Manifestations of Genericity

OUTSTANDING DISSERTATIONS IN LINGUISTICS Edited by Laurence Horn Yale University A ROUTLEDGE SERIES OTHER BOOKS IN ...

Author: Yael Greenberg

16 downloads 895 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

OUTSTANDING DISSERTATIONS IN LINGUISTICS

Edited by Laurence Horn Yale University

A ROUTLEDGE SERIES

OTHER BOOKS IN THIS SERIES: AUDITORY REPRESENTATIONS IN PHONOLOGY Edward S.Flemming THE SYNCHRONIC AND DIACHRONIC PHONOLOGY OF EJECTIVES Paul D.Fallon THE TYPOLOGY OF PARTS OF SPEECH SYSTEMS The Markedness of Adjectives David Beck THE EFFECTS OF PROSODY ON ARTICULATION IN ENGLISH Taehong Cho PARALLELISM AND PROSODY IN THE PROCESSING OF ELLIPSIS SENTENCES Katy Carlson PRODUCTION, PERCEPTION, AND EMERGENT PHONOTACTIC PATTERS A Case of Contrastive Palatalization Alexei Kochetov RADDOPPIAMENTO SINTATTICO IN ITALIAN A Synchronic and Diachronic Cross-Dialectical Study Doris Borrelli PRESUPPOSITION AND DISCOURSE FUNCTIONS OF THE JAPANESE PARTICLE MO Sachiko Shudo

THE SYNTAX OF POSSESSION IN JAPANESE Takae Tsujioka COMPENSATORY LENGTHENING Phonetics, Phonology, Diachrony Darya Kavitskaya THE EFFECTS OF DURATION AND SONORITY ON CONTOUR TONE DISTRIBUTION A Typological Survey and Formal Analysis Jie Zhang EXISTENTIAL FAITHFULNESS A Study of Reduplicative TETU, Feature Movement, and Dissimilation Caro Struijke PRONOUNS AND WORD ORDER IN OLD ENGLISH With Particular Reference to the Indefinite Pronoun Man Linda van Bergen ELLIPSIS AND WA-MARKING IN JAPANESE CONVERSATION John Fry WORKING MEMORY IN SENTENCE COMPREHENSION Processing Hindi Center Embeddings Shravan Vasishth

INPUT-BASED PHONOLOGICAL ACQUISITION Tania S.Zamuner VIETNAMESE TONE A New Analysis Andrea Hoa Pham ORIGINS OF PREDICATES Evidence from Plains Cree Tomio Hirose CAUSES AND CONSEQUENCES OF WORD STRUCTURE Jennifer Hay INTERACTIONS BETWEEN MARKEDNESS AND FAITHFULNESS CONSTRAINTS IN VOWEL SYSTEMS Viola Miglio THE PHONETICS AND PHONOLOGY OF GUTTURALS A Case Study from Jul’hoansi Amanda Miller-Ockhuizen TRUE TO FORM Rising and Falling Declaratives as Questions in English Christine Gunlogson

MANIFESTATIONS OF GENERICITY

by

Yael Greenberg

Routledge New York & London

Published in 2003 by Routledge 29 West 35th Street New York, NY 10001 Published in Great Britain by Routledge 11 New Fetter Lane London EC4P 4EE Copyright © 2003 by Taylor & Francis Books, Inc. Routledge is an imprint of the Taylor & Group This edition published in the Taylor & Francis e-Library, 2005. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” All rights reserved. No part of this book may be reprinted or reproduced or utilized in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system without permission in writing from the publisher.

Library of Congress Cataloging-in-Publication Data Greenberg, Yael. Manifestations of genericity/by Yael Greenberg. p. cm. — (Outstanding dissertations in linguistics) Thesis (Ph.D) — Bar-Ilan University, 2002 ISBN 0-415-96777-5 (Print Edition) (Hardcover : alk. paper) 1. Grammar, Comparative and general—Sentences. 2. Grammar, Comparative and gener al—Syntax. 3. Genericalness (Linguistics) 4. Semantics. 5. Pragmatics. I. Title. II. Series. P295.G67 2003 415—dc21 2003009171

ISBN 0-203-01019-1 Master e-book ISBN ISBN 0-415-96777-5 (Print Edition)

To my son—Ilai—in the hope of a flourishing linguistic future.

And to my whole family, with love.

If thou (dear reader) art bored with this wearisome method of calculation, take pity on me who had to go through with at least seventy repetitions of it, at a very great loss of time; nor wilst be surprised that by now the fifth year is nearly past since I took on Mars. Johannes Kepler, A New Astronomy, 1609 (Translated by Arthur Koestler, The Sleepwalkers, 1959)

Contents

CHAPTER 1

PREFACE

xxii

ACKNOWLEDGMENTS

xxv

COMPARING SENTENCES WITH GENERICALLY INTERPRETED INDEFINITE SINGULAR (IS) AND BARE PLURAL (BP) SUBJECTS: THE TRADITIONAL APPROACH AND ITS PROBLEMS

1

1.

Introduction

1

2.

A brief review of central genericity phenomena and analyses

4

2.1

Genericity in English

4

2.2

The I (characteristic)/D (kind) genericity distinction

5

2.3 2.4 2.5 2.5.1

2.5.2

Relevance of the characteristic/kind distinction to the IS/BP distinction

7

The Basic Semantic Structure of Characteristic (I-) Generics

9

The semantics of the generic operator—the modal approach

12

Tolerance of exceptions, law likeness, and counterfactual support of characteristic generics

12

Kratzer’s (1981) analysis of nongeneric modal verbs and its application to the semantics of Gen

15

x

2.6 3. 3.1 3.2.

Summary

18

Unexpected differences between minimal pairs of IS and BP sentences

18

Differences between IS and BP sentences in temporally modified sentences

19

Differences in the level of “law-likeness” and type of rules expressed by IS and BP sentences

24

3.2.1

The original intuition

24

3.2.2

Problems with the original intuition

26

3.3

3.4

3.5

3.5.1 3.5.2 3.6 4.

5. 5.1

Unexpected differences between minimal pairs of IS and BP sentences with subjects expressing “extremely unnatural classes”

28

Unexpected differences between minimal pairs of IS and BP sentences with VPs expressing “Extremely Unconnected Properties” or extremely unconnected properties

32

The different effect of the presence/absence of contextual support on the felicity of BP and IS sentences

33

“Out of the blue” vs. contextually supported IS and BP sentences

33

Contextual support does not always help: IS sentences in pure inductive scenarios

34

IS and BP sentences which do not differ so much

36

Parallels between the IS/BP distinction in temporally restricted and classic generic constructions

38

The basic intuition

40

Carlson’s (1995) distinction between the “inductive” and the “rules and regulations” approaches to (I-) genericity

40

xi

5.2

The intuitive underlying difference between IS and BP sentences

43

An outline of this work

45

“IN VIRTUE OF” GENERALIZATIONS, AND THE SEMANTICS AND PRAGMATICS OF IS SENTENCES

47

Introduction

47

1.1

Basic Characteristics of IS Sentences

47

1.2

The basic intuition

50

1.3

The Direction of Formalization: “in virtue of” properties restricting the accessibility relation of IS sentences

51

The pragmatic aspects in the semantics of IS sentences and their formalization

54

Ability of previous theories to account for the data

56

2.1

Krifka (1987): modality vs. nonmodality

56

2.2

Unspecified modal base of I- generics—Krifka et al. (1995) and Krifka (1995)

57

Cohen (2001): IS sentences as expressing rules

59

Chierchia’s (1995) and Brennan’s (1993) theories: a formalization of the “in virtue of” intuition

62

Chierchia (1995): “felicity conditions” restricting the accessibility relation

63

Brennan (1993): an accessibility relation “keyed to the properties of the subject”

66

6. CHAPTER 2

1.

1.4 2.

2.3 3.

3.1 3.2 3.2.1

Data concerning root modals

67

3.2.2

Brennan’s intuition

67

3.2.3

The formalization

68

xii

4.

Application of Brennan’s ideas to the requirements on the accessibility relation of IS sentences

70

4.1

Basic data and intuition

70

4.2

“Accessible w.r.t. a property”: subset, and not membership relationship

71

The role of the world of evaluation in choosing the possible “in virtue of” properties

72

4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.4

4.5 5. 5.1 5.2

5.3 5.4

Contextual restrictions on the choice of the “in virtue of” property

72

Brennan’s suggestion for restricting the set of “in virtue of” properties

75

An intuitive characterization of “associated properties”

76

Formal characterization of “associated properties”

77

Integrating the definition of “associated properties” into the truth conditions of IS sentences

78

The problem of exceptions

80

Advantages

83

Normative, legal, epistemic, etc., “flavors” of IS sentences

83

Comparing IS sentences and nongeneric universal sentences with overt deontic modal verbs

85

High level of law-likeness: the “analytic” flavor of IS sentences

88

IS sentences with subjects denoting “extremely unnatural classes”

91

5.4.1

Basic infelicity

91

5.4.2

Contextual support for infelicitous IS sentences

94

xiii

5.4.3

Inductive contexts

96

5.4.4

“Linguistically associated properties” as improving the felicity of IS sentences

97

An additional contextual requirement on IS sentences

99

The problem: “association” of ^S with ^P is not enough

99

The intuition: A second contextual requirement on the ^S property

101

Formalizing the intuition

103

6. 6.1 6.2 6.3 6.3.1

The extensional disjunction

103

6.3.2

Problems with the “extensional disjunction”

105

6.3.3

A “Good possibility” operator over the disjunction

107

6.3.3.1

The intuition

107

6.3.3.2

Defining the right kind of “good possibility”

108

6.4

Further motivations for the second presuppositional requirement on the ^S property

111

An apparently better alternative: limiting ^Q and not ^S

111

6.4.2

Triggering of the presupposition

113

6.4.3

A more efficient evaluation process

118

6.4.1

7. CHAPTER 3

1. 1.1

Conclusion and summary

120

“DESCRIPTIVE” GENERALIZATIONS, AND THE SEMANTICS AND PRAGMATICS OF DESCRIPTIVE BP SENTENCES

125

Introduction

125

Basic properties of BP, as opposed to IS sentences

126

xiv

1.2

Descriptive BP sentences: the intuition and the direction of formalization

128

1.2.1

The intuition

128

1.2.2

Direction of formalization

129

1.2.3

Exceptions to descriptive BP sentences

130

1.2.4

Basic modality/intensionality of descriptive BP sentences

130

The nature of modality/intensionality of descripti ve BP sentences

131

Basic semantic structure, and structure of this chapter

132

1.2.5 1.2.6 2. 2.1

A closer look at descriptive BP sentences: entailments and predictions

134

Truth of the universal statement in the “similar worlds”

134

2.1.1

The data and the intuition

134

2.1.2

The formalization: a Lewis-style definition of maximally similar worlds

137

Truth of the universal statement in the “normal futures”

139

2.2 2.2.1

The data and the intuition

139

2.2.2

The formalization: evaluation in worldinterval pairs

141

The problem of natural, expected, changes, and its solution

144

Truth of the universal statement in the actual world: present and past

147

A unified definition of relevant circumstances and unified truth conditions for descriptive BP sentences

149

Accounting for the basic properties of descriptive BP sentences

153

2.2.3 2.3 2.4

3.

xv

3.1

Descriptive BP sentences as freely expressing “unreasonable generalizations”

154

3.2

Unified flavor of descriptive BP sentences

155

3.3

Descriptive BP sentences as expressing a low degree of law-likeness

156

Advantages and an apparent problem with the underspecification of the superinterval I′

157

4.1

Advantages

157

4.2

An apparent problem

160

The “existence” presupposition of Descriptive BP sentences

161

5.1

The existence requirement: The basic data

161

5.2

The existence requirement is a presupposition

165

5.3

Triggering the presupposition

166

4.

5.

5.3.1 5.3.2

The conventionally triggered “existence presupposition”

167

A mismatch (between the conventionally triggered and the “real” existence presupposition), w.r.t. the relevant interval involved, and its resolution

168

5.3.2.1

The mismatch

168

5.3.2.2

A resolution of the mismatch: triggering existence at both edges of (instead of at) the superinterval I′

171

A mismatch (between the conventionally triggered and the “real” existence presupposition), concerning the relevant world involved, and its resolution

173

5.3.3

5.3.3.1

The mismatch

173

5.3.3.2

A resolution of the mismatch: triggering presupposed existence in the actual world

174

5.4

Existence presuppositions of IS sentences

177

xvi

5.4.1 5.4.2 5.4.3 5.4.4 5.4.4.1

5.4.4.2

6.

Existence presuppositions of IS sentences— basic data

177

What triggers existence in the actual world with epistemic IS sentences

180

The existence presupposition with deontic IS sentences

182

Two further supporting facts

182

The first support: defeasibility of the existence presupposition with epistemic IS sentences

182

The second supporting fact: a difference between epistemic IS and Descriptive BP sentences concerning existence of relevant situations in the actual world

183

The “enough” presupposition of Descriptive BP sentences and the relationship between descriptive generics and the process of inductive inference

185

The “enough” implication: basic data and intuition

186

The “enough” implication is a conversationally triggered presupposition

187

The temporal asymmetry and variability of the “enough” presupposition

191

Inductive inferences and descriptive generalizations

194

6.5

Back to the “enough” presupposition

196

6.6

An apparent counterexample: more cases where “enough” is very few

198

Conclusion and summary

199

TOLERANCE OF EXCEPTIONAL AND IRRELEVANT ENTITIES WITH IS AND DESCRIPTIVE BP SENTENCES

203

6.1 6.2 6.3 6.4

7. CHAPTER 4

xvii

1.

Introduction

203

2.

The basic data: similarities in the way IS and descriptive BP sentences tolerate exceptional and irrelevant entities

204

Capturing the similarities: a Kadmon & Landman-style “domain vague” restriction on the generic quantifier

209

Kadmon and Landman’s 1993 theory of generics

209

Applying K&L’s suggestion to the truth conditions of IS and descriptive BP sentences

211

3.2.1

A vague restriction on the set of individuals

211

3.2.2

A vague restriction also on the set of situations

213

3.

3.1 3.2

3.3 3.4 3.4.1 3.4.2 4.

Advantages of the new mechanism for tolerating exceptional and irrelevant entities

216

Shortcomings of the present mechanism for tolerating exceptions

218

“Abnormality” of the legitimate exceptions is not captured

218

A (newly observed) difference between IS and descriptive BP sentences is not captured

219

Why are descriptive BP and IS sentences different w.r.t. characterizing exceptions?

221

4.1

The intuition

221

4.2

Empirical support for the intuition: IS sentences where the characterization of legitimate exceptions seems impossible

222

The direction of formalization

227

Formalizing the “normality w.r.t. ^S” requirement on the vague restriction of IS sentences

229

The further pragmatic requirement on the vague restriction on the P set of individuals

229

4.3 5.

5.1

xviii

5.2

The further pragmatic requirement on the vague restriction on the set of situations

233

IS sentences with an even more vague restriction

236

Formalizing “normality” with the vague restriction of descriptive BP sentences

237

The further pragmatic requirement on the vague restriction on the set of individuals

237

The further restriction on the vague restriction on the set of situations

242

Conclusion

243

TEMPORALLY RESTRICTED IS AND BP SENTENCES

247

1.

Introduction

247

2.

Temporally restricted generics

248

2.1

The basic data

248

2.2

Differences between BP and IS sentences

249

Real generic or “functional”?

254

3.1

Condoravdi’s (1993, 1997) “functional” reading

254

3.2

Condoravdi’s conditions for nongenericity are relevant for temporally restricted sentences

256

Examining Condoravdi’s conditions for nongenericity

257

Positive indications for genericity

260

A closer examination of the BP/IS distinction in temporally restricted sentences

264

4.1

Two readings of the temporal adverbial

265

4.2

A descriptive generalization

268

How can existing theories of genericity handle the descriptive generalization?

270

5.3 6. 6.1 6.2 7. CHAPTER 5

3.

3.3 3.4 4.

5.

xix

5.1

5.1.1

5.1.2 5.2

6. 6.1

6.2

6.3 6.4 6.4.1 6.4.2 6.5 6.6

7.

Existing theories on the restriction on IS sentences. Or, why are characteristic (I-) genericity and episodicity mutually exclusive?

271

Binding of the situation variable, and similarity of Gen with overt Q-adverbs as explaining the behavior of IS sentences

272

Modality/law-likeness as explaining the behavior of IS sentences

275

Genericity theories on the behavior of BP sentences: why can BP sentences freely express “episodic genericity”?

276

Temporally restricted IS and BP sentences: the solution

279

Similarities between the behavior of IS and BP sentences in temporally restricted and classical generic sentences

279

The intuition: IS sentences with indexically interpreted temporal adverbials express “unreasonable generalizations”

281

Truth conditions of temporally restricted present tense IS and BP sentences

284

IS sentences with the “indexical” and “functional” readings of the temporal adverbials

287

The “reasonable causation” presupposition with temporally restricted IS sentences

287

Support for the proposed solution

291

Why BP sentences are compatible with the default, “indexical” reading of the adverbial

293

Actual times, actual worlds and nonaccidentalness: on the double nature of temporally restricted BP sentences

294

Conclusion

296

xx

CHAPTER 6 CONCLUSION AND DIRECTIONS FOR FURTHER RESEARCH

297

1.

Introduction

297

2.

A brief summary of the semantics and pragmatics of IS and BP sentences

298

2.1

Similarities between IS and BP sentences

298

2.2

The differences between IS and descriptive BP sentences

300

2.2.1

Accessibility relations

300

2.2.2

Presuppositions

302

2.3

Apparent interactions between “in virtue of” and “descriptive” generalizations

305

Directions for further research: why do IS and BP sentences differ in the availability of accessibility relation?

306

3.1

The problem

306

3.2

The semantic difference between IS and BP noun phrases

307

An apparent problem: can BP NPs corresponding to “extremely unnatural classes” be kind referring as well?

309

The interpretation of characterizing sentences with BP subjects

312

Chierchia’s (1998) and Krifka et al.’s (1995) suggestions

312

Similarities between characterizing sentences with BP and proper-name subjects

313

Two formal options for the representation of characterizing BP sentences

316

3.

3.3

3.4 3.4.1 3.4.2 3.4.3 3.4.3.1

Option A: a Delfitto (1996)-style representation

316

xxi

3.4.3.2 4.

Option B: a von Fintel (1994)-style representation for BP sentences

320

Conclusion

321

BIBLIOGRAPHY

325

INDEX

331

Preface

This is a slightly revised version of my Ph.D. dissertation, submitted to Bar-Ilan University in May 2002. I have updated the references, expanded the discussion of several topics (especially in chapter 4 and chapter 5), and made some stylistic changes throughout the text. Naturally, my thoughts about the topics dealt with in this dissertation have continued to develop and change since I finished writing it. The dissertation deals with the manifestation of two distinct kinds of nonaccidental genericity in the grammar of modern English. Empirically, the main concern of the work is a comparison of the semantics and pragmatics of two kinds of generic sentences which at first sight appear almost synonymous, namely those with indefinite singular subjects (IS sentences, for short), such as A dog has four legs, and those with bare plural subjects (BP sentences, for short), such as Dogs bark. I begin with several previous observations (in Lawler (1973), Burton-Roberts (1977) and Declerk (1991)), highlighting some differences between these two kinds of generic sentences, and intuitive attempts to explain these differences in terms of the “analyticity” or “rule-like” statements expressed by IS sentences as opposed to the more “descriptive” or “inductive” nature of BP sentences. Although some modern semantic theories of genericity (e.g. Krifka et al. (1995)) mention these claims, few attempt to define these intuitive differences more precisely or to integrate them into the formal analysis of genericity. The result is that almost all such theories concentrate only on the similarities between minimally contrasting pairs of IS and BP sentences, and assign them an equivalent (quantificational) formalization, as if they were truly synonymous. In this work I show that the range of semantic, pragmatic and distributional differences between IS and BP sentences (differences in felicity and truth value depending on a variety of combinations of subjects and VPs in different contexts, differences in tolerating

PREFACE xxiii

legitimate exceptions, etc.), is significantly larger than what has been observed so far, and thus should no longer be ignored. My main claim is that there is an underlying reason for these differences, namely the fact that the syntactic distinction between the two kinds of sentences encodes a semantic distinction between two basic kinds of quantificational, nonaccidental genericity, which I call “in virtue of” and “descriptive.” Specifically, I show that IS sentences are restricted to express only the former, “in virtue of” generalizations, asserting, very roughly, that the generalization is nonaccidentally true in virtue of some inherent property or factor, associated with the denotation of the subject. In contrast, BP sentences can express both “in virtue of” as well as “descriptive” generalizations, asserting that “there is a pattern here,” i.e. that the generalization is simply nonaccidentally true, without specifying any factor in virtue of which this is true. My intuitions about this semantic distinction are to a large extent inspired by the original intuitive observations found in Lawler (1973), Burton-Roberts (1977) and Declerk (1991), as well as by Carlson’s (1995) distinction between the “rules and regulations” vs. the “inductive” approaches to genericity. The main innovation of this work is in making precise these intuitions using the tools of formal semantics, by taking the underlying difference between the two kinds of generalizations to be a distinction between two basic kinds of modality, i.e. two kinds of accessibility relations which restrict the modalized generic quantifier associated with the two kinds of sentences. I show that by combining the claim about the underlying difference in accessibility relation with independently motivated semantic and pragmatic mechanisms, we can successfully account for the wide variety of both old and newly observed differences between IS and BP sentences. In addition, a number of similarities between IS and BP sentences are successfully accounted for by the quantificational-modal truth conditions assigned to both, and the exceptionstolerance mechanism integrated into these truth conditions (developed on the basis of Kadmon and Landman’s (1993) supervaluationist proposal for a “domain vague quantification”). In the final chapter I examine the relationship between the “in virtue of”/“descriptive” distinction, and Krifka’s (1987) classical I/D genericity distinction (between sentential, nonaccidental genericity, and nominal, kind-referring genericity). I show that the distinction I develop here cannot be reduced to Krifka’s I/D distinction, since both IS and descriptive BP sentences are instances of sentential, nonaccidental (I-) genericity, which should be formally represented as quantificational,

xxiv PREFACE

modal structures. I claim, nonetheless, that an explanatory account of the differences in accessibility relations between these two kinds of sentences can be arrived at by integrating into the theory independently motivated claims about the semantic differences between IS and BP noun phrases, as property and kind-referring expressions, respectively. Thus, the distinction I develop in this work between the two kinds of sentential, (I-) genericity can be taken to be influenced in an indirect and unexpected way by whether nominal, (D-) genericity interacts with it or not. Moreover, the difference I suggest between the two kinds of sentences becomes relevant to the ongoing debate on the interpretation of IS and BP noun phrases (dealt with in e.g. Delfitto (1996), DobrovieSorin and Laca (1996) and Chierchia (1998)).

Acknowledgments

It is my great pleasure to thank Susan Rothstein, my dissertation adviser and, in fact, my intellectual mother. The fifteen years of learning from and working with Susan—observing the way she thinks, teaches and works—taught me not only how to do semantics, but also what it means to be an intellectually mature and independent scholar, and how to express one’s own unique voice and soul through a formal theory. Her numerous comments and suggestions on this dissertation, and our long discussions and arguments (oftentimes so much fun!) were invaluable for helping me to shape what started out as a very vague idea into the theory I present here. Susan’s friendship, support and concern made some of the more difficult parts of these years much easier. I also sincerely thank Greg Carlson, Gennaro Chierchia, Angelika Kratzer and Fred Landman for their generosity and aid, particularly for their many instructive comments. Their work and thinking have had a profound influence on my work, and this dissertation in particular. It was Gennaro who told me, “We all know that genericity is a very complicated phenomenon, but we don’t want our theory of generics to be as complicated as the phenomenon itself!” I hope I managed, at least to some extent, to follow this idea. I am also grateful to Maria Bittner, Veneeta Dayal, Edit Doron, Manfred Krifka, Anita Mittwoch, Orin Percus, Carl Posy and the audience of the 1997 IATL conference, and the SALT XII conference for their help, suggestions, and comments. A special person I want to thank is Ariel Cohen, a rare combination of a serious intellectual rival and a warm, supportive friend. Having many similar intuitions about genericity, but more or less opposite views on how to formalize them, Arik and I have taken part in many long and heated discussions on generics. Arik gave me some of the sharpest comments and criticism on various points of my work, comments which greatly helped me to clarify my own beliefs. At the same time, Arik’s warm support, sense of humor and immediate

xxvi ACKNOWLEDGMENTS

willingness to help made the long, and often lonely, process of writing a dissertation much more enjoyable. A person who had a very significant, though indirect, influence on my work in general, and especially this dissertation, is David Zeilicovici, who taught me Philosophy of Science. In his intense and intriguing way, Dr. Zeilicovici introduced me to the work and thought of Carl Popper. Popper’s ideas about self-criticism and the importance of making one’s own theory as testable and vulnerable as possible became relevant at many stages of working on this dissertation, and I hope will continue to be relevant for me in the future as well. The English department of Bar-Ilan University has been my professional home for the last fifteen years, including the period of working on the dissertation. I want to thank all of the staff, academic and administrative, and especially Jonathan Fine and Joel Walters, for their steadfast and warm support and concern, and for doing anything they could to help me during all these years. I spent five years writing the dissertation in the library of the VanLeer Institute in Jerusalem. I am delighted to have this opportunity to thank the members of the Institute for their hospitality, and especially their wonderful, friendly and supportive librarian, Hanna CaineBrunschvig, I am also grateful for the planning and grants committee of the Council for Higher Education of Israel—the planning and grants committee—for the generous Rotenstreich fellowship I received between 1995 and 1998. It is a great pleasure for me to thank my wonderful family—my parents, Shmuel and Devora Feuerstein; my sisters, Chaya Brodie and Lea Weinreb, and their families; my parents-in-law, Ari and Rachel Greenberg; and above all, my dear and special husband, Yehoshua, and my beloved (and energetic!) boys, Chayim Shalom, Ilai and Levi—for being themselves and for their immense love, concern and tolerance, throughout these long and often difficult years of writing the dissertation. There is simply no way I could have finished writing it without their emotional, practical, and financial support. Our shared efforts to strike the proper balance between family and research enriched me, and placed the work on this dissertation in the right place and perspective in my soul and mind. Finally, I want to thank God.

CHAPTER 1 Comparing Sentences with Generically Interpreted Indefinite Singular (IS) and Bare Plural (BP) Subjects: The Traditional Approach and Its Problems

1. Introduction The main concern of this work is a comparison of the semantics and pragmatics of two kinds of generic sentences, namely those with indefinite singular subjects (IS sentences, henceforth), as in (1a), and those with bare plural subjects (BP sentences, henceforth), as in (1b): (1) a. A dog has four legs. (IS sentence: with an indefinite singular subject) b. Dogs have four legs. (BP sentence: with a bare plural subject) Although at first sight such minimally contrasting sentences look synonymous, there have been past observations which have pointed out certain differences between them, as well as intuitive attempts to explain these differences in terms of the “analyticity” expressed by IS sentences such as (1a), as opposed to the more “descriptive” or “inductive” nature of BP sentences such as (1b) (see e.g. Lawler (1973), Burton Roberts (1977) and Declerk (1991)). Some modern semantic theories of genericity mention these claims, but almost none of these theories relate to them seriously, define them more precisely, or try to integrate them in the formal analysis of genericity. This is probably because, as Krifka et al. (1995) write, the distinction is clear, and (it) manifests itself in striking results, but the underlying reason is not clear. (p. 13, ff#10) In practice, then, almost all modern theories of genericity concentrate only on the similarities between minimally contrasting pairs of IS and

2 MANIFESTATIONS OF GENERICITY

BP sentences as in (la) and (1b), and assign them an equivalent (quantificational) formalization. In this work I show that the range of semantic, pragmatic and distributional differences between minimally contrasting pairs of IS and BP sentences (like (1a) and (1b)) is significantly larger than what has been observed so far, and thus should not be ignored any longer. My main claim is that there is, indeed, an “underlying reason” for all these differences, namely that the syntactic distinction between IS and BP sentences encodes a semantic distinction between two basic kinds of quantificational, nonaccidental genericity, which I call “in virtue of” and “descriptive.” Specifically, I will show that IS sentences are restricted to express only the former “in virtue of” generalizations, asserting, very roughly, that the generalization is nonaccidentally true in virtue of some inherent property, or factor, associated with the denotation of the subject, whereas BP sentences can express both “in virtue of” as well as “descriptive” generalizations, asserting that “there is a pattern here,” i.e. that the generalization is simply nonaccidentally true, without specifying any property or factor in virtue of which this is true. My intuitions about this semantic distinction are largely inspired by the original intuitive observations in Lawler (1973), Burton-Roberts (1977) and Declerk (1991), as well as by Carlson’s (1995) distinction between the “rules and regulations” vs. the “inductive” approaches to genericity. The main innovation of this work is in making precise these intuitions, using the tools of formal semantics, by taking the underlying difference between the two kinds of generalizations to be a distinction between two basic kinds of modality, i.e. two kinds of accessibility relations which restrict the modalized generic operator associated with the two kinds of sentences. I show that by combining the claim about the underlying difference in accessibility relation with independently motivated semantic and pragmatic mechanisms, we can successfully account for the large variety of both old as well as newly observed differences between IS and BP sentences, including differences in entailments, presuppositions and implications, dependency on different contexts, and differences in tolerating legitimate exceptions. In the final chapter of this work I examine the relationship between the “in virtue of”/“descriptive” distinction, and Krifka’s (1987) classic I/ D genericit distinction. The I/D genericity distinction, which was shown to be manifested in a variety of linguistic phenomena, is a distinction between sentential, quantificational genericity, where a nonaccidental, characteristic generalization is made (e.g. by the sentence A dog barks),

MANIFESTATIONS OF GENERICITY 3

and nominal genericity, where some possibly accidental generalization is made through a noun phrase which directly refers to a kind (as with the reference of the subject in the sentence The dodo is extinct). I show that the distinction I develop here, between IS sentences expressing “in virtue of” generalizations and BP sentences under their “descriptive” reading cannot be reduced to Krifka’s I/D distinction, since both are instances of sentential, nonaccidental (I-) genericity. I claim, nonetheless, that an explanatory account of the differences in accessibility relations between these two kinds of sentences can be arrived at by integrating into the theory independently motivated claims about the semantic differences between IS and BP noun phrases (as property and kind referring expressions, respectively). Thus, the distinction I develop in this work between the two kinds of sentential, (I-) genericity can be taken to be influenced in an indirect and unexpected way by whether nominal, (D-) genericity interacts with it or not. In this chapter I review the dominant semantic approach to minimally contrasting IS and BP sentences, which treats these sentences in exactly the same manner, and then point at old and new differences between these two kinds of sentences which indicate the need to supplement this dominant approach with a finer-grained semantic analysis of these sentences. In section (2) of this chapter I present a brief review of relevant genericity phenomena and widely held semantic analyses which have been proposed in the literature for these phenomena. I focus on the equivalent modal-quantificational representation given to minimally contrasting IS and BP sentences such as (la) and (1b), which will provide the starting point for the analysis developed in later chapters. (This section can be skipped by readers who are familiar with the modalquantificational approach to genericity). In section (3) I turn to existing and new observations concerning semantic, pragmatic and distributional differences between IS and BP sentences, which pose a problem for these widely held theories, These concern two kinds of constructions: temporally restricted, or “episodic generic,” constructions, and in more traditional generic constructions (namely simple present sentences with individual-level or habitual predicates). In section (4) I point to a number of parallels between the IS/BP distinction in the temporally modified and the traditional generic constructions, which indicate that the behavior of IS and BP sentences in both constructions should be attributed to the same underlying semantic difference. In section (5) I sketch my intuition concerning this basic difference between IS and BP sentences, which will be formally


clarified in the rest of this work. Finally, in section (6) I provide an outline of the rest of the chapters in this work. 2. A brief review of central genericity phenomena and analyses 1 2.1 Genericity in English As their name suggests, generic sentences, in all languages, express generalizations. In English, as in many other languages, genericity can be manifested through a variety of syntactic constructions. For example, despite their syntactically different subjects, all of the sentences in (2)-(6) are traditionally considered to be generic: (2) Indefinite singular subject: a. A whale is a mammal. b. A blue whale eats plankton. (3) “Bare plural” subject: a. Blue whales eat plankton. b. Blue whales are on the verge of extinction. (4) Definite subject: a. The killer whale is not dangerous to humans. b. The rat reached Australia in 1770. (Krifka et al. (1995), p. 12) (5) Mass noun subject: Wine contains at least 10% alcohol. (6) Proper noun subject: John speaks very quietly.

Notice that the same kind of NP subjects can appear in nongeneric sentences as well: (7)

1

a. A blue whale attacked our ship last week. b. Blue whales are diving under our ship!

For a much more detailed review, see Krifka et al. (1995).


c. The whale is yours, if you want it. d. Wine was spilled on my dress. e. John was speaking very quietly now. The semantic difference between the sentences in e.g. (2)-(6) and those in (7) is very strong. All the sentences in the first group express some kind of generalization. For example, the sentences in (2) and (3) do not talk about particular whales (as in (7a-c)), but rather about whales or blue whales in general, or about the whole species of blue whales. Similarly (5) gives a property of wine in general (as opposed to (7d), which talks about a particular quantity of wine), and (6) talks about what John does in general (as opposed to (7e), which concerns what John did in a particular situation). 2.2 The I (characteristic)/ D (kind) genericity distinction The class of generic sentences is not only syntactically but also semantically heterogeneous. Several theories have pointed out important semantic distinctions in the kind of meanings expressed by generic sentences, which correlate with syntactic distinctions (e.g. with the characterization of the syntactic subject). In his (1987) paper, Krifka (partly in collaboration with Gerstner) introduced one such distinction, originally labeled the D (=definite)/I (=indefinite) genericity distinction, which became a central assumption of most current genericity theories (see e.g. Wilkinson (1991), Schubert and Pelletier (1987, 1988, 1989) ter Meulen (1995) and Link (1995)). I-generics, also called “characterizing” or “characteristic” generics (in Krifka et al. (1995), and henceforth),2 are exemplified in e.g. (2a, b) (3a) and (6) above, and in the similar sentences in (8): (8) a. Dogs bark/A dog barks. b. A blue shark is very dangerous/Blue sharks are very dangerous. c. John walks to school.

2

And also Characteristic Kind Predication (CKP), in e.g. ter-Meulen (1995), or Derived Kind Predication (DKP), in e.g. Link (1995).


Intuitively, such sentences express some general characteristic, or a regularity, by predicating a certain property of all/most individuals and/ or situations. For example, (8a) roughly asserts that all/most individual dogs bark, (8b) that all/most individual blue sharks are dangerous and (8c) that all/most situations where John goes to school are walking situations. In analyzing such sentences semantically, the majority of recent theories take them to indeed involve some kind of quantification over all/most individuals and/or situations. A sentence like (9a), for example, is very roughly represented as in (9b) (where d is a variable for individuals, and “Gen” stands for the generic quantifier), and paraphrased as in (9c): (9)

a. A dog has four legs. b. Gen d dog (d) → has four legs (d) c. All/most dogs have four legs.

In the following sections I examine the suggested semantics of such sentences in detail. In contrast to such characteristic generics, D-generic sentences, also called kind-generic sentences (in Krifka et al. (1995), and henceforth), are exemplified in e.g. (3b) and (4b) above, as well as the similar sentences in (10): (10) a. Dodos are extinct. b. Linguists have more than 8000 books in print. (Krifka et al. (1995) #124b) c. In Alaska, we filmed the grizzly. (Krifka et al. (1995) #124f) d. Rats were reaching Australia in 1770. (Krifka et al. (1995) #28b) e. Man set foot on the moon in 1969. (Krifka et al. (1995) #124g) Unlike the characteristic generics above, these generic sentences do not express a general property or a regularity, and are clearly not predicating a property of all/most instances of the kind. Sentence (10e), for example, cannot be paraphrased as “Every man set foot on the moon in 1969.” Similarly, (10a) does not mean that all/most individual dodos are extinct, and (10d) does not mean that all/most rats reached Australia in 1770. Instead, sentences like (10a-e) seem to assert that a certain property holds directly of the kind itself (and are thus also called Direct Kind


Predication in e.g. ter Meulen (1995), or Proper Kind-Predication (in e.g. Link (1995)). For example, (10d) seems to mean that the kind “rat” reached Australia in 1770 (probably through the arrival of the first individual rats to Australia), (10a) means that the whole kind “dodo” is extinct (since there are no more individual dodos in the world), and (10e) that the kind “Homo sapiens” set foot on the moon in 1969. Most theories of genericity represent such sentences as involving no quantification, but instead as predicating a property directly of kind entity. The subjects of (10a-e), then, are semantically taken to denote kind terms,3 and sentences like (11a) are roughly represented as in (11b), (where capital letters stand for a kind term), and paraphrased as in (11c): (11) a. Blue whales are on the verge of extinction. b. On the verge of extinction (BLUE WHALES) c. The property of being on the verge of extinction is true of the kind blue whales. 2.3 Relevance of the characteristic/kind distinction to the IS/BP distinction Having the characteristic/kind genericity distinction in mind, we can turn to the first generalization concerning IS and BP sentences. Carlson (1977) observed that IS sentences are infelicitous when IS subjects combine with clear “kind level” predicate (like be extinct, be rare), as in (12a) and (12b). When combined with episodic predicates, as in (12c) such sentences are still felicitous, but the IS subject does not have a generic reading, but only an existential one:4 (12)

3

a. # A blue whale is rare. b. # A dodo is extinct. c. A rat reached Australia in 1770.

The interpretation of “kind” is debated. According to Carlson (1977) kinds are primitive kinds of individuals—intensional, abstract entities, which can have realizations in the world as “objects” or “stages.” Chierchia (1998) takes kinds to be collective entities. This view is followed also in Dobrovie-Sorin & Laca (1996). 4 (a) is felicitous, and (b) has a generic reading if the indefinite singular NP has a “taxonomic reading,” where it denotes a certain kind of blue whale.


On the basis of these types of sentences, and similar data, Krifka (1987) (followed by ter Meulen (1995), Wilkinson (1991), Link (1995), Schubert and Pelletier (1987, 1987, 1989), among others) claims that IS sentences can only express characteristic (namely I-) genericity, as in A dog barks/has four legs, and not kind genericity. These theories claim further that this indicates that indefinite singular NP subjects can never be interpreted as denoting kind terms, but instead as predicate expressions, introducing variables into the representation, which are bound by the generic quantifier. In contrast to IS sentences, BP sentences are all perfectly felicitous as generic in the minimally contrasting (13a-c): (13)

a. Blue whales are rare. b. Dodos are extinct. c. Rats reached Australia in 1770.

Several theories (Carlson (1977), and more recently Dobrovie-Sorin & Laca (1996) and Delfitto (1996)) claim that the felicity of BP sentences like (13a-c) indicates that bare plural NPs are always kind denoting, and thus that any generically interpreted BP sentence, even one like Dogs bark, is in fact a kind generic (or a Direct Kind Predication structure), involving no generic quantification. In most theories, however, (e.g. Chierchia (1995, 1998), ter Meulen (1995), Wilkinson (1991, 1995), Link (1995), Schubert & Pelletier (1988), the label “kind genericity” is saved for sentences like (10a-e) and (13a-c), which can only be interpreted as “Direct Kind Predication,” and cannot be understood as involving quantification over all/most entities. Most of these theories thus take sentences like Dogs bark/Dogs are dangerous to express only characteristic (I-) genericity. The implication of this latter claim, which is central to claims I will make in the chapters below, is that most recent theories assign the same basic quantificational representation to minimal pairs like (14a, b) and (15a, b). (14)

(15)

a. Dogs bark. b. A dog barks. a. Dogs have four legs. b. A dog has four legs.


The more detailed semantics, and motivation for this quantificational representation is the focus of the next subsections. As for the interpretation of the bare plural NP subjects, most theories (e.g. Chierchia (1995), ter Meulen (1995), Wilkinson (1991), (1995), Link (1995), Schubert & Pelletier (1988)) claim that such NPs are ambiguous between predicate expressions, just like their IS counterpart (i.e. as introducing variables which are bound by the generic quantifier) in characteristic generics like (14a-15a), and kind referring terms, in Proper Kind-Predication structures like (13a-c).5 In the chapters below I adopt the view that BP subjects in sentences like (14a) and (15a) are represented like their IS counterparts, i.e. as indefinite expressions in quantificational structures. In the last chapter I examine more closely the claims about the possible different interpretation of IS and BP noun phrases, and make use of these claims in sketching the direction for a more explanatory account of the underlying difference between IS and BP sentences. 2.4 The basic semantic structure of characteristic (I-) generics Farkas & Sugioka (1983) and Carlson (1989) were the first theories to suggest that characteristic generics like (14) and (15) involve a “relational” generic quantifier, i.e. a two-place quantifier, which semantically relates a matrix saying what the generalizations is about— and a “related” constituent, where some property is predicated of the generalized entity. For example, (14a) and (14b) have the same schematic representation in (16a), paraphrased in (16b): (16)

5

a. Gen [dogs] [bark] b. For dogs (in general) it holds that they bark

Two further versions of this approach are Chierchia (1998) and Krifka et al. (1995). Chierchia (1998) claims that bare plural NPs are unambiguous kind terms, but in characteristic generic sentences, where a generic quantifier is present, they are type-shifted into indefinite expressions. Krifka et al. (1995) claim that in sentences like (14a) and (15a) the BP subject can still be interpreted as a kind term, and that the quantification ranges over (roughly) all/ most “realizations” of the kinds. In both theories, then, (14a) and (15a) express quantification over instances of the kind. I come back to these two theories in chapter 6.


There are also cases where there are two ways to partition the sentence into matrix and “related constituent,” and as a result we get two possible generic readings. Sentence (17a) (originally cited by Milsark (1974)), and sentence (17b) are examples of such ambiguous sentences: (17)

a. Typhoons arise in this part of the Pacific. b. Computers control modern planes.

(17a) can mean either that it is a property of typhoons in general that they arise in this part of the Pacific, or that it is a property of this part of the Pacific in general that there are typhoons arising in it. Similarly (17b) can either mean that it is a property of computers in general that they control the route of modern planes, or (more reasonably) that it is a property of modern planes in general that they are controlled by (some) computers. The first and second readings of (17a) and (17b) are obtained by the partitions of the sentence in (18a, b) and (19a, b), respectively: (18) a. Gen [typhoons] [arise in this part of the Pacific] b. Gen [this part of the Pacific] [(some) typhoons which arise in it] (19) a. Gen [computers] [control modern planes] b. Gen [modern planes] [controlled by (some) computers] This intuitive “relational” approach to characteristic genericity became widely accepted and was much more precisely formulated following the introduction and adoption of Heim’s (1982) and Kamp’s (1981) theories of indefinite expressions and tripartite quantification. In these theories, indefinite singular and bare plural NPs are interpreted as indefinite expressions which have no quantificational force of their own. Instead, they are predicate expressions introducing variables into the representation, which are bound by default existential closure, or by overt or covert quantificational adverbs (like usually, always or sometimes) in a “restrictive clause” (defining what the quantification is about) and a “nuclear scope” (predicating a property of the quantified entities). Such quantifiers are “unselective” in the sense of Lewis (1975), i.e. they can bind variables of different kinds, specifically both individual variables introduced by indefinites, as well as event/situation variables introduced by verbal predicates. For example, the sentence in (20a), with the quantificational adverb always has the representations in


(20b), where d and s are variables over individuals and situations, respectively: (20) a. A grizzly bear always snores loudly. contextually relevant (s, d)] b. d, s [Grizzly bear (d) [snores loudly (s, d)] Paraphrase: “Every relevant situation involving any individual grizzly bear (e.g. every sleeping/snoring situation) is a situation where that bear snores loudly.” Following these ideas, theories like Wilkinson (1991, 1995), Chierchia (1995), Kratzer (1995), Diesing (1992), Krifka (1987, 1995), Krifka et al. (1995), and Schubert & Pelletier (1989) claimed that characteristic (I-) generics as in (21a) are associated with a nonovert unselective generic quantifier (called Gen) (similar to the overt quantificational adverb usually) and have the semantic structure as in (21b): (21) a. A grizzly bear snores loudly. /Grizzly bears snore loudly. contextually relevant (s, d)] b. Gen d, s [Grizzly bear (d) [snores loudly (s, d)] Assuming the claims made by neo-Davidsonian theories such as Condoravdi (1992), de Swart (1994) and Chierchia (1995), according to which all predicates have a situation variable, this basic semantic structure is widely used not only with stage-level predicates like snore or run quickly (denoting properties which are basically transitory) as in (21), but also with individual-level predicates like be black or be a mammal (denoting permanent, or life-time properties), as in (22a), represented as in (22b):6 (22) a. A grizzly bear is a mammal. /Grizzly bears are mammals. contextually relevant (s, d)] b. Gen d, s [Grizzly bear (d) [mammal (s, d)]

6

As Chierchia (1995) notes, with individual-level predicates as in (22a), the requirement that the situation involving the individual is “contextually relevant” is in fact vacuous. Since such predicates hold of an individual for a long time, any situation involving these predicates is taken to be contextually relevant. Thus, with individual-level predicates we require the restriction that the situation simply contains, or involves the individual.


2.5 The semantics of the generic operator—the modal approach 2.5.1 Tolerance of exceptions, law likeness, and counterfactual support of characteristic generics Assuming the basic semantic structure of IS and BP sentences with stage-level and with individual-level predicates, such as those in (21a) and (22a), the last important issue to be clarified is the nature of the generic quantifier Gen. As all theories correctly note, Gen cannot be a simple extensional universal quantifier, i.e. (21a) cannot be represented as in (23): (23)

d, s [Grizzly bear (d) loudly (s, d)]

contextually relevant (s, d)] [snores

(23) means “Every (relevant) situation (e.g. a snoring situation) involving every grizzly bear is a situation where this bear snores loudly.” Such a paraphrase does not capture the meaning of (21a) because of the well-known fact that unlike simple universal statements, characteristic generics tolerate the existence of exceptions. In other words, the representation in (23) entails that the sentences in (21a) are false if we find even one grizzly bear that does not snore loudly, or even one situation where a certain grizzly bear that usually snores loudly suddenly snores quietly. But, as many theories show (e.g. Dahl (1975), Krifka (1987), Krifka et al. (1995), Asher & Morreau (1995), Wilkinson (1991)), these sentences, and many other characteristic generics, are still true in such scenarios. In fact, most characteristic generics allow for exceptions, which are usually taken to be “less normal” in some sense. One may take this “tolerance of exceptions” property to suggest that the generic quantifier has the quantificational force of “most” instead of “every,” i.e. that (21a) should be represented as in (24): (24) Most d, s [Grizzly bear (d) [snores loudly (s, d)]

contextually relevant (s, d)]

But this kind of representation fails due to another special property of characteristic generics (which also distinguishes them from kind-generic sentences), namely their law-likeness, or “nonaccidental” nature.


Sentence (24) (with most) can be true if it happens to be the case that most actual snoring situations involving most actual grizzly bears are snoring-loudly situations. But this is not enough to make (21a) true. A similar example (inspired by Krifka et al.’s (1995) example, p. 44) is the following: Suppose that there are only eight lions left in the world, and seven of them lost a leg in an accident or a fight, or due to disease. This situation is enough to make (25a), but not the generics in (25b) true: (25)

a. Most lions have three legs. b. A lion has three legs. /Lions have three legs.

Again, the intuition is that the generics in (25b) express a law-like, nonaccidental generalization, and it takes more than the properties of actual individuals, in the actual world, to make them true. Thus, neither the extensional quantifier every nor most can be used to arrive at an exact paraphrase of such sentences This latter intuition has led several researchers (e.g. Dahl (1975), Heim (1982), Krifka (1987, 1988), Krifka et al. (1995), Chierchia (1995, 1998), Asher & Morreau (1995), and Wilkinson (1991)) to claim that characteristic generics are intensional, and more specifically, modalized. To be more precise, the claim is that Gen, the generic quantifier, is a modalized universal operator, quantifying over all accessible possible worlds, as well as individuals and situations. Dahl (1975), for example, claims that “indefinite generic NPs…always involve quantification over possible objects, rather than over actual ones” (p. 108), and suggests interpreting generic sentences as “in all possible worlds, any object x in w, has the property P,” (p. 109) or alternatively as involving universal quantification within the scope of a modal operator. Similarly, Krifka (1988) claims that (I- generics) cannot be used to express facts which hold just coincidentally, but are law-like statements…. For example, if some nut were to clip the wings of every existing blackbird then the sentence A blackbird flies would nevertheless remain true. If one tries to develop semantic analysis in terms of possible world semantics, I-generic sentences cannot be statements with a truth value that can be checked at one index, e.g. the actual word. Instead, we have to take into account a set of indices. Thus, genericity is reconstructed as a modal notion—as some sort of necessity. (p. 297)


This line of thinking is strengthened by the fact that unlike (26a) and (26b), generics like (27a) and (27b) support a counterfactual statement, as in (28): (26) (27) (28)

a. Every paper on generics is long.7 b. Most papers on generics are long. a. A paper on generics is long. b. Papers on generics are long. If this were a paper on generics, it would be long as well.

As claimed in e.g. Stalnaker (1968) and Lewis (1973, 1986b), evaluating the truth of the counterfactual requires considering other possible worlds besides the actual one. If the generic quantifier were extensional, i.e. if the assertion that “every situation involving every paper on generics is a situation where that paper is long” was asserted to hold in the actual world alone, there would be no way to guarantee the entailment of the counterfactual in (28). Instead, the entailment can be

7

There is a reading of (26a) in which it does support the counterfactual in (28), namely one where it expresses a nonaccidental universal statement. Carlson (1989) (and then also Kratzer (1989)) observe that sentences with overt nominal universal quantifiers like (i) are ambiguous between the extensional/accidental (paraphrased in (ia)) and the intensional/nonaccidental reading (paraphrased in (ib)):

(i) Every friend of John’s votes for Socialists (ia) It happens to be the case that Bill, Harry, Jane, Anne and Tom, who are John’s only friends, all vote for Socialists. (ib) It is a requirement that anybody who is a friend of John must vote for Socialists. The second, intensional reading of (i), supports the counterfactual in (ii): (ii) If this was a friend of John, he would vote for Socialists as well Unlike standard generic sentences (like (27a, b), however, such generic universal statements do not tolerate exceptions. In chapter 4 (footnote 4) I show specifically which underlying semantic characteristics lead to the similarities and differences between the two kinds of generic, intensional sentences.


naturally accounted for if the universal statement is asserted to hold in all accessible possible worlds.8 2.5.2 Kratzer’s (1981) analysis of nongeneric modal verbs and its application to the semantics of Gen. The modal treatment of Gen was further formally developed by Heim (1982), Krifka (1987) and Krifka et al. (1995), using the framework developed in Kratzer (1981) for interpreting nongeneric modalized sentences like (29) and (30): (29)

(30)

a. John must still be alive. b. John should finish his homework now. c. I must have this gold ring! a. John may still be alive. b. I cannot play the trombone. (Kratzer (1981), p. 54) c. You can go home now.

Kratzer shows that the semantics of modal verbs as in (29) and (30) involves three parameters: the modal force, the modal base and the ordering source. The modal force determines whether the modal verb is represented as a universal or an existential quantification over worlds, and consequently whether the sentence expresses necessity (as in (29) or possibility (as in (30)). The modal base and ordering source are both “conversational backgrounds,” i.e. functions from worlds to sets of propositions, in other words, sets of sets of worlds). There are various such typical functions: epistemic—from worlds to sets of propositions known in these worlds, circumstantial—from worlds to sets of propositions of a certain kind which hold in these worlds, deontic—from worlds to sets of propositions commanded in these worlds, buletic— from worlds to sets of propositions wished, or wanted in these worlds, normal—from worlds to sets of propositions which form the normal course of events in these worlds, and so on.

8

Although Cohen (1999a) agrees that generics are intensional, he claims that their intensionality is temporal, and not modal I argue against this view, and for the modal analysis of genericity, in chapters 2.


The modal base determines the set of accessible worlds, and the ordering source ranks these worlds as closer or farther away from some “ideal” world. Schematically, modalized sentences as in (29a) expressing necessity, are interpreted as in (31) (where p is an atomic proposition): (31) must p is true in a world w iff w′ [w′ is a member of the set of worlds of the modal base, closest to the ideal world determined by the ordering source] → [p is true in w′] The specification of the modal base and ordering source is contextually determined. Different combinations of modal bases and ordering sources yield various kinds of modalized sentences. For example, in the prominent reading of (29a) p, (“John is still alive”), is asserted to hold in all worlds which are epistemically accessible from w, (i.e. where what is known in w holds), which are also closest to an ideal world where the “normal course of events in w” hold. In (29b) p (“John is finishing his homework now”) is most likely to hold in all worlds where facts of a certain kind hold (e.g. where John is a pupil, and he has homework for tomorrow morning), which are closest to an ideal world where what is commanded holds (e.g. that all pupils have their homework ready on time), etc. Krifka et al. (1995) suggest using Kratzer’s framework of modality in the truth conditions of characteristic generics like (32a) and (32b), as in (33) (ignoring the situation variable for the moment): (32) a. A lion has a bushy tail. b. Lions have a bushy tail. (33) GEN [x1…xi, y1…yi] (Restrictor, matrix) is true in w relative to a modal base Bw and an ordering source ≤w iff: For every x1…xi and every w′ Bw s.t. Restrictor [x1…xi] is true in w′, there is a world w″ in Bw s.t. w″ ≤w w′, and for every world w'″≤w w″, y1…yi Matrix [ {x1},…{xi}, y1…yi]] is true in w′″. (p. 52) (where Bw is a modal base, and w″ ≤w w′ means that w″ is closer to the ideal world determined by the ordering source than w′)


According to this definition, (32a) and (32b) can mean that “everything which is a lion in the worlds of the modal base is such that, in every world which is most normal according to the ordering source, it will have a bushy tail” (p. 52). This kind of representation has three main advantages. Firstly, it captures the “law-likeness,” nonaccidentalness and counterfactual support of sentences like (32a, b) (through the universal quantification over worlds). Secondly, the use of the “normal” sourc e accoun ts for the toler an ce of exce pt ions b), i.e. the fact that the existence of lions without a bushy tail does not falsify (32a) and (32b), despite the universal quantification over lions. The idea is that the universal quantification over individuals is asserted to hold only in the accessible worlds which are “most normal.” In such worlds there are no “abnormal” lions, e.g. those which have lost their tails in an accident or a fight, or those who were born with a different tail, as a result of a genetic mutation. Thus, such “abnormal” lions are not quantified over in the first place.9 The third advantage of the definition in (33) is in naturally accounting for the various “flavors,” or kinds of rules which characteristic (I-) generics can express. This variety can be seen in the following examples: (34)

a. A dog has four legs. (epistemic) b. A pupil is polite to his teachers. (deontic) c. A single parent pays 24% income tax. (legal) d. A bachelor is an unmarried man. (linguistic) e. An even number is non-prime. (mathematical)

This variety of “flavors” is naturally accounted for in Krifka et al.’s suggestion, in very much the same way this variety is accounted by Kratzer with modalized nongeneric sentences like (29) and (30) above, i.e. by employing different combinations of modal bases and ordering sources in the truth conditions.

9

See also Morreau (1996), Asher & Morreau (1995) and Pelletier & Asher (1997) for similar suggestions. In chapters 2 and 5 I show that there are several serious problems with this “normal ordering source” solution to the exceptions problem, and suggest an alternative exceptions-tolerance mechanism, based on Kadmon and Landman’s (1993) “domain vague” quantifiers.


2.6 Summary This brief review presents a certain widely held approach to the semantics of IS and BP sentences like Dogs bark/have four legs and A dog barks/has our legs. The two main claims of this approach, which focus on what IS and BP sentences have in common, are now summarized in (35A) and (35B). These claims will prove to be especially relevant for the arguments made in the following chapters (35) A. The semantic representation of BP sentences which are understood as quantificational, characteristic generics (e.g. Dogs have four legs) is equivalent to the representation of minimally contrasting IS sentences (e.g. A dog has four legs). B. This equivalent representation is a tripartite quantificational structure, headed by a universal, modal and unselective operator, binding world variables, situation variables (introduced by the verbal predicate) and individual variables (introduced by the bare plural or indefinite singular NP subjects). This kind of representation naturally accounts for the law-likeness, tolerance of exceptions and variety of “flavors” which can be expressed by characteristic generics. I now turn to examine the data which indicate that the claim in (35A) must be changed. 3. Unexpected differences between minimal pairs of IS and BP sentences In this section I examine five principal observations concerning differences between minimal pairs of IS and BP characteristic (I-) sentences. These differences pose a problem for the claim in (35A) above, because they indicate that these two kinds of sentences should not be represented equivalently. In section (3.1) I examine differences in the felicity of IS and BP sentences in temporally modified generic constructions. In section (3.2), I examine the differences noted by Lawler (1973), Declerk (1991) and Burton-Roberts (1977) in the level of law-likeness expressed by IS and BP generics, and in their felicity when


combined with VPs expressing “nonessential” properties. I then look at the prominent generic reading of BP sentences as opposed to the prominent and unexpected existential reading of IS whose subjects express “extremely unnatural classes” (in section (3.3)), or those whose VPs express “extremely unconnected properties” (in section (3.4)), In section (3.5) I examine the different effect that the presence and absence of supporting context has on the felicity of IS and BP sentences expressing highly unreasonable generalizations. Finally, in section (3.6) I look at minimal pairs of IS and BP sentences which do not differ so much, and which indicate the need to formally capture not only the differences, but also the similarities between the two kinds of sentences. In the rest of the work I develop the claim that the variety of differences between IS and BP sentences in characteristic (I-) generic sentences are due to a single underlining difference in the nature of modality they involve, which results in two kinds of nonaccidental generalizations that these sentences express. Thus, I maintain the second claim, namely the basically equivalent quantificational-modal semantic structure for IS and BP sentences, but give two distinct definitions of the accessibility relation in this semantic structure.10

3.1 Differences between IS and BP sentences in temporally modified sentences In the sections above we looked at the behavior of IS and BP with VPs expressing individual level or habitual properties (like A dog barks/Dogs bark and A dog has four legs/Dogs have four legs respectively). Such constructions have always been considered classical I-generic constructions. In contrast, sentences with VPs expressing episodic, stage-level predicates such as be closed, be busy, and be happy, are traditionally considered nongeneric. The traditional assumption is that only kind-predication sentences (like Rats reached Australia in 1770), are felicitous as generic with episodic information, whereas characteristic genericity and episodicity are mutually exclusive.

10

In the last chapter of the dissertation I examine the possibility that the difference in accessibility relation between IS and BP sentences should be attributed, after all, to a different interpretation of the two kinds of NP subjects in them.


In other words, the claim is that characteristic generic sentences always express what Chierchia (1995) calls “tendentially stable” properties. Carlson (1989, 1995) even takes nonepisodicity to be a central component in the definition of characteristic genericity. To quote: I will take it for granted that I know what a generic sentence is—any sentence expressing a generalization—and that the opposing category consists of episodic sentences—sentences which relate specific occurrences (1995, p. 224) [emphasis mine]. Typical examples which are presumed to support this view are the existential-only interpretation of the bare plural and indefinite singular subjects in (36), and the oddness of the habitual sentences in (37): (36) (37)

a. Dogs are barking now. b. A dog is barking now. # John walks to school today.

It turns out, however, that unlike this traditional view, BP and IS can both receive a generic interpretation as subjects of episodic predicates,11 i.e. with predicates modified by specific temporal adverbials, as in (38) and (39): (38) (39)

Italian restaurants are closed tonight. A Catholic is in church today.

My claim is that sentences like (38) and (39) are real generic sentences (and not “functional” in Condoravdi’s (1993) terms12), and, crucially, that the kind of genericity involved in both cases is characteristic and not kind-predication genericity. For example, both the BP and the IS subjects combine with predicates of individuals (i.e. nonkind level predicates), and moreover, the interpretation of these subjects seems to be “quantificational,” i.e. something like “all/most (typical) Italian restaurants” (in (38) and “all/most Catholics” (in 39)). In other words, the predicates in these sentences denote properties which are distributed

11

See also McNally (1995) who makes a similar claim. In chapter (5) below I discuss in detail the difference between I-generic and Condoravdi’s (1993) “quasi universal” reading.

12


over all/most members of the sets denoted by the subject. In addition, both the BP sentence in (38), and the IS one in (39) tolerate exceptions. Finally, and most crucially, both IS and BP sentences express nonaccidental generalizations. This is felt very strongly with the IS sentence in (39), but also with the BP one in (38) which, like standard characteristic generics supports counterfactual statements, as in (40): (40) If this was an Italian restaurant, it would be closed tonight. (pointing at a Chinese restaurant) Taking into consideration all these observations, the sentences in (38) and (39) should be analyzed as examples of very bizarre creatures indeed: episodic characteristic-generic sentences. Furthermore, given the widely held theories reviewed above on the semantic equivalence of minimally contrasting IS and BP sentences, these sentences should be given an equivalent (quantificationalmodal) representation. Examining such sentences more closely, however, we can see differences between IS and BP sentences. Firstly, although both sentences express nonaccidental generalizations, IS sentences are felt to express a stronger type of “normative” rule than the BP sentence, which seem to express a weaker, more descriptive generalization. What is more serious is that examining a wider range of temporally restricted sentences we can see that in most cases BP are judged as much better in this construction than their IS counterparts. In fact, as seen in (41), the IS sentence in (39) above is exceptional, in that the judgments about most generically interpreted IS sentences in this construction vary between “odd” and “completely unacceptable.”13 In contrast, the minimally contrasting BP sentences are in general perfectly natural with such temporal adverbials, as seen in (42):

(41)

13

a. # An Italian restaurant is closed tonight. b. # An accountant is very busy this week. c. # A lion is very aggressive today. d. # A child is especially polite today.

These sentences are fine under the existential reading of the IS subject. From here on I will relate only to the generic reading. In addition, people who are “exposed” to long lists of such sentences find them better under their generic readings.


(42)

e. # A Clinton supporter is happy tonight. f. # An American believes in God today. a. Italian restaurants are closed tonight. b. Accountants are very busy this week. c. Lions are very aggressive today. d. Children are especially polite today. e. Clinton supporters are happy tonight. f. Americans believe in God today.

Interestingly, many informants find the status of originally infelicitous IS sentences as in (41) better when they are uttered in context, rather than “out of the blue.” However, not every kind of context can help. Specifically, the IS sentences get better in what I will call a “predictional” scenario, but not in an “inductive” scenario. In contrast, BP sentences can appear easily in both kinds of scenarios.14 The following examples illustrate the different kinds of scenarios, and the different behavior of BP and IS sentences in them:15

(43) The Italian restaurants scenarios: The inductive scenario: John and Mary plan to meet in town and eat in an Italian restaurant. John arrives in town before Mary, and while walking down the street, he sees that Mamma Mia is closed. He then goes to Little Italy, and then to Fonta Bella, but they are also closed. Two more Italian restaurants he tries are also closed. When Mary arrives and meets him, John says: (a) Look, Italian restaurants are closed tonight. Let’s go to an Indian restaurant (b) # Look, an Italian restaurant is closed tonight. Let’s go to an Indian restaurant. The predictional scenario: My friend and I decide to eat tonight in an Italian restaurant, but an hour later my friend calls

14

The original observation of this distinction is due to Susan Rothstein (p.c.) As can be seen from the way I marked the IS sentences, their status is problematic, and worse than that of their BP counterparts even in the predictional scenarios. Nonetheless, most informants indeed find their status indeed significantly better in these scenarios than in the “inductive” ones.

15


me and says, “We’d better go eat in an Indian restaurant tonight. I just remembered that it’s Italy’s Independence Day, (a) Italian restaurants are closed tonight. (b ? An Italian restaurant is closed tonight. (44) The accountants scenarios: The inductive scenario: I need to consult with an accountant, so, I open the Yellow Pages and try to call one. Unfortunately, the secretary tells me her boss is very busy and I should try again next week. So, I try the next accountant on the list. This time the accountant himself answers, and says he is very sorry, but he is very busy now. In the following three or four days I try again and again to set up a meeting with an accountant, but all the accountants I call are busy. At some point, I tell my husband: (a) Accountants are busy this week. I’ll try again next week. (b) # An accountant is busy this week. I’ll try again next week. Predictional scenario: It’s the week the taxes are due, and everybody is dealing with paying their taxes. I listen to a program about economics on the radio, and an expert being interviewed says: “Anybody who wants to consult with an accountant should do it now: a. Accountants are busy this week” b. ? An accountant is busy this week” A comparison of the two kinds of scenarios suggests that IS sentences in temporally modified sentences are more felicitous as generic when the temporal adverbial modifiers are interpreted as referring to a day or week of a certain kind (Italy’s independence Day, the last week of the year, or the like). In contrast, BP sentences are perfectly felicitous with both this reading of the temporal adverbial as well as with the more natural “indexical” interpretation, as simply specifying the temporal location relative to the utterance time (i.e. the day or week of utterance). For example, the BP sentence Italian restaurants are closed tonight can mean that Italian restaurants in general are closed on the night of utterance, without knowing anything more about what kind of night it is (e.g. that it is Italy’s independence Day). Intuitively, the infelicity of temporally restricted IS sentences as in (41) results from their incompatibility with this default “indexical” reading of the temporal adverbial.


As I will show in chapter (5), the traditional approach to characteristic genericity (reviewed in section (2) above) can offer no satisfactory explanation for this behavior of IS sentences, since there is nothing in the semantics of Gen (specifically neither its modal nature nor the universal binding of the situation variable) which is incompatible with an “indexical” reading of temporal modification. In addition, given the classic I-generic properties of BP sentences as in (38) as well as (42) above, the traditional approach cannot explain why they differ so drastically from their minimally contrasting IS sentences in this construction. Instead, I will claim that the behavior of IS sentences in this construction and the way they differ from BP counterparts are a special case of a more general semantic difference between the two kinds of sentences, one which can also be seen in more classical generic constructions, with individual-level or habitual predicates. The next sections are devoted to old and new observations concerning this type of difference. 3.2 Differences in the level of “law-likeness” and type of rules expressed by IS and BP sentences 3.2.1 The original intuition As shown above, the primary motivation for assigning minimally contrasting IS and BP sentences the same semantic structure is the intuition that these sentences express the same meaning in sentences like (45a, b), with habitual or individual level predicates: (45)

a. A dog barks. /has four legs. b. Dogs bark. /have four legs.

Nonetheless, there have been observations in the past about cases where BP and IS sentences differ semantically, as well as intuitive attempts to characterize these differences. Lawler (1973), Burton-Roberts (1977), and Krifka (1987) observed that minimally contrasting IS and BP sentences are not really synonymous. They claimed that generically interpreted IS subjects can felicitously combine only with “inherent” or “essential” properties, and the IS sentences express statements which


are “analytic,” “perceived as analytic” or “normative” in some sense. In contrast, these scholars noted that minimally contrasting BP subjects can combine with no problem with “accidental” or “nonessential” properties, and the BP sentences express “weaker” kinds of generalizations, which have a more descriptive or “inductive” flavor. The following sentences are used as examples of this distinction: (46) (47)

(48) (49)

a. Madrigals are polyphonic. b. Madrigals are popular. a. A madrigal is polyphonic. b. # A madrigal is popular. c. A football hero is popular. a. Rooms are square. b. # A room is square. a. Gentlemen open the door for ladies. b. A gentleman opens the door for ladies.

According to Lawler, the infelicity of the IS sentence (47b) is due to the fact that the IS subjects combine with the property “being popular,” which, unlike “being polyphonic” (as in (47a)), is not taken to be an essential, inherent property of madrigals. The perfect felicity of both (46a) and (46b) shows that, unlike IS subjects, BP subjects can felicitously be interpreted as generic, no matter whether they combine with inherent or non-inherent properties. A similar explanation is given to the contrast in (48a) and (48b). As Nunberg & Pan (1975) point out, (47c) shows that “inherent” is always considered relative to the subject involved. The idea is that this sentence is felicitous since although “being popular” is not taken to be an inherent property of madrigals, it is taken to be an inherent property of football heroes. Burton-Roberts also claims that the felicity limitations on IS sentences are related to the fact that they are restricted to asserting “definitional” generalizations which are perceived as “analytic.” A sentence like (47a), for example, means something like “To be a madrigal is to be polyphonic,” which sounds reasonable. In contrast, the infelicity of sentences like (47b) results from the fact that paraphrasing it as “To be a madrigal is to be popular” is a bizarre analytic statement. In addition, Burton-Roberts claims that an IS sentence as in (49a) carries a normative requirement with it, and will be used most naturally to tell men how to behave with ladies. In contrast, BP sentences express simply descriptive—non-analytic, and non-normative—generalizations,


e.g. (49a) expresses a descriptive or inductive generalization about all or most gentlemen. The “old intuition” about the difference between IS and BP sentences has not received much attention in more recent theories of genericity, and in general there have not been many serious attempts to integrate it into the formal representations of BP and IS generics. Krifka et al. (1995), for example, admit that “the distinction is clear, and manifests itself in striking results, but the underlying reason is not clear” (p. 13, ff#10). An exception is Krifka (1987), who takes the “old intuition” as evidence for his claim that only generically interpreted IS are necessarily characteristic (I-) generic, where the subject introduces a variable bound by a sentential modal Gen operator. The modality of this operator, Krifka suggests, leads to the lawlikeness expressed by IS sentences, and this conflicts with accidental, nonessential properties. In contrast, Krifka claims, the compatibility of BP with predicates expressing accidental properties proves that they can be kind-referring terms in Proper Kind-Predication structures, which are not associated with a modal, law-like generic quantifier. A similar claim was made more recently by Dobrovie-Sorin & Laca (1996). These claims, however, are problematic given the fact that like other BP sentences in this construction, the BP sentences (47a), (48a) and (49a) support counterfactuals statements (as in (50a, b, c) respectively), and thus seem to involve some kind of modality as well: (50) a. If this were a madrigal, it would be popular. b. If this were a room, it would be square. c. If he were a gentleman, he would open the door for ladies. In the chapters below I show in more detail that the differences between minimally contrasting IS and BP sentences cannot be explained in a simple manner in terms of the I/D genericity distinction. 3.2.2 Problems with the original intuition While I believe that the old observations are fundamentally right, on the intuitive level they stand now they are not very helpful. It is not clear, for example, what is the exact meaning of “inherent” and “accidental” properties. Moreover, assuming that habitual BP sentences as in (46a) (48a) and (49a) express “nonaccidental generalizations” (as indicated by


their supporting counterfactuals property), it is not clear what it means for them to predicate “accidental” properties of their subjects. More serious, the claim that IS can only combine with “essential” properties seems to be false. Cohen (2001) correctly points out that there are many cases of felicitous IS sentences which express highly contingent, nonessential properties of their subjects, as in (51): (51)

a. A carpenter earns very little. b. A gold cube weighs 1.2 Kg. c. A Volvo costs NIS 150,000.

All these sentences are clearly not “analytic,” and the properties expressed in them are indeed contingent—they can clearly be false in other possible worlds, and even in the actual world in other times or cultures. We don’t want to paraphrase (51a) as “To be a carpenter is to earn very little,” or (51c) as “To be a Volvo is to cost NIS 150,000.” The sentences in (52) illustrate this point even better: (52)

a. A donkey has short ears. b. A dog has three legs. c. A government minister earns very little.

Sentences (52a)-(52c) are all clearly false. It is clear, then, that the properties expressed by their VPs are not taken to be “essential.” For example, it is clear that we do not take “has three legs” to be an essential property of dogs. Nonetheless, and unlike the sentences in (47b) and (48b) above, these sentences are perfectly felicitous as generics. In addition to all these problems with the “old intuition,” there is also a new body of data concerning semantic and distributional differences between BP and IS sentences which cannot be straightforwardly handled by the “old intuition..” This data will be discussed in the following sections. An examination of this data will eventually help us to refine the “old intuition” and better characterize the difference between IS and BP sentences.


3.3 Unexpected differences between minimal pairs of IS and BP sentences with subjects expressing “extremely unnatural classes” In his discussion of the notion “kinds,” Chierchia (1998) claims that BP are interpreted generically (and not existentially), as opposed to existentially, only in cases where they denote “natural” kinds, which he characterizes as follows: From an intuitive, pretheoretical point of view, kinds are generally thought of as regularities that occur in nature […] To any natural property, like the property of being a dog, there corresponds a kind, viz. the dog kind…. By natural kinds I do not necessarily mean, in the present context, just biological ones, or even “well established” ones. Artifacts (like chairs or cars) or complex things (like intelligent students or spots of ink) can qualify as kinds, to the extent we can impute to them a sufficiently regular behavior. (p. 348) [emphasis mine] However, examining IS subjects expressing “extremely unnatural classes,” and comparing them with minimally contrasting BP subjects (as in (53a-56a) and (53b-56b), respectively) shows that Chierchia’s claim is much more suitable to the former than to the latter.16 Consider, for example, the pairs in (53)-(56):

(53) a. A Norwegian student whose name ends with ‘s’ or ‘j’ wears thick green socks. b. Norwegian students whose names end ‘s’ o r ‘j’ w ear t green socks. (54) a. A tall, left-handed, brown-haired neurologist from France earns more than $ 150,000 a year.

