The Grammar of Q: Q-Particles, Wh-Movement, and Pied-Piping

The Grammar of Q OXFORD STUDIES IN COMPARATIVE SYNTAX Richard Kayne, General Editor The Higher Functional Field: Evid...

Author: Seth Cable

369 downloads 578 Views 1MB 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

The Grammar of Q

OXFORD STUDIES IN COMPARATIVE SYNTAX Richard Kayne, General Editor The Higher Functional Field: Evidence from North Italian Dialects Cecilia Poletto

The Structure of CP and IP: The Cartography of Syntactic Structures, Volume 2 Edited by Luigi Rizzi

The Syntax of Verb-Initial Languages Edited by Andrew Carnie and Eithne Guilfoyle

The Syntax of Anaphora Ken Safir

Parameters and Universals Richard Kayne

Principles and Parameters in a VSO Language: A Case Study in Welsh Ian G. Roberts

Portuguese Syntax: New Comparative Studies Edited by João Costa XP-Adjunction in Universal Grammar: Scrambling and Binding in Hindi-Urdu Ayesha Kidwai Infinitive Constructions: A Syntactic Analysis of Romance Languages Guido Mensching Subject Inversion in Romance and the Theory of Universal Grammar Edited by Aafke Hulk and Jean-Yves Pollock Subjects, Expletives, and the EPP Edited by Peter Svenonius A Unified Theory of Verbal and Nominal Projections Yoshiki Ogawa Functional Structures in DP and IP: The Cartography of Syntactic Structures, Volume 1 Edited by Guglielmo Cinque Syntactic Heads and Word Formation Marit Julien The Syntax of Italian Dialects Christina Tortora The Morphosyntax of Complement-Head Sequences: Clause Structure and Word Order Patterns in Kwa Enoch Oladé Aboh

Structures and Beyond: The Cartography of Syntactic Structures, Volume 3 Edited by Adriana Belletti Movement and Silence Richard S. Kayne Restructuring and Functional Heads: The Cartography of Syntactic Structures, Volume 4 Guglielmo Cinque Scrambling, Remnant Movement and Restructuring in West Germanic Roland Hinterhölzl The Syntax of Ellipsis: Evidence from Dutch Dialects Jeroen van Craenenbroeck Mapping the Left Periphery: The Cartography of Syntactic Structures, Volume 5 Edited by Paola Benincá and Nicola Munaro Mapping Spatial PPs: The Cartography of Syntactic Structures, Volume 6 Edited by Guglielmo Cinque and Luigi Rizzi The Grammar of Q: Q-Particles, Wh-Movement, and Pied-Piping Seth Cable

The Grammar of Q Q-Particles, Wh-Movement, and Pied-Piping Seth Cable

2010

Oxford University Press, Inc., publishes works that further Oxford University’s objective of excellence in research, scholarship, and education. Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam

Copyright © 2010 by Oxford University Press, Inc. Published by Oxford University Press, Inc. 198 Madison Avenue, New York, New York 10016 www.oup.com Oxford is a registered trademark of Oxford University Press. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press. Library of Congress Cataloging-in-Publication Data Cable, Seth. The grammar of Q: Q-particles, WH-movement, and pied-piping / Seth Cable. p. cm.— (Oxford studies in comparative syntax) Includes bibliographical references and index. ISBN 978-0-19-539226-5; 978-0-19-539227-2 (pbk.) 1. Grammar, Comparative and general—Interrogative. 2. Grammar, Comparative and general—Word order. 3. Grammar, Comparative and general—Syntax. 4. Minimalist theory (Linguistics) I. Title. P299.I57C33 2010 415—dc22 2009037248

9 8 7 6 5 4 3 2 1 Printed in the United States of America on acid-free paper

Dedicated to the memory of Johnny Marks (Kóox, Kootex’teek), a great teacher and scholar, without whom this work would not have been possible.

This page intentionally left blank

ACKNOWLEDGMENTS

First and foremost, I wish to thank David Katzeek, Anita Lafferty, John Marks, and Fred White, the Tlingit language consultants for this project. Their generosity, patience, and energy are truly exceptional, and I thank them for all the time and help they have provided me in my study of their language. Learning from them has been a great privilege and a great pleasure, one for which I will always be deeply grateful. Most of the Tlingit data presented here were gathered from interviews conducted at the Sealaska Heritage Institute (SHI) in 2005, 2006, and 2007. Special thanks are owed to Rosita Worl, Yarrow Vaara, Jordan Lachler, and everyone else at SHI. The time I have spent at SHI has always been remarkably productive, enjoyable, and memorable. It is my hope that the information contained here might be of some use to SHI and its programs, so that the resources and knowledge that I have received from SHI may be repaid in some small way. In the course of this work’s development, it has been greatly improved by the criticisms and corrections of the world’s leading scholars of the Tlingit language. For their many contributions (particularly their corrections), I wish to thank Jeff Leer, Richard Dauenhauer, Nora Marks Dauenhauer, Keri Edwards, and James Crippen. Finally, I wish to thank Roby Littlefield for inquiring about certain Tlingit forms with Mary Anderson, and Mary Anderson for teaching both Roby and me these forms. Aatlein gunalchéesh! For providing me their Japanese judgments, I thank the following persons: Sachiko Kato, Shigeru Miyagawa, Junri Shimada, and Shoichi Takahashi. For providing me data on Korean, I thank Dong-Whee Yang. Many thanks are owed to Hideki Kishimoto for providing important data and information regarding Sinhala. I also thank Kai von Fintel for providing his German judgments, and Kirill Shklovsky vii

viii

ACKNOWLEDGMENTS

for his Russian judgments. Finally, Jessica Coon is owed special thanks for all the data she has provided regarding Chol, Tzotzil, and Mayan in general. This work is a revised and greatly reduced version of my doctoral dissertation, submitted to MIT in 2007. In addition to many corrections and rewrites, several significant portions of the dissertation have been removed. For example, the present work does not discuss pronominal resumption in Tlingit, nor does it provide an indepth review of the semantic puzzles surrounding pied-piping. The removal of this material was not for its inadequacies, but simply in order to streamline the argumentation. For this reason, I sometimes refer the reader to Cable (2007) for greater expansion on certain points. Although I claim authorship of this work, its content has been greatly influenced by the comments and criticisms of a number of outstanding individuals. First and foremost among these is David Pesetsky, my principal doctoral advisor, as well as the other members of my doctoral committee: Danny Fox, Irene Heim, and Norvin Richards. The following individuals also provided critically important comments upon earlier versions of this work: Mark Baker, Henry Davis, Keri Edwards, Kai von Fintel, Paul Hagstrom, Fabian Heck, Sabine Iatridou, Angelika Kratzer, Jeff Leer, Lisa Matthewson, Shigeru Miyagawa, and Keren Rice. I would also like to briefly acknowledge here the outstanding work of Fabian Heck (Heck 2004, 2008, 2009), whose 2004 dissertation provided much of the background to my own, and whose 2008 book (particularly its critical discussion of my thesis) spurred many of the revisions and improvements herein. I would like to thank the audiences at the following workshops and conferences for their comments on earlier versions of this work: NELS 37, WSCLA 11, SULA 5, and the 2006 ECO5 Student Syntax Workshop. I would also like to thank the audiences and organizers of the colloquia where this work has been presented: University of Ottawa; University of Massachusetts, Amherst; UCLA; University of British Columbia; University of Victoria; Cornell University; University of Maryland; McGill University; and University of Tromsø. I gratefully acknowledge the financial support of the National Science Foundation. This book is based on work supported under a National Science Foundation Graduate Research Fellowship and a National Science Foundation Dissertation Improvement Grant (BCS-0632431). The most critical support, however, has come from my wife, Summer. Without her, none of this could have come into being.

CONTENTS

Abbreviations

xi

American Tlingit Orthography 1.

xiii

Introduction 3 1.1 The Central Claim and the Main Character 3 1.2 Some Classic Assumptions in the Theory of Wh-Questions 1.3 Tlingit Wh-Questions Force a New Approach 6 1.4 Two Broader Consequences for Grammatical Theory 9 1.5 The Overarching Research Project, and Further Major Consequences 11

4

2. Wh-Fronting and Q-Movement in Tlingit 13 2.1 Introduction 13 2.2 Relevant Background Regarding the Tlingit Language 14 2.3 The Behavior of Wh-Words in Tlingit Wh-Questions 21 2.4 Q-Particles in Tlingit Wh-Questions: The Formal Status of Sá 30 2.5 Wh-Fronting in Tlingit as a Consequence of Q-Movement 36 2.6 The QP-Intervention Condition 43 2.7 A Semantics for Tlingit Wh-Words and Q-Particles 63 3. Applications to Wh-In Situ Languages 84 3.1 Introduction 84 3.2 The Nature of Wh-In Situ Languages 85 3.3 The Semantics of Wh-Indefinites and Wh-Questions in Wh-In Situ Languages 93 3.4 The Theory of LF/Focus Intervention Effects 96

x

CONTENTS

4. Applications to Other Wh-Fronting Languages, Pied-Piping, and Intervention Effects 100 4.1 Introduction 100 4.2 The Generality of the Q-Based Structure: Some Initial Motivation 101 4.3 Some Initial Applications to the Theory of Pied-Piping Structures 115 4.4 Intervention Effects and Superiority Effects in Wh-Fronting Languages 122 5.

Constraints on Pied-Piping and Secondary Wh-Fronting 141 5.1 Introduction 141 5.2 Q/Wh-Agreement and the Constraints on Pied-Piping 142 5.3 Further Results Regarding Pied-Piping 156 5.4 Secondary Wh-Fronting 176 5.5 Massive Pied-Piping and Its Constraints 190

6.

Conclusion 199 6.1 Introduction 199 6.2 The Syntax and Semantics of Other A-Bar Movements 6.3 Free Relatives 206

Notes 211 Bibliography 235 Index 245

200

ABBREVIATIONS

Grammatical Categories 3PL ABS ACC AUX DAT DEC DUB ERG EXCLM FOC GEN HON LINK NOM PART PERF PST Q REFL REL TOP

Third Person Plural Marker / Prefix / Pronoun Absolutive Case Marker / Postposition Accusative Case Marker / Postposition Auxiliary Verb Dative Case Marker / Postposition Declarative Force Marker Dubitative Marker / Particle Ergative Case Marker / Postposition Exclamative Marker / Particle Focus Particle / Marker Genitive Case Marker / Postposition Honorific Marker Linking Morpheme Nominative Case Marker / Postposition Partitive Marker / Particle Perfective Aspect Past Tense Question Particle / Q-Particle / ‘Q’ Reflexive Marker / Pronoun Relative Clause Marker Topic Particle / Marker

xi

xii

ABBREVIATIONS

Grammatical Principles or Phenomena CSC FCA PIC

Coordinate Structure Constraint First Conjunct Agreement Phase Impenetrability Condition

Rules of Semantic Composition FA LC

Function Application Lambda Conversion

Institutions SHI

Sealaska Heritage Institute

AMERICAN TLINGIT ORTHOGRAPHY

Consonants American Orthography d t t’ dz ts ts’ s s’ n dl tl tl’ l l’ j ch

IPA t t‫ހ‬ t‫ތ‬ ࢎts tࢎ‫ހ‬s ࢎts‫ތ‬ s s‫ތ‬ n ࢎt ܾ tࢎ ‫ܾހ‬ ࢎtܾ‫ތ‬ ܾ ܾ‫ތ‬ tࢎ ‫ݕ‬ tࢎ ‫ݕހ‬

Description of Sound voiceless alveolar stop voiceless aspirated alveolar stop ejective alveolar stop voiceless alveolar affricate voiceless aspirated alveolar affricate ejective alveolar affricate voiceless alveolar fricative ejective alveolar fricative alveolar nasal stop voiceless lateral affricate voicless aspirated lateral affricate ejective lateral affricate voiceless lateral fricative ejective lateral fricative voiceless palatal affricate voiceless aspirated palatal affricate

ch‫ތ‬ sh y g gw k

tࢎ ‫ތ ݕ‬ ‫ݕ‬ j k k‫އ‬ k‫ހ‬

ejective palatal affricate voiceless palatal fricative palatal glide voiceless velar stop rounded voiceless velar stop voiceless aspirated velar stop xiii

xiv

kw k’ k’w x xw x’ x’w w g gw k kw k’ k’w x xw x’ x ’w . .w h hw

AMERICAN TLINGIT ORTHOGRAPHY

k‫އހ‬ k‫ތ‬ k‫ތއ‬ x x‫އ‬ x‫ތ‬ x‫ތއ‬ w q q‫އ‬ q‫ހ‬ q‫އހ‬ q‫ތ‬ q‫ތއ‬ χ χ‫އ‬ χ‫ތ‬ χ‫ތއ‬ ‫ݦ‬ ‫އݦ‬ h h‫އ‬

rounded voiceless aspirated velar stop ejective velar stop rounded ejective velar stop voiceless velar fricative rounded voiceless velar fricative ejective velar fricative rounded ejecative velar fricative labio-velar glide voiceless uvular stop rounded voiceless uvular stop voiceless aspirated uvular stop rounded voiceless aspirated uvular stop ejective uvular stop rounded ejective uvular stop voiceless uvular fricative rounded voiceless uvular fricative ejective uvular fricative rounded ejecative uvular fricative glottal stop rounded glottal stop glottal fricative rounded glottal fricative

ì‫ޝ‬ í‫ޝ‬ ì

long low-toned high front long high-toned high front short low-toned high front short high-toned high front long low-toned mid front long high-toned mid front short low-toned mid front short high-toned mid front long low-toned low back long high-toned low back short low-toned low back short high-toned low back long low-toned high back rounded long high-toned high back rounded short low-toned high back rounded short high-toned high back rounded

Vowels ee ée i

í

ei éi e é aa áa a á oo óo u ú

í è‫ޝ‬ é‫ޝ‬ è é ‫ܤ‬Ғ‫ޝ‬ ‫ޝ́ܤ‬ ‫ܤ‬Ғ ‫́ܤ‬ uҒ‫ޝ‬ ú‫ޝ‬ uҒ ú

The Grammar of Q


1

Introduction

1.1 The Central Claim and the Main Character The central claim of this book is that, for a variety of phenomena surrounding wh-operators, the proper locus of explanation is not those wh-operators themselves, but rather a distinct element bearing a special semantic (and sometimes syntactic) relationship to the wh-operator. In many languages, this distinct element is phonologically empty, and for this reason its existence and importance have not been widely recognized. I follow certain previous authors in referring to this special element as a ‘Q(uestion)-particle’ (Hagstrom 1998; Kishimoto 2005), but it should be noted that this label is somewhat misleading. As we will see, this particle is not restricted simply to questions, and it is a distinct entity from the interrogative complementizer head, which also has a tradition of being referred to as ‘Q’. To help unpack these claims, it will be useful to briefly introduce the ‘main character’ of this book, the wh-questions of Tlingit (Na-Dene; Alaska, British Columbia, Yukon). As we will see, it is these structures, illustrated in (1) and schematized in (2), that provide the most direct evidence for the proposals that follow. (1)

Illustrative Examples of Wh-Questions in Tlingit a. Wáa sá sh tudinookw i éesh? how Q he.feels your father How is your father feeling? (Dauenhauer & Dauenhauer 2000: 138) b. Daa sáwé i éesh al’óon? what Q.FOC your father he.hunts.it What is your father hunting? (Dauenhauer & Dauenhauer 2000: 186)

3

4

THE GRAMMAR OF Q

(2)

General Form of a Wh-Question in Tlingit [S . . . [ [ . . . wh-word . . . ] sá ] . . . Main Predicate . . . . ]

The schema in (2) encapsulates the following properties of Tlingit wh-questions. First, the wh-word must precede the main predicate of the wh-question, and is typically initial in the clause. Secondly, the wh-word is followed by the Q-particle sá, which either directly follows the wh-word or directly follows a phrase containing the wh-word.1 Finally, the remaining material of the sentence typically follows the whword, with a strong tendency to follow the verb. Although this structure might not seem shockingly unfamiliar, we will see that, when examined carefully, the form of wh-questions in Tlingit challenges a variety of widely held views regarding the nature of wh-fronting. To get a sense of why this is, let us briefly review some background regarding the theory of whfronting.

1.2 Some Classic Assumptions in the Theory of Wh-Questions Since at least the 1960s, a fundamental question in the theory of wh-questions has been “Why do wh-words have to front in the wh-questions of some languages?” Although there are many specific answers to this question, they all share the following form: (3) Structure Common to Theories of Wh-Fronting Hypothesis 1: Wh-words have a special property, X. Hypothesis 2: The position that wh-words move to has a special property, Y. Hypothesis 3: General principles entail that X must be located at positions bearing property Y.

That is, across many different frameworks, theoreticians generally agree that whwords front in some languages because the wh-word has a ‘special property’ that requires it to be located at the position that it fronts to. To illustrate, I outline in (4) several theories of wh-fronting, each characterized in terms of its particular value for X, its value for Y, and its particular story regarding why X must be located at Y. (4) Some Theories of Wh-Fronting, Characterized in Terms of the Structure in (3) A GB Account (Pesetsky 1982; May 1985; Lasnik & Saito 1992) X = the feature WH ; Y = the feature COMP and the feature [+WH] An LF filter (the ‘WH-Criterion’) requires that “all WHs be in a [+WH] COMP at LF.”

INTRODUCTION

5

An MP Account (Chomsky 2000) X = an interpretable Q-feature [iQ] ; Y = an uninterpretable Q-feature [uQ] and an EPP feature. Agreement between [iQ] and [uQ] is required for convergence. Because of the EPP feature, such Agreement triggers movement of the phrase bearing X to the position bearing Y. A GPSG Account (Bennett 1995) X = the feature [+Q] ; Y = daughter of a root node bearing the feature [+Q] A principle (the ‘Foot Feature Principle’) requires that a root node bearing the feature [+Q], such as the root node of a wh-question, have a daughter which is [+Q]. An LFG Account (Falk 2001) X = the feature WH ; Y = specifier of CP An ID rule requires that a specifier of CP (as opposed to an adjunct of S) bear the feature WH. A Semantic Account (Karttunen 1977) X = existential force ; Y = scope above the ‘proto-question’ In order for a structure to be interpreted as a wh-question, the existential force contributed by the wh-word must have scope above the ‘proto-question’. A Pragmatic/Discourse-Structural Account (Horvath 1986; Kiss 1995) X = Focus ; Y = Focus Position General principles entail that constituents bearing ‘focus’ appear at the designated ‘focus position’.

Interestingly, although virtually every theory of wh-fronting possesses the classic structure under (3), such proposals are immediately faced with a rather fundamental challenge: how to analyze sentences where more than the wh-word undergoes fronting: (5)

a. [Whose book ] did you read? b. [To whom ] did you speak? c. [How long a book ] did he write?

Although it is not often explicitly said, sentences like these directly challenge the analytic structure in (3). After all, if it is a property of the wh-word that motivates the fronting, how did this property come to appear on the larger, fronted phrase, a phrase that does not otherwise inherit the properties of the wh-word? For example, contrasts like those in (6) show that a possessive DP does not inherit the number properties of a wh-possessor. How, then, does the DP inherit the ‘wh-property’ that supposedly triggers the fronting in (5a)?

6 (6)

THE GRAMMAR OF Q

a. Who is / *are coming to your party? b. [ Whose sisters ] are / *is coming to your party?

There is, of course, a commonly accepted answer, one that allows the hypothesis in (3) to ‘preserve the phenomenon’ in (5): the structures in (5) all exhibit something called “pied-piping”. Although details of implementation vary across frameworks, generally speaking, the term “pied-piping” describes cases where an operation targeting the features of a particular lexical item applies to a phrase properly containing the maximal projection of that item. This definition is highlighted in (7).2 (7)

Pied-piping occurs when an operation that targets the features of a lexical item L applies to a phrase properly containing LMAX.

We might, then, contrast the analytic term “pied-piping” with the more descriptive and theory-neutral label “pied-piping structure”, defined in (8). (8)

A pied-piping structure is one where a phrase properly containing the maximal projection of a wh-word (or related operator) has undergone fronting ‘typically associated’ with that operator.

Thus, to claim that pied-piping exists is to claim that it is possible for an operation targeting the features of L to apply to a phrase properly containing the projections of L. Of course, what makes such cases possible—what mechanisms serve to derive pied-piping—is a separate, subsequent question, one that has received much attention (Ross 1967; Sells 1985; Webelhuth 1992; Kayne 1994; Ginzburg & Sag 2000; Grimshaw 2000; Heck 2004, 2008, 2009; Watanabe 2006; Horvath 2007a). Curiously, however, the more fundamental question of whether pied-piping actually exists has not yet received serious attention. This is largely due to the ubiquity of the explanatory structure in (3). After all, if the only analytic option is that the fronting in wh-questions targets a property of wh-words, then the sentences in (5) clearly show that piedpiping exists. Indeed, in some introductory discussions of pied-piping, pied-piping is presented as an observable phenomenon, a datum that must be explained, rather than as a technical solution to an empirical challenge faced by a particular analysis. In summary, despite significant disagreement over more specific issues, the literature on wh-fronting does exhibit a common, classic consensus: (a) the fronting of wh-words in wh-questions directly results from a property borne by the wh-word, and (b) wh-questions where there is fronting of a phrase properly containing the projections of the wh-word reveal the existence of pied-piping.

1.3 Tlingit Wh-Questions Force a New Approach I will argue that the wh-questions of Tlingit strongly challenge this classic consensus. Specifically, we will see that these structures require a model where wh-fronting is not directly triggered by any properties of the wh-word itself. Rather, such fronting

INTRODUCTION

7

is found to target the features of a distinct, formal element, the aforementioned ‘Q-particle’. The basic ideas underlying the analysis are roughly sketched in (9). (9) The Proposed Analysis of Wh-Questions in Tlingit Daa sá i éesh al’óon? what Q your father he.hunts.it What is your father hunting? Base Structure IP DP

Surface Structure CP VP

I éesh

QP

IP

QP1 V Daa sá

DP

Q

daa

sá

i éesh

t1

al’óon

al’óon

QP-Fronting

In outline, the analysis runs as follows. First, as shown in the structure above, the Q-particle sá must c-command the wh-word. Moreover, this Q-particle heads its own projection, labeled ‘QP’. Note that because of the c-command relation between the Q-particle and the wh-word, this QP projection necessarily contains the wh-word. Finally, and most importantly, the ‘rule’ for forming wh-questions in Tlingit is that the QP is fronted, and nothing about the wh-word specifically enters into the rule at all. Nevertheless, because the QP necessarily contains the wh-word, such obligatory fronting of the QP has as a secondary consequence the obligatory appearance of the wh-word in the left periphery as well. Thus, although it is true that a wh-word must appear in the left periphery of a Tlingit wh-question—and so the language is, descriptively speaking, a ‘wh-fronting’ language—this word order is not due to an operation directly triggered by the features of the wh-word. Rather, the movement targets the features of the Q-particle. The left-peripheral position of the wh-word simply follows from its being contained inside the Q-particle’s phrasal projection. (Note that, although the analysis as sketched in (9) is phrased informally in terms of a ‘rule’ of QP-fronting, the actual analysis developed in chapter 2 will be formalized within the framework of ‘probes and goals’, as employed in recent work within the Minimalist Program.) As we will see, some core evidence in support of this analysis is the fact that the well-formedness of a Tlingit wh-question depends only upon the locality of the Q-particle to the left periphery; the locality of the wh-word is irrelevant. This fact is illustrated by patterns like that in (10). (10)

Wh-Operators May Be Inside Islands Iff Q-Particle Is Outside the Island tuwáa sigóo? a. [ [ Wáa kwligeyi CP] xáat NP] sá i how it.is.big.REL fish Q your spirit it.is.glad How big a fish do you want? (A fish that is how big do you want?)

8

THE GRAMMAR OF Q

b. * [ [ Wáa sá kwligeyi CP] xáat NP] i tuwáa sigóo? how Q it.is.big.REL fish your spirit it.is.glad

As we see in (10a), a Tlingit wh-question may contain a wh-word inside a relative clause island, as long as the Q-particle is located outside of the island. If the Q-particle is located inside the relative clause island, as in (10b), then the sentence is illformed (Hagstrom 1998; Kishimoto 2005). This pattern argues that the rules for forming wh-questions in Tlingit are sensitive only to the position of the Q-particle, and therefore it is only the features of the Q-particle that are referenced by those rules (Hagstrom 1998; Kishimoto 2005). Besides its empirical motivation by facts like (10), the analysis in (9) receives indirect support from the ways in which it simplifies other aspects of Tlingit grammar. One of the most important of these relates to the language’s ‘pied-piping structures’, as defined in (8) and illustrated in (11). (11)

Pied-Piping Structures in Tlingit a. Aadóo yaagú sá ysiteen? who boat Q you.saw.it Whose boat did you see? b. Aadóo teen sá yeegoot? who with Q you.went Who did you go with? c. Daakw keitl sá asháa? which dog Q it.barks Which dog is barking?

As witnessed in (11), the key property of Tlingit pied-piping structures is the following: the particle sá always marks the right edge of whatever has been ‘pied-piped’. For this reason, we can adopt the following as our analysis of these structures. (12)

Pied-Piping Structures Without Pied-Piping in Tlingit Aadóo yaagú sá ysiteen? who boat Q you.saw.it Whose boat did you see? Base Structure IP DP


pro

QP

IP

QP1 V

Aadóo yaagú sá pro

Q

DP1 DP2

DP1

Aadóo

yaagú

sá

ysiteen QP-Fronting

t1

ysiteen

INTRODUCTION

9

Under this analysis, the pied-piping structures in (11) are all simply cases where the Q-particle has as its sister a phrase larger than the maximal projection of the whword. For example, a sentence like (11a) possesses a structure where the Q-particle sá is sister to the complex DP aadóo yaagú ‘whose boat’, which properly contains the maximal projection of the wh-word aadóo ‘who’. Consequently, as we see in (12), the structures in (11) can all be derived by normal phrasal movement of the QP, exactly as in the case of simple wh-questions like (9). Crucially, since it is the QP—and not the wh-word—that bears the features triggering ‘wh-fronting’ in Tlingit, we find that the pied-piping structures of Tlingit are not cases of true pied-piping, as defined in (7). That is, under the analysis in (9), the structures in (11) are not cases where movement applies to a phrase larger than the projection of the lexical item that triggers it. Furthermore, since the projections of sá are never properly contained within the fronted constituent of a Tlingit wh-question, we find that there simply are not any true cases of pied-piping in Tlingit. For this reason, the special concept of ‘pied-piping’ can be eliminated without cost from our theory of Tlingit grammar, thus simplifying the overall theory. By adopting the analysis in (9), we need not deviate from the null hypothesis that if an operation (in Tlingit) targets the features of a given lexical item, then it applies only to the maximal projection of that lexical item.

1.4 Two Broader Consequences for Grammatical Theory Besides the advantages that the analysis in (9) brings to the theory of Tlingit grammar, we will see that it also advances a number of broader issues in the syntax of whconstructions. Of course, for this to be the case, the analysis in (9) must not simply be peculiar to Tlingit, but rather must underlie the structure of wh-questions in many other languages. Indeed, a central claim of this book is that the Q-based analysis in (9) actually holds for all wh-fronting languages. Some initial motivation for this ‘universalist’ position can be found in the following thought experiment. Suppose we were to delete the particle sá from the surface structures of Tlingit. The result would be a language that would not look significantly different from more widely studied wh-fronting languages. This suggests that we might think of these latter languages as having the Q-based structure in (9), but lacking phonologically overt Qs.3 Furthermore, in as much as Tlingit whquestions require the analysis in (9), general typological and learning-theoretic considerations would suggest we extend the analysis to similar wh-fronting structures in other languages. Beyond these general considerations, extending the ‘Q-based’ analysis in (9) to other wh-fronting languages brings a number of other potentially valuable results. One of the most immediate is that the concept of ‘pied-piping’ may be eliminated from the theory of grammar. It will be shown that such elimination comes at relatively little cost, as many of the subtler facts that theories of pied-piping seek to capture receive interesting analysis under a Q-based approach, where there is no true pied-piping.

10

THE GRAMMAR OF Q

Another important result concerns certain well-known conditions on the extraction of wh-words. We will find that certain apparent conditions on wh-movement can be seen to result from independently visible conditions on the placement of Q-particles. To begin unpacking this claim, let us first observe an important property of wh-words functioning as indefinites in Tlingit. As in many languages, Q-particles in Tlingit must appear with wh-words functioning as indefinites in declarative clauses. (13)

Tléil aadóo teen *(sá) xwagoot. not who with Q I.went I didn’t go with anyone.

Importantly, when the Q-particle sá appears with wh-indefinites, there are certain conditions on where the particle can appear. For example, it cannot appear between a postposition and its complement; compare (13) and (14). (14)

*Tléil aadóo sá teen xwagoot. not who Q with I.went

We will see in chapter 2 that, as one might suspect, the QP does not undergo movement in declarative sentences like (13) and (14). Therefore the impossibility of sentences like (14) must reflect a pure condition on the placement of the Q-particle, and not any property of the movement relation itself. On its own, this condition on the placement of Q-particles might be of rather limited interest. However, combined with the analysis in (9), it has intriguing consequences. Note that, under the analysis in (9), the condition against structures like (14) alone rules out postposition stranding in Tlingit. As illustrated in (15), postposition stranding must be derived from a structure where a Q-particle appears between a P and its DP complement. However, as we have just seen, such structures are ruled out by independent constraints governing the placement of Q-particles. Consequently, we find that the ban on postposition stranding in Tlingit is due to a property of the Q-particle, and not a property of the movement relation itself. (15)

Wh-Fronting Cannot Strand a Postposition CP IP

QP1 DP

Q

…wh-word…

PP QP t1

P

Impossible PP, Ruled Out by Constraints on Q-Placement

INTRODUCTION

11

We will see in chapter 2 that similar explanations account for the ill-formedness in Tlingit of various left-branch extractions, phenomena which in other languages are also commonly thought to reflect a property of the movement relation. In its inability to strand adpositions and form left-branch extractions, wh-fronting in Tlingit is very similar to that in other wh-fronting languages. Given the evidence that the account in (15) is correct for Tlingit, it follows that we should pursue such an account for those phenomena in all other wh-fronting languages as well. We find, then, that the nature of postposition stranding and left-branch extractions in Tlingit provides additional support for extending the analysis in (9) to all wh-fronting languages. Moreover, under this approach, we see that phenomena commonly attributed to constraints on movement are ultimately due to more basic conditions on the placement of Q-particles.4

1.5 The Overarching Research Project, and Further Major Consequences The two results described above illustrate and motivate the broader research project undertaken in this book. As mentioned, the perspective of this project is that the formal element pronounced as ‘sá’ in Tlingit is a structural component of the wh-questions of all human languages, including all wh-fronting languages. Because of its phonological invisibility in the best-studied wh-fronting languages, the important role played by the Q-particle in a variety of phenomena has not been recognized. Consequently, these phenomena have in previous treatments been somewhat misanalyzed, often in terms of the movement relation between the wh-word and the left periphery. However, the overt appearance of the Q-particle in Tlingit wh-questions and wh-indefinites yields an invaluable empirical tool, which can factor out three possible sources of explanation. These are listed in (16). (16)

The Sources of Explanation Under the Q-Based Analysis a. Conditions on the movement relation between the QP and the left periphery. b. Conditions on the (initial) position of the Q-particle in the clause. c. Conditions on the relation between the Q-particle and the wh-word.

The research reported here attempts to characterize the contribution of each of these three factors to various phenomena related to wh-words. To the extent that such a project proves to be feasible, and interesting results obtained, additional support is found for the Q-based analysis in (9). We have already been introduced to two of the major results of this project; the following is a more complete list. (17)

Principal Results of the Q-Based Approach • A theory of so-called pied-piping structures, wherein the operation of ‘feature percolation’ and even the concept of ‘pied-piping’ itself are eliminated from the theory of grammar. [chapters 2, 4, 5]

12

THE GRAMMAR OF Q

• A semantics for wh-questions that correctly interprets pied-piping structures without recourse to any mechanisms beyond those needed for wh-questions without pied-piping structures. [chapters 2, 4] • A theory of the constraints on pied-piping structures which captures many aspects of their variation across languages. [chapter 5] • A syntax and semantics for multiple wh-questions, which ties the presence of Superiority Effects to the absence of Intervention Effects, and which predicts a previously unnoticed Intervention Effect in English. [chapter 4] • A unified account of the ill-formedness of certain left-branch extractions, as well as of adposition stranding. [chapters 2, 4] • A typology of wh-question formation, which predicts certain morphological features of wh-indefinites from the structure of wh-questions. [chapter 3]

As these results indicate, a general message of this book is that the introduction of the ‘QP’ projection in (9) brings with it a new and highly versatile analytic tool, one that allows many classic puzzles to be reconceived and approached in novel ways. Finally, the work detailed in this book might also offer yet another object lesson in the importance of endangered and understudied languages in the development of linguistic theory. Often, when linguists are asked to explain the importance of research into such languages, it is noted that their study can provide crucial evidence to decide between competing analyses. In this way, careful documentation of these languages is seen to ‘broaden the empirical database’ that theories of language must cover. However, it is often overlooked in these discussions that the study of such languages can also serve to introduce new analyses, ones that may offer entirely new approaches and perspectives to older, seemingly settled issues (cf. Matthewson 1998; Rullmann et al. 2008). That is, rather than shrink the space of potential analyses, careful study of these languages can reveal that the current hypothesis space is too narrow and fails to include hypotheses that would have otherwise never been imagined for more well-studied languages.

2

Wh-Fronting and Q-Movement in Tlingit

2.1 Introduction In this chapter I argue that the ‘Q-based’ analysis from chapter 1, repeated below, provides the best account of several features of the wh-questions of Tlingit. (1)

The Proposed Analysis of Wh-Questions in Tlingit Daa sá i éesh al’óon? what Q your father he.hunts.it What is your father hunting? Base Structure IP

DP


I éesh

QP

QP1

IP

V Daa sá

DP

Q

daa

sá

i éesh

t1

al’óon

al’óon

QP-Fronting

I begin in the following section by providing the reader with some very brief background regarding the Tlingit language, as well as the methodology employed by the present study. Although comparatively long for a section of its type, this section greatly condenses the expansive background chapter of Cable (2007), to which the reader is referred for further information. Following these preliminaries I undertake my defense of the analysis in (1) for the wh-questions of Tlingit. I begin in section 2.3 by arguing that Tlingit is a 13

14

THE GRAMMAR OF Q

‘wh-fronting language’, in the sense that wh-words in its wh-questions must appear at left peripheral positions. Following this, I argue in section 2.4 that the Tlingit particle sá should be categorized as a ‘Q-particle’ alongside the Japanese particle ka and the Sinhala particle da.1 The importance of this categorization is clarified in section 2.5, where the analysis in (1) is contextualized among certain recent proposals regarding the syntax of Japanese and Sinhala wh-questions (Hagstrom 1998; Kishimoto 2005). Section 2.5 also introduces a more formal implementation of the analysis in (1), couched within the framework of ‘probes’ and ‘goals’, as developed within the Minimalist Program (Chomsky 2000). Finally, section 2.5 presents the core arguments supporting (1) over a more traditional analysis where Tlingit wh-words (rather than QPs) are the ‘targets’ for movement. These arguments show that the interrogative C head of a Tlingit wh-question probes and Agrees with only the (projection of the) Q-particle sá, and not with the wh-word itself. Having presented the core arguments for the analysis in (1), I then discuss in section 2.6 some further Tlingit-internal applications of the analysis. I argue that several additional conditions in Tlingit governing the placement of sá would follow from a single generalization, the ‘QP-Intervention Condition’, but only under the assumption that both sá and its sister are contained within a QP projection. Furthermore, I show that, under the analysis in (1), these independent conditions on the placement of sá would alone account for several additional constraints on the form of Tlingit wh-questions. Interestingly, in many other languages, these ‘additional constraints’ are thought to result from conditions on movement. This fact foreshadows later arguments that the analysis in (1) is not unique to Tlingit, but holds for all whfronting languages. Finally, in section 2.7 I provide a semantics for Tlingit wh-questions and whindefinites. This semantics, based on earlier semantic research into wh-words and Q-particles (Hagstrom 1998; Beck 2006), is shown to correctly interpret the perhaps exotic-looking syntactic structures that the analysis in (1) employs. Interestingly, this semantics also captures certain properties of Q-particles crucial to the argumentation from section 2.4. Most interesting of all, however, is that the system can be shown to assign the correct interpretation to the pied-piping structures of Tlingit, without appealing to any special machinery beyond that used for plain wh-questions without pied-piping.

2.2 Relevant Background Regarding the Tlingit Language In this section I provide the reader with some elementary background regarding the Tlingit language. Section 2.2.1 covers key sociohistorical features of the language, including its geographic distribution, genetic classification, and current vitality. Section 2.2.2 provides the reader with a brief summary of prior scholarship on this rather understudied language, and section 2.2.3 introduces the reader to certain of its major grammatical features. Finally, section 2.2.4 describes the methodology employed in this study.

WH-FRONTING AND Q-MOVEMENT IN TLINGIT

15

2.2.1 Sociohistorical Background 2.2.1.1 Geographic Distribution and Genetic Classification of the Tlingit Language Tlingit is spoken primarily in the southeastern panhandle of Alaska, stretching from Yakutat, Alaska (60 degrees N latitude) to Alaska’s southern border (55 degrees N latitude) (Naish 1966; Story 1966; Leer 1991). Tlingit is also spoken in areas of northern British Columbia (Atlin) and in portions of the Yukon Territory (e.g., Carcross, Tagish, Teslin) (Naish 1966; Story 1966; Leer 1991). Despite its considerable geographic spread, there is rather little dialectical diversity within Tlingit (Leer 1991). The language is typically divided into four major dialects, all of which are mutually intelligible: Northern Tlingit, Transitional Southern Tlingit, Sanya-Henya Tlingit, and Tongass Tlingit (Leer 1991). The key differences between these dialects solely concern prosody, the inventory of tones and certain morphophonological alternations (Dauenhauer & Dauenhauer 1987; Leer 1991). All the speakers consulted in this study are speakers of the Northern Tlingit dialect. The Tlingit language is the sole member of the Tlingit family of the Na-Dene language phylum (Thompson 1996; Campbell 1997; Mithun 1999). Besides Tlingit, the Na-Dene phylum also contains the Eyak language and the Athabaskan languages (e.g., Navajo, Apache, Slave) (Thompson 1996; Campbell 1997; Mithun 1999).2 Although it does not seem possible to establish subgroupings of the three families within the Na-Dene phylum, in several respects Tlingit appears to be further removed from the Athabaskan languages than Eyak does (Leer 1991). In addition, recent studies have demonstrated a genetic link between the Na-Dene languages and the Yeniseian languages of Siberia (Vajda 2009). The key genealogical relationships of Tlingit are illustrated in (2). (2)

The Dene-Yeniseian Languages Dene-Yeniseian Yeniseian Ket Yugh Pumpokol

Na-Dene Tlingit Tlingit

Eyak-Athabaskan Eyak Eyak

Athabaskan Navajo Apache Slave Dogrib, etc.

2.2.1.2 Current and Future Vitality of the Language According to the fifteenth edition of Ethnologue, Tlingit is spoken by approximately 845 individuals in an ethnic population of 10,000 (Gordon 2005). It should be noted, however, that this is simply an estimate attributed to Michael Krauss in 1995. As of the summer of 2010, a rigorous documentation of the number of native Tlingit

16

THE GRAMMAR OF Q

speakers has not been done, though scholars privately estimate the number as now being at most between 300 and 400 (James Crippen, personal communication). The youngest native speakers of Tlingit are in their 50s, and there is no known native speaker of the language under the age of 40 (Dauenhauer & Dauenhauer 1987). Consequently, Tlingit is considered a highly endangered language and has been categorized by some as ‘moribund’ (Leer 1991). It should be noted, however, that there is extensive ongoing work aimed at documenting, maintaining, and revitalizing the Tlingit language (Sealaska Heritage Institute 2003). The Tlingit community generally has a positive attitude toward their language, and community interest in the language continues to grow (Gordon 2005). Courses in Tlingit are presently taught in Alaskan public schools and at the University of Alaska, Southeast, and a number of successful immersion camps have been held (Sealaska Heritage Institute 2003). Thanks in part to this activity, some younger adults have acquired a significant degree of fluency and have made efforts to reintroduce the language into family and public life. Particularly strong L2 communities appear to be emerging in areas throughout Southeast Alaska: Sitka, Klukwan, Hoonah, Juneau, and Wrangell. The ever-growing visibility of the language is illustrated by the recent translation of Macbeth into Tlingit, which was performed at the Smithsonian’s National Museum of the American Indian (Quinn 2007). All of this helps to dispel absolute pessimism concerning the survival of the language and builds a certain amount of guarded optimism amongst specialists.

2.2.2 Prior Scholarship Regarding the Tlingit Language The published scholarship concerning Tlingit is largely dominated by the question of its relationship to the Athabaskan languages. For reasons of space and relevance, the extensive literature devoted to this question will not be discussed here. Summaries can be found in Dürr and Renner (1995) and Campbell (1997). Furthermore, for reasons of space I discuss here only (what I consider) the ‘major works’ of Tlingit scholarship. Cable (2007) provides a more extensive bibliographic tour. Study of the Tlingit language began in the early 19th century, during the period of Russian colonization (Leer 1991; Mithun 1999). The most notable work of this period is the grammatical sketch by Veniaminov (1846). The first extensive collection of Tlingit texts is Swanton (1909), and Swanton (1911) provides a comparatively extensive grammatical sketch. However, this latter work is superseded by Boas (1917), widely regarded as the first adequate and fully accurate description of Tlingit phonology and verbal morphology. The ‘modern period’ of Tlingit language research begins with the ground-breaking work of Naish (1966) and Story (1966). Story (1966) surpasses Boas (1917) in its description of the phonology and verbal morphology of Tlingit, and Naish (1966) far exceeds the latter work in its coverage of the language’s syntax. Indeed, Naish (1966) is the first and only extensive work focusing primarily on the syntax of Tlingit. For this reason, it remains today the principal reference regarding the language’s syntax and represents virtually all that is known within this subject.


17

Following their study of Tlingit grammar, Naish and Story collaborated on both a verb dictionary (Story & Naish 1973) and a noun dictionary (Story et al. 1976). Besides its value as a dictionary, Story and Naish (1973) contains a grammatical sketch and numerous example sentences. A later and more extensive noun dictionary for Tlingit is Leer et al. (2001). Story and Naish (1973) remains the only verb dictionary for the language, although a more ‘learner-friendly’ verb dictionary is in development (Edwards, forthcoming). The only other purely grammatical study of the Tlingit language is the PhD dissertation of Jeff Leer (Leer 1991), whose treatment of Tlingit verbal morphology and phonology improves immeasurably upon the work of Story (1966). Although it clarifies a great many issues in the grammar of Tlingit, Leer (1991) focuses primarily on verbal morphology and phonology, and only 20 pages in 500 are given to syntactic description. It should be noted, however, that those 20 pages contain important, novel insights regarding the structure of the language’s noun phrase and clausal architecture. Since the 1960s, a sizeable amount of Tlingit textual material has been collected and published. Worthy of special mention are the four volumes of texts edited by Nora and Dick Dauenhauer (Dauenhauer & Dauenhauer 1987, 1990, 1994; Black, Dauenhauer, & Dauenhauer 2008). Williams, Williams, and Leer (1978) and Nyman & Leer (1993) are also deserving of special mention. Educational materials are available for the study of Tlingit as a second language. Besides the aforementioned dictionaries, there is an introductory textbook with audio CDs (Dauenhauer & Dauenhauer 2000), a phrase book (Dauenhauer & Dauenhauer 2002), and a phrase book with accompanying CDs (Edwards et al. 2005). An intermediate textbook is presently under development (Dauenhauer & Dauenhauer, forthcoming), as well as a more comprehensive verb dictionary (Edwards, forthcoming). All the aforementioned educational materials (except for those yet to be published) are either published or distributed by the Sealaska Heritage Institute (SHI). 2.2.3 Grammatical Sketch In this section I provide a brief overview of the structure of the Tlingit language. Later sections will direct the reader to portions of this sketch where they are relevant for the argumentation. 2.2.3.1 Phonology Throughout this book I represent the sounds of Tlingit using the ‘American orthography’ for the language (Dauenhauer & Dauenhauer 2000). This orthography is used in the Alaskan school system and in all publications by SHI. A different, ‘Canadian orthography’ is used by Tlingit living in Canada and by the Yukon Native Language Center (Nyman & Leer 1993). The front material of this book provides a chart matching the characters of the American Tlingit orthography to their equivalents in IPA. This chart also summarizes the phonemic inventory of the language. Although the phonemic inventory and phonological alternations of the Tlingit language are of much interest, they will not be of direct relevance to the syntactic study that follows. Interested readers are referred to Leer (1991) and Cable (2006b).

18

THE GRAMMAR OF Q

2.2.3.2 Morphology Like nearly all languages of North America, Tlingit may be described as a ‘headmarking’ language (Nichols 1986). Consequently, nominal morphology in Tlingit is comparatively poor—consisting primarily of possessive marking and optional plural marking—while verbal morphology is quite rich and contributes a wide variety of information about the event described. The verbal morphology of Tlingit is strikingly similar in form to that of its Athabaskan relatives. That is, the verbal morphology is almost exclusively prefixal. Moreover, the order of the prefixes in a Tlingit verb is not easily derivable from general principles, and so it is standardly described by use of a stipulative ‘morphological template’. For descriptions and analyses of the ‘templatic verbal morphology’ of Tlingit, see Story (1966), Leer (1991), Dauenhauer & Dauenhauer (2000), and Cable (2006b). The prefixal template of a Tlingit verb consists of approximately 16 different positions. Consequently, a particular surface verbal form of Tlingit may underlyingly contain a sizeable number of prefixes. As in its Athabaskan relatives, these underlying prefixes are greatly affected by various phonological processes, which serve to drastically reduce the pronounced surface form of the verb. These points are illustrated by the verbal forms in (3). Note that, although the underlying forms of the verbal prefix strings differ in only one syllable, the surface forms differ remarkably in their appearance. (3)

Phonological Alternations in the Tlingit Verbal Prefix String a. Surface Form: Daa sá kkwaxáa? Underlying Form: Daa sá ga-u-ga-xa-xáa? Morpheme Gloss: what Q future-irrealis-modal-1sSubj-eat Translation: What will I eat? Daa sá gaxtooxáa? b. Surface Form: Underlying Form: Daa sá ga-u-ga-too-xáa? Morpheme Gloss: what Q future-irrealis-modal-1plSubj-eat Translation: What will we eat?

Given the complexities of Tlingit verbal morphology, and the fact that we are only concerned with its phrasal syntax, I will only provide the roughest of glosses for the Tlingit verbs exemplified throughout. I will not provide a full morphological breakdown of every verbal form, but will instead gloss only the ‘propositional content’ of a given verb, as illustrated in (4). (4)

Propositional Gloss of Verbal Forms in Tlingit Sentences a. Daa sá kkwaxáa? what Q I.will.eat.it What will I eat? b. Daa sá gaxtooxáa? what Q we.will.eat.it What will we eat?

Note that these ‘propositional glosses’ are merely a notational convenience, and do not represent any serious proposals regarding the morphosyntax of Tlingit. Thus,


19

although these glosses contain English pronouns, I do not seriously adopt the ‘Pronominal Argument Hypothesis’ (Jelinek 1984) for Tlingit. Rather, I assume that full DPs in Tlingit can function as verbal arguments, and are not necessarily mere clausal adjuncts.

2.2.3.3 Syntax As mentioned previously, the only works providing focused discussion of Tlingit syntax are Naish (1966) and Leer (1991). Although many aspects of Tlingit syntax have yet to be studied, certain general features are well documented. Since many of the more specific details of Tlingit syntax will be introduced as they become relevant, I discuss here only the most basic facts about its gross syntactic structure. In its syntax, the Tlingit language is not very different from its Athabaskan relatives. Like the other Na-Dene languages, Tlingit is a head-marking language with extensive null anaphora. Also like its Athabaskan relatives, Tlingit largely displays a head-final alignment: the language employs postpositions, and no prepositions; possessors and other nominal complements precede the head noun; demonstratives, adjectives, relative clauses, and other nominal modifiers precede the head noun; auxiliary verbs follow main verbs. In addition to this, the most frequent word order in Tlingit texts is typically OV (Dryer 1985). Unlike its Athabaskan relatives, however, Tlingit has a rather free word order, and freely permits the positioning of major constituents after the verb.3 Generally speaking, any permutation of S, V, O is an allowable sentence of Tlingit, although there are of course discourse-structural effects associated with particular orders (see Leer 1991, chapter 2). This freedom of word order is illustrated in (5).4 (5)

Word Order Freedom in Tlingit5 a. SOV Wé shaawátch xóots awsiteen. that woman.ERG bear she.saw.it The woman saw the bear. b. SVO Wé shaawátch wusiteen xóots.6 that woman.ERG she.saw.it bear The woman saw the bear. c. OVS Xóots awsiteen wé shaawátch. bear she.saw.it that woman.ERG The woman saw the bear. d. OSV Xóots wé shaawátch wusiteen. bear that woman.ERG she.saw.it The woman saw the bear. e. VSO Awsiteen wé shaawátch xóots. she.saw.it that woman.ERG bear The woman saw the bear. f. VOS Awsiteen xóots wé shaawátch. she.saw.it bear that woman.ERG The woman saw the bear.

20

THE GRAMMAR OF Q

2.2.4 Methodology Used in This Study 2.2.4.1 The Nature of the Data Many of the illustrative example sentences found throughout this book are taken from published texts, and most of the grammatical generalizations are supported by textual analysis. In many cases we will see that the textual absence of a putative structural type is robust enough to warrant the conclusion that such structures are ill-formed in Tlingit. Thus, for many of the questions we examine, published texts are not simply a source of positive data, but also provide (implicit) negative data. In the course of this study, the following texts were examined: Boas 1917; Dauenhauer and Dauenhauer 1981, 1987, 1990, 1994, 2000, 2002; Naish 1966; Nyman and Leer 1993; Story 1995; Story and Naish 1973; Swanton 1909; Velten 1939, 1944; Williams, Williams, and Leer 1978. In addition to these published texts, I have also been able to examine some currently unpublished material, including Dauenhauer and Dauenhauer (forthcoming) and Edwards (forthcoming), as well as certain unpublished materials archived at the offices of SHI in Juneau, Alaska. Although all these texts were examined and found to be consistent with the grammatical generalizations proposed here, a ‘core’ set of five book-length texts were analyzed to obtain specific word-order counts for a variety of subjects: Dauenhauer and Dauenhauer 1987, 1990, 2000, 2002, and Nyman and Leer 1993. All the textual analysis reported here was carried out entirely by hand; no digitized files were analyzed by algorithmic means. Besides the analysis of texts, data for this project have been obtained through interviews with native speakers of Tlingit. In the course of this project, four native speakers of Tlingit were consulted, three men—David Katzeek (Juneau), John Marks (Juneau), Fred White (Yakutat)—and one woman—Anita Lafferty (Hoonah). All four individuals speak the Northern Dialect of Tlingit. My access to these speakers was facilitated by the SHI, and more specifically by Keri Edwards, who at the time was a linguist on staff with SHI. Interview sessions with these speakers consisted primarily of the following activities, paradigmatic of field linguistics throughout its history (Samarin 1967; Newman & Ratliff 2001). First, speakers were asked to directly translate short English passages, or English sentences supplied with a clarifying context. Subsequently, speakers were often asked to compare the correctness of novel, constructed examples to ones previously uttered by the speakers or categorized by the speakers as correct. I should note that, through their experiences as language educators and in doing extensive linguistic work with Edwards, these speakers were very comfortable with the task of judging the correctness of novel sentences, and usually offered much more extensive commentary and information than simple judgments of ‘correct’ or ‘incorrect’. Besides these sorts of tasks, further information about the Tlingit language was obtained directly from the speakers’ own observations about their language. Speakers would sometimes draw my attention toward other ways of translating a given English sentence, or other interesting structures in which a given word or phrase might appear.


21

These interview sessions included both group sessions and sessions with individual speakers. In general, at a given time I worked with all the people who were available. With the permission of the speakers, all interview sessions were recorded on a Tascam digital 8-track recorder. I later made transcriptions based on these recordings. For each session, each speaker in that session was later given a copy of the recording and of the transcription. Furthermore, copies of every recording and transcription were given to SHI for inclusion in their Tlingit language archives. 2.2.4.2 The Sealaska Heritage Institute Given its central role in the procurement of the data upon which this work is based, I provide here some information regarding SHI, my relationship to the institute, and their official archives. The Sealaska Heritage Institute is a Native nonprofit organization established for the Tlingit, Haida, and Tsimshian people of Southeast Alaska, whose mission is to “perpetuate and enhance Tlingit, Haida, and Tsimshian Cultures” (Sealaska Heritage Institute 2003). For information concerning the history of SHI, I refer the reader to Sealaska Heritage Institute (2003). Since 1997, SHI has adopted as its “foremost priority” the revitalization of the Tlingit, Haida, and Tsimshian languages. As one small part of its many ongoing programs related to language restoration, SHI provides “logistical support and introductions [for] visiting scholars [whose work] may advance the mission of SHI, provided that the Institute may share in the results” (Sealaska Heritage Institute 2003). It is through this visiting scholar program that I was able to conduct work at SHI in June 2005, May 2006, and in the summer of 2007. Following each of my visits to SHI, I composed for the institute a report detailing the results of my study. These reports present the main descriptive results of the study in clear, nontechnical language, as the intended audience of the reports is the staff at SHI, as well as all those who may study the language in the future. The purpose of these reports is to make the primary descriptive results of the study available to the community and to fulfill my obligation to share with SHI the results of the work I conducted as a visiting scholar. The reports are currently on file in the SHI archives at the SHI offices in Juneau, Alaska. These archives are not generally open to the public, but are accessible with the permission of SHI. Access is granted to individuals by SHI on a case-by-case basis.

2.3 The Behavior of Wh-Words in Tlingit Wh-Questions Having provided a basic introduction to the Tlingit language, I will now seek to establish the principle claim of this chapter, that the structure in (1) is the correct analysis of wh-questions in Tlingit. I begin, in this section, by arguing that Tlingit can be described as a ‘wh-fronting language’, in that the wh-words of a Tlingit wh-question obligatorily appear within the left periphery of the clause. Given the paucity of descriptive work on Tlingit

22

THE GRAMMAR OF Q

syntax, this is an original claim regarding the structure of the language’s wh-questions, and so I take care to defend it at length. I begin in section 2.3.1 by introducing various basic properties of Tlingit wh-questions. Then, in section 2.3.2, I outline a variety of word order constraints found operable in Tlingit wh-questions, and argue that they are best explained by the generalization that wh-words in such questions must be located in the left periphery of the clause. 2.3.1 Basic Properties of Tlingit Wh-Questions A wh-question of Tlingit necessarily contains one of its wh-words, which are listed in (6). (6)

The Wh-Words of Tlingit a. Daa What b. Daakw Which c. Aa Who d. Aadóo Who7 e. Goo Where f. Wáa How, why, what8 g. X’oon How much h. Gwátk When (in the past) i. Gwátgeen When (in the future)

The general form of wh-questions in Tlingit is illustrated by the sentences in (7), and schematized by the structure in (8). (7)

Illustrative Examples of Wh-Questions in Tlingit a. Wáa sá sh tudinookw i éesh? how Q he.feels your father How is your father feeling? (Dauenhauer & Dauenhauer 2000: 138) b. Daa sáwé i éesh al’óon? what Q.FOC your father he.hunts.it What is your father hunting? (Dauenhauer & Dauenhauer 2000: 186)

(8)

General Form of a Wh-Question in Tlingit [S . . . [ [ . . . wh-word . . . ] sá ] . . . Main-Predicate . . . .]

The schema in (8) encapsulates the following properties of wh-questions in Tlingit. First, as we will soon see, the wh-word must precede the main predicate of the wh-question, and is typically initial in the clause. Secondly, the wh-word is followed by the Q-particle sá, which either directly follows the wh-word or directly follows a phrase containing the wh-word; that is, the Q-particle must c-command the wh-word. As shown in (7b), this Q-particle can form a portmanteau with the ‘focus particles’ áwé, áyá, áyú, áhé, the two surfacing together as sáwé, sáyá, sáyú, sáhé.9 Finally, the


23

remaining material of the sentence typically follows the wh-word, with a strong tendency to follow the verb. Because of the freedom of word order in Tlingit, it is not obvious upon casual examination whether the language requires wh-words to occupy a left-peripheral position in wh-questions. Indeed, this issue has not yet been addressed in the published grammatical descriptions of Tlingit. Nevertheless, certain facts indicate that such wh-words are left peripheral in Tlingit wh-questions.10 2.3.2 Evidence That Tlingit Is a Wh-Fronting Language In this section I present evidence that Tlingit is a wh-fronting language. In section 2.3.2.1 I demonstrate that the wh-word of a wh-question must precede the main predicate of the clause. In section 2.3.2.2 I demonstrate that material preceding the whword of the wh-question must be interpreted as a discourse topic, and so is likely to occupy a left-peripheral discourse-structural position. In section 2.3.2.3 I demonstrate that long-distance questions in Tlingit require fronting of the wh-word to a position before the matrix predicate. In section 2.3.2.4 I demonstrate that multiple wh-questions in Tlingit are subject to Superiority Effects. Taken together, I conclude that these facts effectively show Tlingit to be a whfronting language, in that the wh-words of its wh-questions must occupy left-peripheral positions. 2.3.2.1 Obligatory Prepredicate Position of Wh-Operators in Wh-Questions As we saw earlier, word order in Tlingit is generally free, and any permutation of S, V, and O is a well-formed sentence. In a Tlingit wh-question, however, the phrase understood to be the wh-operator must appear left of the main predicate of the clause.11 By the term “predicate” here, I mean either the verb of the clause (if one is present) or the so-called focus particles áwé, áyá, áyú, áhé in their ‘copular use’. Examples of copular use of a focus particle are given in the sentences in (9). (9)

Copular Use of the Focus Particles a. Tás áyá. thread FOC This is thread. (Dauenhauer & Dauenhauer 2000: 77) b. Daa sáwé? what Q.FOC What is that? (Dauenhauer & Dauenhauer 2000: 77)

The requirement that a Tlingit wh-operator precede the predicate is apparent both from patterns within published texts and from the well-formedness judgments of native speakers. The table in (10) demonstrates how this pattern emerges across a range of published texts.

24

THE GRAMMAR OF Q

(10) The Prepredicate Position of Wh-Operators in Wh-Questions

Text

Wh-questions containing an overt predicate

Of such sentences, those in which wh-operator precedes the predicate

Dauenhauer & Dauenhauer 1987 Dauenhauer & Dauenhauer 1990 Dauenhauer & Dauenhauer 2000 Dauenhauer & Dauenhauer 2002 Nyman & Leer 1993 TOTAL

117 31 170 84 114 516

117 31 170 84 114 516

As the table shows, all the wh-questions in the selected corpus containing an overt predicate place the wh-operator before the predicate. This pattern is also confirmed by the grammaticality judgments offered by native speakers. As the following data show, speakers reject as ill-formed any wh-question where the wh-operator follows the main predicate. Such sentences are consistently corrected by speakers to ones in which the wh-operator precedes the predicate. (11)

Wh-Operators in Tlingit Must Precede the Main Predicate yá x’úx’? a. Aadóoch sá kgwatóow who.ERG Q he.will.read.it this book Who will read this book? b. Aadóoch sá yá x’úx’ akwgwatóow? who.ERG Q this book he.will.read.it c. Yá x’úx’ aadóoch sá kgwatóow? this book who.ERG Q he.will.read.it d. * Yá x’úx’ akwgwatóow aadóoch sá? this book he.will.read.it who.ERG Q

(12)

a. Aadóoch sá kawshixít who.ERG Q he.wrote.it Who wrote this book? b. Yá x’úx’ aadóoch sá this book who.ERG Q c. * Yá x’úx’ akawshixít this book he.wrote.it

yá x’úx’? this book kawshixít? he.wrote.it aadóoch sá? who.ERG Q

(13)

a. Aadóoch sá ax sakwnéini aawaxáa? who.ERG Q my bread he.ate.it Who ate my bread? b. Ax sakwnéini aadóoch sá uwaxáa? my bread who.ERG Q he.ate.it c. * Ax sakwnéini aawaxáa aadóoch sá? my bread he.ate.it who.ERG Q

(14)

a. Daa sá kéet axá? what Q killer whale he.eats.it What do killer whales eat?


25

b. Kéet daa sá axá? killer whale what Q he.eats.it c. * Kéet axá daa sá? killer whale he.eats.it what Q (15)

a. Wáa sáyá at kuwanóok? how Q.FOC they.do.it What are those people doing? b. * At kuwanóok Wáa sáyá? they.do.it how Q.FOC

Of course, one might justifiably wonder whether the ill-formedness of the starred sentences above is due not to a rule of obligatory wh-fronting, but to independent semantic conditions on postpredicate NPs. Perhaps postpredicate NPs must possess qualities that wh-words inherently lack, such as definiteness? Recall, however, that wh-words in Tlingit can function as indefinites in declarative clauses (chapter 1). When a wh-word is used as an indefinite, it is not required to appear before the predicate of the clause. This fact is clearly indicated both by textual examination and by the well-formedness judgments of native speakers. The table in (16) demonstrates that the selected corpus of texts supports this grammatical generalization. (16)

Wh-Indefinites May Freely Follow the Main Predicate of the Clause

Text

Sentences containing wh-indefinite and an overt predicate

Of such sentences, those in which wh-indefinite precedes the predicate


74 26 0 6 205 311

63 24 0 6 187 280

As the table shows, not all wh-indefinites in the selected corpus precede the main predicate of their clause. The following two sentences illustrate these textually attested cases of postpredicative wh-indefinites. (17)

Textual Examples of Postpredicative Wh-Indefinites haa uwaxée x’oon sákwshéwé. a. Áa there we.spend.night how.much Q.DUB We stayed there I don’t know how long. (Nyman & Leer 1993: 176; line 190) s kaawashoo daat yáx sá. b. K’e EXCLM they.got.drunk what like Q Well, they got really drunk. (Nyman & Leer 1993: 176; line 591)

This pattern is also confirmed by comments offered by native speakers. Although sentences such as (11d) and (14c) are not acceptable as wh-questions, speakers note that they can function as declarative sentences containing wh-indefinites.

26 (18)

THE GRAMMAR OF Q

Postpredicative Wh-Indefinites12 a. Yá x’úx’ akwgwatóow aadóoch sá. this book he.will.read.it who.ERG Q People will read this book. axá daa sá. b. Kéet killer.whale he.eats.it what Q A killer whale will eat anything. c. Yéi uwatee x’oon táakw sá. he.lived.there how.many winters Q He lived there for a number of years (= many years).

We see, then, that there is no condition requiring wh-indefinites in Tlingit to appear before the main predicate of the clause; such wh-words may freely appear in the postverbal field. I conclude that the inability for wh-operators in wh-questions to appear following the predicate is not due to their lacking some inherent semantic property that postpredicate NPs are required to have. Indeed, the only relevant difference between the wh-words in (16)–(18) and those in (11)–(15) is that the latter function as wh-operators, while the former do not. I conclude that the best explanation for the requirement that wh-operators appear before the predicate of the clause is that such wh-words are fronted into the left periphery. Further evidence for this generalization will be provided in the next few sections. 2.3.2.2 Topic Status of Material Preceding Wh-Operators in Wh-Questions Additional evidence that wh-operators are left peripheral in Tlingit wh-questions may be found in the discourse-structural properties of material preceding such wh-words. As shown by sentences like (11c), it is possible for other XPs to precede the wh-word in a Tlingit wh-question. Placement of an XP before the wh-word, however, creates a structure with special discourse properties: the fronted XP must be construed as a discourse topic. This is suggested both by textual examination and by speaker judgments. Although often accepted by speakers, sentences such as (11c) are remarkably rare in texts. Indeed, the overwhelmingly predominant pattern is for wh-words in wh-questions to precede all other major constituents in the sentence. The table in (19) illustrates. (19)

The Initial Position of Wh-Words in Tlingit Wh-Questions

Text

Wh-questions containing wh-word and a second, major, Of these, those in nonpredicate which the wh-word phrase is initial in the clause

Dauenhauer & Dauenhauer 1987 43 Dauenhauer & Dauenhauer 1990 21 Dauenhauer & Dauenhauer 2000 27 Dauenhauer & Dauenhauer 2002 18 Nyman & Leer 1993 58 TOTAL 167

43 20 27 18 58 166

Of these, those in which the initial position of wh-word does not follow from typical word order 32 11 19 8 44 114


27

In this table, the left-most column indicates the number of wh-questions in the corpus containing some major constituent besides the wh-word and the predicate. The middle column reports how many, from the questions represented in the left column, place the wh-word initially in the clause. Finally, the right column indicates the number of questions in the middle column in which the initial position of the whword does not follow from more general word-order frequencies in Tlingit, such as the fact that subjects tend to precede objects in the language (Dryer 1985). The totals at the bottom indicate an overwhelming preference for wh-questions to begin with wh-words. Consonant with their textual rarity, sentences like (11c) are occasionally judged by speakers to be marginal or ill-formed, a classification that is sometimes revised upon further reflection. The textual rarity of sentences like (11c) would, of course, follow from their possessing special discourse properties, ones that place strong limits on the kind of context in which such structures might be embedded. Such special discourse properties would also account for their occasional rejection by speakers, rejection occurring when the licensing context is difficult for the speaker to imagine or strikes them as far-fetched. It seems likely, then, that sentences like (11c) possess some special discoursestructural property. That this property is the ‘topichood’ of the material preceding the wh-word comports well with a number of other facts. First, in all the naturally occurring instances of noninitial wh-operators I have encountered, the material preceding the wh-operator is a referential expression. The following two examples illustrate the general pattern. (20)

Textually Attested Examples of the Order [ XP . . . Wh-Operator . . . V ] sée daakw aa sáwé? a. Wé i that your daughter which of.them Q.FOC Which one is your daughter? (Dauenhauer & Dauenhauer 1990: 298; line 10) b. I kutaaní wáa sá wootee? your summer how Q it.was How was your summer? (Sealaska Heritage Institute; Tlingit Phrase of the Week; September 6, 2005)13

Note that this pattern is also evident in sentences (11c), (12b), (13b), and (14b).14 Indeed, speakers do not allow fully nonreferential material to precede the wh-operator of a wh-question. (21)

Nonreferential DPs Cannot Precede Wh-Operators sáyá l daa sá uxá? a. Aa who Q.FOC nothing he.eats.it Who ate nothing? b. * L daa sá aa sáyá uxá? nothing who Q.FOC he.eats.it

These data indicate that only referential XPs may precede the wh-operator of a wh-question. Of course, one of the core properties of ‘topics’ is that they can only be

28

THE GRAMMAR OF Q

denoted by referential expressions (Li 1976), and so these data support the notion that any material preceding the wh-operator of a Tlingit wh-question must be construed as a discourse topic. A final suggestive piece of evidence is the translations offered by speakers for sentences like (11c). When these sentences are accepted by native speakers, they are regularly translated into English using hanging topic left dislocation structures, such as (22). (22) The Order [ XP . . . Wh-Operator . . . V ] Translated as Left Dislocation daa sá aawaxáa? a. Ax éesh my father what Q he.ate.it Translated as ‘My father, though, what did he eat?’ sá uwaxáa? b. Yá xáat aadóoch this fish who.ERG Q he.ate.it Translated as ‘That fish—who ate it?’ c. Yá x’úx’ aadóoch sá kgwatóow? this book who.ERG Q he.will.read.it Translated as ‘This book—who will read it?’

