Parameters of Slavic Morphosyntax (Oxford Studies in Comparative Syntax)

PARAMETERS of SLAVIC MORPHOSYNTAX

OXFORD STUDIES IN COMPARATIVE SYNTAX Richard Kayne, General Editor Principles and Parameters of Syntactic Saturation Gert Webelhuth Verb Movement and Expletive Subjects in the Germanic Languages Sten Vikner Parameters and Functional Heads: Essays in Comparative Syntax Edited by Adriana Belletti and Luigi Rizzi Discourse Configurational Languages Edited by Katalin E. Kiss Clause Structure and Language Change Edited by Adrian Battye and Ian Roberts Dialect Variation and Parameter Setting: A Study of Belfast English and Standard English Alison Henry Parameters of Slavic Morphosyntax Steven Franks

PARAMETERS of SLAVIC MORPHOSYNTAX Steven Franks

New York Oxford OXFORD UNIVERSITY PRESS 1995

Oxford University Press Oxford New York Athens Auckland Bangkok Bombay Calcutta Cape Town Dar es Salaam Delhi Florence Hong Kong Istanbul Karachi Kuala Lumpur Madras Madrid Melbourne Mexico City Nairobi Paris Singapore Taipei Tokyo Toronto and associated companies in Berlin Ibadan

Copyright © 1995 by Steven Franks Published by Oxford University Press, Inc. 200 Madison Avenue, New York, New York 10016 Oxford is a registered trademark of Oxford University Press, Inc. 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 Franks, Steven. Parameters of Slavic morphosyntax / Steven Franks, p. cm.—(Oxford studies in comparative syntax) Includes bibliographical references and index. ISBN 0-19-508970-7—ISBN 0-19-508971-5 (pbk.) 1. Slavic languages—Syntax. 2. Slavic languages—Case. 3. Principles and parameters (Linguistics) 4. Slavic languages— Grammar, Generative. I. Title. II Series. PG207.F7 1995 491.8—dc20 94-16800

246897531 Printed in the United States of America on acid-free paper

Foreword

In order to fully appreciate the contribution of Steven Franks's book Parameters of Slavic Morphosyntax to both Slavic and theoretical linguistics, it is necessary to understand the relation between generative theory and Slavic linguistics in the United States since 1957. There is virtually complete agreement now that the publication of Noam Chomsky's Syntactic Structures in 1957 was a revolution: linguistics would never be the same again. In the early years, the focus of attention was discovering the rules of English syntax, the assumption being that it was sufficient to thoroughly analyze a single language in order to understand the essential syntactic structure of any human language. A result of this approach was that the subject matter of syntactic analysis was dictated by and large by the structure of English; the specific problems of highly inflected languages like the Slavic languages (e.g., case) were more or less ignored. It was assumed that case assignment, agreement phenomena, and so on were relatively superficial phenomena that could be accounted for by late, superficial rules. Thus, in Chomsky's Aspects of the Theory of Syntax (1965), only a few pages at the end of the book are devoted to inflectional processes (e.g., p. 170); one German example is discussed. While the rapid development of generative theory all but ignored the kinds of problems that are at the heart of the Slavic languages, the field of Slavic linguistics itself, which was dominated by the theories of Roman Jakobson, proceeded with little or no concern for the goals and accomplishments of generative theory. Jakobsonian theory, with its extensive use of privative binary oppositions, was concerned primarily with phonological analysis and the feature decomposition of grammatical categories; it did not lend itself to syntactic analysis and little if anything of lasting value dealing with Slavic syntax was produced in Jakobson's framework. Thus the concerns of generative theory and Slavic linguistics were far removed: each developed without regard for the other—to the detriment of both. The rapprochement of Slavic linguistics and generative theory began with a

