but

one

object

is

such

"

illustrated

as

syntactic

, the

realization

argument in

amvare

as

the

tree

mapping

possible

the

subject

below

.

in

is

, namely an

intransi

-

( 17 ) eo

_/

/

~" / -"

" " "

el

e2

I R ( el

I , x )

P ( e2 , x )

!

!

SHADOWED

What tion the

is is initial

important

not

possible subevent

to

note

with in

( X : SUBJ

is

that

morire the

event

, although ( die

)

and

structure

a

)

direct

arrivare above

causative ( arrive is

still

construc

possible

certain adjunct phrases, as shown in (18a)- (18d). (18) a. Gianni e marta per una polmonite . " John died from pneumonia ." b . II tetto e crollato per il peso della neve. "The roof collapsed from the weight of the snow ." c . Maria e arrossita per l 'imbarazzo . "Mary blushed out of embarrassment ."

-

) , reference

to with

192

Chapter 9

d . Gianni e annegato per il maltempo . "John drowned from bad weather ." The PP per una polmonite in (18a) is not a true adjunct , but is in fact adding specificity to the relation in the event associated with the AGENTIVE Quale for the verb . Schematically , this is represented by the mapping given in (19) : below .

(19)

As suggestedabove, whereaslexical unaccusativessuch as arrivare are right -headed underlyingly, the polysemy possible with affondare (sink) is due to an underspecifiedevent structure . The lexical form for the verb affondare is given in (20) below. affondare

[El=process 1 E2 = state RESTR = < 0(

-

(20)

-

By its headlessnature, the lexical entry for affondaremakesavailabletwo grammatical constructions, the unaccusativein (12b) and the causative in (12a). The realization as an unaccusativegives rise to a right -headed event structure , by foregrounding the resulting state of the boat, and consequently shadowing the arguments and the subevents in the agentive . Realization as a causativegivesrise to a left-headedevent structure ,

The Lexical Semantics of Causation

193

by foregrounding the AGENTIVEpredicate and projecting all arguments therein. In other words, given an unheaded event structure a we can expressa via anyone of its roles, namely either through the AGENTIVE , or through the FORMALrole. As discussedin chapter 5, this strategy for projecting semantic arguments from underlying forms is needed in order to filter the expressionof "unheaded" arguments. In other words, given the presenceof more than one qualia role, individual qualia "compete" for projection , and the mechanism of headednessacts as a filter to constrain the set of projectable arguments. The quale for the headed event, e*, projects the configuration (or template) associatedwith that quale. In chapter 5, I described how the qualia of a lexical expression must be "saturated" by the syntax. This condition is repeated below in (21), along with the definition of covering.5 (21)

(22)

QUALIA SATURATION: A qualia structure is saturated only if all arguments in the qualia are covered. COVERING : An argument d is coveredonly if : (i) it is linked to a position in s-structure; or (ii ) it is logically dependent on a covered argument {3; or (iii ) it is closeableby virtue of its type .

The realization of an unaccusative takes place by heading the culminating event and by abstracting the only argument in the relation associated with the headed quale, namely the "deep object ," La nave. This is shown below in (23):

194

Chapter 9

(23)

,e] S AXAeO 'AAeS [ayondata ff: A = SHAD /\ F* = sunk (eS,y )]

A = sink_act (eO ",x,y) AF* = sunk(eS,y)]

Projection of theleft-headed eventgivesriseto a causative transitive construction , asillustrated in thetreein (24) below . (24)

S ...-,/ / / "'-"--" " '-" NP [eu*

. '1 '1., nem'Z c.,

[eO ' AXAeO ' AYAes [af f ondato : la nave A = sink _act (eO ' ,x ,y ) 1\ F = sunk (eS,y )]

What these two structures illustrate is how event headednessacts to constrain the form of the semantic representation and its syntactic realization . The strategy within this analysis is to allow for gross potential overgeneration of syntactic mappings from semantic expressions, and constrain the output (or restrict the licensing) of generated forms. This

The Lexical Semantics of Causation

195

is essentially the strategy employed by Chomsky (1986) , where , without constraints , the statement on how an expression is formed would appear much too general to actually match the data . Constraints or parameters on the applicability of the strategy , however , restrict the expressiveness of the rules . The analysis thus far has illustrated how both causative and unaccusative forms of this verb class can usefully be derived from an under lying causative representation . Following the discussion in Pustejovsky and Busa (1995) , let us briefly examine the conditions under which defaulted and non-headed arguments are realized as adjuncts rather than as direct arguments . Chierchia (1989) and Roeper (1987,1993) have discussed the conditions under which the agent (under normal thematic interpretations ) is accessible for control of rationale clauses (e.g., (25) ) . (25) a. Giovanni ha affondato la nave per incassare l 'assicurazione . "J ohn sank the boat to collect insurance ." b . * La nave e affondata per incassare I 'assicurazione . "The boat sank to collect insurance ." Chierchia ( 1989) argues .that the unaccusative form of affondare is a zero reflexivization form of the transitive counterpart , where the under lying causative semantics is preserved , but the causing event is inter preted statively . It will become clear from independent considerations , that this analysis falls under the Default Causative Paradigm presented above , and accounts for a range of acceptable adjuncts to these verbs . From our previous discussion un headedness in chapter 5, given a left -headed predicate , one should in fact expect adjuncts to modify the head :

(26) a. II nemico ha affondato la nave dall 'aereo. "The enemy sank the boat from the plane."

b. La nave e stata affondata dall'aereo. "The boat was sunk from the plane."

c. * La nave e affondata dall'aereo. "The boat sank from the plane." In (26a) , dall 'aereo modifies an event and not the agent , since it denotes the locatum of the event that brought about a sinking . The passive in (26b) , as expected , licenses the same adjunct since the agent in the

196

Chapter9

initial subevent is also accessible. However, in (26c) the sameadjunct is ungrammatical . On the basis of the examples below , it appears that it is not accessibility to the agent that crucially determines the licensing of the adjunct for unaccusatives , but rather accessibility to an event .

(27) a. La nave e affondata per l 'esplosione. "The boat sank from the explosion ." b . ?La nave " The c . * La

boat nave

e affondata sank

from

e affondata

per

la bomba

the

bomb

per

! ' aereo .

.

."

"The boat sank from the plane ."

There are two things to point out concerning these data. First , (27a) contrasts with (26c) in that the per-PP is able to make referenceto the initial event

event itself , while the dalrPP itself

makes reference to the agent of the

.

In some sense, this is analogous to the rule of passivization . Grimshaw

(1990) discussesthe fact that in the passivethe agent has the status of a special kind of adjunct , namely an a-adjunct which is systematically headed by the preposition by. With regard to event structure , headedness has the effect of shadowing the non-headed event , which can only be accessed by a special kind of adjunct , and in the case of unaccusatives , one that can construe a coherent causal chain by selecting the proper . subevent . This seems to be exactly what the preposition per is doing in the examples above. The different degrees of grammaticality of ex-

amples (27a-c) show that it is not a question of binary acceptability, but rather , that there is a range of grammaticality

judgments explained

by the degree to which the adjunction satisfies (i .e., "is coherent with respect to " ) the default causative paradigm. In order to capture this degree of "semanticality ," as argued in chapter 4, we need to appeal to the richer descriptive notions of type coercion and nominal qualia struc -

tures. This is explored in more detail in Pustejovsky and Busa (1995, forthcoming) .6 If unaccusativity

is characterized

as a property

of an expression

satis -

fying the DCP with a right -headed event , then one might expect there to be non-lexical unaccusatives , as long as these conditions are satis-

fied. In fact, this appears to explain the behavior of correre (run) when appearing in a dative goal context, as in (28b).

The Lexical Semantics of Causation

197

(28) a. Gianni ha corso. "John has run ." b. "Gianni e corso a casa. "John ran home." While correre behaveslike an unergative in (28a), it is clear that (28b) is an unaccusative form for the same verb. The verb need not be entered into multiple semantic classes, however, since (28b) is a non-lexical unaccusative, derived by co-composition with the PP-phrase a casa(cf. Pustejovsky and Busa, 1995). By allowing a verb's membershipin a particular semantic class to emerge from the composition of the sentence it appears in, we obviate the need to enumerate separate sensesfor the distinct semantic classesassociatedwith that verb. This ability is as significant as the operation of coercion towards providing an explanation for why these verbs are polysemousin the way that they are. One final example will illustrate the usefulnessof the notion of derived unaccusativity. This involves the meaning shift associatedwith creation verbs such as build, as in (29a). (29) a. Tension is building . b. Mary 's presenceis building tension among the other faculty. It would appear as though build has shifted from its normal status as a constitutive causativeverb, which has no unaccusativeform. (30) a. * The house built jis building quickly . b.

The construction crew built the house quickly .

As shown in Pustejovsky and Busa (1995), by selectingan abstract noun as object , the modeof causation for the verb changesfrom a constitutive to a direct causative, allowing this use of build to satisfy the DCP. That is, while typically the direct object of a constitutive causative makes reference to a default argument as the value of its CaNSTquale (e.g., material as part of the artifact that is built ), the qualia structure for an abstract noun such as tension does not distinguish the contents of the CaNSTand FORMALqualia (cf. Pustejovsky, 1995b); consequently, the possibility of a constitutive causative interpretation disappears as well. The relevant structure for the derived senseof build as used in the VP build tension is shown in (31).

198

(31)

Chapter9

[build tension ] FORMAL physobj 1 =~ rtl[animate -ind .] ARGSTR = ARG . = FORMAL =abstract ARG 2 - LJ r-;:l[tensIon ] QUALIA = FORMAL =exist (e2 ,0) =build _act (el'0,0)] [AGENTIVE create -Icp

Ignoring for now the specificsof the co-composition, the resulting qualia structure for the above VP mirrors that of the lexical structure for a direct causative, repeated below in (32) : 0: (32)

o:_act {el ,x ,y) QUALIA= [ AGENTIVE FORMAL= =Q :-result {e2,y ) ]

This , in turn , appearsto licensethe unaccusativeform for the verb build that we see in sentence(29a). These issuesare explored in more detail in Pustejovsky and Busa (forthcoming). In this section, I have outlined a general strategy for representingboth causativesand unaccusativesas related to an underlying causalrelation . As we saw, this supports the view put forth in Chierchia (1989), where unaccusativity is analyzed as a causative. Furthermore, by making use of underspecifiedrepresentationsin the event structure , the model allows us to capture the polysemy inherent in this class of polymorphic verbs.

9.3

Aspectual

Causatives and Coercion

In chapter 7, it was argued that much of the polysemous behavior of verbs which take multiple subcategorizationscould be explained in terms of deepsemantic selection and type coercion. For aspectualverbs such as beginand finish , shown in (33) and (34) below, it was suggestedthat the verb selectsfor the semantictype of event funct ion . If the appropriate type is absent from the local environment (i.e., if the complement is not an event function), then coercion applies, making use of semantic information associatedwith the complement, in order to reconstruct the proper type. For the two examples given below, the default predicates associated with the TELlO roles of the complements are reading and drinking respectively, although within the right context, any number of interpretations is possible.

TheLexicalSemantics of Causation

199

(33) a. John began to read the book. (vp [+ INF]) b. John began reading the book. (VP[+ PRG]) c. John began the book. (NP) (34) a. Mary finished drinking her beer. (VP[+ PRG]) b. Mary finished her beer. (NP) Whatever the specific value of the embedded "predicate" in these examples, the coercion analysis requires there to be an event function associated with the complement of the verb. In Pustejovsky and Bouillon (1995), this analysis is extended to cover the behavior of French aspectual predicates, and is refined in two respects: (a) constraints on the applicability of coercion, motivated by data in Pustejovsky and Anick (1988) and Godard and Jayez (1993), are presented in order to restrict the overgenerationof semantic expressions; and (b) the nature of the selectional properties of aspectual predicates is given a more formal treat ment, while explaining the polysemy between the control and raising sensesof these predicates. As discussedin chapter 7, a semantic type projects to a canonical syntactic form (csJ) along with the forms allowable from licensed coercion operations. Together, these forms constitute what we called a phrasal paradigm, and the type surfacesas one of three possible forms for begin, and one of two forms for finish , depending on which coercion rules are applicable. There is, however, only one semantic type being selected for , and the clustering of the particular syntactic forms appearing as surface complement types in (33) and (34) are systematically projected by virtue of this semantic type. That is, a verb such as begin, selecting for an event, will paradigmatically allow for the expressionof the grammatical forms shown above, assuming surface syntactic constraints are satisfied. For this reason, the structuring of this kind of linguistic knowledge, where this event type has its syntactic expressionas anyone of the surface types in (33) or (34) was called a phrasal paradigm in chapter 7. In line with the discussion from chapter 7, the NP the bookin (33c) is coerced to the appropriate type required by its governing verb, in this case an event. What makes coercion possible in this case is the availability of the selected type, given as part of the NP's qualia structure, indicating , for example, that the TELICrole for bookis the event function of reading, while the AGENTIVErole is an event function of writing . The result of applying this coercion operator to an NP is effectively to create

200

an

Chapter 9

extension

the

of

case

of

two

event

the

book

the

the

.

There

the

,

not

in

can

( 1995

second

)

mentioned

( 35

)

)

and

a

.

The

b

.

It

c

.

The

.

*

d

( 36

)

a

b

Notice

)

,

*

The

It

a

position

,

war

the

type

selected

This

is

encoded

both

is

above

"

,

the

type

more

se

,

coercion

and

even

-

that

of

detail

in

-

demon

rather

structures

-

the

limit

,

but

appli

on

applicable

to

manner

ability

.

of

as

the

many

such

the

ele

-

Pustejovsky

aspectual

control

Consider

,

,

)

verbs

structures

for

example

to

already

,

the

sentences

by

and

our

control

.

Bosnia

.

not

.

.

with

As

,

only

of

shown

in

( 35c

coercion

in

as

)

the

Pustejovsky

.

Inter

furthermore

,

control

discussion

that

Bouil

constructions

of

how

;

exist

-

to

some

structure

reason

-

reconstruction

aspectual

,

and

are

.

following

in

subject

and

metonymic

the

NPs

nominals

instance

possible

Observe

referential

- denoting

an

a

current

the

possible

event

be

constructions

for

and

,

allow

raising

to

into

- formed

is

begin

finish

Bosnia

.

to

well

this

into

is

( 35b

not

language

raising

of

qualia

.

appear

as

relevant

in

the

principled

in

reaching

construction

with

about

operations

reach

finished

however

grammatical

the

.

raising

such

a

well

)

early

raining

)

type

.

finished

( 35d

)

in

universally

discussed

( 34

began

NPs

of

predicates

rain

might

( 1995

writing

the

types

necessary

the

and

to

book

what

lon

)

began

pleonastic

,

is

began

to

The

*

note

means

to

as

( 33

party

.

of

.

war

.

and

operations

coercive

concerns

began

that

estingly

)

provide

book

virtue

such

In

.

in

( 36

no

in

constructions

above

( 35

of

- tuned

This

observation

raising

by

to

stated

finely

.

generated

constraints

of

are

composition

in

by

The

be

operators

the

" reconstructive

overgeneration

Bouillon

( 35a

,

coercion

.

.

are

however

reconstruction

reading

availability

such

.

operations

appear

,

the

,

things

First

complement

the

semantic

the

metonymic

namely

are

predicates

operations

,

:

interesting

.

a

example

itself

several

of

for

complement

aspectual

The

,

and

coercion

mantics

and

book

called

interpretations

the

are

of

in

the

,

interpretations

of

ments

meaning

complement

cation

that

NP

These

structure

strate

NP

function

selected

with

the

causation

namely

,

for

the

is

it

is

same

no

lexically

accident

aspectual

un

-

The Lexical Semantics of Causation

201

predicate , given that causative and unaccusative forms exist for the same predicate , as with sink . That is, we can view the relation between these two constructions as one involving causation in much the same way as the default causative paradigm . It has long been noted , of course, that

pairs such as that in (37) are related by a causative interpretation of some sort (cf. Perlmutter , 1967, Levin, 1993). (37) a. The movie began. b . Mary began the movie .

The construction in (37a) is essentially an unaccusativeform, while that in (37b) can be easily interpreted as its causative counterpart. This

solution

, however

, need

not

be restricted

to the

event - nominal

casesin (37) above, but can be extended to include the distinction between raising and control senses of the same verb . That is, begin exhibits a logical polysemy between control and raising senses, where the unac cusative

form is associated

structure

is associated

for verbs that sentation

exhibit

with the raising

with

control

the

causative

construction form . The

and raising

behavior

and the control lexical

semantics

is a causative

repre -

.

The alternation displayed above is licensed by the headless nature of the event structure representation of the predicate begin. Whether it surfaces as a raising

verb or a transitive

control

by which subevent is headed . If the initial structure

results . If , however , the final

struction

results

verb will be determined

event is headed , a control

event is headed , a raising

con -

.

In Pustejovsky and Bouillon (1995), it is argued that much of the confusion associated with the coercion data arise from this ambiguity . Although these sensesare distinct , they are logically related types , where only the control sense allows coercion . The idea of analyzing aspectual verbs as essentially ambiguous is not new , but was already proposed

by Perlmutter (1970) for English, and Lamiroy (1987) for French. The traditional method for distinguishing between control and raising verbs involves a battery of diagnostics testing for selection , agentivity and

controllability (cf. Dowty, 1979and Zaenen, 1993). Perhapsthe best indicator of a raising predicate is that it imposes no selectional restrictions

on its subject, as illustrated with the verb seemin (38). (38) a. The lake seems to have frozen .

202

Chapter 9

b . A riot seems to have happened yesterday . c. This fact seems to have escaped Mary 's attention .

The subject in each sentencein (38) is restricted by the embeddedpredicate in the VP selected for by the verb seem. A control predicate ~ on the other hand , imposes clear and obvious restrictions

on the subject

NP (cf. (39));

(39) a. Marytriedto leavetheparty. b. *A riot triedto happen yesterday . Therearealsosyntactic constraints imposed bycontrolpredicates that areabsent in raisingconstructions :7 (40) a. Thereseems to bea riot goingonnow. b. *Thereattempted to bea riot. Perlmutter(1970 ) usesforce -complement constructions as another clearindication of a controlverb. Compare (41a ) with(41b). (41) a. MaryforcedJohnto beginwritinghisthesis . b. *Maryforced it to beginrainingyesterday . Thesentence in (41a) illustrates thatthematrixobjectstands in a control relationto theembedded VP. Observe , however , thattheungram maticalsentence in (41b) illustrates that a "raised " NPcannotsatisfy theselectional constraints imposed byforce . (42) a. I amforcingyouto beginreading thebookbyProust . b. *1amforcingit to beginto rain. Thewell-formedness of objectcomplement coercion with aspectual predicates suchasbegin isconditioned bytheeventsortofthequaliaassociated withtheNPitself. Thus , onlyNPshaving associated transition events will allowcoercion andcontrol . Thisis notto say , however , that begin selects onlyfortransition events . Thereare, of course , perfectly grammatical examples ofprocess orstatecomplements , asshown in (43) below : (43) a. Theacidis beginning to corrode themarble . b. It is beginning to rain.

The Lexical Semantics of Causation

c. d. e. f. g.

203

The snow began to fall at midnight . The war is beginning to reach into Bosnia. John is beginning to bleed. John is beginning to be annoyed by the noise. John is beginning to be ill .

.J

The above examples illustrate the use of beginas a raising verb. The two sensesof the verb beginconform to the observation that Perlmutter originally made, namely, that begin functions as both a raising and a control verb.8 As a Raising verb, the event sort specified as the complement to beginmay be any sort. As a control verb, it appears that the complement must be a TRANSITION .9 Finally , as pointed out in Jacobson(1990), VP-ellipsis can be used as a diagnostic for determining whether a complement is part of a raising or control construction in English; namely, only control complementsenter into this construction. Notice that in (44), the only fully grammatical sentenceinvolves an overt control interpretation of begin, that in (44b). (44) a. * John began to bleed and Mary began, too . b. John began to read the book, and Mary began, too. These considerations together suggest that there are in fact two constructions associatedwith begin, namely control and raising. To seehow these sensesare related, let us presentthe lexical semantics for begin, illustrated in (45) below.

(45)

What emerges from this representation are two typing assignments f01 begin, corresponding to the two senses: (46) a. begin as a Control verb : e , > , <e, b . begin as a Raising verb : <eO ", >

204

Chapter9

Given that event-headednessacts to foreground or 'focus' a single quale of the verbal semantic representation, the effect of heading the final event from the lexical structure in (45), i .e., the FORMALrole, correspondsto the raising interpretation ; what is assertedis simply the initiation of an event, without explicit referenceto causal preconditions of the event. Consider now the semanticsof the control interpretation of begin, and how it interacts with coercion on the complement.

(47) Mary begana book. Assume

the

qualia

structure

for

book

to

be

as

given

in

(

48

)

.

book

ARGSTR

(

48

=

ARG2 ARG

[

)

QUALIA

=

= =

1

info

y x

= =

AGENT

heading

begin

,

trol

relation

(

49

the

we

initial

arrive

at

is

expressed

event

a

shadowing

.

the

~ (

of

y

e

, x , w

write

(

lexical

subject

]

lcp

hold

read

=

in

as

physobJ info

_

FORMAL TELlO

By

: :

. physobj

eT

) , x

, v

)

, x

)

representation

the

associated

culminating

and

event

event

function

,

while

complement

with

the

con

-

.

)

Following the discussionin chapter 7, the qualia can also be seenas type pumping operations, giving rise to new types for coercive environments. Thus for example, the type available to an expressiona with quale Qi of type T, can be seenas allowing the following type inference:

(50) a : a E9 Qi [a, T] : a -::;. T Qi [a , T] (a ) : T

205

The Lexical Semantics of Causation

This says that , given an expression a of type a , there is a coercion possible between 0" and T, which changes the type of a in this com position from a to T. We will illustrate the application of this coercion operation

below, as used in the beginexample above in (47). Becauseof the qualia structure of the complement inherited from that of its head book, the typing requirements specified by the governing verb begin are satisfied just in case the type of the NP is coerced to an event function .

As illustrated below in (51), coercion applies to the complement NP, where reconstruction

with either the TELIC or AGENTIVE qualia will

result in the appropriate type selected by the verb . We illustrate

beg 'iln . (51)

:

the

(e

This corresponds

VP

.

C -) -

- 1- c, 1 ; ;;

within

11

derivation

- 1-

where

the

TELIC b

the

"'.(e~') .eQT --.".- :e~

role k

has been ,

selected

,

.

'

(e - 1- c,C -) wffi a 00 :e begin the book : -

to the selection

tree structure

shown below .

(52)

This states that the TELICrole of the book, AXAeT[read(eT, x , a-book)], is available as an alias for shifting the type of the NP . This metonymically reconstructed type , <e , ' >, is identical to that selected by the verb begin in complement position . After the coercion operation , the derivation proceeds as if a full predicate were present in the syntax .l0 The result of heading the final event in begin gives rise to either direct

event selection, as in (53a) below, or a raising construction, as in (53b). (53) a. The party began at midnight . b. Snow began to fall .

206

Chapter9

In (53a), the typing of the predicateis satisfieddirectly by the eventdescription in subjectposition (ignoringsometechnicaldetails). For the constructionin (53b), I assumethat raisingis accomplished by function composition(FC), in the mannerof Jacobson(1990). In particular, the raisingverb begin , of type <eO " , >, imposesthe type eO " on its complement. Assumingthe VP to fall, for example, in (53b), is <e, E>, then FC (begin,to fall ) returnsthe expressionAP[begin(jall (P)]. Finally, let me turn briefly to the semanticsof the verb finish and how the semantictype selectedby this verb corresponds to a different canonicalform as well as the associatedphrasalparadigm. There are two things to noteaboutfinish. Noticefirst that finish differsfrom begin in that it is not logicallypolysemous , havingonly a controlsense . That is, the raisingexamplesin (54) are ungrammatical . (54) a. *It hasfinishedraining. b. *The sun hasfinishedshiningin my eyes. c. *The acid finishedcorrodingthe marble. This would suggestthat finish is not lexically underspecifiedwith respectto headedness , as is begin, but is alreadyspecifiedwith a head. It is this lexicalspecificationthat givesriseto the controlreadingonly.II Considernow the sentences in (55), which appearto be raising constructions, contradictingthe claim madeaboveabout finish beinglexically specifiedas a controlverb. (55) a. The leaveshavefinishedfalling. b. The paint hasfinisheddrying. Thesedata would suggestthat a raising constructionis possiblewith finish with somenominals. But the sentences in (55) are best analyzed aspseudo -controlcases , and they arerestrictedto a certainwell-defined classofnominals. In general,theseverbsdo not passthe standardraising tests, but nounssuchas paint and leavesare exceptionsbecausethey carry qualia information indicating a kind of "autonomyof behavior" relativeto certainpredicates . Hence,paint, for example,is construedas a pseudo -agentin the control relation becauseof this property; i.e., it can dry on its own. It is interestingto observethat anothertype of intransitiveconstruction is possiblewith finish; namely, if the eventnominalin subject position has an agentivecomponent(cf. (56b) and (56c)), then a control

TheLexicalSemantics of Causation

interpretation

207

is possible in what would otherwise appear to be an in-

transitive (i.e., raising) construction. Pustejovsky and Bouillon (1995) refer

to these

as intransitive

control

constructions

.

(56) a. ??The party finishes at midnight . b . Class

will

finish

at 2 :00 pm .

c. The talk will finish by noon .

(57) a. *The rain will finish by noon. b . The rain will stop by noon . While

classes and talks have an apparent agentivity

and controllable

component to them , parties are less controllable , resulting in the less

acceptable (56a). Since rain is completely uncontrollable, it is ungrammatical in an intransitive

control construction with the verb finish (cf.

(57a)) . The verb stop, however , allows a raising interpretation and per mits the intransitive raising construction in (57b) . This verb is inter esting because it has both control

and raising

senses , yet does not allow

complement coercion at all . Observe that stop appears in the sentences

in (58) with a non-control construction, assumingthe senseof "prevent:" (58) a. John stopped Mary from smoking in his house. b . Mary stopped the man from hitting her . c. John stopped the bomb from exploding . In fact , there is a kind of coercion possible in complement position with stop, essentially reconstructing an ellipsed predicate , as in (59) . (59) a. John stopped the car . (from moving )

b. The referee stopped the clock. (from moving) c. Mary stopped the record. (from playing/ moving) What these data suggest is that the complement type of stop is not an event function , as with begin, but rather simply an event , where the type of the verb is <eO ' , <e , . That is, these verbs are not strict obligatory control verbs , such as try and begin, but impose "available controller "

binding , as with verbs such as want (cf. Chomsky, 1981, Dowty, 1985, and Farkas , 1988) . It is worth noting that , with the complement of stop specified as eO ' , it is clear that coercion is not possible since this is not among the type aliases for the NP complements (cf . Pustejovsky and

Bouillon , 1995 for discussion).12

208

9.4

Chapter 9

Experiencer

Predicates

Giventhe generalstatement of causation presented above , it is interest ing to speculate on whatroleverbalqualiamightplayin otherlexical causatives . Forexample , ashasbeenlongnotedandaswasmentioned in chapter2, the verbkill allowsfor a broadrangeof possible subject types, aslongasthe causalrelationbetween the argumentin subject positionandthe resultingstateis coherent in someway. (60) a. Johnkill. .dMary. b. Thegunk!~ g Mary. c. Thestormkill~ Mary. d. Thewark!ll~ Mary. e. John's shootingMarykilk.dher. In Pustejovsky (1991a ), it wassuggested that kill selects for aneventand coerces its subjectto thistype, in a mannersimilarto theinterpretations of enjoyand begin . Thisis essentially correct , but the detailsaremore complicated , andin fact, moreinteresting . Assume that the lexicalrepresentation for the verbkill is asgivenin (61), wherethe initial subevent is headed . kill

(61)

El = el:prOCess E2= e2:state EVENTSTR = RESTR =
In sentence (60a) above , the process Johnengaged in suchthat Mary diedis left completelyunspecified . In (60d), on the otherhand, the particularactionis completely specified by theeventdescription in subject position. This exampleillustrateshowqualiaunificationcanfully specifywhatis left underspecified with an animateargumentin (60a). Thecases of (60b-d), on the otherhand, arecausallycoherent because of coercionand qualia-basedselection . That is, in (60b), the NP the

209

The Lexical Semantics of Causation

gun must participate in the relation within the AGENTIVE Quale in some

significant way, just as in (60c), the storm is a coherent cause only if Mary stands in a relation to this event , such that the storm could effect such a change of state , e.g., she was in the middle of the storm , etc .

Hence, even sentenceswith an event nominal as subject (as in (60c) and (60d) must respect argument coherencein order for the causal relation to be satisfied (cf . Croft , 1986 for discussion of this idea ) . The above discussion

is important

because the semantics

of DCP verbs

such as kill and break is suggestive of the underlying causative nature of

experiencer predicates, such as those shown below in (62). (62) a. Books QQ@ me. b . The movie frightened Mary .

c. The newspaper~ngere~ John. d . Listening to Mary irritates Alice . We can view these sentences as involving a metonymic reconstruction of the subject to an event , and in particular , to an experiencing event between the surface object and the surface subject . That is, in (62) , it is

(my reading) bookswhich bores me, (Mary 's watching) the movie which frightened her, (my seeing) John's face which scared me, and (Alice 's) listening to Mary which irritates her .13 We can , of course, experience objects in any number of ways . That is, one need not read books in order to be bored by them . One can be bored by looking at them , shopping for them , writing them , or thinking about them . This is not in any way inconsistent with the GL approach . The qualia determine two types of information

in the context of coercion :

i . Type and sort information which the qualia must satisfy ; ii . Specific qualia values which are the explanatory standing

modes in under -

a word .

For words such as film and book, the TELIC quale role value of watch and read respectively are not optional in any sense, but are part of the semantics

of the

words

.

When

an NP

enters

into

a coercive

environ

-

ment , such as here with experiencer verbs , the qualia values act only to determine

the default

assignment

for how the type environment

is recon -

structed . Thus , it seems that linguistic evidence supports an underlying semantic type of an event as the subject , which would directly explain

210

Chapter 9

what the connection between the subject and object of the experiencing relation is. The underlying semantics of psychological predicates is a causative structure where the surface subject is the logical object of an experiencing event . On this view , the lexical representation for the verb

anger has something like the following form, where Exp (x , y) is a sortally restricted relation of experiencing (e.g., hearing, seeing, watching, etc.), and < is a strict partial order of temporal precedence: (63) DVxVyVe[anger(e, y, x)] - + 3e13e23Exp[Exp (Cl, x , y) /\ -,angry (el , X) /\ angry (e2, x ) /\ -' C2< el ] This states that a verb such as anger involves someone who directly experiences something , and as a result becomes angry . What is inter esting about examples such as (62a) and (62b) is that the semantics of the NP in surface subject position contributes information to the inter pretation of what kind of experiencing event is involved . That is, the qualia structure projected by the NP books contributes the particular

manner in which I became bored in (62a), namely the NP's TELIC role of reading . Similarly , our knowledge of movies as something that we watch

and experience

in a particular

manner

is encoded

in the TELIC

role of movie in sentence (62b). The event projected from the noun movie , viz ., watch , in turn

satisifies

the selectional

requirements

of the

verb frighten on its subject . The complete semantic form assigned to this verb is not the default causative paradigm discussed in section 9.2

(i .e., dcp- lcp ), but rather is a causative act which predicates a certain state of the person performing the act , hence, the experience . Thus , unlike the direct causative we encountered in the lcp expression of ar-

gument coherence(cf. Section 9.1 above), an experiencer verb exhibits experiencedcausation, as shown in (65) . (64)

DIRECT CAUSATION:

[QUALIA =[_ FORMAL =o1esult (ez ,y)]] AGENTIVE= a_act (el ,x ,y )

(65)

EXPERIENCED CAUSATION :

AGENTIVE Q_act(el ,X ,...)] [ QUALIA = [FORMAL ==a:-result (e2 ,x)

]

The Lexical Semantics of Causation

211

Experiencer predicates do obey argument coherence, but by controlling or being in control of the causing action . As a specific example of this representation , consider the semantics for the verb anger.

