Time and Modality (Studies in Natural Language and Linguistic Theory)

TIME AND MODALITY Studies in Natural Language and Linguistic Theory VOLUME 75 Managing Editors Marcel den Dikken, Cit...

Author: Jacqueline Guéron | Jacqueline Lecarme

219 downloads 1692 Views 3MB 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

TIME AND MODALITY

Studies in Natural Language and Linguistic Theory VOLUME 75 Managing Editors Marcel den Dikken, City University of New York Liliane Haegeman, University of Lille Joan Maling, Brandeis University Editorial Board Guglielmo Cinque, University of Venice Carol Georgopoulos, University of Utah Jane Grimshaw, Rutgers University Michael Kenstowicz, Massachusetts Institute of Technology Hilda Koopman, University of California, Los Angeles Howard Lasnik, University of Maryland Alec Marantz, Massachusetts Institute of Technology John J. McCarthy, University of Massachusetts, Amherst Ian Roberts, University of Cambridge

For other titles published in this series, go to www.springer.com/series/6559

TIME AND MODALITY

Edited by

Jacqueline Guéron Université Paris 3, France

and Jacqueline Lecarme CNRS - Université Paris 7, France

Jacqueline Guéron Université Paris 3 Sorbonne Nouvelle France

ISBN 978-1-4020-8353-2

Jacqueline Lecarme Laboratoire de Linguistique Formelle CNRS - Université Paris 7 France

e-ISBN 978-1-4020-8354-9

Library of Congress Control Number: 2008927066 © 2008 Springer Science + Business Media, B.V. No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work. Printed on acid-free paper. 9 8 7 6 5 4 3 2 1 springer.com

The contributors to this book dedicate it to Carlota Smith. . . . ‘O fearful meditation: where, alack, Shall Time’s best jewel from time’s chest lie hid? Or what strong hand can hold his swift foot back, Or who his spoil of beauty can forbid? O, none, unless this miracle have might, That in black ink my love may still shine bright’. William Shakespeare, Sonnet LXV

Preface

The papers in this volume were first presented at the International Round Table Time and Modality held in Paris in December of 2005 and organized by the Jeune ´ Equipe B 368 Syntaxe anglaise et syntaxe comparative (Université Paris 3-Sorbonne Nouvelle) and by the Laboratoire de Linguistique Formelle (Centre National de la Recherche Scientifique (CNRS) and Université Paris 7). We wish to thank the institutions that provided financial support for the Round Table: the CNRS, the Laboratoire de Linguistique Formelle, the Conseil Scientifique (Universités Paris 3 and Paris 7) and the Bureau des Relations Internationales (Université Paris 3).

vii

Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 2

1

Tense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Aspect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Perfect and Perfectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Eventuality Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Modals and Modal Verbs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Subject/Speaker Orientation . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 The Temporal Location of Modal Verbs . . . . . . . . . . . . . . . 3.3 The Temporal/Causal Function of Modal Verbs . . . . . . . . 3.4 The Temporal Syntax of Non-Root Modals . . . . . . . . . . . . 3.5 Weak Necessity Modals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 The Role of Past Morphology in Modal Contexts . . . . . . . . . . . . . . . 5 The Subjunctive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Genericity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Copular Clauses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Conclusions and Open Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2 3 3 4 5 5 6 7 8 9 9 10 11 11 12 14

Patterns in the Semantics of Generic Sentences . . . . . . . . . . . . . . . . . . . . . . Greg Carlson 1 The Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 The Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Induction and Stipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Rules and Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 What do Generic Sentences do? . . . . . . . . . . . . . . . . . . . . . 3.3 Inductive Generalizations Again . . . . . . . . . . . . . . . . . . . . .

17 17 18 22 22 23 24

ix

x

Contents

4

Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Patterns and Non-Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Generic Sentences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Restriction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Review of Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Weak and Really Strong Generalizations . . . . . . . . . . . . . . . . . . . . . . 6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

25 25 27 31 33 34 36 37

Intensional Subjects and Indirect Contextual Anchoring . . . . . . . . . . . . . . . Ileana Comorovski 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Specificational Copular Clauses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Higgins’s (1973) Taxonomy of Copular Clauses . . . . . . . . 2.2 The Subject of Specificational Clauses . . . . . . . . . . . . . . . . 2.3 The Copula of Specification . . . . . . . . . . . . . . . . . . . . . . . . . 3 Romanian Questions of the Form Care Copula DP? . . . . . . . . . . . . 3.1 Two Types of Discourse-Linking . . . . . . . . . . . . . . . . . . . . . 3.2 Semantic Restrictions on the Postcopular DP . . . . . . . . . . 3.3 Remarks on the Syntax of Romanian Questions of the Form ‘Care Copula DP?’ . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Indefinite Subjects, Topichood, and Point of View . . . . . . . . . . . . . . 4.1 Indefinite Subjects and Topichood in Constituent Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Topichood, Point of View, and the Conditional Mood . . . . 5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

39

Temporal Orientation in Conditionals (Or, How I Learned to Stop Worrying and Love UFO’s) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bridget Copley 1 UFOs and Other Oddities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Away from an Explanation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Towards an Explanation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 No Real Epistemic Eventives . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Antecedents Have Their Own Modal Flavors . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . On the Temporal Syntax of Non-Root Modals . . . . . . . . . . . . . . . . . . . . . . . . Hamida Demirdache and Myriam Uribe-Etxebarria 1 Crosslinguistic Asymmetries in the Temporal Construals of Non-Root Modals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 The Grammar of Temporal Relations . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Tenses and Aspects as Spatiotemporal Predicates . . . . . . . 2.2 Anaphora . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Condoravdi (2002): Perfect Modals . . . . . . . . . . . . . . . . . . . . . . . . . .

39 40 40 41 42 43 43 44 49 51 52 52 53 56 56 59 60 65 68 69 73 76 76 79

80 82 82 84 87

Contents

xi

4

The Temporal Phrase Structure of Non-Root Modals . . . . . . . . . . . . 89 4.1 Setting the Modal-Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 4.2 Ordering the Modal-Time . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 5 The Temporal Interpretation of English Non-Root Modals . . . . . . . 91 5.1 Present/Future Oriented Epistemic Modals . . . . . . . . . . . . 91 5.2 Aspectually Complex Modals: Progressive and Perfect Modals . . . . . . . . . . . . . . . . . . . . . . 95 6 The Temporal Interpretation of Spanish Non-Root Modals . . . . . . . 104 6.1 Modals Inflected for the Past . . . . . . . . . . . . . . . . . . . . . . . . 104 6.2 Modals Inflected for the Present . . . . . . . . . . . . . . . . . . . . . 109 6.3 Modals Inflected for the Future . . . . . . . . . . . . . . . . . . . . . . 110 7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

How to Say Ought in Foreign: The Composition of Weak Necessity Modals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Kai von Fintel and Sabine Iatridou 1 A Basic Contrast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 2 Weakness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 3 The Crosslinguistic Picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 4 Flavors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 4.1 Epistemic Modality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 4.2 Goal-Oriented Modality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 4.3 Deontic Modality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 5 Counterfactuality? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 6 A Consolation and a Precedent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 7 Scope Confusion? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 8 Ordering Source Promotion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 9 Why Counterfactual Marking? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140 On the Temporal Function of Modal Verbs . . . . . . . . . . . . . . . . . . . . . . . . . . 143 Jacqueline Guéron 1 Modal Verbs Part I: Grammatical Properties . . . . . . . . . . . . . . . . . . . 144 1.1 The Tense of Modals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 1.2 Absence of Agreement Morphology . . . . . . . . . . . . . . . . . . 146 1.3 The Temporal/Causal Function of Modal Verbs . . . . . . . . 149 2 Causality in Grammar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 2.1 Two Syntactic Levels of Interpretation . . . . . . . . . . . . . . . . 149 2.2 Point of View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 2.3 The Instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 2.4 Two Types of Causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 2.5 Metaphysical Causality on Higher vP Levels . . . . . . . . . . . 161 2.6 On the Syntactic Determination of Causality . . . . . . . . . . . 164 2.7 “Metaphysical Intentionality” . . . . . . . . . . . . . . . . . . . . . . . 167

xii

Contents

3 Modal Verbs Part II: Causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 The English Perfect and the Metaphysics of Events . . . . . . . . . . . . . . . . . . . 173 James Higginbotham 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 2 Metaphysical Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 3 Interactions with Sequence of Tense . . . . . . . . . . . . . . . . . . . . . . . . . . 181 4 Shifted Perfects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192 Tense and Modality in Nominals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 Jacqueline Lecarme 1 Tense in Nominals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 1.1 Space and Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 1.2 The D-T Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 1.3 Time and the Noun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 1.4 Interim Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 2 Modality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 2.1 Time and Worlds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 2.2 Generics and Habituals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 2.3 Attitude Verbs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208 2.4 Free Relatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 2.5 Interim Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 3 Evidentiality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 3.1 Visual Evidentiality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212 3.2 Approaches to (Sentential) Evidentiality . . . . . . . . . . . . . . 214 3.3 Extending Kratzer’s Theory . . . . . . . . . . . . . . . . . . . . . . . . . 217 4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 Time With and Without Tense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227 Carlota S. Smith 1 Background and Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228 1.2 Pragmatic Principles for Temporal Interpretation . . . . . . . 230 1.3 Temporal Information in Language: A Classification . . . . 231 2 Tensed Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232 3 Tenseless Languages and Mixed-Temporal Languages . . . . . . . . . . 234 3.1 Mandarin Chinese . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234 3.2 Inferred Temporal Location . . . . . . . . . . . . . . . . . . . . . . . . . 235 3.3 Sentences With Overt Aspectual Viewpoints . . . . . . . . . . . 237 3.4 Other Tenseless Languages . . . . . . . . . . . . . . . . . . . . . . . . . 239 3.5 Mixed-Temporal Languages . . . . . . . . . . . . . . . . . . . . . . . . . 240

Contents

xiii

4 Zero-Marked Sentences: The Neutral Viewpoint . . . . . . . . . . . . . . . 241 5 Formalizing the Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247 The English Konjunktiv II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 Tim Stowell 1 The Syntactic Distribution of K2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252 2 A Sociolinguistic Interlude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255 3 Subjunctive and Indicative Conditionals . . . . . . . . . . . . . . . . . . . . . . 257 4 Counterfactual Subjunctives Versus Other Types of Subjunctives . . 264 5 Further Aspects of English Counterfactual Subjunctives and Sequence of Tense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266 6 Inversion in Conditionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268 7 A Variant Form of the Subjunctive . . . . . . . . . . . . . . . . . . . . . . . . . . . 269 8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 Phasing in Modals: Phases and the Epistemic/Root Distinction . . . . . . . . . 273 Karen Zagona 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273 2 Epistemic and Root Modals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276 2.1 Subject Versus Non-Subject Orientation . . . . . . . . . . . . . . . 277 2.2 Modal Evaluation Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278 2.3 Aspect and Veridicality of the Event . . . . . . . . . . . . . . . . . . 281 3 Modals and Phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 3.1 Toward a Phase-Based Analysis: Lexical Features of Modals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 3.2 Modals as Tense Morphemes . . . . . . . . . . . . . . . . . . . . . . . . 285 4 Perfectivity and Root Modals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293

Contributors

Greg Carlson Linguistics Department, University of Rochester, Rochester, NY 14627-0096, USA [email protected] Ileana Comorovski Departement de Sciences du Langage, Université de Nancy 2, 54015 Nancy, France [email protected] Bridget Copley Structures Formelles du Langage, CNRS-Université Paris 8, 2 rue de la Liberté, 93200 Saint-Denis France, bridget.copley@sfl.cnrs.fr Hamida Demirdache University of Nantes, LLING (EA 3827), Chemin de la Censive du Tertre BP 81227, 44312 Nantes Cedex 3, France, [email protected] Kai von Fintel Department of Linguistics and Philosophy, MIT, 77 Massachusetts Avenue 32-D808, Cambridge, MA 02139, USA, fi[email protected] Jacqueline Guéron Institut du Monde Anglophone, Université Paris 3-Sorbonne Nouvelle, 5 rue de I’Ecole de Médecine, 75006 Paris, France, [email protected] James Higginbotham Department of Linguistics, University of Southern California, 3601 Watt Way, Los Angeles, CA 90089-1693, USA, [email protected] Sabine Iatridou Department of Linguistics and Philosophy, MIT, 77 Massachusetts Avenue 32-D808, Cambridge, MA 02139, USA, [email protected]

xv

xvi

Contributors

Jacqueline Lecarme Laboratoire de Linguistique Formelle, CNRS-Université Paris 7, Case Postale 7031 - 2, place Jussieu - 75251 Paris Cedex 05, France, [email protected] Carlota Smith Department of Linguistics, University of Texas at Austin, Austin, TX, USA [email protected] Timothy Stowell UCLA College of Letters and Science, 2300 Murphy Hall, Los Angeles CA 90095-1438, USA, [email protected] Myriam Uribe-Etxebarria University of the Basque Country, Filologia eta Geografia-Historia Fakultatea, Hizkuntzalaritza eta Euskal Ikasketak Saila, Unibertsitateen Ibilbidea 5, VitoriaGasteiz 01006, Spain, [email protected], [email protected] Karen Zagona Department of Linguistics, University of Washington, A210 Padelford Hall, Box 354340, Seattle, WA 98195-4340, USA, [email protected]

Introduction

This book is a follow-up to its companion volume, The Syntax of Time (Guéron and Lecarme, 2004), which contributed various perspectives on the syntax of tense and the temporal construal of events: models of tense interpretation, construal of verbal forms, temporal aspect versus lexical aspect, the relation between the event and its argument structure, and the interaction of case with aktionsart or tense construal. Similar in approach to the earlier book, Time and Modality explores the grammatical relations between tense and modality. The tense system of the grammar places the states and events the sentence describes in the space-time world of the speaker. But other worlds, based on desire, necessity, or possibility, are also somehow accessible to the speaker. What grammatical mechanisms function like a magic carpet to transport us from the deictic world to other imaginable worlds? Part of the answer lies no doubt in the syntactic versatility of the grammar: the same semantic content describes real or possible worlds depending on its categorial envelope and its syntactic context. Thus in English and Arabic, a modal verb provided the envelope for a future tense morpheme without confusing tense and modality. In English, modal verbs with an incorporated tense morpheme are functional categories merged in the higher Tense domain of the sentence while in Romance and other languages, semantically identical verbs are merged in the lower lexical Verb Phrase domain. Past tenses, whether perfective or imperfective, may be used with either realis temporal value or irrealis conditional value, behaving in the latter case like non-tensed verbs. How are we to account for both the morphosyntactic inseparability and the interpretive independence of tense and modality? In the last colloquium, it was shown that the ordinary Noun Phrase, the most stable of our categories, has a temporal value. Given the grammatical inseparability of tense and modality, we would expect to find that the Noun Phrase can have modal values also. Is this verified in natural language? And if so, what grammatical mechanisms can account for it?

J. Guéron and J. Lecarme (eds.) Time and Modality, 1–15. c Springer Science+Business Media B.V. 2008

1

2

Introduction

Each chapter in this book contributes in some way to the central issues just outlined. Together, they show that the grammar devotes a lot of space to modality, and that modal sentences containing modal verbs occupy a unique position within this grammatical space. Below we have summarized the main issues covered in this book. Given the state of the art it is perhaps inevitable that all of the articles in this volume discuss the facts of English. It is well-known, however, that the grammar of tense and modality in English has special properties not shared by many other languages. Thus lexical finite verbs in English, unlike those of Romance or Germanic languages, do not raise from VP to the Tense node in syntax. Only auxiliary verbs, like have and be, raise to the Tense node. On the other hand, again unlike Romance or Germanic, English modal verbs MUST be merged with tense in syntax and may not appear in VP. It is thus important to examine the grammar of tense and modality in other languages, in view of the all-important question of to what degree the semantics of tense and modality remains constant in the face of sometimes dizzying syntactic variety. The languages discussed here include, in addition to English, the Romance languages French, Italian, Spanish, and Romanian, the Germanic languages Dutch, German, and Old English, the Slavic languages Russian and Croatian, and the Afroasiatic languages Arabic and Somali, in addition to Hungarian, Mandarin Chinese, and Japanese. In several cases, as in the articles by Zagona and Demirdache and Uribe-Uxteberria, there is detailed comparative study.

1 Tense From the study of tenseless languages and mixed-temporal languages such as Mandarin Chinese and Navajo, Smith addresses fundamental questions about the grammaticalization of time: How is temporal information conveyed in language? Do tenses code temporal information directly? Does the universality of temporal interpretations arise from a common syntactic structure necessarily including a Tense Phrase? According to Smith, a few very general grammatical principes account for temporal interpretation in both tensed and tenseless languages. In the absence of tense morphology, aspectual information about boundedness and information about internal temporal properties (Static/Dynamic, Telic/Atelic, Durative/Punctual), coupled with invariant pragmatic principles, suffice to derive temporal interpretation. Smith proposes that two simple pragmatic principles constrain direct temporal interpretation in languages with tense (English, French), and guide indirect temporal interpretation in languages without tense (Mandarin, Navajo): (i) the default interpretation of present tense sentences as located in the Present, (ii) the fact that bounded events cannot be located at Speech Time (‘Bounded Event Constraint’). A more general principle of Simplicity ensures that Present takes precedence over the Future: futurity is never a ‘purely temporal concept’ (Lyons, 1997, 677), and the

Introduction

3

element of uncertainty makes the Future more complex than the Past or the Present in the information they convey. Smith assumes that the syntax of a fully-tensed language includes a Tense Phrase (TP). Every sentence has direct information about temporal location: (i) the default interpretation of present tense sentences as located in the Present, (ii) past tense conveys that the Reference Time precedes the Speech Time. The other types of languages have a syntactic Aspect Phrase but no TP. In these languages temporal location is inferred from aspectual information.

2 Aspect Smith’s discussion focuses on aspectual information about boundedness. In her model, the ‘Bounded Event Constraint’ plays an important role in guiding inference about temporal location: bounded situations cannot be neutrally located at Speech Time. However, inner aspect (aktionsart) and outer aspect (aspect proper) have independent sources of boundedness, and these features bound different objects: inner aspect bounds an event, outer aspect bounds a time interval. What is the role of inner/outer aspect in temporal and modal interpretations? How does perfective aspect alter the veridicality of the event? How does eventuality type correlate with temporal and modal properties?

2.1 Perfect and Perfectivity Perfective aspect is generally understood to mean that the eventuality the sentence describes is evaluated as a whole, including its onset and its endpoint boundaries. The English past tense is sometimes characterized as a (morphologically unmarked) perfective past. On a perfective past reading (Sue read the article), the past time referred to includes the entire event of reading. For Guéron, a verb is perfective if in addition to aspect, its lexical content introduces a boundary on the time interval when V raises to T. It is imperfective otherwise. Perfectivity can also be defined in syntactic terms. Auxiliary have derives a perfective configuration by defining a time interval in T and by selecting as its complement a perfect participle which defines an endpoint on that interval. Modal verbs define a perfective configuration by defining a time interval in T and selecting a bare infinitive complement which denotes a state or event marking the endpoint of the interval. Zagona addresses the effects of aspect on modal interpretations. Perfective aspect can affect the main predication of the clause, altering its status from averidical to counterfactual or presuppositional. This effect varies from language to language. Both implicative and counterfactual readings occur in the Spanish perfective past tense (Juan pudo ganar la carrera ‘Juan could have won the race’ ((i) He may have

4

Introduction

won (ii) But he didn’t win)). English does not allow either reading in modal+main verb sequence, but modals should and could show counterfactuality in combination with have. Both imperfective aspect and epistemic modals are immune to these effects. Zagona’s analysis concludes that the interpretive generalizations are traceable to different syntactic relationships among several constituents of the clause, including subjects, C, Tense, Aspect, and the core verb phrase of the clause. Higginbotham explores the general thesis that the English Perfect is purely aspectual, serving to shift from a predicate of events e to a predicate of events e that is the result of e. The head Perf has therefore 2 argument positions and expresses the relation R(e, e ) satisfied by an ordered pair of events such that e is a result of e. Because, by assumption, results temporally follow the states or events of which they are the results, it follows that the situations characterized by a predicate F(e) are Past with regard to a present result e of e. In this analysis, the Perfect is not involved in Tense at all, except derivatively. In Present Perfect assertions, and in tensed Perfects generally, it is the result that carries the Tense. Focussing on adverbial and quantificational expressions (John crossed the street quickly, Mary occasionally walks to work), Higginbotham concludes that there is also an e-position in adverbs like quickly or occasionally in addition to the e-position that is identified by the Verb. As a consequence, Result and Resultant states of Perfects may be given in terms not only of the events described in the verbal heads, but also of those that arise through modification or quantification. In the example the fire having burnt out slowly warmed the bricks, it is Perfect Aspect, not the e-position in the head burn, upon which Tense operates. Higginbotham explores further issues in the syntax and semantics of Sequence of Tense in English, showing that the general conditions on these constructions support the account of the Perfect as purely aspectual.

2.2 Eventuality Type Lexical aspect is a classification of verbal expressions (VPs) according to the internal temporal properties of the events in their denotations. On Smith’s view, situation type indirectly classifies a clause as expressing a situation with certain internal temporal properties. There are 3 temporal features: Static/Dynamic, Telic/Atelic, Durative/Punctual. These features cluster in the situation type categories State, Activity, Accomplishment, Achievement, Semelfactives: States are Static and Durative. Events are Dynamic, etc. As discussed in the paper by Demirdache and Uribe-Etxebarria, the distribution of the non-shifted versus the future-shifted reading of the modal complement is sensitive to lexical aspect (Zagona, 1990; Condoravdi, 2002; Stowell, 2004). When the complement of the modal verb is a stative predicate, the situation time of the modal complement is construed either as ongoing or as future shifted (Amina must/may/might/could/should be in Ottawa (now/tomorrow)). However, with an eventive modal complement, the situation time of the modal complement is construed as future-shifted relative to the utterance time (Amina must/may/might/can/

Introduction

5

could/should win the marathon). How can we account for the correlation between modal interpretation, temporality, and eventuality type? Guéron argues that within vP, a stative verb describes a simple spatial configuration (inclusion of a Figure in a Ground). At LF, the spatial configuration is predicated of a single point of time (or extended without change over all the points of an interval of time). An eventive verb denotes a change of state. It thus implies two distinct spatial configurations, one which pertains before the change and another which pertains after it. Two non-identical states are therefore predicated of two intrinsically ordered points of time. As a consequence, a state Sn−1 is construed as the cause of a state Sn , and Sn as the effect of Sn−1. Developing this analysis, Guéron defines causality in the grammar and examines the causality particular to modal verbs (section 3.3). Copley’s analysis focusses on the correlations between modal flavor, eventuality type, and temporal orientation in conditionals. As is well known, eventuality type and temporal orientation are correlated: eventives must be future oriented (If it rains, Max will get sick), while statives can be present or futur oriented (If John is sick, Celeste is/will get sick). The current explanation for the ‘present eventive constraint’ (Condoravdi, 2002; Werner, 2003) is that eventives cannot be evaluated at the present moment because the present moment is homogeneous. Copley proposes a more general analysis which also incorporates the ‘future stative constaint’. According to her analysis, ‘metaphysical’ modal flavor (Condoravdi, 2002) correlates with eventivity, epistemic modal flavor correlates with stativity. This analysis also predicts that antecedents have a modal flavor independently of their consequent, a fact that has largely passed unnoticed in the generative linguistic literature.

3 Modals and Modal Verbs The major questions that unite the papers in this volume concern modality and modal verbs: How are modals temporally evaluated? What is the temporal contribution of non-root modal verbs to the temporal interpretation of the sentence? What is the role of the apparent past inflection on could/might on an epistemic reading? Can the relative scope of modals and tense be represented as a structural difference? How do modal verbs interact with inner/outer aspect? Few of these questions have yet been definitely answered. The papers in this volume do not yet offer a complete and coherent picture, but they reflect, in our opinion, some of the best recent efforts in that direction.

3.1 Subject/Speaker Orientation Several properties of epistemic modals versus root modals have been shown to be crosslinguistically general. According to Zagona, epistemic and root modals differ

6

Introduction

in three respects: (i) subject/non subject orientation, (ii) temporal location relative to the tense of the clause, (iii) effects of perfective aspect (illustrated by Spanish and French) (section 2.1). Root modals are ‘subject oriented’ in that a sentence with a root modal is true if the state of obligation/permission holds of the subject, regardless of whether the eventuality occurs or not. ‘Subject orientation’ is taken to imply a grammatical subject-predicate relationship that is located in time by clausal tense. Epistemic modals are ‘speaker oriented’ in that the modal qualifies the speaker’s subjective attitude toward the factuality of the proposition (Sue may be at home). More generally, epistemic modals describe a judgement of someone outside the modal clause: in embedded clauses, epistemic modals can be ‘oriented’ toward an argument of the main clause (It seems to Mary that it must be dinner time). How does the grammar arbitrate the conflict between the subject and the speaker? For Guéron, the question bears on the larger issue of event structure and event perception. Time is continuous as perceived by a consciousness: temporal continuity can be maintained, despite the gaps which separate states and events, if every situation is constructed as located in the scope of a sentient point of view. Guéron proposes (technically) that an interface condition arbitrates the conflict between the two sentient points of view made available by the grammar, that of the subject and that of the speaker: whenever an event description in the syntactic scope of the subject contains no temporal gap, the sentence is construed from the point of view of the subject. If the event description contains gaps, the speaker must assure temporal continuity.

3.2 The Temporal Location of Modal Verbs It has been shown that epistemic and root modals differ in the way they interact with clausal tense (Stowell, 2004, a.o.). Root modals are located in time by the tense of the clause; present or past tense on the modal indicates that the modal state of the subject holds at a present or a past time (Carl can’t/couldn’t move his arm). Epistemic modals are temporally located at the ‘evaluation time’ of the clause (Jack’s wife can’t/couldn’t be very rich). In main clauses, this is usually the speech time; in embedded clauses, it is the time of the main clause event. The question arises as to why this difference should occur, and whether it can be traced to syntactic properties of modals. A natural hypothesis is that the relative scope of modals and tense can be represented as a structural difference: epistemic modals are positioned above Tense and root modals below Tense. As Zagona observes, interpretive differences do correspond to structural differences in some languages (e.g., Catalan, Swedish). However, in other cases, for example with the English finite modals like must and may, all modals have the same syntactic distribution regardless of their reading. In the Minimalist framework, the different interpretations have to be derived by covert movement. In a framework without covert movement (Chomsky, 2001, 2005), some alternative analysis is called for.

Introduction

7

To account for the modals’s dual semantics, Zagona (2006) explored the hypothesis that the two types of modals are merged in different phases (Chomsky, 2001, 2005): root modals are merged and interpreted in the v∗ P phase, and epistemic modals in the CP phase. The descriptive generalization captured by this analysis is that root modals, being merged within the domain of predicate-argument structure, are part of the predicate. Epistemic modals merged outside the v∗ P are part of the left periphery, an adjunct of the Force predicate, with speaker orientation. While this analysis is viable for languages whose epistemic and root modals have different syntactic distributions, it is inconsistent with the long-standing assumption that all English modals have the same distribution. Moreover, it incorrectly associates epistemic modals with the speaker, via interpretation of the modal as an adjunct of the assertion predicate of Force. However, in embedded clauses, epistemic modals can be oriented toward an argument of the main clause (John thought that it must be dinner time). In the present chapter, Zagona offers an account of the epistemic versus root distinction based on a single syntactic position for the finite modals, namely the head of a Tense Phrase. This revision calls into question the generalization that modals are interpreted in relation to either the v∗ P or C, the Force predicate. Instead, the interpretation of the modal is argued to vary according to its own lexical features, and the features of agreeing elements. Concretely, Zagona proposes that modals can have either a valued or an unvalued Tense feature. A modal whose Tense feature is interpretable is in relevant respects like a (displaced) main verb, and can be predicated in syntax of the subject (root reading). A modal whose tense features are uninterpretable is analogous to v rather than V. It is understood as predicated of the v∗ P proposition (epistemic reading).

3.3 The Temporal/Causal Function of Modal Verbs Within a ‘possible worlds’ semantics, modal verbs introduce (sets of ) possible worlds. Guéron’s hypothesis is that the grammatical function of modal verbs is not to introduce alternative worlds, but rather to provide the means by which the hypothetical situation the vP denotes can be introduced into the ongoing deictic world. Guéron first argues that modal verbs are impersonal verbs, which are incompatible with referential agreement (the agreement paradigm for person/number realized by the suffixal alternation between Ø and s). Modal sentences are impersonal in that no checking of a variable person feature takes place: a modal verb selects no subject. Guéron then presents an analysis of modal verbs as causal verbs, and situates modal verbs with respect to other types of causal verbs in English. Causality, Guéron argues, is a function of tense interpretation: in order for a state Sn−1 to count as the cause of a new state Sn , all the conditions necessary for the change of state to take place must be already present in Sn−1 . Causality is intentional when a human subject is construed as a conscious entity and non-intentional or ‘metaphysical’ (Condoravdi, 2002), when it is not triggered by human intention.

8

Introduction

According to Guéron’s analysis, modal verbs select the same type of complement as make/see or faire/voir: a bare infinitive in Modern English, a suffixed infinitive in Romance and Old English. While make/see and faire/voir are transitive causative verbs which agree with their subjects (Mary made John leave), modals are impersonal causative verbs which lack morphological agreement features. The subject of modal Mary must leave is a [+human] argument, which functions as an instrument capable of introducing the event defined by the complement VP into the discourse world.

3.4 The Temporal Syntax of Non-Root Modals Zagona’s paper can be instructively paired with the paper by Demirdache and Uribe-Etxebarria, who also develop an analysis of the temporal interpretation of modals from a cross-linguistic perspective with emphasis on English and Spanish. The paper concentrates on the analysis of non-root modals and seeks to provide answers to the following questions: what is the temporal contribution of non-root modal verbs to the temporal interpretation of the sentences in which they occur? How do we account for the asymmetries displayed by English and Spanish sentences, in particular with regard to the temporal value of the inflectional features of modal verbs? How do modal verbs—be it in English or Spanish—interact with tense and (outer/inner) aspect? Demirdache and Uribe-Etxebarria analyze English modal verbs as impoverished: in some contexts (e.g., Amina sould/might be asleep), the past inflection does not locate either the time of the possibility under discussion or the situation-time of the modal complement. Spanish modal verbs contrast sharply with English modal verbs in that they can appear fully inflected for person agreement as well as for tense and aspect. The present versus past inflection on the modal has an impact on the temporal interpretation of the clause. Demirdache and Uribe-Etxebarria incorporate non-root modals in their phrase structure for tense and aspect (Demirdache and Uribe-Extebarria, 1997, sqq): the functional heads T, M, ASP, V, each introduce a time argument projected onto a specifier in the syntax. The time argument contributed by the modal is an openended interval [t, ∞), following Condoravdi (2002). From this uniform temporal syntax, the asymmetries between English and Spanish can be derived in a principled manner. English non-root modals are tenseless: the outer TP projection embedding the modal is not headed by a semantic tense. The modal-time is thus set to the utterance time via deictic anchoring (as is the case when the modality is epistemic), or by anaphoric anchoring (when the modality is ‘metaphysical’, relating to ‘how the world might have turned out to be’ (Condoravdi, 2002)). Tense morphology on non-root modals in Spanish, contrary to English, is not semantically vacuous. This assumption explains why Spanish past inflected modals allow either a past-oriented epistemic construal or a future oriented metaphysical construal, whereas English past inflected modals allow a future/present oriented epistemic construal.

Introduction

9

3.5 Weak Necessity Modals In the basic framework proposed by Kratzer (1981, 1991), modals quantify over a set of worlds that is calculated from a modal base of accessible worlds and an ordering source which ranks the worlds in the modal base. Different flavors of modality (epistemic, deontic. . . ) come from the interplay of modal base and ordering source. But apparent differences in strength exist between necessity modals like must/have to and ought (You ought to do the dishes but you don’t have to). How can the difference between weak and strong necessity modals be captured in this framework? Von Fintel and Iatridou investigate the weak necessity modal ought and the cross-linguistic expression of this modal concept. According to their analysis, what makes necessity modals like ought semantically weaker is that they have a smaller domain of quantification: strong necessity modals say that p is true in all of the best worlds in the modal base, while weak necessity modals say that p is true in all of the very best (by some additional measure) among the best worlds. In other words, the weak necessity modals signal the existence of a secondary ordering source. While the standard Kratzer framework parametrizes the semantics of modals to two parameters (modal base and ordering source), von Fintel and Iatridou propose that a secondary ordering source is somehow brought into play by counterfactual marking. Weak necessity modals use this secondary ordering source to refine the ranking of the worlds favored by the primary ordering source that strong necessity modals are sensitive to.

4 The Role of Past Morphology in Modal Contexts A recurrent theme throughout almost all of the chapters is the role of past tense morphology in modal contexts. As Guéron observes, in Modern English, a modal verb with ED suffix can be situated either in the present time or in the past time depending on the local reference time. Alternatively, the ED morpheme can situate the modal verb in a counterfactual world provided it is in the scope of an operator like wish or counterfactual if. Iatridou (2000) argued that counterfactuality is a function of past tense morphology: it is thus independent of the grammar of modal verbs. A related issue discussed by several chapter contributors, in particular Demirdache and Uribe-Etxebarria and Zagona, is the role of past inflection of modal verbs. For example, as noted above, the apparent past inflection on could/might in many contexts does not serve to locate either the time of the possibility under discussion (the modal-time) or the situation-time of the modal complement in the past (Amina could/might be asleep/win the race). On an epistemic reading, (e.g., Jack’s wife couldn’t be very rich) couldn’t is a present impossibility (not a past impossibility). Still another piece of the same puzzle is the relation between counterfactual morphology and weak necessity, as investigated by von Fintel and Iatridou: crosslinguistically, a strong necessity modal becomes a weak necessity modal when

10

Introduction

marked with a counterfactual morphology (i.e., Future + Past combination, cf. Iatridou, 2000). In the typical uses of OUGHT, we do not claim that a strong necessity holds in a counterfactual scenario; rather, we claim that a weak necessity holds in the actual world. How does that meaning arise from the combination of counterfactual marking and strong necessity? From the analysis of ‘transparent OUGHT languages’ like Greek (where OUGHT corresponds to the necessity modal prepi in the Past tense in combination with the Future particle tha), von Fintel and Iatridou argue that all that the morphology marks is a change in evaluation parameters, suggesting that counterfactual marking is co-opted in a somewhat meta-linguistic kind of way, so as to promote a secondary ordering of the favored worlds. Some more evidence for this view is provided by the nominal domain. Building on proposals in previous work that show that the tense structure of nominals (DPs) strongly parallels that of clauses (CPs), Lecarme investigates more precisely the non-temporal meanings of nominal tenses in Somali, an Afroasiatic language. The analysis explores the conditions under which a nominal past is interpreted in the modal dimension, contributing either a quantificational reading of the past DP (comparable to English -ever), or an evidential reading (the visible/nonvisible distinction). Lecarme proposes that the common abstract feature underlying the different uses of nominal pasts is a more primitive, domain-general exclusion/dissociation feature (Iatridou, 2000). To account for the link between direct evidentiality and visual perception in nominals, Lecarme proposes that Kratzer’s (1981; 1991) theory of ‘doubly relative’ modality be supplemented by a perceptual component. Along this analysis, past morphology gives rise to the ‘non-actual’, ‘unknown’ or ‘invisible’ modal meanings, depending on different choices of modal base and ordering source.

5 The Subjunctive In this chapter, Stowell explores the syntax and semantics of the English ‘Konjunktiv-II’ (K2) construction (e.g., If you hadn’t ’a/’ve/of said that. . . ) within the context of a broader investigation of subjunctive verb forms in the language. He argues that K2 should be analyzed as a subjunctive perfect form, where the subjunctive is conveyed by the affix –ed, and the perfect functions as a ‘past polarity item’ (Stowell, 1996) signaling the presence of an abstract past tense located either in the same clause or higher in the syntactic structure of the clause. The type of subjunctive mood that occurs in this construction must be distinguished from the mandative subjunctive mood that occurs in the complements of demand/ask class verbs. Both types of subjunctive are licensed strictly locally, in contrast to the subjunctive mood licensed by negation in languages such as French; Stowell suggests that this is related to the modal force conveyed by the subjunctive in these contexts. The particle of/have that occurs within K2 is a ‘subjunctive polarity item’, which has the effect of disambiguating the subjunctive perfect from the indicative past perfect in the informal register that allows K2.

Introduction

11

6 Genericity What is the best model for the semantics of generic sentences? What are the truthconditions of generic sentences in a truth-conditional framework? According to Carlson, the issues that remain central in the discussion of the semantics of generic and habituals since Krifka et al. (1995) concern the treatment of exceptionality, the intensional properties of the interpretations, the unity of meaning, and the strength of negated generics. Carlson (1995) proposed two different models for the semantics of generic sentences: (i) the inductive approach, according to which generics essentially express inductive generalizations, where the base of the generalization (such as ‘Dogs bark, Pencils have erasers’) is some observed set of instances, (ii) the ‘rules and regulations’ approach, according to which the paradigm cases of generics are rules and regulations that we can stipulate (such as ‘Bishops move diagonally’). As is well known, the inductive approach fails on the problem of exceptionality (nonbarking dogs), whereas the ‘rules and regulations’ approach flounders where ‘purely descriptive’ generalizations are involved (Bob eats fast, the traffic on Main Street increases throughout the afternoon). Here, Carlson proposes that both approaches are subsumed by a more fundamental semantic principle, namely that generics and habituals express ‘patterns’ of events. A (mathematical) pattern is any set of non-finite sequences whatsoever that are not random, such as a sequence of odd integers. Instead of a sequence of numbers, Carlson considers a sequence of events construed as a series ordered in time. Randomness, according to Carlson, is precisely what we cannot know about, and what we most certainly do not have any mental capacity to represent. According to this analysis, generic and habitual sentences have quite simple truth conditions: the real world consists of a sequence of events (many of which may overlap one another), and temporally circumscribed portions of the worlds exemplify ‘patterns’ that form the basis for the truth-conditions for sentences. Carlson’s new proposal goes a long way towards accounting for the intensional nature of genericity, exceptionality, and the problem of negated generics, while still leaving us with the challenge of asking ‘how can we know whether a generic sentence is true or not?’

7 Copular Clauses Comorovski focuses on important problems concerning the copula in natural language, more specifically, on the semantic/pragmatic properties of the subjects of specificational clauses (Higgins, 1973). Using mostly French and English data, Comorovski (2007) argued that a semantic/pragmatic characterization of subjects of specificational clauses must make reference both to their intensional semantic type, and to their special relation to the context: semantically, they are of the type of individual concepts <s,e> (i.e., functions from indices to individuals); pragmatically, they must be linked to the context of utterance by a referential expression or by a referential implicit argument they contain (indirect contextual anchoring).

12

Introduction

Given this generalization, Comorovski (2007) suggested that the copula can create an intensional subject position and proposed a definition of the ‘copula of specification’ which takes into account both the intensionality of the subject position and the contextual anchoring of the subject noun phrase. The present article focuses on Romanian copular constituent questions introduced by care ‘which’, ‘what’. Comparing the semantic restrictions on the postcopular DP in Romanian care-questions to the restrictions on the subject of English specificational clauses, Comorovski concludes that the Romanian care-questions like Care e capitala Moldovei? ‘What is the capital of Moldavia?’ (where care is non-D-linked) are the wh-interrogative counterpart of declarative specificational clauses. Comorovski further argues that non-D-linked care is selected exclusively by the ‘copula of specification’. The existence of this selection relation speaks in favour of a lexical treatment of the specificational reading of copular clauses. The paper also discusses the role played in Romanian by the conditional mood in the acceptability of specificational clauses with indefinite subjects: although indefinites cannot generally be subjects of constituent questions, the occurence of indefinite subjects in Romanian care-questions is possible if the verb in the question is in the conditional mood. This distributional pattern of indefinite subjects is explained in terms of the relation between the topic of a constituent question and verbal forms in the conditional mood.

8 Conclusions and Open Problems According to the possible worlds analysis of modality (Lewis, 1986; Kratzer, 1981, 1991), modal sentences introduce (sets of) possible worlds. One justification for such an analysis is the fact that modal words are notoriously ambiguous,or, at the very least, context-sensitive (cf. Kratzer, 1977). However, modal environments still need to be characterized syntactically. In our Introduction to The Syntax of Time, we pointed out that certain compositional procedures in the construal of the syntactic structures underlying tense interpretation are invariable over languages. To some degree this can also be said of the construal of modality. Since modal verbs are ambiguous in isolation, it is difficult to account for variation in modal meaning without a theory of how they are anchored in syntax. It has been shown that scope effects are crucial to differentiating modal interpretations, although accounts differ as to how these effects are achieved in syntax or in Logical Form. Moreover, syntactic analysis provides the advantage of falsifiability without which there can be no reliable discovery procedure. Certain contributions in this volume may be considered as radically syntactic. Thus Zagona bases the different readings of modals on different syntactic relationships among the constituents of the clause, including subjects, C, Tense, Aspect and the core Verb Phrase, such that interpretive generalizations reflect different combinatory possibilities. For Guéron, modal construals are derived by combining the same syntactic elements and relations which are necessary for non-modal, and in

Introduction

13

particular, for causal construals. They involve the speaker, the subject, the agent, the instrument, and their interaction with the modal and the lexical verbs on different levels of sentence structure. Several authors emphasize the effects of perfectivity on both tense and modality interpretation. A number of questions are left open. Copley points out that we still need a theory of modality that can account for the correlation between modal flavor, eventuality type, and temporal orientation. Smith, having shown that certain mechanisms underlying tense interpretation are uniform across languages, including ‘tenseless’ languages, raises the question of the role of temporal morphology in tensed languages. Von Fintel and Iatridou and Lecarme explore the various meanings carried by past morphology, including temporal, conditional, and evidential meanings. They suggest that past morphology might have co-opted capacities that serve the other cognitive systems with which language interfaces. The same grammatical/mental mechanisms would underlie the human capacity to express tense construal and modality. We may ask what the grammar would be like if it contained mechanisms for tense interpretation but no modal expressions (beyond the fact that future morphemes would probably not have developed). Certainly, the order of events would still be recorded, just as as they are on the tombs of the Egyptian Pharaohs in the Valley of the Kings. Events would also be forseen, as they are in any civilization which invents a calendar registering the regularity of natural occurrences in time. What would be missing would be the implications between past, present, and future situations which modality expresses and which go far beyond memory and even prediction. Without modality, how could it be said that the fact that Mary’s blinds are drawn means that she might be sick, and how would this lead the speaker to conclude, and to urge action, by saying that Mary’s mother should go see her right away? On the other hand, where does the uncertainty about Mary’s sickness and her mother’s actual rushing to her side that modality introduces come from? It is not uncertain that King Amenhotep reigned from 1411–1375 B.C. Nor do we consider it uncertain that the sun will rise at 6:40 A.M. tomorrow. A number of authors have emphasized the fact that the present speech instant cannot coincide with the end (or the beginning either) of an event. What then is happening at the speech time? Surely we are not lost in a vacuum; something must be happening. The subject of a sentence may be at one point of a spatio-temporal trajectory (John is walking to school). Or the subject or the speaker may affirm the existence of a state of affairs (John is tired). But speech, which records discontinuous states and events in time, is itself a continuous event in time; so it is not possible to linger on the present state of affairs. The speaker must order the instant with respect to the past and the future in terms of natural continuity or cause and effect. The speaker, poised on one instant of space and time, like Robert Frost stopping by woods on a snowy evening, cannot know where her road is leading nor, given the present situation, can she even know what other situations are simultaneously true now. Mary’s blinds being drawn does not necessarily mean that she is sick. And if the speaker knew whatever else was pertinent now with respect to Mary, as we do

14

Introduction

know everyhing pertinent to the sun rising tomorrow, she would be able to predict whether Mary’s mother will rush to see her, which she cannot do in fact. However, the speaker has knowledge of a number of laws of the discourse world which enable her to hypothesize, to a more or less good approximation, what other situations might be true of the present (or true of the past and pertinent to the present), what the sum of these situations can lead to in terms of future events, and how she can contribute to favoring future events by her speech acts. Without modality, order would still reign, but human beings would not be able to contribute to the ordering of states and events in the discourse world by means of assertive speech acts alone.

References Carlson, G. (1995). Truth conditions and generic sentences: Two contrasting views. In Carlson, G. and Pelletier, F. J., editors, The Generic Book, pages 224–237, Chicago, IL. University of Chicago Press. Chomsky, N. (2001). Derivation by phase. In Kenstowicz, editor, Ken Hale: A Life in Language, pages 1–52, Cambridge, MA. MIT Press. Chomsky, N. (2005). On phases. to appear in Carlos Otero et al. eds. Foundational Issues in Linguistic Theory. Cambridge, MA: The MIT Press. Comorovski, I. (2007). Constituent questions and the copula of specification. In Comorovski, I. and von Heusinger, K., editors, Existence: Semantics and Syntax, volume 84 of Studies in Linguistics and Philosophy. Condoravdi, C. (2002). Temporal interpretation of modals. In Beaver, D. I., Martinez, L. D. C., Clark, B. Z., and Kaufmann, S., editors, The Construction of Meaning, pages 59–88, Palo Alto, CA. CLSI Publications. Demirdache, H. and Uribe-Extebarria, M. (1997). The syntax of temporal relations: A uniform approach to tense and aspect. In Curtis, E., Lyle, J., and Webster, G., editors, Proceedings of the Sixteenth West Coast Conference on Formal Linguistics, volume XIV, pages 145–159. WCCFL. Guéron, J. and Lecarme, J., editors (2004). The Syntax of Time, Cambridge, MA. MIT Press. Higgins, R. F. (1973). The Pseudo-Cleft Construction in English. Ph.D. thesis, MIT. Iatridou, S. (2000). The grammatical ingredients of counterfactuality. Linguistic Inquiry, 31:231–270. Kratzer, A. (1977). What ‘must’ and ‘can’ must and can mean. Linguistics and Philosophy, 1:337–355. Kratzer, A. (1981). The notional category of modality. In Eikmeyer, H.-J. and Rieser, H., editors, Words, Worlds, and Contexts. New Approaches in Word Semantics, pages 38–74, Berlin. De Gruyter. Kratzer, A. (1991). Modality. In von Stechow, A. and Wunderlich, D., editors, Semantics: An International Handbook of Contemporary Research, pages 639–650, Berlin. de Gruyter.

Introduction

15

Krifka, M., Pelletier, F. J., Carlson, G. N., ter Meulen, A., Link, G., and Chierchia, G. (1995). Genericity: An introduction. In Carlson, G. N. and Pelletier, F. J., editors, The Generic Book, pages 1–124, Chicago and London. University of Chicago Press. Lewis, D. K. (1986). On the Plurality of Worlds. Blackwell, Oxford. Lyons, J. (1997). Semantics, volume 1 and 2. Cambridge University Press, Cambridge. Stowell, T. (1996). The phrase structure of tense. In Rooryck, J. and Zaring, L., editors, Phrase Structure and the Lexicon, pages 277–291, Dordrecht. Kluwer. Stowell, T. (2004). Tense and modals. In Guéron, J. and Lecarme, J., editors, The Syntax of Time, pages 621–636, Cambridge, MA. MIT Press. Werner, T. (2003). Deducing the future. In Proceedings of NELS, volume 33, pages 445–462. Zagona, K. (1990). Times as temporal argument structure. Talk given at the Time in Language Conference, MIT, Cambridge, MA. Manuscript, University of Washington, Seattle. Zagona, K. (2006). On the syntactic features of epistemic and root modals. In Fernandez Soriano, O. and Eguren, L., editors, Proceedings of the 16th Colloquium on Generative Grammar.

Patterns in the Semantics of Generic Sentences Greg Carlson

Abstract Generic and habitual sentences have a number of general properties that any analysis will need to give some account of. It is well-known that they tolerate exceptions. They are intensional in their semantics. They seem able to cover a wide notional range of generalizations (e.g. speak of habits, customs, propensities, abilities, etc.). Finally, and less often noted, their negation is unexpectedly strong, amounting to a near-total prohibition (for instance, if sheep don’t eat meat, virtually no sheep may do it). The analysis offered here attempts to provide an account of all these properties simultaneously. Generic and habitual sentences express a pattern of occurrences of events. Patterns are essentially constraints on sequences of events that constrain occurrences, with unconstrained occurrence amounting to chaotic (random) sequencing. Truth and falsity of generic sentences then depends on whether the pattern is exemplified by a subsequence of its events in a world. The exceptions, and the intensionality, result from the definition of what a pattern is. Key words: Generic, pattern, habitual, negation, intensionality, episodic, chaos, event, exception, constraint

1 The Setting Complete chaos is hard for me to imagine.1 If you share my imaginative limitations, though, here is what it might be like: Take the existing universe from inception to some distant future point. Now, slice it up into a series of times, making the times 1 I wish to thank the editors and conference organizers, Jacqueline Gueron and Jacqueline Lecarme for the opportunity to present a version of this material, and their detailed comments. I also wish to thank James Higginbotham, Henry Kyburg, Len Schubert, and Richard Zuber whose comments led to improvements that otherwise would have been among the shortcomings in the paper, for which I remain responsible. This work was supported by the NSF under Grant No. 0328849.

University of Rochester


17

18

G. Carlson

about as small as you like. I’m going to take one tenth of a second as the smallest time-slice, since it tests the limits of my imagination, but it really doesn’t matter precisely what one chooses. Now, imagine a complete state-description of the world for each of those time-slices, as if one is inspecting frames in a film. Take those state-descriptions, and mix them up at random. Do it with dice or coin tosses or a random-number generator, it doesn’t change the present, intuitive point. After mixing them up, run the “film” at real-time speed. You now have a “universe” that is not easily recognizable as our own. At one moment, a blank wall is replaced by (what appears to be) a basket made of leather, by nothing, then by a Picasso, and then by a shoelace, then by the left-hand upper part of a car door, then by. . . What is missing from this reframed universe is of course, stability. It’s a mess. Viewing it as one might a film on a screen, it would be little more than a continuous blur. From one moment to the next, one has no idea what is going on, or what to expect. Chaos is the disconnect between what is and what is next, and what was before. There is simultaneously, everything, and nothing. If you are disturbed by my presumptions here about temporal order (a source of order, after all), that’s fine: take all the state-descriptions as simultaneous, thereby atemporal, and see if that suits your imagination better. It just does not happen to suit mine better. In comparison to this (and I stress, only in comparison), the present world is as orderly as if some despotic dictator were running it. Rubber balls bounce off solid cement surfaces and if dipped even briefly in liquid oxygen, shatter, and don’t bounce far higher instead. Eggs don’t bounce, but break under the same circumstances—they don’t become walruses or flout Grice’s maxims. You can sit on chairs, and wash your hands with soap and water. Any illusion of lack of order in our universe is supported by focusing on what we don’t expect, what we don’t take for granted. This is an entirely distinct observation from the notion that the universe is deterministic. It is just surprisingly orderly when weighed against the genuinely chaotic alternative, however you wish to try and imagine it.

2 The Issues This is about semantics, not metaphysics, and how we might give an account of generic and habitual sentences in a semantics (I normally refer to both simply as “generics” but throughout intend to include what are also somewhat misleadingly called “habitual” sentences as well). The aim is to shed possibly some light on some basic puzzles of genericity in language that semantic researchers have been dealing with in a formal framework for several decades now (see Krifka et al., 1995 for a summary of issues, most of which remain germane at the present time). The central question in a truth-conditional framework is, naturally, what are the truth conditions of generic sentences? Since the work of Lawler (1973) and Dahl (1975), this has proven a very difficult matter to articulate and has spurred considerable work in

Patterns in the Semantics of Generic Sentences

19

applying the tools of quantification, intensionality/possible worlds, nonmonotonic logics, pragmatics, semantic and pragmatic restriction, and a great deal else in service of the problem (e.g. Cohen, 1999; Greenberg, 2003; Asher and Morreau, 1995; Eckardt, 1999; among many others). Each approach, to my mind, offers its own insights. One particularization of the problem is the question of how one deals with exceptions and exceptionality (Carlson, 1999; Greenberg, 2007). If for instance birds fly and dogs bark, what bearing do the nonbarking dogs and flightless birds themselves have on the question of the truth or falsity of these claims? Why aren’t the nonbarking dogs and the nonflying birds regarded as the unexceptional individuals instead? This is a classic question, falling broadly under the “problem of induction” as discussed by Hume, Goodman, and many others. An appealing initial way to think about this question starts from the observation that dogs that bark are more numerous than the silent ones, and likewise for the flying birds vs. the flightless ones. They are more numerous in terms of number of individuals, and number of types as well (e.g. flying birds include robins, pigeons, warblers, eagles. . . , and there are more of these than there are of penguins, ostriches, kiwis. . . ). Then the question becomes one of how predominant some property needs to be, and that is where most of us who have examined this approach get stuck. Another way of thinking about exceptionality (Delgrande, 1988; Eckardt, 1999) is to construct members of a class with a given property either normal, or not normal. It is a clear intuition that nonflying birds are not normal birds (in this respect) and that nonbarking dogs are not normal dogs (in that respect). Or, another means is to treat instances with properties not resulting from a default in a non-monotonic type of system as a model of exceptionality (e.g. Asher and Morreau, 1995). Related to such systems are rule systems which treat exceptions as instances of more specific rules taking precedence over more general rules (e.g. if birds fly, then penguins do not fly takes precedence since it is the more specific proposition). On this view exceptionality is relative, as all properties are determined by some rule or other (if we treat individual specification as a ‘rule’). These views have their strengths and weaknesses, but have not been examined in nearly as much detail as the initial, quantity-based approach, and so an orthodoxy of widely-discussed problems for each has not been as fully developed. But whatever the approach, exceptionality plays a central part of what needs to be accounted for. Another property of generic (and habitual) sentences that plays a slightly less prominent role is that of intensionality. Not only do generics and habituals generally prohibit the intersubstitution of referential equivalents if examined judiciously (Bennett, 1972; Dahl, 1975), but also, like modals, their truth and falsity seems to depend on intensional considerations; this gives them the widely-noted characteristic of supporting contrafactuals (Lewis, 1973) (e.g. if lawyers wear expensive suits, then if John, who is not a lawyer, were to become a lawyer, he would wear an expensive suit—at least probably!). This characteristic is typically understood as questioning whether a generalization is “real” or not. Schubert and Pelletier (1987) note the example of the children of the small town of Rainbow Lake, Alberta. Suppose all eight of them born there in a given year, turned out left-handed. Would

20

G. Carlson

it then be so that “Children born in Rainbow Lake are left-handed”? It could be, say, if there were some unusual environmental condition that led to left-handedness. Most of us, however, would regard this as a coincidence, or an accident. But to resolve this issue we need to go well beyond the fact that each of these children is, in fact, left-handed, to the question of who else would be left-handed were they born at that time and place. This issue of the “reality” of a generalization is a highly familiar question often raised in philosophy of science, mathematics, and other domains. Another side of the whole problem, which has over the years received relatively spotty attention, is that of what counts as a “generic” or a “habitual”. A wide variety of candidates suggest themselves. While habits and broad generalizations which are regularly instantiated naturally take the lead, generics can also be used to express propensities, dispositions, abilities, laws, rules of games, customs, social norms, job descriptions, functions, and other notions. The question is, if (1a) is a “habit”, and (1b), a “disposition”, (1c) a “capacity”, and so forth, does this preclude a semantics common to all? (1)

a. b. c. d. e. f. g. h. i.

John bites his fingernails. Crystal vases break easily. Cheryl high jumps nearly two meters. The President appoints the Chief of Staff. The four players bid in turn from the dealer clockwise. Boy Scouts tell the truth. The janitor keeps the hallways free of fire hazards. Soap removes stains. Birds fly.

Note that the forms of these English sentences are the same in a certain respect— they are in the simple present tense. There is nothing to formally distinguish one notional interpretation from another. Further, any of these notions can be expressed in whatever other ways there are to express these in sentence forms (this includes past tense, future, infinitives, the “used to” past, etc.). We could of course be looking at some idiosyncrasy of English, but the failure to differentiate among these different notions in the grammar would seem to be a persistent feature across languages. While often generics/habituals are expressed by syncretism (as using a progressive or an imperfective), and often find expression in particularized forms, the overall pattern suggests, to me, that there is a unity of meaning among these different notions that requires some account. While current research does not unequivocally support such a conclusion, and a great deal remains to be done, it remains my working assumption (see Dahl, 1985, 1995), and the analysis discussed below is intended to articulate this presumed unity. The final property I wish to highlight, which has not played a large role in previous discussions, is the case of negated generic sentences. The problem with negated sentences is that their truth-conditions do not appear to mirror the properties of the positive sentences. Consider, for instance, the GEN operator as outlined in Krifka et al. (1995), a “default quantifier” meaning something like “most” or “usually”,


21

subject to various contextual restrictions. Now, if we have a negated sentence such as the following: (2)

Sheep do not eat meat.

We may expect a scopal ambiguity between the negation and the GEN operator, e.g.: (3)

∼GENΦ GEN∼ Φ

It is probably unsurprising that such an ambiguity does not appear, given that negation tends not to interact with tense and other elements of the verbal complex; but which of the readings then is the one we in fact find? The first would be interpreted, roughly, as saying that it is not the case that usually holds (subject to other restrictions). This does not seem right as then example (2) “Sheep do not eat meat” would seem to be true if only one sheep in three was a meat-eater, or if some other significant minority of sheep were carnivores. But if one in three were a carnivore, then (2) would seem false. The other reading, with the narrow scope negation, seems to say something like it is the usual case that does not hold. Again, this allows for plenty of meat-eating activity by sheep, as long as it does not predominate. Yet in both cases under these circumstances it would be true that sheep eat meat (just not all of them—an issue we return to later). Neither scoping seems a promising account of the truth-conditions of negated generics. Negated generics are much stronger than their positive counterparts. While the positive counterparts sometimes tolerate exceptionality to a very large degree, the negated sentences tolerate few exceptions indeed—they seem to mean a (near-) total absence of the property in question, a (nearly) complete prohibition. So while (4a) may easily hold if one mosquito in a hundred carries the West Nile virus and the remaining 99 do not, (4b) certainly does not allow that 99 out of 100 do carry it. It seems to say that (practically) none do. (4)

a. b.

Mosquitos carry the West Nile virus. Mosquitos do not carry the West Nile virus.

Again, this exceedingly strong interpretation is not found in English alone. Reports in the descriptive grammars of a variety of languages, in instances where the facts are noted, agree on this very “strong” interpretation of negated generics and habituals. To summarize, the issues that I am taking as central in the discussion of the semantics of generics and habituals concern the treatment of exceptionality, the intensional properties of the interpretations, the unity of meaning, and the strength of negated generics. All analyses thus far presented shed light on these issues in varying ways. I am not at this point interested in comparative arguments, or anything more than presenting an alternative way to think about the semantics of generics, and to suggest an avenue of approach which appears to me, at present, quite promising.

22

G. Carlson

3 Induction and Stipulation 3.1 Rules and Induction In Carlson (1995) the question posed is which is the better model for the semantics of generic sentences: should one take as the basis a model of inductive generalization, and assimilate the remainder to that, or should one take consciously-stipulated rules, such as laws and rules of games, as the basic model and instead extend that analysis to the rest?

While this may appear a false dilemma to some, a fair division of the territory into “inductively-based” and “rule-based” examples is equally problematic, and gains far less than one might anticipate. Both views have their strengths, and their weaknesses. The inductive approach flounders on the problem of exceptionality, something the rule-based account promises to deal with easily. The “Rules and Regulations” approach fails to give adequate account of generalizations whose provenance is acknowledged as fully unknown, or thought to be known in a way that just isn’t so; this is absolutely no issue for the “inductive” approach. The Inductive approach is reasonably intuitive and widely known (if not widely assumed) and I’ll not spend much time on it here. The idea is that one makes a number of observations of phenomena and then draws generalizations from that set of data. If you see Bob get up at dawn on day1 , day2 , . . . , dayn , at some point you’ll conclude that “Bob rises at dawn”. Its truth-conditions depend on deciding when one has observed “enough” instances, and what constitutes “enough”. The semantic truth-conditions of Rules and Regulations, on the other hand, takes its inspiration from games. One goes out and purchases a board game, such as checkers or Monopoly, and quite often one will find the rules of the game listed on the inside of the box cover or otherwise included with the game set. These rules, if not modals (“the first player must/may then. . . ), are generic (“Play proceeds in turn clockwise”; “Each player rolls the die to determine the number of squares to move the token”). These are never episodic sentences, although the play of the game on a given instance can be described in terms of a series of episodic sentences (“Frank moved the token six squares” “Mary collected $15 in rent on Baltic Avenue”). The rules of course are stipulated by those with the authority to write them (an inventor, an international federation, the dealer. . . )—they are not post-hoc derived from observations of players actually engaged in a game. As play of a game proceeds, the rules are “in force” as long as the game lasts and guide behavior.


23

The semantic truth-conditions are initially quite straightforward: the generic sentence is true iff it expresses a proposition (expressed by a sentence) listed on the box top—the axioms of the game. This must immediately be revised to read that it must be entailed by a (sentence or set of sentences expressing a) proposition listed on the box top. What it does best is deal with those cases of generics in which inductive evidence is either lacking or misleading, and naturally deals with the facts about intensionality of generics. However, where the approach loses credibility is where “purely descriptive” generalizations are involved, where one has no idea whatsoever of the underlying governing principles or axioms, and what they may or may not entail, but our judgments about truth and falsity seem secure nevertheless. If Bob eats fast, or if the traffic on Main Street increases throughout the afternoon, can we really explain why? This is precisely where the inductive approach is the strongest, in its agnosticism about the source of a pattern. There is, however, one area not discussed in Carlson (1995) where the Rules and Regulations approach is strong, at least in a promissory way: negated generics. A negated generic would mean “not (entailed by a sentence expressing a proposition) listed on the box top” and thus not permitted by game rules, a “strong” reading. This is where the inductive approach fails quite badly, predicting only the weaker reading which does not seem to be found. However, this all depends upon explicating a theory of the connection between the R&R approach and the instances of the episodics governed by them (e.g. a “causal” connection), a relation which is at best obscure to explicate. On the other hand, at least the outlines of the relation between behavior (or rather, episodic events) and inductive generalization, as well as many of the details, are reasonably clear, at least by comparison. In going back and forth between these two approaches, I do not imagine I will contribute to the longstanding and plentiful literature on the nature and problem of induction, or the nature of laws. What I am trying to do is to frame an analysis of generics and habituals in such a way that is consonant with these concerns, yet focuses on the empirical issues raised by the examination of natural language.

3.2 What do Generic Sentences do? As a preliminary step to considering a semantics for generics, let us ask a slightly different question from what they might mean: ask instead, what they, the rules, do. If we take the point of view suggested by the R&R approach, as well as one suggested by classic views of natural laws, the answer seems fairly straightforward: these things govern activities (however “govern” cashes out in the end). If a group of people decide to play a game of poker, their behavior will be vastly different than if they decide to play a game of volleyball. One way of looking at this is that rules and regulations constrain alternatives; they eliminate or in other ways disfavor some alternatives over others. Again, in games the idea is the clearest. I am going to invite a perspective on this by a partial analogy.

24

G. Carlson

Consider what rules of a game, I’ll use chess here, do. A chess board has 64 squares, and a “move’ is a transition from one square to another, or an ordered pair e.g. <e7, a3>. Assuming is also a “move”, at any given point for a square there are 64 possible moves, and since there are 64 squares, there are 64 × 64 = 4,096 possible moves, and so there would be more than 16 million two-move sequences, etc. OK, sit down and play a game, and I’ll give you the rules as we go along. It would help if I gave an initial arrangement and number of pieces, so fill up the first two rows as is customary. This means for one thing that <e3, f6> is no longer a possible move, on the first move (there being nothing to move at e3), so now note that the number of possible first moves has been reduced from 4,096 to 2,048 possible first moves (i.e. one does not “move” from an unoccupied square). If I further inform you that any move of the form <x, y> where x = y is not possible, it reduces the space of possibilities even more, to 2,016 first moves. If I tell you that you cannot move over or through a piece, or to a space already occupied, it goes down even more, to 384 first moves (neglecting knight moves). And then if I type the pieces for you (queen, king, pawn. . . ) and add things like “bishops move diagonally” or “rooks move on rows and columns”, and insist pieces once typed don’t change their type, with exceptions, introduce an opponent, tell you to alternate moves between opponents, etc. the number of possible first moves becomes what you actually can do: something like 24, a portion of which are inadvisable if you’re playing to win against a decent opponent. What we’ve done is to start out with a highly (not completely, but highly) unstructured possible sequence of events, and what rules have done, being constitutive rules in the example, is to create some order. We have reduced possibilities by first defining something vaguely approaching an unstructured (or “chaotic”) situation and then restricting the space of possibilities to something more limited. Note that this has the property of intensionality as discussed above, because if we know the set of generalizations then we know what moves we WOULD be able to make given any state of the board, no matter the number of previous moves, without limit. For the time being, let’s attempt to apply the kind of thinking we just went through for chess to the matter of ordinary generics. That is, we ask first what defines the space of possibilities, and then devise means by which that space of possibilities gets “narrowed down” to more manageable levels.

3.3 Inductive Generalizations Again The very strength of inductive generalization is its agnosticism about causes. Whereas rules and regulations may be somewhat convincingly thought to have causal force of some sort (as, say when one plays golf or files a tax return, the behavior is in part caused by knowledge of the relevant rules or their entailments), this is not evidently so for the “weaker” inductive generalization. If it’s so that farmers paint barns red, or traffic increases on Main Street during the afternoon hours, simply seeing that it’s so is sufficient without any background knowledge or


25

theorizing. We don’t imagine there are laws or rules specifically governing these traffic or barn-painting behaviors and these would perhaps seem to us a product of a large number of factors plus accidents of history. Thus the inductive generalizations do not sit comfortably under the “governing behavior” intuition as a guide to their fundamental nature, and rules and regulations do not find expression in quantified or inductive generalization as their foremost guiding conception. What, then, might we find as a way of expressing what inductive patterns and rule-based patterns have in common? My sense is that both approaches (and note we have set aside other processes such as abduction and analogy) give way to something more fundamental for a semantics. That is, they express patterns, patterns of events as they unfold in the world.

4 Patterns 4.1 Patterns and Non-Patterns That commonality is, I claim, that generics (and habituals) of all stripes express a pattern. It is plausible to think that a stipulative job description, for instance “A person in this position assists the chairman with financial projections”, expresses a pattern, in this case, of (prescribed) behavior, and a simple inductive statement like “Pencils have erasers” expresses a pattern, in this case, of the composition of pencils. I am putting weight here on the vocabulary word “pattern” as a convincingly plausible way to talk neutrally about the overarching commonality found in the meanings of all manner of generic sentences, such as those in (1). But can we be a bit more precise about what a pattern is? Fortunately, most of the hard work has already been done for us in defining what constitutes a “pattern” (e.g. the classic work of Shannon, 1948; Shannon and Weaver, 1949 for identifying patterns in noise, is but one such example). I have nothing to add to the mass of work in psychology, philosophy, mathematics, statistics, computer science, theories of encryption and code-breaking procedures, and so forth, regarding what constitutes a “pattern”. My goal here is to take some of those insights and apply them to the formulation of a semantics. The key ingredient I return to is that patterns are fundamentally intensional. Wittgenstein noted that a sequence like 1,3,5,7,9,. . . is itself consistent with an unlimited number of possible patterns. Though we naturally assume that the pattern exemplified by this little sequence is the set of odd integers, the question is ultimately a contrafactual one: IF one were to continue the sequence, what would the next number in order be? IF one assumes it is the listing of odd integers, then of course “11” would be next. And, if the sequence were 1,3,5,7,9,11. . . then 13 WOULD be next, and so forth. But suppose the pattern is a slightly more baroque one: it’s the sequence consisting of odd prime numbers, with every fifth one being a non-prime odd number two larger than the one before. So, at this point the sequence

26

G. Carlson

1,3,5,7,9,11,13. . . is how one WOULD continue it, and it is a sequence identical to that produced by previous pattern as well. However, at this point, one WOULD continue in the first instance with 15, and in the second with 17. And IF the two never differed no matter how long one went on, then they would express the same pattern. If one WOULD always continue the same way, patterns are the same. On this view, patterns may be exemplified by finite sequences (or, more generally, finite objects of any sort), but the pattern itself is not finite, and not identical to any particular exemplification. Instead of talking about sequences of numbers, as above, let us instead talk about sequences of events. Take a set of token events E—let’s assume it is not finite for the moment—and construe them as an ordered series, and for the sake of exposition let’s assume that the ordering is strictly temporal, to keep things simple. So we have a series e1 < e2 < e3 < e4 . . . < en < . . .. Next we type the token events according to the resources available in a language (English, in this instance) in any way whatsoever that we wish, according to how we may talk about them. So, imagine the following typing: e1 : Snow falls in Siberia. e2 : Mercury reverses orbit around the sun. e3 : Jennifer Aniston fries an egg. e4 : The nickel in a grocery store cash register turns into a rhinoceros. .... en : A bear drives a bulldozer through a chain-link fence. ... This is intuitively a series of unconnected and in some instances implausible events, and suppose it just goes on and on and on like this. What kind of world is this? It’s clearly one where if you were a resident you’d have no idea whatsoever would be coming next (quick, now, what is e5 going to be?). This would be like the chaos discussed in the first section of the paper. At e5 , who knows what’s going on, if anything, with Mercury, Siberia, and Jennifer, and we can assume nothing about any rhinoceros or nickel or grocery stores or driving bears, etc. It gets even a bit worse when we realize that just because e1 is a snowing event in Siberia, that it’s not necessarily also classified as a precipitation-in-Siberia event, or a snowing-in-Asia event, that is, until we structure the event sequences according to the generalizations about the semantic interpretation of language. In the chaotic worlds dimly imagined here, even that structuring is lost. I am going to assume that the mathematical model for this chaos is the notion of randomness. There are various definitions of randomness, and I have nothing whatsoever to add to this side of things. Here a few: (5)

The program must be at least as long as the sequence. All sequences have an equal probability of occurrence. Given a sequence Σ, one cannot predict what Σ + n will be.

The third choice above probably exemplifies most directly our intuitive sense of randomness (though it is well-known that sequences judged as randomized are, on


27

the whole, not). Suppose one were an observer of such a world as the one consisting of e1 < e2 < e3 < e4 . . . en . . .. At a given time, one will have no idea whatsoever what comes next. Furthermore, whatever help one might try and marshal to figure out what might come next—other people, oracles, computers; seers, prophets and fortunetellers, dreams and visions; weather forecasters, and, perhaps, economics forecasters—the chances of guessing what will come next are as small as they can be (functionally, 0). If you see a boulder fall off a cliff at one moment, you have absolutely no assurance at all that it will be in mid-air at the next moment, never mind that there will be any such boulder or any such cliff or any such thing as gravity to make the boulder fall, at all. For sake of convenience, I am going to talk in terms of probabilities, the second definition, but I don’t believe anything I say here relies upon this choice. In the end, I wish to have constructs (sequences of token events) which are part of the model and which represent what the talk of probabilities is about, and for present purposes, talk of “probabilities” is talk about them. I am also going to talk in terms of “chaos”, rather than “randomness”, as the latter is laden with theoretical significance that, to my understanding, “chaos” is not. If randomness really does represent chaos as used here, it is a point to be decided later. A chaotic sequence is one where P(a|b) = P(b|a) for all sequences a and b, provided P(a) and P(b) are defined for all a and b, or none are defined. Let’s begin to unpack this some. If we think of chaos as a set of non-finite sequences of events C, then any set of nonfinite sequences that is not in the set is a pattern. A pattern P is any non-empty set of non-finite sequences such that P ∩ C = Ø A generic sentence then is true iff there is some sequence that appears in the world in the domain specified that is a part (i.e. a subsequence of events) of some member of Pn where Pn is the pattern denoted by that generic sentence. That is, the portion of the world being spoken of exemplifies that pattern.

4.2 Generic Sentences The basic idea sketched above is fairly general, and in this section we will look at the idea in a little more detail. Let us assume that the set of episodic event-types of the language is defined by the contents of “the VP” as approximately characterized in Diesing (1992), and in Carlson (2003). Episodic sentences are those which are true, or false, according to whether a given episodic event, state, or process appears in the world at the time and domain specified (see Kratzer, 1995; Fernald, 2000; Jäger, 2001; Kuroda, 1972; for discussion of the issue, among others). I’ll assume a neo-Davidsonian representation here. The denotation of a VP (which includes the sentence subject) for “A parrot flew” might be: (6)

∃x [parrot(x) & fly(x, e)]

28

G. Carlson

If this is an episodic sentence, then existential closure will yield the assertion of the existence of an event: (7)

∃e ∃x [parrot(x) & fly(x, e)]

This is true with respect to world and time, if a token event e appears in the sequence of events in the model at the temporal location provide by the tense of the sentence. I assume the world is structured by events bearing relations to one another as presented in Kamp (1979). While Kamp (1979) does not include times in his models (the point is to construct times from events), I’m going to assume, for ease of exposition, that these are laid out on a temporal time-line, and that tense deictically localizes the reference time that an event must occur in, or otherwise be related to. So, suppose the time of utterance u in the parrot example above is “between” the second and third events, and that what is being uttered is “A parrot flew”.

Suppose this is said in a context where what was intended was that the parrot’s flight took place very recently. So let’s say deictically it picks out the interval [t9 , t54 ], designated by the darker line above. This means the sentence will be true iff there is some event classified as a parrot flying included within that interval. What tense does then is to localize events. Events of course can be localized not only temporally but spatially as well, in which case the sequence of events are those, say, overlapping the spatial location (e.g. “in the next room”).2 Now we consider what occurs when an episodic VP is made into a generic (keeping it as simple as possible for purposes of illustration). Then the denotation of the VP is the grounding event-type for a generalization. Here, we need to go into just a little more detail. We are taking it that the eventtypes defined in any natural language are going to be a set with the cardinality of the integers. We assume this because of recursion: it would appear that eating a bagel, beginning to eat a bagel, trying to begin to eat a bagel, beginning to try to begin to eat a bagel. . . .are different event-types; so are eating food, eating food prepared by a chef, eating food prepared by a chef raised in Malaysia, eating asparagus, eating asparagus grown in the northernmost reaches of Malaysia, eating overcooked asparagus prepared by a fat, left-handed chef who was born in the northernmost highly populated area of Malaysia.. . . 2

The treatment of the world as a sequence of episodic events (processes,states) is consistent with the Humean view (Price, 1940) that individuals are abstracted as parameters of such events, rather than the other way around (a nominalist view). I appreciate Jim Higginbotham’s help in making this observation.


29

It appears there is a strong, plausible case for natural language defining an unlimited class of event-types. In a sequence of events e1 < e2 < e3 < e4 . . . en . . ., the types may be drawn from an unlimited class. The types, of course, are what the probability distributions are calculated on. If one has an unlimited cardinality of types, then in the chaotic instance C, 0 probability does not mean an event of the appropriate type does not in fact occur; it only means that it has (functionally) a 0 probability of occurrence and, in an infinite sequence, will occur only a finite number of times. Let us call such things “miracles”. Further, in an infinite sequence of event-types, the probability of the occurrence at a given point of any one event-type approaches 0, but is not a defined number. There is no extremely small probability measure that, when applied to a type in the context of an infinite set of types as equal possibilities, yields a number such that when recursively added infinite times, yields a sum of 1.0. Once again, this does not mean that occurrences are impossible, only that their probability of occurrence is functionally 0. So, in the context of drawing a number from a hat, where the (rather large) hat contains one token each of all the integers, drawing a number between 1 and 20 has no chance of occurring at all. Yet, one might just draw the number 14. But, this would be a “miracle”.3 Before we proceed, let us distinguish several cases, and provide them with a label.4 We first have the case where the probability is undefined because it approaches 0 but its exact value cannot (in principle) be determined; these are the instances of event-types in the context of an infinite set of event-types. Let us call this the “Zero” case. Then there is the case of a finite set of possibilities in the context of an infinite set. Let us call this the “Miracle” case. Both the Zero case and the Miracle case do not preclude the possibility of occurrence; that is, they are compatible with the existential assertion that an event e of the appropriate type, in fact, occurs. Then there is the instance of an event-type T such that no instance of T ever occurs in an infinite sequence of typed events. This is the “Never” case (distinct from the “Zero” case, as the Never case is a negated existential claim and the Zero case a probabilistic one). Suppose S is a generic (or habitual) sentence. Its interpretation is to provide “constraints” on C to create a C such that for an event-type E, P(E) is < 0 (i. e. it is defined, and the value is greater than zero). Think, once again, of constraining chess moves as in the illustration discussed above. Thus if there were a generic statement of the sort “Snow stays on the ground after it falls”, then at the next time slice snow is, to a detectable degree of probability, present on the Siberian ground if it has just snowed there. That is, we can “expect” snow to remain with some degree of stability among all the other instabilities surrounding it—given this, it is the only thing we can reasonably expect. No other event-types achieve this degree of probability. We can accomplish this by dividing the events in each member of 3

I hope that the term “miracle” here is understood, as intended, in an entirely mundane way. I think I have 67 cents in my pocket and it turns out to be 68 cents; a penny has “miraculously” materialized out of nothing into my pocket. An extra piece of gravel appears at a construction site. A car horn honks in the distance. 4 I owe my thanks to Carl Mueller and Peter Farkas for their efforts in clarifying these distinctions for me. Neither is responsible for any of this, though.

30

G. Carlson

C into two types. One is defined by the event type of the “grounding” event-type provided by the interpretation of the VP, E. Then, all other events in any sequence of C are classified as −E. Thus from a chaotic sequence where no probabilities are defined for the types, emerges a structure where p(E) + p(−E) = 1. Any sequence with this structure is a generalization on E, the denotation of a generic sentence. The grounding event-type for a pattern is the only event-type in the pattern defined by the language (−E is not one of the event-types defined in the language) that achieves a probability of 1 (the monotonous world) and is lower-bounded by the probability by something approaching 0, dipping into the range of likely hardly noticeable generalizations. All other event-types defined by the language that are non-grounding are zero cases. (This presumes a structure that does not reflect any definition of entailment—that would be a “later” structuring). What, then, might be the truth-conditions of a generic sentence S? The answer to this is actually quite simple: it is true iff it is a part of a pattern based on the event-type denoted by the VP. It is part of that pattern just in case the token events of the world of evaluation w are found in a sequence of the pattern; it follows then that the temporal relations among the tokens will also hold. GEN(VP) is true iff ∃P ∃s ∈ P [P(|VP|) > 0 & w ≤ s] How one can determine this, in any practical sense, is another matter. The problem is that any sequence of finite length, no matter how apparently well-ordered, could have been generated at random, that is, could be a sequence that is part of a sequence in C. 1, 3, 5, 7, 9, 11, 13, 15 could of course be a sequence of integers from a randomly-generated sequence. And, as we’ve already observed, it can only be used to suggest a pattern (e.g. odd integers), but not to define one. So one is always, under all circumstances, making a guess in a sense about whether a given generalization holds or is accidental. They all could be accidental; in principle, one cannot prove otherwise. As indicated in the commentary in the opening section of the paper, however, such an extreme skeptical position as this makes the order of the world something of a surprise, and a rather huge one at that. The primary feature of this proposal is that patterns are not necessarily absolute. They may come with “noise” as a part of them; the generalization is the “signal” part. Here is an analogy (though it may be lost on some). Consider watching a television set that is not getting very good reception. One can see on the screen mostly what is called colloquially “snow” but within it may be found, say, the characters of your favorite television comedy show speaking to each other. If one tuned in a little more, the definition would become clearer, i.e. the noise or “snow” reduces and the picture would emerge more clearly. Tune in a little more, etc. Each image projects the same pattern. The question is how clearly the pattern is represented by the images projected, not whether the pattern is the same (Fig. 1). The same hold for generalizations. Some come through the clutter more clearly than others. If Stuart smokes every minute of every day, then the pattern described by “Stuart smokes” emerges more clearly than if he smokes that often for only half a day, and that more clearly than if he smokes twice a day out in some hidden spot.


31

Fig. 1 Sitcom on a television with poorer and better reception

If this machine crushes oranges for orange juice, the pattern comes through more clearly if it is engaged in that activity both in the night and day shifts, and less clearly if it is put in some storage room in the back until the orange juice market picks up a bit. However, what might strike many as somewhat odd is that generics can also hold of sequences of events where the type of event (such as smoking, or crushing oranges) never appears. Take an instance of Stuart walking slowly across the street and then scratching his head. This small sequence of events may (or may not) be in the pattern designated by “Stuart smokes”; it may (or may not) be in the pattern designated by “Stuart runs fast”; in “People eat ice cream”, and so forth. It all depends on whether the token events here e1 < e2 that are members of E are, or are not found in one of the sequences designated by Pn .