VI

FOREWORD

number of books and articles in the early 1970s (for details, see R. Brecht and C. Chvany [eds.], Slavic Transformational Syntax, [1974]). But the realization that data from the Slavic languages could make a major contribution to generative theory came from a shift in the theory of generative grammar itself. Starting with Chomsky's Lectures on Government and Binding (1980), the focus of attention shifted from language-particular rules to the search for universal principles of language structure. The major hypothesis underlying this work is that these basic principles are innate. Now, if the basic principles underlying the structure of all human languages are innate, the following question naturally arises: How are we to account for the language diversity that obscures the universal foundation of human language? What has come to be known as Principles and Parameters theory accounts for language diversity in terms of parameterization of the universal principles alluded to above. Upon hearing the language spoken by those who surround him/her, the child "learns" his/her native language by setting certain parameters. For example, simple facts of word order allow the setting of the headedness parameter, which is responsible for the typologically prominent distinction between leftbranching and right-branching languages. The setting of these parameters has wideranging effects on the language's structure. Principles and Parameters theory recognizes that the way to test these hypotheses is to look at a wide variety of languages and to determine whether the differences we observe between them can be accounted for in terms of parametric variation. This has produced a renaissance in comparative grammar. A crucial role in this research is played by the study of closely related languages: Do the differences among them conform to the predictions of Principles and Parameters theory? It is in this context that Slavic linguistics and generative theory have found a common ground. Steven Franks's book Parameters of Slavic Morphosyntax is, to the best of my knowledge, the first attempt to analyze a number of the salient typological syntactic properties of the Slavic languages in the Principles and Parameters framework. It is for this reason that I consider his book so important. It is obvious that it will be a major contribution to Principles and Parameters theory. What may be less obvious is its contribution to the field of Slavic linguistics: it explains the basic notions of Principles and Parameters theory with admirable clarity and will therefore serve as an introduction to the generative study of Slavic syntax for a new generation of Slavic linguists. LEONARD H. BABBY

Pieface

The past decade has seen a remarkable spurt of activity in the areas of theoretical and comparative syntax. Syntacticians in various frameworks have sought to impose theoretically consistent analyses of diverse phenomena from a broad range of languages, in the hope of establishing the mechanisms and limitations of the human linguistic capacity. The impetus for this work was largely due to the "principles and parameters" model of Chomsky's (1981a) Government and Binding (GB) theory, and it is within this general framework that the present monograph is conceived. One language group that has received little attention from this perspective is Slavic. Even a comprehensive and influential study like Baker (1988) mentions Slavic languages only in passing, and with regard to a single construction. Yet this language group offers an especially valuable source of insight into parametric variation. It is my intention to demonstrate this in the pages to follow, where Slavic data will be used to test and refine theoretical proposals. The research reported here rests on the belief that analyses of Slavic languages should be of significance to general linguists in having implications for modifying their theories, and also that such theories ought to assist Slavists by shedding light on how to analyze particular problems in the grammars of Slavic languages. These two goals complement each other; they are opposite sides of a mutual concern with discovering how language works. This book is thus intended for Slavic and general linguists alike. As such, although it assumes a familiarity with the basic concepts of GB theory, it explores recent developments within the theory that should be of interest to Slavists, and it makes Slavic data available to syntacticians, developing new analyses within the confines of GB theory. It is hoped that the result will contribute to our understanding of how the abstract structures and principles of the human language faculty interact to calculate the range of linguistic forms and meanings found in the Slavic languages, and that it will promote communication among linguists investigating diverse aspects of Slavic syntax.

Vlll

PREFACE

The Language Faculty It has long been recognized that the ability to learn language sets human beings apart from the other animals, but we are far from understanding what makes the miracle of language acquisition possible. Given that children accomplish this complex feat in just a few years, regardless of individual intelligence and typically in the absence of explicit instruction, one promising approach is to attribute as many facets of language learning as possible to our innate biological endowment. In other words, the distinctly human capacity to ' 'project'' a grammar from primary language data can be explained in terms of some kind of language-learning cognitive system. It is this special "language faculty" that renders us, through a little understood combination of maturation and stimulation, mentally capable of developing linguistic competence. Linguistic research traditionally consists in the uncovering of systematic patterns within and across languages. Generative grammarians argue that these patterns, in order to satisfy minimal criteria of learnability, must in some way derive from inherent properties of the human language faculty. In this view, as articulated in, for example, Chomsky (1986b), the fundamental goal of cognitive linguistics is to discover the nature of "knowledge of language." A grammar is an internalized system of rules and representations by which we compute and manipulate linguistic structures. By studying the properties of grammars, we may learn about the structure of our language faculty. A grammatical analysis should therefore be regarded as a theory of mind—it makes specific empirical claims about abstract mental representations. Although in principle there is a broad range of evidence open to scrutiny for testing these claims, in practice native speaker intuitions offer the most prolific source of information. That is, facts garnered through introspection into our linguistic knowledge can be analyzed into grammatical systems, and these systems can be studied in turn for general organizational principles. Linguistic theory thus defines a research program for investigating the language faculty, and this methodology for seeking the principles that regulate grammars currently provides our best avenue into the workings of the human mind.