(66)

.I

Notice that the temporal restriction is different from default causatives, and involves a precede and overlap relation between the experiencing processand the resulting experiencedstate. That is, if doing something angers me, then I need not complete the activity before I become angry. The major difference between DCP verbs and experiencer verbs, however, is the necessarycontrol relation that exists between the coerced predicate in the subject position and the individual who appears in object position. That is, just as the coercing control verbs enjoy and beginwere analyzed as selecting for event functions in the complement position, experiencer predicates select for an event function in subject position. For NPs such as the newspaper and the movie, the qualia structure is able to drive a metonymic reconstruction to the type coerced by the predicate, an event function . The event structure associated with the sentencein (62c) , for example, is illustrated below in (67), where a specific processhas been specified, by virtue of the TELIC role from the qualia structure of the newspaper.

(67)

212

Chapter 9

Notice

that

different

the from

( 68

mapping that

to

syntactic

associated

form

with

the

from DCP

the

headed

causatives

event ,

such

as

,

el

,

kill

is

.

) S

~

/

/

- " "

'

"

, ,"' . ...

[ < e , E > ]

; ; \ x ; ; \ e [ read

( e ,x

the

, the

- ll

Vp

e ws

p a p : ~

newspaper

~

/

/

~ "

" " " , "

I

"

[ human

]

anger

It

is

subject

important

verb

to

appears

are

verb

,

to

emerging

( 69

to

NP

be with

)

a . .

note

made

even a the

that

the

available

within

the

controlled

The

sign

startled

sign

~ ngere

in

.

non ( 69a

Mary

.

Mary

.

on type

experiencers versus

sentences

The

conditions through

For

example

- controlled )

and

how

the

qualia

coercion

( 69b

will ,

one

sees

perception )

respectively

of vary

the from what

distinction .

b. ~

That is, the verb startle seemsto select for direct perceptual experiences such as given by direct visual and auditory stimuli , e..g, seeing and hearing. In sentence (69a), the default interpretation of the experiencing relation between Mary and the sign refers to her simply seeing it . The verb anger, on the other hand, requires of the subject that the experiencing event is controlled and intentional ; hence, the predicatesassociated with the reconstructed event functions for the subject of anger will reflect this distinction . In (6gb), for example, Mary must at least read the sign to be angeredby it . Although it is certainly true that the mere presenceof the sign in this location could make Mary angry, that is not a possible interpretation for (6gb) .14 There is a noll-trivial issue of the aspectual properties of experiencer predicates that I have not addressedhere. This is the fact that the resulting aspectual class of an experiencer predicate in context is determined by the reconstructedevent type from the subject position. Consider the following sentences:

213

The Lexical Semantics of Causation

(70) a. The flash of lightning frightened John. b . The electrical storm frightened John .

(71) a. The sudden ring of the telephone frightened John. b . The ringing telephone frightened John .

That is, the point-like property of the event in the subject of (70a) is inherited by the overall experiencing event denoted by the sentence.

Likewise, the bounded processevent in the subject of (7Gb) influencesthe aspectual interpretation

of the whole sentence as well . Similar remarks

hold for the distinction in (71). This is related to the analysis of how complements influence the aspectual properties of their governing verbs , as presented in Verkuyl ( 1993) . Although this phenomenon must be accounted for in the aspectual characterization of the experiencer verbs , I will defer discussion of this issue to Pustejovsky (forthcoming ) . To conclude this section , let us discuss briefly

what the consequence of

this an.alysis of experiencers is for the well -known phenomenon of binding

violati ,ons associatedwith this classof verbs. Following Chomsky (1981), the normal constraints on anaphoric binding will refer to properties of precedence, c- command , and binding domain . As discussed in Belletti

and Rizzi (1985), Pesetsky (1987) , and Grimshaw (1990), frighten and other ~experiencer verbs violate these constraints by allowing anaphoric bindin ,g into a subject which itself c-commands the antecedent , as shown

in (72:).15 (72) a. The pictures of each other frighten the teachers. b . * Each other 's students frighten the teachers . c. The picture of himself frightened John . (73) * The students of each other fear the teachers . The I )sych-predicates patterning like fear behave predictably with respect to binding , while the frighte11rclass data are not accounted for witho .ut specific stipulations on the conditions for anaphoric reference .16 Recall from our previous discussion that experiencer verbs select for an ev.~nt function in subject position , and are coercive on this position . This .explained why control is possible into subject position from the object , as seen in (74) .

(74) a. PROi SeeingMary so happy pleasedJohni.

214

Chapter 9

b . PROi

Assuming kind

this

of

given

To tree

this

structure

have

and

event

this

some

we

of the

of

the

the

relation

in in

becomes

subject

clear

of

an

simply

furthermore It

from

of

()- roles

from

the

, is the

should

be

unilaterally

/

~/ /

"""-"""-"""-""

? Each

event

, where

I

proposition

.

""" e2

the

and . But

possible

upset

' s students

, one

directly they

are

structure

of

these

governing

verb

.

might

that .

into

That

no

the

expect

is , coercion opaque to

the

into

the

statements

subject

subject

on

, but

this

fol -

. This

position

position

( 76 ) illustrate

,

. is

not

.

.

teachers

subject

coherent

referentially anaphors

the

into

.

teachers

upset

possible

notice

because

John the

in

bind

predication of

into

to

specific

is possible

sentences

upset

to

inverse

binding

(j ) ]

ability

interpretation

that

, as the

due of

control

frightened

coercion

ceptable

function

[ajraid

, the

is not

licensing

reason out

otheri

otheri

frighten

is

event

same

a . * Himselfi

the

(x ,j )]

structure

predicate

pointed

b . * Each

Given

( 72c ) . The

I _picture

above

or

acceptable

( 76 )

( 77 )

the

experiencer

ordering

lows

in .

of the

I [see ( j , x ) / \ the

the

,

e < I:X

< el

As

the

surface between

( 75 ) below

structure

( 75 )

expect

( 72 ) to

engages

is shown

quantificational

in

causative

sentence

proposition

.

actually

above

individual

again

Johni

would

shown

this

, consider

frightens

, then

structure

illustrating

simplified

Boston

violations

the

point

in

correct

underlying

object

clarify

is

car

c - command

actual

surface

his

analysis

apparent

the

the

Driving

reconstruct

.

position

the

sentences

in

reconstruction applies , there to

of

verbs ( 76 )

from to

the

to

the

subject

is

nothing

in

the

type

required

such

as

be

ac -

reflexive NP , but

the

qualia by

the

215

The Lexical Semantics of Causation

9.5 Modal Causatives In

this

section

,

ordinated

I

This

is

in

the

an

lexical

causatives

force

syntax

,

(

78

Some

a

.

John

.

Mary

verbs

state

.

79

)

(

a

b

This

verb

,

79

John

.

Mary

,

is

Namely

internal

tree

is

,

l

to

action

if

if

x

a

pos

-

relation

killed

y

;

modally

in

,

as

or

x

order

broke

to

subordinated

it

she

encode

such

add

by

lexical

the

the

in

the

.

the

dishwasher

of

is

resulting

of

the

relation

.

the

semantics

item

causal

much

it

uncertainty

into

a

so

puts

the

directly

of

k

killing

.

studied

in

the

breaking

is

Fillmore

an

verb

the

risk

expressed

watering

Atkins

its

it

by

,

lexical

which

lex

-

syntactically

(

strange

leads

,

1994

)

it

1990

)

and

Atkins

construction

.

(80) a. Mary ri ~ Q deathto saveher SOll . b. Mary I~.k .dher life to saveher SOll . Mary Ii ~ .dillness. Mary ri. ~ .dher health.

(82) a. Johnn ~ Q bankruptcydoingthat.

.

the

with

-

to

a

much

in

along

complement

Fillmore

within

so

putting

complementarity

constructions

Goldberg

by

glass

and

of

unusual

that

tree

the

because

for

a. b.

how

the

causal

state

watering

glass

appear

modality

argument

(81)

show

.

. 17

the

resulting

be

by

the

of

rllik

and

verb

of

That

must

I

-

action

.

there

1987

the

,

!

motivation

,

verb

break

interesting

,

.

and

an

affects

this

certainty

the

,

sub

of

.

~

example

)

.

)

an

An

the

risk

to

the

modally

dimension

arguments

about

78

however

the

in

(

ay

follows

encodes

above

(

may

,

itself

ically

in

~

which

item

form

shown

b

the

as

modal

semantics

,

encode

effects

such

this

brought

uncertainty

as

)

their

event

of

verbs

encode

of

some

items

subordinated

of

of

part

then

the

more

semantics

realization

inherent

,

lexical

modally

encoding

syntactic

Most

how

or

representation

sible

as

briefly

relations

seen

semantic

y

discuss

causal

the

taking

dishwasher

.

its

nominal

behavior

.

expression

to

analyze

grammar

of

the

this

verb

(

cf

.

Fill

-

216

Chapter 9

b. JohnKllikgghis own solvencydoingthat. What is interestingabout theseexamplesis that the direct object in each(a) sentencestandsin complementarydistribution to thosein the (b) sentences , yet they aresemanticallynearparaphrases of eachother. Whateverthe thematicor casestructureof the verb is, it is clearthat the presenceof one argumentactsto completelyshadowthe expression of the other. In their investigationof this word, Fillmore and Atkins (1990) argue that the semanticrolesassociatedwith the verb risk must includethe followingnamedrelations: (83) a. b. c. d. e.

HARM: a potential unwelcome development or result ; VICTIM : the individual who will potentially be harmed ; DEED: the act which brings about the risky situation ; GOAL: that which is achieved by the act ; POSSESSION : something valued by the victim .

According to their analysis , there is no obvious compositional solution to the selection and assignment of the appropriate case roles as exhibited in the above data , and this is seen 88 evidence in support

of a construction

grammar solution . From our perspective , however , the complementary expression

of the HARM and POSSESSION roles in the sentences

above

is indicative of a deeper relation between the roles and the nominaliza tions that express them . Namely , the HARM role always indicates the privation of a possible POSSESSION role , but in complementary semantic distribution

. IS

This minor

observation - that

one argument

of risk is the privative

of

another - points to an interesting fact about the complement types for this verb . That is, there are actually three basic syntactic patterns for

the verb risk (for these arguments), as shown below in (84): (84) a. Mary risked @~ h to save John. b . Mary risked her lif ~ to save John . c. Mary risked losing her life to save John .

Given that theseNPs- which accordingto Fillmore and Atkins' analysis, eachassumea differentcaserole- stand in complementarydistribution to eachother, then it is plausibleto considerthem as different

217

The Lexical Semantics of Causation

expressions of the same underlying role , namely a privative stative event. Assume that the noun death is the privative of life , and illness is the pri vative of health, and so on . The phrasal expression of this role , losing one 's life , is also privative , and it must be explained why the syntac tic variability of the same semantic type is expressible in three different forms

.

I believe that the complement taking behavior of risk follows from the following assumptions : a. The direct object argument is typed a PRIVATIVE STATE. b . There

is no distinction

between

HARM

and

POSSESSION

roles .

c. The verb may coerceits complement, giving rise to variable (nominally dependent) interpretations of the semantic type. Essentially, PRIVATIVE is acting

as a function

over NP denotations

.

d. The semantics of lexical items makes referenceto privativejnon privative pairs . Qualia structure provides the semantic mechanism for reconstructing a privative interpretation for a lexical item . How do these assumptions relate to the three grammatical forms avail able for this unique semantic

type ? We can think

of PRIVATIVE as func -

tioning as a coercion operator over a semantic expression , where , for exam pIe, given the nominal health, a metonymic reconstruction of this stative nominal is returned , which satisfies the privative typing environ ment of risk , without changing the meaning of the NP itself .

(85)

Metonymic Reconstruction by Coercion : a. PRIVATIVE ( health) = losing one's health b. PRIVATIVE ( life) = losing one's life c. PRIVATIVE ( solvency) = losing one's solvency

In (85a), the privative typing requirement is directly satisfied by a number of lexical items , i .e., sickness, illness , which are both lexically pri vative , or by the actual gerundive phrase given . Similarly , the lexical

privative for (85b) is death, and for (85c) is bankruptcy. The modal nature of the resulting state brought about by the action performed in a sentence with risk is encoded as part of the lexical semantics

of the

verb . That

in the semantics

is , a modal

of lexical

causative

statement

must

be allowed

items . We can express the conditional

nature

218

Chapter 9

ofthisresulting statein thequaliastructure oftheverbriskdirectly.

The qualiastructure for risk can be givenas follows:

(86)Ae 1 2AxAe [risk: FORMAL= , 2 [P(e x) V -iP(e 2, x)} A AGENTIVE= [R(e,,

x)J}

Thisstatesthat an individual x, byperforming the activityel, willend up eitherin the state P(e 2, x), or P(e 2, x). Morespecifically, this denotesthe possibilitythat the activityin R(e,,x) resultseitherin the

continuation of the relevantstateholdingof x, P(e 2, x), or its negation,2,P(e x). Thedisjunction intheFORMAL roleistheuncertainty of howthe actionwilleffectthe9subject. Thisfurthermore explains

whytherearethreeexpressions corresponding to the typeof PRIVATIVE

STATE. Namely, the typecanbe satisfied byselecting 2, -P(e x) from theFORMAL quale,andeitherfindsa lexicalized formforthisprivative expression, suchas illness, orit isforced to grammaticalize theprivative ofthe positive form,health,givingriseto losingoneshealth.Finally,

the non-privative, P(e 2, cc),canbe selected, andinterpreted, becauseof coercivereconstruction, as the privative in this context. 2

In order to accountfor the modal nature of the causal relation ex-

pressedintheverbrisk,it wassufficient to allowa disjunctive statement in the qualiastructure,something I haveavoidedthusfar,forobvious reasons.Ifwecanmaintain theinterpretation ofdisjunctive expressions withinqualiaasinterpretedrelativeto theirmodalforce,thenthe nature of the semantic systemwillbe extremely constrained. It is not clear,

however, whatothertypesofmodalcausatives existinnaturallanguage, andwhethertheywillbesoeasilytreatedbya disjunctive analysis. 9.6

Conclusion

In thischapter,I havetriedto illustrate howa generative lexicalap-

proachto semantics is ableto structurethe spaceof possiblelexicalizationsforcausalrelationsin language.Manytopics,therefore,have beenignored,amongthemthe semantics of resultative constructions,

morphologically derived causatives, aswellastherelationto syntactic

causatives.

What we haveseenis a numberof possiblelexicalformsfor the expressionof causation.In particular,weexaminedthree basicstructures satisfyingthe conditionon argumentcoherence, summarized below.

219

The Lexical Semantics of Causation

(87) DIRECT CAUSATION :

[QUALIA =[FORMAL =a~esult (e2 'Y)]] AGENTIVE= o:_act (el 'x ,y )

(88) EXPERIENCEDCAUSATION :

[QU ALIA = AGENTIVE

[

(89)

INDIRECT

=

FORMAL

=

( CONSTITUTIVE

QUALIA

=

closing

this

lexically

FORMAL

=

to

causative

causative

manv

have

co

,

obligations

.

,

refers

to

)

( 91

made

the

a

.

We

had

the

painter

paint

.

We

had

the

painter

repaint

.

We

had

the

house

)

is

painted

,

,

and

how

compares

have

,

and

the

the

event

and

causative

under

acceptable

the

But

a

or

odd

is

.

agreement

clearly

is

notion

contractual

normal

,

subject

somehow

and

in

inter

-

furthermore

control

of

,

but

)

English

. 21

.

house

.

.

the

house

the

with

in

house

the

.

house

of

allowing

grammatical

in

1980

the

because

,

are

,

make

.

.

grammatical

possible

Saksena

is

painted

distinctions

.

)

repaint

house

agreement

( cf

briefly

coherence

make

by

paint

painters

the

forms

generally

verbs

which

painters

the

made

derivational

,

,

b

Both

compare

.

*

.

to

on

have

.

contractual

]

of

( 90a

event

- agency

made

( 90b

]

between

with

c

)

Hindi

while

? We

censed

not

,

We

,

)

useful

about

.

)

]

)

condition

brought

.

Interestingly

as

be

a

We

]

, z

distinction

statement

co

' y

) ' X

the

the

,

- upon

the

( z ( , ely

controllability

play

example

agreed

. )

:

( e2

act __ of

its

b

c

are

at

same

by

( 90

of

For

an

mediated

,

the

also

the

. )

CAUSATION

might

with

, . , X

.

involves

is

pretations

fare

observed

English

- agency

Q

, X

( e2

involving

have

~

in

of

it

,

constructions

Italian

AR

,

causation

( el

a : Jesult part =

=

discussion

- encoded

_ act

)

AGENTIVE CONST

In

Q

a : _ result

with

causative

this

same

the

have

.

force

Such

lexical

types

of

of

make

subtle

causatives

in

notion

some

( breach

to

be

li

-

distinctions

,

languages

but

do

exist

,

e

. g

. ,

220

Chapter 9

How

does

sider

the

this

relate

structural

illustrated

in

( 92 )

a .

b .

II

ha

suo

II

a verb

causative kill

exhibits .

dying ble

, and

not

( 93 )

it

sarily

affected

how into

the

the

in

Italian

crepacuore

.

heart

."

the

on

itself

.

for

' "

"

a

on of

at

.

all

object

Hence

semantics

the

break

condition

the

, some

of

crepacuore

.

."

of

coherence

causatives

arguments

di

interpretation

affected

the

Maria heartbreak

The

affected

( 92 ) does

, as

the

in neces

case

with

( 93 )

~ /

act

/

< cx:

/ '/ """"

2

(e 2, L

e3

, e3)

~

/

/

/

"

x

::::"

"

" "

e4

e5

I die

The

interpretation

large suggestive causal

an

of

issue

to

of

how

relations

_ act

do

causative

justice

syntactic .

I

( e4 , x , j )

to

dead

constructions here

causatives

, and

( e5 , j )

is , this

necessarily

of

discussion differ

course has from

,

structure cannot

is

the

possi

unification

tree fare

by

many

qualia

as

event

in

are

event

such

is , while

denoted

, there to

syntactic

That

, fare

event

due

verb

.

argument a

.

e

,

."

di

obey the

the

semantics

el

-

."

coherence

object

by

kill

Con

.

must

on

these

?

.

cancer

of

die

argument

event

for

causative

no

to

Within

coherence

underspecified

bind

lexical

is

die

Maria

, it .

of

morire

to

die

cancro

die

Maria to

above

exhibits the

illustrates

fatto

morire

, there

interpretations left

ha

Mary

causation

argument ,

to

morire

Mary

discussed

Rather

others

a

caused

, however

not

fatto

encodes

, as

fatto

relations

causative

di

John

Mary

caused

ha

" John

When

ha

causative

transitive

Gianni

caused

divorce

the

morire caused

divorzio

of

.

comportamento

Gianni

coherence

below

fatto

behavior

" The d .

lexicalization to

)

smoking

" His c .

) - ( 92d

fumo

" His

the

alternatives

( 92a

II

to

,

much

been

too merely

lexicalized

-

10

Consequences of a Generative Lexicon

10.1 Co-composition and Verbal Semantics Therepresentations assumed fromprevious discussion forcausative verbs suchasopen , shown in (1) below , aremotivated in largepartto allow headedness fromanunderspecified eventrepresentation to projectto eithera causative or unaccusative construction , asoutlinedin chapter 9.

(1)

Such a representation provides us with a great deal more structure , in terms of granularity of the semantic description for the lexical item , while also connecting directly with the syntactic expressiveness of the semantic types . But notice that the structural

semantic description for

the predicate in this case is quite different from the cases discussed

previously, such as sink, kill , and the other verb classes. Namely, the lexical representation for the predicate being defined , open, itself makes reference to a predicate with the same name in the FORMAL role . The qualia do actually distinguish the verb open from the stative adjective in this representation , by referring to the open _result predicate rather than any particular lexicalized form for it in the language . Nevertheless , what is distinctly different about verbs such as open, close, break, repair , and fix , is the functional dependence that the verb has on the argument denotation for arriving at the appropriate interpretation of the verb in context . To better

understand

this connection

, consider

the following

sentences:

(2) a. Mary opened the letter from her mother. b . The rangers have opened the trail for the summer . c. John opened the door for the guests.

222 Chapter 10

(3)

a. Mary Q!:Q~ the teapot this morning . b . Federica 's television QIQ, @during the Oscars. c. Mary QrQ~ the stick in two .

What each sentence in(2)asserts about its complement isthat the function orpurpose associated with the object of the predicate isavail able for use .form That is,the the object 'sthe TELIC role is shared ,in amodally subordinated , as value of FORMAL role for the main predi cate open . For a noun such as letter , the TELIC is specified as the event function AZAe3x .y[read (eFORMAL ,z,x.y)J;role this expression isembedded into a modal context within the of the predicate open in sentence (2a )letter above .as Assume that expression ofthe lexical semantics for is given in(4the ); relevant letter physobj .= info -lcp (4) QUALlA = TELlO FORMAL R ( =read (xe,,yz),x.y) Inorder to take advantage of the qualia -based information in the comple ment , the verbal semantics for open can be formulated so as to mod ~ lly embed the relevant expression from the semantics of the compleme phrase . A revised statement of the semantics of the verb open ispre sented in ( 5 ), where the modal statement making reference to the 0 b ject 'sTELIC isexplicitly given . (5)

Giventhesetwoexpressions , thequaliastructure resulting fromcocomposition withintheVPcanbeexpressed asin (6).

223

I .-.

--

Consequences of a Generative Lexicon

-

-

What this expressionsaysis the that the result of the open_act is a state asserting the possibility of the activity inherent in the TELICrole of the complement. This is another example of co-composition, where qualia unification results in a specific verbal sensein context, without the need to enumerateindividual senses. The sameverbal representation for open in (5) gives rise to the appropriate (but distinct ) meaning of the verb in (2c) above, where the complement is the NP the door (cf. chapter 6 for discussion) : door

(7) QUALIA =

~~LIC = walk _through ( e,z,y ) [ physobj .aperture -lcp ]

I

In this case, the denotation for the VP [v p open the door] makes referenceto the ability to walk through the aperture as a result of the action of opening: the _door

[El=el:process ] RESTR

=

-

-

< cx:

.

E2 = e2 : state

0 2 physobj .aperture -lcp 0 door QUALIA = [TELIC = 0 = walk _through (e,ztY )]

224

Chapter 10

In some sense, these functionally dependent verbs are modal causatives making referenceto the qualia of their arguments. For this reason, we will call them qualia modal causatives(cf. Pustejovsky, forthcoming) to distinguish them from the classof modal causativessuch as risk. Other examples of this type of underspecification with co-composition include the verbs breakand close, which both make referenceto the TELlOrole of the complement as well, but in different ways. Briefly, the TELICrole for the complementof the verb breakcan be viewed as modally subordinated within the following context:

NP

(9)

0

. ..

QUALIA =

TELIC

=

0 =c/J]

= P ( e2, -' 0 [ 0 ] ) ]

That is, the resulting state of a transitional event of breaking refers to the inability to usethe object for that which it is intended; i .e., its TELlO role.! .I

The above discussionof functionally dependent verbs raises the issue of underspecified meanings more generally. Namely, when is it appropriate to "pack" many meanings into a single lexical representation, either with devicesfor semantic underspecificationor by treating the item functionally , and when should one "unpack" several sensesfor a single lexical item , even when they are logically related? In the caseof verbs such as breakand open, co-compositional operations give rise to the contextualizing effects that the complementshave on the phrase. Similarly, for casesof semantic type selection with multiple syntactic forms, such as begin and believe, the underspecification brought about by coercion seemsgenerally correct and well-motivated by the data. Some casesof nominal polysemy, however, are not as straightforward , and complications arise when trying to account for all logical polysemies in terms of an underspecified representation. In particular , sensepairs related by "grinding" operations, as with the nouns lamb and haddock, are not homogeneouslyrepresentedas single meta-entries, and may involve the application of lexical rules giving rise to senseextensions (cf. Copestake

Consequences

and

Briscoe

involve nouns

10.2

of a Generative

, 1992 , 1995 ) .2 The

" sense packing such

nominal

" are those

as lunch , sonata

225

Lexicon

analyzed

, and

polysemy above

cases that as dot

objects

do seem to , namely

,

book .

Stage-Level Predication

As mentioned in chapter 2 above, the stage-ieveljindividual -level distinction in predicate types first made by Carlson (1977) is an important criterion for distinguishing the semantics of predication . Given the descri ptive possibilities of qualia structure , one might view stage-level predicates as similar in some way to artifacts , and individual predicates as similar to natural kinds . Seen from this perspective , the distinction between

states

such

as be tall

and

be angry

is not

a difference

in event

type , as argued in Pustejovsky (1991b) , where stage-level predicates were analyzed as event transition functions , but rather is due to the presence or absence of reference to the mode of explanation which brings that state about , namely , the AGENTIVE quale .3 Thus , rather than simply typing stage-level and individual level predicates distinctly , as Carlson proposed , the semantics should describe how a state is changeable . What distinguishes stage-levels from the general class of stative predicates is the inherent reference to that factor that brings this state about ; as mentioned above, there is reference to the "coming into being " factor , the AGENTIVE role in the qualia . That is, a stage-level predicate is an artifactual state. In order to make this proposal more explicit , let us develop the similari ty to nominal artifacts more closely. Just as an artifact is defined by virtue of its AGENTIVE quale making reference to a specific event description of how it comes into being , let us assume that a changeable state also makes reference to such an event . .As suggested in chapter 8, subevents , like arguments , may also have a default status in the event

structure . For example, in (10) below, the predicate a is typed as a state , but

reference

to a default

event , e2, binds the argument

in the

stative predication to a relation preceding this state , in other words , the cause

of the

state .

226

Chapter 10

(10)

There are several reasons for thinking reference to default

that a lexical item might make

events , as opposed to the non -expression

of an event

due to lack of headedness. The function of heading an event , recall , is to both focus that event within a larger structure of events, as well as to filter

the expressive

To illustrate

output

from the semantics

the consequences

to the syntax .

of this proposal , consider the sentences

in (11)- (13) , involving the psych-stage-level predicates angry, nervous, and upset .

(11) a. John is ~~~~ at the newspaper. b . John is angry from reading the newspaper .

(12) a. ?There are musicians nervous. b . There are musicians nervous about tonight 's competition .

(13) a. ?There are Americans ~E~~! . b . There are Americans

upset with the way the President

is han -

dling foreign policy . As Diesing ( 1992) points out , the there-insertion construction is a generally reliable test for pulling out the existential interpretation possible

with stage-level predicates. Although (12a) and (13a) are marginal, notice how the PP -adjunct makes reference to the Quale identifying the predicate as an artifactual state . The result in each case is a perfectly

acceptable stage-level predication (cf. (12b) and (13b)). By specifying the AGENTIVE

Quale value in the qualia

and upset, the predicate is appropriately

structure

for angry , nervous ,

identified for an existential in -

terpretation . The lexical semantics for angry is given in (14) below, illustrating

how causative adjuncts such as from reading the newspaper

act to specify

the AGENTIVE role for the predicate .

227

Consequences of a Generative Lexicon

(14)

The event of reading the newspaper is unified with the experiencer pro cess identified as e2 in the structure above. Notice that this is possible only because read is a subtype of experiencer predicates . Unlike its associated causative , anger, the predicate angry remains stative even when reference to the causing event is present through an adjunct expression , as in this

case .

Assuming that this approach to stage-level predication is correct , let us now briefly revisit the problem of resultatives first mentioned in chapter 8, in the context of co-composition . As argued in Pustejovsky (1991a) , the adjectival phrases entering into resultative constructions are just those stage-level predicates which "cohere" with the process predicate present in the phrase . For adjectives such as clean and flat , as

used in (I5b ) and (16b) below, (15) a. Mary waxed the car. b . Mary waxed the car ~l~~ .

(16) a. John hammered the metal. b . John hammered the metal ~

.

composition with the VP can be accomplished by a modified version of qualia unification presented in chapter 8. To illustrate this deriva tion , consider the sentence in (15b) . The lexical representation for the

.I

adjective clean is that given in (17).

228

Chapter 10

EVENTSTR =

(17)

ARGI -

Here I am assuming that R is a predicate that is sortally structured to subsume wax, wipe, and related predicates. The stage-level predicate acts as a function over the VP wax the car, but under qualia unification , the FORMALrole of the adjective unifies with that of the VP, resulting in a derived causative and an aspectually telic interpretation (cf. Dowty, 1979).

(18) e < oc

~_/

el

/

/ -/ "

" " " " """'"

e2

I

I

(wax_act (m , the_car)]

clean(the_car))]

~"he structure in (18) is basically that of the default causative paradigm encountered in chapter 9; the difference, of course, is that it is a derived interpretation and is not lexicalized into one lexical item. In somesense, the stage-level predicative construction correspondingto (I5b ), namely, The car is clean can be thought of as a potential unaccusative. The proper unaccusative for (I5b ) would be (19): ( 19) * The car ~ ~ ~4 clean.

Obviously, the crucial difference here is an aspectual one, where the adjective can denote only a state and not a transition , as with unaccusatives. Stage-level predicates, therefore, facilitate or licensereference to that which brings about that state, but only in specific constructions under co-compositional operations.

229

Consequences of a Generative Lexicon

To close this section , it is interesting to see whether the distinction between stage-level and individual -level predicates as construed above might extend to the nominal domain as well . There is a difference between "role-defining " nominals such as physicist , linguist , and violinist , and "situationally -defined " nominals such as pedestrian , student , passenger , and

customer

. I will

refer

to these

two

classes

as individual

- level

nominals (ILNs) and stage-level nominals (SLNs) , respectively. Observe that , whereas ILN s successfully denote even when the defining characteristic is absent , SLN s denote only if there is existential force on the predicate defining the nominal .

(20) a. The yiolinis_t is eating lunch at the cafeteria. b . We met the physicists from the institute on the trail .

(21) a. The passengersare eating lunch on the plane. b . Bill is helping a customer .

The NPs in (20) still felicitously pick out the individual (s) denoted by the ILN s because of the generic nature of the interpretation . Hence , even though the TELIC Quale for violinist refers to "playing the violin ," it need not be an activity

or state that

holds at the time

of reference .

Similarly , "doing physics " in some official capacity would arguably define an individual as a physicist , but one need not be engaged in this activity in order to be individuated as one. Here again , the generic force of the occupational classification transcends any particular situation which that

individual

is part of .

Notice that the SLNs in (21), however, properly denote only if the activity defined by the noun is existentially closed; that is, in (21a) the passengers, defined in terms of the process of "travelling ," must be engaged in this activity in order to be identified by this term . The same

holds for customer in (21b) , where it is a particular situation (i .e., "shopping " ) that determines the conditions for which this term will properly denote . There are cases, however , where a SLN may appropriately refer even after the situation identifying the individual has occurred , as in (22) .

(22) Hey, you're the passengerfrom my flight yesterday! What is important to point out about such examples , however , is that the situation referred to by the NP the passenger is still existentially quantified over , unlike the generic readings available for ILN s.