4.3 Restriction Generic states, like episodic events, can be localized (and typically are), despite the emphasis in some portions of the literature on the unchangeable, eternal, and universal nature of generalizations. This may be true for such things as physical laws or laws of mathematics, but in our everyday world generic states do come and go, and are often localized in other ways, e.g. by space. It used to be that the largest predator in South America was a bird, but it’s not now. American gentlemen used to wear powdered wigs, but not now; in Rome they wore togas, but not now. This localization is indicated not only by tense, but often by a prefixed “in” framing adverbial phrase: (8)

a. b. c. d. e. f.

In German, word-final stops are devoiced. In Canadian elections, debates are blunt and protracted. In 16th-century Paris, university students shared books. In the North Atlantic, the south-moving ocean current moves near ocean bottom. In chess, bishops move diagonally. In prehistoric times, people ate mostly nuts, berries, and meat.

32

G. Carlson

The function of such “in” phrases (and other adverbials) is to construct sequences of events that hold only of the frame mentioned (e.g. “In German” is the sequence of all events where someone is speaking or otherwise using German; “In chess” includes only those instances of the game being played, etc.). However, “restriction” is more often discussed with respect to generics as a fundamental aspect of their interpretation. Kuroda (1972), Ladusaw (1994) discuss generics in terms of “categorical” vs. “thetic” judgments. The categorical judgments are fundamentally two-place relations: does this belong to that category? Most of the work on restriction, though, is inspired by work on quantification and focus. In Krifka et al. (1995), this is directly represented in the analysis of GEN as a twoplace operator (forming a “tripartite structure”): GEN (Restrictor) (Matrix). The restrictor part functions as it does in quantification, that is, providing a context in which the matrix is interpreted. If this is the structure of the interpretation of a sentence, then the interpretation of some parts need to be assigned to the restrictor, and the remainder to the matrix. This is most evident in cases where there seem to be more than one interpretation available. Consider (9): (9)

a. b. c.

A weatherman delivers the daily weather forecast. When spring arrives, geese migrate to Canada. Meteors appear in the sky around midnight.

In (9a), there is an ambiguity regarding whether one is defining what weathermen in general do, or how the daily forecast gets delivered to us (i.e. by a weatherman). The first interpretation supports contrafactual such as “If John were a weatherman (which he is not, he’s a milk truck driver), then he would deliver the daily weather forecast”, whereas the second does not. In (9b) the question is whether this is a report about the character of geese (“all” of them fly to Canada) or about something that occurs in the spring (you’ll see (some) geese flying that way). In the last instance, the question is whether one is saying that only the odd or exceptional meteorite appears in the sky at a time other than midnight, or whether if you look in the sky at midnight you will see meteorites. These ambiguities can be modeled in terms of what appears in the restrictor and what appears in the matrix. See Krifka et al. (1995) and references therein. The restrictor is the information that is taken as “given” (it may be presupposed), but whatever the precise linguistic status, it does suggest a Bayesian interpretation of conditional probabilities. So, suppose the linguistic structure of (9b) above is, on one interpretation, schematized by the following: Restrictor: [(if) spring arrives] Matrix: [(some) geese migrate north] We can interpret this as saying that the probability of geese migrating GIVEN the arrival of spring is a pattern (so if S is “spring arrives” and G is “(some) geese migrate”, p(G|S) < 0.) What we do then is to restrict our consideration of sequences to the occurrence of S-type events, and then a check to see if G events which are appropriately related to the S events (e.g. occur after, or about the same time as, are two common relations—see Dowty, 1986 for one such discussion). How in practice this is implemented will remain unexplored at the moment. However, it does seem


33

likely that all generic and habitual sentences do contain such restriction as a part of the semantics, with additional restriction possibly provided pragmatically.

4.4 Review of Issues Even at this level of generality, a number of issues arise. Let me at this point merely address the issues discussed in Section 2, and how this might serve as a means of understanding them. I harped on intensionality. The real world, or at any rate our experience with it, individual and collective, is finite. The representation of patterns here, however, involves non-finite sequences of events which necessarily include events of the type the generic is based on occurring with a frequency greater than is to be found among the members of C. All other event-types in that pattern, again in the most general case, may occur with varying degrees of frequency in each sequence (they constitute the “noise”), including frequencies indistinguishable from C. Some of these other event-types will be parts of other patterns (i.e. will constitute the event-type that a generalization is based upon). With the resources that are available, one can construct structured sequences which will be isomorphic to a possible world (this is a brand of situation semantics, after all, in the spirit of Kratzer, 1989). That is, the system will be able to deal with all the intensional phenomena that possible worlds may deal with. The open question then is whether it provides extra power to do a bit more. The treatment of exceptionality, under this proposal, is basically already built into the representations. If we insist on a deterministic universe, then we would stipulate that all events in every sequence must be the grounding for generalization. I see no reason to insist on this, though, and allow that some things “just happen” without necessarily being governed by one generalization or another (the limiting case for this is the set of sequences in C). “Exceptions” are those instances which appear in the pattern under a stronger, “clearer”, representation, that is, one where the frequency of the grounding event-type increases monotonically in the direction of 1.0. In the television analogy mentioned above, these would correspond to the pixels (or groups of them) which fail to convey the originating information (e.g. the black spots in the otherwise seamlessly blue sky). “Accidental” generalizations, the converse of exceptions, are events of the same event-type as some grounding event-type for a pattern, but do not appear in Pn in any sequence that the world (or rather, the portion under discussion, see below) is a part of. Again, in the television analogy, these would correspond, say, to say the nearly vertical straight line of black dots on Paris Hilton’s red throat accessory (or, in another domain, constellations of stars). The treatment of negation is intuitively in line with the observations presented earlier. The truth-conditions for generics involve an existential assertion; negation of that is the negation of an existential assertion. Negation demands the absence of any pattern of the type specified, i.e. for there to be no such pattern at all. Once again, as the television signal gets worse, the pattern on the screen fades, and at

34

G. Carlson

some point is gone. That is the condition negation of a generic points to, when it is gone. If there is any pattern of the appropriate type present at all, however “noisy,” the existential holds. Negation does not demand the total absence of the event-type. It is why it is characterized as having a “strong” meaning, rather than an “absolute” meaning. If someone does not smoke, they might at some point try a cigarette or two to see if they like it; if gazelles do not attack lions, it allows for one of them to do so in an epileptic fit (for instance). This is quite obviously not even a remotely psychologically plausible approach to the semantics of generics, provided we continue to assume the finiteness of the brain. We do not, I believe, have direct access to the token event identities the theory requires. And we most certainly do not have any mental capacities to represent chaos. Chaos is precisely what we cannot know about (though surely there are many ungraspable patterns, as well). Further, the cardinalities of the sequences would appear to preclude any possibility of mental representation (see Partee, 1980). Thus, to construe what has been suggested here psychologically would have to mount at least these challenges. This leaves us then with the challenge of asking, how do we know if a generic sentence is true or not? After all, what we could be witness to is an apparent pattern which is some subsequence of a random sequence. Since one can always say that, under all circumstances, it gives rise to the classic problems of induction that I’m hardly going to add anything to here. If by “know”, one means “know for absolute certain based on evidence” in some philosophically satisfactory sense, then, no, we can never know whether any generic is true or not. All I can say is that we are suspicious creatures who do not like coincidences. It is well-known that in many cases we overattribute patterning where there is none (e.g. games of chance). We have devised various strategies for at least trying to sift the coincidence from the real. For event-types that recur with at least some moderate frequency in our experience, such as habits, customs, etc. an inductive strategy seems in place, one that has attracted a great deal of attention. For much weaker patterns, such as dispositions or propensities or functions, we devise something like criterial conditions or something like standards of normalcy. We also have abductive strategies (If A happens, then B, what generalization would cover that?). And then, let us not forget, most of what we “know” comes from simply being told so. We have no means of knowing New York state law except by being told what the law is. One learns to play a board game mostly by being guided through the rules by someone who already knows how to play, or maybe reading the rules. These also may be differentiated by the role we believe they play, be it a social construction of varying sorts, a natural or an artifactual one.

5 Weak and Really Strong Generalizations A primary difficulty this proposal meets with is the truth-conditions of what I will call weak generalizations. Above I touted this as a strength, as it does provide the basis for the range of generic meanings notionally attested, and it provides for an


35

account of negation which I believe is in line with the facts. However, it sets a seemingly low bar for truth. I would, in general, argue that this is correct. For instance, a common type of exchange might sounds as follows: (10)

A says: Dogs don’t eat tomatoes. B replies: Dogs DO eat tomatoes! I have a dog that just loves them!

The argument here is not over whether it is a prevalent fact about dogs that they like to eat tomatoes, but whether it is a pattern in keeping with the character of dogs for those which do so, to do so. Similarly, we can make sense of the following paradigm of examples, all of which I believe are true, even assuming that one sheep in a dozen is black, and the rest are white: (11)

a. b. c.

Sheep are black and white. Sheep are black or white. Sheep are black, and sheep are white.

There is an ambiguity in all these examples, but the intended reading is the one where the speaker is in control of full knowledge about sheep’s colors. If we take it that “Sheep are black” is false, then the only one of the sentences that could be true (on the intended reading) is the disjunction. However, given that “Sheep are purple” is false, and assuming the speaker knows it, the following seems quite strange as a true assertion, unlike (11b) above: (12)

Sheep are purple or white.

So, in the general case, I believe there is evidence that weak generalizations of this sort should be countenanced. However, let us return to the question of the children of Rainbow Lake discussed in Schubert and Pelletier (1987). Let us take the grounding event-type as “A Rainbow Lake child is born left-handed”. If we take this to be a “principled” generalization, then we would be claiming that there is a pattern Pn such that the events e1 . . . e8 of that type appear within some sequence found in Pn . In other words, there is a sequence that is a member of Pn containing all eight events of left-handed births, where if you were to continue on in that pattern, the probability of left-handedness would be distinguished from any probabilities in C. Yes, this is right, but it would also be correct if all eight children had been born right-handed if considered to be a normal sampling of the population. Similarly, given that any horse that runs fast also runs slowly at times, it would be true that horses run slowly and at the same time that horses run fast. So, while I believe encompassing weak generalizations within the bounds of a semantics of generics is right, there remains some reason to give one pause. The converse appears somewhat more tractable: instances where the generalization appears (nearly) absolute. Rules of games are a good example: the rules do not express patterns which include all manner of unexplained departures. Rather, the rules appear to be universal, without exception, unless of course the exception

36

G. Carlson

is explicitly provided for by other rules. And the rule-system as a whole has extraordinarily low, or no, tolerance for apparent exceptionality. On some random poker hand, no one is able to ignore the fact (if a rule of that style of poker), particularly to one’s advantage, that tens outrank nines, for instance. I believe that the present approach offers a means of understanding this. Essentially, such rule-systems are patterns with the “noise” squeezed out of them. The mechanism for accomplishing this is a restriction on the types of events in the patterns described, mediated by the game’s specialized vocabulary. For instance, while a “move” in chess or a “run” in baseball use common terms, their denotations in a game are strictly defined. A move in chess is not simply the displacement of a piece from one square to another, say, by an earthquake or the tail of a passing dog, but is strictly defined by the rules of the game. And many things are simply undefined by the rules of the game, such as “bathtub” in chess or “major surgery” in basketball. What I’m suggesting then is that the “world” the game rules described is limited just to those sequences in which every event e is described by some eventtype defined by the specialized vocabulary of the game, and none which are not. Thus, every event will fall under some event-type that grounds a generalization of the game. One could also think of laws, social injunctions, systems of ethics, and other consciously-defined constitutive arrangements, in similar terms.

6 Conclusion In this paper it has been proposed that generic and habitual sentences have quite simple truth-conditions associated with them, although the epistemic problem remains. It is proposed that the real world consists of a course or sequence of events (many of which may overlap one another), and that temporally or otherwise circumscribed portions of the world exemplify (or not) patterns that form the basis for the truthconditions of the sentences. The proposal promises success in dealing with the intensional nature of genericity (as do other proposals), of dealing with exceptionality, and the problem of negated generics. It promises a perspective on the notional disunity of the field of generic sentences, and perhaps some insight into the nature of rules and laws. The problem of when we can meaningfully and truthfully assert certain weak generalizations remains an unresolved issue. The most interesting possibility, as far as I am concerned, is the potential consequences for a theory of modality employing such notions as those outlined above. Using the representations of the type suggested, we can in principle distinguish generalizations from one another which are not distinguishable in a possible worlds framework. For example, “2 + 2 = 4” is a different generalization from “3 + 3 = 6”, the former being based on events of adding two to two, and the other based on a different type of event, the adding of three to three. Here, an outline for a means of differentiating them is provided, and further exploration of the possibilities inherent in this approach may prove the above incoherent, as coherent but having an unsuitable structure for representing meanings at a finer grain than as functions from possible worlds to truth values; or, as coherent and as having the right sort


37

of structure, not only to represent the meanings of generics, but also as the basis for further work on modality.

References Asher, N. and M. Morreau (1995). What some generic sentences mean. In G. Carlson and F. J. Pelletier (eds.) The Generic Book. University of Chicago Press, Chicago, IL. 300–338. Bennett, M. (1972). Some Extensions of a Montague Fragment of English. Indiana University Linguistics Club. Carlson, G. (1995). Truth conditions of generic sentences: two contrasting views. In G. Carlson and F. J. Pelletier (eds.). 224–237. Carlson, G. (1999). Evaluating generics. In P. Lasersohn (ed.), Illinois Studies in the Linguistic Sciences, 29:1, pp. 1–11. Carlson, G. (2003). Weak indefinites. In Martine Coene and Yves D’Hulst (eds.) From NP to DP: On the Syntax and Pragma-Semantics of Noun Phrases. Benjamins Publishing. 195–210. Carlson, G. and F. J. Pelletier (eds.) (1995). The Generic Book. University of Chicago Press. Carlson, G. and G. Storto (2006). Sherlock Holmes was in no danger. In B. Birner and G. Ward (eds.) Drawing the Boundaries of Meaning. Benjamins, Amsterdam. 53–70. Chierchia, G. (1995). Individual-level predicates as inherent generics. In G. Carlson and F. J. Pelletier (eds.). 176–223. Cohen, A. (1999). Think Generic: The Meaning and Use of Generic Sentences. Stanford, CA: CSLI. ¨ (1975). On generics. In E. Keenan (ed.) Formal Semantics of Natural Dahl, O. Language. Cambridge; Cambridge University Press. 99–111. ¨ (1985). Tense and Aspect Systems. London: Blackwell. Dahl, O. ¨ (1995). The marking of the generic/episodic distinction in tense-aspect Dahl, O. systems. In G. Carlson and F. J. Pelletier (eds.). 412–425. Davidson, D. (1967). The logical form of action sentences. In Rescher (ed.). The Logic and Decision of Action. University of Pittsburgh Press. 81–95. Diesing, M. (1992). Indefinites. Cambridge, MA: MIT Press. Delgrande J. (1988). ‘An approach to default reasoning based on first-order conditional logic.’ A revised report. Artificial Intelligence 36: 63–90. Dowty, D. (1986). ‘The effects of aspectual class on the temporal structure of discourse: semantics or pragmatics.’ Linguistics and Philosophy 9: 37–61. Eckardt, R. (2000). ‘Normal objects, normal worlds, and the meaning of generic sentences.’ Journal of Semantics 16: 237–278. Fernald, T. (2000). Predicates and temporal arguments. Oxford University Press, Oxford. Greenberg, Y. (2003). Manifestations of Genericity. Routledge, New York.

38

G. Carlson

Greenberg, Y. (2007). Exceptions to generics: Where vagueness, context dependence, and modality interact. Journal of Semantics 24: 131–167 Jäger, G. (2001). ‘Topic-comment structure and the distinction between stage level and individual level predicates.’ Journal of Semantics 18: 83–126. Kamp, H. (1979). Events, instances, and temporal reference. In Bäuerle, Egli, and A. von Stechow (eds.) Semantics from Different Points of View. Springer, Berlin. 376–417. Kratzer, A. (1989). ‘An investigation of the lumps of thought.’ Linguistics and Philosophy 12: 607–53. Kratzer, A. (1995). Stage-level and individual-level predicates. In G. Carlson and F. J. Pelletier (eds.). 125–175. Krifka, M., Pelletier, F.J., Carlson, G., ter Meulen, A., Chierchia, G. and Link, G. (1995). ‘Genericity: An introduction.’ In Carlson. G. and Pelletier, F. J. (eds.). 1–124 Kuroda, S-Y. (1972). ‘The categorical judgment and the thetic judgment.’ Foundations of Language 9: 153–85. Ladusaw, W. (1994). Thetic and categorical, stage and individual, weak and strong. In Proceedings of the Conference on Semantics and Linguistic Theory 4. Cornell University, Dept. of Modern languages and Linguistics, Ithaca, NY. 220–229. Lawler, J. (1973). Studies in English generics. University of Michigan Papers in Linguistics, 1:1. University of Michigan, Ann Arbor, MI. Lewis, D. (1973). Contrafactuals. Oxford: Blackwell. Partee, B. (1980). Semantics – mathematics or psychology? In Bäuerle, Egli, and von Stechow (eds.) Semantics from Different Points of View, Springer, Berlin. 1–14. Portner, P. (1992). Situation Theory and the Semantics of Propositional Expressions, University of Massachusetts Ph.D. dissertation. Price, H. H. (1940). Hume’s Theory of the External World. Oxford at the Clarendon Press. Schubert, L. and F. J. Pelletier (1987). Problems in the representation of the logical form of generics, plurals, and mass terms. In E. LePore (ed.) New Directions in Semantics. London: Academic. 385–451. Shannon, C. E. (1948). A mathematical theory of communication, Bell System Technical Journal 27: 379–423, 623–656. Shannon, C. E. and W. Weaver (1949). The mathematical theory of information. University of Illinois Press, Urbana, IL.

Intensional Subjects and Indirect Contextual Anchoring Ileana Comorovski

Abstract This paper brings evidence in favour of the semantic-pragmatic characterization of subjects of specificational copular clauses proposed in Comorovski (2007). The evidence is based on Romanian questions of the form Care-copulaDP? (‘Which/what copula DP?’), some of which are argued to be specificational questions. It is shown that the subject of these questions is of the type of individual concepts (<s, e>) and that it must contain a referential expression that anchors it to the context of utterance. The subject of a specificational care-question is generally definite; it can be indefinite just in case the question contains a subjective predicate and the copula is in the conditional mood. We show that an indefinite noun phrase cannot be the subject of a constituent question unless the question has a topic other than the subject. We argue that the point of view argument of a subjective predicate can function as a sentence topic. Therefore, if specificational care-question contain a subjective predicate, they allow an indefinite subject since they have a non-subject topic. Key words: Copula, intensional noun phrase, topic, point of view, evidentiality

1 Introduction This paper examines the semantic properties of the subjects of certain interrogative and declarative clauses of the form DP copula DP. The focus of our attention are Romanian copular constituent questions introduced by care (‘which’, ‘what’), such as the question in (1) below. We will consider the semantic properties that the

This is a revised version of a paper that appeared in the proceedings of the 2004 Antwerp Workshop on the Expression of Time and Space (Comorovski, 2005). I wish to thank Jacqueline Guéron and Jacqueline Lecarme for the pleasant atmosphere they created at the Paris round table, which made it a constructive and stimulating event. I also wish to thank them, as well as a Springer anonymous reviewer, for their comments on drafts of this paper. Université Nancy 2/CNRS


39

40

I. Comorovski

postcopular DP must have and show that it must be of an intensional type (<s, e>) and that, moreover, it must be linked to the context of utterance by a referential expression it contains: (1)

Care e [DP<s,e> capitala Moldovei] ? capital-the of-Moldavia CARE is ‘What is the capital of Moldavia?’

The Romanian questions of the form Care copula DP? will be analyzed in light of Higgins’s (1973) taxonomy of copular clauses (summarized in section 2.1 below). One class in Higgins’s taxonomy is that of specificational clauses, illustrated by (2): (2)

[DP The guests] were Julie and Jerry.

Higgins characterizes specificational copular clauses as having a subject that acts as the heading of a list; the complement of the copula specifies the members of the list. We will compare the semantic restrictions on the postcopular DP in Romanian carequestions to the restrictions on the subjects of English specificational clauses. This comparison will lead us to the conclusion that the Romanian copular care-questions are the constituent question counterpart of declarative specificational clauses. The Romanian data in this paper come to support a proposal developed in Comorovski (2007), who argues that the copula can create an intensional subject position. Using mostly French and English data, she proposes that subjects of specificational clauses call for a semantic/pragmatic characterization that makes reference both to their intensional semantic type and to their special relation to the context. The paper is organized as follows: section 2 outlines Higgins’s (1973) taxonomy of copular clauses and summarizes the results obtained by Comorovski (2007); it presents a definition of the copula of specification, which is argued to be the source of the specificational reading of clauses of the form DP copula DP. Section 3 describes the semantic restrictions on the subject DP of Romanian questions of the form Care copula DP?, showing that they are very similar to the semantic restrictions on the subject of English specificational clauses. Section 4 examines specificational sentences with indefinite subjects and analyzes the role played by subjective predicates in licensing indefinite subjects in Romanian specificational care-questions. Section 5 concludes the paper.

2 Specificational Copular Clauses 2.1 Higgins’s (1973) Taxonomy of Copular Clauses Higgins’s taxonomy of copular clauses is summarized in (3) below; the taxonomy concerns clauses in which the copula is flanked by two noun phrases: (3)

a.

predicative:

Alice is a writer.

Intensional Subjects and Indirect Contextual Anchoring

b.

identity (equative):

c.

specificational:

d.

identificational:

41

The Morning Star is the Evening Star. Samuel Clemens is Mark Twain. The author of the book is Alice. The candidates were Alice, Tom, and Frank. That is Alice. That woman is Alice.

The distinction between classes (3a) and (3b) of copular clauses – predicative and identity copular clauses – is a classical one. Classes (3c) and (3d) are introduced by Higgins. Higgins informally characterizes specificational copular clauses as having a subject that acts as the heading of a list; it therefore has descriptive content. The complement of the copula specifies the members of the list. Higgins emphasizes the fact that the subject of specificational clauses is not referential, but has an attributivelike reading (which he calls ‘superscriptional’). Identificational clauses constitute the least clearly characterized class of copular clauses. Their subject, which is a demonstrative pronoun or a noun phrase introduced by a demonstrative determiner, is used deictically. The role of the complement of the copula appears to be twofold: to identify the referent of the subject and to attribute a property to it. English offers few clear-cut tests that help distinguish specificational clauses from the other types of clauses of the form DP copula DP. One test that appears to work fairly consistently is the tag question test proposed by Mikkelsen (2004). Mikkelsen shows that a singular subject of a specificational clause requires the pronoun it in a tag question, even if the subject is [+human]. In contrast; a singular [+human] subject of a predicative or equative clause requires the pronouns he or she in a tag question: (4)

a. b. c.

The (female) winner is Susan, isn’t it/∗ she? (specificational) Susan is a linguist, isn’t she/∗ it? (predicative) Samuel Clemens is Mark Twain, isn’t he/∗ it? (equative)

Thus the tag question test is useful in distinguishing specificational clauses from predicative and equative clauses of the form DP be DP.

2.2 The Subject of Specificational Clauses Higgins sums up his descriptive generalizations concerning specificational subjects in a table that is partly reproduced below: Table 1 (Higgins, 1973: 264) Constituent type Deictic Proper name Definite NP Indefinite NP

Referential + + + ?−

Superscriptional − − + ?−

42

I. Comorovski

Table 1 indicates that specificational subjects cannot be proper names and deicitc expressions, by which Higgins (1973) understood demonstrative pronouns and noun phrases introduced by a demonstrative determiner. Thus standard rigid designators cannot function as specificational subjects. As we see in Table 1, specificational subjects can be definite descriptions. They cannot be definite descriptions used deictically (referentially, in Higgins’s terms), as evidence by Mikkelsen’s (2004) tag question test: (5)

The woman in green is Alice, isn’t she/∗ it?

As argued in Comorovski (2007), definite specificational subjects, even though not referential themselves, must contain a referential expression or a referential implicit argument, as in (6) below. The tag question test shows the copular clauses in (6) to be specificational: (6)

a. b.

Your best friend is Alice, isn’t ∗ she/it? The next candidate is Alice, isn’t ∗ she/it?

The fact that specificational subjects cannot be quantifying noun phrase is already obvious from Higgins’s informal characterization of specificational clauses. We can see from Table 1 that Higgins (1973) was unsure about whether indefinites can or not function as specificational subjects. We will tackle the topic of indefinite subjects in Section 4. In Comorovski (2005, 2007) we have suggested that specificational subjects are of type <s, e>.2 To this generalization, we have added the observation that specificational subjects must contain a referential expression or a referential implicit argument; we have introduced the term ‘indirect contextual anchoring’ for the relation established between an intensional noun phrase and the context of utterance with the help of a referential element that the noun phrase contains. We have given the following characterization of specificational subjects: semantically, they are of type <s, e> (functions from indices to individuals). They must also fulfill a pragmatic condition, which reflects their indirect contextual anchoring: they are linked to the context of utterance by a referential expression or by a referential implicit argument they contain. We state our proposal in (7): (7)

A semantic-pragmatic characterization of specificational subjects: Specificational subjects are of type <s, e> and must be indirectly contextually anchored.

2.3 The Copula of Specification Given the generalization in (7) above, Comorovski (2005, 2007) suggests that the copula can create an intensional subject position and proposes a definition for the 2 The proposal that specificational subjects are of an intensional type has been independently made by Comorovski (2005) and Romero (2005). Romero’s arguments come from the interpretation of concealed questions.


43

copula of specification. The definition takes into account both the intensionality of the subject position and the indirect contextual anchoring of the subject noun phrase. The relevance of both intension and context of utterance leads to the use of the semantics proposed in Kaplan (1977, 1989) for the interpretation of sentences containing indexical expressions. Kaplan’s semantics makes it possible for the denotation of the copula of specification to be stated relative to two parameters: an index (world-time pair <w, t>) and the context of utterance, noted c. The former is necessary given the fact that the subject of a specificational clause is an individual concept; the latter makes it possible to represent the indirect contextual anchoring of the subject noun phrase. Below is the definition of the copula of specification: (8)

[[copulasp ]]M,w,t,c,g = [[λ xλ u[∨u = x]]]M,w,t,c,g Condition: [[∨ u]]M,w(c),t(c),c,g ∈ c (where x is a variable over individuals and u is a variable over individual concepts)

Since specificational subjects are linked to the context of utterance by a referential expression they contain, the individual which is their value at the current index is normally an element of the context of utterance. Given the condition in (8), the assignment function g must pick as a value for u an individual concept that has as a value at the world and time of the context of utterance an individual that is an element of this context. Thus the definition in (8) adequately represents the two characteristics of the specificational copula: the fact that its external argument denotes an individual concept and the fact that, due to its indirect contextual anchoring, the denotation of this argument at the world and time of the context of utterance is an element of the context of utterance.

3 Romanian Questions of the Form Care Copula DP? 3.1 Two Types of Discourse-Linking Romanian care occurring without a following N has been analyzed as an inherently disourse-linked interrogative phrase (Comorovski, 1989, 1996; Dobrovie-Sorin, 1994), i.e. as the correspondent of English which. It has gone unnoticed that there is one environment in which care need not be discourse-linked, namely questions of the form Care copula DP?. The possible non-D-linked reading of care is seen in (1) above. In order to be able to state the semantic restrictions on the postcopular DP, we will make a distinction between ‘weak’ and ‘strong’ discourse-linking. If a whphrase ranges over an already given set, we will call this type of D-linking weak discourse-linking. If, moreover, the constituion of the set over which a wh-phrase ranges is known by both speaker and hearer, we will call this type of D-linking

44

I. Comorovski

strong discourse-linking. The ‘weak’ type of discourse-linking is equivalent to Pesetsky’s (1987) discourse-linking. The ‘strong’ type of discourse-linking is a subcase of Comorovski’s (1996:11f.) discourse-linking. Comorovski (1996) argues that English which and Romanian care are never weakly D-linked. Yet we will see below that copular care-questions present an exception to this generalization.

3.2 Semantic Restrictions on the Postcopular DP If care is strongly discourse-linked, there are no semantic restrictions on the postcopular DP of questions of the form Care copula DP?. If care is non-D-linked or weakly D-linked, the following semantic restrictions operate on the postcopular DP of such questions: (9)

a.

b.

If definite, the postcopular DP cannot be: (i) a rigid designator; (ii) an ‘incomplete definite descriptions’; The postcopular DP cannot be quantifying.

We will examine these restrictions in turn.

3.2.1 Restriction on Rigid Designators In the questions in (10–11) below, care can be (and is most naturally) interpreted as non-D-linked (10a, b) or weakly D-linked (11a, b). The postcopular DPs are all non-rigid. (10)

a.

b.

(11)

a.

b.

Care e capitala Moldovei? what is capital-the of-Moldavia ‘What is the capital of Moldavia?’ Care e adresa lui Ion? what is address-the of John ‘What is John’s address?’ Care e singurul naist român cunoscut ˆın str˘ain˘atate? which is only-the pan-piper Romanian known abroad ‘Which is the only Romanian pan-piper known abroad?’ Care e cel mai important compozitor român? which is the most important composer Romanian ‘Which is the most important Romanian composer?’

In (10a, b) care is non-D-linked: in (10a), care does not range over a given collection of capitals or of cities; in (10b), it does not range over a given collection


45

of addresses. In (11a, b), care is weakly D-linked: question (11a) presupposes the existence of a collection of Romanian pan-pipers and question (11b) presupposes the existence of a collection of Romanian composers. The constitution of these collections need not be known (and most likely is not known) by the participants in the conversation. In contrast, in the questions in (12) below, where the postcopular DPs are all rigid, care can only be interpreted as strongly D-linked: (12)

a.

b.

c.

d.

Care e a˘ sta ? (fam.) which is this ‘Which one is this?’ Care e Ion ? which is John ‘Which one is John? Care es, ti tu ? which are you ‘Which one are you?’ Care e satul a˘ sta ? which is village-the this ‘Which village is this one?’

Here are some contexts in which these questions could be asked: question (12a) could be asked while looking at a person that belongs to a group of people already introduced in the discourse and whose members are known (e.g., by name). Questions (12b, c) could be asked while looking at the picture of a group of people which includes John in the case of (12b), and the hearer in the case of (12c). Question (12d) could be asked by a tourist while driving with his friends through a historical village which is one of a set of villages that he and his friends had planned to visit. The description we have given of some appropriate contexts for the questions in (12) makes it clear that care is interpreted as strongly discourse-linked in these questions, i.e. as ranging over a set whose constitution is known by all the participants in the conversation.

3.2.2 Restriction on Incomplete Definite Descriptions Incomplete definite descriptions are singular definite noun phrases whose descriptive content is not rich enough to uniquely identify the referent of the noun phrase: (13)

The table is covered with books. (Strawson, 1950) (Rom: Masa e acoperit˘a cu c˘art, i.)

Definite descriptions such as the table in (13) can satisfy the uniqueness condition associated with singular definites only if a restricted context is taken into consideration.

46

I. Comorovski

If an incomplete definite description appears in a copular care-question, care can only have a strongly D-linked interpretation: (14)

Care e rˆıul / satul? CARE is river-the / village-the “Which one is the river / the village?”

(14) could be asked in front of a map on which several rivers and villages are represented. The speaker asks for the identification on the map of a river/village that was already referred to in the discourse. Thus the context that “completes” the definite descriptions in (14) and ensures their uniqueness is the discourse that has preceded the question. What is the possible status of the antecedent of the definite description in (14)? Since the speaker asks the hearer of (14) for the identity of a given river/village, it follows that the hearer is supposed to know it, while the speaker does not know it. This circumstance arises if the antecedent of the definite description was introduced by the hearer through the use of a specific indefinite. We assume that the incomplete definite description in (14) is of the same semantic type as its antecedent. The issue of rigidity (hence of semantic type) arises in the case of specific indefinites, much in the same way that it arises in the case of referential definite descriptions.3 Kaplan (1978) analyzes referential definite description as rigid. Yeom (1998) suggests that a specific indefinite functions as a rigid designator in the belief (=information state) of the agent who ‘has an individual in mind’. Thus, both referential definite descriptions and specific indefinites are characterized by direct reference; the basic distinction between them is the cognitive asymmetry presented by specific indefinites, whose interpretation is dependent on the information states of the participants in the conversation. The examples in (14) fully parallel those in (12) with respect to the restriction on non-D-linked care and weakly D-linked care. The generalization that emerges is that if the postcopular DP is rigid, care can only be interpreted as strongly D-linked. 3.2.2.1 Uniqueness. Definite descriptions that are not referential (but are not used attributively either) are not exempt from the uniqueness requirement, which is the core characteristic of definites (see Abbott, 1999, 2004; Birner and Ward, 1994). To be able to occur after the copula in questions introduced by weakly D-inked or nonD-linked care, the uniqueness of a singular definite noun phrase must be guaranteed by linguistic, context-independent means. Let us see what they are.4 3

The terms ‘referential’ and ‘attributive’ are used in Donnellan’s (1966) sense: a definite description is referential iff the speaker ‘uses the description to enable his audience to pick out whom or what he is talking about’; a definite description is attributive iff the speaker ‘states something about whoever or whatever is the so-and-so’. According to Donnellan (1966), the referential/attributive distinction is a semantic one, a view that has been a matter of controversy (see, for instance, Kaplan’s (1978) implementation of the view that the distinction is semantic and Kripke (1977), Neale (1990), and Dekker (1998) for pragmatic approaches to the referential/attributive distinction). 4 There is a partial overlap between definites that can occur in post-copular position in carequestions and Löbner’s (1985) ‘semantic definites’.


47

First, uniqueness can be enforced by a modifier of the noun in the postcopular DP. Among the modifiers able to enforce uniqueness are: (i) ordinal numerals (e.g. ‘third’); (ii) adjectives such as ‘next’, ‘only’; see the example (11a) above, which contains the adjective singur(ul) (‘(the) only’) and example (15) below which contains the adjective urm˘ator (‘next’); (iii) superlatives, as in the example (11b) above. (15)

Care e trenul urm˘ator? which is train-the next ‘Which is the next train?’

The modifiers in classes (i)–(iii) occur only with nouns that denote a presupposed, given set. For instance, question (15) presupposes the existence of a sequence of trains, of which the answer must pick the unique next train. Question (11b) presupposes the existence of a collection of Romanian composers, of which the answer must pick the unique individual who is the most important of the Romanian composers. In these questions, the modifier indicates that a unique individual is picked from the presupposed set; the DP corresponding to care in the answer must identify this individual. In the copular care-questions whose postcopular DP contains one of these modifiers, care is therefore weakly discourse-linked. Second, uniqueness can be enforced by a functional N contained in the postcopular DP, as in (10a, b), repeated below for convenience: (10)

a.

b.

Care e capitala Moldovei? what is capital-the of-Moldavia ‘What is the capital of Moldavia?’ Care e adresa lui Ion? what is address-the of John ‘What is John’s address?’

As we have already pointed out, the occurrences of care in (10a, b) are non-Dlinked. We will include in this class also questions such as (16) below, whose postcopular DP contains a measure N. Measure Ns such as temperature, weight, price are functional. For instance, the N temperature in (16) takes as an argument an explicit or implicit spatio-temporal location and yields as a value the temperature at that location: (16)

Care e temperatura (la voi)? CARE is temperature-the (at you) ‘What is the temperature (where you are now)?’

If the postcopular DP of a question of the form Care copula DP? contains a [+human] functional N, care can only be strongly D-linked, as shown in (17) below.

48

I. Comorovski

Therefore these questions do not fall in the class of care-questions that we are investigating. (17)

Care e sot, ia lui Ion? which is wife-the of John ‘Which one is John’s wife?’

Care is strongly D-linked, e.g. the question is asked at a party.

An interesting case is that of copular care-questions with propositional answers. The postcopular DPs in these questions do not require an explicit indication of uniqueness. Why should this be so? Comorovski (2007) suggests that this fact follows from the way propositions are counted. A complex proposition (e.g. a conjunction of propositions) can be regarded as a single proposition, so the issue of uniqueness does not arise in a clear way. Take, for instance, the question in (18) below. Answer A1 expresses one simple proposition. Answer A2 expresses one complex proposition. In both cases, the singular definite postcopular DP in the question is associated in the answer with a value that is a single (abstract) entity: (18)

discut¸iei Care a fost concluzia de ieri? what has been conclusion-the discussion-theGEN of yesterday ‘What was the conclusion of yesterday’s discussion?’ A1: That everybody should pay a small entrance fee. A2: That everybody should pay a small entrance fee and that several roles in the play should be given to student actors.

Care as it occurs in (18) is non-D-linked. 3.2.3 Indirect Contextual Anchoring An examination of the examples in (10), (11), (15), (16), and (18) shows that, although the postcopular DP in a care-question cannot be directly referring, it is always linked to the context of utterance by a referential expression it contains or by an implicit argument of the N. In Section 2, we have used for this link the term indirect contextual anchoring. The expression contained in the postcopular DP can be a referential ‘possessor’ DP, as in (10a, b), a spatio-temporal location, as in (16) and (18), or a (modified) noun that refers to a given set, as naist român (‘Romanian pan-piper’) and compozitor român (‘Romanian composer’) in (11a, b), and tren (‘train’) in (15). The range of referential elements which link the denotation of the postcopular DP to the context of utterance shows that this context must be conceived of in a broad way. The relevant context can be larger than the immediate physical context. This fact is made clear by questions that ask for a list of kinds: (19)

Q: Care sunt mamiferele acvatice? which are mammals-the aquatic ‘Which are the water mammals?’ A: (Mamiferele acvatice sunt) balena si delfinul. ‘(The water mammals are) the whale and the dolphin.’


49

The postcopular DP in the question in (19) does not contain any referential expression. Since the answer that is requested is a list of kind-denoting terms, the context to which the DP is implicitly anchored is the earth: the question asks for the species of water mammals on earth. Thus, specificational questions that request a list of kind-denoting terms, and declarative specificational sentences that have in postcopular position a list of kind-denoting terms (cf. the answer in (19)) represent a limit case as to what the anchoring context is: it is the context in which the relevant kinds can in principle be instantiated.

3.2.4 Restriction on Quantifying Noun Phrases The second restriction on the postcopular DP in questions of the form Care copula DP? concerns quantifying noun phrase. Example (20) below illustrates the fact that quantifying DPs are barred from this position: (20)

∗

Care a fost fiecare r˘aspuns? CARE was every answer

Before discussing the distribution of indefinite postcopular DPs, we have to establish which of the two DPs that flank the copula is the subject of questions of the form Care copula DP?.

3.3 Remarks on the Syntax of Romanian Questions of the Form ‘Care Copula DP?’ One factor that helps identify the subject of a question of the form Care copula DP? is the neutral word order in the answer. This order is illustrated below for answers to questions in which care is non-D-linked or weakly D-linked: Non-D-linked care: (21)

Q. Care e capitala Moldovei? ‘What is the capital of Moldavia?’ A. Capitala Moldovei e Chis, in˘au. ‘The capital of Moldavia is Chisinau.’

Weakly D-linked care: (22)

Q. Care e cel mai important compozitor român? ‘Which is the most important Romanian composer?’ A. Cel mai important compozitor român e George Enescu. ‘The most important Romanian composer is George Enescu.’

As seen in (21) and (22), the sentence-initial constituent in the answers corresponds to the postcopular DP in the questions. Since the neutral order in Romanian

50

I. Comorovski

declaratives is SVO, it follows that capitala Moldovei (‘the capital of Moldavia’) and cel mai important compozitor român (‘the most important Romanian composer’) are the subjects of the answer sentences in (21) and (22) respectively. Given that the syntax of an answer to a constituent question mirrors the syntax of the corresponding question before wh-fronting has applied, we conclude that the subjects of the questions in (21) and (22) are the postcopular noun phrases. The strong D-linking of care has a bearing on the syntax of questions of the form Care copula DP?. Significantly, if care is strongly D-linked, the postcopular noun phrase can be a bare nominal, as illustrated in (23): (23)

a.

b.

Care (dintre ei) e aviator? which (of them) is pilot ‘Which one of them is a pilot?’ Care (dintre ei) e (un) bun coleg? which (of them) is (a) good colleague ‘Which one (of them) is a good colleague?’

The occurrence of postcopular bare nominals in questions introduced by strongly D-linked care indicates that the subject of these questions is care, and not the postcopular noun phrase, since in Romanian bare nominals cannot function as subjects. On the other hand, they can function as predicate nominals. We will discuss these facts below. The nouns that can occur without a determiner after the copula in care-questions introduced by strongly-D-linked care fall into two classes: (i) Ns that indicate a profession (e.g. aviator (‘pilot’), pescar (‘fisherman’)); (ii) modified [+human] relational Ns (e.g. coleg (‘colleague’), prieten (‘friend’)). Note that these are precisely the classes of Ns that can occur as bare predicate nominals in Romanian declaratives, as illustrated below: (24)

a.

b.

Ion e pescar. John is fisherman ‘John is a fishermant.’ Ion e (un) prieten de n˘adejde. John is (a) friend reliable ‘John is a reliable friend.’

Bare nominals can occur sentence-initially in Romanian, but they do not function as subjects of the sentences they introduce. Bare nominals occur sentence-initially as a result of fronting. The construction resulting from the fronting of a bare predicate nominal is illustrated in (25a); this construction is similar to the one which results from the fronting of a predicative adjective, illustrated in (25b): (25)

a.

Dirijor e Ion. conductor is John ‘It is John who is a conductor.’


b.

51

Talentat e Ion. talented is John ‘It is John who is talented.’

The sentences in (25) sound odd out of context, which is an indication of the fact that the word order they display is not neutral. Moreover, their intonation is different from the intonation of simple declaratives with neutral word order. The corresponding sentences with neutral word order are given in (26): (26)

a.

b.

Ion e dirijor. John is conductor ‘John is a conductor.’ Ion e talentat. John is talented

On the assumption that bare nominals are property-denoting and thereby share the semantic type of predicative adjectives, (25a, b) are sentences in which a propertydenoting complement of the copula has been fronted sentence-initially. The subject of these sentences appears sentence-finally. In copular questions introduced by non-D-linked or weakly-D-linked care, bare nominals cannot appear in the postcopular position. This is a second argument for an analysis of the noun phrase occurring in the postcopular position of these questions as the subject of the question and not as the predicate nominal. In sum, in Romanian questions of the form Care copula DP? in which care is non-D-linked or weakly D-linked, the subject is the postcopular DP.

3.4 Conclusion We have seen that direct reference is ruled out as a property of the subjects of questions of the form Care copula DP? if the question is introduced by non-Dlinked or weakly D-linked care. The subject of a question of this type cannot be a rigid designator. However, the subject is linked to the context of utterance by a referential expression or implicit argument it contains, i.e. it is ‘indirectly contextually anchored’. The semantic-pragmatic restrictions on the subject of the copular care-questions under consideration are the same as those on the subject of English declarative specificational clauses. We conclude that questions of the form Care copula DP? introduced by non-D-linked or weakly D-linked care are an interrogative form of specificational clauses. We take the fact that non-D-linked care occurs only in such questions as an indication of a selection relation: non-D-linked care is selected exclusively by the copula that occurs in specificational questions; in our terms, this is the copula of specification. The existence of this selection relation speaks in favour of a lexical treatment of the specificational reading of copular clauses.

52

I. Comorovski

4 Indefinite Subjects, Topichood, and Point of View 4.1 Indefinite Subjects and Topichood in Constituent Questions In Section 2, we have seen that Higgins (1973) was unsure about whether indefinites can or not function as specificational subjects. Mikkelsen (2004) clearly shows that they can: (27)

a.

b.

A philosopher who seems to share the Kiparskys’ intuitions on some factive predicates is Unger (1972). (Delacruz, 1976:195,4 quoted in Mikkelsen, 2004:230) Another speaker at the conference was the Times columnist Nicholas Kristof [. . .] (The New Yorker, Oct. 27, 2003, p. 86, quoted in Mikkelsen, 2004:173)

The indefinites in the sentences in (27) meet the indirect contextual anchoring condition. The indirect contextual anchoring of the subject of (27a) is ensured by the definite noun phrase the Kiparskys; the subject of (27b) is indirectly contextually anchored by the definite description the conference. The subjects of specificational clauses are also their topic, a fact on which researchers agree. Thus specificational clauses have an ordinary information structure, as the subject of a sentence is usually its topic. It is generally considered in the literature that indefinites, unless specific, cannot be sentence topics. However, Mikkelsen (2004:Ch. 8) convincingly argues that this generalization is not finegrained enough. Mikkelsen shows that if part of the information carried by an indefinite is discourse-old, the indefinite can function as the topic of a specificational sentence. In line with Mikkelsen’s remark, we suggest that the topic status of indefinite specificational subjects is facilitated by their indirect contextual anchoring. Indefinites are the one type of noun phrases that can function as subjects of declarative specificational clauses, but whose occurrence as subjects of specificational care-questions is heavily restricted. Let us look into this declarative/ interrogative asymmetry. Drijkoningen and Kampers-Manhe (2001) advance the following condition on the acceptability of constituent-questions: ‘do not pose questions about entities that you introduce in the discourse while posing the question’. The goal of Drijkoningen and Kampers-Manhe’s condition is to block the occurrence of indefinite subjects in French constituent questions. The fact that it is subjects that are at stake plausibly points to the relevance of information structure: the subject often plays a privileged role with respect to the information structure of a sentence, functioning as the sentence topic. The notion of topichood indirectly creeps into Drijkoningen and Kampers-Manhe’s principle via their use of the preposition ‘about’: 4

Delacruz, E.B. (1976): ‘Factives and Propositional Level Constructions in Montague Grammar’, in B.H. Partee, ed., Montague Grammar, Academic Press, New York.


53

the questions that their principle rules out are questions about new entities. The use of ‘about’ is relevant because aboutness has often played a central role in the definition of topichood (e.g. Portner and Yabushita, 1998). Since indefinites introduce new discourse referents and subjects tend to be topics, Drijkoningen and Kampers-Manhe’s principle effectively rules out constituent questions with indefinite subjects. Notice that if the linear order of elements in a constituent question confers topic status to a non-subject, the subject of the question can be indefinite. This is the case in the question below, in which the direct object is sentence-initial, preceding the interrogative pronoun, and thus becoming the topic of the question: (28)

Noua mea nuvel˘a, oare cˆınd o va citi cineva? new-the my short-story Q-prt. when it will read somebody ‘My new short story, when will somebody read it?’

As we have seen, indefinite subjects that are indirectly contextually anchored can function as sentence topics. But since such a subject introduces a new discourse referent (an individual concept), a constituent question cannot ask information ‘about’ it. Thus, even though an indefinite that is indirectly contextually anchored can function as the topic of a declarative sentence, it cannot function as the topic of a constituent question. As there is no plausible candidate for topichood in copular care-questions with indefinite subjects, and as any question must be about something, care-questions with indefinite subjects are ruled out on the ground of having no topic. Below we will show that the occurrence of indefinite subjects in Romanian carequestions is possible if the question contains a subjective predicate. The data in (28) have led us to explain the possibility of having an indefinite subject in a constituent question by the existence of a non-subject topic in the question. Following this line of thought, we will relate the role of subjective predicates in licensing indefinite subjects of copular care-questions to the requirement that constituent questions have a topic.

4.2 Topichood, Point of View, and the Conditional Mood Specificational care-questions allow indefinite subjects provided: (a) the question contains a subjective predicate, such as a predicate expressing a moral judgment or an aesthetic judgment, and (b) the verb of the question is in the conditional mood. Here are some examples of specificational care-questions with indefinite subjects: (29)

Care ar fi (dup˘a p˘arerea voastr˘a) un loc bun de mers la var˘a? ‘What would be (in your opinion) a good place to go to this summer?’

54

(30)

I. Comorovski

Care ar fi dup˘a ei un rezultat satisf˘ac˘ator? ‘What would be according to them a satisfactory result?’

Questions containing a subjective predicate ask for the point of view of the hearer or of a third party on what the true answer to the question is. The reference to a point of view can be explicitly made by the use of an expression such as dup˘a tine / el (‘according to you/him’), dup˘a p˘arerea voastr˘a (‘in your opinion’), or din punctul t˘au / lor de vedere (‘from your/their point of view’). In the absence of such an expression, the holder of the point of view is understood to be the hearer, as is the case in (29) if the expression in brackets is omitted. Following Lasersohn (2005), we will call the holder of the point of view the ‘judge’. Any subjective predicate has a ‘judge’ argument. We suggest that the topic of the questions in (29–30) is the point of view of the ‘judge’. This hypothesis about the information structure of a specificational care-question seems to us comparable to a proposal that has been put forth by Jäger (2001). Jäger analyzes the information structure of certain statements that were considered in the functionalist literature as topicless or ‘all new information’ statements, e.g. ‘Somebody is knocking at the front door’. These statements contain a non-stative predicate. According to Jäger, the topic of these statements is the event argument of the predicate; the event argument can be left implicit. Note that, just like non-stative predicates have an event argument, subjective predicates have a ‘judge’ argument that can remain implicit. Why is the presence of a subjective predicate obligatory in specificational carequestions with an indefinite subject? As we have seen in the previous subsection, constituent questions are asked ‘about’ a discourse-old entity. Such an entity cannot be denoted by an indefinite subject. But it can be denoted by a ‘judge’ argument, whether overt or implicit. The question can then inquire ‘about’ the point of view of the ‘judge’. Let us turn now to the role of the conditional mood in specificational carequestions with an indefinite subject. We will argue that the occurrence of the conditional morphology is related to the presence of a subjective predicate in the question. To see how this relation comes about, we will first look into the interpretation of the Romanian conditional as an evidential marker. Among the ‘canonical’ interpretations of the conditional mood, Gramatica limbii române (2005) (‘The Grammar of the Romanian Language’, 2005)5 lists a reading which the conditional can have in both independent and embedded clauses, namely that of an evidential marker. The conditional is an indirect evidential: it indicates that the source of the information presented by the speaker is other than his perceptions; the speaker presents his information as reported information and does not commit himself to the truth of what he reports. Thus the Romanian conditional is a reportative evidential.

5

Vol. I, p. 376, vol. II, pp. 679, 688. Editura Academiei Române, Bucures, ti.


55

Here are some examples of sentences in which the conditional mood is used as a reportative evidential. The source of the reported information is given in brackets and can be omitted if it is recoverable from the context: (31)

a.

b.

(Cic˘a) polit, ia ar fi arestat trei mafiot, i. (evid.prt.) police-the aux.cond be.perf. arrested three mafia-men (They say/The rumour goes that) the police have arrested three mafia men. (Potrivit unei surse diplomatice,) pres, edint, ii celor dou˘a (according-to a source diplomatic presidents-the of-the two t, a˘ ri countries According to a diplomatic source,) the presidents of the two countries held talks last week.

Our goal is to relate the interpretation of the conditional mood as a reportative evidential to the obligatory occurrence of the conditional in specificational care-questions with indefinite subjects. As we have seen, these questions contain a subjective predicate, and subjective predicates have a ‘judge’ argument. The utterer of the question requests the ‘judge’ to give what he considers to be the true answer to the question. We can think of the ‘judge’ as of a source of information whose truth is not taken for granted by the utterer of the question. The distancing of the utterer of the question from the truth of the answer is marked in the question by the conditional mood used as a reportative evidential. Indirect evidentials are analyzed by Izvorski (1997) as epistemic modal operators with a presupposition of available indirect evidence for the truth of the proposition the evidential scopes over. In the case of questions, an indirect evidential, such as the reportative conditional, will presuppose the existence of an indirect source of evidence for the information to be given in the answer. We suggest that in specificational care-questions, the indirect source presupposed by the use of the reportative evidential is the ‘judge’. Thus the presupposition associated with the reportative evidential is satisfied by the argument structure of the subjective predicate occurring in the question. The proposed analysis of specificational care-questions with indefinite subjects opens the issue of an observed difference between these questions and other sentences that contain a subjective predicate: the conditional is obligatory in the former, while it is not in the latter. Why should there be such a difference? We will tentatively suggest that the answer is related to the issue of topichood. The ‘judge’ argument plays a crucial role in the acceptability of constituent questions with an indefinite subject, as it provides them with a topic, namely the judge’s point of view. The ’judge’ also acts as the indirect source of evidence presupposed by the conditional of evidentiality. Thus, the occurence of the conditional of evidentiality highlights the presence of the judge argument. No such highlighting is needed in the case of other types of sentences that contain a subjective predicate, as these sentences do not need as a topic the judge’s point of view.

56

I. Comorovski

5 Conclusions This paper has brought evidence in favour of a certain semantic-pragmatic characterization of specificational subjects. The evidence is based on Romanian copular questions of the form Care-copula-DP? in which care is non-D-linked or weakly D-linked. We have argued that these questions are interrogative specificational sentences. It has been shown that the subject of these questions is of the type of individual concepts (<s,e>) and that it must be indirectly contextually anchored. We have seen that indefinites cannot be subjects of constituent questions unless the question has a non-subject topic. Interestingly, in case a specificational carequestion contains a subjective predicate and has the copula in the conditional mood, the subject of the question can be indefinite. To explain this fact, we have suggested that the point of view argument of a subjective predicate can function as the topic of a sentence. In particular, it functions as the topic of specificational care-questions with indefinite subjects.

References Abbott, B. (1999): ‘Support for a Unique Theory of Definite Descriptions’, Proceedings from Semantics and Linguistic Theory IX, 1–15, CLC Publications, Ithaca, NY. Abbott, B. (2004): ‘Definites and Indefinites’, in Handbook of Pragmatics, L. Horn & G. Ward (eds.), Blackwell, Oxford. Birner, B. & G. Ward (1994): ‘Uniqueness, Familiarity, and the Definite Article in English’, Proceedings of the Berkeley Linguistics Society 20, 93–102. Comorovski, I. (1989): Discourse and the Syntax of Multiple Constituent Questions, Ph.D. dissertation, Cornell University, Ithaca, NY. Comorovski, I. (1996): Interrogative Phrases and the Syntax-Semantics Interface, Kluwer, Dordrecht. Comorovski, I. (2005): ‘On certain copular constituent questions in Romanian’, in M. Coene & L. Tasmowski, eds., On Space and Time in Language, 353–377, Clusium, Cluj. Comorovski, I. (2007): ‘Constituent Questions and the Copula of Specification’, in I. Comorovski & K. von Heusinger, eds., Existence: Semantics and Syntax, 49–77, Springer, Dordrecht. Dekker, P. (1998): ‘Speaker’s Reference, Descriptions, and Information Structure’, Journal of Semantics, 15, 305–334. Dobrovie-Sorin, C. (1994): The Syntax of Romanian, Mouton de Gruyter, Berlin. Donnellan, K. (1966): ‘Reference and Definite Descriptions’, Philosophical Review, 75, 281–304. Drijkoningen, F. & B. Kampers-Manhe (2001): On the Interpretation of Postverbal Subject Positions, vol. 21, Utrecht Studies in Romance Linguistics.


57

Higgins, F. R. (1973): The Pseudo-Cleft Construction in English, Ph.D. dissertation, MIT. Izvorski, R. (1997): ‘The Present Perfect as an Epistemic Modal’, in A. Lawson & E. Cho (eds.), Proceedings of SALT VII, CLC Publications, Cornell University, Ithaca, NY. Jäger, G. (2001): ‘Topic-Comment Structure and the Contrast between Stage Level and Individual Level Predicates’, Journal of Semantics, 18, 83–126. Kaplan, D. (1977): ‘On the Logic of Demonstratives’, Journal of Philosophical Logic, 8, 81–98. Kaplan, D. (1978): ‘Dhat’, Syntax and Semantics, P. Cole (ed.), 9, 221–243, Academic, New York. Kaplan, D. (1989): ‘Demonstratives’, in Themes from Kaplan, J. Almong, J. Perry, & H. Wettstein (eds.), 421–563, Oxford University Press. Kripke, S. (1977): Speaker’s Reference and Semantic Reference. In Midwest Studies in Philosophy, vol. II: Studies in the Philosophy of Language, P. A. French, T. E. Uehling, Jr. & H. K. Wettstein (eds.), 255–276, Morris: University of Minnesota, MN. Lasersohn, P. (2005): ‘Context dependence, disagreement, and predicates of personal taste’, Linguistics and Philosophy, 28, 643–686. Löbner, S. (1985): ‘Definites’, Journal of Semantics, 4, 279–326. Mikkelsen, L. (2004): Specifying Who: On the Structure, Meaning, and Use of Specificational Copular Clauses, Ph.D. Dissertation, University of California, Santa Cruz, CA. Neale, S. (1990): Descriptions, MIT Press, Cambridge, MA. Pesetsky, D. (1987): ‘Wh-in situ: Movement and Unselective Binding’, in The Representation of (In)definiteness, E. Reuland & A. Ter Meulen (eds.), MIT Press, Cambridge, MA. Portner, P. & K. Yabushita (1998): ‘The Semantics and Pragmatics of Topic Phrases’, Linguistics and Philosophy, 2, 117–157. Romero, M. (2005): ‘Concealed Questions and Specificational Subjects’, Linguistics and Philosophy, 28, 687–737. Strawson, P. (1950): ‘On Referring’, Mind, 59, 320–344. Yeom, J.-I. (1998): A Presuppositional Analysis of Specific Indefinities: Common Grounds as Structured Information States, Garland, New York.

Temporal Orientation in Conditionals Bridget Copley

Abstract This paper argues for a modal explanation for temporal orientation facts in both antecedents and consequents of conditionals. Future-oriented statives are shown to get their future orientation by means of a different mechanism from futureoriented eventives. Thus, eventuality type and temporal orientation turn out to be correlated more closely than previously thought, and the “present eventive constraint” is not useful in accounting for the temporal orientation facts. Thus we must look for a new kind of explanation. I argue that temporal orientation, and therefore also eventuality type, are correlated with modal flavor, so that the most promising way to explain the temporal orientation facts will be through appealing to the modal facts. Two apparent objections to this kind of account are removed: the proposed existence of epistemic eventives (I argue that these examples are actually derived statives), and the assumption that antecedents and consequents share the same modal flavor (they do not, despite traditional assumptions). Key words: Conditionals, temporal orientation, eventuality type, Aktionsart, epistemic modality, metaphysical modality.

The goal of this paper is to argue for a certain way of explaining temporal orientation in conditionals. We will limit ourselves to indicative English will conditionals, such as the one in (1). (1)

If Zoe gets cranky, Tasha will get cranky.

By “temporal orientation,” I mean the temporal location of the antecedent eventuality (here, the Zoe-get-cranky event) with respect to the utterance time, and similarly

I would like to thank the Jacquelines Lecarme and Guéron, Jean-Daniel Mohier, various English speakers, and the participants at the roundtable. Structures Formelles du Langage, CNRS/Université Paris 8


59

60

B. Copley

the temporal location of the consequent eventuality (here, the Tasha-get-cranky event) with respect to the utterance time. We will also be interested in how these eventualities are situated with respect to each other. We can articulate this goal a little more precisely. In the above example, let’s agree that Zoe get cranky is the “antecedent” (rather than if Zoe gets cranky), and that Tasha get cranky is the “consequent” (rather than Tasha will get cranky; the reasons for scoping out will will become clear later). Now, let us assume that the antecedent and consequent each take a temporal argument. Let t∗ ant be the temporal argument of the antecedent – it will normally be equal to tu , the utterance time – and let t∗ cons be the temporal argument of the consequent. Finally, we define t-sitant as the run time of the eventuality in the antecedent (cf. Klein’s (1997) time of situation or Reichenbach’s (1947) E), and similarly define t-sitcons . Then the goal of this paper is to explain how to relate t-sitant to t∗ ant and t-sitcons to t∗ cons , and furthermore, to determine how the antecedent is temporally related to the consequent. We must be coy about this latter point until we know more, since at the outset, it is not immediately clear which antecedent time might be related to which consequent time, though it is clear enough that the antecedent and consequent are somehow temporally related. In the end, we will not have much to say about this issue, though we will need to say something in order to determine what t∗ cons is.1 We will see that eventuality type – whether something is eventive or stative – correlates with temporal orientation – whether a t-sit is future-oriented or presentoriented with respect to its t∗ (i.e., the temporal argument of its clause). This result will render inadequate the “present eventive constraint,” the currently accepted way to explain temporal orientation in modal contexts. We will instead justify (though not provide) a different kind of explanation, based on a correlation with modal flavor. It turns out that metaphysical modal flavor correlates with eventivity, and epistemic modal flavor correlates with stativity. This result also makes the prediction that antecedents have a modal flavor independently of their consequent, which is true, though strangely, such judgments have been overlooked in earlier literature.

1 UFOs and Other Oddities In this section we will see what the apparent temporal relations are between each t∗ and its corresponding t-sit, and also give a description of how the antecedent and consequent are temporally related. We begin with antecedents. Our starting point is a phenomenon that I will be calling “unexpected future orientation,” or “UFO” for short. The antecedent in (2a) below is considered to have a UFO because it has future orientation of t-sitant with respect to tu (and since we presume t∗ ant to be equated with tu , there is also future orientation of t-sitant with respect to t∗ ant ) – but there is no obvious morphology to Note that t∗ cons is not the same as Klein’s topic time or Reichenbach’s R (reference time). Past temporal morphology in the antecedent or consequent can, for instance, dissociate topic time/R from t∗ cons ; the latter will still be equal to the utterance time.

1

Temporal Orientation in Conditionals

61

mark the futurity. Indeed it is impossible2 to assert the material in the antecedent by saying (2b). (2)

a. b.

If it rains tomorrow, Max will get sick. #It rains tomorrow.

This contrast does not appear in languages universally. The morphological equivalent of (2b) in many languages is perfectly acceptable; (3), for instance, is acceptable in German. (3)

Morgen regnet es. tomorrow rains it ‘It’s going to rain tomorrow.’

Thus in a language like German, one would not necessarily be surprised by future orientation in antecedents using the same morphology. But in English, (2a) is good and (2b) is bad, and that fact requires explanation.3 The UFO we have just seen in (3a) has an eventive predicate. Eventive predicates in antecedents must be future-oriented (FO), and cannot be present-oriented (PO), as shown in (4): (4)

If it rains, Max will get sick. = ‘If it’s raining right now, Max will get sick.’

∗ PO

reading

The situation is a little different when the antecedent has a stative verb phrase as in (5). (5)

a. b.

If John is sick tomorrow, Celeste will get sick. If John is sick right now, Celeste will get sick.

ant. FO ant. PO

The example in (5a) shows that statives in the antecedent can have UFOs; the example in (5b) shows that they need not. This behavior constrasts with that of eventives, which as we have seen necessarily have UFOs. This pattern of judgments is familiar from the behavior of complements of (certain) modals (see Condoravdi (2001), Werner (2002), Werner (2006)), as in (6). 2

It is true that a future event time is possible with a simple present verb in English, if the verb phrase describes an event that is scheduled ahead of time or otherwise predetermined. (i)

Zoe plays chess with Xander tomorrow.

The status of these “futurate” readings will become relevant once we have a theory of the type proposed in this paper. For now, we will not attempt an explanation of these readings (but for the classic treatment see Dowty (1979), and more recently Cipria and Roberts (2000), and Copley (2008)). We will, however, make reference to futurates again below, in the discussion about “settled eventives”. 3 Actually, it is something of a matter of opinion which of (44a) and (44b) requires the explanation. If, based on the facts in other languages, one expects verbs with present (null) morphology to be able to refer to the future, one could instead speak of (44b) as having an “unexpected lack of future orientation.”

62

B. Copley

Similarly to the antecedent cases, eventives under modals can only be FO, while statives under modals can be either PO or FO.4 (6)

a. b. c.

Morgan may go. (eventive, only FO; = ‘may be going’) Morgan may be here right now. (stative, PO) Morgan may be here tomorrow. (stative, FO)

It is not so surprising that these environments should behave similarly to antecedents of conditionals, considering that these are both modal contexts as well. (This common element of modality should also make us wonder what the temporal facts have to do with the modal facts, a question that we will take up again later.) Consequents of conditionals, which of course are in a modal context as much as antecedents and complements of modals are, behave similarly when it comes to temporal orientation. To ascertain the temporal orientation of t-sitcons with respect to t∗ cons , we will need to know what t∗ cons is, i.e., what the temporal argument of the consequent is. I have mentioned above that the question of how to calculate t∗ cons from a time in the antecedent is a tricky one. The answer is different depending on the eventuality type of the antecedent. To begin with, we will hold the eventuality type of the antecedent constant, using only eventive antecedents. Consider the sentence with in (7), for example. The antecedent and consequent are both eventive. What can be said about the temporal relationship between the antecedent and the consequent? (7)

If it rains, Max will get sick.

It appears that the start of the consequent event must begin at least a little bit after the start of the event of the antecedent.5 In uttering (7), one conveys that Max gets sick at least a little bit after the rain begins. Indeed, the inference that the rain causes Max to get sick is (at least) difficult to escape, and may well be inescapable. The need for a causal interpretation would definitely explain the temporal observation; for the contextually salient causal mechanism, the rain should indeed start before the illness starts, in order for the rain to cause the illness. For our purposes at the moment, it is enough to note the temporal fact: there must be at least a little rain before the illness begins.6 4 There is another wrinkle here, that of modal flavor, e.g. the examples in (45) have differing possibilities for epistemic, deontic, etc. readings. These will be considered below; for now, we are just interested in the temporal possibilities. 5 There are cases where the same event is described in both antecedent and consequent, as in (i):

(i)

If Marissa leaves, she will leave quickly.

In such cases, the above generalization does not apply. It is not clear to me when reference to the same event is permitted. 6 Suppose, to the contrary, that one wished to convey that the onset of Max’s illness coincides exactly with the onset of the rain. One could say, for example, (ia) or (ib), using when, to convey this.


63

When the consequent is stative, it can be either FO or PO with respect to the start of the antecedent event. Consider the examples in (8), which have a stative in the consequent. (8)

a. b.

If John gets mad, Celeste will be mad (afterwards). If John gets mad, Celeste will be mad (then).

cons. FO cons. PO

In (8a), where the consequent is future-oriented, John gets mad and then Celeste is mad at some later point. In the reading where the consequent is present-oriented, John gets mad during the time that Celeste is already mad. (These readings have different modal “flavors,” which will be considered later.) Thus statives in consequents can be either FO or PO with respect to the relevant time in the antecedent. This pattern is enough to cause us to posit that t∗ cons in (7) and (8) begins at the same time as the start of t-sitant . With that assumption, we get the expected relation between t∗ cons and t-sitcons : when the consequent is eventive, t-sitcons must be futureoriented with respect to t∗ cons (i.e., it must start at least a little after the beginning of t∗ cons ), and when the consequent is stative, t-sitcons can be either present-oriented or future-oriented with respect to t∗ cons . The situation is similar when the antecedent is stative and PO, as in (9). (9)

a. b. c.

If Xander is cranky now, Zoe will get cranky. eventive cons., FO If Xander is cranky now, Zoe will be cranky (now). stative cons., PO If Xander is cranky now, Zoe will be cranky (the next day). stative cons., FO

The familiar pattern is there: eventive consequents must be future-oriented, and stative consequents can be either present-oriented or future-oriented. The data in (9) are also compatible with the idea that t∗ cons is constrained to start when t-sitant starts. But in fact there would be another possible hypothesis if we were just to look at the data in (9) on its own, since in stative PO antecedents, t-sitant has to overlap t∗ ant . So the data in (9) are also compatible with a hypothesis that t∗ ant (not t-sitant ) is what constrains t∗ cons . This alternative hypothesis turns out to be appropriate when the antecedent has additional tense-aspect morphology (that is, it is what Condoravdi calls a “temporal” predicate), as in (10). (10)

If Xander was cranky yesterday, Zoe will be cranky now/#then/the day after tomorrow.

The time in the antecedent that constrains the temporal location of t∗ cons cannot be t-sitant . If it were, we would expect Zoe’s crankiness to be either overlapping t-sitant or later than it. But it cannot do this (as shown by the non-acceptability of (i)

a. b.

It will start raining exactly when Max gets sick. Exactly when it starts raining, Max will get sick.

But one could not truly say (46). For (46) to be true, Max’s illness has to start at least a bit after the start of the raining.

64

B. Copley

then); instead, it either overlaps or is later than tu – which we assume has the same value as t∗ ant . So the alternative hypothesis, namely that t∗ ant is the antecedent time that matters, works here, and the original hypothesis is not appropriate. On the other hand, the original hypothesis — that t-sitant is the relevant antecedent time — is appropriate when the antecedent is a future-oriented stative, again on the assumption that t∗ ant is tu : (11)

If Xander is cranky tomorrow, Zoe will be cranky #now/then/the next day.

Here, by parallel argumentation to (10), t∗ cons cannot have the same value as tu , but rather, has to be in the future. The two hypotheses about the relationship between t∗ cons and the antecedent are summarized in (12): (12)

a.

b.

Original hypothesis: t∗ cons begins when t-sitant begins. Accounts for eventive antecedents and future-oriented stative antecedents. Compatible with present-oriented stative antecedents. Alternative hypothesis: t∗ cons begins when t∗ ant begins. Accounts for temporal antecedents. Compatible with present-oriented stative antecedents.

It is important to remember that even with this complex picture, there is one clear result: Consequents behave just like antecedents (and like the complements of modals) in relating their t∗ to their t-sit: With eventive consequents, t-sitcons is futureoriented with respect to t∗ cons , and with stative consequents, t-sitcons can be either present-oriented or future-oriented with respect to t∗ cons .7 Incidentally, the existence of this pattern for both the antecedent and consequent of will conditionals suggests that will contributes no extra future or non-past tense meaning of any kind to the relationship between t∗ and t-sit in the consequent – at least, none that it does not also contribute to the antecedent. There are at least two ways we might account for this fact: will might take wide scope, or it might mark agreement with a null modal that takes wide scope. Either way is fine; I will assume the first. We now have the basic facts about the relation between t∗ and t-sit in both antecedents and consequents. We also know that these facts hinge on the eventuality type of the antecedent and consequent: eventives must be FO, while statives can be either FO or PO. Our next move is not to explain these facts, but first, to dismantle a popular type of explanation for these facts. After that, we will interest ourselves in the possibility of a different kind of explanation for temporal orientation within the antecedent and consequent, one that makes reference to modal flavor.

7

Temporal predicates, of course, further alter the relationship between t∗ and t-sit.


65

2 Away from an Explanation Given that we see the same relationship between t∗ and t-sit in antecedents, complements of modals, and consequents – i.e., in modal contexts – it is reasonable to think that it is all for the same reason. The most reasonable reason to propose, and one that has often been proposed for such situations, is what we will be calling the “present eventive constraint” (PEC). It says that both FO and PO readings are in principle possible, but that eventive FO readings are alone in being ruled out. However, we will see that the PEC is not in fact the right way to deal with these environments, since a key assumption motivating the PEC is not valid. This development will make us rethink it as an explanation for this pattern. We have just seen that t-sit in a modal context can apparently be either PO or FO with respect to the t∗ . Consider the matrix will sentences below in (13). It is clear that t-sit can be PO with respect to tu , as in (13a), or FO with respect to tu , as in (13b).8 (13)

a. b.

Zoe will be cranky (right now). Zoe will be cranky at 4 tomorrow.

“epistemic will;” PO FO

Note that (14), which lacks a temporal adverbial, sounds somewhat incomplete when it has an intended future orientation, or as if the temporal specification must be somehow understood from the context. (14)

Zoe will be cranky.

This need for a temporal adverbial is commonly referred to as “anchoring.” Anchoring can be accomplished in a number of ways: with a temporal adverbial as in (15a), with a when clause as in (15b), or with an if clause, as in (15c). (The if clause does something else modal in addition to providing a temporal anchor, of course, but the point here is that it does provide the temporal anchoring.) (15)

a. b. c.

At 4, Zoe will be cranky. When you see Zoe, she will be cranky. If you see Zoe, she will be cranky.

So anchoring must do what (we decided in the last section) the eventive antecedent does: It shifts forward the t∗ of Zoe be cranky. We can confirm that it is t∗ and not t-sit that is shifted forward by considering a temporal predicate, that is, one with temporal-aspectual morphology, as in (16): (16) 8

(At 4/When you see Zoe/If you see Zoe), she will have been cranky.

Some readers may see a modal, epistemic will (13a), and a non-modal, future tense will in (13b). Instead of treating (13b) as having a future tense, we are treating it as having a modal with metaphysical modal flavor. See Condoravdi (2003) for arguments motivating a unitary modal analysis of will. In any case, at the moment we are concerned with the difference in the temporal orientation of the eventualities in (13a) and (13b), though the difference in modal flavor will come to be relevant towards the end of the paper.

66

B. Copley

In the sentence in (16), what is shifted forward is the time that is the argument of Zoe have been cranky, not the time of the crankiness itself (which, according to the meaning of the future perfect, could take place before the anchor). This is additional evidence that the time being anchored is not t-sit, but is rather t∗ . So t-sit in (14) is FO with respect to tu exactly when t∗ is shifted forward by an anchor: It can be shifted forward contextually, or with a temporal adverbial, or with a when-clause, or with an if clause. But if these anchors are absent, t-sit must be PO with respect to tu . This need for anchoring has two important corollaries for us. The first is that t∗ is not always the same as tu ; it can be shifted forward into the future. The second is that t-sit is always PO with respect to t∗ . Because if it were allowed to be FO with respect to its t∗ , the sentence in (16) could have a FO reading in the absence of an anchor, as eventives do. To see the contrast, compare the sentence in (14) (repeated below as (17a)) with that in (17b): (17)

a. b.

Zoe will be cranky. Zoe will leave.

In (17a), there is a strong feeling that a contextually-specified time is needed (i.e., an anchor) if the sentence is to be read with future orientation of t-sit with respect to tu . In (17b), on the other hand, there is future orientation of t-sit with respect to tu even though there is no anchoring. This indicates that t-sit in (17b) really is FO with respect to t∗ , and t∗ is to be identified with tu . We have seen that it is possible to shift the t∗ of an eventive clause forward in at least one case. The t∗ cons is shifted forward by an eventive antecedent, as in (18) below. (18)

If it rains, Max will get cranky.

So there is one future shift for t∗ cons with respect to tu , accomplished by the antecedent, and a second one for t-sitcons with respect to t∗ cons , accomplished by the eventive itself. This second shift is exactly what statives cannot do. Interestingly, it’s quite difficult to shift the reference time for eventives with temporal adverbials. There seems only to be one shift into the future in (19); the getting cranky seems to have to be at 4. (19)

At 4, Max will get cranky.

So in (19), t∗ apparently shares the same value with tu , and at 4 characterizes t-sit, even though it is pronounced at the beginning of the sentence.9 For now, it is puzzling why this should be so; we will return to this case later. We have so far considered stative predicates in complements of will and in consequents of will conditionals. We have seen in these cases that when the t-sit is FO with respect to tu , it cannot be because t-sit is FO with respect to t∗ , but must be 9 Syntactically, what this would mean is that the temporal adverbial at 4 in (47) is interpreted in the vP, and cannot be interpreted up high.


67

because t∗ is FO with respect to tu ; t-sit is always PO with respect to t∗ . This behavior contrasts with that of eventives, in which t-sit is always FO with respect to t∗ (but as with statives, t∗ may or may not be PO with respect to tu .) We can represent these generalizations as in (20) and (21), borrowing the neo-Reichenbachian notation in which “t1 , t2 ” denotes present orientation of t2 with respect to t1 (or equivalently, of t1 with respect to t2 ), and “t1 t2 denotes future orientation of t2 with respect to t1 . (20)

Statives: a. tu , t∗ , t-sit b. tu t∗ , t-sit c. NOT POSSIBLE: tu , t∗

(21)

e.g., Zoe will be cranky now. e.g., Zoe will be cranky at 4. t-sit

Eventives: a. tu , t∗ t-sit e.g., Zoe will leave. e.g., If Max shows up, Zoe will leave b. tu t∗ t-sit (only possible with when or if anchors)

For further evidence for these generalizations, let us consider the behavior of antecedents. Stative antecedents behave as expected. Without an (overt or contextual) adverbial to anchor the stative, t-sitant is PO with respect to tu , but with an adverbial to fix a time in the future, t-sitant is FO with respect to tu . (22)

a. b. c.

If Zoe is cranky,. . . If Zoe is cranky at 4 tomorrow, . . . If, when you see her, Zoe is cranky, . . .

PO (FO with context) FO FO

As before, the need for the anchor to shift t∗ forward is evidence that t-sit is not allowed to shift forward on its own.10 For eventive antecedents, as before, t-sitant is normally FO with respect to tu as shown in (23a), and this does not change with a temporal adverbial, as in (23b). But with a when clause, we get the double future orientation: the onset of the crankiness in (23c) must be after the onset of your seeing Zoe. (23)

a. b. c.

If Zoe gets cranky, . . . If Zoe gets cranky at 4 tomorrow, . . . If, when you see her, Zoe gets cranky, . . .

This behavior is exactly what we expect. Under what conditions is t∗ allowed to shift forward? The examples in (i) seem to indicate that will (or presumably, another modal) must be in the consequent:

10

(i)

a. b.

If John is cranky tomorrow at 4, Mary will be happy. #If John is cranky tomorrow at 4, Mary is happy.

This is further evidence that will has to take wide scope in all the conditionals we have been looking at. It is also reminiscent of theories in which will is a non-past tense, since will here shifts t∗ (as we might expect of a tense) rather than t-sit (as we would expect of aspect). Of course some modal element is required too in order to get the conditional semantics; I assume that will contains the modal element, though nothing in particular hangs on that assumption here.

68

B. Copley

So both stative and eventive predicates can have a t∗ that is PO or FO with respect to tu . But eventives always have a t-sit that is FO with respect to t∗ , and statives always have a t-sit that is PO with respect to the reference time. But if this is always true, then we have no “real” FO statives – that is, no statives that can shift forward by themselves. They only shift forward because of an independent mechanism that allows a future shifting of t∗ , a process that is common to eventives as well. What do these facts mean for the present eventive constraint? We began by noting that it was possible to have future orientation of t-sit with respect to tu in antecedents (these were the “UFOs”), and more generally under modals and in consequents, without any obvious future-orienting morphology. Since it is obligatory for eventives, and optional for statives, the traditional response to this fact has been to generally rule in both present and future orientation of t-sit with respect to tu . Then it is just PO eventives that need to be ruled out. These were ruled out by some version of a “present eventive constraint”. But now we see that, if more attention is paid to t∗ – which can be either presentor future-oriented itself with respect to the tu – it is clear that the relationship between t-sit and tu should be broken down into two relationships, one between tu and t∗ , and one between t∗ and t-sit. Once we do that, we can conclude that eventives always have a t-sit that is FO with respect to t∗ , and statives always have a t-sit that is PO with respect to t∗ . But now it is unsatisfactory to have one constraint to rule out PO eventives and another constraint just to rule out FO statives; two constraints to rule out two of four possibilities is arguably too much theoretical machinery. Moreover, it is not even clear how we would justify a “future stative constraint.” The usual explanation for the PEC goes like this: Eventives can not be evaluated at the present moment because the present moment is homogenous, and cannot have an event take place during it.11 Can such an explanation be extended to account for a putative FSC, i.e., to explain why statives might not be evaluated except at their input time? At this point, it is hard to think of how it could be. Thus the PEC is really not worth pursuing further as an explanation for UFOs and other temporal orientation facts in conditionals and other modal contexts; it cannot explain the unexpected lack of future orientation for statives. But fortunately, there is something associated with present orientation and future orientation that we can mine for an explanation: namely, modal flavor. We now, therefore, turn our expectant hopes towards the possibility of a modal explanation for the temporal facts.

3 Towards an Explanation In (24a) below, a prediction is made that has the flavor of being about how the world turns out, whether we get to check and see it or not. We will refer to this flavor as a “metaphysical flavor;” Zoe’s actually getting cranky is FO with respect to the 11

At any rate, such an explanation needs special pleading to account for the PEC-like effects with future-shifted t∗ , since the future t∗ is not the present moment, but eventives still must have their t-sit shifted into the future with respect to the t∗ . At the very least, it must be made clear how a future t∗ might be treated as though it were tu .