That speakers use English left dislocation to translate these sentences supports their having a special discourse structure that is not possessed by a simple wh-question and that only left dislocation in English is able to simulate.15 There is, then, reason to conclude that any material preceding the wh-operator of a Tlingit wh-question must be interpreted as a discourse topic. This fact itself would most naturally follow from a syntax in which wh-operators are fronted into the left periphery of Tlingit wh-questions. Under such a syntax, any material occurring to the left of a Tlingit wh-operator would either have to occupy a left-peripheral topic position (Rizzi 1997), or else would have to simply be a dislocated, hanging topic. 2.3.2.3 Long-Distance Questions in Tlingit Require Long-Distance Movement A third argument that wh-operators in Tlingit undergo obligatory fronting may be found in the language’s long-distance questions. In Tlingit long-distance questions, the subordinate clause preferably follows the verb it is complement to (see (23a)), though a preverbal order is also possible (see (23b)). (23)

Long-Distance Wh-Questions in Tlingit a. Daa sá oowajée wutoo.oowú? what Q he.thinks we.bought.it What does he think we bought? b. Daa sá wutoo.oowú oowajée? what Q we.bought.it he.thinks

For obvious reasons, the activity of an obligatory wh-fronting rule in Tlingit long-distance questions is easiest to detect when the subordinate clause follows the


29

main verb. In such sentences, the interrogative word must appear to the left of the main verb, and cannot appear downstairs in its base position. (24)

Long-Distance Movement in Tlingit Long-Distance Questions tuwáa sigóo [ t1 yéi ysaneiyí ] ?16 a. [ Daa sá ]1 i what Q your spirit it.is.glad you.do.it What do you want to do? b. * I tuwáa sigóo [ daa sá yéi ysaneiyí ] ? your spirit it.is.glad what Q you.do.it

(25)

a. [ Daa sá ]1 haa kóo at latóowu haa yawsikaa [ t1 wutootoowú ] ? what Q our teacher he.told.us we.read.it What did our teacher tell us to read? b. * Haa kóo at latóowu haa yawsikaa [ daa sá wutootoowú ] ?17 our teacher he.told.us what Q we.read.it

(26)

a. [ Goodéi sá ]1 i shagóonich has oowajée [ t1 wutoo.aadí ] ? where.to Q your parents.ERG they.think we.went Where do your parents think that we went? b. * I shagóonich has oowajée [ goodéi sá wutoo.aadí ] ? your parents.ERG they.think where.to Q we.went

The impossibility of the (b)-sentences in (24)–(26) indicates that wh-operators in Tlingit must be fronted into the left periphery of the wh-question. 2.3.2.4 Superiority Effects in Multiple Wh-Questions A final piece of evidence for wh-fronting in Tlingit wh-questions comes from the language’s multiple wh-questions. As we saw earlier in (5), word order in Tlingit declarative clauses is rather free; both objects and adverbial phrases are generally permitted to precede subjects in a Tlingit declarative clause. In multiple wh-questions, however, such relative freedom of order is not available. Interrogative subjects must obligatorily precede interrogative objects and adverbial phrases. (27)

Superiority Effects in Tlingit Multiple Wh-Questions sá daa sá aawaxáa? a. Aa who Q what Q they.ate.it Who ate what? sá aa sá aawaxáa? b. * Daa what Q who Q they.ate.it

(28)

a. Aa sá goodéi sá woogoot? who Q where.to Q they.went Who went where? b. * Goodéi sá aa sá woogoot? where.to Q who Q they.went

(29)

a. Aa sá Wáa sá kuyawsikaa? who Q how Q they.said.to.someone Who said what?

30

THE GRAMMAR OF Q

b. * Wáa sá aa how Q who

sá kuyawsikaa? Q they.said.to.someone

It thus appears that in a Tlingit multiple wh-question, a wh-word subject must precede any wh-word objects or adverbs. This otherwise mysterious requirement would, of course, follow naturally from the Superiority Condition (Kuno & Robinson 1972; Chomsky 1973), but only under the assumption that Tlingit wh-words undergo obligatory fronting in wh-questions. I conclude, then, that the apparent activity of the Superiority Condition in Tlingit multiple wh-questions provides further evidence that wh-operators in Tlingit obligatorily front to the left periphery of the clause.

2.4 Q-Particles in Tlingit Wh-Questions: The Formal Status of Sá I conclude from the grammatical patterns described in the previous section that the wh-operator of a Tlingit wh-question must occupy a left peripheral position within the clause. In this section I argue that the Tlingit particle sá is most plausibly categorized as a Q-particle. This argument, however, must be rather indirect, as there are no widely accepted diagnostics for applying the term ‘Q-particle’. I will therefore argue that sá is a Q-particle on the basis of its similarity to the particle da in Sinhala and the particle ka in Japanese. Given that da and ka are uncontroversial instances of Q-particles, the overwhelming parallels between sá, da, and ka will demand that sá receive the same categorization. In section 2.4.1 I demonstrate that sá, da, and ka share the property that they must appear both with wh-operators in wh-questions and wh-indefinites in declarative clauses. In section 2.4.2 I show that these particles all must c-command the wh-word they are paired with. In section 2.4.3 I show that they all have the property that they cannot be separated from the edge of the clause by a syntactic island. Finally, in section 2.4.4 I show that sá and da share the property that they cannot appear at the end of matrix clauses, although they can appear clause-finally in subordinate clauses. 2.4.1 The Obligatory Presence of Sá A wh-question in Tlingit must contain the particle sá. If this particle is removed from any of the previous sentences, the result is ill-formed. (30) The Obligatory Presence of Sá in Tlingit Wh-Questions éesh? a. Daa *(sá) aawaxáa i what Q he.ate.it your father What did your father eat? b. Goodéi *(sá) kkwagóot? where.to Q I.will.go Where will I go?

As in many languages, wh-words in Tlingit may also function as indefinites. When they do, the particle sá is still obligatory.


(31)

31

The Obligatory Presence of Sá with Tlingit Wh-Indefinites Tlél goodéi *(sá) xwagoot. not where.to Q I.went I didn’t go anywhere.

The data in (31) demonstrate that sá is required not only by the interrogative force of the clause, but by the wh-word itself. Although this may seem to undercut the force of the label “question particle”, this property also holds for such prototypical ‘Q-particles’ as Japanese ka and Sinhala da.18, 19 (32)

The Obligatory Presence of Da in Sinhala Wh-Questions and Wh-Indefinites a. Chitra monawa *(da) gatte? Chitra what Q bought What did Chitra buy? (Kishimoto 2005: 3–4) b. Mokak *(da) waetuna. what Q fall Something fell. (Hagstrom 1998: 23)

(33)

The Obligatory Presence of Ka in Japanese Wh-Questions and Wh-Indefinites a. John-ga nani-o kaimasita *(ka)?20 John-NOM what-ACC bought.polite Q What did John buy? b. John-ga nani-*(ka)-o katta. John-NOM what-Q-ACC bought John bought something.

The data in (32) and (33) led Hagstrom (1998) to develop a semantics for Q-particles under which they are expected to appear both within wh-questions and with wh-words interpreted as indefinites in declarative clauses. We will see in section 2.7 that this semantic analysis may, with minor modification, be extended to the Tlingit particle sá, and would similarly predict its parallel grammatical behavior. Such a shared semantics would constitute one reason to apply the label “Q-particle” to Tlingit sá.21

2.4.2 The Structural Position of Sá As can be seen from most of the previous sentences, it is common for the particle sá to be located directly to the right of a wh-word. However, this particle can also appear further to the right, detached from the interrogative word. This is evident from sentences such as (17b), (18c), (20a), and (26a). More examples illustrating such rightward positioning of sá appear in (34). (34)

Tlingit Sá Separated From the Wh-Word a. [ Goodéi ] sá kkwagóot? where.to Q I.will.go Where will I go to?

32

THE GRAMMAR OF Q

b. [ Aadóo yaagú ] sá ysiteen? who boat Q you.saw.it Whose boat did you see? c. [ Daakw keitl ] sá asháa? which dog Q it.barks Which dog is barking? gushé [ x’oon k’óox ] sáyú has aawaják. d. Tle then DUB how.many marten Q.FOC they.killed.them I don’t know how many marten they killed. (Nyman & Leer 1993: 56) e. [ Daat tlein ] sáwé tsú wéix yaa nagút. what big Q.FOC too there.at it.is.walking There was something large walking along over there. (Leer 1993: 17) shagóonich?22 f. [ Goodéi wugootx ] sá has oowajée i where.to he.went Q they.think your parents.ERG Where do your parents think that he went?

Upon examination of just the sentences in (34), one might form the simple hypothesis that the particle sá can be freely placed anywhere to the right of the interrogative word. Although this would be the simplest conclusion, the ill-formedness of sentences (35b) and (36b) demonstrates that it cannot be correct. Rather, the correct generalization is that sá must appear either directly to the right of the wh-word, or directly to the right of a phrase containing the wh-word. In other words, the particle sá has to c-command the wh-word. (35)

(36)

Tlingit Sá Must C-Command the Wh-Word a. [ Aadóo jeet ] sá wé sakwnéin aawatee? who hand.to Q that bread he.brought.it Who did he give the bread to? b. * [ Aadóo jeet ] wé sakwnéin sá aawatee? who hand.to that bread Q he.brought.it a. [ Goodéi ] sá has oowajée wugootx i shagóonich? where.to Q they.think he.went your parents.ERG Where do your parents think he went? wugootx sá i shagóonich? b. * [ Goodéi ] has oowajée where.to they.think he.went Q your parents.ERG

The condition that the Q-particle c-command the wh-word also holds for Sinhala da and Japanese ka. This c-command condition is stated explicitly by Kishimoto (2005: 13) for Sinhala da and by Yatsushiro (2001: 182) for Japanese ka. The sentences in (37) illustrate structures where the Sinhala particle da appears detached from its associated wh-word.23 (37)

Sinhala Da Separated From the Wh-Word (Kishimoto 2005: 13) a. Chitra [ mona pota ] da gatte? Chitra what book Q bought What book did Chitra buy?


33

b. Chitra [ kaa-ge amma ] da daekke? Chitra who-GEN mother Q saw Whose mother did Chitra see? c. Chitra [ kauru ekka ] da kataa kalee? Chitra who with Q talk did Who did Chitra talk with?

Such identity of distribution further emphasizes the formal similarity between sá, da, and ka. Moreover, it will be shown in section 2.7 that this apparently syntactic condition on the placement of these particles follows from a particular semantic theory of Q-particles and wh-words. 2.4.3 Q-Particles and Extraction Islands One of the most intriguing similarities between Tlingit sá and Sinhala da concerns their behavior with respect to islands. As described in Hagstrom (1998) and Kishimoto (2005), the wh-operator of a Sinhala wh-question may be contained inside an island if and only if the Q-particle da is merged outside the island. In the case of relative clause islands, the Q-particle must be merged to the right of the head of the relative clause, illustrated by (38).24 (38) Interaction Between Q-Particle and Relative Clause Islands in Sinhala a. Oyaa [ [ Chitra kaa-ta dunna CP] pota NP] da kieuwe? you Chitra who-DAT give book Q read Who did you read the book that Chitra gave? b. * Oyaa [ [ Chitra kaa-ta da dunna CP] pota NP] kieuwe? you Chitra who-DAT Q give book read (Kishimoto 2005: 29)

The same condition can be observed in Tlingit. The wh-operator of a Tlingit whquestion may be contained inside an island if and only if the particle sá is merged outside the island. When this occurs, the entire island is pied-piped into the left periphery of the interrogative clause. In the case of relative clause islands, the particle sá must be merged to the right of the head of the relative clause. (39)

Interaction Between Q-Particle and Relative Clause Islands in Tlingit tuwáa sigóo?25 a. [ [ Wáa kwligeyi CP] xáat NP] sá i how it.is.big.REL fish Q your spirit it.is.happy How big a fish do you want? (= A fish that is how big do you want?) b. * [ [ Wáa sá kwligeyi CP] xáat NP] i tuwáa sigóo? how Q it.is.big.REL fish your spirit it.is.happy c. * [ [ Wáa kwligeyi CP] sá xáat NP] i tuwáa sigóo? how it.is.big.REL Q fish your spirit it.is.happy

(40)

sháax’w sáani NP] sá ash koodlénxaa? a. [ [ Wáa yateeyí CP] how they.are.REL girls Q they.are.tempting.him What kind of girls are tempting him? (= Girls that are how are tempting him?)

34

THE GRAMMAR OF Q

b. * [ [ Wáa sá yateeyí CP] sháax’w sáani NP] ash koodlénxa? how Q they.are.REL girls they.are.tempting.him c. * [ [ Wáa yateeyí CP] sá sháax’w sáani NP] ash koodlénxa? how they.are.REL Q girls they.are.tempting.him (41)

a. [ [

Wáa yateeyí CP] sháax’w sáani NP] sá ash tuwáa gaa yatee? how they.are.REL girls Q his.spirit for they.are What kind of girls are pleasing to his eye? (= Girls that are how are pleasing to his eye?) sháax’w sáani NP] ash tuwáa gaa yatee? b. * [ [ Wáa sá yateeyí CP] how Q they.are.REL girls his.spirit for they.are sá sháax’w sáani NP] ash tuwáa gaa yatee? c. * [ [ Wáa yateeyí CP] how they.are.REL Q girls his.spirit for they.are

The speaker judgments indicated in (39)–(41) are consistent with the available textual data as well. The table in (42) demonstrates that the selected corpus of texts supports this generalization. (42)

The Interaction Between Q-Particle and Relative Clause

Text

Wh-questions where wh-operator is inside a relative clause

Of these, those in which sá appears to the right of the NP modified by the relative


4 6 4 1 5 20

4 6 4 1 5 20

As the totals here clearly indicate, whenever a wh-operator in the selected corpus is contained inside a relative clause, the particle sá appears to the right of both the relative clause and the noun it modifies. Thus the selected corpus supports the generalization that in Tlingit wh-questions, the wh-operator can be located inside of an island if and only if the Q-particle sá is outside the island. To further illustrate the content of this generalization, the following are some textually attested examples of the Tlingit structure witnessed in (39)–(41); note that the relative clauses in the structures below are ‘reduced relatives’. (43)

Interaction Between Q-Particle and Relative Clause Islands in Tlingit a. [ [ Daat yís ] át ] sákwshéiwégé? what for thing Q.DUB Literally: ‘A thing for what is this?’ (Nyman & Leer 1993: 120) b. [ [ Goodáx ] k’anáaxán tlein ] sáyá du kát satéen? where.from fence big Q.FOC its surface.to placed(?) Literally: ‘A big fence from where was placed on it?’ (Nyman & Leer 1993: 150) káa ] sáyá yéi yatee? c. [ [ Goodáx ] where.from man Q.FOC he.is Literally: ‘A man from where was he?’ (Dauenhauer & Dauenhauer 1987: 168)


35

Both the speaker judgments and the textual data above further emphasize the syntactic parallels between Tlingit sá and Sinhala da. In section 2.5 we will see that a uniform syntactic account can be provided for these facts, just so long as both these particles share a syntactic categorization as Q-particles. 2.4.4 Q-Particles at the Right Edge of the Matrix Clause One final parallel between Sinhala da and Tlingit sá is that neither particle may freely appear at the right edge of the matrix clause, following the matrix predicate.26, 27 This generalization is illustrated for Sinhala da in (44). (44)

Sinhala Da Cannot Appear at the Right Edge of a Matrix Clause (Kishimoto 2005) a. Chitra monawa da gatte? Chitra what Q buy What did Chitra buy? b. * Chitra monawa gatta da? Chitra what buy Q (Kishimoto 2005: 3–4)

The speaker judgment data in (45)–(47) demonstrate that this property also holds for Tlingit sá. (45)

Tlingit Sá Cannot Appear at the Right Edge of a Matrix Clause a. Daa sá iyatéen? what Q you.can.see.it What can you see? b. * Daa iyatéen sá? what you.can.see.it Q

(46)

a. Aadóo sá xáat aawaxáa? who Q fish he.ate.it Who ate fish? b. * Aadóo xáat aawaxáa sá? who fish he.ate.it Q

(47)

a. Wáa sá ituwatee? how Q you.feel How do you feel? b. * Wáa ituwatee sá? how you.feel Q

Despite the inability for da and sá to appear at the right edge of matrix clauses, both particles may freely appear at the right edge of subordinate clauses, following a subordinated predicate. This is illustrated for Sinhala da below. (48)

Sinhala Da Can Appear at the Right Edge of a Subordinate Clause Ranjit [ kauru aawa kiyala ] da danne? Ranjit who came that Q know Who does Ranjit know came? (Kishimoto 2005: 13)

36

THE GRAMMAR OF Q

Furthermore, sentences like (49) illustrate that this property also holds of Tlingit sá. (49) Tlingit Sá Can Appear at the Right Edge of a Subordinate Clause [ Goodéi wugootx ] sá has oowajée i shagóonich? where.to he.went Q they.think your parents.ERG Where do your parents think that he went?

Thus we find that both Sinhala da and Tlingit sá share the property that, although they can appear at the right edge of subordinate clauses, they cannot appear at the right edge of a matrix clause. Again, it will be shown in section 2.6 that a uniform account can be provided for these facts, but only if Tlingit sá and Sinhala da are assumed to be the same formal entity, a Q-particle. 2.4.5 Summary: The Formal Unity of Tlingit Sá, Sinhala Da, and Japanese Ka To summarize, we have seen in the preceding sections that the Tlingit particle sá, the Sinhala particle da, and the Japanese particle ka all share the following properties. (50) Properties Shared Between Sá, Da, and Ka i) Obligatory in wh-questions; more generally, required in clauses containing a wh-word. ii) Must c-command the wh-word. iii) In matrix wh-questions, cannot appear inside islands. If merged outside any islands, can ‘save’ matrix wh-questions where the wh-word is located inside an island. iv) Cannot appear at the right of the matrix clause (does not hold for Japanese ka). v) May appear at the right of subordinate clauses. I conclude that Tlingit sá should most likely receive the same analysis as Sinhala da and Japanese ka. As these latter two particles are classically glossed as ‘Q-particles’ (Hagstrom 1998; Kishimoto 2005), I conclude that Tlingit sá should therefore also be glossed as a Q-particle.

2.5 Wh-Fronting in Tlingit as a Consequence of Q-Movement In the preceding sections we have seen that (i) wh-operators occupy a left-peripheral position in Tlingit wh-questions, and that (ii) wh-words in Tlingit are obligatorily c-commanded by a Q-particle. In this section I will argue that the left-peripheral position of these wh-operators is ultimately due to the movement of the Q-particle that c-commands them. That is, I will argue that Tlingit does not actually possess a rule of ‘wh-fronting’ per se. Rather, the generalization that the wh-operator is fronted in a Tlingit wh-question is merely an epiphenomenal consequence of an operation that targets the Q-particle.


37

However, before I can present evidence for this proposal, earlier summarized in (1), I must clarify a number of technical points. The analysis sketched in (1) is vague on several details. This informality was intended both to ease the discussion and to show that the core proposals rely on few architectural assumptions, and so can be exported into a variety of more specific frameworks. However, to make our argumentation here more rigid, we must adopt a more precise statement of our syntactic hypotheses than what appears in (1). In this context, it will help to begin with some recent proposals concerning wh-in situ languages, which will provide our own proposals with some clarifying context. In recent work, an operation of ‘Q-movement’ has been argued to underlie the formation of wh-questions in several wh-in situ languages. For example, various lines of evidence lead Hagstrom (1998) and Kishimoto (2005) to propose the following analysis of wh-questions in Sinhala, couched in the framework of ‘probes’ and ‘goals’, as developed within the Minimalist Program. (51)

Hagstrom (1998) and Kishimoto (2005)’s Analysis of Sinhala Wh-Questions CP CP IP

Q1 CQ Agree / Attract

XP Adjunction XP

…wh-word…

Q1 Covert Movement

The structure in (51) represents the following claims. As we have seen, a wh-word in Sinhala is obligatorily c-commanded by a Q-particle, the particle da. Under the analysis in (51), this particle is adjoined to a phrase containing the wh-word. Furthermore, under this analysis it is the Q-particle da—not the wh-word—which is probed by and Agrees with the interrogative C head of the wh-question. More concretely, the interrogative C head bears an uninterpretable instance of the interpretable Q-feature borne by da. The interrogative C must therefore probe for an interpretable instance of the Q-feature. Upon reaching the adjoined Q-particle, the interrogative C Agrees with the particle, eliminating its own uninterpretable instance of Q. This Agreement then triggers movement of the goal, the Q-particle da, into the projection of C. Because the Q-particle is adjoined to its sister, it may freely detach from its base position. Therefore its movement into the CP, which in Sinhala is typically (though not always) covert, leaves the wh-word and the phrases containing it in their base positions at LF.28 Other lines of evidence lead Hagstrom (1998) to extend the ideas underlying this analysis to wh-questions in Japanese. Hagstrom proposes (52) as the derivation of wh-questions in Japanese.

38

THE GRAMMAR OF Q

(52)

Hagstrom’s (1998) Analysis of Japanese Wh-Questions CP CP IP

CQ

XP

Agree / Attract

Adjunction XP

Q1

Q1 Overt Movement

…wh-word…

Under this analysis, wh-questions in Japanese are essentially identical to those in Sinhala. Their sole difference is that the Q-particle ka in Japanese always moves overtly into the projection of the C, leaving the wh-word and phrases containing it behind. In both languages, however, interrogative C bears a syntactic relationship only with the Q-particle c-commanding the wh-operator. In this context, let us consider again the conclusions of the previous two sections: (i) the Tlingit particle sá is formally identical to Sinhala da and Japanese ka, and (ii) in Tlingit wh-questions, the wh-operator and its Q-particle must appear in the left periphery of the clause. With these facts as background, the structure in (53) immediately suggests itself as an analysis of Tlingit wh-questions. (53)

Fronting of Wh-Word in Tlingit Wh-Question as a Secondary Effect of Q-Movement CP

QP1

CP

Complementation XP

Q CQ

… wh-word…

IP

Agree/ Attract QP1 Overt Movement

Under this analysis, the wh-questions of Tlingit receive a derivation nearly identical to those of Japanese. The principal difference is that, in Tlingit, the Q-particle sá is not adjoined to the phrase containing the wh-operator. Rather, it takes that phrase as complement, thus projecting the category of the phrase minimally containing the Q-particle and its sister. As a projection of Q, it would be natural to assume that this QP also bears the Q-feature probed for by the interrogative C. Furthermore, because this QP properly contains the Q-particle, it is the first node bearing the Q-feature to be


39

probed by the interrogative C. The standard algorithm for probing therefore entails that interrogative C in Tlingit must Agree with this QP projection. As before, this Agreement requires the goal—in this case, the QP—to move into the projection of the interrogative C. Thus the entire QP is attracted into the left periphery of the wh-question. Since this constituent necessarily contains the wh-operator of the wh-question, it follows that such wh-words must occupy left-peripheral positions in wh-questions.29 We see, then, that the analysis in (53) links together the wh-questions of Tlingit, Sinhala, and Japanese in a typology of wh-question formation. Besides this, there are a number of empirical considerations that support the analysis in (53) for Tlingit wh-questions. First, it should be noted that Tlingit wh-questions require both the wh-word and the Q-particle to be fronted. For example, sentence (54a) becomes ill-formed if sá is left downstairs in its base position, as in (54b). Note that the ill-formedness of (54b) is not due simply to a condition against the ‘stranding’ of sá. Such a condition, after all, would not alone rule out the ill-formed (54c). In sentence (54c), the Q-particle sá is not ‘stranded’ since its complement is the unmoved subordinate CP, a possibility that is independently witnessed in sentences like (54d). (54)

No Fronting of Wh-Word Alone a. [ [ Goodéi sá ]1 [ has oowajée [ t1 wugootx ] i shagóonich ] ]? where.to Q they.think he.went your parents.ERG Where do your parents think he went? b. * [ Goodéi1 [ has oowajée [ t1 sá wugootx ] i shagóonich ] ]? where.to they.think Q he.went your parents.ERG c. * [ Goodéi1 [ has oowajée [ t1 wugootx sá ] i shagóonich ] ]? where.to they.think he.went Q your parents.ERG d. [ [ Goodéi wugootx sá ]1 [ has oowajée t1 i shagóonich ] ]? where.to he.went Q they.think your parents.ERG Where do your parents think he went?

Moreover, we see below that the well-formed (55a) becomes ill-formed if the particle sá is fronted without the wh-word, as in (55b). The ill-formedness of (55b) is not simply due to a (prosodic) condition that sá follow some phrasal material in the sentence, as (55c) illustrates. (55)

No Fronting of Q-Particle Alone éesh aawaxáa? a. Daa sá i what Q your father he.ate.it What did your father eat? b. * Sá i éesh daa aawaxáa? Q your father what he.ate.it éesh sá daa aawaxáa? c. * I your father Q what he.ate.it

Importantly, all the data in (54) and (55) follow naturally from the analysis in (53). Under this analysis, a Tlingit wh-question must have the Q-particle sá within its left periphery, thus ruling out (54b, c). Furthermore, the fronting of the QP necessarily brings with it the wh-word associated with sá, as that wh-word is contained within the QP. Thus sentences (55b, c) are also ruled out under the analysis.

40

THE GRAMMAR OF Q

Of course, one might conclude from (54) and (55) that both the wh-operator and the Q-particle are directly attracted into the left periphery, perhaps by separate heads. (56)

Wh-Operator and Q-Particle Both Attracted, but by Separate Heads CP DP1 Daa

CP C

CP Q2 sá

CP C

IP

i éesh t1

t2 aawaxáa?

An immediate problem for the structure in (56), however, arises in the context of multiple wh-questions. Sentences like (57a, b) demonstrate that all the wh-words of a Tlingit multiple wh-question may front together into the left periphery of the clause. (57) Multiple Wh-Fronting in Tlingit Multiple Wh-Questions a. [CP [Aadóo sá ]1 [daa sá ]2 [IP t1 yéi oowajée [t2 du jee yéi teeyí ]] ] ? who Q what Q they.think their hand.at it.is.there Who thinks they have what? b. [CP [Aa sá]1 [daa sá]2 [IP du1 tuwáa sigóo [t2 wutoo.oowú ] ] ] ? who Q what Q their spirit it.is.glad we.bought.it Who wants us to buy what?

We can also see from (57a, b) and (27)–(29) that the order of wh-words and Q-particles in multiple wh-questions is such that each Q-particle immediately follows the wh-word it is associated with. Therefore, if there were separate C heads attracting wh-words and Q-particles of Tlingit, then the left periphery of a Tlingit multiple wh-question must appear as in (58), where the CWH heads are those attracting wh-words and the CQ heads are those attracting Q-particles. (58) Structure Required for Multiple Wh-Fronting, Under the Analysis in (56) [CP CWH-1 [CP CQ1 [CP CWH-2 [CP CQ2 . . . ] ] ] ]

We have already seen that the order of wh-words in a Tlingit multiple wh-question is constrained by superiority; as shown again in (59), wh-subjects must precede wh-objects. (59)

Multiple Wh-Fronting Constrained by Superiority a. * [CP [Daa sá ]2 [aadóo sá ]1 [IP t1 yéi oowajée [ t2 du jee yéi teeyí ] ] ]? what Q who Q they.think their hand.at it.is.there


b. * [CP [ Daa sá ]2 [ aa what Q who

41

sá ]1 [IP du1 tuwáa sigóo [ t2 wutoo.oowú ] ] ] ? Q their spirit it.is.glad we.bought.it

However, if the left periphery of a multiple wh-question has the structure in (58), then the simplest algorithm for probing would incorrectly derive the ill-formed, superiority-violating orders in (59). The lowest CWH head would probe first, attracting the highest wh-word in the IP. Only later will the higher CWH head probe for a wh-word, and by this time, the only ‘visible’ wh-word left in the clause will be the lower wh-word not probed by the first CWH head. The structure in (60) illustrates. (60)

Derivation of the Incorrect Orders in (59), via the Analysis in (56) CPWH-1 Daa2

CPWH-1

...

CWH-1

CPWH-2 aadóo1

CPWH-2

…

CWH-2

IP t1 yéi oowajée t2 du jee yéi teeyí

On the other hand, the analysis proposed in (53) can derive the targeted word order, assuming a theory of ‘Tucking-In’, as in Richards (1997). Under this analysis, a single CQ head probes for both QPs in the multiple wh-question. Following the standard algorithm for probing, this CQ first probes and attracts the highest QP in the clause. Following this attraction, the CQ then continues to probe for additional QPs. It subsequently probes and attracts the lower QP, requiring that the QP front into the CQ projection. However, because of the constraint ‘Shortest Move’, this lower QP has to be merged to as close a position to the CQ as possible. Consequently, the QP ‘tucks in’, and moves to a Spec position lower than that occupied by the higher wh-word. This derivation is sketched in (61). (61)

Derivation of the Correct Orders in (57), via the Analysis in (53) [with ‘Tucking In’] CP QP1

CP

Aadóo sá

QP2 daa sá

First Movement

CP CQ

IP

t1 yéi oowajée t2 du jee yéi teeyí Second Movement, With Tucking In

42

THE GRAMMAR OF Q

Thus the view that there is a single head attracting the entire wh-word + Q complex as a whole is necessitated by the word order facts in (57) and (59). Finally, one might yet resist the analysis in (53) by suggesting that the single C head attracting the wh-word + Q complex also probes for features of the wh-word. That is, we have not yet ruled out that a single “Cwh/Q” head Agrees with both the Q-particle and the wh-operator. In response, however, one might also point out that there is yet no evidence that the C head does Agree with the wh-operator. After all, the left-peripheral position of the wh-operator could very well result from the already demonstrated relationship between the C head and the Q-particle, as proposed in (53). In the absence of evidence that a relation holds between C and the wh-word, it is simplest to assume that it does not. We can, however, press the issue even further, and show that Agreement does not hold between the C and the wh-operator. First, let us entertain a comparatively strong view of syntactic islands, under which they are domains that no syntactic relations may cross, not even probing and Agree. Assuming this view of islands, the acceptability of sentence (62)—where the wh-word is contained within an island— indicates that there is no probing/Agreement between it and the matrix C. (62)

Wh-Operators in Wh-Questions Can Be Internal to Islands [ [ Wáa kwligeyi ] xáat ] sá i tuwáa sigóo? how it.is.big.REL fish Q your spirit it.is.glad How big a fish do you want? (= A fish that is how big do you want?)

Now, one might counter by proposing that the wh-word in (62) is somehow accessible to matrix C. Indeed, one might reject the notion that relative clauses are islands to probing and Agree. However, any such proposal is immediately subject to the following problem: recall the contrast between (62) and (63). (63)

The Q-Particle Sá Cannot Be Internal to Islands in Wh-Questions * [[Wáa sá kwligeyi ] xáat ] i tuwáa sigóo? how Q it.is.big.REL fish your spirit it.is.glad

Sentence (63) differs from (62) only in that the Q-particle sá is directly adjacent to the wh-word. Therefore any analysis which holds that the wh-word is syntactically accessible to matrix C in (62) must equally well hold that the Q-particle is accessible to matrix C in (63). Therefore the impossibility of (63) must follow from something other than the fact that the Q-particle is located inside a syntactic island. What this could be, however, remains unclear.30 The analysis in (53), however, predicts the contrast between (62) and (63), under the assumption that no syntactic relationship may cross into an island.31 The impossibility of (63) follows directly from the placement of the Q inside a relative clause island. When the Q-particle is located outside the island, as in (62), it is accessible to matrix C, and the sentence is well formed. The fact that the wh-word in (62) remains inside the island has no bearing on the well-formedness of the sentence, given that the matrix C bears no syntactic relationship to the wh-operator itself. We find, then, that the contrast between (62) and (63) supports what is, perhaps, the most unusual


43

feature of the analysis in (53): the existence of a relationship between the interrogative C and the Q-particle, but not between the C and the wh-word.32 The preceding arguments demonstrate the empirical support held by the analysis of Tlingit wh-questions in (53), an analysis independently motivated by their similarity to the wh-questions of Sinhala and Japanese. I conclude that (53) is, in essence, the correct analysis of Tlingit wh-questions.

2.6 The QP-Intervention Condition In the preceding section I presented arguments supporting the analysis in (1)/(53) for Tlingit wh-questions. None of those arguments, however, directly addresses one crucial aspect of (1)/(53): the claim that the Q-particle sá takes its sister as complement rather than being adjoined to its sister as in (51) and (52). Of course, this assumption is far simpler than the alternative, in as much as it treats the concomitant fronting of the wh-word as resulting from normal phrasal movement of the QP. Nevertheless, it would be optimal to find some independent evidence that the wh-word is dominated by a QP-projection in Tlingit. We will see in this section that the behavior of sá in certain environments provides this evidence. In outline, this section concerns some further constraints on the placement of sá within the clause. We will see that the particle sá cannot appear (i) between a postposition and its complement, (ii) between a wh-possessor and the possessed NP, or (iii) between a wh-determiner and its NP complement. We will then first consider the possibility that these constraints follow from the fact that QPs must undergo movement in Tlingit, given that the three aforementioned environments tend to be islands for movement. Although an attractive possibility, I will ultimately reject such a movement-based analysis of the patterns in (i)–(iii). In brief, we will see that the generalizations in (i)–(iii) also hold for QPs interpreted as indefinites. Therefore a movement-based analysis of (i)–(iii) must assume that such indefinite QPs also undergo obligatory movement, presumably obligatory QR. However, there is evidence that indefinite QPs do not undergo obligatory QR: indefinite QPs can scope out of islands. Therefore QR is not necessary to fix their scope, and so the putative obligatory movement remains mysterious. In place of a movement-based account, I propose a condition governing the position of QPs, which I dub the ‘QP-Intervention Condition.’ I show that this condition can capture the observed patterns, as well as avoid the problems inherent in a movement-based account. Importantly, this account of the patterns in (i)–(iii) requires that a QP-projection dominate the wh-word in Tlingit. Thus the success of the account provides some additional evidence that Q in Tlingit takes its sister as complement. More importantly, however, we will see that, given the analysis in (1)/(53), the QP-Intervention Condition alone derives various constraints on the form of Tlingit wh-questions, constraints which in other languages are typically analyzed in terms of conditions on the operation of movement. Therefore the discussion here foreshadows my arguments in later chapters that the analysis in (1)/(53) applies not simply to Tlingit, but to all wh-fronting languages.

44

THE GRAMMAR OF Q

2.6.1 Further Constraints on the Placement of Sá Beyond the conditions introduced in section 2.4, there are a number of further, yet unstated constraints governing the placement of sá in a Tlingit wh-question. First, as the sentences in (64) and (65) illustrate, the particle sá cannot, in a Tlingit wh-question, intervene between a postposition and its complement. (64)

No Q Between a Postposition and Its Complement in Tlingit Wh-Question a. Aadóo teen sá yeegoot? who with Q you.went Who did you go with? b. * Aadóo sá teen yeegoot? who Q with you.went

(65)

a. Goodéi sá yeegoot? where.to Q you.went Where did you go? b. * Goo sádéi yeegoot? where Q.to you.went

The judgments above correspond with the patterns found in our Tlingit corpus, as illustrated in (66). (66)

The Placement of Sá with Respect to Post-Positions in Wh-Questions

Text

Postposition marks a phrase containing a wh-operator

Of these, those in which sá appears to the right of the postposition


22 7 44 18 19 110

22 7 44 18 19 110

As the numbers here clearly indicate, whenever a wh-operator in the selected corpus appears (in a phrase) marked by a postposition, the particle sá appears to the right of that postposition. Thus we find that the order ‘[wh . . . Q . . . P]’ is textually unattested. Only the order ‘[wh . . . P]Q’ is found in natural speech. A second constraint on the position of sá is that it cannot, in a Tlingit whquestion, intervene between a possessor and the possessed NP. This is illustrated by speaker judgments like (67)–(70). (67)

No Q Between a Possessor and the Possessed NP in a Tlingit Wh-Question a. Aadóo yaagú sá ysiteen? who boat Q you.saw.it Whose boat did you see?


b. * Aadóo sá who Q

45

yaagú ysiteen? boat you.saw.it

(68)

a. Aadóo x’asheeyí sá iya.aax? who song Q you.heard.it Whose song did you hear? b. * Aadóo sá x’asheeyí iya.aax? who Q song you.heard.it

(69)

a. Aadóo jeet sá iyatee? who hand.to Q you.brought.it Who did you give it to? (= Whose hand did you bring it to?) b. * Aadóo sá jeet iyatee? who Q hand.to you.brought.it

(70)

a. Aadóo xánx’ sáyá yéi iyatee? who area.at Q.FOC you.are.there Who are you living with? (= Whose area are you staying at?) b. * Aadóo sá xánx’ yéi iyatee? who Q area.at you.are.there

As before, the judgments here correspond with the patterns found in our Tlingit corpus. (71)

Placement of Sá With Respect to Possessed Noun Phrases in Wh-Questions

Text

Wh-operator is a possessor modifying a possessed NP

Of these, those in which sá appears to the right of the possessed NP


1 2 3 3 1 10

1 2 3 3 1 10

As the numbers here clearly indicate, whenever a wh-operator in the selected corpus functions as a possessor modifying a possessed NP, the particle sá appears to the right of that possessed NP. Finally, a third constraint on the position of sá is that it cannot, in a Tlingit whquestion, intervene between a determiner and its NP complement. This is illustrated by speaker judgments like (72)–(75). (72) No Q Between a D and Its NP Complement in a Tlingit Wh-Question a. Daakw keitl sá asháa? which dog Q it.barks Which dog is barking? b. * Daakw sá keitl asháa? which Q dog it.barks

46

THE GRAMMAR OF Q

(73) a. X’oon keitl sá ysiteen? how.many dog Q you.saw.them How many dogs did you see? b. * X’oon sá keitl yisiteen? how.many Q dog you.saw.them (74) a. X’oon gaaw sáwé? how.many hour Q.FOC What time is it? (= How many hours is it?) b. * X’oon sáwé gaaw? how.many Q.FOC hour (75) a. Daat gaaw sá ikgwaháa? what hour Q you.will.arrive What time will you get there? b. * Daa sá gaaw ikgwaháa? what Q hour you.will.arrive

Once again, the speaker judgments here correspond with what we find in our Tlingit corpus.

(76)

Placement of Sá With Respect to NP Complements of DP in Wh-Questions

Text

Wh-operator is a determiner taking the following NP as complement

Of these, those in which sá appears to the right of the NP complement


6 0 16 3 5 30

6 0 16 3 5 30

As the numbers in (76) show, whenever a wh-operator in the selected corpus functions as a determiner, the particle sá appears to the right of its NP complement. In summary, then, we can see from the array of data above that in a Tlingit whquestion, the particle sá cannot intervene (i) between a postposition and its complement, (ii) between a possessor and a possessed NP, and (iii) between a determiner and its NP complement. Of course, when these restrictions are observed in the context of wh-questions, as they are here, they might not seem very surprising, particularly given our analysis in (1)/(53). Under that analysis, a Tlingit wh-question requires that the QP be fronted into the left periphery. Therefore placement of the Q-particle between, for example, a postposition and its complement, would, in a wh-question, result in extraction of the complement of P, as illustrated in (77).


47

(77) The Order [ WH . . . Q . . . P ] in Wh-Questions Entails P-Stranding CP C´

QP1 DP

Q

CQ

IP

…wh-word… PP QP

P

t1

Extractions of this form, however, are ill-formed in many languages of the world, a pattern that leads many to view such movements as cross-linguistically marked (Ross 1967; Abels 2003; Heck 2008). Similarly, placement of the Q-particle between a WH-possessor and the possessed NP in a wh-question would entail extraction of the specifier of DP, as shown in (78). (78)

The Order [ WH-POSS . . . Q . . . NP ] in Wh-Questions Entails Possessor Extraction CP QP1 DP

C Q

CQ

IP

…wh-word… DP QP

D

t1

D POSS

NP possessum

Again, however, extractions of SpecDP as in (78) are often ill-formed in languages, leading to a consensus that they violate certain general constraints on movement (Ross 1967; Corver 1990, 2007; Bošković 2005a). Finally, placement of the Q-particle between a WH-determiner and its NP complement would, in a wh-question, entail extraction of the determiner from the DP. (79)

The Order [ WH-DET . . . Q . . . NP ] in Wh-Questions Entails D-Extraction CP C

QP1 D wh-word

Q

CQ

IP

DP QP t1

NP

48

THE GRAMMAR OF Q

Once more, though, such extractions of D from DP as in (79) are in many languages impossible, which has led linguists to conclude that they are also in violation of certain general movement constraints (Ross 1967; Corver 1990, 2007; Bošković 2005a). It is apparent, then, that each of the ill-formed structures in (64)–(75) contains an extraction that independently appears to be ill-formed across languages. Therefore the ill-formedness of those structures may simply follow from whatever constraints on movement disallow such extraction. Given that this is a rather natural and compelling line of explanation, I will dignify it with a name. (80)

The ‘Movement-Based Analysis’ of the Further Constraints on Sá The impossibility for Q to appear in the environments outlined above is due to more basic conditions on the operation of movement. Given the analysis in (53), placement of the Q at those positions would entail movements of QP that would violate independent movement constraints.

Although the movement-based account in (80) might strike one as ‘common sense’, if one probes further into the grammar of Tlingit, it becomes clear that it cannot be the correct account of the facts in (64)–(75). We will see that the movement-based account in (80) is, in fact, too weak, and is not sufficient to capture the full paradigm of facts. This is chiefly because, as we will see, the generalizations governing the placement of Q in (64)–(75) hold even when the QP never moves. To begin our case against the movement-based analysis, let us first note that the generalizations in (64)–(75) still hold when the wh-word/QP in question functions as an indefinite in a declarative clause. As illustrated below, even in noninterrogative clauses, the Tlingit particle sá cannot intervene between a postposition and its complement. (81) No Q Between a Postposition and Its Complement With Tlingit Wh-Indefinites a. Tléil aadóo teen sá xwagoot. not who with Q I.went I didn’t go with anyone. b. * Tléil aadóo sá teen xwagoot. not who Q with I.went (82) a. Tléil goodéi sá xwagoot. not where.to Q I.went I didn’t go anywhere. b. * Tléil goo sádéi xwagoot. not where Q.to I.went (83) a. Hél aadóo een sá axwal’eix.33 not who with Q I.danced I didn’t dance with anyone. b. * Hél aadóo sá een axwal’eix. not who Q with I.danced


49

(84) a. Hél goodéi sá has wu.aat. not where.to Q they.went They didn’t go anywhere. b. * Hél goo sádéi has wu.aat. not where Q.to they.went

As before, the speaker judgments in (81)–(84) are supported by textual analysis. (85)

The Placement of Sá With Respect to Postpositions With Wh-Indefinites

Text

Postposition marks a phrase containing a wh-indefinite

Of these, those in which sá appears to the right of the postposition


10 9 0 0 45 64

10 9 0 0 45 64

The table in (85) is identical to that in (66), except that it tracks the position of sá in declarative clauses containing wh-indefinites. As we saw before, the totals here demonstrate that whenever a wh-indefinite in the selected corpus appears (in a phrase) marked by a postposition, the particle sá appears to the right of that postposition. Thus we find that with wh-indefinites too, only the order ‘[ wh . . . P ] Q’ is found in naturally occurring speech. Similarly, the constraint against placement of sá between a possessor and a possessed NP holds even for wh-indefinites. This is illustrated by speaker judgments like (86)–(90). (86)

No Q Between a Possessor and the Possessed NP With Tlingit Wh-Indefinites a. Tléil aadóo yaagú sá xwsateen. not who boat Q I.saw.it I didn’t see anyone’s boat. b. * Tléil aadóo sá yaagú xwsateen. not who Q boat I.saw.it

(87)

a. Tléil aadóo x’asheeyí sá xwa.aax. not who song Q I.heard.it I didn’t hear anyone’s song. b. * Tléil aadóo sá x’asheeyí xwa.aax. not who Q song I.heard.it

(88)

a. Tléil aadóo jeet sá xwatí. not who hand.to Q I.brought.it I didn’t give it to anyone. (= I did not bring it to anyone’s hand.) b. * Tléil aadóo sá jeet xwatí. not who Q hand.to I.brought.it

50 (89)

(90)

THE GRAMMAR OF Q

a. Tléil aadóo xánx’ sá yéi xat utí. not who area.at Q I.am.there I am not living with anyone. (= I am not staying at anyone’s area.) b. * Tléil aadóo sá xánx’ yéi xat utí. not who Q area.at I.am.there a. Hél aadóo yaagú sá xwsateen.34 not who boat Q I.saw.it I didn’t see anyone’s boat. b. * Hél aadóo sá yaagú xwsateen. not who Q boat I.saw.it

Again, textual analysis independently supports the generalizations gained from the speaker data in (86)–(90). (91)

Placement of Sá With Respect to Possessed Noun Phrases With Wh-Indefinites

Text

Wh-indefinite is a possessor modifying a possessed NP

Of these, those in which sá appears to the right of the possessed NP


1 3 0 0 10 14

1 3 0 0 10 14

The table in (91) is again nearly identical to the table in (71), the only difference being that it tracks the position of sá in declarative clauses containing wh-indefinites. As the totals here again indicate, whenever a wh-indefinite in the selected corpus functions as a possessor modifying a possessed NP, the particle sá appears to the right of that possessed NP, just as in wh-questions. Finally, further investigation reveals that the inability of sá to appear between a wh-D and its NP complement also holds for wh-D’s functioning as indefinites in declarative clauses. (92)

No Q Between a D and Its NP Complement With Tlingit Wh-Indefinites a. Tléil daakw keitl sá ushaa. not which dog Q it.barks None of the dogs are barking. b. * Tléil daakw sá keitl ushaa. not which Q dog it.barks

(93)

a. Yéi uwatee x’oon táakw sá. he.lived.there how.many winter Q He lived there for a number of years. b. * Yéi uwatee x’oon sá táakw. he.lived.there how.many Q winter


51

Additional support for this generalization can again be found in the selected corpus of Tlingit texts. (94)

Placement of Sá With Respect to NP Complements of DP in Wh-Indefinites

Text

Wh-indefinite is a determiner taking the following NP as complement

Of these, those in which sá appears to the right of the NP complement


2 4 0 0 12 18

2 4 0 0 12 18

As the numbers in (94) show, whenever a wh-indefinite in the selected corpus functions as a determiner, the particle sá appears to the right of its NP complement. In summary, we see from the data that—just as for the wh-operators in Tlingit wh-questions—the particle sá cannot, in a declarative clause with a wh-indefinite, intervene (i) between a postposition and its complement, (ii) between a possessor and a possessed NP, and (iii) between a determiner and its NP complement. Let us now consider what the movement-based analysis in (80) has to say regarding the facts in (81)–(94). Recall that according to (80), the constraints on sá placement in (64)–(75) are due to the impossibility of extraction from the base position of the QP. It follows that, in order to capture the similar facts in (81)–(94), the movement-based account must hold that the QPs in these declarative sentences are likewise extracted from their base position. Now, the fact that wh-indefinites in Tlingit can remain postverbal, as in (93a), shows that this putative extraction does not occur overtly. Therefore the movement-based analysis would require that such QPs undergo covert movement of some kind. We must now ask, then, what kind of covert movement could be responsible for the facts in (81)–(94)? The positions in question (SpecDP, CompPP) imply that such movement has nothing to do with case. Moreover, the declarative force of (81)– (94) entails that such movement is not motivated by the need to check a Q-feature in C. The only remaining possibility is that the movement is some kind of QR, a plausible prospect given that these QPs appear to be existential quantifiers. I conclude, then, that the movement-based account in (80) can only work for Tlingit if its whindefinites obligatorily undergo QR. In section 2.6.2, however, we will encounter evidence showing that, contrary to the hypothesis, Tlingit wh-indefinites need not undergo QR. 2.6.2 Evidence Against the Movement-Based Analysis In this section I put forth evidence against the movement-based account of the whindefinite data in (81)–(94). I begin by briefly showing that, except for the environments in (81)–(94), it is generally possible in Tlingit declarative clauses for QPs to

52

THE GRAMMAR OF Q

appear inside extraction islands. Following this, I challenge the claim, crucial for the movement-based analysis, that wh-indefinites in Tlingit undergo obligatory QR. To begin, let us note that according to the movement-based account of (81)–(94), the inability for QPs to occupy either CompPP or SpecDP follows partly from the fact that these positions are islands for extraction. Under the logic of this account, then, it would follow that QPs containing wh-indefinites should generally be unable to appear inside extraction islands. This prediction, however, is not true. There is no general constraint in Tlingit against sá appearing within an island. Recall that relative clauses in Tlingit are islands for extraction, and that in a wh-question the Q-particle cannot appear inside of a relative clause. (95)

Tlingit Relative Clauses Are Extraction Islands (cf. (62)) * [ [ Wáa sá kwligeyi CP] xáat NP] i tuwáa sigóo? how Q it.is.big.REL fish your spirit it.is.happy

Nevertheless, when paired with a wh-indefinite, nothing prevents a QP from being inside a relative. (96) QPs Containing Wh-Indefinites Can Appear Inside Relative Clause Islands Wáa sá yatee [ wé [ l goodéi sá wugoodi ] káa ]? how Q he.is that not where.to Q he.went.REL man How is the man who didn’t go anywhere?

The well-formedness of (96) suggests that the ill-formedness of the illicit structures in (81)–(94) is not due simply to their QPs appearing within an extraction island. Beyond this, however, there are some deep difficulties for one core assumption of the movement-based account in (80), the idea that wh-indefinites in Tlingit obligatorily undergo QR. Let us begin here by asking why QR of Tlingit wh-indefinites should be obligatory. Such obligatory QR would seem to imply that wh-indefinites in Tlingit cannot obtain their scope in situ. After all, if wh-indefinites in Tlingit could obtain their scope in situ, there would be no reason for QR to obligatorily target all such indefinites (Reinhart 1997). Therefore the movement-based analysis predicts that wh-indefinites in Tlingit always move to their scope positions via QR. Given the assumption that QR is sensitive to (adjunct) islands (Chomsky 1975; Reinhart 1997), we therefore predict that Tlingit wh-indefinites should—like strong quantifiers—be unable to scope out of such islands (Reinhart 1997). Let us, then, seek to test this prediction of the movement-based account. As shown in the work of Ruys (1992, 2000) and Matthewson (1999), an effective test can be gained by looking to the interpretation of indefinites contained within conditionals. For example, the well-known fact that English indefinites can scope out of islands is demonstrated by the consistency and coherency of the discourse in (97).


(97)

53

An English Indefinite Scoping Out of the Antecedent of a Conditional a. Dave is my friend. b. Many of Dave’s in-laws don’t like him. c. His brother in law, though, loves him. d. He said to Dave, “If I ever win the lottery, I’ll buy you a house.” e. Therefore, if one of Dave’s in-laws wins the lottery, he’ll get a house.

As noted by Ruys (1992, 2000), in order for the discourse in (97) to be coherent, sentence (97e) must be interpreted so that the existential force of the indefinite “one of Dave’s in-laws” has scope outside the conditional antecedent. If the existential force were interpreted internal to the conditional antecedent, the final sentence would be equivalent to “if any of Dave’s in-laws win the lottery, he’ll get a house”, which is inconsistent with the prior discourse. Thus the coherence of discourses like (97) shows that indefinites in English can scope out of a conditional antecedent, a structure that otherwise constitutes a movement island. This fact, in turn, has led many to abandon the notion that English indefinites must undergo movement (QR) to fix their scope (Reinhart 1997; cf. Schwarzschild 2002). Consequently, if it can be shown that Tlingit allows for discourses comparable to (97), where the final sentence contains a wh-indefinite in the conditional antecedent, then we may conclude that such wh-indefinites need not undergo QR. In order to apply this test, we should first confirm that Tlingit wh-indefinites allow the ‘specific readings’ that are required for indefinites to obtain the exceptional scope seen in (97e) (Fodor & Sag 1982; Kratzer 1998; Matthewson 1999). First, the reader may note that in many of the examples we have seen thus far, the Tlingit whindefinite is apparently interpreted as an NPI or free-choice item ((18a, b), (81)–(92)). Nevertheless, it is possible for wh-indefinites in Tlingit to be interpreted as plain existentials, outside the scope of any other logical operators.35 Listed in (98) are some textually attested examples. (98)

Tlingit Wh-Indefinites as Plain Existentials a. Ax x’agáax’i yéi yatee ch’a aadóoch sá yawudlaagí. my prayer thus it.is just who.ERG Q they.get.it My prayer is that someone learn it. (Dauenhauer & Dauenhauer 1990: 206; line 186) b. Ch’a daat yís sáwé yáat aas áa wsi.aa yáat. just what for Q.FOC here tree some they.grow here For some reason, there are trees growing here. (Nyman & Leer 1993: 4; line 47) c. Wé éexnax.á áwé, daa sáyá aya.áxch. that south.to.one FOC what Q.FOC he.heard.it The [old man] to the south heard something. (Nyman & Leer 1993: 10; line 34) yóo dikéenax.á d. Daa sáwé what Q.FOC yonder far.out.across.one There was something up there. (Nyman & Leer 1993: 14; line 103) e. . . . áwé daa sáwé xwasiteen. foc-part what Q.FOC I.saw.it . . . and I saw something. (Nyman & Leer 1993: 66; line 497)

54

THE GRAMMAR OF Q

f. Ch’a daa sá aagáa kukkwatées’. just what Q it.for I.will.search I’ll look for something there. (Nyman & Leer: 180; line 266)

Furthermore, there is at least one passage within our Tlingit corpus where such a ‘specific wh-indefinite’ appears to take scope outside of an adjunct island. (99) Tlingit Wh-Indefinite Given a Specific Reading [Wáa kunaaliyéi wugoodi sáwé ] wé t’akwanéiyi du dix’kaadé sh k’awdligáy. how it.is.far she.went Q.FOC this baby her back.on it.fussed After she had gone a certain distance, the baby began to fuss on her back. (Nyman & Leer 1993: 226; line 184)

The context from which sentence (99) was taken makes clear that the wh-indefinite here is to be given a specific reading. The speaker does not mean to state that the baby began to fuss as soon as the mother had gone any distance at all, which would be the interpretation if the wh-indefinite had scope inside the temporal adjunct. Rather, the speaker is stating that the baby’s fussing began only after a particular distance had been crossed—that there is a certain distance after which the baby began to fuss. Such an interpretation requires that the wh-indefinite have wide scope, outside the temporal adjunct. From the previous data, it is apparent that Tlingit wh-indefinites can receive a ‘specific’ reading, and so we should be able to obtain speaker judgments regarding whether discourses comparable to (97) are possible. Consider, then, the following Tlingit discourse. (100) A Tlingit Wh-Indefinite Scoping Out of the Antecedent of a Conditional Dave. a. Ax xooní áwé my friend FOC Dave Dave is my friend. b. Shayadihéini du káani yán tlél has du tuwáa ushgú. they.are.many.REL his in-laws not their spirit it.is.glad Many of his in-laws don’t like him. ku.aa wusixán. c. Du káanich his brother-in-law.ERG though he.loves.him His brother-in-law, though, loves him. d. Yéi ayawsikaa, “Dáanaa káa dulxes’ át yaxwadlaagí, hít i jeeyís he.told.him money on one.gambles thing I.win.it house your hand.for kukwa.oo. I.will.buy.it He said to him (Dave), “If I ever win the lottery, I will buy you a house.” e. [ [Daakw aa du káanich sá ] yawudlaagí ], hít ayakgwadláak. which of.them his in-laws.ERG Q they.win.it house he.will.get.it So, if a certain in-law of Dave’s wins the lottery, he’ll get a house.

The Tlingit discourse in (100) was constructed with the help of a native speaker, who recognized the discourse as a sensible story and an accurate translation of the English


55

original in (97).36 Note that if the wh-indefinite in (100e) could only have narrow scope inside the antecedent of the conditional, then the discourse in (100) would be neither internally consistent nor an accurate translation of the original English story. Rather, the consistency and faithfulness of (100) require that the existential force of the wh-indefinite be located outside the antecedent of the conditional. Sentences like (100e) therefore demonstrate that, contrary to the predictions of the movement-based account, it is possible for Tlingit wh-indefinites to scope out of (adjunct) islands. More acutely, these sentences show that, in Tlingit, there are mechanisms other than movement (QR) that determine the scope of an indefinite’s existential force. This, however, entails that the scope of a wh-indefinite in Tlingit need not be fixed by movement (QR), contrary to the assumptions of the movement-based account. That is, because wh-indefinites in Tlingit can obtain their scope without undergoing QR, there is simply no reason for QR to obligatorily target all such indefinites. Of course, one could always maintain the movement-based account of (81)–(94) by simply giving up the notion that QR is the movement operation that obligatorily targets Tlingit wh-indefinites. However, since we have already eliminated all other plausible candidates, we find that the movement-based account must appeal to an as yet unknown form of covert movement. It is therefore most reasonable to conclude that the movement-based account in (80) is simply incorrect. 2.6.3 The Proposed Analysis: The QP-Intervention Condition The preceding section presented evidence against a central assumption of the movement-based account in (80), the claim that Tlingit wh-indefinites obligatorily undergo QR. On the basis of this evidence, I conclude that the movement-based account is not the correct analysis of the contrasts seen in (64)–(94). Therefore, despite their obvious similarity to independently proposed constraints on movement, some other grammatical principles must underlie the pattern of data in (64)–(94). In this section I will put forth one view of what those principles are.37 Let us begin by looking back to the wh-indefinites in (81)–(94). Moreover, let us consider those ill-formed structures where the Q-particle intervenes between either (i) a postposition and its complement, (ii) a possessed NP and its possessor, or (iii) a determiner and its NP complement. Now, recall that according to our central analysis in (1)/(53), the Tlingit Q-particle sá takes its sister as complement and projects a QP. It therefore follows that the ill-formed sentences in (81)–(94) contain structures like (101)–(103). (101) Structures Where Q Intervenes Between P and Its Complement PP P

QP DP …wh-word…

Q

56

THE GRAMMAR OF Q

(102)

Structures Where Q Intervenes Between Possessor and Possessed NP DP QP DP

D Q

D POSS

NP possessum

…wh-word…

(103)

Structures Where Q Intervenes Between D and NP Complement DP QP D

NP Q

wh-word

Interestingly, (101)–(103) all share the following property: in each, a QP intervenes between a functional head and a phrase selected by that functional head. In (101), the QP intervenes between the postposition P and the DP selected by P. In (102), the QP intervenes between the possessive D head POSS and the possessor selected by POSS. In (103), the QP intervenes between the D and the NP it selects. Furthermore, note that none of the well-formed sentences in (81)–(94) have this special property. In the well-formed sentences of (81)–(84), the Q-particle occurs to the right of the postposition, and so the QP it projects does not intervene between the P and its DP complement. (104)

Structures Where Q Appears to the Right of the Postposition QP PP DP

Q P

…wh-word…

Moreover, the PP complement of Q in these sentences is an adjunct, and so is not selected by any higher functional heads. In the well-formed sentences of (86)–(90), the Q-particle occurs to the right of the possessed NP, and so its projection does not intervene between the D and its specifier.


(105)

57

Structures Where Q Appears to the Right of the Possessed NP QP DP

Q

DP

D D POSS

…wh-word…

NP possessum

Moreover, the complement of Q in these sentences is either an adjunct (89), or is selected by a lexical head. Thus the QPs here do not interrupt the selectional relationships of any functional heads. Finally, in the well-formed sentences of (92) and (93), the Q-particle occurs to the right of the NP complement of D, and so its projection likewise does not intervene between D and NP. (106)

Structures Where Q Appears to the Right of the NP Complement of D QP DP D wh-word

Q NP

Here again, in these sentences the complement of Q is either an adjunct (93) or is selected by a lexical head. Thus the QPs here also do not interrupt the selectional relationships of any functional heads. On the basis of these observations, let us propose the generalization in (107). (107)

The QP-Intervention Condition A QP cannot intervene between a functional head F and a phrase selected by F.38 (Such an intervening QP blocks the selectional relation between F and the lower phrase, as illustrated in the following.)39 Lexical Head

Functional Head

LP

L

FP

QP

Q

Selection: OK

F

XP

QP

Q

XP

XXXX QP-Intervention Condition Violated

As we have already seen, this ‘QP-Intervention Condition’ would be sufficient to predict the data in (81)–(94). Furthermore, this condition successfully avoids the

58

THE GRAMMAR OF Q

problems inherent in the movement-based account, in as much as it does not erroneously assume that Tlingit wh-indefinites undergo obligatory QR. However, to truly make the case for the QP-Intervention Condition, we must demonstrate that it makes accurate predictions beyond just the facts given in (81)–(94). First, let us observe that our QP-Intervention Condition predicts the fact, noted earlier, that Tlingit sá cannot appear to the right of a matrix predicate. (108)

Tlingit Sá Cannot Occur to the Right of a Matrix Predicate a. Daa sá iyatéen? what Q you.can.see.it What can you see? b. * Daa iyatéen sá? what you.can.see.it Q

If sá were to occur to the right of the matrix predicate, then there are two logical possibilities concerning its exact position in the clause, neither of which is consistent with the stated properties of Q in Tlingit. First, it could be the case that sá takes the entire matrix CP as complement, as illustrated in (109). (109)

Tlingit Sá Taking Matrix CP as Complement QP CP

Q sá

Daa iyatéen

However, we will see in the next section that our semantics for Tlingit wh-questions would rule out such a structure. Briefly, Q-particles are analyzed as variables that must be bound by higher operators. In structures like (109), however, there is no position for a higher, binding operator to appear. Thus the Q-particle goes unbound, and the sentence cannot be interpreted. The second possibility regarding the position of a clause-final sá is that it takes as complement either the VP or one of the higher projections along the ‘functional spine’ of the clause. In either case, however, the QP-Intervention Condition would be violated. If sá were to take VP as complement, then the QP it projects would intervene between VP and the higher Infl head, as illustrated in (110). (110) Tlingit Sá Taking Matrix VP as Complement IP QP VP

I Q


59

However, Infl head is a functional head, and selects for the VP complement of Q. Therefore (110) violates the QP-Intervention Condition. Similarly, if Q were to take as complement any higher projection, F1, along the ‘functional spine’ of the clause, the QP it projects would intervene between F1P and the next higher functional projection, F2P. (111)

Tlingit Sá Taking as Complement Projections in the Matrix ‘Functional Spine’ F2 P QP F1P

F2 Q

Again, though, F2 is a functional head, and selects for the F1P complement of Q. Therefore (111) also violates the QP-Intervention Condition. We have thus ruled out the ability for Tlingit sá to appear anywhere to the right of the matrix predicate.40 Nevertheless, our account does correctly predict that sá can appear to the right of a subordinate clause, as in (112). (112)

Tlingit Sá Can Appear at the Right Edge of a Subordinate Clause [ Goodéi wugootx ] sá has oowajée i shagóonich? where.to he.went Q they.think your parents.ERG Where do your parents think that he went?

As long as the subordinate CP is either an adjunct or is selected by a lexical head, the QP-Intervention Condition will not be violated if a QP takes a subordinate CP as complement. Moreover, since such Qs occupy a position internal to the matrix CP, our semantics predicts that higher operators will be able to bind them, and so the resulting structures will be interpretable. Thus far we have shown how the QP-Intervention Condition can account for the behavior of sá with wh-indefinites in (81)–(94). Recall, however, that a benefit of the movement-based analysis was that it also captured the parallel behavior of sá in wh-questions (64)–(76). Clearly, the similarity between the facts in (64)–(76) and (81)–(94) demands that a uniform account be adopted rather than one attributing the facts in (64)–(76) to constraints on movement and the facts in (81)–(94) to the QP-Intervention Condition. Interestingly, though, the facts in (64)–(76) indeed follow from the QP-Intervention Condition alone. Let us begin with the ill-formed P-stranding sentences in (64) and (65). According to our analysis in (1)/(53), the left-peripheral constituent of a wh-question is a QP that has been extracted from its base position. Thus the ill-formed sentences in (64) and (65), where extraction of the QP strands a postposition, would at earlier stages of their derivation have a QP intervening between a P and the DP selected by P.