230

Chapter 10

This in

distinction

qualia

identified

,

with

generic the

can

structure the

readings noun

be

brought

where

Quale for

in

the

situational

AGENTIVE

available

, cr .

out

the

ILN

semantics reading

of

the

s are

as of

noun

an

( cf .

associated

a

difference

SLN

can

( 23 ) ) ,

with

the

be

while

TELlO

the role

of

( 24 ) .

pedestrian

QUALIA

( 23 )

=

~ ~ ENTIVE [ FORMAL

=

= walk X

=

~ ~ LIC [ FORMAL

play = X

_ act

( e ,x )

]

violinist

QUALIA ( 24 )

Pustejovsky a

( 1989

nominal

is

( 1990

) makes

with

the

use

TELlO

quale

on

the

where

event

In

section

this

proach the in

in

the

an

is

SLN

in

, other

described

I

briefly

in devices

of

this

the

work to

widely

explored

how

, I diverse

to

the

the

way

existential

Busa

in -

of

the that

closure

( forthcoming

) ,

.4

Coercion

type

grammar are

in

detail

of

Moravcsik

genericity in

an

role

associated

Applied

, the

involving

greater

of

review parts

is

while

distinction

as

quale

studied

.

above

TELlO

typically

statement

nominals

the ,

most

appropriate

the

to

inference

activity

generic of

AGENTIVE

reference

defeasible

Applications

generative language

the

a

provide

Characterizing in

Further

is classes

distinction

10.3

adopted

to

how

that

object

) ]

and

two

seems .

this

an the

( e , x , y : violin

discusses

observation

of of

denote

)

prototypical

the

terpretation

they

a

, 1991a

=

coercion .

believe

The , seen

phenomena

might advantages in

the and

be of

usefully the

application

ap of

constructions

.5

Coercion in Temporal Connectives The standardinterpretation of the temporal connectivesbefore , after, while, and during, involves selectionof an interval or event. Remainingneutral on this issuefor now, it is clear that in order to accountfor the selectionalbehavior exhibitedbelow, we must either posit two lexicalentriesfor theseitems or reconstructthe interpretationsin terms of type coercion. Consider,

Consequencesof a Generative Lexicon

231

for example, the selectional distinctions available for the connectives beforeand after compared to that for the preposition during. (25) a. b. c. d. e. f.

I will call you b. fuI~ John gets here for his appointment. I will call you b. fu!~ my next appointment. My next appointment is John. *1 will call you .Q ~fu~ my next appointment, who is John. *1 will call you Q~fu~ John. Let 's examine Mary ~ fuI. John. (ellipsis)

(26) a. Let 's leave ~.fuI we drink the coffee. b. Let 's leave ~fiN dessert. c. Let 's leave ~.ft~ the coffee. (27) a. *1 will call you ~~~~~ my next student. b. *1 will read it during my coffee. While before and after seem to generally license coercion, during does not . Many of these sentencespoint out that coercion can only facilitate type satisfaction, and pragmatic or contextual factors may be required to determine what the full meaning of a reconstructed interpretation is. The interaction between lexically-encoded linguistic knowledge and pragmatic effects on interpretation is studied in more detail in Asher and Lascarides (1995) and Nunberg (1995) . Functional Prepositions and the Role of Qualia There is an interesting paradigm that has received little attention in the literature , concerning the use of English prepositions such as at and in to specify functional involvement. Consider the following sentences. (28) a. John is ~ his desk. b. John is by his desk. (29) a. The student is ~ the board. b. ??My 2-year old daughter is ~ the board. (30) a. John is ~ the office. b. I left my computer ~ the office. (31) a. Zac is ill school from 9:00 am to 3:00 pm.

232

Chapter 10

b. The computers are finally ill the school.

(32)

a. Mary is ill hospital with a broken leg. b. Mary is ill the hospital visiting her brother.

There are two things to observe about these data and the broader paradi gm . First , the preposition at seems to only select locations that have in their qualia structure a specific TELIC role . For example , blackboards are used for writing on , the desk is used for work , and so on . Secondly , when an individual is predicated of an at- PP , it suggests that the individual is participating in the TELIC role taken from the preposition 's complement . That

is , if I am at my desk , then I am situated

in the specific orientation

which would enable me to perform the activities typically associated with the desk. Similarly , if I am at the board , then I am in the process of or intending to or being forced to write something on the blackboard , i .e., the TELIC of the NP . Notice that if such an activity cannot reasonably

be fulfilled , as with (29b) above, then the sentenceis semantically odd, since my daughter doesn't have the ability to use the board for its proper

function (cf. Li , 1994).6 There is another parameter at play here involving definiteness and

the bare singular predicative readings in (31a) and (32a). Namely, for many TELIC-specified nominals , most with dot object status , such as hospital and school, when they appear as bare singular expressions as

complement to prepositions such as in (and at), they predicate not only the location of the individual

but also the activity associated or events

characterized as the function of that nominal . Thus , being " in hospital " is to be a patient , and being " in school" is to be a student , and so on . Although it is not completely clear what the constraints are on such interpretations , what is clear is that the qualia are contributing essential information to the interpretation of the expression in a way that is not a strict

coercion

.

10 .4

Linguistic

versus Commonsense

Knowledge

From the discussion presented above, the boundary between what we for mally take to be linguistic or lexical knowledge and that which is sometimes referred to as "commonsense knowledge " might appear fuzzier than ever .7 Once we start enriching our lexicon with information that ,

233

Consequencesof a Generative Lexicon

to a linguist , appears better suited for a knowledge base, there may ap-

pear to be no systematic means to judge where to stop (cf. Herzog and Rollinger, 1991, and in particular , Lang, 1991). Yet, the fact that there appears to be a continuum between these two types of knowledge doesn't mean that there are not clear cases of paradigmatic linguistic behavior that are better treated as language specific knowledge , rather than in terms of general inferencing mechanisms . One of the goals of the present work has been not only to argue in favor of richer lexical representations and compositional mechanisms , but also to show that these structures and devices are still language specific in identifiable ways . What is dif ferent about the approach taken here is that there are clear and obvious means to interface lexical knowledge with commonsense and pragmatic inferences

.

For example, recent work by Asher and Lascarides(1995) investigates how discourse structure can affect the selection of lexical senses, focusing in particular on the mechanisms whereby lexical semantics affects and contributes to discourse interpretation . To this aim , they integrate three components

:

(A ) a theory of discourse structure called SDRT(cf. Kamp and Reyle, 1993) , which represents discourse in terms of rhetorical relations that connect together the propositions introduced by the text segments;

(B) an accompanying theory of discourse attachment called DICE (cf. Lascarides and Asher , 1993) , which computes the rhetorical relations holding between constituents , on the basis of the reader 's background information

; and

(C) a formal language for specifying both the syntactic and semantic knowledg for a lexical item , called the LRL, lexical representation

language (cf. Copestake and Briscoe, 1993), making use of a typed feature structure logic . By integrating information

these separate flow

in both

components

directions

, from

discourse to words . For the mapping Lascarides

show how the LRL permits

, they words

are able to model to discourse

, and

the from

from words to discourse , Asher and the rules for computing

rhetorical

relations in DICE to be generalized and simplified , so that a single rule applies to several semantically related lexical items . From discourse to words , they encode two heuristics for lexical disambiguation :

234

Chapter 10

. disambiguate words so that discourse incoherence is avoided ; and . disambiguate words so that rhetorical connections are reinforced . With these heuristics , several cases of lexical disambiguation can be analyzed , that have until now been outside the scope of theories of lexical processing . Asher and Lascarides show how lexical processing can work in service to a theory of discourse attachment . The knowledge resources encoded in a theory of discourse attachment , however , are also useful to lexical processing . Consider the following examples and the ambiguities in them concerning the words plant , bar and dock. (33) a. They ruined the view . b. They improved the view . c. They put a plant there . (34) a. The judge demanded to know where the defendant Ross was. b . The barrister mumbled apologetically , and said that Ross had last been seen drinking heavily . c. The judge told the bailiff to escort Ross from the ~

to the

dock .

They argue that bar in the second example is disambiguated to its "drinking establishment " sense on the basis of constraints on coherent discourse . In contrast , plant in the first example is disambiguated on the basis of strengthening the rhetorical link between the sentences. The inference which leads to this disambiguation is driven by the lexical semantic information associated with the qualia structure for the words . The examples above illustrate the disambiguation of contrastive senses with the help of contextual information . These are distinguished from the cases of logical polysemy studied in this work for reasons outlined in some detail in chapters 3 and 4. There are, however , some types of sense

extension

that

I have

not

discussed

in this

work

, and

which

do

require some mention in this section . These are cases of displaced refer -

ence, studied in Nunberg (1979,1995), Fauconnier (1985), and Jackendoff (1992b). What distinguishes this from logical polysemy is the lexically idiosyncratic nature of the ambiguity , as well as the semi-productive status such extensions have in language . These involve sentences like those given in (35) below . (35) a. I am parked out back .

Consequences of a Generative Lexicon

235

b . Ringo squeezed himself into the parking space. These sentences illustrate

two types of referential

transfer : a type - mis -

match between subject and predicate in (35a), where it is the car that is parked , not the individual ; and a mismatch between verb and object , together with a non-identity between antecedent and anaphor in the

binding relation in (35b). Nunberg refers to such extensions of meaning as predicate transfers . In particular , he argues against a metonymic

analysis, where the subject I in (35a) and the object himself in (35b) are interpreted as my car and his car respectively . Rather , his position is that there are pragmatically -licensed conditions which allow the pred icate to extend its sense, where it is retyped to select for the subjects that are present in the syntax . The operation of predicate transfer allows a name of a property to be mapped into a new name denoting a property to which it functionally

corresponds , as with the meaning of

the predicate parked out back in (35a) above. Predicate transfer is responsible for the ability of this predicate to both refer to the car , but also to take syntactically as its subject , the driver of the car . According

to Nunberg's (1995) formulation of the phenomena, predicate transfer is sub ject to two general conditions ; . the basic and derived dence

to one another

property

must stand in a functional

correspon -

;

. the derived property should be a "noteworthy " feature of its bearer . N unberg argues that reference to predicate transfer allows us to main tain a very strict definition of syntactic identity , thereby ruling out all cases of "sortal crossing ," where a term appears to refer to things of two sorts at the same time , as in examples like Ringo squeezed himself into a tight space; in such a case, the reflexive is strictly coreferential with its antecedent . N unberg claims that these observations enhance the reliability of "zeugma" tests for ambiguity , while also highlighting a theoretical difficulty in distinguishing polysemy and generality . Further more , he claims that these results pose a difficulty for the view argued here that there is a legitimate distinction between logical polysemy and more general operations of sense transfer such as metaphor .8

Consider next the sentencesfirst discussedin Jackendoff (1992b), for the effect of predicate transfer on reflexivization .

236

Chapter10

(36) a. Ringoi squeezedhimselfi into a narrow space. b. Yeatsi did not like to hear himselfi read in an English accent.

Because of the restricted conditions on when such co- predications are allowed, Nunberg suggeststhat these neednot be casesofsortal crossing, but rather instancesof predicate transfer, where the individuals denoted by the subject expressionsare fixed, and it is the predicate which changes its sense. The reliability of the zeugma as a test for determining the polysemy of a word is preservedwith this interpretation . What emergesvery clearly from the recent work on the interface between lexical and non-lexical semantic information is that polysemy is not a single, monolithic phenomenon. Rather, it is the result of both compositional operations in the semantics, such as coercion and cocomposition, and of contextual effects, such as the structure of rhetorical relations in discourse and pragmatic constraints on co-reference. What is necessaryis for researchto tackle the difficult question of how other components in the natural languageinterpretation processinteract with the lexicon to disambiguate and fully determine the semanticsof words in context. This work, emergingfrom very different traditions , illustrates how lexical semantics can be made sensitive to sentencelevel compositional processesas well as discourselevel inferencemechanisms, reacting to the diverse and multiple causesof lexical ambiguity.

10.5 Lexical Inference and Rhetoric If we look at the relation between inferential structures within compositional semantics and the larger structures of inference associatedwith rhetoric , some interesting parallels emerge. These can perhaps best be illustrated by examining the structure of enthymemic in"ferencein the structure of rhetoric ; that is, the conditions which license ellipsis in persuasive discourse. Within Aristotle 's view of how logic and rhetoric interact (cf. Rhetoric), rhetoric is viewed as the application of the general inferential mechanismsof the logic to a particular context, for a specific goal. Inference is put to use for persuasivediscourse, and that which is persuasiveis evaluated only in referenceto some specific individual ; becauseof this , the mode of ellipsis depends on the rhetorical structure employed.

237

Consequencesof a Generative Lexicon

Consider the form of an enthymeme , which is a set of two proposi tions

offered

as an argument , such that

the addition

of a third

results

in a categorical syllogism . When the ellipsed proposition is the major premise , then we have what could be termed enthymemic induction . In this structure , propositions can be viewed as being coerced to a pat tern of coherence to satisfy the inference . The following illustrates this pattern , where the bracketed expression is the ellipsed proposition . A . J ames

is a Texan

B . Therefore

, James

. is tall .

c . [All Texans are tall .] A 1\ [c A ' - t B '] =>- B The assumed belief is that the speaker and hearer share ellipsed propo sitions

as common

belief . The speaker , by using such a rhetorical

form ,

is establishing coherencebetween (i .e., coercing) the propositions for the hearer

.

If the ellipsed proposition is the minor premise , then an enthymemic abduction results . This is illustrated by the following argument . A . Every American loves baseball . B . Therefore

, John

loves

baseball

.

C. [John is American.] [A 0 ' --t B '] !\ B = > 0 The assumed belief here is that the speaker and hearer know that the ellipsed proposition is true . The speaker in this case is using a non-analytic proposition as the major premise . This is the coercive assumption from which , with the common belief of C , the syllogism has rhetorical force . Finally , if the ellipsed proposition is the conclusion , then a simple enthymemic deduction results , shown below . A . All politicians are corrupt . B . Berlusconi is a politician .

C. [Berlusconi is corrupt .] [A Bf ~ Of] !\ B ~

0

238

Chapter10

Common knowledgeprovides the valid rule of syllogistic reasoning. But the speakerdoesnot completethe syllogism, leaving the inferenceas new and topical information conveyedor projected by the discourse. The effect of coercion and co-composition, when projected to discourselevel interpretations , could be viewed as a type of lexical enthemymic inference. Consider the example below, where the qualia contribute information that can be used in the discoursein a method similar to the inferencesdiscussedabove. A. B. C. C'.

Steven Kingbegan a newnovel . [Steven Kingis a writer.] AGENTIVE (novel ) = AzAx .YAeT [write(eT,z,x.y)] Steven Kingbegan to writea newnovel .

As argued in chapter 7, coercion requires that the verb begin selects an event function as its complement, but the interpretation is not fully determined by the lexical semantics alone. The knowledge of what the type of the NP Steven King is, and what qualia values mayor may not contribute to further specifying this interpretation , can be seen as discourse inferences that are biased by the compositional semantics of the sentence. Similar remarks hold for data mentioned in chapter 4 above, where the full interpretation of the type required by the coercion on the complement is logically separate from the type coercion itself. (37) a. Most commercial pilots ~~~ E Kennedy to Logan. b. Most commercial pilots prefer New York to Boston. These comments are merely suggestive, but establishing the contribution of lexical semantics to pragmatic and contextual inferences is an important area of research.

10.6

Summary

In this book, I have presented a framework for the semantic analysis of natural language, motivated by the following theoretical and computational concerns: (1) Explaining the interpretation of words in context;

Consequencesof a Generative Lexicon

239

(2)

for wordsfromfinite Deriving a potentiallyinfinitenumberof senses resources;

(3)

Accounting for the systematic relatedness between word senses in a formal and predictable way ;

(4)

Characterizing natural languages in terms of their polymorphic prop erties ; more specifically , studying what types of polymorphisms exist and how productive they are in natural language ;

(5) Providing

a semantic representation that is applicable cross-categori ally , and is not restricted to verb semantics alone .

By structuring the representation of words in a manner completely par allel to that of syntactic objects , we obviate the need to postulate lexical primitives , since the natural language vocabulary is itself the data set of primitives . Just as the generativity of syntactic structures is the major concern of syntactic theory , accounting for the generativity of senses in a language is the primary goal of lexical and compositional semantics . This , in my view , characterizes what a generative lexicon should be. There are obviously many questions that have been left unanswered in the current work . For example , issues relating to parsing , details of interpretation concerning quantification , and connections to pragmatic and commonsense inferencing mechanisms have not been seriously dealt with here , although they have contributed to the design of the frame work . I have attempted to present a model of semantics for natural language that provides the necessary lexical resources to perform these tasks , while simplifying and , I hope , clarifying the role of the lexicon in linguistic theory .

Notes

Chapter 1.

1

Weinreich (1964) , in making this distinction , was critiquing the somewhat arbi trary

manner

with

which

the

Webster

' s Third

Dictionary

handled

lexical

items

with multiple word senses. The observation shows an insight into the problem of distinguishing accidental sharing of lexical forms from linguistically motivated sense relations , that others did not generally recognize .

Chapter 1.

2

Perhaps the most notable exception to this claim or movement would be Fill more 's recent move towards the encoding of larger units of phrasal structure ,

known as constructions (cf. Fillmore , 1985) . Goldberg (1994) extends this perspective to a fairly broad range of phenomena , showing how constructions can adequately account for cases of polyvalency and polysemy . In some sense , this is not contradictory to my claim above , since larger structural units are essentially

lexicalized rather than subject to general rules of composition (cf. chapter 8 below) .

2.

This is still a contentious

point

and is an issue that is not at all resolved in the

community . Hobbs (1987) and Wilensky ( 1986,1991) , for example, question the distinction between commonsense knowledge and lexical knowledge . Even more fundamentally , some previous research in NLP and computational linguistics ,

such as Wilks (1975) and Schank (1975) saw no utility in distinguishing knowl edge of the world and knowledge

of how we talk about the world . More recently ,

Wilks ' work (cf. Wilks et al., 1988,1993) has pointed to examining how faithfully lexical knowledge

represents our knowledge

of the world , as encoded in machine -

readable formats such as dictionaries , e.g., LDOCE ( cf. Procter , 1978) . Schank has not recently addressed the issue of knowledge representation as it relates to language . Nevertheless , I will suggest below that there are good reasons , both methodological and empirical , for establishing just such a division . Pustejovsky and Bergler ( 1992 ) and Pustejovsky (1993 ) contain a good survey on how this

issue is addressed by the community . Saint-Dizier and Viegas (1995) compare some recent approaches

to addressing

this distinction

more formally . Cf . also

Atkins and Zampolli ( 1994) for approaches to integration of lexicographic and computational

methods .

3. The context within which an utterance is situated must reflect the state of the discourse, as many have recently argued, for example, Heim ( 1982) , Kamp and Reyle ( 1993) , Asher (1993), and recent trends within situation semantics. Just as important , however, is some notion of the purpose of the utterance . Grosz and Sidner (1986) , Pollack (1986) , Cohen, Morgan , and Pollack ( 1990) , and others have rightly stressed the importance of goals in laying the contextual groundwork for the communicative act . Thus , in some sense, the rhetorical structure of the discourse defines what the speaker wants to accomplish (cf. Hovy , 1993) . I will return to this point in chapter 10. 4. See Dixon (1991) and Levin (1993) for discussion. 5. The generalization , as Levin (1993) seesit , bears on verbs denoting both contact and motion (or perhaps more correctly , motion towards contact ) . Notice , however, that (9b) above becomes much better if we contextualize the sense with adjunct modification :

242

Notesto pages1023

(i)?Underthetable,thecatkepttouching at mylegwithits frontpaw. The issueof sensecontextualization willfiguremoreprominently later in our

discussion of co-compositionin chapter 7.

6. Aswewillseebelow,however, eventhisdistinction requiresmodification, since allthreeofthelexically marked activity verbscanphrasally orsententially appear

in transitional(i.e.,telic) contexts.Seechapter9 belowfor further discussion. 7. I arguein Pustejovsky(1995c)that the aspectualdistinctionin the indefiniteNP deletioncasesis in fact sufficientto explainwhichverbslicensethe alternation.

Forthe broaderclassof complement-drop phenomena, however, including the infinitival complement casesin (17)and(18)above,moreisinvolved thansimple

aspect.

8. On details of this, see Dowty (1979).

9. It is usefulhereto pointoutthe distinction betweentemporalculmination, where an eventof whateversort simplystops,and logicalculmination, wherebysomethingis fulfilledor finishedas a resultofthe activity(cf.alsoParsons,1990and Tenny, 1992).

10. Theissueofvariableaspectuality isdiscussed insomedetailin chapter9 interms of polysemy and event coercion.

11. Althoughthere are severaldiagnostics for indicatingmembership in oneor the otherclass,manyof themare simplyinadequate to accordwithspeakerintu-

itions. For example,both there-insertionand progressivetests seem to favor stage-levelpredicatesoverindividual-level ones,but there are manycounterexamplesto the generalization. I discussthe generalissueofstage-level predication below in chapter 10.

12. For crystallographers, of course,thereis a verydifferentexperiencewithsand as a material. They mightverywellreferto individualgrainsof sand as sand.

Theclassification ofa materialorthingin theworldis obviously drivenbyhow

we perceiveit and constructthe realityfor that object,givena certain modeof measurement.See Modrak (1987)and Hacking(1983). 13. I return to a discussionof this issuelater in chapter 7. 14. Chomsky(1957)seesthis as a clearmismatchbetweenthe structurethat Markov

modelsassign to suchsentences andthe underlying representations necessary to them.

understand

15. Anypropertyis of coursepossibleas the complement of easyor difficult when overtcomplementation is used,as illustratedin (i) and (ii): (i) a. That exam was diffiylt to grade.

b. The examhas beeneasyto write,but its goingto be verydifficultfor the students

to take.

(ii) This highwaywillbe easyforthe workcrewto pavein the summer.

Defaultinterpretationsof ellipsedexpressions willalwaysgivewayto overtlin-

guistic expressions.

16. Thisdefinition ignoresthe issueof opacityandnon-substitutability in opaque contexts.Katz (1972)has a usefuldiscussionon the generalnature of semantic

relations, whileBierwisch andSchreuder (1992) discuss theissuesofrepresentationandaccessforthedifferent semantic relations between lexicalitems.Apresjan (1973a)examines the linguistic contextsforsynonymy in particular,and howit isstructurally identified. Sparck Jones(1986) presents a helpful studyof criteriaforsemantic classification andtheidentification oftermsassynonyms.

243

Notes to pages 24- 43

17 .

The issue of presupposition and how lexical meaning relates to projection cannot be given a full discussion in this work . I will return , however, to these issues briefly in the context of factivity in chapter 7. Soames (1989) presents a helpful guide to most of the recent approaches to the treatment of presupposition in nat ural language.

Chapter 1.

2. 3.

3

Hirst treats taxiedasambiguousin this sentenceaswell. Although the uninflected form taxi is certainly ambiguous,the past tensemarkerhereactsto disambiguate the form, thereforewe neednot considerit for disambiguationpurposes. Seeoriginal articles by Katz (1965), Vendler (1963), on the useof "good." But we return to this in the next chapter. In

some

sense

proposed

,

interesting

a

the

raise

is

clear

in

Fodor

that

have

for

to

biguate

.

item

SYNSET

it

,

the

set

of

extensionally

i

)

CAT

(

ii

=

count

)

=

This

6.

of

1980

and

Chapter 1.

exact

most

Pollack

in

et

al

of

(

2

. ,

word

,

,

approach

properly

disam

1990a

to

)

.

On

information

)

)

namely

synonyms

;

,

i

. e

with

will

for

.

. ,

example

-

aid

this

in

view

also

those

,

store

a

words

This

that

would

help

in

:

(

1987

,

1988

)

.

and

)

,

to

Bookman

representations

Sag

(

1994

selection

is

do

reported

on

Memory

-

(

1994

in

Based

vary

several

Retrieval

not

trivial

,

varies

same

is

recently

.

performed

the

Strategies

and

)

are

selection

acknowledge

senses

,

lexical

such

and

techniques

Hirst

like

and

for

search

contrastive

,

-

Pollard

mechanisms

sense

Passing

1987

above

Aspects

of

researchers

of

,

given

,

-

1988

noun

work

word

biguation

Marker

sets

the

sur

,

raised

to

GENUS

with

Not

enumerative

distribution

line

with

HPSG

the

,

sense

Miller

.

border

starts

the

such

system

.

)

]

in

how

had

features

and

)

Mehler

-

of

noun

-

=

approach

Although

cf

associated

line2

codified

(

CAT

1986

that

message

count

SYNSET

5.

item

is

]

CAT

[

-

=

the

)

It

. "

using

complementary

of

critic

selectional

suggested

to

synonyms

sense

SYNSET linel

[

lexical

addition

(

to

and

Theory

,

1963

con

identical

Pinker

Fodor

Semantic

of

have

in

the

-

(

.

disambiguation

McClellan

in

Katz

system

Fodor

forms

sense

papers

ambiguities

a

in

appears

distinguishing

(

,

the

and

almost

and

.

in

rich

of

would

word

senses

strongest

researchers

senses

lexical

lexical

Explorations

fairly

Some

disambiguating

cf

contrastive

a

senses

the

some

employ

(

Katz

ambiguous

Rumelhart

work

that

"

.

since

representing

at

of

( cf

this

essay

much

for

attempts

theory

of

his

that

system

model

issues

,

changed

connectionist

critics

same

Weinreich

It

-

the

the

' t

same

structure

Katz

,

will

early

feature

early

prisingly

a

haven

the

that

structed

4.

things

essentially

basic

quite

and

bit

details

system

strategy

a

for

,

of

(

such

cf

to

disam

course

implementations

techniques

the

from

,

.

-

from

Hendler

as

Waltz

)

4

The context-dependenceof the predicatesmentionedabovewas noticed by Aristotle in the Categoriesand the NicomacheanEthics, where he speaksof the particular aspectof an individual to which is attributed a quality such as goodness; for examplea goodlyre-player or a goodsculptorspeaksto the function of

244

Notes to pages 44- 60

particular ness

aspect

; for

that

in

, 1967

)

. of

chapter

Examples

,

.

These

Street

who

to

a

of in

polysemy chapter

)

and 10 .

and

the

it

or

the

.

with

strategy

meaning

as

.

of

Vendler

adjectival

in

I

good

function

sculpting

interacts

general of

such to

Aristotle

return

to

's

this

in

( 1964 extensions

by

psychological

( 1974

the

matrix

I

( ii )

know

I

The

how

don

't

matrix

just

as

)

* 1 don

Whatever Van

tall

't

filters Valin

member

,

and

the

examines

the

this

question

See

Verma

and

constructions

Morrill

and

positionality

as

the

constituent

tion

, we

See

Pustejovsky

do

.

rich

senses

and

( 1973b

, 1981

study

of

Nunberg

are

and

the

's

) , of

Viegas

( 1994

and

Pustejovsky of

on

to

) ,

proposals

in

classes

perspective

how

pragmatics

recent

explored

related

and

adjectives

the )

,

polysemy

and

Bouillon

.

from

" questions

"

whetherwh

on

the

basis

of

- complementation and

with

tests .

questions

Ob

-

disappears

address )

)

for

make

a

degree

of

the

sentence

dimension further

filter

the

analysis

factive

in

the

verb

of

complementation

a

related of

verb

( 1990

of

the

extensive

,

) , in

survey

a

of

( 1993

and

the

)

syntac

somewhat

understand

) .

re and

Jackendoff

intend

verb

( iii

patterns

manner the

Wheeler

an

the

for

In

to

interpretation

-

related .

I

return

experiencer

to

subject

.

to

this

appears

properties

)

complement

possible .

it , while

( 1991

)

interesting

structure

languages

( 1990

also an

verb

different

relates

question

exclamation

its

the

selectional 7

indirect

reading

Mohanan

( 1994b

a

exlamations

provide

conceptual

structure not

tall

.

it

is

word

general

to

the ) :

below

Carpenter

cor

( BCET

Dictionary

Apresjan

examined "

between

chapter

in

) ,

to

Bouillon

( only

( iii

)

of

associated

in

is

( 1993

underlying

the

) , Longman

related

- linguistic

are

blocks

exclamation

relation

, discusses

cross

between

is .

in

semantics

realizations

manner

he

he

Wilkins

underlying

)

this

sensitivity

( i )

( ii )

that

the

and

tall

factive

know

various Text

is .

in

the

from

English

( ii ) .

he

how

blocks

a

, see

in

negation

it

, for

, and

in

know

in on

" exclamations

negation

taken of

return

adjectives

adjectives

ambiguity

of

( 1979 we

details

adjectives

, polarity

are

( RD

( 1983

, and

Also

) distinguishes

negation that

(i)

these

issue

) , Nunberg

of

) .

,

Digest

Bierwisch

further

( 1995a

book Collection

general

are

For

the

) , Readers

the

sense

) .

of

) .

properties ( 1994

) , where

Elliott

( WSJ

, Katz

( 1995

with

9 .

on

Birmingham

addressed

Pustejovsky

serve

8 .

lyre

how

modes

quality

speaks

the

and

at

a

sculptor

playing

expands

( LDOCE

demonstrated

tic

)

attributed

good

at

remainder the

semantically

subselectional

the

is

a

function

looking

the :

, 1989

extent

involving

7 .

in are

have

lesser

The

and

them

( 1993

6 .

i .e . ,

of

( 1975

English

Others

( iii

or

,

role

by

Journal

represent

see

which

- player

capacity the

Moravcsik

here

sources

Wall

5 .

that

understanding

Contemporary

4 .

to

lyre

6 .

pus

3 .

individual

good

discusses

modification theory

an

a

individual

( 1963

2 .

of

example

distinction of

between

how .

strongly

Although

weak the

this

here

, but

discussion

.

see

rather

is

and

composition an

important

Pustejovsky

strong

com

-

respects distinc

-

( 1994

) .

Notesto pages61- 67

245

Chapter 5

1.

As

an

operation

forming and

2.

on

a Wegner

Both

,

Levin

involving

separates

arguments

The

in

tempted

to

although

it

though of

the ,

Levin

the

and

phrase

Restrictions

type a

su

( i )

on

as

in

the

an

btyping

Mary

~

b

John

dyed

c .

Zac

For

the

the

incorporated

~

colored

verbs

~

dye

of

( 1990

( 1981

)

01

and

On

this

) ( e , x

formal

( i ) , AYAxAe

individual

and the

, in

have :::;

the

2 ,

to of

the

Baker

' s

operations

,

is

For

a

come

from

some

verbs

where

the

predicate the

as

both

( al

-

nature

hammer

and adjectival

the

complement the

is verb

the

adjunct

restrict

For

- like ) ,

the

the

.

process

such by

be ,

semantically

discussion

verbs

might

example

construction are

for

one

For

not

the

contributed

of

closer

semantic

.

7

.

obey

) .

that

there

is

,

dyes

a ,

shadow and

resultative

logically

restricted

:

.

These

seem

,

associated

respectively :

to

be in

white

~

the

event

paint

a

_ color

Pustejovsky

, the

( 1991b

, is

LFrom

between

representation

where

with .

positioned

discussed

Davidsonian

expression

,

relations

resultatives

purely

argument colors

following

color

( e , x

, y

added

the

verb

as

an

) .

build

argument

) ]

captures ) ,

02

the ( e , y

) ,

relational

( e )

event logic

/ \

01

( e , X

essentially

of

roughly of

is .

structure

corresponding

interpretations

the

verb

to

incorporation

arguments

chapter

is

the

emphatic

a

those

:

[ building

view

,

paints

green

from

.

color

we

following

[ build

,

,

-

.

green

and

re

.

of

directly

AYAXAe

functions

,

within

the

relation

Parsons

,

,

denote the

dye

and

specifically

would

dragon

view

_ color

resultatives

More

white

,

is

construction

argument

purple

1995b

into

type

shadow

semantics point

pure

,

to

,

-

further

many

of

chapter

the

and

notice

house

jeans

the

paint

selectional purple

a

the

his

)

fashion

to

~

in

predicate

But

tighter

relation

a . .

even

.

type

entering

state

this

verb

trans

Cardelli

appear

constructions in

verb

.

distinction

verb

athough

shadow

verbs

resulting

This

proposal

,

the

certain

verbs

from

the

theory

,

and

( 1995

differ of

well

all

make

.

discusses

' s

)

Pustejovsky

the

)

the

who

Wunderlich

briefly of

that

can

. as

discussed

as

( cf

arguments ) ,

into

of

terms

shadow

( 1988

default

Rappaport

semantics

predicate itself

in

appear

arguments

wipe

( i )

,

would

see

by

,

class

.

from

( 1987

classification

construction

homogeneous

( cf

of

the

of

directly

lexicon

seen

.

.

fairly

( 1984 arguments

directly

's

be types

) .

preposition

Baker

can

Marantz

internal

German

is

perspective

rethink

1987

and

Wunderlich

in than

generative

the

)

expression

in

translate

a

resultative

the

processes

from

a

coercion polymorphic

,

1995

indirect

from

with

operations

,

type

with

O - assignment

{ } - role

here

,

one

Benthem

versus

described

presented

available

van

( 1988

receiving

derivational

Viewed

A - calculus into

and

a

behavior

syntactic

to

Klein

involved of

analysis

5.