69

utterance time. In (24b), however, the prediction has the flavor of being about what is true now (PO), and there is a suggestion, perhaps, that we are about to check whether that is so. If anything about the situation in (24b) is in the future, it is the finding out, rather than Zoe’s being cranky. We will refer to this flavor as an “epistemic flavor.” (24)

a. b.

Mary will get cranky tomorrow. Mary will be cranky right now.

metaphysical; FO epistemic; PO

In this case, the present orientation of t-sit with respect to t∗ is associated with the epistemic flavor, while the future orientation of t-sit with respect to t∗ is associated with the metaphysical flavor. The question now is whether we can connect the temporal orientation facts to something modal. That is, can we say that statives, in addition to always having a t-sit that is PO with respect to t∗ , are also always epistemic? And that eventives, in addition to always having a t-sit that is FO with respect to t∗ , are always metaphysical? And having demonstrated both of these, can we give an explanation for the temporal facts that is based on the modal facts? Condoravdi proposes (for independent reasons) an argument that, if valid, would vitiate such a line of reasoning. She argues that there are epistemic eventives. If she is correct, we cannot appeal to modal flavor to explain the correlation between temporal orientation and eventuality type. I will argue now, however, that epistemic eventives do not actually exist. It is not that Condoravdi’s examples are not actually epistemic; it is that they are derived statives, and therefore not actually eventive.

3.1 No Real Epistemic Eventives To argue for the existence of epistemic eventives in modal contexts, Condoravdi considers a context where we know that a certain professor will meet with one senior administrator. Then both the discourse in (25) and the discourse in (26) are coherent. (25)

a. b.

It hasn’t been decided yet who he will meet with. He may see the dean. He may see the provost.

(26)

a. b.

It has been decided who he will meet with but I don’t know who it is. He may see the dean. He may see the provost.

Condoravdi’s idea is that the sentences in (25b) and (26b) are assertable whether the future-oriented event is not yet settled, as in the context provided by (25a), or already settled, as in the context provided by (26a). I find that the settled reading can be a bit difficult to get, but the dialogue in (27) highlights that reading. (27)

a. b.

Who does he see tomorrow? Someone. Let’s see. Darn, I can’t find the book. He may see the dean, he may see the provost. I don’t know. Go ask him.

70

B. Copley

When something is assumed settled, like the identity of the person being seen tomorrow, one can check now to see whether it is true or not (at least, that is what the sentences in question presuppose). That is, what we get when something is settled is epistemic modal flavor. For Condoravdi, it is settledness itself that makes metaphysical modality impossible. However, to stop there is to miss a generalization. Settledness is linked to another phenomenon aside from modal flavor, namely the ability to be predicated of the present time, as we will see now. This ability makes “settled eventives” look more like statives. The sentence in (28a), which has a stative, can describe something that is currently taking place. But (28b) cannot be used to describe something taking place at the present moment. (28)

a. b.

Toby likes the dean. (does describe a current Toby-like-the-dean state) #Toby sees the dean. (cannot describe a current Toby-see-the-dean event)

In English, if you want to use an eventive verb phrase to refer to current goings-on, you can, but the meaning is either habitual, as in (29a), or something very much like Condoravdi’s settled case, but without a modal (a simple present futurate), as in (29b). (29)

a. b.

Right now, Toby sees the dean on Mondays. Right now, Toby sees the dean tomorrow.

habitual simple futurate

The important point here is that the “settled” futurate case in (29b) is fine, just like the habitual in (29a). Also like the habitual, it seems to say something about the utterance time tu , even though t-sit is different (i.e., for the habitual, today need not be a Monday for (29a) to be true).12 In the habitual case, if it is truly asserted, what is true right now is that Toby has the habit of seeing the dean on Mondays (or perhaps, what is true right now is that there is a rule that compels him to do so). In the futurate case, if it is truly asserted, what is true at tu is that there is a schedule or plan for Toby to see the dean tomorrow.13 12

In fact, if we consider other things we can do to eventive predicates to form sentences that describe a current state, we can construct parallel examples with may that are also fine. (i)

a. b. c.

Toby sees the dean on Mondays. Toby is seeing the dean right now. Toby is seeing the dean tomorrow.

habitual ongoing progressive futurate progressive

(ii)

Everyone has a special administrator who they see for advice. He may see the dean, he may see the provost, I don’t know. habitual

(iii)

He may be seeing the dean. He may be seeing the provost.

ongoing or futurate progressive

13 See also Copley (2008) for how the schedule or plan interacts with the rest of the denotation, as well as Copley (2005) for thoughts on the nature of the schedule or plan.


71

Both habitual sentences and futurate sentences, then, behave like stative sentences in that they can both be asserted of the present moment. Now, the work of the PEC in distinguishing eventives and statives is done in Condoravdi’s theory by her AT function, given below in (30). (30)

AT(t, w, P) = ∃e[P(w)(e) & τ (e, w) ⊆ t] if P is eventive = ∃e[P(w)(e) & τ (e, w) ◦ t] if P is stative = P(w)(t) if P is temporal

Eventives, treated in the first clause of the definition, are given a run time τ (i.e., tsit) that is included in the interval t. Statives, treated in the second clause, have a run time overlapping the interval t. In a matrix clause, the interval t will have the value now, which for these purposes is an interval that begins at the moment of utterance and extends infinitely into the future.14 Based on Condoravdi’s treatment of the “settled” cases under modals, as in (26b) and (27b) above, how would her theory handle the difference in judgments between (31a), a “settled” or futurate case, and (31b), an ordinary eventive case? (31)

a. b.

Toby sees the dean tomorrow. #It rains tomorrow.

Both have present tense, which for Condoravdi is as in (32). (32)

PRES: λ Pλ w [AT(now, w, P)]

The AT function guarantees that the eventive in both cases is shifted to the future. But what rules out (31b)? Presumably it is that (31b) does not describe a settled event – since we don’t normally presuppose it to be settled that it is going to rain tomorrow – while (31a) does describe a settled event. Condoravdi doesn’t discuss these cases, and treats settledness as a property that only applies to complements of modals. But let’s try to extend her theory by expanding the notion of settledness to include the non-modal cases, and adding the constraint that assertions can only be about things that are already settled: (33)

For any P, w, t: “P” can be asserted at w, t only if P is settled at w, t.

This addition may be the simplest one to make to account for the difference between (32a) and (32b). But it is also problematic. We will need to extend the definition of settledness to include temporal predicates, because one does not assert just “P” in this account, but rather “PRES P.” There shouldn’t be a problem with that. But what about extending the notion of settledness to something with a modal, like will P? Surely we want to say that sentence such as that in (34) can be asserted. (34) 14

It will rain tomorrow.

This function incorporates a version of the PEC, and thus does not address the remarks above about the non-existence of “real” FO statives.

72

B. Copley

In that case, then, according to the requirement in (33), it should be settled that “will P.” But do we really want to say that in order to assert (34), it should be settled that it will rain tomorrow? And that if it is not settled that it will rain tomorrow, that (34) is just as bad as (31b)? One alternative is to define the settledness of will P in some other way to entail that it is assertable even when it is not settled whether P will happen or not. Another alternative is to relax the requirement in (33) in such a way as to exclude the sentence in (34). But each of these alternatives seems to be quite inconvenient. This line of inquiry having hit an obstacle, we should wonder if there is anything else we could try, to explain the difference between (31a) and (31b), and account for the future orientation of settled eventives. We noted above that futurate readings and habitual readings do not predicate the event of their reference time (the present time, when the sentences are in the present tense). Habituals predicate of the reference time something like lawlike behavior, while futurates predicate of the reference time a current rule or schedule. Thus it makes sense to say that the futurate cases, like the habituals, are really derived statives. It is whatever the futurate semantics is, presumably something15 in the semantics of the imperfective, that pushes the occurrence of the event into the future. So when it combines with a modal, as far as the modal is concerned, it is just combining with a stative. And when we assert a futurate, we are asserting a stative of the present, not asserting an eventive of the future. Additional evidence for this idea comes from the fact that futurate eventive antecedents do not behave like ordinary (non-futurate) eventive antecedents, but rather like stative antecedents, in that the time that enters into a temporal relationship with the consequent is t∗ , not t-sit. Consider the conditional in (35). The antecedent could have a non-futurate reading, in which the speaker will give Mary a call after she actually leaves (and nothing is implied about Mary’s current plans), as well as a futurate reading, in which case the speaker is saying she’ll give Mary a call as soon as she finds out that she has plans to leave on Monday. (35)

If Mary leaves on Monday, I’ll give her a call.

The dialogue in (36) further demonstrates the existence of the futurate reading for the antecedent If Mary leaves on Monday; the addition of today to the consequent in fact rules out the non-futurate reading: (36)

a. b. c.

I thought Mary was already out of town. No, she leaves on Monday. Oh, well, if Mary leaves on Monday, I’ll give her a call today.

The difference between the non-futurate and futurate readings is this. As we saw earlier, with the non-futurate eventive antecedent, the t∗ cons begins when t-sitant begins. Here, this means that the speaker calls Mary a little bit after the time when she 15

If this something is a modal, which seems plausible, then we can return to Condoravdi’s idea of settledness as being a property that is only relevant to complements of modals. But this modal would not be an epistemic one, rather one having to do with natural laws/current schedules, so there would still be no epistemic eventives.


73

actually leaves. But with the futurate case, the reference time of the consequent does not start when t-sitant starts, but rather, with t∗ (or equivalently, with tu ) starts. In this, it behaves similarly to stative antecedents, discussed above. If settled eventives are really to be treated as PO statives, the way habituals are treated, there is no more justification for the existence of eventive FO epistemics, and in fact, there is no justification for the existence of any eventive epistemics at all, as eventive PO epistemics clearly do not exist, and so were never at issue. Therefore, all epistemic cases are stative.16

3.2 Antecedents Have Their Own Modal Flavors The way is almost clear for us to make the case that the correlation between eventuality type and temporal orientation should be explained in terms of a correlation of both with modal flavor. Statives have a t-sit that is PO with respect to their t∗ , and they are epistemic in modal flavor; eventives have a t-sit that is FO with respect to their t∗ , and they are metaphysical in modal flavor. One obstacle remains: if modal flavor correlates with eventuality type and temporal orientation in both antecedents and consequents, we would expect an antecedent and a consequent in the same conditional to be able to have different modal flavors from each other. This question is not addressed by Kratzer-style (Kratzer, 1986) possible worlds treatments of conditionals, which assume that a conditional has a single modal flavor (in the absence of overt embedded modals in either clause). Yet this traditional assumption, surpisingly, is not warranted. As it turns out, antecedents have their own modal flavor, and this modal flavor correlates, as we now expect, with eventuality type and temporal orientation. How do we test a sentence (or an utterance) for modal flavor? In general, it is necessary to take a native speaker’s word for it, as for judgments of felicity or truth. 16

It is not true, however, that all statives under modals are epistemic. Modals can have deontic flavor with either statives or eventives. For example, either (ia) (eventive) or (ib) (progressive, i.e. derived stative) are possible as injunctions for what Mary is supposed to do: (i)

a. b.

Mary will sing when the queen walks in. Mary will be singing when the queen walks in.

Under deontic must, as in (iia) and (iib), these possibilities have been called “obligation to do” and “obligation to be” respectively. In (iia) Mary has an obligation to sing, while in (iib) there is a requirement that she be singing at that point. (ii)

a. b.

Mary must sing when the queen walks in. Mary must be singing when the queen walks in.

Even imperatives can occur both with eventives and statives: (iii)

a. b.

Sing when the queen walks in! Be singing when the queen walks in!

I will not consider such cases further here.

74

B. Copley

There is, however, a straightforward test for antecedents.17 Adding it’s true that to the antecedent is good with stative antecedents (which have an epistemic flavor) but bad with eventive antecedents (which have a metaphysical flavor).18 The stative antecedent in (37a), even before the consequent is heard, has an epistemic flavor, and is fine with it’s true that, as shown in (37b). (37)

stative a. If it’s cold out right now,. . . b. If it’s true that it’s cold out right now,. . .

The eventive antecedent in (38a), on the other hand, has a metaphysical flavor, and is not good with it’s true that, as in (38b). (38)

eventive a. If it rains tomorrow,. . . b. #If it’s true that it rains tomorrow,. . .

(38b) may be good for certain speakers of philosophical English, but for most speakers it is terrible. Why is (38b) so bad? Let us suppose that #it rains tomorrow does not express a proposition; therefore it can’t be true or false, and can’t be asserted. Anchored stative antecedents (i.e, those with a future-oriented t∗ ) seem at first blush to behave like eventives, in that they are not possible with “it’s true that”: (39)

anchored “FO” stative a. If it’s raining at 4 tomorrow,. . . b. #If it’s true (now) that it’s raining at 4 tomorrow. . .

There is however a variant of (39b) that is acceptable, namely (40). (40)

If at 4 tomorrow it’s true that it’s raining (at that time). . .

This has an epistemic flavor as well. These data show that anchored statives are possible, but they are not propositions evaluated at the present moment (because if they were, (39b) would be good). Rather, they are propositions evaluated at the future time provided by the temporal adverbial. This explains why (40) is acceptable, and is exactly the result that was obtained in the earlier discussion of the status of these future-oriented statives: t∗ ant , the argument of the antecedent proposition, is FO with respect to tu , and t-sitant is interpreted as PO with respect to t∗ ant . Note that the anchored stative in (39b) is ruled out for a different reason than the FO eventive in (38b), which is (we suspect) itself ruled out for not being a proposition. This difference is confirmed by the fact that the eventive remains unacceptable in (41), unlike the stative in (40). (41)

#If at 4 tomorrow it’s true that it rains (at that time). . .

17 Recall that I also argued for a correlation between epistemic modal flavor and the ability to occur on its own without the modal, so that would consitute a test as well. 18 It’s unlikely that I am the first person to notice this, so I would be grateful to hear about any earlier statements of this fact.


75

So the stative antecedents are things that can be true, either now or at some future t∗ . The FO eventives cannot be true, now or ever, and we thus surmise that they are not propositions at all. It makes sense that stative antecedents, always PO with respect to t∗ ant , would have an epistemic flavor. The reason is that what one knows has to do (presumably) with which propositions one has added to a set of known propositions as of t∗ . It’s not clear yet what the eventive antecedents are – perhaps they are predicates of events – but if they are not propositions, as we suspect, we should not be surprised that they invoke a different flavor of modality, one that does not have to do with adding propositions to a set of known propositions. It is worth confirming that an antecedent and consequent in the same conditional can have different modal flavors, based on their eventuality type. That is, if we look at conditionals of the forms in (42a) and (43a), will they correspond to modal flavors as in (42b) and (43b) respectively? (42)

a. b.

If stative, will eventive =? If epistemic, will metaphysical

(43)

a. b.

If eventive, will stative =? If metaphysical, will epistemic

We exemplify the combinations we want with the stative predicate be sick and the eventive predicate get sick. (We will stay away from the tricky anchored statives.) (44)

a. b.

If John gets sick tomorrow, Mary will be sick then. If John is sick right now, Mary will get sick tomorrow.

FO + PO PO + FO

Now, the question is for each of these readings, what modal flavor is possible in the antecedent, and what modal flavor is possible in the consequent? Speakers’ judgments are as follows. In (44a), the antecedent supposes that something happens (metaphysical flavor), while the consequent conveys that in such a situation, we will find something out that is already the case (epistemic flavor). Conversely, in (44b), the antecedent supposes that we find something out now that is already the case (epistemic flavor), and the consequent then discusses what we are entitled to conclude will happen. In other words: the conditionals in (44) behave as hoped. Thus, it is official that we need a theory that can specify a different modal flavor for both the antecedent and the consequent; this is the expected result, since the correlation between eventuality type and modal flavor predicts that the antecedent and consequent modal flavors are entirely independent of each other. More urgently, we need a theory of modality that can actually account for the correlations between modal flavor, eventuality type, and temporal orientation. Moreover, this theory must be able to account for the fact that both kinds of modal flavor can occur in a single conditional, even though epistemic modality apparently has to do with propositions, and metaphysical modality apparently does not.

76

B. Copley

4 Conclusion At this point we have, I hope, some “proof of concept” that a certain kind of modal theory of temporal orientation in conditionals is an attractive one. To reach this conclusion, I argued that there is a stronger correlation than previously thought between eventuality type and temporal orientation in conditionals, and between both of these and modal flavor. Stativity, present orientation with respect to t∗ , and epistemic modality are correlated; likewise, eventivity, future orientation with respect to t∗ , and metaphysical modality are correlated. (We set aside deontic modality, which can occur with both eventives and statives.) These correlations make the traditional “present eventive constraint” less attractive as a way to explain temporal orientation in conditionals and other modal contexts, since it is not just present-oriented eventives that need to be ruled out, but also future-oriented statives. An explanation of the temporal facts based on the correlation with modal flavor is more attractive, once the objections to this kind of explanation are removed. An interesting result of this investigation was the finding that antecedents have their own modal flavor, independently from that of the consequent. One upshot is that we can no longer talk about “epistemic conditionals,” for instance, unless by that we mean a conditional whose antecedent and consequent are both epistemic. The modal independence of antecedents is predicted the viewpoint presented here because the modal flavor is correlated with the eventuality type of the predicate. The judgments on modal flavor in antecedents corroborate this prediction quite clearly, though oddly, in the mainstream generative linguistic literature, the modal flavor of the antecedent has largely passed unnoticed.

References Cipria, A. and Roberts, C. (2000). Spanish imperfecto and pretérito: Truth conditions and aktionsart effects in a Situation Semantics. Natural Language Semantics, 8:297–347. Condoravdi, C. (2001). Temporal interpretation of modals. In Beaver, D., Kaufmann, S., Clark, B., and Casillas, L., editors, Stanford Papers on Semantics. CSLI Publications, Palo Alto, CA. Condoravdi, C. (2003). Moods and modalities for will and would. Handout of talk given at the 2003 Amsterdam Colloquium. Copley, B. (2005). Ordering and reasoning. In Gajewski, J., Hacquard, V., Nickel, B., and Yalcin, S., editors, New Work on Modality, number 51, pages 7–34. MIT Working Papers in Linguistics. Copley, B. (2008). The Semantics of the Future. Ph.D. thesis, MIT 2002. Outstanding Dissertations in Linguistics, Routledge. Copley, B. (2008). The plan’s the thing: Deconstructing futurates. Linguistic Inquiry, 39(2).


77

Dowty, D. (1979). Word meaning and Montague Grammar. Reidel, Dordrecht. Klein, W. (1997). Time in Language. Routledge, New York. Kratzer, A. (1986). Conditionals. In Papers from the Regional Meetings, Chicago Linguistic Society, pages 1–15. Reichenbach, H. (1947). Elements of Symbolic Logic. Macmillan, New York. Werner, T. (2002). Deducing the future. In Kadowaki, M. and Kawahara, S., editors, Proceedings of NELS 33. Werner, T. (2006). Future and non-future modal sentences. Natural Language Semantics, 14(3):235–255.

On the Temporal Syntax of Non-Root Modals Hamida Demirdache1 and Myriam Uribe-Etxebarria2

Abstract The goal of this paper is to lay the foundations of a crosslinguistically valid model of the temporal construals of non-root modal sentences. The two languages discussed are English and Spanish. We propose a uniform temporal syntax for non-root modals where the heads, T◦ , M◦ , ASP◦ , V◦ , each introduce a time-denoting argument/reference-time projected in the syntax (onto the external specifier of the relevant head). These time intervals (Zeit-Ps) can, just as any DP, enter into anaphoric dependencies where anaphora will be construed as either coreference or binding, and undergo phrasal movement (QR) to higher scope positions. Syntactic movement at LF—be it XP movement of a Zeit-Ps or X◦ movement of a temporal head—can reverse initial temporal scope relations. Movement must target the closest licit landing site that would yield a well-formed temporal output and is a last resort operation generating a temporal output that could not otherwise be generated. We derive the asymmetries between non-root modals (without auxiliary Haber/Have) in Spanish and English from a single assumption: in Spanish, unlike English, past morphology on the modal is temporally interpreted. Key words: Time-denoting arguments, temporal anaphora, temporal scope reversal, epistemic modality, metaphysical modality

We are indebted to Claudia Borgonovo, Sarah Cummins, Jacqueline Guéron, Heidi Harley, Raffaella Folli, Brenda Laca, Jacqueline Lecarme, Tim Stowell and the participants of the Round Table on Tense, Mood & Modality (Paris, 2005) for insightful comments and questions. This research was partially funded by the LLING/EA 3827, the University of the Basque Country (9/UPV 0003313888-2001; 9/UPV 00114.130-16009/2004), the Spanish Ministry of Science and Technology (MICYT)/FEDER (BFF2002-04238-C02-01) and the Basque Government (Program for the Development of Research Nets in Humanities, calls 2006 and 2007, and Research Groups IT-210-07). The work reported here is part of Program #1.2 (“Temporalité: typologie et acquisition”) of the CNRS Fédération Typologie et Universaux Linguistiques (FR 2559).

1 2

University of Nantes/LLING (EA 3827), Nantes, France University of the Basque Country, Gasteiz, Spain


79

80

H. Demirdache and M. Uribe-Etxebarria

1 Crosslinguistic Asymmetries in the Temporal Construals of Non-Root Modals Non-root modals report the speaker’s modal judgment relative to the truth-value of the modal propositional complement. There are two times involved in their interpretation: the time at which the possibility or necessity under discussion holds—which we call the modal-time (MOD-T); and the time at which the situation described by the propositional complement in the scope of the modal holds—its situation or eventuality time (SIT-T/EV-T). This paper seeks to lay the foundations of a crosslinguistically valid model of the temporal construals of non-root modal sentences. The two languages we discuss here are English and Spanish.1 English modals verbs (EMVs) are morphological very impoverished. Although they occur only in finite clauses, they do not show person agreement, nor do they give rise to do-support. Further, many EMVs (e.g. must or ought to) do not exhibit any morphological present vs. past alternation, and the semantic present vs. past contrast displayed by some EMVs (e.g. can/could, may/might) is neutralized in certain contexts. In (1), for instance, the apparent past inflection on could/might does not serve to locate either the time of the possibility under discussion (the modaltime), or the situation-time of the modal complement [AMINA BE ASLEEP/WIN THE RACE ] in the past since the time of the possibility is speech-time and time of the situation of the modal complement is construed as either ongoing at speech-time (1a) or as forward shifted relative to speech-time (1a, b). (1)

a.

b.

Amina could/might be asleep. MOD - T holds at UT - T , SIT - T is ongoing at/future-shifted relative to the present MOD-T. Amina could/might win the race. MOD - T holds at UT - T , SIT - T is future-shifted relative to the present MOD - T .

The situation is very different when we turn to Spanish. Spanish modal verbs (SMVs) contrast sharply with EMVs in that they can appear fully inflected for person agreement as well as for tense and aspect, as illustrated in (2) with the verbs deber (‘must/should’) and poder (‘can/may’). (2)

1

a.

Maddi puede/debe estar dormida/ganar la carrera. Maddi mayPRES.3.SG /mustPRES.3.SG be asleep/win the race.

In this paper, we leave aside root modals and focus only on the non-root readings of the English and Spanish examples discussed. Due to space limitations, we do not address the whole range of differences that distinguish English and Spanish modals. We refer the reader to Stowell (2004), Drubig (2001) and Butler (2004, 2005), for recent analyses of English modals, and to Borgonovo and Cummins (2005), Bosque (1999), Bosque and Torrego (1995), Garc´ıa (2006), Gómez Torrego (1988, 1999) and Laca (2006) among others, for analyses of Spanish modals.

On the Temporal Syntax of Non-Root Modals

b.

c.

81

Maddi pudo/debió estar dormida/ganar la carrera. Maddi mayPERFECTIVE PAST.3.SG /mustPERFECTIVE PAST.3.SG be asleep/win the race. Maddi pod´ıa/deb´ıa estar dormida/ganar la carrera. Maddi mayIMPERFECT PAST.3.SG /mustIMPERFECT PAST.3.SG be asleep/win the race.

The present vs. past inflection on the modal has an impact on the temporal interpretation of the clause. In particular, (2a), with a present modal, allows a temporal interpretation parallel to the English (1a, b): the possibility under discussion holds at speech-time, while the SIT-T of the modal complement is construed either as ongoing (‘be asleep’) or as future shifted (‘be asleep/win the race’) relative to speech-time. This construal, however, is not available with the past modal verbs in (2b, c). That is, the Spanish (2b, c), with past morphology on the modal verb, are not the counterpart of the English examples in (1). The Spanish past modal verbs in (2b, c) yield the two different temporal construals given in (3). Under the reading (3a), the possibility under discussion (that is, the MOD-T) holds at speech-time, while the SIT-T of the propositional complement in the scope of the modal [MADDI WIN THE RACE] is past-shifted relative to the present modal-time. Following Condoravdi (2002), we will say that, on this construal, the modal provides a present temporal perspective about a past situation. In contrast, under the reading in (3b), the modal provides a past temporal perspective about a future (in the past) situation: the time at which the possibility holds is past-shifted relative to speech-time, and the SIT-T of the propositional complement [MADDI WIN THE RACE] is forward-shifted relative to the past modal-time. (3)

a.

b.

Present temporal perspective construal MOD - T holds at UT - T , SIT - T is past-shifted relative to present MOD - T . ‘(It is possible that) Maddi could/might have (already) won the race.’ Past temporal perspective construal MOD - T past-shifted relative to UT - T , SIT - T is forward-shifted relative to past MOD-T. ‘(At some point in the past) Maddi could/might (still) have won the race.’

We have seen that these two construals are not available in English for simple modals with apparent past inflection such as (1). As indicated by the English glosses in (3), to achieve either the present perspective construal about a past situation in (3a), or the past temporal perspective construal in (3b), could/might must combine with perfect auxiliary have. (4) thus allows both of the construals in (3). (4)

Maddi could/might have won the race

This intricate set of facts raises obvious questions concerning the crosslinguistic temporal interpretation of sentences with modal verbs. Namely, what is the temporal contribution of non-root modal verbs to the temporal construal of the sentence in which they occur—be it in English or Spanish? How do modal verbs—interact

82


with tense and (outer/inner) aspect? How do we achieve a unified analysis of the temporal interpretation of non-root modal sentences while at the same time accounting for these crosslinguistic asymmetries? How do we account for the asymmetries displayed by English and Spanish modal sentences, in particular, with regard to the temporal value of inflectional features on modal verbs? This paper seeks to provide answers to these question and, thus, lay the foundations of a crosslinguistically valid model of the temporal representation of non-root modal sentences. The paper is organized as follows. Sections 2 and 3 lay out the theoretical ground for the analysis of non-root modals developed here: section 2 summarizes the framework for temporal representations developed in our previous work, and section 3 introduces two theoretical assumptions adopted from Condoravdi’s (2002) analysis of non-root modals. Section 4 then incorporates non-root modals within our model of temporal representations. Section 5 and 6 are devoted to our analysis of English and Spanish modals, respectively. Section 7 concludes the discussion and summarizes the theoretical consequences of the analysis.

2 The Grammar of Temporal Relations The temporal syntax for modals we propose presumes some familiarity with the framework for temporal interpretation developed in Demirdache and UribeEtxebarria (1997, 2000, 2002, 2004, 2005, 2007, in press; henceforth D&U-E). We present below the core relevant assumptions of this model.

2.1 Tenses and Aspects as Spatiotemporal Predicates Tenses and aspects are dyadic predicates of spatiotemporal ordering taking timedenoting arguments. They uniformly project their time arguments onto the syntax, as Zeit-phrases in the sense of Stowell (1995), and serve to establish ordering/topological relations between their time arguments (ZeitPs): ZeitP1 BEFORE/ AFTER/WITHIN ZeitP2 , as illustrated in (5, 6). This proposal yields the phrase structure for temporal relations in (7) where both tense and aspect are assigned isomorphic structural representations.2 (5) Syntax of tense UT- T

T’

To AST- T WITHIN AFTER / BEFORE

2

(6)

Syntax of aspect ASP - TP

TP

AS - T

ASP ’

ASPo EV- T WITHIN AFTER / BEFORE

Our proposal extends to time adverbs uniformly analyzed as PPs headed by a predicate of spatiotemporal ordering. The reader is referred to D&UE (2004, 2005) for an analysis of time adverbs as temporal modifiers of time-denoting arguments. See footnote 3.


83

(7) The isomorphic syntax of tense and aspect TP REF - T UT- T

T’ To

ASP - P ASP ’ REF - T AST- T ASPo VP EV- T

VP

The phrase structure in (7) breaks down both tense and aspect syntactically into their respective semantic components. The functional heads ASP◦ and T◦ each relate two time arguments. The external argument of T◦ is a reference-time ( UT-T in an independent or matrix clause). T◦ orders this reference-time with respect to its internal argument, the time of the assertion (AST- T). The external argument of ASP◦ is itself also a reference-time (the AST- T). ASP◦ orders this time interval with respect to its internal temporal argument, the time of the event denoted by the VP. Viewpoint aspect serves to focus a subinterval in the temporal contour of the described event (Smith, 1991). This interval is the AST-T-that is, “the time to which the assertion is confined, for which the speaker makes a statement” (Klein, 1995). Why does aspect focus a time span in the temporal contour of the described event? Because aspect, just like tense, is a spatiotemporal predicate ordering two time intervals: the AST-T (its external argument) and the EV-T (its internal argument). This ordering relation can be one of subsequence, precedence or inclusion: (8) Viewpoint aspect Retrospective AST- T after EV- T EV- T AST- T [ ] [ ] >

Progressive within EV- T AST- T [ [ ] ] > EV- T AS - T

Prospective before EV- T AST- T EV- T [ ] [ ] >

AST- T

Tense then orders the AST-T relative to a reference-time (typically, the UT-T in main clauses). This ordering relation will again be one of subsequence, precedence or inclusion, as illustrated in (9). This proposal yields the typology for predicates of spatiotemporal ordering in (10). (9) Tense Past after AST- T AST- T UT- T [ ] [ ] >

UT- T

UT- T

[

Present within AST- T UT- T [ ] ] >

AST- T

Future before AST- T UT- T AST- T [ ] [ ] >

UT- T

84


(10) Tense Aspect

AFTER

WITHIN

BEFORE

(Subsequence)

(Inclusion)

(Precedence)

Past Retrospective/Perfect

Present Progressive

Future Prospective

The time intervals (Zeit-Ps) projected into the syntax as either covert arguments of tenses and aspects, can, just as any DP: (i) enter into anaphoric dependencies where anaphora will be construed as either coreference/covaluation or (semantic) binding; and (ii) enter into scopal relations with other time arguments.3 The proposal that time arguments, just like any DP, can enter into either anaphoric dependencies or scopal relations with other time arguments will play a crucial role in the analysis of the temporal construal of modals developed in sections 4–6. We explicate below the distinction between covaluation and binding when applied to the temporal realm, while at the same time providing empirical arguments for assuming that this distinction in the construal of anaphora extends to the temporal realm.

2.2 Anaphora The role of ASP◦ is to overtly/morphologically specify the ordering relation holding between the EV-T and the AST-T. When ASP◦ is not morphologically overt, the relation between the AST-T and the EV-T is established via anaphora.4 We know that anaphora between individual denoting noun phrases can be established via either coreference/covaluation or (semantic) binding—which can be distinguished in the appropriate context (Reinhart, 1997).5 The default hypothesis is that anaphora between time-denoting noun phrases can likewise be established via either coreference or binding and, further, that the resulting temporal construals can be distinguished. This proposal allows D&UE to derive the aspectual viewpoint of simple tenses—which is either perfective or neutral in the sense of Smith (1991) and Laca (2005) (see D&UE, 2007, in press, in prep). 3 Further, just like regular DPs, time-arguments can be restrictively modified. See D&UE (2004, 2005) for an analysis of time adverbs as temporal modifiers restricting the temporal reference of either the AST-T or the EV -T. 4 The same analysis extends to zero tense. When there is no morphological tense specifying the ordering relation between T◦ ’s external argument (UT-T in main clauses) and internal argument (the AST-T), this relation is established via anaphora. Merely for reasons of simplicity, we assume throughout that when T◦ is null, the AST-T is covalued with the UT-T, as illustrated with the analysis of the tenseless perfect in (41a), section 5.2.3. Note, however, that if the default construal of anaphora is in fact binding (see footnote 12), then zero tense will entail binding of the AST-T by the UT-T. 5 Consider, for instance, the examples in (i). (ia) asserts that Amina is the only individual to love Mr. Grace (Amina’s father), while (ib) asserts that Amina is the only individual to love her own father. See Reinhart (1997).

(i)

Only Amina loves her father. a. Covaluation: Only Amina λX [x loves her father] b. Binding: Only Amina λX [x loves x’s father]

her = Amina


85

2.2.1 Perfective Viewpoint Simple past tense sentences without over morphological aspect are assigned the temporal phrase structure in (11). To orders the UT-T after the AST-T, yielding the temporal construal in (12a). ASPo having no morphological content, the ordering relation between the AST-T and the EV-T is established via anaphora. (11) Past tense without morphological aspect TP UT- T

T’

To ASP - P AFTER ASP ’ AST- Ti ASPo

VP

EV- Ti

(12) a.

VP

UT- T AFTER AST- T :

b.

Covaluation:

Consider first what happens when anaphora in (11) is coreference or covaluation —that is, the assignment of identical semantic/temporal values to two intervals. Covaluation yields an ordering of exhaustive coincidence: the AST-T and the EVT denote the same time interval. Since these two intervals are cotemporal, the EV - T ends up indirectly ordered (via covaluation) in the past relative to UT-T, as shown in (12b). This derives the temporal interpretation of the French simple past in examples such as Amina pleura ‘Amina cried’, which have a perfective interpretation. Covaluation thus explains why the event described by a simple past tense is portrayed in its entirety, from its initial to its final bound—so-called perfective/aoristic aspect. Covaluation entails that the AST-T and the EV-T denote exactly the same time interval. An imperfective viewpoint is then automatically excluded: the AST-T cannot denote a subinterval of the EV-T.

2.2.2 Neutral Viewpoint Aspect We have argued that when there is no morphological aspect overtly specifying the ordering relation between the AST-T and EV-T (as is the case with morphological simple tenses), this relation is established via anaphora. We have seen that covaluation between the AST-T and the EV-T yields an ordering of exhaustive coincidence, the resulting viewpoint is thus perfective. If anaphora were to be established via (semantic) binding, then the resulting viewpoint would in fact be neutral aspect, in the sense of Smith (1991), Laca (2005). To see why, consider (13).

86


(13)

a.

ASP - P

AST- T

ASP ’

λEV-T

VP

EV-T

b. (14) a.

AST- T

VP CRY ( AMINA )

λI [ CRY ( AMINA )(I )]

Imperfective viewpoint: AST-T ⊆ EV-T

b. Perfective viewpoint: AST-T ⊇ EV-T

λ -abstraction over the temporal variable inside the VP (EV-T/ I)6 in (13) creates a predicate which takes the AST-T as external argument. (13) requires that the AST-T have the property of being an interval at which AMINA CRY obtains. Since cry is an activity predicate with the subinterval property (if Amina cried from 1:00 to 2:00, then most subintervals of this time are times at which Amina cried; Dowty 1986), then AMINA CRY is true/holds of the AST-T in either of the temporal configurations in (14). In (14a), the AST-T is contained within the interval during which Amina cries, yielding an imperfective viewpoint. In (14b), the AST- T itself contains the interval during which Amina cries, yielding a perfective viewpoint. In sum, when ASP◦ is empty, semantic binding of the EV- T by the AST- T yields neutral viewpoint. With this proposal, D&UE (2007) explain how the French simple present and future, as well as the imparfait, yield neutral viewpoint aspect (Smith, 1991; Laca, 2005). Note that our analysis correctly predicts that neutral aspect arises only with predicates satisfying the ‘subinterval property’. In particular, achievements in French do not allow the imperfective construal in (14a) and, hence, have perfective viewpoint aspect. And as for accomplishments, only those accomplishment predicates that can be construed as activities (that is, can be shifted/coerced into activity predicates—e.g. déclamer un poème ‘declaim a poem’) allow an imperfective construal with the simple present and future in French.7 6

Note that the event/assertion-times are merely temporal variables, labeled for convenience but we might just as well have chosen I to stand for these variables over intervals. 7 Telic predicates (achievements and accomplishments) cannot yield the imperfective construal in (14b) where the AST-T is itself contained within the EV -T—and, hence, excludes the final bound of the EV -T —because with a telic predicate such as run a mile, AST-T λI [RUN (ONE MILE)(I )] will only be satisfied if the AST -T contains the culmination point/final bound of the running event. Why? Because run a mile does not have the subinterval property: if Amina ran a mile between 1:00 and 2:00, then it is false that she ran a mile in any subinterval of this time. Thanks to Jacqueline EV -T/ AST-T,


87

3 Condoravdi (2002): Perfect Modals Condoravdi’s goal is to provide a unitary account of the semantic contribution of non-root modals to the temporal interpretation of the sentence in which they occur, and then derive the type of modality they express from the given temporal interpretation. In particular, Condoravdi proposes that non-root modals uniformly contribute an open ended interval, [t, ∞), to the temporal interpretation of the clause in which they occur; that is, an interval, starting at an initial bound t and extending without limit into the future. We adopt Condoravdi’s proposal that non-root modals, uniformly contribute a time interval, [t, ∞), to the temporal interpretation of the clause in which they occur. We take this open-ended interval to be a reference-time, projected onto the syntax as the external (time) argument of a modal head. We refer to this time argument as the modal-time (MOD - T). Condoravdi further argues that the ambiguity in the temporal construal of English modals combining with perfect have (illustrated in (4) and repeated below) has its source in the relative scope of the modal and perfect operators. The present perspective construal in (15) arises when the modal scopes over the perfect, as in (16) (where MB is a modal base and τ the running-time of the event e).8 (15)

Amina might have won the race. Present temporal perspective: (It is possible/consistent with what is known that) Amina may/might have (already) won the race (# but she didn’t).

(16)

PRES ( MIGHT MB ( PERF (he win))): λ w∃w [w ∈ MB (w, now) & ∃t [t < [now, ∞) & ∃e [[ he win](w )(e) & τ (e, w ) ⊆ t ]]]

In (16), the modal introduces a reference-time that starts at UT- T, extending indefinitely into the future [now, ∞). The perfect also introduces a reference-time t ordered in the past relative to the modal reference-time (and, thus, relative to UT-T), and required to contain the running-time (and, thus, relative to UT-T), and required to contain the running-time of the winning event. The winning-time is thus past shifted relative to UT- T. Note that since the issue of whether Amina won the race or not has been settled at UT- T (as we are now at a time following the described past event), modality relates to the epistemic state of the speaker, his/hers lack of evidence/knowledge as to the outcome of the described event. Guéron, Brenda Laca and Arnim von Stechow for insightful discussion of our analysis of neutral aspect and its predictions. The immediate question raised by this proposal is how to account for cross-linguistic variation in the construal of simple tenses: why do the simple present and future in French, but not in Spanish or English, have neutral viewpoint aspect? See Laca (2005) and footnote (12) for interesting proposals. 8 With an epistemic modal base, MB (w, now) is the set of worlds compatible with what a subject knows in w at UT-T. With a metaphysical modal base, MB (w,t) consists of the set of metaphysical alternatives of w at t.

88


In contrast, the past perspective construal in (17) arises when the scope of the perfect relative to the modal is reversed: the perfect scopes above the modal, as in (18). (17)

Amina might have won the race. Past temporal perspective: At some point in the past, Amina might (still) have won the race (but she didn’t).

(18)

PRES ( PERF ( MIGHT MB (he win))):

∃e [[he

win](w )(e)

&

τ (e, w )

λ w∃w ∃t [t < now & w ∈ MB (w,t & ⊆ [t , ∞] ]]

The perfect introduces a past reference-time t . Since the perfect has scope over the modal, the reference-time set by the modal starts at the past reference-time introduced by the perfect, extending indefinitely into the future. Past shifting of the interval introduced by the modal is thus due to the perfect. Forward shifting of the running-time of the winning event is, in turn, due to the modal: the later contributes a time interval extending into the future and required to contain the running-time of the winning event. On this construal, the modal thus provides a past temporal perspective about a future situation. This reading is counterfactual.9 Modality does not involve epistemic uncertainty since, at the past time of the possibility, the issue of whether Amina won or not, had not been settled yet. For Condoravdi (2002: 4), the modality is metaphysical, relating to “how the world might have turned out to be: at some point in the past, the world was such that it could evolve into a world in which [s]he won the game”. In sum, on the epistemic reading in (15), the possibility is from the temporal perspective of the present, about a past situation. On the metaphysical reading in (15), the possibility is from the temporal perspective of the past, about a future (in the past) situation.10

9 Condoravdi explains the counterfactuality of this construal as a pragmatic inference arising from the speaker’s choice of a past over a present perspective modal. This reading is only weakly counterfactual in English (as opposed to German)—e.g. “At some point in the past, Amina could/might have won the race, and in fact she did” (see D&UE in press). 10 As reported by Condoravdi (2002) in a footnote, perfect modals can yield a third construal: the future perfect reading illustrated in (i).

(i)

Amina may have arrived by next week.

The time of the possibility in (i) holds at UT-T, as was also the case for (4) under the epistemic present perfect construal in (15, 16). However, whereas in (15, 16), the EV -T of the modal complement is past-shifted relative to the UT-T, in (i), it is past-shifted relative to a future reference-time, the time designated by next week. Due to space limitations, we do not address this third reading here. The reader is referred to D&UE (in press) for a detailed analysis of this construal analyzed as a resultative/continuative future perfect.


89

We will adopt Condoravdi’s proposal that the ambiguity in the construal of perfect modals illustrated in (15) and (17), is a scopal ambiguity. A past temporal perspective arises when the scope of the perfect relative to the modal is reversed (see also Butler, 2004, 2005, and Stowell, 2004).

4 The Temporal Phrase Structure of Non-Root Modals We incorporate non-root modals into our phrase structure for tense and aspect by assuming that they head a maximal projection ( MP), embedded under TP and above ASP - P , as in (19). Under this proposal, there is an outer tense projection embedding the modal projection, but no inner tense projection under the scope of the modal. (19)

TP UT- T

T’ To

MP MOD - T

M’

[t,∞) Mo

ASP - P AST- T

ASP ’

ASPo

VP EV- T

VP

We adopt Condoravdi’s proposal that non-root modals, whether they are present, future or past oriented, uniformly contribute a time span to the temporal interpretation of the sentence in which they occur. This reference-time is an open-ended interval [t, ∞)—that is an interval starting at an initial bound t, and extending without limit into the future. We call the time argument contributed by the modal head, the modal-time (MODT ). This time span is projected into the syntax as the external temporal argument of the modal head. Under this proposal, each head (TP, MP, ASP - P, VP) in (19) introduces a reference-time projected onto a specifier position in the syntax. This proposal raises two immediate questions. How is the modal-time set? How is the modal-time ordered relative to the other time arguments in the clause?

4.1 Setting the Modal-Time The modal in (19) contributes a reference-time to the temporal interpretation of the clause. This interval, [t, ∞), has an initial bound but no final bound. How then is the modal-time set, temporally anchored? We assume that the modal-time can be set

90


either deictically or anaphorically. Deictic anchoring (of the initial subinterval) of the modal-time is anchoring to the UT- T, as shown in (20) below: (20) Deictic anchoring of the modal-time TP UT- Ti

T’ To

MP MOD - T

[ti ,∞)

Mo

M’ ASP - P AST- T

ASP ’

ASPo

VP EV- T

VP

Note that in the phrase structure in (20), the only available controller for the modaltime—more precisely, for its initial bound t—is the UT- T. The modal-time thus picks out an interval starting at the time of speech and extending without limit into the future: [UT- T, ∞).11 Deictic anchoring will ensure that the temporal perspective provided by the modal is from the present onwards into the future, as is the case when the modality is epistemic: (21)

Epistemic modality a. Present temporal perspective about a past situation: (It follows from what is known at UT- T that) Amina may/might have traveled to Europe. b. Present temporal perspective about a future situation: (It follows from what is known at UT- T that) Amina may/might travel to Europe.

Anaphoric anchoring (of the initial subinterval) of the MOD - T will be discussed in section 5.2.3, when we turn to the derivation of the past perspective metaphysical construal discussed in (17–18) above.

4.2 Ordering the Modal-Time We now turn to the question of how the modal-time in (20) is ordered relative to the time arguments under its scope—that is, the AST- T and the EV- T. Note that the 11

The question arises as to whether deictic anchoring involves binding of the initial bound of the by the UT-T, or covaluation between these two times. We remain agnostic on this point in this paper and for reasons of simplicity represent deictic anchoring as coindexation. MOD -T


91

modal phrase ( MP), unlike TP or ASP - P, is not headed by a spatiotemporal predicate establishing an ordering relation between the time argument it takes as external argument (MOD - T) and the time argument immediately under its scope. Recall, however, the assumption defended in section 2.2: when either T◦ or ASPo is not morphologically overt, the ordering relation between the external and internal arguments of the (relevant) temporal head is established via anaphora. Since the head of MP is not a temporal head, we assume that the ordering relation between its external argument (MOD - T) and the time argument under its scope is, likewise, established via anaphora where anaphora is (semantic) binding.12

5 The Temporal Interpretation of English Non-Root Modals Lets now see how the above proposal derives the temporal construal of English nonroot modals. We first discuss bare non-root modal verbs (that is, without perfect have). We then turn to the construals of perfect modals.

5.1 Present/Future Oriented Epistemic Modals Recall from our discussion of (1) vs. (4) above, that bare non-root modal verbs always have a present temporal perspective (the time of the possibility holds at UT- T ) and yield either a non-shifted (ongoing) or a future-shifted construal of the propositional complement under their scope. This is the case irrespective of the morphological makeup of the modal verb (may/can vs. might/could). In particular, the apparent past inflection on could/might does not serve to locate in the past, either the time of the possibility under discussion (the MOD - T) or the SIT- T of the modal complement. In this respect, English modal verbs contrast sharply with their Spanish counterpart discussed in (2) where the past morphology on the modal yields either of the two construals illustrated in (3): past-shifted SIT- T of the modal complement on the present perspective construal in (3a), or past-shifted modal time on the past perspective construal in (3b). On the basis of this contrast between English vs. Spanish, we conclude that English non-root modals are tenseless: the outer TP projection embedding the modal 12

Note that for Reinhart (1997), binding is the default construal of anaphora. Coreference is ‘special’, licensed by a pragmatic rule. Extending this proposal to the temporal realm would predict the default viewpoint of simple tenses (without morphological aspect) to be neutral aspect since the latter arises via binding of the EV -T by the AST-T (see section 2.2.2)—that is, via the default construal of anaphora under Reinhart’s generalization. According to Laca (2005), the default viewpoint for simple tenses is indeed neutral aspect. Laca argues, however, that the construal of a given simple tense is subject to polarization effects arising by Q-implicature. For instance, in both English and Spanish, the simple past is aspectually neutral. The existence of a progressive in these languages entails that the simple past is interpreted, by polarization, as perfective (its imperfective value is negated). Since the progressive cannot combine with states, past states have neutral viewpoint.

92


in the temporal phrase structure given in (19) (or (25)/(32) below) is not headed by a covert semantic tense, be it present tense (as assumed in Condoravdi 2002) or past tense.13 (22)

Amina may/might/could/should be in Ottawa (now/tomorrow).

(23)

Amina may/might/can/could/should win the marathon.

As is well known, the distribution of the non-shifted vs. future-shifted readings of the modal complement is sensitive to lexical aspect (see Zagona, 1990; Condoravdi, 2002; Werner, 2003; or Stowell, 2004). When the complement of the modal verb is a stative predicate, as in (22), the time of the situation ([AMINA BE IN OTTAWA]) can be construed either as future-shifted or as non-shifted (ongoing) relative to UT- T . However, with an eventive modal complement, such as [ AMINA WIN THE MARATHON ] in (23), the situation-time of the modal complement is construed as future-shifted relative to the UT- T. Werner (2003) observes, however, that eventive complements can also yield a future-shifted construal where the SIT- T of the modal complement overlaps with UT- T , as illustrated in (24). ((24a–c) are adapted from Werner, 2003: 119). (24)

a. b. c. d. e.

Nora may/might/could/should wash her car from top to bottom. Nora may/might/could/should shake everybody’s hands. Nora may/might/could/should cross the street slowly. Nora should work 24 hours straight to get her article in on time. Maddi should sing all the songs in the repertory.

(24a) can be uttered felicitously even when the car washing event has already started, (24b) even after Nora has started shaking hands, (24c) while Nora is already in the process of crossing the street, and (24d/e) when the event of singing/working has already started. Werner points out that this construal is licensed only when certain types of modifiers are added to the VP. We return to this point below. Note, however, that achievement predicates allow this construal even without a modifier. Thus, (23) can be felicitously uttered when the running of the marathon has already started. In order to make the demonstration that follows more perspicuous, we adopt the simplified temporal phrase structure in (25) for simple non-root modals where ASP P is not projected. If we ignore, at this first stage, the temporal contribution of aspect, it is merely in order not to obscure the insight underlying our proposal by adding an extra step in the temporal calculus (that is, an extra layer of anaphora since there is no overt aspectual head in (22) or (23)). We will give below a full temporal structure for the sentences in (22) and (23).

13

See also Abusch (1997:23), who likewise argues that non-root modals do not occur in the scope of a tense operator: “present/future oriented modals are semantically tenseless and directly pick up the local evaluation time as a modal perspective, without mediation of tense”.


93

(25) Simplified temporal phrase structure (without ASP - P) TP UT- Ti

T’ To

M MOD - T

[ti , ∞)

M’

Mo

λEV −T SIT- T

VP VP [ AMINA BE IN OTTAWA ] [ AMINA WIN THE MARATHON ]

The outer TP projection embedding the modal in (25) above is not headed by a covert semantic present tense. The modal head in (25) contributes a REF - T to the temporal interpretation of the clause: [t, ∞). This interval must be temporally anchored. The only available controller for the initial bound of the modal-time is the UT- T. The MOD - T is thus set to UT- T via deictic anchoring (see section 4.1). The MOD - T therefore denotes an open-ended interval starting at UT- T, [UT- T, ∞), as illustrated in (26a). (26) a.

b. c.

MOD - T MOD - T

λI [IN OTTAWA (AMINA)(I )] λI [WIN (MARATHON) (AMINA)(I )]

Since there is no temporal head to order the MOD - T relative to the SIT- T of the VP in (25), the ordering is established via semantic binding. λ -abstraction over the temporal variable inside the VP (SIT- T/I) in (25/26b, c) creates a predicate which takes the MOD - T as external argument. Semantic binding entails that the MOD - T ([UT- T, ∞)) have the property of being an interval at which [AMINA BE IN OTTAWA] or [AMINA WIN THE MARATHON] be true. (26b) is satisfied under either of the two temporal construals illustrated in (27) and (28). In (27), the MOD - T (that is, the open-ended interval [UT- T, ∞)) properly contains the time of Amina’s being in Ottawa. This derives the construal of (22) where the SIT- T of the modal complement is future shifted relative to the MOD T /UT- T . (27)

MOD - T

properly contains the time of Amina’s being in Ottawa

The MOD - T in (26b) will also have the property of being an interval at which [AMINA BE IN OTTAWA] holds if a subinterval of the time of Amina’s being in Ottawa (SIT- T) overlaps with UT- T, as depicted in (28). This yields the non-shifted

94


construal of (22) where the SIT- T of the stative complement is understood to hold at the MOD - T/UT- T. (28)

MOD - T

contains a subinterval of the time of Amina’s being in Ottawa

Summarizing, since the MOD - T is an interval starting at UT- T and extending into the future, semantic binding with a stative predicate yields either a future shifted or a non shifted construal of the SIT- T of the modal complement. Let us now consider the derivation for a simple non-root modal with an eventive modal complement (e.g. (23)). λ -abstraction over the temporal variable inside the VP in (25/26c) creates a predicate which takes the MOD - T as external argument. Binding entails that the MOD - T/[UT- T, ∞) have the property of being an interval at which [AMINA WIN THE MARATHON] be true. This property will hold of the MOD - T under either of the temporal construals illustrated in (29) and (30). (29)

MOD - T

properly contains the running time of the winning event

In (29) the running time of the winning event is properly contained within the modal interval, [ UT- T, ∞). This temporal configuration generates the construal where the EV- T of the modal complement is future shifted relative to UT- T . The MOD - T in (26c) will also have the property of being an interval at which [AMINA WIN THE MARATHON] is true if it at least properly contains the interval at which the winning event culminates, as depicted in (30). (30)

MOD - T

contains the interval at which the wining event culminates

This yields the future shifted construal of eventive predicates with simple modal verbs, discussed by Werner and illustrated in (24), where the future shifted EV- T of the modal complement can be construed as overlapping with UT- T. Clearly, the construal illustrated in (30), where only the culmination time of the event is future shifted relative to the UT- T, is not available with all eventdenoting predicates. It appears to be allowed with either achievement predicates (e.g. win, reach the summit, die) or, as pointed out by Werner, with modified activity/accomplishment predicates—that is, with predicates that either end up denoting events unfolding over an extended interval of time (24a)/(24c, d) or events construed iteratively (24b)/(24e). This restriction, however, does not hold in French:


(31)

95

Que fait Amina en ce moment? ‘What is Amina doing now?’ a. Elle doit chanter. ‘She must be singing.’ b. Elle doit travailler (depuis au moins quelques heures maintenant) ‘She must be working (for at least a few hours now).’ c. Elle doit laver sa voiture. ‘She must be washing her car.’

As illustrated in (31a–c), activity predicates in French allow an ongoing construal construal of the eventive modal complement where only the culmination of the event time is future-shifted relative to the present modal time, [UT- T, ∞). We leave the source of this contrast between English and French for further research. For completeness, we give below the full temporal phrase structure for (22/23). (32)

TP UT- Ti

T’ To

MP MOD - T

[ti , ∞)

M’

Mo

λAST −T

ASP - P

AST- T

ASP ’

ASPo

λEV −T EV- T

VP - P VP

[ AMINA BE IN OTTAWA ] [ AMINA WIN THE MARATHON ]

The MOD - T is set via deictic anchoring. It thus denotes an open-ended interval starting at UT- T. The MOD - T, in turn, binds the AST- T. Since ASP◦ in (32) is empty, the AST- T in turn binds the EV- T . The modal-time is thus indirectly required to be an interval at which AMINA BE IN OTTAWA /AMINA WIN THE MARATHON obtains. The MOD - T will satisfy this property under either of the temporal construals illustrated in (27, 28) or (29, 30) above.

5.2 Aspectually Complex Modals: Progressive and Perfect Modals We now show how our analysis derives the temporal construal of modals with progressive and perfect complements.

96


5.2.1 Epistemic Progressive Modals Consider the sentence in (33), with a progressive modal complement. The MOD - T provides a present temporal perspective (epistemic modality) and the SIT- T of the modal complement can be construed as either non-shifted/ongoing (33a) or as future shifted (33b) relative to UT- T. (33)

Alaia may/might be writing a letter. a. Alaia may/might be writing a letter now. b. Alaia may/might be writing a letter tomorrow/ when you get home.

The temporal phrase structure assigned to (33) is given in (34a). Note that the representation in (34a) is identical to the one given in (32) above in all relevant respects except that this time, the head corresponding to ASP◦ is not empty (as was the case in (32)), but is occupied by the spatiotemporal predicate WITHIN (since (33) contains progressive viewpoint aspect). (34)

a.

TP UT- Ti

T’ To

M MOD - T

[ti , ∞)

M’

Mo

λAST −T

ASP - P

AST- T

ASP ’

ASPo ASP - P WITHIN VP EV- T [ ALAIA WRITE A LETTER ]

b. c. MOD - T λAST −T [ WITHIN (AST-T, EV-T )] d.

Lets now compute the temporal interpretation that the phrase structure in (34a) yields. The modal in (34) contributes a reference-time to the temporal interpretation of the clause. The initial bound of the MOD - T is set to UT- T via deictic anchoring. The MOD - T thus denotes an interval starting at UT- T and extending indefinitely into the future, as illustrated in (34b). Since there is no temporal head to order the MOD T relative to the AST- T immediately under its scope, the ordering relation between these two times is established via binding, as in (34c). The head ASP◦ is not null, but headed by the predicate of spatiotemporal ordering WITHIN (progressive aspect). The latter orders the AST- T within the EV- T , as in


97

(34d). The AST-T is thus constrained to denote a subinterval of the EV-T —that is, a time at which the event of writing unfolds/is ongoing that excludes both the initial and the final bounds of the event. λ -abstraction over the temporal variable in the specifier of ASP - P (AST- T) in (34a/c) creates a predicate which takes the MOD - T as external argument. (34c) requires that the MOD - T (itself an open-ended interval starting at UT- T) have the property of being a time at which the AST- T is true. Since the AST- T itself denotes a subinterval of the time of writing (see (34d)), (34c) requires that the MOD - T have the property of being a time at which a subinterval of the time of writing is true—that is, a time at which the event of writing is ongoing, unfolds. The MOD - T will satisfy this property if it contains either a subinterval of the AST- T as illustrated in (35a), or the whole AST- T as in (35b). (35)

a.

b.

c.

Non-shifted/ongoing construal in (33a):

d.

Future-shifted construal in (33b):

Now, since the AST- T denotes a subinterval of the EV- T that excludes the initial bound of the event (see (34d)), no assertion is made about the beginning of the writing event: it may well have started before the initial bound of the MOD - T as illustrated in (35c), yielding the non-shifted/ongoing construal of the modal complement in (33a), or after the initial bound of the MOD - T/UT- T as illustrated in (35b), yielding the future shifted construal of the modal complement in (33b). Since, moreover, the AST- T in (34, 35) does not contain the final bound of the writing event, no assertion is made about the culmination of the writing event. We thus correctly predict that the writing event may well culminate at a future time, or never culminate (e.g. Alaia may/might be writing a letter now/tomorrow but she will never finish it). The phrase structure in (34a) thus automatically yields the epistemic/present perspective construal of progressive modals in English: the possibility under discussion (that is, the MOD - T) holds at speech-time, while the EV- T of the propositional complement in the scope of the modal [ALAIA WRITE A LETTER] can be construed

98


as either non-shifted/ongoing (33a)/(35c) or as future shifted (33b)/(35d) relative to UT- T. 5.2.2 Epistemic Perfect Modals We now show how our temporal syntax for non-root modals derives the epistemic construal of perfect modals (see discussion of (3a) and (15, 16) above) illustrated below. (36)

Present temporal perspective construal a.

(It is possible/it follows from what is known at UT- T that) Amina may/might have (already) travelled to Europe.

b. The perfect modal in (36a) will be assigned the temporal phrase structure in (37a). Note that the head corresponding to ASP◦ is neither empty (as was the case with (32)), nor occupied by the spatiotemporal predicate WITHIN (as was the case with (34)). Rather ASP◦ in (37a) contains the spatiotemporal predicate AFTER (since (36a) contains perfect viewpoint aspect). (37)

a.

TP UT- Ti

T’ To

M MOD - T

[ti , ∞)

M’

Mo

λAST −T

ASP - P

AST- T

ASP ’

ASPo

ASP - P AFTER VP EV- T [ ALAIA TRAVEL TO EUROPE ]

The modal in (37a) contributes a reference time to the temporal interpretation of its clause. Once again, the initial bound of the modal-time is set to UT- T via deictic anchoring. The MOD - T is thus an open-ended interval starting at UT- T: (37)

b.

Since there is no temporal head to order the MOD - T relative to the AST- T immediately under its scope, the ordering relation between these two times is established via binding: the MOD - T binds the AST- T, as in (37a/c). Perfect aspect (that is, the


99

predicate of spatiotemporal ordering AFTER) orders in turn the AST- T after the EVT . The MOD - T thus binds an interval (the AST- T ) that is constrained to follow the EV- T . (37)

c.

MOD - T λAST -T [ AFTER (AST-T, EV-T), TRAVEL TO EUROPE ( AMINA ) (EV-T )]

λ -abstraction over the temporal variable in the specifier of ASP - P (AST- T) in (37a/c) creates a predicate which takes the MOD - T as external argument. The MOD - T is an open-ended interval starting at UT- T: [UT- T, ∞). Since (i) the modal-time binds the AST- T, and (ii) the AST- T has itself been ordered by perfect aspect after the EV- T , (37c) requires that the MOD - T have the property of being an interval itself subsequent to the EV- T, as illustrated in (37d). (37)

d.

The modal-time in (37) thus picks out an interval, [ UT- T, ∞), starting at speechtime and extending without limit into the future. This time is required to have the property of being a time that is subsequent to Amina’s traveling to Europe. The phrase structure in (37a) thus automatically yields the epistemic/present perspective construal of perfect modals in English: the possibility under discussion (that is, the MOD - T) holds at speech-time, while the EV- T of the propositional complement in the scope of the modal [AMINA TRAVEL TO EUROPE ] is past-shifted relative to the present modal-time. 5.2.3 Metaphysical Perfect Modals14 We now turn to the metaphysical construal of perfect modals (see discussion of (3b) and (17, 18) above) illustrated below. Recall that on this construal, the MOD - T is past-shifted relative to UT- T and the situation-time of the modal complement itself shifts into the future relative to this past time, as illustrated in (38b). (38)

Past perspective modal-time, metaphysical construal a.

(At some point in the past) Amina might (still) have won the game (but she didn’t).

b. 14

We have assumed the following scopal hierarchy of functional projections for non-root modals: < MP < ASP-P (see section 4). Stowell (2004) argues, however, that metaphysical modals more closely resemble root than non-root modals, in which case the scopal hierarchy MP < ASP-P could be reversed with the modal projection generated below ASP-P and above VP (see also Butler, 2004, 2005), thus yielding the scopal hierarchy of functional projections TP < ASP-P < MP. This proposal, which we do not consider here for reasons of space, is fully compatible with our analysis of metaphysical perfect modals (See D&UE in press for discussion). TP

100

(39)


a.

PRESENT

< PERFECT < MODAL → past perspective, metaphysical

construal b.

PRESENT

< MODAL < PERFECT → present perspective, epistemic

construal Recall further that Condoravdi correlates the past perspective, metaphysical construal in (38) with the scopal hierarchy in (39a), and the present perspective, epistemic construal (discussed in (36, 37) above) with the hierarchy in (39b). Note, in particular, that under Condoravdi’s analysis, both readings involve a present tense combining with perfect aspect. In contrast, Stowell (2004) argues that the past perspective, metaphysical reading correlates with a semantic past tense construal of the nonfinite perfect scoping above the modal, while the present perspective, epistemic reading correlates with a perfect construal of the modal complement in the scope of a semantically present tense modal. We suggest that the morphosyntax of these readings in Spanish and French provides support for Stowell’s correlation. Recall that the Spanish past modal verbs in (2b, c) above (see also (40a) below) allow both the present perspective/epistemic construal illustrated in (3a)/(36) and the past perspective/metaphysical construal in (3b)/(38). In contrast, the Spanish present perfect in (40b), just like the passé composé (40c), only allows an epistemic reading. (The paradigm in (40) is taken from Borgonovo and Cummins, 2005; henceforth B&C.15 ) (40)

a.

Pedro debió ganar la carrera. Peter MODAL.PRETERIT win the race ‘Peter may have won the race.’ → Epistemic ‘Peter should have won the race.’ → Metaphysical /Counterfactual

b. c.

Pedro ha debido ganar la carrera. Pierre a dû gagner la course. Peter has MODAL . PAST. PART win the race ‘Peter may have won the race.’ → Epistemic

15 There seems to be some dialectal variation with regard to the possibility of interpreting (40b) counterfactually. This reading is available for speakers of peninsular Spanish (see Laca, 2006 for discussion). We leave this question for further research. Note also that a preterit modal combining with a perfect infinitive, as in (i) below, allows either an epistemic or a counterfactual construal. (i), however, is semantically not a past perfect (since there is only relation of anteriority in its meaning). Under the analysis developed below in the text, (i) would involve only one occurrence of the spatiotemporal predicate AFTER at LF. If AFTER occupies T◦ at LF, a metaphysical reading arises (see the derivation of this reading for Spanish past modals given in (49) below in the text). If AFTER occupies ASP◦ , an epistemic reading arises (see the derivation of this reading given in (48) below). For discussion, see Bosque (1999), Bosque and Torrego (1995), Laca (2006) and references therein.

(i)

Debiste

haberlo

matado. killed ‘You must/should have killed him.’ MODAL - PRETERIT.2P have-him


101

Since the past perspective construal in (38) is unavailable with a present perfect, we conclude that this construal does not involve a true present perfect (that is, a semantic present tense combining with perfect aspect). Rather, it is licensed by either a past tense as is the case with the Spanish examples in (40a)/(2b, c), or by a non-finite perfect as the case with the English example in (17)/(38a) semantically interpreted as a past tense. The question then is how or why non-finite have can yield the construal of a simple past tense. Now, under the proposal developed here, there is no outer semantic present tense embedding the English non-root modal (as Condoravdi assumes). That is, the TP projection embedding the modal is not headed by present tense, be it morphosyntactically or semantically: there is no (covert) spatiotemporal predicate (WITHIN) heading TP. On our proposal, the head T◦ is empty. We now argue that, when T◦ is empty, perfect aspect can yield the construal of a past tense via head movement to T ◦ , as we shall see with the derivation in (41b). Recall that within the framework developed here, perfect aspect and past tense instantiate the spatiotemporal predicate AFTER. AFTER ASPECT orders the AST- T after the EV- T, while AFTER TENSE orders the UT- T after the AST- T. With this in mind, consider (41). (41)

The ambiguity of the non-finite perfect a. Present perfect construal

b.

Past tense construal

TP UT- Ti

TP T’

To

UT- T ASP - P

AST- Ti

ASP ’

ASPo VP AFTER EV- T

a’.

T’ To

ASP - P AST- Ti

ASP ’

ASPo VP AFTER EV- Ti

VP

b’.

In (41a), perfect is an aspect ordering the AST-T after the EV-T. Since there is zero tense under T◦ , the ordering relation between the UT-T and the AST-T is established via covaluation ( UT-T = AST-T).16 The temporal phrase structure in (41a) thus yields the present perfect construal in (41a ). In (41b), perfect is generated under ASP◦ and subsequently raises at LF (via head movement) to T◦ and the lower copy is deleted. The predicate AFTER is thus interpreted semantically in its landing site as a tense (ordering the UT-T after the AST-T). Since there is (no longer) an ordering predicate under ASP◦ , the ordering relation between the AST-T and the EV-T is established via covaluation, yielding perfective viewpoint (AST-T = EV-T). Head movement of ASP◦ to T◦ in (41b) thus, ultimately, yields the construal of a simple past tense, as illustrated in (41b). 16

See footnote 4.

VP

102


Summarizing. In non-finite perfect modals, there is no (covert) semantic present tense. The hypothesis defended in D&UE that perfect aspect, just like past tense, is the spatiotemporal predicate AFTER predicts that a non-finite perfect can yield either a (present) perfect construal if the perfect head remains in situ under ASP◦ , or a preterit past construal if it raises to (null) T◦ , at LF.17 We now argue that head movement of PERF ◦ /AFTER to T◦ automatically yields a past perspective metaphysical construal because it triggers past-shifting the modaltime relative to UT-T. To understand why, consider the temporal phrase structure assigned to (38a), after raising of PERF◦ to T◦ . (42) Past perspective, metaphysical construal: Amina might have won the game a. b. TP UT- T

*TP T’

To MP AFTER MOD - T

UT- Ti

T’ To

MP AFTER MOD - T

M’

Mo ASP - P AFTER AST- T ASP ’

[ti , ∞)

Mo

M’ ASP - P

ASP ’ AST- T ASPo VP o ASP VP AFTER EV- T VP EV- T VP AMINA WIN THE GAME

The perfect head AFTER, in (42a), has undergone head movement to T◦ at LF, and is thus interpreted in its landing site—that is, as a simple past tense ordering the UT-T after the EV-T. The only available controller for the initial bound of the MOD-T is the UT-T. The initial bound of the MOD-T is thus set to UT-T (deictic anchoring), as shown in (42b). The phrase structure in (42b) yields, however, an uninterpretable, nonsensical temporal output. That is, T◦ orders its external argument ( UT-T) after its internal argument (MOD-T). Now, since the MOD-T is an open-ended interval extending indefinitely into the future ([UT-T, ∞)), no time can ever be ordered after it. In other words, unless nothing else happens, the derivation crashes. There is, however, one way of rescuing the derivation—namely, scoping out the (closest) time argument, that is, the AST-T, to a position where it binds the MOD-T. Note that within a model where time arguments are projected into the syntax as Zeit-Ps, we expect them to undergo, just as any regular DP/QP, phrasal movement (QR) to higher scope positions, thus reversing initial scope relations. Scoping out the AST-T yields the phrase structure in (43a) where the AST-T has been adjoined to MP. 17 See Hoffman (1976) who first discussed the ambiguity of non-finite perfect have between a present perfect and a simple past tense construal.


(43)

103

a.

b. TP

TP UT- T

UT- Ti

T’ To

MP

AFTER AST- Ti

MOD - T Mo

MP

AFTER AST- Ti

MP

[ti , ∞)

T’ To

MP

MOD - T

M’

[ti , ∞)

ASP - P AST- T

Mo

λEV −T

ASP ’

ASPo

M’

ASP - P ASP ’

VP EV- T

VP AMINA WIN THE GAME

ASPo

VP EV- T

VP

T◦

in (43a) now orders the time argument in its specifier, UT-T, after the time argument in its immediate scope, that is, after the AST-T, as shown in (43c). Moreover, the AST-T is now the closest available controller (c-commanding time argument) for the initial bound of the MOD-T. The initial subinterval of the MOD-T is thus bound by the AST-T, itself a past time since it has been ordered by tense prior to UT-T (43c). The MOD-T thus shifts into the past: it denotes an interval [AST-T, ∞) starting at some past interval and extending without limit into the future, as illustrated in (43d). (43)

c. d. e.

MOD - T

λEV −T [WIN (THE GAME) (AMINA) (EV-T )]

At this stage, the UT-T, AST-T and MOD-T have all been order relative to each other. The MOD-T remains, however, unordered relative to the subordinate EV-T. Since there is no temporal head to establish this ordering relation, it is established via binding. λ -abstraction over the temporal variable inside the VP (that is, the EV-T) yields a predicate which takes the MOD-T as external argument, as in (43e). The MOD-T thus picks out an interval [AST-T, ∞) starting at some past time and extending into the future. Binding constrains this interval to be a time at which Amina wins the game. The assumption that AFTER ASPECT raises to null T◦ thus automatically derives the past perspective metaphysical construal of English modal verbs combining with the perfect auxiliary Have. The default assumption is that raising of PERF◦ to T◦ should be free. Raising to T◦ , however, yields a derivation which can only be rescued by reversing the scope of the AST-T relative to MOD-T. Scope reversal, in turn, automatically entails, first, past-shifting the MOD-T relative to the UT-T and, then,

104


future-shifting the EV-T of the modal complement relative to the (initial past bound of the) MOD-T—yielding a perfect future. Note that we do not assume that QR (scope reversal) of the AST-T is free. Following Fox (2000), QR targets the closest licit landing site that would yield a well-formed temporal output that could not be generated otherwise. This proposal can be extended to head movement. Head movement will be subject to the Fox-type economy constraint: it targets the closest landing site that would yield a well-formed temporal output that could not be generated otherwise. Both these conditions are satisfied in the derivation in (42, 43): T◦ is a licit landing site for perfect ASP◦ (since T◦ , just like ASP◦ , can be headed by a spatiotemporal predicate), and head-movement to T◦ ultimately yields a construal (the past perspective construal) which could not have been generated otherwise. In contrast, under the epistemic construal illustrated in (36), the perfect remains in situ under ASP◦ , the MOD-T is deictically anchored to UT-T, and the EV-T pastshifted relative to the (initial present bound of the) MOD-T—yielding a present perfect construal of the EV-T (see the temporal derivation in (37), section 5.2.2). Our analysis of the epistemic vs. metaphysical construal thus captures Stowell’s generalization that the present perspective, epistemic construal in (36) correlates with a nonfinite perfect interpretation of the modal complement in the scope of a semantically present tense modal, whereas the past perspective construal in (38) correlates with a semantic past tense interpretation of the nonfinite perfect scoping above the modal.

6 The Temporal Interpretation of Spanish Non-Root Modals We will now show how our proposal derives the asymmetries between English and Spanish non-root modals discussed in section 1. Recall from our discussion that Spanish modals differ from English modals in (at least) two respects: (i) Spanish modal verbs are fully inflected for tense and aspect; (ii) past inflectional features on non-root modals do not yield the same construals in Spanish and English. In the following sections, we develop a unified analysis of English vs. Spanish non-root modals, which seeks to account for these asymmetries in a principled manner.

6.1 Modals Inflected for the Past Compare the Spanish example in (44a) with its English counterpart in (44b). (44)

a. b.

Oihana debióPAST /pudoPAST estar dormida/ganar la carrera. Oihana should/could be asleep/win the race.

Although (44a) and (44b) look similar from a morphological point of view (debió/pudo and should/could bear morphological past inflection), they crucially


105

differ in their temporal construals. Whereas the Spanish (44a) allows the construals in (45a, b), the English (44b) only allows the construals in (45c, d). Conversely, the construals in (45c, d) (discussed in section 5.1) are unavailable for the Spanish simple past modals in (44a). (45)

Spanish past modals (without auxiliary HABER PERFECT): a.

Past-shifted SIT- T relative to present MOD-T (epistemic)

b.

Future-shifted SIT- T relative to past MOD-T (metaphysical)

English past modals (without auxiliary HAVE PERFECT): c.

Present/non-shifted SIT- T relative to present MOD-T (epistemic)

d.

Future-shifted SIT- T relative to present MOD-T (epistemic)

The construals in (45a, b) are only possible in English when the modal combines with non finite HAVE PERFECT, as in (46). (46)

Oihana should/could have been asleep/have won the race

We conclude that the temporal syntax of the Spanish non-root modals in (44a) must involve a predicate with the meaning of AFTER responsible for either past-shifting the SIT- T of the modal complement relative to the present MOD-T (MOD-T/UT-T AFTER SIT- T ) yielding the epistemic construal in (45a), or for past-shifting the MOD T yielding the metaphysical construal in (45b). The question arises as to whether this predicate is past tense or perfect aspect. To answer this question, consider the examples in (47) adapted from B&C, which allow for an epistemic interpretation. (47)

a. b. b . b .

Urko deb´ıaIMP.PAST /pod´ıaIMP.PAST estar en casa. ‘Urko must/could have been at home.’ Alaia deb´ıaIMP.PAST trabajar/abrir la puerta. ‘Alaia must have been working/opening the door.’ ‘Alaia must have worked/opened the door (habitually).’

Now, although tense and aspect are morphologically realized on the modal verb in (47), they are both interpreted on the complement of the modal: the SIT- T of the modal complement is past-shifted relative to the present MOD-T and has an imperfective viewpoint, as reflected by the English translations. As B&C point out,

106


the fact that past inflectional features are interpreted on the modal complement18 is particularly clear in the case of eventive predicates such as (47b), where the imperfect yields its two characteristic readings: the durative/ongoing construal in (47b) and the habitual construal in (47b). Since the viewpoint in (47) is imperfective, the spatiotemporal predicate AFTER responsible for past-shifting the SIT- T of the modal complement cannot be perfect aspect. We conclude that the predicate of spatiotemporal ordering AFTER in Spanish past modals must be past tense. In sum, Spanish past inflected modals differ from English past modals (without auxiliary HABER / HAVE PERFECT) in that the past morphology on the modal is not semantically vacuous: it contributes to the temporal interpretation of the sentence a predicate of spatiotemporal generated under T◦ : AFTER TENSE. We will now show how this assumption ultimately derives both the epistemic and the metaphysical construal of (44a). Lets start with the derivation of the epistemic construal. (44a) is assigned the phrase structure in (48a) where T◦ is headed by the spatiotemporal predicate AFTER. The initial bound of the MOD-T is set to UT-T (deictic anchoring). (48)

Deriving the epistemic construal of Spanish past modals Oihana debióPAST /pudoPAST ganar la carrera. a.

b.

*TP UT- Ti

TP T’

UT- Ti

To MP AFTER MOD - T

[ti , ∞)

Mo

T’

To MP AFTER MOD - T

M’

[ti , ∞)

ASP - P AST- T

Mo ASP - P AFTER ASP ’ AST- T

ASP ’

ASPo

VP EV- T

M’

ASPo VP OIHANA GANAR LA CARRERA

VP EV- T

VP

Now, the phrase structure in (48a) yields an uninterpretable temporal output. That is, T◦ orders the UT-T after its internal argument, the MOD-T. Since the MOD-T is an open ended interval extending indefinitely into the future, [ UT-T, ∞), no time can ever be ordered after it. Consequently, unless nothing else happens, the derivation will crash. 18 Notice that the verb conocer, which can signify ambiguously either meet or know, can only signify meet (and not know) in (i) below. B&C take this fact as further evidence that aspect (in this case, perfective) is interpreted on the complement of the modal.

(i)

Pedro debió conocer a Marta en la fiesta Pedro mustPERFECTIVE PAST know Marta at the party ‘Pedro must have met Marta at the party’


107

The derivation in (48a) can, however, be rescued by lowering at LF the spatiotemporal predicate AFTER under T◦ to the closest head, M◦ , as shown in (48b). This movement yields the temporally interpretable output illustrated in (48c). The MOD - T is deictically anchored to UT - T and, thus, denotes an interval starting at UT T and extending indefinitely into the future. The spatiotemporal predicate AFTER under M◦ orders its external argument, the modal-time [UT-T, ∞), after its internal argument, the AST-T: (48)

c.

The derivation in (48b) thus yields the epistemic construal of (44a), where the present MOD-T is ordered AFTER the AST-T. There is, however, an alternative way of rescuing the derivation in (48a) (repeated as (49a)—namely, scoping out the AST-T, to a position where it binds the MOD-T. Scoping out yields the phrase structure in (49b) where the AST-T has been adjoined to MP. (48)

Deriving the metaphysical construal of Spanish past modals Oihana debióPAST /pudoPAST ganar la carrera. a.

b. TP

*TP UT- Ti

UT- Ti

T’

To MP AFTER MOD - T

[ti , ∞)

To MP AFTER AST- Ti

M’

AST- T

[ti , ∞)

Mo

ASP ’

ASPo

VP EV- T

MP

MOD - T

ASP - P

Mo

T’

M’ ASP - P AST- T

ASP ’

ASPo ASP - P VP OIHANA GANAR LA CARRERA EV- T VP

T◦ in (49b) now orders the time argument in its specifier, UT - T , after the time argument in its immediate scope, that is, after the AST-T. Moreover, the AST-T is now the closest available controller (c-commanding time argument) for the initial bound of the MOD-T. The initial subinterval of the MOD-T is thus bound by the AST-T, itself a past time since it is ordered by tense prior to UT-T. The MOD-T thus shifts into the past: it denotes an interval [AST-T, ∞) starting at some past interval and extending without limit into the future, as illustrated in (49c).

(49)

c. d.

MOD - T

λEV −T [WIN (THE RACE) (OIHANA) (EV-T )]

108


Scoping out of the AST-T thus yields the metaphysical construal of (44a) where the MOD-T is past-shifted relative to UT-T and consequently denotes an open-ended interval starting at a past time. The derivation in (49b) then proceeds exactly on a par with the derivation given for the metaphysical construal of English perfect modals in (43b, d). The UT-T, AST-T and MOD-T in (49b, c) have all been ordered relative to each other. The MOD-T remains, however, unordered relative to the subordinate EVT . Since there is no temporal head to establish this ordering relation, it is established via binding, as shown in (49d). λ -abstraction over the temporal variable inside the VP (that is, the EV-T) in (49d) yields a predicate which takes the MOD-T as external argument. The MOD-T thus picks out an interval [AST-T, ∞) starting at some past time and extending into the future. Binding constrains this interval to be a time at which Oihana wins the race. The EV-T of the modal complement is thus construed as future shifted relative to the initial past bound of the MOD-T. The derivation in (49b) yields the metaphysical construal of (44a), where the initial bound of the MOD-T is past shifted relative to UT-T, via scope reversal of the AST - T relative to the MOD - T . Scope reversal, in turn, entails future shifting the SIT- T of the modal complement relative to (the initial past bound of) the MOD-T. Recapitulating. We have argued that past morphology on non-root modals in Spanish is not semantically vacuous: it contributes the spatiotemporal predicate AFTER TENSE to the temporal computation of the sentence in which the modal occurs. Generating AFTER under T◦ , however, yields a temporally uninterpretable output (48a/49a) since the UT-T cannot be ordered after a time that extends indefinitely into the future. Unless nothing else happens, the derivation will crash. There are, however, two alternative ways of rescuing the derivation in (48a/49a)—namely, lowering the predicate AFTER under T◦ to the closest head, M ◦ as in (48b), or scoping out the AST-T to a position where it binds the MOD-T as in (49b). Lowering AFTER TENSE to M◦ yields an epistemic construal for Spanish past-inflected non-root modals, whereas scoping out of the AST-T over the MOD-T yields a metaphysical construal.19 We have proposed a uniform temporal phrase structure for root modals in Spanish and English. We have argued that syntactic movement at LF (be it X◦ movement of a temporal head, or XP movement of a Zeit-P) can alter temporal scope relations. We have illustrated two ‘mirror image’ instances of X◦ movement of a temporal head at LF: (i) raising of AFTER ASPECT in English (42, 43); (ii) lowering of AFTER TENSE in Spanish (48). In both cases movement is motivated and constrained by the same principles (in the spirit of Fox, 2000): (i) movement targets the closest licit landing site that would yield a well formed temporal output; (ii) movement is a last resort operation generating a temporal output that otherwise could not be generated. Under the proposal developed here, Spanish non-root modals (without auxiliary HABER AFTER ) differ from their English counterparts in one crucial respect: the past morphology on the modal is not semantically vacuous in Spanish. 19

For an account of the imperfective viewpoint of the past in (47) and, conversely, of the perfective viewpoint of the past in (44a), see D&UE (in preparation).