60

THE GRAMMAR OF Q

(113)

QP-Intervention Condition Rules Out Tlingit P-Stranding CP C

QP1 DP

Q

IP

CQ

…wh-word… PP QP

P

Impossible PP, Ruled Out by QP-Intervention Condition

t1

We have already seen, however, that such base structures are impossible in Tlingit, and are ruled out by the QP-Intervention Condition. Since the QP-Intervention Condition rules out the base structure that necessarily underlies P-stranding, it thereby rules out P-stranding in Tlingit, and so no special condition against such extractions need be appealed to in the grammar of Tlingit. Next, let us turn to the ill-formed possessor-extraction sentences in (67)–(70). Note that, according to our core analysis in (1)/(53), each of these structures would at earlier stages of their derivation have a QP intervening between a possessor and the possessive D that selects the possessor. Again, however, such configurations are independently ruled out by the QP-Intervention Condition. (114) QP-Intervention Condition Rules Out Tlingit Possessor Extraction

CP QP1 DP

C’ Q

CQ

IP

…wh-word… DP QP t1

D’ D POSS

NP possessum

Impossible DP, Ruled Out by QP-Intervention Condition

Finally, let us consider the ill-formed D-extraction sentences in (72)–(75). Given our analysis in (1)/(53), such structures could only be derived from base structures where a QP intervenes between the D and the NP complement of that D, a configuration again ruled out by the QP-Intervention Condition.41


(115)

61

QP-Intervention Condition Rules Out Tlingit Determiner Extraction CP C

QP1 D wh-word

Q

CQ

IP

DP QP

NP

Impossible DP, Ruled Out by QP-Intervention Condition

t1

In summary, we find that all the ill-formed sentences in (64)–(76) could only be derived from structures that violate the QP-Intervention Condition. Thus the QP-Intervention Condition in (107) is alone sufficient to rule out the ill-formed sentences in (64)–(76), and therefore does provide a uniform account for all the data in (64)–(94). Furthermore, we have seen that it also predicts the fact, observed earlier, that the Tlingit sá can appear to the right of subordinate predicates, but not to the right of matrix predicates. Consequently, I conclude that the QP-Intervention Condition offers the best analysis of the data we have seen in this section. At this point, let us note two further consequences of this analysis. First, it should be observed that, given the success of this analysis, we have further confirmation that the Q-particles of Tlingit take their sisters as complements and project a QP. If we were to assume that Tlingit Q-particles, like those in Japanese, simply adjoined to their sisters, and projected no higher phrasal category, then the statement of the QPIntervention Condition becomes much more difficult. It would have to be rephrased to something like “no Q-particle can be immediately dominated by a node occurring between a functional head F and a phrase F selects for”, a condition which must appeal to more complex tree-geometric relations. Thus the assumption that Q in Tlingit takes its sister as complement not only simplifies the theory of Tlingit wh-fronting, it is also crucial for understanding the wider distributional properties of Tlingit sá. Finally, let us note a potential consequence for the general theory of movement; my comments here will be brief, as I will more fully treat this subject in chapter 4. Under our analysis, the impossibility of the extractions in (77), (78), and (79) is ultimately due not to any constraint on extraction per se, but to independently visible constraints on the placement of Q. Such constraints serve to limit wh-extraction by limiting the structural preconditions for it, ruling out the base structures from which the ill-formed extractions must be derived. Thus, rather than explain the impossibility of such extractions in terms of the ‘islandhood’ of the base positions, we explain the apparent islandhood of those positions in terms of independently visible constraints on the position of Q. Recall, however, that the impossible extractions of the kind seen in (77)–(79) are found to be ill-formed in many languages of the world. Just as we should seek a uniform Tlingit-internal explanation of the facts in (64)–(94), we should likewise seek a uniform account of these facts across languages. Given the evidence supporting the

62

THE GRAMMAR OF Q

QP-Intervention Condition in Tlingit, it therefore is most reasonable to conclude that this condition must also be responsible for the impossibility of the aforementioned extractions in all other wh-fronting languages. Of course, such an analysis is only possible under the view that wh-fronting in all languages proceeds as represented in (1)/(53). The remaining chapters are devoted to exploring this possibility and realizing the full potential of its explanatory force. 2.6.3.1 The Nature of the QP-Intervention Condition In our preceding discussion, the QP-Intervention Condition in (107) was introduced as a special stipulation. However, we should of course seek to eliminate such stipulations from our account. This is particularly so in the case of (107), which under its current phrasing appears rather different from other, more widely accepted grammatical principals. Interestingly, however, when combined with the semantic proposals introduced in the following section, the condition in (107) can be seen to follow from a more general hypothesis concerning selection. Let us begin by recalling the distinction between ‘s-selection’ and ‘c-selection’ (Grimshaw 1981; Pesetsky 1982). In brief, s-selection is selection for a particular semantic type, while c-selection is selection for a particular syntactic category. Thus the c-selectional requirements of a head demand that its sister be of a particular syntactic category, while the s-selection requirements only demand that the sister be of a particular semantic type. In this context, let us suppose that it is only functional heads that c-select for their arguments. Consequently, lexical heads only s-select their arguments (Pesetsky 1982). Given the nature of c-selection and s-selection, it would follow that it is only functional heads that require their arguments to be of a particular syntactic category. The selectional requirements of a lexical head would be satisfied as long as their arguments were of the correct semantic type. Now, in the next section, we will develop a semantic theory of QPs. Interestingly, a crucial consequence of this semantics is the following: a QP will always have the exact same semantic type as the sister of Q. If the sister of Q is of semantic type τ, then the QP as a whole will also be of type τ. This result is illustrated in (116). (116)

QP Always the Same Semantic Type as Sister of Q QP: semantic type = Always the same Q

XP: semantic type =

When we combine this semantics for QPs with our hypothesis concerning selection, we effectively derive the ‘QP-Intervention Condition’ in (107). First, let us show that QPs will be permitted to intervene between a lexical head and any phrase selected by that lexical head. Let L be a lexical head s-selecting for a phrase of semantic type τ. By assumption, L has no c-selectional requirements. Therefore the selectional requirements of L will be satisfied as long as its complement


63

is of type τ, regardless of its syntactic category. Now, recall that a QP where Q takes as sister a phrase of type τ will itself be of type τ. Consequently, such a QP will satisfy the selectional requirements of L. Thus we see that a QP will be able to intervene between any lexical head L and a phrase of the semantic type selected by L. This general result is schematized in (117). (117)

QP Can Satisfy S-Selectional Requirements of Lexical Head LP

L s-selection:

QP: semantic type = s-selection satisfied by QP Q

XP: semantic type =

On the other hand, our system predicts that QPs will not be able to intervene between a functional head F and a phrase selected by F. Recall that, by assumption, a functional head F also c-selects for its arguments. Consequently, unless F actually c-selects for a QP, the selectional requirements of F will not be satisfied by any QP. Consequently, a QP will be unable to intervene between any functional head F and a phrase of the category selected by the functional head. (118)

QP Cannot Satisfy C-Selectional Requirements of Functional Head FP

F c-selection: XP s-selection:

XXX c-selection not satisfied by QP

QP

Q

XP

Thus we find that the content of our QP-Intervention Condition in (107) could follow from a broader difference in the selectional requirements held by lexical and functional categories. Of course, whether or not (107) can be shown to follow from more general principals of grammar does not impact its truth as a surface generalization. Since our ensuing discussion hinges only upon the descriptive adequacy of (107), I leave to future work a more thorough study of what accounts for the generalization as stated.

2.7 A Semantics for Tlingit Wh-Words and Q-Particles In this section I provide a semantics for the wh-words and Q-particles of Tlingit that assigns the correct interpretations to the somewhat exotic tree structures employed by our Q-based account in (1)/(53). Furthermore, we will see that this semantics can alone account for various observed properties of Q. In chapter 3 we will see that this

64

THE GRAMMAR OF Q

semantics can also be applied to the wh-words and Q-particles of Japanese and Sinhala, further emphasizing the formal identity between Tlingit sá, Japanese ka, and Sinhala da. For reasons of space, the discussion here will be rather compact, and will be most comprehensible to those with some familiarity with current work on the semantics of wh-words and Q-particles, particularly Hagstrom (1998), Shimoyama (2001), Kratzer and Shimoyama (2002), and Beck (2006). 2.7.1 Two Semantic Preliminaries In this section I briefly introduce two of the core background assumptions underlying our semantic proposals: (i) the theory of focus semantics, and (ii) our intensional type theory for natural language.

2.7.1.1 Focus Semantics Following much recent work on the semantics of wh-questions, I will propose that Q-particles are ‘focus-sensitive operators’. Such a proposal requires some assumptions regarding the semantics of focus and focus-sensitive operators. For our purposes here, I will adopt a somewhat simplified version of the well-known theory of Rooth (1985, 1992). I assume that besides the regular, ‘normal-semantic’ interpretation function “[[ . ]]”, there exists a special, ‘focus-semantic’ interpretation function, “ [[ . ]]F ”. The value of the focus-semantic interpretation function is defined in terms of the value of the normal-semantic interpretation function, as in (119). (119)

The Nature of the Focus-Semantic Interpretation Function [[ . ]]F a. The focus-semantic value of an unfocused head is simply the set containing its normal-semantic value. [[ X ]]F = { [[ X ]] } b. The focus-semantic value of a focused element (head or phrase), is the set of its ‘focus alternatives’. If the normal semantic value of the focused element is of type τ, then its focus-semantic value is all the elements of type τ. [[ XF ]]F = { x ∈ Dτ : [[ X ]] ∈ Dτ } [[ XPF ]]F = { x ∈ Dτ : [[ XP ]] ∈ Dτ }

The rules outlined in (119) indicate that the value of [[ X ]]F depends upon whether or not X bears intonational focus. If X is not focused, then its focus-semantic value is simply the singleton set of its normal-semantic value. If X is focused, however, then its focus-semantic value is its set of ‘focus alternatives’, the set of all the entities of its semantic type. Note that, since the rule in (119a) applies only to unfocused heads, the rules in (119) do not yet provide a focus-semantic value for unfocused phrases. In order to rectify this, we introduce a special rule known as ‘pointwise semantic composition’. It is defined in (120).


(120)

65

Pointwise Semantic Composition If [[ X ]]F = { f< σ , τ > }, and [[ Y ]]F = A = { g, h, j, . . . }, then [[ X Y ]]F = B< τ , t > = f[A] = { f(g), f(h), f(i), . . . }

In essence, (120) states the following: if the focus-semantic interpretation of X (a head or a phrase) is the singleton set consisting of a function f of type , and the focus-semantic interpretation of Y (a head or a phrase) is a set of entities of type σ, then the result of pointwise composition between X and Y is the set one obtains by applying the function f to all the entities in A. Given this compositional rule, we can effectively compute the focus-semantic value of a sentence, as illustrated in (121). (121)

Focus-Semantic Interpretation of a Sentence = a. [[ DAVEF smokes ]]F = (by (119a)) b. [[ DAVEF ]]F [[ smokes ]]F = (by (119b)) c. [[ DAVEF ]]F { [λx. λw. x smokes in w] } d. { Dave, John, Frank, Bill . . . }{ [λx. λw. x smokes in w] } = (by (120)) e. { [λw. Dave smokes in w] , [λw. John smokes in w] , [λw. Frank smokes in w] , . . . }

Finally, let us review our assumptions regarding the meaning of so-called focussensitive operators, such as “only”. The core property of a focus-sensitive operator is that its semantic contribution depends upon the focus-semantic value of its sister. For example, the meaning of “only” might be represented as in (122). (122)

A Focus-Semantic Interpretation Rule for “Only”42 [[ Only S ]] = [[ S ]]∪ = T, and for all p ∈ [[ S ]]F, if p ≠ [[ S ]] then p∪ = F

According to this semantics, the operator “only” contributes the following information: (a) the normal-semantic value of its sister is true, and (b) for all the propositions in the focus-semantic value of its sister, if they are not equal to the normal semantic value of its sister, then they are false. Given this semantics, we can compute the truth conditions of a sentence like “Only DAVEF smokes”. (123)

Semantic Computation of a Sentence Containing a Focus-Sensitive Operator a. [[ Only [ DAVE smokes ] ]] = (by (122)) b. [[ DAVE smokes ]]∪ = T, and for all p ∈ [[ DAVE smokes ]]F, if p ≠ [[ DAVE smokes ]] then p∪ = F = (by (119), (121)) c. [λw. Dave smokes in w]∪ = T, and for all p ∈ { [λw. John smokes in w] , [λw. Frank smokes in w] . . . } if p ≠ [λw. Dave smokes in w], then p∪ = F = (by meta-logic) d. Dave smokes, but nobody else does (not John, not Frank, etc.)

The definitions introduced in (119), (120), and (122) will constitute the formal theory of focus semantics employed throughout our discussion. For a more complete

66

THE GRAMMAR OF Q

background into systems of this sort, I refer the reader to Rooth (1985, 1992), Kratzer (1991), and Hagstrom (1998). 2.7.1.2 The Intensional Type Theory Throughout our discussion, I will employ a system of intensional semantics akin to that employed by Lewis (1970), Rooth (1985), and Kratzer (1991). Within such a system, the words and phrases of natural language are assigned only intensional types. Thus, for natural language, the most basic types are not the extensional types e and t, but rather the intensional types <se> and <st>. For example, a proper name such as “Dave” will have as its semantic value, not the entity Dave itself, but rather the individual concept ‘[λw. Dave(w)]’. Similarly, the predicate “runs” will have as its semantic value the intensional property ‘[λx<se>. λw. x(w) runs in w ]’. Consequently, a sentence like “Dave runs” will have as its semantic value, not a truth value, but rather the proposition ‘[λw. Dave(w) runs in w ]’. The adoption of this intensional system is necessitated by the central role played by focus-semantics in our account. Ultimately the issue is that the focusalternatives of a given constituent must be computed from the intension of that constituent. This is most easily seen via our semantics for “only” in (122). Note that this semantics assumes that the focus-alternatives of a sentence form a set of propositions, rather than a set of truth values. Indeed, it would be impossible to build a semantics for “only” under the view that the focus-alternatives for a sentence are a set of truth values. Given that the focus-alternatives for a sentence must be a set of propositions, the simplest procedure for computing a sentence’s focus-alternatives must assume that a sentence has a proposition as its normalsemantic value. Thus our purely intensional semantics is necessitated by the simple fact that our semantics for Q-particles proposes them to be focus-sensitive operators. 2.7.2 The Semantics of Wh-Words With the background provided by the preceding sections, we can now put forth our main proposals concerning the semantics of wh-questions and wh-indefinites. I begin in this section with the core ideas surrounding the semantics of wh-words themselves. There has long been observed to be a special link between wh-words and focus in the world’s languages. A particular view of this link is put forth in recent work by Beck (2006), which I adopt here. Following Beck (2006), I will assume that whwords are semantically deficient in a characteristic way: in all languages, wh-words have only a focus-semantic value, their normal-semantic value being undefined. To elaborate, it is assumed that wh-words are assigned a specific logical type and value for animacy, though they are not assigned an actual normal-semantic value. Therefore, given our assumptions regarding the semantics of focus, the focus-semantic value of a focused wh-word is a set of focus-alternatives, each of the same logical type and animacy as the wh-word. For example, the wh-words what (English), daa


67

(Tlingit), nani (Japanese), and mokak (Sinhala) have the following characteristic semantics. (124)

Semantics of WHAT normal-semantics: [[ what / daa / nani / mokak ]] focus-semantics: [[ whatF / daaF / naniF / mokakF ]]F

= undefined = { x<se> : x ∉ human }

Similarly, the wh-words who (English), aadóo (Tlingit), dare (Japanese), and kauru (Sinhala) all have the following characteristic semantics. (125)

Semantics of WHO normal-semantics: focus-semantics:

= [[ who / aadóo / dare / kauru ]] [[ whoF / aadóoF / dareF / kauruF ]]F =

undefined { x<se> : x ∈ human }

There are several benefits to this particular treatment of wh-words. First, as originally shown by Beck (2006) (and as we will see in chapters 3 and 4), this semantics provides an interesting account of so-called LF- or Focus-Intervention Effects across languages. Furthermore, it provides a reason why wh-words must be focused in so many of the world’s languages; if wh-words were not focused, then a semantic crash would necessarily result.43 It should also be noted that this system, unlike those in Hagstrom (1998) and Shimoyama (2001), identifies the ‘alternatives’ denoted by whwords as focus-semantic values. This eliminates the need for special rules of pointwise semantic composition specifically for the values of wh-words, and instead employs the pointwise composition rules that are independently needed for the computation of focus-semantic values. Finally, we will see later that, given plausible ancillary hypotheses, this treatment predicts several of the core grammatical properties of Q. 2.7.3 The Semantics of Q-Particles Our semantics for wh-words states that their (focus-)semantic contribution is a set. Given the close relationship between Q-particles and wh-words in Tlingit, it would be natural to propose that Q is interpreted as an operator over sets, a view independently proposed by Hagstrom (1998), Yatsushiro (2001), Kratzer and Shimoyama (2002), and Beck (2006). More specifically, we will follow Hagstrom (1998) and Yatsushiro (2001) in claiming that Q-particles are variables over choice functions.44 As variables, Q-particles are assumed to carry indices. Thus the Q-particles sá, ka, and da all have the following characteristic semantics. (126)

Semantics of Q45 [[ sái / kai / dai ]]g = g(i) ∈ Dcf

The semantics in (126) states that the meaning of Qi relative to a variable assignment g is the value that g assigns to the index i, which is stipulated to be some element

68

THE GRAMMAR OF Q

from the domain of choice functions. To briefly review, a ‘choice function’ is any function that takes a set as argument and returns a member of that set as its value. Thus all the functions illustrated in (127) qualify as choice functions. (127)

Illustrative Examples of Choice Functions (Reinhart 1997, Winter 1997) A choice function takes a set and returns a member of that set. f({Dave, John, Larry, Phil}) = Larry = the Bible g({the Bible, the phonebook, LSLT}) h({Amherst, Boston, Natick, Worcester}) = Worcester

Given that they are variables, higher operators can—and in fact must—bind the Q-particles. For example, we will see later that the interrogative C head of a whquestion is semantically an interrogative operator binding the choice function variable contributed by Q. In addition to this, I assume that an existential operator over choice functions can be inserted via a rule of existential closure (Reinhart 1992, 1997; Yatsushiro 2001). I assume here that such existential closure occurs at the level of the IP, to bind any free variables occurring within the IP (cf. Reinhart 1997, Kratzer 1998). (128)

Existential Closure at the Level of the IP [CP [IP … Qi … ] ]

[CP

i [IP

… Qi … ] ]

Existential Closure over Choice Function Variables

The final ingredient of our semantics for Q concerns its method of semantic composition. Q-particles are assumed to semantically compose with their sisters via a syncategorematic rule specific to Q-particles (Beck 2006). The normal-semantic value of a Q-particle and its sister is stipulated to be the normal-semantic value of the Q-particle applied to the focus-semantic value of its sister. (129)

Special Composition Rule for Q-Particles [[ Qi XP ]] = [[ Qi ]]([[ XP ]]F)

Note that according to this semantics, a Q-particle takes as argument the focussemantic value of its sister. Therefore, as alluded to earlier, Q-particles are claimed to be focus-sensitive operators. 2.7.4 The Compositional Semantics of Wh-Indefinites Given the machinery thus far introduced, we can now provide a compositional semantics for the wh-indefinites of Tlingit. To begin, let us first observe how our system puts together the meaning of a Tlingit QP. In the simplest case, the wh-word itself is the sister to the Q-particle. (130)

Semantics of a Simple QP [[ [QP DaaF sá1] ]]g

=

(by (129))


69

= (by (124)) [[ sá1 ]]g ([[ DaaF ]]F g) = (by (126)) [[ sá1 ]]g ({ x<se> : x ∉ human }) = g(1) ({ x<se> : x ∉ human }) f ({ [λw. Fido(w)] , [λw. Amherst(w)] , [λw. Connecticut-River(w)], . . . })

As the derivation in (130) shows, relative to the variable assignment g, the semantic value of the QP Daa sá1 ‘What Q’ is the choice function that g assigns to the index ‘1’, taking as argument the set of all nonhuman individual concepts. Of course, given the nature of choice functions, this means that the ultimate semantic value of this QP is some nonhuman individual concept, and so such a QP is of type <se>. The semantic computation becomes a bit more interesting when the Q-particle is not directly adjacent to the wh-word. Such a structure is witnessed by the Tlingit phrase Aadóo yaagú sá, ‘Whose boat Q’, which we might assume to have the structure indicated in (131). The semantic value of (131) is computed in (132). (131)

Assumed Structure of the Complex QP Aadóo Yaagú Sá, ‘Whose boat Q’ QP n-sem: <se> DPb DPF

f-sem: <se,t>

AadóoF

Q1 n-sem: sá

f-sem: <se,t>

DPa

f-sem: ,t>

D f-sem: >,t> POSS

NP

f-sem: ,t>

yaagú

(132) Semantic Computation for Structure (131) = a. [[ [QP [DP AadóoF [ POSS yaagú ] ] sá1 ] ]]g b. [[ sá1 ]]g ([[ [DP AadóoF [ POSS yaagú ] ] ]]F g) = c. g(1)([[ [DP AadóoF [ POSS yaagú ] ] ]]F g) = = d. g(1)([[ [ POSS yaagú ] ]]F g ([[AadóoF]]F g)) e. g(1)(([[ POSS ]]F g([[ yaagú ]]F g)) ([[AadóoF]]F g)) = f. g(1)(({[λP<se,st> [ λx<se> [λw. the P(w) of x(w) in w ]]]} = {[ λx<se> λw. boat(x)(w) ]})([[AadóoF]]F g)) g. g(1)({[ λx<se> [λw. the boat(w) of x(w) in w ]]} ([[AadóoF]]F g)) = h. g(1)({[ λx<se> [λw. the boat(w) of x(w) in w ]]} { x<se> : x ∈ human }) = i. g(1)({ h<se> : ∃x<se> ∈ human . h = λw. the boat(w) of x(w) in w }) = j. f ({ [λw. the boat(w) of Jim(w) in w], [λw. the boat(w) of Tom(w) in w], . . . })

(by (129)) (by (126)) (by (120)) (by (120)) (by Lexicon, (119)) (by (120)) (by (125)) (by (120))

As the computation demonstrates, the semantic value of the Tlingit phrase Aadóo yaagú sá1, ‘Whose boat Q1’, relative to a variable assignment g is the choice function f that g assigns to the index 1, taking as argument all those individual

70

THE GRAMMAR OF Q

concepts which for some human individual concept x<se> are the concept [λw. the boat(w) of x(w) in w ]. Therefore, given the nature of choice functions, the ultimate semantic value of this QP is an element from this set, and so such a QP is type <se>. Having presented the semantic computations for these two types of QPs, we can now see how our remaining hypotheses provide a compositional semantics for Tlingit wh-indefinites. Briefly, the picture is as follows: a Tlingit wh-indefinite simply arises when the QP containing the wh-word remains inside the IP, and so is bound by existential closure. For example, the following illustrates the interpretation our system derives for the Tlingit sentence in (98e), Daa sá xwasiteen, ‘I saw something’. (133)

Semantics of a Simple Wh-Indefinite in Tlingit Daa sá xwasiteen what Q I.saw.it. I saw something. IPc

n-sem: <st>

IPb

1

n-sem: <st>

pro n-sem:<se> (existential closure)

IPa

n-sem: <se,st>

Infl

QP

DPF Daa

f-sem: <se,t>

VP

n-sem: <se,st>

n-sem: <se>

Q1 sá

V n-sem: <se,<se,st>> xwasiteen

n-sem:

[[ IPc ]]g = w. f. I saw [f ({ x<se>: x

human })] in w

An inspection of the tree in (133) reveals that our system assigns to the structure in (133) the proposition ‘[λw. ∃f. I saw [f ({ x<se> : x ∉ human })] in w ]’.46 At a particular world w, this proposition is true iff there is some choice function f such that f applied to the set of nonhuman things yields an entity that the speaker saw in w. Of course, such a choice function exists if and only if there is a nonhuman entity that the speaker saw in w. We conclude, therefore, that our semantics correctly assigns to (133) its observed meaning, a proposition equivalent to ‘there is some nonhuman thing which the speaker saw’. Let us next consider a somewhat more complex case, one where the Q-particle is not directly adjacent to the wh-word. Such a wh-indefinite appears in sentence (86a): Tléil aadóo yaagú sá xwsateen, ‘I didn’t see anyone’s boat’. The following illustrates the interpretation that our semantics assigns to (86a).


(134)

71

Semantics of a Complex Wh-Indefinite in Tlingit Tléil aadóo yaagú sá xwsateen. not who boat Q I.saw.it I didn’t see anyone’s boat.

The tree in (134) reveals that our system derives as the meaning of (134) the following proposition: ‘[λw. ¬∃f . I saw [f ({ h<se> : ∃x<se> ∈ human . h = λw’. the boat(w’) of x(w’) in w’ })](w) in w]’. To break down this complex formula, let us first note that at a particular word w, this proposition is true if there is no choice function f such that the speaker saw the entity which f yields when applied to the set of all individual concepts of the form [λw. the boat(w) of x(w) in w], where x is human. Of course, this is equivalent to stating that there is no individual concept h of the form [λw. the boat(w) of x(w) in w], where x is human, such that the speaker saw h. Finally, note that this is itself equivalent to simply stating that there is no human x such that the speaker saw the boat of x in w. We therefore conclude that our semantics correctly assigns to the structure in (134) its observed meaning, a proposition equivalent to ‘there is no person such that the speaker saw that person’s boat’. We have thus far seen two cases where our semantics assigns the correct meaning to Tlingit sentences containing wh-indefinites. In both cases, we find that existential quantification over choice functions is truth-conditionally equivalent to existential quantification over the set of entities denoted by the wh-word. Of course, this result is most interesting in those cases where the Q-particle is not directly adjacent to the whword itself. In such cases, the argument of the choice function variable is not simply the set contributed by the wh-word, and so the equivalence between existential quantification over that set and over the domain of choice functions is not so intuitively

72

THE GRAMMAR OF Q

obvious. We may rightly wonder, then, whether the result in (134) is but a particular instance of a more general one. That is, is it ever the case in our system that altering the position of the Q-particle alters the proposition denoted by the sentence? The answer to this question is ‘no’. As long as the Q-particle c-commands the wh-word and is c-commanded by the existential quantifier, the same propositional function is assigned to the structure, one that is equivalent to existential quantification over the set denoted by the wh-word. This result will be of some importance to our discussion in chapter 4, as it will allow us to convert otherwise complex formulas into more manageable and recognizable ones. Let us therefore see in detail why it is so. We wish to show that no matter how far the Q-particle is from the wh-word, existential quantification over the domain of choice functions is still equivalent to existential quantification over the set of entities denoted by the wh-word. In other words, we wish to show that any arbitrary structure of the form in (135a), where τ is the semantic type of the wh-word, is equivalent to the structure in (135b), where the wh-word is replaced with a (normal) variable of its type, the Q-particle is removed, and the type-τ variable is existentially bound and restricted to the set denoted by the wh-word. (135)

Structures to be Proven Semantically Equivalent a. ∃1 [A . . . Q1 [B . . . wh-wordτ . . . ] . . . ] b. ∃x ∈ [[ wh-wordF ]]F [A . . . [B . . . xτ . . . ] . . . ]

First, let us note that our semantics entails that the syntactic structure in (135a) will be assigned the following as its truth conditions: ∃f [A . . . f [B . . . { x : x ∈ Dτ & P(x) } . . . ] . . . ], where ‘P(x)’ represents the lexical presuppositions of the wh-word (e.g. ‘human(x)’ for who, and ‘¬human(x)’ for what). Thus the structure in (135a) will be true iff there is some choice function f ∈ Dcf such that the following holds: [A . . . f [B . . . { x : x ∈ Dτ & P(x) } . . . ] . . . ]. Of course, given the nature of choice functions, this last condition holds if and only if there is some R within the set constituting [B . . . { x : x ∈ Dτ & P(x) } . . . ] such that [A . . . R . . . ] holds. Note, however, that since R is a member of [B . . . { x : x ∈ Dτ & P(x) } . . . ], our semantics entails that there is some entity a ∈ { x : x ∈ Dτ & P(x) } such that R = [B . . . a . . . ].47 Thus [A . . . R . . . ] holds iff for some a ∈ { x : x ∈ Dτ & P(x) }, [A . . . [B . . . a . . . ] . . . ] holds. Finally, this later condition holds iff the structure in (135b) is true, which was to be proven. This chain of reasoning is laid out in (136). (136)

Proof of the Equivalence of (135a) and (135b) ∃1 [A . . . Q1 [B . . . wh-wordτ . . . ] . . . ] is true iff ∃f [A . . . f [B . . . { x : x ∈ Dτ & P(x) } . . . ] . . . ] is true iff For some f ∈ Dcf, [A . . . f [B . . . { x : x ∈ Dτ & P(x) }. . . ] . . . ] is true iff For some R ∈ [B . . . { x : x ∈ Dτ & P(x) } . . . ], [A . . . R . . . ] is true iff For some a ∈ { x : x ∈ Dτ & P(x) }, [A . . . [B . . . a . . . ] . . . ] is true iff ∃x ∈ [[ wh-wordF ]]F [A . . . [B . . . xτ . . . ] . . . ] is true.


73

The equivalence between (135a) and (135b) can also be seen via reflection upon the meanings of the QPs themselves and the nature of variable assignments. Recall that Q-particles are variables over choice functions. Therefore, relative to an arbitrary variable assignment g, the semantic value of ‘[QP [XP . . . whτ . . . ] Q ]’ will be some entity A of the normal-semantic type of the XP. Now, consider the structure ‘[XP . . . yτ . . . ]’, where the wh-word is replaced with a (normal) variable of its type (which is also assumed to share the presuppositions of the wh-word). Relative to a variable assignment g, this latter structure is also some entity A′ of the normal-semantic type of the XP. Clearly, then, for any variable assignment g such that [[ [QP [XP . . . whτ . . . ] Q ] ]]g = A, there is some other variable assignment g′ such that [[ [XP . . . yτ . . . ] ]]g′ = A, and vice versa. Therefore, if there is a variable assignment g such that [[ [A . . . Q1 [B . . . wh-wordτ . . . ] . . . ] ]]g = T, then there is a variable assignment g′ such that [[ [A . . . [B . . . yτ . . . ] . . . ] ]]g′ = T, and vice versa. It follows, then, that if ever the structure in (135a) is true, then the structure in (135b) must also be true, and vice versa. As mentioned before, the equivalence between (135a) and (135b) will provide us a powerful tool, as it will permit the simplification of several otherwise complex formulas that our semantics derives. Of course, a more immediate consequence of (136) is that the existential interpretation of wh-indefinites in Tlingit may be obtained from our semantics via existential closure over the choice function variable contributed by Q. Thus our semantic theory illuminates the essential role played by socalled Q-particles in the meaning of wh-indefinites in declarative clauses. The reader may recall, however, that Q-particles also seem to play a crucial role in the semantics of wh-questions (hence, their very moniker). In a moment, we will see how the semantics just proposed can be incorporated into a treatment of Tlingit wh-questions. Before that, however, we will first observe two notable corollaries of our semantics. 2.7.5 The Special Relationship Between Q-Particles and Wh-Words In section 2.4 I presented evidence that the Tlingit particle sá, the Japanese particle ka, and the Sinhala particle da should all be considered instances of the same formal element, the Q-particle. This conclusion was based on a number of distributional similarities between these elements. In this section we will see that certain of these shared properties follow from the semantics just put forth. In particular, we will see that our semantics can derive the following two properties: (a) wh-words must be c-commanded by Q-particles, and (b) Q-particles must c-command wh-words.48 To begin, let us recall that our semantics adopts the leading views of Beck (2006). Interestingly, there are some further components to the Beck (2006) system that will prove to be of use to us. The theory of ‘LF/Focus-Intervention Effects’ put forth by Beck (2006) relies upon two independently plausible assumptions. The first is the ‘Principle of Interpretability’, stated in (137). (137)

Principle of Interpretability (Beck 2006: 16) A sentence must have a normal-semantic value.

74

THE GRAMMAR OF Q

As stated, this principle entails that any sentence that cannot be computed to have a normal-semantic value is ill-formed. The second crucial assumption is somewhat more complex, though equally as important. (138)

Uniqueness of the Q-Particle (Beck 2006: 13) The Q-particle is the only focus-sensitive operator whose meaning does not also take as input the normal-semantic value of its sister.

Let us pause for a moment to consider what the condition in (138) states. First, let us note that, according to the semantics in (129), repeated in (139), the Q-particle takes as argument only the focus-semantic value of its sister. (139)

Special Composition Rule for Q-Particles [[ Qi XP ]] = [[ Qi ]]([[ XP ]]F)

Of course, this insensitivity of Q to the normal-semantics of its sister is needed for our system to work. Recall that we assume wh-words do not have normal-semantic values ((124), (125)). Consequently, the sister of Q will never have a normalsemantic value. Thus, if our semantics ever required us to compute the normalsemantic value of Q’s sister, the derivation would crash. Now, although this insensitivity to normal-semantics is required for Q, it is clearly not a property of other focus-sensitive operators. For example, note how our semantics for “only” in (122), repeated in (140), requires one to also compute the normal-semantic value of the operator’s sister. (140)

A Focus-Semantic Interpretation Rule for “Only” [[ Only S ]] = [[ S ]]∪ = T, and for all p ∈ [[ S ]]F, if p ≠ [[ S ]] then p∪ = F

Thus we find that Q differs from “only” in that computing the meaning of a phrase containing Q does not require that one compute the normal-semantic value of Q’s sister. The principle in (138)—which is crucial for the theory of Beck (2006)—states that, in fact, it is only the Q-particle which has this peculiar insensitivity to normal-semantic values. Interestingly, the principles in (137) and (138) alone predict that wh-words must be c-commanded by Q-particles. Suppose that a wh-word in a given sentence is not c-commanded by a Q-particle. By assumption, then, either (i) the wh-word is c-commanded by a focus-sensitive operator OP that is not Q, or (ii) the wh-word is not c-commanded by any focus-sensitive operator. Let us first consider condition (i). Since OP is not a Q-particle, principle (138) entails that the semantic computation for the entire sentence requires one to compute the normal-semantic value of the sister of OP. However, since OP c-commands the wh-word, it follows that the sister of OP contains the wh-word. Therefore, computing the normal-semantic value of the sister of OP requires one to compute the normal-semantic value of the wh-word, and so the sentence is predicted to be uninterpretable. Now, let us


75

consider condition (ii). Since there is no focus-sensitive operator c-commanding the wh-word at all, computing the normal semantic value for the entire sentence requires that one compute the normal-semantic value of the wh-word. However, since the wh-word does not have a normal-semantic value, the derivation crashes. Resultingly, the sentence cannot be assigned a normal-semantic value, in violation of principle (137). We find, then, that (137) and (138) entail that every wh-word must be c-commanded by a Q. Let us now consider the inverse condition, that every Q must c-command a wh-word. We will see that this condition follows from a particular version of the principle of ‘Full Interpretation’ (Chomsky 1995). Informally, the principle of Full Interpretation requires that semantically interpretable elements be interpreted, that there be no entirely superfluous elements within a sentence. Another way of stating the general ‘intuition’ behind Full Interpretation is that “meanings are used”. That is, the procedure for interpreting a sentence cannot selectively ‘ignore’ the meaningful elements the sentence contains. Therefore we might adopt (141) as one specific version (or subcase) of Full Interpretation. (141)

The Principle of Full Interpretation Given a structure [A X Y ], if X has a normal-semantic value, then there is some replacement Z differing from X only in its normal-semantic value such that [[ [A X Y ] ]] ≠ [[ [A Z Y ] ]].

In essence, the principle in (141) states that if any phrase X has a normal semantic value, then the normal-semantic value of any phrase containing X must partly depend upon X’s normal-semantics. Thus all normal-semantic values must be ‘used’ in the computation of a phrase’s normal-semantic interpretation. If we accept the principle in (141), we thereby predict the need for Q-particles to c-command wh-words. To see this, let us first consider the structure in (142), where the sister to Q does not contain a wh-word, but does contain a focused DP. As the semantic derivation in (142) demonstrates, such a structure is, in principle, interpretable by our system. (142)

Interpretable Structure Where the Q-Particle Does Not C-Command a Wh-Word

76

THE GRAMMAR OF Q

As illustrated in (142), our semantics requires only that Qs have some focused XP within their scope; the semantics alone does not entail that the focused XP be a whword. However, with the addition of the principle in (141), structures like that in (142) may be ruled out. Consider, for example, the structure in (143), which differs from (142) only in the normal-semantic value of the focused DP. (143)

Interpretable Structure Where the Q-Particle Does Not C-Command a Wh-Word

Although the focused DP in (143) differs in its normal-semantic value from that in (142), the normal-semantic value of the entire QP remains unchanged. Furthermore, given our rules for computing focus-semantic values in (119), it follows that if one attempts to replace ‘Frank’ in (142) with any expression of its type, the interpretation of the QP will always be the same, namely, f ({ x : x ∈ De}). Thus we find that the structure in (142) violates the Principle of Full Interpretation in (141). Consequently the focused phrase within a QP can never be just any arbitrary DP. On the other hand, if the focused phrase within a QP is a wh-word, no violation of (141) need result. After all, our assumptions in (124) and (125) state that wh-words lack normal semantic values. As a result, wh-words vacuously satisfy the condition in (141). We may conclude, then, that (141) entails that the focused phrase in a QP must be a wh-word, and so any Q-particle must c-command a wh-word. In summary, we have seen that our semantics for wh-words and Q-particles, when combined with certain independently plausible principles, is able to derive the obligatory co-occurrence and c-command relation between Q-particles and whwords. In as much as these properties are shared by the particles sá, ka, and da, applying our proposed semantics to each of these particles would thereby account for the characteristic behavior of each, which lends further credence to their underlying formal identity. 2.7.6 The Compositional Semantics of Wh-Questions In this section I will show how the system developed thus far can provide a compositional semantics for Tlingit wh-questions. Before laying out our key proposals, I will begin with some preliminary background regarding the semantics of wh-questions.


77

2.7.6.1 The Semantics of Wh-Questions: Basic Background Before we can address how the meaning of a wh-question is computed, we must first adopt some view as to what the meaning of such a question is. Although the answer to this logically prior question remains a matter of much controversy (Hamblin 1958, 1973; Karttunen 1977; Groenendijk & Stokhof 1982, 1984; Heim 1994; Kratzer 2006), here I will simply adopt certain assumptions found in prior work on the semantics of Q (e.g., Hagstrom 1998; Shimoyama 2001; Kratzer & Shimoyama 2002; Beck 2006). That is, I will adopt the classic (though controversial) proposal that the meaning of a wh-question is a set of propositions, those propositions that constitute potential answers to the question (Hamblin 1958, 1973; Karttunen 1977). For example, the meaning of the wh-question in (144a) would be the set of propositions in (144b), understood to be the set of all the potential answers to the wh-question. (144)

The Semantics of Wh-Questions a. What did Dave eat? b. { [λw. Dave(w) ate pizza(w) in w], [λw. Dave(w) ate bread(w) in w], . . . }

Similarly, the meaning of the wh-question in (145a) would be the set of propositions in (145b), again understood to be the set of all its potential answers. (145)

The Semantics of Wh-Questions a. Whose book did Joe read? b. { [λw. Joe(w) read Lou’s book(w) in w], [λw. Joe(w) read Ian’s book(w) in w], ...}

If we adopt this view as to the meaning of wh-questions, it will help to represent those meanings in a particular, canonical fashion. Note, for instance, that the set of potential answers to a wh-question can be an infinite one; this is of course indicated by the ellipses in (144b) and (145b). We therefore require some notation for finitely representing this set of propositions. Throughout our discussion, I will adopt a version of the notation first employed in Karttunen (1977). This notation is illustrated in (146) and (147). (146)

(147)

The Semantics of Wh-Questions a. What did Dave eat? b. λp [ ∃x<se> ∉ human. p = [λw. Dave(w) ate x(w) in w ]] The Semantics of Wh-Questions a. Whose book did Dave read? b. λp [ ∃x<se> ∈ human. p = [λw. Dave(w) read [the book of x](w) in w]]

The lambda notation in (146b) may be informally read as “the set of propositions p such that p is the proposition Dave ate x for some nonhuman thing x.” Intuitively, this is indeed the set of propositions informally represented in (144b). Similarly, the notation in (147b) may be read as “the set of propositions p such that p is the

78

THE GRAMMAR OF Q

proposition Dave read x’s book for some human x.” Again, this set is intuitively the set of potential answers to the question in (145a).

2.7.6.2 The Compositional Semantics of Tlingit Wh-Questions We will begin our semantic discussion of Tlingit wh-questions by introducing two final, additional hypotheses: one semantic and one syntactic. The semantic hypothesis concerns the source of the characteristic ‘interrogativity’ of a wh-question. I assume that every wh-question contains in its left periphery a dedicated Force head, ForceQ, supplying the interrogative force of the clause. More concretely, I assume that this ForceQ is interpreted as an operator, one that binds the choice-function variable introduced by the Q-particle. As an operator, the ForceQ comes paired with an index. Also paired with this ForceQ head is the following syncategorematic rule, which effectuates the binding of the Q-particle by the Force head. (148)

Special Composition Rule for ForceQ [[ ForceQi XP ]]g = λp [ ∃f. p = [[XP]]g(i/f)]

The second syntactic hypothesis concerns the exact position of the QP within the left periphery. Until now we have simply assumed that such QPs occupy SpecCP. However, given the assumed existence of ForceQ, we must now ask which specifier position within the left periphery the QP occupies. Let us assume that the specifier occupied by QP is located below the Force head. More concretely, adopting the theory of the left periphery put forth by Rizzi (1997), we might assume that the QP of a wh-question moves to the Spec of a FocusQ phrase, which is obligatorily selected by the ForceQ head. (149)

The Fine Structure of the Left Periphery QP-movement targets a Spec position just below the interrogative Force head. ForceP ForceQ

FocP QP

The seat of interrogative force

FocP FocQ

IP

QP-movement targets Spec FocP

We now have everything necessary to provide a compositional semantics for Tlingit wh-questions. Let us begin by considering the simple wh-question in (7b), Daa sá i éesh al’óon?, “What is your father hunting?”, repeated in (150).


(150)

79

Simple Wh-Question in Tlingit a. Sentence: Daa sá i éesh al’óon? what Q your father he.hunts.it What is your father hunting? b. Targeted Interpretation: λp [ ∃x<se> ∉ human. p = [λw. [your father](w) is hunting x(w) in w]]

Given the English translation of this sentence, we want our semantic theory to assign as its interpretation the set of propositions represented by the lambda expression in (150b), which picks out the set of propositions p such that p is of the form ‘your father is hunting x’, where x is nonhuman.49 Let us, then, determine whether our semantic system can correctly assign the meaning in (150b) to sentence (150a). Given our preceding hypotheses, (151) is the structure we assign to sentence (150a).50 (151)

The Syntactic Structure of Sentence (150a)

Assuming that the FocQ head here has a trivial semantic value (i.e., [λp. p]), the following derivation computes the meaning that our semantics assigns to (151). (152)

Semantic Interpretation of Structure (151) a. [[ ForceQP ]]g b. [[ ForceQ1 FocPc ]]g c. λp [ ∃f. p = [[FocPc]]g(1/f) ] d. λp [ ∃f. p = [[ QP FocPb ]]g(1/f) ] e. λp [ ∃f. p = [[ FocPb ]]g(1/f)([[QP]]g(1/f)) ] f. λp [ ∃f. p = [[ FocPb ]]g(1/f)(f({ x<se> : x ∉ human })) ] g. λp [ ∃f. p = [λx<se>. λw. [your father](w) is hunting x(w) in w] (f({ x<se> : x ∉ human })) ] h. λp [ ∃f. p = [λw. [your father](w) is hunting [f({ x<se> : x ∉ human })](w) in w]]

= = = = = =

(by (148)) (by FA)51 (by (130)) (by standard rules)

= (by LC)52

As the derivation in (152) demonstrates, our semantic theory assigns to (151) the following meaning: ‘[ λp [ ∃f. p = [ λw . [your father](w) is hunting [ f ({x<se> : x ∉ human}) ](w) in w]]. This lambda notation picks out the set of propositions p

80

THE GRAMMAR OF Q

such that there is some choice function f such that p is the proposition ‘your father is hunting f({ x<se> : x ∉ human })’. Thus we must now determine whether this set of propositions is equivalent to the ‘targeted interpretation’ in (150b). Interestingly, the equivalence between (150b) and (152h) follows from the general equivalence proven in (135), repeated in (153). (153)

Structures Proven Semantically Equivalent a. ∃1 [A . . . Q1 [B . . . wh-wordτ . . . ] . . . ] b. ∃x ∈ [[ wh-wordF ]]F [A . . . [B . . . xτ . . . ] . . . ]

Given this general equivalence, it follows that the formulas in (154a) and (154b) are equivalent (relative to any value for the free propositional variable). (154)

Structures Proven Semantically Equivalent a. ∃f. p = [λw. [your father](w) is hunting [f ({ x<se> : x ∉ human })](w) in w]] b. ∃x<se> ∉ human. p = [λw. [your father](w) is hunting x(w) in w]]

Given the equivalence of the formulas in (154), it follows that the formulas in (155) are equivalent. (155)

a. λp [ ∃f. p = [λw. [your father](w) is hunting [f({ x<se> : x ∉ human })](w) in w]] b. λp [ ∃x<se> ∉ human. p = [λw. [your father](w) is hunting x(w) in w]]

The formula in (155a) is, of course, the meaning computed in (152h) for (151), while the formula in (155b) is simply the targeted meaning in (150b). It therefore follows that these meanings are equivalent, and so our semantics correctly assigns the ‘targeted meaning’ in (150b) to the sentence in (150a). Aside from the argument offered above, we can also perceive at a more intuitive level the equivalence between (155a) and (155b). Consider any proposition p from the set represented by (155a). By definition, there is some choice function f such that p is the proposition ‘your father is hunting x’, where x is the value that f yields when applied to the set of nonhumans. Given the nature of choice functions, this simply means that p is the proposition ‘your father is hunting x’, for some nonhuman x. Consequently, p is also in the set of propositions represented by (155b). Now, let us consider any proposition p from the set represented by (155b). By definition, there is some nonhuman x such that p is the proposition ‘your father is hunting x’. Again, however, by the nature of choice functions, this entails that there is some choice function f such that p is the proposition ‘your father is hunting x,’ where x is the value that f yields when applied to the set of nonhumans. Thus p is also in the set represented by (155a). We have just seen that our semantics correctly interprets those Tlingit wh-questions where Q is directly adjacent to the wh-word. Let us now consider a somewhat more difficult structure, one where Q is not directly adjacent to the wh-word. We will use as our representative example the wh-question in (67a), Aadóo yaagú sá ysiteen?, ‘Whose boat did you see?’ Given its English translation, we want our semantics to


81

assign as its interpretation the set of propositions represented by the formula in (156b), the set of propositions p such that p is of the form ‘you saw x’s boat’, where x is human. (156)

Tlingit Wh-Question Containing a ‘Pied-Piping Structure’ a. Sentence: Aadóo yaagú sá ysiteen? who boat Q you.saw.it Whose boat did you see? b. Targeted Interpretation: λp [∃x<se> ∈ human. p = [λw. you saw [λw′. the boat(w′) of x(w′) in w′](w) in w]]

Let us now determine whether our semantics can correctly assign the meaning in (156b) to sentence (156a). First, our syntactic theory entails that (156a) has the structure in (157). (157)

The Syntactic Structure of Sentence (156a)

The derivation in (158) computes the meaning assigned by our semantics to the structure in (157). (158)

Semantic Interpretation of Structure (157) a. [[ ForceQP ]]g b. [[ ForceQ1 FocPc ]]g c. λp [ ∃f. p = [[FocPc]]g(1/f) ] d. λp [ ∃f. p = [[ QP FocPb ]]g(1/f) ] e. λp [ ∃f. p = [[ FocPb ]]g(1/f)([[QP]]g(1/f)) ] f. λp [ ∃f. p = [[FocPb ]]g(1/f) (f ({ h : ∃x ∈ human. h = λw′. the boat(w′) of x(w′) in w′ })) ] g. λp [∃f. p = [λx . λw. you saw x(w) in w] (f ({ h : ∃x ∈ human . h = λw′. the boat(w′) of x(w′) in w′ })) ] h. λp [∃f. p = λw. you saw [f ({ h : ∃x ∈ human. h = λw′. the boat(w′) of x(w′) in w′ })](w) in w]]

= = (by (148)) = = (by FA) = (by (132)) = (by standard rules) = (by LC)

We find, then, that our semantics assigns (157) the following set of propositions as its meaning: ‘[λp [ ∃f. p = λw. you saw [f ({ h : ∃x ∈ human . h = λw′. the boat(w′) of x(w′)

82

THE GRAMMAR OF Q

in w′ })](w) in w]]’. This rather complex expression denotes the set of propositions p such that p is the proposition ‘you saw x’, where x is the value obtained by applying some choice function f to the set of individual concepts of the form ‘[λw. the boat(w) of x(w) in w]’, where ‘x(w)’ is human. Let us now determine whether this set of propositions is indeed equivalent to the ‘targeted interpretation’ in (156b). First, let us observe that the general equivalence in (153) again entails that these two sets are identical. Given (153), it follows that the formulas in (159) are equivalent. (159)

Structures Proven Semantically Equivalent a. ∃f. p = [λw. you saw [ f ({[ λy [ λw′. the boat(w′) of y(w′) in w′ ]]}{ x : x ∈ human }) ](w) in w] b. ∃x<se> ∈ human. p = [λw. you saw [ λy [ λw′. the boat(w′) of y(w′) in w′ ]](x)(w) in w]

Again, however, the equivalence of the formulas in (159) entails the equivalence of the two sets in (160). (160) a. λp [ ∃f. p = [λw. you saw [ f ({[ λy [ λw′. the boat(w′) of y(w′) in w′ ]]}{ x : x ∈ human }) ](w) in w]] b. λp [ ∃x<se> ∈ human. p = [λw. you saw [ λy [ λw′. the boat(w′) of y(w′) in w′ ]](x) (w) in w]]

Finally, Function Application and Pointwise Composition entail that the formulas in (160) are equal to those in (161). (161) a. λp [ ∃f. p = [λw. you saw [ f ({ h : ∃x ∈ human . h = λw′. the boat(w′) of x(w′) in w′ }) ](w) in w]] b. λp [ ∃x ∈ human. p = [λw. you saw [ λw′. the boat(w′) of x(w′) in w′ ]](w) in w]]

Given that the formulas in (161) are simply the targeted interpretation in (156b) and the meaning computed in (158), it follows that these two sets are equivalent, and so our semantics correctly assigns the ‘targeted meaning’ in (156b) to the sentence in (156a). The equivalence of the formulas in (161) can also be grasped at a more intuitive level. Let us consider any proposition p from the set in (161a). By definition, this proposition p is of the form ‘[λw. you saw y(w) in w]’, where y is the value obtained by applying some choice function f to the set of individual concepts of the form ‘[λw. the boat(w) of x(w) in w]’, where ‘x(w)’ is human. Given the nature of choice functions, this entails that p is of the form ‘[λw. you saw y(w) in w]’, where y is some individual concept of the form ‘[λw. the boat(w) of x(w) in w]’, where ‘x(w)’ is human. But this simply means that p is of the form ‘[λw. you saw [ λw′. the boat(w′) of x(w′) in w′ ](w) in w], for some human ‘x(w)’. Consequently, p is also a member of the set in (161b). Now let us consider any proposition p from the set in (161b). By definition, p is of the form ‘[λw. you saw [ λw′. the boat(w′) of x(w′) in w′ ]](w) in w]’, where ‘x(w)’ is some human. Of course, this simply means that p is of the form ‘[λw. you saw y(w) in w]’, where y is the individual concept ‘[λw. the boat(w) of x(w) in w]’ and ‘x(w)’ is human. Consequently there is a choice function f such that p is of the form ‘[λw. you saw y(w) in w]’, where y is the value obtained by applying f to the set of


83

individual concepts of the form ‘[λw. the boat(w) of x(w) in w]’, where ‘x(w)’ is human. Thus p is also within the set in (161a), and we have thereby shown the equivalence of these two formulas. Having established the equivalence between the ‘targeted interpretation’ in (156b) and the meaning computed in (158), we may conclude that our semantics assigns the correct interpretation to those Tlingit wh-questions where Q is not directly adjacent to the wh-word. I therefore conclude that our proposed semantic system is adequate for all (single) wh-questions in Tlingit.53 2.7.6.3 A Brief Note on the Interpretation of Pied-Piping Structures In the preceding section we saw that our semantics is able to correctly interpret those Tlingit wh-questions where the sister of Q properly contains the wh-word. Recalling our discussion from chapter 1, such structures constitute the ‘pied-piping structures’ of Tlingit, structures where the fronted phrase of the wh-question properly contains the wh-word. We should note here that this result is of potential significance. The interpretation of pied-piping structures poses a difficult challenge to certain, classic approaches to the semantics of wh-questions. In response to this challenge, some have proposed that pied-piping structures undergo syntactic alterations before interpretation (von Stechow 1996), while others have proposed that the interpretation of pied-piping structures requires additional semantic operations (Sharvit 1998). Under such prior approaches, the interpretation of pied-piping structures requires mechanisms beyond those needed for simple wh-questions without pied-piping. Given the sheer ubiquity of pied-piping structures, both within and across languages, such a state of affairs is rather suspicious. In contrast, the pied-piping structures of Tlingit present no difficulties for the semantics proposed here. The interpretation of such structures requires no mechanisms beyond those required for wh-questions without pied-piping. Indeed, within this system, pied-piping structures are semantically unremarkable. After all, such structures differ from non-pied-piping structures only in the position of the Q, and we saw in (135) that variations in the position of Q have no semantic effect. We may conclude, then, that it is an advantage of the semantics proposed here that the piedpiping structures of Tlingit present it with no prima facie difficulties.

3

Applications to Wh-In Situ Languages

3.1 Introduction Chapter 2 introduced a number of proposals concerning the syntax and semantics of Q-particles. Thus far, these hypotheses have been used to capture facts concerning the wh-questions and wh-indefinites of Tlingit. In this chapter we will see how these ideas might also advance understanding of certain phenomena in wh-in situ languages. I begin by introducing the core proposals concerning wh-in situ languages. I claim that the wh-in situ languages comprise at least two distinct syntactic types: (i) languages where Q adjoins to its sister (Q-adjunction languages), and (ii) those where Q takes its sister as complement, as in Tlingit, but QP-movement occurs covertly (Q-projection languages). I then present evidence that Japanese and Korean are languages of the first type, while Sinhala is a language of the second type. In brief, we will see that this proposed typology predicts certain differences in the distribution of the three languages’ Q-particles. Following the introduction of this typology, I then turn to the semantics of whquestions and wh-indefinites in these languages. I show that the semantics developed in chapter 2 may be applied straightforwardly to the structures of these wh-in situ languages. The final section of this chapter introduces our assumptions regarding so-called (LF/Focus)-Intervention Effects (Beck 1996, 2006). I adopt here the approach developed by Beck (2006), in which such ‘Intervention Effects’ follow from the semantics of wh-words and Q-particles. I demonstrate how the analysis captures the Intervention Effects of Japanese and Korean. The discussion here will lay the groundwork for our broader discussion, in chapter 4, of Intervention Effects in wh-fronting languages. 84

APPLICATIONS TO WH-IN SITU LANGUAGES

85

3.2 The Nature of Wh-In Situ Languages This section presents and defends our proposed typology for wh-in situ languages. Section 3.2.1 introduces the proposed distinction between ‘Q-adjunction’ and ‘Q-projection’ languages, identifying Japanese and Korean as instances of the former and Sinhala as an instance of the latter. Section 3.2.2 demonstrates that this hypothesis predicts certain features of Sinhala da. Section 3.2.3 then shows how the hypothesis captures various features of Japanese ka and Korean ka. 3.2.1 Two Kinds of Wh-In Situ Languages In chapter 2, I presented evidence that some languages (e.g., Tlingit) contain the structure in (1). (1)

Wh-Fronting as a Secondary Effect of Q-Movement CP

QP1

CP

Complementation XP

Q CQ

IP

Agree/ Attract

… wh-word…

QP1 Overt Movement

Within this structure, a Q-particle takes its sister as complement, with the result that a QP node immediately dominates the Q-particle and its sister. Consequently, attraction of the Q-feature to the projection of the interrogative C head entails that the entire QP projection is moved. The reader may recall that this structure is nearly identical to Hagstrom’s (1998) analysis of Japanese wh-questions, repeated in (2). (2)

Hagstrom’s (1998) Analysis of Japanese Wh-Questions CP CP IP

CQ

XP

Agree / Attract

Adjunction XP

Q1

Q1 Overt Movement

…wh-word…

86

THE GRAMMAR OF Q

The structures in (1) and (2) differ in only one respect: in (2), the Q does not take its sister as complement, but rather adjoins to it. Consequently, the node immediately dominating Q and its sister is not a QP, but is instead of the same type as the sister of Q. Thus attraction of the Q-feature into the CP entails only that the Q-particle moves, and the sister of Q is left in situ. Since the sister of Q contains the wh-word, such a language is predicted to be wh-in situ. We find, then, that (2) might accurately represent the structure of certain wh-in situ languages. Interestingly, however, if we accept the existence of the structure in (1) and the existence of covert movement, then there is also another imaginable means for deriving wh-in situ structures. (3)

Covert QP-Movement as a Source of Wh-In Situ CP

QP1

CP

Complementation XP

Q CQ

… wh-word…

IP

Agree/ Attract QP1 Covert Movement

Like (2), the structure in (3) differs from (1) in just one respect: in (3), the movement of the QP is covert. Since the QP in (3) is pronounced in its base position, it follows that the wh-word is also pronounced in its base position. Thus (3) might also represent the structure of certain wh-in situ languages. In the following subsections, I will argue that both the structures in (2) and (3) are attested in the world’s languages. Languages whose wh-questions have the structure in (2) will be referred to as ‘Q-adjunction languages’, while those whose wh-questions have the structure in (3) will be labeled ‘Q-projection languages’.1 I will present evidence that Sinhala is a Q-projection language, while Japanese and Korean are Q-adjunction languages.

3.2.2 The Distribution of Sinhala Da In the previous chapter we saw that the Sinhala particle da shares a variety of properties with the Tlingit particle sá. We also saw that by labeling both these particles as ‘Q’, our syntactic and semantic theory predicts a number of these shared properties: (i) their obligatory co-occurrence with wh-words, (ii) their need to c-command a wh-word, and (iii) their behavior with respect to islands.


87

What, though, of the properties observed in section 2.4.4? To recall, neither Sinhala da nor Tlingit sá can appear at the right edge of a matrix clause, while both particles can appear at the right edge of subordinate clauses. The relevant data for Sinhala da are repeated in (4) and (5). (4)

Sinhala Da Cannot Appear at the Right Edge of a Matrix Clause (Kishimoto 2005) a. Chitra monawa da gatte? Chitra what Q buy What did Chitra buy? b. * Chitra monawa gatta da? Chitra what buy Q (Kishimoto 2005: 3–4)

(5)

Sinhala Da Can Appear at the Right Edge of a Subordinate Clause Ranjit [ kauru aawa kiyala ] da danne? Ranjit who came that Q know Who does Ranjit know came? (Kishimoto 2005: 13)

Given that this shared pattern was taken as evidence for a shared syntactic category, it must receive a uniform account in both Tlingit and Sinhala. When we look back to our account of the parallel Tlingit data, we find that our analysis crucially relies upon the assumption that Tlingit is a Q-projection language. Under this assumption, merger of Tlingit sá at the edge of a matrix CP would violate the QP-Intervention Condition. Such a position for sá would entail the existence of a QP intervening between a functional head and its complement. (6)

Tlingit Sá Taking as Complement Projections in the Matrix ‘Functional Spine’ F2P QP F1P

F2 Q

In this context, let us observe that if Tlingit were a Q-adjunction language, nothing would rule out the appearance of a sá matrix-finally. After all, as the structure in (7) illustrates, if Tingit sá did not project a QP, then a matrix-final position would not violate the QP-Intervention Condition. (7)

Q-Particles in Q-Adjunction Languages Can Be in the Matrix ‘Functional Spine’ F2P F1P F1P

F2 Q

88

THE GRAMMAR OF Q

We find, then, that our account of the Tlingit data parallel to (4) ultimately relies upon the assumption that Tlingit is a Q-projection language. Consequently, if we wish to extend this account to the parallel Sinhala data under (4), we must likewise assume that Sinhala is a Q-projection language. Finally, note that the syntactic and semantic assumptions introduced in chapter 2 entail that, within whquestions, a Q-particle must move into the left periphery by LF. Given that the surface form of a Sinhala wh-question places the Q-particle and the wh-word in their base positions, we must conclude that the QP of a Sinhala wh-question moves covertly to CP. We find, then, that when combined with our proposals from chapter 2, the facts in (4) entail that Sinhala wh-questions have the structure in (3). Further evidence for the analysis in (3) can be found in certain additional constraints governing the position of Sinhala da. It appears that like Tlingit sá, Sinhala da cannot appear (i) between a postposition and its complement, (ii) between a possessor and its possessed NP, or (iii) between a determiner and its NP complement. This is stated in the following passage from Kishimoto (2005). (8)

Further Conditions Governing the Distribution of Sinhala Da “It is not possible to place da immediately after a wh-word embedded inside a PP or DP” (Kishimoto 2005: 13).

Although Kishimoto (2005) does not illustrate this generalization with negative data, it is evident from the context that the following data pattern is intended. (9)

No Q Between a Postposition and Its Complement in a Sinhala Wh-Question a. Chitra [ kauru ekka ] da kataa kalee? Chitra who with Q talk did Who did Chitra talk with? (Kishimoto 2005: 13) b. * Chitra [ kauru da ekka] kataa kalee? Chitra who Q with talk did

(10)

No Q Between a Possessor and the Possessed NP in a Sinhala Wh-Question a. Chitra [ kaa-ge amma ] da daekke? Chitra who-GEN mother Q saw Whose mother did Chitra see? (Kishimoto 2005: 13) b. * Chitra [ kaa-ge da amma ] daekke? Chitra who-GEN Q mother saw

(11)

No Q Between a D and Its NP Complement in a Sinhala Wh-Question a. Chitra [ mona pota ] da gatte? Chitra what book Q bought What book did Chitra buy? (Kishimoto 2005: 13) b. * Chitra [ mona da pota ] gatte? Chitra what Q book bought

We saw in chapter 2 that Tlingit sá is also subject to these conditions. Thus, in order to have a unified account of this pattern, we must apply our analysis of it


89

in Tlingit to its realization in Sinhala.2 When we turn to that analysis, however, we find that it again rests on the assumption that Tlingit is a Q-projection language. Under that assumption, the ill-formed structures in (b) all contain violations of the QP-Intervention Condition, while the well-formed structures in (a) do not. Moreover, as we will soon see in greater detail, if Tlingit were not a Q-projection language, then our system would predict that the placement of sá would not follow the pattern above. Given that this pattern is also found for Sinhala da, we must likewise conclude that Sinhala is a Q-projection language. We may therefore conclude that the data in (9)–(11) provide further evidence that Sinhala wh-questions possess the structure in (3): they involve covert movement of a QP projection. 3.2.3 The Distribution of Japanese Ka and Korean Ka We have just seen that further parallels between Tlingit sá and Sinhala da force the view that Sinhala is a Q-projection language. In this section we will see that the converse holds for the languages Japanese and Korean. That is, various differences between the Qs of Tlingit/Sinhala and Japanese/Korean suggest that Japanese and Korean are Q-adjunction languages. We begin in the following subsection by outlining how this perspective can capture the behavior of ka in Japanese and Korean wh-questions. Following this, we turn to wh-indefinites in these two languages, and show that they independently reveal Japanese and Korean to be of the Q-adjunction type. 3.2.3.1 Behavior in Wh-Questions As was briefly noted in chapter 2, the particle ka in Japanese differs from Tlingit sá and Sinhala da in that it can (and indeed must) appear at the right edge of a matrix wh-question. (12)

Japanese Ka Can Appear at the Right Edge of a Matrix Clause John-ga nani-o kaimasita ka ? John-NOM what-ACC bought.polite Q What did John buy?

This property is shared with the homophonous Q-particle in Korean, as illustrated in (13). (13)

Korean Ka Can Appear at the Right Edge of a Matrix Clause Eti-ey sensayng-nim-i ka-sipni-kka? Where.to teacher-HON-NOM go-HON-Q Where did the teacher go?

We observed earlier that the inability for Tlingit sá and Sinhala da to appear matrixfinally ultimately follows from their being Q-projection languages. Similarly, we will

90

THE GRAMMAR OF Q

now see that the ability for Q to appear matrix-finally in Japanese and Korean would follow from their being Q-adjunction languages. Assuming that Japanese and Korean are Q-adjunction languages, there are actually two ways in which Q could come to appear matrix-finally. First, Q could be initially merged in this position. Recall that in Q-adjunction languages, Q does not project a QP when it merges with its sister. Consequently, the QP-Intervention Condition would not be violated if Q were to take as its sister the phrase immediately below the interrogative ForceQ head. (14)

Q-Particles in Q-Adjunction Languages Can Be Initially Merged in the Periphery ForceQP FocP FocP

ForceQ Q ka

… wh-word…

As we will see later in section 3.3, our semantics will interpret the structure in (14) as a wh-question. Thus (14) represents one possible structure underlying the sentences in (12) and (13). Interestingly, though, our account does not predict that Q in these languages must be initially merged as in (14). That is, nothing in our account prevents the merger of Q lower down, within the IP of the wh-question. What, then, does our theory predict for this latter type of structure? The answer, of course, is already provided by the diagram in (2). As we see there, if ever the Q is merged at a lower, clause-internal position, then that Q subsequently undergoes movement to the periphery of the clause. In a Q-adjunction language, such movement targets the Q-particle alone, leaving the wh-word in its base position. Thus the matrix-final position of Q could, in a Q-adjunction language, be derived via movement of the Q-particle from an underlying clause-internal position. We therefore find that, in the wh-questions of Q-adjunction languages, Q-particles can (and indeed must) appear at the edge of the matrix clause. Moreover, we saw in the previous section that Q-particles cannot appear matrix-finally in the wh-questions of Q-projection languages. Therefore, given the pattern of data in (12) and (13), we must conclude that Japanese and Korean are Q-adjunction languages, and that their wh-questions possess the structure in (2).3 In section 3.2.3.2, we will see that the wh-indefinites of these languages provide further evidence for this analysis. 3.2.3.2 Behavior in Wh-Indefinites To begin, let us note that the analysis in (2) predicts that Japanese and Korean wh-indefinites should permit the Q to come between an adposition and its DP complement. Consider the structure in (15).


91

(15) Structures Where Q Appears Between P and Its Complement in Japanese/Korean PP DP DP

P Q

…wh-word…

As seen here, in a Q-adjunction language, a Q-particle does not project the category of the phrase minimally dominating it and its sister. Therefore, in a structure of the form ‘[wh-word . . . Q . . . P]’, there is no QP projection intervening between the P and the DP it selects for. Consequently, such structures do not violate the QP-Intervention Condition, and they are predicted to be well formed. This prediction is indeed accurate, as the sentences in (16) and (17) demonstrate. (16)

Japanese Q Can Appear Between a Postposition and Its Complement a. Taroo-wa doko- ka-e itta. Taro-TOP where-Q-to went Taro went somewhere. b. Taroo-ga [ dono tosi ]-ka-e ryoko sita-rasii. Taro-NOM which city-Q-to travel did-seems Taro seems to have traveled to some city.

(17) Korean Q Can Appear Between a Postposition and Its Complement4 Ku-nun eti-eyn-ka-ey ka-ess-ta. he-TOP where-LINK-Q-to go-PST-DEC He went somewhere.

Furthermore, our account in (2) predicts that Japanese and Korean wh-indefinites should permit the Q to appear between a wh-possessor and the possessed NP. That is, in a Q-adjunction language, nothing rules out structures like that in (18). (18) Structures Where Q Appears Between Possessor and Possessed in Japanese/Korean DP DP DP …wh-word…

D’ Q

D POSS

NP possessum

In this structure, Q is adjoined to its sister, and so no projection of Q intervenes between the possessive D head and the possessor that it selects. Our QP-Intervention

92

THE GRAMMAR OF Q

Condition, then, permits (18). Thus we predict that wh-indefinites in Japanese and Korean should permit Q to appear between the possessor and the possessum. As the following sentences demonstrate, this is again an accurate prediction. (19) Japanese Q Can Appear Between a Possessor and Possessed NP Taroo-wa [ dare-ka-no oniisan ]-ni atta. Taro-TOP who-Q-GEN brother-DAT met Taro met someone’s older brother. (20)

Korean Q Can Appear Between a Possessor and Possessed NP Ku-ka [ nwukwu-in-ka-uy tongsayng ]-ul manna-ess-ta. he-TOP who-LINK-Q-GEN brother-ACC meet-PST-DEC He met someone’s brother.

Finally, note that (without additional assumptions) our theory predicts that Q-particles in Q-adjunction languages should be able to intervene between wh-determiners and their NP complements. After all, nothing stated thus far would rule out structures like that in (21). (21)

Q Appearing Between D and Its NP Complement in a Q-Adjunction Language DP D D wh-word

NP Q

This prediction, however, is incorrect for Japanese and Korean. Even in Japanese and Korean, a Q-particle cannot intervene between a D and its NP complement, as the sentences in (22) and (23) illustrate. (22)

Japanese Q Cannot Appear Between D and Its NP Complement a. Taroo-ga [dono hito ]-ka-o hoomon sita-rasii. Taro-NOM which man-Q-ACC visit did-seem Taro seems to have visited some man. b. * Taroo-ga [dono-ka hito ]-o hoomon sita-rasii. Taro-NOM which-Q man-ACC visit did-seem.

(23)

Korean Q Cannot Appear Between D and Its NP Complement a. Ku-ka [ enu salam ]-in-ka-lul manna-ess-ta. he-NOM which man-LINK-Q-ACC meet-PST-DEC He met some man. b. * Ku-ka [enu-in/eyn-ka salam ]-ul manna-ess-ta. he-NOM which-LINK-Q man-ACC meet-PST-DEC

On the other hand, our account is not necessarily inconsistent with the facts in (22) and (23), as the impossibility of these structures may result from independent


93

factors. To build toward one possible explanation, note that the structure in (21) differs from those in (15) and (18) in that the Q-particle in (21) is adjoined to a head.5 Thus the D-head in (21) is initially merged with Q, rather than with the NP constituting its internal argument. Suppose, however, that selection for the internal argument of a head H must be satisfied no later than at the point where H first externally merges with something.6 Under this assumption, the ill-formedness of (21) would follow. Since initial merger of D in (21) joins it with Q, and Q does not contain the phrase selected by D, a selectional violation occurs. Therefore we find that independent factors relating to selection may be responsible for the ill-formedness of (21). 3.2.4 Section Summary The data presented in section 3.2.3 argue that Japanese and Korean wh-questions receive the ‘Q-adjunction’ analysis in (2). We have also seen that the facts in section 3.2.2 provide evidence that Sinhala wh-questions receive the ‘Q-projection’ analysis in (3). Thus it seems that both (2) and (3) represent attested structures, rendering the so-called wh-in situ languages a heterogeneous class. Finally, we have seen that it is possible to determine whether a given wh-in situ language is of the Q-adjunction or Q-projection type by examining the distribution of its Q-particles; in particular, by determining whether Q can appear between functional heads and the phrases those heads select for.