,

direct

operations type

a

Rappaport

assigned

within

4.

1985

finements

the

within language

and

arguments

3.

types

monomorphic

)

/ \

these

02

to

roles

as

the O - roles

in

Dowty

verb

through as

a

used ( 1989

by ,

set

Chomsky

1991

) .

( e , V ) ]

identified

by

predication

,

as

with

any

other

of

246

6 .

Notes to pages 68- 84

This

is

( 1989

the

view

) , and

espoused 7 .

here

Landman van

only of

9 .

in

a

verbs

allow

(i )

a

Mary

( ii )

~

Mary

tion

11 .

~ ~

( cf .

Jackendoff

,

lexical

or

differ

to

might

be

example

,

it

,

also and

which

strategy

for

dictionaries

.

For

the

( cf .

Pollard

of

variation

fully

alternation

in

more

fully

both

structures

define

overlap

define

in

" exhaustive

given

a set

planning

of

- Ginet

adverb

1987

a

( 1982

, such

part the

as

of

the

subject

read

in

of

events

.

events

for

) , some

ad

-

( i) :

event

in

( ii )

- oriented

,

, ,

a

above

interpreta

-

1994

I

chapter

broadly

a

in

10

I

am

( 1993

below

closest .

) . to

that

and

The

of

the

kinds

are

presented

of

HPSG aspects

addressed of

here

etc

listing

certain

treatment

of ,

structured

style

the

general

many

These

) .

activity use

dictionaries

ignoring ) .

X quote

the

with

a

It

For

particular

our

have

as

or

to

music

notation ,

( forthcoming is

, we ,

. .

Malcolm

some

that

along

construed

Sanfilippo

)

,

and

however of

reference

Hence

feature

Furthermore

Johnston ( 1993

.

case

in

listing

how

world

passage

however

dictionaries

of

the

,

written

make

a

of

maintain ( the

8 , of

visual

adopt ) .

discussed

Sanfilippo

also

reflection distinctions

particular

also

chapter

a

ontology to

a

are

more

our

dictionary

compilation ,

even

in difficult

structure

i .e . ,

only

informative

( e . g . , for

in

the

course are

objects

novel

argue

to

is .

of

these

must

I

and

in

to

dictionaries

exposition

as

in

's

.

gives

are

to

lists

,

the

are ,

dictionaries

Pustejovsky

given this

.

Lexikon some sort

Sag

the

McConnell

reference

of

to

more

term

of and

typology

- scope

AGENTIVE

novels

purpose

are on

and

distinctions

that

the

word of

syntactic

more

discuss

argued

e .g . ,

The German of information

that wide

refer

constructing ,

)

on

possible

refers

orderings

event

classification

) , and

dictionary

new

differences

these

therefore

,

.

Nevertheless

certainly

be

to

differs

structure

Benthem

us

.

) system

event

to

allows

events

) .

.

characterizes

term

12

1972

that

( 1979

of authors

.

while

sortal

above

might

way

the

is

this

collocational

argued

mentioned

interface

observes

semantically

contribute

's 's

suggestion

,

) , Descles

.

~

)

reading

these

van

restrictions

interpretation

! ~

( 1985

's

by

Kamp Kamp

, while

possible the

( 1985

departed

manner

These

the

explore

of

- scope

departed

In

Landman

Higginbotham

the

they

It

richness

wide

Higginbotham gives

)

language

in

how

.

, furthermore

what

( 1993

the

out

shared

of

( 1978

decomposition

.

natural

of

pointed

be

relations

This

) , Mourelatos

structural

events

follow

.

) illustrates

Pulman

Discussion As

well

for

primitive We

of

discussion

logic

are

( 1976

view necessarily

useful )

.

as

, 1984

and

discourse

10 .

"

) , Comrie

The

not

inclusion

of

( 1983

Crouch

a

precedence

part

Allen

) .

( 1983

overlap

temporal

overlap

( 1973

( 1990 , would

's

and

contain

Dahl

) provides

Benthem

precedence

terms

by

, however

( 1991

from

8 .

taken

Guentcheva

,

verbal and

we

.

Notesto pages85- 87

Chapter

247

6

1. Much of the perspective taken here has been influenced directly or indirectly by readings of Aristotle 's Organon and subsequent commentaries on it . 2. Hobbs' (1987) "systems," within the model of commonsense reasoning he assumes, are good approximations for the type of understanding needed to categorize and name things in the world . Nirenburg 's (1989a,1989b) notion of "mi crotheory " is a rich model of what language maps into after superficial linguistic interpretation has occurred . Miller 's ideas regarding the structure of the mental lexicon are also interesting in this respect, since the qualia can be viewed as "distinguished " links between lexical concepts in the lexicon . 3. As in Moravcsik 's (1990) treatment of Aristotle 's aitiae - - the modes of explanation mentioned above - the goal is to explore a sound cognitive view on semantics, largely based on the constructivist assumptions in Aristotle 's metaphysics. Qualia structure , as interpreted in a generative lexicon , is an integrated part of a larger methodological shift from conventional approaches to lexical semantics. The qualia provide the foundation for an inherently polymorphic approach to semantics. But they are only one tool for the characterization of natural language as appropriately polymorphic . What gives the qualia their true generative power is an explicit statement of how they are typed , i.e., what the expression in the quale role actually refers, as well as reference to explicit mechanisms of enriched composition ) such as coercion, selective binding , and co-composition . to be discussed in chapter 7. Hence, as presented in the literature , there is some divergence between our use of qualia structure and Moravcsik 's use of aitiae . GL makes explicit just what the generative devices are that allow qualia to provide different ways of explaining what an object is. What these approaches share is the goal of looking beyond conventional truth -theoretic models of semantics to a model of meaning which admits of a multiplicity of reference in linguistics expressions. For further exposition of this position , see Pustejovsky (forthcoming ) . There is possibly some similarity between aspects of qualia theory and the ontological theory of moments put forth by Smith and Mulligan (1983) . They argue for an ontology consisting of "dependent parts " of objects , which are not prop erly characterizable in standard mereological terms . Many of these "parts " are characteristic properties of an object which might be viewed as qualia associated with it . Another theoretical similarity exists between GL 's use of qualia structure and the notion of "lexical function " as employed by Mel ' cuk and his colleagues (cf. Mel 'cuk (1973a,1988a) , within the Meaning-Text Model . Heylen (1995) , in fact , compares qualia structure and lexical functions as descriptive mechanisms for accounting for semantic relatedness in potential , as well as actual , collocational usage. Finally , some of the proposals given in Bes and Lecomte (1995) are much in line with the methodology here, concerning what level of lexical knowledge is both appropriate and necessary for linguistic theory . 4. It might be argued that the subject always stands in a containment relation to the substance denoted by the complement NP. That is, yeast is contained in beer, tomatoes are contained in the pasta sauce, and so on. There are some interesting counter -examples, however, to this generalization , as (i) illustrates . (i) Methylene Chloride is used in decaffeinated coff~~. This example suggests that the construction is actually making reference to some process involved in bringing about the substance, as opposed to the substance itself , since the chemical in subject position is not present in the resulting liquid .

248

Notes to pages 88- 99

5.Another The notion of defeasibility here isthat asused insentences Lascarides et ale (1994 ). way of viewing the distinction between these is that the space of interpretations available torestricted the NPs New York and Boston is that available tothe more sortally airport proper names inlarger (4a ).than 6.qualia InPustejovsky and Anick (1988 ),the senses are "split "role according toits separate assignments . Under this analysis , the FORMAL takes as value the Figure of a physical object , while the CaNST role assumes the Invert -Figure value ofan aperture .The polysemy was seen as resulting from the foregroundin or backgrounding of a nominal ' s qualia . That is , in (9a ) paint applies tothe FORMAL role of the door , while in (9b ), through will apply to the CaNST interpretation ofthe same NP .The problem ,however ,to with sense splitting inside the qualia structure for a lexical item is that it fails capture the selectional behavior ofthe item for allcontexts . The interpretation of complex such as these given (1994 )overcomes these difficulties ,types as we see below infirst chapter 8.inPustejovsky 7.regularly Pustejovsky and Boguraev (1994 )show how these logical polysemies are infact encoded indictionary definitions for these words . 8.See LeiB (1991 )and Gunter (1992 )for anexplanation offormal mechanisms of type inference within the A -calculus , and Copestake (1993 ), Morrill (1994 ), Buitelaar and Mineur (1994 ) for its application to lexical representation . Shieber (1992 )discusses general issues oftype inference for natural language processing . It should be pointed out that the details of the typing assumptions discussed in those works differ in ways that are beyond the scope of the present discussion . For further details of the typing rules and the use of typed feature structures within generative lexicons ,see Pustejovsky and Johnston (forthcoming ). 9.munication Following a),suggestion by Manfred Krifka and Nicholas Asher (personal com the specific relation that structures the types can be built into the type structure itself .This isessentially the analysis pursued below in8.3. 10 . For expository purposes ,Iignore the other qualia roles until the discussion below . 11 . AGENTIVE Inprevious treatments of the semantics for dotted types ,Iassociated aunique and TELIC role value for each argument in the argument structure (cf.Pustejovsky (1994 ),Viegas and Bouillon (1994 )).for This seemed appropriate since the information for both modes of presentation an object must be ac counted for . For example , on this view , for a two -argument dotted type nominal such asbook ,there is amode ofexplanation for both the coming about of the information and the coming about of the physical object as well . That is , the qualia structure was given as follows : book =yx:phys _ob ]~iOn ARGSTR =[ARG2 ARGI :infOrma ]container information -phys _ obj -lcp QUALIA = TELIC FORM = hold (Ytx ) ==read (P(T ,w,v ,x),x), puhlish AGENT write (T,z:printer ,y)

The problem with this interpretation isthat the predicates that are sui generis to entities that are both physical objects and information bearing objects are actually mistyped here ; that is , both read and write are only understandable in terms of a complex typing system , such as that outlined in the previous chapter . The objects selected for byapredicate such as read and the predicates made

Notesto pages100- 108

249

reference to by objects such as books are the characteristic functions for those entities . 12. Aristotle Metaphysics , Physics II , and Moravcsik (1975). 13. These strategies replace the mapping rules discussed in Pustejovsky (1991b) , where it was less clear what type of semantic representation the mapping to syntax actually allows. These were given as follows : (A ) The semantic participant involved in a predicate opposition is mapped onto the "internal " argument position of the lexical structure . All transitions involve a predicate opposition of some sort . Independent of this principle , Pustejovsky (1988) and Grimshaw (1990) argue that the notions of causer and agent are associated with the initial subevent of an event structure . (B ) The agentive participant in the initial subevent or event structure is mapped onto the external argument position of the lexical structure . The result of both principles A and B applying in the structures above is not general enough, however, to account for the mapping to passive constructions as well as to experiencer constructions (cf. chapter 9 for discussion) .

14. This proposal should be compared to the lexical mapping theory , as outlined in Bresnan and Kanerva (1989) and Bresnan and Moshi (1990) , where semantc arguments are related to their morphosyntactic expressions by means of specific syntactic functions . Along with these principles , they assume a universal hierarchy of thematic roles, as given below: (i) ag > ben > goal > inst > patient > locative Cf . also Alsina and Mchombo (1990) for further discussion. Chapter

7

1. Briefly, we can summarizesomeof the type shifting operationsthat have been suggestedin the literature (cf. Partee (1985), Chierchia(1984), Hendriks (1987), Dowty (1988). (

i

)

Type

a

Shifting

.

b

.

c

.

d

g

2.

3 .

4 .

:

.

iota

pred

.

=

lower

:

:

general

> "

j

P

x

AX

~

P

=

~

lifting

)

]

lift

( j

)

=

[ P

n

u

( j

[ X

LP

>

:

[ P

(

~

P

:

:

>

lower

:

nom

.

functions

j

ident

.

e

f

lift

(

j

X

)

)

=

j

]

]

P

x

:

a

~

AX

<

a

, t

>

[ X

(

a

)

]

There is a long controversyover how to best analyzethe infinitival in such constructions; namely, as either a VP of some sort or as a full sentencewith a PRO-subject. In order to focuson issueof the coercionphenomenathemselves , I will not discussthis issuehere. Cf. Chomsky, 1981and Pollard and Sag, 1994. As we seein 7.4 below, this is in agreementwith the generalstrategy outlined in Chomsky (1986) moving towards constraints making referenceto semantic selectionrather than syntactic configurationsdirectly. Seebelow for discussion . For details seeKlein and Sag (1985) and Gazdar et ale (1985).

250

Notesto pages109-110

5. One recent proposalwhich would disagreewith this conclusionis Chierchiaand Turner (1989), wherepropertiesare allowedas arguments. 6. If we return to the type selectionof considerin (3) above, we can seeanother way of solving this type clash; namely, to allow the NP to shift in denotation accordingto the referencetype ladder. Followingthe proposalsjust discussed , however, there is another possibility; to shift the type of the verb itself, so that it would accept an NP as type e,t >,t >, without the NP shifting at all. Assumethat this could be accomplishedby a type-shifting operator, f sc, (for small clause), which would act to relate the two structures in (ia) and (ib). (i) a. John considers Mary to be an honest person. b. John considers Mary an honest person. Thus , f sc exhibits the shifting shown in (ii ) below. (ii ) a. consider E: b . fsc (consider) E We

then

the

(

iii

)

The

with

[ fsc

of

(

iv

)

(

(

)

the

(

meaning

of

(

Pl

(

ib

)

)

(

P2

is

)

(

P3

)

second

f - ?

shown

assuming

in

a

right

-

person

postulate

the

(

(

-

(

iv

in

phrase

API

)

AXPI

{

below

wrap

,

(

iii

)

,

guaranteeing

.

AY

[ X

along

=

y

with

operation

]

}

)

the

of

(

P2

)

(

P3

)

]

correspond

some

sort

,

-

cf

.

Bach

.

a

.

John

b

.

/

c

.

d

.

e

f

)

sentence

translation

)

operation

interpretation

VPIVP2VP3D

semantic 1977

this

predicative

derivation

ing (

associate

correct

.

.

sc

considers

(

John

consider

'

consider

'

consider

'

consider

consider

(

(

'

'

(

(

)

(

an

an

-

APAxP

{

Ax3z

3z

honest

honest

person

honest

AY

[ X

[ honest

[ honest

=

Y

]

}

(

. person

'

'

AQ3x

'

. person

. person

(

m

(

)

.

)

z

)

)

m

)

( j

)

[ honest

( z

( j

(

/ \

)

/ \

[ m

. person

Ay

[ x

=

zJ

=

]

)

y

( j

]

(

z

)

]

)

'

(

(

m

x

)

)

/ \

( j

Q

(

x

)

J )

(

m

)

( j

)

)

)

)

As the structure abovemakesclear, this rule allowsthe secondinternal argument to be treated predicativelyof the first. This solution is more in keepingwith the proposalin Klein and Sag (1985), sincethe verb is the elementshifting its type. 7. As discussedin Pustejovsky(1993), more in line with the proposalsabove, we can imaginetwo alternative waysof capturing the relatednessbetweenthe above sentencesinvolving want: either the verb shifts its type due to a lexical specification, or the NP shifts with referenceto a type ladder. Adopting a Klein and Sag-style analysis, we can proposean operator fhd (for have-deletion) that would relate the two usesof want in (1). (i) a. Mary wants to haveJohn. b. Mary wants John. It operatesin the following way. Given that the usagein (la ) is already derived from applying IE to the underlyingtype for want, i.e., <S,
Notesto pages 111 - 112

251

We would need to propose a meaning postulate

to generate these equivalences :

(iii ) ~ Pl . . . PnO [! hd ( ) (Pl ) . . . (Pn ) +-+ ( have ' (Pl ) ) . ' . (Pn )] This would allow for the derivation (iv ) a . b. c. d.

in (iv ) .

Mary wants 1.Q!!!! . ! hd (! E (want ' ) ) (Aj) (m ) ~ ! E (want ' ) (Ahave' (j ) ) (m ) ~ m { Ax (want ' (Ahave' (j ) (x * ) ) (x * )] }

This solution accounts for one interpretation of want , but as Dowty ( 1979 ) and ( 1985 ) point out , there might be many possible such operators , depending on the denotation of the complement , as we see later in this chapter . 8 . I assume that the basic rule of function

application

is given simply

as:

(i ) If a is of type a, and {3 is of type , then (3(a ) is of type b . It should be pointed out that the revised rule of application , FAC , is equivalent to the standard formulation in (i ) just in case Eo: is null ; that is , when the grammar provides no type shifting operators that can apply to the argument . 9 . One might consider a mixed solution , making use of type shifting over the verb and type coercion on the complement as well . Briefly , this involves shifting the type , essentially embedding the NP denotation within an empty predicate (cf . Dowty , ( 1979 ) . Let us simply call this operation p and characterize it in (i ) . (i ) p :

e,t >,t >

The meaning (ii ) :

=>-

postulate

<e , t > associated with this type shifting

rule is given below in

(ii ) VPD [pCP ) +-)0 Ax3P [P (P ) (x )]] The result of applying this operator to an NP is effectively to create a metonymic extension of the NP meaning . In the case of the NP John , for example , the operator produces the following : (iii ) p (J\ P [P (j )] ) => J\x3Q [Q (J\P [P (j )] ) (x )] This operator , together with the independently able to generate an " underspecified " translation

motivated Equi operator , fE , is of (iv ) , as shown in (v ) below .

(iv) MarywantsJohn. (v) a. fE (want')("'p(j ))(m) ~ b. fE (want')("'AX3P[P(j )(x)])(m) => c. m{Ax[want'("'Ax3P[P(j )(x)](x*))(x*)]} => d. m{Ax[want'("'3P[P(j )(x*)])(x*)]} ~ e. want'("'3P[P(j )(m)])(m) The expression in (ve) leaves the relation between Mary and John underspecified . This actually happens to be an advantage rather than a problem , since lexical information from the particular elements in composition with the verb may act to specify this relation in unique ways.

252

10

.

Notesto pages113-122

Pollard

and

be

Sag

( 1994

considered

In

a

particular

below

)

)

( i )

from

subject

was

John

promised

and be

order

to

in

Sag

show

GL

,

at

but

.

In

an

general

interesting

we

have

eliminates .

subject

of

( i )

a

the

denied

his

He

denied

having

similar

.

I

began

to

b

.

I

began

the

calls

the

need

of

)

a

.

b

I .

1

to

began

of

In

chapter

In

particular

the

type

.

.

aspectual

For

this

I

discuss

,

I

the not

the

, ,

complement

sense

for

many

" zeroing

as

is

studied

promise

cases

.

of

transformations a

cause in

been

type

treats

specific

" to controller

on

what surface

transformation

long

as

the

responsible

"

meaning

for

the

that

can

zeroing

be

of

the

:

( iia

)

for

" appropriateness

,"

Harris

.

.

variable

the

order

X

zeroing

to

apt

and

admits

transformation

prevent

Xap

that

are

from

applying

the

unclear

in

and

( iiia

) ,

for

rule

, t

be >

, t

> .

of

type

apply given

as

( 1989

to

)

to follows

lexical

generalized ,

where

Verkuyl

'

,

'

the

language

from is

NP

Harris

basic

an

and

-

the

insights

can

.

through and

the

the

type

newspaper

issue

( 1993

)

types

this

section

lattice

.

formally

in

,

quantification at

in

some

length

.

we ,

of discuss

will

not

since

concern

we

are

ourselves

here

interested

.

quantifiers S

,

novel

that

.

semantics

alone

Notice

appropriateness

complernent

inheritance

complement

in

behavior

Hence

ignoring

and

verbal

verbal

the

for

from

been

examples the

.

.

of

framework

structures

have

problem

question with

differs

qualia

I

Kritka

of

this

the

associated

related book

,

,

lexical

issues

their

subsequent

revisits

representation

richer

how

exposition

semantic

should

now

this

some

aspects

never lexicon

values

a

.

.

work

qualia

and

all

in

The

later

consequences

and

book

generative

examine

of

further

e

)

into

shift

.

within

position

the

can

in

the

over

but

hierarchy

primarily

15

the

,

purposes

with

a

coercion

8

complement

14

coerced

have

have

conditions

licenses

book

' of

by of

buy

Harris

view

preserved

For

of

context

,

the

directly

operation

.

book

it

study

,

point

swered

13

is to

.

which

began *

Unfortunately

.

to

( 1965

below

from

appropriatedness

further

)

.

semantic

Harris

different

the

)

.

derived

book

( ii

.

party

data

be

example )

.

in

.

under be

These

sequence

( ib

shift

.

( iii

12

but

can

for

slept slept

read

under

example

be

)

a

Harris

the

)

( iib

a

( ii rule

introduces

in

having

manner that

conditions

in

NP

He

)

,

in

lexical

terms

Harris of

complement

b

.

in )

transformation

a .

suggests

( ii

coercions

members

This

,

party the

underlying

article

properly here

controller

the

attend a

one

the

attend

.

of

particular

secondary

preserved

In

In

attend

would

virtue

type .

to

employ

methodology

in

to

allowed

relevant

considered

to

interpretation

by

and

forms

be

can examined

.

allowed

They

correct

coerced

syntactic

to

that being

involved

allowed be

coercion

phenomenon

- control

be to

."

complement

"

object

to

how

the

the

systematically

11

to

attend

of general

" coercion

Sandy

to

arrive

the

promised

allowed

type

more

- control

Sandy

to

a the

discuss

b

Pollard x

discuss of

they

a . .

)

subset

abbreviates

as the

well

. type

The t

type ,

and

derivation NP

'

the

type

253

Notes to pages 123127

believe : s + (NP + s) (i)

believe(p 2[NP,

John:NP p2[NP,S]:NP S p2[NP,SJ(John):S

S](John)) : NP

16. Insomesense,thiscanbeseenassimilarto thegoalsofthelexicalsubordination

operation in LevinandRapoport(1988).In fact,co-composition provides an elegant solution tothepolysemy involving resultative verbssuchashammer and wipe,aswellaswithverbssuchas rattleinrattledownthehill(cf.Jackendoff, 1990 and Levin, 1993). See chapter 10 for discussion.

17.Asweshallseein chapter10,lightverbspecification characterizes the wayin whicha specificsensein contextof certainverbsis determined by the complement.In manyofthesecases,the verbactsonlyas a generalfunction over qualia-based information fromthecomplement; e.g.,theverbsopen,close,break,

andfix.Thisisformally distinct fromthetypesofconstructions discussed inDi Sciullo and Rosen (1990) and related work, however.

Two other casesdiscussedin Pustejovsky(forthcoming)includemanner co-

composition andfeaturetranscription. Theformerdetermines howanargument to the verbmayitselfspecifythe mannerin whichthat objectis actedupon. Thereare twobasicsubtypesconsidered: (a) wherethe complement specifies manner, e.g.,theverbstryandsample; and(b)wherethesubjectspecifies manner,e.g.,theverbsbreakandothercausatives, aswellaspsychverbs. Featuretranscriptioninvolvesthoseconstructions wherea modifierto an ex-

pression contributes information to further specify a semantic feature ofthat expression that wasleftunspecified. Forexample, adjectives suchas pregnant, whenapplied to termsnotsensitive togender, suchasprofessor, willspecify the expression forthatfeature.Weinreich (1972), in hiscritique oftheKatzand Fodormodelof semanticdescription,calledsuchconstructionsfeature transfers.

Whatis interesting abouttheseoperations is that, althoughtheyare similar to coercions in somerespect,theyarenot typechanging operations; rather,

theymake useofqualia-based information tofurther specify theverbssensein context.

18.Thistypeofmodification mightat firstseemsimilar to theanaphoric island effectsdiscussedin Postal (1971),illustratedin (i) below.

(i)*Johnis anorphanandhemisses them verymuch.

Obviously, in (i)theindexed pronoun isunable to refertotheputative implicit reference to (Johns)parents,andthesentence is ungrammatical underthisinterpretation. Theanaphoric binding in (i)is qualitatively different fromthe selective binding operations above,however. Notice that an adjective suchas fastorgood isabletoidentify partofthequalia structure oftheexpression it is incomposition with.Making anexpression available formodification isnotthe sameas elevating it to a positionof antecedenthood forsubsequent anaphora, asin(i)above. Wecanthinkofthequalia withtheirvalues asa moreexplicit statement ofthetypeofthe nounitself.In thissense,selective bindingallows anadjective to modify a typefragment oftheexpression. Suchaninterpretation would prevent theprojection ofimplicit features expressing aspects oftheFORMALroleforthenounorphan.Notice that,although anaphoric islandeffects are preserved, eventpredicate selection isinfactpossible withorphan, asillustrated in (ii):

(ii) a recent orphan

254

Notes to pages 128134

Thissupportstheviewthat theseareverydifferent phenomena. 19.It mightbeargued thatthemodification byadjectives suchasgood asillustrated

in (67)above is notlicensed bythetypewithin thequalia, butratherbythe qualia themselves. Thatis,therulemightbetterbestatedintermsofqualia

selection:

(i) SELECTIVE BINDING (byquale):

Ifa isq,of type , /3is typeb,andthequalia structure of3,,QS quale, then cx/3 is of type b,of where = /3 fl a(q, ).

has

Although it mayappearthatspecific qualiaareselected for,in general, such statements canberecastintermsoftypes.Iftheformulation ofselective binding in (66)is correct, thenwewould predictthat anyqualesatisfying thetype required bytheadjective wouldbe available forselective interpretation. This

doesseemto be the case,as illustratedin (ii) below. (ii) a goodknife:a knifethat is well-made; (ii) Thats a goodknife,but it doesnt cut verywell.

Ifsuchdistinctions aregrammatical, which theyappear tobe,thentheadjective

is ableto selectthe AGENTIVE qualeofthe noun,supporting the viewthat a type rather than a quale is selected.

20. Asstatedabove, strictlyspeaking themodification isintermsoftypesatisfaction andnota particularqualerole.Hence,eitherTELICor AGENTIVE wouldbe a

possible target of the modification.

21.Onecanimagine someuseofhouse ina sublanguage where thingsaredifferent, forexample, Victorian orEdwardian England, wherea family mighthaveany number ofhouses.Blandings Castle, forexample mightbe an oldhouseof mine,stillin mypossession, whileI mighttypicallylivein a newerhouse.The wordhouse wouldnt havechanged significantly inthissubculture, butenough to reconceptualize it as an objectwitha differentintension.Theissueof how muchofthissemantic shiftiscultural andwhatislexical isbeyond thescope of thepresentinvestigation. Mymethodological assumption throughout hasbeen toattribute specific grammatical effects topredominantly linguistic distinctions, whichbytheirverynature,revealtheparticular conceptualizations oftheworld constructed by a word or phrase.

22. Forsomestates,beingmodified byan adjective suchas oldwillreferto relations

that arepossibly extensionally equivalent in duration to theinterpretation of interpretation forthisreason (i.e.,thereisnoHeisoldasmybrotherreading). The relationshipof brotherhoodor sisterhooddoesnot allowthis. theobjectitself.Forexample, an oldbrother soundsparticularly oddunderthis

23. Kiparsky andKiparsky (1971)attemptto relatethe syntactic behavior ofcomplementsto the semanticsof the predicate,the complement, and the relation

between them.Grimshaw (1979) distinguishes thebehavior ofsyntactic cornplementation in English in termsofsemantic selection overthreecomplement

types,propositions, interrogatives, andexclamations (cf. Bresnan, 1972).As beenpursuedby functionalists, andis centralto the goalsof RoleandRefer-

VanValinandWilkins(1993)pointout,the traditionofsemanticselection has

enceGrammar (RRG) theorists (cf.VanValin, 1993 andthepapers therein).

Notesto pages137-145

255

Wierzbicka (1980,1982,1988) develops an increasingly rich system of semantic selection , with many interesting insights regarding the role of semantic categories for syntactic form . Yet the goal of semantic theory in this respect should be to capture the most elegant generalizations concerning syntactic expressiveness from semantic selection, and not to completely recreate the richness of the syntactic descriptive system within semantics. The observations by Grimshaw (1982) and Jacobson (1992) concerning the limitations of semantic selection are worth noting , and there is no reason to think that syntax is completely parasitic on semantic types .

24. That something besides case assignment is involved can be seen from sentences such as the following : (i) John asked me about my name. (ii ) John asked me my name. In sentence (i) , the preposition is not just case marking the complement but shifting the interpretation of the predicate as well .

25. Groenendijk and Stokhof (1989) discuss the scope distinctions allowed by these verbs in sentences like (i ) and (ii ) below. (i) Peter knows who John loves or who Mary loves. (ii ) Peter ~ o!!ders who John loves or who Mary loves. The sentence in (ii ) is ambiguous where the disjunction may take wide or narrow scope

relative

to wonder

.

Chapter 8 1.

In Pustejovsky (1991) and Pustejovsky and Boguraev (1993) , a distinction is made between fixed and projective inheritance . Following Touretzky (1986) , the fixed inheritance are concepts

structure

in our

model

of a lexical item is defined as follows , where Q and P of lexical

network

. Then

:

(i) DEFINITION : A sequence< Ql , PI , " " Pn > is an inheritance path, which can be read as the conjunction of ordered pairs { < x I , Yi > 11~ i ~ n } . Furthermore

, from

this the set of concepts

that

lie on an inheritance

path

is

defined , as distinguished by a particular quale role (e.g., TELlO vs. FORMAL ) . This is called the conclusion

space for a given quale , q'

(ii ) DEFINITION : The conclusion space of a set of sequences, q, is the set of all pairs < Q , P > such that a sequence < Q , . . . , P > appears in q , where q is one of the qualia for the concept Q .

(iii ) DEFINITION

: The complete conclusion space is the set of all conclusion

spaces defined for each quale for a concept : = q'i ' Finally , by adopting

Touretzky

's operator

I nh - where , for every set of sequences

S, Inh (S) denotes the set of values inheritable from S- the lattice structures shown

above

for

book can be differentiated

as follows

:

256

Notes to pages 146- 161

Let [a ] cfI stand for the denotation of Q: with respect to a model of inheritance over the set of sequences, 4>. Then ,

[book]
/\ Formal (x ) = Inh (physobj ' ) /\ Telic (x ) = Inh (literature ' ) 1\ Agentive (x ) = Inh (literature ' ) ]

By viewing the different facets of meaning of a lexical item (i .e. its qualia ) as inheriting from orthogonally typed lattice structures , this approach hoped to avoid the problems associated with multiple inheritance , although this has not been fully explored within this system . Briscoe et al . ( 1993 ) address the issues relating to multiple inheritance in a more comprehensive manner ; cf . also Klein and van Benthem , 1987 , Evans and Gazdar , 1990 , and Russell et ai ., 1993 . 2 . Details on how nouns with " horizontal relations " such as brother , neighbor , and sister piffer formally from those denoting "hierarchical relations ," such as father , daughter , and son , are discussed in Pustejovsky ( 1995b ) . 3 . Besides the two types of dot objects discussed above , book and newspaper , there is at least one other type we will consider , that involved in defining nouns such as sonata , discussed in 8 .5 . What distinguishes these nominals is the way in which the qualia structure relate the dot elements . For a noun such as book. I the FORMAL role refers only to the two elements , hold (x , y ) , and both AGEN TIVE and TELlO qualia are defined in terms of the dot object , x .y , itself . For newspaper , the dot elements are "split " between qualia values , where the pro ducer sense functions separately from the product sense in the qualia . No Quale role makes reference to the dot object , however , which presumably restricts the noun 's denotational possibilities ; that is , newspaper can never refer simultane ously to the organization and the product on the same predicative level . Finally , the dot object for a noun such as sonata is constructed with the specific relation of " performing ," as is discussed below . In this case, the relation is not saturated by the dot elements as arguments , but makes reference to other arguments as well . Cf . Pustejovsky and Johnston (forthcoming ) for discussion of the formal difference in dot object qualia structures . 4 . The case of (59c ) above is similar to the issue of anaphoric islandhood mentioned in Postal ( 1971 ) , where an adjective seems to be filling the role of a true argument to the noun . Cf . footnote 17 for chapter 7 . 5 . There are some interesting complications , however , suggesting that simple control between the subject and the infinitival phrase is not mediated by constraints on agentive controllability .