109

But what about present or future morphology on non-root modals in Spanish? We address this issue in the next subsections.

6.2 Modals Inflected for the Present Consider the examples in (50), with present morphology on the modal and auxiliary HABER PERFECT: (50)

a. b.

Ainhoa puede haber estado en casa Ainhoa may have been at home Itsaso puede haber le´ıdo el libro Itsaso may have read the book

(50a, b) will be assigned the temporal phrase structure given in (51). Since there is no spatiotemporal predicate heading MP, the ordering relation between the MOD-T and the AST-T is established by binding: the MOD-T binds the AST-T. (51)

Present morphology on the modal TP UT- T

T’ To

MP WITHIN MOD - T

[t, ∞)

M’

Mo

λAST −T

ASP - P

AST- T

ASP ’

ASPo AFTER EV- T

VP VP

[ AINHOA AT HOME ]/ [ ITSASO READ THE BOOK ]

In (51), T◦ orders the UT-T within the MOD-T ([t, ∞)), as illustrated in (52a). ASP ◦AFTER orders the AST- T after the EV- T , as shown in (52b). There is, however, no temporal head under M ◦ to order the MOD - T relative to the AST- T. The ordering between these two times is thus achieved via binding: the MOD - T binds the AST- T (52c). Since the MOD - T binds the AST- T, and the AST- T is itself ordered by PER FECT aspect after the EV - T , (52c) requires that the MOD - T have the property of being an interval that is itself subsequent to the EV-T, as depicted in the temporal configuration in (52d).

110

(52)


a.

b. c.

MOD - T

λAST -T [AFTER (AST-T, EV-T) & AT HOME (AINHOA) (EV-T)]

d. This analysis straightforwardly derives the epistemic construal of the Spanish examples in (50): the EV- T is past shifted relative to the MOD-T, itself ongoing at UT- T .20

6.3 Modals Inflected for the Future Finally, let us consider the analysis of modal sentences where the modal verb is inflected for the future. Recall that future tense is the spatiotemporal predicate BEFORE . The temporal structure of future modals is given in (53). (53)

Future morphology on the modal a. TP UT- T

T’ To

MP BEFORE MOD - T

[t, ∞)

M’

Mo

λAST −T

ASP - P

AST- T

ASP ’

ASPo

VP EV- T

VP

b. In (53), T◦ orders the UT- T before the MOD-T, as illustrated in (53b). The temporal output in (53b), where the UT-T precedes the MOD-T is well formed but incompatible with an epistemic modal base. Consequently, future morphology on the modal cannot give rise to an epistemic construal (see also Laca, 2006 for discussion). 20 Our analysis of English present/future oriented modals and of ‘progressive’ modals, developed in sections 5.1 and 5.2.1 respectively, can be directly extended to their corresponding Spanish counterparts with present tense morphology. Due to space limitation, we do not flesh out these analyses here.


111

7 Conclusions We have derived the asymmetries between non-root modals (without auxiliary HABER/ HAVE AFTER ) in Spanish and English from a single assumption: in Spanish, unlike in English, past morphology on the modal is not semantically vacuous. This assumption ultimately explains why Spanish past inflected modals allow either a past oriented epistemic construal (past shifted SIT- T relative to the present MOD - T ) or a future oriented metaphysical construal (future shifted SIT- T relative to a past-shifted MOD - T), whereas English past inflected modals only allow a future/present oriented epistemic construal (future/non-shifted SIT- T relative to the present MOD - T). We have proposed a uniform temporal syntax for non-root modals—be it in Spanish or English. The heads, T◦ , M◦ , ASP◦ , V ◦ , each introduce a time argument/REF - T projected onto a specifier in the syntax. The REF - T contributed by M◦ (MOD - T) is an open-ended interval, [t, ∞), following Condoravdi (2002). Since there is no temporal head under M◦ , the ordering relation between the external argument of the modal (MOD - T) and the time argument immediately under its scope ( AST- T), is established via anaphora, where anaphora is binding. We have argued that the null assumption is that time-denoting DPs/Zeit-Ps can, just as regular DPs/QPs, undergo phrasal movement (QR) to higher scope positions. We have shown that syntactic movement at LF—be it XP movement of a Zeit-P or X◦ movement of a temporal head—can reverse initial temporal scope relations. Extending Fox (2002), we assume that X◦ movement of a temporal head is subject to the same constraints as XP movement of a time argument: it targets the closest licit landing site that would yield a well-formed temporal output; and is a last resort operation generating a temporal output that could not otherwise be generated. The proposal that Perfect aspect, just like Past tense, is the spatiotemporal predicate AFTER allows us to uniformly derive temporal construals of non-root modals via scope reversal effects created by syntactic XP/X◦ movement. In particular, we have illustrated two ‘mirror image’ instances of X◦ movement of the temporal head AFTER at LF: (i) raising of AFTER ASPECT to T◦ in English, (ii) lowering of AFTER TENSE in Spanish. Lowering of the temporal head AFTER in Spanish yields the epistemic construal of simple past modals (48), whereas raising in English yields the metaphysical construal of zero-tense perfect modals (42, 43). The metaphysical construal, be it in English or Spanish, involves scope reversal of the AST- T relative to the MOD - T. Scope reversal, however, is not stipulated. It is automatically enforced in order to yield a well-formed temporal output, whenever the temporal head AFTER is either generated under T◦ , as is the case with Spanish tensed modals (49), or raised to T◦ at LF as is the case with English tenseless modals (42, 43).

112


References Abusch, D., 1997. Sequence of tense and temporal ‘de re’. Linguistics and Philosophy 20, 1–50. Borgonovo, C., Cummins, S., 2005. Tensed modals. Unpublished manuscript, Laval University, Québec. Bosque, I., 1999. ¿Qué sabe el que sabe hacer algo? ‘Saber’ entre los predicados modales. In: Korta, K., Garc´ıa Murga, F., (Eds.), Palabras. Servicio editorial de la UPV-EHU, Bilbao, pp. 303–324. Bosque, I., Torrego, E., 1995. On Spanish ‘haber’ and tense. Proceedings of Langue et Grammaire I. Presses Universitaires de Vincennes, Paris, pp. 13–29. Butler, J., 2004. Phase Structure, Phrase structure and Quantification. Ph.D. dissertation, University of York, York, Great Britain. Butler, J., 2005. Non-finite temporality and modality. Talk given at Chronos VI, University of Geneva, Geneva. Condoravdi, C., 2002. Temporal interpretation of modals. Modals for the present and for the past. In: Beaver, D., Casillas, L., Kaufmann, S., Clark, B., (Eds.), The Construction of Meaning. CSLI Publications, Stanford, CA, pp. 59–87. Demirdache, H., Uribe-Etxebarria, M., 1997. The syntax of temporal relations: a uniform approach to tense and aspect. In: Curtis, E., Lyle, J., Webster, G. (Eds.), Proceedings of the Sixteenth West Coast Conference on Formal Linguistics. CSLI Publications, Stanford, CA, pp. 145–159. Demirdache, H., Uribe-Etxebarria, M., 2000. The primitives of temporal relations. In: Martin, R., Michaels, D., Uriagereka, J. (Eds.) Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. MIT Press, Cambridge, MA, pp. 157–186. Demirdache, H., Uribe-Etxebarria, M., 2002. Temps, aspects et adverbes. In: Laca, B. (Ed.), Temps et Aspect. De la morphologie a` l’interprétation. Collection Sciences du Langage, Presses Universitaires de Vincennes, Paris, pp. 125–175. Demirdache, H., Uribe-Etxebarria, M., 2004. The syntax of time adverbs. In: Guéron, J., Lecarme, J. (Eds.), The Syntax of Tense and Aspect. MIT Press, Cambridge, MA, pp. 143–179. Demirdache, H., Uribe-Etxebarria, M., 2005. Aspect and temporal modification. In: Kempchinsky, P., Slabakova, R. (Eds.), The Syntax, Semantics and Acquisition of Aspect. Kluwer Academic, Dordrecht/Boston/London, pp. 191–221. Demirdache, H., Uribe-Etxebarria, M., 2007. The syntax of time arguments. In: Zagona, K. (Ed.), Special Issue on Tense. Lingua 117, 330–366. Demirdache, H., Uribe-Etxebarria, M., in press. Scope and anaphora with time arguments: the case of ‘perfect modals’. In: Harley, H. and Folli, R. (Eds.), Special Issue of Lingua. Demirdache, H., Uribe-Etxebarria, M., in preparation. Temporal anaphora and viewpoint aspect. Ms. University of Nantes and University of the Basque Country. Dowty, D. R. 1986. The effects of aspectual class on the temporal structure of discourse: Semantics or pragmatics? Linguistics and Philosophy 9:37–61.


113

Drubig, H. B., 2001. On the syntactic form of epistemic modality. Ms. Universidad de Túbingen. Fox, D., 2000. Economy and Semantic Interpretation. MIT Press, Cambridge, MA. Garc´ıa, L. (Ed.)., 2006. Diccionario de Per´ıfrasis Verbales. Gredos, Madrid. Gómez Torrego, L., 1988. Per´ıfrasis verbales. Arco/Libros, Madrid. Gómez Torrego, L., 1999. Los verbos auxiliares. Las per´ıfrasis verbales de infinitivo. In: Bosque, I. y Demonte, V. (Eds.). Gramática Descriptiva de la Lengua Española. Espasa Calpe, Madrid. Hofmann, T. R., 1976. Past tense replacement and the modal system. In: McCawley, J. (Ed.), Notes from the Linguistic Underground. Syntax and Semantics 7. Academic, New York, pp. 85–100. Kamp, H., Reyle, U., 1993. From Discourse to Logic. Kluwer Academic, Dordrecht/Boston/London. Klein, W., 1995. A time relational analysis of Russian aspect. Language Vol. 71, No. 4, pp. 669–695. Laca, B., 2005. Périphrases aspectuelles et temps grammatical dans les langues romanes. In: Bat-Zeev Schyldkrot, H., Le Querler, N. (Eds.) Les Periphrases Verbales. Lingvisticae Investigationes Supplementa, Benjamins, Amsterdam/ Philadelphia, pp. 47–66. Laca, B., 2006. On the interaction of modal verbs with tense and aspect in Spanish. Ms. U. Paris VIII. Mondadori, F., 1978. Remarks on tense and mood: the perfect future. In: Guenthner, F., Rohrer, C. (Eds.), Studies in Formal Semantics: Intentionality, Temporality, Negation. North-Holland, Amsterdam, pp. 223–248. Reinhart, T., 1997. Strategies of Anaphora Resolution. OTS Working Papers. TL97007. Smith, C., 1991. The Parameter of Aspect. Kluwer Academic, Dordrecht/ Boston/London. Stowell, T., 1995. The Phrase Structure of Tense. In: Zaring, L., Rooryck, J. (Eds.), Phrase Structure and the Lexicon. Kluwer Academic, Dordrecht/Boston/London. pp. 381–396. Stowell, T., 2004. Tense and modals. In: Guéron, J., Lecarme, J. (Eds.), The Syntax of Tense and Aspect. MIT Press, Cambridge, MA, pp. 621–636. Stowell, T., 2005. Sequence of Perfect. Talk given at Chronos VI, University of Geneva, Geneva. Werner, T., 2003. Deducing the future and distinguishing the past: temporal interpretation in modal sentences in English. Ph.D. dissertation, The State University of New Jersey. Zagona, K., 1990. Times as temporal argument structure. Talk given at the Time in Language Conference, MIT, Cambridge, MA. Zagona, K., 2003. Tenses and anaphora. Is there a tense specific theory of coreference? In: Barss, A. (Ed.), Anaphora. A Reference Guide. Blackwell, Oxford, pp. 140–171.

How to Say Ought in Foreign: The Composition of Weak Necessity Modals Kai von Fintel and Sabine Iatridou

Abstract In this article1 , we draw attention to the fact that what English expresses by the use of the weak necessity modal ought, many other languages express by Massachusetts Institute of Technology 1 This paper has been presented at the workshop “Time and Modality: A Round Table on Tense, Mood, and Modality”, Paris, December 2005, at a CUNY linguistics colloquium in May 2006, and at the 6th Workshop on Formal Linguistics in Florianópolis, Brazil, August 2006. We thank the audiences at those presentations, in particular Orin Percus, Tim Stowell, Marcel den Dikken, Anna Szabolcsi, Chris Warnasch, Roberta Pires de Oliveira, Renato Miguel Basso, and Ana Müller. We thank Noam Chomsky, Cleo Condoravdi, and Irene Heim for very helpful conversations about this material. We thank Bridget Copley for sharing with us her recent manuscript “What Should Should


115

116

K. von Fintel and S. Iatridou

combining a strong necessity modal with the morphology that appears in the consequent of a counterfactual conditional. On the hypothesis that there should be a compositional form-to-meaning mapping, we explore the semantics of weak necessity modals and propose how to derive this semantics from the combination of a strong necessity modal and counterfactual marking. Specifically, building on the semantics for weak necessity modals proposed by Sloman, we propose that weak necessity modals are the result of the promotion of a secondary ordering source of a strong necessity modal. This meta-linguistic operation is signaled or effected by counterfactual marking. The fact that it is a strong necessity modal that is counterfactually marked crosslinguistically, shows that even with weak necessity modals the quantificational force is universal. Key words: Modals, epistemic, deontic, goal-oriented, counterfactuals, wishes, ordering source Of all the differences between man and the lower animals, the moral sense or conscience is by far the most important. . . . [I]t is summed up in that short, but imperious word ought, so full of high significance. Charles Darwin, Descent of Man

1 A Basic Contrast Consider the following sign, posted at a summer camp on Cape Cod: (1)

After using the bathroom, everybody ought to wash their hands; employees have to.

From (1), we see that there is a distinction to be made between ought on the one hand and have to on the other. How can ought and have to contrast like this? What distinguishes them? Or for that matter, what distinguishes ought from must, which patterns like have to, as (2) shows? (2)

Everybody ought to wash their hands; employees must.

Mean” (Copley, 2006), which was written independently of our work on ought. We have already learned a lot from her paper and we plan to address some of her observations and proposals in a future version of this paper. We are also grateful to all of our informants, who are individually acknowledged in the text. Thanks to Noam Chomsky for suggesting the epigraph and thanks to Larry Horn for drawing our attention to the New Yorker cartoon*. Any comments are welcome. Mistakes are each other’s. *Cartoon from the New Yorker of July 31, 2006, printed with permission of the New Yorker.

How to Say Ought in Foreign: The Composition of Weak Necessity Modals

117

An intuition that many researchers have tried to capture is that ought (as well as its near equivalent should, about which we remain officially agnostic for the purposes of this paper2) is somehow weaker than have to/must. Some evidence for this relative weakness comes from the fact that (3) is not a contradiction while the examples in (4) are3 : (3)

You ought to do the dishes but you don’t have to.4

(4)

a. b.

#You have to do the dishes but you don’t have to. #You must do the dishes but you don’t have to.

A second piece of evidence for the relative weakness of ought comes from sequences like these:5 (5)

a. You ought to wash your hands—in fact, you have to. b. ??You have to wash your hands—in fact, you ought to.

Because of the apparent difference in weakness between must/have to and ought, ought is often referred to as a “weak necessity modal” (as opposed to “strong” necessity modals like must or have to). This paper is an investigation into ought and the cross-linguistic expressions of this modal concept. Before we turn to the cross-linguistic facts, we will review some possible semantic analyses of weak necessity modals.

2 Weakness As is customary in most linguistic work on modality, we adopt the basic framework proposed by Kratzer (1981, 1991). Modals quantify over a set of worlds that is calculated from a modal base of accessible worlds and an ordering source which ranks the worlds in the modal base. Different flavors of modality (epistemic, goaloriented, deontic, etc.) come from the interplay and contextual resolution of modal base and ordering source. Let us call the worlds in the modal base that are most highly ranked by the ordering source the favored worlds. Let us also introduce the term prejacent, first used by our medieval colleagues, to designate the proposition embedded under the modal.

2

Should shows considerable similarities to ought but also some differences:

(i)

It’s strange that he should/∗ ought to do that.

3 We are of course not the first ones to observe data like these. See for example, Wertheimer (1972: Chapter 3, “The Meaning of the Modals”), Jones and Pörn (1986), McNamara (1996, 2006). 4 The use of have to instead of must is required in this context because only have to scopes under negation. 5 We take this kind of example from Copley (2006).

118


How can the difference between strong and weak necessity modals be captured in this framework? One straightforward idea, inspired by Horn,6 is that while strong necessity modals require the prejacent to be true in all of the favored worlds, weak necessity modals say that the prejacent is true in most of the favored worlds. We will not pursue this idea, for a couple of reasons.7 First, we have some qualms about being able to “count” possible worlds in such a way as to make sense of saying that most of the worlds in a particular set have a certain property. More importantly, we don’t think that the “most” analysis truly captures the meaning of weak necessity modals. We think that a sentence like You ought to do the dishes means not that among the favored worlds, most are worlds where you do the dishes. Rather, it means that among the favored worlds, all the very best ones are worlds where you do the dishes. That is, the ought-claim makes a further distinction as to how good particular worlds among the favored world are. So, the central idea we would want to capture in a semantics for ought is this: ought p says that among the favored worlds, p-worlds are better than non-p-worlds.8 That is the intuition we pursued in our paper on anankastic conditionals (von Fintel and Iatridou, 2005) for goal-oriented uses of strong and weak necessity modals. There, we were inspired by an early proposal by Sloman (1970), who wrote: For instance If you want to get to London by noon, then you ought to go by train picks out the best means without excluding the possibility of others, whereas If you want to get to London by noon then you have to (must, will be obliged to etc.) go by train implies that no other means exists. [p. 390f.] In other words, Sloman proposes that ought says what is best, or better than all alternatives. On the other hand, must picks out the only candidate. For example, (6) says that in all the worlds in which your goal of going to Ashfield is achieved, you have used Route 2: (6)

To go to Ashfield you have to/must use Route 2.

This means that there is no other way of satisfying your goal of going to Ashfield. On the other hand, when we use ought, what is conveyed is that there are several ways of going to Ashfield but that by some measure, Route 2 is the best: (7)

To go to Ashfield, you ought to use Route 2.

6 Copley (2006) attributes the idea to Horn. In Horn’s dissertation (Horn, 1972), weak necessity modals are characterized as occupying the same location on the scale of modal strength as most does on the scale of quantifiers. 7 See Copley for another argument against the “most” analysis. 8 We should note that Kratzer (1991) distinguishes between necessity and weak necessity as well. Her informal characterization is similar to ours: p is a weak necessity iff p is a better possibility than not p. The technical implementation is different from ours and crucially involves not accepting that there is always a set of most favored worlds (what is known as the Limit Assumption in the trade). It appears to us that if one makes the Limit Assumption, Kratzer’s definitions collapse, leaving no distinction between simple necessity and weak necessity.


119

The way we proposed to implement Sloman’s insight was to suggest that must/have to say that all worlds in the modal base where the goal is achieved are worlds where the prejacent is true, while ought to/should say that all worlds in the modal base where the goal is achieved and which are optimal by an additional measure are worlds where the prejacent is true. The weak necessity modals explicitly signal that a secondary measure was used to make further distinctions among the favored worlds. Our conception of weak necessity then makes them universal/necessity modals just as much as strong necessity modals are. What makes them weaker semantically is that they have a smaller domain of quantification: strong necessity modals say that the prejacent is true in all of the favored worlds, while weak necessity modals say that the prejacent is true in all of the very best (by some additional measure) among the favored worlds. In the terms of the Kratzerian framework, we suggested that weak necessity modals are in general sensitive to (at least) two ordering sources. In the goal-oriented case, the first ordering source is simply the goal proposition designated by an (in order) to-adjunct or an if you want to-anankastic conditional. The second, subsidiary ordering source contains considerations such as how fast, how comfortable, how cheap, . . . the means for achieving the goal are. Weak necessity modals are used not just in goal-oriented modal claims, of course. There are epistemic uses and deontic uses: (8)

Morris ought to be in his office. (ambiguous between epistemic and deontic readings)

What are the additional ordering sources in epistemic and deontic cases? We propose that epistemic ought differs from epistemic must/have to in being sensitive not just to the hard and fast evidence available in a situation but also to a set of propositions that describe what is normally the case.9 And in the deontic case, ought might be sensitive to less coercive sets of rules and principles in addition to the laws and regulations that strong necessity modals would be interpreted with respect to.10,11 9

We should note that Kratzer suggested that even the strong epistemic necessity modals are sensitive to shakier assumptions. This was her attempt at explaining the apparent fact that must p seems weaker than a plain assertion of p. We are not entirely convinced that this is right. Perhaps, must p is in fact a strong necessity claim but marks that a deduction has occurred, while only a plain assertion of p is compatible with direct observation. This is something that is explored a little bit further by von Fintel and Gillies (2007). 10 An intuition that deontic weak necessity goes with less coercive rules is laid out by Bybee et al. (1994), who write: “An examination of familiar and well-documented languages suggests that the major distinctions within obligation have to do with gradations of strength of the obligation: that is an obligation may be either strong or weak. If a weak obligation is not fulfilled, the consequences are not too serious; but the consequences of not fulfilling a strong obligation are much more severe. [. . . ] English distinguishes strong obligation, expressed with must and have to, and weak obligation, expressed with should” [p. 186]. 11 It should be noted that the choice of what is a primary ordering source and what is a secondary ordering source is presumably not an accident. In the goal-oriented case we have the designated goal and measures of ways of achieving it, in the epistemic case we have hard and fast evidence and

120


There is obviously much more to be done before we would have a satisfactory theory of weak necessity and we won’t be able to do much of that here. In this paper, our goal is to find some illumination from the way that many other languages express weak necessity.

3 The Crosslinguistic Picture Does something like ought exist in other languages? (Note: For convenience, we will mostly be using capitalized “OUGHT” for the meaning of English ought and its equivalent in different languages. We will reserve “ought” for the English lexical item.) It is not possible to answer the question of the cross-linguistic existence of OUGHT without providing a way to identify OUGHT cross-linguistically, that is, without providing essential ingredients of its meaning. We will try to identify OUGHT in other languages by trying to set up contrasts like those in (1) and (3). We will start with Greek12 : (9)

Tha eprepe na plinis ta piata ala dhen ise ipexreomenos na to FUT must+Past NA wash the dishes but NEG are obliged NA it kanis do ‘You ought to do the dishes but you are not obliged to do it’

(10)

#prepi na plinis ta piata ala dhen ise ipexreomenos na to kanis NA it do must NA wash the dishes but NEG are obliged ‘You must do the dishes but you are not obliged to do it’

(11)

Oli tha eprepe na plenun ta cheria tus ala i servitori ine All FUT must+Past NA wash the hands their but the waiters are ipochreomeni na to kanun obliged NA it do ‘All ought to wash their hands but the waiters are obliged to do it’

(12)

Oli tha eprepe na plenun ta cheria tus ala i servitori prepi All FUT must+PAST NA wash the hands their but the waiters must na ta plinun NA them wash ‘Everybody ought to wash their hands but waiters have to wash them’

What we see as qualifying as OUGHT in Greek is the necessity modal prepi in the Past tense in combination with the Future, the undeclinable particle tha. In fact, we guesswork based on unreliable assumptions about the normal course of events, and in the deontic case we have strict laws and less sanctionable codes of behavior. 12 The item na that occurs in most of our examples from Greek is an INFL area particle, present in all the Balkan Sprachbund languages. Its nature is not relevant to us here. We simply gloss it as NA.


121

can see in (12) that the modal prepi appears twice, once as OUGHT and once as a strong necessity modal. In the absence of the additional morphology on prepi as in (10), the sentence in (9) becomes a contradiction. This means that at least in Greek, the difference between weak and strong necessity is not marked by the choice between different lexical items but by the presence or absence of the Future + Past morphosyntax on one and the same modal (which in its simple form expresses strong necessity). What is this Future + Past combination? It is the morphology that appears on verbs in counterfactuals, specifically, in the consequent of counterfactual conditionals (from Iatridou (2000))13: (13)

An efevge simera tha eftane tin ali evdhomadha if left today FUT arrive/PAST/IMP the other week ‘If he left today, he would arrive next week’

(14)

An ton iche xtipisi to aftokinito tha iche pethani if him had hit the car FUT have+PAST died ‘If the car had hit him he would have died’

The astonishing conclusion is that Greek OUGHT is a strong necessity modal meaning ‘must’ augmented by counterfactual morphology. (Henceforth, we will often use the abbreviation CF to refer to counterfactual morphology.) Next let’s consider French. Here are the sentences we are considering: (15)

Tu devrais faire la vaisselle, mais tu n’es pas obligé you must/COND do the dishes but you not+are not obliged ‘you ought to do the dishes but you are not obliged to do them’

(16)

#Tu dois faire la vaisselle mais tu n’es pas obligé

(17)

Tout le monde devrait se laver les mains mais les serveurs sont everybody must/COND REFL wash the hands but the waiters are obligés obliged ‘Everybody ought to wash their hands but the waiters have to’

The sentence that has OUGHT in its translation is (15). It is not a contradiction, unlike (16). In (15), the modal devoir has the morphology that is traditionally described as “conditional mood”. This morphology is absent in (16). In (17), with the Conditional morphology, the distinction between OUGHT and strong necessity, namely between devoir+COND and plain eˆ tre obligé, can be set up again. Conditional Mood is what appears in the consequent of counterfactual conditionals in French: 13 We can see in (14) that the verb in the counterfactual consequent is also carrying Imperfective morphology, a specification that is missing form the verb prepi in its guise as OUGHT. Imperfective morphology is indeed required in Greek counterfactuals (and in counterfactuals in many other languages). However, there are some (extremely few) verbs in Greek that are not specified for the Imperfective/Perfective distinction and prepi is one of them (the verb meaning have is another, as can be seen in (12)). Therefore, we will not consider the neutrality with respect to the imperfective/perfective distinction an impediment to the conclusion reached in the text.

122

(18)


Il n’est pas soûl. Si il etait soûl, il parlerait plus fort He not+is NEG drunk. If he were drunk he talk/COND more loud

Iatridou (2000) argues that Conditional mood in Romance is nothing but Past+ Future combination, but this is not terribly important for us here. What is important is that the modal that means OUGHT in French carries the same morphology as the verb in a counterfactual consequent. Spanish behaves just like French and Greek: (19)

Deberia limpiar los platos, pero no estoy obligado Must+COND clean the dishes but not am obliged ‘I ought to do the dishes but I am not obliged’

(20)

Tendria que limpiar los platos, pero no estoy obligado Have+COND COMPL clean the dishes but not am obliged ‘I ought to do the dishes but I am not obliged to’

(21)

#Tengo que limpiar los platos pero no estoy obligado Have COMP clean the dishes but not am obliged

(22)

mas No esta borracho. Si estuviera borracho, gritaria if was/SUBJ drunk yell/COND more Not is drunk ‘He is not drunk. If he were drunk he would yell more.’

In (19)/(20), we see that we can set up the by now familiar contrast without generating a contradiction when a necessity modal contains conditional morphology. Sentence (21) shows that in the absence of this morphology we do get a contradiction. Sentence (22) shows that the morphology in question is exactly what appears in a counterfactual consequent. Finally, (23) shows the other way of bringing out the contrast: (23)

Los alumnus de quinto tendrian que/deberian conocer la The students of fifth have/COND COMPL /must/COND know the historia pero los de sexto deben/tienen que conocerla story but those of sixth must/have COMP know+it ‘The students of 5th grade ought to know the story but those of 6th grade have to’.14

Outside of Romance and Greek, we find the same pattern in Slavic. First let us consider Russian.15 In Russian, the morphology on the counterfactual antecedent is Past tense plus the element byl, which we will not gloss here. (24)

Esli by on byl p’jan, to on by shumel if byl he was drunk then he byl make-noise-Past-Imperf ‘If he was drunk, he would be making noise’

14 Karlos Arregi, who is the source of our Spanish data, reports that deben here is slightly dispreferred and tienen is the preferred option. It is unclear to him why this is so. He reports that overall, including contexts outside ours, he prefers to use tener que instead of deber. 15 Russian provided by Tania Ionin.


123

This is exactly the morphology we find when we try to set up the OUGHT versus must/have to contrast. Unlike in the previous languages, here the modal element is participial (or adjectival) and the counterfactual morphology appears on the copula: (25)

Ty dolzhen byl by vymyt’ posudu, no ty ne objazan you required be+PAST by wash-Perf-Inf dishes but you not obligated eto delat’ this do-Inf ‘You ought to wash the dishes, but you don’t have to do it’

And again, (25) is not a contradiction. Next let us take a look at Croatian.16 The CF-morphology is bi + participle: (26)

pijan, viˇse bi vikao Da si if are.2SG drunk more would.2SG yelled.PCPL ‘If you were drunk, you would yell more.’

When we add the CF-morphology to the necessity modal, we get the meaning of OUGHT and lack of a contradiction in sentences like the following: (27)

pospremiti sobu, ali na sreću ne moraˇs Morao bi room but on luck not have.2SG must.PCPL would.2SG clear ‘You ought to tidy up your room, but luckily, you don’t have to.’

In the absence of CF-morphology on the modal there is a contradiction: (28)

#Moraˇs pospremiti sobu, ali na sreću ne moraˇs must clear room but on luck not have.2SG ‘You have to tidy up your room, but luckily, you don’t have to.’

And we can also set up the by now familiar contrast as in 17 (29)

Puˇckoˇskolci bi morali znati algebru, ali elem.school children would.3 PL must.PCPL know algebra but srednjoˇskolci je moraju znati. high school children it must know ‘Elementary school children ought to know algebra, but high school children have to know it.’

We find the same phenomenon in Germanic. Consider Dutch.18 The Dutch counterfactual consequent contains the past tense of the verb zullen, which by itself (i.e. without the past tense) is used as a future marker. We will be glossing it with ‘would’, therefore, as this element can be seen as the past tense of will. (30)

16 17 18

Als ik rijk was, zou ik stoppen met werken. with work If I rich were, would I stop ‘If I were rich I would stop working’

Croatian provided by Martina Gracanin. Bi is a second position clitic, which results in the reversal in order of the participle and bi. Dutch data provided by Janneke Huitink.

124


What happens when this counterfactual morphology combines with the modal necessity modal moeten? As expected, we get the meaning of ought, which does not cause a contradiction when juxtaposed with a negated necessity modal: (31)

Je zou eens Anna Karenina moeten lezen, maar het hoeft19 you would sometime AK must read but it must/NPI niet not

In the absence of the counterfactual morphology, the sentence is a contradiction: (32)

#Je moet AK lezen, maar het hoeft niet. You must AK read, but it must/NPI not

And here is the other contrast: (33)

Iedereen zou Anna Karenina moeten lezen, en/maar mijn studenten Everyone would AK must read and/but my students moeten het lezen. must it read ‘Everyone should read AK and/but my students have to read it’

This is as good an occasion as any to point out that we are not claiming that all the world’s weak necessity modals are formed by CF-marking on strong necessity modals. There are other ways to express weak necessity, in particular through dedicated lexical items, such as English ought. Dutch, for example, as pointed out to us by Marcel den Dikken (pc), has a modal horen that has as part of its lexical meaning weak necessity and it doesn’t need CF-marking to convey that. The item is also lexically restricted to deontic uses, it cannot be used as an epistemic or goal-oriented modal.20 Let us add one more Germanic language to the picture, namely Icelandic21: (34)

Allir ættu a vo sér um hendurnar all.NOM.PL have.cf.3PL to wash.INF themselves at hands.the.ACC.PL en starfsmenn eru skyldugir a gera a . but employee.NOM.PL be.3PL obliged.NOM.PL to do.INF it ‘Everyone ought to wash their hands, but employees are required to do so.’

(35)

u´ ættir a vo upp en u´ ert ekki you.NOM have.cf.2SG to wash.INF up but you.NOM be.2SG not skyldugur a gera a . obliged.NOM.SG to do.INF it ‘You ought to do the dishes, but you’re not required to do them.’

19 The verb hoeven is the form of the necessity modal when it scopes under negation. We gloss it there as must/NPI. 20 Another language that uses both CF-marking on a strong necessity modal and a dedicated lexical item appears to be Swedish, as reported to us by Anna-Sara Malmgren (pc). 21 We thank J´ ohannes G´ısli Jónsson and Chris Warnasch for providing us with these data.


(36)

# u´ a´ tt a vo upp en u´ ert ekki you.NOM have.2SG to wash.INF up but you.NOM be.2SG not a. skyldugur a gera obliged.NOM.SG to do.INF it ‘#You have to do the dishes, but you’re not required to do them.’

(37)

Ef hann sig meira a´ væri hann ekki hreyfi if he.NOM move.cf.3SG himself more then be.cf.3SG he.NOM not jafn reyttur. equally tired.NOM.SG ‘If he were more active then he wouldn’t be so tired.’

125

Outside Indo-European, we find the same phenomenon. Consider Hungarian.22 In a counterfactual, the so-called ‘conditional’ morphology -na/-ne appears: (38)

Nem re’szeg. Ha re’szeg len-ne-0, hangos-abb-an not drunk if drunk be-cond-pres,3sg loud-comp-adv kiabal-na-0 shout-cond-pres,3sg

Then, as before, if we take the CF-morphology and place it on a necessity modal kell, we get exactly what we have seen so far. (39)

El kell-ene-0 mosogat-n-od a ta’nye’r-ok-at, de nem away must-cond-3sg wash-inf-2sg the dish-pl-acc but not vagy musza’j be-pres,2sg must ‘You ought to / should wash the dishes, but you don’t need to / but it’s not necessary’

In the absence of the conditional morphology on the necessity modal the sentence is a contradiction: (40)

#El kell-3sg mosogat-n-od a ta’nye’r-ok-at, de nem vagy away must-3sg wash-inf-2sg the dish-pl-acc but not be-pres,2sg musza’j must ‘You must wash the dishes, but you don’t need to / but it’s not necessary’

And the other place where we have been seeing the contrast can also be set up:

22 The Hungarian data were provided by Aniko Csirmaz. According to Anna Szabolcsi (pc), there may be an additional interesting fact in Hungarian: when the complement of OUGHT is stative, the situation is counterfactual, in that it cannot be changed anymore. When the complement is not stative, no such entailment/implicature arises. We were not able to duplicate this judgment with all of our informants, though.

126

(41)


Az o:to:dik-es-ek-nek tud-ni-uk kell-ene ez-t a to:rte’net-et, the fifth-adj-pl-dat know-inf-3pl must-cond this-acc the story-acc de a hatodik-os-ok-nak musza’j / kell tud-ni-uk but the sixth-adj-pl-dat must / must know-inf-3pl ‘Fifth graders ought to / should know this story, but sixth graders must know it’

We conclude then that it is a cross-linguistically stable fact that the meaning of OUGHT can be conveyed with counterfactual morphology on a strong necessity modal.23 In the perhaps illusory hope that we can get the semantics of OUGHT compositionally and transparently from the combination of counterfactuality with a strong necessity modal, we will from now on be using the term “transparent OUGHT ” to refer to the strong necessity modal + CF-morphology that English ought translates into in languages like the above. The next section addresses the question whether transparent OUGHT has the same range of modal flavors as English ought does.

4 Flavors Which of the common modalities (deontic, epistemic etc.) can ought/OUGHT function as?

4.1 Epistemic Modality Here is an example of ought in an epistemic use: (42)

It’s 3pm. He ought to be in his office.

Let’s say you are on your way to Morris’s office, which is down the hall from mine, and ask me whether I think that Morris is in his office. Neither of us knows whether he is, in fact, there. Under those circumstances, I can utter (43). (43)

It’s 3pm. Given what I know about Morris’s habits, he ought to be in his office. Why don’t you go check?

The same fact is also true for transparent OUGHT. Greek: (44)

Ine 3. Tha’prepe na ine sto grafio tu. Pigene na dhis ‘It is 3. He ought to be in his office. Go see.’

In short, both ought and transparent OUGHT can be used epistemically. 23 In fact, historically, English fits this picture too. According to the OED and other sources— many thanks to Jay Jasanoff for discussion of these points—modal ought was the past indicative as well as the Past Subjunctive of the verb owe (‘possess’, one more case of a verb meaning “possession” becoming a modal) in the Old English period (700–1100). Later on, ought continued as the past subjunctive only, with the past indicative of owe continuing as owed.


127

4.2 Goal-Oriented Modality Next we go to ought as a goal-oriented modal. It seems uncontroversial that such a use is possible: (45)

To go to Ashfield, you ought to take Route 2.

We already discussed this case earlier. The best way to go to Ashfield is the one where some secondary goal is satisfied as well, e.g. avoiding traffic, or a having a scenic drive. Here are some cases of goal-oriented transparent OUGHT: (46)

Pour traverser, tu devrais prendre ce bateau-ci to cross, you must+CF take this boat

(47)

Gia na perasis apenandi tha’prepe na chrisimopiisis aftin edho tin in order to cross other side must+CF NA use this here the varka boat

4.3 Deontic Modality In order to make certain that we are dealing with deontic ought/OUGHT, we can try to make the source of the obligation overt: (48)

?According to the law, people convicted of stealing ought to go to prison

(49)

?Simfona me ton nomo, I kleftes tha’prepe na pane filaki (Greek)

(50)

?Segun la ley, un ladron deberia ir a la carcel According-to the law, a thief must-COND go to the jail

How good are these sentences? We feel that there is something funny about them. The law does not speak like that.24 A theory of ought/OUGHT will have to capture and explain the funniness of its use in deontic contexts like the ones above. There are sentences that possibly come closer to showing that ought can appear as a deontic modal: (51) (52)

You ought to do the dishes. It ought to be the case that bullying is/be illegal.

Here, the two authors are disagreeing for the time being. One author thinks that she could spin an argument that such cases are not really deontic but goal-oriented, something like “to satisfy rules of politeness, you ought to do the dishes”, etc.25 The 24

Wertheimer (1972: pp. 116 and 120) makes the same observation. The idea that OUGHT is primarily a goal-oriented modal is also defended by Finlay in work in progress (Finlay, 2006).

25

128


other author believes that he can see deontic ought as perfectly normal. The crucial point about ought is that it signals the existence of a secondary ordering source. When we report the content of one particular set of rules or principles,26 a kind of megalomania occurs that makes that set of rules the only relevant ordering source and so ought becomes unusable. But as soon as we have more than one set of rules interacting, deontic ought is fine. We will not resolve this debate here. The reader is free to pick a side. *** To conclude this section: We have seen that transparent OUGHT really seems to be very much the same as English ought semantically. The range of uses is entirely parallel, the conveyed meanings appear to be the same. We had a preliminary idea about the meaning of English ought: that it expresses weak necessity, plausibly construed along the lines suggested by Sloman. Now, it is time to face the music: if transparent OUGHT has the same meaning of weak necessity, how does that meaning arise from the combination of counterfactual marking and strong necessity?

5 Counterfactuality? There is much that we could and should say about the morphosyntax, semantics, and pragmatics of CF-morphology, some of which both of us have done in the past (Iatridou, 2000; von Fintel, 1999). For now, we would like to stay at a fairly simple and intuitive level. Counterfactual marking signals that some explicit or implicit assumption has taken us outside the “context set” in Stalnaker’s sense. Consider a CF-marked conditional: (53)

If Peter were in the office, Mary would be happy.

The counterfactual marking could be there because it is taken for granted that Peter is not in his office (counterfactuality in the strict sense), or it could be merely 26 A side-remark on how the content of laws, rules, and regulations is commonly presented: from the perspective of Kratzer’s framework, where ordering sources are given as sets of propositions, satisfaction of which is used to measure the position of a particular world on the ordering scale, one might expect something like this:

(i)

The following is the law: there is no obstruction of driveways, anyone who obstructs a driveway pays a fine, . . .

The content of the law is a set of propositions that the world should ideally make true. Instead, what one usually finds is this: (ii)

The following is the law: there must be no obstruction of driveways, anyone who obstruct a driveway must pay a fine, . . .

That is, the law itself is presented using deontic modals. When you think about it, this is a curious kind of circularity. The solution is presumably that the modals used in the declaration of the law are performatively used. We’ll leave this topic to another occasion or for other researchers.


129

because the speaker wants to admit the possibility that it is taken for granted that Peter is not in the office. Now, in the transparent OUGHT construction, counterfactual marking occurs on a modalized sentence (where the modal is a strong necessity modal). The hope is, of course, that if we compose these two ingredients, the meaning of ought will arise. If this was an ideal world (one without compositionality puzzles—but what fun would that be?), every time we have a strong necessity modal with CF-morphology, the meaning of OUGHT would arise. Unfortunately, or, fortunately, things are not that simple. When you think about it, having counterfactuality on top of an embedded modal should result in the claim that the modal claim holds not necessarily in the actual world but in some, possibly counterfactual, worlds that the higher operator takes us to. So, in our case we would predict that CF + strong necessity should claim that a strong necessity claim holds in some possibly counterfactual world, which is not what we wanted: we wanted as the result that a weak necessity claim holds in the actual world. Interestingly, we find that English behaves exactly as we would have predicted. English has a strong necessity modal that can inflect and occur in an infinitival form, namely have to. This language also has CF-morphology, usually taken to be would. What we find now is that adding would on top of have to does not yield the meaning of a weak necessity OUGHT. Recall the environments in which we diagnosed OUGHT. In those we see that ought cannot be replaced by would have to: (54)

a. b.

∗ #27

(55)

a. b.

∗

Everyone ought to wash their hands; employees must Everyone would have to wash their hands; employees must

You ought to do the dishes but you are not obliged to do them. #You would have to do the dishes but you are not obliged to do them.

The same conclusion holds if we look at goal-oriented modality. Recall that a plain necessity modal like have to conveys that there is only one way to achieve one’s goal, while ought conveys that there is more than one way to achieve the goal but that the suggested means is the best by some measure: (56)

a. b.

To get to the island you have to use this boat. To get to the island you ought to use this boat.

If English had transparent OUGHT, that is, if the combination of necessity + CF always yielded OUGHT, then would have to should be able to convey what ought does. But it does not: (57)

To get to the island, you would have to use this boat.

(57) conveys that there is only one way to get to the island. That is, it patterns with have to, not ought. In other words, English provides us with a case that shows that not every combination of necessity + CF yields OUGHT. It shows precisely the expected 27

We use the symbol “∗#” to avoid determining the nature of the inappropriateness.

130


interpretation from such a combination. The string would have to talks about a necessity that obtains in a counterfactual world.28 In the actual world, there is no modal advice, suggestion or obligation (we will refer to this meaning as WOULD HAVE TO ): (58)

(If Fred had a car) he would have to register it.

(59)

(If Fred wanted to get to the island) he would have to use this boat.

(60)

(If the law were/was different) Fred would have to give up everything in the divorce.

On the other hand, when we use ought, the modal’s force holds in the actual world: (61)

He ought to register the car.

(62)

He ought to use this boat.

(63)

Fred ought to give up everything in the divorce

Now, we come to the surprising phenomenon of transparent OUGHT. In the transparent languages, the exact same string is used in all of (58–60) and (61–63). The following examples are from Greek but the observation holds for the other languages as well: (64)

An o Fred iche aftokinito, tha eprepe na to if the Fred had car, must+CF(=WOULD HAVE TO) NA it dhilosi register

(65)

An o Fred ithele na pai sto nisi, tha eprepe If the Fred wanted to go to-the island, must+CF(=WOULD HAVE TO) na pari aftin tin varka NA take this the boat

(66)

An o nomos itan dhiaforetikos, o Fred tha eprepe If the law were different, the Fred must+CF(=WOULD HAVE TO) na parachorisi ta panda sto dhiazijio NA give up everything in-the divorce

(67)

o Fred tha eprepe na dhilosi to aftokinito the Fred must+CF(=OUGHT) NA register the car

(68)

na pari aftin tin varka o Fred tha eprepe the Fred must+CF(=OUGHT) NA take this the boat

(69)

o Fred tha eprepe na parachorisi ta panda sto dhiazijio the Fred must+CF(=OUGHT) NA give up everything in-the divorce

28 Counterfactual epistemics have difficulties of their own, so we will stay away from them for now.


131

In sentences (64–66), the modal relation holds in a counterfactual world. In (67–69), the modal relation holds in the actual world. This, then, is the picture that emerges: In some languages (the ones we called “transparent OUGHT languages”) the string necessity+CF has two meanings, that of a weak necessity modal in the actual world and that of a strong necessity modal in some counterfactual worlds. For convenience, we will refer to the interpretation in which the modal holds in the actual world as OUGHT. We will refer to the interpretation where the modal holds in the counterfactual world as WOULD HAVE TO . English, on the other hand, has lexicalized the interpretation where the modal holds in the actual world into the item ought. In addition, the English string would have to unambiguously refers to the interpretation where the modal holds in a counterfactual world. It would appear then that in the case of transparent OUGHT, counterfactual marking is not doing its usual job of marking that we are being taken to a counterfactual scenario. Instead, the modal claim continues to be made about the actual world and the effect of the marking is to weaken the strong necessity modal to a weak necessity modal. In the transparent languages, counterfactual marking has two uses in combination with a strong necessity modal: (i) saying that the strong necessity holds in a counterfactual scenario, (ii) saying that a weak necessity holds in the actual world. In English, counterfactual marking on a strong necessity modal only has use (i). To express use (ii), English resorts to the lexical item ought.

6 A Consolation and a Precedent In the previous section we saw that not all combinations of strong necessity + CF yield the meaning OUGHT. For example, English would have to fails to yield such a reading. But the closer inspection to which the English case forced us, also made us realize that in the transparent languages one and the same string (strong necessity + CF) is ambiguous between two interpretations. This is a very crucial point as we will need to address the question of where and how this ambiguity manifests itself. That is, does the string necessity + CF yield two separate LFs or is there one (underdetermined) meaning or LF that can yield the two interpretations seen above by way of the context? In either case, the question arises whether it is English that is weird or the transparent OUGHT languages. As we saw, from a certain point of view, the single interpretation that English gives to would have to is exactly what we expect, so the fact that in transparent OUGHT languages a second interpretation emerges seems unexpected. On the other hand, the ambiguity we found in transparent OUGHT languages is cross-linguistically very stable and maybe once we find a way to explain it, the absence of the OUGHT reading for English would have to will become the unexpected fact.

132


We will try to answer these questions in what follows. In this section, we will show that there is another set of data where we find that for a certain modal meaning, English chooses a designated lexical item, while other languages choose a “transparent” way of conveying that meaning. This seems to support the view that it is English that is the outlier. Consider the English verb wish. In a variety of languages what are called “counterfactual wishes” are done with want + CF-morphology (Iatridou, 2000). That is, where English uses the verb wish for counterfactual wishes, other languages use the verb want augmented by CF-morphology (Iatridou, 2000): (70)

He wished she had a Honda Odyssey

(71)

Il voudrait qu’elle ait une Honda Odyssey

(72)

Tha ithele na iche ena Honda Odyssey

In short, English lexicalizes into the verb wish what other languages express with the verb want + CF, and English lexicalizes into ought what other languages express with the verb must + CF. So like before, the question arises whether internal to English, want + CF can freely occur instead of wish and vice versa. The answer will be ‘no’ for either direction. The English version of want + CF is would want to. This periphrasis cannot substitute for English wish: (73) (74)

I wish that I was taller. ∗I

would want that I were taller.29

Neither can wish substitute for want when the latter is in a counterfactual conditional30: (75) (76)

If he were taller, he would want to have a different bed. ∗ If

he were taller, he wishes (for/ to have/ that he had) a different bed.

So what we have is that is that there are two lexicalizations in English that other languages can express as a verb + CF-morphology. However, in both cases internal to English, the lexicalized items cannot substitute for or be substituted by the relevant English verbs augmented by CF-morphology. 29

We are disregarding here the archaic I would that I were taller. This test was not presented in the section on OUGHT because we have not managed to find consistent data among the native speakers that we consulted. That is, Section 5 showed that would have to cannot appear where ought is good. The question also arises whether ought can appear where would have to is good. This is the same as the question whether ought can appear in a CF consequent, satisfying the morphosemantic requirements of the verb in a CF consequent:

30

(i)

All males who are 18 years old have to register with the Selective Service. It’s a good thing he is not 18. If he were 18, he ought to register with the Selective Service.

Unfortunately, we have found that there is serious disagreement among speakers as to the status of (i) and similar sentences.


133

However, as before, in the “transparent wish” languages, the string want + CF is ambiguous between wanting in a counterfactual world as in (77) and wanting in the actual world as in (78): (77)

An itan psiloteros tha ithele makritero krevati if was taller FUT want+Past longer bed ‘If he was taller he would want a longer bed’

(78)

Tha ithela na imun psiloteri FUT want+Past NA was taller ‘I wish I was taller’

English, on the other hand, lexicalizes into one item the case where the desire is in the actual world (wish31 ) while the periphrastic string is reserved for desire in a counterfactual world (would want to). The parallelism to the case of OUGHT should now be clear: English chooses specialized lexical items for the interpretation where the modal claim holds in the actual world (ought) and where the desire holds in the actual world (wish). We can visualize the situation as in Fig. 1. In both cases, the case of transparent OUGHT and the case of transparent WISH, we see that while in the transparent languages the combination of the basic item and CF-morphology does double duty, in English a dedicated lexical item takes on the meaning where the modal claim holds in the actual world and in English the CF-marked structure carries only one of the meaning that it would have in the transparent languages. The systematicity of the picture suggests to us that English should be treated as the special case. Why should in English the combination of CF-marking and strong necessity not be able to express weak necessity? One possibility is morphological blocking: it is precisely the presence of a dedicated lexical item that blocks the weak necessity meaning for the more complex structure. However, this idea would appear to be immediately falsified by the fact that some transparent languages also have dedicated weak necessity modals (Dutch, for example, has the dedicated deontic weak necessity item horen as we saw earlier). Another possibility is that English is missing a crucial enabling factor without which CF-marking cannot do the job it does in the transparent OUGHT and transparent WISH constructions. It is quite likely that additional factors are, in fact, required. For example, CF-morphology with a necessity modal that is adjectival also fails to yield OUGHT. Sentence (79) is a counterfactual modal (i.e. it patterns with would have to); it does not mean OUGHT: (79)

To get to the island it would be necessary to use this boat.

And lest the reader think that failure to compose OUGHT is a general property of English and that adjectival modals are just a special case of that, consider (80) from Greek. In this sentence, CF-marking with the adjectival necessity modal fails to 31 This shows that the term “counterfactual wishing” is misleading. The desire is in the actual world.

134


transparent languages:

strong necessity + CF

OUGHT modal claim in actual world

WOULD HAVE TO modal claim in counterfactual world

"ought "

"would have to"

"wish"

"would want"

English :

WISH desire in actual world

transparent languages:

WOULD WANT desire in counterfactual world

want + CF

Fig. 1 Comparing English to “transparent” languages

yield OUGHT, the result is just that of a modal in a counterfactual world (i.e. WOULD HAVE TO ): (80)

tha itan anageo na paris aftin tin varka BE+CF necessary NA take this the boat ‘It would be necessary to take this boat’ (not ‘it ought to be necessary to take this boat’)

Actually, with the exception of the Russian participial modals (which may turn out to not be an exception after all), we have no case where a non-verbal modal could yield OUGHT when (the copula is) combined with CF-marking. From what we have seen, only verbal necessity modals turn into OUGHT. So it may not be sufficient to have a necessity modal of any type and CF marking to make OUGHT. Some additional condition must be satisfied. We don’t know what that would be.32

32

Noam Chomsky (pc) pointed out to us a possible generalization: the transparent reading only arises when tense is marked directly on the modal. Why would that be so, if it turns out to be true?


135

Here is where we are now. Our goal is to figure out how the combination of CF-morphology and a strong necessity modal can result in the meaning of a weak necessity modal. It was immediately clear that in the typical uses of OUGHT, we do not claim that a strong necessity holds in a counterfactual scenario. Instead, we claim that a weak necessity holds in the actual world. That throws doubt on the idea that CF-marking is doing its usual job in transparent OUGHT. Then, we saw that CF-marking on a strong necessity modal fails to create OUGHT in at least two cases: (i) English would have to, which may be due to blocking by the lexicalized form ought and is parallel to the failure of would want to mean ‘wish’ even though want with CF-morphology does mean ‘wish’ in many other languages; (ii) CF-marking on non-verbal necessity modals, for which we have no explanation. No matter how we eventually explain these exceptions, we still have no handle on what the CFmorphology is doing to the strong necessity meaning in the transparent OUGHT cases that do work. In the next section, we discuss a possible solution to the puzzle, which we argue cannot work.

7 Scope Confusion? We saw that the combination of CF-morphology and strong necessity modals is ambiguous between a WOULD HAVE TO reading and an OUGHT reading. The former is what one would expect from a modal in a counterfactual scenario. It’s the latter that is puzzling. What if in that case there is a permutation at LF whereby the counterfactuality doesn’t actually take scope over the modal but takes scope under it? We owe this idea to Tim Stowell (pc). We think that such an operation may in fact occur in certain cases. For example, consider the following puzzling use of counterfactual marking in English: (81)

I would have expected him to be here.

This sentence appears ambiguous. There is the entirely predictable reading of what the expectations of the speaker in a certain counterfactual scenario would have been, as in: (82)

If he had promised to attend this meeting, I would have expected him to be here.

But there is also a reading where (81) actually expresses an actual expectation, but one that turned out to be unsatisfied: (83)

I would have expected him to be here. Why isn’t he?

Perhaps, the right analysis of (81) is that the counterfactual marking on expect is out of place and at LF is interpreted on the complement sentence, marking that he is not in fact here.

136


Another case that one might consider such an analysis for is the case of transparent WISH that we introduced earlier. Here, counterfactual marking on want could be seen as expressing not a want in a counterfactual scenario but an actual want towards a counterfactual state of affairs.33 So, this is not a crazy idea. Can we carry it over to transparent OUGHT? No, the analysis fails spectacularly on two connected grounds. The first one is that in OUGHT , it is not a strong necessity modal that makes it to the actual world. That is, if scopal rearrangement brought the necessity modal out of the scope of CFmarking then a sentence with OUGHT should make a strong necessity claim in the actual world. But this is not so, as we have seen. The second reason is that in the transparent OUGHT cases, the complement is simply not marked as counterfactual. When a speaker uses transparent or non-transparent deontic or goal-oriented ought, there is no feeling whatsoever that the event under ought is contrary-to-fact or even unlikely.34 So, we are back to square one. We have no explanation for why counterfactual marking turns the meaning of a strong necessity modal into a weak necessity meaning.

8 Ordering Source Promotion Let’s regroup. We started with a brief discussion of our previous view of the difference between strong and weak necessity modals. In essence, weak necessity modals bring in a secondary ordering of the favored worlds. Strong necessity modals say that the prejacent is true in all of the favored worlds, while weak necessity modals say that the prejacent is true in all of the very best (by some additional measure) among the favored worlds. While the standard Kratzer framework parametrizes the semantics of modals to two parameters (modal base and ordering source), we introduced a pair of ordering sources: (i) the primary one that is the only one that strong necessity modals are sensitive to and (ii) a secondary one which is the one that weak necessity modals use to refine the ranking of the worlds favored by the primary ordering

33

We actually don’t think that this is the right analysis. But we’ll leave the treatment of transparent WISH to some other time. For now, there is some relevant discussion in Iatridou (2000). 34 Of course, there are cases where the prejacent is interpreted as contrary-to-fact: (i)

He ought not to have revealed the secret.

But we assume that here the counterfactuality of the prejacent (he didn’t reveal the secret) is signaled by additional morphological factors. Note by the way that it is not easy to use strong necessity modals to express the same post-fact denunciation of a mistake: (ii)

#He had to/has to not have revealed the secret.

(iii)

#He would have to not have revealed the secret.

We do not know why this is so.


137

source. We built that differential sensitivity into the lexical entries of must/have to and ought. The puzzle we are facing now is that in the transparent OUGHT construction something makes it so that a strong necessity modal suddenly shows sensitivity to the secondary ordering source. And that mysterious something is somehow brought into play by counter-factual marking. Our strategy now will be to first identify what operation needs to happen to make a strong necessity modal sensitive to a secondary ordering source, and then to think about why counterfactual marking brings about that operation. What needs to happen to make a strong necessity modal sensitive to a secondary ordering source? By assumption, the lexical entry of a strong necessity modal only looks at the primary ordering source. So, what we need to do is to take the secondary ordering source and promote it to primary status, without, of course, forgetting the initial primary ordering source. The idea is that saying that to go to Ashfield you ought to take Route 2, because it’s the most scenic way, is the same as saying that to go to Ashfield in the most scenic way, you have to take Route 2. We have promoted the secondary goal of enjoying as much scenery as possible to primary status. It is crucial though that the primary goal of getting to Ashfield is still paramount—the fact that you get the most scenery possible if you go to Serengeti National Park is irrelevant. This makes formalizing the notion of ordering source promotion a bit tricky, as we will now see. The simplest idea might be that we merge the secondary ordering source with the primary ordering source and interpret the strong necessity modal with respect to the newly merged primary ordering source, which now includes the promoted secondary ordering source. But what would “merger of ordering sources” be? In Kratzer’s framework, ordering sources are (functions from evaluation worlds to) sets of propositions, and we assess the status of the worlds in the modal base as to how many of the propositions in the ordering source they make true. So, since the ordering sources are sets of propositions, a natural idea about promotion and merger would be to just take the set union of the two sets of propositions. But that will go wrong and will not produce the same as the weak necessity meaning. Take our goal-oriented example. The primary ordering source is some goal such as “you get to Ashfield” and the secondary ordering source is a goal such as “you experience as much scenery as possible”. Given the right circumstances, that might mean that you ought to (but don’t have to) take Route 2. But when we merge the two ordering sources, we would quite possibly rank as equally optimal worlds where you get to Ashfield in a very scenic way and worlds where you go to Serengeti National Park in the most spectacularly scenic way imaginable. The problem is that the primary goal of getting to Ashfield should not be put on a par with maximizing scenery. Even though scenery maximization has been promoted, it should not be able to trump or even be considered at the same level as getting to Ashfield. The trick then is to make strong necessity modals sensitive to the secondary ordering source by promoting it but without making it count at the same level as the primary ordering source. There are at least two ways of doing this that we can think of. One would involve making modals in general sensitive to an ordered sequence

138


of ordering sources, making strong necessity modals sensitive to ordering sources in a designated initial segment of that sequence, and treat promotion as moving an ordering source into that initial segment.35 A perhaps simpler way of formalizing promotion would involve something very much like set union of the two ordering sources but would only add propositions from the secondary ordering source into the new ordering source if they do not conflict with the primary ordering source. A complicating issue with such an approach would be that there might not be a unique way of getting a newly merged ordering source. (What if the secondary ordering source itself contained two contradictory propositions? Which one would be added to the primary ordering source?) So, at the moment, we don’t know whether we should pursue this second option.36 The first half of our task is done (modulo the missing formal implementation): we understand what needs to happen to make a strong necessity modal sensitive to a secondary ordering source in such a way that it will express weak necessity. We need to promote the secondary ordering source to a status that makes it visible to the strong necessity modal. Now, we need to turn to the second half of our task: why is CF-morphology signaling ordering source promotion?37 35

It would probably be distracting to go through a formal development of that idea. Here are some rough sketches of the notions one might use: (i)

The context provides for each modal, a modal base f and a bipartitioned sequence of ordering sources , < gi+1 , . . ., gk >>

(ii)

Strong necessity modals say that the prejacent is true in all worlds in maxgi (w) (. . . (maxg1 (w) ( f (w)))).

Here maxg (w) is a function computed from an ordering source that identifies the best worlds in a set of worlds. (iii)

Weak necessity modals say that the prejacent is true maxgk (w) (. . . (maxgi+1 (w) (maxgi (w) (. . .(maxg1 (w) ( f (w))))))).

in all worlds

in

(iv)

is An ordering source sequence , < gi+1 , . . ., gk >> changed by ORDERING SOURCE PROMOTION by moving any number of ordering sources from the second tier into the first tier. For example, , < gi+2 , . . ., gk >> is the result of submitting the initial sequence in 1 to a one-step promotion operation.

36

Some of the technical work done by Frank (1996) on the notion of “compatibility restricted set union” would probably be useful to us. 37 Ordering source promotion may happen outside the transparent OUGHT construction as well. Consider an example attributed to Wolfgang Klein by von Stechow et al. (2006). Imagine that to cross Siberia to go to Vladivostok you can take one of two trains: the Russian train or the Chinese train. The Chinese train is significantly more comfortable. Now consider the following two variants: (i)

To go to Vladivostok, you have to take the Chinese train.

(ii)

To go to Vladivostok, you ought to take the Chinese train.

They report that Wolfgang Klein accepts the have to-variant, while Orin Percus only accepts the ought to-variant. What Klein-type speakers can do, in our analysis, is to silently promote


139

9 Why Counterfactual Marking? Why is it CF-morphology that gets put to the use of turning a strong necessity modal into a weak necessity modal in the transparent languages? What does CFmorphology on its more understood uses have in common with this notion of promotion of the secondary ordering source? In the transparent OUGHT cases, we are not moving to counterfactual worlds that differ from the actual world at the ground level of empirical facts: there are no different circumstances there, no different goals, primary or secondary, no different evidence, reliable or shaky. Instead, a parameter of evaluation is changed. We move from one context where a secondary ordering source is invisible to a strong necessity modal to a new context where that secondary ordering source is promoted in such a way as to become visible to the strong necessity modal. Perhaps, then, the counterfactual marking is co-opted here in a somewhat metalinguistic kind of way: “if we were in a context in which the secondary ordering source was promoted, then it would be a strong necessity that. . . ”. This would explain why even though there is CF-morphology, the modal claim is made firmly about the actual world; all that the morphology marks is a change in evaluation parameters. It is probably not an accident that counterfactual marking brings with it an element of tentativeness: the speaker is not saying that the secondary ordering source is something that has to be obeyed. The choice of whether to really promote the secondary ordering source is left open.

10 Conclusion In this paper, we have raised the question of how the semantics of weak necessity modals fits into the general picture of the semantics of modal expression. We have reiterated a tentative suggestion inspired by the old proposal by Sloman. We then brought in the cross-linguistic picture. It turned out that it is a very stable fact across languages that weak necessity can be expressed by taking a strong necessity modal and marking it with counterfactual morphology. We explored this pattern in a number of languages. We then raised the question of whether our ideas about the semantics of weak necessity can help us understand the fact that a strong necessity modal becomes a weak necessity modal when marked with counterfactual morphology. We proposed that what is going on here is the promotion of a secondary ordering source. The counterfactual morphology marks this quasi-meta-linguistic the secondary goal of being comfortable. Percus-type speakers cannot do silent ordering source promotion but either have to mark it by choosing a weak necessity modal or explicitly add comfort to the primary ordering source: (iii)

To go to Vladivostok comfortably you have to take the Chinese train.

140


operation but in a hypothetical way (“if we were to take your secondary goals and make them non-negotiable”, “if this were a normal day (i.e. if we were to take as given assumptions that only hold firmly for normal days)”, etc.).

References Bybee, J. L., Perkins, R. D., and Pagliuca, W. (1994). The Evolution of Grammar: Tense, Aspect, and Modality in the Languages of the World. Chicago University Press. Copley, B. (2006). What should Should mean? Ms of a paper given at the Workshop “Language under Uncertainty: Modals, Evidentials, and Conditionals”, Kyoto University, January 2005. Finlay, S. (2006). (In Order That. . . ) Ought. ms, USC. von Fintel, K. (1999). The presupposition of subjunctive conditionals. In Sauerland, U. and Percus, O., editors, The Interpretive Tract, number 25 in MIT Working Papers in Linguistics, pages 29–44. MITWPL, Cambridge, MA. von Fintel, K. and Gillies, A. S. (2007). An opinionated guide to epistemic modality. ms, MIT and University of Michigan, to appear in Tamar Gendler Szabó and John Hawthorne (eds.) Oxford Studies in Epistemology, Vol.2, Oxford University Press. von Fintel, K. and Iatridou, S. (2005). What to do if you want to go to Harlem: Anankastic conditionals and related matters. ms, MIT. Frank, A. (1996). Context Dependence in Modal Constructions. Ph.D. thesis, Universität Stuttgart. Horn, L. R. (1972). On the Semantic Properties of the Logical Operators in English. Ph.D. thesis, UCLA, CA. Iatridou, S. (2000). The grammatical ingredients of counterfactuality. Linguistic Inquiry, 31(2):231–270. Jones, A. J. and Pörn, I. (1986). Ought and Must. Synthese, 66(1):89–93. Kratzer, A. (1981). The notional category of modality. In Eikmeyer, H. J. and Rieser, H., editors, Words, Worlds, and Contexts. New Approaches in Word Semantics, pages 38–74, Berlin. de Gruyter. Kratzer, A. (1991). Modality. In von Stechow, A. and Wunderlich, D., editors, Semantics: An International Handbook of Contemporary Research, pages 639–650, Berlin. de Gruyter. McNamara, P. (1996). Must I do what I ought? (or will the least I can do do?). In Brown, M. A. and Carmo, J., editors, Deontic Logic, Agency and Normative Systems – DEON’96: Third International Workshop on Deontic Logic in Computer Science, Sesimbra, Portugal, 11–13 January 1996, Workshops in Computing, pages 154–173. Springer, Berlin. McNamara, P. (2006). Deontic logic. In Zalta, E. N., editor, The Stanford Encyclopedia of Philosophy. Sloman, A. (1970). Ought and Better. Mind, 79(315):385–394.


141

von Stechow, A., Krasikova, S., and Penka, D. (2006). Anankastic conditionals again. In Solstad, T., Grønn, A., and Haug, D., editors, A Festschrift for Kjell Johan Sæbø – In Partial Fulfilment of the Requirements for the Celebration of his 50th Birthday, pages 151–171. Oslo. Wertheimer, R. (1972). The Significance of Sense: Meaning, Modality, and Morality. Cornell University Press, Ithaca, NY.

On the Temporal Function of Modal Verbs Jacqueline Guéron

The time is out of joint. O cursed spite That ever I was born to set it right. (Shakespeare, Hamlet) Abstract In semantic approaches to modality inspired by Kratzer (1981, 1991), modal verbs are construed as introducing possible worlds. An examination of the syntactic properties of English modals suggests a somewhat different perspective. On the one hand, English modal verbs belong to the class of auxiliary verbs which map spatial configurations defined in VP onto a temporal interval defined in TP, deriving events and states. On the other hand, English modal auxiliaries, like lexical modals in Romance, Germanic, and Old English, function as causative verbs: they bridge the gap between an input spatial configuration holding at some instant of time and an output configuration holding at the next instant of time by introducing an instrument capable of effecting the change of configuration. The grammar provides for two types of causality, intentional causality, in which the instrument of change is triggered and manipulated by a [+human] Agent, and metaphysical causality (cf. Condoravdi, 2002) in which the instrument of change needs no human trigger. It is claimed that causative verbs lack the lexical content motivating an Agent argument, and are thus necessarily associated with metaphysical causality. The [+human] subject of causative verbs, including modal verbs, is thus not to be construed as an Agent but rather as an instrument of metaphysical causality. Key words: Modal verbs, causality, tense, auxiliary verbs.

Introduction1 In this chapter, I will present an analysis of modal verbs as causal verbs. Section 1 sets out the grammatical properties of modal verbs in English. In section 2, I propose 1

I want to thank the members of the audience at the Colloquium on Time and Modality as well as the participants in the U. Paris 8 Temptypac (Temporalité: Typologie et Acquisition) group and the U. Paris 3 English Linguistics seminar for comments on oral presentations of this work. I also thank J. Lecarme for comments on two written versions of the paper and B. Copley and B. Mahapatra for provocative discussion. Université Paris 3 – Sorbonne Nouvelle


143

144

J. Guéron

that causality enters the grammar as a function of tense interpretation. In section 3, I examine the type of causality particular to modal verbs.

1 Modal Verbs Part I: Grammatical Properties 1.1 The Tense of Modals All English verbs except modal verbs show an alternation between 0 and ED suffixes, or, for “strong” verbs, an alternating root vowel, which distinguishes present from past tense. Unlike all other verbs in English, modal verbs may be situated in the deictic present time even when they bear past tense morphology. The modal verb is, as it was in Old English, a perfecto-present verb whose perfect morphology is compatible with a present tense construal. Like other auxiliaries and unlike lexical verbs, modal verbs merge with Tense in syntax in English. I propose that this definitional syntactic property of an English auxiliary corresponds to a definitional semantic property of auxiliary verbs in general: on the interface level of temporal construal, an auxiliary verb mediates the mapping of the situation vP denotes onto the Assertion time in T. In languages like English, in which vP can be predicated of T directly, auxiliary verbs must contribute information which is not otherwise available. Guéron and Hoekstra (1988) propose that the structural skeleton of the sentence is a T(ense)-Chain headed by Comp and including T, V, and auxiliary and light verbs situated between T and V. (The chain also extends into VP to include PP and small clause projections). The Comp node is associated with the Reference time, of which the Speech time is one instance. The Tense node contains an overt or covert tense morpheme which denotes the Assertion time, calculated with respect to the Speech/Reference time. Thus, if the Speech time is indexed Si , and the Assertion time at T is indexed Ti , then the Assertion time is the Present. If the Assertion time is Tj counterindexed with Si , it denotes the Past. There is no Event time. I claim that vP describes a purely spatial configuration which is not construed as an Event until it is mapped onto the Assertion time, either directly, via V-to-T movement, or indirectly, within a T-chain containing a matrix auxiliary verb. Unlike many linguists, I do not define grammatical Aspect with respect to the Event which vP describes, but rather with respect to the time T denotes. I propose that while the Speech or Reference time in Comp is always associated with an interval of time, the Assertion time defines an interval only if a tense morpheme merges with an aspect morpheme in T. Otherwise, T denotes a point of time (cf. Guéron, 2004). A state described in vP is construed as a simple spatial configuration which can be mapped onto a single point of time in T. But an event denotes a change from one spatial state or configuration to another, so it must be mapped onto an interval of time.

On the Temporal Function of Modal Verbs

145

The hypothesis that an aspect feature or morpheme is necessary to create a time interval in T accounts for the contrast between French “Je te parle” and English “∗ I talk to you”, if we further assume that in English, lexical verbs lack a feature for aspect. While all verbs bear aspect in French, only certain auxiliary verbs bear aspect in English. Auxiliary have has both tense and aspect features. When tense is present, have defines a time interval in T which holds at the Speech time and extends into the past. Modal verbs with no incorporated ED morpheme, such as can or may, bear features for present tense and aspect. These modals define a time interval in T which holds at the Speech time and extends into the past when the VP complement denotes a state and into the future when the complement denotes an event. Modal verbs incorporating an ED morpheme, such as could or might, bear an aspect F and an underdetermined tense feature which defines either a present or a past time depending on a local reference time. The tense of the modal in (1) is ambiguous between a shifted past and a present (double access) construal. (1)

John said Mary could/might/should leave (if she wants/wanted to).

The creation of a temporal interval in T is thus not a property of modal verbs as such (cf. Condoravdi, 2000), but of all verbs which incorporate an aspect morpheme in addition to tense. The ED morpheme can situate the modal verb in a counterfactual world if it is in the scope of an operator like wish or counterfactual if, as in (2a). But any other stative verb in English can do the same (cf. (2b)). As Iatridou (2000) argued, counterfactualty is a function of past tense morphology, so it is independent of the grammar of modality. (2)

a. b.

Mary wishes she could marry John. Mary wishes she knew Sanscrit.

Aspectual structures define either imperfective or perfective aspect. I claim that these construals are not complementary; but, rather, that perfectivity subsumes imperfectivity. An imperfective verb introduces a time interval in T. A perfective verb also introduces a time interval in T and it adds something more, a boundary, to that interval. The greater complexity of perfective aspect is shown by the fact that while there is only one way to derive an imperfective temporal interval, that is, by the merger in T of a verb with both tense and aspect features, languages exhibit a variety of ways to derive a perfective temporal interval. Sometimes the imperfective and perfective verbs do form suppletive pairs, as seems to be the case in Arabic. Alternatively, the perfective verbal form is morphologically more complex than the imperfective form. In Russian, in general simple verbs are imperfective and perfectivity is derived by the idiosyncratic adjunction of a telic particle to the verbal root. In English, perfectivity is derived in syntax by the merger of auxiliary have, which contributes imperfectivity, and a past participle, which introduces a boundary on the imperfective interval. Similarly, modal verbs, which are inherently imperfective, select an

146

J. Guéron

aoristic infinitival complement which bounds the interval they define either at the Assertion time (for states) or at a posterior time (for events). In many languages, a perfect participle has different construals on different syntactic levels: it defines a bounded event in an embedded position but a bounded time interval if it raises to T. In English, the sentence “John has walked in the park today” defines a completed event located within the present time interval, while “John walked in the park yesterday” defines an event, completed or not, situated in a bounded past time interval. Arabic likewise uses the same verbal form as a participle denoting a completed event in a lower syntactic domain and as a perfective verb indicating a bounded past time interval in the matrix T. Russian used the same mechanism in diachrony. So did Old English, but with a twist. The class of Old English lexical perfecto-present verbs, from which the Modern English modal auxiliaries descend, were construed, despite their perfect morphology, as defining an imperfective present tense. An extra morpheme was added to denote past tense.

1.2 Absence of Agreement Morphology All English verbs except modal verbs show a suffixal alternation between 0 and s which constitutes an agreement paradigm for person/number. I account for the absence of agreement morphology in English modal verbs by the hypothesis that agreement morphology is referential in English, while modals are impersonal verbs, incompatible with referential agreement. In the Minimalist framework (Chomsky, 1995), DP raises to T to check the Dfeature of T. I propose that the shared D-feature is a “nominal Tense” F which is [−interpretable] in T but [+interpretable] in DP. A definite DP which checks its D-feature in Spec TP is associated with a biography, an interval of time which starts in the Reference time interval associated with C and extends up to the point or interval of the Assertion time in T. A definite DP in Spec TP thus assures temporal continuity between the Reference time and the Assertion time. A number of arguments support the hypothesis that definite subjects are associated with a biography which participates in the tense construal of the sentence. Lecarme (2004) showed how the tense morphemes in D in the Somali DP interact with sentential tense. Even in English, where friends means “more than one friend”, not “present friend”, evidence for nominal tense as an active part of sentential tense construal is rampant. (i) The double access construal of (3), like that of (1) above, requires John and Mary to have inhabited the discourse world continuously from the past time of John’s speech act to the present time of the asserted pregnancy (or Mary’s ability to leave in (1)). (3)

John said/announced/declared that Mary is pregnant.


147

(ii) Present tense auxiliary have selects a referential subject whose biography must stretch from a past time to the Speech time, as shown by the ungrammaticality of (4a). A metonymous subject, for which a biography is irrelevant, as in (4b), need not check a person F in Spec TP. (4)

a. #Einstein has lived in this house. b. Einstein has transformed modern physics.

(iii) Various modifiers can provide the referent of a DP with a biography that began before and continued up to the Assertion time. While the individuals referred to in (5a–c) are no longer fugitives, a high school student or Sue’s husband, their existence as individuals in the discourse world is continuous from past to present. (cf. Enç , 1987). (5)

a. b. c.

All fugitives are now in jail. Imagine! I went to high school with the old man who is sitting on the sofa. Julie married Sue’s former husband.

An indefinite DP in Spec TP checks the D feature of T but it is not assigned a biography which extends to the Reference time interval, perhaps because its D node lacks referential features. Unless it bears additional specifications, an indefinite nominal is associated solely with the point or interval of the Asssertion time. A definite DP in Spec TP also “agrees” with the finite verb in T: the DP checks its [+interpretable] phi features with the [−interpretable] phi features of V in T. I propose that a DP which checks a referential person F, varying over the three persons, is construed at the interface as an individual endowed with the psychological properties of consciousness, perception, and will. The entire sentence is then presented from the subject’s “point of view”: the event unfolds as the subject perceives it. In the impersonal sentences (6a–c), the subject checks an invariable third person sg. F which lacks referential import. (6)

a. b. c.

It is raining. There is a man at the door. It is dangerous to lean out the window.

Modal sentences are also impersonal. A modal verb selects no subject. Its surface subject is derived by raising in syntax from its small clause complement, as is shown by the traditional test of expletive or idiom chunk subject raising in deontic (7a–c) and epistemic (8a–c). (7)

a. b. c.

There must be seven chairs around this table by four P.M. It must rain soon or the crops will be destroyed. The cat must be let out of the bag.

(8)

a. b.

There must be something wrong with him. It must rain a lot here for the grass to be so green.

148

J. Guéron

c.

Time will fly, like it or not.

In (9) below, the raised definite DP checks its D, or nominal tense, Feature with the D-Feature of T and is assigned a biography. But the absence of person morphology on the verb prevents the DP from checking its person F. Consequently, the subject is not associated with the psychological properties of intension and will and the sentence does not lie in its point of view. (9)

John must leave. a. e must [John leave] b. John must [t leave]

(d-structure) (spell-out)

In Russian, while verbs of ability (moch’) and wanting/wishing (xotet’) agree in person with their referential subject, the core modal verbs denoting necessity and possibility exhibit impersonal 3rd person sg. agreement. (10)

a. b.

Nyzhno/nado idti domoi. (it) is-necessary to go home. Zdec’ mozhno chitat’. Here (it) is possible to read.

In Arabic, modality is expressed either by an impersonal structure “it is necessary that S”, as in French or English, or by an invariable particle which governs an imperfective lexical verb. (11)

qad ?a-cu:d-u ?ila: fa:s may I return+ind. to Fez (I may return to Fez)

Lack of variable phi features in modal verbs or particles in English, Russian, and Arabic alike ensure that a modal sentence is not placed in the scope of the subject. In the Romance languages, where modal verbs do agree in person with their subjects, morphological agreement is not necessarily referential. In French, the same morpheme is used both for referential [+human] and for non-referential or [−human] third person pronouns. English must distinguish these cases. (12)

a. b.

Il pleure. He is crying.

(13)

a. b.

Il (le livre) est sur la table. It/∗ he (the book) is on the table.

(14)

a. b.

Il pleut. It/∗ he is raining


149

1.3 The Temporal/Causal Function of Modal Verbs In semantic approaches to modality inspired by Kratzer (1981, 1991), modal verbs introduce possible worlds. Here I will explore the hypothesis that the semantic function of the modal verb is not to introduce alternative worlds in which the complement proposition is true, but rather to illustrate how the hypothetical situation the vP denotes can be introduced into the ongoing deictic world. I claim that the mapping of an event onto time implies a grammar of causality, and that modal sentences occupy a unique position within that grammar.

2 Causality in Grammar 2.1 Two Syntactic Levels of Interpretation I assume two levels of sentence interpretation, analogous to the phases of Chomsky (2001), the vP-VP level and the CP-TP level. Syntactic constituents are subject to spatial construal in vP and temporal construal in TP (cf. Guéron, 2004, 2005, 2006). V to T movement unifies the two levels in syntax or in Logical Form. Within vP, a stative verb describes a simple spatial configuration of inclusion of a Figure (Fig) in a Ground (Gr), as in (15a, b). At the TP level of construal, the spatial configuration is predicated of a single point of time (or extended without change over the points of an interval of time). (15)

a. b.