3.3 The Semantics of Wh-Indefinites and Wh-Questions in Wh-In Situ Languages In this section we move from the syntax of wh-in situ languages to their semantics. We will quickly confirm that the semantic system introduced in chapter 2 correctly interprets the wh-indefinites and wh-questions of the Q-adjunction languages. Such confirmation will also provide a concrete illustration of how our semantics can be applied to the structures of these languages. 3.3.1 The Semantics of Wh-Indefinites in Wh-In Situ Languages Let us begin by briefly considering the wh-indefinites of in situ languages like Sinhala, which possess the Q-projection structure in (3). Recall that such languages differ from Tlingit only in that their QPs move covertly in wh-questions. Thus the wh-indefinites of such languages will not differ in their structure from the wh-indefinites of Tlingit. Given the adequacy of our semantics for Tlingit, we may conclude that it will also correctly interpret the wh-indefinites of these wh-in situ languages. Now let us turn to the wh-indefinites of the Q-adjunction languages, such as Japanese and Korean. Recall that in such languages, the Q-particle is adjoined to its sister in a wh-indefinite. Note, however, that since no movement of the Q-particle occurs in a wh-indefinite, such structures are otherwise identical to the wh-indefinites of the Q-projection languages. Finally, given that the semantic principles of chapter

94

THE GRAMMAR OF Q

2 make no reference to whether the Q-particle is adjoined to its sister or not, it follows that this semantics will also correctly interpret the wh-indefinites of the Q-adjunction languages. To illustrate this result, the Q-adjunction structure in (24) is interpreted according the semantic system in chapter 2. The reader is invited to confirm that our semantic system can also interpret any of the other examples of wh-indefinites in Q-adjunction languages found throughout our discussion. (24)

Semantics of a Wh-Indefinite in a Q-Adjunction Language7 nani-ka-o katta. Japanese: John-ga John-NOM what-Q-ACC bought John bought something.

3.3.2 The Semantics of Wh-Questions in Wh-In Situ Languages Having shown that our semantic system can interpret the wh-indefinites of both types of wh-in situ languages, we now turn our attention to their wh-questions. Let us begin again with those languages possessing the structure in (3). Recall, again, that such languages differ from Tlingit only in that their QPs move covertly. Thus, at LF—the structural input to semantic interpretation—the wh-questions of these languages will not differ in any relevant way from the wh-questions of Tlingit. Given the adequacy of our semantics for Tlingit wh-questions, we may therefore conclude that it will also correctly interpret the wh-questions of these wh-in situ languages. Our semantics is thereby adequate for the wh-questions of Sinhala. Let us now consider the wh-questions of the Q-adjunction languages. Recall that in the wh-questions of such languages, the Q-particle obligatorily appears at the periphery of the clause, just beneath the interrogative ForceQ head. Recall, also, that the Q-particle obtains this peripheral position either via movement or via base generation. To simplify our discussion here, we will adopt the following additional syntactic assumption.


(25)

95

The Absence of Q-Traces at LF 8 If a Q-particle alone undergoes movement (rather than a QP), then its traces are deleted by LF.

According to (25), there are no lower copies of Q at LF. Consequently, there is no syntactic difference at LF between those wh-questions where Q is base generated at its surface position (14) and those where it moves to that position (2). That is, by LF, all wh-questions in Q-adjunction languages have a structure akin to that in (26). (26) The LF Structure of Wh-Questions in Q-Adjunction Languages nani-o kaimasita ka? Japanese: John-ga John-NOM what-ACC bought Q What did John buy? ForceQP t> FocPb FocPa IPb John

Foc

IPa

f-sem: t>

f-sem: t>

Infl

VP DPF

Q1 ka

f-sem: t>

f-sem: t>

f-sem: <se,t>

ForceQ1

<st>

t>>

f-sem: t>

V f-sem: >t> kaimasita

nani

Let us now determine whether our semantics assigns the correct interpretation to this structure. Given its English translation, we wish to derive as the meaning of (26) the following set of propositions: those propositions of the form ‘John bought x’ where x is some nonhuman. This set, which is intuitively the set of possible answers to question (26), may be represented via the lambda notation in (27).

(27)

The Targeted Interpretation of Structure (26) λp [ ∃x<se> ∉ human. p = [λw. John(w) bought x(w) in w] ]

Assuming that structure (26) is representative, the following derivation demonstrates how our semantic system interprets the wh-questions of a Q-adjunction language. (28)

Semantic Interpretation of Structure (26) a. [[ ForceQP ]]g b. [[FocPbForceQ1]]g c. λp [ ∃f . p = [[FocPb]]g(1/f) ]

= = (by (148))9 =

96

THE GRAMMAR OF Q

d. e. f. g. h. i. j. k. l.

λp [ ∃f . p = [[ FocPa Q1 ]]g(1/f)] λp [ ∃f . p = [ [[ Q1 ]]g(1/f) ( [[ FocPa ]]F g(1/f)) ]] λp [ ∃f . p = [f ( [[ FocPaw ]]F g(1/f))]] λp [ ∃f . p = [f ( [[ [ John [ naniF kaimasita ]] ]]F g(1/f) )] ] λp [ ∃f . p = [f ( [[ naniF kaimasita ]]F g(1/f) ([[John]]F g(1/f)) )] ] λp [ ∃f . p = [f ( ( [[kaimasita ]]F g(1/f)([[naniF]]F g(1/f))) ([[John]]F g(1/f)) )] ] λp [ ∃f . p = [f ( ({λx<se> λy<se> λw. y(w) bought x(w) in w } ([[naniF]]F g(1/f)))({John}) )] ] λp [∃f. p = [f (({λx<se> λy<se> λw. y(w) bought x(w) in w} ({x<se> : x ∉ human}))({John}))]] λp [ ∃f. p = [ f ({ q<st> : ∃x<se> ∉ human. q = [λw. John(w) bought x(w) in w] } )]]

= (by (129)) = (by (126)) = = (by (120)) = (by (120)) = (by (119a)) = (by (124)) = (by (120)) =

We find, then, that our semantic system assigns the following set of propositions as the interpretation of the wh-question in (26): those propositions p such that p is the value some choice function f yields when applied to the set of propositions of the form ‘John bought x’, where x is nonhuman. As a final step in proving the adequacy of our semantics, let us determine whether this latter set is equivalent to that represented in (27). Indeed, such a determination is rather trivial. After all, given that choice functions simply yield members of their arguments, the set computed in (28) is clearly just the set of propositions p such that p is ‘John bought x’, where x is nonhuman. This latter set is, of course, the one in (27).10 We have thus demonstrated that our semantics assigns the correct interpretation to the wh-question structure in (26). Again, assuming this structure to be representative of all wh-question structures in all Q-adjunction languages, we have thereby shown that our semantics is sufficient to correctly interpret the wh-questions of Japanese, Korean, and all other Q-adjunction languages. 3.3.3 Section Summary The results obtained in section 3.3.1 show that our semantics can interpret the wh-indefinites of both Q-adjunction and Q-projection languages. Section 3.3.2 shows that this semantics can also interpret the wh-question structures of both these language types. We may conclude, then, that the semantics from chapter 2 is sufficient to handle both whquestions and wh-indefinites in all wh-in situ languages. Combined with the earlier semantic results from chapter 2, we may conclude that our semantics is sufficient for the wh-indefinites and wh-questions of all the language types predicted by our theory. Let us finally note in passing that the cross-linguistic applicability of this semantics is a potential advantage of our overall theory of wh-questions and wh-indefinites.

3.4 The Theory of LF/Focus-Intervention Effects In chapter 4 we will discuss at length the potential consequences our theory has for the analysis of LF/Focus-Intervention Effects in wh-fronting languages. In this


97

section we will lay the groundwork for that later discussion by introducing the theory of LF/Focus-Intervention Effects that we will assume here. Our discussion here will center on the relatively simple case of LF/Focus-Intervention Effects in Q-adjunction languages such as Korean. Let us begin by reviewing the basic facts regarding LF/Focus-Intervention Effects. In many languages, it is not possible for an in situ wh-word to be in the scope of any of a set of ‘offending operators’. Thus the Korean sentence in (29) is illformed, as the wh-word nuku-lul ‘who-ACC’ is in the scope of the offending operator–man ‘only’. (29)

Intervention Effect in Korean (Beck 2006: 3) * [ Minsu-man nuku-lul po-ss]-ni? Minsu-only who-ACC see-PST-Q Who did only Minsu see?

Such ill-formed structures may be repaired by movement of the wh-word to a position higher than the offending operator. Thus (30) is well formed, as nuku-lul is no longer c-commanded by–man. (30)

Obviation of Intervention Effect via Movement of the Wh-Word (Beck 2006: 3) [ Nuku-lul Minsu-man po-ss]-ni? who-ACC Minsu-only see-PST-Q Who did only Minsu see?

One way of describing the contrast between (29) and (30) is that the none of the ‘offending operators’ can come between, or intervene between, the Q-particle and the wh-word. If such intervention occurs, the resulting sentence is ill-formed, a state of affairs referred to as an ‘(LF- or Focus-) Intervention Effect’. We will assume the analysis of these phenomena developed by Beck (2006). Under that analysis, the ill-formedness of sentences like (29) follows from the principles in (31) and (32), first introduced in chapter 2. (31)

Principle of Interpretability (Beck 2006: 16) A sentence must have a normal-semantic value.

(32)

Uniqueness of the Q-Particle (Beck 2006: 13) The Q-particle is the only focus-sensitive operator whose meaning does not also take as input the normal-semantic value of its sister.

In brief, it is assumed that the ‘offending operators’ triggering Intervention Effects are simply the focus-sensitive operators other than Q. Given (32), all such operators require computation of the normal-semantic value of their sisters. As the LF of a sentence like (29) is assumed to be that in (33), it follows that computation of its meaning entails that one compute the normal-semantic value of the wh-word nuku.

98

THE GRAMMAR OF Q

(33) Deriving Intervention Effects in Q-Adjunction Languages ForceQP ForceQ

FocP FocP

Q ni

IP

Foc

-man ONLY

IP

DP Minsu

IP I

VP DP nuku-lul

V poss

However, because wh-words are assumed not to have normal-semantic values (see chapter 2), it follows that a normal-semantic value cannot be computed for sentence (29), in violation of principle (31). Sentence (29) is therefore uninterpretable, and thus deviant. On the other hand, sentences like (30) are predicted to be semantically interpretable. The fronting of the wh-word in (30) entails that this sentence may be assigned the LF in (34). (34) Obviation of Intervention Effects in Q-Adjunction Languages ForceQP FocP

Force Q

Q ni

FocP IP

Foc IP

DP 1 Nuku-lul -man ONLY

IP

DP Minsu

IP I

VP DP t1

V poss

In this LF structure, the complement of–man ‘only’ does not contain a wh-word.11 Therefore the interpretation of (30) does not require one to compute the normal-semantic value of a wh-word. Thus (30) is interpretable by our system, and is not predicted to be deviant.


99

We see, then, that our semantics correctly predicts the contrast between (29) and (30). Of course, our reasoning here generalizes to many other structural types. Consequently, our semantics predicts that the type of structure in (35) is that which triggers an Intervention Effect. (35) Configuration Resulting in an Intervention Effect [ … Q [ … Offending Operator

[ … [ wh-word ] … ] ] ] No Q-particle

Again, due to (32), interpreting this type of structure requires that one compute the normal-semantic value of the sister to the ‘offending operator’. However, the absence of a Q-particle within its sister entails that one must eventually compute the normalsemantic value of the wh-word. Since wh-words are assumed not to have normalsemantic values, the semantic computation therefore crashes, and deviance results. The following quote nicely states this result of our overall semantic theory: “The system I have introduced requires a wh-phrase to have as its first c-commanding operator a Q operator” (Beck 2006: 16). We have just seen how our adoption of the core hypotheses of Beck (2006) allows us to employ without modification her analysis of the Intervention Effects found in Q-adjunction languages like Korean and Japanese. In chapter 4 we will see that when we combine these proposals with our theory of wh-fronting in (1), we can also capture various features of Intervention Effects found in the wh-fronting languages of the world.

4

Applications to Other Wh-Fronting Languages, Pied-Piping, and Intervention Effects

4.1 Introduction Chapter 2 presented evidence that the structure of Tlingit wh-questions is that in (1). (1)


QP1

CP

Complementation XP

Q CQ

IP

Agree/ Attract

… wh-word…

QP1 Overt Movement

In this chapter and the next, I will argue that this ‘Q-based’ analysis should be extended to the wh-questions of all wh-fronting languages. That is, I will argue that there is no language whose wh-questions display the structure in (2), where the wh-word alone is directly attracted into the left periphery.

100

APPLICATIONS TO OTHER WH-FRONTING LANGUAGES, PIED-PIPING , AND INTERVENTION EFFECTS

(2)

101

Wh-Fronting as Direct Attraction of the Wh-Word CP wh-word 1

CP Cwh

IP

Agree/ Attract wh-word1 Overt Movement

I begin in section 4.2 with a few brief, initial arguments for the abandonment of (2). I first raise some general typological and learning-theoretic considerations that motivate our application of (1) to other wh-fronting languages. Following this, I briefly discuss the Edo language (Niger–Congo; Nigeria), whose wh-questions have a surface form that seems well suited to (1). I next consider how the Q-based analysis in (1) can, in conjunction with the QP-Intervention Condition, predict the ill-formedness of P-stranding and certain left-branch extractions across many wh-fronting languages. I also discuss the analysis of languages in which these extractions are allowable. Finally, I briefly compare the account in (1) to three earlier, similar accounts: Watanabe (1992), Tanaka (1998, 1999), and Sternefeld (2001). Having presented these initial arguments, section 4.3 begins our longer discussion of the ways in which (1) might advance our understanding of pied-piping structures. Here, I focus on the ways in which the syntax and semantics of pied-piping structures are simplified under the Q-based account. Indeed, as we will see, there is a sense in which the very existence of ‘pied-piping’ (as defined in chapter 1) is called into question. More in-depth discussion of pied-piping will come in chapter 5. The final section of this chapter examines two widely studied parameters of variation between the wh-fronting languages: (i) the presence of Superiority Effects, and (ii) the presence of Intervention Effects. We will see that the Q-based account in (1), in conjunction with our semantics from chapter 2, accurately predicts that multiple wh-questions should evince Superiority if and only if they do not evince Intervention Effects (Pesetsky 2000). Furthermore, we will see that our Q-based account captures the fact that ‘pied-pipers’ are universally subject to Intervention Effects (Sauerland & Heck 2003).

4.2 The Generality of the Q-Based Structure: Some Initial Motivation In this section I present three comparatively brief arguments for extending the Q-based analysis in (1) to all wh-fronting languages. Following this, I discuss the similarities and differences between the analysis in (1) and those of Watanabe (1992), Tanaka (1998, 1999), and Sternefeld (2001).

102

THE GRAMMAR OF Q

4.2.1 General Typological and Learning-Theoretic Considerations If we accept that there are languages possessing the structure in (1), then general typological and learning-theoretic considerations suggest that (2) should not be a possible structure of human language. To begin, it should theoretically be possible for Q in some languages to be phonologically null. Indeed, there is a well-established tradition in the syntactic literature that views the wh-questions of English as having a ‘silent’ version of the structure pronounced as ka in Japanese (Baker 1970; Cheng 1991). Thus the general notion that Japanese ka might be ‘structurally present but phonologically absent’ in some languages is as well accepted as any in theoretical syntax. In this context, recall our arguments that Japanese ka and Tlingit sá are the same formal entity. Consequently, it should be possible for the particle pronounced as sá in Tlingit to be phonologically null in some languages. Let us therefore consider a language that is identical to Tlingit, except that its Q is unpronounced. To imagine such a language, simply delete sá from all the Tlingit sentences that we have seen. Interestingly, such a language would appear (to both the linguist and the child learner) exactly like a wh-fronting language of the kind we are familiar with.1 Therefore, having accepted the analysis in (1) for Tlingit wh-questions, as well as the possibility of phonologically empty Q-particles, it would be simplest to also apply (1) to the wh-questions of the more familiar wh-fronting languages. To put the matter more acutely, if we accept that there are languages with the structure in (1), and that there are languages where Q is null, then the logical independence of those properties entails that there should also be languages with the structure in (1) but where Q is null. Since such languages would, on the surface, appear very similar to languages with the structure in (2), we must ask what subtler properties might distinguish them. That is, what evidence could show that a language’s wh-questions have the structure in (2) rather than (1)? Interestingly, as we will see throughout this chapter and the next, many languages purported to have the structure in (2) receive interesting accounts under the Q-based analysis in (1). Indeed, to my knowledge, there is no evidence that would force the analyst to conclude that a particular language has the structure in (2) rather than that in (1). Therefore a linguistic typology admitting of both the structures in (1) and (2) would seem to be dubious. Similarly, adopting the position of the language learner, the absence of any phenomena necessitating the analysis in (2) would entail that a hypothesis space containing both (1) and (2) would create a substantially more difficult learning task. Following this line of thought, I conclude that in no language—not even English— do wh-words bear a direct syntactic relationship with interrogative C, as in (2). Rather, in all languages, interrogative C probes and Agrees with the Q-particles accompanying the wh-words, as in (1). The left-peripheral position of wh-words in some languages is merely an epiphenomenal consequence of the fronting of QP. Pursuing these ideas further, we find that whether a language requires overt fronting of wh-words depends upon two parameters: (i) whether the movement of Q(P) is overt, and (ii) whether the Q-particle takes its sister as complement. Under this view, wh-fronting languages are simply those whose Qs move overtly and take


103

their sisters as complement. A third, independent property affecting the surface appearance of a language’s wh-questions is whether the Q-particles have any phonological content. The table in (3) illustrates the typology emerging from this perspective. As (3) indicates, this perspective invites the notion that (i) English differs from Tlingit only in that English Q-particles are phonologically null, (ii) Tlingit differs from Sinhala only in that QP-movement in Tlingit is overt, and (iii) Sinhala differs from Japanese in that Japanese Q adjoins to its sister and moves overtly. (3)

The Emerging Typology

Language

Movement of Q-Projections: Covert/Overt

Q-Particle Takes Sister as Complement: Yes/No

Phonology of Q-Particle: Null/ Pronounced

English Tlingit Sinhala Japanese

Overt Overt Covert Overt

Yes Yes Yes No

Null Pronounced Pronounced Pronounced

As our discussion proceeds, two additional parameters will be added to the theory described above: (i) whether a language permits multiple wh-questions to contain multiple Qs (section 4.4), and (ii) whether a language requires Q and the wh-word to undergo Agreement (section 5.2). 4.2.2 The Edo Language: Another Wh-Fronting Language With Overt QPs In the following sections I will argue that many of the most familiar wh-fronting languages receive interesting analyses under the assumption that they possess phonologically null Q-particles. Of course, if it turned out that no wh-fronting language besides Tlingit seemed to possess an overt instance of Q, the resulting theory of whfronting would look rather suspicious. However, as I will argue, there are indeed other wh-fronting languages that seem to overtly possess the QP structure in (1). It is important to note that the claim here is not merely that there are wh-fronting languages whose wh-questions contain something we would pretheoretically dub a ‘question particle’. After all, our analysis in (1) assumes the existence of an interrogative C in wh-questions. Thus a prima facie ‘question particle’ might simply be an overt pronunciation of C, and not an instance of Q. Moreover, since the analysis in (2) shares the assumption that wh-questions contain an interrogative C, the existence of languages where interrogative C is overt does not specifically argue for our Q-based account. Thus we must be careful to argue for the following, more specific claim: there are wh-fronting languages whose wh-questions overtly contain an element that is best analyzed as a ‘Q’, in our special sense of the term. The Niger–Congo language Edo, spoken in Nigeria, provides a particularly striking case of such a language.2 As described by Baker (1999), the wh-questions of

104

THE GRAMMAR OF Q

Edo obligatorily contain a particle dè, which Baker glosses as ‘Q’. The sentences in (4) illustrate. (4)

The Structure of Edo Wh-Questions3 a. Dè òmwàn nè Àdésúwà bó!ó ? Q who that Adesuwa comfort Who did Adesuwa comfort? (Baker 1999) b. Dè èmwìn nè Òzó há!é Úyì rè ? Q what that Ozo pay Uyi it What did Ozo pay Uyi? (Baker 1999) c. Dè [ èbé [ òmwán ] ] nè Úyì dé ? Q book whose that Uyi buy Whose book did Uyi buy? (Baker 1999)

In his study of these structures, Baker (1999) first considers the hypothesis that dè is some kind of wh-determiner. Interestingly, the evidence he puts forth against this hypothesis concerns the inability for dè to appear inside of PPs and DPs. As Baker notes, wh-determiners in English and other languages can generally appear in such structures, where they pied-pipe the larger phrase in which they are contained. (5)

Wh-Determiners Inside SpecDP and CompPP in English a. [ [ Which person’s ] book ] did Uyi buy? b. [ To [ which person ] ] did Uyi give the coconut?

In Edo, however, the particle dè cannot appear in either of these environments. As the following datum illustrates, if dè ever intervenes between a possessor and a possessed NP, the sentence is ill-formed. (6)

Edo Dè Cannot Appear Between Wh-Possessor and Possessed NP (cf. (4c)) * [Èbé [ dè òmwán ] ] nè Úyì dé ? book Q whose that Uyi buy (Baker 1999)

Similarly, if dè ever appears between a P and its DP complement, the sentence is illformed. (7)

Edo Dè Cannot Appear Between a P and Its DP Complement * [ Nè [ dè òmwàn ] ] nè Òzó rhié né!né ívìn? to Q who that Ozo give the coconut To whom did Ozo give the coconut?

From these facts, Baker concludes that dè should not be analyzed as a wh-determiner like the English which. Importantly, our Q-based theory of wh-fronting can provide an account of these facts, if we assume that Edo dè is a Q-particle. Because Edo is a wh-fronting language, our analysis entails that it must be a Q-projection language, where the Q-particle takes its sister as complement and projects a QP. Given our QP-Intervention Condition, we predict that Q-particles in Edo will be unable to appear between (i) wh-possessors and


105

their possessed NPs and (ii) prepositions and their DP complements. Consequently, if we suppose that dè is the overt realization of Q in Edo, then we predict the data in (6) and (7).4 Furthermore, let us also observe that Edo wh-questions contain a separate, overt realization of the interrogative C head; each of the sentences above contains the element nè, which generally functions as a complementizer in the language. We must conclude, then, that not only does dè receive an interesting analysis as Q, it cannot receive an analysis as the interrogative C. We find, then, that the special behavior of dè in Edo suggests that it is a ‘Q’ in our sense of the term.5 Thus Edo provides us one other case where the QP structure in (1) is overtly pronounced in a wh-fronting language.6 Moreover, as we have just seen, Edo also demonstrates that it is possible for both the Q-particle and the interrogative C to be overtly pronounced in the same language. Thus these structures provide independent evidence for the distinction between Q-particles and interrogative Cs. The surface form of Edo wh-questions confirms that the structure in (1) is not limited to the Tlingit language. This fact, in turn, bolsters the view that (1) is a structure licensed by UG, and so supports the application of (1) to the wh-questions of Tlingit. Furthermore, the fact that the structure in (1) is transparently found in two unrelated languages on separate continents suggests that it is not an ‘aberration’ specific to a single language family or area, but might instead have a broader distribution in the languages of the world. This latter conclusion, in turn, renders more plausible the notion that (1) is, in fact, common to all the world’s wh-fronting languages.

4.2.3 The Ill-Formedness of Adposition Stranding and Split DPs We have just seen that the ‘Q-based’ analysis in (1) can be motivated for Edo on the grounds that it exhibits grammatical patterns which, in Tlingit, are explained only through that analysis. Beginning in this section, we will see that the same may be true, at a more abstract level, for even the most well-studied wh-fronting languages of the world. To begin, we saw in chapter 2 that Tlingit wh-questions do not permit stranding of postpositions (8), extraction of wh-possessors (9), or extraction of wh-determiners (10). (8)

(9)

No P-Stranding in a Tlingit Wh-Question a. [QP [PP Aadóo teen ] sá ]1 t1 yeegoot? who with Q you.went Who did you go with? b. * [QP Aadóo sá ] [PP t1 teen ] yeegoot? who Q with you.went No Possessor Extraction in a Tlingit Wh-Question a. [QP [DP Aadóo yaagú ] sá ] t1 ysiteen? who boat Q you.saw.it Whose boat did you see? b. * [QP Aadóo sá ] [DP t1 yaagú ] ysiteen? who Q boat you.saw.it

106 (10)

THE GRAMMAR OF Q

No D-Extraction in a Tlingit Wh-Question a. [QP [DP Daakw keitl ] sá ] t1 asháa? which dog Q it.barks Which dog is barking? b. * [QP Daakw sá ] [DP t1 keitl ] asháa? which Q dog it.barks

As we immediately noted, these constraints on Tlingit wh-questions are quite similar to facts seen in many other wh-fronting languages. First, we observe that, aside from certain Germanic and African languages, the great majority of wh-fronting languages in the world do not permit P-stranding. (11)

No P-Stranding in Wh-Fronting Languages (e.g., Russian) a. [PP Ot čego ] sleduet otkazat’sja t1 ? of what follows give.up-self What should one give up. b. * [DP Čego ] sleduet otkazat’sja [PP ot t1 ]? what follows give.up-self of (Abels 2003: 160)

Similarly, a wide variety of wh-fronting languages do not allow wh-possessors to be extracted. (12) No Possessor Extraction in Wh-Fronting Languages (e.g., English) a. [DP Whose book ] did you read t1? b. * [ Whose ] did you read [DP t1 book ]?

Third, it is rather common for wh-fronting languages not to allow wh-determiners to be extracted. (13) No D-Extraction in Wh-Fronting Languages (e.g., English) a. [DP Which book ] did you read t1? b. * [ Which ] did you read [DP t1 book ]?

Given the obvious similarity between the facts in (8)–(10) and those in (11)–(13), we noted in chapter 2 that a unified account for these facts would be optimal. Under such an account, the ill-formedness of the Tlingit sentences in (8)–(10) would follow as one instance of the more general cross-linguistic pattern seen in (11)–(13). Unfortunately we found that certain additional facts from Tlingit seem to prevent such an account. To recall, the predominant view regarding the cross-linguistic pattern in (11)–(13) is that it reflects properties of movement. We saw, however, that such a ‘movement-based analysis’ cannot be extended to the Tlingit data in (8)–(10). Rather, a broader examination of Tlingit syntax reveals that the ill-formedness of (8)–(10) follows from constraints governing the position of Q, namely the QP-Intervention Condition, and not from any constraints on movement itself. It follows that, under the standard view that (11)–(13) reflect constraints on movement, a unified account for all the data in (8)–(13) is not possible. This, of


107

course, is an unacceptable result, and should be remedied by giving up certain of the assumptions that lead to it. Given the strength of the Tlingit-internal evidence for the QP-Intervention Condition, I conclude that the most likely source of error is the assumptions regarding (11)–(13). I conclude, then, a unified account for the data in (8)–(13) requires us to extend our Q-based analysis of (8)–(10) to all other wh-fronting languages. Under such an analysis, the ill-formedness of (11)–(13) is due to the same factor at play in the parallel Tlingit structures (8)–(10). In all these sentences, the ill-formed wh-extraction could only take place from a base structure where a QP intervenes between a functional head F and a phrase that F selects for, in violation of the QP-Intervention Condition. (14)

The Ill-Formedness of P-Stranding in Wh-Fronting Languages

(15)

The Ill-Formedness of Possessor Extraction in Wh-Fronting Languages

(16)

The Ill-Formedness of D-Extraction in Wh-Fronting Languages

We therefore find that the parallelism between the Tlingit data in (8)–(10) and the cross-linguistic patterns in (11)–(13) motivate the view that all wh-fronting structures receive the Q-based analysis in (1). Under this analysis, the ill-formedness of (11)– (13) is not explained in terms of the islandhood of the base position of the wh-word. Rather, the apparent islandhood of those positions is explained in terms of independently visible constraints on the placement of Q. Consequently, the empirical motivation for classifying those positions as syntactic islands is weakened.7 Indeed, within the Q-based account, it seems best not to view those positions as syntactic islands at all.8 Furthermore, in rejecting the status of these positions as islands, we must thereby

108

THE GRAMMAR OF Q

also reject any syntactic principles whose goal is to predict the islandhood of these positions. Therefore I conclude that our Q-based theory of wh-fronting challenges such principles as the ‘Left Branch Condition (LBC)’ (Ross 1967; Corver 1990, 2007). Although our Q-based analysis of (11)–(13) is motivated on conceptual grounds, it might at this stage be criticized for a degree of empirical incompleteness. After all, our discussion has so far ignored a rather well-known fact about the extractions in (11)–(13), a fact that is moreover one of the key data that theories of (11)–(13) seek to capture: in some languages, these extractions are not ill-formed. For example, it is well known that English wh-fronting may freely strand prepositions. (17)

Wh-Fronting in English Can Strand Prepositions Who1 should I give this [PP to t1 ] ?

Furthermore, in the Slavic languages, it is generally possible for wh-possessors to be extracted from DPs. (18)

Wh-Fronting in Russian Can Extract Possessors Ja sprosil [ čju1 ty cital [DP t1 knigu ] ] I asked whose you read book I asked whose book you read. (Heck 2004)

Relatedly, the Slavic languages are also well known to permit extraction of wh-determiners. (19) Wh-Fronting in Russian Can Extract Determiners ty cital [DP t1 knigu ] ] Ja sprosil [ kakuju1 I asked what.kind.of you read book I asked what kind of book you read. (Heck 2004)

The permissibility in certain languages of the extractions in (11)–(13) challenges any theory that seeks to derive their general ill-formedness. We must naturally ask, then, how our own Q-based account fares in light of these facts. How can our account capture the apparent cross-linguistic variation in whether (11)–(13) are ill-formed? In this context, it is important to note that most movement-based accounts of (11)–(13) assume that the syntactic constraints violated by those extractions hold across all human languages. Consequently, the well-formedness of those extractions in some languages is not due to variation in the constraints on movement, but rather to some structural difference in the base position of the wh-word. That is, under most accounts, the well-formedness of, for example, possessor extraction in some languages is due to possessive nominals in those languages having a special structure, one that renders the relevant syntactic principles insufficient to rule out extraction of the possessor.


109

This general approach to the data in (17)–(19) can also be adopted within a Q-based account. In fact, our Q-based account can adopt, without much modification, certain already established analyses of (17)–(19). To see this, let us first consider possessor and determiner extraction in Slavic languages ((18) and (19)). It has been widely noted that languages permitting possessor and determiner extraction generally permit NPs to appear without overt determiners (Uriagereka 1988; Corver 1992; Bošković 2005a). Many scholars have therefore sought to connect these two properties, to derive the permissibility of those extractions from the ability for NPs not to be dominated by DP projections. For example, Bošković (2005a) proposes that extraction of possessors and determiners in these languages is due to their being pure NP adjuncts. That is, instead of the familiar DP structure, the structure of possessives and determiner-marked nominals in these languages is as follows. (20)

Structure of Possessive Nominals in Slavic Languages NP Possessor

NP

N Possessum

(21)

Structure of Determiner Modification in Slavic Languages

NP Determiner

NP

N

Bošković (2005a) shows that a movement-based account of (12) and (13) predicts that languages whose nominals have the structure in (20) and (21) should permit extraction of possessors and determiners. Interestingly, this same result also holds for a Q-based account of (12) and (13). First, note that our Q-based account predicts that languages with the structure in (20) should generally permit Q-particles to intervene between possessors and possessed NPs. After all, within (20), possessors are adjuncts to NP—a lexical category—and not specifiers of a functional DP projection. Thus the QP-Intervention Condition would not be violated by structures like that in (22).

110 (22)

THE GRAMMAR OF Q

An Available Base Structure for Russian Possessor Extraction

NP QP XP

NP Q N Possessum

Wh-Possessor

Moreover, if the structure in (22) were to feed QP-movement, we would thereby derive possessor extraction. Therefore the structure of possessor-extraction sentences like (18) may be as in (23). (23)

The Structure of Possessor-Extraction in Slavic Languages Ja sprosil [ [QP čju ]1 ty cital [NP t1 knigu ] ] I asked whose Q you read book I asked whose book you read.

Similarly, note that languages possessing the structure in (21) should generally allow Q to intervene between a determiner and an NP. Since determiners are simply adjuncts to NP in these languages, the QP-Intervention Condition would allow for structures like that in (24). (24)

An Available Base-Structure for Russian Determiner-Extraction NP QP

Wh-Determiner

NP

Q

N

Of course, if the structure in (24) were to feed QP-movement, we would thereby derive determiner extraction. Thus the structure of determiner-extraction sentences like (19) may be the following. (25)

The Structure of Determiner-Extraction in Russian Ja sprosil [ [QP kakuju ]1 ty cital [NP I asked what.kind.of Q you read I asked whose book you read.

t1 knigu ] ] book

We find, then, that the well-formedness of possessor and determiner extraction in certain languages need not be inconsistent with our overall Q-based approach to the facts in (8)–(13). Indeed, certain well-established analyses of such languages apply equally well within a Q-based account. But what of the possibility of P-stranding


111

in languages like English (see (17))? Are existing perspectives on the nature of P-stranding consistent with the Q-based approach? In a sense, they are. Within the literature on P-stranding, it is commonly held that the possibility of P-stranding in English reflects some special, sui generis property of English Ps. For example, Abels (2003) proposes that the general impossibility of P-stranding follows from the status of P as a ‘phase head’ (Chomsky 2000). Consequently Abels claims that the possibility of P-stranding in English indicates that English P is, somehow, exceptionally not a phase head (Abels 2003: 233). Similarly, van Riemsdijk (1978) claims that the general impossibility of P-stranding follows from the unavailability of SpecPP as a ‘landing site’ for movement. Consequently the possibility of P-stranding in English must follow from an exceptional ability for SpecPP in English to be an available landing site. Throughout our discussion, I will take a similar view regarding the nature of English P-stranding. As in prior accounts, I will assume that English P-stranding follows from an exceptional property of English prepositions, one that (unfortunately) plays little role in the broader grammar of English. I will claim, however, that this special property is that of being a lexical category. First, note that if P in English is a lexical category, then English should generally allow Q to intervene between P and its DP complement. Due to the status of P as a lexical head, such structures will no longer violate the QP-Intervention Condition. Of course, if such structures were to feed QP-movement, we would thereby derive P-stranding in English. The following structure illustrates. (26)

P-Stranding in English Licensed by the Lexical Status of English P CP C’

QP1 DP

Q

IP

CQ

…wh-word… PP QP t1

P

Possible PP, Permitted by QP-Intervention Condition Because ‘P’ in English Is a Lexical Category

We therefore find that the exceptional possibility of P-stranding in languages like English poses no greater challenge to a Q-based theory of P-stranding than it does to a movement-based theory. That is, there is currently no real explanation for why Ps may be stranded in languages like English. The best that current accounts can offer is that P in these languages simply lacks the property that in most other languages prevents P-stranding. Adopting this same logic, given that a Q-based theory of P-stranding derives its impossibility from P being a functional head, such an

112

THE GRAMMAR OF Q

account derives the possibility of English P-stranding from English P being a lexical head. While the stipulative nature of this account could be justly criticized, I submit that it is no more stipulative than any other currently available approach to the distribution of P-stranding across languages. Finally, it is perhaps worth noting, if only in passing, that Huddleston and Pullum (2002: 597–661) argue that the category P in English is much larger than traditionally conceived, and contains many items that have previously been miscategorized as adverbs, adjectives, or verbal participles. Consequently, Huddleston and Pullum (2002: 603) distinguish P in English as a “relatively closed class” (emphasis mine), since it differs markedly from other functional categories (like determiners) in both the sheer number of its members, and the relative rapidity with which it gains members from the lexical classes N, V, and A. Thus, there are some independent grounds for distinguishing P in English from clear-cut cases of functional heads, though it remains to be seen whether these properties also distinguish P in any languages that disallow P-stranding. 4.2.4 Comparison With Earlier Accounts: Watanabe (1992), Tanaka (1998, 1999), and Sternefeld (2001) In previous chapters, explicit comparison has been made between our Q-based analysis of wh-fronting in (1) and the analyses of wh-in situ languages put forth by Hagstrom (1998) and Kishimoto (2005). However, our analysis in (1) also bears a strong similarity to three earlier proposals concerning wh-fronting and pied-piping: those of Sternefeld (2001), Tanaka (1998, 1999), and Watanabe (1992). Of the three, the work of Sternefeld (2001) comes closest to the Q-based account developed here. Indeed, this brief 14-page paper (which I was not aware of at the time Cable (2007) was written) independently develops the Q-based syntax and semantics presented in chapter 2. Although details of implementation differ, the two accounts align remarkably well. Sternefeld (2001) likewise notes many of the syntactic and semantic advantages that are detailed in section 4.3. Given that Sternefeld’s (2001) proposals are essentially those made here, I will not review his particular formulation of them in detail. The principal difference between Sternefeld (2001) and the work on offer here is their scope. Given the brevity of the paper, Sternefeld (2001) offers his account mainly as a solution to some of the basic syntactic and semantic puzzles surrounding pied-piping, and no independent evidence is provided for his correlate of (phonologically empty) Q. The theory of wh-questions put forth by Tanaka (1998, 1999) is also strikingly similar to our Q-based account. Indeed, the two accounts are effectively the same. Tanaka (1998, 1999) proposes that wh-fronting does not directly target the features of the wh-word. Rather, wh-fronting is ultimately movement of a separate phrasal projection dominating the wh-word, which Tanaka labels ‘ωP’. As with our ‘Q’, the ‘ω-head’ of Tanaka (1999, 1998) can be separated from its associated wh-word by phrasal material. Consequently, Tanaka (1998, 1999) proposes that the ‘ωP’ is the mechanism underlying pied-piping, that pied-piping structures are simply those where the sister of the ‘ω-head’ properly contains the wh-phrase.


113

The similarities between the core proposals of Tanaka (1998, 1999) and our own Q-based theory are rather obvious. Where Tanaka’s work differs from the system offered here is chiefly in the use to which the analysis is put. Tanaka (1998, 1999) is primarily concerned with the interpretation and linear order of wh-words in Japanese. Consequently, there is comparatively little development in this work of the ‘QP/ωPanalysis’ of overt wh-fronting. Thus, although the overall proposals of our work are very close, the empirical results obtained are entirely complementary. Another proposal related to our own is that developed by Watanabe (1992). Although substantially different in its details, Watanabe’s account shares a certain logical structure with our own. Indeed, one could view the proposals put forth here as one development of these common core ideas. In brief, Watanabe (1992) proposes the following. In both English and Japanese wh-questions, there is overt fronting of an entity that Watanabe labels ‘Op’ (short for ‘operator’). In Japanese, this ‘Op’ element can be moved away from the wh-word, leaving the wh-word in situ. In English, however, ‘Op’ cannot be detached from the wh-word. Consequently, overt movement of ‘Op’ in English entails overt movement of the wh-word. Thus the wh-questions of Japanese and English both involve overt movement of the same formal element, and their surface differences result from a more basic difference in whether that moved element can detach from its associated wh-word. Our Q-based account is thus quite similar to the proposals of Watanabe (1992). There are, however, several details of implementation that render Watanabe (1992) rather different from the Q-based analysis advocated here. The chief differences lie in the theory of wh-fronting languages. Unlike our account in (1), where the Q/Opelement heads a projection containing the wh-phrase, Watanabe (1992) proposes that the Q/Op-element is a specifier of the wh-phrase. This is illustrated in (27). (27)

The Structure of an English Wh-Word, Following Watanabe (1992) DP Op/Q

D’

D who

As a result, Watanabe must treat all wh-movement as an instance of (true) piedpiping rather than as normal phrasal movement. Consequently Watanabe (1992) admits the existence of true pied-piping, and so cannot adopt the ‘eliminativist’ theory of pied-piping structures that is developed in the remainder of this book. A greater problem for Watanabe (1992), however, concerns the predictions the analysis makes regarding wh-fronting languages. Watanabe (1992) proposes that whfronting results from a need for the Op/Q-element to always remain adjacent to the wh-word. Thus the analysis predicts that no island can ever separate the Op/Q

114

THE GRAMMAR OF Q

element and the wh-word in a wh-fronting language. Finally, given the assumption that ‘feature percolation’ is constrained by syntactic islands, it follows that there can be no pied-piping of islands in any wh-fronting language. This result is summarized in the following quote: “The prediction of our system is that languages like English which must move an entire wh-phrase at S-structure will never exhibit large-scale pied-piping, while languages like Japanese which move pure wh-operators and allow wh-in-situ are potentially able to employ large-scale pied-piping.” (Watanabe 1992: 65). Although it is indeed true that English does not permit the large-scale pied-piping of islands (see chapter 5), it is not true that no wh-fronting language permits such piedpiping. As we have already seen, such pied-piping is readily available in Tlingit. (28) Pied-Piping of Relative Clause Islands in Tlingit [ [ Wáa kwligeyi CP] xáat NP] sá i tuwáa sigóo? how it.is.big.REL fish Q your spirit it.is.happy How big a fish do you want? (= A fish that is how big do you want?)

We find, then, that the assumption in (27), that Q/Op is a specifier of the whphrase, eventually leads the system of Watanabe (1992) to be quite different from that proposed here. As a result of that assumption, Watanabe (1992) is unable to adopt our approach to pied-piping structures, and wrongly predicts the nonexistence of languages like Tlingit. Consequently, the core advantage of our Q-based account over Watanabe’s is the notion (shared with Tanaka (1998, 1999)) that the wh-phrase is complement to the Q/Op/ω-head. Finally, before we leave this discussion, I would like to briefly compare our Q-based account in (1) to a more general idea that has been ‘in the air’ for some time. In the syntactic literature on wh-questions, one often encounters a view that can (vaguely) be stated as follows: the element pronounced as ka in Japanese wh-questions surfaces in English as the ‘wh-morpheme’ in the wh-word. For example, Watanabe (1992) suggests that the Op-element is in English overtly pronounced as the /wh-/ component of the wh-word, which accounts for the inability of English Op to ever move away from the wh-word. This same basic perspective also appears in work by Cole and Hermon (1998), although it receives a rather different implementation. I might also add that several individuals have suggested to me that, rather than analyze English as containing a null version of Tlingit sá, it may be that this ‘Q’ element receives overt pronunciation as ‘wh-’. Although the general idea that ‘Q is wh-’ is an interesting one, in the context of the proposals put forth here, it cannot be made to work. The basic issue concerns the role played by the Q-particle in the theory of pied-piping structures that we develop in the following sections. As alluded to in chapter 1, we will assume that the Q-particle in English plays the same role in pied-piping structures that the Q-particle sá does in Tlingit. As we will see in section 4.3, this entails that in English pied-piping structures, the Q-particle sits just above the fronted phrase, separated from the whword by phrasal material. Such a position for Q, detached from the wh-word, is clearly at odds with the idea that Q is the wh-morpheme. On the other hand, a proponent of the view that ‘Q is wh-’ might address this issue by appealing to covert movement of Q. That is, it might be possible within a


115

‘single-output architecture’ (Bobaljik 2002) to adopt the following view: (i) the Q-particle is initially merged in a low position, adjacent to the wh-word; (ii) the Q-particle then undergoes movement to a higher position, where it triggers the observed ‘pied-piping’; (iii) the pronunciation of the Q-particle adjacent to the whword (as the wh-morpheme) simply reflects pronunciation of its lowest copy.9 Such a proposal, however, immediately invites the following question: what is the mechanism underlying the hypothesized Q-movement? If the mechanism is Agreement with some higher head X, then why not simply view the higher head X as the correlate of Q in English? Indeed, chapter 5 presents an analysis rather similar to this, within which the wh-word of English undergoes syntactic Agreement with the (higher) Q-particle. Under our approach, then, there could be a sense in which the wh-morpheme of English is an instance of Q. Since Q undergoes Agreement with the wh-word in English, it might be possible to view the ‘wh-’ submorpheme as some reflex of this Agreement, possibly an uninterpretable instance of the ‘Q-feature’ on the interrogative word. Indeed, Cable (2007) develops this view in greater detail. While we will not pursue it in this work, it is nevertheless apparent that, in a certain sense, the Q-based theory proposed here could adopt one implementation of the oft-encountered notion that ‘English /wh-/ is Japanese ka.’

4.3 Some Initial Applications to the Theory of Pied-Piping Structures So far I have presented two general arguments for abandoning the ‘classic’ account in (2) and adopting the Q-based account in (1) for all wh-fronting languages: (i) doing so would greatly simplify our typological theory, and thus the hypothesis space learners must navigate, (ii) doing so is the only way to provide a unified account of the impossibility of P-stranding and left-branch extractions across languages. The remainder of this book presents two further, extended arguments in support of our Q-based account. The overall thrust of these arguments is that applying (1) to some of the best-studied wh-fronting languages advances understanding of various phenomena in those languages. The first of these two arguments, which concerns pied-piping, will begin in this section and will continue into chapter 5. I will first argue that our Q-based account can advance understanding of the syntax of pied-piping structures, and even allows us to eliminate the concept of ‘piedpiping’ from the theory of grammar. Following this, I will briefly note that our Q-based account might also advance our theory of the semantics of pied-piping structures, as it provides an especially simple system for their interpretation. 4.3.1 The Elimination of ‘Pied-Piping’ From the Theory of Grammar We observed in chapter 1 that sentences like the following directly challenge the classic theory of wh-fronting in (2).

116 (29)

THE GRAMMAR OF Q

Some Pied-Piping Structures of English a. [ Whose book ] did you read? b. [ To whom ] did you speak? c. [ How long a book ] did he write?

To recall, the problematic property exhibited by these sentences is that the fronted phrase is not a projection of the wh-word, but rather properly contains the projections of the wh-word. Consequently, if we assume, as in (2), that wh-fronting targets features of the wh-word itself, then structures such as those in (29) should not exist. It was as a solution to this basic, empirical problem that the concept of ‘piedpiping’ was introduced into linguistic theory. Recall that throughout our discussion we understand the term ‘pied-piping’ to refer to a particular kind of analysis, defined as follows. (30) Definition of ‘Pied-Piping’ Pied-piping occurs when an operation that targets the features of a lexical item L applies to a phrase properly containing LMAX.

Thus the term ‘pied-piping’ contrasts in meaning with the purely descriptive label ‘pied-piping structure’, which we use to refer to a particular class of surface structures. (31) Definition of ‘Pied-Piping Structure’ A pied-piping structure is one where a phrase properly containing the maximal projection of a wh-word (or related operator) has undergone fronting ‘typically associated’ with that operator.

Following these definitions, there is a clear sense in which one can deny or assert the existence of pied-piping: to assert/deny the existence of pied-piping is simply to assert/deny that there exist cases of the kind in (30). Furthermore, we find that the classic analysis in (2) must, in light of the facts in (29), accept that pied-piping exists. In this context, let us observe that our theory of Tlingit wh-questions has no need of the concept of ‘pied-piping’ in (30). Moreover, and most importantly, this is despite the fact that Tlingit clearly possesses pied-piping structures, as defined in (31). That is, as we saw in chapter 2, and as illustrated below, it is possible in Tlingit for the fronted phrase of a wh-question to properly contain the maximal projection of the wh-word. (32)

Pied-Piping Structures of Tlingit a. [PP Aadóo teen ] sá yeegoot? who with Q you.went Who did you go with? b. [DP Aadóo yaagú ] sá ysiteen? who boat Q you.saw.it Whose boat did you see?


117

c. [DP X’oon keitl ] sá ysiteen? how.many dog Q you.saw.it How many dogs did you see? kwligeyi ] xáat ] sá i tuwáa sigóo? d. [DP[CP Wáa how it.is.big.rel fish Q you.want.it How big a fish do you want? (= A fish that is how big do you want?)

The structures in (32) clearly satisfy the definition in (31) of a ‘pied-piping structure’, and their similarity to English pied-piping structures like (29) is rather obvious. But, in what sense does our theory of Tlingit avoid appeal to (30) in the analysis of (32)? First, recall that in a Tlingit wh-question, the particle sá must always occur directly to the right of the fronted phrase, as shown in (33). (33)

Tlingit QP Cannot Be Inside the Fronted Phrase yeegoot? a. * [PP Aadóo sá teen ] who Q with you.went b. * [DP Aadóo sá yaagú ] ysiteen? who Q boat you.saw.it sá keitl ] ysiteen? c. * [DP X’oon how.many Q dog you.saw.it d. * [DP[CP Wáa sá kwligeyi ] xáat ] i tuwáa sigóo? how Q it.is.big.rel fish you.want.it

Thus the fronted phrase of a Tlingit wh-question never properly contains the QP. Now, according to our analysis in (1), it is the features of the Q—and not the wh-word—that trigger movement in Tlingit wh-questions. Under our analysis, then, the pied-piping structures in (32) are all straightforward cases of simple phrasal movement. As (34) illustrates for (32b), in each structure, movement simply targets the projection of the head whose features trigger the movement (namely, Q). (34)

Derivation of the Tlingit Pied-Piping Structure in (32b) CP CP

QP1 DP

Q sá

CQ

Aadóo yaagú

IP ysiteen QP1

Probe/Agree for Q

Attract / Move

Thus none of the structures in (32) are cases where the moved phrase properly contains the projections of the head triggering the movement. Consequently none of the Tlingit pied-piping structures in (32) are instances of true ‘pied-piping’, as defined in (30).

118

THE GRAMMAR OF Q

Furthermore, given the generalization illustrated in (33), we find that there are not any true cases of pied-piping in Tlingit. Following our definition in (30), true pied-piping in Tlingit wh-questions would have the appearance of the structures in (33), where the projections of the particle sá (the target of movement) are properly contained within the fronted phrase. To clarify the point, such hypothetical structures are illustrated in (35). (35)

A Hypothetical ‘True Pied-Piping’ Structure in Tlingit CP DP1 QP Aadóo

CP DP

sá

D POSS

CQ NP yaagú

IP ysiteen DP1

Probe/Agree for Q

Attract / Move (of DP containing the QP)

Thus, under our Q-based account in (1), the pied-piping structures of Tlingit are not actually cases of pied-piping, and what would be actual cases of pied-piping in the language are ill-formed. For this reason, the concept of “pied-piping” as defined in (30) can be eliminated from our theory of Tlingit grammar. By adopting the analysis in (1), we need not deviate from the null hypothesis that if an operation (in Tlingit) targets the features of a lexical item, then it applies only to the projections of that item. Similarly, if we extend the analysis in (1) to all wh-fronting languages, we need not ever deviate from that null hypothesis. The pied-piping structures of all the beststudied wh-fronting languages could receive an analysis akin to that shown for English in (36). (36)

The Pied-Piping Structures of English, Under the Q-Based Theory a. Whose father’s cousin’s uncle did you meet at the party? b. [QP [ [ [ [ whose ] father’s ] cousin’s ] uncle ] Q ] did you meet at the party?

Under this analysis, a pied-piping structure in English is derived exactly like the pied-piping structures of Tlingit. In such sentences, the (null) Q-particle takes as sister a phrase properly containing the wh-word, which entails that the fronted phrase of the wh-question properly contains the wh-word. Thus one need not view sentences like (29) as cases where movement targets ‘more than’ the phrase whose features trigger it. Under such a Q-based theory, then, one need not ever accept that pied-piping (as defined in (30)) truly exists in English.10 If this same basic result can be maintained for all wh-fronting languages, then one need not ever admit the existence of pied-piping in human language.


119

The ability to eliminate pied-piping from our theory of grammar is advantageous in several respects. First, there is simply the fact that it will reduce the number of phenomena that our theory grammar must explain. More importantly, however, it allows us to eliminate from our theory all those mechanisms whose purpose is to account for pied-piping. Given the common acceptance of pied-piping, the question of what mechanisms underlie it has received much focused attention (Ross 1967; Sells 1985; Webelhuth 1992; Kayne 1994; Grimshaw 2000; Horvath 2007a; Heck 2008). Within this literature, there are two main proposals. The first is that piedpiping structures are derived via special mechanisms of ‘feature percolation’, which transfer the features of a head onto nodes outside the projection of the head (Chomsky 1973; Kayne 1983; Gazdar et al. 1985; Sells 1985; Cowper 1987; Webelhuth 1992; Grimshaw 2000). The second proposal is that the theory of movement be weakened so that it allows movement to merely target phrases that contain somewhere within them the features triggering the movement (Ross 1967; Heck 2008). Both these approaches, however, encounter certain conceptual problems. First, several arguments are put forth by Heck (2008: 65–73) against the concept of ‘feature percolation’. The overarching problem noted by Heck is that this putative operation cannot be reduced to any other, more widely encountered syntactic operations. For example, feature percolation cannot (under current assumptions) be an instance of Agreement, since Agreement cannot insert features into projections where they did not previously exist.11 Suppose, for example, that we were to adopt the following analysis of possessor pied-piping (29a): in such structures, there is whAgreement between the wh-word in SpecDP and the possessive D-head (Cowper 1987; Grimshaw 2000). Any such account would, under current models of Agree, still have to explain the existence of the wh-feature on the possessive D, the very fact that a theory of feature percolation is intended to explain. Another possibility worth considering here is that ‘feature percolation’ is simply feature movement. Under this view, pied-piping occurs when the wh-feature of the wh-word itself undergoes movement that places it outside the projection of the wh-word. As Heck (2008) argues in detail, however, such an account faces certain empirical problems.12 The general problem is that feature percolation, construed as feature movement, seems to violate well-known constraints on movement. Consider again the case of possessor pied-piping in (29a). Let us also assume that such structures are generated by feature movement of [wh] from the wh-possessor to the higher DP. As illustrated in (37), it follows that the [wh]-feature must be extracted from the specifier of DP. (37)

Feature Percolation as Feature Movement, With Wh-Possessors

[wh]DP

Movement / Percolation of Wh-feature From SpecDP

[wh]DP

wh-possessor

D’ POSS possessum

120

THE GRAMMAR OF Q

However, as we saw from data like (38), English does not generally permit movement from SpecDP.13 (38)

No Possessor Extraction in English a. [DP Whose book ] did you read t1? b. * [ Whose ] did you read [DP t1 book ] ?

Furthermore, it was noted earlier that many other wh-fronting languages share with English the pattern in (38): they permit pied-piping by possessors, but do not generally permit extraction of possessors. Consequently the notion that pied-piping structures are generated by feature-movement, as in (37), is in conflict with the tendency against movement from SpecDP. It therefore appears that the putative operation of feature percolation cannot be reduced either to Agreement or to movement. In the absence of other proposals, we must conclude that any theory appealing to ‘feature percolation’ necessarily invokes an additional, primitive syntactic operation. Most worrying of all, however, is that such a primitive feature percolation mechanism appears to have little purpose outside of deriving pied-piping structures. Given these considerations, theories of piedpiping that appeal to ‘feature percolation’ do seem rather suspicious. What, then, of those theories that adopt (2) but do not appeal to feature percolation? Under these approaches, the theory of movement is weakened so as to permit structures where the moved phrase does not itself bear the feature triggering the movement (Ross 1967; Heck 2008).14 Thus, according to such analyses, sentence (29a) is permissible because it satisfies the (weak) condition requiring that the moved phrase contain the wh-word somewhere inside it. Unfortunately, a pervasive problem for this form of analysis is the potential for overgeneration. That is, it is not generally the case that any phrase containing a wh-word may be fronted in an English wh-question, as the ill-formedness of (39b) illustrates.15 (39) a. Who1 did Dave buy [ copies of [ paintings of t1 ] ] ? b. * [ Copies of [ paintings of who ] ] did Dave buy?

Of course, if appeal is made to feature percolation, then the observed limits on piedpiping may be encoded into the percolation mechanism itself, by placing limits on how ‘far’ feature percolation may carry a feature from its lexically associated head. However, without this sort of mechanism, it is difficult to identify the source of anomaly in sentences like (39b), especially since sentences like (39a) show that the embedded wh-words are, in principle, accessible to the matrix interrogative C. In summary, then, we find that the classic analysis in (2) is not only saddled with the additional concept of ‘pied-piping’, but also requires special mechanisms to generate pied-piping structures. To put the matter more acutely, under the classic picture in (2), the pied-piping structures in (29) represent a deviation from the expected grammatical pattern. That is, on its own, (2) predicts that structures like (29) should not exist, as it can only account for such structures via the addition of special mechanisms. This is in stark contrast to the Q-based analysis in (1). Under that analysis, pied-piping structures merely represent cases where the sister of Q is not the


121

wh-phrase itself, but rather a phrase containing the wh-phrase. Importantly, the possibility of such configurations requires no special assumptions or additions to our basic theory. Indeed, our theory would require special assumptions to rule out pied-piping structures. Consequently our Q-based analysis permits us to dispense with the special concept of ‘pied-piping’ in (30), as pied-piping structures can be generated via normal phrasal movement of the QP. Indeed, within our Q-based account, the only distinctive property of pied-piping structures is that the Q is not directly adjacent to the wh-phrase. Such structures are, again, immediately predicted by our account; they have no interesting syntactic or semantic properties, and do not stand out as a special taxonomic class. Thus, in a very real sense, our Q-based account entails that there is no such thing as pied-piping. In addition, we will see that our Q-based theory makes a range of further predictions regarding pied-piping structures. In this chapter, we will see that our analysis predicts that ‘pied-pipers’ are universally subject to Intervention Effects. Moreover, chapter 5 argues that our account may capture several other properties of pied-piping structures, including the rather severe limits placed on pied-piping by languages like English. Thus we find that extending the Q-based analysis to all wh-fronting languages— and thereby doing away with the concept of ‘pied-piping’ in (30)—would seem to yield a simpler overall theory of phrasal movement. 4.3.2 A Note on the Semantics of Pied-Piping Structures In this context I would like to briefly observe a semantic property of our Q-based theory, one that we noted earlier in chapter 2.16 We saw in the final section of chapter 2 that our semantics correctly interprets structures where the sister of Q is a phrase properly containing the wh-word. Moreover, we saw that the interpretation of such structures requires no mechanisms beyond those required for structures where Q and the wh-word are directly adjacent. It is apparent, then, that if we were to extend our Q-based account to all whfronting languages, we would thereby obtain a comparatively simple semantics for pied-piping structures. Under our Q-based syntax/semantics, the interpretation of pied-piping structures would require no mechanisms beyond those required for simple wh-questions without pied-piping. As noted earlier, this result is of potential significance. The interpretation of pied-piping structures poses a difficult challenge to certain, classic approaches to the semantics of wh-questions. In response to this challenge, some have proposed that pied-piping structures undergo syntactic alterations before interpretation (von Stechow 1996), while others have proposed that the interpretation of pied-piping structures requires additional semantic mechanisms (Sharvit 1998). Under such approaches, the mechanisms required to interpret simple wh-questions are not sufficient for the interpretation of wh-questions with pied-piping. However, given the sheer ubiquity of pied-piping structures, it would clearly be preferable for their meaning to immediately follow from the rules set up for simple wh-questions. Our Q-based theory of pied-piping structures clearly meets this natural desideratum. Of course, it shares this property with a number of other, widely known semantic theories

122

THE GRAMMAR OF Q

(Hamblin 1958; Groenendijk & Stokhoff 1982, 1984). What unites all these accounts is the notion that the wh-word is of a simple nonquantificational semantic type. That is, as noted by Cable (2007), the special mechanisms developed by von Stechow (1996) and Sharvit (1998) are required because they assume (following much earlier work) that whwords are of a higher quantificational type. Under such an assumption, a wh-question can only be interpreted if the wh-word has scope over the entire IP of the clause. In a piedpiping structure, however, the wh-word is buried within the fronted phrase, and thus it does not at its surface position have the proper scope. Consequently, special mechanisms are required, either to alter the position of the wh-word prior to interpretation (von Stechow 1996), or to permit it to take exceptional scope from its surface position (Sharvit 1998). Of course, if we assume that wh-words are not of a quantificational type, then they are free from the need to take scope over the IP. Consequently, they can be interpreted in their surface position, deep down within the fronted phrase. Thus a semantics that assumes wh-phrases to be of type e, such as our Q-based account, can provide a rather straightforward semantics for pied-piping structures, one in which their meaning directly follows from the rules required for simple wh-questions without pied-piping.

4.4 Intervention Effects and Superiority Effects in Wh-Fronting Languages In this section we will see how our Q-based account in (1) can capture the crosslinguistic relationship between Intervention Effects and Superiority Effects in the multiple wh-questions of wh-fronting languages. Inasmuch as the account offered here seems like a promising direction, we find further support for the notion that the structure in (1) is common to all wh-fronting languages. We will begin by briefly previewing the phenomena of interest. We saw in chapter 3 that in many languages, an in situ wh-word cannot appear in the scope of certain ‘offending operators’. Although we originally discussed such ‘(LF/Focus) Intervention Effects’ in the context of wh-in situ languages, they can also be observed in some wh-fronting languages. For example, the well-formed German multiple whquestion in (40a) becomes ill-formed if the in situ wh-word is placed in the scope of the offending operator niemanden ‘nobody’, as in (40b). (40)

Intervention Effects With In Situ Wh-Words in German angetroffen? a. Wer hat Hans wo who has Hans where met Who met Hans where? b. ?? Wer hat niemanden wo angetroffen? who has nobody where met Who met nobody where?

Interestingly, although the in situ wh-words of German are subject to Intervention Effects, this is not so for other wh-fronting languages. Famously, English wh-words


123

seem generally impervious to Intervention Effects.17 Unlike in German, wh-words in English can appear within the scope of such typical ‘offending operators’ as nobody and not. (41)

No Intervention Effects With In Situ Wh-Words in English a. Who didn’t read what? b. Which children wanted to show nobody which pictures?

What, then, accounts for this interesting contrast between English and German? As noted by prior scholars (Beck 1996; Pesetsky 2000), our theory of Intervention Effects should be able to derive it from some broader, more fundamental difference between the two languages. In the following discussion we will see how our Q-based account might accomplish this. Moreover, we will see that our account ties this contrast in Intervention Effects to another well-known contrast between English and German: Superiority Effects. As is well known, multiple wh-questions in English typically require that only the structurally highest wh-word undergo overt wh-fronting; lower wh-words must stay in situ.18 (42)

Superiority Effects in English Multiple Wh-Questions a. Who1 bought what2? b. * What2 did who1 buy?

Adapting the terminology of Pesetsky (2000), we may say that English wh-questions exhibit ‘Superiority Effects’. This is in contrast to the multiple wh-questions of German, as seen in (43). (43)

No Superiority Effects in German Multiple Wh-Questions a. Wer1 hat was2 gekauft? who has what bought? Who bought what? b. Was2 hat wer1 gekauft? what has who bought Who bought what?

Unlike the English pattern in (42), German multiple wh-questions permit overt fronting of the structurally lower wh-word. Thus it may be said that German does not exhibit Superiority Effects.19 Taken together, then, we find that English multiple wh-questions exhibit Superiority Effects but fail to exhibit Intervention Effects, while those of German exhibit Intervention Effects but fail to exhibit Superiority Effects. Following earlier work (Beck 1996; Pesetsky 2000), we will take up the view that this is no accident.20 We will see that our Q-based account is able to tie the presence of Superiority Effects in multiple wh-questions with the absence of Intervention Effects, and vice versa. In addition, our Q-based account of (40)–(43) will be found to make the following prediction regarding pied-piping structures.

124

THE GRAMMAR OF Q

(44) Intervention Effects in Pied-Piping Structures In all languages, an Intervention Effect will arise if a ‘pied-piping’ wh-word is c-commanded by an ‘offending operator’ inside the pied-piped constituent.

That is, our account predicts that internal to the fronted phrase of a wh-question, a wh-word is always subject to Intervention Effects. This generalization has been independently shown by Sauerland and Heck (2003) to hold for German wh-questions. The data in (45) illustrate. (45) Intervention Effects in German Pied-Piping (Sauerland & Heck 2003) a. Fritz möchte wissen [ ein wie schnelles Motorrad ] du fahren darfst. Fritz wants to.know a how fast motorbike you drive may Fritz would like to know how fast a motorbike you are allowed to drive. b. *Fritz möchte wissen [kein wie schnelles Motorrad ] du fahren darfst. Fritz wants to.know no how fast motorbike you drive may

We will also see that this generalization holds for pied-piping in English, as shown by data like (46). (46) Intervention Effects in English Pied-Piping a. (?) [ A picture of which president ] does Jim own? b. * [ No picture of which president ] does Jim own?

It is important to observe here that the Intervention Effects associated with ‘piedpiping’ wh-words are found in both German and English. This is, of course, quite unlike the Intervention Effects associated with in situ wh-words. We will see that our Q-based account of (45) and (46) indeed predicts this contrast between these two types of ‘intervention environments’. In the following subsection, we begin to develop our account of these facts. We start with the multiple wh-questions of English. 4.4.1 Intervention Effects and Superiority in English To begin, let us observe a failing of our current semantics for wh-questions. Recall that a crucial component of our system is the interrogative ForceQ head in (47). (47) Special Composition Rule for ForceQ [[ ForceQi XP ]]g = λp [ ∃f . p = [[XP]]g(i / f) ]

According to the semantics in (47), this ForceQ head contributes exactly one existential quantifier to the meaning of the wh-question. As we have seen, this is indeed correct for simple, single wh-questions. For multiple wh-questions, however, this will be insufficient. To see this, let us consider the meaning of a multiple wh-question like ‘Who saw what?’ Intuitively, this question is answered by propositions like Dave saw


125

the table, Frank saw the picture, etc. That is, an answer must name a particular person x and a particular thing y such that x saw y. Consequently, the meaning of this wh-question is the set of propositions p such that there is some human x and some nonhuman y such that p is the proposition ‘x saw y’. These facts are summarized in (48). (48)

The Semantics of Multiple Wh-Questions Who saw what? a. Question: b. Meaning (Set Notation) { [λw. Dave saw the table in w], [λw. Frank saw the picture in w], . . . } c. Meaning (Lambda Notation): λp [ ∃x ∈ human. ∃y ∉ human. p = [λw. x saw y in w]]

Therefore we find that the meaning of a multiple wh-question contains multiple existential quantifiers, one for each wh-word in the question. Consequently the ForceQ head in (47), which introduces only one existential quantifier, will not be sufficient for multiple wh-questions. How are we to remedy this? Here I will pursue an approach similar to that of Dayal (1996) and Pesetsky (2000). I assume that multiple wh-questions require for their interpretation a special interrogative Force head, which I label ‘ForceQ2’. That is, I assume that in addition to the ‘ForceQ’ of (47), the lexicon on English contains a distinct head ‘ForceQ2’, which has the following semantics. (49)

The Head ForceQ2 in the Semantics of English Multiple Wh-Questions [[ ForceQ2 i,j XP ]]g = λp [ ∃f . ∃h . p = [[XP]]g(i / f) (j / h) ]

Thus ForceQ2 differs from ForceQ in that it contributes two existential quantifiers to the meaning of the question. We will see that such a semantics allows our system to correctly interpret sentences like (48a). However, before we extol the virtues of this ‘ForceQ2’, let us first acknowledge one serious criticism against it. Given the semantics in (49), this head contributes only two existential quantifiers. Consequently it will not be enough to interpret, for example, ternary multiple wh-questions, which contain three wh-words. As illustrated in (50), the meanings of such questions contain three existential quantifiers. (50)

The Semantics of Ternary Multiple Wh-Questions a. Question: Who gave what to whom? b. Meaning (Set Notation) { [λw. Dave gave the cat to Sue in w], [λw. Frank gave the tie to Joe in w], . . . } c. Meaning (Lambda Notation): λp [ ∃x ∈ human. ∃y ∉ human. ∃z ∈ human. p = [λw. x gave y to z in w]]

Although this is a serious issue, I will put off discussion of ternary wh-questions until the next subsection. We will see then that a minor change to our account will not only allow it to interpret such questions, but will also predict a certain curious property they exhibit.