(i) OUf house !!~~

a new roof .

(ii ) The soup !!~~.9. more salt . (iii ) The carpet ngg~ vacuuming . The interesting thing about such examples, however, is that we predicate need of only those things that we can or want to affect or act on in some way. Thus , I might say of a lawn , that it needs mowing , but I would not say of a tree , that it needs cutting down , unless there is some reason (i .e., TELlO state ) that precipitates this comment , e.g., it is rotting , it is blocking the view to the city , and so on. Similarly , a house needs a new roof only because of how it impacts

Notesto pages167-169

the integrity

257

of the house , while the carpet needs to be vacuumed

because it is

dirty , and someone (the speaker) wants it clean, and so on. This is not to say that need entails a decomposition or reconstruction into the appropriately modal statement capturing the above intuitions . It does mean , however , that when we predicate something of need, the conditions on licensing this predication must include the ability for an agent to act to bring about the change of state implicitly referred to by the complement NP or gerundive .

6. The

verb

burn has both

causative

and unaccusative

forms , according

to the

analysis in chapter 9 . The representation will change slightly to reflect this polysemy , but not in any significant way for our discussion here . The expression of the underlying agentive argument in the causative readings of such ing - nominals

is optionally defaulted , while the object is not , as the example in (ia) suggests. (i) a. The Allied Forces' burning * (of Dresden) was viewed with horror . b . We witnessed the arson's burning * (of the house). 7 .

Grimshaw (1990) discusses many of these same issues from the point of view of syntactic expressibility . Grimshaw 's goal is to account for the classic problem of complement optionality in NPs in a systematic way . For example , it is well known that verbal argument patterns appear quite systematic compared to the apparent optionality of arguments within the nominal system (cf . (i ) and (ii ) ) :

( i) a. * The doctor examined. b . The doctor 's examination (of the patient ) was successful. (ii ) a. *They attempted . b. Their attempt (to reach the top ) was successful. As Grimshaw points out , however , complements in the NP system are not com pletely optional . She proposes that the argument structure of a lexical item is derived from two independent components of predicate meaning ; the aspectual

structure and thematic structure . Any item lacking an aspectual structure (e.g., the event structure of Pustejovsky (1991b)) will have no argument structure and will take such as analysis complex

only optional arguments . This is Grimshaw 's distinction between nouns exam and examination . Although I generally agree with Grimshaw 's of aspectual structure , the semantics of these nominals is much more and subtle .

To beginwith , the distinction Grimshawmakesin grammaticalbehaviorbetween examand examinationdoesnot completelyaccordwith the data. For example, although it is true that sentence(iiia) and (iiib ) are both grammatical, both (iva) and (ivb) are ungrammatical. (iii ) a. The examination was long . b . The exam was long. (iv ) a. *The examination of the patients was on the table . b. * The examination was on the table . As Grimshaw correctly points out , the noun examination is ambiguous between a process reading and a result reading . Namely, it can refer to the act of the examining , or to the specific results of the examination . Furthermore , observe that , contrary to Grimshaw 's proposal , examination does not require its full argument structure to be satisfied when the genitive agentive NP is present:

258

Notesto pages170-177

(

v

)

a

.

An

unskilled

long

b

.

Any long

is

by

time

no

room

to

evidence

chapter

extends

in

this

area

(

of

presented

that

to

that

systematic

instructor

consequences

objects

nicely

shows

here

the

cover

)

in

this

will

always

area

-

take

)

will

a

always

of

of

(

of

take

data

)

in

within

event

provides

sub

outlined

Italian

and

expression

types

from

forthcoming

nominalizations

argument

semantic

projection

Busa

linguistic

problem

syntactic

.

analysis

cross

the

projection

grammar

(

unskilled

the

dot

suggesting

furthermore

to

an

discuss

as

stantial

examination

.

nominalizations

the

' s

.

examination

a

There

instructor

time

English

in

a

in

NPs

generative

-

this

.

Busa

follows

from

lexical

approach

.

8. Nunes

(

range

(

1993

of

1986

)

)

presents

a

discussion

interpretations

also

presents

nominals

of

possible

some

how

in

the

the

semantics

of

nominalizations

interesting

the

verb

they

constraints

on

the

affects

form

.

the

Lebeaux

formation

of

derived

.

9. As

pointed

out

different

by

from

complex

as

That

is

this

is

the

.

For

symphony

,

,

music

is

TELIC

example

one

in

role

i

. ,

to

event

while

This

a

variable

perform

will

be

is

predicates

in

a

independent

important chapter

9

our

for

value

;

Bouillon

,

in

1995

values

listen

listen

is

in

a

,

the

is

by

and

.

dependent

coercion

aspectual

)

)

in

perform

through

coercion

(

event

is

projectable

object

recovering

that

more

experiencer

relational

the

also

a

event

the

and

very

in

an

(

of

are

defined

listener

Recovering

of

artifacts

is

role

the

without

being

social

conjunction

.

discussion

,

Pustejovsky

a

partial

,

for

In

coercion

with

of

.

listen

to

)

function

TELIC

role

as

and

1991

their

the

the

structure

perform

(

that

audience

due

binding

Miller

in

defining

an

qualia

reconstruction

similar

in

ignore

for

. e

,

,

cannot

the

,

as

artifacts

performed

reflected

metonymic

such

physical

manner

such

psychologists

simple

itself

.

experiencer

.

10. One

issue

the

that

we

propositional

types

that

along

their

,

,

observations

(

i )

ii

)

*

)

John

is

John

,

are

the

that

verb

we

,

,

would

be

expressed

and

cf

.

,

Descles

,

Even

csfs

verbs

differ

has

and

)

,

it

is

this

curious

-

think

respect

.

use

,

where

and

prominent

Guentcheva

,

heteroge

believe

in

a

of

7

behave

the

think

1989

(

nominals

thought

predicate

he

he

think

sortally

proposals

to

not

its

,

account

for

as

well

.

as

1990

conference

forms

Cattell

,

for

a

For

pro

-

related

,

1984

systematic

it

so

complex

predi

thought

,

which

-

are

.

of

on

into

some

propositions

propositions

syntactic

motivated

)

space

and

.

.

enter

give

:

issues

the

conference

and

following

,

to

the

nominalized

our

the

as

to

thinking

( cf

debates

go

go

structure

least

,

go

not

some

belief

at

phrasally

do

and

can

should

should

and

as

including

superstitions

move

(

that

believe

that

.

structure

chapter

forms

their

selection

stative

below

such

to

attitudes

)

a

sortal

in

:

constructions

believe

ii

thinking

unavailable

I

(

believing

is

Furthermore

cate

in

is

finer

begun

and

belief

believe

the

syntactic

verbs

complement

is

discussion

canonical

nominalizations

while

predicate

(

with

attitude

to

here

the

associated

respect

with

cess

address

Following

propositional

with

example

.

were

traditional

neously

adequately

domain

semantic

of

cannot

into

,

.

The

differences

semantic

beliefs

,

subtypes

suppositions

justification

,

for

in

how

distinctions

such

these

a

types

between

them

.

11. This

tween

raises

qualia

an

obvious

structures

question

.

The

concerning

relation

the

,

for

example

interdependencies

,

between

possible

bake

and

be

baker

-

is

Notesto pages178- 188

259

typically construed as that between an activity and the agentive nominal associated with that activity . Thus , the suffix -er is treated as a function over the relation and returns the individual associated with the agentive aspect of the relation (cf. Williams , 1981, Travis and Williams , 1983 and Moortgat , 1988) . The qualia structure for an agentive nominal would have roughly the following form , here illustrated with baker:

(i)

baker FORMAL= human (x ) TELlC = bake(e,x ,y :bread )

[

I]

While English - er functions both as the agentive nominalizer and the instru mental nominalizer , many languages use distinct suffixation for these roles. For example, Comrie and Thompson (1985) point out that in the Californian Indian language Wappo , the suffix - ( e) ma designates purpose as applied to a verb root , such that from the verb ~ (plow) , the instrument / tool ~ cema (plow) is derived . Returning to the example in English above, the instrumental nominal for baking is, of course, the noun oven. What oven and baker share is a more or less common TELIC Quale specification , where baker is as represented in (i ) and oven has the qualia structure in (ii ) .

oven FORMAL = container (x ) TELlO = bake (e,w ,y :bread )

(ii)

1\ loc (e,x )

This is quite unlike the TELIC values for tools such as knife and umbrella , encountered above. The locative relation in (ii ) , however, is a necessary feature for distinguishing the semantics of locative nominalizations from instrumental nominalizations (cf. Comrie and Thompson , 1985 for discussion) . Chapter

9

1. Support for the distinction made above with respect to argument coherence and causation comes from Abusch (1985) , who argues, independently , that accomplishments and causatives are not co-extensive, on the basis of data such as that in (i) , which show that causatives may involve processesa well . ( i) John galloped the horse for three hours.

This view is assumedin the statementof causationabove.

2. Matters may not be as straightforward as this, however. Alsina (1992) argues

3.

that , in somelanguages , the causativemorphemeis a three place predicate, involving the semanticrole of PATIENT as .well as CAUSERand CAUSEE. On this account, morphologicalcausativescannot be reducedto a syntactic derivation, such as that proposedby Incorporation Theory (cf. Baker, 1988). Ackerman (1993) proposesa proto-role accountof related data, and arguesthat the argument-linking accountin Alsina (1992) is unableto explain certain causative marking casesin Hungarian. Cf. also Saksena(1980) and Dowty (1991). A similar proposal is made in Pustejovsky (1988, 1991b), but a slightly different phenomenonis being examined, namely, how the "deep object" of an unaccusativesurfacesas a subject in English.

260

4.

6 .

7 .

8. 9.

( i) a. Mary is working on a book . b. Mary is beginning to work on a book .

Notes to pages 189- 205

261

Notes to pages 206- 209

See Pustejovsky Klein

and

( 1993 ) for

details

Sag ( 1985 ) for how

on the

binding

control

is achieved

relation

. We

follow

generally

in Equi - constructions

such

as

begin . In Pustejovsky and Bouillon ( 1995 ) , Dowty ' s ( 1979 ) analysis of finish is placed within a generative lexical framework , where this verb is analyzed as making ireference to two events : ( 1) that subevent which brings about , sine qua non , the culmination of the event as a whole ; and ( 2 ) an assertion that the entire event has occurred . As Dowty makes clear , this presupposes that the event has a natural division into t \VO subparts . This would seem to indicate that finish is an aspectualizer which type - shifts the complement event into an achievement ( i .e . , a right - headed transition ) . It does this , however , by preserving the integrity of the complement event , for notice how finish - sentences behave in many respects as both accomplishments and achievements : ( i ) a . Mary b . Mary

finished finished

building building

the house the house

in 3 months . at 3 :00 pm today

.

Hence , even lexical accomplishments ( left - headed transitions as achievements ( right - headed transitions ) when complements

) can be interpreted of finish . In order

to capture this intuition while still satisfying Dowty ' s fundamental interpretation of finish , we define a general relation of logical culmination , cut , between an event and one of its subevent :

( i ) Vel Ve2 [cul ( el ) = e2 +-+ -, 3e [e2 ::j el A e2 < e A e ::j ell ] we can

build

this

relation

directly

into

the

event

structure

itself , in which

case

it would be a relation on event trees . This would essentially be a logical culmi nation relation between events ; culcx : ( el , e2 ) ' Now we can express the semantics of finish as a right - headed transition , where the sub event standing in the cul minating relation with the larger event is seen as the AGENTIVE of the overall aspectual event . Furthermore , the FORMAL or result of the aspectual event is the assert ability of the entire transition , of which the AGENTIVE is a part .

(i) AxAe'!;3e13R3P [finish : ES= elculcx:e2 1\ FORMAL = [P (x ) (e2T)]] 1\ AGENTIVE = [R(el , x )]] 12. An interesting paradigm showing how control and raising interpretations differ for verbal and nominal forms is illustrated below in (i) and (ii ) . (i) a. When you go underwater , you stop breathing . (Control ) b . When you die , you stop breathing . (raising )

111_

-

(ii ) a. *When you go underwater , your breathing stops. (no control ) b. When you die , your breathing stops. (only raising ) The nominalization in (ii ) does not allow the intransitive control reading , thereby ruling out (iia ) . The only reading available in the intransitive form is the conventional raising interpretation .

13. These examples are related to the discussion of "Super Equi -NP Deletion ," begun by Grinder (1970) , where sentences such as (i) and (ii ) are discussed: (i) Rogeri thinks it is easy PROi to protect himselfi . (ii ) That washing herselfi in public was enjoyable surprised Louisei .

262

Notes to pages 210- 215

Within the approach taken here ,above the sentence be treated along a similar analysis tothat presented . in(ii)would (i) a. Johnlookedthroughthe glasscarefully . b. *Johnsawthroughthe glasscarefully . (ii) a. What Johndid wasto lookat Bill. b. *What Johndid wasto seeBill. The reconstructions possiblewith startleversusangerseemto pattern along similarlinesto the difference shownabove . 15.

Thereareof coursetwo readingsfor (frighten' type verbs: onewhichis purely causative (x occasions fearin y), andtheotherwhichis beingconsidered here. As Pesetsky(1987 ) notes, the backwards bindingeffectsintroducedbelowobtains only in the secondinterpretation . 16. Williams is somewhat differentfrom Grimshaw 's, but we will here. 17.

(i) a. Mary buttered her bread. b. Mary buttered her bread with an expensive butter from Wisconsin . c. *Mary buttered her bread with butter . (ii ) a. John and Mary danced. b. John and Mary danced a fast waltz . c. * John and Mary danced a dance.

There is another of compositional with ditransitive

type of shadowing which is not purely lexical , but is the result operations in the syntax ; namely , that discussed in chapter 2 verbs such as give and mail .

(iii ) a. John gave a talk to the academy today . b. John gave a talk today . c. * John gave a book today .

263

Notes to pages 216- 219

(iv ) a. Mary mailed a letter to me. b. Mary mailed a letter . c. * Mary mailed a book . Here it is the specific semantics of the complements talk and letter , respectively , which are obviating the obligatory expression of the indirect object . It is possible , therefore , to identify

the following

three types of lexical shadowing :

(A ) Argument Shadowing: Expression of an argument is shadowed by the verbal semantics (as in the cognate constructions ) or by the semantics of the phrase (as with the build - examples .

(B ) Complementary Shadowing: Expression of one argument shadows the expression of another in a complementary fashion (as in the risk -examples). (C) Co-compositional Shadowing: Expression of an argument is made optional by virtue of how the verb "co-composes" with its complement (as with the give a lecture cases) . For further discussion of these issues, cf. Pustejovsky (forthcoming ) . 18 . Privation is but one part of Aristotle 's original theory of oppositions . As pre sented in the Categories Ilb18 , there are four basic classes of opposition : (i ) Correlatives

(between two relatives ) : double vs . ~

(ii ) Contrariety: ~~~ vs. ~

.

.

(iii ) Privation (Privatives to positives) : :Qlindnessvs. ~ ~~ . (iv ) Contradiction (Affirmatives to negatives) : "He is sitting ." vs. "He is not sitting ." For the concept of privation , there is an inherent modality related to Aristotle 's notion of potentiality of a subject . That is , it is not sufficient to merely translate a privative predicate such as sick or blind as not healthy or unable to see, since the individual to which a privative applies must have the ability to be predicated of that term . Privation , therefore , correctly applies to a subject of which the positive should , by its nature , be able to apply . Thus , a dog cannot properly

said to be illiterate , because it is not possibly literate , by its very nature (cf. Horn , 1989 for discussion) . 19 . We ignore , for now , quantification below

over the individual

variables in the expressions

.

20. In Pustejovsky (forthcoming ) , this is explained in terms of a more general process of lexical shadowing

operating

in the grammar .

21. Saksena (1980) discusses this issue at some length . She makes the claim that an "affected" / "non-affected" distinction is more primary than that of AGENT and

PATIENT

in causative

constructions

.

Saksena

argues

that

the

concept

of

AGENT is not a single category, but itself has a dimension of affectedness (cf. Anderson , 1979, 1984) . This view emerges from observations in Hindi , where there are two ways of realizing

a causatives , corresponding

to affected and non -

affected readings, as in (?? and (91) above (cf. also Saksena, 1982) . As with the analysis above for Italian , it should be possible to distinguish

these two readings

in terms

, where

of the

event

structures

associated

with

the sentences

distance

in

a tree structure is correlated with non - affectedness on the arguments . A related statement of this proposal can be found in Pustejovsky ( 1988 ) .

264

Notes to pages 221- 224

Chapter

1.

10

There are many issues raised by a semantics with co-composition and under specified representations . The interesting thing about an operation such as co composition is that specific values for the qualia roles may emerge in composition

through qualia unification (cf. chapter 8) , where otherwise they were not lexically specified for a verb or noun . For example , in the compound book shelf , the TELlO role of the noun shelf is specialized by virtue of the typing restriction imposed by the noun book; that is , the TELlO for the compound is specialized

from "holding " to "holding books" (cf. Levi , 1989). Similarly , within the verbal system , complements can contribute information specifying TELlO values associ ated with the entire VP , although not necessarily associated with the head verb . Adjectives , too , can compound to further specify qualia -related information , such as the AGENTIVE role in the compound adjective American made . Johnston ,

Boguraev , and Pustejovsky (1995) discuss the semantics of compounds , while Pustejovsky ( 1995d ,forthcoming ) and Johnston ( 1995 ) discusses the semantics of underspecification more generally . A related issue concerning semantic specificity

2.

and markedness in morphological operations is discussed in Kiefer (1992) . Although I have not discussed it at any length in this work , another well -studied type of logical polysemy involves the pair of package and grind operations , which can themselves be categorized as coercion operations . Each of the sentences in

(i ) below involves a mass noun which has been packaged by a determiner or a combination of the predicate type and Aktionsarten of the sentence (cf. Verkuyl , 1993 , Dolling , 1992 , and Hendriks , 1987 ) .

(i) a. Mary loves the ~ ~ b . Mary c . John

drank

d . ~~ !!-4 drifted

Examples

John brought .

ate the cheese .

involving

the

beer .

into the tent for hours .

count nouns which

have been ground

by a term

explicitly

typed as applying to mass nouns (i .e., more) are given in (ii ) . These sentences involve

a particular

kind of " fruit -grinding

" which can be seen as a subtype

of

a more general grinding mechanism (cf. Link , 1983 and Bach, 1986) . In these cases, an countable

"food individual

" is typed shifted to " food substance :"

(ii ) a. Mary put apple in the yogurt b . We should

c . Adding

add

some

more

banana to your diet

~~ to this sideof the room. will lowerheartattackrisks.

As Copestake and Briscoe(1992 ) point out, however , the problemof "animal grinding" is moreinvolved , sincethe basictype of the nounis beingshifted, as wellasits countability . Consider the pairsin (iii) and(iv) below. (iii) a. Bill caughta haddo ~k this morning . b. We're havinghaddockfor dinnertoday. (iv ) a. Zac wants a pet !~!!!h for his birthday . b. Sophia loves !~mQ with gravy. Within an approach such as theirs , the semantics for an individual count noun such as haddock is typed underlyingly as a countable animal , as illustrated below :

265

Notes to pages 225- 226

_U

+] -

animal _grinding

.

.

I

J

Copestake and Briscoe ( 1992) argue for generative principles of sensederivation , and propose that a lexical rule such as the grinding operation below accounts for the "edible mass" sense of individual animals .

grinding 1 = RQS = [ animal EDIBLE = +] 0 = RQS = food-8ubstance

This derives the sense of "haddock meat" from the individual fish . Seen in a larger context , this lexical rule is a type coercion operation applying with very specific constraints . Copestake and Briscoe's arguments for generating the mass sense through the application of lexical rules are generally convincing , since the rule is necessary for creative grindings over other lexical items . Within the framework of a generative lexicon , however, another approach is also possible, one involving coercion and selection compositionally , without the appli cation of lexical rules. Polysemous nouns such as haddock could be represented as complex types , where the relation between the dot elements expresses what the lexical rule does for Copestake and Briscoe : Ax.y3R [haddock(x .y ) t\ R (x , y)]. Not surprisingly , the TELlO for a complex type such as haddock or lamb will make reference only to the type which itself has a TELlO value, in this case the type food , whose TELlO is specified as "eating :"

ARG 1 = x : ind _animal ARG2 = y : food - 8tuff

(38)

ind ~ nimal

.food _lcp

FORMAL = R (x ,y ) TELlO = eat (e,w ,y )

It is very possible that the language makes use of both devices, namely complex types such as dot objects as well as the application of lexical rules. Regardless, both types of devices must be seriously constrained by the grammar in order not to overgenerate unwanted forms (cf. Johnston , 1995 for evidence in favor of dot object representations ) . 3. As with the interpretation of qualia structures above) if an event or argument is not mentioned explicitly in the parameter list , then it is assumed that existential closure applies to the variables mentioned in the qualia . The feature structure given in (1) above will correspond to the A-expression given in (1) :

266

Notesto pages 225- 232

( i )

4

.

AXAe23e13y3P3R

The

[ 0; :

semantic

distinction

( forthcoming

) .

the

5

.

in

generative

In

Pustejovsky

of

this

in

analysis

,

can above ,

but

might

they

not

of

knows

that

knows

what

< s , t

>

Mary

's

the

purchase

knows

the

coercion

to and

awaits

Then

, t

.

of

these

phenomena

( el

, y

explored the

, x

) ] ]

in

Busa

semantics

of within

explanation

for

4 .

explains

It

is

which a

the

useful

typed

classes

the

can

an

well

of .

embedded

inter

and

reveal

presumably

- formed

on ,

as

also

constraints

be

in

argument

understand would

syntactic

to

-

interpretation that

for know

on

below

taking

assume

type verb

par lexical

available

as

-

think

generative

propositional

we

inter

to

noun

Under

simple

) ,

The

an

subcategorizations

remarks

apply

further

be

coercion

which

of

,

course

.

the of

car

the

.

car

.

.

in

can

car

.

a

completely

unconstrained

unexplained research

effects

a

buying

answer

leaves

syntactic

coerCIon

> .

bought

knows

how

study

."

a

bought

knows

problem

, t

, < e

it

be

( 1989

Mary

* John

for

above

larger

chapter

multiple

allowing

Mary

* John

,

as

would

sentences

d

John

the

further

the

c . .

is

[ R

.

John

overgenerates

is

Stokhof

some all

John

Thus

It

without

=

made

as in

extension

verbs

and as

b

.

invoked

,

for

these also

s

AGENTIVE

approaches

mentioned

coercion

while

be

a .

e .

ing

,

,

a

and

is ,

account is

typed

expect

are

( i )

be would

typed

coercion

of

would

Groenendijk

used

,

A

SLNs within

" metonymic

type

rogatives

type

terms

That

) ]

.

particular

.

, x

and

Italian

questions

coercion

Following

so

) ,

in this

interrogative

we

and

concealed

examples

an

English

( e2

embedded

framework

alternation

ticipate

these

is

( 1995a of

[P

ILNs

analysis

lexicon

pretation

=

between

Her

nominalizations

FORMAL

why ,

and

a

( ic

)

solution

contribute

to

and

fashion ( id

)

should

are prove

limiting

the

in

these

ungrammatical

cases .

helpful

in

application

This

determin of

-

semantic

.

perhaps

worth

shifting

,

in the

sentence .

each verb

pointed

mentioning out is

by embedded

another Hobbs

et within

al

interesting .

( 1987a an

case , 1987b

interpretation

) .

of

coercion

Notice predicted

how

, the by

involving

sort

complement the

typing

of

(ii) a. Thatchervetoegthe channeltunnel. (theproposal ) b. The boardvetoedan increase in payfor the teachers . (theproposal ) (iii) The organizers havebookedElla into Symphony Hall. (theconcert ) (iv) Hersecretaryhasscheduled Johnfor nextweek. (theappointment ) That is, in (ii) the complements are modallysubordinated within the type of proposal, whichis the type selectedby the verb veto. This producesa weak intensionalcontextfor the complement , sinceit is intensionally"wrapped " by the denotationof anothersort. A similarexplanation accounts for the apparent metonymies in (iii) and (iv)~where , in fact, strict typing is beingobeyed , and sort coercionis providingfor the correctinterpretation(cf. Pustejovsky (1995c ) for furtherdiscussion ). 6. Li (1994 ) discusses spatialprepositions andtheirinterpretation in German , within Lang's (1989 ) theoryof spatialrelationsand Bierwisch 's modelof conceptual structures(cf. alsoBierwischandLang, 1989 ).

267

N dtes to pages 233- 235

7.

One

lexical

ations is

knowledge

is that

of

entries

,

lexicon

words

is

.

The

classes

From

the

of

,

the

relation to assuming

for

a

example

,

generic

is

a .

.

,

{

animal

{

artifact

a

,

,

are

for

The

of divide

a

example

,

typically

attribute to

the

organized

,

,

verbs

relations

.

entails

He

which

( 1991

issue

The

my

( 1988

Lakoff

characterize

them

uniquely

,

a

different

The

the

of

the

it a

of

this

of

ascriptive

.

The

former

modifies

ascribes

class

principles

.

noun

heavy antonym

,

set

nonascriptive

.

value

For

of

adjective

,

the light

,

adjectives

.

which

do way

,

not

the

a

see

set

of

lexical

is

formal

)

snoring

semantics

lexical

( 1990a

musical

entailment He

in

relating

Miller

a

.

since

theories

device ,

predicate ,

musical

sleep

to

formal

WordNet

directly example

is

entails similar

not For

that

to

lexically is

.

one

according

snore

organization

has

more

not

.

( cf

.

Fass ( 1987

so

most

chapter 1988 ) ,

and

ignored

are

, 1993 Sowa

,

to of

no

in

this

patterns

easy

forms

7 ) , ,

been systematic

always

that

( cf

largely

on

assumes

) ,

:

items

and

.

Miller

For and

) .

are

inferencing

to

below

}

of

music

as

correspond listed

}

follows

Thus

structured

on

a

}

condition

unspecified

for

,

each

.

words

some

verb

information

attention

coercion

the

This

metaphor

phenomena

literature of

of

in

postulates

and

Fellbaum

focused

.

meaning

details

"

are ,

.

those

used

WordNet

sleeping

adopt

more

it

,

is

principles

to

example

is

weight

include

instrument

in For

i .e .

,

}

package .

element primes

}

flora

and

package

different

" pertain an

the

noun ,

very

but is

Finally

The

the

adjectives

noun

,

predicates

in

attribute

along

instrument

8.

to

same

Nonascriptive the

heavy

weight

et

is

Rather

}

ascriptive

attribute

adjective

of

refers

bipolar

:

topmost

are

phenomenon

attributes

WordN

classes

which .

actually

object

which

or

the

fields

{ natural

. .

,

semantic

{ natural

features

in

two

of

the

topical relations

relation

These

{ process

function

adjectives

into

the

by ' s

in into

hyperspaces

as

set .

{ state

concept

organization

involves

is

noun

interesting

languages

these

{ plant

}

distinguished

a

Adjectives

}

}

further

example

three

entailment

entity a

of

}

{ food

various

with

some

knowledge

,

function

leaving

organized

hyponymy

into

}

,

WordNet

and

,

is are

{ person

{ cognition

,

representation

}

{ attribute

lexical

In

modifiers

n - dimensional

hierarchy

activity

.

WordNet

of as

separate

fauna

,

Nouns

means

nouns

,

different

explicitly

conventional

example

action

verbs

, .

structure

For

, 000

synonyms

-

relation

.

knowledge

the

forms

fields

{ act

Nouns

the of

partitions

which

semantic

is

consider lexical

64

of

represented

structured

hierarchical et

,

by

are

relation

single

WordN

concept

distinct

-

nouns

principles

,

nouns

the

:

categories

organized

hand

between

equivalent than

are

other

sets

adjectives

three

psychological

fundamental over

, 000

not

and

organization

verbs

on

,

these

while

,

The

lexical

structures

hierarchies Adjectives

actually

verbs

the

39

categories are

,

to

where contains

about

four

nouns

according ) ,

currently

words

view it

designed

, 1990b

into

into

:

is

1990a

WordNet

divided

words

that

,

organized

function

of

way

. are

that

Miller

synonymy

which

the

base

( WordNet

and ( 1984

delineate

,

metonymy

different

, from

Martin , 1993

,

also

,

and

I

polysemy of

as

, 1992 argue

,

much in

types

of

the

and it

. of

the

context

metaphorical

Brugman that

instead

metonymy

and

discussed

) .

have

and

course

certain 1990

)

work

of

is

difficult

Lakoff to

268

Notes to pages 235- 238

maintain any clear separation between lexical and linguistic knowledge, on the one hand , and general conceptual knowledge on the other , based on the modes of sense extension associated with the metaphorical usage of lexical items . This is similar to the position held by Nunberg (1995) as discussed above, but Nunberg would presumably still distinguish between lexically and non-lexically based inferences.

Bibliography

Abraham , W . 1986 . " Unaccusatives manistischen

in German ," Groningen

Abusch . D . 1985 . On Verbs and Time , Doctoral sachusetts

Arbeiten

zur Ger -

28 : 1- 72 . , Amherst

dissertation

, University

of Mas -

, MA .

Ackerman , F . 1992 . " Complex

Predicates

and Morphological

Relatedness : Loca -

tive Alternation in Hungarian ," in I . Sag and A . Szabolcsi (eds.) , Lexical Mat ters , CSLI Lecture Notes , Stanford , distributed by U . of Chicago Press . Ackerman , F . 1993 . " Entailments of Predicates and the Encoding of Causees ," Linguistic Inquiry 24 :535 - 546 . Allen , J . 1983 . " Maintaining Knowledge About Temporal Intervals ," Communi : cations of the A CM 26 :832- 843 . Allen ~ J . 1984 . "Towards a General Theory of Action and Time ," Artificial Intel ligence 23 : 123- 154 . Alsina , A . 1992 . " On the Argument Structure of Causatives ," Linguistic Inquiry 23 : 517 - 555 .

Alsina , A . and S. Mchombo . 1990 . "The Syntax of Applicatives in Chichewa ," Natural Language and Linguistic Theory 8 :493 - 506 . Amsler , R . A . 1980 . The Structure of the Merriam - Webster Pocket Dictionary , Doctoral dissertation , University of Texas , Austin , TX . Anderson

, M . 1979 . Noun

Connecticut

, Storrs

Phrase

Structure

, Doctoral

dissertation

, University

of

.

Anderson , M . 1984 . "Prenominal Genitive NPs ," The Linguistic Review 3 :1- 24 . Anick , P . and S. Bergler . 1992 . " Lexical Structures for Linguistic Inference ," in J . Pustejovsky and S. Bergler (eds .) , Lexical Semantics and Know ledge Repere sentation , Springer Verlag , Berlin . Anick , P . and J . Pustejovsky . 1990 . " An Application of Lexical Semantics to Knowledge Acquisition from Corpora ," in Proceedings of 13th International Conference on Computational Linguistics , Helsinki , Finland , 1990 , pp .7- 12.

Apresjan , J. D . 1973a. "Synonymy and Synonyms," in F . Kiefer (ed.) , 173- 199. Apresjan , J . D . 1973b . " Regular Polysemy ," Linguistics 142 :5- 32 . Apresjan , J . D . 1981 . " Semantic Amalgamation Rules for Russian Verbs ," in P .

Jacobson and H . L . Krag (eds.) , The Slavic Verb, Rosenkilde and Bagger, Copenhagen , 7- 13. Asher , N . and M . Morreau . 1991 . " Common Sense Entailment : A Modal Theory of Nonmonotonic Reasoning ," in Proceedings to the 12th International Joint Conference on Artificial Intelligence , Sydney , Australia . Atkins , B . T . 1987 . " Semantic ID Tags : Corpus Evidence for Dictionary Senses," in Proceedings of 3rd Annual Conference at University of Waterloo , Center for the

New

OED

, pgs . 17 - 36 .