John (Gr) has blue eyes (Fig) There (Gr) is a man in the room (Fig)

An eventive verb denotes a change of state. It thus implies at least two distinct spatial configurations, one which pertains before the change and another which pertains after it. In (16), two non-identical states, Si and Sj, are predicated of two intrinsically ordered points of time. The bracket marks the instant of assertion time which pragmatically bounds the time line, identifying tn and thus tn-1. The states predicated of these points are then themselves ordered as Sn and Sn-1. (Other states predicated of other ordered points of time may precede Sn-1 and Tn-1.) (16)

Points of time States Mapping of s on t

(t t t) tn-1 tn] (s s s) Si Sj (s s s) Sn-1 Sn]

I assume a well-formedness condition like (17) at the syntax-semantics interface. (17)

A state Sn-1 mapped onto a point of time tn-1 must contain whatever is necessary to bring about Sn at Tn.

150

J. Guéron

As a consequence of (17), Sn-1 is construed as the cause of Sn, and Sn as the effect of Sn-1. (16 )

Points of time States Mapping Construal

(t t t) tn-1 (s s s) Si Sn-1 Cause

tn] Sj Sn] Effect

Points of time are intrinsically ordered; states are unordered. Yet the grammar does not allow any random two states to be mapped onto two contiguous points of time.2 Rather, the second state must be licensed by the content of the first. I define causality in the grammar as an asymmetric relation between two states such that the content of the first licenses the existence of the second.

2.2 Point of View Events have initial and final boundaries; a temporal gap thus separates the end of one event from the beginning of the next. A temporal gap also separates Sn-1 from Sn internal to the event. The actual change of state takes place during this gap. Yet speech time is construed by human beings as a continuity, without gaps, as in the expression “time flies”. How can a discontinuous succession of states and events be predicated of a continuous speech time?3 The problem is solved by the phenomenon of “point of view”. Since time is continuous as perceived by a consciousness, temporal continuity can be maintained, despite the gaps which separate states and events, if every situation is construed as located in the scope of a sentient point of view. The grammar provides two sentient points of view, that of the subject which obtains psychological properties by checking a person F in Spec TP, and that of the speaker whose consciousness is implied by the speech act. The struggle between subject and speaker to impose their point of view is part of the drama of the grammar. I propose that the grammar arbitrates the conflict at the interface as in (18): (18)

Interface arbitration: whenever an event description in the syntactic scope of the subject contains no temporal gaps, the sentence is construed from the point of view of the subject. If the event description contains gaps, the speaker must assure temporal continuity.

I will illustrate both cases in this text.

2.3 The Instrument Let us call whatever Sn-1 must contain in order for it to be construed as the cause of Sn an instrument. An instrument must be both efficient and accessible. When VP 2 3

On randomness, see Carlson, this volume. Schwer (2006) discusses the mathematical incoherence of the notion of “discrete continuity”.


151

denotes a physical event, there is no more efficient and accessible instrument than a body part. For example, in the sentence “John hit Bill”, the verb hit implies a body part instrument, a hand. I propose that when V raises to v, V checks its instrument content, or [+interpretable] instrument F, with the [−interpretable] instrument F of v. So we may call the little vP which dominates the VP nucleus the “Instrument Phrase”. The instrument content of V is the source of syntactic transitivity: the instrument selects the object it affects in VP, and a manipulator which handles it in Spec vP.

2.4 Two Types of Causality I claim that the grammar allows for two types of causality, intentional causality and non-intentional or metaphysical causality.4

2.4.1 Intentional Causality A sentence has intentional causality when a [+human] subject which checks a person F in Spec TP, and is thus construed as a conscious entity, targets an object with the intention of introducing a change in that object. Intention precedes the event, but it does not suffice to realise the event. John may want that book but he needs to extend his hand to take it. Nor does an instrument suffice without an intention. In (19a), John’s hand is the instrument which brings about a physical change of state in Mary’s hand. Yet (19b) is ungrammatical. (19)

a. b.

John pinched Mary’s hand. hand pinched Mary’s hand.

∗ John’s

Sn-1 must thus imply both an instrument and an intentional subject. In (20), I deconstruct the construal of (19a) at the interface into two temporal phases. (20)

John pinched Mary’s hand. a. Sn-1: psychological cause in TP (i) John targets Mary, focusing on her hand, with a goal or intention in mind. (ii) John selects his own hand as an instrument for realizing his goal. b. Sn: physical effect in vP:

The goal is realized as Jean’s hand makes contact with Mary’s hand (creating a pinching Figure). The corresponding derivation in (21) is constructed in syntax bottom-up but construed top-down.

4 Thanks to Bridget Copley and Jacqueline Lecarme for discussion of this term, which I have adapted from Condoravdi (2002).

152

J. Guéron

(21)

John pinched Mary’s hand. TP T’

Spec

vP

T Past John (Intender)

v’

Spec

VP

v John (hand) Manipulator

V pinch

DP Mary’s hand (targeted object)

An “Agent” argument such as John in (21) combines two distinct event functions, that of manipulator, assigned in Spec vP, and that of an intentional consciousness, assigned viaperson Fchecking in SpecTP.The“Agent”theta-roleisthusnotasyntactic primitive.Rather,itistheproductofinternalmergerinsyntaxofamanipulatorargument licensed in Spec vP and a psychological argument licensed in Spec TP. There exist manipulators without consciousness, such as the subject in “The storm destroyed the crops” and conscious arguments which are not manipulators, like Experiencer me in “That hurt me”. Neither argument can be construed alone as an Agent. The body-part instrument implied by V may be extended by an artifact tool. By metonymy, the tool may even be credited with bringing about the event. (22)

a. b.

John opened the door with a key. The key opened the door.

In (23a–f), the instrument is indicated in parentheses, along with the artefact which extends it. (23)

a. b. c. d. e. f.

Bill took the book. Max kicked Bill. John kissed Mary. Tim washed his shirt. Sam shot Bill. John built a house.

(hand) (foot) (lips) (hands+soap and water). (hand+gun) (hand+tools+materials)

Intransitive verbs such as walk, sing, etc., imply a covert body part instrument whose movements, directed by an intentional subject, create a Figure such as a walk, a song, etc. In English, the preposition with introduces a tool which extends a body-part instrument, as in (22a). In languages with serial verbs, like Mandarin or certain creoles, the first verb can denote the tool selected in Sn-1 while the second denotes the effect in Sn. The verb take In (24a, b) has the same function as with in English (22a).


(24)

a.

b.

153

wo na dao qie cai. (Mandarin) (from den Dikken and Sybesma, 1998) I take knife cut vegetables (translation: I cut the vegetables with the knife) mi tei faka koti b.eée. I take knife cut bread. (Saramaccan creole) (from Bickerton and Iatridou, 1987) (translation: I cut the bread with the knife)

The instrument Feature of the verb may be realized by a verbal affix. In Japanese, the so-called “lexical causative” morphemes e and as, which are idiomatically assigned to a verbal root, derive a transitive verb from an unaccusative verb (Shibatani, 1975).5 (25) stand move

non-causative tat-u ugok-u

causative tat-e-ru ugok-as-u

2.4.2 Metaphysical Causality A change of state at tn must have a cause at tn-1. The new state has a metaphysical cause whenever it is triggered by some law governing the ordering of events in the universe independent of human will. 2.4.2.1 In the canonical case, the metaphysical cause is a law of nature which affects an inherent property of an object over time.6 (26)

a. b. c. d. e.

The branch broke. The leaves are falling. Mary’s hair turned grey. John became ill. It is raining, snowing, hailing, etc.

If, as we claim, all eventive sentences must be placed within a human point of view, in order to ensure temporal continuity, then the sentences of (26), which lack a selected subject, must lie within the point of view of the speaker. Indeed, such sentences invariably describe events which can be perceived by a conscious observer external to the event space. The hypothesis that metaphysical causality is associated with the speaker’s point of view accounts for the surprising fact that verbs depicting the directed motion of a Figure on a Ground are presented from the perspective of the speaker’s position in space and time, even when the moving Figure is a human being. 5

See also Travis (2000, 2006). The immediate cause of the branch breaking or the leaves falling in (26a, b) is, to be sure, a physical force, like the wind, not a metaphysical force. I claim, however, that the wind is to the metaphysical order of events in (26a, b) as John’s hand is to the intentional order of events in (19a): it is an instrument which mediates between a physical change of state and its higher-level cause. 6

154

(27)

J. Guéron

a. b.

John has not yet arrived. Betsy has disappeared!

I suggest that a cognitive “uncertainty principle” like (28) restricts the form of the grammar: (28)

Uncertainty principle: an individual located at one point of a spatio-temporal trajectory is unable to perceive the initial and end points of the trajectory at that same point of space and time.

Only a conscious being outside the trajectory can perceive the Figure being formed by the mover.7 A verb of motion lacking extra lexical content implying intentionality thus reduces to the primitive concept GO; the SOURCE and GOAL of the trajectory are supplied by the speaker. The “uncertainly principle” in (28) can resolve the “progressive paradox”. In (29), how can John truly be said to be engaged in a cross the street event if the event was never completed? (29)

John was crossing the street when a truck hit him and he died on the spot.

We suggested above that auxiliary verbs mediate the mapping of the situation the vP denotes onto the time in T. I propose, more precisely, that auxiliary verbs translate the spatial information in vP into temporal information, defining, in particular, the shape of time in T as a point or an interval. In (29), ING is a kind of aktionsart operator which targets vP and makes its individual spatial configurations visible for construal. Auxiliary be, which is punctual in English, focuses one spatial configuration and maps it onto a single point of Assertion time. I propose that an operator which focuses one of the series of states which constitute a spatial trajectory places the other states of the trajectory in a complement set construed as part of the presupposition of the sentence. In (29), it is not John, stranded at a single point of space and time, but the speaker, who has access to the presuppositional complement set which includes the end point of the trajectory. The speaker may even impose an interpretation of the goal of the subject’s trajectory which contradicts what the subject himself may think he is doing. (30)

a. b.

When you drive too fast, you are driving to your death. You are wasting my time with your stupid questions.

The principle in (28) also accounts for the fact that a speaker, engaged in a narrative trajectory, cannot map either the initial point or the end point of an event onto the point of Speech time at which she is located (cf. Giorgi and Pianesi, 1997; Copley and Smith, this volume). The sentences of (31) and (32) cannot be mapped onto the 7 It is true, as several members of Temptypac noted, that a [+human] figure engaged in a spatiotemporal trajectory can remember where his trajectory began and can intend it to reach a certain end point, but these intentional properties are weaker than perception, for the intended closure may fail to pertain.


155

deictic present instant of time either in English, which lacks an imperfective present tense, or French, which has one. (31)

a. b.

∗I

(32)

a. b.

∗I

eat at noon. mange a` midi.

∗ Je ∗

(Deictic present tense, said at noon) (idem)

find a coin now. Je trouve une pièce maintenant.

2.4.2.2 The little v node: The mapping of a state Sn onto a point of time tn requires that Sn-1 predicated of tn-1 contain an instrument capable of effecting the change of state. If the little v node identifies the instrument of change, then little v must exist independently of causality type. In the absence of an instrument F in V which selects a manipulator in Spec vP, the v instrument head functions as a free variable. This variable must be bound outside the vP domain by a constituent which implies the point of view of the speaker. Impersonal sentences contain evidential elements which bind the little v node by conveying the speaker’s visual perception or personal judgement of the situation VP describes. In (33), the grammatical sentences contain lexical items which convey either the speaker’s visual perception of a situtation, as in (33a–c), or her personal judgement of its feasibility, as in (33d, e) or its advisability, as in (33f, g). The ungrammatical sentences lack appropriate lexical items. (33)

a. b. c.

d. e. f. g.

John went ∗ (to the shore). La Tour Eiffel se voit ∗ (d’ici) (The Eiffel Tower is visible ∗ (from here)) (i) ∗ Il a e´ té dormi. (it has been slept) (ii) Il a e´ té beaucoup dormi dans ce lit. (It has been slept a lot in this bed, i.e. I can see that this bed has been slept in) It is easy/∗ rectangular [to draw a four-sided figure]. Fireman are available/∗ hungry. It is wrong/∗well-known to steal. It is dangerous/∗scenic to look out the window.

2.4.3 On Distinguishing Causality Type All sentences lacking a referential subject are associated with metaphysical causality; but not all sentences with metaphysical causality lack a referential subject. A referential object may raise to Spec TP to check the D feature of T without checking the person F necessary for intentional causality. As modal verbs cannot check the person F of their subject, there is no ambiguity when the main verb is a modal. But when the main verb has agreement inflection, ambiguity can arise, as in (34a–c) which describe either an intentional or an unintentional change of state.

156

(34)

J. Guéron

a. b. c.

John hit Bill (with his fist/when he fell) John locked himself in the closet (by turning the key/when he closed the door) Jean s’est tué (avec son fusil de chasse/en traversant la rue) John killed himself (with his gun/crossing the street)

Does such ambiguity entail that all sentences basically have metaphysical causality, with intentional causality a pragmatic subcase of metaphysical causality?8 While the above statement is very likely true from the point of view of deterministic science, it does not describe how causality functions in natural language. Two distinct types of causality can in fact be distinguished at the sentence level by construal, by the lexical content of V, and by grammatical mechanisms. 2.4.3.1 A lexical verb cannot be associated with intentional causality unless it denotes an event with internal temporality. Internal temporality is in turn a function of the instrument content of V (or VP). Thus, hit denotes a punctual event in (34a) when it is construed as describing an unintentional physical contact, but it has internal temporality when it implies the existence of an initial state in which the subject targets the object and selects his hand as an instrument to achieve his goal. Similarly for the b. and c. sentences. Verbs which denote an event which lexically implies several stages cannot be construed as unintentional. To murder, to strangle, or to choke someone implies both a body-part instrument and a preparatory stage, as do, on a happier note, to sing, to laugh, or to tell stories. 2.4.3.2 Sentences with indefinite subjects have metaphysical causality. An indefinite subject lacks a biography; so it is associated only with the Assertion time. In order to ensure temporal continuity between the Reference time and the Assertion time, the sentence must be placed within the speaker’s biography. This in turn requires that the sentence define a configuration so distributed over the speaker’s field of perception that it takes her time to perceive it (cf. Guéron, 2005). (35)

a. b.

∗ Two

(36)

a. b.

∗ Water

(37)

a. b.

∗ Books

bells are in the church. Two bells are ringing.

is wetting Bill. Water is running from the sink.

are on the floor. Books are spread all over the floor.

2.4.3.3 Verbs which lexicalize the body part instrument, the tool, or the spatial goal ensure intentionality by collapsing several phases of a spatio-temporal trajectory into one, thus eliminating event-internal gaps which invite speaker intervention. (22b), repeated below, is one example. Others are in (38). (22b) 8

The key opened the door.

Thanks to Bibhuti Mahapatra for raising this question.


(38)

a. b. c. d. e.

John handed Mary the book. John elbowed Jim out of the way. Joe hammered the nail flat. John landed the ship. Jay and Ken shelved the books.

157

(body part instrument) (body part instrument) (instrument + tool) (instrument + goal) (instrument + goal)

2.4.3.4 Generic sentences, which contain a non-deictic tense in T presumably bound by a Generic operator in C, have metaphysical causality. A [+definite] subject which checks the D-F of a non-deictic tense morpheme in T is associated, via the inital C-T link of the sentential T-chain, with the Speech or Reference time and is thus presupposed as existing in the discourse world. But it is not associated with any Assertion time. The sentences in (39a–c) describe an individual-level property of the inanimate subject without asserting that the property is realised. (Example (39b) adapted from Carlson (1995)). (39)

a. b. c.

This car does 170 mph. This machine squeezes oranges and removes the pits. Those bureaucrats bribe easily.

In a habitual sentence, the definite subject has a biography which stretches from the Reference time to a deictic Assertion time. A [+human] subject can check its person F in Spec TP and acquire psychological properties. When tense is punctual, as present tense always is in English, the event vP denotes may not be predicated of any Assertion time interval. However, the speaker can use a “habitual” version of the always-accessible narrative operator, one which construes the Reference time in Comp as defining, not one uninterrupted time interval, but a series of distinct temporal occasions. The event mapped aoristically onto one point of time at T is then predicated of each of the occasions in the series associated with C. When we construe sentences like (40a, b) as “dispositional”, we associate with the [+human] subject the psychological properties of consciousness and will that it acquires by checking a person F in Spec TP. (40)

a. b.

John smokes. Bill watches tv every night.

Purely grammatical mechanisms also distinguish intentional from metaphysical causality. 2.4.3.5 Kuroda (1972), who distinguished categorical sentences, which select a semantic subject, from thetic sentences, which do not, claimed that in Japanese, the subject affix wa identifies the first type of sentence while the affix ga identifies the second type. 2.4.3.6 In many Romance and Germanic languages, auxiliary HAVE selects the past participle of a verb which has an instrument F and normally selects a manipulator argument in Spec vP, while Auxiliary BE selects the participial form of a verb which lacks an instrument F. Have is a transitive verb in all its uses. As an auxiliary, have

158

J. Guéron

selects a temporal subject in Spec TP capable of identifying the implicit manipulator argument of the embedded participle. The verb be is, on the contrary, unaccusative in all its uses. It selects no subject in Spec TP, leaving the v node as a variable to be bound by an evidential element revealing the presence of the speaker. (41)

a. b. c.

E` piovuto (it is rained) E` venuto Gianni (is come Gianni) Gianni e` venuto.

All sentences taking auxiliary BE have metaphysical causality. But, surprisingly, so do many sentences taking HAVE. Folli and Harley (2007) discuss a possible agentive construal of the subjects of (42)–(43). (42)

a.

(43)

b.

Il treno ha fischiato. The train has whistled questo tavolo ha scrichiolato this table has squeaked.

(Italian) (Italian)

In the framework presented here, the criteria for an Agent are very restrictive. Only a manipulator argument which acquires consciousness and intention when it raises to Spec TP can be construed as an Agent. The sounds emitted by inanimate objects are due to their properties, not their intentions. A verb of sound emission does have an instrument content; it is this, I propose, which licenses auxiliary have. However, the lexical content of the verb does not imply a manipulator which handles an instrument in vP, but rather a perceiver located outside the spatial configuration. The temporality of the event of a train whistling, contrary to that of a boy whistling, depends on the perception of the hearer. In (42), whistle does not refer to a tool somewhere in the train, but names a percept, as is clear for squeak in (43). Configurations which are introduced into the discourse world via visual perception lie similarly in the scope of a conscious speaker. The speech mechanism which announces the speaker’s presence in these cases is not judgement, but metaphor. (44)

a. b.

The river runs/goes/descends/falls into the sea. The road drops down sharply/ stops abruptly/ forks in ten minutes.

Other configurations rely on the sense perceptions of smell or touch. (45)

a. b.

This cheese smells bad. This hot water feels good.

We may define auxiliary choice as an inter-phase mechanism which is sensitive to the presence or absence of a lexical instrument F in V. This is independent of the subject’s checking a person F in Spec TP, which is necessary for Agentivity. 2.4.3.7 Targeting marks intentionality. The initial step in intentional causality is the targeting by a subject which checks its person F in spec TP of the [+human] or


159

[−human] object it intends to affect.9 The subcase of “Person-to-person” targeting occurs when a [+human] argument x targets another [+human] argument y in its spatial domain. Argument x then includes argument y within its own temporal and psychological domain. Argument x may check the person F of argument y and affect y psychologically, whether or not x ever touches y in the spatial portion of the event defined in vP (cf. Guéron, 2005). On the TP level, person-to-person Targeting bridges any temporal gap separating two arguments. In English, the particle TO functions as an aktionsart or aspectual link, depending on the level of construal, between the disjoint spatio-temporal phases of an event trajectory. When TO introduces a goal in VP within the syntactic scope of the subject, it marks intentional causality in TP and disallows inanimate subjects. There is no spatio-temporal gap in such sentences, as TO extends the trajectory defined by the verb. (46)

a. b. c.

John gave a book to Mary. John sent a package to Philadelphia. ∗ The war gave a book to Mailer.

In a double-object structure without TO, the event description contains a temporal gap between the moment the subject conceives his intention and the result state of the event. A [+human] subject can bridge this temporal gap by targeting a [+human] object so that the two arguments overlap in time. Targeting licenses Mary in (47a) but not Philadelphia in (47b). (47)

a. b.

John sent Mary a book. sent Philadelphia a package.

∗ John

If the subject does not check a person F, the double-object sentence is construed as causative. A causative sentence with inanimate subject has a metaphysical causality, as we argue below. (48)

a. b.

The war gave Mailer a book. The years have given this copper kettle its patina.

In Spanish, the “expletive” preposition a marks the targeting of a human object by a conscious subject. This a is not found in unaccusative sentences. (49)

Vi a Mafalda. ((I) saw Mafalda)

(50)

∗ Lleg´ o

a Mafalda. (arrived Mafalda)

In French (and Italian), a DP introduced by a has both a [+locative] and a [+human] F. Such a DP functions as a benefactive argument in an unaccusative sentence but as a targeted [+human] goal in a transitive sentence. 9

The importance of targeting was pointed out to me some time ago by Yves D’hulst.

160

(51)

J. Guéron

a.

b.

(52)

a. b.

La tête me tourne. (The head to-me turns) (My head is spinning) Il m’arrive une catastrophe. (It happens to me a catastrophe) (A catastrophe is happening to me) J’ai envoyé un livre a` Marie. (I sent a book to Marie) Je lui ai cassé sa tirelire. (I to her broke her piggybank) (I broke her piggybank (on her))

While “à DP” indicates targeting and therefore intentional causality in an eventive sentence, “par DP” indicates metaphysical causality. According to Kayne (1975), only the “faire a` ” structure in (53b) requires the subject and the prepositional object to be in the same time and place. (53)

a. b.

Jean fera construire cette maison par ses futurs petits-enfants. #Jean fera construire cette maison a` ses futurs petits-enfants.

2.4.3.8 An argument which occupies two positions within a single event trajectory asssures intentional causalty by eliminating temporal gaps. Clitic doubling and inalienable possession exemplify this grammatical strategy. In (54a, b), the subject targets one occurrence of the object spatially and its other occurrence psychologically. (54)

a. b.

Le d´ı un caramelo a Mafalda. ((I) to-her gave a candy to Mafalda) Je lui ai pris la main. (I took to her the/her hand)

Anaphoric clitics, possessive pronouns, and zero anaphora allow the subject to eliminate spatio-temporal gaps by targeting himself. (55)

a. b.

c. d. e.

Me com´ı une manzana. (I ate me an apple) Jean s’en va. (John self + from it goes) (Jean leaves/is leaving) Bill kept his nose to the grindstone. Mary danced her way to fame and fortune. Bill dressed/washed/shaved.

Serial verb languages indicate intentionality by placing the object at two spatial points of the same event. In (56) and (57), the light verb take targets the object in vP while the lexical verb affects it in VP. In (58) I essentially adopt the structure of den Dickken & Sybesma (1998).


(56)

koky so aso do tavo-ji Koku take crab put table-on

161

(Fongbe)

(57)

wo ba Zhang San gan-zou-le (Mandarin) I take Shang San chase away -PERF.

(58)

a. b.

e [vP DP TAKE [VP VERB NP PP]] (d-structure) e [vP DP TAKE [vP NPi [VP VERB NPi PP]]] (spellout)

Person-to-person targeting is a relation of (positive or negative) anaphora, and it obeys locality conditions. In control structures, targeting and binding interact to assure temporal continuity. In (59a), below, the intentional [+human] subject of the matrix sentence targets the [+human] object in order to affect it psychologically by means of a speech act and get it to function as an instrument to realize the event the embedded sentence denotes. The targeted object then bridges another temporal gap by binding the PRO subject of the infinitive. The same referent thus occurs at two different temporal points of a trajectory which starts with the intention of the matrix subject and ends with the action the embedded sentence denotes. In (59b), the conscious subject targets either the hearer or herself, so that she appears at either two or three spatio-temporal points of the trajectory. In (59c), Mary is the targeter in the matrix and the instrument of the embedded event. (59)

a. b. c.

Mary convinced/forced/asked John [PRO to leave]. Mary promised (herself) [PRO to return]. Mary threatened (Jim) [PRO to leave].

2.4.3.9 A note on speaker intervention: Unaccusative sentences have metaphysical causality, yet they may contain intentional adverbs. (60)

a. b.

John left intentionally. Jean est parti exprès.

I claim that the intentional adverbs in (60) express the speaker’s judgement of the subject’s motivation, not that of the subject. When intentionality is lexically determined, it is redundant to add an adverb. (61)

a. b. c.

# John murdered Bill intentionally. # Mary intentionally sang a song. # Susy intentionally handed Bill a book.

2.5 Metaphysical Causality on Higher vP Levels A sentential T-chain may contain a number of vP-type levels between the instrument v and T. Since all of these levels are higher than the vP which contains the manipulator, they all induce metaphysical causality unless another grammatical mechanism is available to impose intentional causality.

162

J. Guéron

One such higher vP level contains aktionsart operators which focus a single state or subevent of the event VP describes. These include progressive BE and the related verbs of (62). (62)

John began/started/stopped crying.

A still higher vP level contains aspectual verbs in the scope of the speaker which place a situation on one or more points of the Assertion time interval, such as English “used to” or Spanish “soler” (cf. Laca, 2004). Between these two levels lies a vP level which contains causative and perception verbs. Romero (2005) argues that it also contained perfecto-present “pre-modal” verbs in Old English. (63)

a. b. c.

John made Bill mow the lawn. John saw Susie cry. pet heo sculen . . . heore bileue cunnen (cited in Lieber, 1982). (that they should+pl their belief can) (that they should know their belief)

On the usual assumption that ‘before’ in the string means ‘higher’ in the structure, the causative/perception vP level is higher than the aktionsart vP level, which is itself higher than the Instrument vP level in which the manipulator is selected. (64)

a. b.

Bill saw John begin to mow the lawn. The teacher made John stop handing Mary notes in class.

Aspectual auxiliaries are in turn located higher than the causative/perception level. (65)

Mary used to make John stop handing Mary notes in class.

Causative and perception verbs do not select an Agent. Rather, they select an instrument in the Spec position of the vP level they define. While the instrument selected by a lexical verb in vP affects a change in an object in space, the instrument selected by perception or causative verbs, which lack spatial content and are merged higher than Instrument little vP, introduces an event in time. The morpheme which denotes a body part instrument depends on the lexical content of the verbal root, but a causative or perception verb does not. Thus Japanese possesses, alongside instrument affixes as and e which vary idiosyncratically with the root verb, a “productive” causative light verb (s)ase (Shibatani, 1975). (66) stand fly

non-causative tat-u tob-u

lexical causative tat-e-ru tob-as-u ∗ tob-e-ru

productive causative tat-ase-ru tob-ase-ru

The temporal instrument selected by a causative or perception verb in its Specifier position derives non-intentional or metaphysical causality, by minimality. The selected subject blocks the raising of the manipulator subject in the lower spec vP


163

to Spec TP. The subject of see in (67a) supplies evidence for the existence of an event and the subject of let in (67b) facilitates the event, but clearly, neither wills the event. (67)

a. b.

I saw John leave the house at 3. I let John leave the house.

In the simplest case, the verb make functions as a kind of copula which asserts a temporal property of its subject. (68)

a. b. c.

Two and two make four. John makes a good husband. John made an error.

When make takes an eventive VP complement, causality is clearly metaphysical when the subject is inanimate, as in (69). (69)

a. b.

Boredom made John leave the party early. That curtain makes the living room look too dark.

In other cases, however, make seems to define an intentional event in the scope of a willful subject, as in (70a). The ambiguity of (70b) would depend on the choice of stative or eventive construal of the complement VP. (70)

a. b.

John made a pie. “This missionary sure makes a good soup” said the cannibals.

Causality is ambiguously intentional or metaphysical in English (71a) and French (71b). (71)

a. b.

John made Bill read the book (by threatening him/with his enthusiastic recommendation). Jean a fait lire le livre a` Bill (par des menaces/en le recommandant chaleureusement).

I have identified an Agent as a manipulator argument in Spec vP which raises to Spec TP and checks its person F, acquiring consciousness and will. If so, the subject of causative verbs, which lack an instrument feature and are merged higher than the instrument vP projection, can be neither a manipulator nor, by this definition, an Agent. Rather, I propose, the subject of a causative verb is, basically, an instrument of metaphysical causality as perceived by the speaker. In (70a) for example, John may well have made the pie under duress. However, (72a, b) do have intentional readings. Moreover, while the fact that causative verbs like make or let do not necessarily take an animate subject supports the hypothesis that these verbs are not lexically agentive (provided with an instrument F), the fact that causative have cannot take a non-human subject or object would seem to go in the other direction.

164

(72)

J. Guéron

a. b.

John/∗ boredom had Mary leave the party early. John had Bill/∗ the car move faster.

Note that have is not a copula, like be, but an asymmetric predicate which lexically denotes spatial inclusion (for some space), and whose object is always construed as lying in the scope of its subject. In vP, the object in the spatial scope of the subject defines possessive have. Above vP, the object in the psychological or temporal scope of the subject defines experiencer have, causative have, or auxiliary have. It seems that whenever have raises to TP in an episodic sentence, even as a possessor, the psychological construal is inevitable and have must check a person F in T. To account for the intentional readings of (71a, b) and (72a, b), I propose that person-to-person targeting by a subject which checks a person F in Spec TP can create a construal of intensional causality in an eventive sentence even when the mediating grammatical verb lacks an instrument F. Checking a person F does not automatically imply Targeting, however. While have in TP always checks the person F of its subject when it is construed as having psychological content, the subject need not target its object. If it does, the sentence has intentional causality. If it does not, the subject is an Experiencer and the sentence has metaphysical causality. Either interpretation is possible in (73a, b) (cf. Harley (1998) for discussion). (73)

a. b.

John had his tooth pulled. John had Mary spit on him.

2.6 On the Syntactic Determination of Causality We have now seen that intentional causality, which seemed at first to be based solely on the presence or absence of an instrument F in V, can be introduced in syntax by person to person targeting in TP. Syntactic determination of causality goes in the other direction when a verb with a lexical instrument content is unable to realize a manipulator argument in syntax and is therefore associated with metaphysical causality at the interface. If, as we claim, auxiliary BE signals metaphysical causality, then the passive sentences in (74b) and (75b) and the French middle in (75c) illustrate metaphysical causality despite the instrument content of the verb. (74)

a. b. c.

Mary washed the shirt. The shirt was washed. This shirt washes easily.

(75)

a. b. c.

Mary a lavé la chemise. La chemise a e´ té/fut lavée. Cette chemise s’est lavée facilement.


165

The fact that a lexically selected subject may fail to appear in syntax shows that the grammar does not contain a Projection Principle which obligatorily realizes lexical theta-roles as syntactic arguments. This is turn weakens the hypothesis that lexical theta-role features exist (cf. Guéron, 2004, among others). However, I assume that the grammar does contain a principle of lexical integrity which implies that a lexically implied instrument must be licensed in syntax. I propose that the internal temporal content of a passive verb is recovered by syntactic raising within the T-chain. We noted above that a non-finite perfect participle is construed in TP as a perfective finite verb. In a passive sentence, the perfect participle focuses the result state of the event vP describes; the remaining initial and internal states become part of the presupposition of the sentence. When the punctual auxiliary be maps the focused state onto one point of time in T, the participle is construed as a perfective verb. The presupposed states raise to T, and to C, where they are projected onto the points of time which precede the focussed point in T. The expression “was washed’ in (74b) is the passive counterpart of the active form “washed” in (74a). BE just functions as a syntactic elevator, which, by providing a person F to the lexical verb under merger, allows the latter to raise to T. In a language like Arabic, in which the perfect participle already incorporates a person F, the auxiliary is not necessary to construct a passive form. In English and other languages, a by-phrase binds the implicit manipulator argument selected by the instrument F of V. The spatial aktionsart F of the Prepositional phrase modifies the entire internal structure of the event up to its lexically defined telos if there is one. (76)

a. b. c. d. e.

They walked by the house. This is a picture by Picasso. I went to Paris by train. He killed the chicken by wringing its neck. Abraham Lincoln studied by candle light.

The by-phrase in the English passive functions much like instrumental case in Russian. According to Matushansky (1998), the only arguments which are always in the Instrumental case in Russian are body-part instruments, tools, and passive byphrases. Matushansky suggests that the “Agent” in the by-phrase and the instrument of action may be grouped together as “intermediary of action”. This is precisely our characterizaton of the interpretive function of little v in the sentence.10 Middle sentences also have metaphysical causality. The middle sentence denotes a change of state determined by an inherent property of its inanimate subject. The internal temporality of the event VP describes cannot be recovered by its generic tense, however. In (74c), the little v implied by the instrument F of V functions as a free variable bound by the easy adverb. Following Hoekstra and Roberts (1993), we associate the adverb easily in (74c) with two arguments, an event argument and an implicit benefactive argument. The benefactive argument binds the manipulator of the event selected by the instrument F of V. The evaluative adverb reveals the point 10

Old English by also had instrument content (cf. Toupin, 2006).

166

J. Guéron

of view of the speaker. Syntax also plays a crucial role. Adverbs located above the instrument level vP make the sentence ungrammatical in English, even if they are evaluative. (77)

This shirt washes easily/∗ painstakingly/∗foolishly/∗adroitly/∗inevitably.

In the absence of an easy adverb, any sentential operator controlled by the speaker in the TP-CP phase can bind the manipulator variable in the little v node: emphasis, negation, interrogation, exclamation, or an epistemic adverb. No person-checking subject intervenes between the speaker and vP. (78)

a. b. c. d. e.

Wow does this shirt wash! Careful, this shirt doesn’t wash. Does the shirt wash at all? But the shirt does wash! This shirt possibly/may/should wash.

I propose that in a Romance middle, clitic SE, with its invariant third person F and merged with V adjoined to T, binds an ARB pro manipulator in Spec vP. SE is necessary: in languages where v +V merge with T in syntax, little v is no longer in the scope of an easy adverb or accessible to a sentence operator on the level of construal. The property of Romance and English middles which interests us most here is their modal content. English middles have a can modalty while French middles have either a can or a must modality. (79)

a.

(80)

b.

This shirt washes well in luke-warm water. Cette chemise se lave a` l’eau tiède. Ces choses ne se disent pas. ∗ Those things don’t say. (cf. One doesn’t say such things.)

(can) (can or ought to)

Where does the modality of a middle sentence come from in the absence of a modal verb? And why is the modality of necessity licensed in French but not in English? The easy adverb or expressive sentence operator which binds the instrument F in little v derives a “can” modality automatically: vP is the level of physical ability. In French, SE in T can also bind a manipulator in Spec vP. The modality of necessity, unlike the modality of ability, is construed on the TP level, for it requires the existence of a time interval in T. Romance SE cannot be construed as an intentional subject in TP, for it lacks a variable person F. I suggest that in Romance the speaker identifies ARB SE as a [+human] instrument capable of effecting the embedded event. The French deontic middle is thus a causative structure whose human instrument is not selected by a grammatical verb on a high vP level, but by the speaker on the TP-CP level.


167

2.7 “Metaphysical Intentionality” “Metaphysical intentionality”, as the term is used here, is an oxymoron, a combination of opposites; it cannot exist outside of fantasy worlds. Yet it has been claimed that passive structures have a syntactically active Agent which binds the PRO subject of a purpose clause in sentences like (81a, b). (81)

a. b.

The ship was sunk [PRO to collect the insurance]. Billy was punished [PRO to teach him not to be rude].

A similar problem is raised by the control structures of (82) discussed in Williams (1974) and Landau (2000), in which PRO has no possible overt binder. If “Africans” bound PRO in (82e), for example, the sentence would be ruled out by Principle C of the binding theory. (82)

a. b. c. d. e.

Grass is green [in order PRO to aid photosynthesis]. These pegs are round [PRO to fit into these holes]. Models are thin [PRO to show off designer clothes]. Jesus died [PRO to save your soul]. Africans have dark skin [PRO to protect them from the sun].

The problem of metaphysical intentionality arises in (82) only if we assume that the subordinating conjunction “(in order) TO” expresses intentionality. It does not arise if we assume that the conjunction refers to what it says, to ordering. I proposed above that TO is an aktionsart/aspect particle which links the disjoint spatio-temporal phases of a single event. The complement of TO denotes the end point of an intentional trajectory when TO is in VP. But when TO links a sentential cause to a sentential effect in the TP domain, as in (81) or (82), then, I claim, “(in order) TO” refers to the metaphysical ordering of events in the discourse world. The ordering is effected by the laws of nature in (82a) and of society in (82b–d).

3 Modal Verbs Part II: Causality Make/ see or faire/ voir and modal verbs all select the same type of complement, a bare infinitive in Modern English and a suffixed infinitive in Romance or Old English. While make/ see or faire/ voir are transitive causative verbs which agree with their subject, modals are impersonal causative verbs which lack morphological agreement features. The subject of modal (83a) and causative (83b) and the object of the control structure (83c), is a [+human] argument, Mary, which functions, in all three sentences, as an instrument capable of introducing the event defined by the complement VP into the discourse world. (83)

a. b. c.

Mary must leave. Mary made John leave. John convinced Mary [PRO to leave].

168

J. Guéron

In order for a state Sn-1 to count as the cause of a new state Sn, all the conditions necessary for the change of state to take place must already be present in Sn-1. In particular, Mary must be capable of making the event happen. The modality of necessity thus presupposes the modality of possibility, and the ability of a subject to function as an instrument. The speaker identifies the subject of the modal sentence as an instrument but she does not target the subject. Targeting would entail that the embedded sentence represents the intentional goal of the speaker. But the speaker has no intentions unless she says “I” or uses imperative mood. Moreover, targeting is a local relation which requires the conscious entity and its target to share the same space. In a modal sentence, the speaker and the subject overlap in time but not in space. What the speaker can do is first, identify the metaphysical direction of change in the discourse world at the Assertion time as it is determined by the laws of nature or of society (as in the circumstantial modal base of Kratzer). (84)

a. b.

The law says you may not take justice into your own hands. You can’t wear that shirt to a formal dinner party.

The laws of the discourse world may also be based on desire, that of a speaker who says “I”, or that of some other participant. (85)

a. b.

You can’t go to that party because I say so. If John wants good cheese, he must go to the South End (von Fintel and Iatridou, 2007).

Having determined the metaphysical direction of change, the speaker selects an accessible goal compatible with this direction and a human instrument capable of realizing the goal. The modal sentence based on MUST or CAN implies an order in the universe in the TP-CP domain, in the same way that a verb of directed movement based on GO implies a trajectory with an end point in the vP domain. And just as a moving figure cannot see the end point of the trajectory in which it is located, so the subject of the modal sentence is unaware of the underlying order of the discourse universe when he is selected to play the role of instrument.11 An epistemic modal requires no external instrument to realize the envisaged state of affairs because it is situated in a mental space which has no temporal gaps. (86) asserts that John’s madness follows naturally and without a temporal gap from other properties of John or from events located in John’s biography. The speaker’s judgement is the evidential source of a sufficient causal link between John’s behavior and his madness. (86)

John must be mad. a. All Englishmen are. b. to say such a mean thing to Mary.

11 He may become aware of his role at some point in the trajectory, however. In our liminal quotation, Hamlet complains bitterly about being an instrument obliged to maintain an order in the discourse world based on filial obligation.


169

A future event may also follow seamlessly from a present state which the speaker judges to include all the conditions necessary to its realization. (87)

John must leave tonight (to judge from this train ticket which I just found).

Although the speaker does not target the subject, she can implicitly target some ARB hearer who shares her time and space in order to get the hearer to be, or find, an instrument of change. (88)

a. b.

There must be twenty chairs in this room by 5 P.M. The baby must have had dinner by the time I return.

Imperfective aspect is necessary for modality. Aspect links the Assertion time to the Reference time, making the subject accessible to the speaker. A second necessary factor is the existence of a scale of values. The speaker’s presence is felt whenever a selection of a value from a scale of values is not attributed to a subject. (89)

a. b.

John thinks that Mary is beautiful. Mary is beautiful.

On the vP level of instrumentality, an event is assumed to be possible and it ranges on a scale going from easy to hard. On the TP-CP level of temporal causality, an event may exist in the mind without existing in the deictic world, so it ranges on a scale going from possible to impossible. The ability to scan a scale of possible values and pick out one of them is a form of focusing, which is the privilege of the speaker.

4 Conclusion I have proposed that intentional causality depends on the existence of instrument content in V (or, with light verbs, in v). The instrument content of V selects a direct object in VP which the instrument affects and a manipulator argument in Spec vP capable of handling it. If the manipulator argument raises to Spec TP and checks a person F with the finite verb, it obtains psychological properties of intention and will and functions as the Agent which triggers the event and assures its temporal continuity. If V lacks an instrument F, then causality is non-intentional or metaphysical. In this case, the speaker must assure the temporal continuity of the event vP describes. As no change of state can take place without an instrument, metaphysical sentences also have a vP Instrument Phrase. The v head functions as a variable which can be bound by a by-phrase in a passive sentence or by an “easy” adverb or a SE clitic in a middle sentence. More complex sentences contain both a spatial manipulator in Spec vP, and a [+/− human] temporal instrument in the specifier of a higher vP projection selected by a causative or perception verb. The causative or perception subject blocks raising of the manipulator from vP to TP. The sentence then has

170

J. Guéron

metaphysical causality, unless person-to-person targeting on the TP level imposes intentional causality. A modal verb behaves much like a causative verb in syntax. In many languages, including, plausibly, Old English, the modal is located on the causative vP level in syntax. However, the modal verb is not transitive but impersonal, and it is not eventive but stative. Lexically, the modal has a privative cognitive content of necessity or possibility that implies an existential gap between what pertains in the discourse world at the speech time and what must or can exist. This cognitive gap has been construed as implying a possible worlds scenario. However, modals belong to the syntactic class of aspectual auxiliaries which facilitate the transformation of a spatial configuration into an event by creating a temporal interval at the deictic Assertion time in the current discourse world. The vP complement of the modal verb doesn’t define a counterfactual or a fantastical situation but rather either a current yet unproven state or an imminent event compatible with the metaphysical laws of the current discourse world and accessible through reasoning or human action. In the latter case, the speaker selects the subject as an accessible and efficient instrument capable of realizing a goal in the current world that the speaker has already identified at the Speech time. Normally the speaker is omniscient: she knows what happened in the past, what is happening now, and what will happen in the future. She also knows the direction of metaphysical causality. But the intervention of the speaker to chose an instrument belonging to the deictic world to realize a specific goal places her inside rather than outside the event trajectory. She then sacrifices her omniscience and cannnot know whether the goal she defined will ever be (or for epistemic modality, is) attained. Consequently the same verb, English will, may be used either as a modal verb by an intervening speaker lacking omniscience or as a future tense by a hands-off speaker enjoying omniscience, without any confusion in the interpretation of the sentence.

References Bickerton, D. & S. Iatridou (1987) “Verb Serialization and Empty Categories”, MS. U. of Amsterdam. Carlson, G. (1995) “Truth Conditions of Generic Sentences: Two Constrasting Views” in G. Carlson & F. Pelletier, eds., The Generic Book, U. of Chicago Press, 224–237. Chomsky, N. (1995) The Minimalist Program, MIT Press. Chomsky, N. (2001) “Derivation by Phase” in M. Kenstowicz, ed., Ken Hale: A Life in Language, Cambridge, MA, MIT Press, 1–52. Condoravdi, C. (2002) “Temporal Interpretation of Modals” in Beaver, D. et al. (eds.), The Construction of Meaning, CSLI Publications, Stanford, CA., 59–87. Copley, B. (2006) “Temporal Orientation in Conditionals” this volume. Copley, B. (2005) “Ordering and Reasoning” New Work on Modality, MITWPL 51, 7–34.


171

den Dikken, M. & R. Sybesma (1998) “Take Serials Light Up the Middle”, ms. U. of Tilburg. Enç , M. (1987) “Anchoring Conditions for Tense”, Linguistic Inquiry, 18,4, 633– 657. von Fintel, K. & S. Iatridou (2007) “Anatomy of a Modal Construction”, Linguistic Inquiry 38, 445–483. Folli, R. & H. Harley (2007) “Teleology and animacy in external arguments” talk at the FIGS conference in Paris. Giorgi, A. & F. Pianesi (1997) Tense and Aspect, Oxford University Press. Guéron, J. (2004) “Tense Construal and the Argument Structure of Auxiliaries” in J. Guéron & J. Lecarme, eds. Guéron, J. (2005) “Tense, Person, and Transitivity” in N. Shir & T. Rapoport, eds. The Syntax of Aspect, Oxford University Press, 89–116. Guéron, J. (2006) “Generic Sentences and Bare Plurals” in S. Vogeleer & L. Tasmowski, eds., Non-definiteness and Plurality, Amsterdam, John Benjamins, 219–234. Guéron, J. & T. Hoekstra (1988) “T-Chains and the Constituent Structure of Auxiliaries” in A. Cardinaletti et al. eds., Proceedings of the Glow Conference in Venice 1987, Dordrecht, Foris, 35–99. Guéron, J. & J. Lecarme, eds. (2004) The Syntax of Time, Cambridge, MA, MIT Press. Harley, H. (1998) “You’re having me on! Aspects of have” in J. Guéron & A. ZribiHertz, eds., La grammaire de la possession, Université Paris X, Nanterre, 195– 226. Hoekstra, T. & I. Roberts (1993) “Middle Constructions in Dutch and English” in E. Reuland & W. Abraham. eds., Knowledge and Language, II, Doredrecht, Kluwer, 183–220. Iatridou, S. (2000) “The Grammatical Ingredients of Counterfactuality” Linguistic Inquiry, 31, 231–270. Kayne, R. (1975) French Syntax, Cambrdige, MA, MIT Press. Kratzer, A. (1981) “The Notional Category of Modality” in Eikmeyer, H. J. & H. Rieser, eds., Words, Worlds, and Contexts, Berlin, de Gruyter, 38–74 Kratzer, A. (1991) “Modality” in A. van Stechow & D. Wunderlich, Handbook of Semantics, Berlin, de Gruyter, 639–650. Kuroda, S.-Y. (1972) “The Categorical Jugement and the Thetic Judgement”, Foundations of Language 9, 153–185. Laca, B. (2004) “Romance ‘Aspectual’ Periphrases: Eventuality Modification versus ‘Syntactic’ Aspect”, in Guéron & Lecarme, eds. Landau, I. (2000) Elements of Control: Structure and Meaning in Infinitival Constuctions, Dordrecht, Kluwer. Lecarme, J. (2004) “Tense in Nominals”, in Guéron & Lecarme, eds. Lieber, R. (1982) “A Note on the History of the English Auxiliary System”, in A. Marantz & T. Stowell, eds., MIT Working Papers in Linguistics, 4, 81–99. Matushansky, O. (1998) “Instrumental Case and Verb Raising in the Russian Copula”, ms. MIT and U. Paris 8 (more extended version in WCCFL 19)

172

J. Guéron

Reichenbach, H. (1947) Elements of Symbolic Logic, New York, Free Press. Romero, C. (2005) L’évolution des verbes modaux dans l’histoire de l’anglais, Doctoral Dissertation, U. Paris 3. Schwer, S. (2006) “En suivant Reichenbach: les S-langages”, Conference for the Temptypac group. Shibatani, M. (1975) A Linguistic Study of Causative Constructions, Indiana U. Linguistics Club. Toupin, F. (2006) “Le préverbe BE- du vieil anglais” Travaux de diachronie”, Tours, Presses Universitaires Franç ois Rabelais, 59–79. Travis, L. (2000) “Event Structure in Syntax” in C. Tenny & J. Pustejovsky, eds., Events as Grammatical Objects, Stanford, CA, CSLI Publications, 145–185. Travis, L. (2006) “Event Semantics in Syntax”, ms. McGill U. von Fintel, K. & S. Iatridou (2005) “Anatomy of a Modal Construction”, Linguistic Inquiry, 38,3, 445–483. Williams, E. (1974) Rule Ordering in Syntax, Diss. MIT.

The English Perfect and the Metaphysics of Events James Higginbotham

Abstract I argue here that the what Otto Jespersen called the “conservative” English Perfect is a purely aspectual predicate, true of occurrent results of prior events, noting also: (i) that nominalizations, including nominalizations of the Perfect, call for extending event positions to nonheads including adverbs and quantifiers, and (ii) that restrictions on the present perfect with prior temporal determinations extend also to embedded clauses under the principles governing tense anaphora, or sequence of tense.

Key words: Perfect, events, result state, tense, aspect.

1 Introduction The general thesis that I explore here is that the English Perfect is purely aspectual, serving to shift from a predicate of events e to a predicate of events that are results of e, as explained more fully below. To the extent that this thesis is defensible, the Perfect is not involved in the Tense system at all, except derivatively (because Tense applies to the Perfect itself). The investigation of the Perfect interlocks with a number of metaphysical questions concerning the nature of events and situations, and further issues in the syntax and semantics of Sequence of Tense in English. In exploring these I shall draw upon some earlier work including Higginbotham (2000, 2002, 2005). The general thesis, that the Perfect concerns aspect rather than Tense, is an old one. Jespersen (1924: 269) proposed that the English Present Perfect characterizes “present results of past events,” and not the past events themselves. Parsons (1990) makes at least one part of this account explicit. Consider a simple assertion such as (1.1): University of Southern California


173

174

(1.1)

J. Higginbotham

Mary has solved the problem.

We assume that the heads line up as in (1.2): (1.2)

[Tense [Perf [VP]]

with V moving to the head Perf, and auxiliary have to Tense. The V solved expresses a 3-place relation S(x, y, e) that is satisfied by a triple (α , β , γ ) just in case γ is an event of α ’s solving β . The head Perf has two argument positions and expresses a relation R(e , e), satisfied by an ordered pair of events such that e is a “result” of e, in a sense or senses to be determined. The interpretation of the Perfect Phrase in construction with VP is then as in (1.3): (1.3)

Result(e , e) & solve(Mary, the problem, e)

I shall assume that the Present expresses the relation ≈ of temporal overlap between (the actual time of) an event and (the actual time of) the speaker’s utterance u. Following existential closure (as originally suggested in Donald Davidson’s work), we obtain (1.4): (1.4)

[∃e ≈ u][∃e][R(e , e) & solve(Mary, the problem,e)]

In Present Perfect assertions, and in Tensed Perfects generally, it is the result that carries the Tense, and is said to surround the time of the clause (the utterance time, in the case of root sentences). But because, by assumption, results temporally follow, and do not overlap, the states or events of which they are the results, it follows that the situations characterized by a predicate F(e) are Past with respect to a present result e of e. Present Perfects are distinct from Pasts (Preterits). Jespersen (1924: 270), himself following earlier observations, notes that: Sentences containing words like yesterday or in 1879 require the simple preterit, so also sentences about people who are dead, except when something is stated about the present effect of their doings, e.g. in Newton has explained the movements of the moon (the movements of the moon have been explained—namely by Newton).

Jespersen further characterized English as relatively “conservative” in these respects; conservative, that is, with respect to his proposed evolution of the Germanic and Romance language families. If this is right, then the Present Perfect is just what it seems to be, namely the Present of a Perfect, as in Jespersen’s famous example (1.5): (1.5)

Now I have eaten enough.

where the Present predicate now must apply to the Present result, not the Past activity. Where the adverb now is predicated of the higher position marked by e , and the inflectional feature -Past is interpreted as expressing the relation ≈ of temporal overlap, the result is (1.6): (1.6)

[∃e ≈ u][∃e] [Now(e ) & R(e , e) & eat enough(I,e)]

The English Perfect and the Metaphysics of Events

175

We next distinguish Results from Resultants (the latter in the sense of Parsons (1990)). Resultant states are states that commence when an event e is over, and continue forever. The meaning of the sentences in which they occur is, as it were, “been there, done that,” salient for instance in (1.7): (1.7)

I have been to Japan.

For a Perfect of Result (not Resultant) consider (1.8): (1.8)

I have spilled my coffee!

The announcement (1.8) is only in order as long as there is spilled coffee around. Likewise, (1.5) is fine when just setting down one’s knife and fork after dinner, and for some time afterwards, but not upon waking up the next morning (unless one has just resolved to go on hunger strike). And so on in like cases, as both Result and Resultant states commence immediately after the events of which they are the Results or Resultants.1 For notation in what follows we use ‘R’ for Result states, and ‘RR’ for Resultant states. Some further preliminary remarks. First, it is evident that Perfect morpheme –en of English can all by itself contribute a result state. The point is seen in examples like those in (1.9): (1.9)

saddled (the saddled horse); botoxed (the botoxed face); treed (the treed cat)

derived from Nouns (I assume that the corresponding Verb is causative on the participle, as in Higginbotham (2000)), and those in (1.10), derived from Verbs: (1.10)

burned (the burned tree); spilled (the spilled coffee); demolished (the demolished house)

Second, the point is underscored by the existence of adverbials that cannot be Verbal modifiers, but may modify Perfective Adjectives, as in (1.11) versus (1.12), or (1.13) versus (1.14): (1.11)

the freshly caught fish, the newly appointed Assistant Professor

(1.12)

∗ we

(1.13) (1.14)

freshly caught the fish, ∗ they newly appointed an Assistant Professor.

John is newly divorced. ∗ John

newly divorced.

1 It is perhaps worth noting that the Perfect head, even when happening to favor in context a Result interpretation, can always be understood as expressing a Resultant, as in (i) for instance:

(i)

I have done many foolish things in my life—I have spilled my coffee, I have lost my wallet, etc.

176

J. Higginbotham

Third and last, the presence of the result is seen in the favored interpretation of expressions like John’s broken arm (not the arm that he broke, but the one that is now broken; i.e., in the result state of a breaking), or the stolen painting (a painting that is now stolen, and so not recovered). Comparable evidence is of course found in the Romance and Germanic languages.2 The Perfect of Result is thus widely attested, and the question is whether all Perfects in English fit the mold. The above syntactic and semantic points leave questions about the nature of the elements—events, situations, Results, and Resultants—that have been invoked in their characterization. In section 2 below these, with consequences the Perfect, will come to the fore. In section 3 I take up the matter of the integration of the Perfect into the English system of Sequence of Tense, and in section 4 I offer a generalization of the puzzles of the Present Perfect, as noted for instance in the quotation above from Jespersen. A number of further issues remain, and I will note some of these in passing.

2 Metaphysical Issues We are recognizing, in ordinary heads, positions for arguments, plus the eventposition (hereafter: E-position), as in solve(x,y,e) in example (1.1) above. I assume further as in Higginbotham (2000) that in heads giving rise to Accomplishment interpretations the E-position is complex, consisting of an ordered pair of positions, one for process and one for telos or end, as in cross the street, where the positions are (x, <e, e >), with the first that reserved for the agent, e for the process of crossing, and e for the end, the onset of being on the other side of the street. Events or situations in this sense are to have to following properties and relations to the syntax: (A) They are the relata of causal relations. To say that striking the match caused it to light is to say that one event e, the striking, caused another e , the lighting. (The relation between process and end as in cross the street is causal as well, but of a rather special kind, namely one where the process—proceeding from the curb more or less toward the other side of the street—continues until the end—being on the other side—comes about.) (B) In gerundive and mixed nominals, and in many derived nominals as well, it is just the E-position in the head that is bound by the Determiner; thus the striking of the match is a definite description as in (2.1): (2.1)

(the e) strike(the match, e)

Moreover, sentences in general allow gerundive nominals, and many allow nominals of the other kinds as well.

2

Boogaart (1999) and references cited therein provides a contemporary survey, with special reference to English and Dutch.


177

(C) Such nominals, and complete IPs in some contexts, may refer to or quantify over events and states. An example of the first kind is (2.2), of the second (2.3): (2.2)

Mary enjoys/remembers [walking to work].

(2.3)

[John fell down] after [Bill hit him].

The expressions in square brackets involve events, not propositions. (D) Events and other situations may be described in any number of ways (they are objects). The event descriptions given by nominalization or IPs admit without exception the substitutivity of identity for singular terms. So, to deploy a well-worn example, the nominals (2.4) and (2.5), if they refer at all, refer to the same thing, even though at one point Oedipus didn’t know that: (2.4)

Oedipus’s marriage to Jocasta

(2.5)

Oedipus’s marriage to his mother

(E) Adverbial modifiers of the simplest sort, the manner adverbials, characteristically modify, by conjoining with, the E-position of the head; adverbial quantifiers may also bind this position. So, in standard manner-adverbial cases like (2.6) the interpretation is as in (2.7) (with perhaps the qualification that the crossing was only relatively quick); and in standard quantificational cases like (2.8) one of the salient interpretations, indicated in (2.9) is that amongst Mary’s journeys to work some reasonable proportion are walks: (2.6)

John crossed the street quickly.

(2.7)

[∃e] [Cross(John, the street,e) & quick(e)]

(2.8)

Mary occasionally walks to work.

(2.9)

[Occasional e: travel to work(Mary,e)] walk to work(Mary,e)

Assembling the above pieces, however, generates a number of problems, of which I concentrate on two, one involving causation, and the other involving complex nominalizations. I will argue that these problems have the same solution, a solution that, while it may appear extravagant, is nevertheless practically a deductive consequence of the basic assumptions above. The problem with causation (noted by many, and folkloric so far as I know) is that, according to the hypothesis that takes (2.6) up as (2.7), the nominalizations in (2.10), one gerundive and one mixed, come out as in (2.11): (2.10)

John’s crossing the street quickly/John’s quick crossing of the street

(2.11)

(the e) (cross the street(John, e) & quick(e))

These nominalizations occur as the subjects and objects of causation, say as in (2.12): (2.12)

The traffic light’s changing to amber caused John’s quick crossing of the street.

178

J. Higginbotham

(or, for the object of (2.12) the infinitival, John to cross the street quickly). But now it may well be that John’s quick crossing of the street was his one and only crossing of the street. In that case we have the identity (2.13): (2.13)

(the e) (cross the street(John, e) & quick(e)) = (the e) cross the street (John, e)

But then we may put equals for equals in (2.12), as in (2.14) (2.14)

The traffic light’s changing to amber caused John’s crossing of the street.

But we can readily conceive that (2.14) is false whilst (2.12) is true, a contradiction. There are many variations on the above theme.3 One can imagine responses that would deny one of (A) or (B), that causal relations are relations between events, or that nominalizations purport to refer to them, or (D), that events are objects. Note that the very success of hypothesis (E), that modifiers such as quickly may be understood as simply conjoining with what they modify, is what brings down the conjunction of the others; and conversely, that abandoning (E) would call (A)– (D) into serious question. Davidson himself suggested that examples such as (2.12) equivocated between statements about causation pure and simple and statements of putative causal explanation, between causes and becauses, as he put it. Certainly, statements A because/causally explains B display properties very different from singular causal statements: they relate sentences (or propositions), and they do not admit the substitutivity of identity for singular terms. But the move deprives the major thesis (A) of much of its credibility. There is, however, another, very different, solution to the puzzle, one that suggests itself as soon as notice is taken of the generality of the process of forming event nominals. Consider again assumption (B), and note that nominals with adverbial heads come in as a special case, as in (2.15): (2.15)

the quickness of John’s crossing of the street

These nominals themselves refer to situations: just as John’s happiness refers to the situation e such that Happy(John, e), so should (2.15) be construed as in (2.16): (2.16)

(the e) [∃e ] [quick(e, e) & cross(John, the street, e )]

for the nominal refers, not to the crossing, but to the quickness of the crossing. But now we may conclude that there is an E-position also in the Adverb, in addition to the position that is identified with that of the Verb.4 In place of (2.7) for (2.6), then, we have (2.17), revealing the extra E-position:

3 The example chosen is one where the alternative descriptions are of the caused, rather than the causing, event. But examples of the latter kind are easily constructed as well, e.g. (i):

(i) 4

John’s quick crossing of the street caused him to be early.

This point is briefly noted in Higginbotham (2005).


(2.17)

179

[∃e][∃e ] [Cross(John, the street, e) & quick(e, e )]

and when we nominalize (2.6), or take the infinitival complement as referring to an event, we may nominalize either with respect to the E-position in the head, or with respect to that in the adverbial. If we nominalize with respect to the syntactic head, then for John’s quick crossing of the street and the other nominals we have the identity corresponding to (2.13), namely (2.18): (2.18)

(the e) [∃e ] [cross the street(John, e) & quick(e, e )] = (the e) cross the street(John, e)

but if we nominalize with respect to the Adverb (as we must do in (2.15)), we have (2.16), referring not to the crossing, but to its quickness. With the object nominal so understood, there is no question of identity, and the implication from (2.12) to (2.14) fails, even on the assumption that John crossed the street just once, in a quick crossing. But must the object nominal be so understood? I am inclined to think not, at least for the case of mixed nominals; that is, there may be a way of understanding (2.14) that makes it true after all (though misleading). For infinitival objects, however, my judgement is that only the interpretation that describes the quickness of the crossing is possible, as in (2.19): (2.19)

The traffic light’s changing to amber caused John to cross the street quickly.

The problem just addressed, and the suggested solution, extend through countless similar examples. Of these I mention one, which I heard delivered by Robert Hambourger in discussion with Davidson at a Colloquium at Columbia University in 1971. Suppose I am going home late at night, and I fear that I am being followed by a mugger. I quickly fumble for my keys. I have two keys for my door, one green and one red. As it happens, the one I grab is the green key. Then: my fear caused me to open the door; and it caused me to open the door with a key; but it did not cause me to open the door with the green key. That is to say: it was not the state e of the key’s being green that caused me to open the door with that key. In the 1971 discussion, Davidson responded to the example with the distinction between “causes” and “becauses,” noted above. In a way the response was correct; what was not noticed was that the “becauses” could be located in the events themselves, through the E-positions in non-heads such as quickly above, or green in Hambourger’s example. Assuming now that at least causal contexts bring out E-positions in non-heads, I turn to a second case where the proposals (A)–(E) come into apparent conflict both with the data and with each other. Consider contexts where complement clauses evidently involve reference to events, as in (2.20) and the like: (2.20)

Mary enjoyed/remembered walking to work this morning.

180

J. Higginbotham

The complement purports to be a definite or indefinite description of an event, namely that of walking to work. The E-position in the complement is accessible to adverbial quantification, as in (2.21): (2.21)

Mary occasionally walks to work.

An assertion of (2.21) will be made against a background according to which Mary’s walks to work are occasional—her travels to work, for instance. Then the truth conditions of (2.21) will be as in (2.22): (2.22)

[Occasional e: travel to work(Mary, e)] walk to work(Mary, e)

But now we may embed such clauses as gerundive complements, as in (2.23): (2.23)

Mary enjoyed/remembered occasionally walking to work.

where the quantificational adverb is interpreted with respect to the clause where it appears. What Mary is said to enjoy or remember is occasionally walking to work— the occasionality of it, as one might say. But the theory allows no such interpretation. The complement is supposed to describe something, but, unlike the complement in (2.21), it is not a predicate but a closed expression (the adverbial has bound the E-position). In light of our discussion above of causal contexts, a solution to this conundrum suggests itself: why not allow an E-position also in the adverbial? It then can become a definite or indefinite description of an event, and we shall obtain truth conditions as in (2.24): (2.24)

[∃e] [Occasional(e, e ): walk to work(Mary, e )] Enjoy/Remember (Mary, e)]5

Indeed, quite apart from contexts such as (2.23) we are forced into recognizing an E-position, or the potential for such a position, in (2.21) anyway, because of the possibility of nominalization: (2.25)

the occasionality of Mary’s walking to work

The general solution, then, to the issues that arise when we put proposals (A)–(E) together is a flourishing of E-positions, and therefore of reference to events, particularly with respect to modifiers, quantifiers, and other non-heads. Viewed in this light, the invocation of such elements as Result and Resultant states (particularly the latter, as they are defined only by what they are the Resultants of) is the least of it. Turning back to the Perfect, it is a consequence of the discussion to this point that Results and Resultants may be given in terms, not only of the events described in the verbal heads, but also of those that come in through modification or quantification. Indeed, causal contexts will exemplify the situation. In (2.26), for example, it is not the fire’s having burned out, but its having burned out slowly that causes the bricks to become warm: 5

I recall this suggestion as having been made in passing by Paul Portner.


(2.26)

181

The fire’s having burned out slowly warmed the bricks.

and indeed the Result of the fire’s burning out—namely, no more fire—and the Result of its burning out slowly—ambient heat—are intuitively different. If this is right, then we have further evidence that it is Perfect Aspect, and not the E-position in the head burn, upon which Tense operates.

3 Interactions with Sequence of Tense I understand the phenomenon of Sequence of Tense in English to comprise anaphoric relations, optional or obligatory, between inflectional heads. The Tenses themselves express binary relations between events (more precisely, their actual times; but we abstract away from this detail for ease of exposition); and we assume the obvious relations of temporal priority and temporal overlap for Past and Nonpast. The paraphrastic Future will is −Past, and would is the +Past inflection of will. The +Past feature, I shall suppose, need not actually express the past, but may function so as to facilitate, in some way, the anaphoric connection. The ambiguity of (3.1) is a case in point: (3.1)

John said that Mary was late.

(3.1) may be taken as a report of a past past-tense utterance, “Mary was late” say, made by John; or a report of a past present-tense utterance, “Mary is late.” In the former case, the speaker of (3.1) is attributing a content to John’s speech that places Mary as late prior to the time of that speech. The truth conditions of an utterance u of (3.1) are then as in (3.2): (3.2)

[∃e < u] Say(John, ∧ [∃e < e] late(Mary, e ), e)

The anaphoric relation consists in the matching of the second coordinate of the embedded Tense with the first coordinate of the superordinate Tense. In the latter case, the speaker is attributing a content that places Mary’s lateness at the time of that speech itself. The +Past feature is not interpreted as expressing priority e] [Pregnant(Mary,e) & e > u],e)

The restriction on the quantifier in the content (complement) clause is of course redundant, as we have (if John speaks truly) the ordering e > u > e, and so in particular e > e. But for (3.4) it is not redundant, as we obtain (3.7): (3.7)

[∃e < u] Say(John, ∧ [∃e ≈ e] [Pregnant(Mary, e ) & e ≈ u], e)

so that the temporal content of the complement overlaps both the time of the subject’s alleged speech and that of the speaker. Transposing the scheme to Perfect complements, we might examine (3.8) and (3.9): (3.8)

John said that Mary has been pregnant.

(3.9)

John said that Mary will have been pregnant.

(3.8) should show double-access, with the Result or Resultant state given in the complement as overlapping the time of the subject’s alleged utterance, and the time of the speaker’s. But pregnant is a stative, hence implicates no Result state; and the peculiar property of Resultant states, that they are temporally unbounded, implies that, in contrast to (3.4), the non-anaphoric copy of the complement −Past has no work to do: if the alleged Resultant state RR(e ) ≈ e, then RR(e ) ≈ u. With Result contexts things are different. Consider (3.10), for instance: (3.10)

The mayor announced that he has signed the legislation.


183

Assertions such as (3.10) are proper on the day that the mayor did the signing, but (so I believe) not the next week, let alone after a longer lapse. But (3.11) will always be assertible: (3.11)

The mayor announced that he signed the legislation.

Examples such as (3.10), then, do provide evidence that the complement Present Perfect is aspectual merely. As for the Future Perfect (3.9), we shall want to ensure that the context is purely temporal, and not the will of “assurance,” whatever that may come to exactly.6 Bearing that point in mind, and so rendering the truth conditions of (3.9) straightforwardly as in (3.12), we have John reported as predicting a Resultant of Mary’s being pregnant whose onset lies in the future of the speaker’s report. (3.12)

[∃e < u] Say(John, ∧ [∃RR(e ) > e] [Pregnant(Mary, e ) & RR(e ) > u], e)

I believe this correct; and similarly for Result states. We may contrast Future Perfect complements above all with the “future in the past” would, as in (3.13)–(3.14) (3.13)

John said that Bill would spill his coffee.

(3.14)

John said that Bill will have spilled his coffee.

Turning now to complements in the Past Perfect, we have the possibility of an APast (where the Past retains its temporal interpretation) and a B-Past (where the complement is not interpreted as Past), together with the Resultant or Result interpretations of the Perfect. The A-Past together with a Resultant interpretation is seen in (3.15): (3.15)

John said that Mary had once been happy.

The temporal relations as shown in (3.16) are appropriate, implying that the content of John’s alleged speech placed the onset of the resultant state prior to that speech: (3.16)

e < u . . . [. . . RR(e ) < e & RR(e ) < u . . .]

But again we have redundancy, as the first embedded temporal location of RR(e ) implies the second. But Result interpretations of the construction are in order as well, where there is no implication that the Result state survives to the present. BPast interpretations are then easily obtained, as for instance when one is telling a story in the middle of which one goes on as in (3.17): (3.17)

. . . then John noticed that I had spilled my coffee. . . ,

the reference time having been reset from the time of speech to some past time determined by the narrative. In this case too, I think, we can observe that the Perfect

6 I have in mind cases such as, “The sum of those digits will be a prime number,” where there is no question of temporality.

184

J. Higginbotham

retains its interpretation: the story line (3.17) is appropriate only if, at the reference time, the coffee is in a spilled (not mopped-up) state. We have seen that, where the effects can be semantically visible, the complement Perfects behave as expected on the general account summarized in (0)–(IV). However, as noted in Ogihara (1995) (also in Iatridou et al. (2005)), the Present Perfect does indeed behave like a Past when it is superordinate. Consider (3.18): (3.18)

John has said that Mary was ill.

If (3.18) were a true Present, like (3.19) below, we would expect it to have a unique interpretation, that in which John is reported as having made a past-tense utterance, “Mary was ill,” or words to that effect. (3.19)

John is saying that Mary was ill.

But (3.18) is in fact ambiguous as between reports of past-tense and presenttense utterances (between A-past and B-past, in terms of the classification (II) above). We must, then, have communication between the complement Past and the superordinate E-position in say, as in (3.20): [∃e ≈ u][∃e : R(e , e)] Say(John, ∧ [∃e < / ≈ e] ill (Mary, e ), e)

(3.20)

with the choice between < (A-Past) and ≈ (B-Past). Of course, a complement Present Perfect may be superordinate with respect to its own complement, as in (3.21): (3.21)

John said that the mayor has made it widely known that he was ill.

The mayor’s announcement could have been either present-tense or past-tense; but the content attributed to John may involve a present Result.

4 Shifted Perfects As Jespersen remarked in the passage I quoted above, the Present Perfect is incompatible with temporal adverbials such as yesterday and in 1879. In this section I offer a partial account of this fact in terms that are, if I read him correctly, rather close to those offered in Portner (2003); I also observe that the phenomenon persists under certain cases of Sequence of Tense, suggesting, again, that the aspectual properties of the Perfect are always visible. The question concerns first of all the generalization that governs the ungrammaticality in examples such as (4.1) (4.1)

∗

I have visited the museum yesterday/last year.

Pancheva and von Stechow (2004) suggest that the generalization is that, in English, the temporal adverbial must apply only to intervals that include the speech time.


185

Thus in Alicia has danced this year, in their view, we intend the whole interval constituting this year up to and including the time of speech, and say that some dancing by Alicia lies temporally within this interval. As last year or yesterday exclude the time of speech, they are unsuitable. However, this cannot be entirely right, because of examples (of a sort that have not much been considered in the literature) such as (4.2) and (4.3): (4.2)

I have visited that museum in the past/before (now).

(4.3)

I have been to Copenhagen twice, a long time ago.

An interval described as “in the past” or “before (now)” presumably cannot include the present. And the parenthetical a long time ago is acceptable in (4.3). The first remark is no mere quibble, as Pancheva and von Stechow intend it the English Present’s overlapping the speech time as crucial in their account of the difference between English and, for example, German, which allows the Present Perfect with adverbials that exclude the time of speech. The second observation raises problems of its own, as the parenthetical in (4.3) must somehow be combined properly, without producing ungrammaticality. I shall not be considering these cases further here (but see footnote 9 below). To fix matters more precisely, consider adverbials that, unlike yesterday, today, and in 1879, do not refer to precise periods of time, but describe matters more loosely. We obtain the partial classification shown in (4.4)–(4.5): (4.4)

Possible with the present perfect: recently, in the past, within/during the last week, lately, yesterday and today, from childhood.

(4.5)

Not possible with the Present Perfect: none too recently, in the distant past, long ago, then, during last week, awhile/a little while back.

It appears that the predicates P possible with the Present Perfect have the property that the past times (hence, times of past situations) of which they are true are closed under temporal succession; i.e., if τ < τ and P(τ ), then P(τ ), for all times τ before the speech time. Thus if Thursday is within the last week, then the Friday following it is as well; if τ is a past interval, and τ is a past interval following it, τ remains a past interval; and so on. But intervals more recent than intervals that are none too recent are some of them recent; so none too recent is not closed under temporal succession. For the moment, then, we may put the condition down as (P): (P)

∗ [e

≈ u) & Per f (e , e) & VP(e) & A(e)] if A is not closed under temporal succession.

I have been supposing that temporal predicates such as recently are true of various sets of points (which may be intervals, or scattered, or even singletons). An alternative, the one I take to have been adopted by Pancheva and von Stechow, is to suppose that they pick out a more or less large period of past time—for recently, the recent past, as one says—so that my predicates Recent(e), which holds if the time τ (e) of e is recent, would be understood as τ (e) ⊆ RecentPast. Evidently, so long as the

186

J. Higginbotham

temporal modifiers can be construed as referring to single intervals, the methods are equivalent. To account for the properties of the English Perfect, Pancheva and von Stechow propose a “competition model,” which selects the Past over the Present Perfect in contexts where there is (in terms of the generalization (P)) a temporal gap between the times falling within the extension of the adverbial and the Present. They suggest that the distinction between English and other Germanic languages for which the puzzle arises, and those for which it does not arise, is correlated with the distinction between those that allow the equivalent of, e.g., John is here since noon and those that do not; but the data here are somewhat obscure, as noted in Giorgi and Pianesi (1998). I what follows, I will offer a conjecture that is independent of questions of correlation, and says nothing of competition. Consider first of all obviously unacceptable examples that constitute clashes of some sort between tense and modifying adverbial, such as (4.6): (4.6)

John is here yesterday.

It will not suffice to observe that they are contradictory: lots of things are contradictory without being intuitively weird in the way that (4.6) is. Nor will it do to suggest that (4.6) is simply ungrammatical. Indeed, as we know that ‘John is here’ allows stretches of Present of arbitrary length, the contradictoriness of (4.6) is by no means obvious. Why, for instance, can I not make the inference in (4.7)? (4.7)

Today there have been no more dinosaurs for a long time; therefore, yesterday there are no more dinosaurs.

An obvious in-between suggestion for (4.6) is that it has something of the status of a category mistake, an expression like blue concept, sleep furiously, or Wednesday is in bed. But even this proposal fails to account for (4.7). The word yesterday, like today, admits extensions: in saying (4.8) I am not speaking just of the day preceding my speech, and it will be obvious to my hearers that this is so. (4.8)

Yesterday children obeyed their parents.

The above phenomena may, however, have an explanation of the sort associated above all with McTaggart (1908). In speech and thought, we set up present, past, and future. We may set them up in different ways, depending upon our interest; but however we do this they exclude one another. To quote from McTaggart: Past, present, and future are incompatible determinations. Every event must be one or the other, but no event can be more than one.

This is essential to the meaning of the terms. (McTaggart, 1908). In (4.6), by virtue of using the word yesterday, the situation of John’s being here is said to be past; but then it is not also present. In this sense, (4.6) is a category mistake. Similarly in the conclusion of (4.7). It does no good that I have asserted in the premise that there have not been any dinosaurs for a very long time; for my very use of


187

yesterday in the conclusion forces the situation of there not being any dinosaurs to be past, and that conflicts with my statement that it is present. Supposing that the above conjecture concerning (4.6) and the like is on track, we may go further. In a routine Present Perfect such as (4.9), in virtue of the fact that recently is closed under temporal succession, we note that both the situation e of my visiting the museum and the onset of the resultant state e are recent: (4.9)

I have visited the museum recently.

But in the paradigm case (4.1), repeated here, supposing as we do that the visit e is confined to the past, the onset of its resultant state e must also be so confined: (4.1)

∗I

have visited the museum yesterday.

(the first moment of the onset is the accumulation point of the time τ (e) of the visit e: on the assumption that yesterday selects a closed interval, this point must belong to yesterday if τ (e) does). Suppose that for a Resultant (or Result) state to be present requires that its onset be present. We then have a contradiction (in McTaggart’s terms), or a category mistake induced by the triadic division of indexically-given time: every situation is past, present, or future, and none is any two of these. I hasten to note that the contradiction in question is not some formal contradiction: the negation of (4.1), ∗ John has not visited the museum yesterday is just as weird a sentence as (4.1) itself. Perhaps in part for this reason, Portner (2003) (to whose view the suggestion in the text is perhaps closest) holds that his rather similar condition (which does not, however, involve resultant states) is “pragmatic,” as he puts it. On the view advanced here it is not pragmatic (it attaches to the expressions in question in virtue of their linguistic form). Rather, the conflict arises because, in English, the trichotomy of past, present, and future applies so as to disallow the expression of quantification over or reference to a Resultant state that is both present (because of the Tense) and, in virtue of the explicit use of the adverbial yesterday, or another that is not closed under temporal precedence, also past. The restriction thus comes in addition to the fundamental, and presumably universal, restriction that rules out John is here yesterday and the like. Of course, none of the above, even if correct, shows that competition models of the sort suggested by Pancheva and von Stechow are wrong. It is worth noting, however, that the puzzle does not persist under Sequence of Tense applied to Past Perfects, either with relative clause objects or complement clauses. Thus consider (4.10): (4.10)

John met a man today who had spilled his coffee yesterday/two days ago.

The Past Perfect (like English ‘would’) requires a licenser (which need not be tense anaphora, as shown by counterfactual contexts). In (4.10) the only licenser around is the superordinate Past, and the complement adverbial yesterday compels a B-Past (“shifted”) interpretation, applying to the Result state. So the second temporal coordinate of the complement INFL is anaphoric to the first coordinate of the superordinate INFL, giving (in relevant detail) (4.11):

188

(4.11)

J. Higginbotham

[∃e < u] . . . [∃e ≈ e][∃e ][R(e , e ) & spill(e ) & yesterday/two days ago (e )]

But now, one can have also (4.12): (4.12)

John met a man today who spilled his coffee yesterday/two days ago.

so that (4.10) and (4.12) are in free variation. It’s not clear, however, why, on a competition model, the competition between the Present Perfect and the Past does not extend to this case. We may propose that the account sketched above is compatible with the fact that (4.10) is acceptable. It is true that, just as the onset of the Resultant state in (4.1) is said to lie in the past of the time of visiting the museum, so the onset of the Result state in the relative clause in (4.10) is said to lie in the past of τ (e), the time of meeting: the cases are formally identical, except that τ (e) has replaced τ (u). But τ (e) is itself past from the point of view of the whole assertion, to which the division past, present, future applies. Hence the adverbials in the relative clause are not restricted to those that, within the temporal perspective provided by that clause, are closed under temporal succession. Similar remarks would apply to cases with complement clauses, as in (4.13): (4.13)

Mary said yesterday that John had visited the museum two days ago.

Mary, being a competent speaker of English, could not have said, “John has visited the museum yesterday.” But if she said, “John visited the museum yesterday,” then I can report her as in (4.13) nonetheless. Despite the absence of a shifted puzzle with the Past Perfect, there is further evidence that distinguishes it from the embedded Past. The B-Past or interpretation of an embedded Past is (as noted above) generally only possible with Statives. Thus Giorgi and Pianesi (1998) remark that interpretations of the like of (4.14), where the complement is a relative Present rather than a relative Past, are not in general possible even for languages where the bare Verb in the Present (unlike English) can be used to report current goings-on: (4.14)

John said that Mary left.

That is, (4.14) cannot be made true in virtue of John’s past utterance of “Mary leaves,” or (in English) “Mary is leaving.” But the Past Perfect under sequence of tense does indeed admit B-past interpretations, as in (4.15) (4.15)

John said that Mary had left.

(John made a past Present Perfect utterance); which is natural enough if, as I am suggesting here, the Perfect, in virtue of moving the application of INFL from the underlying situation, expressed by the main predicate, to the Result or Resultant state, produces a Stative. A true Preterit would not be expected to do that. I conclude this discussion by presenting, first, a diagrammatic representation of cases such as (4.1) on the view taken here, and then observing a case that exemplifies


189

the persistence of the properties of the Present Perfect under Sequence of Tense. The truth conditions of (4.1) are as in (4.16) (4.16)

[∃e ≈ u][∃e][RR(e , e) & I visit the museum(e) & Yesterday(e)]

As the visit e was Yesterday, τ (e) is confined to that day, and the onset of the resultant e of e is also there. By using yesterday, one puts all of yesterday as PAST. As the tense is −Past, τ (e ) is PRESENT; but its onset is PAST, and this is a category mistake, as seen in (4.17): (4.17)

e’ [

e

] u onset of e’

v

v today

yesterday

PAST

PRESENT

...