Principles, Parameters, and Parametric Variation Linguistic analysis reveals the existence of general, unifying principles that regulate the operation and output of grammatical processes. Most work within syntactic theory is directed toward the identification and elaboration of these principles, the content of which will be introduced as needed in the discussion of specific issues in the syntax of Slavic languages. Treating such principles as absolute universals, however, leads to an obvious dilemma: How can the rigidity of universals be reconciled with the variation of individual grammars? One method of accomplishing this is to claim that, although the principles hold universally, details of their imple-

PREFACE

IX

mentation may vary. According to the Principles and Parameters model, we are able to acquire language because the human language faculty contains a universal set of general principles subject to a highly restricted degree of parametric variation (cf. Chomsky, 1981b). Minor parametric distinctions, since they pertain to broadly interacting principles, proliferate throughout the grammar, sometimes with major structural ramifications. Language "learning," at least with respect to the central or "core" aspects of grammar, thus reduces to the acquisition of lexical items and the fixing of the values of these parameters. Although many open questions remain, especially regarding the status of maturational factors, the relative significance of general and language-specific cognitive capacities, the accessibility of input data, and the roles played by markedness theory and the core/periphery dichotomy in language acquisition, this approach has defined a new and promising research agenda for cognitive linguistics, providing impetus for much exciting and productive work. Within the Principles and Parameters paradigm, the comparison of grammars of related languages has come to play a key role. As Chomsky (1981a, 6) observes, while deep study of a single language may yield "principles of explanatory force," study of closely related languages is particularly valuable for the opportunities it affords to identify the parameters that permit variation in the proposed principles. In trying to flesh out and delimit the ways in which grammars may vary, it is often vital to compare similar phenomena within a closely knit group of languages. The Slavic languages, which display both curious differences and suspiciously persistent similarities in their syntax, constitute a particularly fertile testing ground for theoretical proposals about potential parametric variation.

Overview The primary goal of this study is to explore parametric accounts of syntactic diversity among the Slavic languages, unifying analyses of similar phenomena across the different languages. Most of the problems to be addressed deal with issues of phrase structure and case assignment. It will consequently be crucial to consider how the principles of GB's abstract "case theory," which have mostly been proposed on the basis of languages with impoverished morphology, can be extended to accommodate the intricate morphological case phenomena found in languages with rich inflectional systems. Although the grammars of both types of language are superficially disparate, the Principles and Parameters approach mandates that they have a common and innate basis. I will therefore need to develop a general model of case assignment flexible enough to handle a range of constructions within diverse languages. Just such a model is proposed in chapter 2, laying the groundwork for the specific issues to be investigated in subsequent chapters. Chapters 3 through 8 deal mostly with traditional problems in Slavic syntax, some of which have engendered a considerable body of insightful analysis. This work, however, has not generally been carried out within a comprehensive theory of language and, more importantly, typically lacks a concern for explaining variation (although, of course, there are notable exceptions, such as the work of Corbett,

X

PREFACE

Rudin, or Riizicka). Yet the tenets of parametric theory make it essential that Slavic linguists hold their analyses of one language ultimately responsible also for clarifying the data of another, since a crucial measure of the validity of any analysis is its ability to handle variation. The present study therefore attempts to integrate particular investigations of specific problems in the various languages into a conceptually and analytically unified account. My point of departure in the pages to follow will depend on the nature of the construction at hand and on existing analyses, although to some extent I will be concerned with extending available treatments of Russian to the other Slavic languages, with primary focus within South Slavic on Serbo-Croatian and within West Slavic on Polish. Chapter 1 offers an introduction to GB theory, and chapter 2 extends this framework in order to accommodate case phenomena in Slavic, discussing the problem of morphosyntactic features in depth. Chapter 3 considers acrossthe-board dependencies in Russian and Polish in light of the proposed system. Chapters 4 and 5 then examine the morphosyntax of quantified structures, addressing such questions as why numerals sometimes agree with the nouns they modify and sometimes govern them and why verbs sometimes agree with quantified subjects and sometimes do not. Chapter 6 deals with secondary predication. It studies the conditions under which predicate adjectives agree with their antecedents and the factors determining their form and distribution when they do not. Chapter 7 investigates a host of issues relating to null subject phenomena and draws attention to important differences among the Slavic languages, leading to a new approach to null subjects. Chapter 8 is concerned with problems of voice in Slavic and the roles of the reflexive and passive morphemes in the various languages. Finally, chapter 9 offers a brief summary and discussion of the kinds of parameters considered in my attempt to accommodate these dimensions of variation in Slavic morphosyntax. Bloomington, Ind. June 1994