Atkins , B . T . 1991 . " Building a Lexicon : Reconciling Anisomorphic Sense Differ entiations in Machine - Readable Dictionaries ," International Journal of Lexi cography .

Atkins , B . T and B . Levin . 1991. "Admitting Impediments ," in Zernik (ed.) , Lexi cal Acquisition NJ

: Using On -Line Resources to Build

a Lexicon , LEA , Hillsdale ,

.

Atkins , B . T ., J . Kegl and B . Levin . 1986 "Explicit and Implicit Information in Dictionaries ," Advances in Lexicology , Proceedings of the Second Conference of the Centre for the New OED , University of Waterloo , Waterloo , Ontario . Atkins , B . T ., J . Kegl and B . Levin . 1988 . " Anatomy of a Verb Entry : From Linguistic Theory to Lexicographic Practice ," International Journal of Lexi cography 1:84- 126 . Austin , J . L . 1962 . How to Do Things with Words , Harvard University Press , Cambridge , MA .

270

Bibliography

Bach, E. 1981 . "On Time, Tense , andAspect: an Essayin EnglishMetaphysics ," in P. Cole(ed.), RadicalPragmatics , AcademicPress , NewYork, 63-81. Bach, E. 1986 . "TheAlgebraof Events ," in LinguisticsandPhilosophy 9:5-16. Baker, C.L. 1969 . "Concealed Questions ," paperdeliveredat LSA. Baker, M. 1988 . Incorporation : A Theoryof Grammatical FunctionChanging , Universityof ChicagoPress , Chicago . Bar-Hillel, Y. 1953 . "A Quasi -arithmeticalNotationfor SyntacticDescription ," Language 29:47-58. Bartsch, R. 1985 . "The Structureof Word Meanings : Polysemy , Metaphor , Metonymy "in F. LandmanandF. Veltman(eds.) Varietiesof FormalSeman tics, Foris, Dordrecht . Beard, R. 1991 . "Decompositional Composition : TheSemantics of ScopeAmbiguitiesand'Bracketing Paradoxes '," in NaturalLanguage andLinguisticTheory 9:195- 229. Beckwith , R., C. Fellbaum , D. Gross , andG. Miller. 1989 . "WordNet : A Lexical DatabaseOrganized on Psycholinguistic Principles ," in U. Zernik(ed.), Proceedings of theFirst InternationalWorkshop on LexicalAcquisition , at IJCAI, Detroit. Beckwith , R. and G. A. Miller. 1990 . "Implementing a LexicalNetwork ," InternationalJournalof Lexicography 3:302-312. Beierle , C., U. Hedtsttick , U. Pletat, P. H. Schmitt, andJ. Siekmann . 1992 . "An Order-SortedLogicfor knowledge Representation Systems ," Artificial Intelli.qence , 55:149-191. Belletti, A. and L. Rizzi. 1988 . "Psych -verbsand Theta-Theory," NaturalLanguageandLinguisticTheory6:291-352. Benthem , J. van. 1983 . TheLogicof Time, KluwerAcademicPublishers , Dordrecht. Benthem , J. van. 1991 . Language in Action: Categories ,Lambda , andDynamic Logic, Studiesin Logic and the Foundations of Mathematics , Volume130, Noth-Holland,Amsterdam , alsopublishedin 1994 , by MIT Press , Cambridge . Bergler , S. 1991 . ('Thesemantics of collocational patternsfor reportingverbs," in Proceedings of theFifth Conference of theEuropean Chapterof theAssociation for Computational Linguistics , Berlin, April 9-11. Bergler,S. 1992 . EvidentialAnalysisof Reported Speech . PhDdissertation , BrandeisUniversity . Bes, G. and A. Lecomte . 1995 . "SemanticFeaturesin a GenericLexicon ," in P. Saint-DizierandE. Viegas(eds)., Computational LexicalSemantics , CambridgeUniversityPress . Bierwisch , M. 1983 . "Semantische undkonzeptuelle Reprasentationen lexikalischer Einheiten ," in R. RuzickaandW. Motsch(eds)., Untersuchungen zur Seman tik, Berlin, Akademische -Verlag. Bierwisch , M. andE. Lang. 1989 . Dimensional Adjectives . Grammatical Structure and Conceptual Interpretation , SpringerVerlag,Berlin. Bierwisch , M. andR. Schreuder . 1992 . '(FromConcepts to lexicalItems," Cognition 42:23-60. Bobrow, D. G. and T. Winograd . 1977 . "An Overviewof KRL, a Knowledge Representation Language ," CognitiveScience , 1:3-46. Boguraev , B. 1991 . "Buildinga Lexicon : The Contributionof Computers ," in B. Boguraev (ed.), Specialissueoncomputational lexicons , InternationalJournal of Lexicography 4(3). Boguraev , B. andT. Briscoe , eds. 1988 . Computational Lexicography for Natural Language Processing , Harlow,Essex , Longman . Boguraev , B. and J. Pustejovsky . 1990 . "LexicalAmbiguityand The Role of Knowledge Representation in LexiconDesign ," Proceedings of COLING-1990 ,

271

Bibliography

Helsinki. Boguraev, B. and J. Pustejovsky. 1994. "A Richer Characterizationof Dictionary Entries" in B. Atkins and A. Zampolli (eds.), Automating the Lexicon, Oxford University Press. Boguraev, B. and J. Pustejovsky, eds. 1996. CorpusProcessingfor Lexical Acquisition, MIT Press, Cambridge, MA . Bolinger, D. 1967. "Adjective Comparison: A SemanticScale," Journal of English Linguistics 1:2-10. Bolinger

,

D

.

1968

Bolinger

,

D

.

1971a

,

D

Inquiry Bolinger

.

,

Bouillon

, tation ings

P

.

,

.

.

:

through

Viegas

.

Adjectival

P

of

Nominal

Overloading

Trajectories

Structures in

a

study

1994

Knowledge

,

Kluwer

, "

the

Glossa

2 : 119

Progressive

of

the

.

" A

94 1995

.

,

Semi

,"

Verb

" The

:

A

a

- 127

.

Lan

-

Linguistic

' Remind

' , "

Dynamic

Publishers

- Polymorphic

:

Amsterdam

Space

Academic

Constructions

Euralex

, .

E

the

Cross

,

Framework

Boston

Approach - linguistic

to

. the

Perspective

Interpre ,"

-

Proceed

-

.

Semantics

of

Adjectival

Modification

,"

ms

.

ISSCO

,

.

Brachman

by

.

and

of

Geneva

Brachman

1994

Meaning

on

.

Comprehension

of

Bouillon

N

.

the

Note

" Semantic

- 547

Machine

and

Further .

.

: 522 L

" A

- 586

1971b

47

Bookman

" Entailment .

2 : 584

guage

for

.

R

.

Findler

J ,

. ed

Computers ,

1979

.

. ,

Associative

, R

.

J .

and

" On

Academic J

the

Epistemelogical

Press .

Schmolze

Status

Networks , .

: New

1985

York .

"

An

of

Semantic

Representation

and

Networks Use

of

,"

in

Knowledge

. Overview

of

the

KL

- ONE

Knowledge

Representation System," Cognitive Science 9.2. Bresnan , J., ed. 1972. Theory of Complementation in English Syntax, Ph .D . Dissertation , MIT . Bresnan , J., ed. 1982. The Mental Representation of Grammatical Relations , MIT Press, Cambridge , MA . Bresnan , J. and J. Kanerva . 1989. "Locative Inversion in Chichewa," Linguistic Inquiry 20:1-50. Bresnan , J. and L . Moshi . 1990. "Object Asymmetries in Comparative Bantu Syntax ," Linguistic Inquiry 21:147- 185. Bresnan , J. and A . Zaenen. 1990. "Deep Unaccusativity in LFG ," in K . Dziwirek et al., eds., Grammatical Relations : A Cross- Theoretical Perspective, CSLI , Stanford and U . of Chicago, Chicago. Briscoe , E ., A . Copestake, and B . Boguraev . 1990. "Enjoy the Paper: Lexical Semantics via Lexicology ," Proceedings of 13th International Conference on Computational Linguistics , Helsinki , Finland , pp . 42- 47. Briscoe ,T ., V . de Paiva , and A . Copestake (eds.) . 1993. Inheritance , Defaults , and the Lexicon, Cambridge University Press, Cambridge . Brugman , C. 1981. The Story of 'Over ': Polysemy, Semantics, and the Structure of the Lexicon , Master 's Thesis, University of California , Berkeley. Brugman , C. and G. Lakoff . 1988. "Cognitive Topology and Lexical Networks ," in S. I . Small , G. W . Cottrell , and M . K . Tanenhaus, eds., Lexical Ambiguity Resolution , Morgan Kaufmann , San Mateo , CA ., p. 477-508. Buitelaar , P., and A .-M . Mineur . 1994. "Compositionality and Coercion in Categorial Grammar ," in Proceedings of Ninth Amsterdam Colloquium , December 14- 17, 1993, Amsterdam . Burge , T . 1972. "Truth and Mass Terms," Journal of Philosophy 69, 263-282. Burzio , L . 1986. Italian Syntax: A Government and Binding Approach , Reidel , Dordrecht . Cardelli , L . 1988. "A Semantics of Multiple Inheritance ," InfoTm (l,tion and Computation 76:138- 164.

272

Bibliography

Cardelli , L . and P . Wegner . 1985 . " On Understanding Types , Data Abstraction , and Polymorphism ," ACM Computing Surveys , 17:471 - 522 . Carlson , G . 1977 . Reference to Kinds in English , PhD . dissertation , University of Massachusetts

, Amherst

.

Carlson , G . 1984 . "Thematic Roles and Their Role in Semantic Interpretation ," Linguistics 22 :259- 279 . Carlson , G . 1989 . " On the Semantic Composition of English Generic Sentences ," in

G . Chierchiat B . Partee and R. Turner (eds.), Properties , Types and Meaning , II

t Kluwer

, Dordrecht

, 167 - 192 .

Carnap , R . 1956 . Meaning and Necessity , University of Chicago Press , Chicago . Carpenter , B . 1992 . "Typed Feature Structures ," Computational Linguistics , 18 :2. Carrier , J . and J . H . Randall . 1992 . "The Argument Structure and Syntactic Struc ture of Resultatives ," Linguistic Inquiry 23 :173- 234 . Carter t R . 1988a . On Linking : Papers by Richard Carter , B . Levin and C . Tenny ., eds ., MIT Lexicon Project Working Papers 25 , Center for Cognitive Science , MIT , Cambridge , MA . Carter , R . 1988 . " Compositionality and Polysemy ," in B . Levin and C . TennYt (eds .) , On Linkin .fJ: Papers by Richard Carter , MIT Lexicon Proiect Working Papers no . 25 , Dept . of Linguistics and Philosophy , MIT , Cambridge . Cattell , R . 1984 . Composite Predicates in English , Syntax and Semantics 17, Aca demic

Press

, New

York

, NY

.

Centineo , G . 1986 . ~'A Lexical Theory of Auxiliary Selection in Italian ," Davis Working Papers in Linguistics , 1: 1-35 . Charniak , E . and R . Goldman . 1988 . "A Logic for Semantic Interpretation ," in Proceedings of the 26th Annual Meeting of the Association for Computational Linguistics , Buffalo . Chierchia

, G . 1982 . " Bare

Plurals

, Mass

Nouns

, and Nominalization

,"

WCCFL

1,

243 - 255 .

Chierchia , G . 1984 . Topics in the Syntax and Semantics of Infinitives and Gerunds , Doctoral dissertation , University of Massachusetts , Amherst , MA . Chierchia , G . 1989 . " A Semantics for Unaccusatives and its Syntactic Conse quences ," unpublished ms ., Cornell University , Ithaca , NY . Chierchia

, G . 1989 . " Structured

Meanings

, Thematic

Roles , and

Control

," in G .

Chierchia , B . Partee , and R . Turner , eds ., Properties , Types , and Meaning , vol . 2 , Kluwer

Academic

Publishers

, Dordrecht

, The

Netherlands

, 131 - 166 .

Chierchia , G . and S. McConnell -Ginet . 1990 . Meaning and Grammar : An Intro duction to Semantics , MIT Press , Cambridge . Chomsky , N . 1955 . The Logical Structure of Linguistic Theory , University of Chicago Press , first published 1975 . Chomsky , N . 1957 . Syntactic Structures , Mouton , The Hague . Chomsky , N . 1965 . Aspects of the Theory of Syntax , MIT Press , Cambridge . Chomsky , N . 1970 . "Remarks on Nominalization ," in R . Jacobs and P .S. Rosen baum (eds .) , English 7' ranformational Grammar , Ginn , Waltham , MA . Chomsky , N . 1972 . Studies on Semantics in Generative Grammar , Mouton , The Hague . Chomsky , N . 1973 . " Conditions on Transformations ," in S. Anderson and P .

Kiparsky , (eds.) , A Festschrift for Morris Halle , Holt , Rinehart , and Winston , New

York

.

Chomsky , N . 1975 . Reflections on Language , Pantheon , New York . Chomsky , N . 1981 . Lectures on Government and Binding , Faris Publications drecht

Chomsky , N . 1986 . Knowledge New

, Dor -

. York

of Language , its Nature , Origin , and Use, Praeger ,

.

Chomsky , N . 1994a . Language and Thought , Moyer Bell , Wakefield , R .I .

Bibliography

273

Chomsky , N . 1994b . " Language as a Natural Object ," Jacobsen Lecture , Univer sity College , London , May 23 , 1994 . Cohen , P.R ., J . Morgan , and M . Pollack , eds . 1990 . Intentions in Communication , MIT Press , Cambridge , MA . Collins , A . and M . Quillian . 1969 . " Retrieval Time from Semantic Memory ," Jour nal of Verbal Learning and Verbal Behavior , 9 :240- 247 . Comrie , B . 1976 . Aspect , Cambridge University Press , Cambridge . Comrie , B . 1980 . Tense , Cambridge University Press , Cambridge . Comrie , B . and S. Thompson . 1985 . " Causative Verb Formation and Other Verb deriving Morphology ," in T . Shopen (ed .) , Language typology and syntactic description .' Grammatical categories and the lexicon , Cambridge University Press , Cambridge . Conestake .. A . 1991 . " Defaults in the LRL ." in T . Briscoe , A . Copestake , and V . de Paiva , eds ., Proceedings of the Acquilex Workshop on Default Inheritance in the Lexicon , University of Cambridge Computer Laboratory Technical Report No . 238 , 1991 . Copestake , A . 1992 . The Representation of Lexical Semantic Information , CSRP 280 , University of Sussex . Copestake , A . 1995 . "The Representation of group denoting nouns in a lexical knowledge base ," in P . Saint - Dizier and E . Viegas , eds ., Computational Lexical Semantics , Cambridge University Press . Copestake , A . and T . Briscoe 1992 . " Lexical Operations in a Unification -based Framework ," in J . Pustejovsky and S. Bergler , eds ., Lexical Semantics and Knowledge Reperesentation , Springer Verlag , Berlin . Copestake , A . and T . Briscoe 1995 . " Semi -Productive Polysemy and Sense Exten sion ," Journal of Semantics 12. Copestake , A ., A . Sanfilippo , T . Briscoe , and V . de Paiva . 1993 . "TheACQUILEX LKB : An Introduction ." in T . Briscoe , V . de Paiva1 and A . Copestake (eds .) , Inheritance , Defaults , and the Lexicon , Cambridge University Press , Cam bridge . Croft , W .A . 1990 . " A Conceptual Framework for Grammatical Categories : A Taxonomy of Propositional Acts .," Journal of Semantics 7 :245 - 79 . Croft , W . 1991 . Typology and Universals , Cambridge University Press . Croft , W . 1991b . Categories and Relations in Syntax : The Clause -Level Organi zation of Information , University of Chicago Press . Crouch , R . and S. Pulman . 1993 . "Time and Modality in a Natural Language Interface to a Planning System ," Artificial Intelligence 63 :265- 304 . Cruse , D . A . 1986 . Lexical Semantics , Cambridge University Press . Cruse , D . A . 1995 . "Polysemy and related phenomena from a cognitive linguistic viewpoint ," in P . Saint - Dizier and E . Viegas (eds .) , Computational Lexical Se~ antics , Cambridge University Press . Dahl , O . 1981 . " On the Definition of the Telic -Atelic {Bounded -Nonbounded . Dis tinction ," Syntax and Semantics : Tense and Aspect 14, p . 79- 90 . Davidson , D . 1967 . "The Logical Form of Action Sentences ," in N . Rescher , ed ., The Logic of Decision and Action , Pittsburgh University Press , Pittsburgh . Davidson \ D . and G . Harman , eds ., 1972 . Semantics of Natural Language , Reidel , Dordrecht . Decles , J .- P . 1989 . "State , Event , Process , and Topology ," General Linguistics 29 :159- 200 . de Paiva , V . 1993 . " Types and Constraints in the LKB ," in T . Briscoe , V . de Paiva , and A . Copestake (eds .) , Inheritance , Defaults , and the Lexicon , Cambridge University Press , Cambridge . Diesing , M . 1992 . Indefinites , MIT Press , Cambridge .

274

Bibliography

Dini L . and F . Busa . 1994 . " Generative Operations in a Constraint -based Gram mar ," in Proceedings of KONVENS -94 , Vienna , Austria , September 28- 30 . Di Sciullo , A . M . and S. T . Rosen 1990 . " Light and Semi - Light Verb Construc tions ," in K . Dziwirek , P . Farrell and E . Mejfas - Bikandi , eds ., Grammatical Relations : A Cross - Theoretical Perspective , Center for the Study of Language and Information , Stanford University , Stanford , CA , 109- 125 . DiScuillo , A . M . and E . Williams 1987 . On the Definition of Word , MIT Press , Cambridge . Dixon , R . M . W . 1984 . " The Semantic Basis of Syntactic Properties ," BLS 10 ) 583 - 595 . Dixon , R . M . W . 1991 . A New Approach to English Grammar , on Semantic Prin ciples , Oxford University Press . Dolling , J . 1992 . " Flexible Interpretationen durch Sortenverschiebung ," in I . Zim mermann and A . Strigen (eds .) , Fugungspotenzen , Berlin , Akademie Verlag . Dor , D . 1992 . "Towards a Semantic Account of Concealed Questions ," ms . Stan ford University . Dowty , D .R . 1976 . " Montague Grammar and the Lexical Decomposition of Causative Verbs ," in B . Partee , ed ., Montague Grammar , Academic Press , New York , NY , 201-246 . Dowty , D . R . 1979 . Word Meaning and Montague Holland .

Gmmmar , D . Reidel , Dordrecht ,

Dowty , D .R . 1982 . " Grammatical Relations in Montague son and G .K . Pullum , eds ., The Nature of Syntactic Dordrecht . 79- 130 .

Grammar ," . in P . Jacob Representation , Reidel ,

Dowty , D . R . 1985 . " On Some Recent Analyses of Control ," Linguistics losophy 8 :1-41 . Dowty . D . R . 1989 . " On the Semantic Content of the Notion - ~ - ~- ' Thematic - -- -----~--1 G ~ 'Chierchia , B . Partee , and R . Turner , eds . Properties , Types , and Vol . 2, Semantic Issues , Dordrecht . Dowty , D . R . 1991 . " Thematic Proto -Roles and Argument Selection ," 67 :547 - 619 .

and Phi R.ol - -- --~' ." in Meaning ---, Language

Eberle , K . 1988 . " Partial Orderings and Aktionsarten in Discourse Representation Theory ," in Proceedings of COLING - 88 , Budapest , 1988 , pg . 160- 165 . Elliott , D . E . 1974 . " Towards a Grammar of Exclamations ," Foundations of Lan guage 11. En ~, M . 1981 . Tense without Scope, Ph .D . Dissertation , University of Wisconsin , Madison . En ~, M . 1987 . "Anchoring Conditions for Tense ," Linguistic Inquiry 18:633 - 658 . Eschenbach , C . 1993 . " Semantics of Number ," Journal of Semantics , 10:1-23 . Evans , R . and G . Gerald . 1989 . " Inference in DATR ," in the Proceedings of the Fourth European ACL Conference , April 10- 12 , 1989 , Manchester , England . Evans , R . and G . Gerald . 1990 . "The DATR papers : February 1990 ." Cognitive Science Research Paper CSRP 139 , School of Cognitive and Computing Science , University of Sussex , Brighton , England . Evens , M . 1987 . Relational Models of the Lexicon , Cambridge University Press , Cambridge . Farkas , D . F . 1988 . " On obligatory 58 .

Control ," in Linguistics

and Philosophy , 11:27-

Fass , D . 1988a . "An Account of Coherence , Semantic Relations , Metonymy , and Lexical Ambiguity Resolution ," in S.l . Small ~ G .W . Cottrell ~ and M .K . Tanen haus , eds . Lexical Ambiguity Resolution , Morgan Kaufmann , San Mateo , CA . Fass , D . 1993 . " Lexical Semantic Constraints ," in J . Pustejovsky (ed .) Semantics and the Lexicon , Kluwer Academic Publishers , Dordrecht . Fauconnier , G . 1985 . Mental

Spaces, MIT

Press , Cambridge .

Bibliography

275

Fellbaum , C. 1990. "English Verbs as a Semantic Net ," International Journal of Lexicography, 3:278-301. Fillmore , C. 1968. "The Case for Case," in Universals in Linguistic Theory, E. Bach and R . Harms , eds.,. New York , Holt , Rinehart , and Winston . Fillmore , C. 1971. "Types of Lexical Information ," in D . Steinberg and L . Jakobovits , eds., 370-392. Fi11mor~., C. 1977a. "Tonics in Lexical Semantics," in R. W . Cole, ed., Current Issues in Linguistic Theory, Indiana University Press, Bloomington . Fillmore , C. 1977b. "The Case for Case Re-opened," in P. Cole, ed., Syntax and Semantics, Vol. 8, Academic Press, New York . Fillmore , C. 1986. "Pragmatically Controlled Zero Anaphora ," BLS 12, 95- 107. Fillmore , C. 1988. "The Mechanisms of Construction Grammar ," BLS 14, 35-55. Fillmore , C. and B . T . Atkins . 1991. "Toward a Frame-Based Lexicon : The Semantics of Risk ," paper delivered at ACL , Berkeley, CA . Fillmore , C. and B . T . Atkins . 1994. "Starting where the dictionaries stop : the Challend of corpus lexicography ," in B . T . Atkins and A . Zampolli , eds. Computational Approaches to the Lexicon, Oxford University Press, Oxford . Fillmore ., C. and D . T . Langendoen, eds., 1971. Studies in Linguistic Semantics, Holt , Rinehart and Winston , New York , NY . Fisher , C., H . Gleitman , and L . R . Gleitman 1991. "On the Semantic Content of Subcategorization Frames," Cognitive Psychology 23. Flickinger , D . C. Pollard , and T . Wasow 1985. "Structure -Sharing in Lexical Representation ," Proceedings of 23rd Annual Meeting of the ACL , Chicago, IL , pp .262- 267. Fodor , J. 1975. The Language of Thought, Harvard University Press, Cambridge . Foley, - W .. and R . Van Valin 1984. Functional Syntax and Universal Grammar , Camrbidge University Press, Cambridge . Freed, A . F . 1979. The Semantics of English Aspectual Complementation , Dor drecht : Reidel . Gawron , M . 1983. Lexical Semantics and the Semantics of Complementation , Ph .D . dissertation , University of California at Berkeley. Gazdar , G. , E . Klein , G . Pullum , and I . Sag. 1985. Generalized Phrase Structure Grammar , Harvard University Press. Gentner , D . and I . M . France 1988. "The Verb Mutability Effect : Studies of the Combinatorial Semantics," in S. I . Small , G. W . Cottrell , and M . K . Tanenhaus, eds., Lexical Ambiguity Resolution , Morgan Kaufmann , San Mateo , CA . Gleitman , L . 1990. "The Structural Sources of Verb Meaning ," Language Acqui sition , 1, 3-55. Godard , D . and J. Jayez. 1993. "Towards a proper treatment of Coercion Phenomena," in Proceeding of the 1993 European ACL . Goldberg , A . 1992. Argument Structure Constructions . Ph .D . Dissertation , Uni versity of California , Berkeley. Goldberg , A . 1995. Constructions : A Construction Grammar Approach to Argu ment Structure , University of Chicago Press, Chicago. Goodman , N . 1951. The Structure of Appearance , Reidel Publishing , Dordrecht , The Netherlands . Green, G. 1974. Semantics and Syntactic Regularity , Indiana University Press, Bloomington . Grice , H . P. 1971. "Meaning ," in D . Steinberg and L . Jacobovits , eds., Semantics : An Interdisciplinary Reader in Philosophy , Linguistics , and Psychology, Cambridge , Cambridge University Press. Grimshaw , J. 1979. "Complement Selection and the Lexicon " Linguistic Inquiry 10:279-326.

276

Bibliography

Grimshaw, J. 1982. "Subcategorizationand GrammaticalRelations," in A. Zaenen (ed.), Subjectsand other Subjects : Proceedingsof the Harvard Conferenceon GrammaticalRelations, Bloomington, Indiana Linguistics Club, p. 35-55. Grimshaw, J. 1990. Argument Structure, MIT Press, Cambridge, MA . Grimshaw, J. and S. Vikner 1993. "Obligatory Adjuncts and the Structure of Events," in E. Reuland and W. Abraham (eds.), Knowledgeand Language , Vol. II , Kluwer AcademicPublishers, Dordrecht, p. 143-155. Grinder, J. 1970. "SuperEqui-NP Deletion," CLS 6. Groenendijk, J., and M. Stokhof. 1989. "Type-Shifting Rulesand the Semanticsof Interrogatives" , in G. Chierchia, B. Partee, and R. Turner (eds.), Properties, Tpyes, and Meaning, Dordrecht, Reidel. Gropen, J., S. Pinker, M. Hollander, and R. Goldberg 1991. "Affectednessand Direct Objects: the Role of Lexical Semanticsin the Acquisition of Verb Argument Structure," Cognition41:153-195. Gross, D. and K . Miller 1990. "Adjectivesin WordNet," International Journal of Lexicography , 3, pp. 265-277. Grosz, B. and C. Sidner. 1986. "Attention, Intentions, and the Structure of Discourse," ComputationalLinguistics 12:175-204. Gruber, J. S. 1967. "Lookand See," Language43, 937-947. Gruber, J. S. 1976. Lexical Structuresin Syntax and Semantics, North-Holland, Amsterdam. Guentcheva , Z. 1990. "L'oppostion perfectifjimperfectif et la notion d'acheve ment," in J. Fontanille (ed.), Le discours aspectualise , Collection Nouveaux Actes Semiotiques , Benjamins, Paris. Gunter, C. 1992. Semanticsof ProgrammingLanguages , MIT Press, Cambridge. Hacking, I. 1983. Representingand intervening: introductory topics in the philosophy of natural science, CambridgeUniversity Press, New York. Hale, K . and S. J. Keyser. 1986. {'SomeTransitivity Alternations in English," Lexicon Project Working Papers7, Centerfor CognitiveScience , MIT , Cambridge, MA . Hale, K . and S. J. Keyser. 1987. "A View from the Middle," Lexicon Project Working Papers 10, Center for Cognitive Science , MIT , Cambridge, MA . Hale, K .L. and S. J. Keyser 1992. "The Syntactic Character of Thematic Structure," in I . M. Roca, ed., Thematic Structure: Its Role in Grammar, Foris, Berlin, 107-143. Harris, Z. 1951. Methodsin Structural Linguistics, University of ChicagoPress, Chicago. Harris, Z. 1952. "DiscourseAnalysis" Language28:130. Harris, Z. 1957. "Co-Occurrenceand Transformationin Linguistic Structure" Lan.qua.Qe33:283-340. Harris, Z. 1965. "TransformationalTheory," Language41:363-401. Hayes, P. 1979. "Naive PhysicsManifesto," in D. Mitchie (ed.), Expert Systemsin the Micro-electronicAge, Edinburgh University Press, Edinburgh, Scotland. Heim, I . 1979. "ConcealedQuestions", in R. Bauerle, U. Egli, and A. von Stechow (eds.), Semanticsfrom Different Points of View, Springer, Berlin. Hendriks, H. 1987. "Type Changein Semantics : The Scopeof Quantification and Coordination," in E. Klein and J. van Benthem (eds.), Categories , Polymorphism, and Unification, Centrefor CognitiveScience , University of Edinburgh, and Institute for Language, Logic and Information, University of Amsterdam. Heylen, D. 1995. "Lexical Functionsand KnowledgeRepresentation ," in P. SaintDizier and E. Viegas(eds.), ComputationalLexicalSemantics,CambridgeUniversity Press. Herskovits, A. 1987. Languageand Spatial Cognition, CambridgeUniversity Press, Cambridge, UK.

277

Bibliography

Herskovits , A . 1988. "Spatial Expressions and the Plasticity of Meaning ," in B . Rudzka- Ostyn (ed.) , Topics in Cognitive Grammar , John Benjamins , Amster dam Herzo

, 271 .e; ,

- 297

.

O . and

grating

C .- R .

CL

Higginbotham native

to

and

AI

, J .

1983

.

Situation

Higginbotham

,

quiry

14 : 395

J .

1983

- 420

.

Higginbotham

, J . 1985 ,

12 : 465

- 517

J .

Hirst

English ,.

G .

Hirst

1987

,

G .

.

1988

( eds

gies

,

Hobbs

1987

.

Meaning

:

An

:

Extensional

Philosophy and

80

Nominals

,

Inte

-

- 127

-

.

Linguistic

16 : 547

Linguistics

Alter

100

,"

Inquiry ,"

for

- 593

In

-

.

and

an

,"

, T

Philosophy

Computationally , G .W

S . I . Small

of

NP

Reference

Ambiguity

,

of , W

Cam

-

Word

Meaning

, and

,"

,"

. Ringle

,"

K

. Laws

.

Computational

. Edwards ,"

M

. Tanen

Mateo

Discourse

and

.K

-

, CA

in

, eds

.

Strate

-

. , Hills

-

in

Proceedings

of

.

, D . Edwards

Base

M

, San

in

. Lehnert

and

, D

Kaufmann

Coherence

Spreading

, and

.

Mexico

Semantics

with

. Cottrell

, Morgan

Knowledge

. Davies

Resolution

Ambiguity

Associates

. Davies

and

.

in

Processing

Lexical , T

the

Understanding

, New

Aktionarten

University

Resolution

Knowledge

, and

K

Artificial

1987a

.

" Commonsense

Linguistics . Laws

13 .

. 1987b

Intelligence

.

" The

Center

,

TACITUS SRI

Interna

-

. . Stickel

, P . Martin

Intelligence ,

T

.

Structure

1992 .'

, L . R . 1980

Horn

,

L .

R

.

Chicago

.

, D . Edwards : 69 - 142

" Aspect

Role

.

Theta

Grammar

Natural

.

" Projection

Theory

, Foris

and

A

. 1993

. " Interpretation

as

Abduction

,"

.

and

in

" Affixation

1989

the

,"

, Berlin

in ,

Unaccusative

History

of

I .

145

M

.

Roca

- 174

( ed

Hypothesis

Negation

!

Event

Structure

.) ,

Thematic

. ,"

University

CLS of

16 , 134

- 146

Chicago

.

Press

,

.

, A

. van

.

Everaert

1995

, and

Lawrence , E .

1993

.

,"

Artificial

, R . and

R .

Associates

,"

Project

J . in

Jackendoff

147

Tenny

, M

.

Insertion

,

- 385

using

Discourse

Structure

Re

-

.

Bake

Revisited

,"

International

Journal

of

1990 .,

2 ,

.

The Center

" Active

Objects

MIT

in

Parsing for

Syntax

Volume

Cognitive

,

Science

,

Semantics

1988 ,

,

- 1989

MIT

,

,

and

Parsing

Cambridge

,

, R .

1983

,

R .

1985a

,

R .

and

. Semantics .

and

" Multiple

Linguistic

1985b Inquiry

Interpretation

in

Generative

Grammar

, MIT

Press

,

.

Jackendoff

Linguistic

ed

. Semanic

, MA

Language

. ,

Papers

J ackendoff

J ackendoff

Lexical

.