Replacement of yesterday with recently, however, gives (4.18) (4.18)

e’ [

e

] u

onset of e’ v yesterday

v today

...

)

RECENTLY

v

... PRESENT

...

190

J. Higginbotham

The visit e is concluded before the utterance u, and so the onset e of e is before u as well. The adverbial predicate is closed under temporal succession, so that the RECENT period can be as in (4.18). Nothing prevents the PRESENT from subsuming a period that includes the onset of e . The general conditions on English Sequence of Tense allow a −Past as well as a +Past complement to be anaphoric to a −Past superordinate INFL, as in (4.19): (4.19)

John will say that Mary is happy.

In (4.19) the speaker may predict a future, present-tense, utterance by John. Similarly for (4.20): (4.20)

In his speech next year, the President will say that the legislation he will sign next week is working well.

As for +Past complements, we have examples like those in (4.21)–(4.23): (4.21)

John will say two days hence that Mary was happy that day.

(4.22)

John will say two days hence that Mary was happy today.

(4.23)

John will say two days hence that Mary was happy the day before

Replacing the complement Past with the Present Perfect, we have first of all (4.24)– (4.25): (4.24)

John will say two days hence that Mary has been happy that day.

(4.25)

John will say two days hence that Mary has been happy today.7

But also (4.26): (4.26)

∗ John

will say two days hence that Mary has been happy the day before.

I believe that (4.26) represents an anaphoric or shifted violation on the conditions on the Present Perfect. Under Sequence of Tense it will have the truth conditions shown in (4.27): (4.27)

[∃e > u] {Say (John, ∧ [∃e ≈ e][∃e ]{RR(e , e ) & Happy (Mary, e ) & e on the day before the day of e], e) & e is two days after the day of u}

The speaker’s present is unaffected; but we may suppose that the tense anaphora sets up a relative past and present as shown in (4.28):

7

This example is, I think, comparable to cases such as (i), discussed in Higginbotham (2002):

(i)

John will say that Mary is dancing well.

in that they become the more natural the more one considers that John already believes the complement, at the time of saying (i).


191

(4.28)

e’

[

u

e"

] e

onset of e’ v

|

|

today

v

v

tomorrow

two days hence

RELATIVE PAST

...

RELATIVE PRESENT

v

...

PRESENT Briefly: because the day before stands to two days hence just as yesterday to today, the condition (P) on the Present Perfect is violated: the day before is not closed under temporal succession. Very well, assuming the datum (4.26). However, at least one speaker of British English with whom I have discussed these matters with does not particularly object to (4.26). I noted above, as many have observed, that the “Perfect Puzzle” disappears with the Past Perfect in (4.13) and the like. It may be that for some speakers the puzzle disappears altogether under tense anaphora; in other words, that even though the reported utterance in (4.26) could not have been a Present Perfect (John is not predicted to say, “Mary has been happy yesterday”) its content can be reported indirectly with that form. A further, possibly interfering, factor is that the Perfect Puzzle itself can be attenuated, as illustrated above in (4.3). Other illustrations may include (4.29) and the like: (4.29)

John is not happy now, but he HAS been happy, long ago.8

Finally, I note that the context (4.26) does limit the ways in which the day in the complement clause can be designated; in particular, it cannot be designated with a

8

Thus compare (i) to (ii):

(i) John will say in three years that Mary has been happy the year before. (ii) John will say in three years that Mary HAS been happy, the year before. or with the stress pattern and contraction as in (iii). (iii) John will say in three years that Mary’s been HAPPY the year before. I find (iii) out of the question, (ii) somewhat marginal.

192

J. Higginbotham

speaker’s Future indexical, as in (4.30): (4.30)

∗∗ John

will say two days hence that Mary has been happy tomorrow.

For some further discussion and references, see Higginbotham (2006).

5 Conclusion In this article I have examined the scope of the Result Perfect in English, observing en route that, quite apart from bringing in such objects as Result and Resultant states, any account that takes positions for events and situations as implicated in the semantics of human first languages must find them in a variety of causal and other contexts, and even in adverbial and quantificational expressions. It is well of course to look at individual and detailed examples; but the overall picture is that of a very general hypothesis, the evidence for which must rest on a manifold of consequences.9

References Boogaart, R. (1999). Aspect and Temporal Ordering: A Contrastive Analysis of Dutch and English. Doctoral dissertation, University of Amsterdam. Giorgi, A. and Pianesi, F. (1998). “Present Tense, Perfectivity, and the Anchoring Conditions.” Wyner, A. (ed.), Israeli Association for Theoretical Linguistics 5. Bar-Ilan University, Ramat-Gan, Israel. Higginbotham, J. (2000). “Accomplishments.” Proceedings of Glow in Asia II, Nagoya, Japan. Nagoya: Nanzan University. pp. 72–82. Higginbotham, J. (2002). “Why is Sequence of Tense Obligatory?” Gerhard Preyer and Georg Peter (eds.), Logical Form and Language. Oxford: Clarendon Press. pp. 207–227. Higginbotham, J. (2005). “Event Positions: Suppression and Emergence.” Theoretical Linguistics 31, 3, pp. 349–358. Higginbotham, J. (2006). “The Anaphoric Theory of Tense.” Forthcoming in Proceedings, Semantics and Linguistic Theory 16. Iatridou, S., Anagnostopoulou, E., and Izvorski, R. (2005). “Observations About the Form and Meaning of the Perfect.” MS, MIT, Cambridge, MA. Jespersen, O. (1924). The Philosophy of Grammar. Reprint New York: Norton. McTaggart, E. (1908). “The Unreality of Time.” Mind 17, pp. 456–473. 9

Versions of this article, or of parts or longer discussions related to it, were presented at the Paris conference on Tense and Modality, December 2005, and later at Kyoto University and Sendai University Japan; MIT; the University of London; University College Dublin; and the Scuola Normale Superiore, Pisa, Italy. I am grateful to the audiences at these events, and further to Jacqueline Guéron and Jacqueline Lecarme for a number of comments.


193

Ogihara, T. (1995). Tense, Attitudes, and Scope. Dordrecht, Holland: Kluwer. Pancheva, R. and Stechow, Arnim von (2004). “On the Present Perfect Puzzle.” Moulton, K. and Wolf, M. (eds.), Proceedings NELS 34. University of Massachusetts, Amherst, MA. Parsons, T. (1990). Events in the Semantics of English. Cambridge, MA: The MIT Press. Portner, Paul (2003). “The (Temporal) Semantics and the (Modal) Pragmatics of the Perfect.” Linguistics and Philosophy 26, pp. 459–510.

Tense and Modality in Nominals Jacqueline Lecarme

Abstract It has been shown that, at least in certain languages, tense is represented in nominals (Lecarme, 1996, 2004). It is therefore important to consider whether the interplay between tense and modality which is commonly found in clauses exists in the nominal domain as well. This article investigates the non-temporal meanings of nominal tenses in Somali, an Afroasiatic language. It explores the conditions under which nominal past morphology is interpreted in the modal dimension, contributing either a quantificational reading of the past DP (comparable to English -ever in e.g., ‘whenever’) or an evidential reading focusing on the visible / non-visible distinction. It is argued that the common abstract feature underlying the various meanings of past morphology in nominals is a more primitive feature of “exclusion/dissociation” (Iatridou, 2000). To account for the link between direct evidentiality and visual perception in nominals, it is proposed to extend Kratzer’s (1981, 1991) theory of ‘doubly relative’ modality to include a perceptual component. In this revised framework, past morphology gives rise to ‘non-actual’, ‘unknown’ or ‘invisible’ modal meanings, depending on different choices of modal base and ordering source. Key words: Tense, modality, evidentiality, definite articles, demonstratives, visual perception.

Introduction1 The modal use of a past morphology (e.g., in conditional or hypothetical clauses) is a well attested phenomenon in natural language. Informal treatments suggest that 1 This chapter develops ideas outlined in Lecarme (2003), and is based on material presented at the DIP Colloquium (University of Amsterdam, October 2003), the 27th GLOW Colloquium (Aristotle University of Thessaloniki, April 2004) and the International Round Table on Tense, Aspect and Modality (Universités Paris 7 and Paris 3, December 2005). I wish to thank the participants at those events. I am particularly indebted to Jacqueline Guéron for valuable written comments on an earlier version, and to Bascir Kenadid (Bashiir Nuur Keenadiid) for insightful discussion of the Somali data.

Laboratoire de Linguistique Formelle, CNRS-Université Paris Diderot (Paris 7) J. Guéron and J. Lecarme (eds.) Time and Modality, 195–225. c Springer Science+Business Media B.V. 2008

195

196

J. Lecarme

past tense does not receive a temporal interpretation, but rather expresses a modal ‘displacement’. Iatridou (2000) proposes that a unified account of temporal or modal displacement can formally be construed as a set-theoretic exclusion/dissociation relation between a topic time or world and a speaker time or world. In her terms, past morphology realizes an ‘exclusion’ feature whose meaning in the temporal domain is that the topic time (i.e., the time interval (set of times) we are talking about) excludes the utterance time, and whose meaning in the modal domain is that the topic worlds exclude the speaker’s actual world. Izvorski (1997) takes the same approach in her pioneering study of (indirect) evidentiality, suggesting that the ‘exclusion’ mechanism interacts with more specific domains of modality, such as the linguistic expression of a speaker’s source of knowledge (or information). The purpose of this article is to support such accounts, by providing a new perspective. More concretely, I argue that the interaction of tense, modality and evidentiality has a natural parallel in the nominals domain. My study is based on Somali, an East Cushitic language where nominals display an elaborate tense system.2 In the present article, I investigate more precisely the non-temporal meanings of nominal tenses, and show how a nominal past, in addition to being a temporal item, displays properties of a modal item (e.g., apparently contributing a quantificational reading of the past DP), or an evidential item (contributing the visible/non visible distinction). I will approach the issue from two perspectives: (i) Under which conditions is a past nominal interpreted in the modal dimension? and (ii) Can we give a unified account of nominal pasts in their temporal, modal, and evidential meanings? Our discussion will be in three parts. First, I introduce an account of the phenomenon, and discuss the sense in which it can be thought of meaningfully as deserving the designation ‘nominal tense’, that is, ‘grammaticalized location in time’ by Comrie’s (1985) definition. Second, I turn to a discussion of generic/habitual sentences, as well as adverbial and conditional clauses in the present tense, where the nominal past morphology is interpreted not along a temporal, but along a modal dimension. I show that the numerous instances of the interaction of the past tense with the modal domains, for which we have a wide crosslinguistic evidence, have systematic parallels in nominals. This will lead us to the third issue, the use of a nominal past as a direct (i.e., perceptual) evidential. I will explore the possibility that Kratzer’s (1991) notion of ‘doubly relative’ modality might lead to a unified analysis of the tense/direct evidentiality connection, offering, at the same time, a perspective to explain the core facts presented here in a consistent way.

2

See Lecarme (1996, 1999b, 2003, 2004). ‘Nominal tense’ is a well attested morphological phenomenon (see Nordlinger and Sadler (2004) for typological investigation from a wide-ranging survey of languages where the phenomenon occurs). Further elaboration depends on a fine-grained syntactic analysis of the relevant structures and the sentences in which they occur. If the ‘temporal markers’ in some languages are actually nominal aspect, not nominal tense, as argued by Tonhauser (2006) for Guaran´ı, then we might expect languages to allow tensed, aspectual, or tenseless nominals, a further parallel with clauses.

Tense and Modality in Nominals

197

1 Tense in Nominals 1.1 Space and Time To appreciate the role of nominal tense marking in Somali syntax and semantics, it is necessary to introduce some of the basic features of the determiner system (see Lecarme, 1996, 2004). Both the definite article3 and the demonstrative enclitics appear suffixed to the noun (in both cases, the initial consonant-k-/-t- is a gender morpheme which agrees with the noun). Somali demonstratives mark distance in space relative to the speaker, encoding a four-way distinction in proximity with regard to the speech act (1). (1)

Spatial demonstratives: -k/t-án ‘this, these’ -k/t-áas ‘that, those’ -k/t-éer ‘that, those’ (visible middle distance) -k/t-óo ‘that, those yonder’ (far away, still visible)

Besides this, Somali and the closely related East Cushitic languages have a morphology specific to the temporal domain (2). The Somali nominal tense system is, basically, the morphological encoding of past and nonpast (present or generic), which surfaces as a vocalic i/a opposition also found in the verbal system in most Afroasiatic languages. As discussed in Lecarme (1996), the relevant opposition has to be understood as i/ Ø, a being the unmarked, ‘default vowel’ in Afroasiatic.4 (2)

Tensed definite article: [−past] [+past] [+nom] -k/t-u k/t-ii [−nom] -k/t-a k/t-´ıi

Demonstratives and definites can cooccur: n´ın-kan-u, ‘this man’, baabúur-taas-i ‘those trucks’ sh´ıl-kaas-i ‘that accident’. As in other languages with complex determiners (e.g., Germanic), ‘double definiteness’ has the effect of an intensifier.5 3

In Standard Somali, no indefinite article appears on the surface: a man is expressed with the Somali equivalent of man (the indefinite form can mean either non-specific or specific indefinite). 4 In Lecarme (1996, 1999b), the suffixed definite forms are derived by syntactic N-to-D movement. A morphological merger analysis of the sort developed by Embick and Noyer (2001) would rather treat these cases as involving Lowering the entire D, which is internally complex, to the head of its NP complement, leading to a parallel morphological structure root-φ -tense for both the tensed nominal and verbal forms. See Lecarme (2004, 343–344) for more details. 5 This distributional fact is generally not mentioned in the literature (but see Kirk, 1905, for a thorough description, based on the Isaaq dialect of Somali). Since the relevant DPs can occur in non-subject position, the past ending is not to be assimilated to the −i nominative morphology which is marked at the right edge of the phrase under certain conditions (see Anderson et al., 2006, for a recent analysis). Assuming that Somali D is Ø with gender agreement and tense incorporated

198

J. Lecarme

From the fact that both demonstratives and tensed definite articles can be used deictically, should we conclude that they belong to the same category, ‘deictics’? There are several important points worth noting here. First, Somali demonstratives, unlike English this and that, never lose their (deictic) spatial function. Even in their anaphoric uses (e.g., in texts), the demonstratives -an and -aas remain demonstrative in that they pick up a referent in a set whose salience is determined spatially. Second, demonstratives (universally) do not function in narrow syntax,6 only in interpretation of the information they provide: there are no locative elements corresponding to tenses. Somali tense morphology is inflectional, and determines crucial aspects of the narrow syntax of DP. Third, unlike English that/those-phrases, which can have narrow scope readings, Somali demonstratives are scopally inert. Past DPs are able to take narrow scope with regard to operators and quantifiers. I will examine the last two points in more details below.

1.2 The D-T Relation Tense-marking on nominals is an inescapable feature of Somali grammar. Understanding the past/nonpast opposition marked on the definite articles is part of the basic linguistic competence of Somali speakers. As I discussed elsewhere (Lecarme, 1996, 1999b, 2003, 2004), nominal tense is independent of clausal tense, even if they generally coincide (or overlap). This effect is very sharp in relative clauses, where the tense of the antecedent DP need not be the same as the tense of the relative clause. For example, a nonpast DP is characteristically used, as in (3a), when the referent is a famous person, or when the speaker wants to discuss a person or thing in the context of the utterance (3b, c).7

to D (T-to-D), the relative order of the possessive and demonstrative enclitics suggests that demonstratives are merged higher than Spec,TP (presumably, Spec, DP). A full account of how this hierarchical structure is linearized is beyond the scope of this article. 6 Narrow syntax is taken here as the core recursive aspect of the language faculty (Chomsky, 1995, and subsequent work). 7 Here and below, the absence of explicit gloss indicates a nonpast morphology (whether interpretable or not as a present tense). The examples in the article follow the national orthography with small modifications: the main tonal accent of a prosodic word is noted with an acute accent; hyphens are added for morpheme separation. Key to Somali gloss: C/F = complementizer/‘focus marker’ (I analyze the baa/waa morpheme as a declarative root complementizer); defF/M = definite feminine / masculine article, dem = demonstrative, dep. = dependent, dir. = directional particle, expl = expletive, gen = general/generic, Poss = possessive, prog. = progressive, neg. = negation, restr. = restrictive agreement. Pronominal clitics are identified by their person, gender, and number features (upper cases). Referential third object pronouns are Ø in Somali. Low cases = agreement features. Most examples in the text are from my previous papers, from Shabeelnaagood (a Somali play by Hassan Sheikh Mumin), from the BBC Somali Service, and from Google.


(3)

a.

b.

c.

199

N´ın-ka cáan-ka ah ee búug-gani qoray man-defM. fame-defM be and book-defM.dem.past wrote[+nom] (waa Shakespeare F/C Shákisb´ıir) ‘The famous man who wrote this book (is Shakespeare)’ Ma o´ gtahay qóf-ka aan u´ codeeyay? Q know.2s person-defM 1S for vote.past ‘Do you know whom I voted for?’ Arr´ın-ta Xámar i keentáy (baan ku shéegayaa) affair-defF Mogadishu 1S brought F.1S 2S tell.prog ‘(I am telling you/ going to tell you about) the affair which brought me to Mogadishu’

It appears, then, that both nominal and verbal/clausal tenses are computed in relation to the speech time. Thus, assuming that the definite article is a D(eterminer) heading a DP (determiner phrase), the domain of interpretation of the tense marked on D is the DP. In previous work, I proposed that parallel to the C-T relation expressing clausal finiteness, there is a D-T relation expressing ‘nominal finiteness’.8 For the purpose of the present article, I will assume that DPs have a structure (4), where D is formally parallel to C, n∗ is parallel to v∗ and selects the ‘external’ (possessor) argument, n is a light noun, and NP, n∗ P and DP define three syntactic domains for tense interpretation:9 (4)

[DP D[TP T[n∗P n∗ [nP n[NP N]]]]]

This hypothesis immediately accounts for the fact that the past determiners are scope bearing elements with regard to the interpretation of modifiers. Like English before, the (non-intersective) adjective horé ‘before’ is a general ordering predicate that situates or characterizes an entity in either time or space from a reference point. As the examples show, horé has a spatial meaning when used with a spatial demonstrative (5a) or a present nominal tense (5b), and a temporal meaning in the domain of a nominal PAST: in other words, PAST is required for horé be interpreted as ‘former’ (5c). In contrast, only a present nominal is compatible with a modifier like máanta ‘today’. This is best illustrated by (5d). 8 I envisage a parallel treatment of CP and DP, as often implied in the literature (see Hiraiwa, 2005, for a recent syntactic account). On the semantic side, many parallels have been suggested between nominals and clauses: verbs in general are assumed to have neo-Davidsonian meanings, where the VP names a ‘property of eventualities (events and states, cf. Bach, 1981), not a single event. Events are assumed to have the same logical type as individual entities, that is, < e >. ‘Sentence determination’ is obtained by quantifying over the event variable (Krifka, 1992; Partee, 2000; von Stechow, 2002). Building on these studies, I will assume that noun phrases, like sentences, contain implicit quantification over times, and implicit restrictions to times contained in a contextually salient interval. 9 The functional head n∗ is labelled Appl(icative) in Lecarme (2004). In the following, I will use ‘tense’ as ‘morphological tense’ (possibly null), T as the syntactic category T(ense), the locus of computation of tense features ([T]), and PAST as semantic tense (some abstract temporal quantifier).

200

(5)

J. Lecarme

Spatio-temporal modifiers a. Tuulá-dóo horé (éeg!) village-defF.dem before look ‘(Look at) that village in front!’ b. J´ıd-ka hore má ahan ee waa kán labaad street-defM before Neg is and F/C m.dem second ‘It is not the next street ahead, but the second one’ c. (Waa) tuulá-d´ıi hore (oo aad degganayd) C/F village-defF.past before (and 2S lived.2s) ‘(This is) the former village (where you lived)’ d. Máxay kú kala duwán yihiin noló-sh´ıi dád-k´ıi What.3P at between different are.3p life-defF.past people.defM.past hore iyo tán dád-ka maánta? before and f.dem people-defM today In what does the life of former people differ from the one of today’s people?

Nominal tense is ‘inflectional’ in Somali in that the D-T complex determines crucial aspects of the internal syntax of DP, such as Case assignment, agreement and deletion processes (Lecarme 1996, 1999b, 2004). I argued that structural genitive Case involves T in addition to D: elements with inherent case selected by n∗ remain in Spec,n∗ P, pronominal possessors in Spec,TP are assigned Case by T in conjunction with D. Tense also can be used as a feature of agreement (or concord). As (6) shows, Somali adjectives must agree with a definite head noun in gender and tense.10 (6)

Tense agreement (adjectives) a. Xaash´ı-da yar ee cád (buu kéenayaa) paper-defF small and white (F/C.3MS is.bringing) ‘(He is bringing) the small white sheet of paper’ b. Xaash´ı-d´ıi yarayd e´ e caddayd (buu keenay) paper-defF.past small.past.f and white.past.f (F/C.3MS brought) ‘(He brought) the small white sheet of paper’

At the same time, Somali (among other Afroasiatic languages) has a syntactic condition to the effect that noun modifiers in the context of D[Definite] must be marked morphologically for definiteness.

10

On the special morpho-syntactic status of nominal number, see Lecarme (2002).


(7)

201

Tense and definiteness agreement (NP modifiers) a. ardayád soomaal´ı ah student.f Somali.f be ‘a Somali student(f)’ b. Ardayád-da Soomaal´ı-da ah (way imáaneysaa) student.f-defF Somali-defF be (C/F.3F is.coming.f ‘The Somali student(f) (is coming)’ c. Ardayád-d´ıi Soomaal´ı-da ahayd (wáy timid) student.f-defF.past Somali-defF be.past.f (C/F.3F f.came) ‘The Somali student(f) (came)’

In uttering (6b), the language user does not assert that the smallness and whiteness of the sheet of paper is ‘past’ relative to the utterance time: if that were the case the use of past tense would be at best unfelicitous, as under the most natural interpretation the state of the sheet of paper being white and small extends through the time of utterance into the future. Similarly, the use of past in (7c) does not mean that the individual-level predication ‘be Somali’ held of the student (only) prior to the utterance time.11 In addition, the DP modifier in (7b, c) is not interpreted as a definite description. The most telling examples are idioms, since idioms typically do not have compositional meanings. The paradigm for an idiomatic noun+ adjective combination thus will be as follows: (8)

Idioms a. siyaasadó foól xun policies face ugly ‘ugly policies’ (lit. ‘ugly-face’ policies) b. siyaasadá-ha foó-sha xun policies-defM face-defF ugly ‘the ugly policies’ (lit. ‘the face’) c. siyaasad´ı-h´ıi foó-sha xum-aa ee uu policies-defM.past face-defF ugly.past and 3SM guméysi-gu doónayey ´ınuu. . . colonialist-defM[+nom] wanted COMP.3MS ‘the ugly policies with which colonialists wanted to. . . ’

Space prevents me from reviewing all the evidence. What is important here is simply the overall generalization: nominal Tense, like Definiteness, is interpreted syntactic material. [Definite] and [T] pattern together because they are both features on D. Agreement morphemes reflect certain syntactic properties, but do not in any sense contribute these properties to syntax. Assuming the recent minimalist framework (Chomsky, 1995, 2000, 2001), these instances of past tense represent unvalued features (uT) that must be valued valued at the D-phase level, in a local relation to an interpretable PAST (the D-T connection). 11

On the syntactic and semantic difference between adjectives (white) and stative verbs (be white), and between inflected adjectives/predicative nouns and (reduced) relative clauses in Somali, see the discussion in Lecarme (2004).

202

J. Lecarme

1.3 Time and the Noun 1.3.1 On Tense Morphlogy and Temporal Interpretation Assuming that syntax and semantics run in parallel from the bottom to top, three (sets of) times are potentially involved in the temporal interpretation of noun phrases: (i) the time of the predicate (or predication time), that is, the times at which a property like ‘dog’ or ‘president’ is asserted to hold of an individual, (ii) the time of the genitive/possessive relation ‘R’ (Higginbotham, 1983b; Partee, 1983), and (iii) the time of (existence of) the individual.12 The corresponding syntactic domains are NP, n∗ P and DP, respectively. The relevant Somali examples are as follows: (9)

a.

b.

c.

Madaxweyn´ı-h´ıi hore (wuu imaáneyaa) president-defM.past before C/F.3SM is.coming ‘The former president (is coming)’ (Wáxaan ku bárayaa) afáday-d´ıi hore expl.C/F.1S 2S introduce.prog wife-defF.Poss1S-defF.past before ‘(I am introducing to you) my ex-wife’ Weerár-k´ıi ków iyo tobán-k´ıi Sebtémber (ma lá attacks-defM.past one and ten-defM.past September Q with socotaa?) follow.2s ‘(Do you know about) the September 11 attacks?’

As discussed in section 1.2, the spatio-temporal adjective horé, like English before, express temporal precedence when it is bound by a PAST tense. In (9a), horé applies to the predicate, creating a predicate true of individuals that once had the property ‘president’ (horé is needed to convey the aspectual meaning of ex- or former). In (9b), horé applies to the possession relation: the nominal can be used to refer to someone who was formerly ‘my wife’, and is ‘my wife’ no longer. As our examples make it clear in both cases, the time when the property ‘president’ or ‘my wife’ is true of the individual must be a subset of the intervals bound by horé, but in order for those intervals to be temporally located in the past, a nominal PAST is needed. Similarly, in (9c), the running time of the event (the attacks) must be included in the reference time (September 11). The effect of a nominal PAST is to constrain the denotation of the temporal modifier ‘September 11’ to a set of past times. Nominals then are much similar to clauses with regard to temporal interpretation: PAST makes it possible to chain together various elements of the interpretation of noun phrases. Since the core eventuality is anchored at the utterance time, there is no other source of PASTness in our examples. 12

Here and below, I assume that there is no asymmetry between the application of tense to individuals and to events. The individual’s ‘time of existence’ is to be understood as the time associated with underlying existential quantification (making the temporal location of an individual parallel to the temporal location of an event), not the whole time-span of an individual’s existence, which is, in case of animate individuals, their lifetime (Kratzer, 1995; Musan, 1995, 1997).


203

1.3.2 On the Temporal Location of Individuals Most of the previous work on the temporal interpretation of noun phrases (in English and German) has centered on the temporal location of predication times.13 We owe to Enç (1981) the claim that the interpretation of fugitives in sentences such as ‘every fugitive was in jail’ is independent of the tense that is present in the syntax (the possible readings involve individuals who are fugitives now, or were fugitives before they were in jail). Enç (1986) proposed that nouns (like verbs) must be provided with temporal arguments, the value of which are supplied by the utterance context. Important observations made by Musan (1995) led her to question this analysis. According to her, the correct generalization is that only strong (presuppositional) NPs are temporally independent. Cardinal noun phrases (e.g., weak noun phrases in existential there-constructions) can have temporally dependent readings. The temporal location of individuals has received comparatively little attention. Musan (1995, 1997) does provide a pragmatic account of the ‘lifetime effects’ that individual-level predicates like ‘be from America’ or ‘have blue eyes’ impose to their subjects.14 But Musan’s theory aims at explaining in which way the temporal location of individuals is determined or affected by the temporal interpretation of a clause, and crucially depends on an ontology that contains stages (temporal parts) of individuals as basic entities (of type e). On this view, an individual’s ‘time of existence’ is actually the time of existence of a stage of an individual. Deeper insights into the temporality of noun phrases, I argued, can be gained by the study of systems where nominals encode temporal distinctions. In Somali, tense is a feature of any (common) noun, not only nouns that are said to include an event as part of their lexical semantics. The main function of nominal tense in the language is to temporally locate individuals like the table and my brother.15 Syntactic Definiteness and Tense draw the difference between the ‘strong’ pronouns, which have the same distribution as ordinary DPs, and the ‘weak’ pronouns (semantically definite) associated with them. Since pronouns do not contain nouns that can be predicated of a stage, the following examples illustrate a clear case where nominal tense locates an individual temporally, at a past time which overlaps (or coincides with) the time of the event: (10) a. Isá-g´ıi baa hádal-k´ıi qaatáy, oo yiri: HE-defM.past C/F talk-defM.past take.past and m.said ‘HE began to speak, and said:’ b. D´ıbna umá arág isá-g´ıi iyo Back.any at.NEG see HIM.defM.past and lacágtay-diiba. money.f.Poss1S-defF.past.any ‘Afterwards I did not see HIM again, nor my money’ 13 On the temporal location of the genitive/possessive relation, see Burton (1997) and Larson and Cho (1999, 2003). 14 See note 12 above. 15 As expected, proper names, kind-denoting DPs (e.g. naasle´ y-da ‘the mammal(s))’ and DPs that denote abstract entities (e.g. xisaáb-ta ‘the mathematics’) do not reflect tense distinctions.

204

J. Lecarme

Both Enç ’s and Musan’s generalizations can be derived from the nominal tense hypothesis. Assuming that the time variables of nominals are bound DP-internally by a nominal tense operator, the scope paradoxes that motivated Musan’s stage analysis do not arise.16 The cardinal vs. presuppositional distinction is tense-related: if true D (i.e., D + T) relates to referentiality in some sense, an indefinite nonspecific nominal phrase (‘a lot of people’, ‘someone’, etc.) is temporally indefinite. A crucial piece of evidence for this is that nonreferential nominals (nonspecifics, quantified and predicate nominals, etc.) are tenseless in Somali. On this view, the distribution of temporally dependent/independent noun phrases coincides closely with Chomsky’s (2000, 2001, 2008) notion of phase: weak (temporally indefinite) NPs are merged in the VP and interpreted in situ, whereas ‘presuppositional’ (i.e, tense-related) DPs occupy higher positions in the syntax.17

1.4 Interim Discussion To recapitulate the conclusions so far, crucial aspects of temporal interpretation of nominals in Somali are strictly determined by linguistic form, then belong to the syntax and semantics of the language. Depending on the tense marked on the DP, (simple) event and process nominals (e.g. exhibition, trip, ceremony) trigger distinct presuppositions, as reflected by the English translations of the examples. (11)

Presuppositions a. Bandh´ıg-ga máad daawatay? exhibition-defM Q.2S see.2s.past ‘Have you seen the exhibition?’ (still running at UT) b. Bandh´ıg-g´ıi máad daawatay? exhibition-defM.past Q.2S see.2s.past ‘Did you see the exhibition?’ (closed at UT)

Both utterances (11a) and (11b) carry an existence presupposition: the sentences would be used most naturally by a speaker who assumes that the hearer knows that there is a (unique) exhibition—the semantic contribution of the definite article. A second presupposition is associated with tense: by uttering (11b), a speaker makes it clear that she takes for granted that the exhibition is ‘over’ at the time of her utterance. The temporal presupposition is not required to be in the common ground.18 16

See Musan’s (1995, 89–92) arguments against a scope approach of the presuppositional vs. cardinal distinction (Diesing, 1988, 1992), namely that presuppositional noun phrases under the scope of a temporal adverb of quantification still can be temporally independent. 17 In the framework of Chomsky (2001, 2008), a DP with an interpretable tense feature should reach the edge position of the phases v∗ P and CP (i.e., outer Spec, v∗ and Spec, T, if T inherits the edge feature of C) typically associated with ‘surface structure’ interpretation (specificity, new/old information, topicality, etc.). 18 See Lecarme (1996, 1999b) for more examples. The fact that the temporal presupposition is not required to be in the common ground makes it more similar to assertions than a presupposition


205

The presuppositional content is not an implicature either, since conversational implicatures are typically cancellable. The presuppositional fact encoded in the linguistic meaning of nominal expressions—namely that the exhibition is ‘past’ or ‘present’ relative to the time of utterance—is semantic in the sense that it represents what the speaker knows, independently of her communicative intention. On the assumption that the semantic conditions on the temporal interpretation of noun phrases are similar across languages, what distinguishes, say English (or German) from Somali? One possibility is that in languages where the temporal interpretation of nominals does not show up in linguistic form, these temporal interpretations are pragmatic, not semantic. They are accordingly not represented. A more plausible possibility is that the English DP reaches the semantic interface in the same form as the Somali DP, as we would expect if the external systems of interpretation are essentially language-independent. On this view, the temporal arguments of nouns are represented in the syntax as empty categories sensitive to their syntactic environment. Perhaps supporting this conclusion is that nominal tense features are also often phonologically expressed in some fashion. According to Pesetsky and Torrego (2001), the property of nominals called ‘structural case’ is actually unvalued T on D. Determiners across languages are semantically complex elements with inner structure. Languages with ‘double definiteness’ effect provide some evidence that definiteness is comprised of two parts: uniqueness and specificity (in the sense of Enç , 1991).19 In their anaphoric uses, non-proximal demonstratives (e.g., English that/those) are prone to semantic reinterpretations in which the basic deictic meaning has been lost. It is likely that the language faculty does not stipulate which features serves as the past ‘exclusion’ function in the nominal domain. It merely requires some feature of N to perform this function.

2 Modality 2.1 Time and Worlds Crosslinguistically, past tense morphology often does not indicate temporal anteriority (e.g., in English and Japanese, the non-temporal use of a past tense carries a definitive meaning of counterfactuality). Most linguists think that these different uses are closely related and are not merely uses of distinct homonymous lexical items. This lends support to Iatridou’s (2000) idea that past tense is used as an exclusion/dissociation feature, formally expressed as (12). (12)

T(opic)x excludes Cx The variable x ranges over times or worlds (Iatridou, 2000, 246)

under a Stalnakerian analysis, although the presupposition is not directly asserted. This, however, is left for future discussion. 19 Roehrs (2006, on Scandivian) makes the interesting claim that ‘specificity’ originates in a syntactic position lower than D, and that determiners are parallel to auxiliaries in the clause.

206

J. Lecarme

When the variable ranges over times, past tense expresses a relation of precedence between the Topic Time (Klein, 1994) and the Utterance Time. When the variable ranges over worlds, past tense morphology allows the set of topic worlds to excludes the actual world, thus establishing the implicature of counterfactuality. In this section, I argue that the analysis naturally extends to the nominal domain: nominal past is among the devices that make it possible to get access to non-actual worlds for the interpretation of noun phrases. I describe how the generic operator interacts with definite expressions, using first ordinary definites and then free relative examples.

2.2 Generics and Habituals In any sentence expressing a generalization, nominal past does not have a past temporal meaning. Rather, it apparently introduces a modal dimension into the interpretation, by allowing variability of reference of the DP. Consider, for example, the contrast between (13a) and (13b): (13)

Generics a. W´ıil-k´ıi wanaagsani waa ´ınuu wax bartó boy-defM.past good[+nom] C/F COMP.3MS thing learn.pres.dep ‘A (lit. the) good boy must learn something’ (deontic generalization) b. W´ıil-ku / w´ıil-ka wanaagsani waa ´ınuu wax boy-defM[+nom] /boy-defM good[+nom] C/F COMP.3MS thing bartó learn.pres.dep ‘The boy/the good boy must learn something (now)’

(13a) expresses a normative statement, or a lawlike generalization (ideally, a good boy must learn something according to some set of rules). There need not be boys in the context of use: generic quantification is over (deontically) possible individuals, not just the actual ones. Compare with (13b), where the nonpast DP is construed as an episodic subject (‘unique boy who must learn something in the actual world’).20 Similarly, past DPs are required in habitual sentences, involving states or processes that are typically regularly repeated.21

20 Note that the nominal past is the only apparent source of genericity in examples such as (13a). The interpreted modality is not associated with any lexical item in the structure (cf. the English ‘is to’ or ‘has to’ constructions): the waa in construction is formally a verbless sentence, where the root complementizer waa is a pure, non verbal copula and the ´ın-clause a complement CP (see Lecarme 1999a, 286–287). 21 As in English, only stative verbs in Somali have a non-habitual interpretation in the present tense, hence the examples in (14–15) are unambiguously generic/habitual.


(14)

207

Habituals a. Aróor-t´ıi wáxaan toosaa l´ıx-da morning-defF.past expl.F/C.1S wake-up.pres.gen six-defF ‘In the morning I wake up at six’ b. Góor walba dúmar-k´ıi baa a´ wr-ta fura time each women-defM.past F/C camels-defF unload.pres.gen.restr. oo aqallá-d´ıi dhisa and huts-defM.past build.pres.gen.restr. ‘Always the women unload the camels and build the huts’

Habitual sentences are a special case of generalization: over parts of worlds or situations (Kratzer, 1989; Carlson, 1995). Sentence (14a) can be paraphrased most naturally as ‘in general, if there is a morning situation, I wake up at six in that situation’. The temporal noun aroórtii ‘the morning’ is not used to denote a particular time interval here. Likewise, in (14b), where the sentence quantifies over typical situations of the nomadic life in Northern Somalia (the women unload the camels and build the aqal while the men take the stock to water), the past DPs need not be interpreted with the collective subject referring to specific groups of women. The sentence is not necessarily a generalization over instance of observed events: it can be reported in a book. The past DPs are not anaphoric to any antecedent explicitly present in the preceeding speech or text, and they are not ‘discourse anaphoric’ either. Here, past indicates that we are talking about non-actual, non-episodic individuals. A similar effect is observed in ‘inherent’ generics, those describing events that do not require instanciation, as in (15). On the attributive interpretation, ‘the winner’ in (15a) is relativized to the world of evaluation (i.e., ‘whoever is the winner in w’). Past DPs are also required in sentences that express statistical correlations, percentages, prices per kilo, etc., as (15b, c). Here, the distributive interpretations can be reinforced by adding the quantifier -ba ‘each’, as in (15c). (15)

‘Inherent’ generics a. Sánnad kasta qóf-k´ıi gúuleystaa wúxuu year each person-defM.past win.gen.[+nom] expl.F/C.3MS léeyahay kún shillin has.pres thousand shilling ‘Every year, the winner (lit. ‘the winning person’) will receive 1000 shillings’ oo sánnad-k´ıi már la b. Sh´ır-kaas conference-defM.dem and year-defM.past time one qabto holds.dep.pres.gen That conference that is held once a year (= annual conference) c. Kani waa diiqád sahlan oo kú dhacda ilaa sidéed m.dem.[+nom] F/C depression mild and to happens up.to eight hooyo tóban-k´ıi-ba mother ten-defM.past-each ‘This is a mild form of depression that occurs in up to eight out of ten mothers’

208

J. Lecarme

The phenomenon is quite general: similar generalization can be made in most languages about definite and demonstrative descriptions that take narrow scope under quantifiers (e.g., ‘The elegant man does not wear jewelry’). But if the quantificational variability effect is due to the relation between a generic quantifier and the iota definite description operator, as it is generally assumed, why should the DP be past? I propose that all these readings combine a quantificational dimension and a modal dimension. The quantificational dimension of the DP comes from a covert generic operator, as it is generally assumed after Lewis (1975) and Heim (1982). The modal dimension comes from the past dissociative morphology: the quantifier does not quantify over actual individuals, but over possible, non-actual ones. In other words, the -ii morphology is not a quantifier, but a modal determiner).22

2.3 Attitude Verbs In episodic sentences, DPs occuring as objects of attitude verbs expressing ‘ignorance’ or ‘uncertainty’ on the part of the speaker (or attitude holder) are marked as past. The sentences (16a,b) have a ‘subjective’ epistemic reading (cf. Lyons, 1977; Papafragou, 2006) in the sense that they are interpreted as relative to a speaker (or attitude holder)’s knowledge state at the time of utterance: (16)

a.

b.

Jawaáb-t´ıi garán maayo answer-defF.past know Neg.have.pres ‘I do not know the answer’ ´ W´ıx-´ıi dheeráad ah Allaa o´ g! thing-defM.past beyond is God.F/C knows.pres.restr. ‘(Only) God knows what(ever) lies beyond!’

In (16a), nominal past explicitly marks an epistemic possibility: the modal base consists of worlds which, for all the speaker knows at the time of her utterance, could be the actual world—a conjunction of propositions (e.g., ‘the answer is yes’ and ‘the answer is no’) both of which could be true without contradiction, given the speaker’s present state of knowledge. The construction can be paraphrased with an ever-free relative, as in (16b). The overall generalisation is that in Somali, definite descriptions can scopally interact with generic operators and propositional attitude verbs. Definite description that take narrow scope in relation to these operators are overtly ‘dissociated’ from the actual world: the source for the modality can be attributed to the same linguistic device in both cases. Past DPs in either their ‘variation’ or ‘ignorance’ readings are closely related to the free relative examples that I will now discuss.

22 This supports the view that generic sentences are implicitly modalized (Heim, 1982; Diesing, 1992). See Heim and Kratzer (1998, 165–170) for a related discussion.


209

2.4 Free Relatives There is no (uninterpretable) wh-like element in Somali.23 The variable element in plain free relatives (‘I read what you wrote’) is spelled out as qóf ‘person’ and wáx ‘thing’. Plain free relatives, then, are formally relative clauses headed by wáx-a ‘the thing’, qóf -ka ‘the person’, etc. The modal dimension of a relative clause correlates with the presence of the -ii morphology. The following examples have the universal or the free choice readings of English -ever free relatives: (17)

a.

b.

c.

d.

Qóf-k´ıi ilá kulma-ba wáx aanán aháyn buu person-defM.past 1S.with meet-each thing 1S.neg be.inf C/F.3MS i moodaa 1S think.pres.gen ‘Whoever meets me thinks too highly of me’ = The specific person (=qóf -ka) who meets me now W´ıx-´ıi aad heysáan i´ı keena thing-defM.past 2P have.2p 1S.to bring.2p ‘Bring me whatever you have’ = Bring me the specific thing (=wáx-a) that you have now W´ıx-´ıi wár ah e´ e aad na siis´ıd wáxaa la thing-defM.past new be and 2S 1S give.dep.pres expl.F/C one isticmáali doonaa use.inf will ‘We will use whatever information you send us’ aad rabt´ıd qaadó! K´ıi/búug-g´ıi defM.past/book-defM.past 2S want.2s take.imp.2s ‘Take whatever/whatever book you want!’

In every one of these examples, the denotation of the free relative covaries with the situation: the identity of the person who meets me differs in each situation, or different things or books are picked up in different possible continuations of the situation (the quantificational meaning can be reinforced by the polarity sensitive item -ba ‘each’). As in the cases discussed before, the suffix -ii contributes an additional modal dimension, signaling that the individual is not instantiated in the speaker’s actual world. Somali when-clauses also take the form of a relative clause introduced by a definite head noun meaning ‘the moment’, ‘the time’, etc. The presence of -´ıi correlates with the quantificational reading of the adverbial clause.

23 The Somali counterpart of English interrogative complements headed by who, what, when, where, how. . . are relative clauses headed by qóf-ka, c´ıd-da ‘the person’, wáx-a ‘the thing’, s´ıda ‘the manner’, etc. The interpretable interrogative element -ée (e.g. n´ınk-ée ‘which man’?) is not used in complement clauses or free relatives of any kind. See Lecarme (1999a) for a more detailed discussion.

210

(18)

J. Lecarme

When-clauses a. Már-ka la burburiyó Taleban,. . . time-defM one destroy.dep.pres Taleban ‘When (lit. the time) the Taleban will collapse,. . . ’ b. Már-k´ıi la gaboobó wax badán wáa la time-defM.past one becomes-old.dep.pres thing many F/C one illoobaa forgets.pres.gen ‘When (lit. the time) one gets old one forgets many things’

In (18a), ‘the time’ when the Taleban will collapse is a specific referential time, and even if this time is a future time, this time belongs to our world. In contrast, no specific referential time is available in the reading of (18b), where the past tense morphology is interpreted not along a temporal but along a modal dimension. Conditionals are modal in nature, in that they involve quantification over possible worlds/situations (Kratzer, 1981, 1991 a. o.). In Somali, all types of conditional clauses are modalized free relative headed by the past DP hád-d´ıi ‘the time’—interestingly, the present tense form hád-da means ‘now’ (lit. the (present) moment’). (19)

If-clauses a. Nin hád-d´ıi uu seexdó oo sóo man[+nom] time-defF.past 3MS sleeps.dep.pres. and dir. toosó, waa isá-g´ıi u´ n wakes-up.dep.pres. F/C him-defM.past only ‘If (lit. the time) a man goes to sleep and then wakes up, he is only himself’ (i.e. the same as he was before) (Proverb) b. Hád-d´ıi aad rabtó na ráac! time-defF.past 2S want.2s.pre.dep 1P follow.imp ‘Come along with us, if (lit. the time) you want!’

What, then, is the role of the -ii morphology? I suggest that even if the conditional is based on the actual world, the time of the conditional realization (i.e., the denotation of the free relative) is not in the world according to the speaker. It will be useful at this point to make explicit what analysis I have in mind for free relatives in Somali, and free relatives more generally. I am assuming here a now familiar analysis of free relatives along the lines of Iatridou et al. (2001, 224–225), namely that the nominal expression generated as non-specific (more specifically, as a N-head) is merged as a Spec of CP, possibly after extraction from within the VP (e.g., wáx [C [aad heysáan t]] = (17b)) and then projects a DP headed by N (the Somali complementizer of relative clauses is a phonologically non overt form). I propose that the definite determiner is further decomposable, and one of its constituents, the past tense morpheme, is interpreted not along a temporal but along a modal dimension. This account seems compatible with recent treatments of conditional clauses as free relatives of possible worlds (Schlenker, 2004; Bhatt and Pancheva, 2006).


211

2.5 Interim Discussion In the voluminous literature on English free relatives with -ever, it is widely assumed that free relatives are definite descriptions. The status of the morpheme -ever is more controversial. According to Larson’s (1987) analysis, -ever is a universal quantifier. Dayal (1997) argues that whatever-phrases have a modal dimension: in the speaker’s ‘ignorance’ reading, whatever-clauses quantify over epistemic alternatives that differ from the actual world only in the identity of the free relative referent (see also Iatridou and Varlokosta, 1997). According to von Fintel’s (2000) unified analysis, the content of -ever is presuppositional: -ever introduces a presupposition of variation over the denotation of the free relative (a definite expression) across possible worlds. This variation is construed either as ‘indifference’ (‘I grabbed whatever tool was handy’) or ‘ignorance’ (‘There is a lot of garlic in whatever (it is that) Arlo is cooking’), depending on the (counterfactual or epistemic) modal base. But whatever the analysis, a question remains: if the denotation of a definite description varies as a function of the values introduced by a higher quantifier, why do we also need -ever? My proposals come closer to Dayal’s conception of i(dentity)-alternatives as differing from each other only in the denotation of the free relative. From the perspective of the present discussion, what is reflected in the meaning of -ever, like the meaning of ii, is the core meaning of exclusion/dissociation. On this view, the modal properties of English every, who/what/when-ever and French quiconque, quelconque), are related to the temporal polarity items ever and onque of Old English (aefr ylc ‘ever each’)24 and Old French (qui c’onques, que c’onques > quiconque, quelconque). These morphologies should also be viewed synchronically as polar items bound by temporal or modal operators—not the time or world variable itself, but a linguistic sign of exclusion/dissociation.

3 Evidentiality Evidentiality is generally defined as the linguistic encoding of a speaker’s source of information, such as perception, report, or inference (Chafe and Nichols, 1986). Evidential markers found in many languages are the linguistic means to indicate these sources. According to Willett’s (1988) typology, a true evidential system is fundamentally concerned with the distinction between DIRECT (i.e., perceptual) and INDIRECT (i.e., inferential and reportative) evidence types. Systematic interactions exist crosslinguistically between evidentiality and tense, in that the present tense is interpreted as being based on direct evidence. To the extent that there is a close (causal) connection between one’s source of information and one’s knowledge, there is also a close relation between evidentiality and epistemic modality. For instance, epistemic modals like must are often the carriers of indirect (inferential) evidence (Palmer, 1986). Still, 24

See Postma and Rooryck (1996).

212

J. Lecarme

the problem of how exactly to capture the evidential meaning in the possible worlds semantics assumed for epistemic modality remains open. In earlier work, I argued that the study of systems where nominal expressions encode evidential distinctions is obviously relevant to our theoretical understanding of evidentiality, and the nature of the tense/evidential connection. Across languages, DIRECT (basically: visual) evidentiality appears to be the only kind of evidentiality found in nominals. In many (European) languages, the basic distinction visible/invisible is implicitly expressed by the deictic uses of the definite articles, or the system of spatially deictic demonstratives. In other languages, the visible/invisible distinction is superimposed on a proximal-distal system. In languages where determiners also encode temporal distinctions, visibility is associated with the present tense, non-visibility with the past tense.25 I proposed that the classical approaches to evidentiality can be extended to nominals, a domain paradoxically ignored in discussions of evidentiality. I want now to make this proposal more precise.

3.1 Visual Evidentiality In Somali, an evidential past must be used in all the contexts where the referent is not observable in principle, such as (20a). Here, past tense signals that the pen is not in the speaker’s visual field at the time of utterance. The non-past -gu is unacceptable, contradictory-sounding, hence ungrammatical in such examples. Intuitively, the utterance with -gu is incompatible with its ‘inner negation’: this is because the semantics of visual perception comes with the normal use of the present tense, as shown in (20b). (20)

Speaker-oriented visibility / invisibility a. Qálinkáygii/∗-gu méeyey? (—Khaanád-da pen.m.Poss1S-defM.past/∗ -defM.[+nom] Q.is.ms (—Drawer-defF buu kú jiraa.) C/F.3MS in stay.pres.gen) ‘Where is my pen?’ (—In the drawer.) b. Qálinkáy-gu wáa kee? (—Waa kán gudúudan.) pen.m.Poss1S-defM.[+nom] F/C m.Q (—C/F m.dem red) ‘Which one is my pen?’ (—The red one.)

In the context of use, the speaker actually knows that her pen is somehow ‘present’ (e.g., on or under her desk), but in using the -ii morphology, she also adds some crucial piece of information: that she lacks visual experience. In other words, the evidential past provides a negative information, which makes the utterance compatible with the speaker’s current evidence. In Somali as in other languages (e.g., the languages of the Salishan family), a past of evidentiality is characteristically used when talking about a hidden referent. This referent can be close in proximity, as in (21a). In this context, both the speaker 25

See Lecarme (2003) and references cited there. For a more detailed, crosslinguistic investigation of nominal evidentiality, see Imai (2003).


213

and the hearer know that the girl is present in the next room. (21a) is not the simple assertion ‘your daughter is pregnant’: by using an evidential past, the speaker marks that the girl is absent from the immediate visual context. As a further illustration, a speaker who asserts (21b) presupposes that the hearer knows the students. By using a past evidential, she also signals that the students are not in sight—as evidenced by the hearer’s question. The process is not intentional, and not optional either: -ii is unappropriate in both (21a) and (21b) if the girl or the students are visible in the contexts of use. (21)

Talking about invisible referents a. Inántaa-dii u´ ur bay leedahay girl-f.Poss2S-defF.past[+nom] pregnancy C/F.3FS has.3fs ‘(Doctor to Mother) Your daughter is pregnant’. b. Ardaý-dii way jóogayaan. —Kuwée? students-defF.past[+nom]t C/F.3P stay.prog (—defM.pl.Q) ‘(Secretary to Dean) The students are waiting. —Which students?’

Could the -ii morphology be analyzed as anaphorically dependent on a previous ‘discourse’ context where the referent of the DP was in sight? As discussed in Lecarme (2003), this is not a necessary condition: speakers can use an evidential past to refer to individuals, things or places that were not previously in sight nor in the common ground before the time of utterance. The fact that evidential past sometimes may be used to talk about the future suggests that we are dealing here with modal, not temporal, exclusion: (22)

Talking about the near future a. Bal w´ıil-k´ıi baan gáarayaa, nabadgélyo! Well boy-defM.past C/F.1S reach.prog.pres goodbye ‘Well, I am going to the boy, goodbye!’ b. Xáj-k´ıi baan a´ adayaa pilgrimage-defM.past C/F.1S go-to.pres.prog ‘I am going on the Pilgrimage’

Somali has no grammatical evidentials at the sentence level. The difference between perceptual, inferred and reported evidence is obtained compositionally via the evidential values of the nominal tenses—an analogous linguisitic means to disclose the speaker’s source of information. In (23a), both nominals are marked as past, in concord with the past tense marked on the verb. The interpretation conveyed is that of a reportative evidential: (23a) is not a felicitous utterance if the speaker is on the spot of the accident (the sentence out of context has no evidential interpretation). In (23b), both nominals are in the present-tense form. The evidential meaning conveyed is that of (positive) DIRECT evidence. For example, (23b) can be uttered felicitously on the spot of the accident, where both the truck and the road are in sight. On the other hand, (23c) illustates the inferential evidential meaning: (23c) is a possible utterance in a context in which the road is ‘present’, hence visible for the speaker. The interpretation that the rhino is ‘invisible’ or ‘absent’ (as glosses commonly indicate) is conveyed

214

J. Lecarme

by the evidential value of -ii. Thus, both (23a) and (23c) indicate INDIRECT evidence, but differ with respect of the source of evidence: in (23c), the evidence is based on observable results (for example, one might infer that the rhino has crossed the road from seeing traces on the ground). Here, the inference is constructed compositionally from the (positive) DIRECT evidential value of the present tense (i.e., the perceptual meaning). (23)

Evidential constructions a. Baabúur-kii j´ıd-k´ıi buu ká tallaabey truck-defM.past[+nom] road-defM.past C/F.3MS by pass.past ‘The truck crossed the road’ Indirect evidence: report b. Baabúur-ku j´ıd-ka buu ká tallaabey truck-defM[+nom] road-defM[−nom] C/F.3MS by pass.past ‘The truck (visible) crossed the road (visible)’ Direct evidence c. Wiy´ıi-shii j´ıd-ka bay ká tallaabtay rhinoceros-defF.past[+nom] road-defM C/F.3FS by pass.f.past ‘The rhinoceros (invisible) crossed the road (visible)’ Indirect evidence: inference

It appears, then, that past/non-past morphologies have uses that are clearly evidential. The question we must now ask is what ‘past’ and ‘invisible’ interpretations have in common. An answer I have offered in previous work is that they must share some feature: exclusion/dissociation. In this section, I will address more precisely the following questions: what are the factors determining the evidential reading of a past nominal? What gives rise to the visual character of nominal evidentiality? How can the visual meaning be incorporated into the semantics of evidentiality?

3.2 Approaches to (Sentential) Evidentiality There is crosslinguistic evidence that modals and evidentials occupy adjacent segments on the same hierarchy (Oswalt, 1986; Willett, 1988). From an extensive investigation of adverb placement in a number of languages, Cinque (1999) postulates a universal hierarchy of functional projections, in which evidentiality is included as a type of mood. (24)

Cinque’s (1999) functional hierarchy Speech Act MoodP > Evaluative MoodP > Evidential MoodP > Epistemic MoodP > Tense

According to Faller’s (2002) analysis, evidentiality is the linguistic encoding of the speaker’s grounds for making a speech act—in the case of assertion: the speaker’s source of information. Quechua evidential enclitics are illocutionary modifiers which add or modify the sincerity conditions of the act they apply to (for instance, the direct


215

evidential -mi adds a sincerity condition which requires that the speaker has the best possible grounds for believing the asserted proposition). In other words, the meaning of the whole expression is not based on truth-values. But as Faller points out, then we should have to distinguish between ‘pure evidentials’, an illocutionary phenomenon, and other means languages have to express the same distinctions.26 Another problem is that under this analysis, evidentiality is conceived of as part of the theory of action, not knowledge. There is, then, little hope that we will find principles of general interest about how linguistic knowledge is made compatible with knowledge involved in vision, and how linguistic and non-linguistic (e.g., visual) representations interact in the construction of evidential meanings. The modal approach which uses a possible world semantics in the analysis of epistemic modality seems more promising. From a semantic point of view, evidential markers do not contribute to the informative proposition, but disclose the source of information. In Izvorski’s terms,

‘the two dimensions. . . are logically independent. Yet, natural language typically treats propositions based on perceptual evidence on the part of the speaker as propositions asserted by the speaker to be true.’ (1997, 3) Kratzer’s theory of ‘doubly relative’ modality is a profound insight into the nature of evidentiality. As in standard analysis in formal semantics, her theory posits an abstract common logical structure unifying the epistemic and root uses of modals. The lexical semantics of the modal encodes a quantificational force ( or ♦, a relation between sets of worlds). A covert variable next to the modal picks up a contextually salient set of worlds as a restrictor, and this contextually salient set determines how the modal is interpreted. On this view, epistemic accessibility is defined as compatibility relative to a knowledge state: (25) will be true in a world w if and only if it follows from ‘what is known in w’ that John is at home. Speakers do not claim that p (= ‘John is at home’) is true, but that a certain logical relation holds between p and what they know. (25)

(in view of what is known) John must be at home

In Kratzer’s theory, the quantificational part of a modal (the modal relation) is defined along two ‘conversational backgrounds’: the modal base and the ordering source, both functions from worlds to propositions.27 An epistemic modal base (‘what we know’) 26 ‘The framework of speech act theory might also prove to be the right one in analyzing evidentials in other languages, although not necessarily evidentiality in general. It is a reasonable hypothesis that evidentiality that is encoded by markers of tense and modality can more fruitfully be analyzed within a framework such as possible world semantics, which was developed to account for these categories’ (Faller, 2002, 264). 27 Kratzer (1981, 1991) does not explain how a ‘conversational background’ is formally selected from the context of the utterance, but she does suggest that the distinction between modals with epistemic and circumstantial modal bases may correlate with a difference in argument structure. As many have previously noted, there are good arguments that modals with a circumstantial modal base (i.e., root modals) have an implicit agent argument which can act as a syntactic controller, whereas epistemic modals are structurally speaker oriented.

216

J. Lecarme

determines the set of worlds which are epistemically accessible from w. The innovation in Kratzer’s theory is the ordering source, an additional construct for ordering worlds with respect to a reference world. The ordering source (w is at least as close to the ideal as w’) imposes an order on the modal base and let the quantification range only over the closest worlds in the modal base. Thus, if the ordering source for the modal in (25) is the set of propositions which represent ‘the normal course of events’, the ‘most normal world’ may be a world in which John is not at home. This view has crucial consequences on our undertanding of evidentiality. Epistemic modal bases take ordering sources related to information. As Kratzer suggests, in uttering (25), speakers convey that they do not rely on known facts alone. They use other sources of information which are more or less reliable. These other sources potentially include extralinguistic information. This opens the possibility that evidential meanings depend on the kind of ordering source supplied by the context. Izvorski (1997) proposed a formal analysis of indirect evidentiality in this framework.28 Her arguments are based on the analysis of the ‘perfect of evidentiality’ in Bulgarian, Turkish and other languages. As (26a) shows, a present perfect morphology in Bulgarian is interpreted not as a present perfect, but as a perfect of evidentiality (glossed as PE). The indirect evidential reading is linked to the present tense: this reading is unavailable in non-finite environments, as well as in the formation of the past and future perfects (26b). (26)

a.

b.

Ivan izpil vsiˇckoto vino vˇcera Ivan drunk.PE all.the wine yesterday Ivan apparently drank all the wine yesterday (evidential reading) Ivan trjabvada e izpil vsiˇckoto vino vˇcera Ivan must is drunk all.the wine yesterday Ivan must have drunk all the wine yesterday (perfect reading)

According to Izvorski’s analysis, indirect evidentials (PE or apparently) are semantically propositional operators. Sentences of the form EV p are represented in (27): the EV -operator is an universal epistemic modal with a presupposition of available indirect evidence for the truth of the proposition it modifies. This presupposition restricts the domain of quantification of EV, for a given world, to the set of propositions which constitute the available indirect evidence in that world (e.g., ‘There are empty bottles in Ivan’s office’ or ‘Mary said that p’). The domain of quantification is further restricted by a stereotypical ordering source (e.g., ‘If there are empty bottles in someone’s office, that person has drunk the wine’ or ‘Normally, Mary is a reliable source of information’). (27)

a. b.

Assertion. p in view of the speaker’s knowledge state Presupposition: Speaker has indirect evidence for p

In addition, Izvorski is able to derive the presupposition of available indirect evidence from the exclusion/dissociation formalism, thereby defining the formal link between the perfect of evidentiality and the present perfect. For the Present perfect, 28

See also Garrett’s (2002) analysis of the Tibetan indirect evidential.


217

the compositional meaning is that the consequent state of a past eventuality holds at the time of utterance. For the perfect of evidentiality, the compositional meaning is that the worlds in which p is known are epistemically inaccessible from ws , le world of the speaker; the speaker has therefore no direct evidence for p (the modal contribution of the perfect aspect morphology). The accessible worlds are those in which p’ (= ‘there are consequences or results of p’) is true (the modal contribution of the present tense). Together with a stereotypical ordering source, this compositional meaning derives the inference and report interpretations. On this view, evidentiality is a kind of epistemic modality: the presupposition of indirect evidence contributes a proposition P to the modal base, which serves to restrict the domain of the evidential operator.29 In Izvorski’s terms,

‘. . . it is more useful to think of the terms DIRECT and INDIRECT . . . as making a distinction based not on whether or not the evidence is perceptual, but on whether or not the evidence justifies the speaker’s belief in a proposition’ (1997, 3–4). Still, there is a nontrivial, linguistic difference between classical inference on the one hand, and the evidentially modalized counterpart of it. Whether or not the evidence is perceptual does matter: when a speaker utters an evidential statement, the grounds upon which her knowledge is claimed to rest is something more restricted and more specific than the proposition ‘there are consequences or results of p’: crucially, some internal structure corresponds to DIRECT, perceptual experience. Even if the inference is not valid (i.e., if Ivan has not drunk the wine), the particular epistemic state that the speaker is in still includes some perceptual knowledge. Assuming that evidentiality is part of the presuppositional content does not capture that fact. Here, I argue that the treatment of direct evidentiality in possible world semantics is indeed possible, provided that we reinterpret Kratzer’s doubly relative semantics. Informally, we might assume an ordering source that singles out the accessible worlds which come closest to an ideal in which the speaker has direct (visual) evidence: a perceptual ordering source, which relates informal access (e.g. seeing) to current knowledge.

3.3 Extending Kratzer’s Theory There are two necessary factors in the computation of the (either sentential or nominal) evidential meanings. First, evidential expressions must be anchored to the utterance time. At any time, the evidence available to speakers is made compatible with a set of worlds which, for all they know, is the actual world. Deictic anchoring to the present tense is the crucial factor: as discussed earlier (section 2.2), the evidential interpretation of past DPs is unavailable in generic and generalizing sentences, 29

This analysis is similar to Kratzer’s treatment of (modal) conditionals: for each world, the if clause is added to the set of propositions the modal base assigns to that world, thus restricting the accessibility relation.

218

J. Lecarme

where the present tense temporally locates the generic predicate, not the core eventuality. Second, evidentiality has a perceptual component. When uttering an evidential statement, a speaker also claims that her knowledge rests on perceptual grounds. This kind of inference crucially differs from inference from reasoning in that it derives from experience, not a mental construct. Nominals open up a refreshing new way of looking at the phenomenon. As noted earlier, the kind of evidentiality found in nominals is typically DIRECT. At the same time, we can observe a clear correlation between tense and direct (i.e., visual) evidentiality, which is generally not found at the sentence level. Let us consider again the Somali examples where the invisible referent is actually present at the time of utterance. In the context of (28a = 21a), both interlocutors actually ‘know’ that the girl is present in the next room. The referent of the DP is epistemically ‘present’, but evidentially ‘past’, that is, invisible to the speaker (compare with (28b), which is a felicitous utterance if and only if the referent is in sight):30 (28)

a.

b.

Inántáa-dii u´ ur bay leedahay girl.f.Poss2S-defF.past.[+nom] pregnancy C/F.3FS has.3fs (Doctor to Mother):—Your daughter (invisible) is pregnant. Ínan-ka yar u´ yeer! boy-defM small to call ‘Call the little boy! (visible)

The nominal past here can only be understood as involving a shift in the epistemic context of the speaker. Intuitively, exclusion/dissociation operates in a higher layer of structure in terms of Cinque’s 1999 functional hierarchy (24). Past does not exclude the referent of the DP from ‘what is known’ by the speaker, but from a more restricted and more specific set, including the domain of individuals perceived by the speaker/viewer at a given moment. My proposal is therefore, in the spirit of Kratzer, to make explicit the ordering source that the evidential past is sensitive to, and assume that evidential epistemic modality is defined as in (29). (29)

Evidential modality Modal relation: Modal base: epistemic Ordering source: perceptual (in the most ideal or ‘normal’ of her worlds, the speaker has DIRECT (visual) evidence.

The modal base is the set of worlds epistemically accessible from the world of the speaker at the time of utterance. Universal quantification takes place over the epistemically accessible worlds that remain after the ordering source has applied to the set of worlds determined by the modal base. As a result, the accessibility relation supplies a ‘totally realistic’ modal base, that is, the set of propositions that exhaustively describe the perceptual world of the speaker at the time of utterance (e.g., ‘this is 30 See von Tiling’s 1919 study (based on the Isaaq dialect of Somali) for similar examples and contexts.


219

my pen’, ‘my pen is red’, ‘there are empty wine bottles here’. . . ). On this account, the ‘default’ modal force characteristic of evidential statements derives from perception. Evidentiality differs from ‘subjective epistemic modality’ described above (section 2.3) in that it is a subjective neccessity: if our perceptual system is reliable and works properly, what we perceive directly must be true. Our perception of objects and events around us is veridical, a true representation of what those objects are and where they are situated in space. The analysis correctly predicts that evidential statements, like other modal statements, are contingent: our perceptions can be mistaken. But the perception-based knowledge relevant for evidentiality (a linguistic notion) are facts of our world, not necessary truths. Now, why is direct evidentiality visual? It cannot be the case that the tenses ‘encode’ the visual meaning. The visual meaning cannot plausibly be derived from a conversational implicature either, since those implicatures are typically cancellable. The most natural answer is that the hierarchical structure of evidential systems linguistically reflects the epistemic importance of visual perception in human knowledge. This is because visual perception contains an epistemic component (Dretske, 1969, 1990) by which non-linguistic and linguistic representations are mutually compatible.31 From the speaker/viewer’s perspective, the ‘visual’ meaning can be derived as follows. (i) What is brought up by the context is not a proposition (because propositions cannot be perceived), but what can be thought of as the complement of a perception verb: semantically, an object or an event; syntactically, a DP or a bare infinitive— expressions like ‘my pen’ ‘bottles in the office’, ‘John drink the wine’, that is, the uppermost level of perceptual processing (i.e., Dretske’s 1969 ‘non-epistemic seeing’). (ii) The meaning of DIRECT perception does not come from a perception verb (as in perceptual reports)—or from an abstract perceptual predicate or operator.32 The meaning of DIRECT perception comes from the present time, ‘now’. We can think of the deictic present tense as some functional equivalent of demonstrative reference, an indexing mechanism that relates informational access to current knowledge. In evidential contexts, deictic present tense indicates DIRECTness.33 (iii) Anchoring to the utterance time is responsible for matching linguistic with nonlinguistic representations, yielding perceptual knowledge (propositions like ‘the pen is red’, ‘this is my pen’, ‘there are bottles in Ivan’s office. . . ’) (i.e., Dretske’s 1969 ‘primary epistemic seeing’). 31 Beyond linguistics (Jackendoff, 1983), there is an enormous literature concerning the relation between knowledge and (visual) perception in the related fields of philosophy of perception (Dretske, 1969, 1990) and neuropsychology (Milner and Goodale, 1995). 32 On perceptual reports, see Higginbotham (1983a); on the relation between perceptual reports and evidentiality, see Van der Does and Lambalgen (2000). 33 This is a feature of many languages with grammatical evidentials. According to Faller (2002, 123), the Qechua evidential mi encodes direct evidence. The same evidential value is implicated by simple assertions. From a typological perspective (e.g. De Haan (to appear)), many languages only have overt evidentials for indirect evidence types, and in those cases, zero is interpreted as direct evidence. In languages that make the distinction between visual and nonvisual direct evidence, evidence by seeing is the default evidential.

220

J. Lecarme

(iv) Evidential inference crucially rests on perceptual knowledge (i.e., representations in (iii): we come to know that the rhino has crossed the road by seeing traces on the ground, we come to know that Ivan drank the wine by seeing empty bottles in his office, etc. This corresponds to Dretske’s 1969 ‘secondary epistemic seeing’, and crucially differs from inference from reasoning. On this view, we can think of true evidentials, which come with a lexically specified accessibility relation (like attitude verbs), as lexicalizing the ordering source. Assuming a partial ranking of all possible sources gives clues to access the direct (perceptual) meanings and derive the basic distinctions encoded by the evidential systems. The hierarchy of evidential types (DIRECT, INDIRECT INFERENCE, INDIRECT REPORTATIVE ) might be thought of in binary terms: (visually) perceived, (visually) perceived resultant state, lack of personal experience (or non-speaker’s evidence). This has only been an outline of how to spell out an analysis of DIRECT evidentiality in a classical, truth-conditional semantics. I hope this beginning has looked promising enough to motivate interest in further investigation of evidentiality in nominals, and what it reveals about the other systems with which language interacts and interfaces.

4 Conclusions What has been the leading thread troughout this paper is the common abstract meaning underlying the different uses of nominal past, namely exclusion/dissociation. The interaction of modality and temporality has a natural parallel in the nominal domain: past as referential displacement (distinct from the notion of precedence, or chronological order) makes it possible to talk about possible worlds that stand in a certain relation to the actual world. If Kratzer’s theory can be supplemented with a perceptual component, as I proposed, past gives rise to the ‘non-actual’, ‘unknown’, ‘invisible’ modal meanings, depending on different choices of modal bases and ordering sources. From our study of the Somali determiner system, it appears that demonstratives and evidentials encode two different aspects of visual cognition: past versus nonpast is linked to visual perception, visual and episodic memories, knowledge, and reasoning about abstract entities that cannot be perceived. This is consistent with the distinction between two separate visual systems—one for conscious perception and another for the control of action.34 We hope in future work to investigate why these particular components of meaning, ‘past’ and ‘invisible’, correlate in precisely this way, and what this reveals more generally about the interfaces between linguistic meanings and the visual system. If language can be shown to make use of a more primitive, domaingeneral exclusion/dissociation feature, it might be the case that the evidential values of past morphemes could be taken as their core meaning.35 If so, our experience of 34

See Goodale and Milner (1992), Milner and Goodale (1995). I refer here to the cognitive mechanisms underlying temporality and evidentiality, not the historical evolution of grammatical forms. Diachronically, evidential meanings are often acquired by modal, tense and aspect morphemes (Bybee et al., 1994). Among languages with (sentential)

35


221

time may ultimately derive from perceptual mechanisms and processes, which in turn enable us to perceive objects and events.

References Anderson, Stephen, Lea Brown, Alice Gaby, and Jacqueline Lecarme. 2006. Life on the edge: There’s morphology there after all! In Lingue e Linguaggio, volume 1, pages 1–16. Bach, Emmon. 1981. On time, tense, and aspect: An essay in English metaphysics. In Peter Cole (ed.), Radical Pragmatics, pages 62–81. New York: Academic. Bhatt, Rajesh and Roumyana Pancheva. 2006. Conditionals. In Martin Everaert and Henk van Riemsdijk (eds.), The Blackwell Companion to Syntax, volume 1, pages 638–687. Oxford: Blackwell. Burton, Strang. 1997. Past tense on nouns as death, destruction, and loss. In Kiyomi Kusumoto (ed.), NELS, volume 27. Amherst, MA: University of Massachusetts, GLSA. Bybee, Joan, Revere Perkins, and William Pagliuca. 1994. The Evolution of Grammar: Tense, aspect and modality in the languages of the world. Chicago, IL: University of Chicago Press. Carlson, Gregory. 1995. Truth conditions and generic sentences: Two contrasting views. In Gregory Carlson and Francis Jeffry Pelletier (eds.), The Generic Book, pages 224–237. Chicago, IL: University of Chicago Press. Chafe, Wallace and Johanna Nichols (eds.). 1986. Evidentiality: The Linguistic Coding of Epistemology. Norwood, NJ: Ablex Publishing Corporation. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Roger Martin, David Michaels, and Juan Uriagereka (eds.), Step by Step: Essays on minimalist syntax in honor of Howard Lasnik, pages 89–156. Cambridge, MA: MIT Press. Chomsky, Noam. 2001. Derivation by phase. In Kenstowicz (ed.), Ken Hale: A Life in Language, pages 1–52. Cambridge, MA: MIT Press. Chomsky, Noam. 2008. On phases. In Robert Freidin, Carlos Otero, and Maria-Luisa Zubizarreta (eds.), Foundational Issues in Linguistic Theory. Cambridge, MA: MIT Press. Cinque, Guglielmo. 1999. Adverbs and functional heads: A cross-linguistic perspective. Oxford: Oxford University Press. Comrie, Bernard. 1985. Tense. Cambridge: Cambridge University Press. Dayal, Veneeta. 1997. Free relatives and ever: Identity and free choice readings. In Aaron Lawson and Eun Cho (eds.), SALT VII Proceedings, pages 99–116. Ithaca, NY: CLC Publications, Cornell University. De Haan, Ferdinand. 1998. The category of evidentiality. Manuscript, University of New Mexico. De Haan, Ferdinand. to appear. Visual evidentiality and its origins. Diachronica. evidentials, visual evidentials typically derive from already grammaticalized tense or aspect morphemes, or demonstratives (see De Haan, 1998, to appear).

222

J. Lecarme

Diesing, Molly. 1988. Bare plural subjects and the stage/individual contrast. In Manfred Krifka (ed.), Genericity in Natural Language: Proceedings of the 1988 Tübingen Conference, pages 88–42. Tübingen: Seminar für natürlich-sprachliche Systeme. Diesing, Molly. 1992. Indefinites. Cambridge, MA: MIT Press. Dretske, Fred I. 1969. Seeing and Knowing. Chicago, IL: University of Chicago Press. Dretske, Fred I. 1990. Seeing, believing and knowing. In Daniel N. Osherson, Stephen M. Kosslyn, and John M. Hollerbach (eds.), Visual Cognition and Action: An Invitation to Cognitive Science, volume 2, pages 129–148. Cambridge, MA: MIT Press. Embick, David and Rolf Noyer. 2001. Movement operations after syntax. Linguistic Inquiry, 32(4):555–595. Enç , Mürvet. 1981. Tense without Scope: An Analysis of Nouns as Indexicals. Ph.D. thesis, University of Wisconsin, Madison, WI. Enç , Mürvet. 1986. Towards a referential analysis of temporal expressions. Linguistics and Philosophy, 9:405–426. Enç , Mürvet. 1991. The semantics of specificity. Linguistic Inquiry, 22(1):1–25. Faller, Martina. 2002. Semantics and Pragmatics of Evidentials in Cuzco Quechua. Ph.D. thesis, Stanford University, Stanford. von Fintel, Kay. 2000. Whatever. In Brenda Jackson and Tanya Matthews (eds.), Proceedings from the Conference on Semantics and Linguistics (SALT, volume 10. CLC Publications. Garrett, Edward. 2002. Evidentiality and Assertion in Tibetan. Ph.D. thesis, UCLA. Goodale, Melvyn and David Milner. 1992. Separate visual pathways for perception and action. Trends in Neuroscience, 15:20–25. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. thesis, University of Massachusetts, Amherst, MA. Heim, Irene and Angelika Kratzer. 1998. Semantics in Generative Grammar. Oxford: Basil Blackwell. Higginbotham, James. 1983a. The logic of perceptual reports: An extensional alternative to situation semantics. Journal of Philosophy, LXXX:100–127. Higginbotham, James. 1983b. Logical form, binding, and nominals. Linguistic Inquiry, 14:395–420. Higginbotham, James. 2008. Evidentials: Some Preliminary Distinctions. In Robert Stainton and Christopher Viger (eds.), Compositionality, Context, and Semantic Values. Berlin: Springer Verlag. Hiraiwa, Ken. 2005. Dimensions of Symmetry in Syntax: Agreement and Clausal Architecture. Ph.D. thesis, MIT, Cambridge, MA. Iatridou, Sabine. 2000. The grammatical ingredients of counterfactuality. Linguistic Inquiry, 31.2:231–271. Iatridou, Sabine, Elena Anagnostopoulou, and Roumyana Izvorski. 2001. Some observations about the form and meaning of the perfect. In Michael Kenstowicz (ed.), Ken Hale: A Life in Language, pages 189–238. Cambridge, MA: MIT Press. Iatridou, Sabine and Spyridoula Varlokosta. 1997. Pseudoclefts crosslinguistically. Natural Language Semantics, 6(1):3–28.


223

Imai, Shingo. 2003. Spatial Deixis. Ph.D. thesis, State University of New York at Buffalo. Izvorski, Roumyana. 1997. The present perfect as an epistemic modal. In Aaron Lawson and Eun Cho (eds.), SALT VII Proceedings, pages 99–116. Ithaca, NY: CLC Publications. Jackendoff, Ray. 1983. Semantics and Cognition. Cambridge, MA: MIT Press. Kirk, John W. C. 1905. A Grammar of the Somali Language. Cambridge: Cambridge University Press. Klein, Wolfgang. 1994. Time in Language. London: Routledge. Kratzer, Angelika. 1981. The notional category of modality. In Hans-Jürgen Eikmeyer and Hannes Rieser (eds.), Worlds, words, and contexts, pages 38–74. Berlin: de Gruyter. Kratzer, Angelika. 1989. An investigation of the lumps of thought. Linguistic and Philosophy, 12:607–653. Kratzer, Angelika. 1991. Modality. In Arnim von Stechow and Dietrich Wunderlich (eds.), Semantics: An International Handbook of Contemporary Research, pages 639–650. Berlin: de Gruyter. Kratzer, Angelika. 1995. Stage-level and individual-level predicates. In Gregory N. Carlson and Francis Jeffry Pelletier (eds.), The Generic Book. Chicago, IL: University of Chicago Press. Krifka, Manfred. 1992. Thematic relations as links between nominal reference and temporal constitution. In Ivan Sag and Anna Szabolsci (eds.), Lexical Matters. Stanford, CA: CLSI Publications. Larson, Richard K. 1987. Missing prepositions and the analysis of English free relative clauses. Linguistic Inquiry, 16:239–266. Larson, Richard K. and Sungeun Cho. 1999. Temporal adjectives and the structure of possessive DPs. In Sonya Bird, Andrew Carnie, Jason D. Haugen, and Peter Norquest (eds.), WCCFL, volume 18, pages 299–311. Somerville, MA: Cascadilla Press. Larson, Richard K. and Sungeun Cho. 2003. Temporal adjectives and the structure of possessive DPs. Natural Language Semantics, 11:217–247. Lecarme, Jacqueline. 1996. Tense in the nominal system: the somali DP. In Jacqueline Lecarme, Jean Lowenstamm, and Ur Shlonsky (eds.), Studies in Afroasiatic Grammar. The Hague: Holland Academic Graphic. Papers from the 2nd Conference on Afroasiatic Languages, Sophia Antipolis, 1994. Lecarme, Jacqueline. 1999a. Focus in somali. In Georges Rebuschi and Laurie Tuller (eds.), The Grammar of Focus, pages 275–309. Amsterdam: John Benjamins. Lecarme, Jacqueline. 1999b. Nominal tense and tense theory. In Francis Corblin, Carmen Dobrovie-Sorin, and Jean-Marie Marandin (eds.), Empirical Issues in Formal Syntax and Semantics, volume 2, pages 332–354. The Hague: Holland Academic Graphics. Selected Papers from the Colloque de Syntaxe et Sémantique a` Paris (CSSP). Lecarme, Jacqueline. 2002. Gender ‘polarity’: Theoretical aspects of somali nominal morphology. In Paul Boucher (ed.), Many Morphologies, pages 109–141. Somerville, MA: Cascadilla Press.