126

THE GRAMMAR OF Q

Let us, then, examine how (49) permits our system to interpret multiple whquestions. We assume that (binary) multiple wh-questions are headed by ForceQ2, as illustrated in (51).21 (51)

The Syntax and Semantics of Sentence (48a)

As the reader is invited to confirm, the lexical entry for ForceQ2 in (49) entails that our semantic system assigns as the meaning of (48a) the set of propositions indicated above. This is the set of propositions p of the form ‘x saw y’, where (i) x is the value of some choice function f over the set of humans, and (ii) y is the value of some choice function h over the set of nonhumans. Given the nature of choice functions, this is simply the set of propositions p of the form ‘x saw y’, where x is some human and y is some nonhuman. Therefore our semantics correctly assigns to (48a) the meaning in (48c).22 We see, then, that the addition of (49) allows our system to correctly interpret the binary wh-questions of English. Interestingly, we will also see that it predicts the crucial data in (41) and (42). How so? Ultimately the data in (41) and (42) follow from a more abstract prediction of our system: in an English binary whquestion, there are multiple Q-particles, one for each wh-word. This latter prediction follows straightforwardly from our semantics in (49). Since the ForceQ2 head existentially quantifies over two choice-function variables, its complement must contain two Q-particles; otherwise, the structure will violate a general ban on vacuous quantification. So how does the presence of multiple Q-particles in English binary wh-questions capture the data in (41) and (42)? Let us first consider the Superiority Effects in (42). The following diagram summarizes the core of our analysis.


(52)

127

Multiple QPs Predicts Superiority Effects Since all the wh-words are associated with QPs, the usual logic of ‘Attract Closest’ will predict the appearance of Superiority Effects in multiple wh-questions. Superiority-Satisfying Structure: [ [QP Who Q ]1 [ t1 bought [QP what Q ] ] Superiority-Violating Structure * [ [QP What Q ]1 [ did [QP who Q ] buy t1 ] Attract Closest Satisfied Attract Closest Violated!!

Under our Q-based account, the fronting of a wh-word is ultimately fronting of the QP projection dominating that wh-word. Such Q-fronting is, of course, assumed to follow general syntactic principles like ‘Attract-Closest’ (Richards 1997) or the ‘Minimal Link Condition (MLC)’ (Chomsky 1995). Consequently, if there are multiple QPs within a sentence, Attract-Closest/MLC will require that the structurally highest QP undergo Q-movement prior to any lower QPs. Thus only the structurally highest QP will undergo overt movement in English, while any lower QPs must undergo subsequent, covert movement. Given that each wh-word in the sentence is dominated by a QP, it follows that the structurally highest QP will dominate the structurally highest wh-word. Therefore only the structurally highest wh-word will appear overtly fronted in an English binary wh-question. Let us now consider the Intervention Effect data in (41). First, note that our theory assumes that all the QPs in a wh-question move into the left periphery by LF (see (51) and note 21). Consequently, by the time an English binary wh-question is interpreted, all the overtly in situ wh-words have been moved above any ‘offending operators’ that might c-command them on the surface. Thus, by the time they are interpreted, sentences like those in (41) do not contain any wh-words within the scope of any ‘offending operators’, and so they are predicted to be well-formed. Suppose, though, that we were to give up the assumption that in situ QPs move covertly. Interestingly, our account would still predict the data in (41). To see this, let us assume that QPs that are in situ at PF remain in situ at LF. The LF structure of a sentence like (41a) would then be as follows. (52)

The LF Structure of (41a), Assuming That In Situ QP Does Not Covertly Move [ [QP Who Q ]1 [ didn’t t1 read [QP what Q ] ] ]

While the wh-word in (52) is indeed within the scope of the ‘offending operator’ (didn’t), recall from chapter 3 that Intervention Effects are only triggered when sentences contain structures of the type in (53). (53)

Configuration Resulting in an Intervention Effect [ … Q [ … Offending Operator [ … [ wh-word ] … ] ] ] no Q-particle

128

THE GRAMMAR OF Q

That is, Intervention Effects result when the first focus-sensitive operator ccommanding a wh-word is not the Q-particle. Such a structure clearly does not occur in (52). Because the in situ wh-word is paired with its own Q-particle, the first focussensitive operator c-commanding it need not be the ‘offending operator’. Consequently these structures are predicted not to trigger Intervention Effects. We find, then, that our system’s prediction of the data in (41) is rather robust; it rests entirely upon the semantics assigned to ForceQ2, and persists through minor alterations in the surrounding syntactic theory. In summary, we have seen that the addition of ForceQ2 to the lexicon of English has the following consequences: (i) it provides a compositional semantics for binary whquestions; (ii) it predicts that English binary wh-questions should exhibit Superiority Effects (42); and (iii) it predicts that the in situ wh-words of binary wh-questions should be immune to Intervention Effects (41). Thus it is possible in our Q-based account to derive the English data in (41) and (42) from a single core property of the language. 4.4.2 Intervention Effects and Superiority in German We have just seen that our Q-based account can capture the English facts in (41) and (42) via the single assumption that English contains the ‘ForceQ2’ head in (49). What, though, of the complementary facts from German in (40) and (43)? To begin, let us again note a crucial aspect of our analysis of English. Ultimately our account derives the facts in (41) and (42) from the more general prediction that English binary wh-questions contain multiple Q-particles. Note that this prediction is indirectly supported by the surface form of Tlingit multiple wh-questions. As we saw earlier, multiple wh-questions in Tlingit contain multiple instances of sá, one for each wh-word. (54)

Multiple Q-Particles in Tlingit Multiple Wh-Questions sá daa sá aawaxáa? a. Aa who Q what Q they.ate.it Who ate what? b. Aa sá goodéi sá woogoot? who Q where.to Q they.went Who went where? c. Aa sá Wáa sá kuyawsikaa? who Q how Q they.said.to.someone Who said what?

Interestingly, however, not all languages allow multiple Qs in multiple wh-questions. In Navajo, for example, a multiple wh-question must contain only a single instance of the Q-particle lá. (55)

Single Q-Particles in Navajo Multiple Wh-Questions a. Háí-lá ha’át’íí nayiisnii’? who-Q what he.bought.it Who bought what? b. * Háí-lá ha’át’íí-lá nayiisnii’? who-Q what-Q he.bought.it (Barss et al. 1991)


129

It appears, then, that languages may differ in how many Q-particles multiple whquestions are permitted to have. Given the central role that this property plays in our analysis of English, the variation seen in (54) and (55) invites the following hypothesis regarding German. Perhaps the variation between English and German seen in (40)–(43) is ultimately that observed between Tlingit and Navajo in (54) and (55). That is, perhaps German, like Navajo, does not permit multiple wh-questions to have multiple Qs. As we will see, such an analysis of German does indeed predict the data in (40) and (43). Before we come to this, however, let us first determine how our Q-based theory is to analyze languages where multiple Qs are not possible. Recall that the English ForceQ2 head requires a wh-question to contain multiple instances of Q. It follows, of course, that this ForceQ2 head must be absent from those languages where multiple wh-questions cannot have multiple Qs. For these latter languages, then, let us assume that their lexicons contain a separate Force head, one that is, in turn, absent from languages like English and Tlingit. This Force head, which we will label ‘ForceQ+’ is assumed to have the following semantics. (56)

The Head ForceQ+ in the Semantics of German Multiple Wh-Questions [[ ForceQ+ i XP ]]g = λp [ ∃f. ∃h . p = h ([[XP]]F g(i / f)) ]

Thus ForceQ+ shares with ForceQ2 the property of introducing multiple existential quantifiers into the meaning of the wh-question. Unlike ForceQ2, however, the meaning of ForceQ+ already contains the choice function variable bound by one of those quantifiers. Consequently this ForceQ+ head can bind only one Q-particle. Therefore, even though the complement of ForceQ+ may have multiple wh-words, only one of those wh-words can be dominated by a QP.23 Let us then examine the semantics that (56) yields for German multiple whquestions. We assume that the multiple wh-questions of German are headed by ‘ForceQ+’, as illustrated in (57). (57)

The Syntax and Semantics of Sentence (43a)

130

THE GRAMMAR OF Q

The reader is invited to confirm that our semantics in (56) yields as the meaning of (43a) the set of propositions indicated in (57). This is the set of propositions p such that there is some choice function h that yields p when given the following set of propositions: those propositions q of the form ‘x bought y’, where y is some nonhuman and x is the value of some choice function f over the set of humans. Given the nature of choice functions, this is simply the set of all those propositions p of the form ‘x bought y’, where x is some human and y is some nonhuman. Thus the formula derived in (57) is equivalent to that in (58), which quite obviously represents the set of possible answers to (43a). Thus we find that our semantics in (56) and (57) correctly interprets German multiple wh-questions. (58)

Formula Equivalent to that in (57) λp [ ∃x ∈ human . ∃y ∉ human. p = x bought y ]

We find, then, that the addition of the ForceQ+ head in (56) allows our system to correctly interpret German multiple wh-questions.24 We will now see that it also predicts the data in (40) and (43). Let us begin by considering the Superiority data in (43). The following diagram sketches the central ideas. (59) Single QP in Multiple Wh-Question Predicts No Superiority Effects In cases where the highest wh-word is not dominated by the single QP, our system derives the ‘Superiority-violating’ order. a.

Derivation of Superiority-Satisfying Order (43a): Highest Wh-Word Dominated by QP [ ForceQ+ 1 [ wer Q1 ]2 hat [ t2 was gekauft ] ]

b. Derivation of ‘Superiority-Violating’ Order (43b): Lower Wh-Word Dominated by QP [ ForceQ+ 1 [ was Q1 ]2 hat [ wer t2 gekauft ] ]

Recall, again, that under our Q-based account, the fronting of a wh-word is ultimately fronting of a QP projection dominating it. It follows that if a wh-word is not dominated by a QP, then it will not undergo any fronting. More concretely, if a wh-word is not dominated by a QP, then its position will fail to contain a ‘goal’ for probing by interrogative C. Now, via our addition of (56), German multiple wh-questions only contain a single QP. Under the simplest assumptions, this single QP is free to dominate any wh-word in the sentence. Therefore, while this QP could dominate the structurally highest wh-word (59a), it could also dominate any of the lower wh-words (59b). Clearly, if the QP dominates the highest wh-word, then that wh-word will appear overtly fronted, deriving the ‘Superiority-satisfying’ order (59a). However, if the QP dominates any of the lower wh-words, then it will be those wh-words that appear overtly fronted in the sentence, yielding the ‘Superiority-violating’ order (59b).


131

Under our analysis, then, there is no true violation of Superiority in German sentences like (43b). Rather, in such sentences, the structurally higher in situ whwords simply fail to contain the true target of movement, the QP projection. Interestingly, the fact that in situ wh-words in German are not dominated by QPs also predicts the Intervention facts in (40). First, note that the semantics of ‘ForceQ+’ in (56) entails that any in situ wh-word in German fails to be dominated by a QP. After all, (56) entails that German wh-questions only contain a single QP, and this QP necessarily undergoes fronting. Thus, as can be seen in (57), any German wh-word not in the left periphery is thereby not contained within a QP. Since German in situ wh-words are not dominated by QPs, it trivially follows that they are not c-commanded by Qparticles. Therefore, if such wh-words were ever c-commanded by any (non-Q) focus-sensitive operator, such an operator would be the first focus-sensitive operator to c-command the wh-word. Consequently, an instance of the configuration in (53) would result, triggering an Intervention Effect. (60)

Single QPs in Multiple Wh-Questions Predicts Intervention Effects

In summary, we have seen that the addition of ForceQ+ (rather than ForceQ2) to the lexicon of German has the following consequences: (i) it provides a compositional semantics for German multiple wh-questions; (ii) it predicts that German multiple wh-questions should not exhibit Superiority Effects; and (iii) it predicts that the in situ wh-words of German multiple wh-questions should be subject to Intervention Effects. Thus it is possible in our Q-based account to derive the data in (40) and (43) from a single core property of the language. Furthermore, when we compare our analyses of German and English, an interesting picture emerges of the cross-linguistic variation observed in (40)–(43). Under this account, the core, underlying difference between English and German lies in their inventory of Force heads.25 English possesses ForceQ2 and not ForceQ+, which entails that its wh-questions exhibit Superiority Effects, but not Intervention Effects. German, however, possesses ForceQ+ and not ForceQ2, which entails that its wh-questions exhibit Intervention Effects, but not Superiority Effects. We see, then, that our Q-based theory can provide a unified account of the data in (40)–(43), one that derives these two well-known differences between German and English from a single point of variation in their lexicons. 4.4.2.1 Intervention Effects and Superiority in D-Linked Wh-Questions Before we turn to the final section of this chapter, I wish to discuss three issues that arise for our analyses of German and English. The first concerns the behavior of ‘D-linked’ wh-questions in English.

132

THE GRAMMAR OF Q

To begin, our analysis of the data in (40)–(43) easily generalizes beyond simply English and German. Indeed, our Q-based account of (40)–(43) makes the following general prediction. (61) The Complementarity of Superiority Effects and Intervention Effects In any language L, the in situ wh-words of a multiple wh-question of L are subject to Intervention Effects if and only if the multiple wh-questions of L are not subject to Superiority Effects.

As first noted by Pesetsky (2000), the generalization in (61) receives support from the behavior of so-called D-linked wh-questions in English. For our purposes, a D-linked wh-question is one whose answers must make reference only to a conversationally given set of entities. Thus a question like Which of these books did you read? is D-linked, since its answers must name one of a conversationally given set of books.26 As made famous by Pesetsky (1982), such wh-questions present apparent counterexamples to the claim that English multiple wh-questions must satisfy Superiority. (62)

D-Linked Multiple Wh-Questions in English Can Violate Superiority a. [ Which book ]2 did [ which boy ]1 read? b. * What2 did who1 read?

Interestingly, Pesetsky (2000) observes that D-linked wh-questions in English also possess another exceptional property. When such questions violate Superiority, they are also suddenly subject to Intervention Effects. That is, if a structurally lower wh-word is fronted in a D-linked multiple wh-question, as in (62a), then the in situ wh-word is subject to Intervention Effects. (63) D-Linked Multiple Wh-Questions in English Are Subject to Intervention Effects a. * Which book didn’t which boy read? b. Which boy didn’t read which book?

In summary, then, we see that it is precisely when English wh-questions (exceptionally) violate Superiority that Intervention Effects arise for their in situ wh-words. Thus the general pattern in (61) is actually reflected within a particular subcomponent of English. While the facts in (62) and (63) support our prediction in (61), they nevertheless raise important questions for our analysis of English. How is it that D-linking of a question allows it to display the exceptional properties in (62) and (63)? In the remainder of this subsection I will sketch out an answer that is in line with our overall Q-based account. We have claimed that the lexicon of English contains two different interrogative Force heads: ForceQ and ForceQ2. Let us, then, suppose that English also possesses a third interrogative force head: ForceQ-Dlink. The characteristic properties of ForceQ-Dlink are the following: (i) it has exactly the same interpretation as ForceQ+ in (56), and (ii) it can only appear in D-linked wh-questions.27 To make this proposal more concrete, let us assume that ForceQ-Dlink has the semantics in (64).


(64)

133

The Head ForceQ-Dlink in the Semantics of English D-Linked Wh-Questions [[ ForceQ-Dlink i XP ]]g = λp [ ∃f. ∃h . p = h ([[XP]]F g(i / f)) ] if all the propositions in λp [ ∃f. ∃h . p = h ([[XP]]F g(i / f)) ] are ‘familiar’, otherwise undefined.

According to (64), ForceQ-Dlink possesses the normal-semantic value of ForceQ+, but introduces the presupposition that all the propositions in the answer set to the question are ‘familiar’, where ‘familiar’ is taken to mean that they make reference only to discourse-given entities. Assuming that this presupposition amounts to ‘D-linking’ of the question, the semantics in (64) ensures that the ForceQ-Dlink only appears in D-linked wh-questions. Moreover, given that its normal-semantic value is identical to that of German’s ForceQ+, it follows that any wh-questions containing ForceQ-Dlink will have the following properties: (i) they will appear to obviate Superiority, and (ii) their in situ wh-words will be subject to Intervention Effects. Thus the addition of ForceQ-Dlink to the English lexicon would predict that D-linked—and only D-linked— wh-questions in English exhibit the properties in (62) and (63). 4.4.2.2 Intervention Effects and Superiority in Ternary English Wh-Questions We have just seen how D-linked wh-questions in English support the prediction in (61). Interestingly, as first observed by Pesetsky (2000), there is another ‘subarea’ of English where the correlation in (61) can be observed: nonbinary multiple wh-questions. First, we find from sentences like the following that multiple wh-questions in English can freely violate Superiority, just so long as they contain more than two wh-words. (65)

Nonbinary Multiple Wh-Questions in English Can Violate Superiority (Kayne 1983) a. Who1 gave what2 to whom3? b. What2 did who1 give to whom3? c. To whom3 did who1 give what2?

Furthermore, just as we saw for D-linked wh-questions, whenever Superiority is violated in nonbinary wh-questions, the in situ wh-words are subject to Intervention Effects. (66)

Nonbinary Multiple Wh-Questions Are Subject to Intervention Effects (Pesetsky 2000) a. Who1 didn’t give what2 to whom3? b. * What2 didn’t who1 give to whom3? c. * To whom3 didn’t who1 give what2?

Thus we again see that it is precisely when English wh-questions (exceptionally) violate Superiority that Intervention Effects arise for the in situ wh-words, supporting the general prediction in (61).

134

THE GRAMMAR OF Q

But why do nonbinary multiple wh-questions in English display these exceptional properties? What distinguishes them from binary wh-questions, which cannot violate Superiority and do not display Intervention Effects? In the remainder of this subsection I will (very lightly) sketch out an answer that is in line with our overall Q-based account. Unfortunately the answer that I will put forth will be even more vague and programmatic than what was offered for D-linked wh-questions. Nevertheless, it may point the way to a fuller treatment of these facts. First, let us observe that our analysis of English does not yet provide a semantics for nonbinary, non-D-linked wh-questions. The only Force heads we currently have at our disposal for English are ForceQ, ForceQ2, and ForceQ-Dlink. As none of these will be sufficient for nonbinary, non-D-linked wh-questions, we must assume that such questions in English contain yet another, fourth interrogative Force head. What could this Force head be? Let us suppose for the moment that it is none other than the ForceQ+ head of German. That is, let us assume that nonbinary multiple wh-questions in English can contain ForceQ+.28 Clearly such a hypothesis would make the following predictions: (i) nonbinary multiple wh-questions can freely violate Superiority, and (ii) in situ wh-words in such questions will be subject to Intervention Effects. Such an account, then, would easily capture (most of) the data in (65) and (66).29 On the other hand, such an account would seem to make dramatically incorrect predictions for binary multiple wh-questions in English. After all, if ForceQ+ were present in the English lexicon, nothing would seem to prevent it from appearing in binary wh-questions. Consequently, we would fail to predict the observed differences between English and German binary wh-questions in (40)–(43). Therefore, if we wish to introduce ForceQ+ into our theory of English, we must also find some way to prevent ForceQ+ from appearing in English binary wh-questions. In this context, note the following, distinctive property of English: it possesses the ForceQ2 head. Moreover, note the following, distinctive property of ForceQ2: as noted earlier, it can only appear in binary wh-questions. Thus we find that English differs from German in that it possesses a Force head that is tailored specifically for binary multiple wh-questions. This observation invites the following hypothesis. Suppose that the inability for ForceQ+ to appear in English binary wh-questions were due to it being ‘blocked’ by ForceQ2, in a manner akin to morphosyntactic ‘blocking’ (Aronoff 1976). The logic of such an account would run as follows. Although a binary English wh-question containing ForceQ+ would be interpretable, it would have the same interpretation as a nearly identical structure containing ForceQ2. The head ForceQ2, however, is specific to binary wh-questions, while the head ForceQ+ is interpretable in any multiple whquestion. For this reason, a general ‘blocking principle’ that “more specific forms must be used, when they are possible”, would rule out use of ForceQ+ in English binary multiple wh-questions. Such a principle, though, would not ‘block’ ForceQ+ from nonbinary wh-questions, as there is no Force head in English that is tailored to such questions. Similarly, this principle would not ‘block’ ForceQ+ from German binary wh-questions, as German happens to lack the ‘more specific’ ForceQ2 head. Although the details of this ‘blocking account’ remain vague, an account of this sort would permit an analysis whereby ForceQ+ in English is restricted to nonbinary


135

wh-questions. As we have seen, such an analysis would largely capture the exceptional properties of English nonbinary wh-questions in (65) and (66). 4.4.2.3 Intervention Effects in German ‘Separation Structures’ Finally, I would like to briefly discuss a potential failing of our theory of Intervention Effects in German. Throughout our discussion, we have focused on Intervention Effects in German multiple wh-questions. Traditionally, however, the term ‘Intervention Effect’ has also been applied to similar phenomena observed in German ‘separation structures’ (Beck 1996). As the data in (67) illustrate, it is possible in German for the wh-word to be ‘separated’ from the rest of the wh-phrase (67a,c). Importantly, if such separation occurs, then the IP-internal remnant of the wh-phrase cannot be c-commanded by any of the ‘offending operators’ (67b). (67)

Intervention Effects in German Separation Structures a. Wen1 hat Hans [ ___1 von den Musikern ] getroffen? who has Hans of the musicians met Which of the musicians has Hans met? b. * Wen1 haben keine Studenten [ ___1 von den Musikern ] getroffen? who have no students of the musicians met c. [ Wen von den Musikern ] haben keine Studenten getroffen? who of the musicians have no students met Which of the musicians has no student met?

The similarity of the facts in (67) to those in (40) have lead most linguists to consider them instances of the same general phenomenon, that is, Intervention Effects. However, the theory of Intervention Effects we assume here offers no obvious explanation for the facts in (67), a property that it inherits from Beck (2006).30 After all, Intervention Effects are assumed to result from the uninterpretability of the structure in (53), and there is no obvious instance of (53) in (67b). Therefore, unlike earlier theories of Intervention Effects (Beck 1996; Pesetsky 2000), ours does not straight forwardly extend to the data in (67). On the other hand, our inability to predict the data in (67) may actually be a virtue of our account. That is, there is in fact some indication that the data in (67) represent a distinct phenomenon from that seen in (40). To ease our discussion here, I will refer to the phenomenon in (67) as ‘S(eparation)-Intervention Effects’ and the phenomenon in (40) as ‘I(n situ)-Intervention Effects’. Examination of German reveals that the set of ‘offending operators’ triggering S-Intervention Effects is different from those that trigger I-Intervention Effects. First, as originally reported by Pesetsky (2000), S-Intervention Effects in German can be triggered by wh-words. (68)

In Situ Wh-Words Are Interveners for the German Separation Structures “Guenther Grewendorf (personal communication) points out that wh-in-situ appears to generate an intervention effect in German separation constructions. . .” (Pesetsky 2000: 117)

136

THE GRAMMAR OF Q

a. Was1 hat der Professor den Studenten [ ___1 alles ] geraten? what has the professor the students all advised What all did the professor advise to the students? (Pesetsky 2000: 117) b. * Was1 hat wer den Studenten [ ———1 alles ] geraten? what has who the students all advised (Pesetsky 2000: 117) c. * Wen1 hat wer [ ———1 von den Musikern ] getroffen? who-ACC has who-NOM of the musicians met (Kai von Fintel, personal communication)

Interestingly, however, nonbinary wh-questions in German show that wh-words do not trigger I-Intervention Effects. An in situ wh-word can freely be c-commanded by other (in situ) wh-words. (69)

In Situ Wh-Words Are Not Interveners for Other In Situ Wh-Words Wer hat wem was gegeben? who-NOM has whom-DAT what given Who gave what to who? (Kai von Fintel, personal communication)

It is apparent, then, that the ‘offending operators’ triggering S-Intervention Effects are not necessarily the same as those that trigger I-Intervention Effects.31 This fact, in turn, lends credence to the notion that the two phenomena are ultimately distinct, despite their superficial similarity. Thus we may conclude that, rather than present a challenge to our proposed theory, the facts concerning ‘Intervention Effects’ with German separation structures might actually provide further evidence for it. 4.4.3 Intervention Effects in Pied-Piping Structures The preceding sections have shown how our Q-based theory in (1) can capture the distribution of Intervention Effects in the multiple wh-questions of German and English. In this section we will turn our attention to another environment where Intervention Effects arise: pied-piping structures. To begin, let us recall our general theory of pied-piping structures. We assume that in all wh-fronting languages, pied-piping structures have a form homologous to those in Tlingit. That is, in all languages, pied-piping structures are simply cases where the Q-particle of the wh-question takes as complement a phrase properly containing the wh-word. (70)

Q-Based Theory of Pied-Piping Structures a. Whose father’s cousin’s uncle did you meet at the party? b. [QP [ [ [ [ whose ] father’s ] cousin’s ] uncle ] Q ] did you meet at the party?

Thus, in a pied-piping structure, phrasal material intervenes between the focused whword and the focus-sensitive Q-particle that it ‘associates’ with. Consequently, if the wh-word in such structures were to appear within the scope of a focus-sensitive operator within the pied-piped phrase, then a structure of the type in (53) would result. The following diagram illustrates.


(71)

137

Intervention Effects in Pied-Piping Structures [

[QP [ … Offending Operator … [ wh-word ]

… ] Q ]1

[ … t1 … ] ]

Intervention Effect Configuration (53)

Given our theory of Intervention Effects, we predict that structures of the type in (71) will be uninterpretable, and thus ill-formed. Therefore we find that by combining our theory of pied-piping structures with our theory of Intervention Effects, we thereby make the following, general prediction. (72)

Intervention Effects in Pied-Piping Structures In all languages, if a pied-piping wh-word appears within the scope of an ‘offending operator’ within the pied-piped constituent, then an Intervention Effect will be triggered.

But is the prediction in (72) true? Sauerland and Heck (2003) have independently shown that it is indeed borne out for German. They show that, for every ‘offending operator’ triggering an Intervention Effect with in situ wh-words, placement of that operator in configurations like (71) creates an ill-formed structure. For example, we can see in (73) that it is generally possible for a ‘pied-piping’ wh-word in German to be c-commanded by a determiner in the pied-piped constituent (73a). If that determiner, however, is the offending operator kein ‘no’, the sentence becomes ill-formed (73b). (73)

Intervention Effects in German Pied-Piping Structures (Sauerland & Heck 2003) a. Fritz möchte wissen [ ein wie schnelles Motorrad ] du fahren darfst. Fritz wants to.know a how fast motorbike you drive may Fritz would like to know how fast a motorbike you are allowed to drive. b. * Fritz möchte wissen [ kein wie schnelles Motorrad ] du fahren darfst. Fritz wants to.know no how fast motorbike you drive may

Sauerland and Heck (2003) go on to show that parallel data hold for each of the ‘offending operators’ of German. We can also detect evidence for (72) in the pied-piping structures of English. Here, however, the evidence is somewhat more subtle than the striking contrasts observed above for German. This is largely because independent features of English render (72) somewhat difficult to test. The overall issue that, unlike pied-piping in German, the only way that one can pack an offending operator into a pied-piping structure of English is if the fronted phrase undergoes so-called Massive Pied-Piping (Heck 2004, 2008). The phenomenon of ‘Massive Pied-Piping’ will receive a proper introduction and discussion in chapter 5. For our purposes here, however, all we need know is that such pied-piping, illustrated by structures like (74), is already independently ‘marginal’ and ‘unnatural’ in English.

138 (74)

THE GRAMMAR OF Q

Massive Pied-Piping in English a. (?) A picture of which president does Jim own? b. [DP A [NP picture of which president ] ] does Jim own?

Although sentences like (74) are already marginal in English, they become distinctly worse if the ‘massively pied-piping’ wh-word is in the scope of an ‘offending operator’. (75)

Intervention Effects in English Massive Pied-Piping Structures32 a. (?) [DP A [NP picture of which president ] ] does Jim own? b. *[DP No [NP picture of which president ] ] does Jim own? c. *[DP Only [NP PICTURES of which president ] ] does Jim own?

Furthermore, as pointed out by David Pesetsky (personal communication), Massive Pied-Piping in English generally becomes less marginal when it targets a subject, as in (76). (76) Massive Pied-Piping of Subjects in English [DP A [NP picture of which president ] ] hangs in Jim’s office?

Importantly, varying the grammatical function of the pied-piped phrase does nothing to lessen the contrasts in (75). Rather, the contrasts are made even shaper. (77) Intervention Effects in English Massive Pied-Piping Structures a. [DP A [NP picture of which president ] ] hangs in Jim’s office? b. * [DP No [NP picture of which president ] ] hangs in Jim’s office? c. * [DP Only [NP PICTURES of which president ] ] hang in Jim’s office?

It is important to note that the ill-formedness of (75b,c)–(77b,c) is not due to there simply being an offending operator somewhere within the pied-piped phrase. As shown by (78) and (79), the ill-formedness of (75b,c)–(77b,c) is crucially tied to the wh-word being within the scope of the offending operator. (78) No Intervention Effect if the Wh-Word Is Not Within the Scope of the Offending Operator a. [ Which picture [ only of presidents ] ] does Jim own? b. [ Which picture [ only of presidents ] ] hangs behind Jim’s desk? (79) a. [ Which picture [ containing no presidents ] ] does Jim own? b. [ Which picture [ containing no presidents ] ] hangs behind Jim’s desk?

Similarly, we should note that the ill-formedness of (75b,c)–(77b,c) is not due to any ‘pragmatic unnaturalness’ of the questions themselves. As the sentences below demonstrate, such questions are well-formed and natural, just so long as there is no piedpiping of the offending operator.


(80)

139

The Well-Formedness of (75) and (77) Without Pied-Piping of the Operator a. [ Which president ]1 does Jim own [ no picture of t1 ] ? (cf. (75b)) b. [ Which president ]1 does Jim own [ only PICTURES of t1 ] ? (cf. (75c)) c. [NP Pictures of which president ] ] don’t hang in Jim’s office? (cf. (77b))

We find, then, that the pied-piping structures of English provide further evidence for the generalization in (72). These facts also make several additional points concerning the nature of Intervention Effects. First, recall that English does not (generally) exhibit Intervention Effects with its in situ wh-words. Thus, despite the ill-formedness of (75b,c), the following are well formed. (81)

English In Situ Wh-Words Are Immune to Intervention Effects a. Who owns [DP no [NP picture of which president ] ] ? (cf. (75b)) b. Who owns [DP only [NP PICTURES of which president ] ] ? (cf. (75c))

Interestingly, our analysis of English multiple wh-questions predicts the wellformedness of (81a,b). Both surface forms permit analyses where the Q associated with the in situ wh-word appears within the scope of the ‘offending operator’. The structures in (82) illustrate. (82)

English In Situ Wh-Words Are Immune to Intervention Effects a. [QP Who Q ] owns [DP no [NP picture of [QP which president Q ] ] ] ? b. [QP Who Q ] owns [DP only [NP PICTURES of [QP which president Q ] ] ] ?

However, such a local position for Q is not possible in (75b,c). Given our Q-based theory of wh-fronting, the Q-particle must take as complement the fronted phrase of the wh-question. Thus, in (75b,c), the Q must be located outside the scope of the offending operator, yielding a configuration of the type in (71). Thus our Q-based account correctly predicts the contrast between (75b,c) and (81a,b). This is, to my knowledge, a unique prediction of our Q-based account. For example, these data cannot be captured under the syntactic theory of Intervention Effects put forth by Beck (1996). In brief, Beck (1996) proposes that Intervention Effects follow from a syntactic constraint banning covert movement over the offending operators. In order for such an account to capture the contrast between (75b,c) and (81a,b), it must be assumed that (i) pied-piping wh-words in English are covertly extracted from within the pied-piped phrase, and (ii) in situ wh-words in English undergo no covert movement. Such a difference between pied-pipers and in situ wh-words, however, seems hard to independently motivate. More generally, the contrast between (75b,c) and (81a,b) demonstrates that Intervention Effects do not simply result from a wh-word being inside the scope of an offending operator. After all, in both (75b,c) and (81a,b), the wh-word is inside the scope of such an operator. Rather, the real issue is whether a Q-particle is inside the scope of the offending operator, intervening between it and the wh-word. In illformed sentences like (75b,c), the Q-particle cannot be in the scope of the operator, and so the sentences are uninterpretable. However, sentences like those in (81) can be

140

THE GRAMMAR OF Q

parsed as having a Q-particle in the scope of the offending operator, and so the sentences are well formed. Finally, note the following related property of our theory of Intervention Effects in pied-piping structures. Our prediction of the generalization in (72) in no way rests upon any assumptions regarding multiple wh-questions. That is, our derivation of (72) in no way appeals to whether a given language possesses ForceQ2 or not. Thus our theory predicts that (72) will hold across all languages, regardless of whether or not their in situ wh-words are subject to Intervention Effects. As we have seen, this is indeed a correct prediction, as (72) holds for both English and German. We find, then, that our proposed account of (72) rightly (and uniquely) predicts this difference between the Intervention Effects triggered in pied-piping structures and those associated with in situ wh-words.

5

Constraints on Pied-Piping and Secondary Wh-Fronting

5.1 Introduction In the previous chapter we began our argument that the Q-based structure in (1) extends beyond the Tlingit language, and in fact underlies the wh-questions of all wh-fronting languages. (1)


CP

QP1 Complementation XP

Q CQ

IP

Agree/ Attract

… wh-word…

QP1 Overt Movement

In this chapter we continue our argument, focusing primarily on the rather stringent restrictions on pied-piping found in languages like English. Such restricted piedpiping I will label ‘limited pied-piping’. In languages exhibiting limited pied-piping, certain structures are not permitted to dominate the wh-word within the fronted phrase of a wh-question. I will put forth a ‘Q-based’ theory of limited pied-piping, 141

142

THE GRAMMAR OF Q

one in which the restrictions on pied-piping structures ultimately follow from a requirement that Q undergo Agreement with the wh-word. Moreover, we will see that this account lays the groundwork for treatments of two related phenomena: secondary wh-movement (Heck 2008) and massive pied-piping (Heck 2008). Section 5.2 presents our treatment of limited pied-piping. It begins with an introduction to the phenomenon, as well as the questions it raises for our Q-based account. Following this, I provide some background to the leading ideas employed by our analysis, particularly the work of Kratzer and Shimoyama (2002) and Pesetsky and Torrego (2007). With this background in place, I then introduce the central claim of our analysis: languages exhibiting limited pied-piping are those where the Q-particle must Agree with the wh-word inside its complement. I show how this generalization can capture the core aspects of limited pied-piping, as well as a number of other, related phenomena noted in the literature on pied-piping. Having shown how our account derives the core properties of limited pied-piping, I turn in section 5.3 to some further predictions made by our account. I first argue that our Q-based account is able to derive the ‘Generalization on Recursive Pied-Piping’ of Heck (2008). I then discuss the predictions our account makes regarding the piedpiping of main predicates, as well as the relationship between pied-piping and left peripherality. Following this, I argue that our Q-based account correctly predicts that pied-piping should (in principle) be able to occur in free variation with extraction of the wh-word. Finally, section 5.3.5 demonstrates how our Q-based account can capture the fact that limited pied-piping is subject to the Coordinate Structure Constraint. In section 5.4, we momentarily turn away from pied-piping per se to consider a related phenomenon. In the pied-piping structures of many languages, a ‘piedpiping’ wh-word must sometimes undergo movement within the pied-piped phrase. This phenomenon is dubbed ‘secondary wh-movement’ by Heck (2008). I present a Q-based analysis of secondary wh-movement under which the wh-fronting occurs to place the wh-word in a position where it is accessible for Agreement with Q. I then argue that secondary wh-fronting in Chol and Tzotzil (Mayan) may provide interesting support for our Q-based account. The final topic of this chapter is so-called massive pied-piping. Following Heck (2008), the term ‘massive pied-piping’ describes cases where the stringent constraints of limited pied-piping languages appear somewhat weakened. I argue that a particular Q-based account of massive pied-piping can capture certain properties that it seems to exhibit. However, I also note that the properties of most importance to Heck (2008) do not obviously follow under our proposed account. Consequently, massive pied-piping remains an important outstanding puzzle for our Q-based theory.

5.2 Q/Wh-Agreement and the Constraints on Pied-Piping In chapter 4 I argued that our Q-based analysis in (1) seems to provide an attractive theory of pied-piping structures. Under this theory, pied-piping structures are nothing more than cases where the Q takes as its sister a phrase strictly containing the wh-word

CONSTRAINTS ON PIED-PIPING AND SECONDARY WH-FRONTING

143

associating with it. Thus even complex pied-piping structures like (2a) can be analyzed as in (2b), as simply cases of normal phrasal movement of the QP. (2)

The Pied-Piping Structures of English Under the Q-Based Theory a. Whose father’s cousin’s uncle did you meet at the party? b. [QP [ [ [ [ whose ] father’s ] cousin’s ] uncle ] Q ] did you meet at the party?

This general theory of pied-piping structures is, of course, based on their surface form in Tlingit. To recall, Tlingit pied-piping structures transparently possess the structure in (2b), as the Q-particle sá always appears directly to the right of the fronted phrase in a Tlingit wh-question. (3)

The Pied-Piping Structures of Tlingit a. [QP [DP Aadóo yaagú ] sá ] ysiteen? who boat Q you.saw.it Whose boat did you see? b. * [DP [QP Aadóo sá ] yaagú ] ysiteen? who Q boat you.saw.it

We find, then, that our Q-based theory would view all pied-piping structures as homologous to the Tlingit phenomenon in (3). However, when we compare the piedpiping structures of languages like English to their putative correlates in Tlingit, we find that there are some striking differences between them. Generally speaking, the differences lie in the ‘size’ of what can be ‘pied-piped’. Tlingit permits the wh-word in a pied-piping structure to be dominated by structures that English and other wellstudied languages never allow. We have already encountered one rather prominent example of this disparity. Recall that the wh-word of a Tlingit wh-question can be contained inside an island within the fronted phrase (4). I will refer to such structures as “pied-piping past islands”. (4)

Pied-Piping Past Islands in Tlingit tuwáa sigóo? a. [ [ Wáa kwligeyi CP] xáat NP] sá i how it.is.big.REL fish Q your spirit.at it.is.happy Literally: ‘A fish that is how big do you want?’ b. [ [ Daat yís ] át ] sákwshéiwégé? what for thing Q.DUB Literally: ‘A thing for what is this?’ (Nyman & Leer 1993: 120) c. [ [ Goodáx ] k’anáaxán tlein ] sáyá du kát satéen? where.from fence big Q.FOC its surface.to placed(?) Literally: ‘A big fence from where was placed on it?’ (Nyman & Leer 1993: 150) d. [ [ Goodáx ] káa ] sáyá yéi yatee? where.from man Q.foc he.is Literally: ‘A man from where was he?’ (Dauenhauer & Dauenhauer 1987: 168)

Curiously, while Tlingit allows pied-piping past islands, the most commonly studied wh-fronting languages do not. As illustrated in (5), the English correlates of (4) are ill-formed.

144 (5)

THE GRAMMAR OF Q

No Pied-Piping Past Islands in English a. * [DP A fish [CP that is how big ] ] do you want? b. * [DP A book [CP that who wrote ] ] did you buy?

Furthermore, as documented by Heck (2004, 2008), such structures are similarly illformed in many other wh-fronting languages. But it is not merely pied-piping of islands that distinguishes the pied-piping structures of Tlingit. To facilitate our discussion here, let us adopt the following, more general terminology: (6)

A wh-question exhibits pied-piping past X if the wh-word is dominated by an instance of X within the fronted phrase of the wh-question.

With this terminology in place, note that several authors have offered the generalization that English and other wh-fronting languages do not permit pied-piping past lexical categories (Cowper 1987; Webelhuth 1992; Grimshaw 2000).1 That is, in the most commonly studied wh-fronting languages, no wh-operator can be dominated by a lexical category within the fronted phrase of the wh-question. The ill-formed English structures in (7) illustrate.2, 3, 4 (7)

No Pied-Piping Past Lexical Categories in English a. I wonder [[DP whose [NP pictures ] ] John bought ]? b. * I wonder [[NP pictures of whom ] John bought ]? c. * I wonder [[AP proud of whom ] John was ]? d. * I wonder [ [VP eaten what ] John has ]?

Although pied-piping past lexical categories is ill-formed in many languages, it does not appear to be problematic in Tlingit. Indeed, under the plausible assumption that Tlingit relative clauses are adjuncts to NP, such pied-piping is widely exemplified by sentences like those in (4), where the wh-operator is buried within a relative clause.5 We find, then, that our Q-based theory of pied-piping is presented with a difficult challenge. Although it proposes that all pied-piping structures be reduced to the Tlingit phenomenon in (3), those Tlingit structures exhibit properties contrary to the most well-studied cases of pied-piping. Consequently, the Tlingit structures in (3) are not perfectly homologous to the pied-piping structures of English. Now, one could then conclude that the two structures are not homologous at all, and that our Q-based theory of pied-piping is just wrong for languages like English. However, I will argue that such a reaction would be too extreme. Rather, we will see that a slight addition to our Q-based theory will allow it to capture the observed differences between English and Tlingit pied-piping. To again facilitate our discussion here, I will use the term “limited pied-piping”, defined as follows, to describe the pied-piping structures of languages like English. (8)

A limited pied-piping structure is a pied-piping structure where pied-piping past islands and pied-piping past lexical categories is not permitted.


145

Similarly, I will use the term “limited pied-piping language” to refer to languages where all pied-piping structures are instances of limited pied-piping. Now, in order for our Q-based theory to be applied to the limited pied-piping languages, some account must be offered for why those languages do not permit pied-piping past islands or lexical categories. In the remainder of this section, I put forth such an account. I begin, in the following subsection, with some critical background. 5.2.1 Background: Kratzer and Shimoyama (2002) and Pesetsky and Torrego (2007) The leading idea behind our account of limited pied-piping is that wh-words in some languages must undergo Agreement with the c-commanding Q-particle. Interestingly, this notion of ‘Q/wh-Agreement’ is independently proposed in the work of Kratzer and Shimoyama (2002: sect. 9). In brief, Kratzer and Shimoyama propose that in some (but not all) languages, wh-operators bear an uninterpretable instance of the feature [Q].6 In these languages, then, the wh-word and the c-commanding Q-particle must undergo Agreement; failure to do so would leave the uninterpretable [Q] on the wh-word, leading to a crash at LF. However, in those languages where wh-operators do not bear uninterpretable instances of [Q], no Agreement between the Q-particle and any wh-operators need take place. Kratzer and Shimoyama’s ultimate interest in these morphosyntactic hypotheses is their ability to account for certain differences between German (hypothesized to have ‘Q/wh-Agreement’) and Japanese (hypothesized not to). Interestingly, though, these same general hypotheses can be combined with our Q-based theory in (1) to provide an analysis of the limited pied-piping languages. Before this analysis can be presented, however, we must first flesh out the morphosyntactic hypotheses we just sketched. Because Kratzer and Shimoyama’s implementation of these hypotheses employs a syntactic theory different from that assumed here, we must develop our own.7 Such formal implementation will require us to be more explicit regarding the exact nature of feature valuation under Agree. Throughout the remainder of this chapter I will adopt the theory of feature valuation developed by Pesetsky and Torrego (2007).8 The characteristic property of this system is that valuation and interpretability are independent of one another. Consequently, there are four states that a given feature may be in: (i) valued and interpretable, (ii) valued and uninterpretable, (iii) unvalued and interpretable, and (iv) unvalued and uninterpretable. The following table illustrates this idea, as well as the notation we will use to represent each of these four states. (9)

The Independence of Valuation and Interpretability in Pesetsky and Torrego (2007) Feature = F

Interpretable (iF)

Uninterpretable (uF)

Valued (F[val]) Unvalued (F[ ])

iF[val] iF[ ]

uF[val] uF[ ]

146

THE GRAMMAR OF Q

Within Pesetsky and Torrego’s (2007) system, there are two principles that drive syntactic valuation. The first is the requirement that every feature must possess a value by LF. Because of this principle, any unvalued feature F[ ] must probe for a valued instance of itself F[val], at which point the usual mechanics of long-distance Agree apply (Chomsky 2000). The second is the requirement that all uninterpretable features uF must, by LF, be matched to some interpretable instance iF. For further details regarding this theory of feature valuation, I refer the reader to Pesetsky and Torrego (2007). With these ideas in place, let us now incorporate the morphosyntactic hypotheses of Kratzer and Shimoyama (2002) into our broader system. First, we assume that the whwords of some languages (e.g., German, English) bear uninterpretable, valued Q, while the wh-words of other languages (e.g., Japanese, Tlingit) do not bear any instance of Q. (10)

a. German Wh-Word: b. Japanese Wh-Word:

was uQ[+] dare

In languages where wh-words bear uQ[+], this uninterpretable Q-feature must, by LF, be ‘checked’ against some interpretable instance iQ. Given that the only head assumed to carry iQ is the Q-particle itself, the Q-particle in languages like German and English must Agree with the wh-word. In order for this Agreement to take place, however, we must assume that the Q-particles of such languages initially bear unvalued instances of Q.9 That is, in languages where the wh-words bear uQ[+], the Q-particle must in turn bear iQ[ ]. Of course, in languages where wh-words do not bear uQ[+], we can assume that the Q-particle simply bears an interpretable, valued instance of Q: (11)

a. German Q: b. Japanese Q:

iQ[ ] ka iQ[+]

Assuming the initial valuations in (10) and (11), we predict the necessity of Q/whAgreement in languages like German and English, and the absence of such Agreement from languages like Japanese and Tlingit. First, in languages where the Q-particle is lexically assigned iQ[ ], the lack of a value for iQ entails that the particle probe for a valued instance of the feature. Following Chomsky’s (2000) algorithm for probing, the first element bearing Q[val] that the Q-particle probes is the wh-word that it c-commands. Therefore the Q-particle will undergo Agreement with that wh-word, as shown in (12): (12)

Q-Particle Agreeing With Wh-Word in German QP XP YP

Q X

… was uQ[+] … Q/Wh-Agreement

iQ[+]


147

This Q/wh-Agreement has the following results: (i) the unvalued instance of [Q] on the Q-particle receives a value, and (ii) the uninterpretable instance of [Q] on the wh-word is matched to an interpretable instance. Consequently, both the Agreementdriving principles of Pesetsky and Torrego (2007) are satisfied, and the structure is well-formed. In languages where the Q-particle is lexically assigned iQ[+], however, the presence of a value for iQ entails that the Q-particle will not act as a probe. Furthermore, since the wh-words of such languages are assumed not to bear any instance of the Q-feature, nothing will require them to undergo Agreement with the Q-particle. Consequently, in such languages, there is no Q/wh-Agreement.

(13)

No Q/Wh-Agreement in Japanese XP XP YP

Q X

ka iQ[+]

… dare …

Given this implementation of Kratzer and Shimoyama’s (2002) theory of Q/whAgreement, we can now develop a Q-based analysis of the limited pied-piping languages. 5.2.2 The Theory of Limited Pied-Piping Languages To recall, the question we seek to answer is the following: What is responsible for the more constrained variety of pied-piping found in languages like English, where piedpiping structures are subject to constraints that do not hold in languages like Tlingit? Since the constraints governing limited pied-piping concern the ‘locality’ of the whword to the root of the pied-piped phrase, we should naturally seek to derive these constraints from independent ‘locality’ principles of the grammar. Let us, then, consider the following: (14)

The Nature of Limited Pied-Piping If the Q-particle must Agree with the wh-word it c-commands, then a wh-word cannot be dominated in the sister of Q by islands or lexical categories. Thus limited pied-piping languages are those where Q/wh-Agreement must occur.10

That is, I propose that the constraints governing pied-piping in languages like English would follow from a single requirement that the Q-particle and the wh-word Agree. Let us first consider the condition against pied-piping past islands. Recall our assumption from chapter 2 that probing and Agreement cannot apply across islands.

148

THE GRAMMAR OF Q

Under this general assumption, Q/wh-Agreement would, of course, not be able to apply across islands. Therefore, if we assume that limited pied-piping languages require Q/wh-Agreement, we correctly predict that such languages will not permit pied-piping past islands. As illustrated in (15), the domination of the wh-word by an island within the sister of Q would prevent Agreement between the Q-particle and the wh-word. (15)

Inability to Pied-Pipe Past Islands in English a. * [DP A fish [CP that is how big ] ] do you want? b. QP Q iQ[

DP D A

]

NP ISLAND

NP

CP

fish

that is how uQ[+] big

Probing/Agreement Blocked

XXXXXX

More concretely, in languages like German and English—where the Q-particle bears interpretable but unvalued Q (iQ[ ])—an island between the Q-particle and the whword prevents the Q-particle from receiving a value for its Q-feature by LF. As a result, such structures induce a crash at the LF interface. On the other hand, in languages where the Q-particle lexically bears valued Q (iQ[+]), an island between the Q-particle and the wh-word does not affect the LF interpretability of the structure. (16)

Ability to Pied-Pipe Past Islands in Tlingit tuwáa sigóo? a. [DP[CP Wáa kwligeyi ] xáat ] sá i how it.is.big.REL fish Q your spirit.at it.is.glad ‘How big a fish do you want?’ b. QP DP

Q iQ[+] sá

D

NP CP Wáa

kwligeyi

NP xáat

Thus languages without Q/wh-Agreement should permit pied-piping past islands. We might hypothesize then, that Tlingit (like Japanese) does not require Q/whAgreement. Let us now consider the constraint against pied-piping past lexical categories in languages like English. Interestingly, this might follow from certain independent


149

ideas regarding the structure of lexical projections. Let us adopt the following view, taken from recent work in Distributed Morphology. (17)

The Fine Structure of Lexical Categories (Embick & Marantz 2008) Every lexical projection (VP, NP, AP) is complement to a phase head (little-v, little-n, little-a).11 Diagram of the Lexical Projections nP

n

vP NP

Spell-Out Domain

v

aP VP

….

Spell-Out Domain

….

a

AP

Spell-Out Domain

…

Under the assumption that each of the ‘little categorial heads’ shares with little-v the property of being a phase head (Chomsky 2000), it would follow that Q/whAgreement cannot cross into lexical projections. As illustrated in (18), any material inside a lexical projection would occupy a separate spell-out domain from material outside the lexical projection. However, under the original formulation of the Phase Impenetrability Condition (PIC), syntactic operations such as Agree cannot apply to heads in separate spell-out domains (Chomsky 2000).12 Under these assumptions, then, no material inside a lexical projection can Agree with any head outside that lexical projection. Consequently, we predict that Agreement between a Q-particle and a wh-word buried within a lexical projection should be impossible. (18)

Inability for Q/Wh-Agreement to Cross a Lexical Projection QP … n/v/aP… n/v/a

Q

N/V/AP

Spell-Out Domain

… wh-word ….

XXXXXXX Agreement Impossible, Due to the PIC

Thus, in languages like English and German, where the Q-particle bears unvalued ‘iQ[ ]’, it should be impossible for a lexical category to intervene between Q and the whword. Consequently, pied-piping past lexical categories should be ill-formed in these languages. However, for languages like Tlingit and Japanese, where the Q-particle bears valued ‘iQ[+]’, no problem arises if the wh-word and Q-particle are separated by a lexical projection, and so pied-piping past lexical categories should be possible there.

150

THE GRAMMAR OF Q

In summary, broader syntactic assumptions predict that pied-piping past islands and pied-piping past lexical categories should be impossible in only the languages requiring Q/wh-Agreement. It appears, then, that the striking differences between the pied-piping structures of English and Tlingit are not necessarily due to some fundamental disparity in their nature, but instead, to a rather superficial morphosyntactic contrast. We find, then, that our Q-based theory of pied-piping structures offers a sensible perspective on the cross-linguistic variation observed in (4)–(7). But does it actually fare better than a ‘classic account’, appealing to a theory of “pied-piping,” as defined in chapter 1? After all, in order to properly limit pied-piping in English, we have had to supplement our basic Q-based theory in (1) with the following additional assumptions: (a) there are languages where Q and the wh-word must Agree, (b) Agreement cannot cross into islands, (c) lexical projections have the structure in (17), and (d) Agreement cannot cross separate spell-out domains (the PIC). In response to this criticism, it should first be noted that assumptions (a)–(d) have been independently proposed and defended. Thus they are not stipulations introduced ex nihilo here, but do enjoy some external currency and validity. Furthermore, it is worth noting that our account promises a unified treatment of the seemingly sui generis locality conditions on ‘pied-piping’, by reducing them to the independently observable locality conditions on Agree. Finally, I would submit that any theory admitting the existence of “pied-piping” as defined in chapter 1 will almost certainly be more complex than our Q-based account, in as much as it would still appeal to a special theory of pied-piping. Recall from chapter 4 that under the ‘classic picture’ of wh-fronting, pied-piping structures are a deviation from the expected pattern, and the postulation of “pied-piping” represents a weakening of the theory of movement. Even if such a theory offered an elegant treatment of the constraints on English pied-piping, it would still treat pied-piping structures as a kind of ‘aberration’, and would abandon the null hypothesis that movement applies only to projections headed by the features triggering it. In contrast, the Q-based account offers a system where pied-piping structures are fully consistent with this null hypothesis, where they are an organic, natural consequence of the general theory of A-bar movement. It is reasonable, then, to conclude on general grounds that the Q-based account— even paired with the additional assumptions in (a)–(d)—offers serious competition to classic accounts of pied-piping with similar empirical coverage. 5.2.3 Pied-Piping Across Lexical Categories and Across Phrases Having presented my general proposals concerning ‘limited pied-piping languages’, I will in this section return to a crucial empirical assumption of our preceding discussion: English and many of the most commonly studied wh-fronting languages disallow pied-piping past lexical categories. Although I illustrated this claim with but a handful of English data (7), its empirical coverage is much broader. Indeed, this generalization has (in essence) been discovered several times in the literature on piedpiping (Cowper 1987; Webelhuth 1992; Grimshaw 2000), as it accounts for an array


151

of seemingly disparate facts. Therefore, to provide a fuller picture of the empirical coverage provided by our theory of Q/wh-Agreement, I will review some of the phenomena falling under this broader generalization. After this, I will present some additional evidence for another crucial assumption of our account, that Q/wh-Agreement cannot take place across phases. To begin, one of the most striking properties of pied-piping structures is how constrained they are. No language permits a wh-word to pied-pipe any and all phrases containing it. Indeed, it might sometimes seem as if—except for but a few wellknown exceptions—pied-piping is generally impossible. Happily, things are not as dire as they might at first appear. While there is yet no completely unified account of all the constraints and quirks governing pied-piping structures,13 a number of phenomena have been found to follow from the generalization that (in the most commonly studied languages) pied-piping past lexical categories is impossible. For example, we have seen that this generalization captures the pattern in (7), repeated in (19): (19)

Complements of Lexical Heads Cannot Pied-Pipe (in Limited Pied-Piping Languages) a. I wonder [ [DP whose [NP pictures ] ] John bought ]? b. * I wonder [ [NP pictures of whom ] John bought ]? c. * I wonder [ [AP proud of whom ] John was ]? d. * I wonder [ [VP eaten what ] John has ]?

Now, one can certainly imagine other explanations for the impossibility of (19b,c). Indeed, facts such as these have lead some to propose the stronger generalization that no complements of any head can serve as pied-pipers (Kayne 1994; Koopman 2000; Koopman & Szabolcsi 2000). However, this stronger generalization is challenged by the ubiquity of pied-piping by complements of P. That is, contrary to the stronger generalization, it is possible in all the best-studied wh-fronting languages for wh-words to pied-pipe a PP from the complement of P, as briefly illustrated in (20). (20) Complements of P Can Pied-Pipe (in Limited Pied-Piping Languages) a. Icelandic: Ég velti því fyrir mér [ [PP við hvern ] thú talaðir ]. I roll it before me with who you talked I wonder who you talked with. b. Russian: [ v čey mashyne ] priyekhal? in whose car you.arrived Whose car did you arrive in?

To allow for these rather common structures, proponents of the stronger generalization must introduce special assumptions whereby the prima facie complements of P are (at the relevant level of syntax) actually specifiers of P. In contrast, the weaker generalization against pied-piping past lexical categories actually predicts the ubiquity of PP pied-piping. Because P may be regarded as a functional category

152

THE GRAMMAR OF Q

(in most languages), the weaker generalization correctly predicts that wh-words may generally be dominated by PPs within the fronted phrase. Thus a ban on pied-piping past lexical categories correctly predicts that complements of lexical heads cannot ‘pied-pipe’. In addition, it also predicts that modifiers of lexical projections cannot pied-pipe (Webelhuth 1992): (21)

Modifiers of Lexical Heads Cannot Pied-Pipe (in Limited Pied-Piping Languages) a. * [QP[DP The [NP party where ] ] Q ] will John enjoy? b. * [QP[VP Go where ] Q ] will you? c. * [QP[DP A [NP [DegP how big ] party ] ] Q ] will you throw?

Again, since the wh-words in each of these structures are contained within a lexical projection inside the fronted phrase, our Q-based account predicts their ill-formedness.14 A third prediction of our ‘weaker generalization’ concerns pied-piping by possessors. A pervasive phenomenon across limited pied-piping languages is the inability for postnominal possessors to pied-pipe (Heck 2008: 89–94). Sentence (19b) demonstrates this for English, and (22) illustrates this for German. Note that although German generally permits possessors to follow the possessed noun (22a), this cannot occur when the possessor is a pied-piping wh-word (22c). Rather, in such constructions, the pied-piping possessor must appear prenominally (22b). (22)

Postnominal Possessors Cannot Pied-Pipe in German (Heck 2008: 89–94) a. die Bilder des Künstlers the painting the.GEN artist.GEN The artist’s paintings. (Heck 2008: 91) b. Ich weiß [ [ wessen Bilder ] du kaufen würdest. I know whose paintings you buy would I know whose paintings you would buy. (Heck 2008: 91) c. * Ich weiß [ [ Bilder wessen ] du kaufen würdest. I know paintings whose you buy would (Heck 2008: 91)

As discussed by Heck (2004, 2008, 2009), similar data are observed in many other European languages, as well as the Mayan language, Tzotzil.15 Now, recall that in these head-initial languages, it is commonly held that postnominal possessors are complements of the possessed N. Consequently, if they were ever to pied-pipe the possessive phrase, they would be pied-piping past a lexical category. Therefore a ban on pied-piping past lexical categories would correctly predict the general impossibility of such structures.16 Our ‘weaker generalization’ can also capture a related, though subtly different, phenomenon in Hungarian. Szabolcsi (1994) argues that possessors in Hungarian can occupy either of two positions inside the DP: one internal to the NP and one external to it. Importantly, however, both these positions are prenominal; Hungarian does not permit postnominal possessors. Nevertheless, the position a possessor occupies can in part be determined by its case: possessors internal to NP bear nominative, while those external to NP bear dative. Interestingly, only dative-marked possessors can pied-pipe.


153

(23) a. [DP Ki-nek a [NP vendégét ] ] ismertétek? who-DAT the guest you.know ‘Whose guest did you know?’ vendégét ] ] ismertétek? b. * [DP [NP Ki who.NOM guest you.know

Given the evidence that the nominative possessor in (23b) is NP-internal, the ill-formedness of (23b) would follow from our Q-based account. Furthermore, our account provides a unified treatment of both (23) and (22), despite the fact that (23) superficially concerns the case of the pied-piper, while (22) concerns its hierarchical position. To put the matter more acutely, both the wh-possessors in (23) are assumed to occupy specifier positions within the fronted phrase. Consequently, any attempt to capture the contrast in (22) via the ‘stronger generalization’ that piedpipers must be specifiers would not straightforwardly capture the parallel data in (23). Thus the behavior of Hungarian wh-possessors supports our ‘weaker generalization’ against the notion that pied-piping is only possible from specifier positions (Horvath 2007a: sect. 2.2). A parallel argument can be made by certain intriguingly similar data from English. As reported by Horvath (2007a), Culicover (1999) observes the following: (24) a. * [ Who solving the problem ] were you thinking about? b. [ Whose solving the problem ] were you thinking about?

As in (23), the contrast between the sentences in (24) seems to lie in the case of the ‘pied-piper’. In the ill-formed (24a), the gerundive subject bears accusative case, while in the well-formed (24b), it bears genitive. As noted by Horvath (2007a), our account of the Hungarian data in (23) could easily be extended to these English data. Let us assume that accusative-marked gerundive subjects in English occupy the NPinternal position of Hungarian nominative possessors. Furthermore, let us assume that genitive-marked gerundive subjects occupy the NP-external position of Hungarian dative possessors (i.e., SpecDP). Under these plausible assumptions, the data in (24) follow from our condition against pied-piping past lexical categories. Finally, we should again make the point that both the wh-words in (24) are left-peripheral specifiers, and so these data would not follow as easily from a constraint against piedpiping from the complement position. We have seen, then, that an inability to pied-pipe past lexical categories predicts an interesting variety of pied-piping phenomena. We may reasonably conclude, then, that our Q-based theory of limited pied-piping is correct in its prediction of such a constraint. The reader may recall, however, that our Q-based account actually yields an even stronger generalization. That is, via our assumption in (17), the impossibility of pied-piping past lexical categories follows from a more general prediction that piedpiping across separate phases should be impossible. Following the logic presented in (18), in a Q/wh-Agreement language, the wh-word cannot be in a separate phase from the root of the pied-piped phrase. If it were, then Agreement between the wh-word and the Q would be impossible, and the derivation would crash.

154

THE GRAMMAR OF Q

This more general prediction might receive some support from an oft-noted property of CP pied-piping in Basque and Ancash Quechua. While both Basque and Ancash Quechua permit pied-piping of subordinate clauses, it has often been noted that such pied-piped clauses must possess a particular form. In both these languages, a subordinate clause S can only be pied-piped if the wh-word is fronted into the left periphery of S (Heck 2008: 106–107). Thus, as we see in (25), neither Basque nor Ancash Quechua permits subordinate clauses to be pied-piped by wh-words internal to the subordinate IPs. (25)

Pied-Piping of Subordinate CPs in Basque and Ancash Quechua Basque: a. (i) [CP Nor1 [IP joango dela t1 ] ]2 esan du Jonek t2? who go AUX said AUX John Who did John say will go? (ii) * [CP [IP Joango dela nor ] ]2 esan du Jonek t2? go AUX who said AUX John b. Ancash Quechua: (i) [CP Imata1 [IP wawa t1 mikuchun ] ]2-taj Maria t2 munan? what child eat Q Maria want What does Maria want the child to eat? (ii) * [CP [IP wawa imata mikuchun ] ]2-taj Maria t2 munan? child what eat Q Maria want

This pattern might also be observable in putative cases of CP pied-piping in English. Structures like the following, common in colloquial English, have been sporadically noted in the syntactic literature, and are often noted to bear a distinct resemblance to the structures in (25) (Kayne 2000; Horvath 2007a).17 (26)

Possible CP Pied-Piping in English a. [CP What’s in there ]1 do you think t1? b. [CP What did he get ]1 does he think t1? c. [CP Where will we go ]1 does she think t1? d. [CP Who saw John ]1 do you think t1?

In light of the pattern in (25), it is interesting to note that such CP pied-piping cannot occur in English if the wh-word is within the subordinate IP. For example, if the whword is not a subject, it must appear fronted in the pied-piped clause. (27)

English CP Pied-Piping Requires Wh-Fronting a. [CP What did he get ]1 does he think t1? b. * [CP He got what ]1 does he think t1?

Furthermore, if the wh-word is a subject, then there can be no overt complementizer in the pied-piped clause (28). Given the well-known ‘Doubly Filled Comp Constraint’, this suggests that the wh-subject in such structures is (string-vacuously) moved to SpecCP.


155

(28) English CP Pied-Piping Disallows Overt Complementizers a. [CP Who [IP saw John ] ]1 do you think t1? b. * [CP That [IP who saw John ] ]1 do you think t1?

Thus the best candidate in English for CP pied-piping appears to also follow the pattern seen in (25): subordinate clauses can only be pied-piped if the wh-word has been moved into their left periphery. If we assume that Basque, Ancash Quechua, and English are all Q/wh-Agreement languages, then the data in (25)–(28) would follow from our Q-based theory of piedpiping. Since C heads are undoubtedly phase heads, each of the ill-formed structures in (25a(ii)), (25b(ii)), (27b), and (28b) would require Q/wh-Agreement to apply across separate spell-out domains, contrary to the PIC. (29) No Q/Wh-Agreement Across Complementizer Heads QP CP

Q

C

Spell-Out Domain

IP

… wh-word ….

XXXXX Agreement Impossible, by Virtue of the PIC

Thus, in order for a Q-particle to Agree with a wh-word contained within a subordinate CP, that wh-word must undergo movement to the specifier of the CP. As shown in (30), such movement would place the wh-word and the Agreeing Q-particle within the same phase.18 (30) Subordinate CP Pied-Piping Is Possible if the Wh-Word Is in SpecCP QP Q

CP Wh-word1 C

CP

IP

Spell-Out Domain

… t1 …. Q/Wh-Agreement Possible

156

THE GRAMMAR OF Q

We find, then, that the special properties of pied-piped CPs in languages like Basque and Ancash Quechua lend support to our broader prediction that pied-piping past phases should be impossible in Q/wh-Agreement languages. As we have seen, when combined with the hypothesis in (17), this broader prediction entails that piedpiping past lexical categories should likewise be impossible in these languages. This latter prediction, in turn, appears to capture a variety of phenomena that have been observed in the literature on pied-piping.

5.3 Further Results Regarding Pied-Piping Thus far, our discussion has centered on the two core properties of limited pied-piping languages: their inability to pied-pipe past islands, and their inability to pied-pipe past lexical categories. In this section I will discuss some further predictions of our Q-based theory. In some cases these predictions are fully general, and are expected to hold for all wh-fronting languages. In other cases we will see that our system makes different predictions for Q/wh-Agreement and non-Agreement languages. 5.3.1 The Generalization on Recursive Pied-Piping In his extensive study of pied-piping, Heck (2008: 76–88) argues that across languages, pied-piping obeys the following generalization. (31)

The Generalization on Recursive Pied-Piping (Heck 2008: 76) If A can pied-pipe B, and B is in a canonical position to pied-pipe C, then A can pied-pipe C.

This property of pied-piping is nicely exemplified by possessor pied-piping in English. As is well known, possessors can pied-pipe possessive DPs in English (32a). Furthermore, if a possessive DP containing a wh-possessor is itself a possessor, then the larger possessive DP can also be pied-piped by the wh-possessor (32b). This process of embedding may, of course, be further iterated, and there seems to be no principled limit on the depth of embedding (32c). (32)

Recursive Pied-Piping in English Possessive Phrases a. [DP Whose father ] did you meet at the party? b. [DP [DP Whose father’s ] friend ] did you meet at the party? c. [DP [DP [DP [DP [DP Whose ] father’s ] friend’s ] uncle ] ] did you meet at the party?

As discussed by Heck (2008), the possibility of such ‘recursion’ can also be seen in PP pied-piping. It is most easily observed in languages where pied-piping of PPs is obligatory, such as German. In German, the complement of P can pied-pipe the PP in which it is contained (33a). Moreover, if a PP containing a wh-complement is itself the complement of a larger PP, then the larger PP may also be pied-piped by the wh-word (33b). Such embedding can be further iterated, and there again seems to be no principled limit on the depth.