16

Chierchia (1998) introduces examples such as People sitting here or Parts of this machine, which are already discussed in Carlson (1977), and follows Carlson’s claim, based on the behavior of these BP noun phrases in various “opacity tests” that they do not refer to kinds. Somewhat similar examples are discussed in McNally (1995) and Glasby (1995). I return to these examples in chapter 6.


b. Tall, left-handed, brown-haired neurologists from France earn more than $150,000 a year. (55) a. An American linguist born to a famous singer in June, 1954, writes very technical papers. b. American linguists born in June, 1954, to famous singers write very technical papers. (56) a. A relatively small lion called “Leo” eats cornflakes for breakfast. b. Relatively small lions called “Leo” eat cornflakes for breakfast.

Both the BP and the IS subjects in (53)-(56) express “extremely unnatural classes,”17 however there is a clear difference in the interpretation of the sentences, depending on whether these “unnatural classes” are expressed by IS or by BP subjects. The immediate reaction of all informants to the IS sentences in (53a)-(56a) is something like “Ah, this is a story about some Norwegian student/neurologist/ American writer/a lion called “Leo. “In other words, the informants immediately gave the IS subject an existential interpretation.18 In contrast, these informants all understood the BP sentences as generalizations about Norwegian students/neurologists/American linguists/lions called “Leo,” i.e., although the informants thought that these BP sentences expressed bizarre generalizations, they had no problem understanding these sentences as generic.19

17

At this point I am neutral with regard to the debate about the semantic nature of BP, i.e. whether they are indefinites (denoting sets), ambiguous between indefinites and kind referring, or kind referring only). I therefore use the term “classes” (as in extremely (un)natural classes) pre-theoretically. I return to this debate in chapter 6. 18 A similar claim is made independently in Dobrovie-Sorin & Laca (1996). These authors, however, attribute the existential reading of IS sentences to the length of the IS NP (in their terms to “enriching the description” of the set denoted by the IS NP), and offer no explanation for this fact. As I will show below, however, it is the “unnaturalness” of the description and not its length, which matters here, and which is given a formal, explanatory account.


The claim that the relevant factor influencing the interpretation of the IS subjects in (53a)-(56a) is the “extreme unnaturalness” of the subjects is supported by the fact that when we replace these subjects with subjects expressing “natural” classes as in (57a)-(60a), the interpretation of the IS subjects is much more easily generic, and is similar to that of its BP counterpart in (57b)-(60b): (57) a. A Scottish/An Israeli/An Australian soldier/A British football player/A rain-forests researcher/A teacher wears thick green socks. b. Scottish/Israeli/Australian soldiers/British football players/ Rain-forests researchers/Teachers wear thick green socks.

(58) a. A neurologist/A lawyer/A football manager earns more than $150,000 a year. b. Neurologists earn more than $150,000 a year.

19

It is important to emphasize that the existential readings of the IS subjects discussed here, as in (53a-56a), are very different from the ones noted by Carlson (1989), and discussed in section (2) above, as in the prominent reading of (i) and (ii):

(i)

A computer routes a modem plane (prominent reading: Modern planes are generally routed by some computers) (ii) A little red light comes on when you push this button (prominent reading: Generally, when you push this button, there is a little red light which comes on) Unlike Carlson’s examples, where the existential reading is a result of an additional splitting of the sentence, i.e. where there is an overt element in the sentence besides the subject, which is interpreted generically (modern planes, and the situations where you push this button, respectively), in the IS sentences in (53a-56a) above, no such generically interpreted element is present. Sentence (53a), for example, repeated here as (iiia) is clearly not a generalization about thick green socks, as in (iiib):


(59) a. A biologist/A logician/A mathematician writes very technical papers. b. Neuroimonologists write very technical papers. (60) a. A lion/An American child/A movie star eats cornflakes for breakfast. b. American children eat cornflakes for breakfast.

(iii) (a). A Norwegian student whose name ends with ‘s’ or ‘j’ wears green thick socks (b). # Gen s, x [green thick socks (x) C(s, x)] y [Norwegian student (y) wear (s, y, x)] (# “A general property of thick green socks is: some Norwegian student whose name ends with ‘s or j’ wears them”) Instead, the prominent existential readings of these sentences is where the existential quantifier has wide scope over a generic operator binding only the situation variable. For example, sentence (53a) can be paraphrased as “There is some Norwegian student whose name ends with ‘s’ or ‘g’ who regularly (in all contextually relevant situations) wears green thick socks.” In addition, unlike what happens in the example discussed here, where the prominent existential interpretation appears only with the IS, and not with the BP subjects (compare again (53a-56a) with (53b-56b)), in Carlson’s example, the prominent existential interpretation of the subject is the same, no matter whether it is an IS, as in (i) and (ii) above, or a BP as in (i’) and (ii’):

(i’) Computers route modern planes (prominent reading: Modem planes are generally routed by some computers) (ii’) Little red lights come on when you push this button (prominent reading: Generally, when you push this button, some little red lights come on) The existential readings of IS sentences discussed here, then, cannot be covered by the same mechanism suggested by Carlson (1989), namely an alternative “splitting” of the sentence into restriction and scope. They are thus truly “unexpected,” and require a different sort of explanation.


3.4 Unexpected differences between minimal pairs of IS and BP sentences with VPs expressing “Extremely Unconnected Properties” or extremely unconnected properties The unexpected existential reading of IS subjects as opposed to the generic reading of their BP counterparts shows up not only with “extremely unnatural class” subjects (like “Norwegian students whose names end with ‘s’ or ‘j’…”), but also when the subjects express more “natural” classes, as seen in (61)-(63): (61) a. A carpenter in Amherst gives all his sons names ending with ‘a’ or ‘g.’ b. Carpenters in Amherst give all their sons names ending with ‘a’ or ‘g.’ (62) a. An American lawyer writes a thick book every three years about the life of the Jewish sailors in the Middle Ages. b. American lawyers write a thick book every three years about the life of the Jewish sailors in the Middle Ages. (63) a. A famous semanticist sings German arias in the shower. b. Famous semanticists sing German arias in the shower. The existential readings of the IS subjects in (61a)-(63a) seem to result from the fact that the predicates in these sentences seem to be extremely “unconnected” to the classes expressed by the subjects, and not merely from their “accidentalness.” This is still an intuitive observation, which I will define more formally and precisely in chapter 2. But in the meantime we can see that if we preserve the subjects in (61)-(63), and insert other predicates, which are still “accidental” of their subject, but are felt to be more “connected” to them, the IS subjects are much more freely interpreted generically, just like their BP counterparts: (64) a. A carpenter in Amherst earns very little. b. Carpenters in Amherst earn very little. (65) a. An American lawyer pays only 20% income tax a year. b. American lawyers pay only 20% income tax a year. (66) a. A famous semanticist gets at least ten requests for reviews a year. b. Famous semanticists get at least ten requests for reviews in a year.


Crucially, both (61)-(63) and (64)-(66) are the same in that the predicates seem to express highly contingent properties of their subjects, i.e. in both sets of sentences neither of the predicates expresses properties which are known to be essential to the subjects, and there is no problem imagining other situations, worlds, cultures or periods in which e.g. carpenters in Amherst earn a lot, American lawyers pay 46.8 income tax, or where famous semanticists get at least 30 requests for reviews (instead of ten). Rather, what seems to distinguish the predicates in the first set of sentences from those in the second set is indeed the reasonableness of their connection, given our real world knowledge, with the properties expressed by the subjects. In chapters 2 and 3 I formalize this intuition and the reasons for the different behavior of IS and BP subjects with such predicates. In the meantime it is important to keep in mind that an explanation in terms of essential/ accidental properties is not sufficient here. 3.5 The different effect of the presence/absence of contextual support on the felicity of BP and IS sentences 3.5.1 “Out of the blue” vs. contextually supported IS and BP sentences In the previous sections we saw two cases where listeners unexpectedly interpreted IS subjects existentially. However, it is important to notice that the generic reading is not completely blocked for these IS subjects, but rather this reading is more prominent and is given as an immediate reaction by listeners who hear these sentences out of the blue. Nonetheless, a speaker can clearly choose to express a generalization using such IS sentences. Moreover, there are factors which make the generic reading much more salient for such IS sentences even for listeners The clearest such factor is contextual support. Thus, if the IS sentences in (53a)-(56a) above are uttered as continuations to the beginnings in (67a)-(67d), instead of out of the blue, informants report that the prominent existential reading disappears and a prominent generic reading appears:


(67) a. There are very interesting traditions in Norway concerning clothing, professions and last names. For example, a Norwegian student whose name ends with ‘s’ or ‘j’ wears thick green socks. b. The new health minister has bizarre salary criteria, for example, from now on, a tall, left-handed, brown-haired neurologist from Canada earns $150,000 a year. c. Joshua Greenberg was a famous and admired carpenter in Amherst a hundred years ago. From that time until now a carpenter in Amherst gives all his sons names ending with ‘a’ or ‘g.’ d. The annual semantics conference in Germany has always invited all famous semanticists to lecture, and offers, besides interesting discussions in semantics, a varied cultural program and daily performances of opera singers. This may be the source for the joke saying that a famous semanticist sings German arias in the shower. A similar effect of contextual support can be also seen with some of Burton Robert’s original examples. Thus, (68b) is much better than the questionable outof-the-blue (68a): (68)

a. # A room is square. b. In Japan, a room is square. 3.5.2 Contextual support does not always help: IS sentences in pure inductive scenarios

Notice, however, that not any kind of context has this effect on the interpretation of IS sentences, and that even as a speaker, there seem to be some restrictions on the choice of the IS sentences above to express generalizations. In contrast, these restrictions are not present in the case of the minimally contrasting BP sentences, and as before they are freely interpreted as generics. An example of such a context is embedding under the verb notice as in (69a)-(71a). In such a case the salient reading of “problematic” IS subjects (i.e. those expressing “extremely unnatural sets” or those combining with “unconnected properties”) is again existential, despite the fact that they are not uttered out of the blue. As can be seen in


(69b)-(71b)), the minimally contrasting BP subjects continue to have a generic reading with no problem: (69) a. While sitting in Kew Gardens I noticed that an old man wearing a red shirt bends under his bench whenever an airplane flies overhead. b. While sitting in Kew Gardens I noticed that old men wearing red shirts bend under their benches whenever an airplane flies overhead. (70) a. When I was a zookeeper I loved watching the lions. I noticed that a relatively small lion called Leo eats cornflakes for breakfast. b. When I was a zookeeper I loved watching the lions. I noticed that relatively small lions called Leo eat cornflakes for breakfast. (71) a. Are you looking for a book about Jewish sailors in the Middle Ages? Last week I noticed that a thick book with a red paperback cover deals exactly with this topic. b. Are you looking for a book about Jewish sailors in the Middle Ages? Last week I noticed that thick books with red paperback covers deal exactly with this topic. What is it about the verb notice, and about the contexts in (69)-(71) which differ from the contexts in (67) above, and which block the expected generic reading of the IS subjects? Unlike the contexts in (67), those in (69)-(71) do not suggest any prior knowledge about the set denoted by the subjects. Rather, these contexts are purely “inductive” in nature. That is, in each case the speaker seems to reach the generalization after observing several instances of the set denoted by the subjects (people in Kew Gardens, thick books with red paperback covers, etc.) who have the property expressed by the predicate (bending under their benches, dealing with Jewish sailors in the Middle Ages, etc.). The verb notice seems to trigger such pure inductive reasoning scenarios, in which the only relevant information known to the speaker (and the listener), and the basis for the generalization, is what he or she observes. It seems, then, that for some reason pure inductive reasoning scenarios are very well suited to support generic BP sentences, but not IS sentences which express bizarre generalizations. Trying to construct such “pure inductive scenarios” and comparing the behavior of BP and IS sentences in them seems to support this conclusion. Imagine, for example, that as a lion doctor I get to see many


lions in many zoos around the world. Going over the list of diets for the lions I am taking care of, I notice that there are surprisingly many small lions called “Leo” who prefer to eat only cornflakes for breakfast. In contrast, only a few such lions called “Simba” or “Mufasa” in my list like cornflakes. I have no idea why this is the case, but telling my husband about this bizarre correlation, I would tend much more to use the BP sentence in (78b), than the IS one in (72a): (72) a. Wanna hear something funny? A relatively small lion called “Leo” eats cornflakes for breakfast. b. Wanna hear something funny? Relatively small lions called “Leo” eat cornflakes for breakfast. In other words, the “inductive” situation described above seems enough to support the generalization expressed by the BP sentence, but not the one expressed by the IS one. The IS sentence sounds inappropriate here, whereas the BP one is perfectly natural. It seems that the IS sentence expresses a stronger type of generalization, or that we need something else, in addition to observing a number, even a large one, of lions eating cornflakes for breakfast in order to use the IS sentence. Let me emphasize that I do not claim that in “inductive” scenarios, as the one described above, BP sentences are necessarily true, only that, unlike what happens with IS sentences in such contexts, they are perfectly felicitous. For example, looking at more relatively small lions called “Leo,” it may turn out that my generalization about the eating habits of such lions is in fact wrong. Regardless of the question of truth, however, the inductive scenario in this case is a much more appropriate basis for a generalization expressed through a BP than through an IS sentence. 3.6 IS and BP sentences which do not differ so much Let me briefly summarize the data examined above: We saw that there are some cases where IS and BP subjects get different readings in habitual sentences. Specifically, IS sentences expressing bizarre generalizations—those with subjects denoting “extremely unnatural sets” or with VPs denoting “unconnected properties”—unexpectedly get a salient existential interpretation, whereas their BP counterparts are interpreted generically as expected. This existential reading of such IS subjects is especially salient in “out of the blue” utterances (i.e. in the


null context) or in purely inductive contexts. In contrast, such subjects get a more salient generic reading where uttered in different kinds of contexts, intuitively, those assuming some-prior knowledge about the sets denoted by the subjects. It is important to emphasize, though, that the story of differences just sketched is true only of some sorts of IS and BP subjects pairs, specifically those with unnatural subjects, or “unconnected properties.” With other kinds of IS and BP sentences, there seems to be much less difference in distribution and interpretation. Consider once more the pairs in (73) and (74): (73) (74)

a. Dogs have four legs. b. A dog has four legs. a. Whales are mammals. b. A whale is a mammal.

Although there is a slight difference in interpretation between the minimally contrasting IS and BP sentences (the latter still have a more “analytic” “definitional” flavor than the former, which can have a more “descriptive” flavor), both the IS and BP subjects in these sentences are naturally interpreted generically, as expected, although they are all uttered “out of the blue.” In addition, in many cases both IS and BP sentences can easily express generalizations supported by pure inductive contexts. For example, imagine I am a zoologist examining spiders in the rain forest. One day I find a spider of a new species, which I immediately name “Greenberg spiders.” Among other things, this little spider has 23 eyes. The next day I find two more such spiders, and the week after I find 16 more “Greenberg spiders” in a far-away forest, and all have 23 eyes. At this point I send an abstract to the upcoming conference of the RSO (Rain-forest Spiders Organization), presenting my new results. Crucially, my findings about the number of eyes of Greenberg spiders can be stated using both the BP sentence in (75a) or the IS one in (75b), i.e., although the generalization seems to be based on pure inductive reasoning, both the IS and BP sentences are felicitous as generic: (75)

a. Greenberg spiders have 23 eyes. b. A Greenberg spider has 23 eyes.

Similarly, if I watch a documentary film about the communities of ultraOrthodox Jews, and see one woman after another with her hair covered,


I can use both the IS or BP forms in (76) to express a generalization about ultra-Orthodox Jewish women: (76)

a. An ultra-Orthodox Jewish woman covers her hair. b. Ultra-Orthodox Jewish women cover their hair.

One may attribute the similarity between the BP and IS sentences in (75) and (76) to the fact that the predicates in these sentences express “essential” properties of spiders or ultra-Orthodox Jewish women. However, the similarity of IS and BP sentences in inductive scenarios also appears with highly contingent properties, which can easily change. Thus, for example, if I want to buy a new refrigerator, and go to several stores to see the make and price of the refrigerators they have on display, I can summarize my findings naturally by either (77a) or (77b): (77)

a. A refrigerator costs between NIS 4000 and 7000 b. Refrigerators cost between NIS 4000 and 7000

To conclude, besides the need to understand the exact nature of the differences between certain cases of IS and BP sentences, and the reason for these differences, this section indicates that we also need to understand why in other cases BP and IS seem to behave in a similar way. 4. Parallels between the IS/BP distinction in temporally restricted and classic generic constructions In section (3.1) above, I examined the way IS and BP sentences differ when modified by specific temporal adverbials, and claimed that these differences are a special case of a more general difference between the two kinds of sentences, shown also with individual-level or habitual predicates, examined in sections (3.2), (3.3), (3.4) and (3.5) above. Before describing my intuition concerning the source of this general difference, let me first point out how the similarities in the way IS and BP sentences differ in the temporally restricted and the classical generic constructions. There are four such relevant similarities. The first and most general similarity is that in both constructions the distribution of felicitous IS sentences, under the generic interpretation,


is much more restricted than the corresponding BP sentences, which seem to be interpreted generically very easily. Secondly, context improves the status of problematic generic IS sentences in both constructions. Thus, I have shown that IS subjects combining with temporally restricted predicates, as well IS subjects in habitual sentences, denoting “extremely unnatural classes” or combining with “extremely unconnected properties” have a salient existential reading when uttered “out of the blue,” but a salient generic reading when contextually supported. This can be seen again in (78) and (79): (78) a. An Italian restaurant is closed tonight. (salient existential reading) b. It’s Italy’s Independence Day. An Italian restaurant is closed tonight. (a more salient generic reading) (79) a. A Norwegian student whose name ends with ‘s’ or ‘g’ wears thick green socks. (salient existential reading) b. There are interesting traditions in Norway concerning names, professions and clothing. For example, a Norwegian student whose name ends with ‘s’ or ‘g’ wears thick green socks. (a more salient generic reading) A third similarity between the two constructions is that not every kind of contextual support had this effect on IS sentences. Specifically, in both constructions a pure inductive reasoning context was inappropriate as supporting most generic readings of those IS subjects which combine with temporally modified predicates as well as the IS subjects in the “problematic” habitual sentences. In both construction BP sentence were freely interpreted as generic in such inductive contexts. This can be seen again when both kinds of sentences are embedded under the verb notice: (80) a. Walking down the street I noticed that a small Italian restaurant is closed today. (funny as generic; fine as existential) b. Walking down the street I noticed that small Italian restaurants are closed today. (fine as generic) (81) a. While sitting in Kew Gardens I noticed that an old man wearing a red shirt bends under his bench whenever an airplane flies overhead. (funny as generic; fine as existential)


b. While sitting in Kew Gardens I noticed that old men wearing red shirts bend under their benches whenever an airplane flies overhead. (fine as generic) Finally, in both constructions there are cases where the difference between IS and BP sentences seem to be most appropriately described in terms of the “old intuition,” i.e. as a difference between analytic/ normative statements, expressed by IS sentences, as opposed to descriptive generalizations, expressed by BP sentences. Thus, the felicitous generic IS sentences in both (82) and (83) seem indeed to carry some normative, “definitional” or “analytic” flavor, whereas the BP sentences in (84) and (85) seem indeed to express “weaker” descriptive generalizations: (82) An American eats turkey tonight. (normative/analytic) (83) A gentleman opens the door for ladies. (normative/analytic) (84) Italian restaurants are closed tonight. (on the night of utterance) (descriptive) (85) Carpenters in Amherst give all their sons names ending with ‘a’ or ‘g.’ (descriptive) I believe that these similarities between the behavior of IS and BP sentences in the temporally restricted and habitual constructions indicates that the differences between IS and BP in both constructions are indeed a result of the same underlying semantic difference between the two constructions. In the next section I set out my intuition as to what this underlying difference is. 5. The basic intuition 5.1 Carlson’s (1995) distinction between the “inductive” and the “rules and regulations” approaches to (I-) genericity My intuition concerning the difference between IS and BP sentences is to a large extent inspired by a notional distinction described in Carlson (1995). Carlson describes two approaches to characteristic (I-) genericity: the “inductive” one, which sees generics as based on


observations concerning specific instances individuals or situations), and the “rules and regulations” approach,20 which takes generics to reflect abstract causal structures. Formally, the first approach represents generics as involving quantification over particular entities (individuals and situations), whereas the second uses rules as primitives and involves no quantification at all. Carlson is interested in determining which approach is better for representing natural language genericity (although he admits that “one must eventually allow for mixed or intermediate positions” (p. 224)). To do so he examines the empirical advantages and disadvantages of these approaches. Sentences like (92a-d) seem to be best analyzed as expressing some “rules and regulations,” about Mary’s job, the function of this machine, the order of speaking, and the way to play chess: (86)

a. Mary handles the mail from Antarctica. b. This machine crushes oranges. c. The Speaker of the House succeeds the Vice President. d. Bishops move diagonally.

Carlson claims that such sentences seem to pose a problem for the “inductive/quantificational” approach, since they are not necessarily based on observations of particular instances. For example, (86a) can be true even if mail from Antarctica never arrives, (86b) even if this is a new machine which hasn’t crushed a single orange yet, etc. Such sentences indeed seem to express some sort of law, rule or disposition, involving a causal relationship. In contrast, sentences like (87a-d) are naturally accounted for by the “inductive approach”: (87)

a. The sun rises in the East. b. Dogs bark. c. Mary speaks quietly. d. John walks to school.

20 This approach follows the philosophical approaches to dispositions and laws as involving causation. See e.g. Lewis (1986a), Armstrong (1973) and Prior (1985).


These sentences can be naturally described as being based on observations of many specific instances (e.g. many cases where the sun rose in the East, many individual barking dogs, many cases where John walks to school, etc.). One may still claim that sentences like (93a) and (93b) can also be analyzed in terms of the “rules and regulation” approach (as expressing a rule about the sun, or a biological rule about dogs). But this is much harder to do in cases like (93c) and (93d), which according to Carlson are much more likely to express weak descriptive generalizations. It is hard to think of a primitive rule about our world requiring that John walks to school or that Mary speaks quietly. Unlike Carlson, who attempted to choose one approach to analyze the variety of genericity phenomena, I believe no such choice should be made. This is because there is evidence that natural language syntactically encodes both kinds of genericity (or very similar parallels of them). De-Swart (1994), for example, reports an observation originally made by Carlier (1989), according to which in French, the choice between NPs with the indefinite determiner des (as in (88a) and the definite one les (as in (88b)—both generically interpreted makes a semantic difference:21 (88) a. Des agents de police ne se comportenet pas ainsi dans une situation d’alarme “INDEF-PL police officers do not behave like that in an emergency situation”

b. Les agents de police ne se comportenet pas ainsi dans une situation d’alarme “DEF-PL police officers do not behave like that in an emergency situation” De-Swart points out that (94a) and (94b) would be naturally uttered in two different situations. The former “would be uttered to reproach a subordinate with his behavior (whereas the latter) does not have the same normative value, but give us a descriptive generalization” (p. 13). This distinction is reminiscent both of Carlson’s distinction as well as the “old intuition” reviewed above, regarding the differences between IS and BP sentences.

21

I thank Ariel Cohen for bringing these data to my attention.


In addition, as Greg Carlson pointed out to me (p.c.), similar distinctions have been observed in African languages (e.g. Welmer (1974), Arnott (1970) and Westermann (1930)). Welmer (1974), for example, observes that genericity in Swahili can be expresses either with or without a special marking on the verb (‘hu’), as in (89): (89)

ng’ombe hu-la nyasi “Cows eat grass”

With the special marking, the speaker makes a more “in principle” generalization about the nature of the entity denoted by the subject. In contrast, when genericity is expressed without this special marking, the generalization expressed is more “descriptive.” Thus, in uttering, e.g. (89) “one is not referring so much to the daily habit of cows as to the ‘nature of the beast’” (p. 360). My main claim is that a semantic distinction of this sort is syntactically encoded not only in French and Swahili, but in English as well, through the syntactic difference between IS and BP sentences.22 My main concern in the following chapters is to clarify the nature of this distinction using the tools of formal semantics, so it can be used to account for the wide range of differences and similarities between IS and BP sentences, noted above (as well as some more similarities and differences pointed out in the chapters below). The intuitive distinction is described in the next section. 5.2 The intuitive underlying difference between IS and BP sentences Here is what I think is going on with IS and BP sentences. We have two kinds of nonaccidental, modalized generalizations which are syntactically encoded in English, which I will henceforth call “in virtue of” and “descriptive” (these terms closely, although not precisely parallel Carlson’s “rules and regulation” and “inductive,” respectively). IS sentences can express only “in virtue of” generalizations. This means that they can only be asserted if the speaker has in mind, and the listener can accommodate (in Lewis (1983) terms), some relatively specific property associated with the property denoted by the IS subject, in

22

Cohen (1996, 1999b) independently makes a similar claim. I discuss differences between Cohen’s theory and mine in chapters 2, 3, and 5 below.


virtue of which, or because of which, every member of the corresponding set has the predicated property. For example, A boy doesn’t cry will be true, roughly, if there is some property we associate with the set of boys: a genetic property, or a social norm property, e.g. being tough, in virtue of which every member of the set of boys will not cry (in all contextually relevant situations). The choice of the “in virtue of’ property depends in a systematic way on the shared real-world knowledge about the property denoted by the subject (e.g. being a boy), and the one expressed by the VP (e.g. crying). Notice that many times a speaker can assert an IS sentence even if he or she only knows the “sort of property” which leads to the predicated property. So, for example, I can assert A Pentium III computer works very fast, as a computer expert, having in mind a specific sophisticated processor installed in these computers and making them work fast, but also as a layperson, meaning simply that “in virtue of their mechanical properties, Pentium III computers work very fast.” Crucially, however, if no appropriate “in virtue of” property, or “sort of property” is available to the listener on the basis of the shared real world knowledge, he/she will consider the IS sentence infelicitous. BP sentences can express both “in virtue of” generalizations, but, unlike IS sentences, also a second, “descriptive” kind of generalization. Descriptive generalizations assert, on the basis of several actual instances of the set having the predicated property, that “there is some pattern here.” In other words, the generalization is not accidental, i.e. it is not restricted to these actual instances only. In the case of e.g. Boys don’t cry, we can imagine an alien from Mars visiting our planet and watching the behavior of children. At some point, after seeing the behavior of several boys in several “tear-inducing situations,” the alien decides that there is some pattern here, and that the next boy in such a situation will not cry either. Crucially, at this stage he does not know or care, and his statement does not involve, not even in an implicit manner, the property which makes boys not cry. All he claims is that “this is not accidental,” i.e. it is not limited to the observed boys only. BP sentences, then, are perfectly felicitous even if no appropriate “in virtue of” property is available on the basis of the shared real-world knowledge. English syntactically encodes, then, two kinds of non-accidental generalizations. Unlike Carlson (1995) and Cohen (2001), however, I do not take the semantic difference between the two kinds of generalizations to be a difference between quantified structure vs. nonquantified structures, directly asserting the existence of some primitive


“rule.” Instead I am going to represent both kinds of generalizations as quantified, modalized tripartite structures, which differ in the kind of modality they involve, i.e. in the characterization of the set of accessible worlds relative to which the universal statement is evaluated. That is, although I agree with Carlson and Cohen’s claims that IS sentences express some sorts of “rules,” linguistically I do not take such rules to be “primitive irreducible entities” (in Cohen’s terms), but rather suggest that universal, modal statements can capture their meaning very successfully, once the relevant accessibility relation has been appropriately defined. 6. An outline of this work The rest of this work is structured as follows: I start with formally clarifying the semantics of present tense IS and BP sentences with individual level and habitual predicates. Chapter (2) is devoted to the semantics of IS sentences and “in virtue of” generalizations. Chapter (3) is devoted to the semantics of BP sentences, under their descriptive reading (descriptive BP sentences, for short). In both chapters I show first how the different accessibility relations successfully account for the various semantic pragmatic and distributional differences between IS and BP sentences. I also show that when combined with independently motivated conversational mechanisms, the difference in accessibility relation can be used to naturally and successfully account for additional, yet unobserved differences between IS and BP sentences in a variety of other contexts, as well as for the similarities between them in other contexts. In chapter (4), I turn to the way IS and BP sentences tolerate exceptional and irrelevant entities (individuals and situations). I reject the “normal ordering source” solution to the problem of exceptions, reviewed above, and suggest a new mechanism for tolerating exceptions for both IS and descriptive BP sentences, based on Kadmon and Landman’s (1993) suggestion for a “domain vague” restriction on the generic operator. I show how combining this mechanism with the two semantic structures developed in chapters (2) and (3) accounts for both similarities and newly observed differences in the way IS and BP sentences tolerate exceptions. In chapter (5) I show how the analysis of the underlying difference between IS and BP sentences can be used also to account for the


differences between them in temporally restricted constructions examined in section (3.1) above. In the final chapter, chapter (6), I summarize the different accessibility relations and presuppositions associated with the two kinds of generalizations, which were developed in chapters 2–5. I then examine several theories which suggested different interpretations for IS and BP noun phrases (mainly as indefinite and kind terms, respectively), and outline the direction in which this kind of difference can be integrated into a more explanatory theory of the differences in accessibility relations between IS and BP sentences. In this chapter I also relate to the kind of generalizations expressed by, and the accessibility relations available with, simple habitual sentences (like John walks to school).

CHAPTER 2 “In virtue of” Generalizations, and the Semantics and Pragmatics of IS Sentences

1. Introduction 1.1 Basic Characteristics of IS Sentences In the previous chapter we saw that IS sentences have special properties which distinguish them from their BP counterparts. The relevant properties are summarized here: A. IS sentences seem to express some sort of “in-principle” generalizations, with a strong level of “law-likeness,” sometimes called “definitional” or “analytic” (by Burton-Roberts (1977)). Thus, an IS sentence like A table has four legs has a prominent reading, meaning something like “In principle, to be a table is to have four legs.” In contrast the corresponding BP sentence Tables have four legs also has a natural reading expressing some “in-reality pattern, i.e. expressing a mere descriptive generalization about most tables (in addition to the possible “definitional”/“in-principle” reading). B. A related fact noted by Burton-Roberts is that IS sentences often express “normative/deontic” statements. Thus a sentence like A gentleman opens the door for a lady seems to be more normatively loaded than the corresponding BP sentence Gentlemen open the door for a lady—which again has a strong “descriptive” reading.


C. Another related observation is that there are cases where IS sentences are questionable when combining with VPs expressing “nonessential” properties, whereas the minimally contrasting BP sentences are perfectly felicitous with such VPs. Lawler (1973) claims that this “essentiality requirement” with IS sentences is the reason for the oddness of e.g.?? A madrigal is popular, as opposed to the felicity of A madrigal is polyphonic, and that the lack of this “essentiality requirement” with BP sentences is the reason for the felicity of both Madrigals are popular and Madrigals are polyphonic. Similarly, Burton-Roberts (1977) contrasts?? A room is square with Rooms are square (assuming that being square is not an essential property of rooms). Other such contrasts are between?? A man is blond and Men are blond, and between?? An uncle eats marshmallows and Uncles eat marshmallows. D. Nonetheless, as correctly pointed out by Cohen (2001), there are also many cases where IS sentences can freely combine not only with “essential” or “inherent” or stereotypical properties, but also with highly contingent ones, as in A refrigerator costs about NIS 8000. Moreover, IS subjects can sometimes combine with properties which clearly don’t hold of the subject set, and thus express clearly false generalizations, as in A dog has three legs. The felicity of these sentences clearly indicates that a simple explanation of the felicity conditions on IS sentences in terms of the essential or inherent nature of their predicates cannot work. A similar point is made by the perfect felicity of pairs of IS sentences with opposite properties which cannot be both essential of the same subject set, e.g. A lawyer earns very little and A lawyer earns a lot. E. IS sentences with subjects denoting “extremely unnatural properties” get an unexpected prominent existential reading, whereas their minimally contrasting BP subjects get a prominent generic reading (cf. again: A relatively small lion called “Leo” eats cornflakes for breakfast vs. Relatively small lions called “Leo” eat cornflakes for breakfast. F.

In some cases, IS sentences with VPs denoting “extremely unconnected properties” also get an unexpected prominent existential reading. Cf. again A carpenter in Amherst gives all his sons names ending with ‘a’ or ‘g’ vs. Carpenters in Amherst give


all their sons names ending with ‘a’ or ‘g,’ i.e. it seems that although the properties combining with IS can be contingent of the set, they should be reasonably “connected” with it in some sense. G. Notice, though, that the generic reading is not completely blocked for the last two kinds of IS sentences. Firstly, as a speaker it is possible for me to choose to express even funny generalizations using IS sentences. In addition, informants report that as listeners they find the IS sentences in (E) and (F) much better as generic when the sentences are not uttered out of the blue but against supporting context as in e.g.: Joshua Greenberg was a famous and admired carpenter who lived in Amherst at the beginning of the century. From that time until now, a carpenter in Amherst gives all his sons names ending with ‘a’ or ‘g’ (in which the IS sentence is much better as a generic). Context also helps in the case of the questionable IS sentences in (C), as can be seen from the difference between? A room is square and In Japan, a room is square or?? A madrigal is popular and These days, a madrigal is very popular. However, not any kind of context helps. Specifically with “pure inductive context.” and embedding under notice these subjects get again the prominent existential reading, as can be seen from the following sentence: While sitting in Kew Gardens I noticed that an old man wearing a red shirt bends under his bench whenever an airplane flies overhead (cf. While sitting in Kew Gardens I noticed that old men wearing red shirts bend under their benches whenever an airplane flies overhead—in which the generic reading for the BP subject is much more prominent.) The general impression emerging from the data above is that IS sentences have a “double nature”: On the one hand they seem to express generalizations which are more “law-like” and less “accidental” than the ones expressed by BP sentences. In other words, in some sense, the BP sentences seem to describe an actual aspect of our world, or some “in reality pattern,” whereas the IS sentences seem to describe some rule in our world, or some “in principle pattern.” On the other hand, unlike BP sentences, which can felicitously express bizarre generalizations, the felicity of IS sentences heavily depends on real world knowledge, and on what is or isn’t considered reasonable in the actual world. In other words, IS sentences express


generalizations which are less accidental than BP ones, but their felicity is more dependent on real-world knowledge than that of their BP counterparts. The purpose of this chapter is to develop truth conditions which will capture this double nature of IS sentences, and more specifically, which will account for the full range of facts in (A-H) above. 1.2 The basic intuition The main intuition I wish to develop in this chapter is that IS sentences necessarily express what I call “in virtue of” generalizations. This means that an integral part of the meaning of these sentences is having in mind some appropriately chosen property or aspect of our world, in virtue of which the generalization they express is true. In other words, IS sentences express, as part of their semantics, that a certain generalization is nonaccidental, and also convey in virtue of which property or aspect of our world, this generalization is true, where the choice of this “in virtue of” property is pragmatically constrained. Thus, part of fully understanding an IS sentence is to assume that its speaker has some “in virtue of” property in mind, and to try to accommodate it. Thus, fully understanding an IS sentence involves assuming that its speaker has some “in virtue of” property in mind and trying to accommodate it. This “in virtue of” reading is possible for BP sentences too, but it is not their only, or most prominent reading. Let me illustrate what I mean. Imagine you hear, out of the blue, the IS sentence in (1) and its BP counterpart in (2): (1) (2)

A woman in this place does not walk alone outside. Women in this place do not walk alone outside.

My claim is that understanding (1) you not only understand that every (relevant) woman in this place does not walk outside, and that this generalization is not accidental. You also assume that the speaker had some property or aspect of our world, in mind, in virtue of which s/he took the generalization to be true. Understanding the sentence requires you to try to accommodate this “in virtue of” property. Different choices of such a property will lead to different meanings of this sentence. The speaker of (1) can mean, for example, that in virtue of being so modest and religious, the women in this place never walk alone outside. Or it can mean that the women here do not walk


alone outside since they are so beautiful and pleasant, that they are always accompanied by men. Alternatively, (1) can mean that the women in this place don’t walk alone outside since this place is a very dangerous one for women, and there are many murders and rapes. Other such “in virtue of” properties are possible, and without supporting context a listener may accommodate a different “in virtue of” property than the one meant by the speaker. We can imagine, for example, the following dialogue: A: An accountant pays a very low percent of income tax. B: Really? I’d better become an accountant! In virtue of which law or regulation this is true? C: Regulation? No! This is not what I meant. It’s not because of any regulation, but because accountants know so much about taxes that they take advantage of all the legal possibilities and end up paying very little. Misunderstandings, then, may occur, but the crucial claim is that when uttered in context the listener assumes that the speaker indeed has some “in virtue of” property in mind. Without assuming this property the sentence cannot be fully comprehended. Something is missing in its understanding. In contrast, with the BP sentence in (2), it is not at all necessary to have such an “in virtue of” property in mind. A legitimate reading of (2) is one where it simply asserts a descriptive generalization, made, e.g. after observing that no, or very few women in this place walk alone outside. In such a case, the meaning of (2) is something like “there is a pattern here, or “the fact that women in this place do not walk alone outside is not accidental, where crucially, and in contrast to the IS sentence in (1), sentence (2) does not concern, not even in an implicit way, the aspect or property in our world which makes this generalization true. Understanding (2), then, does not require the listener to try to accommodate such an “in virtue of” property. 1.3 The Direction of Formalization: “in virtue of” properties restricting the accessibility relation of IS sentences In this chapter I develop and empirically motivate the intuitive claim that an “in virtue of” property is indeed an integral part of the meaning


of IS sentences. I then deal with making this intuition precise, by integrating an appropriately chosen “in virtue of” property into the truth conditions of the IS sentences. Concerning this last goal, my main claim would be that the “in virtue of” property dictates and restricts the choice of the set of possible worlds which are relevant for the evaluation of IS sentenees, i.e. it restricts the accessibility relation of the modal generic operator of IS sentences. Among other things, this idea is supported by comparing IS sentences with nongeneric sentences and sentences with overt modal operators. Above I claimed that the IS sentence in (1) can mean different things, depending on which “in virtue of” property is chosen. A similar thing happens with (3), cited and discussed by Kratzer (1981): (3)

I cannot play the trombone.

Kratzer writes: When we talk to each other, we hardly ever make explicit in view of which circumstances something should be necessary or possible […] Depending on the situation in which I utter this sentence I may say quite different things. I may mean that I don’t know how to play the trombone […] Or suppose I suffer form asthma. I can hardly breath. In view of my physical condition I am not able to play to trombone, although I know how to do it. I may express this by uttering (3). Or else imagine that I am traveling by sea. The ship sinks and so does my trombone. I manage to get to a lonely island and sadly mumble (3). I could play the trombone in view of my head and my lungs, but the trombone is out of reach. (p. 54) Sentence (3) is similar, then, to (1), in expressing different meanings depending on the “in view of” or “in virtue of” part. This part is taken by Kratzer to express the conversational background, or the accessibility relation in the modalized sentence. A sentence like (3) illustrates Kratzer’s “circumstantial modality,” which differs from epistemic modality in that not all known facts, but only some of the facts in the actual world (my physical condition, the shape of my trombone, my knowledge of trombone playing, etc.) are taken into consideration in forming the conversational background. This idea of circumstantial modality is further developed by Brennan (1993), (reviewed below, section (3.2)) who shows that the


accessibility relations which operate in the semantics of all nongeneric modalized sentences with “root” interpretation are systematically “keyed to the properties of the subjects.” This means that the chosen facts about the actual world, which restrict the accessibility relation, are always facts about the entity/individual denoted by the subject. Thus, unlike more traditional accessibility relations (e.g. epistemic, deontic, or buletic), the circumstantial accessibility relation systematically varies depending on the subject of each sentence. I will claim that IS sentences too involve accessibility relations which are “keyed to the properties of the subjects,” and which thus also systematically vary from sentence to sentence. To do so I will start by following Heim (1982), Krifka (1995), Krifka et al. (1995), Chierchia (1995), Wilkinson (1991), among others, who take IS noun phrases to be indefinite expressions, introducing variables into the representation, and sentences like (1) to be represented as tripartite structures headed by a universal, unselective operator with modal nature, which binds individual, situation, and world variables. The Chierchian-style representation of (1) in (4a) (paraphrased in (4b)), is a typical representative of this approach (where d and s are variables for individuals and situations, respectively, Wf is a variable for accessible worlds (given some modal base f), and C is a contextually filled relation, which produces contextually relevant situations involving individuals in worlds): C(s, d, Wf)] → ¬ (4) a. d, s, wf[woman in this place (d, wf) walk alone outside (d, s, wf) b. Every contextually relevant situation, involving every woman in this place, in all worlds accessible from w w.r.t. f is a situation in which such a woman does not walk alone outside in such an accessible world. My claim is that the special thing about IS sentences, which is missing in (4), is an “in virtue of” property which systematically restricts the set of accessible worlds, and thus constitutes part of the accessibility relation (modal base) of an IS sentence like (1) above. A more accurate paraphrase of (1), then, will be the following: In all (relevant) worlds, in which every woman in this place is so religious and modest (OR so beautiful and charming OR is cautious because of the frequency of violence in this place, etc.), it holds that every contextually relevant situation involving


every woman in this place is a situation in which every such woman does not walk alone outside. 1.4 The pragmatic aspects in the semantics of IS sentences and their formalization I will show, furthermore, that not any arbitrary property can function as the “in virtue of” property, restricting the accessibility relation in IS sentences, and that the choice of such an appropriate property is systematically restricted by our real-world knowledge about the properties denoted by the IS subject and the VP and the context. These real-world knowledge requirements on the choice of the “in virtue of” property are the source of the pragmatic sensitivity of IS sentences, mentioned above, and the reason that, unlike BP sentences, all felicitous IS sentences—both true and false—are restricted to express generalizations which are conceived as “reasonable,” or not too farfetched, given our real-world knowledge. The infelicitous IS sentences are those in which the real-world knowledge requirements on the choice of the “in virtue of” property are not met. Thus, there is a pragmatic/contextual factor which plays a systematic role in the semantics of IS sentences, which we will not find with the corresponding BP sentences. To represent this aspect of IS sentences I will assume that, in addition to the C variable in (4a) above, which restricts situations to contextually relevant ones, we have to add to the representation a variable C for the set of possible, appropriately chosen “in virtue of” properties. The presence of this variable is constant in the truth conditions of all IS sentences, but its content varies from sentence to sentence, and from context to context. Let us call the “in virtue of” property ^S. Taking the IS subject to denote a property ^P, the VP to denote ^Q, and C to be a contextually determined set of properties, the general form of the truth conditions for IS sentences will be (5), where the content of C is not specified yet (where d, s, and w are variables for individuals, situations, and worlds, respectively): (5) An IS sentence, of the form IS VP is true in w iff:


Paraphrase: An IS sentence is true in w iff, there is a property ^S, s.t. in all worlds w′, in which every P individual has S and that ^S is a member of a certain set of properties C, every contextually relevant situation, involving every P individual, in w′, is a situation where every such P individual has Q in w′. For example, a sentence like (1) above will now be represented like (6): (6) A woman in this place does not walk alone is true in w iff

Paraphrase: The sentence is true in w iff, there is a property, such that in all worlds w′ in which every woman in this place has this property, and where this property is a member of a contextually determined set of properties C, every contextually relevant situation involving every such woman in w′ is a situation where such a woman does not walk alone outside in w′. Much of this chapter will be devoted to specifying what the contextually determined set C is, i.e. what are the real-world knowledge requirements that an appropriate “in virtue of,” ^S, property has to meet. Before I deal with this question, though, I will turn (in section (2)) to show that the whole idea of an “in virtue of” property restricting the modal base is indeed empirically motivated. In the next section I examine several theories of genericity, and show that they cannot account for the special properties of IS sentences in a satisfactory way. I then turn (in section (3)) to review two theories which, although do not directly deal with the special properties of IS sentences, inspire the intuition about the “in virtue of” nature of such sentences. I first review Chierchia’s (1995) suggestion of defining modal bases of generic sentences in terms of “felicity conditions,” and then Brennan’s (1993) analysis of nongeneric root modals in terms of “accessibility relation keyed to the properties of the subject” I show that IS sentences share some of the special properties of these sentences, and that assuming the relevant differences between nongeneric and generic sentences, and making the appropriate amendments, the new kind of


accessibility relation can be productively applied to IS sentences as well. I discuss the first contextual restriction on the “in virtue of,” ^S, property (in section (4)), and show (in section (5)), some of the empirical advantages of using it. However, in section (6) I show that, assuming “in virtue of”-style truth conditions, with only the first contextual restriction on ^S, we still cannot account for the infelicity of IS sentences with “unconnected properties” mentioned above (like # A man is blond). In this section, then, I examine an additional kind of contextual requirement (in the form of a presupposition), which formally regulates the “reasonableness” of the connection between ^P (the property denoted by the IS subject) and ^Q (the property denoted by the VP, relative to our real world knowledge. Section (7) summarizes and concludes this chapter. 2. Ability of previous theories to account for the data 2.1 Krifka (1987): modality vs. nonmodality As mentioned in chapter 1, the first theory which attempts to formally distinguish BP from IS sentences is Krifka (1987), who attributes the differences in meaning between the two kinds of sentences to the presence or absence of a modal, unselective operator in the semantic representation. More specifically, Krifka (1987) claims that the “definitional” flavor of IS sentences and their incompatibility with “accidental” properties are a result of a binding by a modal Gen operator, necessarily associated with such sentences. In contrast, the “descriptive” flavor of BP sentences and their perfect compatibility with such “accidental” properties are due to the potential absence of such a modalized operator with BP sentences, since such sentences have the option of being interpreted as non-modalized Proper Kind-Predication structures. More recently, Dobrovie-Sorin & Laca (1996) made a similar claim, and suggest that the questionable status of some IS sentences results from the fact that they express rules which are “unreasonable,” and this conflicts with their necessarily “nomic” nature, i.e. their modality. In contrast, they claim that since BP sentences are not associated with a modalized Gen operator they can


freely combine with any property and can freely express any bizarre or unreasonable rule. But this kind of solution cannot work. Merely being bound by a modal operator, i.e. being evaluated relative to other possible worlds (besides, or in addition to the actual one) cannot account for the special restrictions on IS sentences. This emerges clearly from the behavior of generically interpreted BP sentences, which, just like the IS ones, express nonaccidental generalizations and support counterfactuals without being restricted by the unique pragmatic requirements on IS sentences. The same point is illustrated by the behavior of modalized nongeneric sentences like those in (7): (7) a. The Norwegian students whose names end with ‘s’ or ‘g’ must wear thick green socks. b. The carpenters in Amherst should give their sons names ending with ‘s’ or ‘g.’ Although these sentences are clearly modalized, they are perfectly felicitous (although probably false) with subjects denoting extremely unnatural properties, with “accidental” or “unconnected” properties, and in fact with anything at all. In other words, modalized sentences can in general freely express “unreasonable” statements, or even unreasonable generalizations, and still be perfectly felicitous. Thus, modality by itself cannot account for the unique properties of IS sentences. 2.2 Unspecified modal base of I- generics—Krifka et al. (1995) and Krifka (1995) As mentioned in chapter 1, although Krifka et al. (1995) mention the differences between IS and BP sentences dealt with in this work, they do not integrate these differences into their theory of generics, and thus give IS sentences the same general equivalent modal-quantificational representation as their minimally contrasting BP counterparts. Although this representation is much more detailed and developed than the ones given in Krifka’s (1987) and Dobrovie-Sorin & Laca’s (1996), it makes the wrong predictions with regard to IS sentences which are infelicitous. To see why this is the case, consider again the main types of IS sentences whose truth value and felicity should be accounted for, in (8)-(12):


The problem is how to explain the infelicity (rather than the plain falsity) of IS sentences like those in (11) and (12). Krifka et al.’s (1995) modal definition of characterizing sentences, inspired by Kratzer’s (1981) work on nongeneric modals, given in chapter 1 above, is repeated here. Essentially the same structure is given by Krifka (1995) (where x and y are variables of individuals): (13) GEN [x1…xi, y1…yi] (Restrictor, matrix) is true in w relative to a modal base Bw and an ordering source ≤w iff: For every x1…xi and every w′ Bw s.t. Restrictor [x1…xi] true in w′, there is a world w″ in Bw s.t. w″ ≤w w′, and for every world w′″≤w w″, y1…yi Matrix[ {x1},…{xi}, y1…yi]] is true in w′″. (p. 52) For example, according to this definition A lion has a bushy tail can mean that “everything which is a lion in the worlds of the modal base is such that, in every world which is most normal according to the ordering source, it will have a bushy tail” (52). The main point to note about the definition in (13) is that it does not specify the nature of the accessibility relation, but allows any Kratzerian


combination of a modal base f, e.g. an epistemic one (“what is known”), or circumstantial one (“what is known w.r.t.”), and an ordering source g, which can be stereotypical (“what the normal state of affairs is”), legal (“what is commanded by law”), or deontic (“what is expected by the norms”). As mentioned in chapter 1 this potential variability has some advantages. Firstly, it is very suitable for capturing the various “flavors” of IS sentences (i.e. the various kinds of epistemic, deontic, legal, etc. rules). Secondly, definition (13) makes the right predictions w.r.t. sentences like (8)-(10), e.g. in all worlds where what we know is true and which are closest to an ideal where everything takes its normal course (and no surprises occur) it is indeed true that “every dog has four legs,” and false that “every dog has three legs.” Also, in all these worlds it is not clear whether “every dog has 268 muscles in his tail” is true (unless you are a zoologist). However, the “underspecified” accessibility relation in (13) gives the wrong predictions w.r.t. IS sentences like (11) and (12), since it predicts them to be merely false, and not infelicitous. For example, given what we know and the normal course of events, it does not follow that every tall, famous shoemaker stands on one foot every morning, nor that every man is blond. We will get the same results if we use other modal bases as well (e.g. deontic, stereotypical). That is, the universal generalizations in (11) and (12) do not follow from any known fact, command, stereotype or the like in our world. This means that since there is no logical consequence relationship here, these two sentences are predicted to be false, just like those in (9). But this is wrong: unlike (9a-b), the sentences in (11) and (12) are not simply false but odd, or infelicitous as generic. 2.3 Cohen (2001): IS sentences as expressing rules A crucial question, then, w.r.t. the semantics of IS sentences, is how to distinguish sentences as in (8)-(10) from those in (11)-(12). As already mentioned above, the latter (and similarly the sentences in (E, F) in section (1.1) above) express generalizations, which, given the shared real-world knowledge, common ground, or something of the sort, are perceived as highly unreasonable, whereas sentences like those in (9), although clearly false, and (10), with their unknown truth value, express generalizations which are felt to be reasonable enough. But how can we


capture the distinction between reasonable and unreasonable generalizations? Cohen (2001) answers this question by claiming that sentences like those in (11)-(12), although potentially false, express rules which are members of the “set of potential rules.”1 Unlike Krifka et al., Cohen claims that IS sentences do not express quantificational statements, but rather rules. Following Carlson (1995), he claims that these rules are primitive, irreducible entities (although he does suggest that some formulas may describe rules). Thus, the fact that the rule expressed by (14) is followed, or is “in effect” can be captured by stating that the conditional in (15) holds: (14) (15)

A gentleman opens the door for a lady. opens-doors-for-ladies (x) gentleman (x)

Cohen writes: “What exactly this conditional […] means I will leave unspecified. It may simply mean extensional universal quantification… or perhaps some more intensional version” (p. 12). Formally, Cohen posits an operator, ‘!’, which maps a formula to a rule it describes if there is such a rule, and is undefined otherwise, and a predicate, ineffect, whose extension is the set of all rules which are followed or are in effect, i.e. which are part of the rules, norms, etc. dictating e.g. the appropriate behavior. The representation of (14), then, is (16):

1

In contrast, Cohen (2001) claims, BP sentences can express rules as well as quantificational, descriptive generalizations, headed by a generic quantifier Gen, which Cohen (1999) analyzes as a probability operator (see chapter 3 for a discussion Cohen’s theory). It seems, then, that the general claim, made in Greenberg (1998) and in the present theory, about the existence of two kinds of genericity in natural language, their intuitive characterization, and their manifestations in the different behavior of IS and BP sentences in English, is very similar to the one suggested in Cohen (1996, 1999a, 1999b). The difference between Cohen’s theory and the present one lies in finer-grained intuitions about the two kinds of genericity, and especially in the distinct formal analysis given by the two theories. Whereas Cohen takes BP sentences (under their descriptive reading) to be quantificational, but nonmodal, and IS ones to be nonquantificational, I take both kinds of sentences and both kinds of generalizations to be quantificational and modal statements, differing in their accessibility relation. In chapter 3 I point to some empirical advantages of this latter view. In addition, one of the advantages of this view is in leading to a natural explanation not only of the differences between the two kinds of sentences, but also of the similarities between them.


(16)

in-effect (! (gentleman (x)

opens-doors-for-ladies (x)))

Sentence (16), then, predicates the property of being “in effect” of a certain normative rule. Thus, (14) is true if this rule is indeed “in effect,” i.e. if it is part of the rules which are followed in the relevant domain, society, etc., and false otherwise. Now we can turn back to infelicitous IS sentences, like (11) and (12) above. Cohen claims that an IS sentence is felicitous (although potentially false) only if the rule it denotes is a member of the set of potential rules, which Cohen defines as the set of those rules which “could conceivably hold” (p. 13). Thus, sentence (14) may be false today since the rule requiring gentlemen to open the door for ladies is no longer in effect, but it is felicitous since this rule is a member of the set of potential rules. On the other hand (17a) is ruled out since the rule in (17b) is too odd, and thus is clearly not in the set of potential rules: (17)

a. #A gentleman eats salami. eat-salami(x)) b. ! (gentlemen (x)

I agree with Cohen’s intuition that (17a) is ruled out since it describes a rule which is too odd. The same is true for the infelicitous sentences in (11) and (12) above. The problem is that, w.r.t. to the characterization of “the set of potential rules” the definitions in this theory do not go beyond this intuitive level. How can a language user decide which rules are not odd enough, although the generalizations they express are clearly false (like (9) above), and which of them are “too odd”? In Cohen’s theory there is no systematic procedure for determining this factor, which seems crucial for characterizing the special nature of IS sentences. We can only rely on our intuitions. But if all we have is our intuitions w.r.t. these sentences, we cannot make precise predictions about the felicity of possible combinations of IS subjects with various VPs. Neither can we put the correlation between rules expressed by felicitous IS sentences and “rules which could conceivably hold” to a real test, since it is not clear what the latter term means, and our intuitions are not always very helpful either. I can certainly imagine a society in which eating salami is one of the requirements on the appropriate behavior of gentlemen, or in which rooms should be square. So, maybe we can say that such rules could, in fact, conceivably hold. In contrast, it is not clear that a rule according to which dogs have three legs is a rule which “could conceivably hold,” but


nonetheless A dog has three legs is perfectly felicitous (although clearly false). It would be desirable, then, to find a less intuitive, and more precise, systematic and compositional definition of the felicity conditions on IS sentences, which distinguishes “reasonable,” legitimate from “unreasonable,” illegitimate rules. This cannot be done, however, if all we say about IS sentences denote primitive rules, without giving their compositional semantic and pragmatic analysis. We can, for example, try to account for the infelicity of (17a) by saying that although this is a possible rule which could have hold in a certain future society or world, such a society, or such a world are so far away from our world that such a rule is indeed perceived as unreasonable or too odd. But such an explanation already requires a much richer structure, explicitly making reference to the actual society or world, and relating the meaning of the properties “be gentlemen” and “eating salami” in our world in a systematic way. This cannot be done if all we know about the rules expressed by IS sentences is that they are taken to be some unspecified sort of conditionals. Instead, it should be clearly defined what kind of conditionals IS denote, and exactly what kind of intensionality they are associated with. In the following sections I will give such a definition, and show that using it, we can indeed predict which subject-predicate combinations have corresponding felicitous IS sentences, and which do not. This definition will capture in a precise and formal way what a “reasonable generalization” means, as far as the generalizations expressed by IS sentences are concerned, and the crucial part of it will be specifying the nature of the intensionality and accessibility relation associated with IS sentences. Thus, although I agree with Carlson’s (1995) and Cohen’s (1999b) intuition that IS sentences express some sorts of rules, linguistically I do not take such rules to be “primitive irreducible entities,” but rather suggest that universal, modal statements can capture their meaning very successfully, once the accessibility relation has been appropriately defined. 3. Chierchia’s (1995) and Brennan’s (1993) theories: a formalization of the “in virtue of” intuition As mentioned in section 1 above, my main claim about IS sentences is that their accessibility relation is systematically restricted by some kind of contextually chosen “in virtue of” property or aspect of our world.


This claim is largely inspired by two theories, namely Chierchia (1995), who defines the generic operator (for various kinds of generics) in terms of contextually determined “felicity conditions,” and Brennan (1993), who formulates accessibility relations for nongeneric root modals in terms of “in virtue of” properties. Neither theory deals with the special properties of IS sentences, and moreover, attempting to directly apply the accessibility relations suggested in these theories to the truth conditions of IS sentences will yield the wrong predictions. Nonetheless, reviewing and examining these two theories will prove very productive, and will enable us to identify the steps we must take in order to arrive at the right version of truth conditions for IS sentences. 3.1 Chierchia (1995): “felicity conditions” restricting the accessibility relation As in Krifka et al.’s (1995) and Krifka’s (1987) proposals, Chierchia’s 1995 generic operator is also a modal (as well as universal and unselective) quantifier, which is supposed to be equivalently present in the semantic structure of both IS, BP and simple habitual sentences (like John smokes). Unlike Krifka and Krifka et al.’s “unspecified modal base” approach, however, Chierchia takes the accessibility relation to systematically depend on the property expressed by the VP of the sentence. More specifically, Chierchia’s accessibility relation yields the set of worlds where the felicity conditions for the property expressed by the VP hold. Take, for example, a sentence like (18a), represented as in (18b): (18)

a. Fred smokes. b. Gen s [C(s, fred)] [smokes (s, fred)]

According to Chierchia, the C variable in (18b) restricts not only the relevant situations (e.g. all the contextually relevant smoking situations), but also the worlds relevant for the evaluation of the universal statement to those where the felicity conditions for smoking by John are present, e.g. to those worlds similar to ours where Fred feels like smoking, where he is in a place where there is enough oxygen, where he is not asleep or disabled, etc. Similarly, in a sentence like A bird flies, we should look, roughly, at worlds where the felicity conditions for flying (like having wings) are


present, and where inhibiting factors (like birth defects or disease) are absent. According to Chierchia, this sentence is true, roughly, if any contextually relevant situation involving any bird in all these worlds is a situation where such a bird flies. As Chierchia himself notes, however, such a definition of the accessibility relation is too flexible, since it allows the false A penguin flies to be true as well. This is because if the only restriction on the set of accessible worlds were the felicity conditions for the predicated property, then A penguin flies would be true iff in all worlds where penguins have the property required for flying (and do not have any inhibiting factors) every (relevant) situation involving any penguin will be a flying situation. Thus, with no further restrictions, any generic sentence, including clearly false ones like A cat flies, A human being flies and even A table flies will be considered true. Aware of this risk, Chierchia adds that in addition to the property denoted by the VP, the common noun in the subject of the sentence also adds information which restricts the accessibility relation. In the case of A penguin flies, for example, we will somehow have information saying that “penguins are birds who do not have the structural characteristics necessary for flying are thus excluded from consideration” (p. 196). Chierchia suggests, in a footnote that these two kinds of information should be encoded by different components: Actually, it seems intuitively that there are two kinds of contextual restrictions. One comes, as it were, from the verb; it determines what conditions have to be met for an individual to undertake an action. The other comes from the noun; it determines what features of the species are relevant for the action. This intuition could be accommodated by employing more than one context variable. (p. 196) In other words, according to Chierchia the conditions which are supposed to hold in the accessible worlds have to meet two contextual requirements: First, they should be considered appropriate “felicity conditions” for the action (denoted by the VP), and second, they should be relevant enough for the species (denoted by the subject). If the species is really taken to have the felicity conditions for the action, then we take the generic sentence to be true. I believe that this direction of formulating the accessibility relation is especially suitable for the semantics of IS sentences, and captures the intuition about the “in virtue of a property” aspect of these sentences.


Nonetheless, as it stands now, Chierchia’s suggestion is too general and unspecified for our purposes, and trying to apply it directly to the truth conditions of IS sentences leads to several problems. The first problem is that Chierchia’s theory supplies no precise procedure for determining what is considered a “felicity condition” for a VP property. This may seem trivial and unnecessary, since intuitively we all know what flying, smoking, swimming and the like involve. However, determining the felicity conditions for a wider range of properties, like “eating cornflakes for breakfast,” “wearing thick green socks” or “giving all your children names ending with ‘s’ or ‘j’” is much less clear.2 In addition, Chierchia’s theory does not specify how the second contextual variable is filled, i.e. how the common noun determines which “features of the species” (denoted by it) are relevant for the action. Among other things, it is not clear how Chierchia’s theory distinguishes between false IS sentences like A cat chases dogs and infelicitous ones like # An uncle eats white marshmallows on Sunday, since the two sentences behave very similarly in terms of Chierchia’s criteria. Some cats may have the “features relevant for the action” (of chasing dogs), but such features are not characteristic of the cat species as a whole. Similarly, some uncles may have the “features relevant for the action” of eating white marshmallows on Sunday, but such features are not characteristic of the species of uncles, as a whole. Merely considering “the features of the species relevant for the action,” then, cannot explain why the former sentence is false, whereas the latter is infelicitous. Thus, although Chierchia’s accessibility relation is intended to be more keyed to contextual information about the subject and VP properties than e.g. Krifka et al.’s “unspecified modal base” suggestion, it still cannot make the right distinction between a “reasonable” and an

2

Furthermore, extending the list of predicates to individual level predicates like “be a smoker,” “be square,” or “be blond” complicates things and gives us additional wrong results. We want to distinguish A man has a beard, which is fine as generic, from ??A man is blond which is odd (cf. again the perfectly natural BP version Men are blond). This seems to require, first of all, that both these properties will have different felicity conditions. However, Chierchia characterizes the felicity conditions for all individual-level predicates uniformly as merely “involving the individual” or “being located in any arbitrary location.” This cannot differentiate between the two different types of IS sentences.


“unreasonable” generalization, which seems an important part of the meaning of IS sentences. However, the problems with applying Chierchia’s suggestions to the accessibility relation of IS sentences also suggest the direction in which we should move: We need to formally clarify how precisely the accessibility relation of IS sentences is influenced by our real-world knowledge about the properties denoted by the IS subject and the VP. In the next section I review and examine Brennan’s (1993) analysis, which makes a similar attempt at systematically defining the accessibility relation of nongeneric root modals, using an “in virtue of” property. I will show that integrating Brennan’s formal approach with Chierchia’s ideas about generic sentences in the right way proves to be very productive for the semantics of IS sentences. 3.2 Brennan (1993): an accessibility relation “keyed to the properties of the subject” Unlike traditional accessibility relations (e.g. “what we know,” “what the stereotypes are,” etc.), which give the same output (i.e. the same set of worlds) no matter which sentence they are used with, the modal bases constructed by Brennan for sentences with root modals vary in a systematic way, depending on the subject. In doing so, Brennan formalizes and develops Kratzer’s (1981) original intuition about circumstantial modality, in which the modal sentence is evaluated against only a subset of the facts in our world (as opposed to epistemic modality, in which the whole set of facts is relevant). This direction seems to be good for the accessibility of IS sentences as well, which, according to our original intuition is also some sort of circumstantial modal base, and, as claimed above, should also be restricted (by some “in virtue of” aspect of our world), and flexible (i.e. the relevant “in virtue of” property varies, depending on the material in the sentence). In the following sections I review the relevant parts of Brennan’s theory, examine the compatibility of this theory with the IS data relevant for us, and elaborate on the revisions we need to make in order to apply Brennan’s ideas to the IS data.


3.2.1 Data concerning root modals Brennan (1993) attempts to explain the effect that root modals (i.e. deontic, dispositional, buletic, etc.) as opposed to epistemic modals have on the interpretation of sentences with “symmetric predicates.” This influence can be seen when examining the entailment relations in the following three pairs (Brennan’s p. 127–128): (19) (20) (21) (22)

The governor shook hands with all the prisoners. All the prisoners shook hands with the governor. The governor may shake hands with all the prisoners. All the prisoners may shake hands with the governor. (epistemic reading in both cases: it may be the case that…) (23) The governor may shake hands with all the prisoners. (24) All the prisoners may shake hands with the governor (root reading in both cases: having the right to shake hands) Symmetric predicates, like shake hands with or getting the same score as, are predicates for which we expect that if R(x, y) then it is necessarily true that R(y, x). This expectation holds in the case of (19) and (20), and assuming an epistemic interpretation of may, in the case of (21) and (22) as well. What is unexpected, though, is the failure of the entailment relation between (23) and (24), where may is given a root (and specifically a deontic) interpretation, i.e. the fact that the governor has the right to shake hands with all prisoners does not entail that all prisoners have the right to shake hands with the governor. In other words, for some reason, root, but not epistemic, modals turn symmetric predicates into asymmetric ones 3.2.2 Brennan’s intuition Brennan suggests that it is the unique nature of the accessibility relation associated with root modal which breaks the symmetricality effect. According to Brennan, the modal base of epistemic modals is the usual Kratzerian modal base, a function from worlds to sets of propositions, where crucially, there is no special restriction on the kind of propositions in such a modal base. In contrast, Brennan claims, the modal base restricting root modals are necessarily “about” the subjects, they are “keyed to the properties of the subject” or, somewhat more formally,


they are functions from worlds to sets of—not any arbitrary proposition —but rather propositions which are all “about” the subject. Thus, such modal bases can be viewed, in fact, as functions from worlds to bundles of properties, which are all predicated of the individual denoted by the subject of the sentence.3 Brennan’s intuition can explain the failure of symmetricality effects in (23) and (24) in the following way. Since the modal base of the sentence is “keyed to the properties of the subject,” the governor’s deontically accessible worlds (where he shakes hands with all the prisoners) are not necessarily the same as the deontically accessible words of all the prisoners (where they shake hands with the governor). The sensitivity of root modals to the properties of the subject is, then, the reason that such modals turn symmetric predicates (symmetric in terms of subject-object relationship) into asymmetric predicates.

3.2.3 The formalization Brennan defines a new type of accessibility relation, called “accessible for d” as a relation between a world and a pair consisting of a world (the world of evaluation) and an individual: (25) Where p is an arbitrary property expression restricting a modal: Accessible for d: A world w′ is accessible from a world w for an individual d, <w, d> R w′ iff <w′, d> p Thus, w′ is accessible for d from w iff for some property p, d has this property in w′. Consider, for example, a sentence like (26) (Brennan’s p. 181): (26) Joan lives in Racine and is registered to vote. She may (has the right to) vote in Racine’s mayoral election.

3

This claim is further motivated, according to Brennan, by the correlation between the position of root modals in the syntax (Brennan claims these are VP operators), and the type of entity they are restricted by (property expressions and not propositions, according to Brennan).


Sentence (26) seems to have a meaning along the lines of (27), with an explicit “in virtue of” adjunct: (27) In virtue of living in Racine and being registered to vote, Joan may vote in Racine’s mayoral elections. The properties of living in Racine and having registered to vote are presupposed, according to Brennan, to hold of Joan in the actual world, and form the modal base of the sentence. In Brennan’s words: The modal VP (may vote in Racine’s mayoral election) denotes the property of having the right to vote in Racine’s mayoral elections. In a given context, it’s made clear that Joan has this (modal) property only because she has other properties: namely, she has the property of living in Racine and the property of having registered to vote. Without these properties, it is understood, she probably wouldn’t have the right to vote in Racine’s mayoral election […] Assume for the time being that the two background properties are simply conjoined to form one complex property. It’s this complex property that will stand for p in (28). (p. 182–183) (28) Where p, is { : d lives in Racine in w and d has registered to vote in w} Accessible for d: a world w′ is accessible form the world of utterance w, for an individual d, <w, d> R w′ iff { : d lives in Racine in w and d has registered to vote in w} Now, the modalized predicate “may vote in Racine’s mayoral election” is represented as in (29), where the subscript Cd,w represents the modal base: (29) [may Cd,w vote in Racine’s mayoral election in w] denotes the set of individual d, s.t. there is a world w′, which is Cd,w -accessible for d s.t. d {d: d vote in Racine’s mayoral elections in w′} Applying now the meaning of this predicate to the meaning of “Joan” we get the meaning of the original sentence Joan may vote in Racine’s mayoral elections.


4. Application of Brennan’s ideas to the requirements on the accessibility relation of IS sentences We will now look at various aspects of Brennan’s theory, checking which of them can be applied to the accessibility relation of IS sentences, and making changes when such an application cannot work. 4.1 Basic data and intuition Brennan’s observation about the behavior of symmetric predicates in the presence of (overt, nongeneric) root modals also seems to be true for the behavior of such predicates in IS sentences, where no explicit modal verb is present. Thus, compare the pairs in (30) and (32) to those in (31) and (33): (30) (32) (33)

John did not shake hands with Mary. Mary did not shake hands with John. A Democrat does not shake hands with a Republican. A Republican does not shake hands with a Democrat.

Sentences (30) and (31) behave as expected: they are true in exactly the same circumstances. An indication of this is the infelicity of the conjunction in (34): (34) # John did not shake hands with Mary, but Mary shook hands with John. In contrast, a comparison of (32) and (33) reveals an asymmetry. We can clearly imagine situations where (32) is true and (33) is not, e.g. where Republicans are known, or required to be much more tolerant than Democrats. This is indicated by the felicity of the conjunction in (35): (35) A Democrat does not shake hands with a Republican, but a Republican does shake hands with a Democrat. This pattern can be given an explanation on the same line as Brennan’s explanation of the data in (21) above: IS sentences are modalized, and


their accessibility relation is sensitive to, or “keyed to,” the subject. Thus in (32), the accessible worlds are the deontic or stereotypical worlds of a Democrat, and these worlds may be different from the deontically or stereotypically accessible words of a Republican. This is what allows for the felicity of (35). These IS sentences, then, are evaluated w.r.t. a conversational background similar to Kratzer’s (1981) circumstantial one, i.e. one where only a subset of the facts, norms, stereotypes, etc. are taken into account. Moreover, this subset is determined systematically: it is restricted by the knowledge, stereotypes, norms, etc., which are somehow related to the property denoted by the IS subject. Thus, I will take the accessibility relation of IS sentences to be similar to the one suggested by Brennan for nongeneric root modals, i.e. to be “keyed to,” sensitive to, or simply about the denotation of the subject IS. This means, that as in Brennan, the assumptions about the denotation of the subject IS will systematically restrict the set of accessible worlds. 4.2 “Accessible w.r.t. a property”: subset, and not membership relationship It is clear, though, that we can’t simply copy Brennan’s definition of the modal base. Brennan discusses nongeneric sentences with referential subjects, and thus her modal base is “accessible for an individual”(R <w, d>, w′) (where d stands for an individual). In our case, however, the subject IS, which is bound by the modal operator Gen, denotes a set in a world, i.e. a property. We want, then, the relevant accessible worlds of IS sentences to be accessible from w w.r.t. to a property ^P, denoted by the IS subject (where P is of type >), and not w.r.t. an individual d. For example, we will say that a sentence like A dog has four legs is intuitively evaluated in all worlds which are accessible from our world, w.r.t. to the property of being a dog. We now want to give content to such a statement. What does it mean for a world to be accessible w.r.t. a property denoted by the IS noun phrase? Remember that in Brennan’s theory a world is “accessible for an individual” (i.e. R (<w, d>, w′) holds), when the individual denoted by the subject has a certain property in the output world w’, i.e. when <w′, d> are members of p. For example, in Joan may vote in Racine’s mayoral election the accessibility relation was restricted by the complex property, true of Joan: living in Racine and having registered to vote.