, R . 1972

Cambridge

P . Coopmans

, N .J .

: 341

" English

in

and

.

Pustejovsky

C .

- 169

.

,"

Specification

Generation 63

1989 - 346

Working ,

Discourse

' cuk

2 : 325

on

. ) , Lexical

, Rillsdale

Intelligence

I . Mel

and

( eds

" Automated

Lexicography

Parsing

based

J . Grimshaw

Erlbaum

lations

,

63

.

Its

Horn

MA

Linguistic

of

and

Lexical

Cruces

. Croft

, J ., M

Ingria

LILOG

.

Words

" World

and

, J ., W

Hoekstra

Iison

,"

State

, Cambridge

Erlbaum

. Croft

Artificial

Hovy

,

Semantics

Language

Commonsense

Rout

in

Reports of

Binding

Interpretation -

. " Towards

, J ., W

Hobbs

,

Ohio

Ambiguity

- 3 , Las

tional

Form

,

Polaroid

Metaphysics Hobbs

Understanding .

Journal

Semantics

" Resolving

Natural

J .

Text

Perceptual

" Elucidations

Press

, N J , Lawrence

TINLAP

.

, Berlin

The

Compositional

. ) , Lexical

for

" On .

1991

of ,"

" Logical

Semantic

.

, J . 1982

dale

Logic

Dissertation

and

haus

Hobbs

. A

University

Activation

Hobbs

.

1989

. PhD

bridge

" The

.

.

- Verlag

.

, E . 1985

in

, eds

Semantics

Higginbotham

Hinrichs

Rollinger , Springer

.

Cognition

Theory

3 : 271

" Believing 16 : 445

, MIT

Subcategorization

and - 460

.

Press and

- 295

Intending

, Cambridge the

O - Criterion

, MA ,"

. Natural

. :

Two

Sides

of

the

Same

Coin

,"

278

Bibliography

Jackendoff, R. 1985c. " Multiple

Subcategorization and the Theta - Criterion : The Case of Climb ," Natural Language and Linguistic Theory 3 '271- 295 . Jackendoff , R . 1987 . "The Status of Thematic Relations in Linguistic Theory ," Linguistic Inquiry 18:369 -411 . J ackendoff , R . 1990 . Semantic Structures , MIT Press , Cambridge , MA . Jackendoff , R .S. 1992a . " Babe Ruth Homered His Way into the Hearts of Amer ica ," in T . Stowell and E . Wehrli (eds .) , Syntax and the lexicon , Academic Press , San Diego , 155- 178 . Jackendoff , R . 1992b . " Mme . Tussaud Meets the Binding Theory ," Natural Lan guage and Linguistic Theory 21 :427 -455 . Jackendoff , R . 1993 . "The Conceptual Structure of Intendingand Volitional Ac -

tion ," in H . Campos and P. Kempchinsky (eds.) , Festschrift for Carlos Otero, Georgetown University Press , Georgetown . Jacobs , R . and P . Rosenbaum , eds . 1967 . Readings in English Grammar

, Ginn

, Waltham

, MA

Transformational

.

Jakobson , R . 1970 . Recent Developments in Linguistic Science , Perenial Press . Jacobson , P. 1990 . " Raising as Function Composition ," in Linguistics and Philos ophy 13:423 -476 . Jacobson , P . 1992 . " The Lexical Entailment Theory of Control and the Tough Construction ," in I .A . Sag and A . Szabolcsi , (eds .) , Lexical Matters , CSLI , Stanford

, 269 - 299 .

Jespersen , O . 1924 . The Philosophy of Grammar , Norton , New York . Johnston , M . 1995 . " Semantic Underspecification and Lexical Types : Capturing Polysemy without Lexical Rules ," in Proceedings of ACQUILEX Workshop on Lexical Rules , August 9- 11, 1995 , Cambridgeshire . Johnston , M ., B . Boguraev , and J . Pustejovsky . 1995 . "The Acquisition and In terpretation of Technical Complex Nominals ," in Proceedings of AAAI Sym posium on Lexical Acquisition , March 29- 31 , Stanford . Kamp , H . 1979 . " Some Remarks on the Logic of Change : Part 1" in C . Rohrer ,

(ed.) , Time , Tense, and Quantifiers , Tlibingen , Niemeyer. Kamp , H . and U . Reyle . 1993 . From Discourse lishers

, Dordrecht

to Logic , Kluwer

Academic

Pub -

.

Kamp , H . and A . RoBdeutscher . 1992 . " Remarks on Lexical Structure , DRS Construction , and Lexically Driven Inferences ," Arbeitspapiere des Sonder forschungsbereichs 340 Sprachtheoretische Grundlagen fur die Computerlin guistik , Nr . 21 , Stuttgart . Karttunen , L . 1971 . " Implicative Verbs " . Language 47 :340 -58 . Karttunen , L . 1974 . " Presupposition and Linguistic Context " . Theoretical Lin guistics 1:181-93 . Katz , J . 1964 . " Semantic Theory and the Meaning of ' Good ' ," Journal of Philos ophy61 :739- 766 . Katz , J . 1972 . Semantic Theory , Harper and Row , New York . Katz , J . and J . Fodor . 1963 . "The Structure of a Semantic Theory ," Language 39 : 170 - 210 .

Kautz , H . 1987 . A Formal Theory of Plan Recognition versity of Rochester . Keenan , E . and L . Faltz . 1985 . Boolean Semantics for Publishing , Dordrecht . Kenny , A . 1963 . Action , Emotion , and Will , Routledge Keyser , S. J . and T . Roeper . 1984 . " On the Middle and English ," Linguistic Inquiry 15:381-416 . Keyser , S. J . and T . Roeper . 1992 . " Re : The Abstract guistic Inquiry 23 :89- 125 .

, Ph .D . Dissertation Natural

, Uni -

Language , Reidel

and Kegan Paul , London . Ergative Constructions in Clitic

"

Hypothesis .

Lin -

279

Bib Ii 0gr aphy

Kiefer , F . 1992 . " Hungarian

Derivational

Morphology

, Semantic

Complexity

, and

Semantic Markedness," in I .A . Sag and A . Szabolcsi, (eds.), Lexical Matters , CSLI

, Stanford

, 183 - 208 .

Kilgarriff , A . 1995 . " Inheriting Polysemy ," in P . Saint -Dizier and E . Viegas (eds .) , Computational Lexical Semantics , Cambridge University Press , Cambridge . Kiparsky , P . and C . Kiparsky . 1971 . "Fact ," in D . Steinberg and L . Jakobovitz ,

(eds.) , Semantics, Cambridge University Press, Cambridge , 345-369. Klavans , J ., M . Chodorow , N . Wacholder . 1990 . " From Dictionary to Knowledge Base via Taxonomy ," Proceedings of the 6th Conf . UW Centre for the New OED , Waterloo

, 110 - 132 .

Klein , E . and I . Sag . 1985 . " Type - Driven Translation

," Linguistics

and Philosophy

8 : 163 - 202 .

Klein , E . and J . van Benthem . 1987 . Categories , Polymorphism , and Unification , Centre for Cognitive Science , University of Edinburgh , and Institute for Lan guage , Logic and Information , University of Amsterdam . Kowalski , R . and M . Sergot . 1986 . "A Logic -based Calculus of Events ," New Gen eration Computing 4 :67-95 . Kratzer , A . 1988 . Amherst .

" Stage - level

vs . Individual

- level

Predicates

,"

ms . UMASS

,

Kratzer , A . 1989 . " An Investigation into the Lumps of Thought ," Linguistics and Philosophy , 12:607- 653 . Krifka , M . 1989 . " Nominal Reference , Temporal Constitution , and Quantification in

Event

Semantics

,"

in

R . Bartsch

, J . van

Ebthe

, and

P . van

Emde

Boas

(eds.), Semantics and Contextual Expressions, Foris, Dordrecht , pg. 75- 115. Krifka

, M . 1992 . " Thematic

Relations

as Links

between

Nominal

Reference

and

Temporal Constitution ," in I . Sag and A . Szabolcsi, (eds.), Lexical Matters , CSLI Lecture Notes , University of Chicago Press , Chicago , Ill . Kunze , J . 1991 . Kasusrelationen und semantische Emphase , Akademie Berlin

Verlag ,

.

Ladusaw , W . and D . Dowty 1988 . "Toward a Nongrammatical Account of The matic Roles ," in W . Wilkins , (ed .) , 62- 73 . Lakoff , G . 1968 . " Instrumental Adverbs and the Concept of Deep Structure ," Foundations of Language , 4 :4- 29 . Lakoff G . 1970 . Irregularity in Syntax . Holt , Rinehart , and Winston . Lakoff G . 1971 . " On Generative Semantics ," in Semantics : An Interdisciplinary Reader , D . Steinberg and L . Jakobovits , eds ., Cambridge University Press . Lakoff

, G . 1972 . " Linguistics

and

Natural

Logic

," in D . Davidson

and

G . H . Har -

mon , eds ., The Semantics of Natural Language , Reidel , Dordrecht , 545 -665 . Lakoff G . 1987 . Women , Fire , and Dangerous Objects , University of Chicago Press , Chicago , 1987 . Lambek , J . 1958 . " The Mathematics matica ~ Monthlv 65 . 154 - 170 . also

of Sentence Structure ," American in W . Buskowski , W . Marciszewski

Mathe , and J .

van Benthem (eds.) , 1988, Categorial Grammar , John Benjamins , Amsterdam , 153 - 172 .

Lambek , J. 1961. "The the Calculus of Syntactic Types ," in R . Jakobson (ed.) , Structure of Language and its Mathematical Aspects , Proceedings of the Sym posia in Applied Mathematics XII , American Mathematical Society , 166- 178 Lamiroy , B . 1987 . " Les Verbes de Mouvement Emplois Figures et Extensions Metaphoriques ," Langue Fran ~aise 26 :41-58 . Lamiroy , B . 1987 . " The complementation of aspectual verbs in French ," in Lan guage

, 63 .

Landman , F . 1991 . Structures drecht

.

for Semantics , Kluwer

Academic

Publishers , Dor -

280

Bibliography

Lang , E . 1989 . " The Semantics

of Dimensional

Designation

of Spatial

Objects ,"

in M . Bierwisch and E . Lang (eds.) , Dimensional Adjectives : Grammatical Structures and Conceptual Interpretation , Springer Verlag , Berlin . Lang , E . 1991 . " The LILOC Ontology from a Linguistic Point of View ," in O .

Herzog and C.-R . Rollinger (eds.) , Text Understanding in LILOC : Integrating GL and AI , Springer -Verlag , Berlin . Langacker , R .W . 1991 . Foundations of Cognitive Grammar : Descriptive Applica tion , Vol . 2 , Stanford University Press , Stanford , CA . Lascarides A . and N . Asher . 1993 . " Temporal Interpretation , Discourse Relations and Commonsense Entailment ," in Linguistics and Philosophy 16:437 -494 . Lascarides , A ., E .J . Briscoe , N . Asher and A . Copestake . 1994 . "Persistent associative default unification ," Lin .quistics and Philosophy 17. Lasnik , H . and R , Fiengo . 1974 . " Complement Object Deletion ," Linguistic In Quiry

5 :535 - 571 .

Lebeaux , D . 1986 . " The Interpretation of Derived Nominals ," CLS fJfJ, Part 1, Papers from the General Session , 231- 247 . Lees , R . B . 1966 . The Grammar of English Nominalizations , Mouton , The Hague . Leifi , H . 1991 . " Polymorphic Constructs in Natural and Programming Languages ," Semantics of Programming Languages , Springer , Berlin . Lenders , W . 1989 . " Computergestiitzte Verfahren zur semantischen Beschreibung yon Spracher ," in Handbuch Computerlinguistik , Berlin , De Gruyter , pg . 231244 .

Levi , J . N . 1978 . The Syntax Press

, New

York

and Semantics

Levin , B ., ed . 1985 . Lexical Semantics 1 , MIT

of Complex

Nominals , Academic

.

in Review , Lexicon Project

Working

Papers

.

Levin , B . 1985. "Introduction ," in B . Levin , (ed.) , Lexical Semantics in Review, Lexicon Project Working Papers 1, Center for Cognitive Science , MIT , 1-62 . Levin , B . 1993 . Towards a Lexical Organization of English Verbs , University of Chicago Press , Chicago . Levin , B . and T . R . Rapoport 1988 . " Lexical Subordination ," Proceedings of CLS 24 , 275 - 289 , 1988 .

Levin , B . and M . Rappaport . 1986 . " The Formation of Adjectival Passives ," Lin guistic Inquiry , 17.4 . Levin , B . and M . Rappaport . 1988 . " On the Nature of Unaccusativity ," in Pro ceedings of NELS 1988 . Levin , B . and M . Rappaport Hovav . 1991 . " Wiping the Slate Clean : A Lexical Semantic Exploration ," Cognition 41 :123- 151. Levin , B . and M . Rappaport Hovav . 1995 . Unaccusatives : At the Syntax -Lexical Semantics Interface , MIT Press , Cambridge , MA . Levin , L . 1986 . Operations on Lexical Forms . Ph .D . dissertation , MIT , Cambridge . Li , J . 1994 . Riiumliche Relationen und Objektwissen : am Beispiel 'an ' und 'bei ', Gunter Narr , Tiibingen . Link , G . 1983 . " The Logical Analysis of Plurals and Mass Terms : A Lattice -

Theoretic Approach ," in r . Bauerle, C. Schwarze, and A . von Stechow, (eds.) , Meaning , Use, and Interpretation of Language , Mouton , Berlin . Lloyd , G . E . R . 1968 . Aristotle : The Growth and Structure of his Thought , Cam bridge University Press . LoCascio , V . and C . Vet , eds . 1986 . Temporal Structure in Sentence and Discourse , Foris

, Dordrecht

.

Lytinen , S. L . 1988 . " Are Vague Words Ambiguous ?" in S. I . Small , G . W . Cottrell , and M . K . Tanenhaus , eds ., Lexical Ambiguity Resolution , Morgan Kaufmann , San

Mateo

, CA

.

281

Bibliography

McArthur , T . 1981. LongmanLexicon of ContemporaryEnglish, Longman, Harlow, Essex. McCawlev. J. D. 1968b. "The Role of Semanticsin a Grammar," in E. Bach and R. T . Harms (eds.), Universalsin Linguistic Theory, Holt, Rinehart, and Winston, New York, NY. McCawley, J. D. 1975. "Lexicographyand the Count-Mass Distinction," BLS 1, 314-321. Reprinted in J.D. McCawley(1979), p. 165-173. McCawley, N. 1976. "On ExperiencerCausatives ," in M. Shibatani (ed.), 181-203. McCawley, J. 1979. Adverbs, Vowels , and other Objectsof Wonder, University of ChicagoPress, Chicago. McClure, W . 1990. "A Lexical Semantic Explanation for UnaccusativeMis.matches," in K . Dziwirek, P. Farrel, and E. Mejias-Bikandi (eds.), Grammatical Relations: A Cross- TheoreticalPerspective , CSLI, Stanford. McConnell-Ginet. S. 1982. "Adverbsand Lo.gicalForm: A Linguistically Realistic Theory ," Language 58 : 144- 184 . McKeon , R ., ed . 1968 . The Basic Works of Aristotle , Random Marantz

. A .P . 1984 . On

the Nature

of Grammatical

Relations

House , New York . , MIT

Press , Cam -

bridge , MA . Martin ~ J . 1990 . A Computational Model of Metaphor Interpretation , Academic Press , Cambridge , MA . Martin , J . 1992 . " Conventional Metaphor and the Lexicon ," in J . Pustejovsky and S. Bergler (eds .) , Lexical Semantics and Knowledge Reperesentation , Springer Verlag , Berlin . Mel 'cuk , I . A . 1973a . " Lexical Functions in Lexicographic Description ," BLS 8 , 427 - 444

.

Mel 'cuk , I . A . 1988a . " Semantic Description of Lexical Units in an Explanatory Combinatorial Dictionary : Basic Principles and Heuristic Criteria ," Interna tional Journal of Lexicography 1:165- 188 . Mel ' cuk , I . A . 1988b . Dependency Syntax , SUNY Press . Albany , New York . Miller , G . 1985 . "Dictionaries of the Mind " in Proceedings of the 23rd Annual Meeting of the Association for Computational Linguistics , Chicago . Miller , G . 1989 . " Contexts of Antonymous Adjectives ," Applied Psycholinguistics 10 : 357 - 375 .

Miller , G . 1990a . " Nouns in WordNet : A Lexical tional Journal of Lexicography 3:245- 264 . Miller

, G . , ed . 1990b . " WordNet

: An

On - Line

Inheritance

Lexical

Database

Journal of Lexicography 3 :4 . Miller , G . 1991 . The Science of Words , Scientific

American

Miller

and

, G . , R . Beckwith

, C , Fellbaum

, D . Gross

System ," Interna ," International

Press , New York .

K . J . Miller

. 1990 . " Intro

-

duction to WordNet : An On - line Lexical Database ," International Journal of Lexicography 3 :235 - 244 . Miller , G . and C . Fellbaum 1991 . " Semantic Networks of English ," Cognition 41 : 197 - 229 .

Miller , G . and P . Johnson - Laird . 1976 . Language and Perception , Belknap , Har vard University Press , Cambridge , MA . Mineur , A .- M . and P . Buitelaar . 1995 . "A Compositional Treatment of Polysemous

Arguments in Categorial Grammar ," in S. Peters and K . van Deemter (eds.) , , . . .

Ambiguity and Underspecification ~ CSLI , Stanford University . Minsky , M . 1975 . "A Framework for Representing Knowledge " in P . Winston , ed ., The Psychology of Computer Vision , McGraw - Hill , New York . Mitchell , J . C . 1988 . " Polymorphic Type Inference and Containment ," Information and Computation 76 . Mitchell , J . C . 1991 . "Type Inference with Simple Subtypes ," Journal of Func tional Programming 1:245 - 285 .

282

Bibliography

Modrak, D. 1987. Aristotle: the Power of Perception, University of ChicagoPress, Chicago. Moens, M. 1987. Tense, Aspectand TemporalReference . Ph.D. Dissertation, Edinburgh. Moens, M. and M. Steedman . 1988."TemporalOntologyand TemporalReference ," ComputationalLinguistics 14:15-28. Montague, R. 1970. "Universal Grammar," Theoria 36:373-398. Reprinted in Thomason1974, pp. 222-246. Montague, R. 1973. "The Proper in OrdinarY - Treatment of Quantification . --------'" -En-glish," in K . Hintikka, J. Moravcsikand P. Suppes(eds.), Approachesto Natural Language , Dordrecht, Kluwer, pp. 221-242. Reprinted in Thomason1974, pp. 247-270. Moortgat, M. 1988. CategorialInvestigations: Logical and Linguistic Aspectsof the LambekCalculus, Foris, Dordrecht. Moravcsik, J. M. 1975. "Aitia as GenerativeFactor in Aristotle's Philosophy," Dialogue14:622-36. Moravcsik, J. M. 1981. "How do Words get their Meanings?," Journal of Philosophy 78:5-24. Moravcsik, J. M. 1990. Thoughtand Language , Routledge, London. Morrill , G. 1992. Type-Logical Grammar, Onderzoeksinstituutvoor Taal en Spraak, Utrecht. -

Morrill

.

,

G

drecht

Morrill

.

1994

,

G

. ,

and

Theories

B

of

Mourelatos

,

losophy

- Logical

A

and - 212

.

Grammar

Carpenter

Grammar .

2 : 415

Syntax 191

Type

,

Kluwer

Academmic

Publishers

,

Dor

-

.

P

.

D

- 434 Semantics

. ,"

.

1978 .

1990

.

" Compositionality

Linguistics .

Also

" Events published

14

;

Tense

and ,

Processes in and

,

Philosophy and P

.

States

Tedeschi

Aspect

Implicational 13

and ,

Academic

:4 , " A

Logics

,

and

. Linguistics .

and

Zaenen

Press

( eds ,

New

York

. ) ,

Phi

-

1981

.

,

,

NY

.

Nebel, B. 1990. Reasoningand Revision in Hybrid RepresentationSystems , SpringerVerlag, Berlin. Nirenburg, S. 1989a. "Lexiconsfor Computer Programs and Lexicons for People," in Proceedingsof the 5th Annual Conferenceof the UW Centrefor the New Oxford English Dictionary, pp. 43-66. St. Catherine's College~ Oxford, England. Nirenburg, S. 1989b. "Knowledge -BasedMachineTranslation," Machine Translation (specialissueon knowledge - basedMT ), 4:1. Nirenburg, S. and L. Levin. 1989. "KnowledgeRepresentationand Support," Machine Translation (specialissueon knowledge -basedMT ), 4:1. Nirenburg, S. and V . Raskin. 1987. "The SubworldConceptLexiconand the Lexicon ManagementSystem," ComputationalLinguistics 13:3. Nunberg, G. 1978. The pragmatics of Reference . Indiana University Linguistics Club: Bloomington, Indiana. Nunberg, G. 1979. "The Non-uniquenessof SemanticSolutions: Polysemy," Lin guistics and Philosophy3:143-184. Nunberg, G. 1995. "Transfersof Meaning," Journal of Semantics12. Nunberg, G. and A. Zaenen. 19~2. "SystematicPolysemyin Lexicologyand Lexicography," in H. Tommola, K . Varantola, T . Tolonen, and J. Schopp(eds.), Proceedingsof Euralex II , Tampere, Finland, University of Tampere. Nunes, M. 1993. "Argument Linking in English Derived Nominals," in R. van Valin, Jr. (ed.), Advancesin Role and ReferenceGrammar, John Benjamins Publishing, Philadelphia. Quillian, M. R. 1968. "SemanticMemory," in M. Minsky (ed.), SemanticInformation Processing , MIT Press, Cambridge, MA.

283

Bibliography

Ostler,N. and B. T. Atkins.1991.PredictableMeaningShift: SomeLinguistic Propertiesof LexicalImplicationRules, in J. Pustejovskyand S. Bergler (eds.),LexicalSemanticsandKnowledge Representation, ACLSIGWorkshop Proceedings,

76-87.

Ostler,N. and B. T. Atkins.1992.PredictableMeaningShift:SomeLinguistic Propertiesof LexicalImplicationRules, in J. Pustejovskyand S. Bergler

(eds.),LexicalSemanticsand Knowledge Reperesentation, SpringerVerlag,

Berlin.

Parsons,T. 1985.UnderlyingEventsin the LogicalAnalysisof English,in E.

LePoreand B.P. McLaughlin(eds.), Actionsand Events,BasilBlackwell,Oxford, 235-267.

Parsons,T. 1989.The Progressive in English:Events,States,and Processes, Linguistics and Philosophy 12:213-241.

Parsons,T. 1990.Eventsin the Semanticsof English,MIT Press, Cambridge, MA.

Partee,B. 1992.SyntacticCategories andSemantic Type, in M.RosnerandR.

Johnson(eds.),Computational Linguistics andFormalSemantics,Cambridge

University

Press

Partee,B. andM.Rooth.1983.Generalized Conjunction andTypeAmbiguity, in Meaning,Use,andInterpretation ofLanguage, Buerle,Schwarze, andvon Stechow eds., Walter de Gruyter.

Passonneau,R. 1988.A Computational Modelof the Semanticsof Tenseand Aspect, Computational Linguistics, 14.2.

Paul, H. 1880.Prinzipiender Sprachgeschichte, 9th Edition,1975,Niemeyer, Tbingen.

Pelletier,F. J. andL. K. Schubert.1988.Problemsin the Representation ofthe Logical FormofGenerics, Plurals,andMassNouns,in B. LePore(ed), New Directions in Semantics, Academic Press, London.

Pelletier,F. J. andL.K.Schubert. 1989.MassExpressions,inD.GabbayandF.

Guenthner(eds.),Handbook ofPhilosophical Logic,Vol.IV,Reidel,Dordrecht. Perlmutter,D. M. 1970.On the TwoVerbsBegin, in R. Jacobsand P. Rosenbaum (eds.), 107-119.

Perlmutter,D. M. 1978.ImpersonalPassivesandthe Unaccusative Hypothesis, BLS 4, 157-189.

Pinker,5. 1984.Language Learnability andLanguage Development, HarvardUniversity Press, Cambridge, MA.

Pinker,5. 1989.Learnability andCognition: TheAcquisition ofArgument Structure, MIT Press, Cambridge, MA.

Pinker,S.andJ. Mehler.1988.Connections andSymbols, MITPress,Cambridge, MA.

Pollack,M. 1986.A ModelofPlanInference that distinguishes betweenthe Beliefsof Actorsand Observers, in Proceedings24th Annual Conferenceof the Association of Computational Linguistics,New York, 207-214.

Pollard,C. and I. Sag.1987.Information-Based Syntaxand Semantics,CSLI Lecture Notes Number 13, Stanford,

CA.

Pollard,C. andI. Sag.1994.Head-Driven PhraseStructureGrammar,University of Chicago Press and Stanford CSLI, Chicago.

Postal,P. 1970.On the SurfaceVerbRemind, LinguisticInquiry1:37-120.

Postal,P. 1971.Cross-over Phenomena, Holt,Rinehart, andWinston, NewYork. Prince,A. andP. Smolensky 1994.Optimality Theory, ms.JohnsHopkins University and Rutgers University. Pustejovsky, J. 1988.The Geometry of Events,in Studiesin Generative Approachesto Aspect,C. Tenny,ed.,Lexicon ProjectWorking Papers24,MIT, Cambridge,

MA.

284

Bibliography

1991a . "The

Generative Lexicon ," Computational Linguistics ,

Bibliography

285

Roeper , T. 1987 . "ImplicitArguments andtheHead Complement Relation ," Linguistic Inquiry18. Roeper , T. 1993 . "ExplicitSyntax in theLexicon : theRepresentation of Nomi nalizations ," in J. Pustejovsky (ed.), Semantics andtheLexicon , Kluwer Aca demic Publishers , Dordrecht . Rooth , M. 1985 . Association withFocus , Ph.D. Dissertation , University ofMas sachusetts , Amherst . Rudanko , J. 1989 . Complementation andCase Grammar : A Syntactic andSemanticStudyof Selected Patterns of Complementation in Present -DayEnglish , SUNY Press , Albany , NY. Ruhl,C. 1989 . OnMonosemy : A Study in Linguistic Semantics , State University ofNewYorkPress , Albany , NY. Rumelhart , D.E. andJ.L. McClellan . 1986 . Parallel distributed processing : explorations in themicrostructure ofcognition , MITPress , Cambridge , MA. Russell , G., J. Carroll , andS. Warwick . 1990 . "Multiple Default Inheritance in a Unification -Based Lexicon ," in Proceedings of Workshop onInheritance in Natural Language , Tilburg , August . Russell , G., A. Ballim , J. Carroll , andS. Warwick -Armstrong . 1993 . "A Practical Approach to Multiple Default Inheritance forUnification -Based Lexicons ," in T. Briscoe , V. dePaiva , andA. Copestake (eds .), Inheritance , Defaults , and theLexicon , Cambridge University Press , Cambridge . Ryle , G. 1966 . TheConcept oftheMind , Hutchinson , London . Sag , I. A. andC. Pollard . 1991 . "AnIntegrated Theory ofComplement Control ,') Language 67:63-113 . Sag , I. A. andA. Szabolcsi , eds . 1992 . Lexical Matters , CSLILecture Notes 24, Center fortheStudy ofLanguage andInformation , Stanford University , Stan ford,CA. Saint-Dizier , P. 1995 . "Constraint Propagation Techniques forLexical Semantic Descriptions ," in P. Saint -DizierandE. Viegas , eds ., Computational Lexical Semantics , Cambridge University Press . Saint -Dizier , P. andE. Viegas , eds . 1995 . Computational Lexical Semantics , Cam bridge University Press , Cambridge . Saksena , A. 1980 . "TheAffected Agent ," Language 56:812 -826 . Saksena , A. 1982 . "Contact in Causation ," Language 58:820 -831 . Sanfilippo , A. 1990 . Grammatical Relations , Thematic Roles , andVerb Semantics . Ph.D. Dissertation , Centre forCognitive Science , University ofEdinburgh . Sanfilippo , A. 1991a . "Thematic andAspectual Information in VerbSemantics ,)' unpublished ms., University ofCambridge , Cambridge , UK. Sanfilippo , A. 1991b . "Argument Selection andSelection Change : AnIntegrated Approach ," CLS27. Sanfilippo , A. 1993 . "LKBEncoding of Lexical Knowledge ," in T. Briscoe , V. dePaiva , andA. Copestake (eds .), Inheritance , Defaults , andtheLexicon , Cambridge University Press , Cambridge . Sanfilippo , A. K. Benkerimi , andD. Dwehus . 1994 . "VirtualPolysemy ," in Proceedings of COLING -1994 , Nantes , France . Scha , R. 1983 . "Logical Foundations forQuestion Answering ," MS12.331Philips Research Laboratories , Eindhoven , TheNetherlands . Schank , R. C. 1972 . "Conceptual Dependency : A Theory of Natural Language Understanding ," Cognitive Psychology 3:552 -631 . Schank , R.C. 1973 . "Identification ofConceptualizations Underlying Natural Language " in R.C. Schank andK.M. Colby(eds .), Computer Models of Thought andLanguage , W.R. Freeman , SanFrancisco , CA, 187 -247 . Schank , R. 1975 . Conceptual Information Processing , Amsterdam , North -Holland . Searle , J. 1979 . Expression andMeaning , Cambridge , Cambridge University Press .

286

Bibliography

Seuren,P. 1985.DiscourseSemantics, Blackwell Publishers,Oxford.

Shibatani, M.,ed.1976.Japanese Generative Grammar, Academic Press,New York.

Shieber, S.1992.Gonstraint-based grammar formalisms: parsing andtypeinferencefor naturalandcomputer languages, MITPress,Cambridge, MA. Small, S. I., G. W. Cottrell, and M. K. Tanenhaus, eds. 1988. Lexical Ambiguity Resolution,MorganKaufmann,San Mateo, CA. SparckJones,K. 1986.Synonymy andSemanticClassification.. Ph.D.Disserta-

tionwithnewForeword. Edinburgh Information Technology Series(EDITS). Smith, B.andK.Mulligan. 1983.Framework forFormal Ontology,Topoi 2:7385. Edinburgh: Edinburgh University Press.

Smolka,G. 1988.A FeatureLogicwithSubsorts,Wissenschaftliches Zentrum der IBM Deutschland,LILOG-Report33.

Soames, 5. 1989.Presupposition, inD.Gabbay adF. Guenthner (eds.),Handbook of Philosophical LogicVol. IV, 553-616. Sowa,J. 1984.ConceptualStructures,Addison-Wesley, Boston.

Sowa, J. 1993.Lexical Structures andConceptual Structures, inJ. Pustejovsky (ed.), Semanticsand the Lexicon,Kluwer,Dordrecht. Steinberg, D. andL. Jakobovits, eds.1971.Semantics, Cambridge University Press, Cambridge. Strachey, C. 1967.Fundamental Concepts in Programming Languages, Lecture Notesfor International Summer School in Computer Programming, Copenhagen. Talmy, L.1975.Semantics andSyntax ofMotion,inJ. P.Kimball (ed),Syntax and Semantics 4, AcademicPress, New York, 181-238. Talmy,L. 1976.SemanticCausative Types, in M.Shibatani(ed),43-116. Talmy, L.1985.Lexicalization Patterns:Semantic Structure inLexical Forms,in T. Shopen(ed), LanguageTypology andSyntacticDescription 3, Grammatical

Categories andtheLexicon, Cambridge University Press,Cambridge, UK,57149. Tedeschi, P. andA. Zaenen,eds.1981.TenseandAspect,SyntaxandSemantics 14, Academic Press, New York, NY.

Tenny, C.,ed.1988. Studies in Generative Approaches toAspect, Lexicon Project Working Papers24,CenterforCognitive Science, MIT,Cambridge, MA.

Tenny, C. 1992.The Aspectual Interface Hypothesis,in I. A. SagandA. Szabolcsi (eds.), 1-27.

Thomason, R., ed. 1974.FormalPhilosophy: TheCollected Papersof Richard Montague,New Haven,Yale UniversityPress.