157

(33) Recursive Pied-Piping in German PPs a. Fritz fragt sich [CP [PP mit wem ] sie gesprochen hat ]. Fritz asks REFL with whom she spoken has Fritz wonders who she talked to. (Heck 2008: 85) b. Fritz weiß [CP [PP bis [PP zu [DP welchem Punkt ] ] ] er gehen kann ]. Fritz knows until to which point he go can Fritz knows until which point he can go to. (Heck 2008: 85)

Finally, just as we would expect from the general statement in (31), the possibility of recursive pied-piping is not sensitive to the exact position of the pied-piping phrase. That is, possessor DPs can also pied-pipe larger PP complements, as we see in (34). (34) Recursive Pied-Piping of PPs by Possessors Ich frage mich [CP [PP bis [PP zu [DP wessen Geburtstag ] ] ] ich warten muss ]. I ask REFL until to whose birthday I wait must I wonder whose birthday I have to wait until. (Heck 2004: 127)

Let us now confirm that our Q-based theory can predict the generalization in (31). First, recall that within our theory of pied-piping structures, the informal notion that ‘A can pied-pipe B’ amounts more precisely to the claim that ‘A can be dominated by B in the fronted phrase of a pied-piping structure’. With this in mind, we restate the generalization in (31) so that it reads as in (35). (35) Recursive Pied-Piping Within the Q-Based Account

If A can be dominated by B in the fronted phrase of a pied-piping structure, and B can be dominated by C in the fronted phrase of a pied-piping structure, then A can be dominated by C in the fronted phrase of a pied-piping structure. Non-Agreement languages can easily be shown to satisfy (35). First, note that the only time a non-Agreement language disallows X from being dominated by Y in a pied-piping structure is when Y is not a permissible sister to Q. Let us, then, assume that B can be dominated by C in a pied-piping structure, and that B contains A. This is diagramed in the figure in (36). (36) Recursive Pied-Piping in Non-Agreement Languages QP C B A …wh-word…

Q

158

THE GRAMMAR OF Q

If (36) is well-formed, it follows that C is a possible complement to Q. Thus, if A is contained within B, it follows that A can be dominated by C in a pied-piping structure as well. In this way, our Q-based theory predicts that non-Agreement languages will satisfy the generalization in (35). Let us now confirm that we predict Q/wh-Agreement languages to also satisfy (35). As we will presently see, our account derives this prediction from the basic transitivity of ‘accessibility for Agreement’.19 First, note that if a wh-word in A can be dominated by B in a pied-piping structure, then there must be no ‘barrier’ to Q/wh-Agreement within B (i.e., B contains no islands and no lexical projections).

(37)

Recursive Pied-Piping in the Q/wh-Agreement Languages, Part 1 Wh-word at position A can pied-pipe the larger phrase B.

QP B

Q

A … wh-word…

Agreement Possible: Thus, No ‘Barriers’ to Q/Wh-Agreement Within B

Similarly, if a wh-word in B can be dominated by C in the fronted phrase, then there must be no ‘barriers’ to Q/wh-Agreement within C. (38)

Recursive Pied-Piping in the Q/Wh-Agreement Languages, Part 2 Wh-word at position B can pied-pipe the larger phrase C.

QP C

Q

B … wh-word…

Agreement Possible: Thus, No ‘Barriers’ to Q/Wh-Agreement Within C

Finally, since there are no barriers to Q/wh-Agreement within either B or C, no such barriers would separate a wh-word at A from a Q-particle that is sister to C. Therefore it should also be possible for A to be dominated by C within a pied-piping structure.


(39)

159

Recursive Pied-Piping in the Q/Wh-Agreement Languages, Part 3 QP C

Q

B A … wh-word…

No ‘Barriers’ to Q/Wh-Agreement Within B or C: Thus, Agreement Possible

We find, then, that our Q-based account predicts that across languages, piedpiping should obey the generalization in (31)/(35), even in those languages where pied-piping is otherwise subject to rather stringent locality conditions. 5.3.2 Pied-Piping of Auxiliaries and Predicates in Main Clauses and in Subordinate Clauses Our Q-based account makes a number of interrelated predictions regarding the pied-piping of auxiliaries and predicates in main clauses and subordinate clauses. To begin, it is impossible in English to pied-pipe past a matrix copula or auxiliary verb. (40)

No Pied-Piping Past Matrix Copula or Auxiliary in English a. * [ [ Is [DP whose doctor ] ]1 [ Dave t1 ]? b. * [ [ Is [AP proud of whom ] ]1 [ Dave t1 ]? c. * [ [ Is [VP eating what ] ]1 [ Dave t1 ]? d. * [ [ Has [VP seen what ] ]1 [ Dave t1 ]?

Given that English is a limited pied-piping language, the ill-formedness of (40b–d) easily follows from the more general ban on pied-piping past lexical categories. Since the main predicate in (40b–d) is a lexical category (AP or VP), pied-piping past the main predicate (proud, eating, seen) would be an instance of illicit pied-piping past a lexical category. Importantly, though, this reasoning would not obviously apply to (40a). In (40a), the main predicate is not a lexical category, but rather a functional one (DP). Furthermore, the wh-word in (40a) is not dominated by a lexical category within the main predicate. Consequently, pied-piping past the main predicate should, in principle, be possible in (40a). Indeed, (41a) confirms that such pied-piping is possible, just so long as the copula is not also pied-piped. (41) Pied-Piping of Main Predicates in English a. [DP Whose doctor ] is Dave?

160

THE GRAMMAR OF Q

b. * [AP Proud of whom ] is Dave? c. * [VP Eating what ] is Dave? d. * [VP Seen what ] has Dave?

As expected, (41a) contrasts with (41b–d), where the wh-word is dominated by a lexical category within the main predicate of the clause. What, then, accounts for the ill-formedness of (40a) and the contrast between (40a) and (41a)? Interestingly, these facts can be seen to follow from our QPIntervention Condition, repeated in (42). (42) QP-Intervention Condition A QP cannot intervene between a functional head F and a phrase selected by F.

Let us adopt the picture in (43) regarding the syntax of copulas and auxiliary verbs. (43) The Syntax of Sentences Containing Copulas or Auxiliary Verbs

IP DPj Subject

IP I

F1P F1

… FnP Fn

VP V Copula, Auxiliary

VP/AP/DP tj

Main Predicate

That is, as illustrated in (43), we assume that copulas and auxiliary verbs are initially merged as complements to the lowest functional projection in the ‘functional spine’ of the clause. Further, we assume that the main predicate of the clause is merged as a complement to the copula/auxiliary. Finally, we assume that both the copula and the auxiliary verb are truly verbal in nature. That is, they are V heads, and thus instances of a lexical category. With these assumptions in place, it is apparent that the ill-formedness of (40a) follows from the QP-Intervention Condition in (42). Since the copula is pied-piped in (40a), it follows that its projection must be sister to the Q. Consequently, a QP intervenes between the projection of the copula and the functional head (Fn) which selects for it. As illustrated in (44), such a configuration would violate (42).20


161

(44) Pied-Piping of a Matrix Copula or Auxiliary Violates the QP-Intervention Condition CP CP

QP1 VP V Is

Q

CQ

IP

DP

DP Dave

IP

whose doctor

I

… F nP

Impossible FnP, Ruled out by (42)

QP

Fn

t1

Importantly, however, (42) would not rule out the structure in (41a). Since the copula is a lexical category, it should be possible for a QP to intervene between it and the main predicate it selects. Furthermore, unlike (41b–d), the wh-word is not dominated by a lexical category within the fronted phrase, and so Q/wh-Agreement is possible. This reasoning is illustrated in (45). (45) Pied-Piping of a Main Predicate in English CP QP1 DP DP Whose

CP Q

DP

IP

CQ DP Dave

… FnP

doctor Fn No violation of (42)

VP V is

QP t1

It is clear from the logic of the preceding discussion that our Q-based theory of pied-piping predicts the following generalization(s). (46) No Pied-Piping of Highest Verbal Category in Main Clause No language permits pied-piping of the highest verbal category in the main clause: a. If a main clause contains an auxiliary or copula, no language will permit that auxiliary or copula to be pied-piped. b. If a main clause contains no auxiliary or copula, then no language will permit pied-piping of the main predicate.

162

THE GRAMMAR OF Q

The argument surrounding (44) makes clear the prediction in (46a). The generalization in (46b) follows from the natural assumption that in the absence of any copula or auxiliary, the main predicate is the complement of the lowest functional head Fn. That is, as illustrated in (47b), sentences like (47a) also violate the QP-Intervention Condition. (47) No Pied-Piping of Main Predicates in the Absence of Auxiliaries a. * [ [VP Eats what ]1 Dave t1? b. CP QP1 VP V Eats

CP Q

DP what

CQ

IP DP Dave

IP I

… Fn P

Impossible F nP, Ruled out by (42)

Fn

QP t1

Thus far we have seen that the predictions in (46) are true for the Q/wh-Agreement language English. Moreover, it is apparent from the work of Heck (2008) that (46) is also true of many other Q/wh-Agreement languages. Indeed, throughout Heck’s extensive study of pied-piping structures across the world, there is (to my knowledge) no counterexample to the generalizations in (46). However, many of the languages that Heck (2008) examines might be classifiable as Q/wh-Agreement languages. Is there clear evidence that (46) does indeed also hold for non-Agreement languages like Tlingit? In this context, it is useful to again note the ill-formedness of the Tlingit sentences in (48). (48) No Pied-Piping of Main Predicates in Tlingit a. * Daa iyatéen sá? what you.can.see.it Q b. * Aadóo xáat aawaxáa sá? who fish he.ate.it Q c. * Wáa ituwatee sá? how you.feel Q

Given our Q-based analysis of Tlingit wh-questions, each of the structures in (48) contain pied-piping past a main predicate in a clause containing no copula or auxiliary. As already noted in chapter 2, our QP-Intervention Condition predicts the ill-formedness of these structures.


163

Thus we find that Tlingit supports the prediction in (46b) for non-Agreement languages. Support for (46a) comes from the ill-formedness of structures like (49b). (49) No Pied-Piping Past Matrix Copula in Tlingit naax ] sá isitee? a. [DP Daakw aa which PART clan Q you.are Which clan are you? (Dauenhauer & Dauenhauer 2000: 86) b. * [VP Daakw aa naax isitee ] sá which PART clan you.are Q

As shown in (49a), Tlingit permits pied-piping of main predicates when they are complement to its copular verb si-tee, ‘to be’. In such sentences, however, the copular verb can never itself be pied-piped (49b). Tlingit, then, supports the prediction that (46) holds of even the non-Agreement languages.21 In addition to the generalization in (46), our account of the data in (41) entails the general prediction in (50). (50) Pied-Piping of Complements to Auxiliary Verbs and Copulas

In all languages, it is (in principal) possible to pied-pipe a main predicate when it is complement to a copula or auxiliary verb. In Q/wh-Agreement languages, however, such pied-piping can only occur if the wh-word is not dominated by a lexical projection in the main predicate. Again, (50) follows from the fact that pied-piping of a main predicate P will not violate the QP-Intervention Condition when P is complement to a lexical head, such as an auxiliary or copula (45). Is the generalization in (50) accurate? We have already seen that the Tlingit data in (49) support the prediction for non-Agreement languages, and the English data in (41) support the prediction for Q/wh-Agreement languages. Furthermore, the extensive study of Heck (2008) again lends support to this prediction; throughout Heck’s study, there are no (indisputable) counterexamples to (50).22 Thus far in our discussion, we have limited our attention (and our generalizations) to main predicates and auxiliaries. The reason for this is that, as discussed below, our Q-based account makes rather different predictions for subordinate clauses. (51) Pied-Piping of Subordinate Predicate, Auxiliary, and Copula It is possible to pied-pipe a subordinate predicate, auxiliary, and copula, just so long as: a. The entire subordinate clause is pied-piped. b. The subordinate clause is selected by a lexical head. c. In Q/wh-Agreement languages, no lexical projection or phase boundary dominates the wh-word in the pied-piped clause.

The diagram in (52) illustrates how our account makes the prediction in (51).

164

THE GRAMMAR OF Q

(52) Pied-Piping of Subordinate Predicate, Auxiliary, and Copula … CP1 QPj CP2

… Q

LP L

QP

… Wh-Word … tj

As illustrated above, let us assume that (a) an entire subordinate clause CP2 is pied-piped, and (b) CP2 is selected by a lexical head L. Since CP2 is pied-piped, it must be sister to the Q. Thus a QP intervenes between the projection of CP2 and the head L selecting for CP2. By assumption, however, this head L is a lexical head, and so the entire structure satisfies the QP-Intervention Condition. Therefore, if the language in question is a non-Agreement language (like Tlingit), our principles predict that (52) should be well-formed. The accuracy of this prediction is confirmed by the acceptability in Tlingit of CP pied-piping sentences like the following. (53)

Pied-Piping of Subordinate Predicates in Tlingit (cf. (48)) [QP [CP Goodéi wugootx sá ] ]1 has oowajée t1 i shagóonich ? where.to he.went Q they.think your parents.ERG Where do your parents think he went?

Thus, while our account correctly predicts that pied-piping the highest verbal category of a main clause should not be possible ((48)/(49)), it also correctly predicts that it should (in principle) be possible to pied-pipe the highest verbal category of a subordinate clause (53). Of course, if the language in question requires Q/wh-Agreement, then there are further constraints on pied-piping of subordinate predicates and auxiliaries. That is, as stated in (51c), such pied-piping can only take place if the wh-word is not dominated by any lexical projections or phase boundaries in the pied-piped clause. This, of course, follows from the general assumptions regarding Agreement that were presented in section 5.2. As noted in that section, support for the prediction in (51c) might be found in the constraints on clausal pied-piping in languages like Basque and Ancash Quechua. (54)

Pied-Piping of Subordinate CPs in Basque and Ancash Quechua a. Basque: (i) Jonek t2? [CP Nor1 [IP joango dela t1 ] ]2 esan du who go AUX said AUX John Who did John say will go? (ii) * [CP [IP Joango dela nor ] ]2 esan du Jonek t2? go AUX who said AUX John


165

b. Ancash Quechua: (i) [CP Imata1 [IP wawa t1 mikuchun ] ]2-taj Maria t2 munan? what child eat Q Maria want What does Maria want the child to eat? (ii) * [CP [IP wawa imata mikuchun ] ]2-taj Maria t2 munan? child what eat Q Maria want

As illustrated in (54), in these languages—which we might analyze as Q/wh-Agreement languages—it is not possible to pied-pipe a subordinate clause unless the whword moves to SpecCP, a position where it is no longer dominated by any lexical projections or the phase boundary induced by the subordinate C. In summary, we have found that our Q-based account makes a number of interrelated, seemingly accurate predictions regarding the pied-piping of predicates, auxiliaries, and copulas. More specifically, our account predicts some observed asymmetries between the pied-piping of these categories in main clauses and in subordinate clauses. 5.3.3 The Relationship Between Pied-Piping and the Left Edge In section 5.2.3, I argued against accounts that would rule out pied-piping from any complement position, and thereby force pied-pipers to be specifiers. Since the apparent relationship between pied-piping and specifier position is a significant and reoccurring topic in the literature on pied-piping, I will, in this section, sketch out the perspective offered by our Q-based account. Let us begin by noting that our Q-based account employs exactly the same mechanisms to derive pied-piping by specifiers and pied-piping by complements of P. That is, our Q-based analysis of sentences like (55a) appeals to no notions beyond those used in the analysis of cases like (55b). (55) Pied-Piping of DP and PP in Russian knigu ] ty cital ] a. Ja sprosil [ [DP čju I asked whose book you read I asked whose book you read. b. [ v čey mashyne ] priyekhal? in whose car you.arrived Whose car did you arrive in?

Interestingly, such ‘analytic uniformity’ does not hold for most other theories of pied-piping (Sells 1985; Cowper 1987; Kayne 1994; Koopman 2000; Grimshaw 2000; Koopman & Szabolcsi 2000). Under most other accounts, the analysis of structures like (55b) requires mechanisms not needed for the analysis of (55a). For example, Kayne (1994) and Koopman (2000) hold that only specifiers can truly pied-pipe. Consequently structures like (55b) can only be analyzed via appeal to covert movement operations. On the other hand, Grimshaw (2000) proposes that complements of P can pied-pipe because PP is an ‘extended projection’ of D, and so can inherit D’s wh-feature. However, since her system assumes that a phrase can never be an extended

166

THE GRAMMAR OF Q

projection of its specifier, it follows that some other mechanism (i.e., Spec-Head Agreement) must be responsible for pied-piping structures like (55a). While such distinct treatments of (55a) and (55b) might seem inelegant at first, they do hold one potential advantage over the uniform treatment of our Q-based account. Indeed, it is largely by design that these other accounts treat pied-piping by CompPP as a distinct phenomenon from pied-piping by specifiers. Such accounts generally seek to predict that, except for CompPP, all pied-piping must take place from the specifier position. Importantly, this prediction is in fact true, at least for the best-studied languages. As is clear from the cross-linguistic studies of Heck (2008) and Horvath (2007a), in these languages it is generally the case that only P permits pied-piping from its complement. In all cases but pied-piping of PP, a pied-piper must be a specifier. Given this tendency, which Heck (2008, 2009) dubs the “Edge Generalization”, we must therefore question whether our uniform treatment of piedpiping is in fact accurate. Does our Q-based account fail to capture an important property of pied-piping structures? In fact, it does not. Rather, there is, under our Q-based account, a straightforward explanation for this pattern: P happens to be the only functional (non-lexical) category that takes interrogative words as complements. To see this more clearly, let us first consider the class of functional categories, which we might reasonably assume to be the following: C, I, D, Deg, P.23 Now, consider the class of wh-words, which we might reasonably assume to be represented by the following: who, what, which, where, why, how. Placing these two sets side by side, we observe that the only member from the former that can take as complement a member of the latter is the category P. It follows, then, that if a wh-word ever occupies a complement position, and is not complement to P, then it must be complement to some lexical head. Consequently, in the best studied wh-fronting languages, such a wh-word will not be able to pied-pipe. In short, the reason why pied-pipers tend to be specifiers (in the Q/wh-Agreement languages) is not that there is some special importance of the specifier position per se. Rather, this tendency simply follows from the twin facts that (i) these languages only permit pied-piping past functional categories, and (ii) there is only one functional category, P, that takes wh-phrases as complements. Thus, for the Q/wh-Agreement languages, if the wh-word of a pied-piping structure is ever contained within a phrase other than PP, it must be within the specifier of that phrase. For this reason, in the great majority of pied-piping structures (in the Q/wh-Agreement languages), the wh-word will be a left-peripheral specifier. Thus, while our Q-based theory does predict that, in the best-studied languages, piedpipers will almost always be specifiers, it does not derive this tendency from a general ban on pied-piping from complement position. Rather, it simply follows from the independent fact that P is the only functional (non-lexical) head to take wh-complements. 5.3.4 The Optionality of Pied-Piping In this section we will examine the predictions that our Q-based account makes regarding the optionality of pied-piping. As we will see, there are some respects in which our account fares well, but there are also many open challenges.


167

To begin, our Q-based theory predicts that pied-piping can, in principle, be optional. That is, it predicts that pied-piping of a phrase XP by a wh-word W can, in principle, be in free variation with subextraction of W from within XP. To see this, let us first note that nothing within our Q-based theory requires that Q be ‘as close as possible’ to the wh-word it associates with. Suppose that a wh-word is contained within a structure XP, which is in turn contained within a larger structure YP. (56) Optionality of Pied-Piping in the Q-Based Account YP

QP YP

Q

QP XP

Q

…wh-word…

XP

…wh-word…

Moreover, let us suppose that neither XP nor YP contain any barriers to Q/wh-Agreement. Finally, let us suppose that placement of Q as sister to either XP or YP would not violate the QP-Intervention Condition. Under these assumptions, our Q-based theory licenses both the structures in (56). That is, placement of Q as sister to XP is predicted to be as well formed as placement of Q as sister to the larger structure YP. Given our Q-based theory of wh-fronting in (1), the general well-formedness of both structures in (56) entails that pied-piping of a larger phrase YP and subextraction from within YP should be freely alternating options. That is, under our Q-based account, the mere ability to extract an XP from within some constituent YP does not alone entail that pied-piping of YP should not be possible. Piedpiping and subextraction are predicted to, in principle, occur in free variation with one another. This predicted optionality of pied-piping seems to be a property of many attested cases of pied-piping. For example, it is a property of the pied-piping of subordinate CPs, as the data in (57)–(59) illustrate. (57) Free Variation Between Long-Distance Movement and CP-Pied-Piping in Tlingit sá ]1 [ has oowajée [CP t1 wugootx ] i shagóonich ] ]? a. [ [QP Goodéi where.to Q they.think he.went your parents.ERG Where do your parents think he went? b. [QP [CP Goodéi wugootx sá ] ]1 has oowajée t1 i shagóonich? where.to he.went Q they.think your parents.ERG Where do your parents think he went? (58) Free Variation Between Long-Distance Movement and CP-Pied-Piping in Basque a. Se1 pentzate su [CP t1 idatzi rabela Jonek ]? what you.think written has What do you think Jon wrote? (Arregi 2003a: 117)

Jon.ERG

168

THE GRAMMAR OF Q

b. [CP

Se idatzi rabela Jonek]1 what written has Jon.ERG What do you think Jon wrote? (Arregi 2003a: 118)

pentzate su you.think

t1?

(59) Free Variation Between Long-Distance Movement and CP-Pied-Piping in Quechua a. [ Ima-ta ]1-taj Maria-ka [CP Juzi t1 mikushka ]-ta krin? what-ACC-Q Maria-TOP José ate-ACC What does Maria believe that José ate? (Cole & Hermon 1981) b. [CP Ima-ta wawa mikuchun ]-taj Maria kri? what-ACC child ate Q Maria believes What does Maria believe that the child ate? (Cole & Hermon 1981)

believes

This predicted optionality is also a key empirical difference between the Q-based approach advocated here and the theory of pied-piping developed by Heck (2004, 2008, 2009). A core prediction of Heck’s account is the following generalization. (60)

The Repair Generalization (Heck 2008: 117) Pied-piping of YP by a wh-phrase W is possible only if both the following hold: a. Movement of W out of YP is blocked. b. Pied-piping of XP out of YP is blocked, where XP is dominated by YP and dominates W.

This ‘Repair Generalization’ in (60) is in clear opposition to the predictions of our Q-based account, which in principle permits pied-piping of YP even when extraction of W from within YP is possible. Thus (60) is also in conflict with any data that seem to support this prediction of our Q-based account, such as that in (57)–(59). Listed below are some further cases where pied-piping and subextraction of the wh-word seem to be in completely free variation. (61)

P-Stranding Versus PP-Pied-Piping in Icelandic (Heck 2008: 152) a. Ég velti því fyrir mér [ hvern1 thú talaðir [PP við t1 ] ]. I roll it before me who you talked with I wonder who you talked with. b. Ég velti því fyrir mér [ [PP við hvern ] thú talaðir ]. I roll it before me with who you talked I wonder who you talked with.

(62)

P-Stranding Versus PP-Pied-Piping in German (Heck 2004: 184) a. Fritz möchte wissen du [PP t1 mit ] gerechnet hast ]. [ wo1 Fritz wants to.know where you with calculated have Fritz wants to know what you expected (to happen). b. Fritz möchte wissen [ [PP wo-mit ] du gerechnet hast ]. Fritz wants to.know where-with you calculated have Fritz wants to know what you expected (to happen).

(63)

P-Stranding Versus PP-Pied-Piping in Irish (Heck 2004: 301–303)24 a. Cé 1 a raibh tú ag caint [PP t1 leis ]? who C were you at talking with Who were you talking with?


b. [PP Cé leis ] a raibh tú ag who with C were you at Who were you talking with?

169

caint? talking

(64)

DP Splits in German (and Other Germanic Languages) (Heck 2008: 153–154) a. Fritz fragt [ was1 du [DP t1 für Leute ] einlädst ]. Fritz asks what you for people invite Fritz asks what kind of people you invite. was für Leute ] du einlädst ]. b. Fritz fragt [ [DP Fritz asks what for people you invite Fritz asks what kind of people you invite.

(65)

DP-Splits in French (and Other Romance Languages) (Heck 2008: 155) a. Je sais [ combien1 Marie a décidé d’engager [DP t1 de personnes ]. I know how.many Mary has decided to.employ of persons I know how many people Mary has decided to employ. décidé d’engager. b. Je sais [DP combien de personnes ] Marie a I know how.many of persons Mary has decided to.employ How many people has Mary decided to employ?

(66)

DP-Splits in Russian (and Other Slavic Languages) (Heck 2004: 187) a. Ja ty cital [DP t1 knigu ] ] sprosil [ čju1 I asked whose you read book I asked whose book you read. b. Ja sprosil [ [DP čju knigu ] ty cital ] I asked whose book you read I asked whose book you read.

(67)

DP-Splits in Greek (Heck 2008: 155) a. Anarotieme [ tinos1 echis I.wonder whose you.have I wonder whose book you read. b. Anarotieme [ [DP tinos vivlio ] I.wonder whose book I wonder whose book you read.

diavasi [DP t1 read echis you.have

vivlio ] ]. book

diavasi ]. read

(68)

DP-Splits in Mohawk (Heck 2004: 187) a. Ka nikáyΛ1 íhsere’ ahshnínu’ ne [DP t1 ká’sere ]? which you.think you.buy NE car Which car do you want to buy? ahshnínu’? b. [DP Ka nikáyΛ ká’sere ] íhsere’ which car you.think you.buy Which car do you want to buy?

(69)

DegP-Splits in German (and Other Germanic Languages) (Heck 2004: 188) a. Ich frage mich [ [ wieviel Grade ]1 der Ofen [DegP t1 zu heiss ] war ]. I ask myself how.many degrees the oven too hot was I wonder how many degrees too hot the oven was.

170

THE GRAMMAR OF Q

b. Ich frage mich [ [DegP wieviel Grade zu heiss ] der Ofen war ]. I ask myself how.many degrees too hot the oven was I wonder how many degrees to hot the over was.

In each of the sentence pairs in (61)–(69), we find that subextraction of the whword in the (a)-sentence exists side by side with pied-piping of a larger structure in the (b)-sentence. Therefore these data appear to challenge the ‘Repair Generalization’ in (60). We might, then, conclude that these data support our Q-based account over the system developed by Heck (2004, 2008, 2009). Such a conclusion, however, would be premature. First, it should be noted that Heck (2004, 2008) is well aware of these data, and suggests several programmatic ways that they might be approached within his framework (Heck 2008: 273–294). Furthermore, many of these data actually challenge our own Q-based theory of pied-piping. To see this, let us begin by noting that our Q-based account only predicts that it should, in principle, be possible for pied-piping and subextraction to be in free variation. Our account does not predict that pied-piping will always be optional. In fact, there are many circumstances where our Q-based account predicts that subextraction will block pied-piping. This is especially so in the Q/wh-Agreement languages. Suppose, for example, that we are examining a Q/wh-Agreement language, and we observe that extraction of a wh-phrase from within YP is possible. (70)

Extraction of Wh-Phrase from YP in a Q/Wh-Agreement Language CP QP1

…

XP

Q

YP

… wh-word… t1

Given the QP-Intervention Condition, it follows that the sister of Q, XP, must be selected by a lexical head. Consequently, QP is complement to a lexical projection. (71)

Extraction of Wh-Phrase from YP Entails That QP Is Initially Complement to a Lexical Head CP QP1 XP

… Q

YP

… wh-word… LP L

QP t1


171

Now, since QP is complement to a lexical projection LP, it follows that YP dominates LP, which dominates XP. Consequently, if Q were ever sister to YP, then a lexical projection would intervene between the Q and the wh-word within XP. (72)

Pied-Piping of YP

CP …

QP 1 YP

Q

t1

LP L

QP

XP … wh-word … XXX Q/Wh-Agreement Blocked by LP

Therefore, unless the wh-word is somehow able to move to a position in YP above LP, Q/wh-Agreement will not be possible, and pied-piping of YP will be ill-formed. We see, then, that our Q-based account predicts that in some cases (particularly in the Q/wh-Agreement languages), the ability to extract W from YP will block pied-piping of YP. Unfortunately, such cases will include many of the structures in (61)–(69). For example, given the analysis of P-stranding put forth in chapter 4, our Q-based account is challenged by the apparent optionality of PP pied-piping in English and Icelandic. (73)

P-Stranding Versus PP Pied-Piping in Icelandic (Heck 2008: 152) a. Ég velti því fyrir mér [ hvern1 thú talaðir [PP við t1 ] ]. I roll it before me who you talked with I wonder who you talked with. hvern ] thú talaðir ]. b. Ég velti því fyrir mér [ [PP við I roll it before me with who you talked I wonder who you talked with.

(74)

P-Stranding Versus PP Pied-Piping in English a. [PP With whom ] did she leave? b. Who1 did she leave [PP with t1 ]?

Under our analysis in chapter 4, P-stranding is possible in English and (by extension) Icelandic because P in these languages is actually a lexical category. However, since both English and Icelandic are presumably Q/wh-Agreement languages, our theory from section 5.2 predicts that pied-piping past P should be ungrammatical. Clearly, however, such structures seem to be possible in these languages.

172

THE GRAMMAR OF Q

Another problematic case of optionality is the DP splits in (66)–(68). The Russian example in (66) is repeated in (75). (75)

DP-Splits in Russian (and Other Slavic Languages) (Heck 2004: 187) a. Ja sprosil [ čju1 ty cital [ t1 knigu ] ] I asked whose you read book I asked whose book you read. b. Ja sprosil [ [ čju knigu ] ty cital ] I asked whose book you read I asked whose book you read.

In chapter 4 we captured the possibility of (75a) by assuming that possessors in Russian are simply adjuncts to NP, as illustrated in (76). (76) Structure of Possessive Nominals in Slavic Languages

NP Possessor

NP

N Possessum

However, if wh-possessors in Russian are always contained within NPs, then the possibility of possessor pied-piping in (75b) is unexpected. Such structures would require pied-piping past a lexical category (NP), and so should not be possible in a Q/wh-Agreement language like Russian. There are two imaginable ways in which our Q-based account can approach these challenges. The first is to posit that the problematic pied-piping structures are instances of so-called massive pied-piping (Heck 2008). The phenomenon of ‘massive pied-piping’ will be treated in more detail in section 5.5. In brief, ‘massive piedpiping’ refers to the ability for Q/wh-Agreement languages to tolerate pied-piping past lexical categories in nonembedded clauses. For example, pied-piping past NP is noticeably more tolerable in matrix questions (77a) than it is in embedded questions (77b). (77) Massive Pied-Piping in English a. ?? [ Pictures of whom ] did John buy? b. * I asked [ [ pictures of whom ] John bought.

Of course, such an analysis would clearly not work for the pied-piping structures in (73b) and (75b), which do appear in embedded clauses. However, it might be the right analysis for PP pied-piping in English (74a). As noted by Heck (2008) inter alia, pied-piping of PPs in English is noticeably worse in embedded clauses.


173

(78) Pied-Piping of PPs in English Embedded Clauses (Heck 2008: 123) * Egbert wonders [ [PP with whom ] she left ].

Consequently, Heck (2008: 124) proposes that cases like (74a) in English are likely to be instances of ‘massive pied-piping’.25 Since massive pied-piping by definition permits pied-piping past lexical categories (in Q/wh-Agreement languages), we find that the data in (74a) are not necessarily inconsistent with our view that P in English is a lexical category. But what of the data in (73b) and (75b)? Another approach to pursue might be to posit structural ambiguity. For the Russian data in (75), we might propose that Russian possessive nominals can have either the structure in (76) or a classic ‘DP-structure’ as in English. When the possessor is a mere adjunct to NP, extraction of the possessor occurs, as in (75a). However, when the possessor occupies SpecDP, pied-piping of the entire possessive phrase occurs, as in (75b).26 A similar ‘ambiguity approach’ could be applied to the Icelandic data in (73). Let us suppose that in Icelandic, prepositions have an ambiguous status and can be categorized as either functional or lexical heads. When P enters the derivation as a lexical head, P-stranding as in (73a) is derived. However, when P enters the derivation as a functional head, P-stranding is prevented, and PP pied-piping is required (73b). Of course, these proposals are all merely programmatic suggestions, and it is far from clear whether they can ultimately be made to work. We find, then, that while our Q-based account predicts that pied-piping should sometimes be optional, our account actually has some difficulty capturing many attested examples of optional pied-piping. Given these considerations, I conclude that it is far from clear which of the two accounts under consideration—Heck’s (2008) account or our own—fare best with respect to the optionality of pied-piping. While the data in (61)–(69) challenge Heck’s (2008) prediction that ‘true’ optionality should never occur, our own account does not straightforwardly capture those data either.27 As one final note, let us observe that our Q-based account is not entirely inconsistent with the ‘Repair Generalization’ in (60). That is, while our account does not on its own predict (60), certain assumptions could be added in order to derive it. As noted by Heck (2008: 52–53), one could imagine a Q-based account in which some principle requires that Q be merged as close as possible to the wh-word.28 Clearly, within such an account, the two structures in (56) would not be allowed to coexist, and so pied-piping would never be in free variation with subextraction. Thus I leave to future research the question of whether Heck’s ‘Repair Generalization’ should be captured under a Q-based account, and if not, how best to capture the substantial evidence that Heck marshals in support of it. 5.3.5 Pied-Piping Is Subject to the Coordinate Structure Constraint Our account is able to capture an interesting interaction between pied-piping and coordination. As first noted by Postal (1972), a single wh-word inside a conjunct cannot (in English) pied-pipe the entire coordinate structure.

174

THE GRAMMAR OF Q

(79)

No-Pied-Piping of Conjunction by a Single Wh-Word (in English)29 a. (?)[ Bill and who ] did you meet? b. ?? [ Who and Bill ] did you meet?

(80)

a. (?) [ [ John’s books ] and [ whose paintings ] ] did you sell? b. ?? [ [ whose paintings ] and [ John’s books ] ] did you sell?

Interestingly, however, if both conjuncts contain wh-words, then the two wh-words together can pied-pipe the entire coordinate structure. (81)

Pied-Piping of Conjunction Is Possible if Both Conjuncts Contain Wh-Words a. [ Where and when ] did you see him? b. [ [ Whose books ] and [ whose paintings ] ] did you sell?

As noted by many, these facts bear a striking resemblance to the Coordinate Structure Constraint (CSC), which prevents movement from targeting a single constituent within a coordinate structure. (82)

Illustration of the Coordinate Structure Constraint a. * Who1 did you [ [ meet Bob ] and [ thank t1 ] ]? b. Who1 did you [ [ meet t1 ] and [ thank t1 ] ]?

Naturally, then, previous studies have sought to reduce the facts in (79)–(81) to the CSC. Under an account that assumes ‘feature percolation’, such reduction is straightforward: simply assume that ‘feature percolation’ is subject to the CSC (Heck 2008: 310).30 In a system like ours, however, which does not employ feature percolation, how are these facts to be captured? One possibility might be to posit that the CSC is ultimately about Agreement rather than movement.31 Consider the following constraint. (83)

The Coordinate Structure Constraint for Agreement If an Agreement relation holds between A and B, and B is contained within conjunct D of a coordinate structure “C and D”, then an Agreement relation also simultaneously holds between A and some element within C.

First, note that, under the assumption that all movement is a reflex of Agreement, the condition in (83) is enough to capture the data in (82). In brief, (82a) is impossible because it would require the matrix interrogative C head to bear an Agree relation only with an element of the second conjunct, and not also the first. However, (82b) is possible because in this structure the interrogative C head bears an Agree relation with elements within both conjuncts. Now let us see how the condition in (83) might account for the data in (79)–(81). Since English is a Q/wh-Agreement language, Agreement must hold between the Q and the wh-word it associates with. Moreover, under our Q-based account, the Q-particle associated with a wh-word always dominates the fronted phrase of the wh-question. As illustrated in (84), it therefore follows that (79) and (80) will violate (83).


175

(84) No-Pied-Piping of Conjunction by a Single Wh-Word (in English) QP ConjP

DP1

Q

and

DP2

John’s books whose paintings

XXXXX AGREEMENT-BASED CSC (83) VIOLATED Agreement between Q and member of DP2, but No Agreement between Q and a member of DP1

According to this analysis, the problem with the structures in (79) and (80) is that the Q-particle sitting above the pied-piped coordination only Agrees into one of the two conjuncts. It follows, then, that if a wh-word is added to the other conjunct of these structures, then they should be well-formed. (85) Pied-Piping of Conjunction Is Possible if Both Conjuncts Contain Wh-Words QP ConjP

DP1

and

Whose books

Q

DP2

whose paintings

AGREEMENT-BASED CSC (83) RESPECTED Agreement between Q and member of DP2, and Between Q and a member of DP1

Thus we find that, when combined with the statement of the CSC in (83), our analysis accurately predicts the contrast between (79)/(80) and (81). Furthermore, this account successfully derives these data from the same condition responsible for the similar, classic CSC facts in (82). Finally, let us consider what our Q-based account predicts for non-Agreement languages like Japanese. Clearly, since our account ties the impossibility of configurations like (84) to the necessity of Agreement between the Q and the wh-word, it follows that such configurations should be licit in non-Agreement languages. It has indeed been noted that wh-questions in Japanese seem to violate the CSC (Cheung

176

THE GRAMMAR OF Q

2003).32 The wh-operator of a Japanese wh-question can be located within a coordination, without the other conjunct of the coordination containing a wh-word. (86)

Apparent Violations of the CSC in Japanese Taro-wa [ niku to nani ]-o katta ka? Taro-TOP meat and what-ACC buy Q What is the thing x such that Taro bought meat and x?

Assuming that the Q-particle ka in (86) is initially merged as an adjunct to the entire coordinate structure, we find that our analysis of (79)–(81) provides a straightforward account of these facts. (87)

Base Structure of the Apparent Violations of the CSC in Japanese Con jP Con jP

DP 1 niku

to

Q ka DP2 nani

Under this analysis, the Agreement between the interrogative C head and the Q-particle ka does not violate the CSC, because the Q-particle is not contained within a conjunct of the coordinate structure. Moreover, no violation of the CSC is incurred by the presence of only a single wh-word within the coordinate structure, because there is no Agreement between Q and the wh-word in Japanese. We find, then, that when combined with an Agreement-based statement of the CSC (83), our Q-based theory of pied-piping can account for the interesting contrast between the English data in (79)–(82) and the Japanese data in (86).

5.4 Secondary Wh-Fronting This section examines cases where a wh-word undergoes movement within the fronted phrase of a pied-piping structure, a phenomenon dubbed ‘secondary wh-fronting’ by Heck (2008). I first present the proposed Q-based theory of secondary wh-fronting, illustrating the account with examples from English. Following this, I argue that possessor pied-piping and preposition inversion in the Mayan languages Chol and Tzotzil provide interesting evidence in support of our Q-based account. 5.4.1 The Basic Nature of Secondary Wh-Fronting In the preceding sections we examined the locality restrictions that (parametrically) govern the distance between Q and the wh-word it associates with. We have


177

seen that these locality restrictions often prevent a wh-word W occupying a position A inside a larger phrase B from pied-piping B. Interestingly, many languages possess a clever ‘technique’ for nevertheless allowing W to pied-pipe B. In these languages, W may be moved from its base position A to a higher position inside B, from which it can pied-pipe B without violating the locality restrictions on pied-piping. For example, the ill-formed structures in (88a) and (89a) become well-formed if the pied-piper is located at the left edge of the fronted phrase. (88) Secondary Wh-Fronting in English a. (?) [ Pictures of who ] did John buy? b. [ Whose1 pictures t1 ] did John buy? (89) a. * [ A [ how big ] boat ] did John buy? b. [ [ How big ]1 a t1 boat ] did John buy?

Following Heck (2004, 2008), I will use the term ‘secondary wh-fronting’ to describe cases where a wh-word has undergone movement inside a pied-piping structure. In this section we will develop a Q-based account of this phenomenon, one in which the ‘engine’ driving secondary wh-fronting is the need for the wh-word to be in a position where it is visible for Agreement with the Q-particle. For example, we already possess within our account the means for deriving the pattern in (88). We have already seen in section 5.2.2 that the ill-formedness of (88a) follows from the impossibility of Q/wh-Agreement when the wh-word is complement to NP (90a). Moreover, we also saw that the well-formedness of (88b) follows from the possibility of Q/wh-Agreement when the wh-word occupies SpecDP (90b). (90) The Nature of Secondary Wh-Fronting in Possessive DPs a. No Pied-Piping Without Possessor Raising

QP DP D

Q NP

N Pictures

PP of who

XXXXX No Q/Wh-Agreem ent Across a Lexical Projection

178

THE GRAMMAR OF Q

b. Pied-Piping Possible With Possessor Raising QP DP

Q

DP1

DP

Whose

D

NP N pictures

t1

No Lexical Projection Intervenes Between Q and WH: Q/Wh-Agreement Is Possible

Thus, in this sense, the need for Q in English to Agree with a wh-word ‘drives’ the secondary wh-fronting of possessors. In a similar way, the obligatory ‘DegP inversion’ exemplified in (89) could result from the need to position the DegP so that it may Agree with its c-commanding Q-particle. The ill-formedness of sentences like (89a) would follow from the whword being buried within the modified NP, as illustrated in (91a). However, as shown in (91b), the secondary fronting of the wh-word in (89b) would put the wh-word in a position where it is visible for Agreement with Q. (91)

The Nature of Secondary Wh-Fronting in Degree Phrases a. No Pied-Piping Without DegP Inversion QP DP

Q

D a

NP DegP

Deg how

NP AP

boat

big XXXXXXXXXX No Q/Wh-Agreement Across a Lexical Projection

b. Pied-Piping Possible With DegP Inversion QP DP

Q DP

DegP1 Deg How

AP big

D a

NP t1

NP boat



179

It seems, then, that our Q-based theory could view ‘secondary wh-fronting’ as a means for facilitating Agreement between the Q-particle and the wh-word it c-commands. Note, however, that there is a certain lacuna in this proposed account. While the account would predict that wh-words internal to lexical projections must move in order to pied-pipe, we have not yet accounted for the possibility of such movement. Now, in certain cases this gap is innocuous, since the phrase-internal movement is an independent, freely available option in the language. For example, the general ability for possessors in English to move to SpecDP accounts for the possibility of the secondary wh-fronting in (88b) and (90b). In other cases, however, this gap is a bit more problematic, particularly when the phrase-internal movement undergone by the piedpiper is not a freely available option. In fact, in some cases of secondary wh-fronting, the phrase-internal movement of the pied-piper can only occur with wh-operators. The secondary wh-fronting in (89b) and (91b) is a potential example of this. Although such ‘DegP inversion’ is obligatory when the Deg head is a wh-operator, it is impossible in most other cases.33 (92) Impossibility of DegP Inversion With Non-Wh Deg a. I ate [DP a [DegP very nice ] meal ]. b. * I ate [DP [DegP very nice ]1 [DP a t1 meal ] ].

Our account of the data in (89b) is therefore somewhat incomplete, in that it lacks any explanation of how the ‘rescuing’ movement in (91b) is triggered. Let us, then, begin to fill this lacuna by developing a general approach to those cases where the movement seen in secondary wh-fronting is only allowable for whoperators. We will take English DegP inversion as our paradigmatic example. Since the fronting of the DegP in English is tied to it being a wh-operator, it would be natural to view this movement as triggered by the wh-feature. Let us, then, suppose that English possesses a special indefinite article awh, which bears an unvalued instance of the feature Wh. (93) The Wh-Indefinite Awh in English { Def[-] , Num[sg] } { Def[-] , Num[sg] , Wh[ ] }

= =

a awh

Given our theory of feature valuation under Agree (section 5.2.1), when the determiner awh is merged to its NP complement, it must Agree with something that bears a valued instance of Wh. Assuming that wh-words all bear valued instances of Wh, this Agreement can take place when the complement of awh contains the Deg head how. Finally, as a reflex of the Agreement between awh and the DegP headed by how, the DegP moves into the specifier of the DP. The following structure illustrates.

180

THE GRAMMAR OF Q

(94) DegP Inversion Triggered by Merger of Awh DP DP

DegP 1 Deg How

AP big

D awh

NP

Agree t1

NP boat

Movement Triggered by Agreement Between ‘Awh’ and the DegP Headed by ‘How’

Clearly, from the logic of this proposal, it follows that any DP-internal fronting of DegP can only occur if the DegP is headed by a wh-operator.34 Thus our account correctly predicts the impossibility of (92b). Our discussion here nicely illuminates a rather subtle property of our Q-based theory of secondary wh-fronting, one that warrants some spotlighting. Under the account proposed above, the proximal cause of DegP inversion is an Agreement relation holding between the fronted DegP and a special D-head bearing unvalued Wh. Nevertheless, the necessity of DegP inversion in English ‘degree questions’ is due to the impossibility of Q/wh-Agreement without it. We find, then, that while our theory does propose that secondary wh-fronting is a consequence of obligatory Q/wh-Agreement, it is only an indirect consequence. That is, secondary wh-fronting is not directly triggered by Q/wh-Agreement under our account, but is rather the structural precondition for such Agreement. This property also, of course, holds for those cases of secondary wh-fronting where the phrase-internal fronting is a free, independently available option in the language. Even though the phrase-internal fronting in these other cases is not triggered by the wh-feature, it is locally triggered by some feature of the wh-word. For example, in the case of possessor fronting in English, we might suppose that the fronting of the possessor to SpecDP is due to phi-Agreement between the possessive D-head and the possessor. In the following two subsections we will present some evidence in support of this Q-based approach to secondary wh-fronting. Before we come to these arguments, however, let us briefly note an interesting consequence of our proposals concerning DegP inversion. Recall that our analysis derives the possibility of DegP inversion from the existence of a special indefinite article awh, which is lexically specified as bearing unvalued Wh. Our account therefore provides an explanation of the impossibility of DegP inversion within other indefinite DPs. Note the following contrast between (95a) and (95b,c). (95)

Secondary DegP-Fronting Crucially Tied to Presence of Awh a. [ How big [ a boat ] ] did John buy? b. * [ How big [ some boat ] ] did John buy?


181

c. * [ How big [ three boats ] ] did Jon buy?

The impossibility of (95b,c) would follow from the absence of putative somewh and threewh in English. Interestingly, it is unclear how these facts could follow from any semantic or syntactic difference between the normal indefinite article a and the indefinite determiners some and three. Thus it seems correct to tie the possibility of DegP inversion to an idiosyncratic, unpredictable lexical property of the D-head rather than view it as a productively driven reflex of pied-piping (cf. Heck 2004, 2008). 5.4.2 Supporting Evidence from Possessor Pied-Piping in Mayan A unique property of our theory of secondary wh-fronting is that the movement of the wh-word within the pied-piped phrase is a different type of movement from the movement of the entire pied-piped phrase. That is, according to our Q-based account, the movement of the pied-piped phrase is always an instance of ‘Q-movement’, movement triggered by the Q-feature of the Q-particle. In contrast, secondary wh-fronting is directly triggered by some (other) feature of the wh-word, and so is not an instance of ‘Q-movement’.35 Interestingly, however, the prevailing view regarding secondary wh-fronting is that both the phraseinternal fronting of the wh-word and the fronting of the pied-piped phrase are movements of the same type; both are directly triggered by the wh-feature of the wh-word itself (Heck 2004, 2008). As first observed by Coon (2009), the Mayan languages Chol and Tzotzil provide some evidence that our Q-based account is correct to distinguish the movement seen in secondary wh-fronting from the movement of the pied-piped phrase. Following Coon (2009), all the data below are taken from Chol; the entirely parallel Tzotzil data (also discussed by Coon) can be found in Aissen (1996). Wh-questions in Chol exhibit secondary wh-fronting whenever a wh-possessor pied-pipes the possessive DP. In such wh-questions, the wh-word must front to a prenominal position (97), even though such a position is not otherwise permitted for Chol possessors (96). (96) Regular Possessors in Chol Must Be Postnominal a. Tyi puli [ iyotyoty aj-Maria ] burned house Maria Maria’s house burned. b. * Tyi puli [ aj-Maria1 [ iyotyoty t1 ] ] burned Maria house (97) Pied-Piping Wh-Possessors in Chol Must Be Prenominal a. [ Maxki1 [ iyotyoty t1 ] ] tyi puli? who house burned Whose house burned? b. * [ Iyotyoty maxki ] tyi puli? house who burned

182

THE GRAMMAR OF Q

Furthermore, when a wh-possessor is recursively embedded within another possessor, Chol requires that the wh-possessor appear at the left periphery of the entire complex possessive DP. From this position, the wh-possessor can then pied-pipe the entire DP. (98) Pied-Piping of DP by Possessors of Possessors in Chol a. [ Maxki1 [ ijol [ iyotyoty t1 ] ] ] tyi puli? who roof house burned Whose house’s roof burned? b. * [ Ijol [ iyotyoty maxki ] ] tyi puli? roof house who burned c. * [ Ijol [ maxki1 [ iyotyoty t1 ] ] ] tyi puli? roof who house burned

Crucially, as pointed out by both Coon (2009) and Aissen (1996), in examples like (98a), there can be no pied-piping by the wh-possessor strictly within the larger pied-piped possessive DP. That is, movement of the wh-word maxki, ‘who’, to the left periphery of the possessive DP cannot pied-pipe the embedded NP headed by iyotyoty, ‘house’. Such pied-piping would produce the ‘roll-up’ structure in (99), which is judged in both languages to be ill-formed. (99)

No Pied-Piping of Embedded Possessive DP by Embedded Possessor * [ [ Maxki1 [ iyotyoty t1 ] ]2 ijol t2 ] tyi puli? who house roof burned DP DP D

NP N ijol

DP DP D

NP N iyotyoty

DP maxki

XXXXXXXXXXXXX IMPOSSIBLE PIED-PIPING

As first observed by Coon (2009), the contrast between (98a) and (99) is problematic for accounts that treat both secondary wh-fronting and the fronting of pied-piped phrases as instances of ‘wh-movement’. After all, it is clear from sentences like (97a) and (98a) that wh-fronting in these languages generally allows wh-possessors to pied-pipe possessive DPs. Why, then, can such pied-piping not occur when wh-fronting occurs internal to


183

a larger, pied-piped DP? As Coon notes, there is no straightforward answer within the prevailing theories of pied-piping, which can only rule out the roll-up structure in (99) by external stipulation.36 On the other hand, Coon (2009) argues that the Q-based theory of wh-fronting advocated here predicts the impossibility of (99). Abstracting away from Coon’s own particular implementation of the Q-based theory for Chol, let us assume that the secondary wh-fronting found with Mayan wh-possessors is, like English DegP inversion, an instance of ‘real’ wh-movement, triggered by unvalued Wh[ ] on a higher D-head. Following our general theory of secondary wh-fronting, the appearance of this Wh-bearing D-head is required in order to put the wh-possessor in a position where it can Agree with the Q-particle sitting above the possessive DP. As we see in (100), if the Wh-D does not appear within a possessive DP, then the wh-possessor will remain in a position where Q/wh-Agreement cannot occur. (100)

Secondary Wh-Fronting of Possessors in Chol (and Tzotzil) No Wh-D = No Possessor Fronting = No Q/Wh-Agreement = Ungrammaticality QP DP

Q

D

NP N Iyotyoty

DP maxki

XXXXX No Q/Wh-Agreement Across a Lexical Projection

On the other hand, the presence of Wh-D within a possessive DP entails that the wh-possessor must front to SpecDP, which puts the wh-word in a position where Q/wh-Agreement is possible. (101)

Secondary Wh-Fronting of Possessors in Chol (and Tzotzil) Presence of Wh-D = Fronting of Possessor = Q/Wh-Agreem ent = Convergence QP DP

Q DP

DP 1 Maxki D Wh[ ]

N iyotyoty

NP t1

Wh-movement triggered by D Wh[ ] No Lexical Projection In tervenes between Q and WH: Q/Wh-Agreement Is Possible

184

THE GRAMMAR OF Q

Under this analysis, structures like (98a), with ‘long-distance’ possessor fronting, are derived via the insertion of multiple instances of DWh[ ]. Such long-distance fronting of the wh-possessor is likewise required in order put it in a position where no lexical projections intervene between it and the Q-particle. (102)

Obligatory Long-Distance Possessor-Fronting in Chol (and Tzotzil) QP DP

Q

DP 1 Maxki

DP NP

D Wh[ ] N ijol

DP DP

t1 SuccessiveCyclic WhMovement

DWh[ ]

NP

N iyotyoty

t1

No Lexical Projection In tervenes Between Q and WH: Q/Wh-Agreement Is Possible

Importantly, however, our Q-based account is unable to derive roll-up structures like (99). Recall that under our Q-based theory, there are no real instances of pied-piping. Whenever a phrase larger than the projection of the wh-word is fronted, such fronting is ipso facto not triggered by the wh-feature of the wh-word, but rather by the Q-features of a c-commanding Q-particle. Thus the only conceivable means by which the roll-up structure in (99) could be created is by QP-movement internal to the pied-piped DP. (103)

Roll-Up Structures Like (99) Within the Q-Based Account QP3 DP

Q3

QP 2 DP DP 1 Maxki

DP Q2

DP

iyotyoty

D

NP N ijol

t1

Wh-Movement to Spec DP

QP-Movement to Spec DP

t2


185

Such a structure, however, suffers from a critical problem: it is uninterpretable. The issue is that the lower Q-particle inside the DP closes off the focus-alternatives projected by the wh-word maxki, ‘whose’, preventing them from being passed up to the higher Q-particle outside the DP. Thus the higher Q-particle never receives an argument of the appropriate type, and a ‘semantic crash’ results. This observation can be immediately generalized to a broader explanation of the impossibility of the roll-up structure in (99). Under our Q-based theory, it will never be possible for a wh-word undergoing ‘secondary wh-fronting’ to pied-pipe subconstituents of the larger phrase it is pied-piping. More precisely, it will never be possible for the pied-piper W of a phrase XP to undergo movement internal to XP where it pied-pipes a subconstituent YP of XP. That is, any configuration of the form in (104) is predicted by our Q-based theory to be impossible.37 (104)

General Impossibility of Pied-Piping in Secondary Wh-Fronting QP1 XP

Q1

Higher Q-Particle Never Receives Focus-Alternatives

ZP

QP2 YP

ZP

Q2

t2

W wh-word Pied-piped YP Internal to pied-piped XP

As with the particular structure in (103), the problem with (104) is semantic. If W pied-pipes XP, then XP must be sister to a Q-particle Q1 which takes as argument the focus-alternatives that are ultimately contributed by W. Moreover, if W also piedpipes YP, then YP must also be sister to a Q-particle Q2 which takes as argument the focus-alternatives that are contributed by W. However, given that Q2 is located inside XP, it follows that the focus-alternative contributed by W will not be able to ‘project’ up to XP, and thus Q1 will not receive an argument of the appropriate semantic type. A semantic crash results, and the structure is predicted to be deviant. We find, then, that our Q-based theory rightly predicts the impossibility of the ‘roll-up’ structure in (99). More generally, it predicts that secondary wh-fronting— wh-fronting of a pied-piper internal to the pied-piped constituent—should never itself exhibit pied-piping. Importantly, these predictions are inextricably linked to the

186

THE GRAMMAR OF Q

distinction our theory makes between the type of movement seen in secondary whfronting (real wh-movement) and the type undergone by the pied-piped phrase (Q-movement). 5.4.3 Supporting Evidence from P(reposition)-Inversion in Mayan Our Q-based theory of secondary wh-fronting is able to capture a curious pattern of P(reposition)-inversion in Chol and Tzotzil. As before, all the following data are taken from Chol.38 The entirely parallel Tzotzil data can be found in Heck (2008: 90) and Aissen (1996). As in many languages, prepositions in Chol cannot be stranded. Instead, whenever the complement of P is a wh-word, the entire PP is pied-piped into the left periphery. Interestingly, in those cases where the complement of P contains a wh-possessor, the wh-possessor must be inverted to a position before the P, as in (105a). (105)

P-Inversion With Wh-Possessors in Chol a. [PP Maxki1 [PP tyi [NP iyotyoty t1 ] ] ] tyi majliyety? whose to house you.went Whose house did you go to? b. * [PP Tyi [NP iyotyoty maxki ] ] tyi majliyety? to house whose you.went c. * [PP Tyi [NP maxki iyotyoty ] ] tyi majliyety? to whose house you.went

Moreover, this P-inversion can only target the wh-possessor of the possessive DP. The entire possessive DP can never itself undergo movement to the left of the preposition. (106)

P-Inversion Does Not Apply to the Entire Complement of P * [PP [NP Maxki iyotyoty ]1 [PP tyi t1 ] ] tyi majliyety? whose house to you.went

Besides being an intrinsically interesting pattern, the Mayan data in (105) and (106) raise the following challenge to theories of secondary wh-fronting. Although this pattern is fully general for both Chol and Tzotzil, it is notably absent from other, more familiar wh-fronting languages. For example, although Russian permits both possessor extraction (66a) and PP pied-piping (20b), it does not require, or even allow, the kind of P-inversion seen in (105a). (107)

No P-Inversion With Wh-Possessors in Russian39 a. [PP V [DP čjej mashyne ] ] priexal? in whose car you.arrived Whose car did you arrive in? b. * [PP Čjej1 [PP v [DP t1 mashyne ] ] ] priexal? whose in car you.arrived


187

Thus a full account of the Mayan P-inversion seen in (105a) should also predict the absence of such secondary wh-fronting in languages like Russian. Interestingly, our Q-based theory of secondary wh-fronting provides an analysis of the Mayan pattern in (105) that successfully ties its P-inversion to an independently visible property of these languages, one that, moreover, distinguishes them from Russian and most other wh-fronting languages. A striking property of both Chol and Tzotzil is that neither permits P to take full DPs as complements. Rather, as described by Jessica Coon (personal communication), Ps in these languages must take bare NP complements.40 (108)

Prepositions in Chol Must Take Bare NPs as Complements a. Tsajñoñ ila [ tyi [NP otyoty ] ] I.went here to house I went to this house. (Literally: “I went to house here”) b. * Tsajñoñ [ tyi [DP ili otyoty ] ] I.went to this house

(109)

a. Tyi kwuts’u pisil ix [ tyi [NP ja’ ] ] I.washed clothes there in river I washed my clothes in this river. (Literally: “I washed my clothes in river there.”) b. * Tyi kwuts’u pisil [ tyi [DP ixä ja’ ] ] I. washed clothes in that river

As shown in (108) and (109), the Chol preposition tyi, ‘to’, can only take as complement bare NPs like otyoty, ‘house’, and ja’, ‘river’. If tyi, ‘to’, takes as complement an unambiguous instance of a Chol DP, such as ili otyoty, ‘this house’, or ixä ja’, ‘that river’, then the resulting structure is ill-formed. This inability to take full DP complements distinguishes the Ps of Chol and Tzotzil from the Ps of Russian and most other languages. Structures like those in (110) demonstrate that the prepositions of Russian can take full DPs as complements.41 (110)

Prepositions in Russian Can Take Full DPs as Complements ètot magazin ] ] a. I zashel [ v [DP I entered in this store I entered into this store. b. I zashel [ v [DP tot magazin ] ] I entered in that store I entered into that store.

As we will see, our Q-based account can derive from this basic difference between Mayan and Russian prepositions the contrast seen between (105) and (107). To see this, let us first examine how our Q-based theory would analyze the P-inversion of Chol and Tzotzil. To begin, let us consider the potential consequences of the fact that Chol and Tzotzil P cannot take full DP complements. From this fact, it follows that possessive nominals that

188

THE GRAMMAR OF Q

are complement to P must be bare NPs, and cannot have the DP functional projections of other possessive nominals. (111)

The Structure of Possessive Complements to P in Chol PP P tyi

NP N possessum

DP possessor

Given our theory of Q/wh-Agreement, a Q particle taking PP as complement in Chol will be unable to Agree with a wh-possessor that stays in situ inside the complement of the N. Moreover, as we saw in the previous subsection, Chol typically facilitates Agreement between Q and wh-possessors by fronting the wh-possessor to SpecDP, removing it from the NP projected by the possessum. Unfortunately, when the possessive nominal is complement to P, there is no DP projection into which the whpossessor can move. So, how can Q/wh-Agreement be achieved when a wh-possessor is contained within a complement of P? Note that if the wh-possessor were fronted into the PP-projection itself, then the wh-word would be accessible for Agreement with Q. Let us, then, suppose that prepositions in Chol can optionally bear an instance of unvalued Wh. Following our general theory of secondary wh-fronting, a P bearing Wh[ ] in Chol will trigger wh-movement of the wh-possessor into SpecPP. The crucial result of this wh-movement is that the wh-possessor now occupies a position where no lexical projections intervene between it and Q, and thus Q/wh-Agreement can occur. (112)

Obligatory P-Inversion With Wh-Possessors in Chol QP PP DP1 Maxki

Q PP NP

PWh[ ] tyi N iyotyoty

t1

Secondary Wh-Fronting


We find, then, that our Q-based account can derive the necessity of P-inversion with Chol/Tzotzil wh-possessors from the inability for P in these languages to take


189

full DP complements. Furthermore, our system correctly predicts that this P-inversion will never target the entire possessive NP itself (106). Structures like (106) would necessarily be instances of ‘roll-up pied-piping’, as in (103), which we have shown to be ruled out by our account. Because the phrase moved to SpecPP in (106) is not a projection of the wh-word, it follows that such fronting is not wh-movement, but rather Q-movement. Thus such a structure would necessarily involve the interposition of a Q between the wh-possessor and the Q external to the pied-piped PP. Consequently, (106) is predicted to be uninterpretable. Let us now examine the predictions our account makes for PP pied-piping in Russian. As demonstrated by (110), it is possible for Ps in Russian to take full DPs as complements. Thus, as the structure in (113) illustrates, it should be possible for wh-possessors in Russian to pied-pipe a larger PP without having to move to the Spec of PP.42 (113)

P-Inversion Is Not Necessary for PP-Pied-Piping by Wh-Possessors in Russian QP PP P v

Q DP

DP

NP

ey

mashyne


Interestingly, because Mayan style P-inversion is not necessary in Russian structures like (113), we can thereby derive from general economy principles that such P-inversion not possible either. That is, as is clear from (113), inserting the feature Wh[ ] on P is not required in Russian for structures like (113) to converge at the interpretative interfaces. If we assume that insertion of formal features must have some ‘output effect’ that is otherwise unavailable (Chomsky 1995; Reinhart 1995, 2006; Fox 2000), then we therefore predict that it will not be possible in Russian to insert Wh[ ] on P. Therefore our account correctly predicts that P-inversion structures like (107b) will be ill-formed in Russian. We find, then, that our Q-based theory of secondary wh-fronting correctly restricts the P-inversion seen in Mayan to only those languages, and does not generally predict that any language allowing possessor extraction should exhibit the pattern. Furthermore, the property that requires Mayan languages to exhibit the P-inversion in (105) seems to be a comparatively exceptional one; no other languages to my knowledge prevent adpositions from taking full DP complements. Thus the rather extraordinary movement of wh-possessors to SpecPP in Chol and Tzotzil is derived from a property independently known to be exceptional, and therefore is correctly predicted to be a comparatively rare phenomenon in the languages of the world.

190

THE GRAMMAR OF Q

5.5 Massive Pied-Piping and Its Constraints In the previous section we examined cases where the locality constraints on piedpiping in Q/wh-Agreement languages are obviated by movement of the wh-word within the pied-piped phrase. Interestingly, many Q/wh-Agreement languages possess a further means for obviating their stricter locality constraints on pied-piping. As has been observed numerous times in the literature, there appear to be environments where these stricter locality constraints are relaxed. For example, it has often been observed that the constraint against pied-piping from CompNP is weaker in matrix questions than in subordinate questions (Ross 1979; Sells 1985; Webelhuth 1992; Kayne 1994; Heck 2008). (114) Pied-Piping Past NP Projections in Matrix Versus Subordinate Clauses a. (?) [NP Pictures of which president ] does Jim own? b. * I wonder [CP [NP pictures of which president ] Jim owns ]?

Similarly, this constraint also seems to be weaker in nonrestrictive relative clauses than in restrictive relatives (Bresnan 1976; Emonds 1976, 1979; Jackendoff 1977; Nanni & Stillings 1978; Ishihara 1984; Sells 1985; Safir 1986; Fabb 1990; Grimshaw 2000; Borsley 1992; Horvath 2007a; Heck 2008). (115)

Pied-Piping Past NP Projections in Restrictive Versus Nonrestrictive Relatives a. (?) This book, [CP [DP the [NP reviews of which ]] were awful ], is really quite nice. b. * No book [CP [DP the [NP reviews of which ]] are awful ] is really quite nice.

Following Heck (2004, 2008), I use the term “massive pied-piping” to refer to structures like (114a) and (115a), where the stricter constraints of the limited pied-piping languages can (marginally) be violated. In this last section we will explore how this phenomenon can be approached within our Q-based theory. As first articulated by Heck (2004, 2008), any theory of massive pied-piping should accomplish two highly important goals. First, it should offer a generalization characterizing the environments where massive pied-piping can occur. Secondly, it should explain why massive pied-piping can occur in precisely those environments. Although numerous authors had observed the basic phenomenon of massive piedpiping, it was not until Heck (2004, 2008) that a general theory meeting these two basic criteria was developed. Indeed, Heck (2004, 2008) is the first work to both (i) observe that the weakening of the constraints on pied-piping in matrix clauses is identical to the weakening of those constraints in nonrestrictive relatives, and (ii) provide a unified account of these contrasts. A major insight of Heck’s (2004, 2008) theory of massive pied-piping is his characterization of the environments where it can take place. This characterization is summarized in (116). (116)

The Generalization on Massive Pied-Piping (Heck 2008: 160) Massive pied-piping is only possible within nonsubordinated CPs.


191

That the generalization in (116) predicts the contrast in (114) between matrix and subordinate questions is rather obvious. Somewhat less obvious, however, is that it also predicts the contrast in (115) between restrictive and nonrestrictive relative clauses. In order to derive this contrast from (116), Heck (2008) assumes an analysis of nonrestrictive relatives where they are ‘parentheticals’, and thus not truly subconstituents of the larger sentence where they are interposed. Under such an analysis, nonrestrictive relatives are therefore CPs that are not dominated by any other phrasal nodes, and so classify as being nonsubordinated CPs. Thus (116) predicts that they should also permit massive pied-piping. Given its ability to unify the phenomena in (114) and (115), let us therefore adopt the generalization in (116) as the proper characterization of the environments permitting massive pied-piping. Of course, we must now ask why this generalization should hold. Why should the stricter locality constraints in Q/wh-Agreement languages appear to be (marginally) weaker in nonsubordinated CPs? Although I am unable to offer an entirely principled account, I would like to observe here that several properties of massive pied-piping can be seen to follow from the stipulation in (117).43 (117)

Optionality of Q/Wh-Agreement in Nonsubordinated CPs In all Q/wh-Agreement languages, it is (marginally) possible in nonsubordinated CPs to employ Q-particles that have valued iQ. Thus, in the nonsubordinated CPs of Q/wh-Agreement languages, it is (marginally) possible for a Q-particle not to Agree with any wh-word.

That is, I claim that ‘massive pied-piping’ occurs when the Q-particle of a Q/ wh-Agreement language is (marginally) permitted not to undergo Agreement with any wh-word. Under this view, the stipulation in (117) trivially predicts that massive pied-piping will be restricted to nonsubordinated CPs (116). Importantly, however, (117) can also capture several other properties of massive pied-piping. First, we derive the fact that the constraint against pied-piping from CompNP is weakened in nonsubordinated CPs ((114) and (115)). Recall that the illformedness of (114b) and (115b) is due to (i) the requirement that Q Agree with a wh-word within its domain, and (ii) the inability for Q to Agree with a wh-word buried inside a lexical projection. Given the stipulation in (117), however, it follows that Q/wh-Agreement can (marginally) fail to take place in matrix questions and nonrestrictive relatives. (118) The Structure of Massive Pied-Piping Past NP [QP [NP Pictures of which president ] QiQ[+] ] does Jim own?