My claim is that with IS sentences too there is some property, appropriately related to the ^P property denoted by the IS noun phrase, which systematically restricts the accessibility relation. I will henceforth call this restricting property ^S (instead of Brennan’s p), or the “in virtue of” property. But, since we are not dealing with referential subjects, but with IS ones, which themselves denote properties, we clearly cannot have the same kind of membership relationship as in Brennan’s cases Rather, we will use the subset relation and say that w′ is accessible from w w.r.t. to the property ^P (denoted by the IS) if the set P is a subset of the set S in w′, where ^S meets some requirements in w. That is, we will take the accessibility relation of IS sentences to be something like (36), where the real world requirements on the choice of ^S are not specified yet: (36) w′ is accessible from w w.r.t. the property ^P (denoted by the IS subject) iff for some other property, ^S (which meets some requirements in w): P S in w′ or in other words: iff d P(d, w′) → S(d, w′) For example, A dog has four legs will be evaluated relative to all worlds w′ which are accessible from the world of evaluation w.r.t. the property of being a dog, e.g. relative to all worlds w′ where the set of dogs is a subset of the set of creatures with a four-legged genetic makeup, or more simply, relative to all worlds w′ in which every dog has a four legged genetic makeup.4

4.3 The role of the world of evaluation in choosing the possible “in virtue of” properties 4.3.1 Contextual restrictions on the choice of the “in virtue of” property We now reach a crucial point in our definition of the new accessibility relation, namely specifying the real world requirements that “in virtue of” ^S property has to meet w.r.t. the ^P property denoted by the IS subject. In the definition of the “in virtue of a property” accessibility


relation in (36) these real-world requirements are not yet specified. But if we do not specify such requirements then in principle any property whatsoever can stand for ^S in (36), even if given our real world knowledge it is completely disconnected from ^P (denoted by the IS subject). This can wrongly lead all IS sentences, including clearly false ones, to be evaluated as true. For example, in the case of the false A dog has three legs we can take ^S to be the property of cutting off one’s own leg, thus taking this sentence to mean (roughly) that in all worlds where every dog cuts off his own leg, it has three legs. Similarly, in the case of the false and infelicitous # A man is blond, we can take ^S to be “has a genetic makeup for blond hair,” although this property is not connected to being a man, thus wrongly predicting this sentence to be true and felicitous. Intuitively, then, the ^S property restricting the accessibility relation of IS sentences should not be any arbitrary property, but should meet

4

There is another relevant difference between Brennan’s modal base and the one I will use for IS sentences: In Brennan’s theory the modal base is “property level,” that is, not only is it restricted by some property of the individual denoted by the subject, but also formally, it is a function from worlds into a property, and not into a proposition (or a set of propositions). Thus, Brennan defines <w, d> R w′ as true iff <w′, d> p, and not iff w′ p(d). Brennan’s main motivation for this move is her claim that there is a correlation between the syntactic position of the modal auxiliary and the semantics of the modal base it is associated with. Thus, Brennan claims, the “property level” modal base of root modals is intended to reflect the fact in the syntax such modals operate on the VP, and have a narrow scope w.r.t. the subject (whereas the “propositional level” modal base of epistemic modals is supposed to reflect the fact that such modals are sentential operators). I will not adopt this aspect of Brennan’s formalization to the modal base of IS sentences, mainly because, although I take the generic quantifier binding IS sentence to be similar to root modal auxiliaries in being also “keyed to the properties of the subject,” it is not equivalent to them. I follow Wilkinson (1991), Krifka et al. (1995) and Chierchia (1995), among others, in taking Gen to be an unselective sentential operator, whose semantic structure is similar to that of classic quantificational adverbs like always and usually (see the introduction for evidence for this view). This means that, unlike root modal auxiliaries, Gen binds the subject too, and doesn’t have narrow scope with respect to it. (In this sense, Gen is more similar to what Brennan (following Carlson (1977)) calls “quantificational modals,” although Brennan claims these modals form a separate interpretive class, and cannot have a “root” semantics.) Thus, although I claim that the modal base of IS sentences is property oriented, i.e. it is restricted by a property that every member of the IS set has, I will take it to be “propositional level,” and not “property level.”


some requirements it meets in the actual world. Formally, we can express this idea by requiring ^S, the “in virtue of” property, to be a member of some contextually chosen set of properties C. More specifically, a key factor in choosing an “in virtue of” ^S property which can be a member of C is the reasonableness of its connection with ^P (the property denoted by the IS subject), given our real-world knowledge. Our contextual variable C, then, should be defined as including only properties which have an appropriate relationship to the ^P property, relative to the world of evaluation w, i.e. as CR(^P,w ). The general form of the truth conditions, then, is (37), where the exact definition of CR(^ P,w) is not yet specified (the problem of exceptions is ignored at this point, and will be dealt with below): (37) Let indefinite singular NP denote ^P, and VP denote ^Q. Let ^S be a property An IS sentence of the form NP VP is true in w iff:

Paraphrase: An IS sentence is true in a world w iff, there is some property ^S s.t. in all worlds w′ where for every individual d it holds that if d has P in w′, it has S in w′, (where ^S is a member of a set of properties appropriately connected to ^P in w)), every contextually relevant situation in w′, involving every P individual in w′, is a Q situation involving that P individual in w′.5

For example A dog has four legs will true in w iff, there is some property ^S, s.t. in all worlds where every dog has S (where ^S is a

5

An almost equivalent, and an apparently simpler formalization is (i), where the universal quantification over individuals appears only once, instead of twice as in (37) above: (i)

Despite the relative simplicity of (i) over (37) above, I continue to use (37), for reasons which will become clear below.


member of a contextually chosen set of properties with the appropriate relationship with being a dog relative to w), every contextually relevant situation involving any dog is a situation where that dog has four legs. 4.3.2 Brennan’s suggestion for restricting the set of “in virtue of” properties We now have to define the appropriate relationship of ^S with ^P relative to our real world knowledge, i.e. the nature of R(^P, w). Above we intuitively claimed that ^S should be “associated” or “connected enough” with ^P. Brennan (1993) has a very similar intuition about the “in virtue of” property restricting the modal base of (nongeneric) root modals. Such a property, claims Brennan, cannot be any arbitrary property of e.g. Joan. Rather, it is a property which is presupposed to hold of Joan in the actual world. So, Joan must take this course (in virtue of being a graduate student) is evaluated relative to worlds where Joan has the property of being graduate student, and crucially, this property is presupposed to really hold of Joan in the actual world. Do we want the same kind of requirement on the modal base of IS sentences? If so, then the “in virtue of” property restricting the modal base will be required to be true of every member of the set corresponding to the IS property (^P) in the actual world, i.e. the P set will have to be a subset of the set corresponding to the “in virtue of” property in the actual world. This means that (38a) will be true if and only if (38b) will be presupposed to be true as well. (38)

a. b.

d P(d, w′) → S(d, w′) d P(d, w) → S(d, w)

For example, if in A bird flies we take the “in virtue of” property” to be “having wings” then we have to presuppose that in the actual worlds every bird has wings. Using again Kratzer’s (1981) terminology, this amounts to saying that the modal base of IS sentences should be realistic (i.e. taking the input world of evaluation (w) to be a member of the set of output, accessible, worlds). Krifka et al. (1995) have already argued against such a requirement on the accessibility relation of generics. The same kind of argumentation also holds when adopting the new (“in virtue of”) perspective on IS sentences.


The argument concerns the ability of IS sentences to express deontic necessities. We know that many IS sentences, like A boy doesn’t cry, express such deontic or normative generalizations. Intuitively, such sentences are true in virtue of following some social norm (e.g. requiring boys to be tough). When uttering these sentences, however, we clearly do not have to presuppose that every boy in the actual world follows the social norm, i.e. we can felicitously utter this sentence while assuming that in the actual world very few boys, if any, try to behave according to this “toughness” norm. Thus, although it seems clear that it is indeed some aspect of the real world which allows us to associate the property of boyhood with the property of trying to be tough, (rather than e.g. with the property of undergoing an operation for removing the tear duct), it is not the presupposition about truth in the actual world which makes the difference. Instead, the former property is taken to be more associated in the actual world with being a boy than the latter. I will now attempt to clarify the nature of this “association” relation between properties. 4.3.3 An intuitive characterization of “associated properties” Intuitively, what I have mind is this: Take a property of being a boy. In our world we associate various properties with boys. For example, the property of being between three and seventeen years old, of being a male, of loving sports, of being tough, etc. Now it is clear that we don’t take all these properties to be true of all boys in the actual world. So, for example, of the properties I just mentioned, being male will be true of every boy in the actual world (and in fact, in all worlds where English works as it does in our world). The property of being between three and seventeen years of age will probably be true of every boy only in the worlds where, for example, western standards about boyhood hold, but not, for example, in worlds where standards of other cultures holds (where someone is considered a man when he is fourteen). The properties of loving sports and being tough are clearly not true of every boy in the actual world, but rather in the worlds where some e.g. western cultural stereotypes or norms hold. Similarly, in the actual world we associate with many properties a cluster of other properties, not all of them actually true of every member of the corresponding set in the actual world. For example, in our world we associate with the property of being a dog various other properties


like having a tail, having four legs, being loyal, and having teeth. We associate with the property of being a judge properties like having studied law, being an honest person, being rich, and being respectable. If we knew more about judges we might also associate with them the properties of paying 45.8% income tax a year, of working 37 hours a week, etc. Here too, not all associated properties are necessarily true of all dogs or judges in the actual world. Perhaps only 20% of the actual dogs are indeed loyal. Perhaps only 70%, or even only 20%, of the actual judges pay 45.8% income tax, because in reality the rest of them cheat on their tax returns. On the other hand, even if all current supremecourt judges in the actual world wear big blue shirts, we will not take this property to be associated with the property of being a supremecourt judge, since we do not have any known fact, norm, stereotype, etc., which associates this property with being a supreme-court judge. Thus, the cluster of properties associated with dogs, boys, lawyers, tables, and semantic papers clearly depend on the beliefs, knowledge, stereotypes, norms, etc. we have in the actual world, and it is clear that in other possible worlds—where other standards hold—people could associate with these properties completely different properties. For example, we can clearly imagine a possible world where dogs are associated with properties such as being pink, having three legs, not having a tail, and being extremely wild and dangerous. Similarly we can imagine a world where boys are associated with the properties of being between two and fourteen years old, of studying all day, and being very sensitive, etc. 4.3.4 Formal characterization of “associated properties” The intuitive discussion of associated properties makes it clear that identifying these properties depends not on whether these properties actually hold of the individuals in a certain set in our world, but rather on whether they hold of such individuals according to the knowledge, beliefs, stereotypes, norms, legal commands, etc. we have in our world. More precisely, the associated properties are those where the idea that they hold of the individuals in a certain set follows from these facts, norms, stereotypes, etc. That is, a property ^S is associated with a property ^P in our world, iff we can find some Kratzerian modal base f e.g. epistemic (“what we know in w”), deontic (“what is required in w”), circumstantial (“what we know w.r.t….”), legal (“what the law requires”), stereotypical (“what the stereotypes in w say”) etc., such that


in all worlds—let’s call them w” accessible w.r.t. such a modal base f, it is true that d [P(d)] [S(d)]. In other words, ^S is associated with ^P in d [P(d)] →[S(d)] follows from the propositions in some w iff Kratzerian f, i.e. iff d [P(d)] → [S(d)] is a necessity in the world of evaluation w.r.t. some f.. In contrast, if we cannot find any f(i.e. no set of facts, norms, stereotypes, legal commands, etc.) from which d [P(d)] →[S(d)] follows, then ^S is not associated with ^P. Formally: (39) Association of properties: ^S is associated with ^P in w iff there is some Kratzerian function f (from worlds to sets of propositions (i.e. to sets of sets of worlds) s.t.

worlds w″—Kratzerially accessible w.r.t. f, (d)] is true.6

i.e. iff in all d [P(d)] → [S

This definition ensures that only nonaccidental properties will be “associated.” Since even if every supreme-court judge in our world wears blue shirts, there is no Kratzerian accessibility relation (“what we know,” “what the law provides,” “what the norms are,” etc.) which will make “Every supremecourt judge wears a blue shirt” a necessity in our world. In other words, regardless which Kratzerian function we choose, there is no guarantee that this universal statement will be true in all the Kratzerially accessible worlds.7 4.4 Integrating the definition of “associated properties” into the truth conditions of IS sentences Sentence (39) gives the formal definition of associated properties. I believe that when we think about properties which play the role of the

6

Angelika Kratzer (p.c.) pointed out that her theories on modality (e.g. (1981)) do not supply any empirical constraints on possible conversational backgrounds. Thus, there is no closed, fixed list of “Kratzerian” functions, and hence existential quantification over such functions may be risky. What I mean by this term, however, is the kind of functions used by Kratzer in her analysis of modality, which are, more precisely, those conversational background which yield sets of propositions which are usually known, commanded, stereotyped etc. by the whole community of language users.


“in virtue of,” ^S properties in the truth conditions of IS sentences, we chose exactly those which meet the requirement in (39). More formally, the “association with ^P in w” defined in (39) is the right requirement on the context variable CR(^P, w). in (39) above, i.e. the set or cluster of properties restricting the accessibility relation of IS sentences is in fact the set or cluster of properties associated with ^P in w. The truth conditions of IS sentences look, then, as in (40): (40) An IS sentence, of the form IS VP is true in w iff:

And S CR(^P, w). iff there is some Kratzerian function f (from worlds to sets of sets of worlds) s.t.

Paraphrase: An IS sentence is true in a world w iff, there is some property ^S s.t. in all worlds w′ where every P individual has S (and where ^S is a member of a set CR(^P,w) of properties appropriately connected to ^P in w) it holds that every contextually relevant situation in w′, involving every P individual in w′, is a Q situation involving that P individual in w′. And ^S is a member of CR(^P,w). (i.e . is appropriately connected to ^P in w) iff “every P is S′ follows from what we

7 There are two additional issues concerning associated properties, which I am not going to deal with at this point. Firstly, can we find a way to predict the associated property of a complex property ^P compositionally, i.e. from the associated properties of the constituents properties of ^P? At this point let me only mention here Kamp & Partee’s (1995) claim that it is not always possible to calculate the prototype and stereotypes of complex expressions compositionally (they mention the property “male nurse” as an example). Since stereotypes of properties are a subset of our “associated properties,” this claim seems to be relevant for associated properties in general. The second question is whether associated properties of a property ^P should be considered presuppositions of ^P, i.e. is it generally the case that the property “be tough” is a presupposition of “be a boy,” or that “barking” is presupposed by “be a dog”? These two questions are of course connected. If associated properties are presuppositions of other ^P properties, than calculating such presuppositions of complex properties out of the presuppositions of the constituents properties should be dealt with by theories examining what is called the presupposition projection problem (see e.g. Morgan (1969)). This problem is beyond the scope


know in w, what is commanded in w, what the stereotypes in w are, etc.

This means that the semantics of IS sentences involves, in fact, two levels of modality: The basic set of accessible worlds (w′ in (40)) is defined as those in which “every P has S” holds. But the choice of S depends, in turn, on a more traditional, Kratzerian accessibility relation f, which gives set of worlds where what is known, commanded, etc. hold (w″ in the representation). This “double modality” captures the intuition that a speaker of an IS sentence not only asserts that a certain generalization is nonaccidental (i.e. say that “every P is Q” holds in a set of accessible worlds w′), but also that it is true in virtue of another nonaccidental generalization —“every P is S”—which should hold in a set of Kratzerially accessible worlds—w″.8 4.5 The problem of exceptions According to (40), the truth conditions of (41a) are (41b), paraphrased in (41c): (41) a. A dog has four legs. b. A dog has four legs is true in w iff:

And S CR(^DOG, w), iff there is some Kratzerian function f (from worlds to sets of sets of worlds) s.t.

According to our definition, an ^S property is associated with being a dog iff there is some set of known facts, commands, stereotypes, etc. in our world that “Every dog has S” follows from. In order for (41a) to be

of this dissertation. At this stage, then, I will not attempt to answer these questions, but rather continue to talk about associated properties in the way developed above.


true, we need to find some such ^S property associated with being a dog in w, s.t. (roughly) in all worlds where every dog has this property, he has four legs. Intuitively, “has a four legged genetic makeup” seems to be an appropriate ^S property.9 However, strictly speaking, things are less simple. If “has four legged genetic makeup” is the right ^S property in the truth conditions (41b), then (41a) is true if (42a), paraphrased as in (42b) holds (ignoring the situation variable for a minute): (42) a. w[ d[dog (d, w) → [has four legged genetic makeup (d, w)]]→ [ d dog (d, w) → has four legs (d, w)] b. Paraphrase: “A dog has four legs is true in w iff Every world where every dog has a four-legged genetic makeup is a world where every dog has four legs. But clearly (42a) is false, since, as has been pointed out by numerous scholars, generics like (41a) allow for exceptions. Thus, in the actual world, for example, although all dogs have a four-legged genetic makeup, not all dogs have four legs. Some have undergone an accident, and have only two or three legs. Some were forced to participate in bizarre scientific experiments, and as a result have five legs, and one dog wanted to prove how brave it was, so he cut off his own leg. Besides these “actual world exceptions,” there are also other kinds of potential counterexamples to (42a). We can imagine, for example, bizarre worlds where all creatures with a genetic makeup for n number of legs, end up with an n+1 number of legs, as result of a regular policy

8 Greg Carlson (p.c.) pointed out to me the following potential problem with definition (40): In this definition the IS generic (e.g. A dog has 4 legs) is based on a nonaccidental statement (the statement that every P member has S, e.g. that Every dog has a 4-legged genetic makeup). But this latter statement can also be expressed as an IS generic (e.g. A dog has a 4-legged genetic makeup), and this may lead to an unwelcome circularity in the semantics of such generics. I believe, however, that no circularity arises. Firstly, because although the “in virtue of” universal statement is indeed nonaccidental, there is no necessity to express it in natural language as an IS generic. If we want to express it in natural language at all, we can also do it using a BP generic (e.g. Dogs have a 4legged genetic) or explicit “every” and a modal verb (e.g. Every dog must have a 4-legged genetic makeup). More importantly, there is no risk of circularity since the truth conditions for the “in virtue of” universal statement, and that of the IS generics are explicitly defined to be different. Specifically, the universal statements in each case are required to hold in different sets of worlds.


of transplanting operations for the whole population. This world, where all dogs with a fourlegged genetic makeup end up with five legs, makes (42a) false as well. Such “abnormal” exceptions are discussed often in the literature. Krifka (1995) and Chierchia (1995) suggest accounting for such exceptions using some “normality” constraint on the accessible worlds. Specifically, as already mentioned in chapter 1 above, Krifka (1995) suggest adding a “normal” ordering source to the Kratzerian modal base, restricting the accessible worlds to those closest to an ideal world where “everything takes it normal course,” or where no surprises occur. In these worlds, accidents, bizarre scientific experiments and overly brave dogs do not exist, since intuitively they constitute a break in the “normal course of events.” Although I follow Krifka (1995) in assuming that the exceptions to IS generics are the “less normal” P individuals, and thus, that some “normality restriction” indeed plays a role in the semantics of IS sentences, I think that imposing such a restriction on the set of accessible worlds leads to several serious problems (which I discuss in section (6) below, and in chapter 4). Thus, I will not adopt the “normal ordering source” approach, and suggest instead another tool by which generic sentences tolerate “abnormal” exceptions. This new tool is sensitive to the presence/absence of an “in virtue of” property in the semantics of the generic sentence, and using it will enable accounting for some new observations concerning fine-grained distinctions between the behavior of IS and BP sentences w.r.t. exceptions. Developing this new mechanism for tolerating exceptions, then, depends on first understanding the differences in the semantics of BP and IS sentences (in terms of the presence/absence of an “in virtue of” property in the semantic representation). Thus, I will postpone the criticism of the “normal ordering source” solution, and the introduction of the new mechanism for tolerating exceptions to chapter 4. In the present and the following chapters, I will not attempt to solve the problem of exceptions, and will informally paraphrase the universal quantification over individuals in the consequence of the truth conditions

9

But not necessarily, of course. The only requirement is that an “in virtue of” property is an integral property of the semantics of the IS sentences. It is clear that even somebody who knows nothing about genetics will take A dog has four legs to be true. In such a case he will accommodate another “in virtue of” property which will make the sentence true.


as “every (non-exceptional) P individual.” For example, an IS sentence like (41a), will be intuitively paraphrased as in (43): (43) “A dog has four legs” is true in w iff there is some property ^S, s.t. in all worlds w′ it holds that If every dog in w′ has S in w′ (and ^S is associated with being a dog in w), then every contextually relevant situation in w′, involving every (non-exceptional) dog in w′, is a situation where such dog has four legs. We can now correctly judge (41a) to be true. We can indeed find such a property ^S, which is associated with being a dog, namely having a four legged genetic makeup, s.t. in all worlds where every dog has this property, every situation involving a non-exceptional dog is a situation where it has four legs. 5. Advantages The version of the truth conditions in (40) formalizes the basic intuition that IS sentences express “in virtue of” nonaccidental generalizations, and that the relevant “in virtue of” property (^S) is systematically restricted, i.e. required to be “associated” with the property denoted by the subject (^P), on the basis of context and real-world knowledge. These truth conditions are not complete yet, and below I continue developing them (mainly adding a further pragmatic requirement on the ^S property in section (6) of this chapter, and introducing the mechanism for tolerating exceptions in chapter 4). Nonetheless, we are already in a position to point out some advantages in using this new direction of restricting the accessibility relation. 5.1 Normative, legal, epistemic, etc., “flavors” of IS sentences As noted above, IS sentences may have various “flavors” and seem to express various kinds of rules. In contrast, minimally contrasting BP sentences have a reading which is more descriptive and unified (although they can also have the more “flavored” reading). The contrast has been pointed out by Burton-Roberts (1973) for the pair in (44a, b). The pairs in (45) and (46) manifest the same pattern:


(44) a. A gentleman opens the door for a lady. (social rule) b. Gentlemen open the door for a lady. (Reading 1: descriptive. Reading 2: deontic/social) (45) a. A single parent does not pay more than 23% income tax. (legal rule) b. Single parents do not pay more than 23% income tax. (Reading 1: descriptive. Reading 2: legal) (46) a. A Jew fasts on Yom Kippur. (religious rule) b. Jews fast on Yom Kippur. (Reading 1: descriptive, Reading 2: religious) Notice that Krifka et al. (1995) already explain the variety of flavors with similar generic sentences in modal terms, as depending on the variety of modal bases which can play a role in the interpretation of the sentences. In contrast, according to our truth conditions in (40) above, the modal base of IS sentences does not vary. It is fixed: it is always a function from worlds to the set of worlds where every P individual has S. What varies is the choice of ^S (the property associated with ^P in w), and this variation leads to the fact that each IS sentence is eventually interpreted in a different set of possible worlds (w’ in (40)), and consequently to the variety of “flavors.” For example, (44a) will be evaluated in all worlds in which every gentleman has the S property of following the 19th-century politeness code, (45a), in all worlds where every single parent has the S property of following the taxing regulations, (46a) in all worlds where every Jew has the S property following the special commandments concerning Yom Kippur, etc. The choice of each such ^S property depends on a more traditional, Kratzerian modal base: In the case of (44) S property is required to be true of every gentleman in all 19th-century deontically accessible worlds w′.’ Thus, the set of worlds w′, where every gentleman follows the 19th-century politeness code is a superset of the 19th-century deontically accessible worlds (w”). In the case of (45), the S property holds of every single parent in all legally accessible worlds w′.’ So, the set of accessible worlds w′ is a superset of these legally accessible worlds. And in the case of (46) S will hold of every Jew in all religiously accessible worlds, and thus the set of worlds w′ is a superset of these religiously accessible worlds.


Thus, for each choice of ^S, the set of accessible worlds w′ is indirectly dependent on a different set of Kratzerially accessible worlds: epistemic, deontic, religious, social, legal, stereotypical, etc. More formally, for each choice of ^S, the set of accessible worlds is taken to be a superset of a different set of Kratzerially accessible worlds. This is what allows IS sentences express generalization of different “flavors.” 5.2 Comparing IS sentences and nongeneric universal sentences with overt deontic modal verbs Although IS sentences seem to express a less descriptive and more “rule-like” generalizations than their BP counterparts, they are frequently understood as more descriptive and less rule-like than quantified, nongeneric sentences with an explicit deontic modal verb. For example, (47a) clearly expresses a deontic rule required in the actual world, but makes no predictions w.r.t. the actual behavior of actual ultraOrthodox Jewish women. In contrast, the truth of (47b) seems to say something about ultra-Orthodox Jewish women in the actual world (notice, though, that (47b) is still less descriptive, and more “rule like,” than the corresponding BP sentence in (47c)):10 (47) a. Every ultra-Orthodox Jewish woman must/should cover her hair. b. An ultra-Orthodox Jewish woman covers her hair. c. Ultra-Orthodox Jewish women cover their hair.

This distinction is similar to the one noted by Heim (1982) (quoted in Krifka et al. (1995)) according to which the IS sentence in (48b), but not the quantified one in (48a), implies that dog owners indeed pay taxes on their dog: (48)

a. Every dog owner must pay tax on it. b. A dog owner pays tax on it.

In Kratzer’s (1981) framework, which Heim (1982) adopts, sentences like (47a) and (48a) are clearly interpreted w.r.t. to a deontic,

10

I thank Susan Rothstein (p.c.) for pointing this out.


nonrealistic accessibility relation, whereas, according to Heim, the implications of (47b) and (48b) show that generic sentences always have a realistic accessibility relation. As already mentioned above, Krifka et al. (1995) show that this claim cannot be true for all generics, since there clear examples of generics which can express purely deontic rules (like (44a) and (46a)). But the possible “real world” predictions of (47b) and (48b) (as opposed to their absence in (47a) and (48a)) should be nonetheless explained. I believe that sentences like (47b) and (48b) are, in fact, ambiguous, having (at least) two readings, which can be distinguished by imagining two different contexts of utterance. Firstly, they can be understood as expressing legal or deontic rules. With this “deontic” reading, which is more prominent with the imperative intonation and the contexts in (49a) and (49b), they are not significantly different from the explicitly modalized (47a) and (48a), i.e. they express legal or religious/deontic requirements: (49) a. A dog owner pays tax on it. (written in the “municipality regulations”) b. An ultra-Orthodox Jewish woman covers her hair! (said by the chief Rabbi of the ultra-Orthodox Jewish community in his weekly sermon) But, indeed, (47b) and (48b) can express not only such “purely legal” or “purely deontic” rules, but also, what I call “epistemic/stereotypical legal,” or “epistemic/stereotypical deontic,” generalizations. Imagine, for example, that my best friend wants to get himself a dog. Being aware of his poor financial situation, I utter (48b), since I know, or believe, that the rule requiring dog owners to pay tax on them is indeed followed (e.g. it is strongly enforced). Similarly, someone may utter (47b), not as a requirement imposed on ultra-Orthodox Jewish women, but having the stereotype, or even (wrongly) believing that all ultraOrthodox Jewish woman obey the traditional or religious norms (either willingly, since they are themselves so religious, or because women who don’t follow the norms cannot stay in the community). In the present framework the existence of the two kinds of readings will be explained as follows: The truth conditions of (47b) on both readings are (50):


(50) An ultra-Orthodox Jewish woman covers her hair is true in w iff

Where S CR(^ ULTRA - ORTHODOX JE WISH WOMAN,w). iff there is some Kratzerian function f (from worlds to sets of sets of worlds) s.t.

Moreover, in both readings of (47b), ^S, the “in virtue of” property, which is chosen is the same, and can be taken to be “follow the ultraOrthodox Jewish dress code for women.” On both readings, then, (47a) means: “In all worlds where every woman follows the ultra-Orthodox Jewish dress code for woman, every contextually relevant situation involving every non-exceptional ultra-Orthodox Jewish woman is a situation where that woman covers her hair.” The difference between the two readings is that in the “purely deontic” reading (as in (49a)), ^S is deontically associated, whereas in the “stereotypically/epistemically deontic” reading it is stereotypically or epistemically associated. This means that in “purely deontic” reading the condition on the choice of ^S is (51), whereas in the stereotypical deontic reading it is (52): w′ [w′ Rdeontic w] → [ d ultra-Orthodox Jewish woman (d, w′) → follows the ultraOrthodox Jewish dress code (d, w′)] (where Rdeontic is the accessibility relation which produces as value the set of worlds where what is commanded holds) w' [w' Rstereotypical / epistemic w] → [ d ultra-Orthodox Jewish (52) woman (d, w') → follows the ultra-Orthodox Jewish dress code (d, w')] (where R stereotypical / epistemic is the accessibility relation which produces as value the set of worlds where the stereotypes concerning w/facts true in w hold)

(51)

s, d Thus, the set of worlds in which the universal generalization C(s, d, w′)] → cover her hair [ultra- Orthodox Jewish woman (d, w′) (s, d, w′)] is asserted to be true can be taken to be a superset of two possibly different sets of worlds: Firstly, it can be taken to be a superset of the set of deontically accessible worlds (“In all worlds where every


ultra-Orthodox Jewish woman follows the ultra-Orthodox Jewish dress code, where this universal statement is itself a deontic necessity in our world, i.e. true in all worlds deontically accessible from w), or the superset of the stereotypically accessible ones (“In all worlds in which every ultra-Orthodox Jewish woman follows the ultra-Orthodox Jewish dress code, where this universal statement is itself an epistemic/ stereotypical necessity in our world, i.e. true in all worlds epistemically/ stereotypically accessible from w). Intuitively these two views of the accessible worlds lead to the two readings of (47a), i.e. to the fact that in the “purely deontic” reading the generalization concerning hair covering of ultra-Orthodox Jewish women is understood as true in virtue of a deontic/religious rule, whereas in the stereotypical/epistemic reading, it is understood as true in virtue of some stereotype or believed fact about our world. 5.3 High level of law-likeness: the “analytic” flavor of IS sentences As pointed out in section (1.1) above, the contrast in terms of lawlikeness between IS and BP sentences is felt even when the accessibility relation is not deontic, i.e. even when the IS sentence expresses only an epistemic rule or generalization. Compare, for example, the pairs in (53) and (54) (53) a. A computer is expensive. (Burton-Roberts’ paraphrase: “To be a computer is be expensive.”) b. Computers are expensive. (“It is normally/usually the case in reality that computers are expensive.”) (54) a. A Koko spider has 23 eyes. (“To be a Koko spider is to have 23 eyes.”) b. Koko spiders have 23 eyes (“It is normally/usually the case that Koko spiders 23 eyes.”) My claim is that the strong law-like flavor of IS sentences results from the special circumstantial nature (in Kratzer’s (1981) sense) of the accessibility relation, as defined in the truth conditions above, or, in other words, from the special kind of similarity between the actual world and the accessible worlds. Kratzer’s original distinction between epistemic and circumstantial modality can be illustrated by comparing the following two sentences:


(55)

a. Hydrangeas can grow here. b. There might be hydrangeas growing here.

Kratzer explains the difference between (55a) and (55b) by describing the following scenario: Suppose I acquire a piece of land in a far away country and discover that soil and climate are very much like at home, where hydrangeas prosper everywhere. Since hydrangeas are my favorite plant I wonder whether they would grow in this place and inquire about it. The answer is (55a). In such a situation, the proposition expressed by (55a) is true. It is true regardless of whether it is or isn’t likely that there are already hydrangeas in the country we are considering. All that matters is climate, soil, the special properties of hydrangeas, and the like. Suppose now that the country we are in never had any contact whatsoever with Asia or America, and the vegetation is altogether different from ours. Given this evidence, my utterance of (55b) would express a false proposition. What counts here is the complete evidence available, and that evidence is not compatible with the existence of hydrangeas. (p. 17–18) Thus, unlike the epistemic (55b), in the circumstantial (55a), we consider only a subset of the facts—facts of a certain kind (about the soil, climate the nature of hydrangeas), and neglect all the other facts. Sentence (55a) is felt to express a more “in principle” possibility about hydrangeas growing here than (55b), which expresses something concerning the dispositions of hydrangeas. This correlation between the circumstantial modal base and the “dispositional” or “in principle” reading seems to hold in other examples of circumstantial modals as well, as noted by Kratzer (e.g. in This car goes 220 miles an hour, This material is breakable etc.) Returning now to IS sentences, there is a sense in which the accessibility relation defined in the truth conditions in (40) falls under Kratzer’s original characterization of a “circumstantial” base . T he wor ld s in whi universal statement “Every P is Q” (e.g. “Every dog has four legs”) is asserted to be true preserve a rather narrow, or specific aspect of the actual world.11 These worlds are not required to parallel or resemble actuality in a wide and varied array of facts.


Moreover, the relevant specific aspect of the actual world, which is copied in the accessible worlds of IS sentences, is highly constrained, and in fact more constrained than what we find with Kratzer’s (1981) and Brennan’s (1993) definitions of circumstantial modality: In Kratzer’s theory this narrow aspect of the world can be any subset of, possibly accidental, facts about w. In Brennan’s development of Kratzer’s theory, this relevant “narrow aspect” of the actual world is more restricted: it is not any arbitrary subset of facts, but rather one which is “about” the subjects (in her terms: a (possibly complex) property which in the actual world is true of the individual denoted by the subject). In the present theory of IS sentences the relevant “narrow aspect” of the actual world, which is copied into the accessible worlds, is even more restricted: it involves a nonaccidental property associated with the property denoted by the subject, i.e. a universal statement of the form “every P is S,” which is itself conceived as nonaccidental in our world, i.e. as some sort of necessity or rule about our world. Thus, the accessibility relation in (40) above systematically produces a subset of the epistemic, deontic, legal, or stereotypical rules about the actual world. The generalizations expressed by IS sentences, then, are “rulebased”: they are asserted to be true in all worlds where “every P is S” holds, provided that “every P is S” is taken to be some epistemic, deontic, legal, religious, etc., rule in our world. It is this special “rule-based” circumstantial modality, which seems to give the generalizations expressed by IS sentences their “in principle” and strong law-like flavor.

11

Note also that Kratzer’s circumstantial conversational background (like Brennan’s “in virtue of a property” modal base) always contains a subset of the facts is the actual world, i.e. it is systematically realistic (the input world of evaluation is a member of the output set of worlds). The nonrealistic effects are achieved by adding nonrealistic ordering sources (deontic, stereotypical, etc.) This is not the case in the analysis I develop, since I do not use ordering sources at all, and allow the modal base of IS sentences to be nonrealistic (possibly containing subsets of commands, norms, stereotypes, etc).


5.4. IS sentences with subjects denoting “extremely unnatural classes” 5.4.1. Basic infelicity In the sections above we saw that although modal theories of generics like Krifka (1995) and Chierchia (1995) manage to distinguish between true and false IS generics (like (8) and (9), respectively, repeated here as (56) and (57)), they cannot distinguish the false IS generics from those IS sentences which are odd or infelicitous as generics, e.g. from (1 la, b) repeated here as (58a, b), whose subjects denote “extremely unnatural classes”: (56) A dog has four legs. (57) A dog has three legs. (58) a. #A Norwegian student whose name ends with ‘s’ or ‘g’ wears thick green socks. (cf. again: Norwegian students whose names end with ‘s’ or ‘g’ wear thick green socks). b. # A tall, famous shoemaker stands on one foot every other morning. (cf. again: Tall, famous shoemakers stand on one foot every other morning.) Unlike Krifka’s (1995) and Chierchia’s (1995) theories, the truth conditions in (40) above make the right distinctions quite naturally. Firstly, bracketing the problem of exceptions, we already saw above that sentence (56) is correctly represented as true (in virtue of having the four-leg genetic makeup, which is indeed associated with being a dog in w). Sentences like (57) are also correctly predicted to be false. Sentence (57) is true according to the truth conditions, iff there is a property ^S, belonging to the cluster of properties associated with being a dog in w, such that in all worlds in which every dog has this property, every situation involving every dog is a three-legs situation. But, among all those many properties associated with being a dog in w, i.e. those which are taken to be true of every dog in all epistemically, deontically, stereotypically, etc. accessible worlds, we can find no property which


leads to having three legs. Thus the sentence is correctly predicted to be false. The situation is different when the subject of the IS sentence denotes “extremely unnatural classes.” Take for example (58a). According to the truth conditions in (41) above, this sentence is represented as in (59): (59) A Norwegian student whose name ends with ‘s’ or ‘g’ wears thick green socks is true iff:

where ^S C R(^NORWEGIAN WHOSE NAME ENDS WITH ‘S’ OR ‘G,’ W) . iff there is some Kratzerian function f (from worlds to sets of sets of worlds) s.t.

In order to evaluate the sentence, we have to find a property ^S, associated with being a Norwegian student whose name ends with ‘s’ or ‘g,’ and demand that in all worlds in which every such student has ^S, every relevant situation involving such a (non-exceptional) student is a situation where he wears thick green socks. The problem with (50) is that “extremely unnatural classes,” like the one denoted by the subject of (58a), are felt to be so unnatural exactly because we associate no properties with them. That is, we have no known facts, stereotypes or norms concerning these classes, and we don’t assume that belonging to such a class (and to similar ones like being a tall, famous shoemaker, being a browneyed dog born on June 5, 1987, etc.) implies having other nontrivial properties. Note that this is not because the ^P properties in such sentences are complex properties, since there are many complex properties, which are perceived as very “natural.” In this sense I think that Chierchia’s (1998) characterization of natural kinds (quoted in chapter 1, and repeated here) suits our purposes very well. Chierchia writes: By natural kinds I do not necessarily mean […] just biological ones, or even ‘wellestablished’ ones. Artifacts (like chair or cars) or complex things (like intelligent students or spots of ink) can


qualify as (natural) kinds to the extent we can impute them a sufficiently regular behavior. (p. 6) Here Chierchia speaks about kinds denoted by BP NPs, but his intuition is very similar to the one I have about natural properties denoted by IS NPs. Following Chierchia’s intuition, a property is “natural” if we can impute a sufficiently regular behavior to it. In our terms, this happens if we associate other properties with this property, i.e. if we have some norms, stereotypes, known facts, etc. concerning this property. In this sense, the property of being a Norwegian student whose name ends with ‘s’ or ‘g’ is indeed a very unnatural property, since normally we associate no other nontrivial properties with this property, i.e. there is no set of norms, stereotypes or facts which tell us that every such student has some other property, besides itself belonging to the set of such students. The same is true w.r.t. the property of being a lion called Nimb, a tall, handsome postman who was born on June 5, 1958, etc. Thus, unlike the false (57), in which there are lots of properties associated with ^P (being a dog), but none of which made the universal statement true, in the case of (58), no appropriate ^S property can be found in the first place. Intuitively, in such a case we cannot start evaluating the sentence. In other words, finding an appropriate ^S property, namely one associated with ^P in w, is a necessary condition for characterizing the set of accessible worlds w′. We define this set as the set of worlds where every P individual has S, where ^S is associated with ^P in w. If no appropriate ^S property exists, no accessible world w′ exists either. But in this case we cannot start checking the truth of the universal statement in all accessible worlds. We can say that the infelicity of IS sentences like (58a, b) results from the failure of an existence presupposition, conventionally triggered by the universal quantification over worlds. Similar existence presuppositions are very common with universal quantification over individuals (see, e.g., de Jong & Verkuyl (1984)), as (60): (60)

Every male student got below 60 on the exam.

Assume that the context in which (60) is uttered is the context of Professor Smith’s class. The domain for quantification is the domain of male students in this class, and we presuppose that this domain is not empty. If there are no male students in this class, the presupposition fails, and (60) is infelicitous.


In a similar manner we can claim that the universal quantification over worlds which fit a certain characterization (namely those where every P individual has an associated S) conventionally triggers a presupposition that such worlds exist. Thus, if the cluster C(^P,W) is empty since no property is associated with ^P, there is no world which fits the characterization, and the presupposition fails, leading to infelicity.12 What all that means is that the truth conditions in (40) above should be supplemented with the conventionally triggered presupposition in (61): (61) Presupposition (conventionally triggered by the universal quantification over worlds w'): C(^P,w) is not empty (i.e. there are some properties associated with ^P, so ^P is a natural enough property). 5.4.2 Contextual support for infelicitous IS sentences As shown in section (1.1), the generic reading is not completely blocked for IS sentences like (58a, b) above. Informants report that the potential generic reading shows up much more prominently when these sentences are not uttered “out of the blue” (as in (62a, b)), but given supporting context, as in (63a, b): (62) a. # A Norwegian student whose name ends with ‘s’ or ‘g’ eats lots of bananas every other Monday. (prominently existential) b. # A tall, left-handed neurologist from Hadassah Hospital earns more than $300,000 a year (prominently existential) (63) a. There are very interesting traditions in Norway concerning final letters of the name and certain diets. For example, a Norwegian student whose name ends with ‘s’ or ‘g’ eats lots of bananas every other Monday b. The new health minister is a tall, left-handed neurologist from Canada. It seems that this somehow affects the criteria for 12

Another possible way to phrase the problem is to say that if the cluster C(^P,W) is empty, the conjunction [S(d, w′) ^S CR(^P,w).], and consequently also antecedent of the whole universal statement are false. In such a case the whole universal statement is trivially true, which leads, in turn, to infelicity.


determining salaries. For example, from now on a tall, lefthanded neurologist from Hadassah Hospital earns more than $300,000 a year. This contrast can now be explained in the following way. We said that part of the meaning of IS sentences is the presupposition that the set CR(^p, w) is not empty, that is, that there is at least one property ^S, nontrivially associated with ^P in w. For this presupposition to hold we should find a set of known facts, stereotypes, commands, etc. (i.e. a Kratzerian modal base) from which “every P is S” follows. Now, when an IS sentence is uttered out of the blue, the relevant set of known facts, stereotypes, norms, etc. available for the listener are only the ones shared by the community of language users, i.e. in the null context the basis for associating properties with ^P is the shared real world knowledge. The problem with sentences like (58a, b) and (62a, b) is that given this shared real world knowledge, no property is associated with the “extremely unnatural” ^P properties, and so these sentences become infelicitous. However, language users may associate properties with other properties, not only on the basis of the shared realworld knowledge, but also on the basis of private knowledge, private stereotypes or private norms. For example, the community of language users may take “tall, left-handed neurologists from Hadassah Hospital” to be an unnatural class, since it has no knowledge, stereotypes, or norms concerning this class, and thus does not take the members of this class to share any interesting, nontrivial property. However, specific individuals (e.g. the ones who work at Hadassah Hospital) may have some private knowledge, stereotypes, or norms about this class, and thus take it to be “natural” (in Chierchia’s (1998) terms they “impute such a class with a sufficiently regular behavior”). What happens in (63a, b) is that the IS sentences are not uttered out of the blue (i.e. against the shared real world knowledge), but in a context where knowledge, stereotypes, etc. about the ^P property are made available by the speaker and can be accommodated by the listener. Such contexts turn the previously “extremely unnatural classes” denoted by the subjects of the IS sentences into more natural ones, by “imputing them with a sufficiently regular behavior” (in Chierchia’s (1998) terms), i.e. by associating some properties with them (e.g. “obeying the traditions concerning clothing” in (56a), and “being similar to the new health minister” in (56b)). In such cases, then, the


presupposition requiring CR(^P,w) to be nonempty is met, and the sentences become felicitous as generic. 5.4.3 Inductive contexts Context, then, can improve the generic reading of IS sentences which are infelicitous when uttered out of the blue. However, as noted above, not any context can do this. This can be seen again in (64) and (65) (again the BP counterparts have prominent generic reading): (64) a. While sitting in Kew Gardens I noticed that an old man wearing a red shirt bends under his bench whenever an airplane flies overhead. (cf. While sitting in Kew Gardens I noticed that old men wearing red shirts bend under their benches whenever an airplane flies overhead). (65) When I was a zookeeper I loved watching the lions. I noticed that a relatively small lion called Leo eats cornflakes for breakfast. (cf. When I was a zookeeper I loved watching the lions. I noticed that relatively small lions called Leo eat cornflakes for breakfast. In chapter 1 we explained that embedding under the verb notice seems to trigger a “pure” inductive reasoning context, i.e. a context where the generalization is based only on observing several (or even many) instances of the ^P set having the predicated property. The generalization, then, is that with such a pure inductive context the IS sentences with “extremely unnatural classes” subjects had the same existential prominent reading as with “out of the blue” utterances (e.g. in (62)), and unlike the contextually supported ones, in (63) above. The reason for this is easily seen now. When our shared real world knowledge, stereotypes, norms, etc., associate no properties with the “extremely unnatural” ^P property, inductive reasoning cannot help us either. There is nothing about observing several old men wearing red shirts bending under their benches when an airplane goes by, which can, by itself, help us generally associate some property with the property of “being an old man wearing a red shirt.”


5.4.4 “Linguistically associated properties” as improving the felicity of IS sentences IS sentences like those in (58) and (62), then, are only felicitous in special contexts, and not “out of the blue” or in “inductive” contexts, since shared realworld knowledge does not associate properties with “extremely unnatural properties.” But compare now the infelicitous (58a) (repeated here as (66)) with (67a) and (67b): (66) # A Norwegian student whose name ends with ‘s’ or ‘g’ wears thick green socks/publishes every other month an excellent paper in Phonology. (67) a. A Norwegian student whose name ends with ‘s’ or ‘g’ is a person from Norway, who is a student, and who has a name whose final letters are ‘s’ or ‘g.’ b. A Norwegian student whose name ends with ‘s’ or ‘g’ is a person from Norway, who is a student, and who has a name whose final letters are ‘f’ or ‘l.’ Sentence (67a) as well as the clearly false (67b) are significantly better as generics than (66), although they have the same IS subject, denoting what we called an “extremely unnatural class,” and although they too are uttered out of the blue. The “in virtue of” truth conditions in (61) above, can now explain this unexpected difference naturally. Firstly, the felicity of (67a, b) indicates that unlike what we said above, the cluster of properties CR(^ NORWEGIAN WHOSE NAME ENDS WITH ‘S’ OR ‘G,’W). is not necessarily empty in the null context, i.e. there are cases where the shared real-world knowledge does associate properties with the apparently “unnatural “properties “be a Norwegian student whose name ends with ‘s’ and ‘g,’” But what are these “in virtue of” properties? Cohen (2001) shows that many infelicitous IS sentences (like (68)) become better when rephrased in a form of a Aristotelian definition, i.e. as statement in the form of a genus and differentia, as in (69) (where uncle is the genus and garrulous is the differentia): (68)

# An uncle is garrulous.


(69)

An uncle is a garrulous relative.

In section (6) below I deal with infelicitous IS sentences like (68). What is relevant for us now, however, is that like (69), sentences (67a) and (67b) above are also felt to express linguistic definitions (and not e.g. deontic, epistemic or legal rules).13 Recall now that in section (5.1) above we showed that the various “flavors” of IS sentences are a result of the various “in virtue of” properties associated with the denotations of the IS subject. Assuming this is true, the “linguistic” “flavor” of (67a, b) suggests that their “in virtue of” ^S, property is “linguistically” (rather than epistemically, deontically, etc.) associated with the denotation of the IS subjects. This means that this “in virtue of” property is taken to be true of every Norwegian student whose name ends with ‘s’ or ‘g’ in all linguistically accessible worlds, i.e. the worlds where language is interpreted as it is in our world. This “linguistically associated” property can be defined as ^λd. has all the properties entailed by the meaning of ^P (d). Such a property can be indeed true of every member of P in all linguistically accessible worlds, even if all these members of P share no other properties in all other epistemic, deontic, legal, etc. worlds (i.e. even if ^P correspond to an “extremely unnatural class”). According to the definition of association, then, repeated in (70), this property can be taken to be associated with “be a Norwegian student whose name ends with ‘s’ or ‘g,’” since (71) is true: CR(^P,w). iff there is some (70) Association of properties ^S Kratzerian function f (from worlds to sets of sets of worlds) s.t.

13

Let me emphasize that, although this linguistic/“definitional” “flavor” seems indeed to result from the fact that, like (69), sentences (67a, b) are also phrased in terms of a genus and differentia, the correlation between this form of statements and having definitional flavor is not my concern here. In other words, I do not attempt to claim here that this form of genus and differentia is a necessary and/or sufficient condition for a statement to be perceived as a definition, nor to formulate any such claim more precisely, The important point for me is that the intuition about the “flavor” of (67a) and (67b) (as about Cohen’s (69)) is itself is very strong: these sentences are clearly perceived as expressing linguistic definitions and not deontic, legal, religious or epistemic rules.


w′ [w′ is a world where language is interpreted as in our world] → [ d [Norwegian student whose name ends with ‘s’ or ‘g’ (d, w′)] → [has all the properties entailed by the meaning of “Norwegian student whose name ends with ‘s’ or ‘g’”(d, w’)] Taking now the property “has all the properties entailed by the meaning of ‘Norwegian student whose name ends with “s” or “g”’” to be the appropriate ^S property in the truth conditions of (67a) and (67b) above, now correctly predicts that (a) both sentences will be equally felicitous (since in both of them the presupposition that CR(^P,w) is nonempty is met), and (b) that (67a) is true, and (67b) is false. 6. An additional contextual requirement on IS sentences 6.1 The problem: “association” of ^S with ^P is not enough We have seen that the “in virtue of” truth conditions in (61) above can account for a number of special properties of IS sentences: the fact that they are taken to express “in principle” or “law-like” rules, the fact that these rules can be of different “flavors” or kinds, and the fact that context affects the felicity of IS sentences with subjects denoting “extremely unnatural classes,” like (72a, b). In addition, the truth conditions make the right predictions w.r.t. clearly false IS sentences like (73a, b), and, ignoring in the meantime the problem of exceptions (whose solution is postponed to chapter (4)), w.r.t. the true (74a, b): (72) a. # A Norwegian student whose name ends with ‘s’ or ‘g’ wears thick green socks. b. # A tall, handsome postman who was born in June 1975 drives wildly. (73) a. A dog has three legs. b. A donkey has short ears.


(74) a. A dog has four legs. b. A donkey has long ears. Despite all these successes, though, the truth conditions in (61) above are not complete, since they make the wrong predictions w.r.t. sentences like (75a, b), which are infelicitous as generic: (75) a. # A man is blond. b. # A grandmother is addicted to blue marshmallows. (cf. again the felicitous: Men are blond/Grandmothers are addicted to blue marshmallows) Above we attributed the infelicity of (72a, b) to the fact that their subjects denote “extremely unnatural classes,” which in the null context are not associated with other properties. We claimed that the contextual requirements on ^S (i.e. association with ^P) cannot be met, and that this leads to infelicity. The problem is that, unlike the subjects in (72), the subjects of (75a) and (75b) denote very natural properties. Being a man, or being a grandmother are natural in that these are properties which we have many known facts, stereotypes, norms, etc. about, and therefore we naturally and very easily associate many other properties with them, like having a Y chromosome, loving sports, and having a hairy chest (with “be a man”), and spoiling their grandchildren, and being old and wise (with “be a grandmother”). But this means that according to our truth conditions, sentences like (75a) and (75b) are predicted to have exactly the same status as the false-but-felicitous (73a) and (73b). Take, for example, (75a), which according to our truth conditions in (61) above is true iff the conditions in (76) hold: (76) A man is blond is true in w iff

And S CR(^MAN, w). iff there is some Kratzerian function f (from worlds to sets of sets of worlds) s.t. Presupposition: CR(^MAN, w) is not empty (i.e. the property of being a man is a natural property)


According to (76), “A man is blond” is true iff there is a property associated with being a man, i.e. true of every man in all worlds which are epistemically, stereotypically, legally, deontically, etc., accessible from our world, such that in all worlds in which every man has this property, every situation involving a (non-exceptional) man is a situation where such a man is blond. Now, there are indeed many properties associated with being a man in w, so the presupposition that CR(^MAN, w) is nonempty is met, but among all of them, none leads to blondness. This means that (75a) is predicated to be felicitous and false, just like (73a) in which none of the many properties associated with being a dog is taken to lead to having three legs. The problem is that whereas (75a) is indeed taken to be false, in reality it is different from the felicitous (73a) in being infelicitous as well. 6.2 The intuition: A second contextual requirement on the ^S property The infelicity of (75a,b), then, suggests that the requirement on the association of ^S with ^P is not enough. The intuition is that, given our real-world knowledge, sentences (75a, b) above express, again, rather unreasonable generalizations. But what is the source of this “unreasonableness”? The problem does not seem to lie in our real-world knowledge about ^P itself (with its associated properties), but is rather related to our real-world knowledge about the connection between ^P, the property expressed by the IS subject, and ^Q, the property expressed by the VP. This is supported by the fact that changing the VPs of such sentences while maintaining the same subjects leads to felicity, as seen in (77), (78) and (79): (77)

a. # A man is blond. b. A man has a beard.

(78) a. # A carpenter in Amherst gives all his sons names ending with ‘a’ or ‘g.’ b. A carpenter in Amherst pays 37% income tax. (79) a. # A child in this kindergarten has a sister who doesn’t like pickles. b. A child in this kindergarten wears a long-sleeved shirt.


A possible way to capture the problem with the infelicitous IS cases here is to develop the intuition, raised in chapter 1 above, that the VPs of these sentences express ^Q properties which are too “unconnected” (given the realworld knowledge) with the ^P properties, expressed by the subjects. The intuition I am going to develop, however, is a bit more complicated. Instead of focusing directly on the reasonable relationship between ^P and ^Q, I will concentrate, in a somewhat indirect way, on the relationship between any ^S property associated with ^P and the ^Q property. The justification for preferring the more complicated intuition over the simpler one will be given in section (6.4) below. The rest of this section and section (6.3) are devoted to formally clarifying the more complicated intuition. The basic intuition is that for an IS sentence to express a reasonable “in virtue of” generalization, and to be felicitous, we need to impose a further type of presuppositional requirement on the choice of the “in virtue of” ^S property. Unlike the first presuppositional requirement on ^S, which has to do with the pragmatic-based connection between ^S and ^P (the denotation of the subject), this second requirement has to do with the pragmatic-based connection between ^S and ^Q (the denotation of the VP). Specifically, we need to find an ^S property which, besides being associated with ^P, is also taken to be a “reasonable causer” for properties of the sort of ^Q. If this requirement cannot be met the IS sentence is infelicitous.14 Compare, for example (73a) (“A dog has three legs”) and (75a) (“A man is blond”). Both sentences are the same in that neither has a property associated with ^P (being a dog and being a man) which is taken to be a “reasonable causer” of the ^Q property itself (i.e. of having three legs and of being blond, respectively). Nonetheless, when we think about properties of the sort of ^Q, we can see that there is a difference between the two sentences: We can easily find a property associated with being a dog which is a reasonable causer for having a certain number of legs (i.e. a property of the sort of “having three legs”), namely having the regular genetic makeup for a certain, specific, and fixed number of legs. In contrast, among all the properties we associate with being a man there is no property which we take to be a 14

Note that this intuition is reminiscent of Carlson’s (1995) idea, reviewed in chapter 1 above, about the abstract causal rules which are involved in generics, according to the “rules and regulation” approach. In making this intuition precise, though, I will avoid explicit use of the notion of causation itself and make use of other, more standard, formal tools.


reasonable causer for having a certain specific and fixed hair color, or for any other property of the sort of “being blond.” 6.3 Formalizing the intuition I will now discuss the way to capture the above intuition and the distinction between the felicitous and infelicitous IS sentences, where the formal tool I use is indeed a second presupposition on the choice of the ^S property, added to the truth conditions of such sentences. As will become clear from the discussion below, it is a complicated matter to understand what exactly it is that all felicitous IS sentences share and all felicitous ones do not. In other words, neither group is homogeneous in an obvious way. After going through several versions of presupposition I arrive at a version which not only slices the cake in the right way, but is also triggered in a natural way. 6.3.1 The extensional disjunction A possible way to capture the intuition above is to assume that IS sentences have the following additional presupposition15: (80) Presupposition:

Paraphrase: Either every contextually relevant situation involving a P individual with S is a Q situation involving such an individual, or every such contextually relevant situation is not a Q situation involving such an individual. Ignoring for the moment the problem of exceptions, adopting the presupposition in (80) makes the right predictions w.r.t. the false and felicitous (73a) and the infelicitous (75a), respectively: (81) Presupposition of (73a):

15

I thank Gennaro Chierchia for suggesting the direction of this formalization.


^S is s.t.

Paraphrase: Either every contextually relevant situation involving every dog with S is a situation where that dog has three legs, or that every such situation, involving every dog with S is a situation where that dog does not have three legs. (82) Presupposition of (75a): ^S is s.t.

Paraphrase: Either every contextually relevant situation involving every man with S is a situation where that man is blond, or that every such situation involving every man with S is a situation where that man is not blond. Taking ^S to be “^having the regular genetic makeup for a certain, specific number of legs,” which is a property associated with ^P, and ignoring for the moment the existence of dogs who lost a leg in an accident or as a result of a mutation (i.e. “abnormal” exceptions) the presupposition in (81) is met. This because the second, “negative” disjunct in the disjunction is true: every contextually relevant situation involving a (non-exceptional) dog with such a genetic makeup indeed does not have three legs. Thus, although (73a) expresses the wrong generalization, it expresses, nonetheless, a right “sort of” generalization. The presupposition in (82), however, is not met. No property associated with being a man (e.g. having a Y chromosome, having low voice, loving football, etc.) will make either the first, “positive,” nor the second, “negative” disjuncts true: Neither is it true that every (nonexceptional) man with any such associated ^S is blond, nor that every man with such an ^S is not blond. Men are not the kind of species who have some genetic makeup, or any other factor leading to a specific hair color, but are known to have a variety of hair colors. The presupposition in (82) then is not met, and (75a) is correctly predicted to be infelicitous. The same problem seems to arise with (78a) above (# A carpenter in Amherst gives all his sons names ending with ‘a’ or ‘g’). No property


associated with being a carpenter in Amherst (i.e. that every such carpenter has in all epistemically, deontically, legally, etc., accessible worlds) leads to naming children with names ending with certain, specific letters of the alphabet (‘s’ or ‘g,’ or not ‘s’ or ‘g’). 6.3.2 Problems with the “extensional disjunction” The “extensional disjunction” presupposition above, however, faces two serious problems. Firstly, as already mentioned above, this presupposition can distinguish between IS sentences like (73a) and (75a) only if we systematically ignore exceptions. If we don’t, this presupposition fails in the case of the felicitous (73a) (“A dog has four legs”), since in fact not all dogs with the regular genetic makeup for a certain number of legs do not have three legs (some dogs who had a mutation or an accident do have three legs). An additional problem with the disjunction in (80)) is that it gives the wrong results w.r.t. felicitous IS sentences like (83) and (84): (83)

A carpenter in Amherst pays 17.6 % income tax.

(84)

A child in this kindergarten wears a long-sleeved shirt.

We said above that A dog has three legs is false, since it expresses the “wrong generalization” about the number of legs. This fact also makes it felicitous, since there is a true alternative generalization around (namely that all dogs with the associated ^S have some other number of legs, i.e. four), i.e. a property of the sort of ^Q is caused by an ^S property associated with ^P. On the other hand with # A man is blond there seemed to be simply no generalization around about a genetic makeup (or any other factor) associated with being a man which leads to a property of the sort of ^Q, i.e. to a specific hair color. Intuitively, this is why ^Q in this case seems so “unconnected” with ^P. But in this last respect the felicitous (83) and (84) are just like the infelicitous (75a) and (78a), in that neither expresses the “wrong generalization”: there is in fact no true generalization about the percentage of income tax that carpenters in Amherst pay, or even are required to pay. In actuality each carpenter is required to pay, and indeed pays, a different percentage, depending on his or her income. Similarly, there is no epistemic, or even deontic generalization about the


length of the sleeves in this kindergarten: some children wear short-sleeved shirts and some wear long-sleeved shirts. More formally, when we think about properties associated with being a carpenter in Amherst, or a child in this kindergarten, neither of the presuppositions in (85) and (86) is met: (85) Presupposition of (83): ^S is s.t.

(86) Presupposition of (85): ^S is s.t.

The problem with both (85) and (86) is the same. There are various properties we may associate with the properties “being a carpenter in Amherst” or “being a child in this kindergarten,” i.e. many kinds of necessities concerning them, e.g. obeying the municipal regulations of Amherst, and obeying the regulations of this kindergarten, respectively. However, there is no regulation, or any other factor which actually requires all carpenters in Amherst to pay a certain percentage of income tax, or all children in this kindergarten to wear a certain sort of shirt. In fact, the tax regulations in Amherst are, as in many other places, keyed to the level of income, so each carpenter is required to pay a different percentage. Similarly, we know of no regulation in this kindergarten which has to do with the kind of shirt that the children have to wear. So no property associated with being a child in this kindergarten will cause all such children to wear a certain sort of shirt. Thus, in both (85) and (86), neither the “positive” nor the “negative” disjuncts is true, and as a result both presuppositions fail, wrongly predicting both these sentences to be infelicitous.


6.3.3 A “good possibility” operator over the disjunction 6.3.3.1 The intuition Let’s look again at the felicitous and infelicitous IS sentences we have examined so far: Felicitous (87)

a. A dog has four legs. b. A dog has three legs. c. A carpenter in Amherst pays 37% income tax. d. A child in this kindergarten wears a long-sleeved shirt.

Infelicitous (as generic) (88) a. # A man is blond. b. # A child in this kindergarten has a sister who doesn’t like pickles. c. # A carpenter in Amherst gives all his sons names ending with ‘a’ or ‘g.’ In the previous section we saw that the “extensional disjunction” presupposition fails in the case of both (87a, b) (due to the problem of exceptions in the actual world), and (87c, d) (since no property associated with ^P leads to properties of the sort of ^Q in the actual world). Thus, the “extensional disjunction” presupposition wrongly predicts the sentences in (87) to be infelicitous, just like the ones in (88). In reality, however, we can feel a difference between the two kinds of sentences. Although the disjunction of universal statements is false in the actual world for both kinds of sentences, it is felt to be much more reasonable or less far-fetched in the former cases than in the latter ones. For example, although neither the requirement to pay a certain percentage of income tax nor the requirement to give one’s son a name ending with specific letters characterize all carpenters in Amherst in the actual world, the latter characterization seems much more far-fetched and bizarre than the former. Take again, for example, the property of obeying the regulations or norms in Amherst (which is deontically associated with being a carpenter in Amherst). We can rather easily think about reasonable


scenarios which will lead to the truth of the first characterization: the Amherst municipality wants to encourage carpenters, so it allows them to pay a very low income tax. Or perhaps there are too many carpenters in Amherst already, and the municipality wants to discourage other carpenters from moving in by imposing a very high income tax. Or perhaps there is an ancient tax law in Amherst where every profession indeed pays a fixed percentage of income tax: carpenters pay 37%, lawyers pay 45%, academics pay 89% etc. Given our real-world knowledge, then, the relationship between professions or preferences of municipalities on the one hand, and tax regulations in various places on the other hand, is not a far-fetched connection. It is much more difficult to find a story which will explain the second characterization of carpenters in Amherst: Given our real world knowledge, the connection between professions and norms concerning the final letters of children is rather unusual. Intuitively, this is why (87c) is felt to express a much more reasonable generalization than (88c). Thus, although neither characterization is true in the actual world, the first seems more reasonable, less far-fetched, or simply a better possibility than the second. I suggest, then, that it is neither truth in the actual world, nor mere possibility, but the relation of good possibility which is relevant for the disjunction of universal statements, i.e. that the second presupposition of IS sentences is something like (89): (89)

Presupposition ^S is s.t. There is a good possibility that

6.3.3.2 Defining the right kind of “good possibility” The notion of “good possibility” suggests that the kind of modality we are interested in is graded modality. In Kratzer’s (1981) terms, this notion is represented as existential quantification over a set of worlds, which is restricted not only by the basic modal base, or simple accessibility from the world of evaluation, but also involves similarity or closeness to some ideal version of our world, given by a second accessibility relation, (an “ordering source”), which produces the set of those accessible worlds which are “closest” to the ideal world. Sentence


(89), then, can be more precisely written as in (90), where w′ is the world which is Kratzerially accessible and closest to that idealized world: (90) Presupposition ^S is s.t. There is a world w′, (Kratzerially accessible from w, and closest to an ideal version of our world) s.t.

For example, unlike the “extensional disjunction” version, this presupposition is met in the case of the felicitous A dog has four legs since, although the claim that there is a specific number of legs that every dog with a four-legged genetic makeup has is clearly false (due to the existence of mutations, accidents and the like), we can take this claim to be true in some idealization of our world, where accidents, mutations, etc. are neutralized. At first sight, this way of defining the ideal world may seem equivalent to Krifka’s (1995), and Krifka et al.’s (1995) “normal ordering source” solution to the problem of exceptions, reviewed in section (4.5) above, in which the accessible worlds (which are universally quantified over by the generic operator) are further restricted to be closest to an ideal “most normal” world, where no unexpected things (like accidents, mutations, etc.) occur. However, there is an independent problem with the “normal ordering source” solution, which is relevant for our case as well, and which will force us to define the ideal world in a finer-grained way. The problem is that if “normality” is defined relative to the overall course of events in the actual world, then it is not clear in what sense accidents and mutations are “not normal” since their existence is also part of the definition of “the normal course of events” in our world. In other words, a world where there are no mutations or accidents will be considered in fact a very abnormal world, relative to our actual world. But then, if mutations and accidents are part of any word which is “normal” relative to our world, the presupposition of the felicitous A dog has four legs is still wrongly predicted to fail. The idealized version of the world that we are interested in, then, is not a one where everything is normal, and nothing unexpected happens (since in such a world we will still get accidents and mutations, and hence both four-legged and three-legged dogs). Intuitively, both the


four-leggedness of dogs, as well as the existence of mutations are “normal” processes in our world. But in actuality these two processes sometimes interact, so that the more general one (e.g. having four legs as a result of having a four-legged generic makeup) is blocked, or inhibited by the less general one (i.e. by having mutations, accidents, etc.). We can think, however, about an idealized versions of our world where these two processes do not interact. In other words, instead of thinking about an idealized version of our actual world, where everything is “normal,” we can think about a class, or a set of idealizations, of our actual world, where things are normal w.r.t. to some general process, i.e. where in each such idealization a certain general principle is allowed to work without being interrupted or inhibited by other, less general, factors. In one such world, for example, the intention of people to behave in a certain manner (e.g. to dress in a certain way) will eventually lead to this certain behavior, provided nothing blocks this intention. In another such world, genetic makeup for a certain property will yield this property, without being inhibited by accidents, mutations, etc. In this idealized version of our world every dog with a four-legged makeup indeed has four legs, so the presupposition in (90) can be met and the felicity of (87a), as well as of (87b) is explained. 16 , 17 Defining the ideal world in the presupposition in (90) in this way also correctly accounts now for the felicity of IS sentences like (87d) A child in this kindergarten wears a long-sleeved shirt as opposed to the infelicity of (88b) # A child in this kindergarten has a sister who does not like pickles. The idea is that there is no general principle in our world which reasonably connects property associated with attending a certain kindergarten, and having a sister who doesn’t like a certain sort

16

The problem raised here, with using a “normal” ordering source to solve the problem of exceptions to generics is similar to the problem raised by Vlach (1981) (quoted by Landman (1992)) against Dowty’s attempt to use “inertia worlds” (worlds where things take their normal course of events w.r.t. a world and an interval) for solving the “imperfective paradox” with progressive accomplishments (as in Mary was crossing the street/John was building a house). Dowty uses inertia worlds to eliminate “unexpected occurrences” (like trucks hitting Mary while she is crossing the street), by claiming that such occurrences are not expected to happen in worlds where nothing unexpected happens. Vlach, however, points out that in many situations (e.g. when a truck is already very close to Mary) such occurrences do not constitute a break or an interruption no new line in the natural course of events, but are in fact the natural course of events.


of food (e.g. pickles). Thus, we do not expect this characterization to hold in an idealization of the actual world where a general principle holds without being violated. In contrast, our real-world knowledge tells us that many educational institutes require their pupils to dress in a certain way (in order to show that they belong, as a result of some required social or religious norms, or even because of the climate). Given this real-world knowledge we can treat a world where every child in this kindergarten with some specific associated property (like obeying the regulations) is required to wear a certain kind of shirt as not too far-fetched, i.e. as closest to an idealization of our world where some general principle holds without being inhibited. 6.4 Further motivations for the second presuppositional requirement on the ^S property 6.4.1 An apparently better alternative: limiting ^Q and not ^S The “good possibility” disjunction in (89), phrased more formally in (90), then, slices the cake correctly, i.e. successfully distinguishes between the class of felicitous and infelicitous IS sentences (as in (87) and (88) above, respectively). The intuition behind this disjunction can be summarized in (91): (91) Intuition about IS sentences: IS sentences are felicitous only when given our real world, there is a good possibility (i.e. it is reasonable) that ^S, some property associated with ^P in w, leads to a property of the sort of ^Q.

The solution I provide here for the problem of exceptions in the presupposition on IS sentences (i.e. of relativizing normality to only one general process, instead of to the overall course of events), is somewhat similar to Landman’s (1992) solution to the problem raised by Vlach. Roughly, Landman defines worlds which are reasonable options for a certain event in a world. In these worlds an event in a world is allowed to develop and continue solely on the basis of what is internal to the event itself in the evaluation world (e.g. Mary’s intention and ability to cross the street), without considering outside factors which might block it (like trucks).


A question which is naturally raised at this point, however, is whether we should not make the simpler requirement (mentioned in section (6.2) above) that ^P (rather than the combination of ^P and ^S together) reasonably leads to properties of the sort of ^Q. In other words, we may want to replace the intuition in (91) with the simpler intuition in (92), and consequently, the presupposition in (89) above, with the simpler presupposition in (93): (92) A simpler intuition about IS sentences IS sentences are felicitous only when, given our real world, ^Q (the property expressed by the VP) is a reasonable “sort of property” to be predicated of ^P, (the property expressed by the IS subject). (93) A simpler presupposition of IS sentences There is a good possibility that

Paraphrase: There is a good possibility that either every contextually relevant situation involving a P individual is a Q situation, or that every such situation involving a P individual is not a Q situation. In the simpler presupposition we do not concentrate on some associated property, ^S, and check its reasonable connection with properties of the sort of ^Q. Rather we concentrate on ^Q itself, and check whether it is a reasonable sort of property to be true of all P individuals.

17

At first sight it may seem that the ordering source I suggest here may be useful not only in solving the problem of exceptions for the presupposition of IS sentences, but also for the truth condition of such sentences (e.g. for excluding three-legged dogs from the universal generalization expressed by A dog has four legs). In chapter 4, however, which is wholly devoted to the exceptions problem, I show that this cannot be done. Thus, we continue to use the finergrained ordering source in the presupposition, but develop another mechanism for tolerating exceptions for the universal statement in the truth conditions, namely, a development of Kadmon and Landman (1993) “domain vague” restriction on the generic quantifier. The presupposition developed here, however, will play an indirect role in the definition of the new exceptionstolerance mechanism.