Touretzky, D.1986.TheMathematics ofInheritance Systems, Morgan Kaufmann, Los Altos, CA. Travis,L. and E. Williams 1983.Externalization of Arguments in MalayoPolynesianLanguages The LinguisticReview2.

Trier, J. 1931. Der deutsche Wortschatzim Sinnbezirkdes Verstandes:Die

Geschichte einessprachlichen Feldes.BandI. Heidelberg. VanValin,ft. D. Jr. 1987.The Unaccusative Hypothesisvs. LexicalSemantics:

Syntacticvs.Semantic Approaches to VerbClassification, NELS17,641-661.

Van Valin,ft.D.Jr. 1990.Semantic Parameters ofSplitIntransitivity, Language 66:221-260. VanValin,RD. Jr. 1993.A Synopsis ofRoleandReference Grammar,in ft. van Valin, Jr. (ed) Advances in Role and Reference Grammar, JohnBenjamins Publishing, Philadelphia. VanValin,RD. Jr. andD.Wilkins. 1993.Predicting SyntaxfromSemantics, in ft. van Valin,Jr. (ed.)Advances in Roleand Reference Grammar,John

287

Bibliography

Benjamins Publishing , Philadelphia . Van Voorst , J . 1986 . Event Structure . Ph .D . Dissertation Ottawa

, Ontario

, University

of Ottawa ,

.

Van Voorst , J . 1992 . "The Aspectual Semantics of Psychological Verbs ," Linguis tics and Philosophy 15:65- 92 . Vendler , Z . 1967 . Linguistics in Philosophy , Cornell University Press , Ithaca . Vendler , Z . 1984 . " Agency and Causation ," Midwest Studies in Philosophy 9:371 ~ .Q .4 -

Verkuyl , H . 1972 . On the Compositional Nature of the Aspects , Reidel , Dordrecht . Verkuyl , H . . 1973 . " Temporal Prepositions as Quantifiers ," in F . Kiefer and N .

Ruwet (eds.) , 582-615. Verkuyl , H . 1989 . "Aspectual Classes and Aspectual Composition ," Linguistics and Philosophy 12 :39-94 . Verkuyl , H . 1993 . A Theory of Aspectuality , Cambridge University Press , Camr bridge . Verkuyl , H . and J . Zwarts . 1988 . "Time and Space in Conceptual and Logical Semantics : the notion of Path ," Linguistics 30 :483 -511 . Verma ) M . and K . P . Mohanan 1991 . Experiencer Subjects in South Asian Lan guages , CSLI ) University of Chicago Press . Viegas , E . 1995 . "The Semantics - Pragmatics of Lexicalization ," in P . Saint - Dizier and E . Viegas (eds .) , Computational Lexical Semantics , Cambridge University Press

.

Vikner , C . and P . Hansen . 1994 . " Knowledge Bases and Generative Lexicons ," paper presented at Workshop on Lexical Semantics , November 11, University of Copenhagen . Vlach , R . 1981 . "The Semantics of the Progressive ," Syntax and Semantics : Tense and Aspect 14, 271 - 92 . Vossen P . 1991 . "The End of the Chain : Where Does Decomposition of Lexical Knowledge Lead Us Eventually ?," unpublished ms ., University of Amsterdam , Amsterdam Walker

.

, D . E . , and R . A . Amsler

. 1986 . " The

Use of Machine

- Readable

Dictionaries

in Sublanguage Analysis ," in R . Grishman and R . Kittredge , eds ., Analyzing Language in Restricted Domains , Lawrence Erlbaum : Hillsdale , N J . Walker , D ., A . Zampolli , and N . Calzolari , eds . 1994 . Automating the Lexicon , Oxford University Press . Weinreich , U . 1959 . "Travels through Semantic Space ," Word 14:346 - 366 . Weinreich , U . 1963 . " On the Semantic Structure of Language ," in J . Greenberg

(ed.) , Universal of Language, MIT Press, Cambridge , MA . Weinreich , U . 1964 . " Webster 's Third : A Critique of its Semantics ," International Journal of American Linguistics 30 :405 -409 . Weinreich , U . 1972 . Explorations in Semantic Theory , Mouton , The Hague . Wheeler , R . S. 1990 . " Sense and Subsense : The Meanings of the English Verb Understand ," InS., Utah State University . Whorf , B . L . 1956 . Language , Thought , and Reality , MIT Press , Cambridge , MA . Wierzbicka

, A . 1972 . Semantic

Wierzbicka

, A . 1980 . Lingua

demic

Press , New

York

Primitives

, Athenaum

Mentalis : The Semantics

, Frankfurt

of Natural

.

Language , Aca -

, NY .

Wierzbicka , A . 1982 . "Why Can You Have a Drink When You Can 't Have an Eat ?," Language 58 :753- 799 . Wierzbicka , A . 1988 . The Semantics of Grammar , Amsterdam , John Benjamins . Wilensky , R . 1986 . " Some Problems and Proposals for Knowledge Representa tion ," Cognitive Science Report 40 , University of California , Berkeley . Wilensky , R . 1991 . " Extending the Lexicon by Exploiting Subregularities ," Report

UCB / CSD 91/ 618, Computer Science Department , University of California ,

288

Bibliography

Berkeley . Wilks , . 1975a . " A Preferential Pattern Seeking Semantics for Natural Inference ," Artificial Intelligence , 6 :53- 74 . Wilks , . 1975b . " An Intelligent Analyser munications of the ACM , 18 :264- 274 . Wilks , Y . 1978 . "Making

, Y . , D . Fass , C . M . Guo , J . McDonald

Tractable

Machine

for English ," Com -

Preferences More Active ," Artificial

97 .

Wilks

and Understander

Dictionary

, T . Plate

Language

, and

Intelligence B . Slator

a8 a Resource for Computational

10:75-

. 1988 . " A

Semantics ," in

B . Boguraev and T . Briscoe (eds.), Computational Lexicography for Natural Language Wilks

Processing , Harlow , Essex , Longman .

, Y . , D . Fas , C . M . Guo , J . McDonald

, T . Plate

, and B . Slator

. 1993 . " Provid

-

ing Machine Tractable Dictionary Tools," in J. Pustejovsky (ed.) , Semantics and Williams

the Lexicon , Kluwer Academic Publishers , Dordrecht . , E . 1980 . " Predication ," Linguistic Inquiry 11 :203 - 238 .

Williams

, E . 1981 . " Argument

1 : 81 - 114 .

Structure

Williams , E . 1982 . " Another Argument guistic Inquiry 13: 160- 163 .

and Morphology that

," The Linguistic

Passive is Transformational

Review ," Lin -

Williams , E . 1983 . " Against Small Clauses ," Linguistic Inquiry 14:287-308 . Williams , E . 1983 . " Semantic vs . Syntactic Categories ," Linguistics and Philoso phy 6 :423 -446 . Williams , E . 1987 . " Implicit Arguments , the Binding Theory , and Control ," Nat ural Language and Linguistic Theory 5 :151- 180 . Williams , E . 1994 . Thematic Structure in Syntax , MIT Press , Cambridge . Woods

, W . 1975 . " What

' s in a link : foundations

of semantic

networks

," in D . G .

Bobrow and A . Collins (eds.) , Representation and Understanding : Studies in Cognitive

Science , Academic

von Wright , G . 1963 . Norm Paul , London

Press , New York .

and Action : A Logical Inquiry , Routledge

and Kegan

.

Wunderlich , D . 1987 . " An Investigation of Lexical German be-Verbs ," Linguistics 25 , 283 -331 .

Composition

: The

Case of

Wunderlich , D . 1991 . "How Do Prepositional Phrases Fit into Compositional Syntax and Semantics ," Linguistics 29 :591 -621 . Wunderlich , D . 1994 . " Models of Lexical Decomposition ," in E . Weigand and F .

Hundsnurscher (eds.) , Proceedings of the International

Conference on 'Lexi -

cology and Lexical Semantics ', Munster , Tubingen , Niemeyer . Zaenen , A . 1993 . " Unaccusativity in Dutch : The Interface of Syntax

and Lexi -

cal Semantics," in J. Pustejovsky (ed.) , Semantics and the Lexicon, Kluwer , Dordrecht

Zubizarreta

.

, M . L . 1987 . Levels of Representation

Faris , Dordrecht

Zucchi , A . 1993 . The Language Publishers

in the Lexicon and in the Syntax ,

.

, Dordrecht

.

of Propositions

and Events , Kluwer

Academic

Author

Index

A Abraham, W ., 188 A busch, D., 259 Ackerman, F., 259 Allen, J., 69, 183, 246 Alsina, A., 249, 259 Anderson, M., 263 Anick

,

P

. ,

Apresjan

55

,

Aristotle

,

244

,

Asher

,

,

Atkins

.

. ,

,

B

.

,

247

,

234

,

. ,

91

,

20

,

T

,

242

30

233

62

,

17

246

N

231

,

J

,

76

,

248

,

241

121

5

,

,

,

199

,

77

,

247

244

47

167

,

,

,

,

48

85

176

,

247

,

,

,

122

97

,

243

,

177

,

184

,

215

,

216

,

D

248

,

241

B Bach

Baker

,

E

,

. ,

C

15

. L

,

. ,

16

,

184

,

250

,

264

50

. Ballim, A., 255 Beierle, C., 114 Belletti, A., 213 Benthem, J. van, 57, 58, 69, 92, 184, 245, 246, 255 Bergler, S., 241 Bes, G., 247 Bierwisch, M., 18, 152, 242, 244, 266 Boguraev, B., 5, 44, 51, 90, 91, 116, 141, 248, 255, 264 Bookman, L., 243 Bouillon, P., 116, 199, 200, 201, 207, 248, 258, 260 Brachman, R. J., 24, 58 Bresnan, J., 10, 63, 249, 254 Briscoe, T ., 5, 78, 90, 225, 233, 247, 255, 264 Brugman, C., 268 B uitelaar) P., 248 Burge, T ., 75 Busa) F., 68, 74, 187, 188, 189, 195, 196, 197, 198, 266 Baker

,

M

. ,

245

,

Chomsky, N., 8, 36, 40, 55, 63, 103, 108, 117, 118, 132, 135, 137, 181, 184, 194, 207, 213, 242, 245, 249, 260 Cohen, P.R., 24, 241 Collins, A., 58 Comrie, B., 159, 246, 259 Copestake , A., 78, 81, 90, 94, 114, 141, 189, 224, 233, 247, 248, 264 Cottrell, G. W , 36 Croft, W ., 209 Crouch, R., 246 Cruse) D. A., 23, 24

259

c Cardelli, L., 58, 111, 245 Carlson, G., 15, 225 Carnap, R., 58 Carpenter, B., 81, 94, 145, 244 Carroll, J., 255 Cattell, R., 258 Charniak, E., 7 Chierchia, G., 106, 189, 195, 198, 249

.. Dahl, 0 ., 246 Davidson, D., 67, 165 Descles , J.-P., 246, 258 Di Scuillo, A.-M., 253, 260 Diesing, M., 226 Dixon, R. M. W., 19, 22, 23, 135, 241 Dolling, J., 264 Dor, D., 50 Dowty, D. R., 13, 14, 16, 25, 54, 56, 71, 107, 108, 109, 110, 174, 201, 207, 228, 242, 245, 249, 251, 259, 260

E Eberle,K., 183 Elliott, D., 52, 244 En<;, M., 159 Eschenbach , C., 18, 152 F

Faltz, L., 124 Farkas , D., 207 Fags , D., 268 Fauconnier , G.,, 234 Fellbaum, C., 7, 267 Fillmore, C., 6, 9, 10, 11, 215, 216, 241

Fisher, C., 63 Fodor, J., 25, 58, 243 Freed , A., 260 G

Gazdar

,

Geach

G

)

leitman

Gleitman

Go

P

,

. To

107

,

,

H

,

L

,

106

0 '

. ,

63

63

249

290

Author Index

Godard

, D . , 199

Goldberg , A ., 215 , 241 Goldman

, A ., 7

Grimshaw

, J . , 41 , 50 , 52 , 63 , 68 , 87 ,

137 , 165 , 173 , 196 , 213 , 249 , 257 , 262 Grinder

, J . , 261

Groenendijk Grosz

, J ., 107 , 138 , 255 , 266

, B . , 241

Gruber

Kiefer

, F . , 264

Kiparsky , C ., 25 , 50 , 254 Kiparsky , P ., 25 , 50 , 254 Klein

, E ., 57 , 58 , 106 , 107 , 245 , 249 ,

250 , 255 , 260 Kowalski

, R . , 183

Kratzer

, A . , 15

Krifka

, M . , 15 , 16 , 18 , 162 , 248 , 252

Kunze

, J ., 9

, J . S . , 6 , 261

Guentcheva Gunter

, Z . , 246 , 258

L

, C . , 92 , 114 , 248

Lakoff H

, G . , 9 , 54 , 58 , 135 , 268

Lambek

, J . , 92

Larniroy , B ., 201 Hacking , I ., 242

Landman

Hale

Lang , E ., 233 , 266

, K ., 9

Hansen

, P . , 99 , Z . , 8 , 252

Lascarides 247

, I . , 50 , 241

Lebeaux

, D ., 258

Lecomte

, A . , 247

Harris Heim

Hendler

, J . , 243

Hendriks

, H . , 249 , 264

Herskovits

, A . , 47

Herzog , 0 ., 233 Higginbotham , J ., 165 , 246 Hinrichs Hirst

Hobbs

, J . , 7 , 19 , 24 , 121 , 184 , 241 ,

, A . , 30 , 121 , 231 , 233 , 234 ,

LeiB

, H . , 106 , 248

Levi

, J . N . , 264

Levin

, B . , 2 , 6 , 8 , 10 , 47 , 54 , 56 , 64 ,

122 , 125 , 188 , 189 , 201 , 241 , 245 , 252

, E . , 80

, G . , 29 , 30 , 34 , 243

Li , J . , 232 , 266 Link

, G . , 16 , 17 , 162 , 264

Lyons , J ., 23

246 , 266 Horn

, F . , 246

, L . , 263 M

Hovy , E ., 241 J

Marantz

, A . P . , 245

Martin

Jackendoff

, R . , 25 , 80 , 126 , 234 , 235 ,

244 , 246 , 252 Jacobson

. P . . 203 . 206 . 260

McCawley , J ., 109 McConnell Mchombo

Jayez , J ., 199

Mel

Johnson

, M . , 54

Miller

Johnson

- Laird

Johnston 265

, P ., 6

, M . , 131 , 246 , 248 , 259 , 264 ,

, J . , 268

' cuk

Kamp , H ., 69 , 159 , 160 , 184 , 233 , 241 , 246

, S . , 246

, I . A . , 247

, G . , 6 , 7 , 23 , 24 , 243 , 247 , 267

Mineur

, A .- M . , 248

Modrak

, D . , 242

Moens

, M . , 68 , 71

Mohanan

K

- Ginet , S . , 249

, K . P . , 244

Montague , R ., 8, 56 Moortgat , M ., 92 , 258 Moravcsik

,

J . M .,

76 ,

85 ,

230 ,

247 , 248 Kanerva

, J . , 249

Karttunen Katz Kautz

, L . , 25

, J . , 25 , 43 , 56 , 58 , 242 , 243 , 244

Morgan , J ., 24 , 241 Morrill Moshi

, G . , 153 , 154 , 244 , 248 , L . , 249

, H . , 24

Keenan , E . , 24 , 124 Kegl , J . , 47

Kenny , A ., 13 Keyser , S. J ., 9

Mourelatos

, A . P . D . , 13 , 246

Mulligan , K ., 247

244 ,

Author Index

291

N

Saksena, A., 219, 259, 263 Sanfihippo, A., 81, 189, 246

Nirenburg, 5., 246

Scha, R., 6

Nunberg, G., 54, 57, 231, 234, 235, 236, 268

Schank, R. C., 58, 241 Schmolze, J., 24

Nunes, M., 258

Schreuder, R., 242 Schubert,

0

L. K., 8, 17

Searle, J., 56 Sergot,

M., 183

Ostler, N. D. M., 48

Seuren, P., 25

P

Shieber, S., 106, 248 Sidner, C., 241

Parsons,T., 67, 165,184,245 Partee,B., 106,107,111,249

Small,S. I., 36 Smith, B., 247 Soames, 5., 243

Pelletier,

F. J., 8, 17

Perimutter,D. M.,201,202,203 Pesetsky, D., 137, 213 Pinker, S., 63, 138, 243 Pollack, J., 29, 243 Pollack, M., 24, 241

Pollard, C., 243, 246, 249, 251 Postal,

P., 253, 256

Prince,

A., 243

Procter, P., 241

Sowa,J., 268

Sparck Jones, K., 242 Steedman, M., 68, 71 Stokhof, M., 107, 138, 255, 266 Strachey, C., 55, 106 Strawson,

24

T

Talmy, L., 72, 80, 125

Pulman, S., 246

Tanenhaus, M. K., 36

Pustejovsky,J., 39, 44, 55, 62, 68, 69, 71, 78, 81, 90, 104, 105, 122, 123, 186, 187, 188, 189, 195, 207 208,

Thompson,S., 259 Touretzky,D., 255 Travis,L., 258

213, 250, 263 Q

V

Quillian, M., 58

Van Valin, R., 188, 244, 254 Vendler, Z., 3, 13, 43, 243, 244

R

Rapoport,

Turner, R., 249

Verkuyl, H., 8, 13, 15, 16, 80, 157, T., 125, 188, 252

Rappaport,M.,6, 56,122,189,245 Reichenbach,,

184

Reyle,U.,159,160,184,233,241

Rizzi, L., 213 Roeper, T., 195 Rollinger, C.-R., 233

159, 165, 213, 252, 264

Verma,M.,244 Viegas, E., 241, 248

Vikner, C.,99 Vikner, 5., 41 w

Rooth, M., 106, 111

Waltz, D., 29, 243

Rosen, S., 87, 253

Warwick-Armstrong, 5., 255

Rof3deutscher,A., 184

Wegner, P., 58, 111, 245

Rudanko, J., 260 Russell, G., 255

Weinreich, U., 1, 27, 28, 241, 243, 253 Wheeler, R., 244

Ryle, G., 13

Wierzbicka, A., 19, 135

S

Wilensky, R., 241 Wilkins, D., 244, 254

Sag, I., 106, 107, 243, 246, 249, 250,

Wilks, Y., 19, 56, 58, 183, 241 Williams, E., 63, 258, 260, 262

251, 260 Saint-Dizier,

P., 241

Woods,

B., 8

292

Wunderlich, D., 65, 245

z Zaenen, A ., 188, 201 Zampolli, A ., 5, 241 Zwarts, J., 80

Author Index

Subject Index

A

true

, 63 , 64 , 66

Aristotle

abduction, 121, 236, 237 adjectives, 20 attributive , 20

, 21 structure

of , 49

classification

artifactual aspect

of , 22

durative

, 74

frame

adverbials

state

aspectual

classes

aspectual

shifts

progressive

, 13

knowledge

bare

plural

bare

objects

singular

, 232

selective

, 3 , 61 ,

Binding

, 97

in

, 76 , 247 , 3 , 12 , 68

, 1 , 23

arguments

a- adjunct , 196 , 190

, 245

goal , 12 , 66 direct

case

, 221 modality

, 216

roles

, 63 , 245

, 63 , 65 , 66 , 245

form

, 224

( csf ) ~ 132 ~ 134 ,

Categorial

Grammar

, 92

categorical

syllogism

, 237

causal

inference formal

causal

,

semantics

reasoning

in

AI

in

computational

causal

linguistics

relations

headless

modal

, 187

of , 218

, 215 - headed

events

, 187

of , 183

causation

, 9

constitutive

, 197

, 186 , 210 , 219

experienced indirect

, 187

events

taxonomy

direct

, 183

, 183 , 184

events

- headed

right

, 184

, 183

, 183

lexicalization

, 63 , 64 , 65 , 66 , 91 , 245

indirect

, 200

with

syntactic

left

argument , 63

verb

canonical 199

in

argument coherence , 185 , 209 , 220 of affected object , 220 of experiencer verbs , 211 argument selection , 3 argument structure , 3 , 58 , 61 , 62

default external

environment

, 1, 28 , 29 , 30 , 31 , 36 ,

anaphora , 18 anaphoric binding , 18 of subject , 213

internal

coercion

c

contrastive , 1 , 27 , 28 , 29 , 30 , 56 lexical , 27 , 55

default

, 108

as causative

39

antonymy

128 ,

, 130

Theory

representation

classes , 10

ambiguity complementary

127 ,

break

, 119

alternation

105 ,

book

for verbs , 186 Aktionsarten

86 ,

131 , 134 , 253 , 254 subselective

nouns

, 56 , 14

binding

227

shadow

, 16 , 15

, 16

background

for adjectives (Stage-Level Predicates) ,

aliases

, 225 , 226

, 68

B

point adverbials , 14 temporal adverbials , 13, 109 agentive nominals , 88 agentive role

aitiae

, 243

, 226

adverbs , temporal

-for

Ethics

Organon , 246 Rhetoric , 236 artifacts , 97

tough - movement , 21 , 23 adjuncts , 64 , 65 causative

, 85

, 77

Nicomachean

subject -control , 21 taxonomic

factors

Metaphysics

, 21

relational relational

, 243

generative

polysemous , 49 , 127 predicative , 20 unary

, 17 , 20 , 243

Categories

, 210 , 219 ( constitutive

) , 186 , 219

294

SubjectIndex

causation

(

continued

)

lexicalization

of

modes

of

,

causative

,

184

,

186

220

197

constructions

modal

causatives

qualia

modal

causative

,

,

,

68

215

causatives

verbs

.

,

64

causativejinchoative

alternation

,

64

,

188

co

-

composition

,

122

,

role

in

,

,

coercion

,

,

,

230

entailment , 23, 24 enthymeme , 237 enthymemic inference , 236 abduction , 237 deduction , 237

121

,

knowledge

202

,

,

commonsense

commonsense

,

4

metaphysics

complement

drop

type

,

Equi - NP sentences , 107

19

event

242

,

event

60

9

relation

Construction

,

context

,

,

containee

56

,

contextual

215

,

,

37

,

30

intransitive

control

pseudo

-

control

,

32

constructions

,

data

,

,

cospecification

,

count

nouns

,

8

,

207

123

,

subevents

, 16 , 68

precedence in , 72

tree

structure

exhaustive

42

17

,

18

over , 71 in , 72

temporal event

206

40

, 220

ordering restriction relative prominence

241

control

corpus

, 3 , 58 , 61

head , 69 head of . 72

216

alternations

priming

, 73

focusing mechanisms , 16

75

Grammar

/

structure

coherence within extended , 67

constitutive

container

, 16 , 72

in DCP , 183

1

59

conative

headedness

head configurations ,

,

11

118

compositionality

59

,

,

,

induction , 237 lexical , 238

238

knowledge

,

,

63

syllogism

,

105

238

verbs

common

strong

,

223

,

,

,

weak

86

E

aspectual

Coherence

complex

,

238

,

discourse

in

80

236

coercion

in

rule

61

connectives

minimal

with

,

3

temporal

role

3

unification

predication

in

,

,

qualia

-

2

242

discourse

with

co

134

disambiguation , 29, 30 connectionist approach, 243 of contrastive senses, 234 pragmatically constrained , 29 sense enumeration lexicons, 36 sortally constrained , 30 discourse, 7 ellipsis in persuasive discourse, 236 inference, 236 lexical inference, 236 persuasive, 236 discourse attachment , 234 displaced reference, 234

, 69

ordered overlap , 70

exhaustive

ordered

exhaustive

overlap part of , 69

part

of , 69 , 71

events count

/

mass

covering

alternations

,

,

37

accomplishments

193

activities D

culmination

Davidsonian

lexical

decomposition

ded

uction

,

Default

183

,

187

default

,

233

, 225

processes Paradigm

195

,

201

,

,

,

67

237

reasoning

defeasibility

,

88

,

247

121

(

DCP

of ,

logical , 13, 242 temporal , 242

57

Causative

default

DICE

representation

,

, 14

, 11 , 12 , 13 , 14 , 16

)

,

states telic

, 68

, 12 , 15 , 16 , 68 , 15 , 74

transitions

, 11 , 68

accomplishments achievements

, 12- 14, 16 , 73

, 12 , 14 , 16 , 73

295

Subject Index

existential closure, 65, 82 experiencer predicates argument coherence of, 211 aspectual properties of, 212 binding violations of , 213 c-command violations , 214 temporal restriction in , 211 experiencer verbs, 209 F factivity , 25, 50, 51 figure / ground alternation , 37 force-complement constructions , 202 formal role disjunctive statement in , 218 for verbs, 186 function application with coercion (FAC ) , 111 function composition (FC ) , 206 Functional Completeness, 63

G generalized quantifiers , 107, 118, 252 genus term , 34, 243

grammar context-free, 55 families of grammars , 57 transformational , 55 grammatical alternations , 12 grammaticality degrees of, 196

H head of event structure, 72 hearer, 41 homonymy, 1 hyponymy, 23, 24

I ideal meaning , 47 imperfective paradox ) 16 inchoative verbs, 64 indefinite NP deletion , 10, 242 individual -level predicates , 15, 20, 79, 225, 242 individual -level nominals (ILNs ) , 229 induction , 237 inference. 7. 19, 85, 87, 236 and rhetoric , 236 causal, 184

in discourse, 236 lexical , 234, 236 inferencing techniques, primitives -based, 19 inferential structures , 236 inheritance networks , 24 inheritance structure , 3 interrogative complements , 137 extensional , 138 intensional , 138 Italian , 188 K knowledge commonsense, 232 lexical , 5, 232

L language classes monomorphic , 2, 55, 56 unrestricted polymorphic , 2, 56 weakly-polymorphic , 56, 57 context -free, 55 finite -state descriptions , 55 lexical conceptual paradigm (lcp) , 62, 91, 92, 93, 118, 119, 133 lcp type constructor , 118 lexical decomposition , 6, 24 lexical disambiguation heuristics , 233 lexical inheritance structure , 58, 61 lexical insertion , 19 lexical processing, 234 lexical redundancy rules, 55 Lexical Representation Language (LRL ) , 233 lexical rules, 265 lexical selection, 27 lexical semantics and discourse inference, 233, 236 logical polysemy, 28, 29, 31, 32, 56 container / containee alternation , 31 control / raising , 201 count / mass alternation , 31 figure / ground alternation , 31 mass/ count alternation , 8 material / product alternation , 64 of verbs, 188 place/ people alternation , 31 process/ result alternation , 31 product / producer alternation , 31

296

Subject Index

M

polysemy

manner specification , 11 mapping principles , 190 mass nouns, 8, 14, 17, 18 meaning postulates , 55, 56, 108, 110 meronymy , 23, 24 meta-entry , 62, 91 metaphor , 4 metonymic reconstruction ., 200, 217 of subject , 209 of object , 119 metonymy , 1, 4, 54 logical metonymy , 54 modality within qualia structure , 222 modes of explanation , 76, 85, 86 modes of meaning , 244 Montague Grammar , 110

0 object drop, 11 open as causativeverb~ , 222 representation with modality , 223 Orthogonal Parameter Binding , 68 p parametric polymorphism :I 106, 115 part -of relation , 87 passive, 196 passivization , 196 phrasal paradigm , 199 polymorphism , 3

1 ,

10 , 27

, 49

verbal

, 127

, 73

pragmatic

priming

predicate

, 30

transfer

effects

on

, 235

reflexivization

, 235

prepositions functional

, 231

presupposition

, 23

primitives

as

, 24

, 25

, 58

privative

state

, 217

coercion

PRO

, 218

operator

, 21

interpretation

PRO

- form

,

infinitivals

,

136

, 16 , 20

projection

from

Projection

Principle

, 63

relation

, 206

in

7

136

progressive

pseudo

N natural kinds, 97 natural languageprocessing ,5 natural languageunderstanding, 7 newspaper , logical polysemy, nouns, 16 abstract, 19 concrete, 19 dollble figure-ground, 91 group denoting, 17, 18 individual denoting, 17, 18 predicative, 18 process/result, 93, 96 relational, 18 horizontal, 18 vertical, 18

,

adjectival

, 242

qualia

,

131

- agent control

psychological

predicates

,

210

, 213

Q

qualia

saturation

qualia

structure

agentive

,

role

role role

telic

role

, 76

, 85

, 86

telic

purpose qualia

, 76

, 76

direct

, 75

, 247

, 97 , 85

, 98

, 95

, 99

, 99 telic

, 99

unification

question

, 61

, 76 , 86

constitutive formal

193

, 3 , 58

,

,

124

- answering

100 , 227

systems

, 51

R

rationale control

clause of ,

reasonIng

195

,

syllogistic

, 238

referential

transfer

, 235

reflexives in

experiencer

constructions

resultative emphatic pure

, 213

constructions

,

resultatives resultatives

15 , 245

, 245 , 245

s

SDRTt

233

selectional

classes

selectional

restrictions

, 8 , 19

interpretation

semantic

transformations

, 86

semantic

well

, 6

- formedness

,

, 44

semantic

7

Subject Index

297

semanticality , 40 , 41 , 42 degrees of , 196

recursive, 92 typing judgments, 92 rule of abstraction, 92 rule of appJication, 92 rule of projection, 92

sense

complementary , 38 contrastive fixed , 46

, 38

generators

u

, 46

sense enumeration

lexicons

, 29 , 34 , 35 ,

37 , 38 , 39 , 42 , 45 , 46

sense

extension

sense

in context

242

, 46

sense splitting

, 60 , 78 , 86 , 98 , 110 ,

, 248

shadowing , 191 sortal crossing , 235 , 236 sortal specification , 19 speaker , 41 and

rhetorical

devices

, 237

split intransitivity , 188 stage - level predicates , 15 , 20 , 79 , 225 , 226 , 242

stage-level nominals (SLNs) , 229

unaccusat ives , 68 , 189 grammaticalization of , 192 , 193 lexical representation of , 189 non -lexical , 196 , 197 unaccusativity , 188 derived , 197 underspecification , 87 , 2.24 lexical , 188 of event structure , 189 semantic , 73 , 188 with co-composition , 224 with verbal polysemy , 188 v

psych -st age-level , 226 stative

predicates

, 225

Structuralism , 55 Super Equi - NP Deletion , 261 synonymy

, 1 , 23

T temporal text

connectives

understanding

thematic

roles

there

- insertion

theta

criterion

type type

, 230 , 7

, 6 , 99 , 242 , 63 , 64

cluster

, 93

coercion , 2 , 57 , 58 , 59 , 61 , 86 , 105 , 107 , 134

type

hierarchy

type

ladder

type

latt

, 81 , 90 , 106 , 249

ice , 61

multiple

types

single

type

type

pumping

type

shifting

, 115

, 115 , 119 , 204 , 16 , 106 , 107

functions

, 249

operators

, 108 , 111

type

system

typed types complex deep

, 2 , 92

feature

structures

, 81 , 90

, 55 type

dotted inference

, 112

type

, 92 , 94 , 95 , 100

, 248

verb classes , 2, 6 verbal alternations , 8 , 10, 33 inchoative / causative , 33 verbs active process , 80 argument inversion , 74 aspect ual , 198 in control structures , 200 in raising constructions , 200 causatives , 53 , 69 , 74 , 168 aspectual , 198 constitutive , 103 grammaticalization of , 192 modal , 215 , 218 , 224 non - lexical , 183 qualia modal , 224 change of state , 186 contact , 241 control , 202 counterfactives , 26 DCP , 209 experiencer , 209 , 211 experiencers , 53 factives , 25 intransitives , 9 motion , 80 , 241 of creation , 13 , 186 passive process , 80 performance verbs , 13 raising , 202 raising / control alternation right -headed , 186 , 189

, 74

298

Subject Index

verbs (continued ) risk - verbs , 4 unaccusative unaccusatives

, 74 , 69

unergative , 197 unergatives , 189 VP -ellipsis with

control

verbs

, 203

w weakly polymorphic languages, 2 word meaning , primitive -b85ed theories , 57 relational -based theories , 57

z zeugma as test for ambiguity, 235