As illustrated in (118), (117) allows the use of valued iQ[+] in matrix clauses. Consequently, Agreement between Q and the wh-word in the pied-piped phrase need not occur. Therefore, the fact that the wh-word is buried within a lexical projection in (118) does not impact the well-formedness of the clause. We find, then, that (117) predicts the contrasts seen in (114) and (115). We also predict that in such nonembedded environments, it should be possible for Q and the

192

THE GRAMMAR OF Q

wh-word to be separated by multiple NP projections. Thus we correctly predict the possibility of structures like the following. (119)

Massive Pied-Piping Past Multiple NPs in Matrix CP a. [DP The [NP height of the [NP lettering on the [NP covers of which books ]]]] is against regulations? b. These books, [CP [DP the [NP height of the [NP lettering on the [NP covers of which ]]]] is against regulations ] , must be returned.

We have just seen that (117) predicts that pied-piping past NP should be marginally possible in nonsubordinated CPs. Generally speaking, however, our reasoning demonstrates that any constraint on pied-piping that derives from conditions on Q/wh-Agreement should be weaker in nonsubordinated CPs. Thus we should find that for all lexical categories LP, it is marginally possible to pied-pipe past LP in such environments. The following show that this prediction is borne out for VPs. (120) Pied-Piping Past VP in Matrix Versus Subordinate Clauses a. (?) [CP To [VP criticize who ] ] would be a mistake? b. * I wonder [CP [CP to [VP criticize who ] ] would be a mistake ]. (121) Pied-Piping Past VP in Restrictive Versus Nonrestrictive Relatives a. (?) John, [CP [CP to [VP criticize whom ] ] would be a mistake ], is having some trouble. b. * No one [CP [CP to [VP criticize whom ]] would be a mistake ] should be criticized.

Furthermore, data like the following show that our prediction is also borne out for APs. (122)

Pied-Piping Past AP Projections in Matrix Versus Subordinate Clauses a. (?) [CP To be [AP proud of who] ] would be a mistake? b. * I wonder [CP [CP to be [AP proud of who ] ] would be a mistake ].

(123)

Pied-Piping Past AP Projections in Restrictive Versus Nonrestrictive Relatives a. (?) John, [CP [CP to be [AP proud of whom ] ] would be a mistake ], is in some trouble. b. * Every man [CP [CP to be [AP proud of whom ] ] would be a mistake ] is a real jerk.

Finally, recall that the possibility of P-stranding in English led us to propose in chapter 4 that P in English is a lexical head. Consequently, the stipulation in (117) predicts that pied-piping of PP in English should be noticeably better in nonsubordinated clauses than in embedded questions and restrictive relatives. As the following data demonstrate, this prediction is borne out.


(124)

Pied-Piping of English PPs in Matrix Versus Subordinate Clauses a. (?) [PP With whom ] did she leave? b. * Egbert wonders [ [PP with whom ] she left ].

(125)

Pied-Piping of English PPs in Restrictive Versus Nonrestrictive Relatives a. (?) John, [CP [PP with whom ] she left ], is a very nice guy. b. ?? No man [CP [PP with whom ] she left ] is a very nice guy.

193

The data in (114)–(125) support our prediction that the prohibition against piedpiping past lexical categories will be weakened in the nonsubordinated CPs of limited pied-piping languages. Furthermore, recall from section 5.3.5 that the sensitivity of limited pied-piping to the CSC derives from the fact that Q/wh-Agreement is governed by the CSC. It follows, then, that (117) predicts that pied-piping in nonsubordinated clauses will marginally permit violations of the CSC. This prediction appears to be borne out. (126)

Massive Pied-Piping in Matrix Questions Is Not as Sensitive to the CSC a. (?) [ Dave and who ] did you see at the party? b. * I wonder [ [ Dave and who ] you saw at the party ].

(127)

Massive Pied-Piping in Nonrestrictive Relatives Is Not as Sensitive to the CSC a. (?) John, [ [ his mother and whom ] I saw at the party ], was really drunk. b. * The man [ [ his mother and whom ] I saw at the party ] was really drunk.

Thus far we have seen a variety of data demonstrating that, in the limited piedpiping languages, pied-piping in nonsubordinated clauses is marginally less constrained than pied-piping in subordinate clauses. One might naturally ask, then, whether piedpiping in nonsubordinate clauses is entirely unconstrained in these languages. That is, one might conclude from the data above that all constraints on pied-piping are weakened in nonsubordinated contexts. Such a conclusion, however, would be incorrect. As the following data illustrate, even massive pied-piping is still subject to the constraint that the highest verbal category of a main clause cannot be pied-piped (section 5.3.2). (128)

No Massive Pied-Piping of Main Copula or Auxiliaries a. * [ [ Is [DP whose doctor ] ]1 [ Dave t1 ]? b. * [ [ Is [AP proud of whom ] ]1 [ Dave t1 ]? c. * [ [ Is [VP eating what ] ]1 [ Dave t1 ]? d. * [ [ Has [VP seen what ] ]1 [ Dave t1 ]?

(129)

No Massive Pied-Piping of Main Predicate in Absence of Copula or Auxiliary * [ [VP Eats what ]1 Dave t1?

As the data indicate, there is no sense in which the sentences in (128) and (129) are even marginally acceptable in colloquial English. This is in interesting contrast to sentences like the following.

194 (130)

(131)

THE GRAMMAR OF Q

Massive Pied-Piping of Main Predicates When Copula Is Present a. ?? [AP Angry at whom ] could John never be? b. ?? Robert, [ [AP angry at whom ] John could never be ], has really underperformed. Massive Pied-Piping of Main Predicates When Copula Is Present a. ?? [DP A doctor of what ] was John during the war? b. ?? Podiatry, [ [DP a doctor of which ] John was during the war ], is a wonderful field.

Here, we find that massive pied-piping of main predicates is possible in English as long as there is some higher verbal category (such as a copula) selecting the predicate.44 These data show that even in massive pied-piping, it is impossible to pied-pipe the highest verbal category of a main clause. Given the overall freedom of massive pied-piping, we should naturally ask why it should remain subject to this particular constraint. Interestingly, our Q-based account provides an answer. Recall that the stipulation in (117) predicts only that massive pied-piping should be free of the constraints that derive from conditions on Q/wh-Agreement. It follows that any constraints on pied-piping that are independent of Q/wh-Agreement should still apply to massive pied-piping. Furthermore, recall from section 5.3.2 that the ill-formedness of (128) and (129) follows purely from the QP-Intervention Condition, and so is independent of the conditions on Q/wh-Agreement (46). Therefore, even though the Q in (128) and (129) need not Agree with the wh-word, such structures still violate the QP-Intervention Condition, and so are predicted to be ill-formed. This contrasts with the structures in (130) and (131), which do not violate the QP-Intervention Effect, and so are (marginally) well-formed. We find, then, that our stipulation in (117) correctly predicts that massive piedpiping is not free from every constraint on pied-piping. Rather, it is only the constraints that follow from properties of Q/wh-Agreement that are (marginally) weakened in nonsubordinated CPs. Importantly, this calls to mind the cross-linguistic variation in pied-piping structures observed in sections 5.2 and 5.3.2. As we saw there, there are languages such as Tlingit where pied-piping appears to be less constrained than it is in English. As we noted in section 5.3.2, however, pied-piping in such languages is not entirely unconstrained; it is still subject to constraints that follow from the QP-Intervention Condition. We find, then, that the distinction our theory draws between the ‘Agreement-based’ constraints on pied-piping and those that follow from the QP-Intervention Condition appears to be a rather robust one. It surfaces not only in the cross-linguistic variation seen in pied-piping structures, but also in the intralinguistic variation in pied-piping environments. In both cases, the same sets of constraints group together. To the extent that our Q-based account is able to capture these natural classes, it receives interesting empirical support. To conclude our discussion of massive pied-piping, I would like to point out a number of problems that face our Q-based account of the phenomenon. In short, besides the pattern in (128) and (129), there are certain other constraints on massive pied-piping in English, ones that do not easily follow from our Q-based account. It is not currently known how general these restrictions are. If further study determines


195

that they are a general property of massive pied-piping, then they stand as a difficult challenge to the Q-based account offered here. It has long been noted that English does not permit massive pied-piping past specifiers (Nanni & Stillings 1978; Ishihara 1984; Sells 1985; Heck 2008). Otherwise acceptable massive pied-piping structures in English are ill-formed if a specifier c-commands the wh-word within the fronted phrase.45 (132)

No Massive Pied-Piping Past Specifiers a. Their parties, [ [to be invited to one of which] is an honor], are legendary. b. * Their parties, [ [for us to be invited to one of which] is an honor], are legendary.

Similar to this is the observation that wh-words in a specifier position cannot massively pied-pipe (Nanni & Stillings 1978; Kayne 1983; Sells 1985; Heck 2008). That is, if a wh-word is contained within a specifier inside a phrase XP, then XP cannot be massively pied-piped by that wh-word.46 (133)

Specifiers Cannot Massively Pied-Pipe a. Their parties, [ [to be invited to one of which] was an honor ], are legendary. b. * Their parties, [ [for which to be reported in Time] was an honor], are legendary.

Finally, we might note a third generalization, similar to that in (133). Data like that in (134) suggest that English does not permit massive pied-piping by adjuncts. That is, if a wh-word is contained within an adjunct inside a phrase XP, then XP cannot be massively pied-piped by that wh-word. (134)

Adjuncts Cannot Massively Pied-Pipe a. * [DP A [NP [ how big ] fish ] did you catch? b. * [DP A [NP journey to NY how ] ] did he make? c. * [DP The [NP party where ] ] did you attend?

The first thing to note about the contrasts in (132)–(134) is that none of them currently follow from our Q-based theory of massive pied-piping. Nothing within our Q-based theory would rule out structures like the following, where the Q-particle undergoes no Q/wh-Agreement with the wh-word. (135)

Q-Based Theory of Massive Pied-Piping Fails to Predict (132)–(134) a. [QP [CP For us to be invited to one of [DP which parties ] ] QiQ[+] ] is an honor? b. [QP [CP For whom to be invited to one of [DP their parties ] ] QiQ[+] ] is an honor? c. [QP [DP A [NP [DegP how big ] fish ] QiQ[+] ] did you catch?

Interestingly, while our Q-based account does not yet predict these facts, we might nevertheless discern a wider generalization from which they would follow. Note that if we abandon the distinction between specifiers and adjuncts (Kayne 1994), all three generalizations may be derived from the following, more general statement.

196 (136)

THE GRAMMAR OF Q

No Specifiers Along the Path of Massive Pied-Piping If there is no Q/wh-Agreement in an English clause, then no node in the path from Q to the wh-word can contain a specifier position.47

Before we examine some further predictions of (136), let us confirm that it indeed derives the data in (132)–(134). Let us begin with the generalization in (132), that there can be no massive pied-piping past a specifier. The diagram in (137) illustrates the argument that follows. (137)

Massive Pied-Piping Past Specifiers Violates (136) QP XP

Q

YP SPEC

A node in the path between Q and wh-word contains Spec position, contrary to (136) ZP

… wh-word…

Assume that a wh-word W massively pied-pipes a constituent XP. By the stipulation in (117), it follows that (i) XP is sister to a Q-particle, and (ii) there is no Agreement relation between Q and W. Now suppose that there is a specifier SPEC inside XP that c-commands W. Since SPEC c-commands W, the phrase YP, of which SPEC is the specifier, must dominate W. Moreover, since YP is contained within XP, it follows that YP does not dominate the Q itself. Thus, by the definition of ‘path’ (Pesetsky 1982), it follows that YP is in the path between the Q and W. However, since YP contains a specifier position, it follows that the principle in (136) is violated, and so the structure is ill-formed. Next, let us confirm that (136) derives the generalizations in (133) and (134) that specifiers (i.e., specifiers and adjuncts) cannot massively pied-pipe. (138)

Massive Pied-Piping by Specifiers Violates (136) QP XP

Q

YP SPEC

…wh-word…

A node in the path between Q and wh-word contains Spec position, contrary to (136) ZP


197

Assume that a wh-word W massively pied-pipes a constituent XP. By the stipulation (117), it follows that (i) XP is sister to a Q-particle, and (ii) there is no Agreement relation between Q and W. Now suppose that W is contained within a specifier (i.e., specifier or adjunct), SPEC. Since W is dominated by SPEC, it is dominated by the phrase YP, of which SPEC is the specifier. Moreover, since YP is contained within XP, it follows that YP does not dominate Q. Thus YP is in the path between Q and W. However, since YP contains a specifier position, it follows that (136) is violated, and so the structure is ill-formed.48 We have seen, then, that the (admittedly mysterious) stipulation in (136) is able to derive each of the three generalizations in (132)–(134). Of course, despite these successes, the stipulation in (136) does not bear any natural resemblance to any other properties of Q that we have independently introduced. For this reason, both (136) and the data it covers remain difficult challenges to our Q-based theory of massive pied-piping. Nevertheless, let us conclude our discussion here by briefly observing how two additional features of massive pied-piping can follow from (136). As noted by Heck (2004, 2008), English does not permit massive pied-piping of finite CPs, despite the fact that it does permit massive pied-piping of non-finite CPs. (139)

No Massive Pied-Piping of Finite Subordinate Clauses in English a. (?) [CP To hire which candidate ] would be a mistake? b. * [CP That we hired which candidate ] was a mistake?

Interestingly, the inability to massively pied-pipe finite clauses in English may be a simple consequence of the generalizations in (132) and (133). Since finite clauses in English obligatorily contain subjects, a wh-word contained within a finite CP must either be (i) subject of the finite clause, or (ii) c-commanded by the subject. If the former is true, then massive pied-piping of the CP by the wh-word would violate the generalization in (133); if the latter is true, then such massive pied-piping would violate the generalization in (132). Thus (132) and (133) together rule out massive pied-piping of finite CPs in English. On the other hand, if we assume that infinitival CPs in English are truly subjectless,49 then (132) and (133) would fail to rule out pied-piping of infinitival CPs as in (139a). Finally, since both (132) and (133) follow from (136), we can view the pattern in (139) as potentially another consequence of that stipulation. The stipulation in (136) might also be responsible for another feature of massive pied-piping in English. Curiously, unlike pied-piping in Tlingit, not even massive pied-piping in English permits the pied-piping of relative clause islands. (140)

No Massive Pied-Piping of Relative Clause Islands in English a. * [DP The book [CP that Mary gave who ] ] did you steal? b. * Dave, [CP [DP the book [CP that Mary gave whom ] I stole ], is an avid reader.

Although the complete impossibility of structures like (140) in English is problematic for our Q-based account, we may nevertheless understand it as a particular case of the (equally problematic) pattern in (134). Recall from the data in (134) that

198

THE GRAMMAR OF Q

English does not permit massive pied-piping past adjuncts. It is worth noting that the relative clause island in (140) is an adjunct to NP that it modifies. Thus the impossibility for wh-words to massively pied-pipe relative clause islands in English could be seen as a particular case of the pattern in (134). Finally, since (134) is itself a consequence of (136), we find that the pattern in (140) might ultimately follow from that latter stipulation. In summary, the generalizations regarding massive pied-piping in (132)–(134), (139), and (140) do not to follow from any independently known principles of our Q-based theory. Nevertheless, we have seen that each of them would follow from the broader generalization in (136). Unfortunately, however, the broader condition in (136) remains a bald stipulation, and is irreducible to anything independently known about Q-particles. Moreover, (136) seems to be rather parochial to English, given that pied-piping in non-Agreement languages like Tlingit is rather obviously not subject to it. For the moment, then, we find that our Q-based theory has no special insight to offer into the question of why English massive pied-piping is subject to the conditions above. Consequently, I consider the facts in (132)–(140) to be a difficult empirical challenge for our Q-based theory of massive pied-piping.

6

Conclusion

6.1 Introduction Throughout the preceding chapters, we have developed and defended a somewhat novel approach to the syntax and semantics of wh-questions. We have seen that this Q-based theory brings with it several interesting results, particularly the ability to eliminate ‘pied-piping’ from the theory of grammar. As these results indicate, the introduction of the QP structure in (1) provides a new and rather versatile analytic tool, one that opens up many new analytic vistas for the linguist to explore. (1)


CP

QP1 Complementation XP

Q CQ

IP

Agree/ Attract

… wh-word…

QP1 Overt Movement

199

200

THE GRAMMAR OF Q

Of course, there are many open problems and challenges that our Q-based account faces, several of which were encountered in chapter 5. The solution to these problems is far from trivial, and quite a few seem to directly undermine the strength of our proposals. However, I hope that the preceding chapters have shown that the analysis in (1) is a viable and interesting hypothesis, one that has rather far-reaching consequences for the general theory of movement. In this final chapter we will briefly explore two additional ways in which the project begun with this study could be continued further. We begin in section 6.2, with an examination of the consequences our account has for the analysis of other types of A-bar movements. Following this, we briefly examine the way in which our Q-based theory might inform the study of free relatives.

6.2 The Syntax and Semantics of Other A-Bar Movements The central theoretical claim of this book has been that various phenomena surrounding wh-fronting directly result from the properties of Q-particles rather than from properties of the movement relation itself. Three prominent examples of such phenomena are (i) the inability to strand adpositions, (ii) the inability to extract possessors and determiners, and most notably, (iii) the existence of pied-piping structures. One of our central goals throughout has been to argue that these three phenomena do not, contrary to perception, reflect general properties of movement per se. It is important to note, however, that wh-fronting is not the only movement construction to exhibit these three properties. Indeed, it seems that any form of A-bar movement can be seen to exhibit them. For example, in English, we find that focusmovement is also able to ‘pied-pipe’. (2)

Pied-Piping in English Focus-Movement I’ve read John’s book, but [ DAVE’s book ], I haven’t read.

In the underlined portion of sentence (2), the possessive phrase Dave’s book undergoes focus-movement. Importantly, however, the preceding context makes clear that only the subconstituent Dave bears focus in this sentence. If we assume that the focus-movement in (2) is triggered by the focus feature, then (2) is a structure where the fronted phrase properly contains the projection bearing the features triggering the fronting. Thus sentence (2) seems to show that focus-movement allows ‘pied-piping’. Similarly, from the ill-formedness of sentences like (3), we find that focus-movement is unable to extract possessors. (3)

English Focus-Movement Cannot Extract Possessors * I’ve read John’s book, but [ DAVE’s ]1, I haven’t read [ t1 book ].

The movement of relative pronouns in English relative clauses also seems to exhibit these two properties. Structures like the following indicate that such movement permits ‘pied-piping’.

CONCLUSION

(4)

201

Pied-Piping in English Relative Clauses The man [CP [DP [ whose ] father ]1 I met t1 ]

In the noun phrase in (4), the relative clause has been formed by fronting the entire possessive whose father. Again, if the fronting seen in a relative clause targets the features of the relative operator, then structures like (4) demonstrate that relativization permits pied-piping. Similarly, the ill-formedness of structures like (5) show that relativization in English is unable to extract possessors. (5)

English Relativization Cannot Extract Possessors * The man [CP [DP whose ]1 I met [ t1 father ] ]

Therefore it certainly appears as if these three phenomena are general properties of all A-bar movement constructions, and are not strictly confined to the putative QPmovement of wh-questions. Indeed, the reader might recall that this is ultimately the reason why these three phenomena are commonly held to be properties of the movement relation itself, rather than (say) some idiosyncratic restriction on wh-fronting. One might rightly worry, then, whether our Q-based theory of these properties does not incorrectly predict that they should not hold for any other A-bar movement constructions. That is, at first blush, our Q-based theory of possessor-extraction might seem to predict that structures like (3) and (5) should be well-formed, since they would not be instances of wh-fronting/QP-movement. If this were the case, then our Q-based theory in (1) would seem to be a huge step backwards, abandoning the immensely successful program of cross-constructional generalization begun by Ross (1967). Let us ask, then, how our Q-based theory might incorporate the facts in (2)–(5). Given that our theory derives these phenomena from properties of Q, the most natural conclusion to draw is that any movement construction exhibiting them is also an instance of QP-movement. That is, the facts in (2)–(5) force the view that wh-fronting is not the only construction where a QP is fronted. Rather, besides wh-fronting, numerous other movement constructions—including focus-movement and relativization—are also cases where an instance of the category ‘Q’ undergoes movement. To begin spelling out this idea more concretely, let us recall from chapter 2 that the presence of so-called Q-particles with wh-indefinites in declarative clauses shows that Q is not inherently linked to questions. That is, although its name calls to mind the notion of a ‘Question-Particle’, our Q-element (and Q-feature) really has no deep connection with interrogativity per se. Let us expand on this view, and further adopt the position that ‘Q’, as we have been using the term, is simply a syntactic category label. Consequently, the Q which we have been studying throughout this work might simply be a single instance of a more general category. For example, we might hypothesize that the syntactic category ‘Q’ also contains heads that we may dub ‘QFOC’ and ‘QREL’. Consequently, let us rename the Q-particle found in wh-questions and wh-indefinites as ‘QQ’. Given that focus-movement in English evinces the properties seen in (2) and (3), we will assume that such movement constructions are formed via movement of a QFOCP, which dominates the focused constituent of the fronted phrase, as illustrated in (6).

202 (6)

THE GRAMMAR OF Q

Focus-Movement as QP-Movement

[ QP [DP [ DAVE’s ] book ] QFOC ]1

, I didn’t read t1.

Similarly, because English relativization evinces the properties seen in (4) and (5), we will assume that such constructions are formed via movement of a QRELP, which dominates the relative operator. (7)

Relativization as QP-Movement

The man [CP

[QP [DP [ whose ] father ] QREL ]1

[IP I met t1 ] ]

Within the broader contours of our Q-based account, the analyses in (6) and (7) would predict that both focus-movement and relativization should exhibit the three properties of (i) inability to strand adpositions, (ii) inability to extract possessors and determiners, and (iii) possibility of pied-piping structures. The analyses in (6) and (7) illustrate a more general approach to A-bar movement that our Q-based theory entails. Although the arguments throughout this book center on wh-fronting, our Q-based theory should be more broadly understood as a general theory of all those movement constructions that exhibit, for example, piedpiping. Of course, we will have to leave to future study the full development of Q-based accounts of these other movement constructions. Nevertheless, it should be apparent that our Q-based theory of wh-fronting need not be an embarrassing step backwards from the results gained by Ross (1967). Rather, it invites a new understanding of (some of) those generalizations, one where they are not properties of the movement relation per se, but rather of the preconditions for movement in many of the most well-studied movement constructions. Although the analyses sketched above demonstrate how our Q-based theory might handle the facts in (2)–(5), we should nevertheless ask if there is any independent evidence for these Q-based analyses. Furthermore, what role do the putative Qs play in the semantics of these movement constructions? What, exactly, is the QFOC ‘doing’ in a focus-movement construction? The Q-based analysis of focus-movement in (6) receives some independent support from recent work by Horvath (2000, 2007b).1 This work is principally concerned with the question of whether the feature ‘Focus’ is present in the syntactic representation of a sentence. Importantly, however, the main empirical claim of Horvath (2000, 2007b) is that so-called focus-movement does not actually target the features of the focused phrase inside the fronted constituent. Rather, such movement is triggered by

CONCLUSION

203

the features of a (phonologically null) focus-sensitive operator that c-commands the focused phrase.2 The evidence that Horvath marshals in support of her analysis comes from Hungarian. According to standard descriptions, the focused phrases in a Hungarian sentence must undergo focus-movement to a special preverbal focus position. This is illustrated in (8). (8)

Obligatory ‘Focus-Movement’ in Hungarian (Horvath 2007b) Question: Kinek mutattad be Jánost? who.DAT you.introduced.him John.ACC Who did you introduce John to? a. [ AZ UNOKAHUGOMNAK ] mutattam be Jánost. the my.niece.DAT I.introduced.him John.ACC I introduced John to MY NIECE. b. *Bemutattam Jánost [ AZ UNOKAHUGOMNAK ] I.introduced.him John.ACC the my.niece.DAT

As we see above, the preceding question indicates that the capitalized phrases in (8) are understood as bearing focus. Moreover, the ill-formedness of (8b) demonstrates that such focused phrases cannot occupy postverbal positions in Hungarian. Rather, they must undergo focus-movement to a preverbal focus position, as in (8a). Thus Hungarian seems to be a language where focused phrases must obligatorily undergo focus-movement. Because of the ubiquity of focus-movement in Hungarian, it has often been cited by linguists as evidence that movement can be triggered by the feature ‘Focus’. However, as Horvath (2000, 2007b) argues at length, Hungarian ‘focus-movement’ does not seem to be triggered by the feature ‘Focus’ per se. That is, contrary to the widespread impression resulting from facts like (8), not all focused phrases in Hungarian can undergo so-called focus-movement. For example, a focused DP that associates with the focus-sensitive operator még. . . is ‘even’ cannot undergo the movement seen in (8). (9)

Foci Associating With ‘Even’ Cannot Undergo Focus-Movement (Horvath 2007b) a. Mari elkésett még [ AZ ESKŰVŐJERŐL ] is. Mary.NOM she.was.late yet the her.wedding.from also Mary was even late to HER WEDDING. b. * Mari még [ AZ ESKŰVŐJERŐL ] is késett el Mary.NOM yet the her.wedding.from also she.was.late

As we see in (9), even though the phrase ‘to her wedding’ is focused in (9a), it need not undergo focus-movement to the preverbal focus position. Indeed, as we see in (9b), such a focused phrase cannot undergo such movement, despite the fact that it bears the feature ‘Focus’. Further evidence that so-called focus movement does not directly target focus comes from wh-questions. Contrary to the common understanding of the facts in (8), it is not always the case that the focused answer to a wh-question in Hungarian must undergo focus-movement. Indeed, if the wh-question is given a so-called

204

THE GRAMMAR OF Q

‘mention-some’ reading, where the answer to the wh-question need not be exhaustive, then the focused phrase in the answer cannot undergo the movement seen in (8). (10) Partial Answers Cannot Undergo Focus-Movement (Horvath 2007b) a. Question: Hol tudhatnám meg a vonatok menetrendjét? where I.can.know the train schedule.ACC Where can I find out about the train schedule? b. Answer: Megtudhatod (például) [ AZ INTERNETEN ] . . . you.can.know for.example the internet.on You can find out about it, for example, on the Internet . . . (in addition to possibly other places as well)

In (10b), the phrase az interneten is understood as the answer to the wh-question in (10a), and therefore bears focus. However, unlike in (8), this focused answer is not required to appear before the verb. Horvath (2000, 2007b) argues that what distinguishes the well-formed (10b) from the ill-formed (8b) is the way in which the answer is interpreted. Because of the implicit context, sentence (8b) must be understood as providing an exhaustive listing of the true answers to the question. On the other hand, the answer in (10b) can be understood as giving only one of the many possible true answers. Therefore, it seems that the movement seen in (8) is not obligatory when the focused phrase is understood to be only a partial answer to the question. Indeed, if the focused phrase in (10b) were fronted to the preverbal focus position of (8a), then the sentence would be understood as an exhaustive answer. From the data in (9) and (10), we find that the mere presence of the feature ‘Focus’ on a phrase XP is not sufficient for XP to trigger so-called focus-movement in Hungarian. Therefore we must conclude with Horvath (2000, 2007b) that so-called focus-movement does not target the feature ‘Focus’ per se. Of course, this conclusion immediately raises the question, “What features and/or phrases does ‘focusmovement’ target?” Although many answers are imaginable here, the one pursued by Horvath (2000, 2007b) is strikingly akin to our Q-based theory of focus-movement in (6). Horvath observes that the phrases undergoing focus-movement are always understood to exhaustively identify the entities of which the remainder of the sentence is true. Thus a sentence like (8a) is true iff the speaker’s niece is the only individual to which the speaker introduced John. Crucially, such ‘exhaustivity’ seems to be necessary for focus-movement to take place. Whenever a focused phrase is not understood to exhaustively identify in this way—such as in (9) and (10)—then focus-movement becomes impossible. In this sense, it appears as if focus-movement takes place iff the focus is interpreted as if it associates with a phonologically null version of ‘only’. According to Horvath (2000, 2007b), this is more than a matter of appearance. That is, Horvath proposes that Hungarian does indeed possess a phonologically null variant of the particle csak ‘only’, which she dubs ‘EI-OP’ (for ‘Exhaustive Identification Operator’). Importantly, Horvath proposes that so-called focus movement is actually triggered by the features of this EI-OP, and has nothing to do with the Focus feature itself. Thus the movement of the focused phrase in (8a) receives the analysis below.

CONCLUSION

205

(11) Focus Movement as Movement of EI-OP in Horvath (2000, 2007b)

[ CP [ EI-OP [ AZ UNOKAHUGOMNAK ] ]1 CEI-OP [ mutattam be Jánost ]. the my.niece.DAT I.introduced.him John.ACC Agreement/Movement

According to this analysis, the movement in (8a) actually targets the features of the EI-OP, which c-commands the focused phrase. Because of the semantics of EI-OP, the fronted phrase in (8a) is understood to exhaustively identify the true answers to the question. Of course, just as with its overt cousin czak ‘only’, a focused phrase in Hungarian need not necessarily be c-commanded by EI-OP. Therefore Horvath’s analysis in (11) predicts that some focused phrases in Hungarian need not undergo ‘focus-movement’, and that such unmoved foci will be interpreted nonexhaustively. Thus the analysis in (11) correctly predicts that so-called focus-movement correlates not with focus per se, but with an exhaustive interpretation of the focused phrase. Finally, what about the prima facie appearance that focus-movement is triggered by ‘Focus’? According to Horvath (2000, 2007b), this is ultimately due to the special relationship between EI-OP and Focus. Because EI-OP is simply a null version of csak ‘only’, it is a focus-sensitive operator. Therefore it must always c-command some focused phrase, and so movement of the EI-OP will always entail movement of a focused phrase. Therefore, because of the phonological invisibility of the EI-OP, linguists have incorrectly concluded that the movement triggered by its features are triggered by ‘Focus’ instead. It is interesting to note the strong similarity between the analysis in (11) and our own Q-based theory of focus-movement in (6). Indeed, if we were to simply identify our QFOC with Horvath’s EI-OP, the two analyses would be the same. More acutely, if we assume that QFOC possesses the exhaustive semantics of only (and the EI-OP), then our analysis in (6) could account for the Hungarian facts in precisely the way that (11) does. In this sense, then, we find that the evidence discussed by Horvath (2000, 2007b) provides interesting support for our Q-based analysis in (6). Furthermore, Hovath’s (2000, 2007b) analysis potentially answers the question of what role the putative ‘QFOC’ plays in the semantics of the focus-movement construction.3 More generally, we have seen in this section that it is possible and productive to view other A-bar movements as also being instances of the QP-movement postulated for wh-fronting in (1). Under such analyses, the A-bar movement in question is ultimately phrasal movement of a Q-particle. Importantly, such analyses allow us to provide a uniform account of the facts in (2)–(5), one where they receive the same analysis as the parallel facts for wh-fronting discussed throughout this work. Thus our Q-based theory need not commit us to the patently false claim that these properties will hold only for wh-fronting.

206

THE GRAMMAR OF Q

6.3 Free Relatives Another area where our Q-based account might yield results is the theory of so-called free relatives. A fundamental question in the theory of free relatives concerns the apparent disconnect between their internal syntactic form and their external syntactic distribution. Across languages, free relatives tend to have the internal appearance of subordinate questions rather than of adnominal relatives (Rooryck 1994; Jacobson 1995; Bury & Neeleman 1999; van Riemsdijk 2007; but cf. Citko 2009). For example, in English, the wh-word what can function as the wh-operator in a free relative (12a) and in a subordinate interrogative (12b), but not in a normal adnominal relative clause (12c). (12) Similarities Between Free Relatives and Subordinate Questions in English a. I ruined [ what you were cooking ]. b. I know [ what you were cooking ]. c. * I ruined [ the food [ what you were cooking ] ].

Other languages provide more striking examples of the alignment between free relatives and subordinate interrogatives. Indeed, one rather clear example comes from Tlingit. The following sentences illustrate the free relatives, subordinate questions, and adnominal relatives of Tlingit. (13) Free Relatives in Tlingit Át gasa.aaxí aadooch sá has du een kawuneegí. to.them let.them.listen who.ERG Q them.with they.speak Let them listen to whoever tells them. (Dauenhauer & Dauenhauer 1990: 224) (14) Subordinate Questions in Tlingit Hél has wuduskú [ waa sá has kawdayaayí ] not they.know how Q it.happened.to.them No one knew what happened to them. (Dauenhauer & Dauenhauer 1987: 294) (15) Adnominal Relative Clauses in Tlingit Wáa sá yatee [ wé [ l goodéi sá woogoodi ] how Q he.is that nowhere he.went.REL How is the man who didn’t go anywhere?

káa ]? man

As discussed in detail by Cable (2006a), there are numerous ways in which Tlingit free relatives (13) pattern with subordinate interrogatives (14) rather than adnominal relatives (15). The most obvious concerns the presence of a wh-operator. As we see in (13), free relatives in Tlingit are formed via movement of an overt wh-operator. Of course, such an overt wh-operator is also a crucial component of Tlingit subordinate interrogatives (14). Importantly, however, overt wh-operators do not appear in Tlingit adnominal relatives like (15). Rather, adnominal relatives in Tlingit are formed via null relative operators, and do not contain the overt wh-operators of either free relatives or subordinate questions. Furthermore, it should be stated that

CONCLUSION

207

the verb in a Tlingit free relative bears morphology that is otherwise characteristic of verbs in subordinate interrogatives, and it lacks the morphology characteristic of verbs in adnominal relative clauses. I refer the reader to Cable (2006a) for more details. We find, then, that free relatives across languages have the internal syntactic form of subordinate questions. Crucially, however, free relatives do not have the external syntactic distribution of subordinate questions. Generally speaking, the external distribution of a given free relative will match that of the phrase fronted into its left periphery (Bresnan & Grimshaw 1978; Groos & van Riemsdijk 1981). For example, when the fronted phrase of the free relative is a DP, the free relative will be able to appear in only those positions where a DP is permitted, as illustrated in (16). (16)

Categorical Matching Effects in Free Relatives a. I visited [ [DP whatever city ] you went to ]. b. * I went [ [DP whatever city ] you went to ].

In the sentences in (16), the fronted phrase in the free relative is the DP whatever city. Consequently the entire free relative can follow the verb visit, which selects for a DP, but not the verb go, which requires an adverbial. On the other hand, if the fronted phrase of the free relative is a PP, then it will have the opposite distribution. (17)

Categorical Matching Effects in Free Relatives a. * I visited [ [PP to whatever city ] you went ]. b. I went [ [PP to whatever city ] you went ].

In (17), the fronted phrase in the free relative is the PP to whatever city. Consequently this free relative cannot be complement to the verb visit (17a), which selects for a DP, but it can appear with the verb go, which can be modified by an adverbial PP (17b). In summary, free relatives exhibit the following puzzling combination of properties: they possess the internal syntactic form of wh-questions, but they have the external syntactic distribution of whatever phrase is fronted into their left periphery. Understanding this combination of properties is the most central problem in the theory of free relatives. How can something with the internal appearance of a subordinate question have the external distribution of whatever happens to be in its left periphery? There are, of course, a wide variety of answers to this fundamental question (Bresnan & Grimshaw 1978; Groos & van Riemsdijk 1981; Rooryck 1994; Jacobson 1995; Bury & Neeleman 1999; van Riemsdijk 2007; Citko 2009). Nevertheless, our Q-based theory of wh-questions in (1) might provide a novel perspective on this issue. First, recall that our QP-Intervention Condition from chapter 2 entails that QPs are (in certain cases) ‘transparent’ for selection. That is, a core property of QP projections is that they may intervene between heads and the phrases that those heads select for. Now, let us entertain the hypothesis that, just as QPs are ‘transparent’ for selection, any phrasal node bearing the feature ‘Q’ will likewise be transparent. That is, let us suppose that any phrase XP bearing the feature ‘Q’ will be able

208

THE GRAMMAR OF Q

to intervene between a head H and the phrases selected by H.4 This idea is illustrated in (18). (18)

The Q-Feature Renders Any XP Transparent for Selection HP H

XPQ YP

XPQ XQ

ZP

XP Q Is ‘Transparent’ for Selection – Selection Possible Between H and Either YP or ZP

Let us now consider what the picture in (18) would predict regarding subordinate interrogative CPs. Following our analysis of wh-questions in (1), all such CPs would bear the ‘Q’ feature. Thus the hypothesis in (18) entails that subordinate questions will be transparent for selection. (19)

Interrogative CPs Will Be Transparent for Selection

HP H

CPQ YP

CPQ CQ

ZP

CPQ Is ‘Transparent’ for Selection – Selection Possible Between H and Either YP or ZP

As can be seen, (18) predicts that the left-peripheral phrase in an interrogative CP should be able to satisfy the selectional requirements of the sister to CP. If we assume that the distributional properties of a phrase follow entirely from what selectional requirements it can satisfy, we predict that subordinate questions will be able to appear in any position that their left-peripheral phrases can appear in.

CONCLUSION

(20)

209

The Nature of Matching Effects in Free Relatives VP V visited

CPQ QP

CPQ

DP whatever city

you went to CPQ and QP Are ‘Transparent’ for Selection – Selection Possible Between V and DP in SpecCP

VP V went

CPQ QP

CPQ

PP

you went

to whatever city

CPQ and QP Are ‘Transparent’ for Selection – Selection Possible Between V and PP in SpecCP

In summary, the hypothesis in (18), combined with our Q-based theory of whquestions, predicts that subordinate questions should be able to exhibit the external syntactic distribution of the phrases occupying their left periphery. Of course, the very existence of free relatives could be taken as evidence that this prediction is borne out. That is, the hypothesis in (18) could provide a new analysis of free relatives, one that offers a unique answer to the ‘fundamental question’ regarding these structures. Under this analysis, free relatives have the internal syntactic form of interrogative CPs because they are interrogative CPs (Rooryck 1994; Jacobson 1995; van Riemsdijk 2007). However, because such CPs bear instances of the feature ‘Q’, they can exhibit the external syntactic distribution of their left-peripheral phrases. Thanks to the special property of Q-particles illustrated in (18), the specifiers of Q-bearing CPs are able to satisfy the selectional requirements of heads external to the CP. Consequently, interrogative CPs will appear to ‘match’ the external behavior of their left-peripheral specifiers. Of course, any serious attempt to work out the analysis sketched here will have to provide some explanation of the principle stipulated in (18). Assuming that this can be done, we find that our Q-based theory in (1) might provide a syntactic theory for free relatives, one where they may be analyzed simply as interrogative CPs, and where their peculiar external distribution follows entirely from independently motivated principles regarding Q.


NOTES

Chapter 1 1. As a side-note here, the reader should observe that the presence of the Q-particle sá can sometimes be obscured by its forming a portmanteau with the ‘focus particles’ áwé, áyá, áyú, and áhé, the two surfacing together as sáwé, sáyá, sáyú, and sáhé. This pattern can be seen, for example, in sentences like (1b). 2. Note that, following the definition in (7), I do not include under the rubric of “pied-piping” all instances of phrasal movement. That is, I accept as uncontroversial the existence of a mechanism of feature projection, which places the features of a head onto the projections of that head. What is at issue is any mechanism that places the features of a head onto nodes outside the projections of that head. This is a significant distinction, because feature projection is arguably indispensable, while the latter sort of devices are of little utility outside of deriving pied-piping structures. 3. Of course, such a language would exhibit one property that would saliently distinguish it from the best-studied wh-fronting languages. Given the possibility of structures like (10) in Tlingit, our hypothetical language would appear to allow the pied-piping of islands, a pattern not permitted in the best-studied wh-fronting languages. This variation will be discussed in chapter 5, where I propose that it follows from a rather superficial difference in the morphosyntax of the languages’ wh-words. 4. On the other hand, one cannot dispute that there are extraction types besides wh-fronting that are unable to extract from PPs or left branches. Indeed, this is the ultimate reason why the inability to extract from those two environments is commonly thought to be a property of movement in general, and not simply some idiosyncratic restriction on wh-fronting. One might worry, then, whether the account in (15) does not incorrectly predict that extractions from PP and left branches should be 211

212

NOTES TO PAGES

11–22

well-formed for all other types of movement constructions, like focus-movement and relativization. If so, then our account in (15) would seem to be a step backwards, abandoning the immensely successful program of cross-constructional generalizations begun by Ross (1967). Discussion of this important issue appears in chapter 6. There we will see that the Q-based theory sketched in (15) can capture the cross-constructional ill-formedness of such extractions by assuming that all the constructions in question also involve some subvariant of the Q-movement illustrated in (15). That is, besides the Q-particle found in wh-questions, there also exist separate, featurally distinct instances of the category ‘Q’ in focus-movement constructions, relative clauses, etc. As we will see later, this idea receives some independent support from recent work on focus-movement by Horvath (2000, 2007b), who argues that so-called focusmovement is actually movement of a (null) focus-sensitive operator sitting just above the fronted phrase. Chapter 2 1. Since the transliteration conventions for Sinhala can vary between authors, I will briefly explain the conventions I follow here. Certain authors variably transliterate the Sinhala character representing the vowel /a/, transliterating it as “ə” in contexts where the vowel is reduced, and as “a” in all other contexts. Thus certain authors transliterate the Sinhala Q-particle as “də”. Throughout this book I follow the simpler convention of uniformly transliterating this character as “a”; thus I transliterate the Sinhala Q-particle as “da”. Aside from this, I make no changes in the transliterations of the authors whose data I cite. 2. Throughout this book I follow Thompson (1996) in his use of the term ‘NaDene’ to mean a language phylum containing Tlingit, Eyak, and Athabaskan. In other work, this phylum is sometimes referred to as ‘Athabaskan-Eyak-Tlingit’ (Leer 2000), since the term ‘Na-Dene’ originally denoted a hypothetical (and likely erroneous) grouping that includes the neighboring language Haida. 3. Indeed, in some texts, (S)VO order slightly outnumbers (S)OV order. This is reported by Dryer (1985) and also conforms to my own experience. 4. While the data in (5) were elicited, Cable (2007) also provides textually attested examples of each of the six logically possible word orders. 5. Keri Edwards (personal communication) reports that some speakers find the SVO and VSO orders in (5b,e) to be highly awkward, characterizing them as ‘backwards’ and only said in moments of great excitement. A generalization that would cover this pattern might be that for such speakers, if the object follows the verb (V > O), then the subject must follow the object (O > S). 6. The reader may note that the verbal form in (5b) differs from that in (5a). This is due to a morphophonological rule that deletes third obviative object agreement when the verb is directly preceded by an NP marked by the optional ergative postposition. The effect of this rule appears in many of the examples throughout our discussion. 7. It is not known to me whether there is any difference in meaning or grammar between aa and aadoo. Their origins are also unknown to me, as well as which—if any—is the older form.

NOTES TO PAGES

22–29

213

8. Wáa is translatable into English as ‘what’ when it is the object of a propositional attitude verb. 9. I borrow the label ‘focus particle’ from Leer (1991). It is not clear to me, however, whether ‘focus particle’ is the best label for these particles. Story (1995) notes that the particles can serve equally well to either ‘background’ or ‘foreground’ material. My own suspicion is that these particles can simply follow any element in any left-peripheral position, whether Topic or Focus. Such a particle has been independently reported for the neighboring language Haida (Enrico 2003: 193, 246, 251– 252), where it actually seems cognate with the Tlingit particle. I should note that such an account of these particles is essentially that proposed by Leer (1991), though it seems out of sorts with the label ‘focus particle’. Finally, Dauenhauer & Dauenhauer (1990) take the view that these particles are semantically empty, and can simply be optionally added to any prosodic phrase in the sentence. 10. Much of the evidence that follows is consistent with an analysis where whwords in Tlingit wh-questions must be fronted to an immediately preverbal focus position, akin to wh-questions in other so-called ‘discourse configurational languages’ (Kiss 1995). It has been argued by some authors that such immediately preverbal focus positions are not left-peripheral positions (Arregi 2003b). Note, however, that sentences such as (7b) indicate that Tlingit does not require wh-words in wh-questions to occupy an immediately preverbal position. This, combined with the data below, indicates that wh-fronting in Tlingit is to a left-peripheral CP position. 11. Throughout this chapter I use the term ‘wh-operator’ in a purely informal, descriptive sense, as (roughly) ‘the wh-word representing the information being sought by the speaker’. As will be clear from the semantics proposed in section 2.7, I do not believe that such wh-words are ‘operators’ in any real semantic sense. 12. Keri Edwards (personal communication) reports that some speakers find these sentences to be unacceptable, and require the wh-indefinites to appear before the main predicate. I would hypothesize that for such speakers, there are additional conditions on postpredicate placement that independently rule out the appearance of postpredicative wh-indefinites. It is worth noting that these speakers do seem to exhibit a more restricted postverbal field (see note 5). 13. The Sealaska Heritage Institute regularly posts a ‘Tlingit Phrase of the Week’. This and others may be found at http://www.sealaskaheritage.org/programs/ tlingit_phrase_of_week.htm. 14. The appearance of the generic NP kéet in sentence (14b) does not necessarily upset the generalization, given that generic NPs can be classified as referential terms, denoting kinds (Carlson & Pelletier 1995). 15. Of course, the possibility exists that these speakers were simply trying to mirror the syntax of the original Tlingit in their English translations. I find this explanation doubtful, however. One speaker who was quite consistent in using left dislocation in his translations of these sentences would nevertheless translate other non-English word orders as standard SVO English sentences. 16. Sentence (24a) illustrates the Tlingit idiom for ‘to want’. Since we will encounter this expression numerous times, a few words should be said about it here. In Tlingit, one expresses the proposition “X wants Y”—where Y can be a CP or a DP—with an idiom literally meaning “Y is glad in X’s mind-face (spirit)” (Leer

214

NOTES TO PAGES

29–32

1991). This idiom can also be interpreted as “X likes Y”, which may in fact be the original meaning. This idiom has undergone a certain amount of grammaticalization and phonetic reduction, but its original structure can be seen in sentences like the following: (i) Has du tuwáx’ gu.áwe gé xat sigóo gé. their spirit.at DUB.FOC Q I.am.glad Q I wonder if they like me. (Naish 1966: 63) 17. Sentence (25b) can reportedly be interpreted to mean “Our teacher told us what we read.” Thus the asterisk here is intended only to represent that the sentence cannot be interpreted as a matrix wh-question. 18. Throughout this book we will see that the syntactic label ‘Q’ must be understood as an arbitrary name for a particular class of objects, ones that do not necessarily have any inherent connection to ‘questions’ per se. 19. It should be noted, however, that there are a number of particles besides da and ka with which wh-indefinites in Sinhala and Japanese may appear. This is not so for Tlingit. 20. In highly colloquial Japanese, it is possible to drop ka in matrix wh-questions like (33a) (Lasnik & Saito 1992; Yoshida & Yoshida 1996). However, there are stringent conditions governing this ‘particle drop’, and under one current account, such sentences contain an unpronounced ka (Ko 2005). 21. One salient difference, however, between Tlingit sá and the other two particles is that sá can only appear in sentences containing wh-words. The particle sá simply has no use outside of its obligatory co-occurrence with wh-words. This is unlike Japanese ka and Sinhala da, which can function both as markers of polar questions and as disjunctive operators (Hagstrom 1998). In this context, however, note that polar (yes/no) questions in Tlingit are formed via insertion of the particle gé, as illustrated by the following sentence: (i) Lingít gé x’eeya.áxch? Tlingit y/n you.understand.it Do you speak Tlingit? Thus, in Tlingit, wh-questions and yes/no questions are formed via two distinct particles. I suspend judgment here as to whether the particle gé should also be regarded as an instance of Q. Nevertheless, given the distinction between gé and sá in Tlingit, I assume that the use of da/ka in Sinhala/Japanese polar questions reflects the existence of a separate, homophonous ‘yes/no’ particle. Thus the apparent difference noted above might be only apparent, as the actual, underlying correlates of sá in Japanese and Sinhala likewise appear only in wh-questions. 22. Note that sentences like (34f) demonstrate that subordinate CPs in Tlingit may be pied-piped. 23. Japanese speakers tend not to allow the particle ka to be detached from the wh-word when it functions as a wh-indefinite (cf. Yatsushiro 2001). In wh-questions,

NOTES TO PAGES

32–42

215

however, the particle ka must appear detached, where it c-commands the wh-word from a clause-peripheral position. This ‘detachment’ of ka in wh-questions will be analyzed in chapter 3. As for the condition that ka in wh-indefinites be as close to the wh-word as possible, I currently have no principled account of this fact (which does seem limited to the wh-indefinites of Japanese). 24. Because Japanese ka obligatorily appears at the end of the interrogative clause, it cannot be easily determined whether this property also holds of the Japanese Q-particle. However, Hagstrom (1998: 40) argues that the behavior of the emphasis marker ittai in Japanese provides indirect evidence that it does. Thus, accepting the evidence put forth by Hagstrom (1998), we find that this pattern also unites Japanese ka with Sinhala da and Tlingit sá. 25. Like many languages, Tlingit possesses only a small, closed class of adjectives, and so most nominal modification is accomplished with relative clauses. Thus questions regarding the degree to which some NP possesses a given property (e.g., “How ADJ a NP”) must in Tlingit be asked using a structure in which the wh-word is buried within a relative clause. 26. This property clearly does not hold of the Japanese particle ka. 27. Hagstrom (1998) and Kishimoto (2005) describe some limited cases where Sinhala da may appear at the right edge of the matrix clause, although they disagree about what characterizes these cases. We should also note here that Sinhala da can generally appear matrix-finally in yes/no questions. (i) Chitra ee pota kieuwa da? Chitra that book read yes/no Did Chitra read that book? (Hagstrom 1998: 21) Recall from note 21, however, that I assume that the particles required in yes/no questions are distinct (though possibly homophonous to) the Q-particles appearing in wh-questions and wh-indefinites. Thus the behavior of da in Sinhala yes/no questions does not bear on the identity between Tlingit sá and the particle da in Sinhala wh-questions. In this context, it is also worth noting that the Tlingit yes/no particle gé can appear clause-finally. 28. Note that the Q-particle in (51) is not part of the functional projection of the wh-word itself. As we saw in (37), the sister of Q may contain lexical heads selecting for the wh-head. Thus the analysis in (51) must be distinguished from the competing claim that wh-in situ involves pure ‘feature-movement’ of [+wh] (Chomsky 1995). 29. Again, we should note that, just as in (51), the Q-particle in (53) is not part of the functional projection of the wh-word (see (34)). Thus the proposal in (53) must be distinguished from the less interesting claim that the wh-feature of a wh-word heads its own projection within the functional projection of the wh-word. 30. Note that the contrast between (62) and (63) also effectively refutes the analysis in (56). 31. One might offer the following objection to the idea that no syntactic relation may cross an island. Given that wh-words in Tlingit obligatorily co-occur with Q-particles, there is presumably some syntactic relation between them. Therefore the

216

NOTES TO PAGES

42–59

well-formedness of (62) indicates that at least this relation may cross an island. However, we will see in section 2.7 that there is not necessarily any syntactic relation between Q and the wh-word, since their obligatory co-occurrence independently follows from their semantics. 32. Similarly, it is the contrast between sentences (38a) and (38b) that most strongly motivates the Hagstrom/Kishimoto analysis of Sinhala wh-questions in (51). In both cases, the fact that only the position of Q affects the well-formedness of the question indicates that only Q bears a relation to the matrix interrogative C. 33. Sentences (83) and (84) reflect the well-formedness judgments provided by Mary Anderson of Atlin to Roby Littlefield. Mary Anderson speaks the Interior Tlingit subdialect of Northern Tlingit. 34. Sentence (90) reflects the judgments provided by Mary Anderson of Atlin to Roby Littlefield. 35. I believe, however, that some speakers greatly prefer the NPI reading of the whindefinite, to the point that the NPI reading is essentially obligatory in the environments that license it. 36. Keri Edwards (personal communication) reports that some speakers find the translation of (100e) incorrect, and can only interpret it to mean “if any of Dave’s in-laws win the lottery . . .”. Thus, for such speakers, it may indeed be that wh-indefinites cannot scope out of adjunct islands. However, it may also be that such speakers share the tendency, observed in note 35, to interpret wh-indefinites as NPIs in those environments that allow such an interpretation, including the antecedent of a conditional. Although the strong preference to interpret wh-indefinites as NPIs should not be taken lightly, the data in (98)–(100) show that such interpretations are (at least for some speakers) not obligatory. 37. Sandra Chung (personal communication) has suggested that the constraints governing the placement of Tlingit sá may be prosodic in nature. This particle is a clitic in Tlingit, and in other well-known languages, prosodic constraints on clitics have a similar effect on their distribution. Most famously, in the South Slavic languages, it has been proposed that prosodic constraints account for the inability for second-position clitics to intervene between (a) possessors and possessed NPs, (b) determiners and their complements, and (c) prepositions and their complements. Although this is a highly intriguing suggestion that should be independently pursued, I will put it aside here. As we will see, the account proposed here will be able to capture facts about Tlingit that I believe such a prosodic account would be unable to (e.g., the facts regarding the clause-final positioning of sá). 38. In this context, I use the term ‘intervene’ as follows: a QP intervenes between F and XP if either (i) QP dominates XP but does not dominate F, or (ii) if QP dominates F but does not dominate XP. 39. Although introduced here as a special stipulation, we will see later in this section how this condition might follow from a more general hypothesis concerning selection. 40. The reader may recall that this constraint also governs the Q-particle da in Sinhala, a wh-in situ language. In the next chapter, we will see how our QP-based theory of wh-questions can treat this fact.

NOTES TO PAGES

60–89

217

41. Note that there is a further problem with the structure in (115): the DP projection appears to be exocentric, in as much as it does not immediately dominate a D-head. 42. The meaning stated in (122) is only offered as an approximation of the semantics of ‘only’. Although satisfactory for our purposes, the meaning in (122) is actually too strong, as it would incorrectly predict that the truth of “only XP” rules out all the logical entailments of XP. 43. Of course, this account renders problematic those languages where it seems that wh-operators need not be obligatorily focused, such as English. As the reader will later see, difficult questions will also arise regarding the tendency for wh-indefinites to be unfocused in languages such as German. 44. Properly speaking, although Hagstrom (1998) considers this proposal, he ultimately rejects it, opting for a theory in which the Q-particles are operators while the traces of the Q-particles are variables over choice functions. 45. Throughout this discussion I adopt the label cf as a means for abbreviating the logical type of the choice function. Furthermore, I assume a cross-categorical definition for choice functions of the kind used by Winter (1997). 46. Note, however, that this tree abstracts away from certain complexities concerning the interpretation of extensional closure in an intensional system of the kind we adopt here. 47. The fact that there is some a ∈ { x : x ∈ Dτ & P(x) } such that R = [B . . . a . . . ] follows ultimately from the fact that wh-words in our system only ever semantically compose with their sisters via pointwise semantic composition. 48. Recall that we observed earlier, under note 31, that this result undermines one possible objection to our arguments for the analysis in (1)/(53). 49. I ignore here the complexities surrounding the indexical your in these sentences. 50. The reader may note that (151) also assumes the Heim and Kratzer (1998) theory concerning the interpretation of structures with movement. According to this theory, movement of a phrase XP to a position Y results in the insertion of a lambda operator directly below Y, binding the trace of the XP. 51. The abbreviation ‘FA’ here stands for the rule of ‘Function Application’. 52. The abbreviation ‘LC’ here stands for the rule of ‘Lambda Conversion’. 53. Multiple wh-questions will be treated in chapter 4. Chapter 3 1. The term ‘Q-projection language’ will also be used to refer to all languages where the Q-particle takes its sister as complement, including wh-fronting languages like Tlingit. 2. On the other hand, it may be that a uniform account of these facts would not be desirable. Kishimoto (2005) adopts a movement-based account of (9)–(11), assuming the Q-adjunction analysis in (2). Recall that such a movement-based account was rejected for Tlingit, largely because this pattern is also observed in Tlingit wh-indefinites. Interestingly, Hideki Kishimoto (personal communication) reports that the pattern in (9)–(11) is not observed for Sinhala wh-indefinites. When appearing with

218

NOTES TO PAGES

89–103

Sinhala wh-indefinites, the particle da can come between functional heads and the phrases those heads select. Such facts indeed lend credence to a movement-based account of (9)–(11), and so challenge the analysis proposed here. 3. It should be acknowledged that many have argued that wh-words in Japanese (and Korean) undergo full phrasal movement in wh-questions. Such arguments challenge our analysis in (2). Hagstrom (1998), which is the original source for the analysis in (2), includes a substantial discussion of the alleged evidence for wh-movement in these languages, and shows how it might nevertheless be captured by the account in (2). I refer the reader to Hagstrom (1998) for a thorough discussion of these important issues. 4. The reader may note that the Q in (17) is separated from the wh-word by a ‘linking’ morpheme eyn. This ‘linker’ is obligatory in Korean wh-indefinites. Unlike Japanese ka, Korean ka cannot directly combine with wh-indefinites. As the reader will observe, there are other linking morphemes besides eyn, and the choice of morpheme depends upon the syntactic position of the wh-indefinite. 5. Another possibility, mentioned by native speakers of both languages, is that Q-particles in Japanese/Korean can only cliticize onto nominal categories, and the words I identify above as D-heads (dono/enu) are actually adjectives. 6. After all, within a Bare Phrase Structure system, some principles must entail that the phrase initially merging with a head H must be the internal argument of H, rather than its external argument. Presumably these principles could also entail that the only thing that may undergo initial merger with a head is its internal argument. 7. Note that (24) is agnostic as to the structural position (and nature) of the case markers in Japanese, which can be assumed to be semantically vacuous. 8. The stipulation in (25) is not absolutely required for our system to correctly interpret wh-questions with pure Q-movement. See Cable (2008) for an analysis where (25) is dispensed with. 9. Example numbers here refer to those from chapter 2, where the rules in question were introduced. 10. Also note that the equivalence of (27) and (28) follows from the general equivalence proved in (135) in chapter 2. 11. Although the complement of–man in (34) does contain the trace of the fronted wh-word, it might be assumed that such traces possess normal-semantic values as individual variables. Chapter 4 1. Of course, given that relative clause islands in Tlingit may be ‘pied-piped’ (chapter 2), it is apparent that the class of pied-piping structures are wider in Tlingit than in languages like English, where structurally parallel sentences are not possible. However, we will see in chapter 5 that this difference may be due to an independent morphosyntactic difference between the wh-words of English and Tlingit. 2. In addition, Cable (2007) discusses several Tupí languages of Central and South America, arguing that they are also wh-fronting languages with overt QPs. In this context, one should also note the analysis of Finnish questions developed by Holmberg (2008). Given that the behavior of Q-particles in Finnish questions is similar to

NOTES TO PAGES

103–105

219

that found in Tlingit, Holmberg (2008) independently develops an analysis rather similar to our Q-based account in (1). 3. Baker (1999) glosses the word òmwàn as ‘person’ and the word èmwìn as ‘thing’. However, this glossing seems to be chosen primarily because these words can be used as indefinites meaning, respectively, ‘someone’ and ‘something’. Therefore I consider it possible that these words are wh-words, their indefinite meaning following from the existence of wh-indefinites in the language. 4. Note also that Edo is, like most other languages of its area, a head-initial language. Consequently, our Q-based analysis correctly predicts that the particle dè must—unlike the Q-particle of Tlingit—precede the wh-phrase. 5. Baker (1999) himself proposes that dè is the head of some higher CP projection that takes the matrix wh-question as complement, like the ‘ForceQ’ head we introduced in chapter 2. Such a view would account for all the data in (4)–(7). On the other hand, as noted by Baker (personal communication), such an account would fail to generalize to the Tlingit particle sá, since Tlingit sá occurs to the right of the wh-phrase. Thus, if we wish to capture the similarities between Tlingit sá and Edo dè noted here, we must adopt the Q-based analysis of Edo whquestions. 6. An outstanding challenge to our analysis is that dè can freely combine with any full NP, and not just wh-words. In such sentences, the particle seems to contribute the meaning of a wh-determiner. (i)

Dè òmó nè Àdésúwà bó!ó ? Q child that Adesuwa comfort Which child did Adesuwa comfort?

However, as noted previously, Baker (1999) provides strong arguments against the analysis of dè as a wh-D. The following is a possible account of these facts that retains the parallelism between Tlingit sá and Edo dè. Suppose that Edo dè differs from Tlingit sá in that it need not take only focus-semantic values as its argument. That is, suppose that Edo dè is interpreted as a choice function taking as argument the normal-semantic value of its sister. Consequently dè could combine directly with set-denoting expressions like NPs. Under this view, sentences like (i) could be interpreted as follows: (ii)

[ λp. ∃f. p = [ λw. Adesuwa comforted f({x: x is a child}) in w ] ]

Note that such a meaning indeed reflects the English translation provided to (i). A second possibility to consider is that Edo might simply possess a phonologically empty wh-determiner. Thus it may indeed be correct to assign Edo dè the exact same meaning as Tlingit sá, with sentences like (i) receiving the following analysis: òmó ] ] nè Àdésúwà bó!ó ? (iii) [QP Dè [DP Q which child that Adesuwa comfort

220

NOTES TO PAGES

107–118

7. It is worth noting in this context that Abels (2003) demonstrates that PP is not generally a syntactic island, even in languages that disallow P-stranding. For example, although Russian does not permit extraction of the complement of P (11), we can see from sentences like the following that it does permit extraction from within the complement of P. (i)

[Protiv kakoj točki zrenija ]1 ty ešče ne slyšal [PP ob argumentah t1 ]? against which point view you yet not heard about arguments Which point of view have you not yet heard about arguments against? (Abels 2003: 161)

From facts like this, Abels (2003) concludes that the ill-formedness of P-stranding cannot be due to the general islandhood of CompPP. 8. On the other hand, it is indisputable that wh-fronting is not the only type of extraction for which the positions in (11)–(13) seem to be islands. Indeed, this is a central reason why the positions in (11)–(13) are thought to be general extraction islands. For example, the inability to extract possessors holds not only for English wh-fronting, but also for focus-movement, relativization, etc. (i) English Focus-Movement Cannot Extract Possessors (a) I’ve read John’s book, but [ DAVE’s book ]1 I haven’t read t1. (b) * I’ve read John’s book, but DAVE’s I haven’t read [ t1 book ]. One might worry, then, whether our Q-based account does not incorrectly predict that the extractions in (12) and (13) should be well-formed in other constructions of English. As will be further discussed in chapter 6, our Q-based account avoids this consequence if we assume that the extractions in question are all some subvariant of the Q-movement seen in wh-questions. That is, besides the Q found in wh-questions, there also exist separate, featurally distinct instances of the category ‘Q’ in focus-movement constructions, relative clauses, etc. As we will see later, this idea receives some independent support from recent work on focus-movement by Horvath (2000, 2007b). 9. This view is essentially that proposed in recent work by Safir (2009). Interestingly, the objections that I raise against this general view do not actually go through for the specific system developed by Safir (2009). 10. On the other hand, it is undeniable that pied-piping structures occur in constructions other than wh-fronting. For example, sentences like the following seem to exhibit pied-piping in focus movement. (i) I’ve read John’s book, but [ DAVE’s book ] I haven’t read. One might rightly worry, then, whether our Q-based account is not too parochial, and fails to derive the possibility of pied-piping structures in A-bar constructions beyond wh-fronting. This issue will receive extended discussion in chapter 6. There we will see how our Q-based account can be extended to other types of A-bar movement that are observed to permit pied-piping structures. In general, the idea we will pursue is that

NOTES TO PAGES

118–123

221

mentioned in note 8: the movements in question are all subvariants of the Q-movement seen in wh-questions. 11. Certain earlier work, however, does assume theories of agreement where such ‘feature transfer’ could take place (e.g., Grimshaw 2000). Furthermore, Pesetsky and Torrego (2001) put forth an Agreement-based theory of pied-piping that may circumvent this problem. 12. Heck (2004) also notes that such an account faces the conceptual problem that it countenances ‘feature movement’, in the sense of Chomsky (1995). In most current work, the possibility of ‘feature movement’ has been rejected in favor of mechanisms like long-distance Agree. 13. Of course, under our Q-based account, the facts in (38) do not follow from constraints on movement per se. However, anyone adopting our Q-based account of (38) would ipso facto be rejecting the ‘feature movement’ analysis of pied-piping in (37), and vice versa. 14. To be precise, Heck (2004, 2008) proposes a hybrid theory, where the labor of deriving pied-piping structures is divided between a limited mechanism of feature percolation (identified as ‘feature movement’) and a limited degree of ‘nonlocality’ between the moved phrase and the feature inside it promoting the movement. Such a theory is able to avoid the problems faced by accounts that appeal to only one of these two general forms of analysis. 15. Note that this is also a prima facie problem for our Q-based theory of piedpiping structures, one that will receive extensive attention in chapter 5. 16. The semantic issues discussed here are treated in greater depth by Cable (2007). They are also discussed by Sternefeld (2001). 17. As we will discuss later, there are some environments in English where whwords are subject to Intervention Effects. For our current purposes, however, we will temporarily put such cases aside. 18. As we will discuss later, there are some environments in English where multiple wh-questions seem not to be subject to Superiority. For our current purposes, however, we will again temporarily put such cases aside. 19. I should note here, however, that the empirical status of Superiority Effects in German remains controversial. For example, as first argued by Fanselow (1991, 1997), Superiority Effects do seem to surface in German if the structurally lower wh-word undergoes long-distance movement from a subordinate clause. That is, sentences like (i(b)) are decidedly ill-formed in comparison to sentences like (i(a)), a fact that is quite reminiscent of the English facts in (42). (i)

Possible Superiority Effects in German Long-Distance Movement glaubte, [ dass Peter ihr wen2 vorstellte ]? a. Wer1 who-NOM believed that Peter her who-ACC introduced Who believed that Peter introduced her to whom? b. * Wen2 glaubte wer1, [ dass Peter ihr t2 vorstellte ]? who-ACC believed who-NOM that Peter her introduced

However, Pesetsky (2000: 76–83) argues that structures like (i(b)) are ill-formed for reasons independent of Superiority.