S. F.

Acknowledgmen ts

This book is the product of many years' labor. The actual writing began in 1990, while I was on a Mellon post-doc at the University of Pennsylvania, but the thinking follows a line of argumentation about Slavic morphosyntax that I have been developing for a dozen years. During that time, it has been my good fortune to interact with linguists too numerous to name. Here I can merely acknowledge the impact a handful of otherwise innocent individuals have had on the contents of this tome. First and foremost, I wish to thank Wayles Browne, my dissertation advisor at Cornell, for his unflagging efforts at keeping me honest. Your knowledge of Slavic linguistics is peerless; your readiness to share that knowledge, boundless. Two other esteemed Slavic linguists with whom I have been lucky enough to study are Len Babby and Charlie Townsend, both of Princeton. Without your influence I would never have written this book. Several colleagues and valued friends who have helped me better to understand the material in these pages are Katarzyna Dziwirek, George Fowler, Gerry Greenberg, Catherine Rudin, and especially Ljiljana Progovac. I am indeed honored to know people like you. During my years at Indiana University, many students have contributed to my ongoing comparison of morphosyntactic structures in the various Slavic languages. Among these I would like to single out Stephen Dickey, Annie Joly Sperling, Curt Woolhiser, Michael Yadroff, and above all Martina Lindseth. You took my ideas seriously enough to disagree with them, and were usually right. Thanks for keeping me on my toes! Much of the material in this book has appeared elsewhere in various guises. I am indebted to my coauthors Katarzyna Dziwirek, Gerry Greenberg, and Norbert Hornstein for allowing me to "borrow" from our joint publications. I thank Elsevier Science Publishers for material that originally apeared in Lingua as Franks (1990c) and Franks (1992), AATSEEL and the editors of Slavic and East European Journal for permission to use portions of Greenberg and Franks (1991), and the MIT Press for granting permission to use material that appeared in Linguistic Inquiry as Franks (1993). I also appreciate the willingness of the International Journal of Slavic

Xll

ACKNOWLEDGMENTS

Linguistics and Poetics, Chicago Linguistic Society, and the Journal of Slavic Linguistics to let me draw freely from the following publications: Franks (in press); Franks (1982), Franks and House (1982), and Franks and Greenberg (1998); Franks and Dziwirek (1993). Finally, portions of Franks (1990a), which appeared in Russian Linguistics, and Franks and Hornstein (1992), which appeared in the volume Control and Grammar, are reprinted by permission of Kluwer Academic Publishers. Specific places where portions of chapters have been published are also acknowledged in those chapters. Finally, my family has suffered many hours of patient exile while I struggled with the manuscript. I wish I could tell my parents, Cyril and Violet, who have always been supportive beyond credulity; my enduring wife, Karen, who understands everything that is not in this book; and my miraculous children, Julia, David, and Elisabeth, who are the joy of my existence, that it will never happen again. All I can say is that even linguistics without you would be meaningless. You have blessed my life with love.

Contents

1. Introduction 1.1. Theoretical Preliminaries 1.1.1. The Modular Conception of Grammar 1.1.2. Levels and Components 1.1.3. Subsystems of Principles 1.1.4. The Categorial Component 1.2. Case Theory 1.2.1. Principles of Abstract Case 1.2.2. Case Assignment 2. Matrices, Indices, and Morphosyntactic Features 2.1. A Model of Morphological Case 2.1.1. Case Submatrices 2.1.2. Coindexation 2.1.2.1. PROJECTION 2.1.2.2. AGREEMENT

3 3 4 4 5 9 10 10 12 16 16 16 19 22 23

2.1.2.3. THETA-ROLES AND CASE

26

2.1.3. Phase Structural Case 2.2. Russian Case Features 2.2.7. Excursus on Jakobson's Features 2.2.2. Revising Jakobson's System