The simpler presupposition in (93) makes exactly the same (correct) predictions w.r.t. the distinction between felicitous IS sentences (as in (87) above), and infelicitous ones (as (88) above), as the original presupposition in (91). For example, using this simpler presupposition we correctly predict A man is blond to be infelicitous, since there is no good possibility, relative to any general process in our world, that a specific hair color characterizes men. More precisely, we do not expect there to be an idealization of our world, where some general principle is not violated, and where every man is blond, or every man is not blond. On the surface there seems to be no need to relate also to an ^S property associated with being a man. However, despite the equivalent empirical predictions of the two presuppositions, and the relative simplicity of the one in (93), my claim is that it is justified to prefer the original, more complicated presupposition in (89) above, which imposes pragmatic requirements on the choice of ^S. Two such justifications appear in sections (6.4.2) and (6.4.3) below. 6.4.2 Triggering of the presupposition One reason to prefer the original presupposition in (89) over the simpler one in (93) lies in the fact that although both presuppositions make the right predictions, the former can be much more easily shown to be triggered than the latter. There is now a general consensus (see e.g. Chierchia & McConnell Ginet (1990) and Kadmon (2001)) that presuppositions cannot suddenly appear in the semantics of a certain construction. Rather, they are introduced or triggered by something. This “something” is frequently a component in the grammar or semantic structure of the specific construction, e.g. a universal operator, which triggers a presupposition of the nonemptiness of the domain (like the first presupposition of IS sentences, discussed in section (5.3.1) above). However, according to e.g. Kempson (1975) and Wilson (1975), and more recently Chierchia & McConnell Ginet (1991) and Kadmon (2001) (who follow Stalnaker’s (1974) idea of “pragmatic presuppositions”), some presuppositions are triggered by non-linguistic, but rather pragmatic premises, e.g. by Grician maxims, i.e. they are conversationally (rather than conventionally) triggered. An example of a conversationally triggered presupposition, cited by Kadmon (2001), is seen in (94a-c):


(94)

a. I’m out of petrol. b. There is a gas station around the corner. c. Gas stations sell petrol

Kadmon claims that (94c) is presupposed by (94b), since “B’s utterance doesn’t merely imply (94c), but it rather implies that B takes it for granted that (94c)” (p. 211). An indication of this is the fact that (94b) is odd or infelicitous without (94c), i.e. (94c) is a condition on the felicity of (94b). However, there is no element in (94b) alone which can trigger (94c), and indeed if (94b) is said in isolation without (94a), the presupposition in (94c) is defeated. In this case, then, the presupposition cannot be conventional but has to be conversationally triggered. Kadmon claims that it is the Grician maxim of relevance which triggers (94c), that is “it is based on the hearer’s attempt to imagine what it is that makes B’s utterance relevant” (p. 211) We can now turn back to compare the triggering of the presupposition in (89) with that of its simpler counterpart in (93). Although the simpler presupposition in (93) makes the right predictions, it does not seem to be triggered by any independent conventional or conversational mechanism. In other words, it is not clear why the ^Q properties denoted by VP of IS sentences must be reasonable “sort of properties” predicated of ^P (the property denoted by the subject). After all, we do not find any such requirement with corresponding BP sentences, or sentences with proper-name subjects, which can felicitously appear with any funny or bizarre property whatsoever. If we want to keep this simpler presupposition, then, we will have to stipulate its existence. In contrast, we can show that there is a natural way to trigger the original, more complicated, presupposition on the choice of the ^S property (namely that it can reasonably lead to properties of the sort of ^Q). This can be done by deriving the presupposition in (89) from two independently motivated and existing mechanisms, namely the requirement on the association of ^S with ^P, together with an existing conversational principle of cooperation. Consider, for example, the false IS sentence in (95), and its truth conditions in (96): (95) A Pentium III computer works very slowly. (96) A Pentium III computer works very slowly is true in w iff


Where ^S CR(^PENTIUM COMPUTER III COMPUTER, w). iff there is some Kratzerian function f (from worlds to sets of sets of worlds) s.t. A (rough) Paraphrase: A Pentium III computer works very slowly is true in w iff there is a property ^S, s.t. in all worlds where every Pentium III computer has S, and ^S is associated with being a Pentium III computer in w, every Pentium III computer works very slowly (in all relevant situations). Where ^S is associated with being a Pentium III computer if it holds of every such a computer in a set of worlds Kratzerially accessible from w, i.e. if “Every Pentium III computer has S” constitutes some epistemic, deontic, stereotypical, etc. necessity in our world, based on the shared real-world knowledge. Two contextual requirements combine in evaluating (95). The first is a general principle of cooperation: We assume, based on Grice’s (1975) maxim of cooperation, that the speaker of (95) intends to make a true statement (maxim of quality). This, of course, does not mean that we, as listeners, agree that the sentence is true, only that we rely on the speaker as adhering to Grice’s maxim of quality (“say only what you believe is true”). Given the truth conditions in (96), this means that we believe that the speaker of (95) has in mind an ^S property which leads to the truth of the universal statement s,d [Pentium III computer (d, w′) C (s, d, w′)] → work very slowly (s, d, w′)], i.e. which s/he takes to cause low operating speed of these computers. The second contextual assumption is specific to IS sentences: As claimed in detail in section (5) above, the ^S property, which serves as the “in virtue of” property, cannot be any arbitrary property, but should be associated with ^P in w, and moreover, associated with ^P on the basis of the real-world knowledge shared by the community of language users, including both the speaker, and us, as listeners. But on the surface there is a conflict between the two contextual requirements, when false IS sentences like (95) are evaluated. Assuming that the speaker intends to make a true statement in (95), the ^S property he has in mind is probably something like “^has a very slow processor” or “^has very limited memory” But given the shared real-world


knowledge, neither these properties, nor any similar property, is associated by us (as listeners) with Pentium III computers. In other words, assuming that the speaker intends to make a true statement, he seems to choose an “in virtue of” property which, given the shared realworld knowledge is not associated with ^P. Alternatively, assuming the ^S property that the speaker has in mind is associated by us as well with being a Pentium III computer, we can no longer assume that this speaker intends to make a true statement. This is because, assuming that s/he shares the same realworld knowledge with us, there is no ^S property associated with being a Pentium III computer on the basis of shared real-world knowledge which leads to the truth of the generalization. The resolution of this conflict is captured by the presuppositional requirement on the choice of ^S, in (89) above. The general idea is that when hearing a false IS sentence like (95), we still believe that the speaker and ourselves have the same real-world knowledge assumptions, but that the speaker drew wrong conclusions on the basis of these assumptions. Our disagreement with the speaker, then, is not on the assumptions (which are supposed to be shared by everyone), but on the conclusions. Intuitively, then, we have to characterize the ^S property in such a way that it is associated with ^P (in our case—with being a Pentium III computer) by both the speaker and listeners of the sentence, but which is, at the same time, flexible enough to allow a disputable truth value. The presuppositional requirement on the choice of ^S in (89) above leads to the desired result. Given this requirement, ^S in the truth conditions in (99) can be characterized as “having the regular processor of Pentium III computers,” or even more generally as “having the regular mechanical property of Pentium III computers responsible for working speed,” i.e. as a property which can reasonably lead to having a certain working speed of Pentium III computers. The assumption that such a property is associated with Pentium III computers is indeed shared by both us (as listeners) and the presumed truthful speaker of (95), but the conclusions we draw are different. For the speaker, this associated property leads to a low working speed of Pentium III computers (since s/he probably associates such computers with a very limited processor, or memory); we, as listeners, take this property to lead to a high working speed (since we associate such computers with a very strong processor, or enlarged memory). We (as listeners), then, share the assumptions of the speaker, but disagree with his/her conclusions, and take the sentence in (98) to be false but nonetheless felicitous.


The claim that the source of the presuppositional requirement on ^S is the need to assume that the speaker of the sentence is speaking truthfully can help us to understand why the disjunction in the presupposition is put under a “good possibility” operator. The idea is that, even if we, as listeners, don’t know of any associated ^S, which actually leads to a property of the sort of ^Q in the actual world (e.g. which connects between being a carpenter in Amherst and paying a specific percentage of income tax, in the case of the felicitous A carpenter in Amherst pays 17.6% income tax), we can still credit the speaker as speaking truthfully if we can imagine a not-too-far-fetched scenario which may connect the two properties, and assume that the speaker had such a scenario in mind when uttering the sentence. We predict, then, that when there is no way to resolve the apparent conflict between the requirement on association of ^S with ^Q and the assumption that the speaker is truthfully speaking, the IS sentence will be odd. This is exactly what happens in the case of the infelicitous #A man is blond. With this sentence there is no reasonable way to assume that the speaker of this sentence shares the same real-world knowledge with us (i.e. that he associates the same properties with “being a man” as we do), and at the same time that this speaker intends to make a true statement. While we managed to characterize the ^S property in A Pentium III computer works slowly in a way which makes it both clearly associated with ^P, and at the same time leads to a disputable truth value, we cannot do the same with A man is blond. Not only is there no property we associate with being a man which can reasonably lead all men to being blond, nor do we associate with being a man any other, more general property (like “having the regular genetic makeup for a specific hair color”), of which we can say “Well, there is a-not-too-farfetched scenario that the speaker has in mind, where he takes this ^S property to lead every man to be blond.” Thus, when evaluating this sentence we are either forced to assume that the speaker does not share the same real-world knowledge assumptions that we, as listeners, have (i.e. that s/he associates properties with men which we do not), or that, based on the shared real-world knowledge it is far-fetched, or unreasonable that s/he intends to make a true statement. Since in either of these alternatives an independently motivated principle is violated, this sentence is odd, and is taken to express a highly unreasonable generalization. The presuppositional requirement on the choice of ^S is the way to ensure, then, that the “association with ^P” requirement on the accessibility relation, and the Grician maxim of quality are both met.


To conclude, preferring the more complicated presupposition on the choice of the ^S property allows us to maintain the claim that the only special thing about IS sentences (as opposed to BP ones), which is responsible for the intuition that they are only felicitous when expressing “reasonable” generalizations, is the presence of an “in virtue of” property associated with ^P in the accessibility relation. The “reasonable causation” presupposition on ^S is naturally derived, in fact, from that “association with ^P” requirement and the Grician maxim of quality. In contrast, in preferring the simpler presupposition (in (95) above), we are forced to stipulate that there are, in fact, two distinct sources of the “reasonableness” requirement on IS sentences: both the “association” requirement on ^S as well as the requirement on properties of the sort of ^Q. 6.4.3 A more efficient evaluation process Another reason to prefer the original, more complicated presupposition in (89) over the simpler one in (93) is that the evaluation process of false-but-felicitous IS sentences, implied by using truth conditions with the simpler presupposition, is intuitively very inefficient. Consider the truth conditions of the false (97) in (98), with the simple version of the presupposition: (97) A donkey has short ears. (98) A donkey has short ears is true in w iff:

And S CR(^DONKEY, w). iff there is some Kratzerian function f (from worlds to sets of sets of worlds) s.t. w″ [w″ Rf w] → [ d [donkey (d, w″)] → [S(d, w″)]] Presupposition (simple version): There is a good possibility that:

(i.e. having a certain length of ears is a reasonable property to be predicated of all donkeys.) Given the truth conditions in (98), when evaluating the sentence we have to find an ^S property associated with being a donkey which will


make the universal statement [ s, d [donkey (d, w′) C(s, d, w′)] → has short ears (s, d, w′)] true. Crucially, there is no further restriction on ^S. This allows for the possibility that a listener of (97) will go over a very long list of properties associated with donkeys, most of them having no connection to the length of ears, and check whether any of them makes the universal statement true, i.e. according to (98), a speaker of (97) is likely to have the following “dialogue” in his mind: (99)

Evaluation scenario, with the simpler presupposition:

We have to find a property associated with donkeys, in virtue of which donkeys have short ears. Maybe it is in virtue of having four legs (associated with donkeys). No, this is not true. Having four legs does not reasonably lead to having short ears. Maybe is it because of being very stubborn (also associated with donkeys). No, this does not reasonably lead to having short ears either. Maybe (102) is true in virtue of being a mammal. No, there are many mammals with long ears (e.g. rabbits). Perhaps it is the property of eating grass. No, this property does not lead to having four legs either. Neither does having a tail. Neither does not being able to fly. Neither does kicking one’s master when angry (etc., etc., etc.) Well, it seems that no property associated with donkeys is a suitable “in virtue of’ property here. We have to conclude that the sentence is probably false. In contrast, if we use the original, more complicated, presuppositional requirement on the choice of ^S, in (89) above, as in (100), the evaluation process is more likely to look as in (104): (100) Presupposition (original): ^S (a property associated with ^P in w) is s.t. there is a good possibility that:

(101) Evaluation scenario, with the original, more complicated presupposition: We have to find an ^S property associated with donkeys, which reasonably leads to having short ears, or which leads to not having short ears (i.e. which leads to the length of the ears), and check whether it leads to having short ears. But the only relevant property associated with donkeys, i.e. having a genetic makeup for the length of ears, leads to having long, and not short ears. Thus the sentence is false.


Under the original presupposition, when evaluating an IS sentence we restrict ourselves from the start to those ^S properties which are relevant (as reasonable causers) for the sort of property denoted by ^Q, i.e. for the length of ears, and do not consider any irrelevant properties, like the number of ears, color, character, etc.). The evaluation process of (104) is much more efficient, and intuitively more reasonable. I presented, then, two arguments why we should prefer the original presupposition in (89) to the simpler one in (93). In addition, in chapter 4, where I introduce the mechanism for tolerating exceptions of IS and BP sentences, I show that by preferring the original, more complicated presuppositional requirement on the choice of the “in virtue of,” ^S, property, rather than the simpler requirement on ^Q, we can offer an explanation for a newly observed special characteristic of IS sentences w.r.t. exceptions, and for the way they differ in this respect from their descriptive BP sentences counterparts. 7. Conclusion and summary With the original presupposition, then, and while still ignoring the problem of exceptions to the main generalization, the truth conditions of IS sentences are in (102): (102) Truth conditions of IS sentences (ignoring in the meantime the problem of exceptions) Let IS NP denote a property ^P, VP denote a (verbal) property ^Q, and ^S is a property An IS sentence of the form IS VP is true in w iff:

where S CR(^P,w). iff there is s ome Kratzerian function f (from worlds to sets of sets of worlds) s.t.

Presupposition: (conventionally triggered by universal quantification over worlds): CR(^p,w) is not empty. Presupposition (triggered by a combination of Grice’s maxim of quality, and the condition on ^S) :^S is s.t. There is a good possibility, i.e. there is a world w′, (Kratzerially accessible from w, and closest to an ideal version of our world) s.t.


(where in the ideal world some general principle which works in our world is not inhibited or blocked by any other, less general, factor). The main claim I made in this chapter, which is now formally captured by these truth conditions, is that the kind of nonaccidental generalizations expressed by IS sentences have an “in virtue of ^S” nature. I claimed that the role of the ^S property, which serves as the “in virtue of” property is to systematically restrict the set of accessible worlds in which the generalization is evaluated. Moreover, I claimed that this ^S property cannot be any arbitrary, or farfetched property, but that it has to meet some requirements it meets in the actual world. Together with the amended presupposition introduced in the previous section, there are in fact two contextual restrictions on ^S. The first is its association with the ^P property (denoted by the subject), and the second (which is derived from the first one and Grice’s maxim of quality) is its “reasonable causation” relation with properties of the sort of ^Q (denoted by the VP). In this sense we have captured, in fact, Chierchia’s (1995) intuition, quoted in section (3.1) above, and repeated here, concerning the double restriction on the set of conditions (“felicity conditions”) which define the accessible worlds of generics: Actually, it seems intuitively that there are two kinds of contextual restrictions [on the set of felicity conditions]. One comes, as it were, from the verb; it determines what conditions have to be met for an individual to undertake an action. The other comes from the noun; it determines what features of the species are relevant for the action. (p. 196) In addition to capturing the “in virtue of” nature of IS sentences, the truth conditions in (102) were developed so they can account for a variety of semantic, distributional and pragmatic properties of these sentences, some of which received unsatisfactory, or even no explanation before, and some which were not observed in the past.


Among other things, these truth conditions account for the fact that IS sentences express “rule-like,” “analytic” generalizations, and at the same time, these “rules” are of different modal “flavors.” The truth conditions can also account for the fact that IS sentences can freely combine with highly contingent properties, and felicitously express clearly false generalizations, but on the other hand, those IS sentences with subjects denoting “extremely unnatural properties” or those with “highly unconnected properties” are infelicitous as generic when uttered “out of the blue.” I have shown that the key for distinguishing felicitous from infelicitous IS sentences lies in the notion of “reasonable generalizations” and supplied an exact and compositional definition of “reasonable generalizations” as far as IS sentences are concerned, in terms of the double contextual restriction of ^S, the “in virtue of” property. I have also shown how the effect of various contexts (epistemic, deontic, linguistic, inductive, etc.) on the felicity of IS sentences is correctly predicted by the contextual restrictions on ^S. Assuming that ^S must be associated with the subject property ^P (in Brennan’s terms, that the accessibility relation is “keyed to the properties of the subjects) can also account for the fact that with IS subjects symmetric predicates behave as asymmetric ones. I think that besides the explanation of these specific facts, the most important thing captured by these truth conditions is the very delicate interplay between law-likeness and actuality in the meaning of IS sentences. As stated above, on the one hand IS sentences seem to express a rather strong level of lawlikeness, and nonaccidentalness, i.e. they are very “nonactual” generalizations, and on the other they are very sensitive to our real-world knowledge, i.e. to information specific to the actual world (this causes their infelicity when they express generalizations which are unreasonable in terms of our information about the real world. The strong law-likeness of IS sentences is captured by their “doubly modal” truth conditions (asserting the truth of a nonaccidental generalization, which is based on another nonaccidental generalization), and by defining their accessibility relation as “circumstantial” in nature (i.e. as an accessibility relation which preserves a narrow aspect of the actual world). On the other hand, the fact that the felicity of such sentences so strongly depends on real world knowledge is captured by the real world knowledge limitations imposed on the choice of the accessibility relation and on the choice of “in virtue of” property. We have gained, then, some understanding as to the nature of “in virtue of” generalizations, but it is not until we contrast these


generalizations with their counterparts, i.e. descriptive generalizations, and the law-likeness/actuality balance in them, that we can fully understand and appreciate this balance with “in virtue of” generalizations expressed by IS sentences. Thus, I turn now to chapter 3, which analyzes BP sentences and inductive generalizations, and which compares the two kinds of generalizations in a more detailed and deeper manner and from a wider perspective. In chapter 4 I show how the two kinds of generalizations correlate with two behavior patterns w.r.t. exceptions. In that chapter I also introduce the mechanism for tolerating exceptions for the two kinds of generalizations, which completes the truth conditions of IS and BP sentences.

124

CHAPTER 3 “Descriptive” Generalizations, and the Semantics and Pragmatics of Descriptive BP Sentences

1. Introduction The main claim made in this work is that the differences between BP and IS sentences should be explained in terms of two different kinds of nonaccidental generalizations encoded by these two kinds of sentences. In the previous chapter I discussed “in virtue of” generalizations, expressed by IS sentences, and defined their truth conditions, and especially their accessibility relation, determined by an appropriately chosen “in virtue of” property. The purpose of this chapter is to clarify the nature of the second kind of nonaccidental generalization: descriptive generalizations. As claimed above, such generalizations are naturally expressed by BP sentences,1 and cannot be expressed by IS sentences. Thus the main concern of this chapter is an examination of BP sentences. Since many BP sentences also have a natural “in virtue of” reading, I will primarily focus on those BP sentences whose prominent reading is “descriptive” or which lack a natural “in virtue of” reading (and have no felicitous, minimally contrasting IS counterparts). I will call BP sentences under their descriptive reading descriptive BP sentences.

1

Another kind of sentence which naturally expresses descriptive generalizations is simple habituals (e.g. John walks to school). I discuss such sentences in chapter 6 below.


1.1 Basic properties of BP, as opposed to IS sentences The truth conditions I developed in the previous chapter for “in virtue of” generalizations account for several special distributional and semantic properties of IS sentences. In contrast to these sentences, descriptive BP sentences have the following distributional and semantic properties: Property A: Unlike IS sentences, descriptive BP sentences have a prominent (if not an only) generic reading when their BP subjects denote “extremely unnatural classes,” e.g. (1). a. Norwegian students whose names end with ‘s’ or ‘g’ wear thick green socks. (cf. A Norwegian student whose name ends with ‘s’ or ‘g’ wears thick green socks—with a prominent existential reading) b. Tall, brown-haired neurologists at Hadassah Hospital earn $ 100,000 a year. (cf. A tall, brown-haired neurologist at Hadassah Hospital earns $100,000 a year—with a prominent existential reading) c. Short, middle-aged men called Jim stand on one foot every morning. (cf. A short, middle-aged man called Jim stands on one foot every morning—with a prominent existential reading) d. Famous books written by middle-aged American zoologists contain an odd number of letters (cf. A famous book written by a middle-aged American zoologist contains an odd number of letters—with a prominent existential reading) Property B: Unlike their IS counterparts, BP sentences with VPs denoting “extremely unconnected properties,” as in (2), are felicitous as generic: (2) a. Uncles eat marshmallows. (cf. # An uncle eats marshmallows) b. Men are blond. (cf. # A man is blond.) c. Rooms are square. (#A room is square.) These two properties indicate that unlike IS sentences, BP sentences are not restricted to express only “reasonable” generalizations, but can also freely express highly improbable ones.2


Property C: Unlike IS sentences, which express various kinds of rules, descriptive BP sentences have a “uniform” flavor. For example, all the BP sentences in (3) describe the same kind of “descriptive” generalization. In contrast, the IS sentences in (4) have various flavors: epistemic, normative, legal, definitional, etc.: (3) a. Third daughters are 43 years old. b. Rooms are square. c. Tall postmen called Titus drive wildly. d. Red-covered books about brown cats contain five or eleven typos. e. Men are blond. (4) a. A dog has four legs. (epistemic) b. A boy does not cry. (deontic) c. A married woman covers her hair. (religiously deontic) d. A single parent with more than two children pays 20% income tax. (legal) e. A bachelor is an unmarried man. (linguistic) The difference can be also seen in Burton-Roberts’ (1977) original minimal pair presented in (5): (5) a. A gentleman opens the door for a lady. (a normative/deontic flavor) b. Gentlemen open the door for a lady. (a more “descriptive” flavor, although the deontic one is possible as well)

2

Note, however, that as shown by Carlson (1977), Chierchia (1995) and McNally (1995), BP sentences where the denotation of the subject is restricted to specific locations of entities are infelicitous as generic (or at least have a prominent existential reading), as seen in (i)-(iii):

(i) (ii) (iii)

People on that bench speak very loudly. Coins in my pocket are wet. Parts of this machine are broken.

I briefly discuss such BP sentences in chapter 6 below. In sections (5) and (6) of this chapter I discuss other cases of BP sentences which are infelicitous under the generic reading.


Property D: Even if we compare BP sentences (as in (6)) to minimally contrasting IS sentences with an epistemic flavor (as in (7)), we feel that there is a difference between the two: (6) Tables have four legs. (paraphrase: there is some pattern about tables in our world—this is the case in reality) (7) A table has four legs. (paraphrase: “in principle,” this is (part of) the definition of a table) The BP sentence in (6) is felt to be more “descriptive” than its IS counterpart, in that the generalization it expresses is felt to describe an actual pattern. In contrast, the IS sentence in (7) is felt to express a more “in principle” rule, i.e. a generalization which is more “definitional/ analytic,” and which is felt to express a higher level of law-likeness. 1.2. Descriptive BP sentences: the intuition and the direction of formalization 1.2.1 The intuition The intuition I will develop in this work is that, unlike IS sentences which assert that a certain generalization is nonaccidentally true in virtue of some property (appropriately chosen on the basis of our realworld knowledge of the ^P and ^Q properties), descriptive BP sentences like (1) and (2) above simply assert that the generalization constitutes a pattern, and is not accidental. For example, the BP sentence Red tables are heavy asserts, roughly, that the generalization “All/most red tables are heavy” is not accidental—i.e. that it not restricted to the actual set of circumstances, but holds in other possible sets of circumstances as well. As have already been mentioned in chapter 1 above, this descriptive reading is especially appropriate as a conclusion of an inductive inference. If, for example, I find that a surprisingly large number of brown-haired librarians are very quiet, I may utter Brown haired librarians are very quiet. By this statement I am trying to convey that not only every (relevant) brown-haired librarian is very quiet, but that I believe that this is not accidental, i.e. that this generalization is not


restricted to the actual circumstances and that we expect it to be true in all those circumstances which are not too different from the actual one. Crucially, and unlike what happens with “in virtue of’ generalizations, I do not imply, and this statement does not concern, not even in an implicit way, what aspect of our world, or of brown-haired librarians stands behind their quietness. My listeners or I may, of course, have all kinds of guesses as to the reason behind this pattern, but unlike what happens with IS sentences, such considerations are not semantically or pragmatically encoded in the linguistic construction. Unlike what happens with IS sentences, then, the speaker of a BP sentence is not committed to have in mind such an “in virtue of” aspect, and the listener is not required to accommodate such an aspect as part of fully understanding the meaning of the sentences. 1.2.2 Direction of formalization Making this intuition precise, I start by adopting the widely held claim (made by e.g. Wilkinson (1991), Krifka (1995) and Chierchia (1995), and reviewed in chapter 1 above) according to which BP noun phrases in characteristic (I-) generic sentences behave like their IS noun phrase counterparts, i.e. as indefinite expressions, which introduce variables into the semantic representation.3 I will assume that, as with IS sentences, the variable introduced by the BP subject, as well as the situation variable introduced by the predicate, are bound by a generic operator, which is universal, and modal. A sentence like (8a), for example, will mean, very roughly, something like (8b): (8) a. Brown cats are very shy. b. Every contextually relevant situation involving any brown cat, in all appropriately accessible worlds, is a situation where this brown cat is very.

3

This view is also compatible with the claim (made by e.g. Chierchia (1998)) that BPs are originally interpreted as kind-referring expressions, but are typeshifted into indefinite expressions in the presence of a Gen operator. In chapter 6 below I examine other possible representations of characteristic (I-) BP sentences.


1.2.3 Exceptions to descriptive BP sentences An important component missing from this paraphrase is, of course, the mechanism for tolerating exceptions. In fact, for (8a) to be true we don’t really need every brown cat to be shy, only a majority of these cats. One may take this intuition to mean that the quantifier associated with descriptive BP sentences really has the quantificational force of “most,” i.e. more than half. The problem with such an approach, however, is that it cannot account for another strong intuition about descriptive BP sentences like (8a), namely that for this sentence to be true, those brown cats who are not shy should not be merely be a minority of the brown cats set, but should be also taken to be, in some sense, less normal than the brown cats who are shy. Put differently, they should be considered “exceptional” in some other respect as well as not being shy. To capture this intuition I will follow the more traditional view of generics (e.g. Krifka (1995), Chierchia (1995, 1998), and Wilkinson (1991)), and take descriptive BP sentences to involve, indeed, some kind of universal quantifier over the non-exceptional, or “normal” P individuals (e.g. over non-exceptional brown cats). I will postpone the discussion of the mechanism for excluding the exceptional members until chapter 4, where I compare the behavior of IS and descriptive BP sentences w.r.t. exceptions. In the meantime, I will paraphrase BP sentences like (8a) as “Every relevant situation involving every nonexceptional cat, in all accessible worlds, is a situation where that cat is shy.” 1.2.4 Basic modality/intensionality of descriptive BP sentences A second crucial claim which I make about the semantics of descriptive BP sentences is that despite their lower degree of law-likeness, and despite the fact that they can freely express highly unreasonable generalizations, these sentences are modalized, and should not be given an extensional, non-modal interpretations (as do e.g. Krifka (1987) or Dobrovie-Sorin & Laca (1996)). The main reason for this claim is the fact that these sentences support counterfactuals. For example, if (9a-c) are true, then the corresponding counterfactuals in (10a-c) are true as well:


(9)

a. Men are blond. b. Tall, brown-haired neurologists at Hadassah Hospital earn $ 100,000 a year. c. Water bottles are empty.

(10) a. If you were a man, you would be blond too. b. If John were a tall, brown-haired neurologist at Hadassah Hospital, he would earn $100,000 a year as well. c. If this were a water bottle, it would be empty. The descriptive BP sentences are perfectly felicitous (although probably false), and the corresponding counterfactuals are also felicitous. As shown by, e.g. Lewis (1973, 1986b), the evaluation of counterfactuals requires examining possible worlds other than the actual ones. Thus, the entailment of the counterfactuals in (10) indicates that descriptive BP are indeed modalized,4 i.e. they involve a reference to possible sets of circumstances besides the actual one. 1.2.5 The nature of modality/intensionality of descriptive BP sentences Both IS and descriptive BP sentences, then, are modalized. Their modality, however, is different. In this chapter I claim that the various distributional and semantic properties of BP sentences, which distinguish them from their IS counterparts, are all derived from the different accessibility relation associated with them, or more specifically, from differences in the degree to which the similarity of the accessible worlds to the actual one is specified. In the previous chapter I claimed that the accessibility relation of IS sentences is “circumstantial” in nature, and is restricted by a property, ^S (chosen on the basis of our real world knowledge of the ^P and ^Q property). An IS sentence is evaluated in all worlds where every P

4

Cohen (1999) rejects a modal analysis of Gen. He claims that the intensionality of descriptive generics is temporal, rather then modal, i.e. it should be defined in terms of possible branching future histories rather than in terms of possible worlds. In sections 2.2.2 and 2.4 I discuss data (concerning counterfactual entailments of present tense descriptive BP sentences into the present, and past) which constitute counterexamples to such a claim.


individual has S, where this condition itself is presupposed to be an epistemic, deontic, legal, etc. rule about the actual world. The required similarity of the accessible worlds to the actual one, then, is a very narrow one, concentrating on only one (nonaccidental) aspect of the ^P property in the actual world. This is what yields the “in virtue of” flavor of IS sentences. In contrast, descriptive BP sentences do not assert that the generalization is true in virtue of some specific aspect, but simply assert that the generalization is not accidental, i.e. holds in other possible circumstances or worlds similar to ours, besides the actual one. Crucially the similarity of these accessible worlds to the actual one is much less specified. If, for example, you assume the truth of the sentence New York taxis are yellow (under its descriptive reading, e.g. after observing many such yellow taxies), then you can also assume that if this car were a New York taxi, then, all things being equal, it would be yellow too, where crucially, there is no systematic specification of which are the things which are exactly equal. This kind of similarity of the accessible worlds to ours, then, is much more “wide perspective” in nature and is far vaguer than with IS sentences, and does not depend on choosing a specific aspect of the ^P property in the actual world as the criterion for similarity between the worlds, and as responsible for the truth of the generalization. I will show below that Lewis’s (1973, 1986b) characterization of the vague kind of similarity used in the evaluation of counterfactual statement is very useful in making this intuition precise. 1.2.6 Basic semantic structure, and structure of this chapter In addition, I will show that besides reference to accessible possible worlds, different from the actual ones, but appropriately related to them, the semantics of (present tense) descriptive BP sentences also involves reference to time intervals, different from the present interval, but appropriately related to it. The truth of the universal statement associated with descriptive BP sentences, then, will be evaluated w.r.t. to indices, namely world-intervals pairs, and the general semantic structure of descriptive BP sentences, will be shown to be basically the one in (11): (11) A BP sentence of the form BP VP is true at <w, I> (where BP denotes ^P, and VP denotes ^Q) iff:


w' [w' is appropriately accessible from w] → I' [I' is d,s [(d <w',I'>) C(s,d w') loc appropriately related to I (s,I'>)] → [Q(s,d, <w'I'>)]] Paraphrase: A BP sentence is true in a world w and an interval I, iff Every contextually relevant situation involving every (nonexceptional) P member at every world w′, appropriately accessible from w, at some interval I′, appropriately related to I, is a Q situation involving this P member at that world w’ and interval I′. A great deal of this chapter is devoted to formally clarifying the “wide perspective” and vague nature of the accessibility relation (which produces the accessible worlds w’ in (11)), and the definition of relevant time interval (I′ in (12)), and to motivating their role in the semantics of descriptive BP sentences. The structure of this chapter is the following: In section (2), I examine the predictions that true and felicitous descriptive BP sentences (in the simple present) make w.r.t. the truth of the universal statement in various types of indices: worlds-time pairs. This examination will be used to build the basic definition of the truth conditions of descriptive BP sentences. The following sections all deal with the advantages and implications of these basic truth conditions. In section (3), I show that by assuming the “wide-perspective” accessibility relation in the truth condition, we can account for the four basic differences between descriptive BP sentences and their IS counterparts (mentioned in section (1.1) above). In section (4) I discuss advantages of the definition of the time interval I′ (in (11), as well as an apparent problem with this definition. I then turn to show that, when combined with independently motivated conversational principles, the truth conditions developed for descriptive BP sentences in the previous sections can also account for two presuppositions of these sentences. In section (5) I deal with “existence” presuppositions of descriptive BP sentences. I also show the need to integrate such existence presuppositions into the definition of IS sentences. In section (6) I examine the connection between the assertions made by descriptive BP sentences, and the processes of inductive inference. I conclude that despite the close connection between the two, the latter kinds of processes do not constitute an integral part of the


Semantics of descriptive BP sentences, but rather a natural basis for the assertions made by them. The connection with inductive inference is encoded by defining an additional, conversationally triggered, presupposition (called the “enough” presupposition), which successfully captures facts concerning the infelicity of descriptive BP sentences in a variety of contexts. Section (7) summarizes this chapter. 2. A closer look at descriptive BP sentences: entailments and predictions The intuition we pursue is that a simple present BP sentence asserts that a certain generalization is not accidental, i.e. it holds not only in actuality, but is also expected to hold in some other sets of circumstances, similar to ours. But what are exactly those similar sets of circumstances and what exactly do we mean when we talk about the actual one? I will attempt to clarify these notions by taking a closer look at the predictions and entailments of descriptive BP sentences. 2.1 Truth of the universal statement in the “similar worlds” 2.1.1 The data and the intuition As stated above, the principal indication that descriptive BP sentences express nonaccidental universal generalizations comes from their counterfactual supporting property, seen again in the entailment from (12) to (13a-c): (12) Brown cats are shy. (13) a. If this (pointing to a white cat) were a (non-exceptional) brown cat, it would shy. b. If this (pointing to a black dog) were a (non-exceptional) brown cat, it would be shy. c. If this (pointing to a yellow giraffe) were a (non-exceptional) brown cat, it would be shy.


In actuality this cat is white, this dog is black and this giraffe is yellow. A set of circumstances where this white cat, this dog and this giraffe are brown cats is no longer an actual set of circumstances. It is a possible, hypothesized set of circumstances, what is usually called a possible world. We see that no matter which individual we point to, the truth of (12) entails that if this individual were a (non-exceptional) brown cat, it would be shy. This means that the BP sentence asserts that anything which is a (non-exceptional) brown cat is shy, i.e. the BP asserts the truth of a universal statement in a set of possible, hypothesized worlds. If this universal statement (about every non-exceptional brown cat) were asserted to be true in this world alone, then there would be no way to guarantee that the non-actual brown cats (i.e. the white cat, black dog and yellow giraffe) were shy as well. It seems clear, however, that not every possible hypothesized world is relevant here. We would not insist that the generalization is true in worlds in which, e.g. all cats have undergone significant mutations, making them very different from what they are in our world, or in worlds in which all brown cats are in fact dyed brown (the only natural color of cats being white). There is an infinite number of possible worlds, and some of them (even most of them) are so bizarre, so funny or so different from our world that we do not expect the generalization to hold in them. Thus, we want to consider only those worlds which are similar to our world in certain respects. On first sight, however, it is not clear how to define this kind of similarity. In what ways exactly should the relevant accessible worlds be similar to ours? Which aspects of our world should be preserved in them, and which should not? Remember that with the “in virtue of” generalizations we could define the similarity quite well, based on our real-world knowledge of the properties involved and the connection between them (i.e. the properties associated with ^P in w, and their possible causal connections with properties of the sort of ^Q). In other words, we could characterize rather precisely which aspects of our worlds should be preserved in the accessible worlds. Moreover, the possibility of getting at such a characterization was a necessary condition on the felicity of IS sentences. Here, on the other hand, things are much less clear. For example, the BP sentence in (14a) and (15a) supports the counterfactual in (14b) and (15b): (14) a. Brown-haired librarians are nervous.


b. If you were a (non-exceptional) brown-haired librarian, you would be nervous, too. (15) a. Uncles love marshmallows. b. If you were a (non-exceptional) uncle, you would love marshmallows as well. Nonetheless, if I hear or utter (14a) I do not necessarily know, presuppose or imply which aspect of the world makes brown-haired librarians quiet, or which aspects of being an uncle in our world causes the consumption of marshmallows. Consequently, although it is clear that these sentences imply that the generalizations are true in other possible worlds besides ours, which are also similar to ours, it is not clear how to define the similarity, i.e. what aspects of our world should hold in these similar worlds: Genetic factors? Biological factors related to hair color? Sociological factors related to being a librarian or being an uncle? Psychological factors? Social norms? There is one aspect of our actual world, however, which we can clearly rule out as a criterion for similarity, i.e. an aspect which cannot and should not be part of what constitutes the similarity between the actual world and the relevant worlds. The counterfactuals above indicate that the relevant possible worlds in which the generalization is expected to hold must be allowed to contain P individuals who are different from the actual P individuals. Imagine, for example, that (14a) is indeed true and that in our world Jane, Anna, Miriam and Sara are the only brown-haired librarians, and all are indeed nervous. The entailment from (14a) to (14b) shows that the generalization about all/ most brown-haired librarians being nervous also hold in worlds which are different from our actual world. In these worlds we should find brown-haired librarians who are not our actual Jane, Anna, Miriam and Sara, since if these were still the only existing brown-haired librarians, there would be no guarantee that a counterfactual like (14b) would be true. It seems, then, that the relevant accessible worlds are allowed to have a different population of P individuals than the one in the actual world. 5 In other words, the set of P individuals in the accessible worlds does not necessarily contain exactly the same members which exist in w, but rather more, fewer, or completely different individual members.

5

This may seem to imply that the actual world is not among the accessible worlds. However, below I develop the definition of accessibility relation so the actual world is included.


In addition, these worlds also must contain non-actual relevant situations, specifically, those contextually relevant situations involving a P member. This is indicated by the fact that (16a) supports the counterfactual in (16b): (16) a. In this school, children talk quietly during meals. b. Speaker A: Listen! All the children in this school are talking very loudly now! Speaker B: This is because this is not a meal situation. If this was a meal situation, then the children would talk quietly. The requirements on the accessible worlds are, then, that they be similar to our world, and that the P set of individuals and the set of relevant situations involving these individuals are not identical to the set of these individuals and situations in the actual world. Besides that the similarity of the accessible worlds to the actual one seems very vague. I suggest that this vagueness concerning the accessible worlds which is felt with descriptive BP sentences is inherent to these sentences, and is very similar to the vagueness described by Lewis (1973, 1986b) for ordinary counterfactuals, and represented by Lewis though the relation of overall comparative similarity between worlds. In the next section I briefly review Lewis’s use of this tool with counterfactual constructions and show how adopting it for the accessibility relation of BP sentences correctly captures the intuitions about such sentences. 2.1.2 The formalization: a Lewis-style definition of maximally similar worlds The two requirements on the accessible worlds of descriptive BP sentences (being similar in a vague way to our worlds, but containing non-actual P individuals and relevant situations) closely resemble Lewis’s requirements on the worlds relevant for the evaluation of ordinary counterfactuals. For example, according to Lewis, in evaluating a sentence like If kangaroos had no tail they would topple over, we should look at all those worlds in which kangaroos have no tail, but which are otherwise not too far away from our world. How similar should these worlds be? Should we look at worlds in which the truth of the antecedent (“kangaroos have no tail”) is the only thing different from our world, everything else being exactly the same? Lewis’s answer is strongly negative. There are simply no such worlds:


We dream of considering a world where the antecedent holds but everything else is just as it actually is, the truth of the antecedent being the only difference between that world and ours. No hope. Differences never come singly, but in infinite multitudes. Take, if you can, a world different from ours only in that Caesar did not cross the Rubicon. Are his predicament and ambitions there just as they actually are? The regularities of his character? The psychological laws exemplified by his decision? The order of the day in camp? […]. Hold everything else fixed after making one change, and you will not have a possible world at all. (1986b, p. 5) Thus, instead of defining such an impossible total similarity, the consequent of the counterfactual should be true in those worlds which, are similar to our world in an overall and inherently vague way, with the exception of the truth of the antecedent and everything naturally needed to allow for its truth. Let me quote Lewis again: It may be objected that […] comparative similarity is hopelessly imprecise unless some definite respect of comparison has been specified. Imprecise it may be; but that is all to the good. Counterfactuals are imprecise too. Two imprecise concepts may be rigidly fastened to one another […]. Imprecise though comparative similarity may be, we do judge the comparative similarity of complex things like cities or people or philosophies— and we do it often without benefit of any definite respect of comparison stated in advance. (1986b, p. 6) I believe that the accessibility relation of descriptive BP sentences should be defined in such Lewisian terms. For example, the possible worlds relevant for the evaluation of Brown-haired librarians are very quiet are those worlds which are maximally similar to our world, except for possibly containing brown-haired librarians, and relevant situations involving them, different from the actual ones (in addition to or instead of the actual ones), and except from all the consequences that such changes normally bring with them. I call such worlds “Lewisian similar worlds,” or WLewisian for short, and define them as in (18): (18) A world WLewisian is a member of the set of “Lewisian-similar world” iff it is maximally similar to w, except from what is needed to allow for the fact that the P set of individuals and the


set of relevant situations involving them are not identical to the sets of P individuals and relevant situations existing in w, respectively. These Lewisian-similar worlds, then, are the first kinds of worlds relevant for the truth of descriptive BP sentences, i.e. in which the universal statement is predicted to hold. 2.2 Truth of the universal statement in the “normal futures” 2.2.1 The data and the intuition Cohen (1999), who takes descriptive BP sentences to express futureoriented probability statements, claims that such sentences make predictions into the future. This claim seems to make sense. Consider again the BP sentences in (19) and (20): (19) Uncles love marshmallows. (20) Tall, left-handed neurologists at Hadassah Hospital earn $ 100,000 a year. If (19) is true, then we expect that the (non-exceptional) uncles I will meet tomorrow, and five years from now, will love marshmallows too. Similarly, if I take (20) to be true, then I expect that this tall, left-handed neurologist, who will start working at Hadassah Hospital next month, will earn $100,000 a year. My claim is that these expectation result from the fact that descriptive BP sentences entail the truth of the universal statement not only in the maximally similar, Lewisian worlds, but also in the future of our actual present world. This can be seen when considering the following scenario: Scenario: Imagine a school in which all (non-exceptional) teachers leave school at 13.00. The principal, however, dissatisfied with this situation. He wants the teachers to stay later, so they can help pupils with their homework, do extra administrative work, etc. It is now a


weekend, and the principal has just decided that from now on teachers will not leave school before 14:30 p.m. Knowing this principal, and the dependency of the teachers on him, I expect that from this coming Monday onward, every (non-exceptional) teacher will not leave the school before 14:30 p.m. Imagine now that I am talking to a friend who is interested in coming to teach in this school, and asks me about it. It is now Sunday, and I just spoke with the principal, who told me about the change he wants to make. In this scenario I utter (20): (20)

Teachers in this school leave school at 13:00 p.m.

If I utter (20) in this scenario, expecting that starting from tomorrow all (non-exceptional) teachers will start leaving the school at 14:30 p.m., instead of at 13:00, then (20) will be clearly false. For it to be true, I have to believe that the (non-exceptional) teachers will continue to leave school at 13:00 in the future. descriptive BP sentences, then, make predictions into the future, and their truth depends on the truth of the universal statement in the future. Akin to Lewisian-similar worlds, the future is not an actual set of circumstances, but a possible one. In fact, there is an infinite number of possible hypothesized future sets of circumstances, and, as with the maximally similar worlds, not all of them are relevant for the truth of the BP sentence. Sentence (21) is false in the context above, since, if nothing unexpected happens, if there are no big surprises, from now on (non-exceptional) teachers will not leave school at 13:00 but rather at 14:30. We can clearly imagine, however, possible future circumstances, where the teachers suddenly decide that they don’t care about the regulations, where they become aggressive and claim that they will nonetheless leave at 13:00, or where aliens from Mars control their minds and force them to leave school at 13:00. None of these possible futures, however, makes (21) true in the context described above. The reason seems to be that they are all too remote from our actual present, and are not expected to occur given what we know about our world now. The kinds of possible future circumstances, then, which are relevant for the truth of descriptive BP sentences can be intuitively characterized as the “normal futures”: those which are not too far-fetched, and where things continue as expected, given what we know about our world in the present.


2.2.2 The formalization: evaluation in world-interval pairs Intuitively, the two kinds of accessible worlds we have characterized, “Lewisian worlds” and “normal futures,” are not very different from one other. Specifically, these two kinds of worlds share two characteristics: First, both are defined as similar to our world (not too far-fetched), and second, both allow for the existence of non-actual P individuals and relevant situations, namely future ones (e.g. future, nonactual uncles, future, non-actual brown-haired librarians, etc.). It is thus tempting to try to use the definition of “Lewisian worlds” in (18) above to define “normal futures” as well. But this cannot work for two reasons. First, although we expect that there will be non-actual P individuals and relevant situations in the “normal futures,” these futures cannot be taken to be “maximally similar” to w, since by definition they allow for many other changes as well. If Uncles love marshmallows is true, for example, then we would expect “Every relevant situation involving every non-exceptional uncle is a lovingmarshmallows situation” to be true in all future circumstances where nothing unexpected occurs, and which are not too far-fetched relative to our actual present. In such future circumstances we indeed may have nonactual uncles and relevant situations involving them, but other things happen as well: John’s Ph.D. dissertation is finished; my sister gives birth to another baby; the peace process becomes mired, and then revives; my baby cuts two more teeth; my grandfather loses two more teeth, the crosswalk in front of my house is repainted, etc., etc. Second, unlike the definition of the “Lewisian similar worlds” in (18), the discussion above suggests that the definition of the “normal futures” cannot remain on the pure modal, world-based level alone, but requires the introduction of a temporal domain. We must quantify over the possible sets of circumstances which occur after the present, or after the time of utterance, and we must define these sets of circumstances on the basis of what holds in the actual present (i.e. now). Cohen (1999) defines the relevant set of futures relevant for the evaluation of descriptive BP sentences using temporal notions alone. He claims that in general the intensionality of descriptive BP sentences is temporal, and not modal, and that they are mainly future oriented, asserting what is likely to happen. Following Thomasson (1970), Cohen assumes that time is branching forward (i.e. in the direction of the future), and claims that the only non-actual, possible circumstances relevant for the evaluation of descriptive BP sentences are (a subset of)


the possible future “histories” (i.e. future time slices or intervals), continuing our actual present (with no reference to possible worlds). I disagree with both of Cohen’s claims. First, it is not true that descriptive BP sentences make predictions only into the future. Second, I believe that modality (i.e. world-based intensionality) is an inherent and productive part of the semantics of descriptive BP sentences. These two views are supported by the counterfactual supporting properties of BP sentences. Above we saw that present tense BP sentences like (14a), repeated here as (22a), support present tense counterfactuals like (14b), repeated here as (22b). This phenomenon is seen even more clearly in the entailment from (23a) to (23b): (22) a. Brown-haired librarians are nervous. b. If you were a (non-exceptional) brown-haired librarian, you would be nervous, too. (23) a. Brown-haired librarians take a two-day vacation in the beginning of May. b. If you were a (non-exceptional) brown-haired librarian, you would take a two-day vacation now as well. (said in the beginning of May) Present tense descriptive BP sentences, then, make predictions not only into the future (saying what is likely to happen in it), but also into possible, non-actual, sets of circumstances which are temporally located in the present (and, as I will show below, to the past as well). The latter set of circumstances cannot be formalized in temporal notions only, as possible time slices, or “histories” since temporally they are all located in the same “history”—the present one (and unlike the future, the present is not branching). Instead, as claimed above, these relevant possible circumstances require a modal, world-based, definition (as “Lewisian”-style worlds, maximally similar to the actual world). (This conclusion will be further supported in section (2.3) below, by the existence of present tense descriptive BP sentences which support even past counterfactuals). In contrast to the “Lewisian circumstances,” the definition of the “normal futures” should clearly employ not only modal, but also temporal notions. It would be desirable to define these two kinds of possible circumstances using a unified framework. My suggestion is to define both of them as pairs of worlds and time intervals. To do that, let us first revise the definition of the “Lewisian-similar circumstances” above.


Instead of “Lewisian, maximally similar worlds,” we will talk about all these worlds in the present interval. Instead of (18) above, then, we will get (24a), where intervals are standardly defined as in (24b), as in Dowty (1979), p. 139: (24) a. “Lewisian-similar circumstance”: all worlds, in the present interval I which are maximally similar to w, except from what is needed to allow for the fact that the P set of individuals and the set of relevant situations involving them are not identical to the sets of P individuals and relevant situations existing in w, respectively. b. Intervals: Let T, which we intuitively regard as the set of moments of time, be the set of real numbers. Let ≤ be the standard dense linear ordering on T. I is an interval iff I and I T and for all moments t 1 , t 2 , t 3 I, if t 1, t 3 I (i ntervals h ave no interval gaps). t 1 ≤ t 2 ≤ t 3 then t2 Returning back to the “normal futures,” we will also defined them as worlds-intervals pairs, specifically, as a set of worlds, temporally located in the interval continuing the present (i.e. the future interval), and which develop normally w.r.t. the actual present (i.e. they do not contain surprises or overly bizarre developments). Following Chierchia’s (1995) suggestion, I will define this set of worlds using the notion of inertia worlds to our world in the present. Inertia worlds, originally defined by Dowty (1979), are those worlds where nothing unexpected happens w.r.t. the present, i.e. where things take their normal course of events w.r.t. the actual world in the present. Dowty defines a function inr which “assigns to every index (consisting of a world and a time interval), the set of worlds which are exactly like the given world up to the time in question, and in which the future course of events after this time develops in ways most compatible with the past course of events” (p. 148), formally: inr () iff w is a member of the set (25) Inertia worlds: w′ of worlds which are exactly like w until I (including I), and in which the course of events in I′, the interval continuing I, develops in ways most compatible with the normal course of events until I (Dowty (1979), p. 149). The “normal futures” intuitively discussed above can be defined now as in (26):


(26) “Normal futures”: all inertia worlds to <w, I> (the actual world in the present interval), at the interval continuing the present interval I 2.2.3 The problem of natural, expected, changes, and its solution We claimed that the truth of the descriptive BP sentence entails the truth of the universal statement in (a) all “Lewisian similar circumstances,” as well as in (b) all inertia worlds to our actual present, at the interval continuing this present, i.e. in all “normal futures” where no unexpected changes happen. There is an apparent problem, though, with assuring entailment in the latter type of circumstances. The problem is that not all changes are caused by “unexpected” factors and constitute a break in the “normal course of events.” In Vlach’s (1981) terms, some such changes are expected, given our actual world in the present, and constitute, in fact, the natural course of events itself. Consider, for example, (27a) and (27b), and the corresponding universal statements in (28a) and (28b): (27) a. Decisions made by the foreign-affairs department are given high priority. b. Pentium VI computers are very expensive. (28) a. Every relevant situation involving every (non-exceptional) decision made by foreign-affairs department is a situation where this decision is given high. b. Every relevant situation involving every (non-exceptional) Pentium VI computer is a situation where this computer is very expensive. According to what we said above, the truth of (27a, b) entails the truth of (28a, b) in two kinds of possible circumstances: all the worlds which are maximally similar to ours, at the present interval (except from having non-actual decisions, and non-actual Pentium VI computers, etc.), and in addition, all “normal futures,” i.e. all worlds, in the future interval, where no unexpected changes occur (relative to our world at the present).


But with respect to the latter set of circumstances, the predictions are not necessarily borne out. Imagine, for example, a scenario in which various people in the office feel that the high priority given to decisions made by the foreignaffairs department is not a good thing. These are very powerful people, and they have already started to consider how to change this undesirable situation. If everything takes its normal course and nothing unexpected happens, eventually they will succeed, sometime in the future, even in the near future, to change the situation, i.e. to make the universal generalization in (28a) false. Similarly, imagine that the next generation of Pentium computers has already been developed, and will be distributed in several days. We know that in the normal course of events in our world when a new generation of computers is on the market, the previous generation becomes much cheaper. Thus, if everything takes its normal course, and nothing unexpected happens, (28b) will be false in the “normal futures.” Given our claims above, the BP sentences in (27a) and (27b) are predicted to be false in these scenarios, since both universal statements are false as a result of changes which are expected, given the normal course of events, i.e. false in the “normal futures.” But this prediction is not borne out in reality. At the time of the utterance, when these natural changes haven’t yet occurred, (27a) and (27b) are clearly true.6 Note, however, that this does not mean that the pattern expressed by the universal generalizations will not be true in the future interval I′ at all. All that this means is that at some point in I′ the pattern stops. In (27a), for example, we expect the universal statement in (28b) to continue to be true after the time of utterance, until the powerful people succeed in their plans. Similarly (27b) predicts that the universal statement in (28b) will continue to be true for some time in the future, i.e. after the time of utterance, until the new computers are on the market. Thus, it seems that we can correctly characterize “normal futures” like this: we continue to require that the universal statement is true in the future, in all inertia worlds to <w, I>. At the same time, however,

6

Cohen (1997) raises a similar problem, and uses it to argue against the use of Dowty’s inertia worlds in the truth conditions of descriptive generics. There are also researchers who claim that Dowty’s definition of inertia worlds makes the wrong predictions in the case of the progressive construction as well, because of roughly the same kind of problem which we now pointed out with regard to descriptive generics. Landman (1992) critically examines Dowty’s inertia-world solution for the imperfective paradox with progressive accomplishments, as in (i):


we do not require that the relevant future interval continue forever, but rather we allow it to be restricted. Thus, instead of taking this interval to be the (infinitely long) interval continuing the present, we take it to be

(i) Mary was crossing the street. According to Dowty (i) is true iff in all inertia worlds to <w, I> the event of Mary’s crossing the street is culminated in the future interval I′, i.e. iff Mary manages to cross in all worlds where nothing unexpected happens, Landman notes that this analysis correctly predicts (ii) to be true: (ii) Mary was crossing the street, when a thunderbolt from heaven struck her dead. (Landman’s example 12) This is because in all worlds where things take their normal course, Mary still manages to cross the street. However, as pointed out by Vlach (1981), this inertiaworlds solution cannot work in the case of (iii): (iii) Mary was crossing the street, when the truck hit her. Landman writes: “Take the interval at which the truck is seconds away from Mary. (iii) is true. Yet the truck is there, and the truck is also there in all inertia worlds. If nothing unexpected happens, the truck will hit her. It would be a miracle for her to escape. Hence, Dowty incorrectly predicts that (iii) is false at this interval” (p. 10). Vlach expresses the same idea by saying that in (iii) the truck hitting Mary “is not an interruption of the natural course of events. It is the natural course of events” (p. 286). The problem of “natural changes” with descriptive generics is simpler than with progressive accomplishments, primarily since the analysis of the former need not relate to culmination in the future. As I show below, this problem can be solved in the case of descriptive generics by taking the length of the future interval to be unspecified. The apparent problem just discussed, where Dowty’s “inertia worlds” are applied to the definition of descriptive generics, resembles the problems with attempts to use a “normal ordering source” as a solution of the exceptions problem of generics. In chapter 2 (section (6)) I discussed these problems and rejected this solution (see also chapter 4). Here, however, there is no need to reject the “inertia worlds”


some (possibly finite) interval continuing the present. Our definition of “normal futures,” then, should look as in (29): (29) Normal futures (revised): all inertia worlds to <w, I> (the actual world in the present interval), in some interval continuing the present interval I. 2.3 Truth of the universal statement in the actual world: present and past We saw that present tense descriptive BP sentences make predictions into two kinds of possible circumstances: the Lewisian circumstances in the present and the normal futures. In addition, the truth of such sentences also depends on the truth of the universal statement in the actual world itself. Consider for example, the BP sentence in (30): (30)

Tall members of the club have blue eyes.

Assume now that this club was founded a week ago. By now there are already 1000 members registered. Even if all non-exceptional tall club members have blue eyes in the normal futures and the Lewisian worlds, if no actual tall club member in the present has blue eyes, or only a minority of members have blue eyes, then the BP sentence in (30) is clearly false as a descriptive generalization (although it can have an “in virtue of” reading, where it expresses a deontic requirement, which is not fulfilled in the actual world). A necessary (although not sufficient) condition for (30) to be true as a descriptive generalization is that the universal statement “Every relevant situation involving every (non-exceptional) member of the club is a situation in which this member has blue eyes” holds in the actual world and the present interval (i.e. in the actual present), in other words, in <w, I>. This conclusion is in accordance with the intuition described in section (1.2.1) above, according to which descriptive BP sentences assert about a certain generalization “this is not accidental,” i.e. that a certain generalization is true both in actuality (the actual present, as we

definition, since the underspecification of the future interval I solves the problem. (See also chapter 2, ff. 16)


just saw), and in other similar worlds (the “Lewisian” worlds and the normal futures). In addition to the actual present, many descriptive BP sentences also make predictions into the actual past, as in (31): (31)

Papers published in these journals are very long.

The truth of (31) as a descriptive generalization implies that not only that all the non-exceptional papers published in the present volumes of these journals are very long, but that the papers published in previous volumes are as well. Notice that the universal statement need not necessarily be true for an infinitely long past interval, i.e. into the distant past, but for some possibly finite past interval, ending with the present. The same holds for the predictions made by (32): (32)

Math students wear red shirts.

If (32) is true, we can expect the universal statement (“Every relevant situation involving every (non-exceptional) math student is a situation where s/he wears a red shirt”) to be true not only right now, but also sometimes in the past. Again, this past period is not necessarily infinitely long. It is not necessarily implied, for example, that the universal statement holds from the time of the first math students in the world. It can clearly be the case that the pattern started only twenty or five years ago, or even last year. This potential finiteness of the past interval is very similar to the potential finiteness of the future interval (relevant for the definition of the inertia worlds), discussed above. Here too we can simply talk about “some interval located before the present.” We can now integrate these predictions of descriptive BP sentences into the present and past of the actual world by requiring that these sentences entail the truth of the universal statement in what I will call “actuality”: (33) Actuality: The actual world, in some (possibly finite) interval ending with (and including) the present interval I


2.4 A unified definition of relevant circumstances and unified truth conditions for descriptive BP sentences To summarize our findings: A present tense descriptive BP sentence entails that the universal statement (“Every relevant situation involving every (non-exceptional) P individual is a Q situation involving this P individual”) holds in the three kinds of world-time pairs, as defined in sections (2.1), (2.2) and (2.3). above. The definitions of these three kinds of world-time pairs are repeated now in (34a, b, c), respectively: (34) a. “Lewisian-similar circumstance” (discussed in section (2.1)):7 All worlds, in the present interval I, which are maximally similar to w, except from what is needed to allow for the fact that the P set of individuals and the set of relevant situations involving them are not identical to the sets of P individuals and relevant situations existing in w, respectively. b. Normal futures (discussed in section (2.2)): All inertia worlds to <w, I> (the actual world in the present interval), in the interval continuing I. c. Actuality (discussed in section (2.3)): The actual world, in some interval ending with (and including the present interval I We will now attempt to arrive at a unified definition of these world-time pairs, and through the definition, at a precise statement of the truth conditions of such sentences. We will do this in two steps. First, we can find a single definition for the “normal futures” in (34b) and for “actuality” in (34c). We do this by exploiting the fact that Dowty’s original definition of inertia worlds requires these worlds to be identical to the input world w, until the present interval. Thus, we do not need to make explicit reference to “actuality” (i.e. to w in the interval ending with the present), since the inertia worlds to <w, I>, in the interval ending with I are identical to it, and differ from w only in the future interval continuing I. Thus, we can require the truth of the universal statement in all the inertia worlds to <w, I> at a larger time

7

And further developed in section (2.2).


interval containing both the past and the future intervals, i.e. at the interval surrounding the present interval I, or, more simply some proper I′). The universal superinterval I' of I, (i.e. the interval I′, s.t. I statement, then, must hold in the indices defined in (35): I′ (i.e. where I is (35) All inertia worlds to <w, I>, at I′, where I the present interval, and a (proper) subinterval of I′). Our second step will be to integrate the “Lewisian-similar circumstances” (in 33a) into the unified definition. The fact that present tense BP sentences support present tense counterfactuals led us above to define these circumstances as the Lewisian-similar worlds, in the present interval I. But, in fact such sentences also support past and future counterfactuals. The simple present (36), for example, supports not only the present counterfactual in (37), but also the past counterfactual in (38), and the future counterfactual in (39): (36) Medium-size stars emit X-rays in the middle of the month. (37) If this (pointing to a large star) were a (non-exceptional) medium-size star, it would emit X-rays now (said in the middle of the month). (38) If this (pointing to a large star) were a (non-exceptional) medium-size star, it would have emitted X-rays two weeks ago (said at the end of the month).8 (39) If this (pointing to a large star) were a (non-exceptional) medium-size star, it emit X-rays in two weeks (said at the end of the month).

Note that in a scenario where the laws of nature or the whole organization of the universe changes at the moment of utterance, or soon after it, the future counterfactual in (39) will no longer be entailed by (36). In addition, we clearly expect (39) to hold two weeks from now, if there have been several “normal” changes in our world, (e.g. where my

8

The entailment of this past counterfactual further supports my claim that the intensionality of descriptive generics cannot be defined in terms of quantification over future branching histories alone (as suggested in Cohen (1999)), but should involve also quantification over possible worlds in the (nonbranching) present or past history (or interval). In other words, the intensionality of such sentences is not only temporal, but also modal.


father has come back from his trip, my son cut another tooth, my grandfather has lost a tooth, etc.). These two facts suggest that the present tense BP sentence make predictions into normal futures. But, crucially, these are not only the “normal futures” of our actual world in the present, but rather the normal futures of the Lewisian-similar worlds in the present, i.e. the normal futures of those worlds in the present which are maximally similar to the actual world except that they contain possibly non-actual medium-size stars (and relevant situations involving them). Based on these data we can now require that the universal statement holds in all worlds which are inertia worlds to Lewisian worlds at the present, at some interval surrounding the present, i.e. at a (proper) superinterval of the present interval I: All inertia worlds to <WLewisian , I> at I′, where I I′ (i.e. where I is a (proper) subinterval of I′) (where WLewisian is a world maximally similar to w except from what is needed to allow for the fact that the P set of individuals and the set of relevant situations are not identical to the sets of P individuals and relevant situations existing in w, respectively). We will now unite the two sets of indices in (35) and (40) by defining a function inrmax from world-interval pairs into sets of worlds. If you feed a world-interval pair <w, I> into this function you get a union of two sets as an output: the set of inertia worlds to <w, I> and the set of inertia worlds to the worlds maximally similar to w at I (i.e. the inertia worlds to <WLewisian, I>. It would be better, of course, if we could use the definition of the inertia worlds to the set of Lewisian worlds at I to also get the inertia worlds to the actual world at I. Unfortunately, this seems impossible. Although, according to Lewis’s “centering assumption,” the actual world is indeed maximally similar to itself, it is not a world in which what is needed to allow for the existence of P members different from the actual ones indeed holds. This is because part of the characterization of the actual world is the properties of the actual individuals in it. Since the antecedent of the maximal similarity relation is false in w, the actual world w is not itself a member of the set the Lewisian worlds: } (41) inrmax (<w, I>): { Paraphrase: Inrmax is a function which takes world interval pairs, and gives a set of worlds as a value. This set is a union of


the worlds which are inertia worlds to <w, I> and those which are inertia worlds to <WLewisian , I> (i-e. inertia worlds to the worlds which are maximally similar to w, except for what is needed to allow for the fact that the P set of individuals and the set of relevant situations are not identical to the sets of P individuals and relevant situations existing in w, respectively). We can now build the truth conditions of simple present descriptive BP sentences. Such sentences will be evaluated relative to a world w and time interval I. They entail the truth of the corresponding universal statement in all those worlds in Inrmax, at some interval, I′, surrounding the present interval I. The truth conditions of such sentences, are, then, in (42): (42) A simple present BP sentence is true in <w, I>9 iff

Paraphrase: A BP sentence is true in a world w and an interval I iff, in all maximally similar inertia worlds w′ (i.e. inertia worlds to <w, I>, and inertia worlds to <WLewisian, I>), there is some interval I′, surrounding the present interval I, and every contextually relevant situation at w′, involving every P individual in w′, and located at that surrounding interval I′, is a Q situation involving these P individuals, in w′. We have reached the truth conditions in (42) by examining the kinds of circumstances (world-time pairs) where universal statements are predicted to hold by corresponding descriptive BP sentences. I will now show that by positing these truth conditions we can account for a variety of additional distributional, semantic and pragmatic properties of these sentences.

9

The present interval I can be taken to be the “extended now” in Abusch’s (1991) terminology.


3. Accounting for the basic properties of descriptive BP sentences I begin with the accessibility relation of descriptive BP sentences, as defined as the inrmax function in (41) above, and the way it differs from the accessibility relation of IS sentences (expressing “in virtue of” generalizations), defined in chapter 2 above, and repeated in (43): (43) Accessible worlds of IS sentences: w′ is accessible from w iff where ^S CR^P,w). iff there is some Kratzerian function f (from worlds to sets of sets of worlds) s.t. Presupposition CR(^ P w) is nonempty Presupposition ^S is s.t. There is a good possibility that

Paraphrase: w′ is accessible from w iff it is a member of the set of worlds in which every P is S, where S is not an arbitrary property, but meets two conditions. Firstly, it is associated with ^P in w (i.e. the universal statement “every P is S” is some sort of Kratzerian necessity (epistemic, deontic, legal, etc.) in w, based on the shared real-world knowledge). Secondly, it is a sort of property which is a reasonable causer in w for the sort of property of ^Q (again, based on the shared real-world knowledge). Comparing the accessibility relations in (41) (of descriptive BP sentences) and (43) (of IS ones) can lead now to a natural account of the four basic differences between the two kinds of sentences, described in chapter 1 and section (1.1) of this chapter, above.


3.1 Descriptive BP sentences as freely expressing “unreasonable generalizations” We saw above that the distribution of felicitous descriptive BP sentences is less restricted than that of IS sentences counterparts. Specifically, unlike IS sentences, descriptive BP sentences are perfectly felicitous as generic when expressing highly unreasonable generalizations. As extensively argued in chapter 2 above, the infelicity of IS sentences expressing “unreasonable generalizations” results from the real-world knowledge requirements on the “in virtue of” ^S, property, which restrict the accessibility relation. If the “association with ^P” requirement cannot be met (since ^P is an extremely “unnatural” property), as happens in (44a), the accessibility relation cannot be appropriately (i.e. nontrivially) defined, and the result is infelicity. Alternatively, even if ^P is a natural enough property, if no associated ^S can be taken to reasonably lead to properties of the sort of ^Q, as in (44b), then the sentence is infelicitous as a result of a presupposition failure (triggered by the “association” requirement and Grice’s (1975) maxim of quality). (44) a. A tall, left-handed neurologist whose name ends with ‘c’ drives wildly. (prominent existential reading) b. # An uncle is addicted to pink marshmallows. As already noted above, in both such cases the minimally contrasting BP sentences are perfectly felicitous: (45) a. Tall, left-handed neurologists whose name ends with ‘c’ drive wildly. (a prominent, if not the only, generic reading) b. Uncles are addicted to pink marshmallows. This felicity is due to the fact that in contrast to IS sentences, descriptive BP sentences are not required to express “reasonable generalizations.” Unlike the accessibility relation in (43), the choice of the accessible worlds in (41) does not depend on any presupposed realworld knowledge ^P and ^Q. Instead, the accessible worlds are those which form the “normal course of events” in our worlds, or the ones


which are maximally similar to it. These similarity criteria can always be appropriately defined, and thus the felicity of descriptive BP sentences, even those expressing highly bizarre generalizations, is guaranteed. 3.2 Unified flavor of descriptive BP sentences An additional difference between descriptive BP and IS sentences, noted above, is that the former (like (46a-e)) have a uniform “descriptive” flavor, as opposed to the variety of epistemic, deontic, legal, and stereotypical flavors of IS sentences, as in (47a-e): (46) a. Third daughters are 43 years old. b. Rooms are square. c. Tall postmen called Titus drive wildly. d. Red-covered books about brown cats contain five or eleven typos. e. Men are blond. (47) a. A dog has four legs. (epistemic) b. A boy does not cry. (deontic) c. A married woman covers her hair. (religious) d. A single parent with more than two children pays only 20% income tax. (legal) e. A bachelor is an unmarried man. (linguistic) In chapter 2, I argued that the dependency of the accessibility relation in (43) on the choice of ^S leads to the fact that the set of accessible worlds varies from sentence to sentence, and possibly even varies for a given sentence in different contexts. This is because the choice of ^S as associated with ^P depends on a variety of possible more traditional, Kratzerian conversational backgrounds. In (47a), for example, the universal statement is evaluated in the set of worlds which is a superset of all epistemically accessible worlds. In (47b) it is evaluated in the set of worlds which is a superset of the deontically accessible worlds, etc. This is what creates the various “flavors,” or the various kinds of rules, expressed by IS sentences. In contrast to IS sentences, the kind of similarity of the accessible worlds relevant for the evaluation of descriptive BP sentences to the actual world is fixed and does not vary from sentence to sentence. We


always get the same kind of worlds: inertia worlds to ours and inertia worlds to those worlds maximally similar to ours. What varies is only the existence of different P individuals and relevant situations in these similar worlds. For example, in the evaluation of Brown-haired librarians are very quiet we will consider those similar worlds where brown-haired librarians, different from actual librarians, exist, while in the evaluation White cats are very shy, we will consider these similar worlds where white cats different from actual cats exist. Aside from this relatively minor variation, the set of accessible worlds relevant for the evaluation of all descriptive BP sentences is uniform for all such sentences. It is this uniformity which leads to the “uniform flavor” of all descriptive BP sentences. 3.3 Descriptive BP sentences as expressing a low degree of law-likeness Finally, following the observations of Lawler (1973), Burton-Roberts (1977) and Declerk (1991), I have shown above that although both BP and IS sentences assert the truth of a non-accidental generalization, the former assert more actual, descriptive patterns, which are less “definitional” or “law like” than the patterns asserted by corresponding IS sentences. Put differently, BP sentences are felt to express “in-reality patterns,” whereas IS sentences are felt to express more “inprinciple rules.” The distinction is seen most clearly when we compare these BP sentences with IS sentences expressing epistemic rules, as in (48a, b) and (49a, b): (48)

a. Dogs bark. b. A dog barks.

(49)

a. Tables have four legs. b. A table has four legs.

Both the BP and IS sentences here describe a non-accidental generalization, a generalization which is true in a set of possible worlds. Moreover, in both sentences the accessibility relation is realistic, i.e. the generalization is supposed to be true in the actual world as well. But nonetheless there is a difference between the two: The similarity of the accessible worlds in the case of descriptive BP generalizations is a vague and “wide perspective” similarity. This means that the accessible


worlds preserve a rather wide, nonspecific array of characteristics of the actual world, including both accidental and more principled and general characteristics of our actual set of circumstances. In fact, all we specify in the accessibility relation is the minor possible differences between the possible worlds and the actual ones (i.e. the existence of more, fewer, or completely different individuals and relevant situations). The assertion is, then, roughly “in all circumstances which are similar enough to actuality, preserving most of its characteristics, the generalization (“every P is Q”) holds.” This is why descriptive BP sentences are felt to express “actual patterns,” i.e. generalizations which, although nonaccidental, are felt to be “about the real, actual world.” In contrast, as explained in chapter 2, the accessibility relation of “in virtue of” generalizations, even of epistemic ones, preserves a specific aspect of the actual world, and ignores all other, irrelevant aspects of the actual world. Moreover, this relevant aspect of the actual world is not an arbitrary set of accidental facts, but a universal statement of the form “every P is S”, which itself constitutes a non-accidental generalization, or necessity in our world. The assertion is, then, roughly, that in all circumstances where a certain specific rule about our world (of the form “every P is S”) holds, the generalization (“Every P is Q”) holds as well. This seems to be the source of their strong law-likeness and definitional/ analytic flavor. 4. Advantages and an apparent problem with the underspecification of the superinterval I′ 4.1 Advantages In the truth conditions in (42), the universal statement is required to hold in all accessible worlds (those in inrmax) at some interval I′ surrounding the present interval I, where as explained above, no initial or final limit of I′ is specified in the semantic structure. This underspecification allows for a variety of possibilities for I′. The principal kinds of possibilities are schematically summarized in (50). In all of these cases the interval I′ (represented as the light gray rectangle) can be considered a proper superinterval of “now,” i.e. the present interval I (represented as the white vertical rectangle). In contrast, (51a, b) are cases where I′ is clearly not a superinterval of I:


(50)

(a)

(b)

(c)

(d)

(51)

(a)

(b) Assume that the horizontal line in (50a-d) and (51a,b) is the course of time in all the relevant accessible worlds (i.e. the inertia worlds to our world, and the inertia worlds to the worlds maximally similar to ours). Our truth conditions in (42) above now predict that a descriptive BP sentence will be true iff the universal statement holds in each of the possible cases of I′ in (50a-d), since in all of them I′ is indeed a superinterval of the present interval I, and false in (51a,b), where I′ is not a superinterval of I. These predictions seem to be borne out. Take, for example, the BP sentences in (52a) and (53a), and their corresponding universal statements in (52b) and (53b): (52) a. Tall members of the club have blue eyes. b. Every relevant situation involving every (non-exceptional) tall member of this club is a situation where this member has blue eyes. (53) a. Pentium VI computers are very expensive b. Every relevant situation involving every (non-exceptional) Pentium VI is a situation where this computer is very expensive.