224

J. Lecarme

Lecarme, Jacqueline. 2003. Nominal tense and evidentiality. In Jacqueline Guéron and Liliane Tasmowski (eds.), Tense and Point of View. Presses de l’Université Paris X-Nanterre. Lecarme, Jacqueline. 2004. Tense in nominals. In Jacqueline Guéron and Jacqueline Lecarme (eds.), The Syntax of Time, pages 441–475. Cambridge, MA: MIT Press. Lewis, David. 1975. Adverbs of quantification. In Edward Keenan (ed.), Formal Semantics in Natural Language, pages 3–15. Cambridge: Cambridge University Press. Lyons, John. 1977. Semantics, volume 1 and 2. Cambridge: Cambridge University Press. Milner, David and Melvyn A. Goodale. 1995. The Visual Brain in Action, volume 27 of Oxford Psychology Series. Oxford: Oxford University Press. Musan, Renate. 1995. On the temporal interpretation of noun phrases. Ph.D. thesis, MIT, Cambridge, MA. Musan, Renate. 1997. Tense, predicates, and lifetime effects. Natural Language Semantic, 5:271–301. Nordlinger, Rachel and Louisa Sadler. 2004. Nominal tense in a crosslinguistic perspective. Language, 80(4):776–806. Oswalt, Robert L. 1986. The evidential system of Kashaya. In Wallace Chafe and Johanna Nichols (eds.), Evidentiality: The Linguistic Coding of Epistemology, volume 20 of Advances in Discourse Processes, pages 29–45. Norwood, NJ: Ablex Publishing Corporation. Palmer, Franck R. 1986. Mood and Modality. Cambridge: Cambridge University Press. Papafragou, Anna. 2006. Epistemic modality and truth conditions. Lingua, 116:1688– 1702. Partee, Barbara. 1983. Genitives—a case study. published as an appendix to Theo Jansen, Compositionality. In Johan van Benthem and Alice ter Meulen (eds.), Handbook of Logic and Language, pages 464–470. Cambridge, MA: MIT Press. Also published by North Holland, Amsterdam. Partee, Barbara. 2000. Some remarks on linguistic uses of the notion ‘event’. In Carol Tenny and James Pustejowsky (eds.), Events as Grammatical Objects, pages 483– 495. Stanford: CLSI Publications. Pesetsky, David and Esther Torrego. 2001. T-to-C: Causes and consequences. In Michael Kenstowicz (ed.), Ken Hale: A Life in Language, pages 355–426. Cambridge, MA: MIT Press. Postma, Gertjan and Johann Rooryck. 1996. Modality and Possession in NPs. In Kiyomi Kusumoto (ed.), Proceedings of NELS, volume 26, pages 87–100. Cambridge, MA: Harvard University and MIT. Roehrs, Dorian. 2006. The morpho-syntax of the Germanic Noun Phrase: Determiners move into the Determiner Phrase. Ph.D. thesis, Indiana University, Bloomington, IN. Schlenker, Philippe. 2004. Conditionals as definite descriptions (a referential analysis). In Ruth Kempson and Klaus von Heusinger (eds.), Research on Language and Computation, volume 2, pages 417–462. Special issue on ‘Choice Functions in Semantics’.


225

von Stechow, Arnim. 2002. Temporal prepositional phrases with quantifiers—some additions to Pratt and Francez (2001). Linguistics and Philosophy, 25(5–6):755– 800. von Tiling, Rainer Maria. 1919. Die Vokale des bestimmten Artikels im Somali. Zeitschrift für Kolonialsprachen, 9(3):132–166. Tonhauser, Judith. 2006. The temporal semantics of noun phrases: Evidence from Guaran´ı. Ph.D. thesis, Stanford University, Stanford, CA. Van der Does, Jaap and Michiel Van Lambalgen. 2000. A logic of vision. Linguistics and Philosophy, 23(1):1–92. Willett, Thomas. 1988. A cross-linguistics survey of the grammaticalization of evidentiality. Studies in Language, 12:51–97.

Time With and Without Tense Carlota S. Smith

Abstract How is temporal information conveyed in language? Do tenses code temporal information directly? Does the universality of temporal interpretations arise from a common syntactic structure necessarily including a Tense Phrase (TP)? From the study of tenseless languages and mixed-temporal languages such as Mandarin Chinese and Navajo, this article addresses fundamental questions about the grammaticalization of time. It is proposed that two simple pragmatic principles constrain direct temporal interpretation in languages with tense (English, French), and guide indirect temporal interpretation in languages without tense: (i) the default intepretation of present tense sentences as located in the present, (ii) the ‘Bounded Event Constraint’, i.e. the fact that bounded events cannot be located at Speech Time. A more general principle of Simplicity ensures that Present takes precedence over the Future (as futurity is never a ‘purely temporal’ concept). According to this view, a few very general grammatical principles account for temporal interpretation in both tensed and tenseless languages. The syntax of fully-tensed languages includes a TP, conveying direct information about temporal location. The other types of languages have an Aspect Phrase, but no TP. Aspectual information about boundedness and information about internal temporal properties (Static/Dynamic, Telic/Atelic, Durative/Punctual), coupled with the invariant pragmatic principles, suffice to derive temporal interpretation.

Key words: Time, tense, aspect, syntactic structures, situations, Mandarin Chinese, Navajo. How is temporal information conveyed in language? In languages with tense it is direct; without tense, inference allows the receiver to arrive at an indirect temporal interpretation. I will discuss tensed and tenseless languages, proposing a unified

The University of Texas at Austin


227

228

C.S. Smith

approach that applies to both. I show that a few very general pragmatic principles account for temporal interpretation, direct and indirect.1 I assume that understanding a sentence requires that the receiver locate an event or state, spatially and temporally: time is one of the basic coordinates for truth conditional assessment. Sentences in all languages convey information that allows us to determine the temporal location of the situation expressed. One would like to understand how this happens. The pragmatic principles that I suggest constrain direct temporal interpretation and guide indirect. In languages with tense, tense gives direct temporal information; however certain apparent possibilities do not arise, due to the pragmatic constraints. In languages without tense, inference allows temporal interpretation. The key point in such languages is that aspectual information guides temporal interpretation. Temporal adverbs are always optional; I am mainly concerned with temporal inference that occurs without them. There are striking commonalties in the semantics of tensed and tenseless languages. I’ll show that both tensed and tenseless languages require the notion of Reference Time, roughly as proposed by Reichenbach (1947), in addition to Speech Time and Event Time. In section 1 of this article I introduce the background assumptions on which I rely, and the general ideas which inform the analysis. I propose three pragmatic principles and a classification of languages according to the way they express temporal information. In sections 2–4 I discuss temporal interpretation in English, a tensed language, and in several other languages including Mandarin Chinese and Navajo; section 5 gives a formal sketch of the account in the framework of Discourse Representation Theory; section 6 concludes.

1 Background and Principles 1.1 Background I sketch very briefly the approach to temporal location and aspect that I will assume. Temporal location: I take an approach based on Reichenbach (1947). Locating a situation in time linguistically involves three times and two relations. The times are Speech Time, the moment of speech; Situation Time, the time at which an event or state occurs or holds2 ; and Reference Time, the temporal standpoint or perspective from which a situation is presented. Speech Time is related to Reference Time, and Reference Time to Situation Time, by the relations of simultaneity and sequence. Aspect: Aspectual systems have two components, I assume, situation type and viewpoint (Smith, 1991/1997). The two components interact in the sentences of 1 I would like to thank the audience at the Colloque for interesting questions and discussion. In this article I take the position that a language is tenseless if it does not have overt tense morphemes. * The editors thank Richard Meier and Emilie Destruel for kindly proofreading Carlota Smith’s article, and Marie-Claude Paris for checking the Chinese examples for errors of transcription. 2 Reichenbach’s term was ‘Event Time’.

Time With and Without Tense

229

a language. This discussion focuses on aspectual information about boundedness, which as we will see allows temporal inference. The notion of situation type is based on the categories proposed in Vendler (1967). Situation type indirectly classifies a clause as expressing a situation with certain internal temporal properties. There are three temporal features: Static-Dynamic, TelicAtelic, Durative-Punctual. These features cluster in the situation type categories State, Activity, Accomplishment, Achievement, Semelfactive.3 The labels are shorthand for clusters of features. The verb and its arguments convey situation type, together with adverbs. A clause is associated by rule with a temporal schema that gives its defining properties. Aspectual information is conveyed in all languages, so far as I know. The term ‘situation’ includes states and events; other terms are ‘event structure’ and ‘eventuality’ (Bach, 1986). Telic events and punctual events are intrinsically bounded. The very notion of these events involves a set terminal point: telic events have a change of state or final natural endpoint and the single stage constitutes a bound for punctual events. The property of intrinsic bounding cuts across the situation type categories. Aspectual viewpoints make visible for semantic interpretation all or part of a situation. In the two-component theory all clauses have an aspectual viewpoint. Viewpoint is usually expressed by a morpheme associated with the verb. Perfective viewpoints make events visible as bounded, including endpoints. Bounds may also be stated independently with adverbial or other information (Depraetere, 1995). Imperfective viewpoints make situations visible without information as to endpoints, unbounded. Neutral viewpoints are flexible, giving enough information to allow a bounded or an unbounded interpretation (Smith, 1991/1997). The neutral viewpoint appears in clauses without an overt viewpoint morpheme; I will call such clauses ‘zero-marked’, following Klein et al. (2000). Boundedness information may thus be conveyed by viewpoint and/or situation type. The boundedness of a situation determines the nature of its location at Situation Time. Bounded situations occur within the Situation Time interval: for instance, in Mary walked to school the event of walking occurs within the past interval talked about. Unbounded situations – ongoing events and states – overlap or surround the Situation Time interval, for instance, in Mary was walking to school the event extends beyond the interval talked about in the sentence. (1)

Bounded events (E) are included in the SitT interval: SitT ⊆E a. Leigh built a sandcastle. John left. Unbounded events and states (S) overlap the SitT interval: E/S O SitT b. John was working. Leigh was at school.

3 I have added the Semelfactive situation type to those originally proposed by Vendler (Smith, 1991/7). The Semelfactive is implicit in his discussion of certain Achievements, single-stage, punctual events that do not involve a change of state. The cluster of temporal properties that characterize the situation types are as follows: States are Static and Durative; Events are Dynamic. Activities are Atelic, Durative; Semelfactives are Atelic, Punctual; Accomplishments are Telic, Durative; Achievements are Telic, Punctual.

230

C.S. Smith

1.2 Pragmatic Principles for Temporal Interpretation I now turn to the pragmatic principles that underlie this account of temporal interpretation. They are familiar, though are not always invoked for this purpose. The first two, the Deictic Principle and the Bounded Event Constraint, are linguistic in nature; the third, the Simplicity Principle of Interpretation is quite general, applying to many different kinds of information. Time is a single unbounded dimension that stretches indefinitely into the past and future. To locate events and states in time we need an orientation point. Prototypical linguistic communication provides it: the speaker is the center of linguistic communication, and Speech Time is the default orientation point. This is the Deictic Principle, basic to linguistic communication. Following the deictic principle we take Speech Time to be the Present, and locate other times with reference to Speech Time. The Past precedes, the Future follows. The pattern discussed here appears in discourse of many types. There are two other patterns: situations may be related to each other or to a previous time, as in narrative and description (Smith, 2003). The Deictic Principle locates situations with respect to Speech Time. The principle allows any type of situation to be located in the Past, Present, or Future (I use initial capitals for times, lower case for tenses). We do not locate all situations freely, however. There is a well-known constraint that involves the aspectual notion of boundedness. Bounded situations are discrete entities: they are closed, with an initial and final endpoint, or they are punctual. They cannot be neutrally located at Speech Time.4 Situations in the Present must be open and unbounded, without endpoints: they include ongoing events (John is talking, Mary is drawing a circle); particular states (Agnes is excited); and general states (Louis often feeds the cat). The explanation for the constraint is at once pragmatic and semantic. In taking the temporal perspective of the Present, speakers obey the tacit convention that communication is instantaneous, occurring at an idealized moment of speech. A bounded event in its entirety is incompatible with a report of a bounded event, because the bounds would go beyond the present moment (Kamp and Reyle 1993: 536–537). Similar ideas are expressed in Lyons (1977), Giorgi and Pianesi (1997). (This convention concerns communication and does not disturb the general notion of the Present as an interval). Thus the second principle excludes the possibility that a bounded situation can be located at Speech Time. I call it the Bounded Event Constraint. Boundedness plays an important role in guiding inference about temporal location, a point developed below. The Bounded Event Constraint holds for languages generally, so far as we know; it is realized differently according to the resources of a given language. The Deictic Principle and the Bounded Event Constraint are both peculiar to linguistic communication. The third principle that I will draw on is a general simplicity principle of interpretation. It concerns the need for enriching limited information that is provided to a receiver – of language, and other types of information as well. People often 4 There are exceptions, e.g. performative utterances; and apparent exceptions, such as stage directions. For discussion see Smith (2003).


231

utter sentences that underdetermine an interpretation, saying the minimum that is necessary. The receiver enriches the interpretation, filling it out with additional information (cf Grice’s second Maxim of Quantity (1975); the Informativeness Principle of Levinson (1983), the R-principle of Horn (1984)). But such enrichment proceeds sparingly. By a very general principle of information-processing, when people encounter incomplete information they prefer to make the simplest possible completion. A clause with no explicit temporal information is incomplete as to temporal location. The effect isn’t limited to language. For instance, Kanisza (1976) discusses visual perception and shows that people interpret visual information along these same lines. He gives examples of how complex figures are perceived as simple gestalts, making the point that perceivers add as little as possible. Similar principles are used to constrain computational reasoning procedures. The three principles: (2)

The Deictic Principle Speech Time is the central orientation point for language. The Present time is located at Speech Time; the Past precedes it, the Future follows.

(3)

The Bounded Event Constraint Bounded situations may not be located in the Present.

(4)

The Simplicity Principle of Interpretation Choose the interpretation that requires least information added or inferred.

For English and other tensed languages, these three principles affect the interpretation of the present tense. For tenseless languages, they underlie the default pattern of temporal interpretation, which is, very simply this: unbounded situations are located in the Present, bounded situations in the Past. The Future requires explicit information. See below for discussion and examples.

1.3 Temporal Information in Language: A Classification Languages can be classified according to how they convey temporal information. I propose a three-way classification that allows for the variation that we find among languages. I suggest that we recognize fully tensed languages, mixed-temporal languages, and tenseless languages. The differences between them are important for the semantic-pragmatic account that I develop here. The notion of tense can be taken broadly or narrowly. In recent cross-linguistic work, Dahl, Bybee and others have proposed a broad Tense-Aspect-Mood category that can be expressed linguistically in a variety of ways (Bybee et al., 1994; Dahl and Velupillai, 2005). I will explore a narrower view, essentially following Comrie (1976). I shall say that tense is a morpheme that expresses temporal information, a verbal inflection or auxiliary. The tense morpheme is obligatory, part of the grammatical ‘spine’ of a sentence. As such, tense has grammatical ramifications: it is involved in agreement, case, anaphora, and the finite/non-finite distinction. All main clauses have an obligatory tense morpheme, so that all main clauses convey temporal information.

232

C.S. Smith

I take this to be the hallmark of a fully tensed language. English, French, German, Hindi, are tensed languages. (They differ as to how aspect is realized.) Mixed-temporal languages have some of the characteristics of tensed languages. They have inflectional morphemes and/or temporal particles and clitics that give direct temporal information, but they are syntactically optional. Thus a given sentence may or may not convey temporal information; Navajo and other Athabaskan languages are of this type.5 Finally, there are languages without temporal inflections or particles, such as Mandarin Chinese and Thai, some Mayan languages, and probably others.6 I will refer to them as tenseless languages. Temporal adverbs are not included in these categories; so far as I know they are optionally available in all languages. I assume that the syntax of a fully-tensed language includes the TensePhrase functional category. The other types of language have a syntactic AspectPhrase category but no TensePhrase. These languages introduce some temporal information, but the information must be supplemented by pragmatic inference, as developed in detail below. I do not posit syntactic structure that corresponds to the pragmatics of temporal inference. Thus although all languages convey information that allows temporal location, they do so with different syntactic structures and semantics. In the discussion I consider tensed languages first and then focus on sentences without direct temporal information in mixed-temporal and tenseless languages.

2 Tensed Languages Tense codes temporal information directly. In the neo-Reichenbach account that I assume, tense gives information about three times and their ordering relations of sequence or simultaneity. Present tense conveys that all three times are simultaneous; past tense conveys that Reference Time precedes Speech Time. The future conveys that Reference Time follows Speech Time, always with the element of uncertainty that inheres in the future. Reference Time and Situation Time are simultaneous with simple tenses, in sequence with other tenses and certain syntactic structures.7 Tense alone locates only relationally. Adverbs usually specify RT except in distinguishable cases (e.g. with perfect tenses, future-in-past, etc.). I will focus on tense and the aspectual information that allows temporal interpretation, with some reference to temporal adverbs.

5

I cannot offer a complete cross-linguistic account here; there are undoubtedly differences among mixed-temporal languages that are worth exploring. 6 For information about tenseless languages see the World Atlas of Language Structures (Haspelmath et al. 2005). 7 The perfect conveys that Situation Time precedes Reference Time. A complement clause can convey that Situation Time precedes or follows, e.g. Max said that Nora (had) left, Max said that Nora was leaving soon.


233

Neo-Reichenbach accounts of tense are given by Hinrichs (1986), Smith (1991/1997), Kamp and Reyle (1993), Klein (1994).8 The pragmatic principles constrain temporal interpretation in English and other tensed languages. There are two cases in English: the default interpretation of present tense sentences as located in the Present; and aspectual coercion in present tense simple (non-progressive) event sentences. Present tense sentences are flexible in interpretation with temporal adverbials. Present sentences are temporally located in the Present or Future, depending on the adverb with which they appear e.g., Mary is working now/ tomorrow.9 Now consider present tense sentences without adverbials, as in (5). (5)

a. b.

Mary is working. Carl owns the farm.

Given the facts about adverbials, we might expect sentences like this to be indeterminate between a Present and Future reading, but they are not: the default is Present. The interpretation is predicted by the Deictic and Simplicity Principles. By the Deictic Principle, the Present is the preferred temporal location. By the Simplicity Principle, the Present is simpler than the Future in terms of the information conveyed: the Future always has an element of uncertainty. Temporally the Future follows Speech Time, but it is unlike other times in having this additional element. As Lyons puts it “Futurity is never a purely temporal concept; it necessarily includes an element of prediction or some related notion” (1977: 677). The future is ‘open’: we cannot know what will happen but can only predict, with various degrees of certainty (Yavas¸ , 1982). This uncertainty is explicit in the branching-time schema proposed by Dowty (1977), Landman (1992) for future time: one cannot be sure which of the branches possible will be the one that actually occurs. The element of uncertainty makes the Future more complex than the Past or the Present in terms of the information it conveys.10 Now consider the aspectual interpretation of event sentences with present tense. Present tense event sentences with the progressive, and statives, are not problematic: (6)

Mary is talking. Leigh believes in ghosts. Sam is in the garden.

8 Klein’s account of tense differs in several ways from Reichenbach’s; his notion of Topic Time, however, is close to Reference Time. Differences arise in accounting for the perfect, which involves both temporal ordering (anteriority), and embedded futures. Bohnemeyer (2003) argues convincingly that Klein’s approach cannot deal with relative tenses except as special cases. 9 Present tense also appears in the historical present – present tense and past adverb –, which locates a situation in the past. The historical present needs an appropriate context, beyond the scope of this discussion. 10 The uncertainty of prediction makes it a kind of modality, as Yavas¸ (1982), Enc ¸ (1996) point out. An account of the meaning of the future must allow for different possibilities. In current formal semantic theories of modality, such differences are modeled as possible worlds (Kratzer, 1981, von Fintel, 2006).

234

C.S. Smith

The sentences of (6) express ongoing situations located in the Present. But there is an apparent problem with perfective event sentences: sentences with the simple, non-progressive verb form, which codes the perfective aspectual viewpoint. The problem is simply that such sentences should not be possible with present tense. Recall that, according to the Bounded Event Constraint, bounded events are not located in the Present. Bounded events cannot be located at Speech Time, yet the perfective viewpoint focuses events with initial and final boundaries. Aspectual coercion resolves the problem: simple present tense event clauses are semantically stative in English. They express a generalization, a pattern of events rather than a particular event (Krifka et al., 1995). The basic event verb and its arguments are coerced from eventive to stative: such sentences are derived statives: (7)

English derived statives Kim walks to work. Leigh (usually) sleeps late. Lions eat meat.

Since states surround Situation Time, at Speech Time they can extend indefinitely into the past or future, as in ‘timeless’ or ‘temporally indefinite’ interpretations of generalizations. Tensed languages may honor the Bounded Event Constraint in more than one way. In French, for instance, the présent is imperfective: events are unbounded. In Russian and Polish, the present perfective locates events in the Future (Comrie, 1976; Vetters and Skibinska, 1997). I now turn to an account of temporal interpretation in tenseless and mixedtemporal languages. I will show that the notions of Speech Time, Reference Time, and Event Time are relevant for these languages as well as for tensed languages. In languages that have sentences without direct temporal information, aspectual viewpoint codes the relation between Reference Time and Situation Time. The relation between Speech Time and Reference Time, which is necessary for temporal location, is pragmatically inferred. I begin with Mandarin Chinese.

3 Tenseless Languages and Mixed-Temporal Languages 3.1 Mandarin Chinese In tenseless languages a sentence need not have direct temporal information – although temporal adverbs that give such information are always possible. I will show that aspect enables inference about temporal location. Before the discussion I state the basic pattern of default temporal location; it holds for Mandarin and, so far as I know, generally for sentences without direct temporal information.


(8)

235

Temporal location pattern – a default Unbounded situations, Present Bounded events, Past

This pattern is explained by the three pragmatic principles stated above. Unbounded situations are located in the Present, by the Deictic Principle and the Simplicity Principle: the simplest deictic interpretation is Present. Bounded events are located in the Past, as has often been observed. The explanation uses all three principles. By the Deictic Principle and the Bounded Event Constraint, bounded events are oriented to Speech Time but cannot be located in the Present. They might then be located in the Past or the Future. By the Simplicity Principle, they are located in the Past. Recall that the Past is simpler in terms of information conveyed than the Future: the Past doesn’t have the element of uncertainty that is always part of the Future. To locate situations in the Future, and to override the defaults given in (8), explicit temporal information is needed. One source of such information is the temporal adverb. Other sources include future-oriented verbs such as plan, expect; adverbs or other information in the sentence or context.

3.2 Inferred Temporal Location In this section I show how aspectual information allows the inference of temporal location. We have seen that temporal location in tensed languages requires three times: Speech Time, Reference Time, and Situation Time. The same three are needed to account for temporal location in languages without tense. Aspect conveys the relation between Reference Time and Situation Time; the relation between Speech Time and Reference Time is inferred. I demonstrate below for Mandarin Chinese. Mandarin is tenseless, I take it (cf. Chao, 1948; Hu et al., 2001; Lin, 2003); the language has a rich aspectual system. There are several overt aspectual viewpoints: two perfective morphemes (-le and –guo) and a group of Resultative Verb Complements (RVCs) which are also perfective.11 The language has two imperfective viewpoints (zai and –zhe). In Mandarin overt viewpoints are optional; the neutral viewpoint occurs in zero-marked clauses. See Smith (1991/1997), Smith and Erbaugh (2001, 2005) for detailed discussion of Mandarin. The grammar of Mandarin requires the notions of Reference Time. I present several kinds of evidence for this claim. The first comes from a comparison of the perfective viewpoint suffixes of Mandarin. The –le perfective conveys that the event talked about occurred at some time, unstated directly; the –guo perfective conveys that the event occurred prior to a given time. The examples illustrate, from Chao 1968: grammatical morphemes are given in capital letters in the glosses.

11

Resultative Verb Complements are verbal suffixes that convey perfective viewpoint, and in addition have content in their own right; see Smith (1991/1997) for discussion.

236

(9)

C.S. Smith

a.

b.

W˘o shu¯aiduàn-le tu˘ı I break-LE leg I broke my leg (it’s still in a cast). W˘o shu¯aiduàn-guo tu˘ı I break-GUO leg I broke my leg (it has healed since).

Reference Time explains the contrast of 9a and 9b: the viewpoints code different relations between SitT and RT. The -le perfective conveys that SitT is the same as RT. In contrast, the -guo perfective conveys that SitT is prior to RT; -guo is essentially a perfect.12 Adverbials also provide evidence for Reference Time in Mandarin. There are two temporal adverbs, y¨ıjing and ca´ı, that code a sequential relation between RT and SitT. Both convey that the event expressed is prior to a given time, the temporal standpoint: SitT precedes RT. The examples illustrate; 10a has the sentential particle le, arguably a different morpheme from the -le perfective suffix (Li and Thompson, 1981): (10)

a.

b.

Zuóti¯an w˘anshang t¯a y˘ıjing z˘ou le yesterday evening s/he already leave LE. Yesterday evening s/he had already left. W˘o ca´ı daò. I only-just arrive. I have just arrived.

These adverbs are similar to already in English. Reference Time is needed to model the relations between situations expressed in complex sentences and in texts. In complex sentences, for instance, the situation in one clause may be located relative to the situation in the other clause. The locating clause provides RT, as in (11), from Mangione and Li (1993). In this sentence DE is a nominalizing morpheme: the understood subject of the second verb (z˘ou) is the same as the overt subject of the first verb (ch¯ı).13 (11)

T¯a ch¯ı-le fàn cái z˘ou de. S/he eat-LE rice only-then go DE. Only after eating did s/he go.

The structurally determined RT for ‘going’ is the time of ‘eating’ (1993: 67). The wider context may also provide information locating situations as overlapping or in sequence. Overlapping situations share Reference Time; those in sequence do not. Thus RT provides a locus for relating situations in a principled manner, explicated for English in Hinrichs (1986). 12

The other differences between the Mandarin perfective morphemes are not relevant here (see Smith (1991/1997), Lin (1993)). 13 The morpheme DE is a hallmark of this construction, known as ‘shi..de’ construction; the verb shi may be omitted, as it is here. The second event – going – is grammatically nominalized. DE also appears in complex nominals as essentially a relative clause marker, and has other functions as well. This section includes examples of DE as a full and reduced relative clause marker.


237

Finally, shifted deictic forms give a different type of evidence for Reference Time. Deictics such as here, now orient to Speech Time; but in English and in Mandarin they can shift to another orientation time. Just as deictics in their basic uses suggest the temporal standpoint of the Present, shifted deictics suggest a different temporal standpoint. The example is a fragment from a 1997 novel. The speaker compares a past time with her rough life at an earlier time; the shifted deictic is xiànzài (now).14 (12)

. . . xi˘angdào g˘ei nàge sh¯a qi¯an d¯ao kèrén dà b¯azhang y¯ı b˘ıjiào . . . think give that kill 1,000 knife guest big palm one compare zh¯en sh`ı bù-sh¯en hu`ıshou. W˘o xiànzài s˘uox`ıng nénggòu jià g¯ei B˘ı really be boundless comparison. I now simply able marry to Bi Xi¯ansh¯eng. . . Mr. . . . thinking back to the time when I slapped that violent killer guest, really a boundless difference. Simply, now I was able to marry Mr Bi.

Here the temporal standpoint is the Past. To account for these facts, I shall say that the aspectual viewpoint morphemes of Mandarin code the Reference Time–Situation Time relation: it is part of the semantic meaning conveyed by the forms. No grammatical forms relate Reference Time to Speech Time, however. This is the relation that locates a situation temporally. In tenseless languages it is determined indirectly, by inference, as I will show directly. Recall that the key factor in inferred temporal location is the boundedness or unboundedness of the situation expressed. The default pattern is repeated here: (8)

Temporal inference pattern Unbounded situations, Present Bounded events, Past

Overt aspectual viewpoint is optional in Mandarin. Thus there are two cases, sentences with overt aspectual viewpoints and zero-marked sentences; the latter have the neutral viewpoint. I first give examples of sentences with overt viewpoints, which follow the pattern above and can be understood with the notions already introduced, and then consider sentences with the neutral viewpoint.

3.3 Sentences With Overt Aspectual Viewpoints Imperfective viewpoints focus situations as unbounded, with no information as to endpoints. They are taken as temporally located in the Present by default; DE in these examples is similar to a relative clause morpheme. 14 This and subsequent examples are given more fully, with their provenance, in Smith and Erbaugh (2005).

238

(13)

C.S. Smith

zai (progressive) a. sh`ısh´ı-shàng zhè-zh˘ong mósh˘ısh˘ı za`ı cha¯ox´ı k¯exué fact-on, this-kind model ZAI copy science In fact, this model is already copying the natural sciences. b.

(14)

y¯ıqiè do¯u zài zhàob¯an hél˘ıhuà de mósh`ı. everything all ZAI imitate rationalization DE model. ‘Education in Hong Kong is already too systematized; everything is all about indiscriminately imitating some model of ‘rationalization”.

-zhe (stative imperfective) bú dà p´ıngzhˇeng de Yánzhe bi¯an fàng-zhej˘ı kuài x˘ıy¯ıb˘an, DE edging side set-ZHE several CL washboard, not very level lu˘ansh´ı d`ımiàn sh˘anyào-zhe y¯ı dài yòu y¯ı dài cobblestone floor surface shine-ZHE one generation also one generation zhùm´ın su˘o tàch¯u de gu¯angzé. resident SUO tread DE glow. ‘(describing a courtyard) Several washboards are set down along its sides, where the rough cobblestone surface shines with a gloss trodden smooth by generation after generation of residents’.

The Deictic and Simplicity Principles predict this interpretation. Perfective viewpoints focus events with bounds, and are located in the Past by default; (15) is an example; it has an RVC suffix, which conveys the perfective. (15)

Bounded event in the Past . . . zhè sh`ı wh˘o hé du¯o we`ı niánq¯ıng xuézhˇe jia¯otán hoù su˘o . . . this be I and many CL young scholar exchange-talk after SUO dé-daò de jiélùn. reach-RVC DE conclusion ‘This is the conclusion which many young scholars and I reached after exchanging views’.

As noted in section 2.1, perfectives cannot be located in the Present, by the Deictic Principle and the Bounded Event Constraint. They are located in the Past rather than the Future by the Simplicity Principle. The examples of (9) above illustrate the same point.


239

The default interpretations can be overridden when there is information to the contrary. Two cases are illustrated in (16): (16)

a.

b.

Unbounded event in the Past W˘o y¯ı tóu g¯ongzuò, y¯ı tóu wén zhe y¯ı zhèn y¯ı zhèn gu¯o lˇı I one head work, one head smell-ZHE one burst one burst pan in huáng yú zài wˇomen a¯ nhu¯i sh`ı méi yˇou de. de yúxi¯ang, Anhui be not exist DE. DE fish fragrance, yellow fish in our ‘One part of me was working, one part was smelling wave after wave of the fragrant smell of the fish in the pan. Yellow croaker fish was something we didn’t have back home in Anhui. . . ’. Bounded event in the Future Ne`ıd`ı ji¯ang yú bˇen zh¯oul`ıu quánmiàn j`ın yòng j´ıyòngj´ıq`ı mainland will to this Saturday completely forbid use disposable de f¯apàoji¯ao c¯anjù DE polystyrene food-containers ‘This coming Saturday the mainland [from Hong Kong] will completely ban the use of disposable polystyrene food-containers’.

In (16a) the larger context, an autobiographical text, sets the narration in the Past; in (16b) the future ji¯ang (will) and the temporal adverb indicates Future. The default can also be overridden by information that locates unbounded events in the Future, and states in the Past and Future.

3.4 Other Tenseless Languages Tenseless languages other than Mandarin have the same pattern of temporal interpretation based on inference from overt aspectual viewpoints. I give examples from Thai and Yukatek Mayan. As we now expect, sentences with an imperfective viewpoint are taken as Present; sentences with a perfective viewpoint (marked in (18) as CM for ‘Completive’) are taken as Past.15 (17)

15

Thai a. nit kamla tææ klonge Nid IMPF compose poem Nid is composing a poem b. lûuk nòy kamla uˆ an nâa-rák yûu child Noy ADV fat cute IMPF ‘Noy’s baby is fat and cute’

The Thai examples are due to Sudmuk (2001), Iwasaki and Ingkaphirom (2005); the Yukatek Mayan examples are from Brody (1998). Mayan languages vary considerably in the temporal domain; this is an interesting topic for future research.

240

C.S. Smith

c.

(18)

nit sâa bâan ya mây sét Nid build house PERF ‘Nid built a house’

Yucatec Mayan a. táan in pak’ –ik-ø DUR 1A plant.IMPF.TR-3b ‘I am planting it’ b. t-in pak’-aj-ø CM-plant-CM-TR-3b ‘I planted it’ c. T-a ch’a-(a) le in chóoy-o’ CM-2A take-PERF-TR DEM my bucket-DEM ‘You took my bucket’

These examples from three languages show that the pattern given of interpretation is quite general in tenseless languages. The same pattern is found in mixed-temporal languages.

3.5 Mixed-Temporal Languages The category of ‘mixed-temporal’ language suggested above distinguishes languages which have characteristics of both tensed and tenseless languages. They may have temporal verbal inflections, particles, or other temporal forms (in addition to adverbs). These forms are optional, so not all sentences have direct temporal information. Navajo is such a language. Some clauses have direct temporal information: there is a future verb prefix, and past and future temporal particles. The future prefix is a member of a set of prefixes known as ‘modes’; one mode prefix appears in all event clauses but they are not available for stative clauses. The future is the only mode prefix that is temporal: the others convey aspectual and modal information.16 Semantically the future prefix is privative: its absence conveys no information about futurity. The past and future particles are optional. Thus Navajo permits sentences with no direct temporal information. In Navajo sentences that lack direct temporal information, temporal inference follows the default pattern identified above. Aspectual viewpoints allow temporal interpretation, as (19) illustrates: the examples are from Smith et al. (2003), which discusses the Navajo system in some detail. (19)

Navajo a. Biih yish’née´ h into-3 1psubj-IMPF-crawl

16 There are 7 mode morphemes in Navajo. 3 convey aspectual viewpoints: Perfective, Imperfective, and Progressive. 2 convey aspectual situation type information; the Customary and Iterative modes; the remaining modes are the Future and the Optative.


b.

c.

241

‘I’m crawling into it’ Nlé´ı dziłba¸a¸hgóo´ hoołti¸¸ił that-one-over-there mountainside-along 3-3psubj-PROG-rain ‘It’s raining there along the mountainside’ Hooghan binishishnish hogan on-3-1psubj-PERF-work ‘I did some work on a hogan’

Other mixed-temporal language include Hua, a Papuan language. In Hua the main distinction is also Future-Nonfuture (Haiman, 1980). Hopi, a Uto-Aztecan language, also has a Future–Nonfuture system (Malotki, 1983).

4 Zero-Marked Sentences: The Neutral Viewpoint In Mandarin and a number of other languages, a sentence may appear without an overt aspectual viewpoint. Such clauses are zero-marked. In the two-component approach to aspect, all sentences have an aspectual viewpoint: viewpoint makes all or part of a situation visible for semantic interpretation. Zero-marked clauses have the neutral viewpoint. This viewpoint is flexible, allowing open and closed interpretations. It requires only that part of a situation be visible; this accounts for the indeterminacy of zero-marked clauses, since the event or state may or may not extend beyond the Situation Time interval. There is a default pragmatic interpretation for zero-marked clauses, which is based on situation type, or event structure. This default supplements the weak semantic information of neutral viewpoint. The default inference depends on whether boundedness is expressed by temporal features of the situation of the clause. Telic and single-stage events (Achievements and Semelfactives) are intrinsically bounded; States and atelic events (Activities) are unbounded. This information is part of semantic representation. The temporal features are associated with the event or state entity introduced in a clause. (20)

Temporal Schema Principle In a zero-marked clause, interpret boundedness according to the temporal features of the event or state entity.

This default pattern supplements the weak semantic information of neutral viewpoint. It is a special case of the Simplicity Principle. The default provides the input to temporal interpretation, which follows the pragmatic principles in inferring that bounded events are taken as Past, unbounded situations are taken as Present. In Mandarin, zero-marked clauses are optional for event clauses and required for statives of all kinds. The overt viewpoint morphemes are not available for statives. The boundedness interpretation of these clauses depend on the event or state entity, as in (20). In clauses with intrinsically bounded situations, the situations are taken as bounded; otherwise the situations are taken as unbounded. With this inferred

242

C.S. Smith

boundedness information, temporal location inference proceeds as it does for clauses with overt aspectual viewpoints. The examples illustrate: (21)

Unbounded: states and atelic events a. State xi¯angg˘ang méiy˘ou b`ıgu¯an z`ısho˘u de tiáojiàn Hong Kong not-have close self-self DE situation. b.

c.

‘Hong Kong does not have the option of closing its doors’. Generalizing stative F¯apàoji¯ao c¯anjù f¯angbiàn, chéngbˇen d¯ı dàn we¯ıhài Polystyrene food-container convenient, price low, but harm j´ı dà, extremely great, ‘Polystyrene food-containers are convenient and cheap, but extremely harmful.’ Activity zài d´ıs¯ık¯e jiézòu zh¯ong, Xióng kuáng fàng de qián hòu yáob˘ai, at disco rhythm inside, Bear wild style DE front back sway, Xióng chén j¯ın zài z`ıjˇı chuàngzào de wˇud˘ao zh¯ong. DE dance in. Bear deeply immerse in own create ‘Inside the disco rhythms, Bear sways wildly back and forth, deeply immersed in a dance of his own creation’.

(22)

Bounded events a. Wáng Jizh`ı f¯am´ıng zh¯ongwén d˘az`ıj¯ı Wang Jizhi invent Chinese word processor ‘Wang Jizhi invented the Chinese word processor’ b.

za`ı q˘u-dé zu`ı chù de chéngg¯ong zh¯ıhoù, N´ı gu¯angnán yoù at take-receive most early DE success after, Ni Guangnan also t´ıch¯u x¯ın de yánji¯u k¯et´ı bring up new DE research project, ‘After his early successes, Ni Guangnan proposed new research projects.’

These examples involve two steps of default inference: the Temporal Schema Principle provides the inference of boundedness or unboundedness, and the Temporal Inference Pattern provides the inference of temporal location. Zero-marked clauses appear in other tenseless languages; some have the same pattern in Mandarin, others do not. Thai has a slightly different pattern as its written and spoken forms. In written Thai temporal interpretation is based on the property of intrinsic boundedness, as in Mandarin. In conversational Thai the property of duration determines whether a situation is located in the Present or the Past. States and durative events – telic and atelic – are taken as Present, while single-stage, punctual events are taken as Past, as in these examples:


(23)

243

Thai a. State n´ıt mii khaamsuk Nid have happiness b.

c.

d.

‘Nid is happy’ Durative events n´ıt hua rór Nid laugh ‘Nid is laughing’ n´ıt sâa bâan Nid build house ‘Nid is building a house’ Punctual event n´ıt phop naalilkaa Nid find watch ‘Nid found a watch’

For this information about Thai I am indebted to Sudmuk (2001). Mixed-temporal languages also allow zero-marked clauses. In Navajo, for instance, such clauses appear when the mode morpheme does not convey an aspectual viewpoint. Temporal interpretation of such clauses in Navajo are also based on event structure: durative clauses are taken as simultaneous, punctual clauses taken as sequential (Smith and Perkins, 2005). Although the interpretation of zero-marked clauses is similar across languages, we cannot assume that it is identical.

5 Formalizing the Analysis The account of temporal inference proposed above has semantic and pragmatic components. It deals with sentences that do not convey direct temporal information. The account is semantic in using information conveyed by the aspectual forms of a sentence: viewpoint morphemes, and the composite of verb and argument that realizes situation type, or event structure. It is pragmatic as well: the default inferences about temporal location are pragmatic in nature. In this section I briefly sketch a formal account for Mandarin Chinese in the framework of Discourse Representation Theory (Kamp and Reyle, 1993). The framework includes syntactic, semantic and pragmatic information, and information from context. Construction rules introduce into the Discourse Representation Structure (DRS) the semantic information conveyed by linguistic forms. The construction rules apply to a fully-annotated syntactic structure. Syntax: For Mandarin Chinese, I assume that aspectual viewpoints are in the AspP node of the syntactic surface-structure tree. I assume an AspectPhrase node rather than a TensePhrase in the syntactic structure of tenseless languages like Mandarin,

244

C.S. Smith

following Cole and Wang (1996). This is the first step of formal implementation for the account given above: the viewpoints introduce Reference Time and Situation Time and their relation, while pragmatic inference relates Reference Time to Speech Time. The construction rules are semantic: they enter into the DRS the information associated with linguistic forms. For the temporal interpretation of tenseless sentences, we need rules that introduce the information conveyed by aspectual forms; tensed sentences, of course, require rules for interpreting tense.17 Aspectual information involves situation type and viewpoint. It introduces situation entities, times, and conditions that characterize and relate them. Situation type: situation entities are composed by rule for each clause according to the verb and its arguments. The rules introduce a situation entity E into the DRS with conditions that associate with E the identifying cluster of temporal properties. For instance, (24a) gives an Achievement sentence in Mandarin and (24b) a construction rule for interpreting the aspectual information: (24)

a.

b.

w˘o zhaòdaò w˘ode l¯uy¯ınj¯ı le I find+arrive(RVC) my tape recorder LE I found my tape-recorder Construction rule CL [X NPdef Y VAch RVC (-LE) (NPdef ) Z] => E; +Dyn, +Punct, +Telic

The left-hand side of the rule specifies that it apply to a clause with a definite subject NP, a verb of the Achievement class with an RVC and an optional perfective morpheme -le, and an optional definite object. The right-hand introduces a situation entity E into the DRS, with the conditions that it is dynamic, punctual, and telic. These are the characteristics of an Achievement event. Similar rules are required for the other situation types, and for aspectual coercion, shifts from one situation type to another. Viewpoint information: On the analysis proposed above, aspectual viewpoints in Mandarin introduce two times, Reference Time and Situation Time, and their relation. The times are annotated as RT (t2 ) and SitT (t3 ). The viewpoints also introduce boundedness information, which is represented by conditions relating the SitT interval and the situation entity E. Perfective viewpoints make visible a bounded event, introducing a condition that the SitT interval t3 includes E: E ⊆ t3 . Imperfective viewpoints make visible a situation without bounds, introducing a condition that E overlaps SitT: E O t3 . The neutral viewpoint introduces a condition providing that a portion of the situation is semantically visible. I assume that the event or state entity E has been introduced into the DRS by the verb and its arguments. (25) gives construction rules for the viewpoint morphemes of Mandarin; for each morpheme, the rule introduces the information associated with that morpheme into the DRS. Boundedness information is stated in terms of the relation between the situation entity and the Situation Time interval t3 . Bounded events are included in t3 , other situations overlap t3 . The rather cumbersome statement for the neutral viewpoint provides 17

For a sketch of such rules see Smith (2006).


245

for the indeterminate information conveyed, merely that some part of the event or state is semantically visible. (25)

Construction rule for the viewpoint morphemes of Mandarin a. Cl [X AspP [-le] Y] → t2 , t3 ; t3 = t2 ; E at t3 ; t3 ⊆ E b. Cl [X AspP [–guo] Y] → t2 , t3 ; t3 < t2 ; E at t3 ; t3 ⊆ E c. Cl [X AspP [zai] Y] → t2 , t3 ; t3 = t2 ; E at t3 ; E O t3 d. Cl [X AspP [-zhe] Y] → t2 , t3 ; t3 = t2 ; E at t3 ; E O t3 e. Cl [X AspP [ø] Y] → t2 , t3 ; t2 = t3 ; E at t3 ; t3 ⊆ e]; ∃ e ∃ e [e ∈ E & l(e ) < e & t1 ⊆ e]

These rules provide that two times, t2 and t3 are introduced in the DRS with each aspectual viewpoint. The third time involved in temporal meaning, t1 for Speech Time (SpT), is entered automatically into the DRS for each clause, as posited by Kamp and Reyle (1993). The pragmatic inference rule for temporal location – (27) below – relates t1 and t2 , Speech Time and Reference Time. The neutral viewpoint does not give boundedness information, so that an additional inference rule is needed for zero-marked cases. Recall the Temporal Schema Principle, which says that boundedness is inferred from the intrinsic boundedness property of a situation entity. Telic and punctual events are intrinsically bounded, other situations are not. The rule looks at the temporal properties associated with the situation entity E in the DRS; if they include the properties [+ telic] or [+ punctual], boundedness is inferred. If other temporal properties are associated with E, unboundedness is inferred.18 The inferences about boundedness take the form of statements about the relation between E and the Situation Time interval t3 . (26)

Default boundedness inference rule: Temporal Schema Principle a. E; [+telic] ∨ [+punctual] → t3 ⊆ E b. E; [±dynamic]; [-punctual] ∨ [-telic] → E O t3 .

As before, we represent boundedness by including E in the SitT interval, and unboundedness by providing that E overlap the SitT interval. This new condition is more informative than the condition given in (25e) above. It is only a default and can be overridden by additional information. The information provided by this rule serves as the input to the temporal inference rule below. I can now state a simple rule of default temporal inference. The rule infers temporal location from the boundedness of the situation entity. Bounded events are included in the t3 interval and trigger the inference of Past; unbounded situations overlap the t3 interval and trigger an inference of Present. The boundedness information conveyed by viewpoint is represented by the relation between E, the situation entity, and t3 , SitT; from this the rule infers the relation between t1 and t2 (SpT and RT). The inference rule looks at the boundedness property of the situation entity E and introduces a new condition into the DRS that relates t1 to t2 . 18 Recall that 3 two-valued temporal features characterize situation entities: [±dynamic] or dynamic-static; [±telic] or telic-atelic; [± punctual] or punctual-durative.

246

(27)

C.S. Smith

Default temporal inference rule a. If E O t3 → t1 = t2 b. If t3 ⊆ E; → t1 < t2

The relation between Speech Time (t1 ) and Reference Time (t1 ) is the inference of temporal location. Sentences with overt aspectual viewpoints convey boundedness information semantically as provided in Rule 25; zero-marked sentences have the additional step of inference given in Rule 26.

6 Conclusion Temporal interpretation, I have shown, is guided by a very few pragmatic principles. The principles constrain temporal location in tensed languages and guide temporal inference in other cases. I propose three classes of language according to how temporal information is conveyed: tensed, tenseless, and mixed-temporal. In tensed languages every sentence has direct information about temporal location; mixedtemporal languages may have grammatical temporal forms, but they are optional and do not appear in every clause. Tenseless languages do not have temporal information in every clause. For the last two cases temporal location is inferred from aspectual information. The notions of Speech Time, Reference Time, and their relations are needed to explain temporal meaning across languages. Tense conveys this information directly. Decoupling the two relations involving Reference Time is the key to temporal interpretation in tenseless and mixed-temporal languages.19 The relation between Reference Time and Situation Time is conveyed semantically by aspectual viewpoint. The relation between Reference Time and Speech Time is pragmatically inferred. There are two simple default principles, which can be overridden by adverbs and other temporal information in the sentence or context. Boundedness is the main factor in temporal inference. It is expressed directly by overt aspectual viewpoints. In zero-marked clauses, both boundedness and temporal location are inferred from the event structure associated with a sentence. Further research is needed to test and develop the general claims I make about types of temporal information in languages of the world. There may be a another type of language, one that has neither tense nor overt aspectual viewpoints. Apparently Maybrat, a language of Indonesia, is such a language (Dol, 1999). Maybrat is tenseless, with optional temporal and aspectual adverbs. I would hypothesize that the pragmatic principles hold for this language as well, and that event structure information allows temporal inference. If so, we would expect that intrinsically bounded events are taken as Past by default, others are taken as Present; and that overt information is needed for Future temporal location.

19

More precisely, in sentences that do not have temporal adverbs or other forms that convey direct temporal information.


247

References Bach, Emmon, 1986. The algebra of events. Linguistics and Philosophy 9:5–16. Bohnemeyer, Jürgen, 2003. Relative tenses vs. aspect: the case reopened. Semantics of Under-represented Languages 2, handout. University of British Columbia. Brody, Michal, 1998. A sketch of the grammatical categories of Yukatek Maya with particular focus on the verbal system. MA thesis, University of Texas. Bybee, Joan, Revere Perkins, & William Pagliuca, 1994. The Evolution of Grammar: Tense, Aspect, and Modality in Languages of the World. Chicago, IL: University of Chicago Press. Chao, Yuen Ren, 1948. Mandarin primer: An intensive course in spoken Chinese. Cambridge, MA: Harvard University Press. Chao, Yuen Ren, 1968. A Grammar of Spoken Chinese. Berkeley, CA: University of California Press. Cole, Peter, & Chengchi Wang, 1996. Antecedents and blockers of long-distance reflexives: The case of Chinese ziji. Linguistic Inquiry 27:357–390. Comrie, Bernard, 1986. Tense. Cambridge: Cambridge University Press. Dahl, Oesten, & Viveka Velupillai, 2005. Tense and aspect: Introduction. World Atlas of Linguistic Structures. M. Haspelmath, M. Dryer, D. Gil, & B. Comrie (eds.). Oxford: Oxford University Press. Depraetere, Ilse, 1995. On the necessity of distinguishing between (un)boundedness and (a)telicity, Linguistics & Philosophy 18:1–19. Dol, Philomena, H., 1999. A Grammar of Maybrat: A Language of the Bird’s Head, Irian Jaya, Indonesia. Leiden: Grafisch Bedrijf, Universiteit Leiden. Dowty, David, 1977. Toward a semantic analysis of verb aspect and the English imperfective progressive. Linguistics & Philosophy 1:45–77. Enç , Murvet, 1996. Tense and modality. In S. Lappin (ed), The Handbook of Contemporary Semantic Theory. Oxford: Blackwell. pp. 345–358. von Fintel, Kai, 2006. Modality and Language. D.M. Borchert (ed.), Encylopedia of Philosophy, Second Edition. Detroit, MI: Macmillan Reference USA. Giorgi, Alessandra, & Fabio Pianesi, 1997. Tense and Aspect. Oxford: Oxford University Press. Grice, H.P., 1975. Logic and Conversation. In P. Cole & J. Morgan (eds.), Syntax and Semantics: Speech Acts. New York: Academic. Haiman, John, 1980. Hua: A Papuan language of the eastern highlands of New Guinea. Amsterdam: Benjamins. Haspelmath, Martin, Matthew Dryer, David Gil, & Bernard Comrie (eds.), 2005. World Atlas of Language Structure. Oxford: Oxford University Press. Hinrichs, Erhard, 1986. Temporal anaphora in discourses of English. Linguistics & Philosophy 9: 63–82. Horn, Laurence, 1984. Toward a new taxonomy for pragmatic inference: Q-based and R-based implicature. In Meaning, Form and Use in Context: Linguistic Applications. Deborah Schiffrin (ed.). Washington DC: Georgetown University Press.

248

C.S. Smith

Hu, Jianhua, Haihua Pan, and Liejiong Xu, 2001. Is there a finite-nonfinite distinction in Chinese? Linguistics 39: 1117–1148. Iwasaki, Shoichi & Preeya Ingkaphirom, 2005. A Reference Grammar of Thai. Cambridge: Cambridge University Press. Kamp, Hans, & Uwe Reyle, 1993. From Discourse to Logic. Dordrecht, The Netherlands: Kluwer Academic. Kanizsa, Gaetano, 1976. Subjective contours. Scientific American 234: 48–52. Klein, Wolfgang, 1994. Time in Language. London: Routledge. Klein, Wolfgang, Ping Li, & Henriette Hendriks, 2000. Aspect and assertion in Mandarin. Natural Language & Linguistic Theory 18: 723–770. Kratzer, Angelika, 1981. The notional category of modality. H.J. Eikmeyer & H. Rieser (eds.), New Approaches in Word Semantics. Berlin: de Gruyter. pp. 38–74. Krifka, Manfred, Francis J. Pelletier, Gregory Carlson, Alice ter Meulen, Gennaro Chierchia, & Godehard Link 1995. Genericity: An introduction. In The Generic Book, G. Carlson and F.J. Pelletier (eds.). Chicago, IL: University of Chicago Press. Landman, Fred, 1992. The progressive. Natural Language Semantics 1:1–32. Levinson, Stephen, 1983. Pragmatics. Cambridge: Cambridge University Press. Li, Charles & Sandra Thompson, 1981. Mandarin Chinese: A functional reference grammar. University of California Press. Lin, Jo-Wang, 2003. Temporal reference in Mandarin Chinese. Journal of East Asian Linguistics 12: 254–311. Lyons, John, 1977. Semantics. Cambridge: Cambridge University Press. Malotki, Ekkehard, 1983. Hopi Time: A Linguistic Analysis of Temporal Concepts in the Hopi Language. Berlin: Mouton. Mangione, L. and Dingxuan Li, 1993. A compositional analysis of -guo and -le. Journal of Chinese Linguistics 21: 65–122. Reichenbach, Hans, 1947. Elements of Symbolic Logic. London: Macmillan. Smith, Carlota S., 1991/1997. The Parameter of Aspect. Dordrecht, The Netherlands: Kluwer. Smith, Carlota S., 2003. Modes of Discourse. Cambridge: Cambridge University Press. Smith, Carlota S., 2006. The pragmatics and semantics of temporal meaning. Proceedings of the 2004 Texas Linguistics Conference: Issues at the Semantics – Pragmatic Interface, edited by Pascal Denis, Eric McCready, Alexis Palmer, and Brian Reese. Somerville: Cascadilla Proceedings Project. pp. 92–106. Smith, Carlota S, Ellavina Perkins, Theodore Fernald, 2003. Temporal interpretation in Navajo. Proceedings of the 2nd SULA Conference. Amherst, MA: Amherst Working papers in Linguistics, pp. 175–192. Smith, Carlota S. & Ellavina Perkins, 2005. Temporal inference of zero-marked clauses in Navajo. Proceedings of the 4th SULA Conference. Amherst, MA: Amherst Working Papers in Linguistics. Smith, Carlota S. & Mary S. Erbaugh, 2001. Temporal information in Mandarin. International Symposium of Chinese Grammar for the New Millenium. Shan Zhourao & Xu Liejiong (eds.). Hangzhou: Zhejiang Jiaoyu Chubanshe. Smith, Carlota S. & Mary S. Erbaugh, 2005. Temporal interpretation in Mandarin Chinese. Linguistics 43: 713–756.


249

Sudmuk, Cholchitha, 2001. Temporal information in Thai. Unpublished paper, University of Texas. Vendler, Zeno, 1967. Verbs and Times. Ithaca, NY: Cornell University Press. Vetters, Carl, & Elzbieta Skibinska, 1997. Le futur: une question de temps ou de mode? Remarques générales et analyse du “présent-futur” perfectif polonais. In A. Borillo, C. Vetters, & M. Vuillaume (eds.), Regards sur L’aspect. Amsterdam: Rodopi, 247–266. Yavas¸, Feryal, 1982. Future reference in Turkish. Linguistics 20: 411–429.

The English Konjunktiv II Tim Stowell

Abstract This chapter examines the syntax and semantics of the English Konjunktiv II (K2) construction. Use of the K2 is restricted to an informal register of English; it is replaced by the past perfect in the standard, more formal, register. It occurs only in a subset of the syntactic environments in which the past perfect occurs, however, and is associated with a strongly counterfactual interpretation (primarily counterfactual conditionals and complements to wish-class verbs). Morpho-syntactically, K2 resembles the past perfect, insofar as it contains the preterit form of the auxiliary have (namely, had) followed by the past participle. K2 differs from the past perfect, however, in that an additional particle occurs between had and the past participle; this particle is phonologically a weak enclitic attached to had, and has been analyzed variously as a reduced form of the auxiliary verb have (−’ve) or as a preposition (of ) or particle (a); thus the past perfect form had gone corresponds to the K2 had’ve gone or had of gone. I analyze K2 syntactically as a subjunctive perfect form, where subjunctive mood is conveyed by the preterit affix −ed, and the perfect functions as a past polarity item signaling the presence of a covert past tense. The type of subjunctive mood that occurs in the K2 is distinct from the mandative subjunctive mood that occurs in the complements of demand/ask class verbs. Both types of subjunctive are licensed strictly locally, in contrast to the subjunctive mood licensed by negation in languages such as French. I suggest that this is related to the modal force of the subjunctive in these contexts. The particle of/have in the K2 is a subjunctive polarity item, disambiguating the subjunctive perfect from the indicative past perfect. Key words: Conditional, counterfactual, Konjunctiv II, imperfect, modals, past, perfect, sequence of tense, subjunctive.

University of California, Los Angeles


251

252

T. Stowell

Introducing K2 In this article, I examine the syntax of a little-studied tense-mood construction in English that appears to correspond to a past subjunctive in certain other Indo-European languages; I have dubbed this the “English Konjunktiv II” (or K2 for short) in honor of what I believe to be its closest cross-linguistic counterpart, the German Konjunktiv II (Konjunktiv-Zwei). Although the English K2 behaves in some respects like a past subjunctive, it does not behave syntactically or semantically like a simple combination of past tense and subjunctive mood, at least in terms of how these terms are used in English grammar. In particular, its morphosyntax is at first blush highly irregular, and its semantics corresponds to that of a strongly counterfactual mood. The English K2 is also of interest from a sociolinguistic perspective, insofar as it belongs to an informal spoken register, and is generally avoided in formal written English. K2 is discussed briefly by Kayne (1997), who cites examples such as the following: (1)

a. b. c.

If you hadn’t If you hadn’t If you hadn’t

’a said that. . . ’ve said that of said that. . .

The boldface ’a, ’ve, and of in examples (1a–c) are alternative orthographical representations of a syntactic formative whose syntactic categorial status is open to debate. On one account, it is a phonological reduction of the nonfinite form of the auxiliary verb have in (2): (2)

(%∗ )

If you hadn’t

have said that. . .

Kayne judges (2) to be ungrammatical; my own judgment is that (2) is perfectly acceptable, though I appear to be in the minority among English speakers on this point. An alternative account, advocated by Kayne (1997), is that the boldface formative is a particle entirely unrelated to have; on Kayne’s account, it is a prepositional complementizer introducing a nonfinite participial CP. I will generally use of (instead of have, ’ve, or a) in constructed example sentences containing K2, but when citing naturalistic examples of K2 drawn from the web, I will leave the original orthography unchanged. I will also use of (rather than have) when referring to this particle in the text of the article, without intending to favor any particular analysis of its syntactic category. I will defer discussion of the latter issue to Section 3, where I propose an analysis of the internal syntax of K2. In the meantime, I will focus on a theoretically neutral description of the external syntactic distribution and of the construction, and its associated semantics.

1 The Syntactic Distribution of K2 Because the English K2 is restricted to an informal register, it has a corresponding form in the standard register, which can also be used in formal (written) contexts; that form is the past perfect. K2 differs minimally from the past perfect in its

The English Konjunktiv II

253

morpho-syntactic composition: whereas the past perfect consists of the preterit form of the auxiliary have (had) followed by the past participle as in (3a), K2 in addition contains of (have/’ve/a) lodged in between them, as in (3b). (3)

a. b.

If John had left, . . . If John had of left, . . .

In both cases, had and the past participle may also be separated from each other by the negative particle (not/n’t) and/or one or more adverbs, as in (4a); in the case of K2, negation and most adverbs may precede of, though many orders are possible; a proper treatment of this is beyond the scope of this paper. (4)

a. b. c. d. e. f.

If John had suddenly left, . . . If John hadn’t (suddenly) left, . . . If John had of suddenly left, . . . ?If John had suddenly of left If John hadn’t of left, . . . ?If John had of not left, . . .

Although all occurrences of K2 in the informal register correspond to the past perfect in the standard register, the converse is not the case: not all occurrences of the past perfect in the standard register can be replaced by K2 in the informal register. As far as I have been able to determine, K2 is restricted to two major types of subordinate clauses, both of which have a strongly counterfactual semantics: (5)

a. b.

Counterfactual conditional clauses Counterfactual complements to wish, rather, would have preferred, etc.

K2 is quite common in informal written English, at least as revealed by searches of written texts on the internet, especially in contexts such as web chats and discussion boards. Surprisingly, it appears to have been in use for at least a couple of hundred years; the earliest example that I have come across is attributed to George Washington, though I suspect that it is much older than this. Here is a sampling of naturalistic examples of K2 culled from the internet, thanks to the assistance of Google, the linguist’s new friend. Orthography is left unchanged throughout: (6)

Counterfactual conditionals a. If she had’ve frozen it, she’d’ve blackmailed the Marshall’s for it’s return. b. You might not feel better than if you hadn’t’ve done the clumsy thing to begin with. c. If they had’ve charged $1 per ride, each kid would probably have spent $5 anywa. d. What if there had a been a system in place in which millions of mobile phone subscribers could have received SMS text warnings over their phones?

254

T. Stowell

e. f. g. h. i. j. k.

l.

m. (7)

If I had a been tougher maybe there would have been more drugs for me and I wouldn’t have had to take such a risky route. If there had a been a plan, Iraq might have been moving toward self-rule by now. Listen lady,” he said in a high voice, ”if I had of been there I would of. . . (—Flannery O’Connor) The funny thing was that the burglar alarm was on, so if there had of been anyone there it would have set it off. Even if they had of charged her, the charges would have no doubt been dismissed. If they had of asked me what was wrong with my mum, I could of described to them what happened when my mum was ill. If you had have thought for a moment before clicking the installer excitedly, you might have figured this out on your own. You didn’t, and had to blog about it. The lack of shower came as quite a surprise, although if there had have been one, I’m sure that the towel policy would have been even more irritating. . . I honestly think if they had have put him in the show from the beginning, he still would have made it to the end.

Counterfactual complements to wish, would rather, etc. a. I wish he had’ve come tonight instead. b. We wish we had’ve had the opportunity to move sooner. c. As I am using it, I think to myself, I wish you had’ve installed this feature, and then a couple of minutes later I find a button that does it! d. I wish I had a bought an APPLE II Back then!!! e. I wish you hadn’t a done it. f. I wish he had a gone a bit more in-depth. g. I don’t know what enticed me to open that door, but I wish I hadn’t of. h. I wish I hadn’t of gotten switched to my current class, at least he was fun to look at. i. You wish you had of been the one to invent adult diapers. j. You’re gonna take this and go straight to bed or wish you had of! k. I wish he had of whipped me. I would have felt better. l. I wish I hadn’t have had to have kept that secret from my step mum. m. I wish I hadn’t have had such a shitty relationship with him over the last 30 years. n. I wish I hadn’t have gotten drunk with a bunch of people I didn’t know and almost had sex with a guy that I’d only known for a week. o. There are other things, but it’s all Erica fucking him on that stupid chair. I know, I wish I hadn’t have mentioned it either.. . . p. I would have rather he had of died then [sic] to end his life in an invalid state. q. I’d rather she had of focused on one or the other, instead of both.


r. s. t.

255

I would much rather they had of been, rather than finding out they have been “claimed” by another. This is a thread on ILE I just made, which I forgot to logout to post, and would rather I had have done. Personally I’m not keen on the design of Rent, I’d rather there had have been no set than the confusing set that there is.

In the standard register, the past perfect is used in a number of other, noncounterfactual contexts. In particular, the past perfect is used in contexts such as the following, to convey that an event or situation is located prior to another past time: (8)

a. b. c.

John had already left when Bill arrived. The Queen of Hearts claimed that Alice had stolen the tarts. The detective interviewed a woman who had witnessed the crime.

These are contexts where the usage of the past perfect arguably involves the phenomenon of “sequence of tense”; this raises a number of complex issues that I do not want to go into here. Regardless of this issue, the key fact to observe is that the past perfect cannot be replaced by K2 in the informal register; such examples are completely ungrammatical: (9)

a. b. c.

∗ John

had already of left when Bill arrived. Queen of Hearts claimed that Alice had of stolen the tarts. ∗ The detective interviewed a woman who had of witnessed the crime. ∗ The

Rather, K2 can only be used in counterfactual contexts such as those illustrated in (6) and (7); in contexts such as (8/9), the informal register must use the past perfect rather than K2.

2 A Sociolinguistic Interlude It is sometimes alleged by conservative prescriptive grammarians (experts and amateurs alike) that formal written Standard English is inherently superior to non-standard registers and dialects on aspects of vocabulary or grammar where the dialects or registers differ. For example, it is sometimes pointed out that the use of whom in conservative Standard English preserves a distinction between accusative and nominative case that is supposedly lost in modern dialects and registers where whom is replaced by who. The restrictions on the usage of the English K2 illustrated in (4–7) provide a nice counterpoint to this type of argument, since in this case, the nonstandard informal register preserves a morphological distinction between counterfactual and non-counterfactual semantics that is absent (or lost) in Standard English. Another type of argument that is sometimes advanced in supporting the supposed superiority of Standard English over its nonstandard competitors is the idea that Standard English conforms to systematic rules that the nonstandard dialects and registers ignore. Here too K2 provides the basis for an interesting refutation of this style of

256

T. Stowell

argument. In Standard English, the finite past tense form of the copula exhibits no distinction between indicative and subjunctive mood when the subject of the clause is second person or plural; were is used in both cases. When the subject of the clause is first or third person singular, however, the indicative past tense form (was) differs from the subjunctive form (were). This subjunctive form were is used only in counterfactual contexts; more specifically, it is used in precisely those contexts where K2 occurs in the informal register: (10)

a. b. c. d.

If I were a terrorist, I would keep a low profile. If Beckham were on the team, we would have won the game. I wish I were a member of the royal family. I’d rather he weren’t so optimistic.

(11)

a. b. c.

John was/∗ were already here when Bill arrived. The Queen of Hearts claimed that Alice was/∗ were a thief. The detective interviewed a woman who was/∗ were at the scene of the crime.

This distinction is made only for the copular verb be; for all other verbs, the preterit past and subjunctive forms are identical in the standard dialect. The web is full of naturalistic examples parallel to those in (10): (12)

a. b. c. d. e. f. g.

But if I were Microsoft CEO Steve Ballmer, I wouldn’t be thinking conservatively. If I were Cheney I’d be sweating a little. Oh I Wish I Were a Little Bar Of Soap. Sometimes I wish I were gay. I wish there were more Yoga studios around here. He said he would rather there were real exposes of the casting couch rather than such “framed” encounters. I would rather he were with her. I would rather that she rested in his arms.

The use of the subjunctive form were is characteristic of an elevated, educated, and traditional register; subjunctive were with first and third person singular subjects is widely replaced by was in natural and informal speech, and in much published prose. In this less formal group of registers or dialects, the distinction between indicative was and subjunctive were is apparently lost; was is used not only to convey past tense, as in (11), but also in counterfactual contexts parallel to those in (10) and (11), where there is no past-shifting tense interpretation, as in (13); such examples are excluded in the formal register. (13)

a. b. c. d.

If I was a terrorist, I would keep a low profile. If Beckham was on the team, we would have won the game. I wish I was a member of the royal family. I’d rather he wasn’t so optimistic.

In Section 5, I will discuss one fact suggesting that was may not really be functioning as a subjunctive here, but at this point I will assume that it is.


257

A prescriptive guardian of traditional formal English might be tempted to claim that the formal register adheres to a conditioning rule governing the were/was distinction that the informal register lacks; but this claim is disproved by the fact that the same rule governs the distribution of K2 in the informal register. Rather, the appropriate way of describing the difference between the registers is to say that each register draws a morphological distinction that is neutralized in the other register: the standard register eliminates the morphological distinction between the indicative and subjunctive forms of the past perfect, and the informal register eliminates the morphological distinction between the indicative and subjunctive forms of the copula. The registers do not differ from each other, however, in terms of the grammatical rule regulating the distribution of these forms; each register relies on the same semantic distinction between counterfactual and non-counterfactual environments to determine the distribution of those subjunctive forms for which it provides an overt morphological contrast between subjunctive and indicative. Both registers maintain a distinction between the indicative and the subjunctive, though the overt expression of this depends on morphological contrasts that are not always present.

3 Subjunctive and Indicative Conditionals I have described the contrast between the past perfect and K2 in the informal register as an indicative/subjunctive contrast; this implies that K2 is the subjunctive form of the past perfect. However, it does not have the semantics of the past perfect. Rather, it has the semantics of a counterfactual past. This reflects a general property of tense morphology in counterfactual contexts. To see why this is so, we must consider the contrast between two types of conditional clauses; I will refer to these as indicative and subjunctive conditionals, more or less following traditional usage. The examples in (14) are indicative conditionals, and those in (15) are subjunctive conditionals, referred to as “counterfactual conditionals” above: (14)

a. b. c. d.

If your bird is infected with this virus, it will surely die. If John leaves next week, he will not arrive in time. If John still loves you, he is a fool. If John stole the money, he made a bad mistake.

(15)

a. b. c. d.

If your bird was/were infected with this virus, it would surely die. If John left next week, he would not arrive in time. If John still loved you, he would be a fool. If John had stolen the money, he would have made a bad mistake.

The indicative conditional often involves the occurrence of will in the consequent clause, especially when the consequent clause conveys a prediction, as in (14a–b), but when the consequent clause simply conveys a logical or contingent deduction, a nonmodal indicative tense can occur instead, as in (14c–d). In the subjunctive conditionals in (15), the consequent clause always contains the modal verb would. This usage of would is traditionally classified as a conditional mood form, though in other

258

T. Stowell

contexts it is often described as a past-tense variant of will. Abusch (1988) characterized the contrast between will and would as follows: will and would share a common modal stem, which she named woll; this stem combines with the present tense affix to form will and with the past tense affix –ed to form would. I will adopt this account here. The use of would in the consequent clause of a conditional is usually (though not always) a reliable diagnostic of the subjunctive conditional, and its absence in the consequent clause of a conditional is always a reliable diagnostic of the indicative conditional (with the proviso that certain other past-tense modals, such as could, can also be used in the consequents of subjunctive conditionals, a fact that is often overlooked). Subjunctive conditionals necessarily involve the occurrence of –ed in the protasis as well, as in (15), regardless of whether past tense semantics is involved (as we shall see); in indicative conditionals, –ed may or may not occur, though it when it does occur it always conveys past tense semantics. I will address the tense semantics shortly. The precise characterization of the semantic contrast between the indicative and subjunctive conditionals has been the subject of much debate. It is sometimes assumed that subjunctive conditionals are counterfactual, though Iatridou (2000) has suggested that they are something closer to hypothetical; she describes them as conveying as a “future less vivid”. In contrast, the neutral conditional conveys a “future more vivid”. Iatridou mainly considers eventive verbs, which exclude a present-tense reading in the simple present tense indicative and thus force a future-shifted reading, as in (14b); stative predicates in the simple present tense generally allow, and favor, a presenttense interpretation. The present tense with the stative predicates in (14a) and (14c) conveys a “present more vivid” and the past tense with these predicates in (15a) and (15c) conveys a “present less vivid”. Steering clear of a formal analysis of the semantics of the indicative/subjunctive distinction, the difference between the two types of conditionals can be characterized in terms of the following conditions on their presuppositions, essentially as in Karttunen and Peters (1979): (16)

a. b.

The indicative conditional can be used only if the speaker does not presuppose that the protasis is false. The subjunctive conditional can be used only if the speaker does not presuppose that the protasis is true.

Spelling it out more explicitly, the indicative conditional does not require that the speaker presupposes the protasis to be true; it only requires that the speaker does not presuppose it to be false. Conversely, the subjunctive conditional does not require that the speaker presuppose the protasis to be false; it only requires that the speaker does not presuppose it to be true. Thus, the two types of conditionals differ not in terms of what they presuppose, but rather in terms of what they do not presuppose. This distinction accounts for Iatridou’s intuitions about greater and lesser degrees of “vividness”, in the following way. If the speaker is certain that the protasis is true, the subjunctive conditional is excluded, and only the indicative conditional is possible; if the speaker is certain that the protasis is false, then the indicative is excluded and


259

only the subjunctive conditional is possible. In situations of uncertainty, the speaker’s usage will generally depend on what he or she believes is most likely to be the case, though the usage may be influenced by pragmatic factors; for example, politeness may dictate using the form that corresponds to the speaker’s beliefs about the addressee’s presuppositions. Let us now focus on conditional clauses where the protasis conveys a hypothetical event located in the past. Here the contrast between indicative and subjunctive conditionals is illustrated by the contrast between (14d) and (15d), repeated below, and by (17a) vs. (17b): (14) (15)

d. d.

If John stole the money, he made a bad mistake. If John had stolen the money, he would have made a bad mistake.

(17)

a. b.

If your bird was infected with this virus, it will surely die. If your bird had been infected with this virus, it would probably have died.

In the indicative conditionals in (14d) and (17a), past tense semantics in the protasis is conveyed by the use of the simple past tense affix −ed; in the subjunctive conditionals in (15d) and (17b), past tense semantics in the protasis is conveyed by the use of the past perfect. Iatridou (2000) suggests that, in past-tense subjunctive conditionals like those in (15d) and (17b), the past tense affix –ed in the protasis conveys lesser vividness, and the perfect conveys conventional past tense (past-shifting) semantics. I take this to indicate that past tense suffix –ed is, in general, ambiguous between preterit (past) tense, as in (14d) and (17a), and subjunctive mood, as in (15d) and (17b). Iatridou seeks to provide a unified semantics for semantic past tense and what I am calling subjunctive mood here, which she characterizes in terms of distance from the world and time of the actual speech act; I will not address this issue here. As Ippolito (2003, 2007) has observed, there is another type of conditional clause that is more strongly counterfactual than the subjunctive conditionals considered thus far; this type of subjunctive uses the past perfect (or subjunctive perfect) to convey a temporal semantics that seems to involve no past tense semantics. This can be used in reference to a counterfactual situation located at a future time, as in (18a); but it can also be used with a simple present-tense interpretation in reference to a counterfactual situation located at the time of utterance, as in (18b). (18)

a. b.

If John had left tomorrow instead of today, he would still have arrived in Paris in time for the meeting. If John had been with us on this trip, we wouldn’t be stranded now.

When the past perfect (or subjunctive perfect) is used in this way, the speaker must presuppose that the protasis is false; in other words, these conditionals are strongly counterfactual. Even if the morphological past tense affix –ed in (18) represents subjunctive mood rather than preterit past tense, so that these past perfects are really subjunctive

260

T. Stowell

perfects, this still leaves the use of the perfect and its semantic effect of strong counterfactuality unaccounted for. What semantic ingredient does the perfect in (18) add to what is conveyed by the finite subjunctive (–ed) that results in the overall effect of strong counterfactuality? I have suggested elsewhere, following Hoffman (1966), that the nonfinite perfect exhibits an ambiguity of temporal interpretation involving “sequence of tense” that is more usually associated with the finite morphological past tense suffix –ed. This arises in contexts such as the following, where the nonfinite perfect is temporally ambiguous, allowing either a past-shifted or “simultaneous” tense interpretation relative to the time of the main clause intensional predicate, as in (19a), just like the finite morphological past does in (19b): (19)

a. b.

Caesar believed his wife to have been in Rome at that time. Caesar believed that his wife was in Rome at that time.

If the nonfinite perfect can behave like the finite morphological past tense suffix –ed in exhibiting this kind of temporal ambiguity, perhaps it also exhibits the ambiguity we have seen in conditional clauses, between preterit past and subjunctive mood. If the perfect in (18) conveys subjunctive mood, then the subjunctive perfect here can be thought of as a kind of double subjunctive. Viewed from the perspective of Iatridou’s approach, these constructions can be seen as involving two degrees of remoteness from the actual world. But Ippolito (2004) has observed that the imperfect in Italian can be used to convey strong counterfactuality in conditionals in a similar way. In Italian imperfect conditionals, the imperfect verb form occurs in both the protasis and the consequent, as illustrated in (20): (20)

Se arrivavi prima, vedevi il film dall’inizio. if (you) arriveImperfect earlier, (you) seeIMP the movie from the beginning ≈ If you had arrived earlier, you would have seen the movie from the beginning.

The occurrence of the imperfect in both the protasis and the consequent clause here is reminiscent of the occurrence of the past tense suffix –ed in both clauses of English subjunctive conditionals (recall that would in the consequent clause of a subjunctive conditional is a fusion of woll and –ed.). The Italian imperfect presumably lacks the morpho-syntactic complexity of the English past perfect (or subjunctive perfect), and it is less plausible to assume that strong counterfactuality arises from a “double subjunctive” per se. I will therefore tentatively reject the hypothesis that the perfect in (18) conveys subjunctive mood. Ippolito proposes an account of the semantics of these imperfect conditionals that I will not discuss here in detail because of space limitations. Suffice it to say that one aspect of her idea is that the imperfect here has a kind of past-tense interpretation, but it does not locate the event conveyed by the main verb at a past time (or at any time); rather, it locates the evaluation of a modal inherent in the semantics conditional at a past time. This implies that the past-shifting tense conveyed by the imperfect takes scope over the entire conditional, even though the imperfect occurs syntactically


261

within the conditional. On her account, an inference of counterfactuality arises from the speaker’s choice to use this type of conditional-in-the-past instead of a conditional evaluated purely at the time of utterance. I have suggested elsewhere (Stowell, 1995b, 2007a) that the English morphological past tense affix –ed never directly expresses a past-shifting semantics; rather, it is a “past polarity item” licensed by a covert past-shifting formative (a silent past tense, if you like) located in a higher syntactic position in the clause. When –ed occurs in a main clause, it signals the presence of a silent past tense in the main clause, but when it occurs in a subordinate clause, as in (19b), the silent past tense that licenses it can in principle be located either in the subordinate clause or in the main clause. In the former case, the so-called “past-shifted” interpretation arises; in the latter case, the so-called “simultaneous” interpretation arises, since there is no covert past tense in the subordinate clause. I have also suggested (Stowell, 2007b) that the nonfinite perfect in (19a) is a past polarity item, and that this accounts for the parallel ambiguity between “past shifted” and “simultaneous” interpretations that it exhibits. In the same way, the imperfect in languages such as Italian should be treated as a “past polarity item”; in main clauses, it conveys a past-shifting tense, but when it occurs in complement clauses analogous to that in (19b), it allows a simultaneous reading, since the imperfect can be licensed by a past tense in the main clause. In the case of Ippolito’s strongly counterfactual imperfect conditionals, the imperfect occurs syntactically within the protasis (and the consequent) of the conditional, but the silent past tense that licenses it is located outside the conditional, taking scope over the entire conditional. Ippolito approaches the interpretation of the past perfect (subjunctive perfect) in (18) in a similar way. If the subjunctive mood conveys no more in (18) than what it conveys in (17), regulated by the usage conditions in (16), then perhaps the silent wide-scope past tense above the conditional in (18) is signaled by the use of the nonfinite perfect, as in sequence of tense contexts such as that in (19a). Viewed from this perspective, the contrast between the past perfect in the strongly counterfactual subjunctive conditionals in (18) and the past perfect in the “less vivid” subjunctive conditionals in (15d) and (17b) is analogous to the contrast between the “simultaneous” and “past shifted” interpretations of (19b). In both types of subjunctive conditional containing a past perfect in the protasis, the perfect signals the presence of a covert past tense licensing it. If the covert past tense is located within the protasis, the past tense locates the event or situation in the protasis in the past, but if the covert past tense is located outside the conditional, the conditional as a whole is strongly counterfactual, but the protasis clause is tenseless, accounting for its apparent present-tense or future interpretation. The only remaining question is why English uses the past perfect (or rather, the subjunctive perfect) for this type of strongly counterfactual conditional, rather than the usual English counterpart to the imperfect—the simple morphological past tense affix –ed. The answer, I suggest, may lie in the fact that English generally lacks a subjunctive that is morphologically distinct from the indicative preterit past. The simple morphological past in the protasis of a conditional can be understood to convey either subjunctive mood or past tense, but not both, so the perfect is used instead

262

T. Stowell

to unambiguously convey the presence of a past-shifting tense somewhere in the structure (in this case, above the conditional). The subjunctive perfect, rather than the indicative present perfect, must be used, because the usage condition on indicative conditionals in (15a) is inconsistent with the counterfactual inference triggered by the past tense conditional. There is another logical possibility here, still involving the idea that these forms are subjunctive, namely to assume that the past-shifting semantic tense is encoded here by the morphological past, and that subjunctive mood is encoded here by the perfect. But this would complicate the morpho-syntactic analysis of both the morphological past and the perfect, and the hierarchical syntactic relation between them, and I see no compelling reason to adopt this complication. Let us now return to K2. The subjunctive perfect in (18) corresponds to K2 in the informal register: (21)

a. b.

If John had of left tomorrow instead of today, he would still have arrived in Paris in time for the meeting. If John had of been with us on this trip, we wouldn’t be stranded now.

If the morphological past tense in the protasis conveys subjunctive mood, and the perfect functions as a past polarity item signaling the presence of a past tense above the conditional, then what does the particle of /have convey? I suggested at the beginning of this section that the contrast within the informal register between the past perfect and K2 involves an indicative/subjunctive alternation. This might be taken to indicate that of /have encodes subjunctive mood. But subjunctive mood is already encoded by the morphological finite past tense in these conditionals. If this is so, then what does of /have encode? I suggest that it encodes a kind of harmonic subjunctive mood marking on the perfect itself. On this view, the informal register exhibits two forms of the perfect, one of which occurs in indicative clauses (have + -en), and the other of which occurs only in subjunctive clauses (have + of /have + -en). In this sense, we can think of of /have as a kind of subjunctive polarity item. A possible problem with Ippolito’s account of the Italian imperfect and the English subjunctive perfect is that the English subjunctive perfect (K2 in the informal register) may occur with strongly counterfactual semantics not only in conditional clauses but also in complements to wish-class intensional predicates. Here the subjunctive perfect and K2 can be used in reference to counterfactual events or situations that are not located in the past, as in (22a–d), similar to what happens in (18). (22)

a. b. c. d.