30 41 42 48

3. Across-the-Board Dependencies 3.1. Case-Matching 3.2. Parallelism in Prominence

61 62 64

XIV

CONTENTS

3.2.1. The Nature of the Parallelism 3.2.2. The Scope of Prominence Effects 3.2.3. Some Possible Extensions 3.3. The Parasitic Gap Analysis 3.3.1. Similarities 3.3.2. Differences 3.4. Conclusion 4. Quantified Structures: Russian versus Serbo-Croatian 4.1. Case Properties of Numeral Phrases 4.1.1. Babby's Analysis of Russian 4.1.2. Extending the Analysis: Gen(Q) Is Inherent in Serbo-Croatian 4.1.3. Characterizing the Structural!Inherent Dichotomy 4.2. The Category of Numeral Phrases 4.2.1. Pesetsky's Analysis of Russian 4.2.2. Extending the Analysis: Serbo-Croatian Numeral Phrases Are NPs 4.2.3. Some Semantic Issues 4.3. The Distribution of Numeral Phrases 4.3.1. The Internal Subject Hypothesis 4.3.2. The NP/QP Dichotomy Revisited 5. Quantified Structures: Polish and Other Puzzles 5.1. West Slavic and the Accusative Restriction 5.1.1. Polish Numeral Phrases 5.1.2. Numeral Phrases in Other Languages 5.2. More Quantified Expressions 5.2.7. Distributive Fo-Phrases and the Structure of DP

64 69 75 77 77 83 86 93 94 94 97 103 106 106 113 115 118 118 119 130 131 131 135 139 139

5.2.1.1. GOVERNMENT PATTERNS AND INITIAL PROPOSALS

140

5.2.1.2. THE STRUCTURAL DATIVE ANALYSIS

146

5.2.1.3. THE DP ANALYSIS

149

5.2.1.4. SERBO-CROATIAN PO-PRASES

157

5.2.1.5. POLISH PO-PHRASES

160

5.2.1.6. OTHER WEST SLAVIC LANGUAGES

5.2.2. Approximative Inversion 5.2.3. Frozen Quantifiers 5.3. Bare Genitives 5.3.1. Empty Quantifier Structures 5.3.1.1. PARTITIVES 5.3.1.2. GENITIVE INITIAL SENTENCES 5.3.2. The Genitive of Negation 5.3.2.1. RUSSIAN VERSUS POLISH 5.3.2.2. CONCLUSION: A CURIOUS CORRELATION

164

165 174 179 180 180 186 195 196 205

CONTENTS

6. Secondary Predication 6.1. Predicate Adjectives 6.7.7. Three Types of Predicate Adjective 6.1.2. On Agreement 6.1.3. Clausal Functional Projections

XV

220 220 221 224 226

6.2. Secondary Predication and Control 6.2.7. Case Transmission

234 235

6.2.3.1. TWO THEORIES 6.2.3.2. GOVERNED PRO 6.2.3.3. A SUBJECT-ORIENTATION APPROACH 6.2.3.4. FURTHER PREDICTIONS 6.3. The Second Dative 6.3.7. The Agreement Analysis

238 240 245 247 249 249

6.2.2. Factors Blocking Case Transmission 6.2.3. Control Theory and Agreement

6.3.2. A Phrase Structural Analysis 6.3.3. Gerunds and Participles

6.4. Parametric Variation

6.4.1. Non-Agreeing Semipredicatives 6.4.2. Dative Subjects 6.4.2.1. INFINITIVES 6.4.2.2. IMPERSONALS 6.4.3. Secondary Predication in Polish

7. Null Subject Phenomena

236 238

256 259

267 267 269 270 272 276

287

7.1. The "Pro-Drop" Parameter 7.7.7. Types of Null Subjects

287 289

7.2. Null Thematic Subjects 7.2.7. Case and Agreement

306 306

7.1.2. Parametric Approaches 7.1.3. Summary of Slavic Facts 7.1.4. Two Sides of Visibility

7.2.2. Ellipsis 7.2.3. Why Russian Is Different

290 297 304

307 309

7.3. Expletives and Visibility 7J.7. Null Expletives Do Not Need Case

312 312

7.4. Overt Expletives in Slavic

319

7.3.2. Overt Expletives Need Case at S-Structure 7.3.3. Different Kinds of Null Expletives

7.4.1. Russian 7.4.2. South and West Slavic

7.5. Conclusion: Arbitrary Third Plural Subjects

313 318

319 322

323

XVI

CONTENTS

8. Voice Alternations 8.1. Voice 8.1.1. Predicate-Argument Structure 8.1.2. Standard Passive Constructions 8.2. Null Subjects and Passive 8.2.1. Si-Constructions in Italian 8.2.2. Some Slavic Variations 8.3. Passive Morphology, Case, and Theta-Theory 8.3.1. Expletive Subjects and Passive 8.3.2. Case Absorption Issues 8.3.3. Theta-Theoretic Issues 8.3.4. Phrase Structure Issues 8.4. Epilogue: The Dispositional Reflexive Construction 9. Summary and Conclusions

333 334 334 336 339 339 340 347 347 347 351 355 364 374

REFERENCES

379

NAME INDEX

395

SUBJECT INDEX

399

PARAMETERS of SLAVIC MORPHOSYNTAX

This page intentionally left blank

1 Introduction

This chapter establishes the theoretical background for the problems in Slavic syntax to be examined in the subsequent chapters. I begin by surveying the general concepts of Government and Binding (GB) theory. A brief presentation of the principles of abstract case as embodied in GB theory follows, establishing the background for the formalism for representing and assigning morphological case proposed in chapter 2.