Case (50a) is the classical one. According to this case, (52a), for example is true iff (52b) holds both long before and long after the utterance time in all the accessible worlds (including the actual world in the past and the “normal futures” to our world), i.e. all the (nonexceptional) actual past, actual present and the future tall members of this club indeed have blue eyes, as well as the ones who in our world are not club members. In such a case (52a) is indeed true. A case like (50b) was discussed in section (2.2.3) above. To reiterate, in such a case, (53a) for example, is true as long as all the (nonexceptional) actual and possible Pentium VI computers up to and including the present, are very expensive, even if two days from now, if nothing unforeseen happens, they will all become very cheap (e.g. if the new generation of Pentium computers hits the market). Case (50c) represents the opposite of (50b). Imagine that until now all (non-exceptional) Pentium VI computers have been very cheap. Two days ago they all became very expensive, and we believe that, if no surprises occur nothing unforeseen occurs all such (non-exceptional) computers will continue to be very expensive in the future as well, and that if these Pentium I computers were in fact Pentium VI they would be expensive (both now and in the future) as well. In such a case, as predicted by the truth conditions, (53a) is true, although the universal statement in (53b) was false very shortly before the utterance time. Finally, in (50d) the interval I′ is also a superinterval of I, although it is very short. In this case, a descriptive BP sentence is predicted to be true even if the universal statement is true only very shortly before and very shortly after the time of utterance. The prediction is borne out, for example in a case where kindergarten run by Wizo is on strike today, and where I believe that this is not accidental, i.e. that being on strike today would also be true of the three YMCA kindergartens, had they also been run by Wizo. In such a case, the sentence Wizo kindergartens are on strike is true if uttered during that day, even if the strike only started yesterday, and crucially, even if I believe it will stop the day after tomorrow. In other words, the underspecification of I′ correctly predicts that descriptive BP sentences can express even generalizations which are episodic and highly temporally restricted, as long as we take them to be non-accidental. More cases of “episodic generic” BP sentences are discussed in chapter 5 below. The predictions are also borne out the case of (51a) and (5 1b), where I′ does not surround I. If, for example, all Pentium computers were expensive until a week ago, but they are now cheap, and I expect them to continue being cheap in the future where no surprises occur nothing


unforeseen occurs, then the sentence “Pentium VI computers are expensive” [again] is false. Similarly, even if all Pentium VI computers start being expensive tomorrow, if until now all of them were cheap, and if I expect every non-actual such computer to be cheap now, then, this sentence is again false. 4.2 An apparent problem The truth conditions, then, seem to make correct predictions w.r.t. the truth of the universal statement in the superinterval I′. There is one case, however, where the predictions are apparently not borne out, as illustrated in the two schemes in (54a) and (54b): (54)a.

b. Consider again (53a) (Pentium VI computers are expensive). Imagine that in both (54a) and (54b) this sentence is uttered on June 7. As before, the light gray area represents the period where the universal statement (roughly “every (non-exceptional) Pentium VI computer is expensive”) holds. We can see that in both (54a) and (54b) there is no superinterval of the present interval in which this universal statement is true. In this sense (53a) is predicted to be false in both schematic cases. But in reality there is an intuitive difference between these two cases. In (54a) where all actual and possible Pentium computers were expensive between January and March, cheap between March and October, and are expected to become expensive again between October and December, (53a) is clearly false at the utterance time (June 7). On the other hand, in (54b), where all Pentium VI computers are indeed expensive throughout the whole year, except for a five-day sale


surrounding the utterance time, (53a) can much more easily be considered to be true. Intuitively, the reason for the difference between (54a) and (54b) is the fact that, relative to the bigger interval surrounding the present (i.e. the period from January to December), the Pentium computers in the five-day period where the universal statement is false in (54b)) can be considered much more easily “exceptional” than those in the five-month period (in (54a)). Another way to understand the pattern here is to compare the behavior of generics in (54a, b) to that of nongeneric sentences with progressive VPs, expressing activities like John was running for an hour (between 8.00 and 9.00). As pointed out by Landman (1991), this sentence is true even if John stopped to rest for five minutes during that hour, but not if he stopped for half an hour. Landman suggests that systematically, activities hold throughout a certain interval, with possible short “pause stages.” Similarly, we can claim that the requirement that a universal generalization is true throughout an interval I′ is fulfilled even when there are some short “pause stages” (as in (54b)) where the universal statement is false (and where the length of these pause stages is contextually determined). 5. The “existence” presupposition of descriptive BP sentences As we saw above, the distribution of felicitous descriptive BP sentences is less restricted than that of their IS counterparts. Nonetheless, there are also situations where the former kind of sentences become infelicitous. I now turn to deal with such cases, and show that when combined with several independently motivated claims and principles, the truth conditions developed in section (2.4) above make exactly the right predictions concerning these infelicitous cases, in a natural way. In this section I deal with one kind of infelicity, caused by a failure of what I will label the “existence” presupposition of descriptive BP sentences. In subsection (5.4) I show that IS sentences have similar presuppositions. 5.1 The existence requirement: The basic data Cohen (1999) argues that BP sentences can express descriptive generalizations even if there are no existing relevant situations, or


existing relevant individuals, to support the generalization. For example, Cohen claims that (55) can be true as a descriptive generalization, even if no emergencies have yet occurred. (55) Members of this club help each other in emergencies (where no emergencies have yet occurred) Cohen writes: (55) may be true even if no emergencies have ever occurred, say in a state of affairs where an obligation to help each other is included in the club’s constitution. Note that we are only concerned here with the descriptive readings of generics, and not with their prescriptive readings […] Suppose emergencies do eventually occur, but it turns out that club members fail to help each other. In (this case) the sentence would […] be false under the descriptive reading. (p. 222–223) I disagree with Cohen’s claim. If no emergencies have yet occurred, then (55) is neither true nor false on its descriptive reading, but is not appropriate as expressing a descriptive generalization at all, no matter what our expectations are concerning the behavior of the club members in the future. This disagreement about the status of (55) reflects a deeper disagreement about the nature of descriptive BP sentences. Cohen takes such sentences to express probability, future-oriented statements, asserting (roughly) what is likely to be (in most cases) in all (homogeneous) future histories continuing our present: (55) does not require that club members actually help each other in emergencies, merely that they are likely to do so. That is to say, in all sufficiently long histories containing emergencies club members will help each other in most cases. (p. 233) [emphasis mine] My view, on the other hand, is that descriptive BP sentences intuitively assert that “there is a pattern here” where “here” is the actual set of circumstances, i.e. that a certain generalization holds both in actuality, and in other similar possible circumstances (defined above), Intuitively, in situation where no relevant instances exist, or have existed, in


actuality we cannot start evaluating the descriptive generalization. [unless actuality = the situation where no relevant instances exist] Other examples may illustrate this point more clearly. Imagine a situation where, according to the constitution of this club, members pay their taxes for the first time on the first day of the third month after their registration. If we are only two months after founding this club, the sentence in (56) will be inappropriate as a descriptive generalization (although it can clearly have a deontic/legal “in virtue of” prescriptive reading). (56)

Members of this club pay their taxes on time.

This is because no member has had a chance of paying any taxes yet. Thus, even if it is true that it is likely that in all relevant futures the members will pay their taxes on time, the sentence in (56) cannot be evaluated as a true or false descriptive generalization. Sentences (55) and (56) are examples of cases where there are no existing relevant situations to support the generalization. The lack of descriptive reading is seen even more clearly where there are no existing individuals (i.e. P members). For example, the sentences in (57a-d) are all odd or infelicitous as descriptive generalizations : (57) a. # Tall members of this club have names ending with ‘t’. (where this club was founded week and no members have registered yet) b. # Astronauts working on Mars read Oliver Sacks’s books. (where there are still no astronauts working on Mars) c. # Copies of my dissertation have two pages missing. (where I have not yet printed even copy of my dissertation) d. # Letters from the king to the queen are hard to read (where the king has not yet a single letter to the queen) The infelicitous examples in (55)-(57) all illustrate cases where relevant instances (relevant situations or P individuals) exist only in the future (but not in the present or past of our world). However, we find a similar infelicity where we have existence of relevant instances only in the past (but not in the present of our world or in the expected future), as in (58a-c):


(58) a. # Dodos have brown feathers. (when dodos are extinct, i.e. no more dodos exist, and none are expected to exist in the future) b. # Members of this club have blue eyes. (where this club existed a hundred years ago, and so all of its members have died) c. # Letters from the king to the queen are written in blue ink (where the king is dead and all the existing letters were burned a minute ago)10 As before, the infelicity is felt not only w.r.t. P members, but also w.r.t. relevant situations involving them, as seen in (59), which is infelicitous if there are no more “first parties”: (59) # Members of this club get very excited at the last two parties of the club. (where there will be no more parties) On the basis of the data in (55)-(57) and (58)-(59) it looks as though the problem with the infelicitous BP sentences we examined is that there are no existing relevant instances (P individuals and relevant situations involving them) in the present. But upon examining a wider array of facts, we see that such a characterization is not accurate. In fact, we find perfectly felicitous descriptive BP sentences where there are relevant instances existing in the past of the actual world and in the (normal) future, but not in the present. Consider, for example, (60a) and (60b):

(60)

a. Women members of this club are very smart. b. Letters from the king to the queen are hard to read.

10

In the above situations, (58a-c) become felicitous if we change their tense to the past, as in (ia-c):

(i) a. Dodos had brown feathers. (when no more Dodos exist) b. Members of this club had blue eyes. (where this club existed a hundred ago, and so all of its members have died) c. Letters from the king to the queen were written in blue ink (where the king is dead and all the existing letters were burned a minute ago)


Imagine that in the past (e.g. several months ago) there were many women members of this club, and many letters from the king to the queen, but right now there are no women members, and all the letters the king wrote to the queen have been accidentally burned. If we believe that in the normal futures there will be no more women in this club (since e.g. the constitution forbids the existence of women members), and no more letters from the king to the queen (e.g. since the king is dead), then (60a) and (60b) are infelicitous. These sentences will be perfectly felicitous, however (although not necessarily true), in a situation where we expect that in the future new women will join the club, and where we expect the king to continue writing letters to the queen. There may be, then, a period, even a long one, surrounding the present, where no relevant instances exist (i.e. a “hole” inside the period of existence), and nonetheless the BP sentence will felicitously express a descriptive generalization. 5.2 The existence requirement is a presupposition What is the reason for the infelicity of descriptive BP sentences (like (57–59)) where the “existence requirement” is not met? My claim is that the reason is a presupposition failure, i.e. that the existence of relevant instances—individuals and relevant situations involving them— is presupposed by descriptive generalizations. This claim is supported by the fact that this “existence requirement” survives in the “family of sentences” test. For example, not only the original BP sentence in (57b), but also all the sentences in (61) presuppose that there are astronauts working on Mars (and are infelicitous if no such astronauts yet exist): (61) a. It is not the case that astronauts working on Mars read Oliver Sacks’s books. b. Is it true that astronauts working on Mars read Oliver Sacks’s books? c. If astronauts working on Mars read Oliver Sacks’s books, they can e-mail him and he will send them copies. Similarly, the existence implication of (57c) above survives the family of sentences test in (62). In other words, (62a-c) are odd if my dissertation is not finished yet, so no copies of it have been printed:


(62) a. It is not the case that copies of my dissertation have pages missing. b. Is it the case that copies of my dissertation have pages missing? c. If copies of my dissertation have pages missing, this can only be the fault of John. The same is true w.r.t. (56) above, where it is the existence of relevant situations, i.e. situations of paying taxes, which is presupposed. The sentences in (63a-c) all presuppose the existence of relevant, payingtaxes situations, i.e. they are odd if no such situations have yet existed (e.g. if the first opportunity for paying taxes will only occur in a month): (63) a. It is not the case that members of this club pay their taxes on time. b. Is it true that members of this club pay their taxes on time? c. If members of this club pay their taxes on time, they are probably honest. We find similar results in cases where we have existence only in the past, as in (64a-c), which are odd under the descriptive reading, assuming that dodos are already extinct: (64) a. It is not the case that dodos have brown feathers. b. Is it true that dodos have brown feathers? c. If dodos have brown feathers, it is because they eat brown seeds. 5.3 Triggering the presupposition The “existence requirement” is presupposed, then, by descriptive BP sentences. How can we guarantee that such sentences will indeed systematically have this presupposition? It is clear that we cannot simply add this presupposition to the truth conditions of such generalizations. As already mentioned in chapter 2 above, we assume that presuppositions are conventionally triggered by a component in the semantic structure or conversationally triggered by an independent pragmatic (conversational) mechanism, or, sometimes by a combination


of the two. In what follows I will claim that the “existence” presupposition of descriptive BP sentences is triggered by a combination of the truth conditional requirement that the universal statement holds at a superinterval of the present interval, (conventional triggering) together with the independently argued special characterization of the accessibility relation of descriptive BP sentences, and the Grician maxim of relevance (conversational triggering). 5.3.1 The conventionally triggered “existence presupposition” Let me repeat, first, the truth conditions of descriptive BP sentences: (65) Truth conditions of simple present descriptive BP sentences A BP sentence is true in <w, I> iff

We may start explaining the “existence” presupposition observed above by suggesting that it is conventionally triggered by the universal quantification over worlds, individuals and situations in the truth conditions. Such a suggestion is clearly on the right track. As shown in chapter (2) above, it is standard to assume (see e.g. de Jong & Verkuyl (1984)) that there are some relevant existing instances in order to evaluate the truth of a universal statement in a nontrivial way (since if there are no instances the universal statement is trivially true).11 Assuming such a conventional triggering, the presupposition of (65) is (66):

11

Interestingly, de Jong and Verkuyl (1984) make a distinction between “contingent” universal statements like (i) and “law-like” ones, like (ii), (a distinction which is independently discussed in Carlson (1989), and formally analyzed in Kratzer (1989)), and claim that only the former presuppose the existence of Ps:

(i) (ii)

All ravens are black. All seats are taken.


(66) Conventionally triggered existence presupposition of the truth conditions

Paraphrase: There are at least some accessible worlds w′, which are members of the set inrmax (<w, I>, and a superinterval of I - I′, where there are at least some P individuals and some relevant situations involving them, which are located at that superinterval I′. However, the conventionally triggered “existence” presupposition in (66) fails to capture the observations in sections (5.1) and (5.2) in two points. That is, there are two kinds of mismatches between this presupposition and what seems to be the “real” existence presupposition of descriptive BP sentences. In the next two sections I point out these two mismatches, and show that they can be naturally resolved once we combine the presupposition in (66) with two other independently motivated claims. 5.3.2 A mismatch (between the conventionally triggered and the “real” existence presupposition), w.r.t. the relevant interval involved, and its resolution 5.3.2.1 The mismatch The first mismatch concerns the characterization of the interval where relevant instances are required to exist. The problem is that the

As far as the “existence” implication concerns, then, the modal quantifier associated with descriptive BP sentences (e.g. with Brownhaired librarians are very quiet) is closer to the extensional universal quantifier in (ii), than to the intensional one in (i). The reason for this is the role that the actual world plays in defining the intensionality (modality) of the generic operator with descriptive BP sentences, as explained below.


conventionally triggered presupposition in (66) requires existence at some superinterval of the present, but such a requirement is too weak: it is easily met in the case of the infelicitous descriptive BP sentences, observed in section (5.1) above, like (57b) (#Astronauts on Mars read Oliver Sacks’s books) and (58a) (#Dodos have brown feathers). In both these cases there are indeed some existing instances at some (proper) superinterval surrounding the present. In (57b) these instances exist in the final subinterval of the superinterval (i.e. in the normal futures), and in (58a) they exist in the initial subinterval of this superinterval (i.e. in the past of our actual present). The presupposition in (66), then, wrongly predicts cases like (57b) and (58a) above to be felicitous. What is, then, the “real” existence presupposition of descriptive BP sentences, which should replace the conventionally triggered one in (66) above? At first sight we might replace the requirement on existence at the superinterval (as in (66)), with a requirement existence of instances throughout the surrounding interval I′, i.e. at every (reasonably long) subinterval of I′. This requirement is stronger than the one made by the conventional presupposition, and it correctly excludes infelicitous cases like (57) and (58) above, since in neither do we find existing instances throughout the superinterval I′. At the same time, however, this requirement seems to be too strong, since it wrongly excludes felicitous cases where there is a “hole” in the interval surrounding the present, i.e. where no relevant instances exist in a large subinterval of I′. For example, requiring that P individuals exist throughout the interval surrounding the present predicts (60b) (Letters from the king to the queen are hard to read) to be infelicitous in a scenario where the king has written, and will be writing letters, although right now no letters exist. In reality, however, and unlike the sentences in (57) and (58) above, this sentence is perfectly felicitous in this scenario. Moreover, unlike what happens with the requirement on truth, which allows the sentences to be true in cases where the universal statement is false for very short periods, i.e. which allows for “short pause stages” or “little holes” (as illustrated in the scheme in (54a) above), in the case of the “existence requirement” the “hole” is allowed to be rather big, i.e. the descriptive BP sentences can be felicitous even if there is a very long period in which no relevant instances exist. Consider, for example, (67): (67)

Yellow snakes climb on high trees during earthquakes.


Assume a scenario where there is a big “hole” in the superinterval I′ in terms of the truth of the universal statement, e.g. where many years ago, during every earthquake situation, indeed every non-exceptional yellow snake climbed on a high tree, but in the last eight years or so this has no longer been true. In this scenario, even if we think that the universal statement will be true again in a year or two, (67) is false as a descriptive generalization. But now look what happens in case there is a big “hole” in the superinterval I′ in terms of existence of relevant instances. If, for example, there haven’t been any earthquakes for the last eight years or so, and we expect that there will be earthquakes in the future, then (67) is perfectly felicitous (although not necessarily true). This difference between the requirement on truth and the requirement on felicity strongly suggests that requiring the existence of relevant instances throughout I′, or even in every reasonably long subinterval of I′, is the wrong direction to take. Instead of requiring existence at the superinterval I′, or throughout it, then, it seems that descriptive BP sentences presuppose existence of relevant instances (at least) at the “edges” of the superinterval I′, or, more precisely, in both the initial and final subintervals of I′. In other words, we want to have relevant situations and P individuals at least in the past (of the actual world) and in the normal futures (and, of course, possibly also in the present). Formally, rather than the conventionally triggered presupposition in (66) repeated here, we can now phrase the real presupposition of descriptive BP sentences as in (68) (I underline the difference between the two). The definitions of initial and final subintervals in (69) and (70) is taken from Dowty (1979, p. 140): (68) Conventionally triggered existence presupposition of descriptive BP sentences

(69) Real “existence” presupposition of descriptive BP sentences

(70) I is an initial subinterval of J iff I is a subinterval of J, and there I such that t≤t′. is t (J—I) which there is t′


(71) I is a final subinterval of J iff I is a subinterval of J, and there is I such that t′≤ t. no t (J—I) for which there is t′ Sentence (68) seems indeed to be the right way to formalize the real existence presupposition of descriptive BP sentences. Since it is now clear that this presupposition is not merely conventionally triggered by the presence of the universal operator in the truth conditions, our next task is to identify what other factors combine with this operator to trigger (68). Put differently, we need to clarify why the universal quantification which binds instances located at a superinterval of the present in the truth condition triggers the presupposed existence of such instances at both edges of that interval. In the next section I show that this question is naturally answered by combining the conventional presupposition with an independently motivated conversational mechanism. 5.3.2.2 A resolution of the mismatch: triggering existence at both edges of (instead of at) the superinterval I′ To identify the additional factor which triggers existence of instances at both edges of (instead of at) I′ note, first, that similar existence can be found not only with BP sentences, and not only with predicates in the simple present but also with past tense extensional universal statements like (71) and (72): (71) Between 1950 and 1980, every member of the club received a present for the holidays. (72) Between 1950 and 1980, every speech made by McNeil was quoted in the newspapers. (71) is false if in the period mentioned, there were club members who did not receive a present for the holidays. But imagine a situation in which the club was only founded in 1966, i.e. there were no club members before. In this context, assuming that the speaker knows this, the sentence is odd, or infelicitous. Similarly, (72) is false if not every speech made by McNeil between 1950 and 1980 was quoted in the newspapers, but infelicitous if McNeil died in 1968, or stopped making speeches at that time.


The reason for infelicity in both sentences is the same. The universal quantifier (lexicalized as “every” in (69) and (70)) conventionally triggers an existence presupposition, i.e. it is only relevant or appropriate to make a universal statement about all P individuals if there are P individuals around. In addition, the speaker explicitly indicates that the period where this universal statement is true is between 1950 and 1980. We assume that the speaker adheres to Grice’s maxim of relevance, i.e. that s/he believes and implies that that the universal statement is relevant (nontrivially true) in that period. In a context where it is clear for the speaker and the listener that such P individuals exist in a shorter period (e.g. only from 1950 to 1968), the maxim of relevance is violated, since the universal statement is in fact relevant (nontrivially true) for a shorter period than the one indicated by the speaker.12 My claim is that essentially the same thing happens with descriptive BP sentences. These sentences express universal statements which are asserted to hold in a set of possible worlds (those in inrmax ) at some superinterval I′ of the present interval. Unlike what happens with (71) and (72), neither the universal operator nor the truth of the universal statement at a certain time interval (the superinterval of I) are

12

A possible exception is a case where the speaker takes the indicated and explicitly referred-to period to be relevant for some other reason. Imagine, for example that I lived in the US between 1960 and 1975, and now (in 2001) I utter (i):

(i) When I lived in the U.S. I had to pay lots of money every time I was hospitalized. Imagine now I was hospitalized several times between 1960 and 1970, but not even once between 1970 and 1975. In such a case the sentence is felicitous, although there are no existing relevant situations (of hospitalization) during a large portion of the indicated period. This means that, unlike what happens with (76) and (75) above, the maxim of relevance is not violated. The reason is that in this case the relevance of the period 1960 to 1975 does not lie in the existence of relevant situations, but in being a period of living in the US., i.e. the speaker is trying to say something about some aspect of living in the US, and this is what makes the whole period relevant although relevant instances exist only in part of it.


lexicalized (through “every” and a time adverbial), but nonetheless, both components are inherent parts of the assertion made by of descriptive BP sentences, i.e. part of their semantic structure. Thus, in a situation where it is clear that the existence presupposition is met only in the past (as in e.g. Dodos have brown feathers) or only in the future (as in Astronauts on Mars love Oliver Sacks’s books), it is not appropriate to assert that the universal statement is (nontrivially) true in a superinterval of the present, since it actually holds only in the interval before or after the time of utterance. In this case the speaker again violates Grice’s maxim of relevance: s/he implies that the universal statement (nontrivially) holds in a period, when in fact s/he knows that it holds (nontrivially) in a shorter period. This seems to be the source of the infelicity of the descriptive BP sentences in (57)-(59) above. The existence presupposition in (68), then, is not only conventionally, but also conversationally, triggered. The assertion that the universal statement holds at some superinterval of I, I′, is combined with the Grician conversational maxim of relevance trigger the presupposition that there are relevant instances existing (at least) at both edges of the superinterval I′ , i.e. (at least) at the initial and final subintervals of I′. This definition guarantees existence at the superinterval as a whole. 5.3.3 A mismatch (between the conventionally triggered and the “real” existence presupposition), concerning the relevant world involved, and its resolution 5.3.3.1 The mismatch Having explained the first mismatch between the conventionally triggered and the “real” existence presupposition, concerning the definition of the relevant interval, we can now turn to the second mismatch between them, which concerns the definition of the appropriate world. The conventionally triggered presupposition in (66), repeated here, requires that there be existing instances (P individuals and relevant situations involving them) in some accessible world, where crucially, this world can be any arbitrarily chosen world in the set inrmax (<w, I>):


(66) Conventionally triggered existence presupposition of descriptive BP sentences

However, in reality, descriptive BP sentences seem to presuppose that instances need to exist in the actual world and its normal future. For example, for the sentence Astronauts on Mars read Oliver Sacks’s books to be felicitous as a descriptive generic, it is not enough for there to be astronauts on Mars in the present of some possible world similar to ours (or in the past and normal future of such a similar world). Instead there should be astronauts on Mars in our world (and its normal futures). Put in other words, the existence of hypothetical, possible astronauts is not enough. For this sentence to be felicitous as a descriptive generalization we need there to be real, actual astronauts on Mars. Similarly, for the sentence Dodos have brown feathers to be felicitous as a descriptive generalization, it is not enough that there are non-actual dodos, at the present interval I of some world similar to ours (or alternatively in the past and normal future of such a possible world). If there are no existing actual dodos in our world (right now, or in the past and the expected normal future of our actual world), this sentence is again infelicitous as a descriptive generic. 5.3.3.2 A resolution of the mismatch: triggering presupposed existence in the actual world The conventional presupposition, then, is too weak for our purposes. We must find what is it that forces the universal quantification over all accessible worlds in the truth conditions to presuppose existence of instances in the actual world (and its normal future), rather then in some arbitrarily chosen accessible world which is not necessarily the actual one. I will now show that it is the special nature of the accessibility relation of descriptive BP sentences which triggers the presupposed existence of instances in the actual world. This accessibility relation, developed in section (2.4) above, is repeated now in (73): (73) inrmax (<w, I>): {

}


Paraphrase: Inrmax is a function which takes world interval pairs, and gives a set of worlds as a value. This output set is a union of the worlds which are inertia worlds to <w, I> and those which are inertia worlds to <WLewisian, I> (i.e. inertia worlds to the worlds which are maximally similar to w, except from what is needed to allow for the fact that the P set of individuals and the set of relevant situations are not identical to the sets of P individuals and relevant situations existing in w, respectively). In the sections above I showed that what is special about the characterization of the accessible worlds in (73) is their “wide perspective” similarity to the actual world (due to the use of the Lewisian “maximal similarity” and Dowty’s notion of “inertia worlds”). I showed that characterizing the accessible worlds in such a way helps to account for several special properties of descriptive BP sentences (as opposed to their IS counterparts). Now we can point to an additional advantage, concerning the existence presupposition of these sentences. The actual worlds plays such a central part in the definition of this accessibility relation, so that if no instances (individuals and situations) exist in the actual world, the definition in (73) guarantees that none exist in any accessible world. In other words, due to the way the accessibility relation is defined, the conventionally triggered presupposition that there are existing instances in some accessible world fails exactly in cases where no instances exist in the actual world, or in its normal futures. To see why this is so consider first the case of nonexistence in the actual past (and or present): If no instances (individuals and situations) exist in the past of the actual world (e.g. if there are as yet no actual astronauts on Mars), then, firstly, none will exist in the initial subinterval of I′ (i.e. the past, and/or present) of the inertia worlds to <w, I>, since these worlds are, by definition, identical to the actual world until now. Secondly, in such a case, no instances (e.g. astronauts on Mars) will exist in the past and/or present of the other kind of accessible worlds, namely those worlds which are defined as “maximally similar to ours, except from what is needed to allow for the fact that the P set of individuals and the set of relevant situations are not identical to the sets of P individuals and relevant situations existing in w, respectively.” If, for example, there are no individual astronauts on Mars in w, then there


can’t be such individuals in the set of astronauts on Mars which is defined as “not identical to the set of astronauts on Mars existing in w.” Put differently, if in some set of possible worlds there are astronauts on Mars, whereas in the past or present of the actual world there are as yet no such astronauts, then the possible worlds in this set cannot be considered anymore “maximally similar to ours” in the way that the definition in (73) requires. Let us turn now to nonexistence in the present and/or the (normal) futures of our world: If no instances are expected to exist in the future of our world, i.e. in the final subinterval of I′ (e.g. if dodos are extinct, they are not expected to exist in the “normal futures” of our world), then, firstly, no such instances will exist in the inertia worlds to ours at that final subinterval (which are, by definition, the “normal futures”)Secondly, in such a case no instances will exist in the final subinterval of the inertia worlds to the worlds maximally similar to ours, i.e. in the “normal futures” of the worlds maximally similar to ours. For example, if dodos are no longer expected to exist, a world in which such dodos suddenly appear can no longer be considered maximally similar to ours. We can see, then, that although the universal quantification over worlds in the truth conditions developed above for descriptive BP sentences is not enough to trigger the presupposed existence of instances in the actual world, we get the latter presupposition by combining the conventional presupposition of the truth conditions with the independently-argued-for definition of the accessibility relation of descriptive generalizations. This accessibility relation, and especially the central role of the actual world in it predicts that the conventional presupposition fails exactly when we would expect it to.13 In this sense, the observations above concerning the “existence” presupposition can be taken as supporting the truth conditions and accessibility relation that we developed for descriptive BP sentences in the sections above.

13

An alternative way to explain the presupposed existence in the present (and/ or past) of the actual world is to claim that if no instances exist then the whole basis for the descriptive generalization (namely the observation of instances) no longer exists, and this leads to infelicity (see section 6 for a similar argument concerning the “enough” presupposition of descriptive generics). Such a claim, however, cannot explain the presupposed existence of instances in the normal futures of the actual world.


5.4 Existence presuppositions of IS sentences In our discussion of the difference between the truth conditions of IS and descriptive BP sentences, we have concentrated on the different accessibility relations associated with these constructions. Note, however, that there is another difference between the two: Our formulation of the truth conditions of descriptive BP sentences (in (42) above) includes a temporal component: the requirement that the universal statement holds at some (proper) superinterval of the present (and as a result, the presupposition that there are existing instances at both “edges” of that superinterval). No such temporal component appears in the truth conditions we developed in chapter 2 for IS sentences, and repeated here: (74)

Original truth conditions of IS sentences: An IS sentence is true at w iff

In these original truth conditions, the IS sentence is evaluated relative to a world only, and the universal statement in (74) is required to hold in only a set of appropriately accessible worlds. It seems, however, that a uniform representation, where both IS and descriptive BP sentences are evaluated relative to world-interval pairs, is more desirable. We should integrate, then, intervals into the truth conditions of IS sentences as well. I will now discuss some data indicating that such a move is not only desirable for uniformity reasons, but is empirically motivated as well. The data concerns “existence” presuppositions with IS sentences. 5.4.1 Existence presuppositions of IS sentences-basic data At first glance, it seems that “existence” presuppositions are not relevant for IS sentences, since, unlike descriptive BP sentences, they express rules and can be felicitously uttered where there are no P individuals and relevant situations involving them, as (75) and (76): (75)

A 110-year-old Malaysian king chooses his descents alone.


(76) A single parent of more than ten children who earns more than $2 million year pays 23% income tax. Sentence (75) expresses a deontic “in virtue of” generalization. It is perfectly felicitous even if no 110-year-old Malaysian king exists now, and if were no such kings in the past. Sentence (76) expresses a legal “in virtue of” generalization. It is felicitous even if there is no single parent of more than ten children who earns more than $2 million a year, and there haven’t been any such single parents in the past. There are, however, some IS sentences where we find existence presuppositions very similar to the ones found in descriptive BP sentences, specifically IS sentences expressing epistemic “in virtue of” generalizations e.g. (77) and (78): (77)

A cloned dog is very aggressive.

(78)

A female dinosaur lays eggs every three years.

Imagine now that, although there are already cloned sheep in the world, no one has yet managed to clone dogs. The first cloned dogs are likely to appear (i.e. to be cloned in the laboratories) in approximately a year from now. In this scenario, where cloned dogs exist in the “normal futures” only, (77) is neither true nor false, but inappropriate. Similarly the present tense (7b) is odd if dinosaurs (including female dinosaurs) are extinct, i.e. if female dinosaurs existed only in the past, but not in the present or (normal) future. These “existence” implications survive in the “family of sentences test” in (79) and (80): (79) a. It is not the case that a cloned dog is very aggressive. b. Is it true that a cloned dog is very aggressive? c. If a cloned dog is very aggressive, it must be because it is not a natural dog. (80) a. It is not the case that a female dinosaur lays eggs every three years. b. Is it true that a female dinosaur lays eggs every three years? c. If a female dinosaur lays eggs every three years, then this must be due to the special of dinosaurs.


The data suggest, then, that present tense epistemic IS sentences have existence presuppositions which are very similar, if not identical, to the existence presuppositions of descriptive BP sentences. Here too the requirement is that there are existing P individuals and relevant situations involving them at both “edges,” i.e. some initial and final subintervals of some interval I′, surrounding the present, in the actual world, and its normal futures. This requires us to add time intervals to the truth conditions of IS sentences, in a similar way to their descriptive BP counterparts. From now on, then, simple present IS sentences have the following truth conditions: (81) Revised truth conditions for simple present IS sentences: Let IS NP denote a property ^P, VP denote a (verbal) property ^Q, and ^S is a property A simple present IS sentence of the form IS VP is true in <w, I> iff:

“There is a property ^S, s.t. in all worlds where every P individual has S (and ^S is associated with ^P in w), there is some superinterval of the present I′, and every contextually relevant situation involving any P individual, which is located at I′, is a Q situation involving such an individual.” Due to the universal quantification and the Grician maxim of relevance, discussed above, these truth conditions have, now, an “existence presupposition,” similar to that found with descriptive BP sentences: (82) “Existence presupposition of IS sentences”:

“There are some worlds where every P individual has S (where ^S is associated with ^P in w), and some superinterval of the present interval, I′, and there are some P individuals in these worlds, and relevant situations involving them in these worlds, which are located at the initial and final subinterval of I′.”


In chapter 2 above, I discussed the presupposition of existence of some accessible worlds (this presupposition is violated when the IS subjects expresses an extremely unnatural class as in # A Norwegian student whose name ends with ‘s’ or ‘g’ wears thick green socks). Until now, however, we did not discuss the presupposed existence of individuals and situations. 5.4.2 What triggers existence in the actual world with epistemic IS sentences If (82) is indeed a presupposition of IS sentences, triggered by the universal quantification in the truth conditions (and the Grician maxim of relevance), then it is expected to be presupposed by all such sentences. In reality, however, we can feel the effects of this presupposition (w.r.t. individuals and situations) only with a subset of IS sentences, i.e. only with epistemic IS sentences (as in (77) and (78)), but not with deontic IS sentences (as in (75) and (76)). In addition, we face a problem similar to the one we faced with descriptive BP sentences, above: The presupposition in (82) requires the existence of some individuals and situations only in some accessible worlds, and there is nothing in it which explicitly and directly guarantees that one of these worlds is the actual world. In reality, however, we can see that the presupposition fails (and we get infelicity) when no instances exist in the actual world (with epistemic IS sentences). The solution to these two problems has to do with the “association with ^P” requirement on the ^S property (i.e. ^S CR(^P,w) in (83)), which systematically restricts the accessibility relation of IS sentences, and the special nature of epistemicity. To repeat, the accessible worlds of IS sentences are all those worlds where “every P individual has S” is true, where ^S needs to be associated with ^P in w, as defined in (83): (83) Association of properties: ^S CR( ^P, w). iff there is some Kratzerian function f (from worlds to sets of sets of worlds) s.t.

Paraphrase: ^S is associated with ^P in w iff it holds of every P individual in all worlds which are epistemically, deontically, legally, stereotypically, etc. accessible from the actual world.


Now, in case that the IS sentence has an “epistemic” flavor (as in (78a, b)), i.e. it is understood as expressing an epistemic rule, ^S is taken to be “epistemically associated,” i.e. it meets the condition in (83’): (83’) ^S is s.t.

Paraphrase: ^S is s.t. every P individual has it in all worlds w” where what is known in w holds. But now look at what happens when no P individuals exist in the actual world, e.g. where there are no longer any female dinosaurs. In such a case, the fact that no P individuals exist, is part of our knowledge about the actual world. Hence, in such a case P individuals do not exist—not only in the actual world itself—but also in all worlds where what is known in our world holds. This means that if no P individuals exist in our world, then none exist in any epistemically accessible world w.” In such a case, however, the condition on association of properties in (83) is trivially met (if no female dinosaurs exist in any epistemically accessible world, then the condition that in all such worlds every female dinosaur has S is trivially true). Hence in such a case (83) is inappropriate. But since (83) is a condition on the general accessibility relation of IS sentences, if (83) is trivial, and inappropriately made, then we cannot properly define the general accessibility relation either (e.g. we cannot define an accessible world w′, where every dodo has an associated property S), and the IS sentence becomes infelicitous. Essentially the same thing happens in (77), in case no cloned dogs exist in the actual world. We can see, then, that requirements on the set of accessible worlds of IS sentences, i.e. the fact that it depends on another modalized condition, together with the special nature of epistemic modality, lead to infelicity of epistemic IS sentences when no P individuals exist in the actual world. The role of the actual world with the accessibility relation of such IS sentences is less direct than with the accessibility relation of descriptive BP sentences, discussed above, but in both cases it leads to similar results (i.e. the failure of the existence presupposition) when no instances exist in this world.


5.4.3 The existence presupposition with deontic IS sentences We can now also understand why the absence of existing instances in the actual world does not lead to the failure of the existence presupposition in (82), when the IS sentences are deontic, as in (75) and (76), above. In contrast to (77) and (78), the ^S properties relevant for the truth of (75) and (76)) are not epistemically, but rather deontically associated with ^P. In such a case, the condition on association of properties in (83) above is that in all deontically accessible worlds, every P individual has S. The point is that this condition can be easily and nontrivially met even if no P instances exist in the actual world. For example, even if in the actual world there is no (and hasn’t been) a 110-year-old Malaysian king, such a king can clearly exist in worlds which are deontically accessible from w (i.e. where what is commanded in w holds). Thus, even in such a case the condition on association of properties (in (83) above) can be nontrivially met. This in turn means that the general accessibility relation of the deontic IS sentence can be properly defined as well, and as a result deontic IS sentences as in (75) and (76) are felicitous. 5. 4.4 Two further supporting facts Having understood the nature of the “existence” presuppositions of IS and descriptive BP sentences, we can now turn to account for two further pieces of data. As I will now show, the explanations of both pieces of data follow directly from the claims and definitions made above, thus further supporting the truth conditions, accessibility relations and presuppositions developed so far. 5.4.4.1 The first support: defeasibility of the existence presupposition with epistemic IS sentences The first supporting fact concerns contexts where the existence presupposition is canceled with epistemic IS sentences. Imagine the following scenarios: Three weeks ago several researchers found some birds in Australia which look very much like dodos. These people published a paper claiming that dodos are not in fact extinct.


Many other people, however, object to this claim, and believe that these birds are not dodos at all. Imagine now that I am one of the students working with these researchers in Australia, and I just found nests of these birds, with yellow, round eggs in them. I immediately call my professor, who is the dodo expert, to tell him what I just found, and he replies by uttering (84): (84) If this is what you found in the nests then I’m less sure that these birds are dodos. A dodo lays long, brown eggs. Sentence (84) is perfectly felicitous in this scenario. But why? In this scenario, the nonexistence of dodos in the actual world is under debate. Thus, unlike what usually happens, the nonexistence of dodos in our world is not part of the common knowledge about the actual world. It is still epistemically possible (even if a bit far-fetched) that dodos exist in our world. Since possibility indicates truth in some accessible worlds, in such a scenario there are still some epistemically accessible worlds where dodos exist. The result is that the condition on association of properties (in (83) above) can be nontrivially met, and the IS sentence is felicitous. 5.4.4.2 The second supporting fact: a difference between epistemic IS and descriptive BP sentences concerning existence of relevant situations in the actual world I showed above that both descriptive BP and epistemic IS sentences are infelicitous when no P individuals exist in the actual world. However, when we examine contexts where there are no existing relevant situations in the actual world, we can see that such contexts have a different effect on the felicity of these two kinds of sentences. Consider, for example, the following scenario: Scenario A: No cloned hens exist yet. The first ones will exist 3 months from now. In this scenario, both the present tense IS in (85a), and the minimally contrasting BP sentence—on its descriptive reading—are infelicitous: (85) a. A cloned hen lays infertile eggs. Don’t even think of buying such a hen. b. Cloned hens lay infertile eggs. Don’t even think of buying such hens.


Given our claims above, this infelicity is expected (since no individuals exist in the actual world in the past and/or present). But now consider these two sentences in a second scenario: Scenario B: The first 30 cloned hens have already existed for five days. Thus, there are already cloned hens in the world, and we expect more cloned hens in the normal futures. However, these hens are still chicks. None of them has laid an egg yet, and we have to wait for about six week until these hens start laying eggs. In this second scenario we find a difference between (85a) and (85b). The IS (85a) (“A cloned hen lays infertile eggs”) is now felicitous (if we imagine, for example, that although no hen has laid eggs yet, the speaker of (85b) is one of the scientists working in the cloning laboratory, who has in mind some special genetic property of cloned hens, which will lead to laying infertile eggs). In contrast, if we utter (85b) (“Cloned hens lay infertile eggs”) as a descriptive generalization this sentence will be odd, or infelicitous in this second scenario. This difference between descriptive BP and epistemic IS sentences follows from the difference we discussed above, between the role that the actual world plays in the nontrivial definition of the “in virtue of” and “descriptive” accessibility relations. I showed above that the role of the actual world is much more direct with descriptive BP sentences than with epistemic IS sentences. In particular, I have shown that with descriptive BP sentences, due to the “wide perspective” similarity to the actual world required of the accessible ones, if no individuals or situations exist in the actual world, then none will exist in any accessible world, and as a result the “existence” presupposition will fail. In contrast, we just saw that although the existence presupposition has the same effect on epistemic IS sentences w.r.t. the existence of P individuals, they behave differently as far as the existence of relevant situations is concerned. This is because the role of the actual world in the accessibility relation is much less direct. More specifically, I claimed above that the existence presupposition of epistemic IS sentences fails when the condition on association of properties (namely the conditional “in all epistemically accessible worlds every P individual has S” as in (83’) above) is trivially met. Since this condition is phrased as a universal quantification over P individuals it is indeed trivially met when there are no existing P individuals in any world where what is known in w holds, and this, in turn, happens when no P individual exists in the actual world w (e.g. there are no cloned hens in the world).


In contrast, even if there are no existing relevant situations in the actual world, this association relation is nontrivially met as long as there are exiting P individuals. For example, if there are existing cloned hens, then the conditional “in all epistemically accessible worlds every cloned hen has S” can be nontrivially met, even if no situations of these hens laying eggs exist yet. Thus, in such a case, the association condition, and consequently also the general accessibility relation of epistemic IS sentences can be properly and nontrivially defined, and the sentence is felicitous. In other words, the nontriviality of the association relation, which is the key for the nontriviality of the accessibility relation of IS sentences, and for their felicity, is not dependent on the existence of relevant situations in the actual world, only on the existence of P individuals. To summarize, the variety of rather subtle distinctions in felicity pointed out above, concerning the existence presupposition with IS and descriptive BP sentences, is naturally accounted for by the truth conditions and accessibility relations developed above, together with some independently motivated assumptions and claims. In the next section I turn to the second presupposition of descriptive BP sentences, which I call the “enough” presupposition. 6. The “enough” presupposition of descriptive BP sentences and the relationship between descriptive generics and the process of inductive inference My claim throughout this chapter has been that the intuitive meaning of descriptive generics is, “there is a pattern here or this is nonaccidental.” The truth conditions developed so far for descriptive BP generalizations capture this intuitive meaning by, very roughly, requiring the universal generalization “every nonexceptional P is Q” to hold both in the actual index as well as in the similar enough indices, or circumstances. Until now, however, the truth conditions did not relate at all to another aspect of descriptive generalizations, namely their connection with the process of inductive inference, i.e. the process by which someone observes a large enough sample of Ps which are Qs (i.e. Ps Qs), and concludes that indeed “this is nonaccidental or there is a pattern here.” This inductive aspect seems to be strongly connected to the meaning of descriptive generalizations, and constitutes part of their intuitive characterization. In fact, the first theories (Burton-Roberts (1977), Declerk (1991), Carlson (1995), etc.) which attempted to


characterize the intuitive difference between IS and BP generics, used the term “inductive readings” or “inductive generics” for the meaning of BP sentences (as opposed to the definitional/analytic flavor of IS sentences). My purpose in this section is to formally clarify the relationship between descriptive generalizations and inductive inferences, that is, show to what extent, if at all, and how exactly, should the process of inductive inference be part of the truth conditions for descriptive BP sentences. I will do this by examining an additional requirement on the felicity of descriptive BP sentences, different from the “existence” presupposition discussed in the previous section, which, if not met, leads to another kind of infelicity. 6.1 The “enough” implication: basic data and intuition Consider the following sentence: (86)

Students registering for this course have brown hair.

Sentence (86) is felicitous (although not necessarily true) if there have already been 67 students who registered for this course. It is odd, or infelicitous, however, if there are only three such students, even if all three have brown hair. It would be much better to say in this second scenario “The students who registered for this course have brown hair.” We get the same intuition with the sentences in (87)-(90). These BP sentences, expressing descriptive generalizations, are odd in the first scenario, where there are very few P members, even if all of them have the Q property, but fine where there are many of them: (87) Green-covered books about the semantics of genericity contain no typos. (a) Only three such books have ever been published. (b) More than 100 such books have already been published. (88) Jews in Nevada are tall. (a) There are only three Jews living in Nevada. (and this is the average number of Jews before the present time as well) (b) There are 3000 Jews living in Nevada.


(89)

Rivers in this country are dirty. (a) There are two rivers in this country. (b) There are twenty rivers in this country .

(90)

New-model Volvos have a scratch on one of their wheels. (a) There are only four such cars. b) There are 3000 such cars.

We get similar results if there are very few relevant situations around. Sentence (91) is odd in the first scenario, even if there are many individual male math teachers: (91) Male teachers of mathematics arrive at school this year 6.5 minutes early. (a) There have been only two school days this year. (i.e. (91) is uttered after the second day of the school year) (b) There have already been 79 school days this year. In all of the (a) cases in (86–91) there are some P individuals, and relevant situations involving them existing at the time of utterance, i.e. in the present interval, and also before and after it (i.e. in the past and “normal futures”). Thus, the oddness of these sentences in these scenarios does not seem to result from the failure of the existence presupposition. Instead, the oddness seems to result from the fact that there are simply not enough P individuals or relevant situations involving them around to support the generalization. It seems, then, that we should add to the truth conditions of descriptive BP sentences something like the intuitively phrased implication in (92): (92) “Enough” implication (intuitive): For a descriptive BP sentence to be felicitous, there should be enough P individuals and relevant situations involving them around. 6.2 The “enough” implication is a conversationally triggered presupposition An apparent potential advantage of adding something like (92) to the truth conditions is that we may be able to use it to subsume the


“existence” presupposition above. If (92) is a presupposition of BP sentences, then it is clear that the requirement that there be enough P individuals and relevant situations involving them around fails if there are no P individuals and relevant situations involving them existing at all. But is (92) indeed a presupposition of descriptive BP sentences? On the one hand, if this requirement is not met, the descriptive BP sentences are infelicitous. Moreover, as with ordinary presuppositions, the implication that there are “enough” P individuals and relevant situations involving them around seems to be somehow taken for granted or “backgrounded” by the BP sentence. For example, (93b) seems to be taken for granted by a speaker and listener of (93a): (93) a. Letters from John to his lawyer contain an odd number of letters. b. There are enough letters from John to his lawyer. (and not only two or three) However, on the surface, and unlike the “existence” presupposition discussed above, there is nothing in the semantic structure of descriptive BP sentences like (93a) to trigger a presupposition like (93b) (e.g. do you mean i.e. here? universal quantification is perfectly felicitous even if there are only three entities to quantify over). Moreover, unlike the “existence” presupposition, the implication that there are “enough” relevant instances around does not always survive in the entire “family of sentences,” and this may suggest that it is not a presupposition of descriptive BP sentences. Consider, for example, (57b) and (61a-c), repeated here as (94) and (95a-c), respectively: (94) Astronauts working on Mars read Oliver Sacks’s books. (95) a. It is not the case that astronauts working on Mars read Oliver Sacks’s books. b. Do astronauts working on Mars read Oliver Sacks’s books? c. If astronauts working on Mars read Oliver Sacks’s books, they can e-mail him and he will send them copies. We saw above that the affirmative (94), the negative, the interrogative and the conditional in (95a-c) all presuppose that there are astronauts working on Mars (either at present, or in the past and normal futures). But things are different when we consider the “enough” presupposition, i.e.


the assumption that there are “enough,” e.g. more than two or three such astronauts. This presupposition has to be met with the affirmative (94), and the negative (95a), but not with the interrogative (95b) and the conditional (95c). For example, it would be odd to utter (91) and (92a) if Bob and Mike are the only and first astronauts working on Mars (no matter whether they read Oliver Sacks’s books or not). Using a definite description (“the astronauts working on Mars…”) is much more appropriate in such a case. But Oliver Sacks’s literary agent, for example, can felicitously utter (95b) and (95c), while planning the campaign for Sacks’ new books, having no idea how many astronauts are working on Mars. In what follows I will claim that the “enough” implication is, after all, a presupposition of descriptive BP sentences. I will also claim, however, that it cannot, and should not, be confused with the “existence” presupposition, or used to subsume it. There are several reasons for this. Firstly, the two presuppositions behave differently in different contexts, as in the interrogative and conditional (95b) and (95c). Below I will discuss an additional distributional difference between the two presuppositions. In addition to the different distribution of the existence and “enough” presuppositions, I will claim that the two are triggered differently. In section (5.4) above I claimed that conventional and conversational mechanisms combine to trigger the “existence” presupposition. In contrast, I will claim, the “enough” presupposition is purely conversationally triggered. In this sense, the fact that this implication does not survive in the entire “family of sentences” in (95) does not pose a problem for its presuppositional status. I have already mentioned above (in chapter 2 section (6.4.1)) that according to e.g. Kempson (1975) and Wilson (1975), and more recently Chierchia & MecConnell Ginet (1990) and Kadmon (2001) (who follow Stalnaker’s (1974) idea of “pragmatic presuppositions”) conversationally triggered presuppositions are introduced by non-linguistic and pragmatic premises, e.g. by Grician maxims. For these researchers, conversational presuppositions are just like their “hard-core” or conventional counterparts in being taken for granted or “backgrounded,” but unlike the latter they are more easily defeasible (in Chierchia & McConnell Ginet’s terminology), or more strongly context-dependent (in Kadmon’s terminology). The pragmatic triggering and the defeasibility of conversationally triggered presuppositions are, of course, connected. In Chierchia & McConnell Ginet’s words: “Presuppositions are defeasible


if they are triggered by defeasible assumptions about human action and conversation” (p. 313).14 Stalnaker’s (1974) example of conversational presuppositions in (96)-(98) illustrates this defeasibility. Stalnaker observes that although sentences with factive predicates usually presuppose the truth of their sentential complements, in some cases these presuppositions can be canceled. For example, although (96b) is a clear presupposition of (96a), as indicated by its survival in the “family of sentences” (97a-c), it is not necessarily presupposed by (98), with the first-person subject: (96) a. John discovered that Bill is a spy. (presupposition: Bill is a spy) b. Presupposition: Bill is a spy. (97) a. It is not the case that John discovered that Bill is a spy. b. Is it the case that John discovered that Bill is a spy? c. If John discovers that Bill is a spy, there will be trouble. (98) If I discover that John is a spy, I will never invite him over again. Stalnaker takes these phenomena to suggest that presuppositions involve pragmatic and contextual factors, and are in this sense “pragmatic” phenomena. Chierchia & McConnell Ginet take this line further and suggest that the failure of (98) to presuppose (96b) indicates that in (96a), (96b) is triggered by Grician maxims of cooperation. Similarly, Kadmon claims being taken for granted, or “backgrounded” is the main characterizing property of presuppositions. We can now turn back to the presuppositional status of the “enough” implication. Above we pointed out three observations relevant for this implication: (A) The “enough” implication is a condition on the felicity of affirmative descriptive BP sentences, which is felt to be “backgrounded” by these sentences.

14

Formally, Chierchia & McConell-Ginet (1990) give a dynamic semantic definition which can systematically derive conversationally triggered presuppositions, and which is seen in (i):


(B) The “enough” implication does not necessarily survive in the entire “family of sentences” test. (C) There is no element in the semantic structure of descriptive BP sentences which triggers the “enough” implication, i.e. it is not triggered conventionally.

Now, if we assume that the “enough” implication is a conversationally triggered presupposition, these three facts can be accounted for, and the failure of the “enough” implication in the “family of sentences” test can be seen as another case of “defeasibility” or “stronger context dependence” which exists with such presuppositions. 6.3 The temporal asymmetry and variability of the “enough” presupposition What is, then, the conversational mechanism which triggers the “enough” presupposition? Two additional observations can show us the preferred direction to develop. Firstly, if we again compare the “existence” and “enough” presuppositions, we see that in fact the two are required to hold in different intervals. The former is temporally “symmetrical.” As discussed above, the presupposition that there are existing P individuals and relevant situations is relevant for both “edges,” i.e. both the initial and final subintervals of I′, (a (proper) superinterval of the present interval I). Things are different with the “enough” presupposition. Consider, for example, (99) and the two schematic scenarios in (100a) and (100b):

(i) S presupposes p relative to a premise set P iff every context c s.t. P is also s.t. c* entails p″. (p. 311) c* The variable c* in (i) is the context change potential of S. P is “some set of premises about human actions in general and conversation in particular […] P might include, for example, that speakers generally adhere to Grician maxims” (p. 311).


(99)

Students participating in this project have brown hair

(100) a.

b. As before, the horizontal line in (100a, b) represents the course of time, and the vertical rectangle represents the present interval. The short vertical lines represent the students participating in this project. In both scenarios there are students existing in the interval surrounding the present interval, i.e. both in the past and in the future (so the existence presupposition is met in both cases). Nonetheless, there is a difference in the felicity of (99) in the two scenarios. Sentence (99) is odd in a scenario, represented as in (100a), where in the past there have only been two students who participated in this project, even if there will be many more in the future. It is felicitous, however, in the scenario in (l00b), where there have already been twenty students participating in this project, even if there will only be two in the future (or, alternatively, where the project has just started and there are twenty students participating now, and only two will participate in the future). Unlike the “existence” presupposition, then, the “enough” presupposition is temporally asymmetric, and does not care about what happens after the present (i.e. in the future). Its main concern is that there will be “enough” relevant instances in the past, or in the present of our actual world. In addition, the meaning of “enough” in the presupposition is inherently vague. In fact, there are cases of descriptive BP sentences which are felicitous even if there are very few existing P members, as in (101): (101) Horses owned by Mr. Smith have two tails. (where Mr. Smith has owned four horses up to now) The felicity of (101) on this scenario seems to constitute a counterexample to the basic observation about the “enough” presupposition. We can still save this presupposition by claiming that for the sentence in (101) three or four existing P individuals is “enough.”


That is, we can claim that the requirement phrased in (92) above still holds, but that the meaning of “enough,” i.e. of how many P individuals or relevant situations are considered “enough,” varies from case to case. I think this kind of solution is legitimate and true. The meaning of “enough” indeed varies from sentence to sentence. Understanding this variation can now lead us to understanding the trigger of the “enough” presupposition and its role in the truth conditions. The question we have to answer now is what determines this variation: What is the factor w.r.t. which the meaning of “enough Ps” varies? Compare (86) above (Students registering for this course have brown hair) with (101) (Horses owned by Mr. Smith have two tails). In (86) three students is not enough. If there were 120 students that would be enough. “Enough students” for what? Enough students to conclude that the generalization is really non accidental, i.e. not only limited to the actual world. Intuitively, having brown hair is such a common property that the fact that the three students in this course have brown hair may be simply accidental, i.e. restricted to these actual students. But 120 students with brown hair is already enough to suggest that “there is a pattern here” or that “this is nonaccidental,” i.e. that probably other, non-actual students in this course will have brown hair as well. In contrast to “having brown hair,” “having two tails” is a very rare property, i.e. its ratio in the population (of horses, at least) is usually very low. If all four horses that Mr. Smith has owned have this property then there is already a possibility that “this is not accidental.” Perhaps Mr. Smith is a zoologist who is interested in mutations of horses. Or maybe he believes that horses with two tails run much faster than horses with only one tail. A speaker of this sentence (under its descriptive reading) is not committed to have in mind why this is the case, but based on what is observed there is a possibility that “this is not accidental,” i.e. there is some pattern here. Finally, note that in a context where the ratio of having brown hair is very low in the relevant population, sentence (86) can be felicitous. For example, if this is a Nigerian university, where almost all students are Nigerian, and where normally Nigerian individuals have black hair, then if the three students in his course all have brown hair, (101a) is much better. What is relevant here, then, is the ratio of individuals with Qs, in the general, or contextually restricted population of Ps: if this ratio is generally very low then the BP sentence can be true even if there are very few existing Ps which are Qs. What these intuitions suggest is that the factor w.r.t. which the meaning of “enough” varies is simply whether the number of P


individuals and relevant situations involving them is enough for the average language user to conclude that the generalization in the actual world is indeed non-accidental, i.e. not limited to the actual set of circumstances alone. In other words, the meaning of “enough” varies depending on whether it is an appropriate basis for concluding that the generalization in the actual world is non-accidental. Intuitively, the problem with (86) above is that three students is not an appropriate basis for concluding that “this is not accidental.” Similarly, three greencovered books about genericity are clearly not an appropriate basis for concluding (87), i.e. for concluding that the generalization that “all nonexceptional green-covered books about genericity contain no typos” is non-accidental. 6.4 Inductive inferences and descriptive generalizations The “enough” implication, with its focus on relevant instances in the past or present, and with the variability of “enough,” seems strongly connected with the process of inductive inference. In such a process an individual observes a sample of P individuals, which are Qs (Ps Qs), in the past, and/or in the present, which is big enough to conclude “this is nonaccidental,” i.e. to conclude that we should expect to see this generalization elsewhere, in other circumstances. But crucially, descriptive generalizations in general and descriptive BP sentences in particular do not express inductive inferences. If descriptive BP sentences were expressing inductive inferences, a sentence like “Brown-haired librarians are quiet” would mean something like “The number of observed brown-haired librarians in the actual circumstance who are quiet is big enough to determine that ‘every non-exceptional brown-haired librarian is quiet’ is true as well in other, similar enough circumstances.” But this is not what this sentence means. Again, it means, roughly that “every non-exceptional brown-haired librarian is quiet” is true in our circumstance and in close enough circumstances. The whole part of “the sample is big enough” belongs not to the meaning of the sentence, but to the way to determine its truth. This is similar to the fact that observing that John is 1.90 meters tall, or that he bumps his head on the ceiling, does not belong to the meaning of “John is tall,” but to the way to determine its truth.


In this sense, both BP sentences like (102a) and sentences like (102b) are different from nongeneric modal sentences like (102c). (102)

a. Brown-haired librarians are pleasant. b. John is tall. c. John must be in Tel-Aviv now.

Unlike (102a) and (102b), in (102c) the evidence for the truth of “John is in Tel-Aviv now” is part of the meaning of the sentence (see Kratzer (1981)). A good indication for this difference is the meaning of (103ac), where sentences (102a-c) are embedded under the verb hope: (103)

a. I hope that brown-haired librarians are pleasant. b. I hope that John is tall. c. I hope that John must be in Tel-Aviv now.

Sentence (103b) does not mean that “I hope that I observe that John is taller than 1.80 meters,” but “I hope that John is a member of that set/ group of tall individuals.” In contrast, sentence (103c) means “I hope that it follows from the facts/John’s instructions, etc. that John is in TelAviv now,” i.e. the basis for the conclusion that “John is in Tel-Aviv” is an integral part of the meaning of the sentence. Despite the modal nature of (102a), in this respect it is like (102b) and unlike (102c). Sentence (103a) means roughly “I hope that the generalization ‘All non-exceptional/most brown-haired librarians are pleasant” is true both in our actual world and in similar circumstances” and not “I hope that I (will) observe a sample of brown-haired librarians who are pleasant, which is large enough to determine that this is not accidental.” Thus, the inductive inference is not an integral part of descriptive BP sentences. The complement position of hope shows this clearly, since hope is nonfactive, i.e. it does not entail the truth of its sentential complement. The claim that the inductive inference process is the basis for arriving at the truth of descriptive BP sentences and not an integral part of its meaning predicts that in cases where the truth of the descriptive BP sentence is not relevant, as with nonfactive predicates, the inductive inference will not show up. This prediction is indeed borne out.


6.5 Back to the “enough” presupposition We can now turn back to the “enough” presupposition and clarify its relationship with inductive inferences and with the meaning of descriptive BP sentences. Note first, that in a context where the inductive process does not show up, as with (103a), where the BP sentence is embedded under hope, the “enough” implication does not show up either. A speaker of (103a) does not necessarily take it for granted that there are “enough” brown-haired librarians in the past and/or present, and we, as listeners, will take this sentence to be felicitous even if there are or have been very few such librarians. (In this sense the “enough” presupposition is, again, distributionally different from the “existence” presupposition, which survives even in (103a). Sentence (103a) is odd if no brown-haired librarians exist now, and there haven’t been any such librarians in the past.) The close connection between the process of inductive inference, and the “enough” implication can be now explained as follows: The inductive inference process, in which we observe a large enough number of Ps Qs, is not an integral part of the meaning of descriptive BP sentences, but a basis for concluding the truth of these sentences. This is similar to the fact that observing that John is taller than 1.80 is not an integral part of John is tall, but a basis for concluding the truth of this sentence. But this similarity is not complete. There are many possible kinds of situations which can function as a basis for concluding that John is tall: observing his height, his clothes, the fact that he bumps his head on the ceiling, the length of his legs, etc. Thus, there are many possible situations in which uttering this sentence is appropriate. In contrast, the inductive inference process seems to be much more strongly connected to the meaning of descriptive generalizations. If we have no “real world knowledge about some “in virtue of” factor which might lead to the ^Q property, then the only way we can conclude that a certain generalization constitutes a pattern, and is non-accidental is through observing enough Ps having ^Q, i.e. through inductive inference. In other words, inductive inference seems to be the most natural, if not the only, kind of basis for concluding merely that “this is not accidental.” Thus, in a context where it is clear that the speaker could not have possibly observed a large enough number of Ps Qs, it would be odd for him/her to claim that “this is not accidental.”


The clearest situation where this is the case is when it is clear that there are simply not enough Ps in the world, i.e. there are very few of them. This is so since if there are not enough Ps, then clearly the speaker could not have observed enough Ps Qs. For example, if there are very few students in this course, then the speaker of Students in this course have brown hair could not have observed a large enough number of students in this course with brown hair to conclude that “this is nonaccidental.” In other words, s/he could not have the necessary basis for making the assertion. This seems to be the source of the infelicity of (86)-(91) above. These intuitions can be phrased more precisely like this: If observing a large enough number of Ps Qs is the necessary appropriate basis for making a descriptive generalization, then when it is clear that there are not enough Ps in the actual world, the speaker of the descriptive BP sentence necessarily violates Grice’s (1975) principle of cooperation. Specifically, such a speaker violates the second part of the Grician maxim of Quality, quoted in (104): (104) Grice’s Maxim of Quality: Say only what you believe is true and adequately supported. (my emphasis) The sentences in (86)-(91) above are odd, then, because of a necessary violation of this maxim. The claim that the process of inductive inference constitutes the basis for the assertion expressed by descriptive BP sentences, and not an integral part of their meaning, and the connection of this process with the “enough” presupposition, can now explain the fact noted above, namely that this presupposition survives with affirmative and negative sentences (as in (95a, b) above), but not with interrogative and conditional ones (as in (95c, d)), and not with nonfactive verbs, (as in (103a)). Unlike what happens in the former two constructions, in the latter three the truth of the descriptive BP sentence is not relevant, and thus neither is the process of inductive inference. Thus, a speaker of sentences like (95a), (95b) and (103a) is not implying that s/he is making his assertion on the basis of inductive inference, and thus, in asserting such sentences s/he does not imply that there were “enough” P individuals and relevant situations involving them to conclude that the assertion is true, i.e. that the universal statement is indeed nonaccidental. Having in mind the observations and discussions in the above sections, we can now phrase the “enough” presupposition as in (105):


(105) “Enough” presupposition (triggered by Grice’s maxim of quality): There are enough P individuals and relevant situations involving them in <w, I’initial> in the past, and/or present in the actual world) for it to be reasonable (for the average language user) to conclude that the universal statement is non-accidental. 6.6 An apparent counterexample: more cases where “enough” is very few Consider the sentences in (106a, b): (106)

a. Compaq computers model Aero 4/25 work quickly. b. Koko spiders have eight eyes.

Imagine that I am working in a computer shop, and a client is asking me about the properties of the new model of laptop computers manufactured by Compaq. Now, I saw only one or two such computers today, and they both worked quickly. In fact, these are the first two computers of that model in existence. On the basis of these one or two observations, I can utter (106a) felicitously. Similarly, if walking in the jungle I observe three spiders of a new species, which I label “Koko,” and all three have eight eyes, I can felicitously utter (106b). In both cases, although the ^Q properties expressed by the VPs are clearly not considered rare (as in (101) above), the inductive inference process seems to be successful with very few observations. Why is this? Intuitively the two scenarios above are examples of cases where “if you’ve seen one, you’ve seen them all,” which are very common and appropriate with artifacts and biological species. That is, if you have seen one (or two) Compaq computers or Koko spiders, you do not need to observe more in order to conclude that all of them will behave the same, since you assume that they are all structured in the same way. Note, however, that the “if you’ve seen one, you’ve seen them all” scenario is not appropriate for all BP sentences with subjects denoting artifacts and biological species. Compare the sentences in (106) to their minimally contrasting pairs in (107): (107)

a. Compaq computers model Aero 4/25 are dirty. b. Koko spiders have a big scratch on one of their left legs.


These sentences are odd in a situation where I observed only one or two such computers which are dirty, or only one or two Koko spiders which have a big scratch on their leg. W.r.t. the properties in (107a) (107b), then, we cannot say “if you’ve seen one, you’ve seen them all.” In order for such a scenario to be appropriate, then, we need a certain combination, i.e. we need both the ^P property to correspond to an artifact or biological kind/species, and the ^Q property to be part of the design or structure which the member of this artifact or biological species are known to share. Our real-world knowledge tells us that being dirty, or having a scratch are not part of the “design” of computers and spiders, respectively. Therefore, for (107a) and (107b) to be true as a descriptive generalization, i.e. in order to conclude that “this is non-accidental,” one observation, or even two, will not be not enough. What this suggests is that in the “if you’ve seen one you’ve seen them all” scenario, the prominent reading of the BP sentences in (107a) and (107b) is not a descriptive reading, but rather, an “in virtue of” one, which is based on, or triggered by, partial inductive evidence. In both cases, the prediction that the next Ps will be Qs is not solely based on the observation of actual facts, and on pure inductive inference. Instead, the prediction (“this is non-accidental”) is based on a combination of the single observation together with the real world knowledge about the ^P and ^Q properties. The felicity of (106a, b) above, then, does not undermine the “enough” presupposition of descriptive BP sentence. The more general lesson to learn from these cases is that although inductive inference processes are the classic basis for descriptive generalizations, there are cases where they can also serve as a partial basis for “in virtue of” generalizations. Thus, the crucial difference between the “descriptive” and “in virtue of” generalizations is not in the basis for the assertion, but in the kind of assertion made (roughly between “‘Every P is Q’ is not accidental” vs. “in virtue of having S, every P has Q”). 7. Conclusion and summary Together with the “enough” and the “existence” presuppositions, then, the truth conditions of descriptive BP sentences look now as in (108): (108) A simple present descriptive BP sentence is true in <w, I> iff


Existence Presupposition

“Enough” presupposition There are enough P individuals and relevant situations involving them in <w, I′ initial > (i.e. in the past, and/or present in the actual world) for it to be reasonable for the average language user to conclude that the universal statement is non-accidental. The truth conditions in (108) capture the basic and original intuition about descriptive BP sentences, according to which such sentences assert of a certain generalization (universal statement) that “there is a pattern here” or that “this is not accidental,” i.e. that the truth of the universal statement is not limited to the actual set of circumstances (our world, now) but is expected to hold in other hypothesized sets of circumstances, similar to the actual one, where crucially, and unlike what happens with IS sentences, no specific “in virtue of” property is implied to be involved. This intuition is captured in the truth conditions by taking the accessibility relation to be realistic, i.e. the universal statement is explicitly required to be true in actuality, since the actual world (w) is a value to the accessibility function, and not only an input to it. In addition, the accessibility function produces as values other possible worlds, all of them are similar to the actual world at the present in a rather vague and “wide perspective” way: we get those worlds which are maximally similar to ours (overall) (except from what is needed to allow for the fact that we may have non-actual members of the P set and the set of individuals involving them), and those worlds in which “things take their normal course” w.r.t. the here and now. These characteristics of the “descriptive” accessibility relation account for several basic properties of descriptive BP sentences, namely their felicity when expressing highly “unreasonable” generalizations, their uniform flavor, and their low degree of “law-likeness” or analyticity.


In addition to the accessibility relation, another central component in the truth conditions above is the time interval I′—defined as some (proper) superinterval of the present interval I. I have shown that the underspecification of this superinterval helps to account nicely for a wide variety of cases where descriptive BP sentences are true and false. I showed how the assertion that the universal statement has to hold at some (proper) superinterval of I—I′ in all accessible worlds, including the actual one, together with the independentlyargued-for Grician maxims of relevance, naturally trigger the “existence” presupposition” of descriptive BP sentences (requiring existing P individuals and relevant situations involving them at least at the “edges” of the surrounding interval). Here too, the underspecification of I′ correctly accounts for a variety of cases where such sentences are felicitous and infelicitous. The fact that we find similar “existence” presuppositions with epistemic IS sentences led me to add such temporal components to the truth conditions of IS sentences as well. I have also shown that several differences concerning the felicity/infelicity of IS and descriptive BP sentences can be naturally derived from the independently motivated difference in their accessibility relations. Thus far, then, BP and IS sentences have exactly the same semantic structure, and the difference in accessibility relations between them can account for the variety of semantic and presuppositional differences between them. Finally, the truth conditions in (108) clarify the relationship between inductive inference and descriptive generalization, achieved through the “enough” presupposition. Adding this conversationally triggered presupposition accounts for the intuition that although descriptive generalizations do not express inductive inference processes, such processes are strongly associated with descriptive generalizations, and form, in fact, their natural basis. Thus, the possibility to use inductive reasoning as the basis for the assertion is presupposed by (speakers and listeners of) these sentences.

202

CHAPTER 4 Tolerance of Exceptional and Irrelevant Entities with IS and Descriptive BP Sentences

1. Introduction This chapter develops a mechanism for tolerating exceptions for both IS and descriptive BP sentences, based on Kadmon & Landman’s (1993) suggestion for a domain vague restriction on the generic quantifier. This suggestion has some important advantages over another popular solution for the exceptions problem, namely the “normal ordering source solution” suggested by e.g. Krifka (1995) and Krifka et al. (1995). Aside from developing this basic mechanism, in this chapter I also discuss a newly observed difference between IS and descriptive BP sentences, concerning the degree to which the properties of the legitimate exceptions can be characterized in advance. I show that this difference results from the different kinds of generalizations expressed by the two kinds of sentences, and more precisely, from the two different accessibility relations and accompanying presuppositions of the two kinds of sentences (essentially, the presence or absence of an “in virtue of” property restricting the accessibility relations, and the presuppositions on the choice of this property). To capture this difference, I impose on the basic mechanism for tolerating exceptions (namely the vague restriction on Gen), two further different pragmatic requirements which are systematically influenced by the two different accessibility relations and presuppositions associated with the two kinds of sentences. The chapter is structured as follows: I discuss the basic data concerning the similarities in the way IS and descriptive BP sentences tolerate exceptions in section (2). In section (3) I show how these similarities can be naturally accounted for by adopting a development of Kadmon & Landman’s (1993) “domain vague”


analysis. In that section I also point out the advantages of this mechanism over the “normal ordering source” solution, as well as two shortcomings of this mechanism, concerning a difference between IS and descriptive BP sentences in the degree to which the properties of the exceptions can be specified. In section (4) I claim that this latter difference is systematically related to the difference in accessibility relations between IS and descriptive sentences, motivated in chapters 1 and 2 above. Sections (5) and (6) are devoted to supplementing the equivalent domain vague restriction on the generic quantifier with two different pragmatic restrictions, which are systematically influenced by the difference in accessibility relations. 2. The basic data: similarities in the way IS and descriptive BP sentences tolerate exceptional and irrelevant entities There are several similarities between IS and descriptive BP sentences, e.g. (la) and (1b), respectively, concerning the tolerance of exceptions. (1) a. A Pentium III computer works very quickly. b. Brown-haired postmen whose names end with ‘g’ or ‘t’ drive wildly. c. The first similarity, noted by numerous writers, is the basic tolerance of exceptional individuals, e.g. the fact that the existence of several Pentium III computers which don’t work very quickly, or several brown-haired postmen whose name ends with ‘g’ or ‘t’ who do not drive wildly, is not enough to falsify (1a) and (1b), respectively. Second, with both IS and descriptive BP sentences, such exceptional individuals are taken to be “abnormal” in some sense (see e.g. Asher & Morreau (1995), Pelletier & Asher 1997, Eckardt (1999)). I take “abnormal in some sense” to mean that an exception to a generic sentence is assumed to be exceptional in some other sense as well, besides not having the Q property, e.g. we will assume that those slow Pentium III computers have some other exceptional property (perhaps they have been damaged by a virus, or broken). Similarly, assuming (1b) to be true, we will take those postmen who do not drive wildly to be legitimate exceptions to this sentence to the extent we can assume that they are “abnormal” or “exceptional in some other sense.”