222

NOTES TO PAGES

123–132

On the other hand, Featherston (2005) reports that German speakers do voice a slight preference for (43a) over (43b). I assume here that such preferences indicate that Superiority-violating structures like (43b) are slightly more difficult to process than those satisfying Superiority (43a), as independently argued by Sag et al. (2006). However, I assume such differences in processing ease are a distinct phenomenon from the categorical grammaticality judgments in (42), which indicate that Superiority-violating structures are not generated by English grammar. 20. While Pesetsky (2000) makes an explicit connection between Intervention and Superiority Effects, this claim is not made explicitly by Beck (1996). Nevertheless, Beck’s (1996) analysis of the contrast between (40) and (41) does implicitly relate this contrast to that in (42) and (43). Note that Beck (1996) ties the presence of Intervention Effects in a language to the presence of (overt) A-scrambling. Interestingly, the presence of A-scrambling is claimed by Fanselow (1991, 1997) to entail an absence of Superiority Effects. Thus, taken together, Beck (1996) and Fanselow (1991, 1997) share the prediction that Intervention Effects and Superiority Effects in multiple wh-questions should be mutually exclusive. 21. Note that the structure in (51) assumes, as is commonly done, that all the wh-words of an English multiple wh-question move into the left periphery by LF. 22. Note that the equivalence between (48c) and the formula derived in (51) also follows from the more general equivalence proven in chapter 2 under (135) and (136). 23. Another important difference between (56) and (49) is that ‘ForceQ+’ in (56) combines with the focus-semantic value of its complement. Thus, unlike ‘ForceQ2’ in (49), this ‘ForceQ+’ is itself a focus-sensitive operator. 24. Moreover, the reader may note that the ForceQ+ head is sufficient to interpret wh-questions containing any number of wh-words. Thus, unlike English ForceQ2, German ForceQ+ is sufficient to interpret ternary wh-questions, and generally any wh-question containing more than two wh-words. 25. In this way, the account we offer here is akin to that of Pesetsky (2000), which similarly derives the variation in (40)–(43) from a hypothesized difference in the range of interrogative C heads that German and English possess. 26. For a more complete explanation of the term ‘D-linked’, I refer the reader to Pesetsky (1982, 1987, 2000). Note that Pesetsky (1982, 1987, 2000) assumes that ‘D-linkedness’ is a property of wh-words, while I assume here that it is a property of wh-questions. Some evidence in favor of the latter notion is the fact, first observed by Comorovski (1996), that violations of Superiority like those in (62a) are again impossible if one of the two wh-phrases is replaced with a simple wh-word. (i) a. [ Which book ]2 did [ which boy ]1 read? b. * [ Which book ]2 did [ who ]1 read? c. * [ What ]2 did [ which boy ]1 read? These facts cohere best with an account similar to the one offered here, where D-linkedness is a global property of the wh-question, and are more difficult to capture if ‘D-linking’ is assumed to be a local property of wh-words.

NOTES TO PAGES

132–144

223

27. Note that we do not state that ForceQ-Dlink appears in all D-linked wh-questions, only that ForceQ-Dlink cannot appear in a question unless the question is D-linked. This distinction is crucial for predicting the data in (63). Since, as described above, the presence of ForceQ-Dlink is the only means by which a (binary) wh-question in English can violate Superiority, that head must be present in sentences like (63a). Consequently, in such sentences, in situ wh-words will be subject to Intervention Effects. However, for D-linked wh-questions where Superiority is respected (63b), nothing in the structure of these sentences entails that they contain ForceQ-Dlink. Thus such questions could instead contain ForceQ2, and we correctly predict that their in situ wh-words are not subject to Intervention Effects. If our system required instead that all D-linked wh-questions contain the head ForceQ-Dlink, then the difference between (63a) and (63b) could not be captured. 28. As observed in note 24, ForceQ+ is sufficient to correctly interpret multiple wh-questions containing any number of wh-words. 29. Note, however, that it would fail to fully predict the data in (66). Crucially it is only when nonbinary wh-questions violate Superiority that their in situ wh-words are subject to Intervention Effects. The account proposed here, however, wrongly predicts that the in situ wh-words of any nonbinary multiple wh-question will be subject to Intervention Effects, and so it wrongly predicts that (66a) should be uninterpretable. 30. To be precise, Beck (2006) does put forth an account of the data in (67). Beck (2006) proposes that in such separation structures, the wh-word is interpreted as part of the IP-internal DP remnant. Such a lower interpretation of the wh-word would indeed allow our account to derive these data as a subcase of the phenomenon in (40). Unfortunately this analysis would rely upon a special, sui generis stipulation regarding the interpretation of separation structures. 31. We should also note that our theory of I-Intervention Effects correctly predicts the well-formedness of structures like (69). 32. The judgments reported in (75)–(82) summarize those collected from five native speakers of English. Only one of these speakers failed to perceive a contrast between (75a, b, and c) or between (77a, b, and c). Chapter 5 1. Properly speaking, it is only Grimshaw (2000) that explicitly states this generalization. Webelhuth (1992) states that there is no pied-piping past theta-assigners, while Cowper (1987) states that there is no pied-piping past categories that can be lexically specified as ‘[+Wh]’. However, in the context of their respective theories, these latter two generalizations are equivalent to there being no pied-piping past N, V, or A. 2. One might legitimately doubt whether this is the correct generalization to draw from the handful of data in (7). I return to this issue in section 5.2.3. 3. Interestingly, the ill-formedness of subordinate questions like (7b) weakens if they are used instead as matrix questions (i) or appositive relative clauses (ii). (i) Pictures of whom has John bought? (ii) My father, pictures of whom John has bought, is very famous.

224

NOTES TO PAGES

144–146

In section 5.5 we will explore this phenomenon in more depth. For the moment, however, I will follow Heck (2008, 2009) in temporarily putting aside such cases of ‘massive pied-piping’. That is, I follow Heck (2008, 2009) in his assessment that the pied-piping in (7b–d) is indeed ill-formed in English, and that there is something special about matrix environments that (marginally) improves their acceptability. 4. Recall, however, our claim in chapter 4 that prepositions in English are actually lexical heads. Under this assumption, the generalization that English disallows pied-piping past lexical categories would predict that English should disallow piedpiping of PPs. Although this might strike many readers as a blatantly false prediction, it has often been noted that PP pied-piping is actually a rather marginal structure of English (or, more accurately, those registers of English that permit P-stranding). The skeptical reader is asked to suspend judgment until our discussion of ‘massive pied-piping’ in section 5.5. There we will see that our overall system actually makes the more nuanced (and accurate) prediction that pied-piping of PPs in (colloquial) English is noticeably better in matrix clauses than in subordinate clauses. 5. It is difficult to find more direct evidence that Tlingit permits pied-piping past lexical categories. It is impossible to construct direct correlates to the English sentences in (7b–d). First, Tlingit has only a very small number of (putative) ‘adjectives’, none of which appear with complements. Furthermore, nominal arguments of Ns in Tlingit do not appear to ever remain within the NP projection. Finally, Tlingit has no process of VP-fronting. On the other hand, clausal arguments of Ns like neek ‘rumor, news’ do seem to stay inside the NP projection. Thus one could examine whether Tlingit permits piedpiping past neek by a wh-word within its CP complement. If this is indeed the case, it would provide additional evidence that Tlingit permits pied-piping past lexical categories. As of yet, I have not been able to conduct this test with speakers. 6. Furthermore, Kratzer and Shimoyama (2002) propose that the presence of [Q] on a language’s wh-word is tied to the phonological form of the wh-word. That is, they claim that a language’s wh-words bear [Q] iff they all share some morphophonological feature in common (e.g., the ‘wh-’ submorpheme of English whwords). Cable (2007) likewise adopts this proposal, and thereby attempts to predict from the form of the wh-words whether a language is a ‘limited pied-piping language’. In this work, however, I will not adopt this assumption. That is, I will simply stipulate for a given language whether its wh-words bear [Q]. I leave to future work the question of whether this property of the wh-word can be predicted by independent means. 7. The original formal implementation in Kratzer and Shimoyama (2002) is couched in terms of “greedy” feature movement, and so is not compatible with the syntactic system that I assume here. 8. This system for feature valuation is also based on earlier work by Brody (1997). 9. It may at first blush seem suspicious for a Q-particle to initially bear an unvalued Q-feature. However, our proposals regarding Q are rather parallel to the

NOTES TO PAGES

146–152

225

proposals regarding tense in Pesetsky and Torrego (2007), where the tense node of the clause initially bears unvalued tense, and receives its tense value only under Agreement with the uninterpretable (but valued) tense feature of the verb. 10. It is interesting to note that, although Japanese is not a wh-fronting language, the hypothesis in (14) predicts that, given its putative lack of Q/wh-Agreement (13), it will permit wh-words to be dominated in the sister of Q by islands. This prediction is borne out by the well-known possibility in Japanese wh-questions for the whoperator to be buried inside of an island. (i) Kimi-wa DP [CP dare- ga kaita ] hono-o ] yomi-masi-ta ka? you-TOP who-NOM wrote book-ACC read.HON-PST Q What person is such that you read books that they wrote? (Hagstrom 1998: 40) Hagstrom (1998: 40–45) argues that such structures are derived via movement of the Q-particle ka from a base position outside the island, as shown in (ii). (ii) Kimi-wa DP [CP dare-ga kaita ] hono-o ] t1 yomi-masi-ta ka1 ? Thus, given the analysis in (ii), it appears that Japanese does allow wh-words to be dominated by islands within the sister of Q, as predicted by our hypothesis in (14). Furthermore, this fact raises the following, more general point. The ‘LF–piedpiping’ analysis of sentences like (i) (Nishigauchi 1990) has often been criticized on the grounds that its hypothesized pied-piping violates the observed constraints on pied-piping in languages like English (Fiengo et al. 1988). However, the behavior of pied-piping in Tlingit and the accompanying theory in (14) undermine these criticisms. A proponent of Nishigauchi’s (1990) analysis could claim that the differences between their hypothesized pied-piping and the pied-piping seen in English follow simply from the fact that Japanese is not a Q/wh-Agreement language. 11. A somewhat similar notion, that every nonphase is complement to a phase head, appears in recent work of Boeckx (2009). 12. Note, however, that there is much controversy over the exact nature, and therefore statement, of the PIC. For instance, Bošković(2005b) argues that Agreement can apply across different spell-out domains. Some evidence in support of this includes the phenomenon in Icelandic of verbal Agreement with nominative casemarked objects: (i) Honum mundu sennilega hafa [VP líkað þeir ] him.DAT would.3PL probably have liked them.NOM He would probably have liked them. (Heck 2008: 44) 13. However, for some recent thorough treatments, see Horvath (2007a) and Heck (2008). 14. Curiously, we can see from sentences like the following (first introduced in chapter 4) that the correlates of (21c) in German are well-formed.

226

NOTES TO PAGES

152–154

(i) Possible Pied-Piping by a Modifier of NP in German Fritz möchte wissen [ ein wie schnelles Motorrad ] du fahren darfst. Fritz wants to.know a how fast motorbike you drive may Fritz would like to know how fast a motorbike you are allowed to drive. To my knowledge, there has never been proposed a fully satisfactory account of this difference between English and German. I choose here to adopt an account similar to that in Heck (2004), where the initial determiner ein, ‘a’, in such structures has actually undergone head-movement to a distinct functional head F above DP. Under this kind of analysis, the inversion of the DegP seen in English sentences like (ii) actually does take place in German, but is subsequently obscured by the fronting of the determiner head to the higher functional position above the DegP. The following structures illustrate the general idea. (ii)

German Degree Questions Contain Obscured DegP Inversion [FP F [DP [DegP How fast ]1 a [NP t1 car ] ] ] can you drive? [FP [F [D Ein ]1 F ] [DP [DegP wie schnelles ]2 t1 [NP t2 Motorrad ] ] ] du fahren darfst.

We will later see in section 5.4 how our Q-based theory of limited pied-piping licenses the ‘DegP Inversion’ overtly seen in English (ii). That account will apply equally well to the postulated German structure in (ii). 15. As reported by Coon (2009), this pattern is also found in the Mayan language Chol. Moreover, Coon (2009) argues that a Q-based theory of pied-piping like the one developed here can capture certain complex features of possessor pied-piping in Chol. A related (although slightly different) account of the Chol facts will be provided in section 5.4. 16. As noted by Heck (2008: 243–244), there are some languages that permit pied-piping by postnominal possessors. Greek, illustrated below, is one such language. (i)

Anarotieme [ [DP to vivlio tinos ]1 mu ipes pos dhiavases t1 ] I.wonder the book whose you said that you.read I wonder whose book you said you read. (Heck 2008: 244)

Following Heck (2008), I will adopt the view that such languages allow SpecDP to be rightward. That is, in these structures the postnominal possessor has not actually remained within the lexical projection NP, but rather has moved up into a (rightward) specifier of D. 17. However, it remains quite controversial whether these structures are indeed instances of CP pied-piping, and thus possess the structure indicated in (26). In English, they appear to be much more limited than the Basque and Quechua structures in (25). For example, the English structures seem to degrade when the ‘matrix’ subject is an R-expression (i), and the ‘matrix’ verbs allowing such putative CP pied-piping seem to be rather restricted ((ii) and (iii)).

NOTES TO PAGES

154–163

227

(i) ?? [ CP What’s in there]1 does James think? (ii) [CP What did he get ]1 did he say / *report? (iii) [CP What did he get ]1 does he think / *believe? Of course, if the structures in (26)–(28) are not CP pied-piping, and CP pied-piping does not exist in English, then our Q-based theory owes some account of its impossibility in English. In this context it might be worth considering again the work of Tanaka (1999). As mentioned in chapter 4, Tanaka (1999) defends a view very similar to our Q-based theory of pied-piping. Interestingly, for his own theory-internal reasons, Tanaka (1999) must rule out covert pied-piping of CPs in Japanese. In order to accomplish this, he introduces a stipulation whereby his correlate of our Q-particle (i.e., his ‘ω-head’) cannot take CPs as complements. If extended to English, this stipulation would also rule out overt CP pied-piping in English. Of course, given the wellformedness of CP pied-piping in Tlingit and other languages, one must assume that this property of Q idiosyncratically varies across languages. 18. Note that we need not assume that the actual trigger for movement of the whword is Agreement with Q. That is, the movement to SpecCP in (30) is assumed to be triggered by Agreement between the wh-word and the subordinate C. Such movement, however, is a necessary precondition for pied-piping of the subordinate CP. These details will be made clearer in section 5.4, which is specifically concerned with the ability for wh-fronting to feed pied-piping. 19. Heck (2004, 2008) also derives (31) from the basic transitivity of Agreement accessibility. 20. It is also apparent from (44) that such a structure would, in English, involve illicit pied-piping past a lexical projection, the projection of the verbal copula/ auxiliary. However, as noted in the following discussion, the fact that such structures are also ruled out by the more general QP-Intervention Condition is of greater interest, as it predicts that such structures should be impossible even in non-Agreement languages. 21. Although Tlingit does possess a verbal auxiliary, no phonological material is ever allowed to intervene between it and the main verb. Indeed, in some grammars this element is analyzed as a verbal suffix (Boas 1917: 83). Because of this tight phonological relation to the main verb, it is not possible to test (46a) for the Tlingit verbal auxiliary. 22. There are, however, two potential counterexamples. First, Heck (2008) notes that German (a Q/wh-Agreement language) allows pied-piping of past participles by a degree head modifying an adverb. (i) Fritz weiß [ [ wie schön geschrieben ] man haben muss . . . Fritz knows how well written one have must Fritz knows how well one must have written. . . (Heck 2008: 157) If these adverbials are contained within VP, then structures like (i) would seem to falsify (50). However, another possibility is that they are adjuncts to vP, in which case they are entirely consistent with (50):

228

NOTES TO PAGES

163–173

(ii) Fritz weiß [ [vP wie schön [VP geschrieben ] ] man haben muss . . . The second counterexample comes from Basque. Heck (2008: 134) notes that it is possible in Basque for complements of past participles to pied-pipe the entire participle phrase. (iii) [VP Nork idatzi ]-a da liburu hori? who.ERG written-ABS is book that By whom is that book written? (Heck 2008: 134) Given our claim that Basque is a Q/wh-Agreement language (section 5.2.3), this would appear to counterexemplify the generalization for Q/wh-Agreement languages in (50). There are two possible ways our Q-based account can approach these data. The most radical is to adopt the position of Arregi (2003b) that Basque is, contrary to appearance, not a wh-movement language. Another possibility is that examples like (iii) involve movement of the wh-word to Spec-vP: (iv) [QP [vP Nork1 [ v [VP t1 idatzi ] ] ] Q ] . . . Such ‘secondary wh-fronting’, similar to that seen in Basque examples like (25a), would position the wh-word within the same phase as the Agreeing Q-particle. Thus, if such movement were to take place, our account would indeed predict the possibility of structures like (iii). Of course, this analysis would raise the question of why such secondary whfronting cannot save English structures like (41b–d). Given the theory of secondary wh-fronting presented later in this chapter, a mechanical solution would be to stipulate that little-v/little-n/little-a generally cannot bear the feature ‘Wh[ ]’. Basque’s divergent behavior would be due to its exceptionally allowing little-v to bear ‘Wh[ ]’. 23. To my knowledge, all other purported functional heads (e.g., Foc, Agr, Num, Loc) are a result of ‘exploding’ the five basic functional categories listed here. 24. The reader may observe that in both German (62) and Irish (63), there is some kind of inversion between the wh-word and the preposition when the PP is pied-piped. Our Q-based account has no special explanation of this inversion. Note, however, that the example of Icelandic in (61) demonstrates that the optionality of PP pied-piping is not necessarily tied to such inversion. 25. Similarly, Heck (2008: 146) presents evidence that CP pied-piping in Basque (58) is an instance of ‘massive pied-piping’. Since Heck (2008) predicts that massive pied-piping (unlike ‘normal’ pied-piping) is optional, he thereby captures the data in (58) and (74). 26. Note that this ‘ambiguity account’ could also be adopted by Heck (2008) to capture (75). Indeed, Heck (2008: 289) explicitly considers such approaches, labeling them ‘pseudo-optionality’. 27. Although our account does easily capture the cases of CP pied-piping in (57)–(59).

NOTES TO PAGES

173–179

229

28. In fact, such a principle is employed by Hagstrom (1998), in his Q-based analysis of Japanese. 29. Note that such pied-piping is somewhat improved if the wh-word is in the second conjunct rather than the first. Thus (79a) and (80a) are noticeably better than (79b) and (80b). I have no account of this apparent contrast. 30. Technically, Heck (2008) proposes that certain cases of pied-piping—the cases of so-called massive pied-piping—are derived via an operation of feature movement, which qua movement, is naturally sensitive to the CSC. Note, however, that this account would analyze the pied-piping in (81) as an instance of ‘massive pied-piping’. Massive pied-piping, though, is by definition restricted to matrix clauses (see section 5.5), and so this account predicts that the pied-piping in (81) is not allowable in subordinate clauses. This, however, appears to be untrue. (i) I wonder [ [ where and when ] you saw him ]. (ii) I wonder [ [ whose books and whose magazines ] you sold ]. Heck (2008: 311) proposes that (i) and (ii) do not actually involve pied-piping of a coordination, but instead involve coordination of two separate clauses, followed by deletion within the first clause. It seems, though, that there is little justification for treating (i) and (ii) as a separate phenomenon from that in (79)–(81). 31. Note, however, that this assumption is highly controversial, and is directly challenged by the phenomenon of ‘First Conjunct Agreement (FCA)’. Babyonyshev (1996) argues that FCA shows that Agreement is not sensitive to the CSC in the way suggested by (83). On the other hand, Aoun et al. (1994, 1999) argue for a theory of FCA that is consistent with our condition in (83). 32. However, Haraguchi (1973) observes that head-internal relative clauses in Japanese seem to observe the CSC. Given the evidence that Japanese relatives are not derived by movement (Matsumoto 1997; Davis 2006), the data Haraguchi (1973) examines stand as a much broader puzzle. 33. Note, however, that this DegP inversion is also required for some Deg heads other than the wh-head how. For example, demonstrative Deg heads in English must also undergo this inversion. (i)

DegP Inversion With Demonstrative Deg Heads a. [DP [DegP That big ]1 [DP a t1 dissertation ] ] will never be widely read. b. * [DP A [DegP that big ] dissertation ] will never be widely read.

We see then that, contrary to our later discussion, DegP inversion is not crucially tied to the wh-feature of the Deg-head how. On the other hand, we will see in the following sections some clearer cases where secondary wh-fronting is crucially tied to the wh-feature. Since the facts surrounding these later examples are rather complicated, I choose not to employ them in this introductory section. For that reason, I opt to (somewhat misleadingly) use the simple case of DegP inversion in English to illustrate the basics of our account.

230

NOTES TO PAGES

180–185

34. Note, however, that something must also ensure that awh can only Agree with DegPs, and not, for example, postnominal possessors. If awh could Agree with postnominal possessors, we would incorrectly derive structures like (i). (i) * [DP Who1 [ awh picture of t1 ] ] did you buy? To my knowledge, this is also a weakness of all other theories of DegP inversion (e.g., Kennedy & Merchant 2000). As a possible solution, we might suppose that awh has some additional featural differences from ‘normal a’ that render it specifically ‘DegP-hungry’. 35. Indeed, we will presently see that secondary wh-fronting is predicted by our system to never be an instance of Q-movement. 36. Within a theory where pied-piping structures are derived via feature-percolation, such a stipulation might be that a wh-word can only ‘percolate’ its wh-feature once within a derivation (Coon 2009). Thus (99) is ruled out because the wh-possessor must percolate its wh-feature once when it moves to the specifier of iyotyoty, ‘house’, and then again when the pied-piped DP moves to the specifier of ijol, ‘roof’. Note, however, that this stipulation is problematic in a number of respects, as it would seem to rule out the cases of ‘recursive’ pied-piping in section 5.3.1. 37. Note that this prediction does not rule out cases of ‘recursive pied-piping’ like the English structures below. (i) [ [ whose dog’s ] leash ] is on the couch? Although they are identical in surface structure to the outlawed Chol structures in (99) and (103), they are not true cases of ‘roll-up’ pied-piping. This is because the fronting of the possessive phrase ‘whose dog’ is an independently available movement type in English, one that is not tied to the wh-feature of the possessor but rather targets the phi-features of the possessive phrase as a whole. Thus, the fronting of ‘whose dog’ in (i) is not necessarily a case of ‘pied- piping’ by ‘whose’, and so is not necessarily an instance of Q-movement. Similarly, Heck (2008: 49–50) raises some potential counterexamples to our prediction in (104), the most challenging of which are German structures like the following. (ii) Mann [CP [CP den1 [ zu t1 überzeugen ] ]2 dir t2 vorzunehmen ]3 ich dir t3 empfehle ]man who to convince you plan I you advise Man that I advise you to plan to convince. (Heck 2008: 50) In these structures, the relative pronoun den appears to first undergo secondary wh-fronting to the left edge of the embedded infinitive zu überzeugen, ‘to convince’. Importantly, the entire infinitive then seems to undergo secondary whw-fronting to the left edge of the larger embedded infinitive dir vorzunehmen, ‘you plan’, from which position it then appears to pied-pipe the entire subordinate clause den zu überzeugen dir vorzunehmen. It would appear, then, that these German structures involve a ‘roll-up’ structure like the kind in (104).

NOTES TO PAGES

185–194

231

However, before we can definitively come to this conclusion, we must be sure that the fronting of the infinitive den zu überzeugen, ‘who to convince’, is not a freely available movement option in German. If it is, then we need not assume that such fronting in (ii) is an instance of Q-movement, and so we need not conclude that (ii) involves ‘roll-up’. Heck (2008) does not discuss this possibility, the crucial test of which is whether structures like the following are independently possible as clauses in German. (iii) [CP [CP Hans1 [ zu t1 überzeugen ] ]2 dir t2 vorzunehmen ]3 ich dir t3 empfehle Hans to convince you plan I you advise I advise you to plan to convince Hans. 38. The Chol data here were provided to me by Jessica Coon (personal communication). 39. I thank Kirill Shklovsky for offering me his Russian judgments. 40. The Chol data here were provided to me by Jessica Coon (personal communication). That Tzotzil behaves in a parallel fashion was reported to me by Jessica Coon (personal communication), citing John Haviland (personal communication). 41. Note that this argument assumes that demonstratives in Russian must take NPs as complements rather than adjoin to them, contrary to our earlier proposals in chapter 4. We will later explain (in note 42) how the following account can be made consistent with our earlier proposals concerning possessor extraction in Russian. 42. Observe, however, that the structure in (113) assumes that possessive nominals in Russian can be DPs rather than bare NPs, contrary to our proposals in chapter 4. On the other hand, we might follow the suggestion made in section 5.3.4 that possessive nominals in Russian have the option of being DPs, as well as bare NPs. However, the reader will observe that the logic of the following account assumes that possessive nominals in CompPP are always DPs (and never bare NPs) in Russian. We might render this assumption consistent with our wider proposals by assuming that P in Russian selects for DPs. Consequently, possessive nominals in Russian must be DPs when complement to P. 43. It should be noted that Heck’s own (2008) theory of massive pied-piping requires a stipulation of equal complexity to (117) in order to derive generalization (116). Although I do not wish to review the details of Heck’s account here, the interested reader should note that Heck (2008: 332) ultimately derives (116) from a stipulation that features having undergone pure feature-movement are invisible for selection (although not any other syntactic operations). 44. Curiously, such massive pied-piping does not seem to be possible when the higher verbal category is an auxiliary. (i) No Massive Pied-Piping of Main Predicate When Auxiliary Is Present a. * [ Made what ] has John for his little sister? b. * [ Making what ] is John for his little sister?

232

NOTES TO PAGES

194–205

I have no explanation of this apparent contrast between the pied-piping in (i) and that in (130) and (131). 45. Heck (2008: 173–179) presents evidence that this constraint also holds in Italian and Russian. 46. Heck (2008: 305) presents evidence that this generalization also holds for Italian. 47. I use the term ‘path’ here in the sense of Pesetsky (1982). For a node A and a node B, where A c-commands B, the path between A and B is the set of nodes consisting of A, B, and the nodes dominating B but not dominating A. 48. As we see from (138), the chief benefit of our abandoning the distinction between specifiers and adjuncts is that it permits us to derive both (133) and (134) from the stipulation in (136). This abandonment, however, does come at a price. If we do not distinguish adjuncts from specifiers, then the generalization in (132) predicts that massive pied-piping past adjuncts should be as ill-formed as massive pied-piping past specifiers. As the following sentences show, however, this prediction is not so. (i)

Massive Pied-Piping Past Adjuncts Is Possible a. (?) [ To quickly fire who ] is important? b. (?) [ Nice pictures of which president ] did you buy?

Thus, unlike what we see for specifiers in (132), it is possible for massive pied-pipers to be c-commanded by adjuncts. I currently have no solution for this straightforward problem. 49. Admittedly, this is quite a ‘big if’, given the strong evidence that English infinitivals contain phonologically null PRO in the subject position. For exactly this reason, Heck (2008: 173) restricts his versions of (132)–(134) to overt specifiers. A similar adjustment could be made to our (132)–(134) and (136). I have avoided doing so simply to keep this exposition brief. Chapter 6 1. Similarly, our Q-based analysis of relativization in (7) is independently proposed by Sternefeld (2001). Sternefeld (2001) also develops a semantics for relative clauses based on this syntax. 2. However, unlike our analysis in (6), Horvath (2000, 2007b) claims that this focus-sensitive operator is adjoined to the fronted phrase, rather than heading that phrase. Thus the analysis in Horvath (2000, 2007b) differs from (6) in that it must still appeal to some mechanism of pied-piping. 3. Similarly, the work of Sternefeld (2001) demonstrates what role our putative ‘QREL’ might play in the semantics of relative clauses. We should note, however, that as pointed out by Heck (2008: 39), it might be difficult to extend a Q-based account to constructions such as topicalization, scrambling, and ‘Wackernagel-movement’ in German, each of which exhibits pied-piping and an inability to strand prepositions. The potential difficulty is that these movements seem not to possess any identifiable semantics. Therefore a Q-based account would have to assume that the Qs associated

NOTES TO PAGES

205–208

233

with these movements are semantically empty, which seems to weaken the Q-based theory. 4. Note that this hypothesis would not follow from the theory of the QP-Intervention Condition sketched in section 2.6.3.1. Thus the following proposals must assume a different theory of why the QP-Intervention Condition holds.

BIBLIOGRAPHY

Abels, Klaus. 2003. Successive cyclicity, anti-locality, and adposition stranding. Doctoral dissertation. Storrs, CT: University of Connecticut. Aissen, Judith. 1996. Pied piping, abstract agreement and functional projections in Tzotzil. Natural Language and Linguistic Theory 14: 447–491. Aoun, Joseph, Elabbas Benmamoun, and Dominique Sportiche. 1994. Agreement, word order and conjunction in some varieties of Arabic. Linguistic Inquiry 25: 195–220. Aoun, Joseph, Elabbas Benmamoun, and Dominique Sportiche. 1999. Further remarks on first conjunct agreement. Linguistic Inquiry 30: 669–681. Aronoff, Mark. 1976. Word formation in generative grammar. Cambridge, MA: MIT Press. Arregi, Karlos. 2003a. Clausal pied-piping. Natural Language Semantics 11: 115–143. Arregi, Karlos. 2003b. Focus on Basque movements. Doctoral dissertation. Cambridge, MA: Massachusetts Institute of Technology. Babyonyshev, Maria. 1996. Structural connections in syntax and processing: Studies in Russian and Japanese. Doctoral dissertation. Cambridge, MA: Massachusetts Institute of Technology. Baker, C. L. 1970. Notes on the description of English questions: The role of an abstract question morpheme. Foundations of Language 6: 197–219. Baker, Mark. 1999. On the interplay of the universal and the particular: Case studies in Edo. In Proceedings of the 35th Annual Meeting of the Chicago Linguistics Society, ed. by Sabrina J. Billings, John P. Boyle, and Aaron M. Griffith. Chicago: University of Chicago Press. Barss, Andrew, Ken Hale, Ellavina Tsosie Perkins, and Margret Speas. 1991. Logical form and barriers and Navajo. In Logical structure and linguistic structure, ed. by C.-T. James Huang and Robert May. Dordrecht: Kluwer Academic. Beck, Sigrid. 1996. Quantified structures as barriers for LF movement. Natural Language Semantics 4: 1–56. Beck, Sigrid. 2006. Intervention effects follow from focus interpretation. Natural Language Semantics 14: 1–56.

235

236

BIBLIOGRAPHY

Bennett, Paul. 1995. A course in Generalized Phrase Structure Grammar. London: University College London Press. Black, Lydia T., Nora Marks Dauenhauer, and Richard Dauenhauer. 2008. Classics of Tlingit oral literature, Volume 4: Anóoshi Lingít aaní ká, Russians in Tlingit America: The Battles of Sitka, 1802 and 1804. Juneau, AK: Sealaska Heritage Foundation. Boas, Franz. 1917. Grammatical notes on the language of the Tlingit Indians. University of Pennsylvania University Museum Anthropological Publications 8:1. Bobaljik, Jonathan. 2002. A-chains at the PF-interface: Copies and ‘covert’ movement. Natural Language and Linguistic Theory 20: 197–267. Boeckx, Cedric. 2009. On the locus of asymmetry in UG. Catalan Journal of Linguistics 8: 41–53. Borsley, Robert. 1992. More on the difference between English restrictive and non-restrictive relative clauses. Journal of Linguistics 28: 139–148. Bošković, Željko. 2005a. On the locality of left branch extraction and the structure of NP. Studia Linguistica 59: 1–45. Bošković, Željko. 2005b. On the locality of Merge and Agree. Cambridge, MA: MIT Working Papers in Linguistics. Bresnan, Joan. 1976. On the form and function of transformations. Linguistic Inquiry 7: 3–40. Bresnan, Joan, and Jane Grimshaw. 1978. The syntax of free relatives in English. Linguistic Inquiry 9: 331–391. Brody, Michael. 1997. Perfect chains. In Elements of grammar, ed. by Liliane Haegeman. Dordrecht: Kluwer Academic. Bury, Dirk, and Ad Neeleman. 1999. Projection of free relatives. Manuscript. University College London. Cable, Seth. 2006a. The interrogative words of Tlingit: An informal grammatical study. Manuscript. University of Massachusetts, Amherst. Retrieved July 25, 2009, from http:// people.umass.edu/scable/papers/Tlingit-Interrogatives-Grammar.pdf. Cable, Seth. 2006b. Syncope in the verbal prefixes of Tlingit: Meter and surface phonotactics. Muenchen: Lincom Europa. Cable, Seth. 2007. The grammar of Q: Q-particles and the nature of wh-fronting, as revealed by the wh questions of Tlingit. Doctoral dissertation. Cambridge, MA: Massachusetts Institute of Technology. Cable, Seth. 2008. Question particles and the nature of wh-fronting. In Quantification: A cross-linguistic perspective, ed. by Lisa Matthewson. North Holland Linguistic Series, Linguistic Variations, Vol. 64. Bingly, UK: Emerald. Campbell, Lyle. 1997. American Indian languages: The historical linguistics of Native America. Oxford: Oxford University Press. Carlson, Gregory N., and Francis Jeffry Pelletier. 1995. The generic book. Chicago: University of Chicago Press. Cheng, Lisa Lai-Shen. 1991. On the typology of wh-questions. Doctoral dissertation. Cambridge, MA: Massachusetts Institute of Technology. Cheung, Lawrence Y.-L. 2003. On certain (non-)violations of coordinate structure constraints. Snippets 12: 9–11. Chomsky, Noam. 1973. Conditions on transformations. In A festschrift for Morris Halle, ed. by Stephen Anderson and Paul Kiparsky. New York: Holt, Reinhart & Winston. Chomsky, Noam. 1975. Questions of form and interpretation. Linguistic Analysis 1: 75–109. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Step by step: Essays on minimalism in honor of Howard Lasnik, ed. by Roger Martin, David Michaels, and Juan Uriagereka. Cambridge, MA: MIT Press.

BIBLIOGRAPHY

237

Citko, Barbara. 2009. What don’t wh-questions, free relatives and correlatives have in common? In Correlatives Cross-Linguistically, ed. by Anikó Lipták. Amsterdam: John Benjamins Publishing. Cole, Peter, and Gabriella Hermon. 1981. Subjecthood and islandhood: Evidence from Quechua. Linguistic Inquiry 12: 1–30. Cole, Peter, and Gabriella Hermon. 1998. The typology of wh-movement: Wh-questions in Malay. Syntax 1: 221–258. Comorovski, Ileana. 1996. Interrogative phrases and the syntax-semantics interface. Dordrecht: Kluwer Academic. Coon, Jessica. 2009. Interrogative possessors and the problem with pied-piping in Chol. Linguistic Inquiry 40: 165–175. Corver, Norbert. 1990. The syntax of left branch extractions. Katholieke Universiteit Brabant. Corver, Norbert. 1992. Left branch extraction. In Proceedings of the Northeast Linguistics Society 22. Amherst, MA: Graduate Linguistic Student Publications. Corver, Norbert. 2007. Subextraction. In The Blackwell syntax companion (SYNCOM), ed. by Martin Everaert and Henk van Riemsdijk. Oxford: Blackwell. Cowper, Elizabeth. 1987. Pied-piping, feature percolation and the structure of the noun phrase. Canadian Journal of Linguistics 32: 321–338. Culicover, Peter. 1999. Syntactic nuts. Oxford: Oxford University Press. Dauenhauer, Nora Marks, and Richard Dauenhauer. 1981. Because we cherish you: Sealaska elders speak to the future. Juneau, AK: Sealaska Heritage Institute. Dauenhauer, Nora Marks, and Richard Dauenhauer. 1987. Classics of Tlingit oral literature, Volume 1: Haa shuká, Our ancestors: Tlingit oral narratives. Juneau, AK: Sealaska Heritage Institute. Dauenhauer, Nora Marks, and Richard Dauenhauer. 1990. Classics of Tlingit oral literature, Volume 2: Haa tuwunáagu yís, For healing our spirit: Tlingit oratory. Juneau, AK: Sealaska Heritage Institute. Dauenhauer, Nora Marks, and Richard Dauenhauer. 1994. Classics of Tlingit oral literature, Volume 3: Haa kusteeyí, Our culture: Tlingit life stories. Juneau, AK: Sealaska Heritage Institute. Dauenhauer, Nora Marks, and Richard Dauenhauer. 2000. Beginning Tlingit. Juneau, AK: Sealaska Heritage Institute. Dauenhauer, Nora Marks, and Richard Dauenhauer. 2002. Lingít x’éináx sá! Say it in Tlingit: A Tlingit phrase book. Juneau, AK: Sealaska Heritage Institute. Dauenhauer, Nora Marks, and Richard Dauenhauer. Forthcoming. Intermediate Tlingit. Juneau, AK: Sealaska Heritage Institute. Davis, Christopher. 2006. Evidence against movement in Japanese relative clauses. Handout from a talk presented at 2006 ECO5. University of Massachusetts, Amherst. Retrieved July 25, 2009 from: http://people.umass.edu/cmdavis/papers/daviseco5-handout.pdf. Dayal, Veneeta. 1996. Locality in wh-quantification: Questions and relative clauses in Hindi. Dordrecht: Kluwer Academic. Dryer, Matthew. 1985. Tlingit: An object-initial language? Canadian Journal of Linguistics 30: 1–13. Dürr, Michael, and Egon Renner. 1995. The history of the Na-Dene controversy: A sketch. In Language and culture in native North America: Studies in honor of Heinz-Jürgen Pinnow, ed. by Michael Dürr, Egon Renner, and Wolfgang Oleschinski. Muenchen: Lincom Europa. Edwards, Keri. Forthcoming. Tlingit verb dictionary project. University of Alaska, Fairbanks.

238

BIBLIOGRAPHY

Edwards, Keri, John Marks, Nora Marks Dauenhauer, and Richard Dauenhauer. 2005. Lingít x’éináx áx! Hear it in Tlingit: A Tlingit mini phrase book & CD. Juneau, AK: Sealaska Heritage Institute. Embick, David, and Alec Marantz. 2008. Architecture and blocking. Linguistic Inquiry 39: 1–53. Emonds, Joseph. 1976. A transformational approach to English syntax. New York: Academic Press. Emonds, Joseph. 1979. Appositive relatives have no properties. Linguistic Inquiry 10: 211–243. Enrico, John. 2003. Haida syntax. Lincoln, NE: University of Nebraska Press. Fabb, Nigel. 1990. The difference between English restrictive and non-restrictive relative clauses. Journal of Linguistics 26: 57–77. Falk, Yehuda N. 2001. Lexical-Functional Grammar: An introduction to parallel constraintbased syntax. Stanford, CA: CSLI Publications. Fanselow, Gisbert. 1991. Minimale syntax. Habilitation thesis. University of Passau. Fanselow, Gisbert. 1997. Minimal link effects in German (and other languages).” Manuscript. University of Potsdam. Featherston, Sam. 2005. Magnitude estimation and what it can do for your syntax: Some whconstraints in German. Lingua 115: 1525–1550. Fiengo, Robert, C.-T. James Huang, Howard Lasnik, and Tanya Reinhart. 1988. The syntax of wh-in-situ. In Proceedings of the Seventh West Coast Conference on Formal Linguistics (WCCFL 7), ed. by Hagit Borer. Stanford, CA: CSLI Publications. Fodor, Janet Dean, and Ivan A. Sag. 1982. Referential and quantificational indefinites. Linguistics and Philosophy 5: 355–398. Fox, Danny. 2000. Economy and semantic representation. Cambridge, MA: MIT Press. Gazdar, Gerald, Ewan Klein, Geoffrey Pullum, and Ivan Sag. 1985. Generalized phrase structure grammar. Oxford: Blackwell. Ginzburg, Jonathan, and Ivan Sag. 2000. Interrogative investigations: The form, meaning, and use of English interrogatives. Stanford, CA: CLSI Publications. Gordon, Raymond G., Jr. 2005. Ethnologue: Languages of the world, 15th ed. Dallas, TX: SIL International. Grimshaw, Jane. 1981. Form, function and the language acquisition device. In The logical problem of language acquisition, ed. by C. L. Baker and John McCarthy. Cambridge, MA: MIT Press. Grimshaw, Jane. 2000. Locality and extended projections. In Lexical specification and insertion, ed. by Peter Coopmans, Marin Everaert, and Jane Grimshaw. Amsterdam: John Benjamins Publishing. Groenendijk, Jeroen, and Martin Stokhof. 1982. Semantic analysis of WH-complements. Linguistics and Philosophy 5: 175–233. Groenendijk, Jeroen, and Martin Stokhof. 1984. Studies on the semantics of questions and the pragmatics of answers. Doctoral dissertation. University of Amsterdam. Groos, Anneke, and Henk van Riemsdijk. 1981. Matching effects in free relatives: A parameter of core grammar. In Theory of markedness in generative grammar, ed. by Andrea Belletti. Pisa: Scuola Normale Superiora. Hagstrom, Paul. 1998. Decomposing questions. Doctoral dissertation. Cambridge, MA: Massachusetts Institute of Technology. Hamblin, C. L. 1958. Questions. Australasian Journal of Philosophy 36: 159–168. Hamblin, C. L. 1973. Questions in Montague English. Foundations of Language 10: 41–53. Haraguchi, Shosuke. 1973. “Dragging” reconsidered. Linguistic Inquiry 4: 95–97. Heck, Fabian. 2004. A Theory of Pied Piping. Doctoral dissertation. Universität Tübingen.

BIBLIOGRAPHY

239

Heck, Fabian. 2008. On pied-piping: Wh-movement and beyond. Berlin: Mouton de Gruyter. Heck, Fabian. 2009. On certain properties of pied-piping. Linguistic Inquiry 40: 75–111. Heim, Irene. 1994. Interrogative semantics and Karttunen’s semantics for ‘know.’ In Proceedings of the 9th Annual Conference of the Israel Association for Theoretical Linguistics, ed. by Rhonna Buchall and Anita Mittwoch. Somerville, MA: Cascadilla Press. Heim, Irene, and Angelika Kratzer. 1998. Semantics in generative grammar. Oxford: Blackwell. Holmberg, Anders. 2008. The syntax of the Finnish question particle. Manuscript. University of Newcastle. Retrieved July 25, 2009, from http://ling.auf.net/lingbuzz/000596. Horvath, Julia. 1986. FOCUS in the theory of grammar and the syntax of Hungarian. Dordrecht: Foris. Horvath, Julia. 2000. Interfaces vs. the computational system in the syntax of focus. In Interface strategies, ed. by Hans Bennis, Martin Everaert, and Eric Reuland. Amsterdam: Royal Netherlands Academy of Arts and Sciences. Horvath, Julia. 2007a. Pied-piping. In The Blackwell syntax companion (SYNCOM), ed. by Martin Everaert and Henk van Riemsdijk. Oxford: Blackwell. Horvath, Julia. 2007b. Separating ‘focus movement’ from focus. In Phrasal and Clausal Architecture: Syntactic Derivation and Interpretation; In Honor of Joseph E. Emonds, ed. by Simin Karimi, Vida Samiian and Wendy K. Wilkins. Amsterdam: John Benjamins Publishing. Huddleston, Rodney D. and Geoffrey K. Pullum. 2002. The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. Ishihara, Roberta. 1984. Clausal pied-piping: A problem for GB. Natural Language and Linguistic Theory 2: 397–418. Jackendoff, Ray. 1977. X-bar syntax: A study of phrase structure. Cambridge, MA: MIT Press. Jacobson, Pauline. 1995. On the quantificational force of English free relatives. In Quantificational in natural languages, ed. by Emmon Bach, Eloise Jelinek, Angelika Kratzer, and Barbara Partee. Dordrecht: Kluwer Academic. Jelinek, Eloise. 1984. Empty categories, case and configurationality. Natural Language and Linguistic Theory 2: 39–76. Karttunen, Lauri. 1977. Syntax and semantics of questions. Linguistics and Philosophy 1: 3–44. Kayne, Richard. 1983. Connectedness. Linguistic Inquiry 14: 223–249. Kayne , Richard. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press. Kayne, Richard. 2000. Parameters and universals. Oxford: Oxford University Press. Kennedy, Christopher, and Jason Merchant. 2000. Attributive comparative deletion. Natural Language and Linguistic Theory 18: 89–146. Kishimoto, Hideki. 2005. Wh-in-situ and movement in Sinhala questions. Natural Language and Linguistic Theory 23: 1–51. Kiss, Katalin E. 1995. Discourse configurational languages. Oxford: Oxford University Press. Ko, Heejeong. 2005. Syntax of why-in-situ: Merge in [Spec, CP] in the overt syntax. Natural Language and Linguistic Theory 23: 867–916. Koopman, Hilda. 2000. Prepositions, postpositions, circumpositions and particles: The structure of Dutch PPs. In The Syntax of Specifiers and Heads: Collected Essays of Hilda J. Koopman. London: Routledge. Koopman, Hilda, and Anna Szabolcsi. 2000. Verbal complexes. Cambridge, MA: MIT Press. Kratzer, Angelika. 1991. The representation of focus. In Semantics: An international handbook of contemporary research, ed. by Arnim von Stechow and Dieter Wunderlich. New York: Walter de Gruyter.

240

BIBLIOGRAPHY

Kratzer, Angelika. 1998. Scope or psuedoscope? Are there wide-scope indefinites? In Events in Grammar, ed. by Susan Rothstein. Dordrecht: Kluwer. Kratzer, Angelika. 2006. Exhaustive questions. Handout from a talk presented at the Massachusetts Institute of Technology on May 12, 2006. University of Massachusetts, Amherst. Kratzer, Angelika, and Junko Shimoyama. 2002. Indeterminate pronouns: The view from Japanese. In Proceedings of the 3rd Tokyo Conference on Psycholinguistics, ed. by Yukio Otsu. Tokyo: Hituzi Syobo. Kuno, Susumu, and Jane J. Robinson. 1972. Multiple wh-questions. Linguistic Inquiry 3: 463–487. Lasnik, Howard, and Mamoru Saito. 1992. Move-alpha: Conditions on its application and output. Cambridge, MA: MIT Press. Leer, Jeff. 1991. The schetic categories of the Tlingit verb. Doctoral dissertation. University of Chicago. Leer, Jeff. 1993. The Tlingit anaphoric system and its roots in Na-Dene. Manuscript. Fairbanks, AK: Alaska Native Language Center. Leer, Jeff. 2000. The negative/irrealis category in Athabaskan-Eyak-Tlingit. In The Athabaskan languages: Perspectives on a Native American language family, ed. by Theodore Fernald and Paul Platero. Oxford: Oxford University Press. Leer, Jeff, Doug Hitch, and John Ritter. 2001. Interior Tlingit noun dictionary: The dialects spoken by Tlingit elders of Carcross and Teslin, Yukon, and Atlin, British Columbia. Whitehorse: Yukon Native Language Centre. Lewis, David. 1970. General semantics. Synthese 22: 18–67. Li, Charles N. 1976. Subject and topic. New York: Academic Press. Matsumoto, Yoshiko. 1997. Noun modifying constructions in Japanese. Amsterdam: John Benjamins Publishing. Matthewson, Lisa. 1998. Determiner systems and quantificational strategies: Evidence from Salish. The Hague: Holland Academic Graphics. Matthewson, Lisa. 1999. On the interpretation of wide-scope indefinites. Natural Language Semantics 7: 79–134. May, Robert. 1985. Logical form: Its structure and derivation. Cambridge, MA: MIT Press. Mithun, Mariane. 1999. The languages of native North America. Cambridge: Cambridge University Press. Naish, Constance. 1966. A syntactic study of Tlingit. Language data: Amerindian Series 6. Dallas, TX: SIL International. Nanni, Deborah, and Justine Stillings. 1978. Three remarks on pied piping. Linguistic Inquiry 9: 310–318. Newman, Paul, and Martha Ratliff. 2001. Linguistic fieldwork. Cambridge; Cambridge University Press. Nichols, Johanna. 1986. Head-marking and dependent-marking grammar. Language 62: 56–119. Nishigauchi, Taisuke. 1990. Quantification in the theory of grammar. Dordrecht: Kluwer Academic. Nyman, Elizabeth, and Jeff Leer. 1993. Gágiwdul.àt: Brought forth to reconfirm. The legacy of a Taku River Tlingit clan. Fairbanks, AK: Alaska Native Language Center. Pesetsky, David. 1982. Paths and categories. Doctoral dissertation. Cambridge, MA: Massachusetts Institute of Technology. Pesetsky, David. 1987. Wh-in-Situ: Movement and unselective binding. In The representation of (in)definiteness, ed. by Eric Reuland and Alice G. B. ter Meulen. Cambridge, MA: MIT Press. Pesetsky, David. 2000. Phrasal movement and its kin. Cambridge, MA: MIT Press.

BIBLIOGRAPHY

241

Pesetsky, David, and Esther Torrego. 2001. T-to-C movement: Causes and consequences. In Ken Hale: A life in language, ed. by Michael Kenstowicz. Cambridge, MA: MIT Press. Pesetsky, David, and Esther Torrego. 2007. The syntax of valuation and the interpretability of features. In Phrasal and clausal architecture: Syntactic derivation and interpretation, ed. by Simin Karimi, Vida Samiian, and Wendy K. Wilkins. Amsterdam: John Benjamins Publishing. Postal, Paul. 1972. Two remarks on dragging. Linguistic Inquiry 3: 130–136. Quinn, Steven. 2007. Perseverance to do ‘Macbeth’ in Tlingit.” Associated Press article. Retrieved July 25, 2009, from http://www.sealaskaheritage.org/news/news_article_tlingit _macbeth_2_2007.htm. Reinhart, Tanya. 1992. Wh-in-situ: An apparent paradox. In Proceedings of the Eighth Amsterdam Colloquium, ed. by Paul Dekker and Martin Stokhof. Amsterdam: ILLC. Reinhart, Tanya. 1995. Interface strategies. Utrecht: OTS Working Papers in Linguistics. Reinhart, Tanya. 1997. Quantifier scope: How labor is divided between QR and choice functions. Linguistics and Philosophy 20: 335–397. Reinhart, Tanya. 2006. Interface strategies: optimal and costly computations. Cambridge, MA: MIT Press. Richards, Norvin. 1997. What moves where in which language? Doctoral dissertation. Cambridge, MA: Massachusetts Institute of Technology. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of grammar, ed. by Liliane Haegeman. Dordrecht: Kluwer Academic. Rooryck, Johan. 1994. Generalized transformation and the wh- cycle: Free relatives as bare wh-CPs. Groninger Arbeiten zur Germanistischen Linguistik 37: 195–208. Rooth, Mats. 1985. Association with focus. Doctoral Dissertation. University of Massachusetts, Amherst. Rooth, Mats. 1992. A theory of focus interpretation. Natural Language Semantics 1: 75–116. Ross, John Robert. 1967. Constraints on variables in syntax. Doctoral dissertation. Cambridge, MA: Massachusetts Institute of Technology. Ross, John Robert. 1979. Wem der Kasus schlagt. Linguistiche Berichte 63: 26–32. Rullmann, Hotze, Lisa Matthewson, and Henry Davis. 2008. Modals as distributive indefinites. Natural Language Semantics 16: 317–357. Ruys, Eddy. 1992. The Scope of Indefinites. Doctoral dissertation. University of Utrecht. Ruys, Eddy. 2000. Weak crossover as a scope phenomenon. Linguistic Inquiry 31: 513–539. Safir, Kenneth. 1986. Relative clauses in a theory of binding and levels. Linguistic Inquiry 17: 663–689. Safir, Kenneth. 2009. Viable syntax: Rethinking minimalist architecture. Manuscript. New Brunswick, NJ: Rutgers University. Sag, Ivan A., Inbal Arnon, Bruno Estigarribia, Philip Hofmeister, T. Florian Jaeger, Jeanette Pettibone, and Neal Snider. 2006. Processing accounts for Superiority effects. Manuscript. Stanford University. Retrieved July 25, 2009, from http://lingo.stanford.edu/sag/ papers/squib.pdf. Samarin, William J. 1967. Field linguistics: A guide to linguistic fieldwork. New York: Holt, Rinehart & Winston. Sauerland, Uli, and Fabian Heck. 2003. LF-intervention effects in pied-piping. In Proceedings of NELS 33, ed. by Makoto Kadowaki and Shigeto Kawahara. Amherst, MA: GLSA. Schwarzschild, Roger. 2002. Singleton indefinites. Journal of Semantics 19: 289–314. Sells, Peter. 1985. Pied-piping and the feature [WH]. Manuscript. Stanford University. Sealaska Heritage Institute. 2003. Sealaska Heritage Institute website. Retrieved July 25, 2009, from http://www.sealaskaheritage.org.

242

BIBLIOGRAPHY

Sharvit, Yael. 1998. Possessive wh-expressions and reconstruction. In Proceedings of NELS 28, ed. by Pius Tamanji and Kiyomi Kusumoto. Amherst, MA: GLSA. Shimoyama, Junko. 2001. Wh-constructions in Japanese. Doctoral dissertation. University of Massachusetts, Amherst. Sternefeld, Wolfgang. 2001. Partial movement constructions, pied-piping and higher order choice functions. In Audiatur vox wapientiae—A festschrift for Arnim von Stechow, ed. by Caroline Fery and Wolfgang Sternefeld. Berlin: Akademie Verlag. Story, Gillian. 1966. A morphological study of Tlingit. Language data: Amerindian Series 7. Dallas, TX: SIL International. Story, Gillian. 1995. An analyzed Tlingit procedural text. In Language and culture in native North America: Studies in honor of Heinz-Jürgen Pinnow, ed. by Michael Dürr, Egon Renner, and Wolfgang Oleschinski. Muenchen: Lincom Europa. Story, Gillian, and Constance Naish. 1973. Tlingit verb dictionary. Fairbanks, AK: Alaska Native Language Center. Story, Gillian, Constance Naish, Henry Davis, and Jeff Leer. 1976. Tlingit noun dictionary. Fairbanks, AK: Sheldon Jackson College. Swanton, John, R. 1909. Tlingit myths and texts. Washington, DC: Bureau of American Ethnology. Reprinted 1970. Hamburg, MI: Native American Book Publishers. Swanton, John R. 1911. Tlingit. In Handbook of American Indian languages, Part 1, ed. by Franz Boas. Washington, DC: Bureau of American Ethnology. Szabolcsi, Anna. 1994. The noun phrase. In The syntactic structure of Hungarian, ed. by Ferenc Kiefer, Katalin E. Kiss, and Lyn Frazier. San Diego: Academic Press. Tanaka, Hidekazu. 1998. Conditions on logical form derivations and representations. Doctoral dissertation. Montreal: McGill University. Tanaka, Hidekazu. 1999. LF wh-islands and the minimal scope principle. Natural Language and Linguistic Theory 17: 371–402. Thompson, Chad. 1996. The Na-Dene middle voice: An impersonal source of the D-element. International Journal of American Linguistics 62: 351–378. Uriagereka, Juan. 1988. On government. Doctoral dissertation. Storrs, CT: University of Connecticut. Vajda, Edward. 2009. A Siberian link with the Na-Dene. Archeological Papers of the University of Alaska 6: 75–156. van Riemsdijk, Henk. 1978. A case study in syntactic markedness: The binding nature of prepositional phrases. Lisse: Peter de Ridder Press. van Riemsdijk, Henk. 2007. Free relatives. In The Blackwell syntax companion (SYNCOM), ed. by Martin Everaert and Henk van Riemsdijk. Oxford: Blackwell. Velten, Henry. 1939. Two southern Tlingit tales. International Journal of American Linguistics 5: 65–74. Velten, Henry. 1944. Three Tlingit stories. International Journal of American Linguistics 10: 168–180. Veniaminov, Ioann. 1846. Zamechaniya o Koloshenskom i kad’yakskom Yasykakh. St. Petersburg: Academia Nauk. von Stechow, Arnim. 1996. Against LF pied piping. Natural Language Semantics 4: 57–110. Watanabe, Akira. 1992. Wh-in-situ, subjacency and chain formation. Cambridge, MA: MIT Working Papers in Linguistics. Watanabe, Akira. 2006. The pied-piper feature. In Wh-movement: Moving on, ed. by Lisa Lai-Shen Cheng, and Norbert Corver. Cambridge, MA: MIT Press. Webelhuth, Gert. 1992. Principles and parameters of syntactic saturation. Oxford: Oxford University Press.

BIBLIOGRAPHY

243

Williams, Frank, Emma Williams, and Jeff Leer. 1978. Tongass texts. Fairbanks, AK: Alaska Native Language Center. Winter, Yoad. 1997. Choice functions and the scopal semantics of indefinites. Linguistics and Philosophy 20: 399–467. Yatsushiro, Kazuko. 2001. The distribution of mo and ka and its implications. In Formal approaches to Japanese linguistics 3, ed. by Maria Christina Cuervo, Daniel Harbour, Ken Hiraiwa, and Shinichiro Ishihara. MIT Working Papers in Linguistics 41. Cambridge, MA: MIT Working Papers in Linguistics. Yoshida, Keiko, and Tomoyuki Yoshida. 1996. Question marker drop in Japanese. ICU Language Research Bulletin 11: 37–54.


INDEX

adposition stranding in English, 108, 110–112 in German, 168 in Icelandic, 168, 171 in Irish, 168–169 theory of, 10–11, 55–62, 105–112 Ancash Quechua pied-piping of complementizer phrases (CPs), 154–156, 164–165, 168 Basque pied-piping of complementizer phrases (CPs), 154–156, 164, 167–168 choice function, 68 Chol, 181–189 P-inversion in, 186–189 pied-piping of possessors, 181–189 of prepositional phrases (PPs), 186–189 roll-up pied-piping structures, impossibility of, 181–186 secondary wh-fronting, 181–189 Coordinate Structure Constraint in Japanese, 175–176 in pied-piping, 142, 173–176, 193

DPs, splitting of. See Split DPs Edo, 103–105 pied-piping of possessors, 104 Q-particles Interaction with prepositions, 104 Interaction with possessors, 104 English adposition stranding in, 108, 110–112 free relatives, 206–209 LF-Focus Intervention Effects, 123–128, 131–135, 136–140 pied-piping of auxiliaries, 159–162, 193–194 of complementizer phrases (CPs), 154–155 of coordinate structures, 173–175 DegP inversion in, 177–181 in focus movement, 200–202 of gerunds, 153 massive pied-piping in, 190–198 of nominalizations, 153 pied-piping past islands, 144, 148 pied-piping past lexical categories 144, 149, 151–153, 176–181, 190–198

245

246

INDEX

English (continued) of possessors, 153, 174–175, 177–178 of predicates, 159–160, 162, 193–194 of prepositional phrases (PPs), 171–173, 192–193 in relative clauses, 190–198, 201–202 secondary wh-fronting in, 176–181 Superiority Effects, 123–128, 131–135 feature percolation vs. feature projection, 211 n.1 as Agreement, 119 as feature movement, 119–120 focus semantics, 64–66 focus-movement, 200–205 in Hungarian, 202–205 pied-piping in, 200–205 free relatives in English, 206–209 matching effects in, 207–209 in Tlingit, 206–207 French split DPs in, 169 Full Interpretation, 75 German adposition stranding in, 168 Agreement between Q-particles and wh-words, 145–147 LF-Focus Intervention Effects, 122–124, 128–131, 135–137 pied-piping of DegPs, 169–170 of possessors, 152 of prepositional phrases (PPs), 168 recursive pied-piping, 156–157 separation constructions with DegPs, 169–170 with was . . . für, 169 Intervention Effects in, 135–136 optional alternation with pied-piping, 169–170 split DegPs, 169–170 Superiority Effects, 123, 128–131 Greek pied-piping of possessors, 169 split DPs, 169 Hungarian focus movement, 202–205

pied-piping of possessors, 152–153 Icelandic, adposition stranding in, 168, 171 pied-piping of prepositional phrases (PPs), 151, 168, 171 intensional type theory, 66 Intervention Effects. See LF-Focus Intervention Effects Irish adposition stranding in, 168–169 pied-piping of prepositional phrases (PPs), 168–169 Japanese Coordinate Structure Constraint in, 175–176 Q-particles Agreement with wh-words, 145–147 interaction with determiners, 92–93 interaction with possessors, 92 interaction with postpositions, 91 position in main clauses, 89–93 presence with wh-words, 31 structural relation with wh-word, 33 wh-indefinites illustrative examples of, 31, 91–93 semantics of, 93–94 wh-questions semantics of, 94–96 Korean LF-Focus Intervention Effects in, 97–99 Q-particles interaction with determiners, 92–93 interaction with possessors, 92 interaction with postpositions, 91 position in main clauses, 89–93 wh-indefinites, 91–93 left branch extraction. See Possessor extraction, Split DPs lexical categories, syntax of, 149 LF-Focus Intervention Effects in D-linked questions, 131–133 in English, 123–128, 131–135, 136–140 in German, 122–124, 128–131, 135–136, 136–137 in Korean, 96–99 in pied-piping, 123–124, 136–140

INDEX

in separation constructions, 135–136 in ternary wh-questions, 133–135 methodology of study, 20–21 Mohawk split DPs, 169 Na-Dene language phylum, 15 P-stranding. See adposition stranding Phase Impenetrability Condition, 149, 153–156 pied-piping of auxiliaries, 159–165, 193–194 of complementizer phrases (CPs) in Ancash Quechua, 154–156, 164–165, 168 in Basque, 154–156, 164, 167–168 in English, 154–156 in Tlingit, 32, 36, 39, 164, 167 of coordinate structures, 173–175, 193 coordinate structure constraint in, 173–175, 193 definition of, 6, 116 elimination of, 8–9, 115–121 as empirical problem, 4–6 in focus-movement, 200–205 of gerunds, 153 left-edge, relevance of, 165–166 LF-Focus Intervention Effects in, 136–140 limited pied-piping, 141–142, 144, 147–148 limited pied-piping language, 141–142, 145, 147–148 massive pied-piping, 137–138, 142, 172–173, 190–198 of adjuncts, 195–198 of complement clauses, 197 of coordinate structures, 193 generalization on, 190 of lexical categories, 191–193 of predicates, 194 of relative clauses, 197–198 of specifiers, 195–198 of modifiers, 152 optionality of, 166–173 pied-piping past islands in Tlingit, 143, 148 in English, 144, 148

247

pied-piping past lexical categories in Tlingit, 144, 149 in English, 144, 149, 150–153 pied-piping structure definition of, 6, 116 of possessors in Chol, 181–186 in Edo, 104 in English, 5, 106, 118, 120, 156, 177–178, 200–202 in German, 152, 157, in Greek, 169 in Hungarian, 152–153 in Russian, 172–173 in Tlingit, 44–45, 47, 57, 81–83, 105 in Tzotzil, 181–186 of predicates, 159–165, 192, 193–194 of prepositional phrases or postpositional phrases (PPs) in Chol, 186–189 in English, 5, 116, 171–173, 192–193, 207 in German, 157, 168 in Icelandic, 151, 168, 171 in Irish, 168–169 in Russian, 106, 151, 165, 186, 189 in Tlingit, 8, 31, 44, 56, 105, 116 in Tzotzil, 186–189 recursive pied-piping, 142, 156–159 in relative clauses, 190–198, 200–202 repair generalization, 168 secondary wh-fronting in, 142, 176–189 in Chol, 181–189 DegP Inversion in, 177–181 in English, 176–181 P-inversion in, 186–189 roll-up pied-piping structures, impossibility of, 184–185 in Tzotzil, 181–189 semantics of, 80–83, 121–122 possessor extraction in Greek, 169 in Russian, 108–111, 169, 172 theory of, 55–62, 105–111 preposition stranding. See adposition stranding Principle of Interpretability, 73, 97 Q-adjunction languages definition of, 86

248

INDEX

Q-based syntax for focus movement, 200–205 of wh-in-situ languages, 37–38, 85–86 learning theoretic considerations for, 9, 102–103 main empirical results of, 11–12 for relative clauses, 200–202 for wh-movement, 6–9, 13, 38, 85, 100, 141, 199 Q-particle Agreement with wh-words, 145–147 in focus movement. See Q-based syntax in Japanese. See Japanese semantics of, 67–68 semantic effect of position, 71–73 in relative clauses. See Q-based syntax in Sinhala. See Sinhala in Tlingit. See Tlingit uniqueness of, 74, 97 in wh-question formation. See Q-based syntax Q-projection languages definition of, 86 Quechua. See Ancash Quechua QP Intervention Condition, 55–63 relativization, 200–202 free relatives. See free relatives pied-piping in, 202 roll-up pied-piping structures, impossibility of, 184–186 Russian pied-piping of possessors, 172–173 of prepositional phrases (PPs), 106, 151, 165, 186, 189 split DPs, 108–111, 169, 172–173 Sealaska Heritage Institute, 16–17, 20–21 selection S-selection, 62 C-selection, 62 Sinhala Q-Particles interaction with determiners, 88–89 interaction with islands, 33 interaction with possessors, 88–89 interaction with postpositions, 88–89 presence with wh-words, 31 structural relation with wh-word, 32–33

position in main clauses, 35, 87–88 position in subordinate clauses, 35–36, 87–88 wh-indefinites illustrative examples of, 31 semantics of, 93 wh-questions semantics of, 94 split DPs in French, 169 in German, 169 in Greek, 169 in Mohawk, 169 possessor extraction. See possessor extraction in Russian, 108–111, 169, 172–173 theory of, 108–111 Superiority Effects in D-linked questions, 131–133 in English, 123–128, 131–135 in German, 123, 128–131 in ternary wh-questions, 133–135 in Tlingit, 29–30, 128 Tlingit dialectics, 15 free relatives, 206 geneological relationships, 15 geographic distribution, 15 impossibility of adposition stranding, 43–62, 105 impossibility of determiner extraction, 43–62, 106 impossibility of possessor extraction, 43–62, 105 morphology, 18–19 phonology, 17 pied-piping of complementizer phrases (CPs), 32, 36, 39, 164, 167 illustrative examples of, 7–8, 31–32, 33–36, 116, 143 pied-piping past islands, 143, 148 pied-piping past lexical categories, 144, 149–150 of possessors, 44–45, 47, 57, 81–83, 105 of predicates, 162–164 of postpositional phrases, 8, 31, 44, 56, 105, 116

INDEX

semantics of, 80–83, 121–122 syntax of, 115–118 prior scholarship, 16–17 Q-particles interaction with determiners, 45–46, 50–51 interaction with islands, 7–8, 33–35 interaction with possessors, 44–45, 49–50 interaction with postpositions, 44, 48–49 presence with wh-words, 30–31, 73–76 structural relation with wh-word, 31–32, 73–76 position in main clauses, 35, 58–59 position in subordinate clauses, 36, 59 similarities to Q-particles in Japanese and Sinhala, 36 relative clauses, 206–207 revitalization of, 15–16 Superiority Effects, 29–30, 128 vitality of, 15–16 word order freedom, 19 wh-indefinites illustrative examples of, 10, 31, 48–51, 53–54 receiving specific readings, 53–54 scoping out of islands, 54–55 semantics of, 70 wh-words inventory of, 22 as indefinites, 10, 31, 48–51, 53–54 wh-questions evidence for Q-based analysis of, 36–43 fronting of wh-word in, 23–26 long-dstance movement in, 28–29

249

multiple wh-questions, 29–30 schema of structure, 4, 22 semantics of, 76–83 topicalization in, 26–28 wh-indefinites in Japanese, 31, 91–93 in Korean, 91–93 in Sinhala, 31 in Tlingit. See Tlingit semantics of, 70, 93–94 wh-questions classic theory of, 4–6, 101 D-linked wh-questions LF-Focus Intervention Effects in, 131–133 Superiority Effects in, 131–133 in Chol. See Chol. in Edo. See Edo. in English. See English. in German. See German. in Japanese. See Japanese. in Korean. See Korean. multiple wh-questions, 122–136 Q-based theory of. See Q-based syntax in Russian. See Russian. in Sinhala. See Sinhala ternary wh-questions LF-Focus Intervention Effects in, 133–135 Superiority Effects in, 133–135 in Tlingit. See Tlingit wh-words as indefinites. See wh-indefinites. in questions. See wh-questions semantics of, 66–67

The Grammar of Q: Q-Particles, Wh-Movement, and Pied-Piping

The Grammar of Q: Q-Particles, Wh-Movement, and Pied-Piping (Oxford Studies in Comparative Syntax)

The Phantasma of Q

The Book of Q

The Phantasma of Q - - - -

The Phantasma of Q - - - -

Q

Q

Q

Q

Q

q²

Q

Q

Q

The Adventures of Q-Bert

The Q Continuum 02 Q-Zone

The Q Continuum 01 Q-Space

The Adventures of Q-Bert

Hide and Q

The Grammar of Graphics

The Philosophy of Grammar

The Grammar of Names

The grammar of graphics

The Grammar of Graphics

The Grammar of Names

The Grammar of Justice

The Grammar of Names

The Grammar of Names

The Grammar of Words

The Philosophy of Grammar

The Grammar of Q: Q-Particles, Wh-Movement, and Pied-Piping