1.1. Theoretical Preliminaries Generative grammar holds that the projection problem can be solved by postulating a sufficiently rich and specific language faculty. That is, there is a complex and dedicated system1 that enables children eventually to acquire a mature grammar solely on the basis of exposure to an unstructured and unreliable fragment of the adult language. It is this deficiency in the primary linguistic data that leads linguists to conclude that "much of the child's final ability is determined by genetically encoded principles, which are triggered or activated by environmental stimulus" (Homstein and Lightfoot, 1981, 13). As pointed out by Hornstein and Lightfoot (1981, 9), there are really three aspects of the poverty of stimulus argument to be contended with. First of all, the speech a child is actually exposed to "does not consist uniformly of complete grammatical sentences," in that it is fraught with the effects of performance factors. Second, a child is ultimately "able to deal with an infinite range of novel sentences," far exceeding those that might in fact have been heard. These first two aspects, Hornstein and Lightfoot contend, would require an inductive theory of language acquisition to be extremely elaborate. Third, for much of what a speaker knows of the structure of his or her language, there is no evidence at all in the primary linguistic data. Linguists are able to devise complicated hypothetical sentences and elicit consistent grammatical judgments, even though the child in the course of acquiring a

3

4

PARAMETERS OF SLAVIC MORPHOSYNTAX

grammar may never have experienced similar sentences. Moreover, speakers are able to reject sentences as ill-formed without ever having been instructed that they are ungrammatical. This third aspect of the deficiency of data poses an insurmountable problem for any account of language "learning" based exclusively on experience, since the very evidence from which induction might be made is lacking. It is for this reason that linguists postulate a language faculty that embodies innate properties facilitating the acquisition of grammar. The initial configuration of the language faculty is known as Universal Grammar (UG). As a "language acquisition device," UG must be sufficiently rich to allow children to acquire the grammar of any human language, but restrictive enough to explain how they are able to construct a grammar relying on deficient data. The Principles and Parameters theory is one type of solution to this dilemma. In discovering the abstract principles of UG and establishing the ways in which they may be parameterized, however, one is of course torn between the desire to describe a wide range of phenomena and the need to explain such phenomena (cf. Chomsky, 1981b, 43). A model of grammar that can resolve this tension must have a rich deductive structure, with highly integrated principles and limited options. In this section, the basic properties of the GB framework, which aims to satisfy these criteria, are sketched out. 1.1.1. The Modular Conception of Grammar One way of dealing with the complexity of UG, regarded as a set of systems of principles and parameters, is to divide it into modules. This is the approach taken within GB theory. There are two distinct aspects of viewing UG as a collection of modules. First, a grammar consists of several rule systems, each a separate subcomponent pertaining to a distinct level of representation. Second, there are subsystems of principles that interact with these rule systems. The core grammar of a language is then the specification of some of the options of these modular subtheories. Each of these is examined in turn. 1.1.2. Levels and Components The so-called T-model of core grammar, which reflects standard assumptions about the organization of grammar (cf. e.g. Chomsky and Lasnik, 1977, 431), is shown in (1). (1)

D—Structure S—Structure

Phonological Form (PF)

Logical Form (LF)

INTRODUCTION

5

Each of these four levels of representation is meant to embody different kinds of information about the structure of a sentence. A grammar must provide every sentence with a structure at every level. Crucially, the model in (1) makes the empirical claim that PF and LF are autonomous subsystems. Associated with the levels in (1) are various components. The PF component contains rules that operate on S-structures to derive PF representations. Similarly, the LF component contains rules that generate LF representations from S-structure ones. So, for example, a constituent might delete in PF without having any effect on the sentence's LF representation, or move in LF without altering its PF representation. The transformational component maps D-structures into S-structures. In the detailed grammars that characterized early transformational-generative grammar, all sorts of complex rules were proposed, often obfuscated by complicated conditions on these rules. The subsequent shift in the focus of research from the properties of rules to the regularities of rule systems, however, gradually led to the reduction of the transformational component to a single rule, Move a. This has been made possible through the introduction of general principles that regulate the operation and output of grammatical processes. Movement leaves behind a coindexed trace that, together with the moved element, forms a "chain." Traces are syntactically present but phonologically null or "empty" categories. These silent elements are required by the principles of the theory of thematic roles, to be discussed in the next section. All of the levels of representation in (1) are phrase structure trees, which are themselves the products of a phrase structure grammar or "categorial" component. Just as with transformational rules, although earlier analyses often specified a detailed categorial component, it has now been shown that most, if not all, of its properties follow from independently motivated principles of grammar. Lastly, a grammar must include a lexical component, or "lexicon." The lexicon contains the lexical items that serve as terminal elements of the phrase structure trees. In it must be specified all idiosyncratic information about lexical items (words, morphemes, and sometimes even phrases; cf. Di Sciullo and Williams, 1987), including phonological, semantic, and morphosyntactic information, as well as all special properties that help to determine their distribution.