A popular way to capture this “abnormality” of the exceptions, suggested, e.g. in Krifka et al. (1995), is to restrict the set of accessible worlds where the generalization is evaluated using a “normal” or “stereotypical” ordering source (in the sense of Kratzer (1981)), which yields only those accessible worlds which are “most normal,” or where “nothing unexpected happens” from the point of view of the actual world. Since, according to this approach, the actual world itself, with all its accidents, mutations, viruses, etc., is not considered “most normal,” the existence of e.g. actual damaged Pentium computers which work slowly does not falsify the generalization. However, the use of the undefined and vague notion of “normality” in the truth conditions of generics leads to unjustified circularities and wrong predictions. One problem, which I discussed in chapter 2, section (6), is that it is not clear how we can treat mutations, accidents, viruses, etc. as abnormal, since a world completely free of such phenomena will be considered truly abnormal, from the point of view of our world. A similar problem, noted by Cohen (1999), lies in the fact that, using a “normal” ordering source restriction on the accessible worlds, IS sentences like A bird flies or A cat has a tail are true, roughly, if in all accessible worlds, where things take their normal course of events w.r.t. the actual worlds, every bird flies and every cat has a tail. But part of the “normal course of events in w” is for birds to fly and for cats to have a tail. These two properties seem to be part of the definition of normality in our world, i.e. a world where these two species have different properties will be considered an abnormal world. Thus, using the “normal” ordering source A bird flies turns out to express the circular statement: “every bird flies in all most normal worlds, where among other things, birds fly.” Below I discuss another kind of problem with the “normal ordering source” suggestion.1

1

It would be tempting, at this point, to try to use the finer-grained ordering source I developed for the second presupposition of IS sentences (see chapter 2 section (6.3.3)), namely relativizing “normality” to a single general process, instead of trying to construct overall normality. This solution clearly avoids the first problem with the “normal” ordering source, and probably the second one as well. However, using this finer-grained ordering source will still not help us solve the problem discussed below in this section, just like using the “


A third similarity between IS and descriptive BP sentences is that both can tolerate not only exceptional, but also contextually irrelevant individuals. This observation, in fact, undermines the widely held claim (see e.g. Dahl (1975), Krifka (1987), Krifka et al. (1995), Condoravdi (1997)) that unlike subjects of explicitly quantified statements, subjects of generic sentences cannot be understood as contextually restricted. Krifka’s (1997) original data, which is supposed to support this view is found in (2): (2) (Context: There are lions and tigers in the cage) a. Every/some lions has a mane. /Some/most/the lions have a mane b. A lion has a mane/Lions have a mane Krifka (1987) notes that unlike the subjects in (2a), which can be understood as every/some etc. lions in this cage, the subjects in (2b) do not relate to lions in the cage only, but generalize over lions in general. However, there are many other cases where IS or BP generics are easily contextually restricted. For example, in the contexts mentioned in (3a-c), the IS and BP generic sentences are understood as generalizing over professors in this university, books and periodicals in this library and shirts in this shop, respectively: (3) a. (Context: There are professors and students in this university.) A professor wears a tie/Professors were a tie.2 b. (Context: There are books and periodicals in this library.) Books/a book can be borrowed for a week, but periodicals/a periodical can only borrowed for one day. c. (Context: You should go to Jack’s shop. There are beautiful shirts and skirts there.) A shirt cost less then NIS 100/Shirts cost less than NIS 100

normal” ordering source will not (since both turn the accessibility relation of the generics into nonrealistic). See also section (3.3) below. See also Eckardt (1999), who observes more problems in the attempt to account for the tolerance o exceptions by restricting the set of worlds. 2 Thanks to Fred Landman (p.c.) for pointing out this example.


These sentences, then, tolerate not only exceptional, but also contextually irrelevant individuals, (e.g. professors from other universities, books in other libraries, and shirts in other shops). Fourth, both IS and descriptive BP sentences tolerate not only the existence of exceptional and irrelevant individuals, but also the existence of irrelevant and exceptional situations. In the case of (1a), for example, an irrelevant situation is one where a relevant and perfectly normal Pentium computer is turned off, and a legitimately exceptional situation is a one where Pentium computers (which are turned on) do not work quickly since they are in the outer space, or in a place where it is exceptionally hot, or cold, etc. It is important to keep the exceptional individuals and exceptional situations apart: we do not want to treat a computer which is undamaged and virus-free as an “abnormal” computer simply because it is temporarily located in very hot place and thus does not operate properly. Instead, we will treat the computer as normal and non-exceptional, but the situation this computer is in as abnormal and exceptional. Note also that when the VP denotes an individual-level property, neither kind of sentence gives us irrelevant or exceptional situations (only exceptional and irrelevant individuals). For example, unlike what happens with e.g. A dog barks or Dogs bark, where we allow for many situations where (normal, relevant) dogs do not bark (e.g. sleeping situations, situations where the dog is exhausted, etc.), with individual level predicates, as in A dog has four legs/ Dogs have four legs we seem to require that in all situations containing a (normal and relevant) dog, such a dog will have four legs. Fifth, with both IS and descriptive BP sentences considering irrelevant, but not exceptional, entities as counterexamples to the generalization is interpreted as a misunderstanding on the part of the listener. For example, if, hearing (2b), some listener points at a professor from some other university, who does not wear a tie, or at a situation where a professor from this university is taking a shower, and tells the speaker of this sentence “these are counterexamples to your generalization,” the listener’s reaction will be considered inappropriate, or odd. We will conclude that this listener did not understand what the speaker was trying to say. The speaker only talked about professors in this university, and about relevant situations, where people wear things, so non-tie-wearing professors from another university or showering situations are simply irrelevant for the generalization. On the other hand, such a reaction will be considered perfectly appropriate if the listener points out a non-tie-wearing professor in this university, or to situations


where a (relevant) professor walks on campus. In such a case we will take the listener to understand the speaker, but to disagree with him. Finally, both descriptive BP and epistemic IS sentences make predictions about actual individuals and situations. For example, assuming A Pentium computer works quickly to be true, I will expect my neighbor’s Pentium computer, which I assume to be normal (no viruses, no damage, etc.) to work quickly, and also this actual situation where it is turned on—which I take to be a normal situation—to be a situation where it works quickly. We get similar predictions with descriptive BP sentences. Note that such predictions concerning entities in the actual word cannot be accounted for if we apply the “normal ordering source” solution to the problem of exceptions (nor the finergrained version of this solution, developed in chapter 2, section (6.3.3) above) since such a move turns the accessibility relation of Descriptive BP and epistemic IS sentences into nonrealistic (in the sense of Kratzer (1981)), i.e. as not having the actual world (with its actual entities) as a member. To summarize, both IS and descriptive BP sentences can tolerate the existence of four kinds of non-Q entities (illustrated in (4) below): irrelevant and exceptional individuals, and in the case of stage-level predicates, exceptional and irrelevant situations as well. With both kinds of sentences, exceptional entities, but not irrelevant ones, are perceived as “abnormal” in some sense. In both kinds of sentences, considering irrelevant but not exceptional entities as counterexamples to the generalization is interpreted as misunderstanding the speaker. Finally, both descriptive BP and epistemic IS sentences make predictions into the actual world, i.e. concerning those actual individuals and situations are taken to be relevant and non-exceptional. (4) (While talking about a certain school) A first grader finishes school at 13:00. Irrelevant individuals: first graders from other schools Irrelevant situations: days which are not school days Exceptional individuals: first graders (from this school) who finish school before 13:00 because e.g. they are part of a regular program for gifted children which starts every day at 12:30, or because they live far away and have to get out early to make a carpool, or regularly finish school at 2:30 because they participate in an in-school enrichment program, etc. Exceptional situations: extremely stormy (school) days, with heavy snow, etc. (where the principal decides to let the first


graders go home earlier), a day where the mayor comes to visit this school (so all the first graders finish school earlier or later than 13:00), etc. 3. Capturing the similarities: a Kadmon & Landman-style “domain vague” restriction on the generic quantifier 3.1 Kadmon and Landman’s 1993 theory of generics In their 1993 paper, Kadmon & Landman (henceforth, K&L) take the generic operator Gen to be a universal modalized quantifier, combining with the common noun to create a (nominal) generalized quantifier. The claim in K&L which is most relevant for our purposes is as follows: The reason IS sentences (like An owl hunts mice) tolerate exceptions, whereas standard universal sentences (like Every owl hunts mice) do not is due to the fact that although both kinds of quantification are contextually restricted, the restriction on the generic quantifier is inherently vague. To quote: Regular universal statements are restricted by context in a way that is supposed to be precise (even though in a particular context it may happen not to be clear what the precise restriction is). When interpreting such a statement we are entitled to assume that there is a well-defined set of objects that the universal quantification ranges over, and if we don’t know what the speaker intends that set to be, we tend to accommodate such a set (in the sense of Lewis, 1983). An object in that set can never be a legitimate exception to the generalization made. The inherent vagueness of generic statements explains why they are different regarding exceptions. For a generic statement there is no welldefined set of objects that the universal statement ranges over. We don’t expect the context of utterance to make clear what the objects are exactly that the generalization expressed applies to. And we don’t attempt to accommodate a precise set of objects. Hence, when we encounter objects that do not fall under the generalization expressed, there is always the possibility that they are not among the objects that the generalization is supposed to


apply to, and we are therefore able to regard them as legitimate exceptions” (p. 409). […]What we would like to propose, then, is that it is an integral part of the nature of generic statements that the restricting set of properties is vague…Saying An owl hunts mice is just like saying ‘every (possible) owl with the right properties hunts mice,’ while, crucially not committing yourself to what the right properties are […] In short, we claim then that (this sentence) means something roughly like ‘All normal (possible) owls hunt mice,’ where what counts as normal is inherently vague. (p. 408) [emphasis in the original] Formally, K&L claim that there is a vague set of properties restricting the generic quantifier, which interacts with the set denoted by the common noun (i.e. the first argument of the GQ), to produce a (vague) subset of this set (e.g. a (vague) subset of the set of owls, in the case of An owl). This vague set of properties restricting the generic GQ consists of a pair . The first set of properties in this pair, v0, is a precise set of properties (which is consistent, and the properties in it are compatible with the property OWL), i.e. in a given context, it contains the properties which are already known to be part of the restriction. Following the supervaluations approach to vagueness (developed, e.g. in Fine (1975), Kamp (1975) and Landman (1991)), the second set of properties in the pair, V, is a set of precisifications of v0, i.e. ways in which vagueness can be eliminated. This is a set of consistent sets of properties, such that each set of properties v V is (a) a superset of v0, and (b) contains properties which are compatible with the property OWL. This set V of precisifications “represents all the different ways of making the restriction completely precise, which are compatible with what is already known about the restriction in the context” (p. 411). A restriction is domain precise if the set of objects determined by each of the precisifications in V is the same as the set of objects determined by the properties in the precise part v0, i.e. if the precisifications add, in fact, no properties, to the ones in the precise part, which are known from the context. The restriction is domain vague, on the other hand, if we do not have this equivalence, i.e. if the restriction contains more properties than those in v0, which are in the precisifications. Formally, if B is the set of objects denoted by the common noun (e.g. the set of owls), and Dv ,B is a domain of objects (a subset of the set of objects denoted by the common NP), then the following holds:


(5) is domain precise on B iff for every v V, Dv,B=DV0, B is domain vague on B iff is not domain precise on B. (p. 411) Now “every” is considered a “domain precise” quantifier since in every context c its restriction is domain precise in the sense of (5) (i.e. all restricting properties are supposed to be supplied by the context or speaker, and accommodated by the listener). In contrast, Gen is a domain vague quantifier, since there is at least some context c where the denotation of the restricting set of properties is vague on the denotation of the CN, i.e. where “there are distinct domains determined by the precisifications compatible with what is known about the restriction in c, and there isn’t a unique precise set that the quantification ranges over” (p. 411–12). 3.2 Applying K&L’s suggestion to the truth conditions of IS and descriptive BP sentences 3.2.1 A vague restriction on the set of individuals K&L take Gen to be a nominal generalized quantifier, and thus in their original suggestion the mechanism for tolerating exceptions is a N, which yields a (vague) restriction on the generalized quantifier subset of the set of individuals denoted by the common noun. Throughout this work, however, I follow the works of Wilkinson (1991), Krifka et al. (1995), Chierchia (1995, 1998), etc., in taking Gen to be a sentential operator, binding both d variables introduced by the common noun, and s variables introduced by the VP. Ignoring the difference in accessibility relations, the truth conditions of both IS and descriptive BP sentences, developed in chapters 2 and 3 above, appear now in (6): (6) Basic truth conditions for IS and descriptive BP sentences A BP or an IS sentence is true at <w, I> iff


Paraphrase: In all appropriately accessible worlds w′ there is a superinterval of the present I′, s.t. any contextually relevant situation involving any P individual in w′, which is located in I′ is a Q situation involving this individual in w′. Let me show first how we can adopt K&L’s suggestion to account for the fact that generic sentences tolerate irrelevant and exceptional individuals. We will impose K&L’s vague restriction on the P property in the restrictive clause of Gen. Instead of taking the set of individuals in worlds which is quantified over by Gen to be given by P, we will take it to be given by a vague subset of P, by imposing a vague restriction on the original P relation. Instead of writing P(d, w′), then, as in (6), we will write PXp(d, w′), where Xp denotes a vague set of properties, which is systematically related by the P property (denoted by the common noun), as defined in the previous section. In the case of e.g. First graders finish school at 13:00, we will not universally quantify over all individuals in the set of first graders, but rather over individuals in a vague subset of this set, represented as first graderX FIRST GRADER. The vague set of properties Xp consists of the pair , where v0 is a (possibly empty) consistent set of properties, and the properties in it are compatible with the denotation of P. In the case of our example about first graders, for example, v0 has in it the property “studying in this school” (so first graders from other schools are considered irrelevant). In other cases there may be no contextual restriction, so the set v0 is empty (so, with e.g. Dogs have a tail, all dogs are considered relevant). Unlike v0, V- the vague part of the restriction is a set of sets of properties s.t. each property v V is (a) a superset of v0 (e.g. a superset of the set of relevant properties of the first graders), and (b) contains properties which are compatible with the property P (e.g. compatible with being a first grader, e.g. not being a second grader). (Note that the elements in will have to be sets of relations between individuals and worlds (i.e. properties), and not just sets of individuals, as in K&L’s original suggestion). We now take xp to always be domain vague on the denotation of P. In the case of e.g. A first grader finishes school at 13:00 (as well as the minimally contrasting BP First graders finish school at 1:.00), the restriction X FIRSTGRADER is domain vague on the denotation of “first grader” in the sense that there is at least one context where the set of individuals who have the properties in (the possibly empty) v0, e.g. the


set of individuals studying in this specific school is not equivalent to the set of individuals who have the properties in V. In other words, there is at least one context where, to use K&L’s words “there are distinct domains (of individuals) determined by the precisifications […] and there isn’t a unique precise set that the quantification ranged over” (p. 411–12). Thus, the property λdλw.first grader X FIRST GRADER(d, w) is interpreted as a vague subset of first graders in a world, i.e. the subset of first graders who have both the properties in v0 (directly supplied by the context, e.g. the property of studying in this school), as well as all the properties in one of the unchosen precisifications of v0 in V. Such a representation now captures the intuition that there are indeed two kinds of first graders that do not falsify the generalization: irrelevant ones, e.g. those who do not study in this school (not quantified over due to the precise part in X FIRST GRADER—v0), and exceptional ones, e.g. those who are very gifted and are allowed to leave school before their classmates (not quantified over due to the vague part in X FIRST GRADER-V).3 3.2.2 A vague restriction also on the set of situations We also noted in section (2) above that the way IS and descriptive BP sentences tolerate exceptional and irrelevant individuals is strikingly

3

We can now also understand the difference between generic sentences like A dog has four legs or Dogs have four legs and explicitly universally quantified sentences like (i) or (ii) (cited by Carlson (1989)):

(i) (ii)

Every dog has four legs. Every friend of John’s votes for Socialists.

As Carlson (1989) and Kratzer (1989) show, sentences like (i) and (ii) can express not only an extensional, accidental generalization (universally quantifying over all actual dogs, or actual friends of John), but also “nonaccidental” or “intensional” non-accidental generalizations, universally quantifying over all possible dogs, and all possible friends of John’s, respectively. Even with this “intensional” reading, however, such sentences are different from ordinary generics with IS or BP subjects, in that they do not tolerate exceptions. For example, unlike the IS A dog has four legs or the descriptive BP Dogs


similar to the way such sentences tolerate exceptional and irrelevant situations. To capture this similarity we will impose on the set of situations in the restrictive clause a vague restricting set properties of situations, similar to the vague restricting set of properties of individuals we imposed on the P set of individuals. Note, however, that since the characterization of situations in the restrictive clause of the truth conditions in (6) is, following Chierchia 1995, already defined as the set of situations which are contextually restricted situations (and which involve an individual in a world), given by the relation C(s, d, w′) in the truth conditions, imposing on this set a K&L-style vague restriction will unnecessarily yield a double contextual restriction on situations (since the precise part in a K&L-style vague restriction supplies contextual information as well). To avoid this problem I will change first the Chierchian relation C(s, d, w′) in the restrictive clause of (6) (“contextually relevant situations involving individuals in w′”) to the more neutral relation Involve(s, d, w′) (“situations involving (or containing) individuals in w′”). The two relations are the same, except the latter is “stripped of” the contextual information. This contextual information will now be supplied as part of the K&L-style vague restriction on the Involve

have four legs, the sentence Every dog has four legs, is falsified by the presence of dogs with more or less than four legs, even under its generic, nonaccidental reading. We can account for these facts by claiming that the quantifier of both standard generics and universal sentences like (i) and (ii) (under their intensional reading) is modalized, but that it is domain vague in the former, and not in the latter kind of sentences., I suggest the following schema to capture the differences: (iii)


relation, and will be understood as restricting the set of situations involving (relevant and nonexceptional) P individuals in w′. In defining this restriction on the set of situations we should let the Q property, denoted by the VP, play a systematic role. This is because the properties of the relevant situations are systematically determined by our realworld knowledge about Q.4 For example, the situations which are relevant for evaluating First graders snore quietly and First graders finish school at 13:00 are clearly different: the former are likely to be sleeping situations (thus situations where the first graders are awake are considered irrelevant), whereas the latter are likely to be school-day situations (thus vacation situations are considered irrelevant). Finally, in case Q is individual level (as in A dog has four legs), we take the restriction on the set of situations to be empty, so in fact we quantify over all situations involving a dog individual in w′, and no situation is considered irrelevant or exceptional.5 To capture, then, the systematic role of the Q property in the vague restriction on the set of situations involving an individual in a world, which produces a vague subset of this set, we will take this vague restriction to be keyed to Q. We will replace the relation C(s, d, w′) in the restrictive clause of the quantifier with the relation InvolveYQ(s, d, w′). This relation will produce a vague subset of the set of situations involving a P individual in a world w, namely those situations which are relevant and non-exceptional. YQ is a vague set of properties of situations. We will take it to be a pair of sets of properties of situations, which functions in a similar fashion to the pair , denoted by the restriction on the P property—Xp—defined above. Thus k in YQ is the precise part of the restriction. It contains properties of situations, compatible with the Q property and supplied by context and our realworld knowledge of the Q property (i.e. the property of being a sleeping situation, in the case of First graders snore loudly). Unlike k0, the vague part of YQ—K—is a set

4

In the case of conjoined VPs, it seems that each VP will contribute a different restriction (I thank Greg Carlson for bringing this point to my attention). Cohen (1996) provides an alternative-based algorithm for calculating the properties of irrelevant entities on the basis of the denotation of the VP. 5 Thus, in the case of individual-level predicates we get only irrelevant individuals. In contrast, with stage-level predicates as in First graders finish school at 13:00 we want to distinguish between irrelevant individual (like first graders from other schools) and irrelevant situations (like situations where a relevant first grader (i.e. a one from this school) is on vacation).


of precisifications on k0, i.e. a set of sets of properties of situations. Thus the restriction YQ excludes both situations which are irrelevant w.r.t. the Q property, as well as exceptional situations (e.g. situations in extremely dry nights, situations where the first grader has a cold and can’t breathe properly, etc.). Finally, in instances when Q is individual level (as in A dog has four legs), we take both k0 and K in YQ to be empty, so that we quantify over all situations involving a dog individual in w′, and no situation is considered irrelevant or exceptional. To sum up, instead of the truth conditions in (6) above, we now have (7): (7) Basic truth conditions for IS and descriptive BP sentences (including a mechanism for tolerating exceptional and irrelevant individuals) A BP or an IS sentence is true at <w, I> iff

Paraphrase: In all accessible worlds w′ there is a superinterval of the present I′, s.t. any relevant and non-exceptional situation involving any relevant and non-exceptional P individual in w′, which is located at I′ is a Q situation involving this individual in w′. 3.3 Advantages of the new mechanism for tolerating exceptional and irrelevant entities In addition to successfully excluding irrelevant and exceptional individuals and situations from the domain of universal quantification, the K&L-style restriction applied above to the truth conditions of IS and descriptive BP sentences offers two additional advantages. First, unlike the “normal ordering source” solution,6 using the present tolerating exceptions mechanism allows us to account for the predictions, discussed in section (1) above, made by both descriptive BP and epistemic IS sentences w.r.t. actual individuals and situations. This is because, unlike the former mechanism, the present one is not a

6

As well as the finer-grained ordering source developed in chapter 2 above. See also Eckardt (1999), who also suggests that the exceptions puzzle be solved by restricting the set of individuals, and not the set of worlds.

7


restriction on the set of worlds, but a direct restriction on the sets of P individuals and situations involving them,7 so the realistic accessibility relation of these two kinds of sentences remains unaltered. Thus, instead of interpreting a sentence like “Bears snore loudly” as “In all accessible, most normal worlds (not including the actual one), every relevant situation involving every bear is a snoring-loudly situation,” we interpret it as, roughly, “In all accessible worlds (including the actual one) every relevant and non-exceptional situation involving every relevant and non-exceptional bear is a snoring-loudly situation.” In addition, using the present mechanism captures the intuitive distinction between irrelevant and exceptional entities, namely that considering the former, but not the latter, as counterexamples to the generalization is interpreted as a misunderstanding on the part of the listener (see again section (1) above). The K&L-style vague restrictions capture this distinction through the distinction between the precise sets of properties v0 and k0, which exclude irrelevant entities from the domain of quantification, and the vague sets of precisifications V and K, which exclude exceptional entities. The idea is that the properties in the precise part of the restrictions are those that the speaker is attempting to precisely convey, and which are thus supposed to be accommodated by the listener. Thus, encountering an entity (individual or situation) which lacks one of the properties in v0 or k0, the listener is supposed to understand that such an entity is not universally quantified over in the first place. This is why considering such an entity as a counterexample to the universal quantification is an indication that the listener failed to accommodate what the speaker had in mind (i.e. that s/ he misunderstood the speaker). In contrast, since the exceptional entities are those which lack a property in one of the precisifications, and since no single precisification is chosen by the speaker, no such single precisification is supposed to be accommodated by the listener, which means that there is no precise set of entities which is assumed to be excluded from the domain of quantification. This is why it is legitimate on the part of the listener to suspect that a certain entity falsifies the universal quantification, i.e. that it constitutes a counterexample to the generalization.


3.4 Shortcomings of the present mechanism for tolerating exceptions As it stands, there are two facts that the present mechanism cannot explain. Both problems result from the fact that the vague parts of the restrictions Xp and YQ turn out in fact to be too vague: in both there is nothing which limits the kind of properties which are allowed to be members of the precisifications.8 Consequently, there is nothing which limits the kind of properties that the exceptional individuals and situations have. In other words, a prediction of the present mechanism is that as far as the characterization of the exceptions is concerned, “anything goes”; there is nothing we can predict about which individuals and situations will be considered legitimate exceptions, and which will be considered illegitimate ones. This prediction fails in two points. 3.4.1. “Abnormality” of the legitimate exceptions is not captured The first problem is that the basic mechanism developed above does not capture the fact, noted in section (2) above, that with both IS and BP sentences the exceptional entities (individuals and situations) are considered legitimate (i.e. exceptional entities which do not falsify the generalization) to the extent they are assumed to be “abnormal” or “exceptional in some other sense” besides not having the ^Q property. This fact is not captured by the basic mechanism, since there is nothing in the mechanism which guarantees that all the properties in the set of precisifications are “normal” properties of individuals and situations, and thus nothing which guarantees that the properties not included in the restriction, which characterize the legitimately exceptional individuals and situations, are “abnormal” in some sense. Practically speaking, this means that we may wrongly predict that the property of having a tail (a normal property of dogs) is enough for turning a dog who has three legs into a legitimate exception to Dogs have four legs, just like the property of undergoing an accident (an “abnormal” property).


3.4.2 A (newly observed) difference between IS and descriptive BP sentences is not captured Unlike the first problem, the second has to do with a yet unobserved difference between IS and descriptive BP sentences, concerning the degree to which the properties of the legitimate exceptions can be characterized in advance. Take first IS sentences like (8a) (A sparrow has four toes on each leg). Not only do the legitimate exceptions to this sentence have to be “abnormal” in some sense, but there also seems to be a further limitation on these properties, so not any “abnormal sparrow” (i.e. a sparrow with any abnormal property) will count as a legitimate exception to this sentence. Specifically, although the sparrows in both (8b) and (8c) are clearly considered “abnormal” sparrows, merely having the “abnormal” properties in (8c) is not enough to turn a sparrow with fewer than four toes on a leg to a legitimate exception to (8a). In contrast, having the properties in (8b) is enough to do so: (8) a. A sparrow has four toes on each leg b. “Abnormal sparrows” which are legitimate exceptions to (8a): Sparrows which were caught by a cat (and as a result lost a toe)/ Sparrows which participated in a raging battle for their territory/ Sparrows which have undergone a mutation in the gene responsible for the number of toes/Sparrows which were part of a scientific experiment investigating how sparrows manage to survive after they have one removed c. “Abnormal sparrows” which are illegitimate exceptions to (8a): Sparrows with a problem in their vocal cords/Sparrows which sleep extremely heavily/Sparrows which love reading papers in semantics/Sparrows whose left wing is colored red/Sparrows which have five names ending with ‘g’

8

The only thing which limits the properties in the sets of precisifications is K&L’s original requirement that these properties be compatible with the P property (denoted by the common noun). As I show below, however, this limitation is not sufficient.


My claim is that the ability to distinguish between legitimate and illegitimate exceptions to (8a) systematically holds with all IS sentences, in all contexts of utterance. The situation regarding descriptive BP sentences is much more in accordance with K&L’s original intuition that “what counts as ‘normal’ inherently vague” (p. 408), i.e. with such sentences we can tell little or nothing about which “abnormal” properties will legitimize an exception and which will not. For example, unlike what happens with (8a), it is much harder, if not impossible, to tell which of the “abnormal” uncles in (9b) will be considered legitimate exceptions to the BP (9a), and which will not (assuming that all the uncles are not addicted to marshmallows): (9) a. Uncles are addicted to marshmallows b. Uncles with “abnormal” properties Uncles who are exceptionally tall/Uncles who are exceptionally short/Uncles who are addicted to hard cheese/Uncles who spend lots of time preparing yellow, round tables with five legs/ Uncles who are missing two fingers on their right hand/ Uncles who have 93 nieces and seven nephews/Uncles who have six names The difference just noted between IS and descriptive BP sentences seems to be quite general, and, when the predicate of the sentence is stage level, it also applies to the predictions concerning exceptional situations, as can be seen from the comparison between (10) and (11): (10)

A Pentium computer works very quickly.

(11)

Carpenters drink only water on cloudy Sundays.

With (10) we can have in mind the kind of abnormal situations where (normal) Pentium computers will not work quickly (e.g. very hot or very cold or “outer space” situations), whereas with (11) it is not at all clear in which “abnormal” cloudy Sunday situations are those in which (normal) carpenters do not drink only water (those which are both cloudy and very warm? When it rains very heavily? Those which fall on the first day of the month? etc.). We cannot account for these differences between IS and descriptive BP sentences by using exactly the same tolerating exceptions mechanism with both, as we did above. More importantly, with the present mechanism, where no limitation is imposed on the properties in


the precisifications on v0 and k0, there is no way to account for the predictions concerning legitimate and illegitimate exceptions to IS sentences. 4. Why are descriptive BP and IS sentences different w.r.t. characterizing exceptions? 4.1 The intuition I believe that the source of the difference between the two sentences lies in the different kinds of generalizations they express, and specifically in the presence of the ^S, the “in virtue of” property in the semantic structure of IS sentences, and the absence of this property from the semantic structure of descriptive BP sentences. The intuition is that we cannot really tell what the legitimate exceptions to the generalization will look like (i.e. which properties they will have) if we have no idea what kinds of factors stand behind the generalization, i.e. in virtue of what it is non-accidentally true. On the other hand, once we have in mind in virtue of which property the generalization is asserted to be true, we can be clearer as to the characterization of the legitimate exceptions to this generalization. We do that by hypothesizing which exceptional, or less normal, factors might block this “in virtue of” property from operating, and thus inhibit its results (namely the Q property). For example, if I assert that all non-exceptional dogs have four legs in virtue of the genetic makeup for number of legs they have, then I can tell more or less which dogs will be considered legitimately exceptional, by imagining which exceptional factors in the world might block the natural outcome of this “in virtue of” property, e.g. an accident, a mutation, etc., and by taking all dogs with such properties to be legitimate exceptions to the generalization. Since, as extensively claimed above, the choice of an “in virtue of” property is an integral part of the assertion made by IS sentences, we can systematically characterize the properties which block ^S, and take these properties to characterize the legitimate exceptions to all such sentences.9 In contrast, if, on the basis of inductive reasoning, all we are saying is that “the generalization is not accidental,” i.e. that the generalization is true in other possible, similar, circumstances as well, besides ours, without knowing or specifying which factor or property is responsible


for this pattern, we will not necessarily know or imply what kind of thing will stop this pattern, and consequently will not know, or imply what abnormal properties will legitimize an exception to the generalization. 4.2 Empirical support for the intuition: IS sentences where the characterization of legitimate exceptions seems impossible What supports the intuition above is the contrast between IS sentences like (8a) and (9a) above, or similarly (12a, b) below, where we can predict the properties of legitimate exceptions, and those like (13a, b), where we unexpectedly seem to feel complete vagueness concerning the properties of the legitimate exceptions (apparently similar to the complete vagueness felt with clearly descriptive BP sentences like Uncles are addicted to marshmallows): (12)

a. A dog has four legs. b. A dog is loyal to its master.

(13)

a. A woman in this place does not walk alone outside. b. An accountant in this city does not pay taxes at all.

9

This intuitive procedure is similar to the one described by Chierchia (1995). Chierchia takes a generic like (i) to be paraphrased as in (ii):

(i) A bird flies. (ii) In all worlds where the felicity conditions for flying (like having wings, etc.) are present, and where inhibiting factors (like birth defects, diseases, etc.) are absent, every contextually relevant situation involving every bird is a situation where such a bird flies (p. 196). Once you determine, then, the felicity conditions for the action (in our terms, the “in virtue of” property), you can also determine what the “inhibiting factors” are. Note that in Chierchia’s suggestion both these factors are part of the restriction on the world variable (i.e. they are part of the characterization of the accessibility relation).


We saw above that to some extent we can characterize the legitimate exceptions to sentences like (12a) and (12b). In contrast, out of context, it is not at all clear what are, and what are not, legitimate exceptions to (13a). Women who are not religious enough? Women who are too ugly? Women who are very brave? Woman who have a famous relative? Women who got special permission to walk alone outside? The same is true for the legitimate and illegitimate exceptions to (13b). Are these accountants who don’t cheat? Accountants who earn lots of money? These who don’t have the right connections with the mayor? Those who have another job? Those who just began their career? Out of the blue, and unlike what happens with (12a) and (12b), then, it seems that we are in total darkness w.r.t. to the properties of the legitimate (and illegitimate) exceptions of (13a) and (13b). Why is this the case ? The reason for the difference between the IS sentences in (12) and those in (13) lies in the ability to choose a single, prominent “in virtue of” property. Remember that this cannot be any arbitrary property, but rather, it has to meet two pragmatic requirements: being associated with the ^P property in w, and being a “reasonable causer” for properties of the sort of ^Q. With the former sentences (e.g. A dog has four legs), our real-world knowledge about dogs and canine physiology limits our choice and allows only one such property for every sentence. In (12a) this is something like “having the normal genetic makeup for number of legs” and in (12b) something like “having the normal genetic makeup (or evolutionary history leading to) mental/emotional character.” In contrast, when both (13a) and (13b) are uttered “out of the blue,” there will be no single prominent “in virtue of” property which will be chosen by all listeners. Instead, with both sentences, there are several such possible appropriate properties. (13a), for example can have one of the paraphrases in (14): (14) a. In virtue of being so modest/religious, a woman in this place does not walk alone outside. b. In virtue of being so beautiful and charming, a woman in this place does not walk alone outside.

In the system I suggest below, however, I separate the two: the “in virtue of” property continues to be part of the restriction on the world variable, but the “inhibiting factors” are part of the restriction on the P set of individuals and the set of situations involving them, in the restrictive clause.


c. In virtue of living in such a dangerous place, where murder and rape are so common, a woman in this place does not walk alone outside. d. In virtue of the fact that women in this place have so many children whom they have to take care of and who follow them everywhere, a woman in this place does not walk alone outside. Similarly, (13b), can have (at least) the rough paraphrases in (15): (15) a. In virtue of being covered by the local legislation, which encourages accountants, an accountant in this city does not pay any taxes. b. In virtue of being deeply dishonest and cheating all the time, an accountant in this city does not pay taxes at all. c. In virtue of earning almost nothing (since accountants are not needed here) an accountant in this city does not pay taxes at all. d. In virtue of having the right connections with the mayor, an accountant in this city does not pay taxes at all. I claimed above that the ability to characterize the properties of the legitimate exceptions is directly dependent on the ability to know in virtue of what the generalization is asserted to be non-accidentally true. The idea is that one cannot really predict which properties inhibit “not paying taxes” or “not walking alone outside,” and consequently, the properties of the legitimate exceptions to (13a) and (13b), if one is unaware of the reason for not paying taxes or not walking alone outside in the first place. Unlike what happens in (12a) and (12b) above, the latter type of “out of the blue” utterances each have are several possible appropriate “in virtue of” properties, and therefore there are several corresponding sets of legitimate exceptions. For example, women who know karate, or have a gun, can be considered legitimate exceptions for (13a) if I choose (14c) (“in virtue of living in such a dangerous place”), but not if I choose (14a) (“in virtue of being so religious/modest”). On the other hand, women who received special dispensation from the Pope (or the Chief Rabbi) to walk alone outside will be considered legitimate exceptions if we choose the latter “in virtue of” property.


Similarly, accountants who have another job may be considered legitimate exceptions to (13b), if we choose (15a) (“in virtue of being covered by the local legislation”), but not if we choose (15b) (“in virtue of being dishonest”). On the other hand, accountants who work under the direct supervision of the security manager may be legitimate exceptions in the latter case (since they can’t cheat there without getting caught), but not in the former (they will still be covered by the local legislation). Thus, the vagueness concerning the characterization of the legitimate exceptions, felt with IS sentences like (13a, b), results from the fact that in the “out of the blue” utterance, we, as listeners, cannot determine which of the possible appropriate properties is taken by the speaker to function as the “in virtue of” property of the sentence. But this vagueness is supposed to be resolved when the sentence is uttered in context, since we take the speaker of the IS sentences as committed to have a certain “in virtue of” property in mind, and thus we, as listeners, are supposed to accommodate this property. In this sense, the apparent vagueness of IS sentences like (13a, b) can be taken to be an unresolved contextual ambiguity, similar to the vagueness and ambiguity of nongeneric modal sentences like (16), cited by Kratzer (1981): (16)

I cannot play the trombone.

I repeat here Kratzer’s explanation (cited above in chapter 2): When we talk to each other, we hardly ever make explicit in view of which circumstances something should be necessary or possible […] Depending on the situation in which I utter this sentence I may say quite different things. I may mean that I don’t know how to play the trombone […] Or suppose I suffer from asthma. I can hardly breathe. In view of my physical condition I am not able to play the trombone, although I know how to do it. I may express this by uttering (16). Or else imagine that I am traveling by sea. The ship sinks and so does my trombone. I manage to get to a lonely island and sadly mumble (16). I could play the trombone in view of my head and my lungs, but the trombone is out of reach. (54) These different possible conversational backgrounds for (16) lead to different possible readings of this sentence, i.e. different propositions are expressed by the sentence, where in context, there is always one


conversational background which is chosen, and thus one proposition expressed: A proposition is expressed by the utterance of (the modalized sentence) if, and only if, there is one conversational background for this utterance. (p. 44) The vagueness or ambiguity of the modalized (16), then, is supposed to be resolved in context. The same is true for IS sentences like (13a) and (13b), where in context the listener is supposed to accommodate a single “in virtue of” property out of the several possible properties. In such a case, the vagueness concerning the properties of exceptions is supposed to be resolved as well, and the speaker can characterize the relevant legitimate exceptions in the same way we do with unambiguous IS sentences like A dog has four legs.10 We can see now that the vagueness concerning the properties of the exceptions, felt with IS sentences like A woman in this place does not walk alone outside, is very different from the vagueness felt with descriptive BP sentences like Uncles are addicted to marshmallows. In the latter sentence, having in mind a certain factor or property underlying the pattern (of, e.g., uncles who are addicted to marshmallows) is not part of the semantic structure, i.e. it is not part of what a speaker of such a sentence is systematically committed to conveying, and what the listener is systematically committed to accommodating in every context. Consequently, there is no systematic specification of the kind of exceptional properties which might block the operation of this underlying factor.11

10

In K&L’s terminology, we can say that as far as the restriction on the quantification over worlds is concerned, with sentences like I cannot play the trombone and A woman in this place does not walk alone outside the quantifier over worlds is domain precise, although there may be cases where it is unclear what the precise restriction is. In other words, the speaker clearly has a certain, unique set of worlds in mind, but the listener may fail to accommodate this set. This is similar to the domain precise universal quantifier over individuals (as in Kratzer’s (1989) example Every tree laden with beautiful apples), where the speaker has a precise restricting set of properties of individuals in mind (e.g. “in my orchard”), and hence also a unique set of trees in mind, but the listener may fail to accommodate it.


4.3 The direction of formalization Formally, the way to capture the above-described intuitive difference between IS and descriptive BP sentences seems to be the following: Although semantically, IS and descriptive BP sentences have the same mechanism for tolerating exceptions, as developed above, namely the vague restrictions on the sets of individuals and situations, there are two further different pragmatic limitations imposed on the sets of precisifications in these vague restrictions. In the case of descriptive BP sentences, all we need to ensure is that the individuals and situations which are not quantified over (and are thus considered legitimate exceptions) are, in some sense, “abnormal.” In other words, we must ensure that all the entities which have the properties in any of the

11

Notice that there are many BP sentences where the vagueness concerning the properties of the exceptions seems much weaker, as in (i):

(i)

a. Dogs have four legs. b. Dogs bark.

With respect to characterizing the properties of the legitimate exceptions, there seem to be no difference between these BP sentences, and the minimally contrasting IS sentences in (ii): (ii)

a. A dog has four legs. b. A dog barks.

Just as in (iia) and (iib), we take the legitimate exceptional individuals to (ia) to be e.g. dogs with a mutation, those who have undergone an accident, cut their own leg, etc., and the exceptions to (ib) to be those dogs with a problem in their vocal cords, with plaster on their mouths, who are particularly cowardly, etc. The similarity seems to result from the fact that the prominent reading of such BP sentences is the “in virtue of” reading. Indeed, the difference w.r.t. to characterizing properties of the exceptions is seen only with BP sentences for which the minimally contrasting IS sentences are infelicitous (like Uncles are addicted to marshmallows, Third sons are thirty-eight years old or Red-covered books contain an odd number of typos). These sentence are indeed unambiguously descriptive, and have no available “in virtue of” readings.


(unchosen) precisifications v in V are considered “normal.” This is phrased more precisely in (17): (17) The further pragmatic limitations on the basic mechanism for tolerating exceptions with descriptive BP sentences (“normality”) a. All relevant P individuals who have the properties in any precisification v in V are considered “normal” individuals. b. All relevant situations involving such P individuals who have the properties in any precisification k in K are considered “normal” situations. In contrast, the legitimate exceptions to IS sentences are not merely all “abnormal” individuals and situations, but only those abnormal individuals and situations which have properties inhibiting ^S. (To repeat, although both dogs with no vocal cords and dogs which amputate their own leg can be considered “abnormal dogs,” only the latter will be taken as legitimate exceptions to A dog has four legs.) Thus, we need a finer-grained notion of normality, normality w.r.t. ^S, which allows only those entities which do not have “abnormal” properties inhibiting ^S to be quantified over. Those entities who have properties which are considered “abnormal” w.r.t. ^S will not be quantified over in the first place and will be considered legitimate exceptions. We will achieve this restriction by imposing a pragmatic requirement on all the properties in each precisification v in V, not including the properties directly supplied by context, i.e. those in v0 (which, according to K&L’s original definition, is always a subset of any of the sets v in V). What we want, then, is to make the following restriction on any property in each v-v0 in V (i.e. on the properties which characterize the nonexceptional entities, excluding those contextually supplied properties which characterize the relevant entities): (18) The further pragmatic limitations on the basic mechanism for tolerating exceptions with IS sentences (“Normality w.r.t. ^S”): a. For all properties v in any v-v0 in V, it holds that v is a complement of a property which inhibits the “normal operation” of ^S. b. For all properties k in any k-k0 in K, it holds that k is a complement of a property which inhibits the “normal operation” of ^S.


For example, in the case of A dog has four legs we will end up quantifying over all dogs who did not undergo an accident, did not amputate their own leg, etc., since accidents, and amputating one’s leg block the “operation” of the “in virtue of”, ^S property (which is, most reasonably, something like “has a genetic makeup for four legs”). We can see that the notion of “normality” is part of both intuitive pragmatic requirements in (17) and (18). However, in attempting to formalize these requirements below I will avoid using the adjective “normal” itself. As becomes clear from the discussion of the “normal” ordering source above, using “normality” without further clarifications may lead to circularities and wrong predictions. In addition, it is problematic to use the inherently vague adjective “normal” in a formal system which gives a precise definition of vagueness. Put differently, following K&L, the present mechanism for tolerating exceptions is designed to produce the inherent vagueness of generics. It produces this vagueness, however, in a very precise and systematic way, by imposing a restriction on the sets of individuals and situations, which contain many alternative precisifications, i.e. alternative sets of properties. If we integrate into this system the requirement that these properties are themselves “normal” (or “normal w.r.t. ^S”), where this term is itself inherently vague and undefined, we unnecessarily get double vagueness of generics. These two problems suggest that instead of using the problematic adjective “normal” in the further pragmatic requirement on the properties in the vague restriction, we should find a better-defined way to make sure the intuitive requirements in (17) and (18) are met. The next two sections are devoted to doing exactly that. I start with the further pragmatic requirement on IS sentences. 5. Formalizing the “normality w.r.t. ^S” requirement on the vague restriction of IS sentences 5.1 The further pragmatic requirement on the vague restriction on the P set of individuals In the case of IS sentences, the requirement in (18) can be derived from two independently motivated pragmatic mechanisms. The first is the


“reasonable causation” presupposition on the choice of the ^S property, developed in chapter 2, section (6), and repeated here: (19) ^S is s.t. There is a world w′, (Kratzerially accessible from w, and closest to an ideal version of our world) s.t.

(where in the ideal world some general principle which works in our world is not inhibited or blocked by any other, less general, factor). Remember that this presupposition is in fact an additional restriction on the choice of the “in virtue of” ^S property of IS sentences (besides the “association with ^P” requirement), which was developed above in order to distinguish between e.g. the false-but-felicitous A dog has three legs and the infelicitous #A man is blond. The idea was that, besides being associated with ^P (i.e. being reasonably connected to ^P) the “in virtue of” ^S property should be a reasonably connected to the ^Q property denoted by the VP, as well. The presupposition captures this intuition by requiring that ^S should be a property which is a reasonable causer for properties of the sort of ^Q, i.e. that there is a good possibility (w.r.t. some general process in the actual world), that every P individual with S has Q, or that every such individual does not have Q. This presupposition is indeed met in the case of the false and felicitous A dog has three legs (since we can think about a sort of property associated with dogs which can reasonably lead all dogs to have three legs, or all dogs not to have three legs, i.e. to have certain number of legs), but fails in the case of the infelicitous #A man is blond (since among all the properties associated with being a man none can be taken to reasonably lead all men to have a specific hair color). In addition to the presupposition in (19), a second pragmatic principle we will use here is based on the Grician principle of cooperation (we assume the speaker to intend making a true statement), and the intuition that, when attempting to characterize the properties of the legitimate exceptions to an IS sentence we are in fact attempting to accommodate what kinds of exceptions the speaker of the sentence has in mind. This means that, for the purpose of characterizing the properties of the exceptions, we too (as listeners) assume the sentence to be true, even when we are actually sure that it is false (as in A dog has three legs), or when we are actually not sure that it is true (as in A dog has 23 muscles


in its tail). In the latter case, for example, we can predict that if the sentence is true, then the legitimate exceptions to it are, e.g. dogs who had a mutation in the gene responsible for the anatomy of the tail, or which had their tail amputated, etc. The key words in these intuitive characterizations of legitimate exceptions are “had the sentence been true.” Of course, the assumption that the sentence is true is not maintained when we turn to evaluate the truth value of the sentence. As far as the characterization of exceptions is concerned, however, this assumption is maintained, and it allows us to relate to only the first half of the presupposition in (19). In the case of, e.g., A dog has four legs, by assuming the speaker to make a true statement, we assume that there is indeed a good possibility that every dog with (an associated) ^S has four legs (and not the second, negative part of the presupposition, namely that every such dog does not have four legs). Since the natural candidate for such an ^S property, is e.g. is the epistemically associated property “having a four legged genetic makeup,” then for the purpose of characterizing the exceptions to A dog has four legs, we assume that (20) is true: (20) Where w′ is accessible from w w.r.t. an epistemic modal base, and closest to an ideal world, where some general principle which works in our world is not blocked by any other less general factor. Assuming the truth of (20), we can now define a set B of properties that no dog has in any such close enough, idealized world w′: (21) B is s.t.

Paraphrase: Every property B in B is s.t. there is no similar world w′ (defined as in (20)) where even one dog with a fourlegged genetic makeup has it. Assuming the truth of (20), namely that every dog with a four-legged genetic makeup in w′ has four legs, the properties B in B, i.e. the ones that no dog with a four-legged genetic makeup has in w′, can be characterized as those properties which block, or inhibit the general, or


reasonable connection between being a dog with a four-legged genetic makeup and having four legs. We can think about a variety of such properties that no dog with a four-legged genetic makeup (who has four legs) has in the idealized world: undergoing a mutation or an accident, amputating one’s own leg, participating in some cruel scientific experiment which check how dogs cope with three or less legs, etc. These are, of course, precisely the properties which characterize the legitimate exceptions to the IS sentence A dog has four legs, namely those which are “abnormal w.r.t. ^S” (as the intuitive requirement in (18a) above demands). In other words, these are exactly the properties that we want to exclude from the vague restriction, so no individual which is universally quantified over will have them. To exclude these “blocking properties” from the restriction, we first define a set of properties B′, which contains properties which are complements to the blocking properties in B: (22)

B′ is s.t.

If the properties in B are, for example, “undergoing an accident” and “amputating one’s own leg,” then those in B′ are equal to “do not undergo an accident” and “do not amputate one’s own leg,” respectively.12 We now want to require that any set of properties in the vague restriction V (excluding the properties in v0, which are already supplied by context) contains only properties which are members of the set of properties. In the case of A dog has four legs, for example, assuming the truth of (20), and the definitions of the sets of properties B and B′ in (21) and (22), respectively, we require that (23) holds: (23) The pragmatic requirement on the vague restriction Xp with IS sentences

Paraphrase: Any set of properties v-v0 in V (i.e. any precisification in V, excluding the precise set of contextually supplied properties) is a subset of the set of properties B′. The requirement in (23) guarantees that we end up quantifying only over those relevant P individuals who do not have any property which inhibits, or blocks the “normal,” reasonable “operation” of the “in virtue of” property ^S. In the case of A dog has four legs, for example, we


quantify over all (relevant) dogs who did not undergo an accident, did not amputate their own legs, etc. Since any dog with a “blocking” property is not quantified over in the first place, its presence does not falsify the universal statement, so it is considered a legitimate exception to the generalization, as desired. In contrast to properties like “undergoing an accident” or “amputating one’s own leg,” a property like “having a problem in the vocal cords” cannot be used to characterize a legitimate exception to A dog has four legs, since although it is an “abnormal” property, it cannot be defined a member of B′, i.e. as blocking the reasonable connection between being a dog with a four-legged genetic makeup and having four legs. Finally, note that although we limited the kind of properties of the non-exceptional individuals, the characterization of the non-exceptional individuals is still vague, and, unlike the characterization of the relevant individuals, given by the set of properties v0, it is not the case that in every context there is a closed list of properties that the speaker has in mind, and the listener must accommodate. This is because there are still various possible sets of properties (V-V0 in V) which characterize the non-exceptional individuals (where each such set of properties is required to be a subset of B′), and neither the speaker nor the listener are committed to choose a single such set (i.e. a single and unique subset of B′). 5.2 The further pragmatic requirement on the vague restriction on the set of situations We now turn to show how we characterize the properties of legitimate exceptional situations to IS sentences. As claimed above, when the VP denotes an individual-level predicate (as in A dog has four legs), we do not consider exceptional (or irrelevant) situations, and thus the sets of properties of situations k0 and k in K (in the vague restriction YQ) are empty. But in the case of IS sentences with stage-level predicates like A first grader finishes school at 13:00, A cat chases mice or An ultra-Orthodox Jewish woman covers her hair, we can exclude exceptional situations by systematically

12 Alternatively, assuming that any property B in B is of the form λd.B(d), we can define the set B′ as in (i): (i)


excluding their properties from the vague restriction on the set of situations universally quantified over. As with exceptional individuals, we take the properties which legitimize an exceptional situation to be those which inhibit the general “normal operation” of ^S. We exclude such a situation from the universal quantification by requiring that the vague restriction will contain only complements of these “inhibiting ^S” properties. We now derive this requirement in essentially the same way as we did with the requirement on the properties of individuals in V, namely from the “reasonable causation” presupposition on IS sentences, and from the fact that, for the purpose of characterizing exceptions (and based on the Grician maxim of quality), we assume the IS sentence to be true. Consider for example the presupposition of the (deontic) IS sentence An ultra-Orthodox Jewish woman covers her hair, in (24) (where UOJ stands for “ultra-Orthodox Jewish”): (24)

Where w′ is accessible from w w.r.t. some Kratzerian f, and closest to an ideal world, where some general principle which works in our word is not blocked by any other less general factor. Sentence (24) requires that ^S (a property associated with being an ultraOrthodox Jewish woman) is a property which, relative to some general process in our world, may reasonably lead every such woman who has it to cover her hair, or it would lead all such women not to cover their hair. In addition, we assume the speaker of the IS sentence to speak truthfully. Thus, for the purpose of characterizing the exceptions, we assume the positive half of (24) to be true. In other words, we assume that there is indeed an ^S property, associated with being an ultraOrthodox Jewish woman, which indeed reasonably leads to covering the hair. A reasonable ^S in this case would be the deontically associated property “obeying the ultra-Orthodox Jewish dress code”: (25)


Where w′ is accessible from w w.r.t. a deontic modal base and closest to an ideal world, where some general principle which works in our word is not blocked by any other less general factor. That is, we assume that there is some world which is deontically accessible (e.g. everything commanded holds), and where some process which is considered general in our world (e.g. the principle according to which the intention of a person to do something results in doing it), is not inhibited by less general processes, and in that idealized version of our world every relevant situation, involving every ultra-Orthodox Jewish woman who obeys the ultra-Orthodox Jewish dress code is a situation where this woman covers her hair. As defined in the previous section, in this accessible and similar world no individual ultra-Orthodox Jewish woman who obeys this dress code will have a property in the set of properties B (e.g. a rare skin problem which inhibits covering her hair). We now add that no contextually relevant situation involving such a woman (e.g. a situation where the woman is walking in public) has a property of situations in the set of properties of situations which we can label T. We can define this set of properties of situations T, then, as in (26): (26) T is s.t.

Paraphrase: Every property of situations T in T is s.t. in no similar world w′ (as defined in (25)), there is even one contextually relevant situation involving an ultra-Orthodox Jewish woman, who obeys the ultra-Orthodox dress code, which is a T situation involving such a woman. A situation with a T property in (26) is, for example, a one where a relevant and non-exceptional ultra-Orthodox Jewish woman who obeys the ultra-Orthodox Jewish dress code and who is walking in public takes off her hair cover in order to help somebody who is severely wounded. Another such situation would be where such a woman is on an important top-secret mission and has to hide her ultra-Orthodox Jewish identity, so she uncovers her hair, or where her Rabbi orders her to uncover her hair for some reason, etc. Being involved in such situations indeed blocks the natural, reasonable connection between


being an ultra-Orthodox Jewish woman who obeys the ultra-Orthodox Jewish dress code, and covering the hair. We now define in (27) a set of properties T′, which contains complements of properties T in T. (27)

T′ is a s.t.

Assuming the truth of (25), and the definitions of the sets of properties of situations T and T′, we require in (28) that any set of properties in the vague restriction K (excluding the properties in k0, which are already supplied by context), contains only properties in T′, i.e. properties of the form “not having a blocking property of situations T in T” (as intuitively phrased in (18b) above). (28) The pragmatic requirement on the vague restriction YQ with IS sentences

Paraphrase: Any set of properties of situations k-k0 in K (i.e. any precisification in K, excluding the precise set of contextually supplied properties of situations) is a subset of the set of properties of situations T′. 5.3 IS sentences with an even more vague restriction The pragmatic requirement on the properties in the vague sets V and K in the restriction can now help explaining the intuitive observations above, namely that with IS sentences like (13b), repeated here as (29), the characterization of the legitimate exceptions seems more vague than in the corresponding (24), repeated here as (30): (29)

An accountant in this place does not pay income tax.

(30)

A dog has four legs.

I claimed above that “out of the blue” it is not clear which accountants are, and which are not, considered legitimate exceptions to (29) (Those who don’t cheat? Those who don’t have the right connections with the Mayor? Those who earn a lot of money? Those who have a second job?). I suggested that the reason for this lack of clarity is that, unlike


what happens with sentences like (30), where there is one prominent “in virtue of” property, with sentences like (29), several “in virtue of” properties are reasonably possible. If we want to predict which property will inhibit an accountant from “not paying taxes,” and thus turn it into a legitimate exception, we must know why the accountant does not pay taxes in the first place. The lack of clarity about the “in virtue of” property leads to a lack of clarity about the properties of the legitimate exceptions. We can now be more precise about this intuition. Assuming that there is a “normal,” reasonable connection between being an accountant with S and paying no taxes, then according to the requirement on the vague restriction Xp in (23) above, the properties of the legitimate exceptions (i.e. of those accountants who are not universally quantified over) are those which can be taken to inhibit, or block this “normal” connection. But the problem is that there are several possible S properties in this case. This is because even if we take the speaker of (29) to intend making a true statement, there are still several equally possible, “in virtue of” ^S properties which s/he might have in mind, all of which are associated with being an accountant and all of which may reasonably affect the accountant’s tax-paying behavior (e.g. the accountant’s honesty or average salary, the laws s/he is obeying, the type of connections s/he has with the mayor, etc.). Thus, there are several possible properties which can stand for ^S, and each choice of S leads to a different sort of B (the set of properties blocking the “normal operation” of ^S). Since the characterization of the properties in B is in fact the characterization of the properties of the legitimate exceptional individuals, then until the vagueness concerning the choice of ^S is resolved by the context, there are several possible distinct ways of characterizing these individuals. 6. Formalizing “normality” with the vague restriction of descriptive BP sentences 6.1 The further pragmatic requirement on the vague restriction on the set of individuals Above we formalized the intuition that with IS sentences, having in mind an ^S property in virtue of which we assert that the generalization


is non-accidentally true helps characterize the properties of legitimate exceptions to that non-accidental generalization. The exceptions to IS sentences are taken to be “abnormal w.r.t. ^S,” so the properties which characterize them are those which block the “normal operation” of ^S. Unlike IS sentences, though, where the choice of ^S is an integral part of the truth conditions, no such “in virtue of” property systematically operates in the evaluation of descriptive BP sentences. Based on observations in the actual world we merely assert that the (restricted) universal generalization is non-accidental, i.e. it is true in other possible worlds, similar (in a vague way) to ours, besides ours, without attempting to systematically convey to our listeners in virtue of which factor this is true. The intuition we developed above says that this lack of an “in virtue of” factor leads to the greater vagueness of descriptive BP sentences (like Uncles are addicted to marshmallows) w.r.t. properties of the exceptions. All we can say about the legitimate exceptions is that they are “abnormal” in some sense, without further relativizing this “abnormality” to some “in virtue of” factor. In this section we will give more precise content to the “abnormality” demand, by imposing a further pragmatic requirement on the basic mechanism for tolerating exceptions with descriptive BP sentences. To do that, consider first what happens if we put no further limitation on the basic vague restriction of a descriptive BP sentence like (31a) and (31b): (31) a. Brown-covered books contain an odd number of typos. b. First graders hate math teachers whose names end with ‘t’ or ‘g.’ Take, for example, sentence (31a). Let’s assume that in the context of utterance we are only talking about English books, so books written in other languages are considered irrelevant for the generalization. v0 in the restriction on the set of brown-covered, then, contains the property “written in English”. But now note what happens if no limitation whatsoever is imposed on the properties in the vague part of Xp, as done in our basic mechanism above. In such a case, “anything goes” in the set of precisifications, i.e. any arbitrary property can be a member of the precisifications in Xp, including properties like “containing less than five pages” “containing exactly 2876 pages” or “having a 13-letter title.” This in turn predicts that brown-covered books in English with more than five pages, with more or less than 2876 pages, or with more or less than 13 letters in


their title are not necessarily part of the domain of quantification, and, consequently can be considered legitimate exceptions to (31a) (if they contain an even number of typos). But this prediction seems wrong: these kind of books are not likely to be considered legitimate exceptions to (31a), since we take them to be the majority of brown-covered books in English in our world. Intuitively, in cases where the majority of P individuals do not have the Q property in the world of evaluation, we cannot take the BP sentence to express a true descriptive generalization, i.e. we will take the descriptive BP sentence to be false.13 We find exactly the same problem with (31b). Imagine that (31b) is uttered in the context of a certain school, so the property in v0 is “studying in this school.” Now, if no limitation is imposed on the vague part of Xp, is done in our basic mechanism above, then any arbitrary property can be a member of the precisifications in Xp, and can then characterize the non-exceptional first graders in this school, who are universally quantified over. In principle, then, we can quantify over first graders who have the property of “being the eldest son of the current President of Israel.” This, in turn, predicts that in such a case, all those first graders in this school who do not happen to be the eldest son of the President of Israel, are not quantified over, and are considered legitimate exceptions to (31b) (in case they love math teachers whose names end with ‘t’ or ‘g’). As with (31a), however, this prediction is counterintuitive, since we end up taking the vast majority of first graders in this school as legitimate exceptions to the generalization, which intuitively makes (31b) simply false. The intuition we want to capture, then, is that the legitimately exceptional P individuals constitute an insignificant minority of relevant P individuals in w, or, in other words, that the number of the nonexceptional P individuals, those who are quantified over by the universal quantifier, i.e. who have the properties in any of the precisifications v in V, is not significantly smaller in w, than the number of relevant P individuals as a whole. This is said more formally in (32):

13

Let me emphasize that we require the legitimate exceptions to be the minority in the world of evaluation, but not necessarily at the time of evaluation, or at all times (intervals) in that world. Thus, I can clearly truthfully make the descriptive generalization Owls hunt mice, even if I know that at some point in time, even now, most owls are very sick (e.g. they have been poisoned), and so do not hunt mice. The generalization can be still true since I base myself on what I take to be the behavior of most owls in the actual world in general, and not at some specific time interval in this world.


(32) The further pragmatic limitation on the vague restriction on the set of P individuals with descriptive BP sentences: For any set of properties v in V it holds that |∩v P in w| is not significantly smaller than |∩v0 P in w| Paraphrase: For any set of properties v in V it holds that cardinality of the set of P individuals who have all properties in v in w is not significantly smaller than cardinality of the set of P individuals who have all properties in v0 in w.

the the the the

The definition in (32) also captures the basic vagueness concerning the characterization of the legitimate exceptions to a descriptive generic, namely the intuition that although these exceptions are indeed assumed to be exceptional, or “abnormal in some other sense” besides not having the Q property, unlike what happens with IS sentences, here we are not necessarily or systematically able to specify what this “other sense” is. There are, in fact, two sources to this stronger vagueness (w.r.t. IS sentences). Let us begin with IS sentences like A dog has four legs. Once we have in mind an “in virtue of” ^S property for such a sentence we can tell rather clearly what kind of individual dogs can be considered legitimate exceptions to this sentence (namely, those which have properties which inhibit the natural connection between being a dog with ^S and having four legs), and which kinds of dogs cannot be considered legitimate exceptions (all those dogs who do not have any of these “inhibiting ^S” properties). In contrast, take a descriptive BP sentence like New York taxis are yellow. All we can say about the legitimate exceptions to this sentence is that the properties which legitimize their exceptionality (i.e. not being yellow) is an “abnormal” properties of NY taxis, i.e. a property of some minority of NY taxis in our world. It is not clear, however, which properties of the minority of NY taxis are the correct ones here. It is clear that we do not want all properties of the minority to necessarily legitimize exceptions, since in uttering e.g. New York taxis are yellow we are certainly allowed to include in the generalization taxis with two holes in the front window, or taxis owned by people whose names end with ‘t.’ These are taxis with properties of the minority of NY taxis, but we do not want to claim NY taxis with such properties are necessarily and automatically not quantified over, and are not part of the pattern of


being yellow. Thus, unlike what happens with IS sentences, the characterization of legitimate exceptions with descriptive BP sentences is partial: we can tell which kind of properties will definitely not legitimize an exception (namely properties of the majority of the P set in w), but there is no systematic way to know which properties (of the minority) will clearly legitimize such an exception. Thus, there are indeed many alternative sets of properties of the minority in the vague part of the restriction, and no systematic way to choose between them.14 The second source of “stronger vagueness” felt with descriptive BP sentences lies in the different kinds of real-world knowledge that people make use of in characterizing the exceptions to IS and descriptive BP sentences. In the case of IS sentences, there is some systematic and shared real-world knowledge about the causal connections that hold in our world between the ^P, ^S and ^Q properties (e.g. about the causal connections between being a dog, having a fourlegged genetic makeup, and having four legs). Thus, we can systematically predict many (although not all) of the individual properties in the set of properties which blocks this causal connection (e.g. having a mutation, an accident, etc.). In contrast, although with descriptive BP sentences we can at least tell which set of properties will not legitimize an exception— namely the set of properties of the majority of relevant P individuals— we do not always have shared real-world knowledge about which properties will probably be in this set and which will not. This is especially true in cases where the BP subjects denote an extremely “unnatural” class, like Brown-eyed students whose name ends with ‘s.’ What are the properties which characterize the majority of browneyed students whose name ends with ‘s’ (which will not characterize a legitimate exception to this generalization)? We can be certain about only very few such properties (like “being older than 14 years of age”). Thus, when uttering a descriptive BP sentence like Brown-eyed students whose name ends with ‘s’ love formal semantics we can say almost nothing about which individuals will clearly be considered legitimate exceptions to this generalization and which will not. Unlike what happens with IS sentences, then, with descriptive BP sentences there is

14

In supervaluationist terminology, we characterize only the negative extension of the second-order property “be a property which counts for the present purposes,” but not the positive extension. Thus any precisification v in V may contain different properties of the majority.


no systematic real-world knowledge which will delimit the vagueness, i.e. which will delimit the number of alternative possible ways of legitimizing exceptions. Instead, with these sentences we are left with many distinct alternative ways. 6.2 The further restriction on the vague restriction on the set of situations Finally, we can use a “majority requirement,” similar to the one used to exclude exceptional individuals, to define the legitimately exceptional situations to descriptive BP sentences as well. As noted above, legitimately exceptional situations are only excluded when the VP expresses a stage-level property, as in e.g. (33a) and (33b): (33) a. Brown-haired first graders get to school ten minutes late. b. Young letter carriers whose names ends with ‘t’ drive carefully. Take (33a). We want to exclude not only legitimately exceptional individuals like Bill—a brown-haired first grader in this school who does not arrive at school ten minutes late, and is considered “abnormal” (since he lacks a property in one of the precisifications), but also legitimately exceptional situations. Exceptional situations are e.g. a relevant (e.g. a school-day) situation where, for some unknown reason, John—a relevant brown-haired first grader who is considered perfectly normal—gets to school on time, or a situation where John is 20 minutes (instead of 10 minutes) late. Such situations are considered legitimately exceptional (i.e. they do not falsify the descriptive generalization) if there are not too many of them in our world, i.e. if they constitute an insignificant minority among the set of relevant situations in our world. As with our characterization of exceptional individuals above, here too we require that the number of all relevant situations in our world involving (relevant and non-exceptional) individuals, which have all the properties in any of the (unchosen) precisifications k in K (and are thus universally quantified over) is not significantly smaller than the number of relevant situations in w, involving such individuals (i.e. those who have all the properties in k0). This is seen more formally in (34):


(34) The further pragmatic limitation on the vague restriction on the set of situations involving P individuals with descriptive BP sentences: For any k in K it holds that

Paraphrase: For any k in K it holds that the cardinality of the set of situations involving a (relevant and non-exceptional) P individual in w, who have all the properties in k, is not significantly smaller than the cardinality of the set of such situations in w which have all the properties in k0. For example, in the case of (33b), all we can require is those situations in our world where some relevant and non-exceptional young letter carrier whose name ends with ‘t’ is driving, and which are characterized by the properties in any of the (unchosen) precisifications, constitute the significant majority of all situations in our world where such a letter carrier is busy driving. This, in turn, means, that all such situations which lack one of the properties in any of the precisifications, and are hence not quantified over in the first place, constitute the minority of the relevant situations. These are the situations which can be considered legitimate exceptions to the descriptive generalization. 7. Conclusion In this chapter I completed the truth conditions of IS and descriptive BP sentences by adding to both of them the same kind of basic mechanism for tolerating exceptional and irrelevant entities, namely the Kadmon & Landman’s 1993 -style vague restriction on the sets of individuals and situations restricting the generic quantifier. Using this mechanism naturally captures several similarities between IS and descriptive BP sentences: their basic tolerance of exceptional individuals and situations, their tolerance of contextually irrelevant individuals and situations, the difference in the kind of vagueness in the characterization of the


exceptional and irrelevant entities, and the predictions concerning actual individuals and situations (with epistemic IS sentences, and with all descriptive BP sentences). To capture the “abnormality” of the legitimate exceptions, and the difference between the two kinds of sentences constructions concerning characterizing the properties of the exceptions, I supplemented the basic vague restrictions in the truth conditions of each construction with two further, and different, pragmatic requirements. With IS sentences the intuition I developed was that you can predict which individuals and situations will function as legitimate exceptions once you have in mind the property in virtue of which the generalization is supposed to be true. Thus, despite the basic vagueness concerning the characterization of the exceptions, their properties can be characterized as “abnormal w.r.t. ^S,” which we more formally defined as “inhibiting the reasonable, normal connection between ^S and ^Q property.” In formalizing this intuition I made use of the independently motivated second presupposition on IS sentences (requiring the ^S is a reasonable “sort of property” leading to the sort of property of ^Q), as well as the Grician maxim of quality (of assuming the speaker to intend making a true generalization). The resulting further pragmatic requirement on the restriction on Gen guarantees the characterization of exceptions is still vague, since the restriction still contains alternative sets of properties, and not a single, unique such set. In contrast, since the truth and well-formedness of descriptive BP sentences does not include a systematic choice of an “in virtue of” property, the vagueness concerning the characterization of exceptions is not systematically delimited in the same way. Instead, the “abnormality” of the legitimate exceptions is captured by taking the properties of the non-exceptional individuals and situations (in V and K) to yield the majority of relevant individuals and situations in the world of evaluation. Put differently, in the case of descriptive BP sentences, “abnormal” is “having some property which places you in some unchosen minority-set of the relevant set of entities in the actual world.” We now see that although the basic mechanism for tolerating exceptions is a restriction on the set of individuals and situations in the restrictive clause of Gen—and not on the set of possible worlds, as in the “normal ordering source” suggestion—the difference between the way the set of accessible worlds is defined with IS and descriptive BP sentences does play an indirect role in the distinct characterization of exceptions to these two sentences, since the further pragmatic limitations


on the vague restriction is systematically influenced by the presence/ absence of an “in virtue of” property in the accessibility relation of the sentence. Thus, unlike Kadmon & Landman’s (1993) original suggestion, where the “law-likeness” and the tolerance of exceptions are treated as two separate, unconnected properties of the generic quantifier (namely being modalized and domain vague), in the present theory the finer-grained definition of the former property systematically influences the finer-grained definition of the latter.

246

CHAPTER 5 Temporally Restricted IS and BP Sentences

1. Introduction In chapter 1, I pointed out a number of semantic and distributional differences between minimally contrasting BP and IS sentences, which pose a problem for the widely held approach which claims that such sentences should be given an equivalent representation. In chapters 2, 3, and 4, I dealt with these differences in classical generic constructions (i.e. with individual-level and habitual predicates) and developed an analysis which formally captured the intuition that the wide range of differences noted is due to two kinds of generalizations expressed by the two kinds of sentences (formalized by using two kinds of accessibility relations in the basically equivalent truth conditions). In this chapter, I turn to the differences between IS and BP sentences in another construction discussed in chapter 1, namely sentences modified by specific temporal adverbials (like tonight or this week). My principal claim is that this set of differences between IS and BP sentences is of essentially the same nature as the set of differences between IS and BP sentences in the classical generic constructions. The purpose of this chapter, then, is to sketch the way in which the behavior of IS and BP sentences in what I call “temporally restricted generics” can be formally captured through the two kinds of accessibility relations and presuppositions developed above for IS and BP sentences in classical generic constructions. Section (2) describes the behavior of IS and BP sentences in “temporally restricted generic” sentences (described in part in chapter 1 above). In section (3) I will argue in detail that such sentences are indeed generic, and not “functional” (in Condoravdi’s (1993, 1997) terms). In section (4) I sketch the descriptive generalization concerning the connection between the behavior of IS


and BP sentences and the interpretation of the temporal adverbial, and in section (5) argue that there is no simple explanation for this generalization in the framework of existing theories of genericity. In section (6) I suggest an explanation for the descriptive generalization in terms of the different accessibility relations and presuppositions of IS and BP sentences, developed in the chapters above. Section (7) concludes this chapter. 2. Temporally restricted generics 2.1 The basic data As pointed out in chapter 1 above, the standard assumption in the literature of genericity is that characteristic (I -) genericity (i.e. quantificational genericity, as opposed to Proper Kind-Predication and episodicity are mutually exclusive, in other words that I-generic sentences always attribute long-lasting, or tendentially stable, properties to their subjects. This claim is explicitly made in e.g. Krifka (1987), Krifka et al. (1995), Chierchia (1995), Cohen (1997, 1999a) and Carlson (1982), (1989, 1995), and is implicit in most other theories of genericity. Typical examples brought to support this view are the existential-only interpretation of the BP and IS subjects in (1) and (2), and the oddness of (3), with an habitual predicate (which is standardly taken to involve generic quantification), all of which are modified by a specific temporal adverbial: (1) (2) (3)

Dogs are barking now. A dog is barking now. # John walks to school this week

As shown in chapter 1, however, there are cases where, in addition to the existential reading, BP and IS subjects can also get a generic interpretation when combined with episodic predicates, i.e. with predicates modified by specific temporal adverbial modifiers. Such sentences are the main concern of this chapter, and examples of these sentences are brought again in (4)–(7): (4)

Italian restaurants are closed tonight.


(5) (6) (7)

Jews are in synagogue today. An American eats turkey tonight. A Catholic is in church today.

Crucially, in these examples (as well as in all the following examples), both the BP and the IS subjects combine with predicates of individuals (i.e. non-kindlevel predicates). Moreover, unlike classical kind predication structures (like Dodos are extinct or Rats reached Australia in 1779), the interpretation of these subjects seems to be “quantificational,” i.e. the sentences is interpreted to mean something along the lines of: “All/most (typical/relevant) Italian restaurants, Jews, Americans, or Catholics” (in (4), (5), (6) and (7), respectively). Restated, the predicates denote properties which are distributed over all/ most members of the sets denoted by the subject. Thus, according to most theories, these sentences (including the BP ones) should be represented as characteristic (I-) generic, involving generic quantification (and not as Proper Kind-Predication structures). 2.2 Differences between BP and IS sentences The sentences in (4)–(7) show that both IS and BP sentences can be interpreted as generic when temporally modified. As discussed in chapter 1, however, examination of a wider set of such sentences reveals that in most other cases BP are judged as much better in this construction than their IS counterparts. In fact, as seen in (8), the sentences in (6) and (7) are exceptional, in that the judgments about most generically interpreted IS sentences in this construction vary between “odd” and “completely unacceptable” (although they are fine under the existential reading). In contrast, the minimally contrasting BP sentences are in general perfectly natural as generic with such temporal adverbials, as seen in (9) (more examples of “episodic generic” BP sentences are cited in McNally (1995)): (8)

a. # An Italian restaurant is closed tonight. b. # An accountant is very busy this week. c. # A lion is very aggressive today. d. # A child is especially polite today. e. # A Clinton supporter is happy tonight.