I wish John had been with us now. I wish John had left tomorrow instead of today. I wish John had of been with us now. I wish John had of left tomorrow instead of today.

On the other hand, it can also be used in reference to counterfactual events or situations that are located in the past, as in (23): (23)

a. b.

I wish John had (of) been with us last week. I wish John had (of) left last week.


263

Given (22), it is unclear whether the subjunctive perfect plays any role in locating the events/situations in the past in (23), or whether all of the wish-complements in (22–23) are tenseless from a semantic point of view. If strong counterfactuality arises in subjunctive perfect (and K2) conditionals because the perfect signals the presence of a covert past tense scoping above the conditional, what kind of structural counterpart is available in (22), where no conditional is present for the past tense to scope over? It is plausible to suppose that a covert modal operator occurs within the subjunctive complement to the wishpredicate that functions similarly to the conditional modal in terms of its scopal interaction with the covert past tense that licenses the perfect in (22–23). It should be noted in this context that there is a semantic effect associated with the subjunctive perfect conditionals in (18) and (21) that is also in play with the examples in (22). In (18a) and (21a), the use of the subjunctive perfect in reference to a counterfactual event located in the future is natural only when the counterfactual future event is being contrasted, explicitly or implicitly, with an alternative actual event that took place in the past. Thus, (21a), repeated here— (21)

a.

If John had of left tomorrow instead of yesterday, he would still have arrived in Paris in time for the meeting.

—contrasts with (24a), which is anomalous, or ungrammatical; the same contrast can be observed with wish-complements; thus, (22b–d) contrasts with the anomalous (24b): (24)

a. ?If John had (of) left tomorrow instead of next week, he would have arrived on time. b. ?I wish John had (of) left tomorrow instead of next week.

Although the contrast between (18a/21a) and (24) involves the temporal location of the contrasting actual event, the critical factor turns out to be somewhat more abstract. This can be seen by the fact that the examples in (25) are perfectly well formed: (25)

a. b.

If John had (of) been with us now instead of Sam, we would have been in a stronger position. I wish John had (of) been with us now instead of Sam.

Even though the examples in (25) do not involve actual contrasting alternatives located in the past, they behave like the examples in (18a/21a) rather than like the examples in (24). The reason for this is that in (25), as in (18a/21a), the time at which the counterfactual event or situation ceased to be a possibility is located in the past. Ippolito (2003, 2007) proposes an account of this kind of effect (involving different types of examples) embedded within her theory of conditional modality and its interaction with tense; I will not pursue this point here.

264

T. Stowell

4 Counterfactual Subjunctives Versus Other Types of Subjunctives Let us assume, then, that the English morphological past tense suffix –ed conveys subjunctive mood in conditionals and wish-complements, and that K2 in these contexts is (unambiguously) a subjunctive perfect. But these are not the only kinds of subjunctive forms in English. It is common to describe the verbs in the subordinate clauses in (26), as being subjunctive: (26)

a. b.

We demand that he leave immediately. We insist that they be hired.

This type of subjunctive is morphologically identical to the uninflected (nonfinite) root form of the verb. It is characteristic of American English; British English typically uses indicative or modal verb forms in these contexts instead: (27)

a. b.

We demand that he leaves immediately. We insist that they should be hired.

Leaving aside the British/American contrast, the question arises how to characterize the difference between the type of subjunctive that occurs in contexts like (26) and the type that occurs in subjunctive conditionals and wish-complements. These forms are not interchangeable, at least in unaffected Modern English: (28)

a. b. c.

If I were/was/∗ be a terrorist, I would keep a low profile. I wish John were/was/∗ be here. We demand that they be/∗ were/∗ was hired.

(29)

a. b. c. d.

If John left/∗ leave tomorrow, he would arrive in Paris on Friday. I wish John lived/∗live nearby. If John had (of) left/∗ leave/∗ have left tomorrow instead of today, . . . I wish John had (of) left/∗ leave/∗ have left tomorrow instead of today, . . .

Should both forms be described as subjunctive? If so, how should we distinguish them from each other? At this point I will adopt the following terminological distinction: I will refer to the subjunctives in (26) as mandative subjunctives, and (at the risk of being misleading) I will refer to the subjunctives in conditionals and wishcomplements as counterfactual subjunctives. Note that counterfactual subjunctives include both subjunctive perfects (including K2) and non-perfect subjunctives encoded by −ed (including both was and were in the case of the copula). Both types of subjunctives correspond to subjunctive forms in at least some other languages. The counterfactual subjunctive corresponds closely to the German Konjunktiv II, among others, while the mandative subjunctive corresponds to the subjunctive in languages such as French and Spanish. If both types of verb forms are subjunctive, what property or properties do they share, and what distinguishes between them?


265

Guided in part by the spirit of Iatridou’s project of providing a unified semantics for past tense and what I am now calling the counterfactual subjunctive, it is tempting to describe the grammatical contrast between the mandative and counterfactual subjunctives in terms of tense, analyzing mandative subjunctives as present-tense subjunctives, and counterfactual subjunctives as past-tense subjunctives. One reason why this is appealing is that the mandative subjunctive often corresponds to a present subjunctive or infinitive in other languages, whereas the counterfactual subjunctive often involves a morphological element (such as English –ed) that is also used to convey simple preterit past tense. But we have already seen that the use of the morphological past tense suffix –ed in weakly counterfactual subjunctives is not generally associated with past-tense semantics. Only if the perfect is also present, in the subjunctive perfect (including K2), is semantic past tense involved, suggesting that, when –ed conveys subjunctive mood, the past tense is conveyed by the perfect rather than by –ed. Since the critical morpho-syntactic element distinguishing counterfactual subjunctives in general from mandative subjunctives is the presence of –ed, it is misleading to characterize the difference between the two types of subjunctive in terms of tense. When we look outside of English, the number of different types of subjunctives expands further; in many languages, subjunctive clauses can be used in contexts where neither the English counterfactual subjunctive forms (including K2) nor the English mandative subjunctive forms can be used. For example, in French, the subjunctive is licensed under the domain of negation, but in English, neither the counterfactual subjunctive (including K2) nor the mandative subjunctive is licensed in this environment: (30)

a. b. b. b.

∗ Nobody

believed that he were/be here. said that he live here. ∗ Nobody said that you had of left. ∗ John didn’t think he had of invited you. ∗ Nobody

Furthermore, in some languages, such as Spanish, subjunctives contrast with indicatives when they occur in relative clauses, conveying an intensional, or de dicto, sense, in contrast to the de re interpretation of indicative relatives. Nothing of the sort is possible with the English counterfactual subjunctives, including K2: (31)

a. b.

Max said that he would marry a woman who was (∗ were) intelligent. Max (would have) said that he would only marry a woman who had (∗ of) earned a lot of money by the time she reached the age of 30.

Summarizing the main point of this section, K2 does behave like a subjunctive, but it is a particular subtype of subjunctive (the counterfactual subjunctive), whose syntactic distribution is different from that of the mandative subjunctive, and severely limited in comparison with subjunctives in many other languages.

266

T. Stowell

5 Further Aspects of English Counterfactual Subjunctives and Sequence of Tense In the previous section, I argued that the counterfactual subjunctive should not be analyzed as a past tense counterpart of the mandative (“present”) subjunctive, since the distinguishing property of the counterfactual subjunctive is the presence of –ed, which is used in these contexts to convey counterfactual subjunctive mood rather than semantic past tense per se. But in Section 3, I argued that, when the subjunctive perfect (including K2 in the informal register) is used in counterfactual conditionals and wish-complements, –ed conveys subjunctive mood, and the perfect signals the presence of a covert past tense. Thus, there is a kind of past subjunctive in English, but this past subjunctive is not the counterfactual subjunctive in general, but rather the subjunctive perfect (including K2) in particular, insofar as it is only the subjunctive perfect that conveys both subjunctive mood and past tense. If the subjunctive perfect is a past subjunctive, its present tense counterpart is the non-perfect counterfactual subjunctive (-ed), as in (28a, b) and (29a, b), and not the type of subjunctive that shows up in mandative contexts. This can be shown in two ways. The first involves cases where the mandative complement has a past tense interpretation. The semantics of mandative predicates precludes an indexical past-tense interpretation in the complement, since they involve demands or requests involving future events or situations. Nevertheless a relative past-shifted interpretation is marginally possible in examples like (32): (32)

a. b.

I demand that he have left the party before I arrive. I ask that you have written up your assignments by lunchtime.

To my ear, (32a, b) sound awkward, and perhaps only marginally grammatical, for reasons that I do not understand. In any case, they are far better than (33a, b), with the finite subjunctive perfect (including K2): (33)

a. b.

∗I ∗I

demand that he had (of) left the party before I arrive. ask that you had (of) written up your assignments by lunchtime.

Thus, there are actually two distinct types of subjunctive perfects: counterfactual subjunctive perfects (K2 in the informal register) and mandative subjunctive perfects like those in (32). Another way of showing that the counterfactual subjunctive perfect is not a past-tense counterpart to the present-tense mandative subjunctive concerns environments where we might expect sequence-of-tense effects to arise. In particular, the counterfactual subjunctive perfect (including K2) cannot occur in the subjunctive complement of a mandative matrix verb, even if the matrix clause contains past tense, as in (34a, b), and even if the matrix clause is itself the subjunctive protasis of a counterfactual conditional, as in (34c): (34)

a. b.

∗ We

∗ We

demanded that he had (of) left immediately. requested that she had (of) been hired.


c.

267

∗ If they had (of) demanded that you had (of) left, we would have objected.

In these environments, the present subjunctive always shows up instead: (35)

a. b. c.

We demanded that he leave immediately. We requested that she be hired. If they had (of) demanded that you leave, we would have objected.

This would, prima facie, be surprising, if counterfactual subjunctives were (morphologically) past tense variants of mandative subjunctives, since English exhibits sequence of tense, and Spanish exhibits a subjunctive counterpart to indicative sequence of tense in this kind of environment. Interestingly, the mandative subjunctive perfect is also blocked here, at least on the relevant “simultaneous” sequence-of-tense interpretation: (36)

a. b. c.

∗ We

demanded that he have left immediately. requested that she have been hired. ∗ If they had (of) demanded that you have left, we would have objected. ∗ We

Apparently, the perfect here cannot be licensed by a main clause past tense external to the mandative subjunctive complement; its licensing is strictly local. A related point concerns the locality of subjunctive mood licensing. Quer (1998) contrasts subjunctives licensed by lexical verbs like ‘want’ with those licensed by operators such as negation, noting that whereas negation-licensed subjunctive mood may recur in complements of attitude verbs embedded within negative main clauses, lexically licensed subjunctives are strictly local, and occur only in the immediate clausal complement of the matrix licensing verb. He conjectures that the difference in locality is determined by the nature of the element licensing the subjunctive: lexical licensors select subjunctive mood only locally, whereas licensors that are logical operators (such as negation) license subjunctive non-locally. With wish-complements, counterfactual subjunctive mood licensing is local, as Quer’s account leads us to expect; this is true of both perfect and nonperfect forms: (37)

a. b.

∗I ∗I

wish John had of thought that Bill had of left. wish John were of the opinion that Bill were a nice guy.

Interestingly, however, subjunctive conditionals also involve strictly local licensing of subjunctive mood, contrary to the expectations of Quer’s account (assuming that conditionals involve operator licensing rather than lexical licensing): (38)

a. b. c.

∗ If

Sam had of claimed (that Bill had of believed) that Sue had of been dishonest. . . ∗ If you had of been told by someone that I had of been a liar, would you have defended me? ∗ If Sam were to say that Sue were ugly, I would clobber him.

This suggests that a factor other than the distinction between lexical verbs and logical operators is at work here. A plausible alternative is that local subjunctive licensing

268

T. Stowell

generally involves a modal operator, which is plausibly present in the complements of all lexical verbs selecting subjunctive complements, including deontic necessity modals selected by mandative verbs of the demand/ask class, bouletic modals selected by want-class verbs, and the type of modal that shows up in counterfactual environments, including both the complements of wish-predicates and subjunctive conditionals. Although counterfactual subjunctive mood is not licensed non-locally in (38), conventional sequence-of-tense marking may occur in these same contexts: (39)

a. b. c.

If Max had of claimed that Bill believed that Sue was dishonest, . . . If Sam had of claimed that Sue was ugly, I would clobber him. If Sam were told by someone that I was a liar, would you defend me?

(40)

a.

I wish he had of thought that I was as friendly as other people think I am. I wish Sam had of believed that he was intelligent. I wish he had of been of the opinion that I was his friend.

b. c.

To my ear, these examples allow simultaneous (SOT) interpretations of the boldface copular verbs only when the K2 in the protasis has a past tense interpretation, suggesting it is the past-shifting tense licensed by the perfect in the K2 that licenses this. With respect to these strict locality effects, the past perfect in the formal register behaves like K2 in being ungrammatical in non-local licensing environments when it conveys simultaneity, presumably confirming our claim that, whenever the past perfect is used to convey counterfacuality in conditionals and complements to wish/rather, it is in its subjunctive form.

6 Inversion in Conditionals English allows inversion of certain auxiliary verbs in the protases of subjunctive conditionals, in which case the inverted auxiliary verb precludes the occurrence of if, as is well known: (41)

a. b.

If John were to rejoin the team, we would win. If John had not left early, we might have won.

(42)

a. b.

Were John to rejoin the team, we would win. Had John not left early, we might have won.

The English present subjunctive is now archaic in conditional contexts, but it likewise allowed inversion: (43)

a. b.

He is happiest, be he king or peasant, who finds peace in his home. (translation of Johann Wolfgang von Goethe) Be he ne’er so vile, This day shall gentle his condition. (Shakespeare, Henry V)


269

Actually, were and had are the only auxiliary verbs (except for should, in some dialects) that can undergo inversion in conditionals. Even the simple past form was (which, I suggested above, is ambiguous between indicative past and subjunctive in most informal registers and in dialects that lack subjunctive were) does not allow it: (44)

a. b.

If John was to rejoin the team, we might win. If John didn’t belong to the team, we would be having problems.

(45)

a. b.

∗ Was ∗ Did

John to rejoin the team, we might win. John not belong to the team, we would be having problems.

The contrast between (42a) and (45a) might indicate that was in (45a), unlike were in (42a), is not a true subjunctive, contrary to what I claimed in Section 2. One way that we might try to distinguish between were and was, while preserving the central insight of Section 2, would be to exploit the notion of a “subjunctive polarity item” invoked above in reference to of/have in K2. Specifically, one might claim that, whereas were is a true subjunctive mood marker with modal force, was is a subjunctive polarity item that must be licensed by a (covert) subjunctive modal. The idea, then, would be that whereas the true subjunctive were can displace if in conditional inversion constructions, the polarity item was disallows this option, since in the inversion construction the subjunctive polarity item would be raised above its covert subjunctive licensor. K2 contains subjunctive had, and it freely allows inversion; the following examples are drawn from naturalistic data on the internet: (46)

a. b. c.

d. e. f.

Had you not of posted about my “complaint”, then you wouldn’t have hijacked this thread with me. Yes Anna i do love you and doubt i could have survived the year at Kingston had you not of been there. Respecting your houses Sir, they will shorely be built agreeable to your directions, and would have been had I not have heard from you at all as I had. . . (Archive of George Washington’s writings) I would also state that they would have been removed sooner had I not have been interfered with by unprincipled white men. He could have saved his life had he only of given her what she wanted. I think God, were He to exist, would be a “cat person”.

These facts are, of course, consistent with the view that the –ed affix in K2 conveys true subjunctive mood.

7 A Variant Form of the Subjunctive The modal perfect form of the conditional would (would have or would of ) occurs in the same environments as K2 in some English dialects; naturalistic examples from the web are provided in (47). A possible parallel with German K2 arises here too,

270

T. Stowell

since German often prefers to use the subjunctive form würde +infinitive instead of the normal K2. (47)

a. b. c. d. e. f. g.

If you would of discussed this issue with him, I’m sure he would justified his actions. If you would of printed the entire context things would of looked much different. If he would of got his personal problems solved I would of loved to see the guy at Wrestlemania 12. If he would of wrestled today, he would be more popular than Stone Cold or The Rock. Don’t worry about the things you wish you would have brought. I just wish there would of been better games to take advantage of the power. Deep in my heart I wish there would of been a way to tell him that.

This form behaves like K2 in many respects, and in fact is homophonous with K2 when the finite auxiliary verb (had or would) is phonologically reduced as a clitic attached to a preceding subject pronoun (or, marginally, of a vowel-final name or noun): (48)

a. b. c. d. e.

I wish I’d of been there. If there’d of been more people at the meeting, we’d of had a quorum. ∗ I wish John’d of left. ?If Lee’d of been at the party, she’d of had a good time. ?I wish Sue’d of told me that she was angry.

Like K2, these conditional forms do not occur in mandative subjunctive contexts, and seem to be restricted in terms of locality (on the relevant reading of simultaneity): (49)

a. b. c.

∗ We

demanded that he would of left. you would of believed that I would of been your friend, . . . ∗ I wish you would told me that you would of been angry at me. ∗ If

I will refer to this form as the conditional K2. The existence of this variant form raises the question whether the English K2 should simply be treated as a variant form of the conditional. This is doubtful, however, since the (non-conditional) K2 is completely excluded from the consequent clause of a conditional: (50)

a. b.

∗ If

he had of got his personal problems solved I had of loved to see the guy at Wrestlemania 12. ∗ If you had of used a trust, you had of avoided probate.

I will not pursue a detailed investigation of the internal syntax of these forms here. Nevertheless the apparent parallel between this construction non-standard English dialects and its counterpart in Standard German is striking, and perhaps indicative of a shared origin.


271

8 Conclusion In this paper I have explored the syntax and semantics of the English K2 construction within the context of a broader investigation of subjunctive verb forms in the language. I have argued that the K2 should be analyzed as a subjunctive perfect form, where the subjunctive is conveyed by the affix –ed, and the perfect functions as a past polarity item signaling the presence of a past tense located either in the same clause or higher in the tree. The type of subjunctive mood that occurs in this construction (which I have referred to, somewhat misleadingly, as the counterfactual subjunctive) must be distinguished from the mandative subjunctive mood that occurs in the complements of demand/ask class verbs. Both types of subjunctive are licensed strictly locally, in contrast to the subjunctive mood licensed by negation in languages such as French; I have suggested that this is related to the modal force of the subjunctive in these contexts. The particle of /have is a subjunctive polarity item, disambiguating the subjunctive perfect from the indicative past perfect.

References Abusch, Dorit (1988). Sequence of tense, intensionality and scope. WCCFL 7, Stanford University. Hoffman, T. Ronald (1966). Past Tense Replacement and the Modal System. In A. Oettinger (ed.), Mathematical Linguistics and Automatic Translation. Harvard University, Harvard Computational Laboratory, Cambridge, MA, VII-1-21. Reprinted in McCawley (1976), Notes from the Linguistic Underground. (Syntax and Semantics, 7.) New York: Academic, pp. 85–100. Iatridou, Sabine (2000). The Grammatical Ingredients of Counterfactuality. Linguistic Inquiry 31, 231–270. Ippolito, Michela (2003). Presuppositions and Implicatures in Counterfactuals. Natural Language Semantics 11, 145–186. Ippolito, Michela (2004) Imperfect Modality. In Jacqueline Guéron and Jacqueline Lecarme (eds.), The Syntax of Time. MIT Press, Cambridge, MA. Ippolito, Michela (2007). Semantic Composition and Presupposition Projection in Subjunctive Conditionals. Linguistics and Philosophy 29, 631–672. Karttunen, Lauri, and Stanley Peters (1979). Conventional Implicature. In Choon Kyu Oh and David A. Dinneen (eds.), Syntax and Semantics 11: Presupposition, Academic, New York, 1–57. Kayne, Richard (1997). The English Complementizer of. The Journal of Comparative Germanic Linguistics 1, 43–54. Quer, Joseph (1998). Mood at the Interface. Holland Academic Graphics, The Hague. Stowell, Tim (1995a). The phrase structure of tense. In Johan Rooryck and Laurie Zaring (eds.), Phrase Structure and the Lexicon, Dordrecht: Kluwer 277–91. Stowell, Tim (1995b). What is the meaning of the present and past tenses? In Pier Marco Bertinetto, Valeria Bianchi and Mario Squartini (eds.), Temporal Reference:

272

T. Stowell

Aspect and Actionality Vol. 1: Semantic and Syntactic Perspectives. Rosenberg & Sellier, Torino, pp. 381–396. Stowell, Tim (2007a). The Syntactic Expression of Tense. Lingua 117, pp. 437–463. Stowell, Tim (2007b). Sequence of Perfect, in Louis de Saussure, Jacques Moeschler and Genoveva Puskás (eds.), Recent advances in the syntax and semantics of tense, mood and aspect, in the series Trends in Linguistics, Mouton De Gruyter, Dordrecht, The Netherlands.

Phasing in Modals: Phases and the Epistemic/Root Distinction Karen Zagona

Abstract Recent work on modals has shown that cross-linguistically, root readings of modal verbs differ from epistemic readings with respect to: (a) subject/non-subject orientation; (b) interaction with finite tense; and (c) effects of perfective aspect on veridicality. This paper considers how the division of clauses into phases may shed light on these properties of modals. In languages whose epistemic and root modals have different distribution, the differences in reading can be derived from the modal’s interpretive relationship to the phase in which the modal is merged. However, for the finite modals of English, which are arguably all Tense items, it is proposed that the divergent properties noted above can be traced to features of the modal, which in turn affects the grammatical relationship between the modal and other clausal constituents. The crucial distinction is proposed to be (un)interpretability of the tense feature of the modal. This in turn affects interpreted temporal location, predication and interaction with aspect.

Key words: Root modal, epistemic modal, modal evaluation time, perfective aspect, phases.

1 Introduction It has been shown in recent literature that epistemic and root modals differ in how they interact with clausal tense. Root modals are within the scope of tense; that is, root modals can be located in the past, present or future time according to the tense of Aspects of this paper were presented at the Tense, Mood and Modality Colloquium, in December 2005 and at the Colloquium on Generative Grammar, Universidad Autónoma, Madrid, April 2006. I am grateful to audiences at those conferences, and to Jacqueline Guéron and Jacqueline Lecarme for comments.

University of Washington


273

274

K. Zagona

the clause. Epistemic modals, however, are outside the scope of the finite tense of the containing clause (Stowell, 2004; Boogaart, 2004; Borgonovo and Cummins, 2007). The question arises as to why this difference should occur, and in particular, whether it can be traced to syntactic properties of modals. A natural hypothesis to explore is that the difference is positional: epistemic modals occur above Tense and root modals below Tense. A complication for this approach is that the relative height of modals and tense would have to be derived by covert movement in some languages, since in some languages modals have the same syntactic distribution regardless of whether they have a root or epistemic reading. According to Palmer (2001), this situation is common cross-linguistically—the finite modals in English being a case in point.1 In the (pre-Minimalist) principles and parameters framework, a natural account of these facts would be via (covert) movement. One framework that provides a theoretical basis for such movement is the universal hierarchy of functional categories of Cinque (1999). Cinque proposes distinct functional heads in clause structure for a range of elements, including modals of different types. The hierarchy characterizes the position in which a given type of constituent is interpreted. However, the universal hierarchy is not sufficiently restrictive to account for scope interactions between modals and other clausal constituents, because, as Cinque shows, movement from specifier (and potentially head) positions in the hierarchy is possible (Zagona, 2007a). As a result, impermissible scope interactions are not excluded, and their impossibility has to be due to other factors.2 Minimialist assumptions (Chomsky, 1995) militate against the idea that the relative scope modals and tense can be represented as a structural difference. Since there is no covert movement, the relevant structural distinction should be apparent on the basis of surface position. Another consideration is that a movement analysis would have to make crucial use of head movement, a process that may be outside narrow syntax, since it violates the extension condition. The phase-based analysis of clauses (Chomsky, 2001, 2006) suggests a different means of differentiating between the two classes of modals. Clauses consist of two phases: v∗ P (a phase that corresponds to the full argument structure of the verb) and CP, which includes v∗ P together with Tense and the functional projections of the “left periphery”. The v∗ P phase is constructed first; its features are valued, uninterpretable features are deleted, and the phase is spelled out. Once a phase is spelled out, only features on the phase edge are accessible to probe-goal relations in the next phase. Next v∗ P is merged with Tense and the phase head C, again features are valued; uninterpretable features are deleted, and the CP phase is spelled out. Given the division of clauses into phases, a natural line of investigation is the idea that some differences in the syntax and semantics of modals follow from the 1

Some languages provide evidence for a structural distinction in where epistemic and root modals are merged, such as Catalan (Picallo, 1990) and Swedish (Platzack, 1979); in other cases both types of modals have the same distribution, as main (raising) verbs, as argued for French (Reed, 2005) or auxiliaries, as in English (Chomsky, 1975:230–232; Pollock, 1989). 2 There is a related theoretical issue that arises with respect to the universal hierarchy, as to whether it reflects the syntactic processes by which the relations are derived. It is possible that it reveals typological generalizations about clausal anatomy but not the mechanisms (syntactic operations) that underlie them. The syntactic relations that involve modals are therefore of primary interest.

Phasing in Modals: Phases and the Epistemic/Root Distinction

275

properties of the phase in which the modal is merged. Zagona (2007a) explores the hypothesis that root modals are interpreted in the v∗ P phase and epistemic modals in the CP phase. It is argued that modals can be merged in either phase, according to the inflectional features that are added to the lexical item as it enters the syntax. That analysis assumes that a modal could (in fact must) be merged in the phase with whose head it agrees. The two positions for modals are as shown in (1a) and (1b): (1)

Mary may sing. a. Root reading: merged in vP: C [TP Tense [vP may [vP Mary v [ sing ]]]] b. Epistemic reading: merged in TP: C [TP may [TP Tense [vP Mary v [ sing ]]]]

This analysis is viable for languages whose epistemic modals behave like auxiliaries and root modals like main verbs. It does not account for languages whose epistemic and root modals have the same syntactic distribution, as has been assumed for English since the inception of generative grammar—the head of TP (formerly INFL or AUX): (2)

[ C . . . [TENSE may] v∗ P ]

The goal of this article is to outline an analysis of epistemic/root modal syntax for languages like English in which only a single modal position is motivated: the position of Tense, as in (2). If in fact epistemic and root modals both occupy the same position, differences between the two subclasses should then be traceable to their probe-goal relations, which in turn define grammatical relationships between the modal and other constituents. The interpretation of the modal will be argued to vary according to its own features and those of constituents with which it agrees. The analysis builds on a traditional syntactic generalization regarding the root/epistemic distinction with respect to the modal’s relationship to the clausal subject. Root modals have been analyzed as “subject-oriented”, while epistemics are sometimes described as “speaker-oriented”. I will argue that epistemic modals are syntactically “situation-oriented”. That is, they take the propositional content of the v∗ P phase as their argument. This is similar in spirit to early analyses of epistemics as raising verbs. The discussion is organized as follows. Section 2 discusses properties of epistemic and root modals; Section 3 takes as a starting point a phase-based analysis of modals as in Zagona (2007a), according to which modals are merged in v∗ P or CP. The next issue addressed is how that analysis might be modified to accommodate a situation in which all modals are merged in the same position. It is proposed that, whether modals occupy a unique position in clauses or not, their crucial syntactic relationship is with v∗ P, and the root and epistemic readings differ according to the features that are added to the modal as it enters the derivation, which in turn affects feature valuing and phase interpretation. Section 4 proposes an implementation of this idea and discusses how the core temporal properties of the two types of readings follow from the differences in features and agreement.

276

K. Zagona

2 Epistemic and Root Modals After a preliminary notional description of the root/epistemic distinction, this section summarizes three respects in which epistemic and root modals differ: subject/nonsubject orientation (section 2.1), temporal location relative to the tense of the clause (section 2.2) and effects of perfective aspect, illustrated by Spanish and French (section 2.3). Epistemic modals are said to qualify the truth value of a sentence.3 The non-modal sentences in (3) make an unqualified assertion of the situations that they describe. The sentences in (4) do not; they express a subjective evaluation of factuality—that the situation is necessarily so or possibly so.4 (3)

a. b.

It is time for dinner. It is raining.

(4)

Epistemic: a. It may be time for dinner. b. It must be raining.

Root modals are often described in terms of their specific notional contribution to the sentence: permission, obligation, duty, ability or will. Root modals do not qualify the assertive force of the sentence, but instead qualify the situation itself. This is illustrated by comparison of the non-modal sentences in (5) and their counterparts in (6) with a root reading of the modal: (5)

a. b. c.

Terry ran that race in an hour. Kim is having dessert before she goes out. Fred sits in the front row.

(6)

a. b. c.

Terry should have run that race in an hour. Kim may have dessert before she goes out. Fred must sit in the front row.

The sentences in (5) make factual statements of the occurrence of an event in the past (5a) or the present (5b, 5c). The corresponding sentences with root modals in (6) make statements of occurrence of the modal condition: in (6a), a past obligation or expectation for Terry to run the race in an hour; (6b), a state of permission or ability of Kim to have dessert. The sentence is true if the state of permission or ability holds, regardless of the occurrence of dessert-eating. Likewise (6c) asserts a situation of externally induced obligation, not a situation of sitting.

3

Palmer includes as a single class both epistemic and evidential modality. Evidential modality expresses the source of knowledge about a proposition, for example whether an event was seen, heard, or reported by a third party. 4 The subjective evaluation implies that a stronger, objective evaluation isn’t possible, although this appears to be pragmatically conditioned. A subjective statement can also mean that the speaker just declines to provide the truth value, or declines to commit to knowledge of it.


277

2.1 Subject Versus Non-Subject Orientation Several properties of epistemic modals versus root modals have been shown to hold cross-linguistically. The first of these is that root modals are often described as “subject-oriented”. In (7a), the obligation or permission is associated with the doctor; in (7b), it is associated with John (examples from Cinque (1999:88): (7)

a. b.

The doctor may/must examine John. John may/must be examined by the doctor.

The term “subject orientation” suggests that the modal is predicated of the subject. In (7a) for example, the doctor can be described as having an ability or need to examine John; in (7b), John has the ability or need to be examined. This is consistent with the generalization that the root modal holds at the time specified by the tense of the clause.5 Recall from the discussion of the examples in (6) that a sentence with a root modal is true if the state (of obligation, permission) holds of the subject, regardless of whether the eventuality associated with the “main verb” and its arguments occurs or not. The modal is therefore the primary predicate of the sentence. Nevertheless, the “subject orientation” of the modal is not a thematic relation. The external argument is thematically related to v, while the modal imposes additional selectional restrictions on it. Modals of permission require an animate subject (John may leave soon/??The rocket may take off soon.) Root modals are generally impossible with non-referential subjects, as shown in (8a), (8b), although (8c) may have a root reading of external obligation: (8)

a. b. c.

(∗ It is expected.) It should rain hard in coastal areas. There may be a pie for dessert. (∗ It is permitted.) There should be a cloud on the horizon. (It is expected.)

A root reading of (8c) could be construed in a situation where a painting teacher is describing requirements for a landscape (imposed by some authorities).6 Root modals are generally incompatible with clausal subjects.

5 On deontic readings of root modals, there is understood to be another “participant” in the root modal state. According to Palmer (2001), deontic modality relates to obligation or permission emanating from an external source. For example in: ‘The doctor must examine John’, someone other than the doctor or John imposes the obligation on the doctor. It is not clear whether or not this additional participant is syntactically represented as an implicit argument. It may be that it is inferred from other information in the clause. Because some root modal readings (such as the ability reading, in: Sue can read Sanskrit) do not implicate an external source, I will assume that the external source is not central to the licensing of root readings. 6 Periphrastic root modals are also sometimes possible with non-argument subjects:

(i)

There has to be a cloud on the horizon; I require it;

(ii)

∗

It is able to rain hard in Seattle.

278

(9)

K. Zagona

(Root readings intended:) a. It ∗ may/∗ can/(∗ )should/∗ must be annoying that the car won’t start. b. It ∗ may/∗ can/?should/∗must be legal to jaywalk.

Summarizing to this point, root modals are “subject oriented” in that they introduce a state of the subject. This state is understood to hold at the time specified by tense. The modal imposes selectional restrictions on the subject, but does have a thematic relation to it. Epistemic modals are sometimes described as “speaker” oriented in that the modal qualifies the speaker’s subjective attitude toward the factuality of the proposition. In: Sue may be at home, the assertion of the proposition (Sue’s being at home) is qualified. However, speaker orientation does not always obtain. In embedded clauses, epistemic modals can be “oriented” toward an argument of the main clause, as in (10) and (11): (10)

It seems to Mary that it must be dinnertime.

(11)

John thought that it must be dinnertime.

In (10), the judgment is that of Mary, in (11), that of John.7 In general, then, epistemic modals describe a judgment of someone outside the modal clause.

2.2 Modal Evaluation Time Root modals are “located” in time by the tense of the clause. In a past tense, the modal state of the subject holds at past time; in a present tense clause, the modal state of the subject holds at the present time. English modals can’t and couldn’t in (12) show the temporal location of the modal on the root reading (examples from Stowell (2004))8 : (12)

a. b.

Carl can’t move his arm. (present state of ability) Carl couldn’t move his arm. (past state of ability)

7

Cinque (1999) notes that some languages appear to show a distinction between epistemic modality (speaker’s judgment) and ‘alethic’ modality, which refers to strictly logical possibilities (propositions that are true in at least one possible world) and logical necessities (propositions that are true in all possible worlds). There may also be instances in which the modal judgment is that of an unspecified individual or people in general, as in (i): (i)

It seems that the policy must be revised.

The necessity of revising the policy may be the view of the unspecified experiencer of seem. 8 Note that English modals like might and could are ambiguous between past and non-past. Sentences in (i) illustrate could with a present (root) reading: (i)

How will I get home? You could walk.

Stowell’s examples crucially show that root modals differ from epistemic modals in terms of the availability of a past/non-past alternation fixed by clausal tense.


279

Epistemic modals are temporally located at the “evaluation time” of the clause— this is Speech time in main clauses, and in some subordinate clauses is the event of the matrix clause. Stowell (2004) shows that in main clauses, the evaluation time is the time of speech (‘now’). On an epistemic reading of (13), couldn’t is a present impossibility: (13)

Jack’s wife couldn’t be very rich.

(present impossibility/∗past impossibility)

In embedded clauses, an epistemic modal is evaluated at the time of evaluation of the clause: (14)

Caesar knew that his wife might be in Rome. (past state of possibility)

The time of the possibility might is past. Stowell shows that this is due to the past evaluation time of the clause, not to a [Past] in the embedded clause. According to Stowell’s argument, if the embedded clause in (14) had its own independent tense, the time of possibility should in principle precede the time of evaluation for the clause— the time of Caesar’s knowledge. As expected under Stowell’s analysis, if the main clause is changed to present tense, might is no longer a past (epistemic) possibility: (15)

Fred knows that his wife might be in Rome. (present state of possibility)

As is expected under Stowell’s generalization, complement clauses that have optional sequence of tense have past epistemic readings only if the main clause has a past tense verb, since only this context admits a sequence-of-tense analysis of the embedded clause. Factive verbs such as realize, please do not force sequence-of-tense on their complements: (16)

a. b.

Terry realized that Sue likes broccoli. It pleased Mary that dinner is being served now.

Given the availability of an independent tense in the complement clause, both epistemic modals should have “independent” (=Speech time) and “dependent” (main clause event time) evaluations. These readings occur, as shown in (17): (17)

a.

b.

Terry realized that Sue might like broccoli. (i) Terry realized that there was a possibility (then) of Sue’s liking broccoli. (ii) Terry realized that there is a possibility (at Speech time) of Sue’s liking broccoli. It pleased Mary that dinner could be being served. (i) It pleased Mary that there was (then) a possibility of dinner being served. (ii) It pleased Mary that there is (at Speech time) a possibility of dinner being served.

In both (17a) and (17b), the time of the possibility corresponds to the time of evaluation of the embedded clause: (i) on a dependent reading, the possibility holds at the

280

K. Zagona

time of the main clause event; (ii) on an independent reading, the possibility holds at Speech time. Crucially, the dependent reading disappears if the main clause is not a past tense: (18)

a. b.

Terry realizes that Sue might like broccoli. (∗ past possibility) It pleases Mary that dinner could be being served. (∗ past possibility)

In these sentences, the time of the embedded epistemic modal is Speech time only. Summarizing, root modals are temporally located at a past or non-past time specified by the finite tense of the clause, while epistemic modals are temporally located at the “time of evaluation” of the clause: in main clauses, this is usually Speech time; in embedded clauses it is the time of the main clause event. This generalization is supported by the fact that when the evaluation time of a clause is affected by some other factor, the time at which the modal holds is also affected.9 The discussion of modal evaluation time to this point has been concerned with differences in the temporal location of the two types of modals. Borgonovo and Cummins (2007) provide evidence that the temporal generalizations are not attributable to tense inflection of the modal. Spanish and French are languages with full inflectional paradigms on modals, yet epistemic and root readings differ in temporal reference (examples from Borgonovo and Cummins, 2007): (19)

a. b.

(20)

a. b.

Pedro deb´ıa estar en casa. P. must-Past-Imp be-INF at home. Pierre devait eˆ tre a` la maison. P. must-Past-Imp be-INF at home. ‘P. must have been at home.’ (epistemic) Pedro deb´ıa pagar la cuenta. P. must-Pastimperf. pay-INF the bill Pierre devait payer la facture. P. must-Pastimperf. pay-INF the bill ‘P. was supposed to pay the bill. (root)

In both (19) and (20), the modal verb has past tense inflection. On the epistemic reading (shown in (19)), the modal does not hold at a past time, but rather at Speech time.10 On the root reading (shown in (20)), the modal verb is understood as holding in the past. Epistemic modals such as the one in (19) then have a “mismatch” between their morphological form and their temporal reference: the modal has a past feature that applies only to its complement v∗ P, not to the modal itself. In (19), for example, the situation of Pedro’s being at home is past. 9

In connected discourse there is also variation in the temporal location of main clause epistemics, as shown for Dutch in Boogaart (2004). Boogaart shows that in narratives with a sequence of past tense main clauses, an epistemic modal has a past evaluation time, that is, it is understood as a past possibility or necessity. 10 to be an epistemic reading of (19b) that could be paraphrased as ‘It was possible/probable that Pierre was at home (given the evidence available then)’. She points out that contexts are needed to capture differences in epistemic flavors. Since (19b) has a human subject, the salience of the ability reading may make judgments especially delicate.


281

The occurrence of “displaced” inflections is not unique to modals. Pollock (1989) makes a similar observation about expletive do that in English is inserted in negative and other contexts that require an auxiliary: (21)

John did not go.

In (21), the past event is go, although it is not-finite in form. In periphrastic progressives such as: was going, the verb form is likewise non-finite; its past time is given on the auxiliary. Although these cases differ in some ways from sequences with modals, they illustrate that tense morphology is not always interpreted on the verb that bears the inflection.

2.3 Aspect and Veridicality of the Event We turn now to the temporal properties of the main predication of the clause. It was shown above for both epistemic and root modals (see discussion of (3)–(6)) that the occurrence of the main predication is not established as a component of the evaluation of the sentences as a whole. That is, the situation is ‘averidical’: its truth value is not known. For epistemic modals this is related to the generalization that the evaluation itself is qualified by the modal; on root readings it is related to the modal’s qualification of the situation. The discussion below will show that perfective aspect also has a “qualifying” effect on the situation: in past perfective clauses, the relationship between a root modal and v∗ P is altered in such a way that the occurrence of the predication is established. Although readings vary cross-linguistically, the general pattern is that perfective aspect can have the effect of deriving either an implicative (entailed) or counterfactual reading. Many languages make a morphological distinction between perfective and imperfective aspect (Comrie, 1976; Smith, 1997). Perfective aspect is generally understood to mean that the perfective eventuality is evaluated as a whole, including its onset and its endpoint boundaries. A perfective past reading is illustrated in: Sue read the article, where the past time referred to includes the entire event of reading; in other words, the event ending is reached during the interval referred to in the sentence. Imperfective aspect does not necessarily include the endpoints of the event, although imperfective aspect gives rise to readings that sometimes do and sometimes do not include event boundaries (Zagona, 2007b). Following Smith (1997), progressive morphology is an instance of imperfective morphology; in John was reading the book, for example, a process of reading is understood to have occurred in the past, although the endpoint of the event is not part of the occurrence. Borgonovo and Cummins (2007) investigate in detail the interactions between modals and perfective and imperfective past tense in Spanish and French, languages that mark aspect morphologically. Spanish is illustrated below (examples from Borgonovo and Cummins, 2007):

282

(22)

(23)

K. Zagona

Juan pudo ganar la carrera. J. can-Prf win the race a. ‘Juan could have won the race.’ (He didn’t.) b. ‘Juan managed to win the race.’ Juan debió ganar la carrera. J. should-Prf win the race a. ‘Juan should have won the race. (He didn’t.) b. ‘Juan was forced to win the race.’

(root) (root)

(root) (root)

On the root reading of the modal, the event of winning the race can be counterfactual or is ‘implicative’ (entailed). This contrasts with the imperfective past, where the event is neither implicative nor counterfactual: (24)

(25)

Juan deb´ıa pagar la cuenta. Juan must-Imp pay the bill ‘Juan was obligated to pay the bill.’ Juan pod´ıa pagar la cuenta. Juan can-Imp pay the bill ‘Juan was able to pay the bill.’

Contrasting with the above pattern, on epistemic readings of the modal, the event is uniformly averidical. This is illustrated for poder ‘may’ in (26): (26)

a.

b.

Juan pudo ganar la carrera. Juan may-Perf win the race ‘It is possible that Juan won the race.’ Juan pod´ıa pagar la cuenta. Juan may-Imp pay the bill ‘It is possible that Juan paid/used to pay the bill.

Summarizing, perfective aspect affects the status of the event in sentences with root modals. On an epistemic reading, viewpoint aspect has no effects on veridicality: the event’s occurrence or non-occurrence is not known. On root readings, the event can be implicative or counterfactual. English does not show perfective effects when a finite (root) modal combines with a main verb. In (27) and (28), the event is neither counterfactual nor implicative, even if the modal is construed as a past ability or obligation: (27) (28)

a. b. a. b.

Terry could pay the bill. Sue said that Terry could pay the bill. Terry should pay the bill. Sue said that Terry should pay the bill.

(ability/permission) (past ability/permission) (obligation) (past obligation)

This is surprising, given that the English past tense is sometimes characterized as a (morphologically unmarked) perfective past. For certain modals, the addition of


283

perfect have produces the perfective counterfactual reading: (29)

a. b.

John should have paid the bill. (past obligation: counterfactual) John could have paid the bill. (past ability)

This does not extend to all modals. Might does not have a root reading (Stowell, 2004); the imperative-like root reading of must does not combine with have: (30)

a. b.

Fred must win the race (obligation) Fred must have won the race (∗ obligation)

Summarizing, in the context of a root modal, perfective aspect can affect the argumenttaking predicate of the clause, altering its status from averidical to counterfactual or presuppositional. This effect varies from language to language. Both implicative and counterfactual readings occur in the Spanish perfective past tense. According to Borgonovo and Cummins, French has only the implicative reading. English does not allow either reading in modal+main verb sequences; modals should and could show counterfactuality in combination with have. However they do not have implicative readings of the type illustrated for Spanish, with readings of ‘managed to’ was forced to’ in (22) and (23). Finally, both imperfective aspect and epistemic modals are immune to these effects.

3 Modals and Phases This section will lay the groundwork for a phase-based account of the epistemic/root contrasts summarized above in section 2. The discussion will take as a starting point the analysis on which the two types of modals are merged in different phases: root modals in the v∗ P phase, and epistemic modals in the CP phase (Zagona, 2007a). In section 3.1 the mechanics of this approach are outlined. It is claimed that the lexical items that give rise to the two readings are the same, and the modals differ in their inflectional features and in the phase heads by which they are licensed. The descriptive generalization that is captured by this analysis is that root modals are merged lower, within the domain of predicate-argument structure, and are an extended part of the situation predicated of the subject. Root modals modify or qualify that situation. Epistemic modals, being merged outside v∗ P, are constituents of the left periphery, and are licensed in relation to the ‘assertion’ features of C. In section 3.2, this approach is revised in such a way as to accommodate modals that occupy a single syntactic slot—the head of TP. This revision calls into question the generalization that epistemic modals are interpreted in relation to either v or C. It is argued that this generalization should be abandoned in any case, since it does not well characterize the epistemic modal as a predicate, which requires syntactic licensing in relation to some subject. It is proposed that epistemic and root readings of modals are due to the availability of two different agree relations between the modal, and constituents of v∗ P. Reviving one traditional view, the two types of modals can be differentiated in

284

K. Zagona

terms of whether they are predicated of DP (root readings), or of the proposition: v∗ P (epistemic readings).

3.1 Toward a Phase-Based Analysis: Lexical Features of Modals The idea to be explored here as a starting point is that root and epistemic modals are single lexical items that differ only in where they are merged, not in intrinsic semantic features. As discussed briefly in section 1, the position in which a modal is merged can vary with optional features that are added to the modal as it enters the syntax. This implies that the epistemic/root distinction can be traced to grammatical processes rather than intrinsic semantic differences. As a starting point for the discussion, consider an alternative position, on which the epistemic/root distinction is due to substantive features of the modal lexical items. If root and epistemic modals were semantically distinct roots, or homophones, it would be surprising to find the same instances of homophony in many languages, as is in fact found (Palmer, 2001:86ff.) Among the languages that use ‘homophonous’ forms, there is variation in whether the two readings are syntactically (distributionally and morphologically) identical. In English, the finite modals are analyzed as occupying a single position and sharing the same morphological defectivity. French and Spanish modals are non-distinct distributionally and inflectionally. Reed (2005) argues that French modals are “main verbs”, based on bi-clausal characteristics (double negation and double viewpoint aspect) that differentiate them from auxiliary and aspectual verbs. Similar considerations hold for Spanish (Borgonovo and Cummins, 2007). On the other hand, Catalan (Picallo, 1990) and Swedish (Platzack, 1979) show evidence of a distributional split, with epistemic modals merged outside v∗ P, and root modals internal to v∗ P. If these descriptive generalizations are right, then the epistemic/root distinction doesn’t derive from differences in category (or subcategory). If it did, only languages with such a syntactic distinction would be expected to display an epistemic/root ambiguity for the relevant forms, contrary to fact.11 Note that if lexical roots were actually ambiguous as illustrated below, (31)

a. b.

may1 may2

[possible] [epistemic] [possible] [deontic]

this would not provide an immediate explanation for the properties discussed in section 2. It is not clear why the two items should interact differently with tense and aspect. It is not obvious why an epistemic modal cannot be evaluated at a time specified as [+Past] by Tense, for example. These considerations suggest that the differences between readings of modals are due to syntactic factors. In Zagona (2007a) it is proposed that inflectional features are responsible for the merging of a modal in either the v∗ P phase or the CP phase. 11 English periphrastic modals (need to, have to; able to, etc.) are not systematically ambiguous between root and epistemic readings. Even considering these as well, it could not be said that the epistemic/root distinction correlates with the auxiliary/main predicate distinction.


285

Given a particular choice of inflectional features, the modal would be constrained as to where in clause structure it could be merged, which in turn determines the construal of the modal. The relevant inflectional feature was proposed to be [person]. This feature both permits the modal to be merged in the lower phase, which associates it with the propositional content of vP. In this context it also values features of Tense and is temporally located by Tense. A modal lacking a person feature could not be merged within vP, since without a person feature, the person feature of Tense could not be valued, leading to a crash. A modal that lacks a person feature could therefore only be merged above TP, and interpreted in relation to C. If merged at this higher level, the modal could only be interpreted in relation to the Force head. The semantic relation was proposed to be a modifier of the assertion feature of Force. These two configurations are shown in (33), corresponding to readings of the modal may in (32): (32)

Fred may eat the last cookie.

(33)

a. b.

[ vP v* [ may [ eat the last cookie ]]] (Agree) [ CP C [ may [ T . . . ]] (Agree)

The analysis thus claimed that inflectional features of the modal determine its position, which in turn affects its status as a predicate, its subject orientation and its temporal location relative to finite tense. A root modal is a modal that is low enough in the structure to be a secondary predicate, predicated of the external argument, and that can value features of v∗ and T, like any finite verb. An epistemic modal is a modal that lacks these features and therefore can’t occur in the v∗ P phase. If merged above Tense, it is part of the evaluation phase of the clause, predicated of the “assertion” feature—a verbal feature in C, which explains why the epistemic modal seems to be interpreted as an attitude of the speaker, since the speaker is the implicit participant in the assertion event.

3.2 Modals as Tense Morphemes The analysis outlined above differentiates root from epistemic readings on the basis of their position relative to phase heads, v∗ and C. Here, I will consider how the properties associated with the different readings might be derived if modals are merged in a single position as Tense morphemes, as in (34): (34)

a. b.

Fred may eat the last cookie. C [ [T may ] [v∗ P Fred v∗ eat the last cookie ]

The most immediate (and obvious) effect of the proposed structure in (34) is that root and epistemic readings can’t be differentiated in terms of the phase in which the modal is merged: Tense has an invariant structural relationship to both C and to v∗ P. Under this assumption it is necessary to abandon the generalization that subject orientation

286

K. Zagona

versus speaker orientation can be traced to a structural relationship between the modal and one phase head or the other. This is not a drawback of the assumed structure in (34), since there is reason to question the underlying empirical generalization. It was shown in section 2.1 above that in embedded clauses, epistemic modals sometimes lack speaker-orientation. Since it is not essential for epistemic construal, it may be that it is not expressed in syntactic representations at all. If the modal is not subject oriented, it may be that pragmatic considerations identify either another argument or the speaker as experiencing the epistemic judgment. The question remains as to how subject orientation arises in a structure like (34), and what conditions trigger it (root readings) or preclude it (epistemic readings). In the minimalist framework, Chomsky (2006) observes that there are two kinds of grammatical relations: set membership (given by Merge) and probe-goal relations. Since set membership is not at issue (Tense and its complement are a constituent regardless of the reading of the modal) we are driven to consider the possibility that root and epistemic readings follow from different probe-goal relations between Tense and its complement. To isolate these relations, let us start by distinguishing a derivation without any modal or other auxiliaries from a derivation with do or modals. The relevant distinction is whether the v∗ P phase contains a finite verb or not. The two cases are shown in (35)–(36): (35)

a. b.

Sue bought a pen. C Tense [ Sue v* [ bought DP ]]

(36)

a. b.

Sue did (not) buy a pen. C [T did ] (not) [ Sue v* [ buy DP ]]

c.

X C [T did (not) [ Sue v* [ buy DP ]]

In (35), the main verb (bought) has interpretable features: [+finite], [+past]. The features of the probe are valued and deleted, and the temporal reading of the v∗ P phase can be spelled out. In (36), the verb buy is a bare infinitive. It cannot value features of v∗ ; the phase head v∗ remains unvalued for its temporal interpretation. If not for the verb in T, the derivation would crash. The same situation arises if T contains a modal rather than do, as in (37): (37)

Fred may eat the last cookie.

The derivation of (37) is identical in relevant respects to (36): the modal “substitutes for” a main verb with respect to providing features that value features of v∗ : (38)

C [ t may ] [ Fred v* [ eat the last cookie ]]

Summarizing to this point, the probe-goal relationship between the modal and v∗ is not an idiosyncratic property of modals in tense. Non-modal do has the same relationship to v∗ .


287

Let us now turn to the contrast between root and epistemic readings. As discussed above, it should be assumed that a single lexical item gives rise to either meaning: both the lexical root and the category are identical. What differs is the features that are added to the modal as it enters the derivation. In the minimalist framework, features can be either ‘interpretable’ or ‘uninterpretable’. The former can survive the derivation; they are interpretable at the interface. Uninterpretable features must be valued and deleted. I propose that modals can have either valued or unvalued Tense features. Stated informally, this amounts to distinguishing between modals as fully ‘lexical’ elements and quasi-functional elements. The distinction is not due to category or to the particular inflectional features that the modal bears, but to the character of its features. It is expected that such ambiguity should be possible in the position of Tense, since this category is a semantically contentful one, but is also a dependent phase head, entering into agree relations that derive from C. The alternation between interpretable and uninterpretable features affects both the temporal evaluation of the modal, and in turn, syntactic predication (i.e., the subject/non-subject orientation of the modal). If the modal has interpretable tense features, a root reading results: (39)

C [ T may ] [ Fred v* [ eat the last cookie ]] [ -past] [ u past ]

In (39), the modal is analogous to V: it has an Interpretable [-Past] feature. This feature values the uninterpretable feature of v∗ , which is in turn deleted. The Interpretable [-Past] feature on the modal produces an interpretation of present location for the modal. Once the subject moves to Spec, TP, a predication relation also holds between the modal and the subject. Because the modal has interpretable semantic content (unlike do), it is not forced to delete, and is the semantic locus of predication also. Suppose now the modal has only an uninterpretable [-Past] feature. On this option, the modal cannot value the [uPast] feature of v∗ . (40)

[ eat the last cookie ]] C [ T may ] [ Fred v* [ u past ] [ u past ]

The only source of an interpretable temporal feature is C: the the “Speech time” of the clause: (41)

C NOW

[ T may ] [ Fred v* [ eat the last cookie ] [ u past ] [ u past ]

In (41), the external evaluation time of the clause, standardly associated with C, transmits its value to the modal and to the v∗ P phase. The modal, as well as the propositional content of v∗ P, is evaluated at Speechtime. The absence of an interpretable tense feature on the modal also affects the grammatical relation between the modal and DP, blocking an independent predication relation and therefore obviating the modal’s subject orientation. This is expected on the assumption that set

288

K. Zagona

membership and probe-goal relations are the (only) two bases of grammatical relations. The absence of an interpretable (valued) tense feature on the modal may block the modal from valuing Case of DP, so DP is not an active goal of the modal. The In this instance the modal is not predicated of DP. The modal does, however, formally agree with v∗ P, as shown in (40). This relation can give rise to predication, with the v∗ P proposition (or eventuality) functioning as the subject. This accords with an intuitive description of what is asserted: the proposition corresponding to v∗ P is asserted to be possible.12 On the root reading, the modal is predicated of DP; the predication relation is established via the probe-goal relation that values phi-features and Case. On the epistemic reading, the predication relation is established via the probe-goal relation triggered by C, which values Tense features of both the modal and v∗ . Summarizing, this section took as a starting point the hypothesis that epistemic and root modals differ in terms of which phase of a finite clause they are interpreted in. This distinction is problematic in two ways. First, for English it is inconsistent with the long-standing assumption that modals have a single distribution. Second, it incorrectly associates epistemic modals with the speaker, via interpretation of the modal as an adjunct of the assertion predicate of Force. An alternative was introduced, according to which English finite modals are categorially Tense morphemes, and like pleonastic do, value tense features of v∗ when the main verb is not inflected. The distinction between root and epistemic modals was proposed to derive from the type of features that are expressed on the modal: interpretable or uninterpretable. The choice of feature type is equivalent to analyzing the modal as V versus v: V has valued (interpretable) features; v has unvalued features. It determines whether the temporal feature of the modal deletes or remains as a feature of the interpretation at the interface. This is the proposed source of the temporal difference between epistemic and root modals. A modal whose features are interpretable is in relevant respects like a (displaced) main verb, and can be syntactically predicated of the subject. A modal whose features are uninterpretable is like a phase head; its features are deleted once valued. Since the v is not a finite predicate at the interface, it cannot be predicated of the Nominative subject. It is understood as predicated of the v∗ P proposition.

4 Perfectivity and Root Modals This section briefly addresses the effects of perfective aspect on the construal of the main verb with root modals. It was illustrated in section 2 that in the perfective past, Spanish has a counterfactual reading of the main predication, and (like French) has an implicative reading. The fact that these readings of the main predicate occur with root modals suggests that the reading is due to the interaction of interpretable features with 12

Platzack (1979) notes that this is essentially the claim of Kiparsky (1970), Perlmutter (1970), and Ross (1969), who treat epistemic modals as intransitive verbs with sentential subjects. At the time, this implied a blurring of the distinction between auxiliaries and main verbs, a consequence that is not necessary under present assumptions.


289

a perfective value for v∗ P. Below, an account of these readings will be sketched, and we then address why the same readings are not systematically possible in English. The effects perfective aspect are illustrated in (22), repeated below: (22)

Juan pudo ganar la carrera. J. can-Prf win the race a. ‘Juan could have won the race.’ (He didn’t.) b. ‘Juan managed to win the race.’

(root) (root)

One difference between the English structures discussed above in section 3 and the structure for sentences like (22) is that English modals are tense morphemes that merge with v∗ P, while the modal pudo ‘could’ in (22) is a main (raising) verb. For purposes of illustration however, we can abstract away from this difference, and focus only on features of the modal and those of its complement infinitival v∗ P. As a first step, we set aside for the moment the effects of perfective aspect, and consider the relationship between the modal and the v∗ P. As discussed above in section 3.2, the root reading of a modal is associated with an interpretable tense feature, which values the tense of its complement: (42)

Modal . . . [ DP v* V-Fin DP ] [+Past] [u Past]

Once the [uPast] feature of v∗ is valued, it is deleted. In (42), the v∗ P phase has no temporal location. The modal represents a past state of the subject; V ganar ‘win’ is non-finite. The v∗ P phase is spelled out as a “hypothetical situation”, since it is not temporally located. Perfective aspect adds to v∗ P an additional finite feature associated with the endpoint of the time interval licensed by TP. However, this finite relation is dependent on v, in the same way that the finite feature of T is dependent on C. The additional proposed structure is shown in (43): (43)

Modal . . . [ DP v* Fin V-Fin DP ] [+Past] [u Past] [u Past]

Both v∗ and the Fin head are valued by the modal. In (43), the infinitival main verb cannot value the [uPast] feature of v∗ , since it is not specified for [±Past]: perfective aspect is a dependent probe, like Tense. Here, the root modal has interpretable features, and it values both heads. Consider now the effects on veridicality. In (43) the v∗ P phase has a temporal location, unlike the structure discussed above for (42). This is due to the Finite head, which inherits a temporal value from Modal+v. The v∗ P is implicative; it is a past situation. The secondary finite head in a perfective clause has the effect of focusing on the end-phase of a situation. The implicative reading says that the endpoint is reached, so the modal and the entire sequence of subevents of v∗ P are asserted as a whole: the modal and infinitive are reanalyzed as subevents of a complex situation whose endpoint is located in the past. Both epistemic modals and imperfective aspects block the relevant restructuring. Epistemic modality blocks the implicative because it lacks the

290

K. Zagona

interpretable feature to value a perfective head. Recall that the ‘Now’ time value of C values v∗ in clauses with epistemic modals. The ‘Now’ value of C is incompatible with perfectivity, which calls for [+Past] to achieve an endpoint reading. Imperfective aspect blocks the implicative reading because it lacks the secondary finite head within v∗ P. The counterfactual reading seems to involve a secondary imperfectivization— that is, a suspension of the realization of the endpoint of the event, as in progressives. The absence of these readings in English could have several causes, since English lacks morphologically distinguished perfective aspect, and restructuring, and may lack person feature agreement with the raised DP in past tenses. The analysis proposed here suggests these avenues of further investigation.

5 Conclusion It is often assumed that modal verbs are one of several expressions of ‘modality’, which, along with marked forms of mood (subjunctive) and sentence type (imperative, interrogative), are grammatical means by which the speaker expresses something other than a factual statement. (Lyons, 1969; Cinque, 1997; Giorgi and Pianesi, 1997; Palmer, 2001). Modal verbs are similarly discussed in notional terms. Lyons, for example, describes Modality as marking “the speaker’s commitment with respect to the factual status of what he is saying (his emphatic certainty, his uncertainty or doubt, etc.). . . (1969:307).” The opposite view is taken by Jespersen (1924). Jespersen argues that mood is a syntactic category; he emphasizes that there is no dependable correspondence between the formal and notional categories of mood. The discussion in this article suggests how the syntactic category of modality can underlie the interpretive generalizations. The latter seem to be traceable to different syntactic relationships among several constituents of clauses, including subjects, C, Tense, Aspect and the core verb phrase of the clause.

References Boogaart, R. (2004). Temporal relations in modal constructions. Paper presented at Chronos VI, Geneva, 22–24 Sept., 2004. Borgonovo, C., & Cummins, S. (2007). Past Tense Modals in Spanish and French. In L. Eguren & O. Fernandez Soriano (Eds.) Proceedings of the 16th Colloquium on Generative Grammar. Amsterdam: John Benjamins. In Press. Chomsky, N. (1975). The Logical Structure of Linguistic Theory. Chicago, IL: University of Chicago Press. Chomsky, N. (1995). The Minimalist Program. Cambridge,MA: MIT Press. Chomsky, N. (2001). Derivation by Phase. In M. Kenstowicz, (Ed.), Ken Hale: A Life in Language, (pp. 1–52), Cambridge, MA: MIT Press. Chomsky, N. (2006). On phases. In R. Freidin, C.-P. Otero & M.L. Zubizaretta (Eds.), Foundational Issues in Linguistic Theory. Cambridge, MA: MIT Press. In Press.


291

Cinque, G. (1999). Adverbs and Functional Heads: A Cross-Linguistic perspective. Oxford: Oxford University Press. Comrie, B. (1976). Aspect. Cambridge: Cambridge University Press. Kiparsky, P. (1970). Semantic rules in grammar. In H. Benediktsson (Ed.), The Nordic Languages and Modern Linguistics, (pp. 262–285) Reykjavik: Visindafelag Islendinga. Giorgi, A., & Pianesi, F. (1997). Tense and Aspect: From semantics to morphosyntax. Oxford: Oxford University Press. Jespersen, O. (1924). The Philosophy of Grammar. New York, H. Holt and Company. Lyons, J. (1969). Introduction to Theoretical Linguistics. Cambridge: Cambridge University Press. Palmer, F.R. (2001). Mood and Modality. (Second Edition). Cambridge: Cambridge University Press. Perlmutter, D. (1970). The two verbs begin. In R. Jacobs & P. Rosenbaum (Eds.), Readings in English Transformational Grammar, (pp. 107–119). Waltham, MA: Ginn and Company. Picallo, M.C. (1990). Modal Verbs in Catalan. NLLT 8, 285–312. Platzack, C. (1979). The Semantic Interpretation of Aspect and Aktionsarten: A study of internal time reference in Swedish. Dordrecht: Foris. Reed, L. (2005). The Syntax of Non-Periphrastic Modal Verbs in French. Manuscript: Pennsylvania State University. Ross, J. (1969). Auxiliaries as Main Verbs. In W. Todd (Ed.), Studies in Philosophical Linguistics, (pp. 77–102). Evanston, IL: Great Expectations Press. Smith, C. (1997). The Parameter of Aspect. Dordrecht: Kluwer. Stowell, T. (2004). Tense and Modals. In. J. Guéron and J. Lecarme The Syntax of Time (pp. 621–635). Cambridge, MA: MIT Press. Travis, L. (1991). Inner aspect and the structure of VP, NELS 22, Amherst MA: GLSA. Zagona, K. (2007a). On the Syntactic Features of Epistemic and Root Modals. In L. Eguren & O. Fernandez Soriano (Eds.), Proceedings of the 16th Colloquium on Generative Grammar. Amsterdam: John Benjamins. In Press. Zagona, K. (2007b). Perfective aspect and Contained Perfectivity. To appear in Lingua.

Index

Abusch, D., 258 Agent argument, 152 Anchoring, 8, 11, 12, 42, 43, 48, 52, 65, 66, 90, 93, 95, 96, 98, 102, 106, 217, 219 Arabic, 1, 2, 145, 146, 148, 165 Aspect imperfective, 106, 145, 169, 281, 283, 289, 290 inner (aktionsart), 159, 167 lexical, 145 outer, 101 perfective, 145, 234, 276, 281–283, 288–290 Aspect Phrase (AspP), 3, 92, 243, 245 Aspectual viewpoint imperfective, 105, 106, 229 neutral, 229, 235, 237, 241, 244, 245 perfective, 101, 229, 234, 235, 238, 239, 244 Assertion-time, 144, 146, 147, 149, 154, 156, 157, 162, 168–170 Boogaart, R., 274 Bouletic modal, 268 Bounded Event Constraint, 2, 3, 230, 231, 234, 235, 238 Bounded situation, 3 Boundedness, 2, 3, 229, 230, 237, 241, 242, 244–246 Carlson, G., 11, 19, 22, 23, 27, 157, 207 Causality, 5, 7, 149–151, 153, 155–165, 167, 169, 170 Causative verbs, 8, 162, 163, 167 Chaos, 17, 18, 26, 27, 34, 235 Chomsky, N., 6, 7, 146, 149, 201, 204, 274, 286 Cinque, G., 214, 218, 274, 277, 290 Conditional K, 270

Conditional mood, 12, 53–55, 122, 257 Conditional(s), 116, 118, 119, 125, 132, 195, 196, 210, 220, 228, 251, 253, 257–264, 267–270 Condoravdi, C., 5, 7, 8, 61, 63, 69–71, 81, 82, 87–89, 92, 100, 101, 111, 145 Constraint, 29, 68, 71, 111, 228, 230, 231, 234, 238 Copula, 11, 12, 39–44, 46, 47, 49–51, 123, 134, 163, 164, 256, 257, 264 Copular clauses, 11, 12, 40–42 Counterfactuality and contrasting alternatives, 263 conditionals, 121, 128, 253, 260, 261, 266, 268 mood, 252, 266 subjunctive, 264, 266, 267 Davidson, D., 174, 178, 179 Dayal, V., 211 De dicto/de re distinction, 265 Definiteness, 197, 200, 201, 203, 205 Deictic principle, 230, 231, 233, 235, 238 Deictics, 198, 237 Demonstratives, 197, 198, 205, 208, 212, 219, 220 Deontic modality, 127 Deontic necessity modal, 268 Determiner phrase (DP), 41, 42, 146–148, 160, 161, 196, 198–202, 204–206, 208, 210, 213, 218, 219, 284, 286–290 Discourse-linking, 43, 44 Dretske, F., 219, 220 Economy constraints, 104 English, 101, 103–106, 108, 111, 120, 124, 126, 128, 129, 131–133, 135, 144–146,

293

294 148, 152, 154, 155, 157, 159, 162, 163, 165–167, 173–176, 181, 184–188, 190–192, 198, 199, 202–205, 209, 211, 231–234, 236, 237, 252, 253, 257, 260–262, 264–271, 274, 275, 281–284, 288–290 Episodic, 22, 23, 27, 28, 31, 164, 206, 208, 220 Epistemic, 104–108, 110, 111, 116, 117, 119, 124, 147, 166, 168, 208, 211, 214, 218, 220, 273, 276, 280–282, 287, 288 Event perception, 158 position, 146, 176, 179, 180 structure, 159, 160, 165, 166, 229, 241, 243, 246 Eventive verbs, 5, 70, 258 Eventuality type, 3–5, 60, 62, 64, 69, 73, 75 Evidential(ity), 155, 158, 168, 196, 211–216, 218–220 Evidentiality, in nominals, 196, 212, 213 Exception, 11, 19, 21, 22, 24, 32, 33, 35, 36, 44, 134, 135, 177 Exclusion feature, 196 Faller, M., 214, 215 French, 2, 6, 10, 11, 40, 52, 85, 86, 94, 95, 100, 121, 122, 145, 148, 155, 159, 163, 164, 166, 211, 232, 234, 264, 265, 271, 276, 280, 281, 283, 284, 288 French modals, 284 Futurates, 70–73 Future, 102–104, 106–111, 120–123, 145, 169, 170, 181, 186–188, 190, 201, 213, 220, 230–235, 238–240, 246, 261, 263, 266 Future less vivid, 258 Future perfect, 66, 182, 183, 216 Generic quantifier, 180, 208, 217 sentences, 11, 18–20, 22, 23, 25, 27, 29, 30, 33, 34, 157, 196 German, 2, 31, 32, 61, 185, 203, 205, 232, 269, 270 German Konjunctiv II, 252, 264 Giorgi, A., 154, 186, 188, 230, 290 Greek, 10, 120–122, 126, 127, 130, 133 Habitual(s), 11, 18–21, 23, 25, 29, 33, 70–73, 106, 157, 196, 206, 207 Iatridou, S., 9, 10, 13, 118, 121, 122, 128, 132, 145, 153, 168, 184, 196, 205, 210, 211, 258–260, 265 Imperatives, 73, 283, 290

Index Imperfective conditionals, 1 Italian imperfect, 260, 262 Instrument, 8, 150–153, 155–158, 161–170 Intensional (noun phrase), 12, 42, 43, 164 Intensionality, 12, 19, 23, 24, 33, 43 Inversion in conditional clauses, 268, 269 Italian, 2, 159, 260, 261 Jespersen, O., 173, 174, 176, 184, 290 Kamp, H., 28, 230, 233, 243, 245 Kayne, R., 160, 252 Klein,W., 60, 83, 206, 229, 233 Konjunktiv–Zwei, 252 Kratzer, A., 9, 10, 27, 33, 73, 117, 136, 137, 149, 168, 196, 207, 210, 215–218, 220 Larson, R., 211 Lewis, D.K., 19, 208 Lexical licensing vs. operator licensing of subjunctive, 267 Mandarin Chinese, 2, 228, 232, 234, 235, 243 McTaggart, E.,, 186, 187 Metaphysical modality, 75 Modal operator and subjunctive licensing, 267 in subjunctive clause, 262 Modal verbs may, 5, 91 will, 59–61 would, 130, 131 Modal-time, 8, 9, 80, 81, 87, 89–91, 93–99, 102–110 Modality Epistemic, 6, 8, 9, 70, 75, 87, 90, 96, 126, 170, 289 Metaphysical, 5, 8, 9, 60, 70, 73, 88, 105 possible worlds analysis of, 12, 210–212 subjective epistemic, 219 Modalized free relatives, 209, 210 Modals contexts, 9, 10, 60, 62, 65, 68, 69, 117, 215 deontic, 9 determiners, 205 English, 1, 2, 4, 7, 8, 80–82, 87, 91, 99, 278, 289 epistemic—, 4, 5, 7, 55, 60, 73, 91, 110, 211, 212, 215–219, 274–280, 283–286, 288–290 non-root, 5, 8, 80–82, 87, 89, 91, 92, 94, 98 root, 5–7, 283, 285, 289 strong necessity, 9

Index temporal interpretation of, 3, 5, 8, 81, 87, 89, 91, 93, 96, 98 weak necessity, 9, 10 Movement X-movement, 79 XP-movement, 79 Musan, R., 203, 204 Navajo, 2, 228, 232, 240, 243 Negation, 10, 21, 33–35, 166, 187, 212, 253, 265, 267, 271, 284 Neo-Reichenbachianism, 67, 232 Ogihara, T., 184 Old English, 2, 8, 126, 143, 144, 146, 162, 167, 170, 211 Ordering source, 9, 10, 117, 119, 128, 136, 137, 215–218, 220 Pancheva, R., 184–187, 210 Parsons, T., 173, 175 Past modals, 81, 100, 105–107, 111 perfect, 3, 10, 183, 187, 188, 191, 252, 253, 255, 257, 259–262, 268, 271, 281 polarity, 261, 262, 271 polarity item, 10, 261 Past-shifting tense interpretation, 256 Pattern, 116, 122, 129, 133, 139, 201, 230, 231, 234, 235, 237, 239–242, 281, 282 Perception, visual, 10 Perfect aspect, 3, 4, 6, 85, 98–102, 105, 106, 109, 111, 181, 217 modals, 87, 89, 91, 95, 98, 99, 102, 108, 111 of evidentiality, 216, 217 Perfectivity, 3, 13, 145, 288, 290 Pianesi, F., 154, 186, 188, 230, 290 Point of view, 6, 23, 52–55, 104, 131, 147, 148, 150, 153, 155, 156, 188, 215, 263 Polarity, 209, 211, 261, 262 Portner, P., 53, 184, 187 Possible worlds, 1, 7, 12, 19, 33, 36, 73, 118, 149, 170, 210–212, 220 Present eventive constraint, 5, 60, 65, 68 modals, 81, 95, 99 tense, 2, 3, 20, 71, 72, 92, 93, 100–102, 104, 144–147, 155, 157, 184, 190, 196, 210–214, 216–219, 231–234, 258, 259, 261, 265, 266, 278, 279 Present perfect, see Perfect Preterit, 100, 174, 188, 253, 259 Preterit past, 102, 256, 259–261, 265

295 Progressive modals, 95–97, 110 Propositions, 6, 7, 19, 23, 48, 55, 74, 75, 119, 137, 138, 177, 178, 208, 215–219, 278, 284, 288 Quer, R., 267 Register (informal register), 10, 130, 252, 253, 255–257, 262, 266, 269 Reichenbach, H., 60, 172, 228, 232, 233 Result state, 4, 159, 165, 175, 176, 182, 183, 187, 188 Resultant state, 4, 175, 180, 182, 183, 187, 188, 220 Reyle, U., 113, 230, 233, 243, 245 Romanian, 2, 12, 39, 40, 43–45, 47–51, 53, 54 Root modals, 108, 273–278, 280–283, 288 Run time, 60, 71 Sequence of tense, 173, 176, 181, 184, 187–190, 268, 279 Settledness, 70–72 Simultaneous tense interpretation, 260 Situation bounded, 3, 229–231, 241 time, 4, 8, 9, 60, 80, 92, 237 type, 4, 228, 229, 241, 243, 244 Somali, 2, 10, 146, 196–198, 200–205, 207–210, 212, 213, 218, 220 Spanish, 2, 3, 6, 8, 80, 81, 91, 100, 101, 104–106, 108–110, 122, 159, 162, 264, 265, 267, 276, 280, 281, 283, 288 Spanish modals, 8, 80, 82, 104, 284 Spatiotemporal predicate AFTER, 82, 84, 85, 98 BEFORE, 82, 84 WITHIN, 82, 84, 96, 98 Standard English, 255, 256, 270 Stative verbs, 5, 61 Stechow, A. von, 138, 184–187 Subinterval property, 86 Subject orientation, 5, 6, 277, 285–287 predicate, 6, 7, 40, 53–55 Subjunctive counterfactual, 261, 264–268, 271 mandative, 10, 264–267, 270, 271 mood, 10, 252, 256, 259–262, 264–269, 271 perfect, 10, 259–267, 271 polarity, 10, 262, 269, 271 were, 256, 269 Temporal adverbials, 66, 184, 233

296 anaphora, 79, 112 binding, 84, 85 covaluation, 84, 85, 101 orientation, 5, 13, 59, 60, 62, 64, 68, 69, 73, 75, 170 predicates, 71, 185 QR (Quantifier Raising), 79, 102, 104 scope reversal, 79, 103, 104, 108 syntax, 8, 82, 98, 105, 202, 204, 205 Temporal argument, in nominals, 60, 62, 83, 89, 203, 205 Tense future, 1, 110 in nominals, 10, 197 past, 1, 3, 6, 9, 10, 20, 85, 92, 101, 102, 105, 106, 120, 122, 123, 144–146, 184, 196, 232, 233, 235, 239 sequence of, 4, 181, 190, 255, 260, 261, 266–268 Tense Phrase (TP), 2, 3, 7, 275, 283, 285, 287, 289 Tensed determiners, 50, 197, 199 Tensed languages, 13, 228, 231–235, 246 Tenseless languages, 2, 13, 227, 228, 231, 232, 234, 235, 237, 239, 240, 242, 243, 246 Thai, 232, 239, 242, 243 Time, 53

Index arguments, 82–84, 89, 90, 102 future, 74, 97, 210, 233, 259, 273 past, 3, 6, 9, 88, 99, 103, 107, 108, 145–147, 183, 185, 202, 203, 207, 237, 255, 260, 278, 280, 281 present, 9, 18, 70, 72, 144, 146, 219, 231, 278 Topic, 12, 42, 52–54, 196, 206 Unbounded situation, 229, 231, 235, 237, 241, 245 Uninterpretable features, 274, 287 Verbs ask/demand, 251, 268, 271 want, 268 wish(es), 251 Viewpoint aspect, 83, 86, 96, 98, 112, 282, 284 Visual perception, 155, 158, 212, 219, 220, 231 Woll, 258, 260 Yukatek Mayan, 239 Zeit Phrase, 82

Time and Modality (Studies in Natural Language and Linguistic Theory)

Ergativity: Emerging Issues (Studies in Natural Language and Linguistic Theory)

Universals of Language Today (Studies in Natural Language and Linguistic Theory)

Natural language parsing and linguistic theories

Locality Domains in the Spanish Determiner Phrase (Studies in Natural Language and Linguistic Theory)

Topics in Kwa Syntax (Studies in Natural Language and Linguistic Theory)

Linguistic Diversity And Language Theories (Studies in Language Companion Series)

Inner Aspect: The Articulation of VP (Studies in Natural Language and Linguistic Theory, 80)

Handbook of Word-Formation (Studies in Natural Language and Linguistic Theory)

Balkan Sprachbund Morpho-Syntactic Features (Studies in Natural Language and Linguistic Theory)

Linguistic Diversity And Language Theories (Studies in Language Companion Series)

Linking: The Geometry of Argument Structure (Studies in Natural Language and Linguistic Theory)

The Linearization of Affixes: Evidence from Nuu-chah-nulth (Studies in Natural Language and Linguistic Theory)

Challenges in Natural Language Processing (Studies in Natural Language Processing)

Language, Cohesion and Form (Studies in Natural Language Processing)

Language and Linguistic Nonviolence

Building Natural Language Generation Systems (Studies in Natural Language Processing)

Building Natural Language Generation Systems (Studies in Natural Language Processing)

A Linguistic Theory of Translation (Language and Language Learning)

Language Teaching: Linguistic Theory in Practice

Language Teaching: Linguistic Theory in Practice

Exploring time, tense, and aspect in natural language database interfaces

Exploring time, tense and aspect in natural language database interfaces

Exploring Time, Tense and Aspect in Natural Language Database Interfaces (Natural Language Processing, 6)

Studies in African Linguistic Typology (Typological Studies in Language)

Romanticism and Linguistic Theory: William Hazlitt, Language, and Literature (Transitions)

Linguistic Universals and Language Change

Linguistic Universals and Language Change

Time and Modality (John Locke Lecture)

Time and Modality (Studies in Natural Language and Linguistic Theory)

Ergativity: Emerging Issues (Studies in Natural Language and Linguistic Theory)

Universals of Language Today (Studies in Natural Language and Linguistic Theory)

Natural language parsing and linguistic theories

Locality Domains in the Spanish Determiner Phrase (Studies in Natural Language and Linguistic Theory)

Topics in Kwa Syntax (Studies in Natural Language and Linguistic Theory)

Linguistic Diversity And Language Theories (Studies in Language Companion Series)

Inner Aspect: The Articulation of VP (Studies in Natural Language and Linguistic Theory, 80)

Handbook of Word-Formation (Studies in Natural Language and Linguistic Theory)

Balkan Sprachbund Morpho-Syntactic Features (Studies in Natural Language and Linguistic Theory)

Linguistic Diversity And Language Theories (Studies in Language Companion Series)

Linking: The Geometry of Argument Structure (Studies in Natural Language and Linguistic Theory)

The Linearization of Affixes: Evidence from Nuu-chah-nulth (Studies in Natural Language and Linguistic Theory)

Challenges in Natural Language Processing (Studies in Natural Language Processing)

Language, Cohesion and Form (Studies in Natural Language Processing)

Language and Linguistic Nonviolence

Building Natural Language Generation Systems (Studies in Natural Language Processing)

Building Natural Language Generation Systems (Studies in Natural Language Processing)

A Linguistic Theory of Translation (Language and Language Learning)

Language Teaching: Linguistic Theory in Practice

Language Teaching: Linguistic Theory in Practice

Exploring time, tense, and aspect in natural language database interfaces

Exploring time, tense and aspect in natural language database interfaces

Exploring Time, Tense and Aspect in Natural Language Database Interfaces (Natural Language Processing, 6)

Studies in African Linguistic Typology (Typological Studies in Language)

Romanticism and Linguistic Theory: William Hazlitt, Language, and Literature (Transitions)

Linguistic Universals and Language Change

Linguistic Universals and Language Change

Time and Modality (John Locke Lecture)

Recommend Documents