1.1.3. Subsystems of Principles Research into rule systems has made it possible to formulate increasingly general and abstract principles, the study of which is central to GB theory. Because they seem to pertain to different aspects of the grammar, they may be viewed as falling into discrete subsystems. This, then, is the second way in which UG can be said to be organized in modules. Chomsky (1981a, 5) enumerates the following subsystems of principles: bounding theory, binding theory, control theory, government theory, theta-theory, and case theory. These contain principles that interact with rules of grammar to restrict their operation. The assumption of principles in UG allows for extremely general rules, eliminating the need to append specific conditions.

6

PARAMETERS OF SLAVIC MORPHOSYNTAX

Bounding theory deals with how far an item may be moved in the syntax (i.e. on the way from D-structure to S-structure). It consists of a single principle— Subjacency. Subjacency states that an element may move over at most one "bounding node," where what counts as a bounding node may vary from language to language, but is generally seen to include S and/or S', NP, and in many languages also PP (cf. e.g. van Riemsdijk, 1978; Sportiche, 1981; and Rizzi, 1982). Although this variation in bounding nodes has often been cited as a paradigm example of parameterization, Rizzi (1989) has recently observed that this is an atypical and thus somewhat suspect instance of parametric variation. Binding theory delimits the possible antecedents for pronouns (e.g. him) and anaphors (e.g. himself), stipulating a domain (not necessarily identical; cf. e.g. Huang, 1983) in which anaphors must be bound and pronouns cannot be. Referring expressions (R-expressions; e.g. John or the man) and variables (traces of wh and other operator movement), on the other hand, cannot have an argument antecedent in any syntactic domain (cf. Chomsky, 1981a, 188). These three principles are summarized in (2), where the precise nature of "domain D" is left unspecified.2 (2)

a. Principle A: An anaphor must be bound in domain D. b. Principle B: A pronoun must be free in domain D. c. Principle C: An R-expression must be free everywhere.

"Bound" is defined as "coindexed with a c-commanding antecedent," and "free" essentially means "not bound." One standard definition of "c-command" is that the first branching node dominating the c-commanding element also dominates the c-commanded one, as in Reinhart (1983). Another common one, now known as "mcommand," is that all maximal (i.e. phrasal) projections are shared, as proposed in Aoun and Sportiche (1983). Chomsky (1986a, 8) points out that these are distinct and independently necessary notions, in that while m-command is required in the definition of government, c-command still seems useful for the binding theory. It should also be noted that within GB the terms "anaphor" and "pronoun" are technically defined in terms of the features [±anaphoric, ±pronominal], as are all nominal expressions. (3) a. anaphor = [ + anaphoric, -pronominal] b. pronoun = [-anaphoric, -(-pronominal] c. R-expression, variable = [ — anaphoric, —pronominal] The binding principles in (2) are thus properly stated in terms of the features in (3).3 This accords with the view that all elements in a phrase structure tree are formally matrices of morphosyntactic features, and that all relevant operations and conditions must be stated in terms of these features. Control theory is the set of principles that determine the referent of the empty category PRO, an abstract and caseless element that typically occupies the positions of subject of infinitive and gerund. Most attempts to devise a specific control theory within GB involve a reworking of the binding theory to incorporate control (cf. e.g. Bouchard, 1984; Manzini, 1983; Sportiche, 1983). It has also been assimilated to the generalized binding theory of Aoun (1986) in Hornstein and Lightfoot (1987). I shall