(9)

a. Italian restaurants are closed tonight. b. Accountants are very busy this week. c. Lions are very aggressive today. d. Children are especially polite today. e. Clinton supporters are happy tonight.

Despite this general incompatibility of IS sentences with specific temporal adverbials, exemplified in (8), there are some factors which considerably improve their status. One is the presence of certain kinds of modal expressions, like would, should or the deontic must: (10)

a. An Italian restaurant must be closed today. b. A lion would be very aggressive today. c. A police officer should be nervous today. d. A child should behave especially politely today. e. A police officer must be especially alert today.

Note that it cannot be the mere modality of these modal expressions which improves the status of the IS sentences, since the presence of e.g. the modal must, under its epistemic interpretation, makes very little or even no improvement at all, as seen in (11) (the epistemic reading of must becomes salient if must is deaccented): (11)

a. # An Italian restaurant must be closed b. # A lion must be very aggressive today. c. # A police officer must be nervous today. d. # A child must be especially polite today. e. # A police officer must be especially alert today.

The epistemic reading of the modal must is even more salient when we use nonhuman subjects, as in (11b), for which the deontic reading is pragmatically less natural (since we do not tend to think of e.g. lions


(unless they are tamed) as obeying some social or moral standard or rule).1 Another such example is (12). Sentence (13) shows that the minimally contrasting BP sentence is perfectly felicitous: (12) (In view of what we know about the unexpected great damage caused by earthquakes all over the world) ??/* An earthquake must be very powerful today. (13) (In view of what we know about the unexpected great damage caused by earthquakes all over the world) Earthquakes must be very powerful today. Turning back to the original data, we can see that another “improving factor” for the originally infelicitous IS sentences is the addition of modifying “normative” adjectives. For example, the sentences in (14), are much better than those in (8): (14)

a. A decent Italian restaurant is closed today.

1

At first sight, the difference between (10) and (11) seems to be connected to the correlation between root (e.g. deontic) and epistemic modals, on the one hand, and the universal and existential reading of IS subject, on the other, as noted in Kratzer (1995). Kratzer’s example is (i):

(i)

A car must be in the garage.

Kratzer points out that the IS subject can be interpreted universally when the modal is taken as deontic, but as existential-only when the modal is understood epistemically, and claims that this correlation indicates that root modals are (at least optionally) unselective operators, binding the x variable introduced by indefinites, whereas epistemic modals cannot be unselective, and can only quantify over worlds. If we apply Kratzer’s explanation to the contrast between (10) and (11), we might claim that the universal reading of the IS in (10) is due to binding by the unselective deontic modal, whereas the lack of such a reading in (11) is because the unselective epistemic modal cannot bind the individual variable introduced by the IS subject. However, there are two main problems with this suggestion. First, it cannot explain the different behavior of temporally restricted IS and BP subjects, namely the fact that unlike the IS subjects, BP subjects with epistemic modals are freely interpreted as generics, as can be seen in (13).


b. A true Clinton supporter is happy tonight. c. A decent accountant is busy this week. d. A well-behaved child is especially polite today. e. A professional police officer is alert today. The last “improving factor” for the IS sentences, as discussed in chapter 1, is context. Most informants find the IS sentences better when uttered with supporting context, instead of “out of the blue.” Interestingly, however, not any context can help here. Specifically, the IS sentences get better in what I will call a “predictional” scenario, but not in an “inductive” scenario. In contrast, BP sentences can appear easily in both kinds of scenarios. The following examples illustrate the different kinds of scenarios, and the different behavior of BP and IS sentences in them:2 (15) The Italian restaurants scenarios: The inductive scenario: John and Mary plan to meet in town and eat in an Italian restaurant. John arrives in town before Mary, and while walking down the street, he sees that Mamma

In addition, an independent problem with Kratzer’s claim is the universal/ generic interpretation of the IS subject in (ii), with the modal may, under its root reading:

(ii)

A car may be in the garage.

The subject here gets a universal reading, just like the one in (i). In this case, however, the modal is existential, and thus cannot universally bind an x variable (since possibility modals, like may are existential operators, not universal operators). Instead, it seems that the universal reading of the subject in (ii) is a result of binding by an independent implicit generic operator, which is present in the sentence in addition to the overt modal. If this is the case, the universal reading of the subject in (i) too is not a necessary indication of binding by the overt modal, and can also be attributed to the presence of an implicit generic operator (see Brennan (1993) who independently makes a similar claim). 2

Some informants still find the IS sentences in the predictional scenarios worse than their BP sentences counterparts. Nonetheless, the IS sentences are still taken to be better in this kind of scenario than in the “inductive” one, where they are clearly infelicitous.


Mia is closed. He then goes to Little Italy, and then to Fonta Bella, but they are also closed. He tries two more Italian restaurants, but they are also closed. When Mary arrives to meet him, John says: (a) Look, Italian restaurants are closed tonight. Let’s go to an Indian restaurant. (b) ??/* Look, an Italian restaurant is closed tonight. Let’s go to an Indian restaurant. The predictional Italian restaurants scenario: My friend and I decide to eat tonight in an Italian restaurant, but an hour later my friend calls me and says, “We better go eat at an Indian restaurant tonight. I just remembered that it’s Italy’s Independence Day, and (a) Italian restaurants are closed tonight. (b) ? An Italian restaurant is closed tonight. (16) The accountants scenarios: The inductive scenario: I need to consult an accountant, so I open the Yellow Pages and try to call one. Unfortunately, the secretary tells me her boss is very busy, and I should try him again next week. So, I try the next accountant on the list. This time the accountant himself answers, and says he is very sorry, but he is very busy now. During the following three or four days I try again and again to set up a meeting with an accountant, but all the accountants I call are busy. At some point, I tell my husband: (a) Accountants are busy this week. I’ll try again next week. (b) ??/* An accountant is busy this week, I’ll try again next week. Predictional scenario: It’s the week when taxes are due and everybody is dealing with their taxes. I listen to a program about economics on the radio, and one of the experts says: “Anybody who wants to consult with an accountant should do it now, a. Accountants are busy this week b. (?) An accountant is busy this week (17) The Clinton supporters scenarios: The inductive scenario: I am a tourist who tonight just arrived in New York. It’s election time, but I’m not really interested in


who is ahead in the polls. In the cab on the way from the airport, I keep seeing people running around, carrying signs and wearing shirts saying “Clinton,” and they are all dancing and screaming with joy. I see several such groups that night. When my husband calls me from Israel and asks me what’s new I say: a. I don’t know. Probably something happened, Clinton supporters are happy tonight. b. ??/* I don’t know. Probably something happened, a Clinton supporter is happy tonight Predictional scenario: My family and I are supporters of Dole. Tonight there was an important public debate between Dole and Clinton. Clinton gave a masterly performance and Dole a terrible one. After the debate my grandmother says bitterly: a. Clinton supporters are happy tonight! b. (?) A Clinton supporter is happy tonight! 3. Real generic or “functional”? 3.1 Condoravdi’s (1993, 1997) “functional” reading Before turning to a closer examination of the unexpected existence of “episodic generic” sentences, and the differences between BP and IS in these sentences, we must ensure that these sentences are indeed generic. If, as most theories hold, tendential stability is a necessary characteristic of quantificational, I-genericity, or, in other words, episodicity and Igenericity are mutually exclusive, it may be that sentences like those in (4), (5) and (9) are not generic at all. This is the view taken by Condoravdi (1993, 1997), who coined the terms “functional” for the interpretation of BP subjects in sentences like (18b), which on first sight are very similar to those examined above:3

3

Condoravdi’s examples differ from the example examined in section (2) in two respects: (a) Condoravdi only discusses sentences with BP subjects, whereas I also discuss temporally restricted IS sentences, and (b) The temporal restriction is contextual in Condoravdi’s example (i.e. part of the background introduced by previous sentences), whereas it is an explicit part of the sentence in the cases I discuss.


(18) a. In 1987 there was a ghost haunting the campus. b. Students were aware of the danger. (Condoravdi 1997 p. 57) Condoravdi (1997) claims that “although the bare plural received a universal interpretation in (18b), it is not generic in any obvious way” (p. 57) [emphasis mine], and she brings five arguments for this claim, which are relevant for our data as well. First, Condoravdi claims, unlike real generic sentences, (18b) “…not express a regularity about the occurrence of awareness in other situations in which a ghost was haunting the campus. The individual level predicate is not understood as expressing a characteristic property” (p. 57). Second, Condoravdi shows that the subject in (18b) is necessarily contextually restricted to students in the campus, and not to students in general. She follows Krifka (1987) and Krifka et al. (1995) in claiming that such contextual restriction is never possible with generically interpreted IS and BP, and thus concludes that (18b) cannot be generic. A third argument brought by Condoravdi is that (18b) expresses a generalization about actual students. An indication that (18b) indeed talks about actual individuals is the fact that it implies the existence of students in the actual world. This can be seen from the contradiction between (19b) and (19c): (19) a. In 1985 there was a ghost haunting the campus. b. Students with police connections were aware of the danger. c. # But there were no students with police connections on campus in 1985. (p. 62) Condoravdi writes that (19b) certainly implies that there were students with police connections on the campus is 1985. Continuing the discourse with […] (19c) leads to a contradiction. If the generic interpretation were the only interpretation for (19b) […], no implication of existence would be guaranteed. (p. 62)

4 Condoravdi claims that a similar thing happens with explicit adverbs like usually or typically.


A fourth argument that Condoravdi brings against the assumed genericity of sentences like (18b) is that the BP subject in this sentence takes wide scope w.r.t. to deontic modals, as in (20b):4 (20)

a. A ghost is haunting the campus. b. Students should be aware of the danger. c. Unfortunately, they are not.

Condoravdi claims that in (20b) “the students are still taken to be the actual students” (p. 66). She takes this as an indication that the BP is not in the scope of the deontic modal, which in turn shows that “the truth of sentences like (20b) depends on nothing but the actual world” (p. 66). Finally, Condoravdi shows that such “functional” readings with episodic or stage level predicates occur only with BP subjects, but not with other indefinites, e.g. minimally contrasting IS subjects. Since, according to her “no such discrepancy between the singular indefinite and the bare plural occurs in standard case of generic quantification” (p. 73), the existence of such a discrepancy in episodic contexts “is another piece of evidence that the usual notion of genericity is not involved in the contextually restricted universal reading (of BP sentences like (19b))” (p. 74). 3.2 Condoravdi’s conditions for nongenericity are relevant for temporally restricted sentences Let us turn back to the temporally restricted BP sentences examined in section (2) above, and repeated here: (9)

a. Italian restaurants are closed tonight. b. Accountants are very busy this week. c. Lions are very aggressive today. d. Children are especially polite today. e. Clinton supporters are happy tonight

The sentences in (4)-(7) above and in (9) seem to share, at least potentially, the properties which Condoravdi takes as necessary signs of nongenericity. None expresses a regularity, since they are all temporally restricted (in fact they are even more problematic than (18b), since they are restricted to much shorter intervals than a year, such as one day or


night). At least some of them can be easily understood as contextually restricted to e.g. Italian restaurants in this town (in the case of (9a)), or to the lions in a certain national reserve (in the case of (9c)). These sentences too seem to talk about actual individuals, e.g. actual Italian restaurants, accountants, lions, etc. and they too have existential implications, as can be seen from the infelicity of (21b): (21) a. Esperanto restaurants are closed tonight. /An Esperanto restaurant is closed b. # But in fact there are no existing Esperanto restaurants in the world. In addition, as in Condoravdi’s examples, the BP subjects of such sentences seem to take wide scope w.r.t. deontic modals: (22)

a. Italian restaurants should be closed tonight. b. Actually, they are not.

Finally, as seen in section (2) above, there is in most cases a discrepancy between such BP sentences and their minimal pairs with IS subjects: (23)

a. Italian restaurants are closed tonight b. # An Italian restaurant is closed tonight

If, then, Condoravdi is right in claiming that real generic sentences never have these properties, then it may be that what I called above “temporally restricted generics” are not generic at all. However, examining Condoravdi’s tests reveals a different picture. 3.3 Examining Condoravdi’s conditions for nongenericity Let us start with contextual restriction. Note first that contrary to Condoravdi’s claim, sentences like her original one in (18b), are, in fact, not necessarily contextually restricted. Thus, the subjects in (24b) (noted by Maria Bittner (p.c.)) and (24c) seem to be about scientists or parapsychology students in general, all over the world, and not necessarily on a certain campus:


(24) a. In 1985 there was a ghost haunting the campus. b. Scientists got interested in the phenomenon. c. Parapsychology students got interested in the phenomenon. Similarly, sentences like those in (9) above, which are explicitly temporally restricted, are, in fact, not necessarily contextually restricted to a certain location. Another example is the subject of (25), which is not restricted to earthquakes in a certain area, but to earthquakes all over the world: (25)

Earthquakes were especially powerful yesterday.

Even more important is that even the contextually restricted sentences above can be easily treated as generic: As noted in chapter 4, contrary to the claim put forth in Krifka (1987) and Krifka et al. (1995), which Condoravdi follows here, clear generic sentences can be, in fact, contextually restricted. For example, the subject in (26a), which seems to be a clear generic sentence (noted by Fred Landman (p.c.)), and the subject in (26b) are naturally understood as restricted to professors in this university: (26) a. There are students and professors in this university. Professors wear/A professor wears a tie. b. This library lends books and periodicals. Books/A book can be borrowed for up to a month (but periodicals/a periodical can only be borrowed overnight). c. There are shirts, skirts and scarves in this shops. Shirts and skirts are rather, but scarves cost less then NIS 50. What this means that a contextual restriction is not a necessary indication of nongenericity, and thus the fact that temporally restricted sentences are, in some cases, contextually restricted, should not be used as argument against their genericity.5 Let us turn now to the existential implications of temporally restricted sentences like (9). As seen in the previous section Condoravdi’s view is that real generics do not have such implications. But in fact, many pasttense generics, involving both IS and BP subjects, have this property: (27) a. In the Middle Ages, astronauts/an astronauts earned lots of money.


b. # But in fact there were no astronauts in the Middle Ages. The infelicity of (27b) indicates that (27a), which is a clear generic sentence, presupposes the existence of astronauts in the Middle Ages. In addition, as shown in chapter 3 above, present tense Descriptive BP sentences systematically presuppose the existence of instances (individuals and relevant situations involving them) in the actual world (although not necessarily in the immediate present). Condoravdi’s claim that the wide scope of the BP subjects w.r.t. deontic modals is an indication for the nongenericity of the subject is also problematic. In fact, in many standard generic sentences, with both BP and IS subjects, the subject outscopes a deontic modal verb. One rather clear case can be seen when we use a deontic possibility modal verb, as in (28): (28)

First graders/A first grader may bring sweets to class

Both IS and BP subjects get a universal/generic reading in (28). If these subject were in the scope of the deontic modal “may,” their interpretation would be existential (since possibility modals are existential quantifiers). Thus, although just as in (22a) above, these

5

Condoravdi (1997) admits that in some cases contextual restriction of real generics is possible. She cites an example very similar to (26) above:

(i)

a. People in this university dress formally b. Professors wear a tie (p. 113)

Condoravdi claims that such cases are possible when modal subordination is involved, i.e. when “the restricting set of variables is itself within the scope of a generic operator” (p. 113). But in fact, we can find contextually restricted generics (as in (iib) even when the restricting clause (in e.g. (iia) is not in the scope of a generic operator, i.e. not a generic but a clearly extensional assertion: (ii) a. This library holds 11,800 books, and 578 periodicals b. Books/A book can be borrowed for one weeks only, but periodicals can only be borrowed overnight.


subjects seem to take wide scope w.r.t. the deontic modal, we would not like to claim, though, that these sentences are not generic. Finally, the fact that there is a discrepancy between many IS and BP minimal pairs in temporally restricted sentences (as seen again in (23a)(23b)), should not be used as an indication that the temporally restricted BP sentences are not generic. This is because, unlike Condoravdi’s claim that “no such discrepancy between the singular indefinite and the bare plural occurs in standard case of generic quantification” (p. 73), we saw in the chapters above many cases of classic, standard generic sentences which differ in their felicity depending on whether the subject is a BP or an IS (starting from Lawler’s classical examples Madrigals are popular and #A madrigal is popular). In fact, my claim in the following sections will be that the discrepancy between IS and BP sentences in temporally restricted sentences is of exactly the same type as the one we find with (29a) and (29b), which are undoubtedly generic: (29)

a. Uncles are garrulous b. # An uncle is garrulous

In contrast to Condoravdi’s theory, then, the discrepancy between temporally restricted IS and BP sentences would be used as an argument for, and not against, the genericity of the latter. To sum up, besides the episodicity of (18b), (as well as (7)-(10)), which is indeed surprising and is a target of examination in this work, there is no indication that such sentences are significantly different from classic quantificational, I- generic sentences. 3.4 Positive indications for genericity Moreover, we have positive indications that these sentences are in fact Igeneric. In particular, these sentences and classic I-generic sentences share three properties which have been identified as central for the characterization of I-genericity. Specifically, like other classic I-generics, the sentences in (4)-(9) allow for exceptions, i.e. they are true even if not all members of the set denoted by the subject have the predicated property. Thus, for example, (9a) is true even if we find an Italian restaurant which is open tonight. Second, like other classic generics, all these sentences obey Cohen’s (1997, 1999a) “homogeneity constraint.” Cohen shows that for a generic sentence to be true, it is not enough that most members of the set


have the predicated property, even if they have the predicated property regularly. Rather, the predicated property has to hold for most members of the set in a “homogeneous” manner, without excluding any salient subset of the set. An example of Cohen’s constraint is a situation where 75% of the primary school teachers are female. Crucially, this situation supports the truth of (30a), which includes the word most, but not the truth of the generic BP sentence in (30b) or the IS sentence in (30c): (30) a. Most primary school teachers are female. b. Primary school teachers are female. (Cohen (1999), p. 228) c. A primary school teacher is female. We get exactly the same pattern with the temporally restricted sentences above. For example, imagine a situation where 75% of the Chinese restaurants are in Chinatown, and all of them are closed tonight, while all other Chinese restaurants in the rest of the city are open. This kind of situation easily supports the truth of (31a), but clearly not that of (31b): (31)

a. Most Chinese restaurants are closed tonight. b. Chinese restaurants are closed tonight.

Finally, and most crucially, like other generic sentences, both the IS sentences (in the predictional scenarios) and the BP sentences (in both the predictional and the inductive scenarios) in the temporally restricted constructions support counterfactuals. For example, (32a) and (33a) support the truth of (32b) and (33b), respectively: (32) a. Italian restaurants are closed tonight. b. If this were an Italian restaurant it would probably be closed tonight as well (uttered while pointing at an Indian restaurant) (33) a. An American eats turkey tonight. b. If they were Americans they would probably eat turkey tonight as well (uttered while pointing at an Israeli family eating fish) This last point deserves emphasis. As pointed out in chapter 1, and emphasized throughout, the ability to support counterfactuals is a central property of I-generic sentences, and a clear indication of their “nonaccidentalness.” As shown above, following the work of e.g. Lewis (1973, 1986b) and Stalnaker (1968), this property was taken by scholars


including Dahl (1975), Heim (1982), Krifka (1987), and Chierchia (1995, 1998) as one of the main motivations to represent I-generic sentences as involving a modal operator, requiring consideration of other possible worlds besides the actual one for their evaluation. Thus, although it is certainly true that temporally restricted sentences like (9) above generalize over actual individuals, it is not true that they generalize only over actual individuals. Considering nonactual individuals in other possible worlds is part of the semantics of such sentences. Indeed, informants comment that despite their episodicity and potential contextual restriction, both the IS sentences in the “predictional scenarios” and the BP sentences in the “inductive scenarios” express generalizations, with some sort of “nonaccidentalness” which is not present in purely extensional sentences like “All/Most Italian restaurants are closed tonight.” This kind of “restricted” non-accidentalness is somewhat confusing and differs at first sight from that of classic generic sentences like Birds fly. The main difference seems to lie in the fact that the “nonaccidentalness” of classic generics like Birds fly manifests itself both by supporting counterfactuals (“If this was a bird it would probably fly”), but also in generalizing over possible entities (birds, in this case), and not necessarily of actual ones. Thus, this sentence can be true even if at the present time there is no bird alive in our world, or if all birds alive now happen to lose their ability to fly. This fact was also noted in the discussion of present tense IS and Descriptive BP sentences in chapter 3 above, which showed that such sentences can be true even if no P individual in the present has the Q property, and felicitous even if no relevant instances exist in the present (as long as they exist in the interval surrounding the present, i.e. past and normal futures). This is not exactly what we find with the temporally restricted generics examined in this section, like Italian restaurants are closed tonight, which despite its counterfactual supporting property, does seem to necessarily talk about the Italian restaurants which exist tonight. This sentence is false if no actual Italian restaurants are closed tonight, and infelicitous if no Italian restaurant exists in the actual world at the present interval. Such temporally restricted generics, then, express generalizations which are, at the same time, “non-accidental,” but also necessarily express generalizations about the actual world and time. Below I clarify this “double nature” of temporally restricted BP sentences, and compare it with the kind of non-accidentalness expressed by the corresponding IS sentences. At this point, however, it suffices to take the counterfactual supporting property of (4)-(9) as an indication of


some sort of non-accidental generalization that these sentences express, and for the presence of a modalized generic quantifier.6 I conclude, then, that apart from their temporal restriction, the temporally modified sentences in (9) behave like classic I-generic sentences in several important respects.7,8 Thus, they indeed pose a problem for the traditional view that I-genericity and episodicity are mutually exclusive, and therefore they merit closer examination.

6

Unlike the temporally restricted sentences in e.g. (9) and (27a), it is not clear to me that all of Condoravdi’s examples, and, in general, all temporally restricted BP sentences, similarly support counterfactuals. Thus, it is not clear to me that (ia), and the similar (iia) support the counterfactuals in (ib) and (iib):

(i)

a. In 1985 there was a ghost haunting the campus. Students were aware of the danger. b. If John was a student in this Campus, he would probably be aware of the danger too. (ii) a. Wow! People on this bus were really noisy! (uttered after getting off a city bus) b. If Mary (who is at home now) were on this bus, she would probably be noisy as well. If indeed such sentences cannot support counterfactuals, I am not sure I want to consider them as generic, and it may be that a different theory should be built for them (maybe Condoravdi’s original theory). At this point I will continue to concentrate on the counterfactual supporting sentences like (9), and leave the discussion of sentences like (i) and (ii) to later research. 7 As pointed out by G.Chierchia (p.c.), in Italian, sentences like Italian restaurants are closed tonight appear in the imperfective aspect, which, as claimed in e.g. Chierchia (1998) and Delfitto (1996), is an indication of genericity and intensionality. 8 In contrast to temporally restricted sentences like the ones discussed above, however, generically interpreted BP sentences with progressive predicates may require different treatment, as nonquantificational Direct Kind Predication


4. A closer examination of the BP/IS distinction in temporally restricted sentences Let me return to the differences between IS and BP sentences when temporal adverbials are present. Remember that the basic pattern seen in section (1.1) was that BP sentences are perfectly natural with such adverbials, whereas in most cases, such IS sentences are infelicitous, at least to some extent, unless some specific “improving factors” are present. In this section I develop a descriptive generalization concerning these facts.

structures. Consider, for example, the sentences in (i) and (ii), brought to my attention by Angelika Kratzer (p.c.):

(i)

a. Children are knowing more and more about sex these days. b. * A child is knowing more and more about sex these days (ii) a. Elm trees are dying now. b. An elm tree is dying now. (existential) Unlike episodic BP generic like the ones discussed above, which indeed seem to involve (restricted) universal quantification over individuals (e.g. Italian restaurants are closed tonight is naturally paraphrased as “All (nonexceptional) Italian restaurants are closed tonight”), with BP sentences like (ia) and (iia) we do not seem to get such a universal quantifier, and a Direct Kind Predication reading seems more appropriate. Sentence (ia), for example, does not mean that for every child it holds that s/he knows more and more about sex these days, but instead that the kind “children” in general (gradually) knows more and more about sex these days. Similarly, (iia) does not mean that every nonexceptional elm tree is dying now, but that the kind elm trees is in the process of gradually dying (so perhaps only a few elm trees are dying now, more will die in a month, more in two months, etc.). These intuitions may indicate that the ungrammaticality (as generics) of the IS sentences in (ib) and (iib) results not from a problem in the type of the “in virtue of” modality, but rather from the lack of a Direct Kind Predication reading, together with the fact that the generic quantifier is incompatible with the progressive (see e.g. Wilkinson (1994)).


4.1 Two readings of the temporal adverbial As already pointed out chapter 1, understanding the felicity/infelicity variations observed with temporally restricted IS sentences seems to depend on distinguishing two readings of temporal adverbials like tonight. Such adverbials have their default, “indexical” interpretation, where they simply denote temporal locations on the night, day, week, etc. in which the utterance is made. This default reading of adverbials like tonight, which I call the “indexical reading,” is the one we most naturally get in simple sentences like (34a), written more formally in (34b). In this case the temporal adverbial specifies that the temporal location of the existentially closed situation variable in the night of utterance.9 (34) a. Mary is at the movies tonight. b. s [at-the-movies (s) & Ag(s)= Mary & loc.(s)=night of the utterance] However, such adverbials have another interpretation, which I term the “functional” reading, as denoting values to functions from years, weeks, etc. to temporal locations. “Italian Independence Day,” “John’s birthday,” or “Sunday” are such “temporal” functions. Under this “value to function” interpretation, besides the temporal location it denotes, the adverbial carries with it some property or qualification of the temporal location. In simple “out of the blue” sentences, such as (34), we normally do not get this second reading. In and of itself, (34) has nothing which will force a typical listener to take tonight as having some additional significant property like also being a holiday, specific day of the week, birthday, or any other important date. But there are some linguistic and pragmatic factors which trigger such a “functional” reading. One is the presence of certain kinds of modal verbs, such as would and should. Thus, compare the following pairs of sentences: (35) (36)

a. Mary would be happy tonight. b. Mary is happy tonight. a. All movie theaters should be closed today. b. All movie theaters are closed today,


The nonmodalized (b) sentences are just like (34), i.e. in order to evaluate them, all we need to know about tonight or today is that the temporal location of the sentence (or, the situation variable) is within the night and day of utterance, respectively. I may utter (35b) and (36b), for example, after watching Mary’s behavior tonight or observing all movie theaters tonight, and without knowing anything special about tonight or today. However, this kind of interpretation does not seem to be sufficient in the case of the (a) sentences. It would be very strange for me, as a speaker, to utter (35a) and (36a), if I know nothing special about tonight or today. In other words, if I utter (35a), it is probably because I know that there is something special about tonight, in virtue of which, or because of which, Mary is happy on this night. And if I say (36a), I probably have in mind something special about today, explaining why all movie houses should be closed. In the case of (35a) this “something special,” what I earlier called the “temporal function,” can be Mary’s birthday, the first night of Mary’s vacation, the night when Mary won the lottery, etc. In the case of (36a), it can be Memorial Day, the day of the annual conference of theater owners, the day where the union of theater owners declared a strike, or any other reasonable function. Again, it is important to emphasize that, unlike what happens with nonmodalized sentences like (34), (35b) and (36b), having in mind such a temporal function seems to be a necessary part of the meaning of (35a) and (36b), without which the sentences cannot be felicitously uttered, and without accommodating such a function, the sentence cannot be fully understood. What supports this claim is the oddness of the (a) answers in the following pieces of discourse: (37) John: Mary would be happy today. Bill: Why? What’s special about today? a. # John: I have no idea. I think this is a regular day. Do YOU think there’s something special? b. John: It’s her birthday/It’s the first night of her vacation. / She won the lottery tonight, etc. (38) John: All movie theaters should be closed today. Bill: Why? What’s special about today?

9

For now, Kamp’s (1971) double indexing method seems appropriate for giving a fully formalized representation of such sentences.


a. # John: Special? I know of nothing special today. Do YOU think there’s something special? b. John: Today is Memorial Day. /The annual conference of theater owners is today/The union declared a strike today/etc.10 In contrast, compare the perfect felicity of both the (a) and the (b) answers in the case of a nonmodalized sentence: (39) John: Mary is happy today. Bill: Why? What’s special about today? a. John: Special? I know of nothing special today. Do YOU think there’s something special? b. John: It’s her birthday/It’s the first night of her vacation/She just won the lottery, etc. The (b) answers in (37) and (38) specify the temporal function of which the night or day of utterance is a value. The (a) answers assert that that no such function is known to the speaker. The infelicity of the latter answers in the case of (37) and (38), vs. its felicity in the case of (39), suggests that unlike what happens in nonmodalized sentences, the presence of the modal expressions would and should forces a “functional” reading of temporal adverbials, or at least makes this reading very prominent.

10

There are two cases where the (a) answers will be felicitous, and thus two exceptions to the claim I make that the functional reading of temporal adverbials is necessarily part of the meaning of the sentence when would and should are present. The first exception, brought to my attention by Fred Landman, is the case of a bet. I may utter Mary would be happy tonight simply as a bet, without thinking there is really something special about today/tonight. The second case is when by uttering (33) and (34) I simply repeat what I heard from somebody else (e.g. (They said on the radio that) All movie theaters should be closed today). In this case too, the (a) answer is felicitous, but in this kind of scenario the speaker of the sentence in fact assumes the role of a listener, who tries to accommodate (or claims s/he does not know) a reasonable temporal function for the sentence. Although both these cases potentially exist, they are not very prominent in an interpretation of sentences like (37) and (38). The prominent and much more natural reading is still one where the temporal function is part of the meaning of such sentences.


4.2 A descriptive generalization The distinction between the default, “indexical” and the “functional” reading of temporal indexical adverbials, described in the section above, can now help us in our first attempt towards understanding the behavior of temporally restricted IS and BP sentences. We can now make the following descriptive generalization: (40) A descriptive generalization a. IS sentences are only compatible with the “functional” reading of temporal indexical adverbials. b. BP sentences are compatible with both the “indexical” and the “functional” readings of temporal indexical adverbials. In other words, we now claim that in order for an IS sentence like An Italian restaurant is closed tonight to be felicitous, it is not enough to know the temporal location of “tonight” relative to the time of utterance. Instead, we also must know some significant property of “tonight,” or in our terms, the temporal adverbial must be understood functionally (e.g. as Italy’s Independence Day, the last week of the fiscal year, etc.). In contrast, BP sentences can be used as generics even if we know nothing about the temporal indexical (e.g. tonight) apart from its temporal location relative to the utterance time. Given that the “indexical” reading is the normal and default reading of such indexical adverbials, whereas the “functional” reading is their marked, nondefault reading, which arises only when certain linguistic or pragmatic factors are present, generalization (36) can immediately explain the general felicity of BP sentences with such temporal indexical adverbials, vs. the general oddness of “out of the blue” IS sentences in this construction (cf. again (8) and (9) above). To extend the argument further, all of the cases described in section (2.2) above, where the status of IS sentences in this construction is improved, are cases where the “functional” reading of the temporal adverbials is made available by various linguistic or pragmatic factors. Thus, for example, the reason the sentences in (10), with the addition of would, should, or the deontic must, are felicitous, is because of the fact, seen in (33) and (34), that the presence of such modals forces the “functional” reading of the temporal indexical, or at least makes it very prominent. As seen in (11b) and (12b), repeated here, this cannot due to the mere modality of these expressions, since the same sentences with


the addition of the epistemic must (and for the same purpose the epistemic may) are again questionable (as before, the epistemic reading is more salient when must is deaccented): (41)

a. ??/* A lion must be very aggressive today. b. ??/* An earthquake must be causing great damage today

The reason for this contrast between would and should on the one hand, and must under its epistemic interpretation, on the other, seems to lie in the fact that unlike would and should, the epistemic must is naturally compatible with the “indexical” reading of temporal indexical adverbials, and does not prefer or force the “functional” reading. This can be seen from the felicity of both the (a) and (b) types of answers to (39), repeated here: (39) John: Mary must be happy today. (epistemic) Bill: Why? What’s special about today? a. John: I have no idea. I just saw her singing and dancing, so I figured that out. b. John: It’s her birthday/It’s the first night of her vacation/She just won the lottery, etc. As seen in (39), sentences with epistemic modals can express two kinds of modalized inferences. The first is the evidentive modalized inference, seen in (a), where the known facts (in this case Mary’s singing and dancing) are taken as evidence for the atomic statement (in this case Mary is happy today). In this case, the speaker need not know anything else about today, i.e. this reading is perfectly compatible with the “indexical” reading of the temporal adverbial. The second kind of inference which epistemic modalized sentences can express are “causal” inferences, as in (b), where the known facts are taken as a cause or reason for the atomic sentence (e.g. Mary is happy because it is her birthday/ because of winning the lottery, etc.). When a temporal indexical is present, it must be understood under its “functional” reading, and the temporal function is taken, in fact, as part of the cause or reason for the atomic sentence. Thus, when a sentence with the epistemic must is uttered “out of the blue,” with no supporting context, the prominent interpretation of the temporal adverbial will be, as usual, the “indexical” one.11 The next “improving factor” for the temporally restricted IS sentences, which can be explained by generalization (40), is the addition


of modifying normative adjectives, as in (13) above. The idea here is that that the addition of such adjectives introduces implicit deontic modality into the sentences, which, like the addition of explicit deontic modals, triggers the “functional” reading of the temporal adverbial. There are also some cases where the “functional” reading of the temporal adverbials, and consequently the improved status of the IS sentences, is triggered simply by an explicit reference to the temporal function (e.g. Italy’s Independence Day), as in the “predictional scenarios” in (15)-(17), above. Finally, in (6) and (7) above, the “functional” reading is triggered by our realworld knowledge, which tells us, for example, that it is an American tradition to eat turkey on a certain holiday (Thanksgiving), or a Catholic habit to go to church on a certain day of the week (namely Sunday). 5. How can existing theories of genericity handle the descriptive generalization? It seems, then, that the descriptive generalization in (36), repeated here, can account in a natural way for the various differences between IS and BP sentences in “temporally restricted generics.” (40) A descriptive generalization a. IS sentences are compatible only with the “functional” reading of temporal indexical adverbials. b. BP sentences are compatible with both the “indexical” and the “functional” readings of temporal indexical adverbials. The next step is to explain this generalization. In the following two subsections I show that no satisfactory explanation of clause (a) and clause (b) of the generalization explicitly or implicitly exists in traditional and more current theories of genericity.

11

See Brennan (1993) for similar claims about the different influence of root and epistemic modals on the interpretation of temporal adverbials.


5.1 Existing theories on the restriction on IS sentences. Or, why are characteristic (I-) genericity and episodicity mutually exclusive? Since most theories represent generically interpreted IS sentences as characteristic, quantificational (I-) genericity, explaining clause (a) of the above generalization comes down to explaining the claims concerning the general incompatibility of such I-genericity and episodicity. The idea here is that when the temporal adverbial is interpreted functionally, the resulting sentence is not really episodic. For example, when “today” is understood as “Italy’s Independence Day,” the sentence An Italian restaurant is closed today is understood as making a generalization about the behavior of Italian restaurants in general on Italian Independence Day. In contrast, when the temporal adverbial is understood indexically, the sentence has to express “episodic I-genericity,” which, by definition, is an illegal creature. Most theories of genericity take the incompatibility of (I-) genericity and episodicity as a primitive property. These theories claim that for a sentence to be considered I-generic, it will necessarily express a regularity, or a habit, and clearly not an episodic generalization. A quotation from Carlson (1995), cited in chapter 1 above, and repeated here, is an example of such a view: I will take it for granted that I know what a generic sentence is— any sentence expressing a generalization—and that the opposing category consists of episodic sentences—which relate specific occurrences. (224) [emphasis mine] Another example of this approach is Chierchia (1995): …Another feature of generics is that they tend to last. When someone has a habit or a disposition, we expect it to occupy a significant portion of his lifespan […] Laws, routines, habits and the like are without doubt tendentially stable in time. This fact should be somehow built into the semantics of Gen. (p. 196) Even if we ignore bracket for the moment the fact that BP sentences can freely express generic statements which do not “tend to last,” it would be better to attempt to derive the suggested “tendential stability” of the generic quantifier which is taken to be associated with IS sentences from


some independent property, instead of simply stipulating it. Two such attempts are discussed in sections (5.1.1) and (5.1.2) below. The first attributes the nonepisodicity of Gen to its binding of the situation variable (and similarity to overt quantificational adverbs), and the second attributes it to its modal nature. 5.1.1 Binding of the situation variable, and similarity of Gen with overt Q- adverbs as explaining the behavior of IS sentences The first approach, taken by e.g. Krifka (1988), attributes the incompatibility with episodic predicates to the (universal) binding of the situation variable by the generic operator. In this approach, the situation variable cannot be at the same time bounded by a (universal) quantifier and understood as located in a specific point in time. In Krifka’s words: Episodic (that is, non-stative) sentences can be plausibly considered as sentences which say something about a specific situation […] Generic sentences, on the other hand, make no claim about specific situations, but about every situation of a certain type. Thus, universal or generic quantification over specific situations can be considered to be the very essence of stativity. (p. 306) Theoretically speaking, however, it is not clear that Krifka’s claim is necessarily correct. We can, in principle, think about “every possible (and contextually relevant) situation, located tonight,” i.e. we can take the property of “being located on the night of utterance” to characterize the type of all situations which we quantify over. More empirically, a prediction of Krifka’s approach is that when combined with episodic predicates (or with the “indexical” reading of the temporal adverbials), generic IS sentences will behave the same as sentences with Q-adverbs (like always or usually), since the latter are also claimed to involve quantification over situations of a certain type. At first sight this last prediction seems to be borne out. For example, both the IS sentence in (41a) and that in (41b), with a definite or quantified subject and an explicit Q-adverb, are infelicitous: (41)

a. # A lion is aggressive today.


b. # Simba/Every lion is always/usually aggressive today. Moreover, when the temporal adverbial is clearly given a functional interpretation, sentences with explicit Q adverbs are as felicitous as their IS counterparts: (42)

a. It’s Sunday. A Catholic is in church today. b. It’s Sunday. Mary is usually/always in church today.

However, concentrating on the similarity between Gen and Q-adverbs has its limits, since there also cases where temporally restricted IS sentences and those with overt Q adverbs differ, as seen from a comparison between the pairs in (43)-(46): (43) (Context: Last night the labor union suddenly declared a strike) a. A union member is on strike today. b. # A union member is always/usually on strike today. # John is always/usually on strike today. (44) (Context: Kenneth Starr admitted this morning that President Clinton was basically telling the truth regarding Monica Lewinsky) a. A true American is proud of his President today. b. # A true American is usually/always proud of his President today. # John is usually/always proud of his President today. (45) (Context: Yesterday was Thanksgiving) a. A true American was eating turkey yesterday. b. # A true American was usually/always eating yesterday. # John was usually/always eating Turkey yesterday. (46) (Context: Yesterday was Italy’s Independence Day) a. ? A decent Italian restaurant was closed yesterday. b. # A decent Italian restaurant was usually/always closed yesterday. # Mama Mia was usually/always closed yesterday. In all of these examples, the sentences with overt Q advs. (with a definite or an IS subject) are clearly much worse than their IS counterparts.12 This is unexpected if the only reason for the behavior of IS sentences is the similarity of Gen with overt Q advs. There is a


second problem in the hypothesis that it is the similarity with Q advs. (namely the quantification over situations) which make IS sentences incompatible with the temporal location reading of the temporal indexical. An implicit prediction of this hypothesis is that sentences with an overt Q-adv. will behave the same w.r.t. their compatibility with the temporal modifier, no matter whether their subject is an IS or a BP noun phrase. However, this prediction is not borne out in reality. As seen in (47), when these two kinds of sentences are uttered in an “inductive scenario” context, the BP sentences (in (47a)) are still much better than the IS ones (in 47b)): (47) (It is Saturday night and my friend wants to buy some bread and milk) a. ?Well, you can try. I noticed that grocery stores are sometimes open tonight. b. # Well you can try. I noticed that a grocery store is sometimes open tonight. This difference is unexpected given the hypothesis that the source of incompatibility of IS sentences with temporal adverbials is only the Qadvsnature of the Gen operator binding the IS. Thus, although Gen and frequency adverbs are similar in binding the situation variable, the differences noted above suggest that the incompatibility of IS sentences with the “indexical” reading of temporal adverbials cannot be explained in terms of this similarity. Instead, there seems to be something special about IS sentences which is responsible

12

There is one reading where the (b) sentences are better, namely when they mean something like “A decent Italian restaurant was closed on most of the situations tonight” or “A true American was proud of his President for most/all/ some of the time yesterday.” This, however, is not the reading we are interested in, and the IS sentences in (43a-46a) do not need to be interpreted this way in order to be felicitous. In addition, there are also minimal pairs (of IS sentences with and without an overt Q-adv.) where there is no way to “save” the infelicity of the latter, as (i) and (ii):

(i) (ii)

A watermelon costs NIS 5 today. # A watermelon always/usually costs NIS 5 today

I discuss such IS sentences below.


for their behavior, and which should be separated from the mere quantification over situations. 5.1.2 Modality/law-likeness as explaining the behavior of IS sentences Another approach (e.g. Krifka (1987), Dobrovie-Sorin & Laca (1996)) argues that the incompatibility of I-genericity and episodicity results from the incompatibility between the fact that Gen is modalized, and the fact that episodic, or “dynamic,” predicates express accidental properties. Although I agree that there is indeed some conflict between the lawlikeness of IS sentences and the “accidentalness” of episodic predicates, it is not merely “modality” in itself which is incompatible with episodicity. This can be seen from a closer examination of both the BP and the IS sentences presented in section (1.1) above. First, these BP sentences are compatible with an episodic predicate (and the “indexical” reading of the temporal adverbials) although they are clearly modalized as well, as can be seen from their counterfactual supporting property. In addition, merely being under the scope of a modal operator is not itself sufficient to contrast with episodicity. In other words, as claimed above, there is nothing theoretically wrong with an assertion like (48): (48) “Every contextually relevant situation whose temporal location is on the night of utterance, involving any Italian restaurant in all accessible worlds, is a situation where this Italian restaurant is closed in these accessible worlds.” In this case only the world axis varies, whereas the time axis stays constant. As claimed in Gamut (1991), this possibility clearly exists in natural language, as shown by the felicity of counterfactuals such as (49): (49) If I hadn’t gone to the party tonight, I wouldn’t have met you. Here too we look at “tonight in all possible worlds” (those maximally similar to ours), where I didn’t go to the party. As discussed earlier, IS sentences are infelicitous under such a reading, and additional information about the properties of the temporal modifier is needed in order for such sentences to be acceptable. Thus, it is not enough to derive the infelicity of the sentences in (8), and the


general nonepisodicity of I-generic sentences, from the mere modality of Gen. Something else must account for these facts. I will claim below that this additional element is the special type of modality of Gen, as expressed in the definition of the accessibility relation of IS sentences, developed in the chapters above. 5.2 Genericity theories on the behavior of BP sentences: why can BP sentences freely express “episodic genericity”? As shown in chapter 1, most theories of genericity give BP sentences a quantificational-modal representation which is equivalent to the representation of their minimally contrasting IS counterparts, and reserve the Proper KindPredication reading for BP sentences where a quantificational reading is impossible. These theories, then, cannot explain why BP sentences are compatible with episodic or temporally restricted predicates but their minimally contrasting IS counterparts are not. One way to explain this difference is to give the two constructions a different semantic representation. This is the approach taken by Krifka (1987), and more recently by Dobrovie-Sorin & Laca (1996).13 Both studies claim that BP sentences, even those which can be paraphrased as quantificational, and which combine with non-kind level predicates (like (50a)), can (or should) be represented as nonquantificational, Proper-Kind-Predication structures, where a property

13

As mentioned in previous chapters, Cohen (1999, 2001) also claims that IS and BP sentences are semantically different. His intuition, which is very similar to the one I develop in this work, is that IS sentences express rules, whereas BP sentences express “descriptive generalizations.” Formally, Cohen takes BP sentences to be quantificational structures, headed by a probability operator, and IS sentences to be non-quantificational structures, expressing primitive rules. Cohen does not deal with the incompatibility of IS sentences with episodic predicates, but intuitively it can be claimed (as does Chierchia (1995)) that rules should be “tendentially stable” and not episodic. As for temporally restricted BP sentences, Cohen claims that such sentences cannot exist, since episodic information is incompatible with the semantics of the probability operator which he takes to be associated with BP sentences. In this chapter, however, I gave many examples of such “episodic generic” BP sentences. A discussion of (parts of) Cohen’s theories, and a comparison to the present theory, can be found in chapters 2 and 3 above.


applies directly to the kind, just like the classic Proper Kind-Predication structures in (50b): (50)

a. Dogs have four legs. b. Dodos are extinct.

According to this approach, the difference between the temporally restricted IS and BP in (51a) and (51b), respectively, should be explained in basically the same way as the difference between the IS and the clear proper kind BP sentence in (52a) and (52b), respectively: (51) (52)

a. # An accountant is busy today. b. Accountants are busy today. a. */# A whale is rare. b. Whales are rare.

At first sight this idea seems to be helpful in explaining the properties of temporally restricted BP sentences like (51b), specifically the fact that they can occur with the “indexical” reading of temporal adverbials. The idea here is that, unlike the Gen operator, which is incompatible with episodic properties, Proper Kind-Predication structures, which are not associated with Gen, are fine with “indexical” temporal restriction. What seems to supports such an idea is the independently observed fact (e.g. Krifka (1987)) that clear Proper KindPredication structures can always be temporally restricted, as seen from (53): (53) a. Rhinos were declared an endangered species on June 12th, 1986. b. The dodo was/Dodos became extinct in March 1768. c. The rat/Rats reached Australia in October 1856. In all three cases, the interpretation of the sentences is not quantificational, i.e. they do not mean “All/most rhinos were declared an endangered species on June 12th,” “All/most dodos became extinct in March 1768,” or “All/most rats reached Australia in October 1856.” Instead, these sentences only have a Proper Kind-Predication reading. Consequently, Krifka (1987), Dobrovie-Sorin & Laca (1996), and other scholars represent such sentences as Proper KindPredication and not as involving Gen. We have, then, independent evidence that Proper Kind-Predication structures can be freely temporally restricted and episodic, i.e. they can


be compatible with the “indexical” reading of the temporal adverbial. This may support the idea that the similar compatibility of BP sentences like Italian restaurants are closed tonight with the “indexical” reading of the temporal adverbials can be also attributed to a Proper KindPredication interpretation, and to the lack of the Gen operator. Despite its intuitive appeal, there are two main problems with this suggestion. First, if we analyze the BP sentences in (9) as non-quantificational, Proper Kind-Predication structures, we should be able to explain the fact that, unlike the classical Proper Kind-Predication sentences in (50b) and (53), these sentences share several properties with quantificational, I-generic sentences. Thus, as mentioned in the previous sections, the BP subjects in (51b) (and 9 above) are interpreted quantificationally, as “all/ most Italian restaurants/accountants/earthquakes,” etc. In addition, these sentences obey Cohen’s (1999a) “homogeneity constraint,” which is also a property of quantificational generics, and which does not seem to be at all relevant for classical Proper KindPredication sentences as in (50b) and (53). It is not clear how a Proper KindPredication analysis of BP sentences like (51b) would account for these facts. Furthermore, this approach cannot provide a satisfactory explanation for the difference between sentences like (51a) and (51b), i.e. for the fact that BP sentences freely combine with the “indexical” reading of the temporal adverbial, but IS sentences do not. Attributing the felicity of (51b) to a non-quantificational semantic structure is not in itself sufficient to explain this difference, since, as we saw above, it is not the mere quantificational nature of IS sentences which causes infelicity with temporally restricted predicates. Neither can the difference between (51a) and (51b) be explained in terms of modality of Gen, vs. the nonmodality Proper Kind-Predication, since (a) as we saw above, modal quantification in general is not incompatible with episodicity, and (b) unlike the clear Proper Kind-Predication sentences in (50b) and (53), BP sentences like (51b) are clearly modalized (as indicated by their counterfactual-supporting nature). Thus, merely taking Italian restaurants are closed tonight to be a Proper Kind-Predication structure is not only problematic, but cannot in itself improve our understanding of its felicity relative to the behavior of its IS counterpart. In other words, even if we decide to represent such sentences as non-quantificational, this in itself would not account for why they differ from the minimally contrasting #An Italian restaurant is closed tonight.


Instead, my claim is that the descriptive generalization above, concerning the difference between IS and BP sentences, should be explained as a special case of the more general difference between the two kinds of modalized, non-accidental generalizations expressed by these sentences. What is at stake here, then, is the type of modality, or accessibility relation, associated with IS and BP sentences, and not merely whether the semantic structure of the sentence is quantificational or not.14 This idea is developed in the next section. 6. Temporally restricted IS and BP sentences: the solution 6.1 Similarities between the behavior of IS and BP sentences in temporally restricted and classical generic sentences Let me reintroduce the descriptive generalization concerning the behavior of temporally restricted IS and BP sentences: (40) Temporally restricted IS and BP sentences: A descriptive generalization a. IS sentences are only compatible with the “functional” reading of temporal adverbials e.g. where tonight is understood as “a night of a certain sort”). b. BP sentences are compatible with both the “indexical” and the “functional” readings of temporal adverbials (e.g. both with “the night when the sentence is uttered” as well as with “a night of a certain sort”).

14

In chapter 6 below, I examine the possibility that it is the difference between the indefinite interpretation of IS noun phrases vs. the (possible) kind-referring interpretation of BPs which triggers the presence of the two kinds of accessibility relations. Thus, it is possible that BP sentences should be represented, after all, as proper-kind predication structures, but under such an interpretation they continue to involve universal modal quantification over situations. At this stage I still follow the widely held claim that the basic semantic structure of IS and BP sentences is the same, and take the accessibility relation to be the only difference between them.


In the section above I showed that this generalization cannot be satisfactorily explained by existing theories of genericity. Now I will claim that it can be explained as a special case of the more general difference, developed in the chapters above, between the two kinds of generalization expressed by IS and BP sentences. In other words, the generalization in (40) is a special case of the generalization in (54): (54) IS and BP sentences: a. IS sentences can only express “in virtue of” generalizations (where a contextually chosen ^S property, associated with the property denoted by the IS subject (^P), is taken to be responsible for having the ^Q property, denoted by the VP). b. BP sentences can express both “in virtue of” or “descriptive” generalizations (where the connection between ^P and ^Q is simply being asserted to be non-accidental, without specifying an “in virtue of” factor). The generalization in (40) is based on comparing the behavior of IS and BP sentences in temporally restricted sentences, and the generalization in (54) is based on the behavior of these two kinds of sentences in classical generic constructions (primarily with individual-level and habitual predicates). In chapter 1 I pointed out a number of similarities between the behavior of IS and BP sentences in both constructions. To reiterate, in both temporally restricted, and classical generic constructions, BP sentences are more easily interpreted as generic than are minimally contrasting IS sentences. In both constructions, context can improve the status of problematic “out of the blue” IS sentences (i.e. those with a prominent existential reading, which are infelicitous as generic), as seen again in (55b) and (56b): (55) a. An Italian restaurant is closed tonight. (salient existential reading) b. It’s Italy’s Independence Day. An Italian restaurant is closed tonight. (the generic reading is more salient) (56) a. A Norwegian student whose name ends with ‘s’ or ‘g’ wears thick green socks. (salient existential reading) b. There are interesting traditions in Norway concerning names, professions and clothing. For example, a Norwegian


student whose names end with ‘s’ or ‘g’ wears thick green socks. (the generic reading is more salient) Third, in both the temporally restricted and the classical generic constructions, not every kind of contextual support can improve the felicity of IS sentences. Specifically, in both constructions a “pure inductive” reasoning context is inappropriate as supporting the generic reading of the originally “infelicitous as generic” IS sentences, while BP sentences are freely interpreted as generic in such inductive contexts. This can be seen again when both kinds of sentences are embedded under the verb notice: (57) a. Walking down the street I noticed that a small Italian restaurant is closed today. (infelicitous as generic) b. Walking down the street I noticed that small Italian restaurants are closed today. (fine as generic) (58) a. While sitting in Kew Gardens I noticed that an old man wearing a red shirt bends under his bench whenever an airplane flies overhead. (infelicitous as generic) b. While sitting in Kew Gardens I noticed that old men wearing red shirts bend under their benches whenever an airplane flies overhead. (fine as generic) Given these similarities between the behavior of IS and BP sentences in the two kinds of constructions, explaining this behavior in a unified manner seems desirable. 6.2 The intuition: IS sentences with indexically interpreted temporal adverbials express “unreasonable generalizations” The behavior of IS and BP sentences in temporally restricted and classical generic constructions are similar in yet another way: In both constructions the generalizations expressed by infelicitous IS sentences seem to be incompatible with our real-world knowledge in a very similar, and specific way. Let me explain. In chapter 2 I examined two kinds of IS sentences which are infelicitous as generic, manifested in (59a) and (60a) (cf.


again the perfectly felicitous minimally contrasting BP sentences in (59b) and (60b): (59) (60)

a. # A small lion called Simba likes cornflakes for breakfast. b. Small lions called Simba like cornflakes for breakfast. a. # An uncle is addicted to marshmallows. b. Uncles are addicted to marshmallows.

I concentrated on the intuition that such sentences express generalizations, which, given our real-world knowledge, are taken to be highly unreasonable, and devoted much space to formally clarifying what “highly unreasonable generalizations” means, as far as IS sentences are concerned. The idea here was that the truth conditions of IS sentences systematically contain a property which functions as the “in virtue of” property for the whole generalization. However, the property which plays the role of this “in virtue of” property (which I call ^S) cannot be any arbitrary property, but must meet two real-world-knowledge requirements: being associated with the ^P property (denoted by the subject), and reasonably leading to properties of the sort of ^Q (denoted by the VP). IS sentences like (59a), which have subjects denoting “extremely unnatural classes,” are infelicitous as generic because the first requirement on ^S cannot be met: our real-world knowledge does not associate with such ^Ps any property at all (and in fact, this is why such properties are felt to be so “unnatural”). IS sentences like (60a) are infelicitous as generic because the second requirement on ^S (which has a presuppositional status) is not met: With such sentences, although there are properties associated with the ^P property, none of these properties is taken to be reasonably leading to properties of the sort of ^Q. In our example in (60a), although there are many properties we may associate with being an uncle, none is taken to be of the sort which leads to things like “being addicted to marshmallows.” In contrast to IS sentences, BP sentences can also felicitously express highly unreasonable generalizations, as seen in (59b) and (60b). The claim above was that this is because they can express not only “in virtue of” generalizations, where the pragmatic requirements on ^S have to be met, but also “descriptive” generalizations, asserting simply that the generalization is not accidental, where no such real-world knowledge requirements are imposed (or more precisely, presupposed). Let us now return to temporally restricted IS and BP sentences. My main claim is that IS sentences are incompatible with the “indexical”


reading of temporal adverbials like tonight because with this reading the generalization expressed is also taken to be “highly unreasonable.” Moreover, such generalizations are “unreasonable” in the same way that IS sentences like (58a) (“An uncle is addicted to marshmallows”) are. In both cases the second requirement on the “in virtue of” property cannot be met. To see why this is the case, consider again the temporally restricted IS sentences in (61): (61)

#An Italian restaurant is closed tonight.

On the “indexical” reading of “tonight,” (61) can be very roughly paraphrased as “In all accessible worlds, every Italian restaurant is closed on the night when the sentence is uttered.” But this kind of statement is indeed highly unreasonable: there is nothing in our realworld knowledge about Italian restaurants which will reasonably lead every such restaurant to be closed on an interval, when all we know about that interval is that it falls on the night when the sentence is uttered. In other words, our general real-world knowledge about Italian restaurants is totally independent of when John, Mary, Bill or I utter sentences about such restaurants. In other words, sentences like (61) are infelicitous as generic since with the “indexical” reading of the temporal adverbial, the second realworld-knowledge requirement on the “in virtue of” ^S property cannot be met: Although the IS subject “Italian restaurant” denotes a “natural property,” i.e. although our realworld knowledge associates many properties with “being an Italian restaurant” (e.g. being popular, being run by Italians, serving pasta and pizza, having Italian names, etc.), none of these properties can be thought of as a reasonable causers for things like “being closed on the night which is temporally located when the sentence is uttered.” In contrast, the second requirement on the “in virtue of” ^S property can be met if the temporal adverbial has its functional reading, i.e. if we have in mind some property of “tonight” besides its temporal location relative to the utterance time. For example, if “tonight” is understood as the night of Italy’s Independence Day, then we can think about properties associated with being an Italian restaurant which can reasonably lead every such restaurant to be closed on Italy’s Independence Day, or another day of that sort. Intuitively this is the reason for the relative felicity of IS sentences under the “functional” reading of the temporal adverbial. Unlike IS sentences, the descriptive generalization above tells us that BP sentences are compatible with both the “indexical” and the


“functional” readings of the temporal adverbial (i.e. Italian restaurants are closed tonight is felicitous as generic even if “tonight” simply means “the night of utterance” and we know nothing else about it). The intuitive reason for this can be seen clearly now: Such sentences have a natural descriptive reading, naturally based on inductive evidence, where they merely assert that the generalization is non-accidental, and where, crucially, there are no presupposed restrictions, guaranteeing that the generalization is a reasonable one in terms of our real-world knowledge about the denotations of the subject and VP. Such a sentence asserts (roughly) that being closed on the night when the sentence is uttered holds of all actual Italian restaurants, and moreover that this is a pattern, i.e. a non-accidental fact, so we would also expect all nonactual Italian restaurants in similar possible circumstances (i.e. in similar possible worlds) to be closed on this night. In the following sections I propose a more precise form for these intuitive explanations. 6.3 Truth conditions of temporally restricted present tense IS and BP sentences The intuition developed above attributes the difference between IS and BP sentences (in terms of their compatibility with specific temporal adverbials) to the presence or absence of an “in virtue of” ^S property in the truth conditions, and especially to the presence or absence of a presupposition limiting the choice of ^S relative to the ^Q property denoted by the VP. Before we can show more formally the effect of the presence or absence of this presupposition, we must first clarify the role that the temporal adverbial (e.g. tonight) plays in the truth conditions of both sentences. Specifically, the question which must be answered is what this adverbial really modifies. Bear in mind that, bracketing the difference in accessibility relation, the truth conditions developed in the chapters above for standard IS and BP sentences like A dog has four legs, and Dogs have four legs are the ones in (62): (62) Basic truth conditions for IS and descriptive BP sentences: A BP or an IS sentence is true at <w, I> iff


Paraphrase: In all accessible worlds w′ there is a superinterval of the present I′, s.t. any relevant and non-exceptional situation involving any relevant and non-exceptional P individual in w′, which is located at I′ is a Q situation involving this individual in w′. Applying these truth conditions now to temporally restricted generics like (63) or (64), with either the “indexical” or the “functional” reading of the temporal adverbial tonight, we seem to face two primary options. (63) (64)

Italian restaurants are closed tonight. # An Italian restaurant is closed tonight.

At first sight, one may take tonight to modify directly the denotation of the VP, namely the ^Q property (closed), so we end up with a complex ^Q property (closed tonight). The motivation for this option lies in the linear order of the temporal adverbial, right after the adjective closed. In this option, the truth conditions of both the IS and descriptive BP sentence, on either reading of tonight will come out as in (65): (65) Basic truth conditions of temporally restricted generics, Option A:

Paraphrase: In all accessible worlds w′ there is a superinterval of the present I′, s.t. any (relevant and non-exceptional) situation involving any (relevant and non-exceptional) Italian restaurant in w′, which is located at I′, is a situation where this Italian restaurant is closed, and is located tonight Bracketing for a moment the irrelevant and exceptional entities, (65) says, very roughly, that any possible situation involving any possible Italian restaurant is a situation where such an Italian restaurant is closed tonight. But this seems to be the wrong interpretation of both (63) and (64) (on either the “indexical” or “functional” readings of tonight). Instead, what we mean when we utter (63) and (64) is, very roughly, that


any possible situation tonight, involving any possible Italian restaurant, is a situation where such an Italian restaurant is closed. In other words, in uttering (63) and (64) we are not really interested in the properties of Italian restaurants in all possible situations—and say that in all these situations any Italian restaurant is closed tonight. Rather, we are only interested in the properties that such restaurants have in all “tonight situations,” and say that in all such “tonight situations” all these restaurants are closed. Situations on other nights are simply irrelevant to the truth of the generalization in the first place. More precisely, in the true interpretation of temporally restricted generics, the temporal adverbial is not a direct modifier of the ^Q property, and does not form a complex property with it. Rather, such adverbials seem to temporally modify the tense of the whole sentence (the interval where the sentence is evaluated). The truth conditions of (63) and (64), then, are more likely to look like (66), than like (65): (66) Basic truth conditions of temporally restricted generics, Option B:

Paraphrase—In all accessible worlds w′ there is a superinterval of the present I′ which is located tonight, s.t. any (relevant and nonexceptional) situation involving any (relevant and nonexceptional) Italian restaurant in w′, located at I′, is a situation where this Italian restaurant is closed. To conclude, tonight in (63) and (64), does not directly modify the ^Q property in the nuclear scope, but rather the tense of the sentence—i.e. the (super)interval I′ where the universal generalization is asserted to hold. Notice, however, that because of this temporal modification of I′, the situation variable in both the restrictive clause and the nuclear scope is also understood as temporally located “tonight.” I.e., the temporal adverbial does eventually modify, in an indirect way, the ^Q property in (63) and (64), so we end up interpreting it as closed tonight. We can now turn to the compatibility of IS and descriptive BP sentences with the two readings of temporal adverbials like tonight.


6.4 IS sentences with the “indexical” and “functional” readings of the temporal adverbials 6.4.1 The “reasonable causation” presupposition with temporally restricted IS sentences Let me start first with IS sentences. The general truth conditions of such sentences, developed in the chapters above, are repeated here: (66) An IS sentence is true at <w, I> iff

Given our discussion in the section above, the truth conditions of the temporally restricted An Italian restaurant is closed tonight are (67) on the indexical reading of the temporal adverbial, and (68), on its “functional” reading: (67) An Italian restaurant is closed tonight (indexical reading) is true at <w, I> iff

Paraphrase: An Italian restaurant is closed tonight (“indexical” reading) is true at <w, I> if and only if there is a property ^S, s.t. in all worlds w′ where every Italian restaurant has S (and ^S is associated in w with being an Italian restaurant), there is a superinterval of the present I′, which is temporally located on the night of utterance, and every (relevant and non-exceptional) situation, involving any (relevant and non-exceptional) Italian restaurant in w′, which is located at I′ (i.e. located on the night of utterance) is a situation where this Italian restaurant is closed.


(68) An Italian restaurant is closed tonight (“functional” reading) is true at <w, I> iff

Paraphrase: An Italian restaurant is closed tonight (“functional” reading) is true at <w, I> if and only if there is a property ^S, s.t. in all worlds w′ where every Italian restaurant has S (and ^S is associated in w with being an Italian restaurant), there is a superinterval of the present I′, which is temporally located on the night of Italy’s Independence Day, and every (relevant and non-exceptional) situation, involving any (relevant and non-exceptional) Italian restaurant in w’, which located on I′ (i.e. located at the night of Italy’s Independence Day) is a situation where this Italian restaurant is closed. We can now turn to the presupposition on the ^S property which is relevant to the difference between the felicity of the IS sentence given the two readings of tonight. To reiterate, this presupposition guarantees that at least one of the ^S properties associated with ^P is taken to be a “reasonable causer” to properties of the sort of ^Q: (69) “Reasonable causation” presupposition of IS sentences ^S is s.t. there is a good possibility, i.e. there is a world w′, (Kratzerially accessible from w, and closest to an ideal version of our world) s.t.

(where in the ideal world some general principle which works in our word is not inhibited or blocked by any other, less general, factor). A direct application of this presupposition to (64) (An Italian restaurant is closed tonight) will give us (70):


(70) Presupposition of (64) (An Italian restaurant is closed tonight) ^S is s.t. there is a world w′ (defined as in (69)) where

Paraphrase: There is a good possibility (relative to some general principle in our world) that either every relevant situation involving every Italian restaurant with S is a closed situation, or that every such situation involving every Italian restaurant with S is not a closed situation. Note, however, that (70) does not appear to be the correct presupposition of (64), with either the “indexical” or the “functional” readings of tonight. This is because when we check whether (64) makes a “reasonable generalization” or not, we are really not interested in any contextually relevant situation involving an Italian restaurant (as (70) says), but only with those relevant situations which are located tonight. Put differently, the presupposition in (70) seems to check whether the connection between being an Italian restaurant with ^S and being closed in general is a reasonable, not-far-fetched connection, whereas we are really interested in the reasonableness of the connection between being an Italian restaurant with S and being closed tonight. One way to fix this problem is to require that in cases of temporally restricted sentences like (64), the specification of the temporal location is part of being considered a contextually relevant situation. In other words, when we define the relation C(d, s, w′) for the presupposition of a sentence like An Italian restaurant is closed tonight, we consider only those situations which are located on the appropriate night. Those situations which are located at other times are simply irrelevant for the presupposition. Similarly, when we define this relation for a sentence like An accountant is busy this week only those situations located on the appropriate week are considered relevant.15 We can now draw a distinction between the “indexical” and the “functional” readings of the temporal adverbial. Consider again An Italian restaurant is closed tonight. With the “indexical” reading of tonight, the presupposition in (70) is paraphrased as in (71): (71) Presupposition of (64) under the “indexical” reading of tonight


There is a good possibility that either every contextually relevant situation involving any Italian restaurant with S and which is being located on the night of utterance is a closed situation, or that every such situation involving every Italian restaurant with S is not a closed situation. Given our real-world knowledge of Italian restaurants, this presupposition is not met: being closed on the night where the sentence An Italian restaurant is closed tonight is uttered by some speaker, or not being closed on such a night cannot be reasonably derived from any property we generally associate with Italian restaurant. This is because, again, the knowledge, stereotypes, norms, etc. we have about such restaurants are not reasonably connected to the time when an arbitrary speaker may choose to utter this sentence. If being located on the night of utterance is the only thing we know about “tonight,” then “being closed tonight” is simply too disconnected from our general real-world knowledge about Italian restaurants, and there is no general process in our world which will help us reasonably connect the “being closed tonight” with “Italian restaurants” in a not-too-far-fetched way. This seems to be the source of the feeling that, with this reading of tonight, the sentence expresses a highly unreasonable generalization or rule, and consequently, this is the source of its infelicity (as generic) with this reading. In contrast, with the “functional” reading of the temporal modifier, the presupposition in (70) can be paraphrased as in (72): (72) Presupposition of (70) under the functional reading of tonight There is a good possibility that either every relevant situation involving any Italian restaurant with S and being located on the

15

Another way to solve this problem is to redefine the C variable in the presupposition on all IS sentences (in general) so that it systematically makes reference to the temporal location of the interval, in addition to worlds (even in case of no temporal modification), as in (i):

(i) ^S, is s.t. there is a good possibility that, i.e. there is a world w′ (defined as in (68))


night of Italy’s Independence Day is a closed situation or that every such situation is not a closed situation.

In contrast to the presupposition in (71), this presupposition can be met. In other words, among the properties we associate with being an Italian restaurant, we can find properties which may lead the restaurants—in a not too far fetched way—to being closed on the night of Italy’s Independence Day, or to being open on such a night (e.g. being owned by Italians, who always celebrate Italy’s independence day by closing their business, or being popular places especially on Italy’s independence day, etc.). This seems to be the reason for the fact that under the “functional” reading of tonight, (64) is felt to express a more reasonable generalization (although not necessarily a true one), and consequently for the relative felicity of this, and similar, temporally restricted IS sentences. 6.4.2 Support for the proposed so the proposed solution Two pieces of data support the line of argumentation presented above. First, there are cases where even under the “functional” reading of the temporal adverbial like tonight, where we have in mind some property of the relevant interval apart from the fact that it is located on the night of utterance, IS sentences such as (64) are still infelicitous as generic. Compare, for example, (73), with the much odder-sounding (74): (73) It’s Italy’s independence day. An Italian restaurant is closed tonight. (74) Today is my aunt’s parrot’s birthday. # An Italian restaurant is closed tonight.

In this way, in temporally modified sentences like An Italian restaurant is closed tonight the contextually relevant situations are automatically defined as located on the night of utterance (under the “indexical” reading of tonight), or located at e.g. the night of Italy’s Independence Day (under the “functional” reading of tonight).


In both cases tonight is understood functionally, as a day of a certain sort. The reason (74) is worse than (73) is that even under the “functional” reading the IS sentence still expresses an “unreasonable generalization.” More precisely, among all properties associated with being an Italian restaurant, there is no property which, given our realworld knowledge, leads in a not-too-far-fetched way to being closed on the day where my aunt’s parrot has its birthday, or to not being closed on such a day. In other words, the presupposition in (75) is not met. (75) Presupposition of (74) under the functional reading of tonight There is a good possibility that either every relevant situation involving any Italian restaurant with S and being located on the night where my aunt’s parrot has his birthday is a closed situation, or that every such situation is not a closed situation. Note that the only way to improve the status of (74) is to imagine a farfetched story which connects the fact that this is the parrot’s birthday with the fact that all (relevant and non-exceptional) Italian restaurants are closed tonight, or all Italian restaurants are not closed tonight (perhaps my aunt owns all the relevant Italian restaurants, and all the staff has been invited to the parrot’s birthday party, etc.). Without inventing such a story, the sentence is very odd. A second piece of data in support of the argumentation above is the fact that there are some rare cases where temporally restricted IS sentences are perfectly felicitous under the “indexical” reading of the temporal adverbial. An example is (76): (76)

A watermelon costs NIS 5 today.

In order for (76) to be felicitous we need not interpret “today” as “a day of this kind,” i.e. we need not know any relevant properties about today which are related to watermelons, apart from the fact that it is the day when the sentence is uttered. Moreover, (76) can be felicitously uttered in “purely inductive” contexts (e.g. by someone coming back from the local market, having checked the prices of different fruits and vegetables). The reason for the felicity of (76) seems to lie in our real-world knowledge about the verb “cost,” and the way prices of fruits and vegetables go up and down. The presupposition of (76) is (77): (77) Presupposition of (76), “indexical” reading


There is a good possibility that either every relevant situation, located on the day of utterance, and involving a watermelon with S is a situation where the watermelon costs NIS 5, or that every such situation involving any watermelon with S does not cost NIS 5. This presupposition can be rather easily met since we know from our realworld knowledge about watermelons and other produce that the prices of these entities regularly change from day to day. In other words, among the properties we associate with watermelons, one property is something like “being sold in the markets, and having prices which change from day to day”—properties which can reasonably lead to “being sold for a specific price (e.g. NIS 5) on an arbitrary day.” Thus, even if we know nothing more about “today” than its being the day of utterance, a sentence like (76) does not express a “highly unreasonable” generalization, and is felicitous. 6.5 Why BP sentences are compatible with the default, “indexical” reading of the adverbial Finally, the reason for the general felicity of temporally restricted BP sentences, under the “indexical” reading of the temporal adverbial, is the fact that they can express descriptive generalizations, where no realworld knowledge about the connection between the P and (temporally modified) Q property is presupposed. The truth conditions of (63), repeated here as (78a), for example, with the “indexical” reading of tonight, are (78b): (78) a. Italian restaurants are closed tonight. b. “Italian restaurants are closed tonight” is true at <w, I> iff

Paraphrase: In all worlds similar enough to our world now, there is a superinterval of now I′, which is located on the night of utterance, and every (relevant and non-exceptional) situation, involving any (relevant and non-exceptional) Italian restaurant in w′, which is located on I′ (i.e. located on the night of utterance) is a situation where this Italian restaurant is closed.


There is nothing in these truth conditions, and nothing in anything presupposed by (78a) which is problematic. Sentence (78a) is not required to express a “reasonable generalization,” given our real-world knowledge. It merely asserts (likely based on observing a large number of Italian restaurants which are closed tonight) that being closed on the night of utterance is not an accidental property of Italian restaurants, but is something we would expect to find with non-actual Italian restaurants as well (e.g. with this Chinese restaurant had it been, counterfactually, a (non-exceptional) Italian restaurant). 6.6 Actual times, actual worlds and nonaccidentalness: on the double nature of temporally restricted BP sentences We can now also explain the “double nature” of temporally restricted BP sentences like (9a), discussed in section (2) above and repeated here as (79a), which distinguishes them from standard generics like (79b): (79)

a. Italian restaurants are closed tonight. b. Birds fly.

To reiterate, (79a) is like (79b) in supporting counterfactuals, but unlike (79b) in that it necessarily generalizes about actual Italian restaurants existing tonight. Whereas (79b) can be true even if all actual birds happen to lose their flying ability, and felicitous even if right now there is no bird in the actual world, (79a) is false if no actual Italian restaurant is closed tonight, and infelicitous if no Italian restaurant exists in the actual world at the present interval. Unlike standard generics like (79b), then, temporally restricted generics express generalizations which are at the same time both “nonaccidental” and “actual, i.e. which generalize over possible individuals, but also necessarily about individuals in the actual world and time. The difference between (79a) and (79b) can be explained now in the following way. Given the truth conditions for descriptive BP sentences developed in the chapters above, both (79a) and (79b) generalize over all (relevant and nonexceptional) P individuals in a set of possible worlds which has the actual world as a member, at some interval I′ surrounding the time of utterance, I. In addition, both presuppose the existence of P individuals in the initial and final subintervals of I′.


In (79b), however, there is no specification as to the length of I′, and it can be very long. Thus, as shown in chapter 3, a BP sentence like (79b) can be perfectly felicitous even if at the present there are no birds in the world (as long as there are birds in both the past and the “normal future” of ), and perfectly true even if all or most birds existing at the present do not fly (as long as they do not constitute the majority of birds in the actual world at the surrounding interval I′). In contrast, in (79a), the length of I′ is explicitly specified, and crucially, it is very short. The initial and final boundaries of I′ cannot exceed tonight, i.e. the night of utterance. Thus, (79a) presupposes that Italian restaurants exist in the actual world on the night of utterance, and is false if none of the Italian restaurants are closed on the night of utterance. Thus, due to the presence of explicit specification of I′ in (79a), and the absence of such specification in (79b), the former is understood as necessarily generalizing over actual and present individuals (e.g. Italian restaurants which exist in the actual world at the time of utterance), whereas the latter is not. But crucially, this does not turn (79a) into nongeneric, since it does not generalize only over actual individuals, but also over possible ones (e.g. possible Italian restaurants), as indicated by its counterfactual supporting properties. This leads to a more general conclusion about generics. Condoravdi 1997 writes: “Generic generalizations have two hallmarks that distinguish them from actual, universal generalizations: they are not sensitive to the way things actually are but they are sensitive to the alternative ways things might be or ought to be” (p. 7, emphasis in the original). Contra Condoravdi, I believe that an integral part of the semantics of descriptive generics (of which one reading of BP sentences is an example) is that they are necessarily sensitive to “the way things actually are,” sometimes even sensitive to the way things are in the present (as in the temporally restricted generics). I believe that the core characteristics of generics, which distinguish them from actual statements, is not whether the generalizations they express cover actual entities, and not even whether these generalizations are asserted to be temporary, but rather whether these generalizations are not limited to actual entities, and also cover individuals in possible circumstances (which may or may not be temporally restricted), which are appropriately related to the actual world.


7. Conclusion In this chapter I presented new data indicating that contrary to widely held claims, there are temporally restricted/episodic I- generic sentences, and that the differences between temporally restricted IS and BP sentences cannot be satisfactorily accounted for by existing theories of genericity. Specifically I showed that these differences cannot be explained with a simple D/I genericity distinction, and that the incompatibility of IS sentences with the “indexical” reading of temporal adverbials (i.e. with episodic information) cannot be attributed to the quantification over situations, or to the mere fact that the generic quantifier is modalized, but rather to the difference between the specific type of modality associated with the two kinds of sentences. My main claim, then, is that the behavior of temporally restricted IS and BP sentences is a specific case of the more general pattern concerning IS and BP sentences in classical generic constructions, formally clarified in chapters 2, 3, and 4 above. More specifically, I argue that episodic information is incompatible with “in virtue of” generalizations expressed by IS sentences, due to the pragmatic, realworld knowledge requirements which generally accompany such generalizations and which inhibit IS sentences from expressing “unreasonable generalizations” (given our real-world knowledge). In contrast, episodic information is perfectly compatible with “descriptive” generalizations, and since BP sentences can also express this kind of generalization, they can be indeed “episodic generic.” An additional, general conclusion of this chapter is that the “nonaccidentalness” of generics, which is rightly taken to be their core characteristic, does not mean that such sentences cannot not generalize over actual entities, only that they are not limited to actuality, i.e. that they necessarily generalize over entities in other, possible worlds as well. Thus, temporally restricted BP sentences can be considered generic, since they support counterfactuals and express nonaccidental generalizations, although unlike standard generics they also necessarily make assertions about actual individuals, in a restricted time interval.

CHAPTER 6 Conclusion and Directions for Further Research

1. Introduction Discussions of genericity in the literature often include the statement that “Generics express non-accidental generalizations.” I strongly agree with this statement, but at the same time I believe it is ambiguous, that is, that people can, and do, use it unconsciously to make two different statements. On the one hand, this statement can simply mean that generics assert that a certain generalization is non-accidental; on the other, it can be used to express something stronger, namely that generics reflect a certain non-accidental generalization, which is supplied by our real-world knowledge. The main claim I have made in this work is that although both minimally contrasting BP and IS sentences (e.g. Dogs bark and A dog barks, respectively) indeed “express nonaccidental generalizations,” they can do so in two different ways, which correspond to the two intuitive meanings of “Generics express non-accidental generalizations”: Whereas in uttering a BP sentence a speaker can simply assert that a certain generalization is not accidental, in uttering an IS sentence, the speaker relies on knowledge of certain non-accidental generalizations about our world, and tries to convey this knowledge to the listener. More specifically, I argue that the distinction between minimally contrasting IS and BP sentences syntactically encodes a semantic distinction between two principal kinds of characterizing, non-accidental genericity, which I labeled “descriptive” and “in virtue of” genericity. Both IS and BP sentences are capable of expressing “in virtue of” generalizations, asserting, very roughly, that the generalization is nonaccidentally true in virtue of some contextually determined property associated with the denotation of the subject. In contrast, only BP


sentences also express “descriptive” generalizations, asserting, very roughly, that the generalization is merely non-accidentally true, without implying or specifying the relevant property or factor in virtue of which this non-accidental truth holds. The original intuitions about the existence of two kinds of nonaccidental genericity are not new (and can be found in e.g. Lawler (1973), Burton-Roberts (1977) Declerk (1991) and Carlson (1995)). The main innovation of this work is in defining the intuitive distinction in a precise and formal way, attributing it to the difference between two main types of accessibility relations which can restrict the modalized generic operator, and then successfully using this formal difference to account for a wide variety of old, as well as newly observed, semantic, pragmatic and distributional differences between minimally contrasting IS and BP sentences. Apart from the difference in the availability of accessibility relations, both IS and BP sentences were given the same semantic representation, a move which enabled us to account for a number of similarities between them. In section (2) I briefly summarize how the variety of similarities and differences between IS and BP sentences were derived from the formal similarities and differences I defined between the two kinds of generalizations. In section (3) I turn to sketch a direction for further research, concentrating on an explanatory account of the difference in availability of accessibility relations between IS and BP sentences. 2. A brief summary of the semantics and pragmatics of IS and BP sentences 2.1 Similarities between IS and BP sentences I formalized the two kinds of generalizations (“in virtue of” and “descriptive”) as having the same basic semantic structure, basically a tripartite quantificational structure, headed by a universal operator (over individual and situation variables), under the scope of a universal modal quantifier (over accessible worlds). More precisely, both generalizations, and thus both IS and BP sentences, have the truth conditions in (1):


(1) The basically equivalent truth conditions for IS and BP sentences (i.e. for both “in virtue of” and “descriptive” generics): Simple present IS and BP sentences are true in <w, I> iff (where ^P is the property denoted by the subject, and ^Q is the property denoted by the VP):

Paraphrase: Simple present IS and BP sentences are true in the world of evaluation (the actual world), at the interval of utterance (“now”) if and only if, in all worlds w′, accessible from w, it holds that there is a superinterval of I, I′ (i.e. an interval “surrounding” the present), s.t. every (contextually relevant and non-exceptional) situation, located at I′, involving any (contextually relevant, and non-exceptional) P individual, is a Q situation involving that individual. This basically equivalent semantic structure enabled us in the previous chapters to account for a number of similarities between IS and clearly descriptive BP sentences, such as (2) and (3), respectively: (2) a. A woman in this place does not walk alone outside. b. A dog barks. (3) a. Uncles stand on one foot every other Monday morning. b. Famous red-covered books contain an odd number of typos. First, following theories such as Krifka (1987, 1982), Wilkinson (1991) and Chierchia (1995), to name just a few, the universal quantifier over individual and situation variables in the truth conditions captures the fact that unlike “particular sentences” (like “John is walking home”), and kind generics (like “Dodos are extinct”), both IS and BP sentences in (2) and (3) are felt to express (restricted) universal quantification over individuals and situations. Second, both kinds of sentences in (2) and (3) express non-accidental generalizations about possible, as well as (or instead of) actual individual and situations, and both support counterfactual statements. This was captured by putting the universal generalization in the scope of a universal modal operator, quantifying over accessible worlds.


Finally, despite the universal quantification, both IS and Descriptive BP sentences as in (2) and (3) tolerate exceptional and irrelevant entities (individuals and situations), and moreover do that in the same basic way (e.g. in both cases the exceptional individuals and situations are taken to be “less normal,” and characterized in a more vague way than the irrelevant individuals and situations). This was captured by integrating into the semantic structure in (la) the same basic mechanism for tolerating exceptional and irrelevant entities, developed on the basis of Kadmon & Landman’s (1993) “domain vague restriction” suggestion, and represented through the superscripts Xp,YQ in (1). 2.2 The differences between IS and descriptive BP sentences 2.2.1 Accessibility relations “In virtue of” generalizations, then, expressed by both IS and BP sentences, and the “descriptive” generalization, expressed by BP sentences only, share several important properties, and have the same basic semantic structure. To account for the differences between them I first defined the accessibility relation restricting the modalized quantifier associated with them in two different ways, instead of treating it uniformly, as is usually done (as in e.g. Krifka et al. (1995), Wilkinson (1991) and Chierchia (1995)). This is seen in (4) and (5): (4) “In virtue of” accessibility relation (available for both IS and BP sentences): w′ is accessible from w iff

i.e. w′ is accessible from w iff, for some property ^S, every P individual has S in w′, and ^S is a member of the cluster of properties associated with ^P in w. Where ^S CR(^P,w). iff there is some Kratzerian function f ( from worlds to sets of sets of worlds) (epistemic, deontic, etc.) s.t.


In other words, ^S is associated with ^P in w iff “every P individual has S” follows from what is known, commanded, stereotyped, etc. in w (i.e. if “Every P individual has S” is itself considered a non-accidental generalization, or necessity, in w). (5) “Descriptive” accessibility relation (available for BP sentences only) w′ is accessible from w iff it is a member of inermax (<w, inr<w, I> w′ I>), i.e. a member of the set : {w′: w′ inr <WLewisian, I>} In other words, w′ is accessible from w iff it is a member of the union set of the worlds which are inertia worlds to <w, I> and those which are inertia worlds to <W Lewisian I> (i.e. inertia worlds to the worlds which are maximally similar to w, except from what is needed to allow the fact that the P set of individuals and the set of relevant situations involving them are not identical to the sets of P individuals and relevant situations existing in w, respectively). The main idea here is that although both kinds of generalizations are asserted to be “non-accidental,” i.e. asserted to hold in a set of possible worlds (rather than in the actual world alone), they differ in the way in which and the degree to which the similarity of these possible worlds to the actual one is specified. With the “descriptive” accessibility relation, this similarity remains rather vague: we want the inertia worlds to our world in the present, i.e. those worlds where nothing unexpected happens, and those which are maximally similar to these inertia worlds (except from allowing for the existence of more, fewer or other, non-actual individuals and situations). The similarity of the accessible worlds to ours, then, is defined in a very “wide-perspective” way. In fact, we keep almost everything similar to the actual world, and specify only what is different (e.g. unexpected things, and the existence of non-actual individuals and situations). In contrast, with the “in virtue of” accessibility relation, the similarity is much more focused and specified: we are only interested in those worlds which are like ours w.r.t. some aspect (property) which is epistemically, deontically, etc. associated with the denotation of the IS subject. This is, then, a much more “narrow-perspective” similarity, which resembles Kratzer’s (1981) circumstantial modality.


In the chapters above I showed how this difference in accessibility relations captures the old observations (in Lawler (1973), BurtonRoberts (1977), Declerk (1991), etc.) that IS sentences are oftentimes felt to express “definitional,” “rule-like” or “in principle” generalizations, whereas their BP counterparts can also express “less law-like” or more “in-reality” generalizations, as well as the new observation that IS sentences express a variety of kinds of rules (epistemic, deontic, legal, linguistic, etc.), whereas BP sentences also have a reading which expresses a uniform (“descriptive”) type of generalization. 2.2.2 Presuppositions The two accessibility relations, combined with some independently argued for conversational principles (Grician maxims), also lead to several differences in the kind of real-world knowledge about the denotations of the subject and predicate (and the connection between them) which is assumed to exist with the two kinds of generalizations. The different presuppositions are repeated in (6) and (7): (6) Presuppositions of “in virtue of” generalizations a. “Natural classes” presupposition (conventionally triggered by universal quantification over worlds): is not empty In other words, the cluster of properties associated with ^P is not empty (so ^P is a “natural” property, with which we associate other properties). b. “Reasonable causation” presupposition (triggered by a combination of Grice’s maxim of quality, and the “association” condition on ^S) ^S is s.t. there is a world w′ (Kratzerially accessible from w, and closest to an ideal version of our world) s.t.

(where in the ideal world some general principle which works in our world is not inhibited or blocked by any other, less general, factor), In other words, there is a good possibility that (i.e. it is not too far-fetched to assume that), relative to some general principle in


our world, having ^P and ^S leads to having properties of the sort of ^Q. c. Existence presupposition (triggered by the universal quantifiers on worlds situations and individuals (conventionally), and by the Grician maxim of relevance (conversationally))

In other words, there is at least some world w′, where every P individual has (an associated) S property, and some superinterval of the present interval, I′, and there are some P individuals in this world, and some relevant situations involving them in this world, which are located at both the initial and final subinterval of I′. (7) Presuppositions of descriptive generalizations a. Existence presupposition triggered by the universal quantifiers on worlds, situations and individuals (conventionally), and by the Grician maxim of relevance (conversationally)

In other words, there is at least some world w′, which is a member of the set of inrmax worlds, and some superinterval of the present interval, I′, and there are some P individuals in these worlds, and some relevant situations involving them in these worlds, which are located at both the initial and final subinterval of I′. b. “Enough” presupposition (triggered by Grice’s maxim of quality) There are enough P individuals and relevant situations involving them in <w, I′ initial> (in the past, and/or present in the actual world) for it to be reasonable for the average language user to conclude that the universal statement is non-accidental. Throughout this work, I showed that the felicity conditions on IS sentences are much tighter than those on minimally contrasting BP sentences. Far more than their BP counterparts, IS sentences are restricted to expressing generalizations which are “reasonable” given our real-world knowledge; this, in turn, is because IS sentences are


restricted to expressing “in virtue of” generalizations, whereas BP sentences can also express “descriptive” generalizations. The “in virtue of” accessibility relation in (4) and the accompanying presuppositions in (6) explicate what “reasonable” generalizations mean as far as IS generics are concerned. The role of presupposed real-world knowledge about the P, S and Q properties (denoted by the subject and VP, respectively) is indeed defined in a much stricter way in (6), than in (7). Specifically, as shown in detail in the chapters above, the presuppositions in (6) account for the fact that, unlike BP sentences, IS sentences in “out of the blue” utterances are infelicitous as generic when their subjects denote “extremely unnatural properties,” where their VPs denote “extremely unconnected properties,” and in most cases where they are modified by a specific temporal adverbial (like tonight). In contrast to IS sentences, all BP sentences are felicitous as generic, even when they express “highly unreasonable generalizations”—with funny subjects and VPs, or with specific temporal modifiers—since no real-world knowledge is presupposed about the P, Q and S properties and the connection between them. However, BP sentences with a clear “descriptive” readings are infelicitous as generic when uttered in a context where there are very few past or present P individuals, or relevant situations involving them around (and as a result, where there is no appropriate “inductive” basis for the “descriptive” generalization). Thus, although inductive inference is not part of what descriptive generics assert, the “enough” presupposition formalizes the intuition that it is the only appropriate basis for truthfully asserting that a certain generalization is indeed non-accidental, i.e. for the assertion made by descriptive generics. In addition, both “in virtue of” and “descriptive” generics have an “existence” presupposition (in (6c) and (7b)). In both cases, the existence presupposition is triggered by the presence of universal quantification over worlds, individuals and situations in an interval (conventional triggering), together with the Grician maxim of relevance (conversational triggering), which leads to the requirement of existence at both “edges” of the interval. However, the difference in the accessibly relation, integrated into the presupposition, leads to several differences, e.g. the fact that, whereas all descriptive BP sentences presuppose the existence of individuals and situations in the actual world, this is true only for epistemic IS sentences (but not e.g. deontic ones). Finally, although both IS and Descriptive BP sentences tolerate exceptional and irrelevant individuals and situations in the same basic


way, they differ in their ability to characterize and predict the properties of these exceptional entities. Specifically, with IS sentences these properties can be characterized in advance (to some extent), and they systematically vary depending on the ^Q and ^S properties, whereas with Descriptive BP sentences the set of these characterizing properties is much more vague, and their characterization is not systematically needed. This difference was attributed to the different pragmatic limitations (developed in chapter 4) imposed on the vague set of properties restricting the individual and situation variables. I showed how these pragmatic limitations are systematically influenced by the presence or absence of an “in virtue of” property in the truth conditions and the presuppositions on its choice. Thus, although the basic tolerance of exceptions is achieved though a restrictionfs on the set of individuals and situations, and not though a direct restriction on the set of accessible worlds (as suggested in e.g. Krifka et al. (1995)), it systematically interacts with the two kinds of accessibility defined for IS and Descriptive BP sentences. The tolerance of exceptions of generic sentences is thus taken to be connected to their (specific kind of) intensionality and lawlikeness. 2.3 Apparent interactions between “in virtue of” and “descriptive” generalizations As pointed out in chapter 1 above, the distinction between “in virtue of” and “descriptive” non-accidental generalizations, and its systematic correlation with the IS/BP distinction, is rarely discussed or formally treated in the literature on genericity, and as a result the usual claim is that minimally contrasting IS and BP sentences should be represented equivalently. In the chapters above I pointed out several reasons for this fact. The first is the similarities noted in section (2.1) between the two kinds of generalizations, and the two kinds of sentence (both seem to involve modal universal quantification over situations, both tolerate exceptions in a similar way, etc.). The second reason is the fact that BP sentences can in principle express both kinds of generalizations. Thus, it is may be hard to distinguish between, e.g. Boys don’t cry and A boy doesn’t cry, since on one reading of the former, the two sentences indeed make an equivalent assertion.


Third, there is some tendency in the literature to confuse the basis for the assertion with its content. For example, in many cases, generalizations expressed by IS sentences (which, according to the present theory are “in virtue of” generalizations only) seem to be based on inductive inference, just like their minimally contrasting BP counterparts. Take, for example, the minimally contrasting A dog has four legs and Dogs have four legs. Clearly, people originally arrived at the truth of both sentences on the basis of observing so many dogs with four legs (i.e. on the basis of inductive inference), and not on the basis of knowing what the genetic makeup of dogs is. Nonetheless, despite the similar basis for the IS and BP generalizations, they can make different assertions: The IS asserts that it is in virtue of some property we associate with dogs that this generalization is non-accidentally true, whereas the BP sentence has an additional reading, asserting that “this is merely non-accidental” (i.e. it is expected to be true as well in other, similar worlds), with no integral reference to an “in virtue of” factor. This concludes the summary of the present theory. In the next section I turn to a potential theoretical shortcoming of the theory, and sketch the direction of remedying it in further research. 3. Directions for further research: why do IS and BP sentences differ in the availability of accessibility relation? 3.1 The problem My main claim made is that the underlying difference between the semantics of IS and BP sentences lies in the type of accessibility relations which are able to restrict their Gen operator (and their accompanying presuppositions): both “in virtue of” and “descriptive” accessibility relations with BP sentences, but only “in virtue of” accessibility relations with IS sentences. In addition, the two kinds of sentences are equally represented as tripartite universal quantification over individuals and situations, under the scope of a universal modalized quantifier. As shown above, the advantage of this claim is that it succeeds in accounting for the variety of similarities between IS and BP sentences (due to the equivalent basic semantic structure), while at the same time


it explanatorily accounts for the variety of differences between them by attributing them to a single factor (the difference in the available accessibility relations). But there is also a serious disadvantage, namely that this original claim itself is unexplained in the analysis above. Specifically, if IS and BP sentences have the same basic semantic structure, then it is not clear what leads to this difference in the availability of accessibility relations. I will not attempt to fully solve this problem here. In this section, however, I outline one direction of doing so. 3.2 The semantic difference between IS and BP noun phrases Since the only surface difference between minimally contrasting IS and BP sentences is the form of the NP subject, the obvious direction to take in explaining the difference in the accessibility relation between these sentences is to derive it from independent semantic differences between the two kinds of noun phrases involved, namely indefinite singular vs. bare plural NPs. As pointed out in chapters 1 and 5 above, an independent difference has been long argued for in the literature. To repeat, following the theories of Kamp (1981) and Heim (1982), the claim is that indefinite singular NPs are always interpreted as predicates, i.e. as set (or property) expressions, introducing a variable into the representation (which is then bound by Gen). The situation is different with BP NPs. Following Carlson’s (1977) original claims, all theories agree that such NPs have a reading of a kind denoting (i.e. a referential) expression in sentences which are clearly “Proper-Kind Predication” structures, and which cannot be given a modal, quantificational paraphrase, like Dodos are extinct or Rats reached Australia in 1770. Following Carlson (1977), and more recently Chierchia (1998), kinds are taken to be a special sort of individuals, type e (e.g. Chierchia (1998) takes kinds linguistically to be the individual which constitutes the totality of instantiations of the kind). There is disagreement, however, w.r.t. whether this kind-referring reading is the only, or merely a potential, reading for BP sentences. In the chapters above I followed the widely held approach (taken by e.g. Wilkinson (1991, 1995), ter Meulen (1995) and Link (1995)), according to which BP NPs are ambiguous between the predicate (indefinite) and the kind referring reading, where crucially, in clearly characterizing (I-) generics, like Dogs bark, (which are the main concern of this work),


they only have their indefinite reading, and are thus interpreted exactly like their IS NPs counterparts. Some theories, however, (e.g. Carlson (1977) and more recently Chierchia (1998) and Dobrovie-Sorin & Laca (1996)) claim that BPs are unambiguously kind referring. One of the pieces of evidence in support of this claim, originally used by Carlson (1977), and more recently by Chierchia (1998), is the contrast between the behavior of IS and BP NPs in non-generic sentences, where scope ambiguities are expected. As the data in (8)-(10) show, these expectations are borne out only with the IS NPs, whereas the BP NPs appear to be scopally inert (the examples are in Chierchia (1998), 371): (8)

(Opacity): a. Miles wants to meet policemen (only opaque) b. Miles wants to meet a policeman (both opaque and transparent) (9) (Differentiated scope) a. Miles killed rabbits repeatedly. (wide scope for repeatedly, i.e. different rabbits were killed) b. Miles killed a rabbit repeatedly. (narrow scope for repeatedly, i.e. the same rabbit is killed repeatedly) (10) (Scope w.r.t. negation) a. John didn’t see rabbits. (narrow scope of “rabbits”) b. John didn’t see a rabbit. (both wide and narrow scope for the IS NP) As Chierchia correctly points out, these differences would not be expected if BPs were ambiguous, i.e. if, besides their kind referring reading, they had an indefinite reading, equivalent to the readings of their IS counterparts. The behavior of the BP NPs in (8a-10a) is naturally expected if we assume that they are kind referring only. Turning back to the explanation of the different accessibility relations of IS and BP sentences, the general line we will explore is that it is the kind referring vs. indefinite difference between BP and IS noun phrases, respectively, which is somehow connected to the difference in accessibility relations between minimally contrasting IS and BP sentences. Before further clarifying this connection, though, let me deal with an apparent problem for this general line of thought.


3.3 An apparent problem: can BP NPs corresponding to “extremely unnatural classes” be kind referring as well? In the chapters above, we attributed the variety of differences between minimally contrasting IS and BP sentences like (11a, b) and (12a, b), such as felicity, “flavors,” characterization of exceptions, existence presuppositions, etc., to an underlying difference in accessibility relations which can restrict their Gen operator: (11) a. Dogs bark. b. Norwegian students whose names end with ‘s’ or ‘g’ wear thick green socks. (12) a. A dog barks. b. # A Norwegian student whose name ends with ‘s’ or ‘g’ wears thick green socks. Within the present line of thought, this underlying difference in accessibility relation should somehow be derived from the fact that BP NPs (as in both (11a, b)) are kind referring, whereas IS NPs (as in (12a, b)) are indefinites. This line of thought, however, is undermined by a claim made by Chierchia (1998), namely that the kind referring reading is, in fact, not available for all BP noun phrases. Specifically, Chierchia argues that only those BPs which correspond to natural properties, i.e. which “pick a class of objects that displays a sufficiently regular behavior” (p. 372), can denote kinds. To support this claim, Chierchia points out that unlike BPs corresponding to “natural properties” like machines or people, BPs like parts of this machine, boys sitting here or people in the next room do not behave as scopally inert (i.e. as referring expressions), in the “scopelessness” tests (as can be seen in (13)-(15)). He argues that this is because the latter BPs do not correspond to “natural properties,” and thus cannot denote kinds or be interpreted as referential expressions. Instead, such BP noun phrases behave like the IS noun phrases in (8a-10a) above, i.e. as existentially closed indefinites: (13) Opacity a. John is looking for machines. (only opaque) b. John is looking for parts of this machine. (both opaque and transparent)


(14) Differentiated scope a. John killed people repeatedly. (wide scope for repeatedly) b. John killed people sitting here repeatedly. (narrow scope for repeatedly) (15) (Scope w.r.t. negation) a. John didn’t see machines. (only narrow scope for the BP) b. John didn’t see parts of this machine. (wide or narrow scope for the BP) (Chierchia (1998), p. 373) If the reason for the behavior of the BPs in (13b)-(15b) is indeed the fact that their “unnaturalness” prevents them from denoting kinds, as Chierchia argues, then we should expect that the kind referring reading would be also blocked for BPs like Norwegian students whose name end with ‘s’ or ‘g’ or Green-covered books with a 7-letter title. These BP NPs also correspond to highly unnatural properties, to which we do not “impute a sufficiently regular behavior” (in Chierchia’s terms), or in our terms, to which we do not associate other properties. But if this is the case, we can no longer argue that it is the kind referring interpretation of the BP subject in (11b) which somehow leads to the “descriptive” accessibility relation we find with this sentence (and consequently to the difference between this sentence and its minimally contrasting IS counterpart in (12b)). In other words, if only “natural” BP NPs can be kind referring, as Chierchia (1998) argues, then the difference in availability of accessibility relations between IS and BP sentences can be attributed to the different semantics of the corresponding IS and BP NPs only with sentences like (11a) and (12a) (with “natural” BPs), but not with (11b) and (12b) (with “unnatural ones”). My claim is, however, that the reason for the absence of the kind referring behavior of the BPs in (13a)–(15a) is not their mere “unnaturalness,” as Chierchia (1998) suggests, but rather their diectic nature, i.e. the fact that they pick out classes of individuals which are restricted to a specific, diectically interpreted place (“people sitting here/ in the next room”) or entity (“parts of this machine”).1 In contrast, I claim that BP NPs, which correspond to highly unnatural properties, but which do not contain diectic expressions (like Norwegian students whose name end with ‘s’ or ‘g’ or Green-covered books with a six-letter title), can be freely interpreted as kind referring.

1

See McNally (1995) for a similar claim.


What supports this claim is that, as seen in (16)–(18), unlike the former BPs (in 13a–15a), the latter behave as scopally inert in the series of “scopelessness” tests, just like the perfectly natural ones in (13b)–(15b) above: (16) Opacity a. John wants to find green-covered books with a six-letter title. b. John wants to meet Norwegian students whose name ends with ‘s’ or ‘g.’ (17) Differentiated scope a. John burned green-covered books with a six-letter title repeatedly. b. John killed Norwegian students whose name ends with ‘s’ or ‘g’ repeatedly. (18) (Scope w.r.t. negation) a. John didn’t see green-covered books with a six-letter title. b. John didn’t see Norwegian students whose name ends with ‘s’ or ‘g.’ In both (16a) and (16b) the BP NPs only have an opaque reading. For example, to use Chierchia’s phrasing (p. 370), it would be odd (or pragmatically uncooperative) to use (16b) in a situation where all that John wants is to meet, say, Bill, and Henry, who, unbeknownst to John, are in fact Norwegian students whose name ends with ‘s’ or ‘g.’ Similarly, in both (17a) and (17b), the adverb repeatedly has wide scope w.r.t. the BPs (i.e. different books are burned, and different Norwegian students are killed). Finally, in both (18a) and (18b), the BPs have narrow scope w.r.t. negation (so no specific books or students are talked about). From this we conclude that, if indeed the “scopelessness” phenomena in (8a-10a) above indicate a kind referring interpretation (as Carlson (1977), and Chierchia (1998) argue), then BPs corresponding to both “natural” as well as highly “unnatural” properties (as in both (11a) and (11b) above) are interpreted as kind referring. With this conclusion in mind, we can return to the connection between the kind referring vs. indefinite reading of the NP, on the one hand and the availability of accessibility relation restricting the Gen operator, on the other.


3.4 The interpretation of characterizing sentences with BP subjects 3.4.1 Chierchia’s (1998) and Krifka et al.’s (1995) suggestions Assuming indeed that BP noun phrases denote kinds, and that IS NPs denote property expressions, our next step is to understand how this difference in the interpretation of the NPs affects the interpretation of minimally contrasting characterizing IS and BP sentences, like (19a) and (19b), which until now were both represented as in (20) (ignoring for the moment the difference in accessibility relation and the mechanism for tolerating exceptional and irrelevant entities): (19) a. A dog barks. b. Dogs bark. (20)

We will continue to assume that (20) is the correct representation of the IS sentence in (19a), and that the interpretation of the BP sentence in (19b) should continue involving quantification over situations under the scope of the modal quantifier over worlds. The change we have to make concerns the interpretation of the BP noun phrase “dogs” in (19b). There are several options in this respect. Chierchia (1998), for example, claims that although BP NPs are unambiguously kind referring, in clearly characterizing sentences like (19b), i.e. in the presence of the Gen operator they are type shifted into predicate expressions, denoting sets of instances of the kind, as seen in (21a). Krifka et al. (1995) make a similar suggestion (in (21b)): dog (d) C(s, d)] [bark (s, d)] (Chierchia (21) a. Gen d, s [ 1998) Paraphrase: Any contextually relevant situation involving any instance of the dog kind a barking situation (in all accessible worlds). b. Gen d [ IS (d, DOGS)] [bark(d)] (Krifka et al. 1995)


Paraphrase every individual in the set of realizations of the kind “dogs” is in the set of barking individuals (in all accessible worlds). According to these two representations,2 the interpretation of the subject in (19a) and (19b) is almost the same. In both cases it is has the same role in the representation, namely as a property expression restricting the Gen operator. The difference lies only in whether this is the property of being a dog (with the IS sentences), or the property of being an instance of the dog kind (with the BP sentences). Assuming that (21) is the only possible representation for BP sentences like (19a), there is little hope that we will understand what leads to the different accessibility relations available with these two sentences. We need to look for an interpretation of (19a) which is differs more substantially from the one in (20). 3.4.2 Similarities between characterizing sentences with BP and proper-name subjects Our quest for a different interpretation for BP sentences seems to be facilitated by the similarity between the behavior of BP sentences and that of habitual sentences with proper-name subjects (like John walks to school). Such habituals are analyzed by e.g. Chierchia (1995), Krifka et al. (1995), etc., as expressing characterizing genericity, i.e. as headed by a generic, modalized operator.3 (Chierchia (1995), for example, represents Fred smokes as Gen s [C(s, j] [smoke (s, j)], paraphrased as “Every contextually relevant situations involving Fred (in all accessible worlds), is a situation where Fred smokes”). The crucial point for us is that as far as the accessibility relation of Gen is concerned, such habituals with proper-name subjects are similar to BP sentences. Specifically, they pattern like BP sentences and differ from IS sentences

2

See also Cohen (1996, 2001) for a similar suggestion. Thus, for example, John walks to school tolerates exceptional situations (where John is ill, tired, has a ride, etc., so he doesn’t walk to school), and supports the counterfactual If this were a (non-exceptional) relevant situation involving John (e.g. a going to school situation), it would be a walking to school situation. 3


in their ability to express both “descriptive” and “in virtue of” generalizations. For example, following Lawler (1973) and Burton-Roberts (1977) we noted in chapter 3 that, unlike IS sentences, which only have a “definitional”, “rulelike” reading (as in 22), all BP sentences (as in (23a)) also have a prominent “weak descriptive” reading (as in (23a), although many also have a stronger, more “rule-like” reading (as seen in (23b)). In this sense, habitual sentences with proper-name subjects behave like the BP sentences (as seen in (24a, b)) (see Carlson (1995), for a similar claim concerning sentences like (24a, b)): (22) IS subject A table has four legs. (Only the stronger “rule-like” reading (“to be a table is to have four legs”) is available) (23) BP subject a. Uncles are addicted to pink marshmallows. (Only the “weak descriptive” meaning (“there is a pattern here”), is available) b. Tables have four legs. (Both a “weak descriptive” reading and a stronger, more “rule-like” reading reading are available) (24) Proper-name subject. a. John walks to school. (Prominent reading: “weak descriptive” generalization (see also Carlson (1995)) b. Mary handles the mail from Antarctica. (Both a “weak descriptive” or a stronger “rule-like” reading (with regard to Mary’s job description)) Another indication of the similarity of BP sentences to those with propername subjects is the potential vagueness in the characterization of exceptions, as opposed to the systematically more precise characterization of exceptions with IS sentences, as seen in (25), (26) and (27): (25) IS subject A Pentium III computer works very quickly. (It is rather clear which “abnormal” Pentium III computers and relevant situations involving them may be considered legitimate exceptions to (25) (e.g. broken


computers and situations where the computer is in an extremely hot place), and which are not (e.g. computers painted in green and red and situations where the computer is placed on top of seven tables).) (26) BP subject Carpenters drink only water on cloudy Sundays. (It is not at all clear in advance which “abnormal” carpenters and cloudy Sunday situations are considered legitimate exceptions to (26) and which are not.) (27) Proper-name subjects John drinks only water on cloudy Sundays. (It is not at all clear in advance which “abnormal” cloudy Sunday situations involving John are considered legitimate exceptions to (27) and which are not.) These facts seem to indicate that habituals with proper names (as in (29b)) pattern like BP sentences (29a)), but unlike IS sentences (as in (28)), are compatible with both an “in virtue of” and a “descriptive” accessibility relation: (28) Only the “in virtue of” accessibility relation is available A dog barks. (29) Both the “descriptive” and the “in virtue of” accessibility relations are available a. Dogs bark. b. Fido barks. This similarity can guide us in our attempt to derive the difference in accessibility relations between IS and BP sentences from the semantic differences between IS and BP noun phrases. Specifically, we should seek a way to allow the role that the BP subject has in the interpretation of (29a) to more closely resemble the role that the proper name “John” has in the interpretation of (29b), than the role that the IS subject has in the interpretation of (28). There are now two formal options to capture this similarity.


3.4.3 Two formal options for the representation of characterizing BP sentences The two options presented below, as well as the intuitions behind them, should be viewed as preliminary suggestions only. Further research is needed to turn them into fully developed and testable options, which I cannot undertake here. Neither will I attempt to choose which of these options is more promising, for arriving at an explanatory account of the differences between IS and BP sentences. 3.4.3.1 Option A: a Delfitto (1996)-style representation 4 The first option is to assume that the reason for the difference in accessibility relations between (28) and (29a, b) is derived from the fact that in the former sentence the subject functions as a property expression, whereas in the latter two sentences the subject is a referring expression (denoting an individual). This means that in the interpretation of the BP (29a), the kind-denoting BP noun phrase “dogs” will remain kind denoting, and will not be coerced (through some typeshifting mechanism) into a property expression (e.g. the property of being a realization of the kind), as in Chierchia’s (1998), and Krifka et al.’s (1995) suggestions in (21) above. Delfitto (1996) has independently suggested such a representation of BP sentences, according to which (29a) will be represented roughly as the “modalized kind predication” in (30a). Applying this idea to the truth conditions of generics developed above we get something like (30b): (30) a. λd. Gen s [C(s, d)] [bark(s, d)] (DOGS) (capital letters indicate a kind term) Paraphrase: The set of individuals for which every contextually relevant situation (in all accessible worlds) is a barking situation has the kind “dogs” as a member. (In other words, the property of habitually barking holds nonaccidentally of the kind dog.)

4

I am grateful to Angelika Kratzer and Gennaro Chierchia (p.c.) for suggesting these two alternative directions to me.


b. w′ [w′ set of worlds accessible from w] → [ I′ I I′ λd. s [Involve (s, d, w′) loc (s, I')] → [bark (s, d, w′)] (DOGS)] Paraphrase: In all accessible worlds from w—w′, there is a superinterval of the present I′, s.t. the set of individuals for which, every situations in these worlds, at I′ is a barking situation, has the kind “dogs” as a member. How can the difference between representing the subject as an individual (as in (30)), or as a property (as in (20 above)) explain the difference in availability of accessibility relation between BP and IS sentences? To repeat, the difference between the two accessibility relations lies in the degree to which the similarity of the accessible worlds to the actual one is specified: With the “descriptive” accessibility relation (developed for Descriptive BP sentences) this similarity remains very vague (we want the inertia worlds to our world in the present, and those which are maximally similar to these inertia worlds), whereas with the “in virtue of” accessibility relation the similarity is much more focused and specified (we are only interested in those worlds which are like ours w.r.t. some aspect (property) which is epistemically, deontically, etc. associated with the denotation of the IS subject). The former accessibility relation can always be successfully defined, whereas the latter can be defined only to the extent our real-world knowledge associates properties with the denotation of the IS subject. The intuition which I want to develop is that these differences in accessibility relations are influenced by the difference in the way we think about natural properties and natural individuals. The main idea here is that properties are considered natural only to the extent that we have some previous real-world knowledge about them (and thus we associate them with other properties); this condition, however, is not required in the case of “natural individuals.” In other words, unlike a property, an individual can be considered perfectly natural even if we have no prior real-world knowledge about him/her, i.e. even if there are no facts or stereotypes, commands or norms regarding him/her which have previously been associated with him/her by our real-world knowledge. This implies that such real-world knowledge is not required in the first place in order to treat an individual as a natural one. This idea seems rather convincing when we think about individuals referred


to through proper name (like “Terner”). Even if I have no idea which individual is referred to by this name, and I have no knowledge about or properties associated with this individual, I have no problem considering this individual a natural one. Crucially, I believe that this intuition is (at least potentially) true even when the individual is referred to through a definite description (like The tall man with the blue hat/the spy) or a kind-referring bare plural (dogs, blond uncles with six names). Although these latter descriptions contain expressions, of which we naturally have real-world knowledge, their use is only to help pick out a specific individual, i.e. an object or a “totality of instantiations” (in Donnelan’s (1966) terms, these referring expressions are used, or at least can be used, referentially and not attributively). To summarize, then, the intuition is that properties, but not individuals, are considered natural only to the extent they “come with” or trigger some previous realworld knowledge (i.e. associated properties), or—to use Chierchia’s (1998) words—only to the extent we impute with them a sufficiently regular behavior. Note that I am not claiming that natural individuals cannot be associated with properties or trigger real-world knowledge, only that, unlike properties, individuals can be considered natural without such an association or real-world knowledge. Thus, many people may associate various properties with the individuals denoted by the proper name and definite descriptive “Shakespeare” and “the man who works in this office,” respectively. Crucially, however, even a person who has never heard the name “Shakespeare,” and has no idea who Shakespeare was, or a person who knows nothing about this office will have no problem considering the individuals denoted by these descriptions as “natural individuals.” In the case of properties, on the other hand, the association of properties is necessary and systematically required in order to consider them natural (e.g. the property “being a brown-haired person born in early July” is considered so unnatural precisely because we associate no other nontrivial properties with it).5 If these intuitions about the difference between natural properties and individuals are correct, they can help explain the difference in accessibility relations [relations?] in the following intuitive way. When uttering a characterizing generic, we are making a non-accidental generalization, i.e. we are claiming that a certain generalization holds in other possible worlds, accessible from the actual one (besides the actual one). From the point of view of generics, the decision regarding which worlds will be considered accessible and which will not is somehow


sensitive to the kind of natural entity individual or property—that the generalization is about. Since natural properties (denoted by IS subjects) systematically come with their associated properties, i.e. some real-world knowledge about them is systematically and prominently present, the similarity of the accessible worlds to ours systematically makes use of this assumed realworld knowledge and defines the worlds similar to ours in terms of some shared real-world knowledge associated with the property restricting the generic operator. In contrast, since natural individuals are not associated with properties (at least, not systematically), when we make a generalization about such an entity, and assert that this generalization holds in other worlds as well, the similarity of these other worlds to ours cannot rely on such associated properties, and hence can always be defined in a default and vague way, as with the “descriptive” accessibility relation. Thus, BP sentences as well as habitual sentences with proper-name (or definite) subjects are always compatible with the “descriptive” accessibility relation, and can always express descriptive generalizations. In addition, in many cases, where it is reasonable to associate “in virtue of” properties with the denotation of the subject, the “in virtue of” accessibility relation is also available. These are, of course, only preliminary intuitions. Further research should concentrate on giving these intuitions more precise content and should clarify questions concerning the formal representation in (30b) above (e.g.: Can predicates of “normal” individuals, like “bark” indeed apply to kind individuals with no mismatch? How can we ensure the tolerance of exceptional and irrelevant entities with this representation?). 5

Notice that the distinction between “natural properties” and “natural individuals” which I suggest in the linguistic domain is different from, though related to, the classic distinction between natural/unnatural kinds in the philosophical domain (see e.g. Cocchiarella (1976), Putnam (1970, 1975) and Kripke (1972)). “Natural kinds” in the latter approach are only those abstract entities which have some essential properties, (in our terms, properties which hold of them in all relevant possible worlds, and not only the actual one). They are usually considered to include only “well-established” kinds. I suggest, however, that in the linguistic domain all entities—individuals, kinds and properties—can be associated with properties (i.e. can be taken to have some characteristic in possible, and not only actual worlds), but, unlike individuals, properties must be associated with other properties (i.e. to have some “essence”) in order to be considered “natural.”


3.4.3.2 Option B: a von Fintel (1994)-style representation for BP sentences As an alternative to the representation of BP sentences in (30) above, we can claim that the reason for the difference between (28) and (29a, b) is connected to the fact that the subject in (28) is not only interpreted as a property, but in addition this property directly restricts the modalized generic operator. In contrast, although the originally kindreferring BP subject in (29a) is “coerced” into a property expression, this property does not restrict the generic operator directly. With this representation, the kind-referring denotation of dogs, as in (29a), is allowed to be coerced into a property expression (e.g. the property of being a realization of the kind, as in Chierchia’s (1998) and Krifka et al.’s (1995) suggestions), but the generic quantifier will not be directly restricted by it, and will not bind individual variables introduced by it. To formally capture this alternative we can employ von Fintel’s (1994) suggestion, where quantificational adverbs are defined as generalized quantifiers over sets of (minimal) situations, instead of as unselective quantifiers (over pairs of individuals and situations), as in the Kamp (1981) and Heim (1982) frameworks. In von Fintel’s theory, property (or set) expressions denoted by indefinite noun phrases do not directly restrict the quantifier, but are interpreted as existentially quantified, and restrict the characterization of the set of (minimal) situations quantified over. For example, von Fintel paraphrases the sentence in (31a), as in (31b) (see von Fintel (1994) for details and motivations of this proposal): (31) a. A ‘u’ usually follows a ‘q.’ b. Most minimal situations in which there is a q followed by some letter, are part of (or—extended into) a minimal situation in which there is a q followed by a u. (von Fintel, p. 64) In representing the BP sentence in (29a) along these lines, we now get (32), where the quantifier (in our case the modalized Gen) binds minimal situations involving some individual which realizes the kind dogs (instead of pairs of situations and individuals, as in Chierchia’s (1998) and Krifka et al.’s (1995) suggestion in (21) above):


(32)

Paraphrase: In all worlds w′ accessible from w, there is some superinterval I′ of the present, s.t. every minimal situation, involving some individual realizing the kind “dogs,” is part of (or can be extended to) a minimal barking situation involving some (relevant and non-exceptional) realization of the kind “dog”. Following the intuition developed above that properties are considered “natural” only to the extent we associate other properties with them, it is possible that the “in virtue of a property” accessibility relation is forced only in a case where a property expression directly restricts the modalized generic operator (as with IS sentences). Thus, even though the kind-referring denotation of BP sentences is coerced into a property expression in (32), the fact that this property does not directly restrict Gen allows the accessibility relation to be more flexible, namely to vary between an “in virtue of” and a “descriptive” one.6 Here, too, further research should concentrate on both clarifying the reasoning behind this suggestion, as well as on the formalization itself. One problem which needs to be solved, for example, is that, assuming that (32) is the right representation of BP sentences like Dogs bark, some additional mechanism will have to make sure that IS sentences like A dog barks cannot be represented the same way. Another question, this one concerning both formal options (in (30b) and (32)), is whether the general line of thought behind these options is still compatible with the claim that BP noun phases are, after all, ambiguous between a kindreferring reading (where they have the role in (30b) or (32)) and an indefinite reading, like their IS counterparts. 4. Conclusion I began this work with a review of Krifka’s (1987) distinction between kind (D-) and characterizing (I-) genericity, which is, essentially, a distinction between nominal and sentential genericity. This distinction has been proved to be very helpful in accounting for a variety of linguistic phenomena, and has become a standard assumption in all current theories of genericity.


In the body of this work I argued that in addition to the D/I distinction, there is another basic distinction in the generic domain which, once adopted, can also account for a variety of phenomena. This is the distinction between “in virtue of‘and “descriptive” generalizations, linguistically encoded through the contrast between IS and BP sentences. I argue against a simple account of the differences between IS and BP sentences as a difference between quantificational modal (I-) genericity and Proper-Kind Predication (D-genericity), and claim instead that both types of sentences express non-accidental generalizations, and should be interpreted as involving modal generic quantification, i.e. both are instances of I-genericity. I claim that the underlying difference responsible for the variety of pragmatic, semantic and distributional differences between IS and BP sentences is the accessibility relation restricting this modal generic operator and the accompanying presuppositions. This discussion suggests that, although the distinction between IS and BP sentences does not correlate in a simple manner with the classic D/I genericity distinction (since both are I-generic sentences), the latter distinction is nonetheless involved in the former in an indirect and unexpected way. Specifically, the distinction I found between the two types of sentential (I-) genericity (in terms of the accessibility relations available) is affected by whether nominal (i.e. D-) genericity interacts

6

A third way to integrate a kind-referring term into the quantificational-modal structure is to assume the line suggested in Krifka et al. (1995), Chierchia (1998) and Cohen (2001), mentioned above, where with BP sentences we start with modalquantification over situations involving the kind (e.g. the kind DOGS), as in (ia), and then, move, though some sort of type-shifting operation, to the representation in (ib), where we get quantification over individuals realizing the kind:

(i)

a. Gen s [C(s, DOGS)] [bark (s, DOGS)]] b. d, s [R(d, DOGS) C(s, d)] [bark (s, x)]

The crucial point for us, however, is that the decision on the type of modality of Gen (namely “in virtue of” or “descriptive”) will be made before type shifting has taken place, i.e. at the stage where the BP sentence is interpreted as in (ia), where the BP NP is still interpreted as a kind, i.e. as an individual. Only after that the kind term will be shifted into a property expression (of being the realization o the kind). But, of course, this suggestion too needs much independent motivation.


with sentential genericity (as seems to happen when the subject is a BP), or not (as happens when the subject is an IS). Following the ideas of Delfitto (1996) and von Fintel (1994), I suggested two directions for formalizing the interaction between nominal and sentential genericity in sentences with BP subjects. I also sketched an intuitive explanation for the connection between the interpretation of the subject and availability of the two kinds of accessibility relations in the interpretation of the sentence. A direction for further research lies in developing and formally clarifying these issues, as well as in giving a fully compositional analysis of IS and BP sentences. I believe, however, that the differences in type of modality (i.e. accessibility relation), which the present work has showed to be associated with BP and IS generic sentences, should be taken into account by any theory which deals with the interpretation of bare plural and indefinite singular noun phrases.

324

Bibliography

Abusch, D. (1991), “The Present under Past as De Re Interpretation,” in D. Bates (ed.) The Proceedings of the 7th West Coast Conference on Formal Linguistics, The Stanford Linguistics Association. Armstrong, D.M. (1973), Belief, Truth and Knowledge, Cambridge University Press, Cambridge. Arnott, D.W. (1970), The Nominal and Verbal Systems of Fula, Clarendon Press, Oxford. Asher, N. and M.Morreau (1995), “What Some Generic Sentences Mean,” in G. Carlson and F.J.Pelletier (eds.), The Generic Book, The University of Chicago Press, Chicago. Brennan, V. (1993), “Root and Epistemic Modal Auxiliary Verbs,” Ph.D. dissertation, University of Massachusetts, Amherst. Burton-Roberts, N. (1977), “Generic Sentences and Analyticity,” Studies in Language 1 . Carlier, A. (1989), “Généricité du syntagme nominal sujet et modalités,” Travaux de Linguistique 19 . Carlson, G. (1977), “Reference to Kinds in English,” Ph.D. dissertation, University of Massachusetts, Amherst. ——. (1982), “Generic Terms and Generic Sentences,” in Journal of Philosophical Logic 11 . ——. (1989), “On the Semantic Composition of English Generic Sentences,” in G.Chierchia, B.Partee and R.Turner (eds.), Properties, Types and Meaning, Kluwer Academic Publishers. ——. (1995), “Truth Conditions of Generic Sentences: Two Contrasting Views,” in G.Carlson and F.J.Pelletier (eds.), The Generic Book, The University of Chicago Press, Chicago. Chierchia, G. (1995), “Individual Level Predicates as Inherent Generics,” in G. Carlson and F.J.Pelletier (eds.), The Generic Book, The University of Chicago Press, Chicago. ——. (1998), “Reference to Kinds Across Languages,” Natural Language and Linguistic Theory. 6 . Chierchia, G. and S.McConnell Ginet (1990), Meaning and Grammar, MIT Press, Cambridge, Mass. Chocchiarella, N, (1976), “On the Logic of Natural Kinds,” in Philosophy of Science. 43 . Cohen, A. (1996), Think Generic: The Meaning and Use of Generic Sentences, Carnegie Mellon University dissertation. Published 1999, Stanford CSLI.

326 BIBLIOGRAPHY

——. (1997), “Generics and Frequency Adverbs as Probability Judgments,” in Proceedings of the Tenth Amsterdam Colloquium, the Netherlands. ——. (1999), “Generics, Frequency Adverbs and Probability,” Linguistics and Philosophy 22 . ——. (2001), “On the Generic Use of Indefinite Singulars,” Journal of Semantics. 18 . Condoravdi, C. (1992), “Individual Level Predicates in Conditional Clauses,” Talk given at the LSA Annual Meeting. Philadelphia, PA. ——. (1993), “Weakly and Strongly Novel Noun Phrases,” in Proceedings of the 11th WCCFL. ——. (1997), Descriptions in Context , Garland, New York. Dahl, O. (1975), “On Generics,” in E.Keenan (ed.), Formal Semantics of Natural Language, Cambridge University Press, Cambridge. Declerk, R. (1991), “The Origins of Genericity,” Linguistics 29 . Delfitto, D. (1997), “Aspect, Genericity and Bare Plurals,” ms. Utrecht University. de Swart, H. (1994), “(In)definites and Genericity,” ms. University of Groningen. Diesing, M. (1992), Indefinites, MIT Press, Cambridge, Mass. Dobrovie-Sorin, C. and B.Laca (1996), “Generic Bare NPs,” ms. Universite Paris 7 and Université de Strasbourg. Donnelan, K. (1966), “Reference and Definite Descriptions,” The Philosophical Review 75 . Dowty, D. (1979), Word Meaning and Montegue Grammar. Kluwer, Dordecht. Eckardt, R. (1999), “Normal Objects, Normal Worlds, and the Meaning of Generic Sentences,” Journal of Semantics 16 , Oxford University Press. Farkas, D.F and Y.Sugioka. (1985), “Restrictive If/When Clauses,” Linguistics and Philosophy 5 . Fine, K. (1975), “Truth, Vagueness and Logic,” in Synthese 30 . von Fintel, K. (1994), Restrictions on Quantifier Domains, Ph.D. dissertation, University of Massachusetts, Amherst. Gamut, L.T.F. (1991), Logic, Language and Meaning, The University of Chicago Press, Chicago. Glasby, S. (1997), “I-Level Predicates that Allow Existential Readings for Bare Plurals,” in Proceedings of SALT 7 , Cornell Linguistics Circle, Cornell University. Greenberg, Y. (1998), “Temporally Restricted Generics,” in Proceedings of SALT. 8 , Cornell Linguistics Circle, Cornell University. ——. (2002), “Two Kinds of Quantificational Modalized Genericity, and the Interpretation of Bare Plural and Indefinite Singular NPs”, in B. Jackson (ed.), Proceedings of SALT. XII , CLC publications, Cornell university. Grice, H.P. (1975), “Logic and Conversation,” in P.Pole and J.L.Morgan (eds.), Syntax and Semantics. 3 : Speech Acts, Academic Press, New York.

BIBLIOGRAPHY 327

Heim, I. (1982), The Semantics of Definite and Indefinite NPs. Ph.D. dissertation, University of Massachusetts, Amherst. Published by Garland, NewYork, 1989. Jong, F.de and H.Verkuyl (1984), “Generalized Quantifiers: The Properness of their Strength,” In J.van Benthem and A.ter-Meulen (eds.), Generalized Quantifiers in Natural Language, Foris, Dordecht. Kadmon, N. and F.Landman (1993), “Any,” Linguistics and Philosophy 16 . Kadmon, N. (2001), Formal Pragmatics, Blackwell, Oxford. Kamp, H. (1971), “Formal properties of ‘Now,’” Theoria 37. ——. (1975), “Two Theories about Adjectives,” in E. Keenan (ed.) Formal Semantics of Natural Language, Cambridge, UK. ——. (1981), “A Theory of Truth and Discourse Representation,” in J. Groenendijk, T.Jenssen, and M.Stokhof (eds.), Formal Methods in the Study of Language. Mathematical Center Tracts 135, Amsterdam. Kamp, H. and B.Partee (1995), “Prototype Theory and Compositionality,” Cognition 57. Kempson, R.M. (1975), Presuppositions and the Delimitation of Semantics. Cambridge University Press, Cambridge. Kratzer, A. (1981), “The Notional Category of Modality,” in H.J.Eikmeyer and H.Reiser (eds.), Worlds and Contexts. New Approaches to World Semantics. de Gruyter, Berlin. ——. (1989), “An Investigation into the Lumps of Thought,” Linguistics and Philosophy 12 . ——. (1995), “Stage Level and Individual Level Predicates,” in G.Carlson and F.J.Pelletier (eds.), The Generic Book, The University of Chicago Press, Chicago. Krifka, M. (1987), An Outline of Genericity [partly in collaboration with C. Gerstner], SNS-Bericht, University of Tübingen. ——. (1988), “The Relational Theory of Genericity,” in M.Krifka (ed.), Genericity in Natural Language, 285–312 SNS-Bericht 88–42 , University of Tübingen. ——. (1995), “Focus and the Interpretation of Generic Sentences,” in G. Carlson and F.J.Pelletier (eds.), The Generic Book, The University of Chicago Press, Chicago. Krifka M., Pelletier, F.J., Carlson G,. ter Meulen A., Link, J. and G.Chierchia ——(1995), “Genericity: An Introduction,” in G.Carlson and F.J. Pelletier (eds.), The Generic Book, The University of Chicago Press, Chicago. Kripke, S, A., (1972), “Naming and Necessity,” in D.Davidson, and G.Harman (eds.), Semantics of Natural Language, Reidel, Dordecht. Landman, F. (1991), Structures for Semantics, Kluwer, Dordecht. ——. (1992), “The Progressive,” Natural Language Semantics. 1 . Langendoen, D.T., and H.B.Savin (1971), “The Projection Problem for Presuppositions,” in C.J.Fillmore and D.T.Langendoen (eds.), Studies in Linguistic Semantics, Holt, Reinhart and Winston, New York.

328 BIBLIOGRAPHY

Lawler, J. (1973), Studies in English Generics, University of Michigan Papers in Linguistics 1:1 , University of Michigan Press, Ann Arbor. Lewis, D. (1973), Counterfactuals, Harvard University Press, Cambridge, Mass. ——. (1975), “Adverbs of quantification,” In E.Keenan, (ed.), Formal Semantics of Natural Languages, Cambridge University Press, Cambridge. ——. (1983), “Scorekeeping in a Language Game,” in R.Bauerle, U.Egli, and A.von Stechow (eds), Semantics from Different Points of View, SpringerVerlag, Berlin. ——. (1986a), “Causal Explanation,” in Philosophical Papers . Oxford University Press. ——. (1986b), “Counterfactuals and Comparative Similarity,” in Philosophical Papers. Oxford University Press. Link, G. (1995), “Generic Information and Dependent Generics,” in G.Carlson and F.J.Pelletier (eds.), The Generic Book, The University of Chicago Press, Chicago. McNally, L. (1995), “Stativity and Theticity,” in S.Rothstein (ed.), Events and Grammar, Kluwer Academic Publishers, Great Britain. Milsark, G. (1974), Existential Sentences in English, Ph.D. Dissertation, Massachusetts Institute of Technology. Distributed by Indiana University Linguistics Club. Morgan, J.L. (1969), “On the Treatment of Presuppositions in Transformational Grammar,” in R.I.Binnick, A.Davidson, G.Green, and J.Morgan (eds.), Papers from the fifth Regional Meeting of the Chicago Linguistics Society, Chicago. Morreau, M. (1996), “Allowed Arguments,” in Proceedings of IJCAI’96, Morgan Kaufmann, Los Altos, USA. Nunberg, G. and C.Pan (1975), “Inferring Quantification in Generic Sentences,” CLS 11 , Chicago Linguistic Society, Chicago. Pelletier, F.J. & N.Asher (1997), “Generics and Default,” in J.van Benthem & A.ter Meulen (eds.), Handbook of Logic and Language, North Holland, Amsterdam and MIT Press, Cambridge, MA. Prior, E.W. (1985), Dispositions, Aberdeen University Press, Aberdeen. Putnam, H. (1970), “Is Semantics Possible?,” in H.E.Kiefer and M.K.Munitz (eds.), Language, Belief and Metaphysics, State University of New York Press, Albany, NY. ——. (1975), “The Meaning of ‘Meaning,’” in K.Gunderson, (ed.), Language, Mind and Knowledge, University of Minnesota Press, Minneapolis. Schubert, L.K. and F.J.Pelletier (1987), “Problems in the Representation of the Logical Form of Generics, Plurals and Mass Nouns,” in E.LePore, ed., New Directions in Semantics, Academic Press, London. ——. (1988), “An Outlook on Generic Statements,” in M.Krifka (ed.), Genericity in Natural Language, SNS-Bricht 88–42 , University of Tübingen.

BIBLIOGRAPHY 329

——. (1989), “Generically Speaking, Or: Using Discourse Representation Theory to Interpret Generics,” in G.Chierchia, B.Partee and R.Turner (eds.), Properties, Types and Meaning, vol 2: Semantic Issues, Kluwer, Dordecht. Stalnaker, R. (1968), “A Theory of Conditional,” in N.Rescher (ed.), Studies in Logical Theory, Blackwell, London. ——. (1974), “Pragmatic Presuppositions,” in M.Munitz and P.Unger (eds.), Semantic and Philosophy, New York University Press, New York. ter Meulen, A. (1995), “Semantic Constraints on Type Shifting Anaphora,” in G.Carlson and F.J.Pelletier (eds.), The Generic Book, The University of Chicago Press, Chicago. Thomasson, R, (1970), “Indeterministic Time and Truth Value Gaps,” Theoria 36 . Vlach, F. (1981), “The Semantics of the Progressive,” in P.Tedelschi, and A.Zaenen (eds.), Syntax and Semantics 14: Tense and Aspect, Academic Press, New York. Welmer, W.G., (1974), African Language Structures, University of California Press, Berkeley. Westermann, D. (1930), A Study of the Ewe Language, Oxford University Press, Oxford. Wilkinson, K. (1991), “Studies in the Semantics of Generic Noun Phrases,” Ph.D. dissertation, University of Massachusetts, Amherst. ——. (1994), “Comparative Progressives and Bare Plural Subjects,” ms. BenGurion University of the Negev. Wilson, D. (1975), Presuppositions and Non-Truth Conditional Meaning, Academic Press, New York.

330

Index

Accommodation: 42, 49–50, 64, 95, 128, 209, 216, 224–226, 230–233, 266 Analytic: 24–25, 39, 47, 88–90, 121, 127, 156, 186, 200 Asher, N.: 16ff, 203

Support: 32–35, 38, 93–95, 280 Variable (C): 65, 73, 213–215 Contingent: 26, 32, 37, 47, 121, 167ff Counterfactuals: 11–14, 19, 26, 56, 134–138, 141–142, 260–156, 274–277, 294–295, 299, 312ff

Brennan, V.: 53, 55, 62, 65–73, 74–76, 122, 251ff, 269ff Burton-Roberts, N.: xxvi–1, 17, xxvi–25, 42-47, 83, 88, 126, 155, 184, 297–301, 313

Dahl, O.: 11–13, 205, 261 Declerk, R.: xxvi–1, 17, 155, 184, 297, 301 de Jong, F.: 93, 167 Delfitto, D.: 7, 262ff, 315–316 de Swart, H.: 10, 41 Diesing, M.: 10 Dobrovie-Sorin, C.: 6ff, 7, 26, 28ff, 56–57, 129, 274–275, 277, 307 Donnelan, K.: 318 Dowty, D: 110ff, 142, 145ff, 148, 170, 174

Carlson, G.: 1, 6–8, 40–44, 59, 62, 184, 247, 271, 297, 306–313 Characterizing, (quantificational (I-)) Genericity: 2, 4–12, 17, 22, 40, 57, 255–260, 267–273, 305, 319 Chierchia, G.: 7–12, 27–28, 53, 62–65, 90–95, 113, 121, 128–129, 189–190, 211–213, 221ff, 247, 305–315, 318 Cohen, A.: 14ff, 26, 42ff, 43, 58–62, 97, 130ff, 139–141, 145ff, 161, 205, 215, 247, 258–260, 275ff, 312ff Condoravdi, C.: 10ff, 19, 205, 247, 254–258, 295 Contextual Restriction: 55–56, 64, 73, 121–122, 211, 213, 254, 257–258, 261

Eckardt, R.: 203, 205ff, 216ff Episodic: 2, 6, 19–20, 158, 247-249, 254, 259, 263, 270–277, 295–296 Essential properties: 26, 47, 318ff Farkas, D.: 8 Fine: 209 Good possibility: 106–113, 117–120, 229–230, 258–259, 289ff, 287–292,

331

332 INDEX

Grician Maxims: 113–117, 120–121, 153, 166, 171, 172–173, 179, 190–189, 196–197, 200, 229, 233, 243, 302–304 Habitual: 2, 19, 26, 36–39, 95, 62, 245–247, 312–315 Heim, I.: 9, 12, 53, 85, 261, 306 Homogeneity: 161, 260, 277 Indefinites: 8–9, 28ff, 44, 53, 128(ff), 249ff, 254, 307–310, 321 Individual-level predicates: 2, 10–11, 24, 38, 65, 206, 233, 245, 279 Inductive: xxvi–1, 22–24, 33–36, 40–42, 48, 95–95, 127, 184–186, 194–199, 252–254, 280–283, 304–305 Inertia worlds: 110ff, 143–144, 145ff, 146–156, 175, 300–301, 316 Kadmon, N.: 16ff, 44, 112ff, 113–114, 189–190, 208–221, 300 Kamp, H.: 9, 78ff, 209, 264ff, 306 Kempson, R.M.: 113, 189 Kinds: 6, 27, 92, 247 Kratzer, A.: 10, 13ff, 14–16, 51, 78ff, 89, 167ff, 194, 203, 212ff, 224, 249ff Krifka, M.: 1–7, 10–16, 25–26, 53, 56–58, 62, 81–82, 84–85, 90–91, 128- 129, 201–206, 254, 271–277, 304, 310–315, 319–321 Laca, B.: 6ff, 7, 26, 28f, 56–57, 129, 274–275, 277, 307 Landman, F.: 44, 8ff, 160, 145ff, 208–221, 300 Link, G.: 4, 6–7, 307 McConnell Ginet, S.: 11, 189–190 Milsark, G.: 9

Modal base: 14–16, 53, 55, 57–58, 65–68, 74, 82, 84, 89, 108, 230, 234 Modality Circumstantial: 14, 51–53, 58, 65, 70, 77, 88–90, 122, 130, 301 Deontic: 14–16, 53, 58, 77, 83, 90–91, 105–115, 121, 126, 131, 152, 154–156, 177, 181, 300, 306, 316 Epistemic: 14–16, 51–53, 5–67, 77, 83–91, 105–115, 121, 126, 131, 152, 154–156, 177–180, 200, 206–208, 215, 230, 249, 267–269, 300–301, 304 316, Root—53, 55, 62, 65–74, 249–251, 269 Stereotypical: 14, 77, 83–88, 91, 121, 131, 152, 154–156, 177, 181, 300, 316 Modalized kind prediction: 315 Morgan, J.L.: 79 Morreau, M.: 16ff, 203 Natural entities: 316–319 Necessity: 13–14, 77–78, 87, 90, 115, 152, 156, 300 Normality: 11, 14–16, 44, 58, 81–87, 88, 103, 109–112, 129, 138–151, 158, 174–175, 200, 203–205, 209, 215, 216–221, 227–233, 237–244, 299, 314 Ordering source: 14–16, 44, 57–58, 82, 89ff, 108–112, 145, 201–205, 206, 244 Partee, B.: 78ff Pelletier, F.J.: 4, 7–8, 10, 17, 203 Possibility: 13–14, 89, 152, 183, 251ff Precisifications: 210–212, 216–220, 227–228, 235–243 Presuppositions: “enough”: 186–197, 199–201, 303

INDEX 333

“family of sentences” test of: 165, 167, 177, 188–190–303 conventionally triggered: 93, 165–176, 302–303 conversationally triggered: 134, 165–176, 186, 188–191, 195–198, 302–303 defeasibility of: 181, 189–191 of existence: 93, 132, 192, 160–184, 186, 176–184, 199–201, 249–251, 258, 294 Probability: 59ff, 139, 161, 275ff. Progressive: 8ff, 145ff, 160, 262ff, 263ff Proper-Kind Predication, (Predicational/(D- genericity): 2, 4–8, 19–19, 56, 244–245, 273–275, 305, 320 Rules: 1, 20, 24, 40–44, 49, 161–165, 83–90, 97, 121, 127, 154–156, 275ff, 301, 313 Schubert, L.K.: 4, 7–8, 10 Scopelessness: 309–310 Similarity (of accessible worlds to the actual one) Comparative: 130, 136–138 Maximal: 136–152, 154–155, 157, 175, 200, 275, 300, 316 Narrow perspective of: 89–91, 131, 301 Vagueness of: 131–132, 136–138, 155, 200, 301, 304, 316, 319 Wide perspective of: 131–132, 155–156, 174, 183, 200, 301 Stage-level predicates: 10–11, 19, 208, 215, 220, 233, 242, 254 Stalneker, R.: 13, 113, 189–190, 261 Sugioka Y.: 8 Supervaluations: 206, 238 Tendential stability: 19, 247, 271, 275 ter Meulen, A.: 4, 6–7, 307

Thomasson, R.: 141 Verkuyl, H.: 93, 167 Vlach, F.: 8, 143, 145ff von Fintel, K.: 319–321 Wilkinson, K.: 4, 7, 10–12, 53, 73ff, 128–129, 211, 264, 299–300, 307 Wilson, D. 113, 189

Manifestations of Genericity

Projection genericity of space curves

Cutaneous Manifestations of Endocrine Diseases

Pulmonary Manifestations of Pediatric Diseases

Pulmonary Manifestations of Pediatric Diseases

Mucocutaneous Manifestations of Viral Diseases

Islamikaze: Manifestations of Islamic Martyrology

Mucocutaneous Manifestations of Viral Diseases

Head and Neck Manifestations of Systemic Disease

Thyroid Disorders with Cutaneous Manifestations

Head and Neck Manifestations of Systemic Disease

Vascular Manifestations of Systemic Autoimmune Diseases

Microbiology and Aging: Clinical Manifestations

Migraine: Manifestations, Pathogenesis, and Management

Color Atlas of Local and Systemic Manifestations of Cardiovascular Disease

Prader-Willi Syndrome: Development and Manifestations

Taiwan's Modernization: Americanization and Modernizing Confucian Manifestations

Principes d'animation et d'organisation des manifestations culturelles

Allergy Frontiers - Volume 3: Clinical Manifestations

AIDS and Other Manifestations of HIV Infection, Fourth Edition

Pulmonary Manifestations of Rheumatic Disease, An Issue of Clinics in Chest Medicine (The Clinics: Internal Medicine)

Victorian Gothic: Literary and Cultural Manifestations in the Nineteenth-Century

Electrical Trauma: The Pathophysiology, Manifestations and Clinical Management

Manifestations dermatologiques des maladies infectieuses metaboliques et toxiques

Endocrine Manifestations of Systemic Autoimmune Diseases, Volume 9 (Handbook of Systemic Autoimmune Diseases)

Radiologic-Pathologic Correlations from Head to Toe Understanding the Manifestations of Disease

Our Mothers, Our Powers, Our Texts: Manifestations of Aje in Africana Literature (Blacks in the Diaspora)

The Neurological Manifestations of Pediatric Infectious Diseases and Immunodeficiency Syndromes (Infectious Disease)

Tracing Modernity: Manifestations of the Modern in Architecture and the City

Pulmonary Manifestations of Rheumatic Disease, Clinics in Chest Medicine, Vol 31, Issue 3 Sep 10

Tracing modernity: manifestations of the modern in architecture and the city

Manifestations of Genericity

Projection genericity of space curves

Cutaneous Manifestations of Endocrine Diseases

Pulmonary Manifestations of Pediatric Diseases

Pulmonary Manifestations of Pediatric Diseases

Mucocutaneous Manifestations of Viral Diseases

Islamikaze: Manifestations of Islamic Martyrology

Mucocutaneous Manifestations of Viral Diseases

Head and Neck Manifestations of Systemic Disease

Thyroid Disorders with Cutaneous Manifestations

Head and Neck Manifestations of Systemic Disease

Vascular Manifestations of Systemic Autoimmune Diseases

Microbiology and Aging: Clinical Manifestations

Migraine: Manifestations, Pathogenesis, and Management

Color Atlas of Local and Systemic Manifestations of Cardiovascular Disease

Prader-Willi Syndrome: Development and Manifestations

Taiwan's Modernization: Americanization and Modernizing Confucian Manifestations

Principes d'animation et d'organisation des manifestations culturelles

Allergy Frontiers - Volume 3: Clinical Manifestations

AIDS and Other Manifestations of HIV Infection, Fourth Edition

Pulmonary Manifestations of Rheumatic Disease, An Issue of Clinics in Chest Medicine (The Clinics: Internal Medicine)

Victorian Gothic: Literary and Cultural Manifestations in the Nineteenth-Century

Electrical Trauma: The Pathophysiology, Manifestations and Clinical Management

Manifestations dermatologiques des maladies infectieuses metaboliques et toxiques

Endocrine Manifestations of Systemic Autoimmune Diseases, Volume 9 (Handbook of Systemic Autoimmune Diseases)

Radiologic-Pathologic Correlations from Head to Toe Understanding the Manifestations of Disease

Our Mothers, Our Powers, Our Texts: Manifestations of Aje in Africana Literature (Blacks in the Diaspora)

The Neurological Manifestations of Pediatric Infectious Diseases and Immunodeficiency Syndromes (Infectious Disease)

Tracing Modernity: Manifestations of the Modern in Architecture and the City

Pulmonary Manifestations of Rheumatic Disease, Clinics in Chest Medicine, Vol 31, Issue 3 Sep 10

Tracing modernity: manifestations of the modern in architecture and the city

Recommend Documents