INTRODUCTION

7

adopt a similar approach in chapter 6, where the interaction between control and secondary predication is examined. The concept of government is more of a theme that unites the different subsystems of GB than a theory in and of itself. Government involves the relationship between a head of a phrase and its dependents. This relationship, familiar from traditional grammar, runs throughout the subsystems of principles in GB. Chomsky (1981a) argues that government is relevant for the theories of binding, theta-roles and case, while others employ it to determine obligatory control (e.g. Bouchard, 1984; Manzini, 1983; Franks and Hornstein, 1992) and the bounding nodes for subjacency (e.g. Kayne, 1981a). Various formalizations of the notion of government exist, mostly based on the c-command relationship, depending on how and in which modules it is being applied. One common definition of government is provided by Chomsky (1986a, 9), who roughly states that one node governs another node if it m-commands it and there is no intervening barrier (with considerable debate as to what constitutes a "barrier" and how "intervening" is to be construed). There is general consensus, however, about the existence of one principle of broad explanatory power that comes under the purview of government theory. This is the Empty Category Principle (ECP), which requires the trace left behind by Move a (i.e. [ - pronominal] empty categories) to be "properly" governed. The effect of this condition is to prohibit movement from subject and adjunct positions, which are generally not properly governed, although special mechanisms are available in some languages for circumventing this. Quite a variety of formulations of proper government have been proposed, depending on the range of phenomena the ECP is intended to account for as well as general theoretical assumptions. Most of them regard proper government as an instance of government in which the governor must be a lexical category. Theta-theory and especially case theory will play much more significant roles in this study than any of the other modules. These are basically concerned with how semantic roles, known in GB as theta-roles, and abstract case are assigned to arguments. I conclude this section with a sketch of the major properties of thetatheory, deferring the discussion of case theory until section 1.2. Chomsky (1986b, 98 ff.) argues that a reasonable property of LF is that it obey a ' 'principle of full interpretation,'' meaning that every element must be appropriately licensed and, in particular, play some semantic role.4 Individual arguments are associated with particular theta-roles, such as agent, theme (or patient), experiencer, instrument, and goal, and verbs are regarded as having a predicate-argument structure in that they take arguments with specific theta-roles (cf. Grimshaw, 1990, or Jackendoff, 1990a, for theories of argument structure). At LF, then, the theta-roles borne by the arguments of a verb must satisfy the requirements of that verb's predicate-argument structure. This one-to-one pairing between roles and arguments is the Theta-Criterion, which can be stated as follows (Chomsky, 1981a, 36): (4) Each argument bears one and only one theta-role, and each theta-role is assigned to one and only one argument. This principle is not merely a criterion of adequacy for LF, however, but holds rather at all levels of the syntax. The requirement that the pairing of theta-roles and

8

PARAMETERS OF SLAVIC MORPHOSYNTAX

syntactic positions mandated by (4) remain consistent throughout the derivation is known as the Projection Principle, one statement of which is as follows (Pesetsky, 1982, 17):5 (5) Representations at each syntactic level (LF, D-structure, S-structure) are projected from the Lexicon in that they observe the theta-marking properties of lexical items. The conjunction of the Theta-Criterion and the Projection Principle has a variety of pervasive effects on the grammar. Together, they restrict syntactic representations and constrain the operation of rules in important ways. For one thing, these principles help to predict the existence and inventory of empty categories that has been postulated on other grounds (cf. e.g. Chomsky, 1975). In particular, the association of a theta-role with a position implies that the position be occupied, even if not by any overt element. Moreover, because the Projection Principle extends this association to all syntactic levels, phonologically empty categories will also be present at all levels. The result is that movement from a theta-position necessarily leaves behind a trace. In addition, in order not to violate the Theta-Criterion, movement must always be to a non-theta position, otherwise the argument chain would be assigned multiple theta-roles. Consider NP-movement in the simple passive and raising structures in (6), where boldface e represents an empty category, in this instance the trace of the book. (6) a. the bookj was being read e, (by Bill) b. the bookj seems B! to be interesting The subject of neither a passive nor raising verb is in a theta-position, since the agent in (6a) can be expressed using the by-phrase and (6b) can be paraphrased as it seems that the book is interesting. Hence, movement is possible to these positions. Of course, this conclusion only makes sense if the book and e are regarded as a single entity for the purposes of theta-theory. That is, the Theta-Criterion is properly defined on chains of coindexed elements sharing a single theta-role. Only one member of a chain, the one occupying the argument's original D-structure position, will actually be in a position to which a theta-role is assigned. In this sense, then, D-structure is a pure representation of the association between grammatical functions (e.g. subject, object, indirect object) and theta-roles (e.g. agent, theme, goal). In sum, theta-theory, with its two guiding principles (4) and (5), has great explanatory value, deriving among other things the following properties of the levels in (1): (7)

a. S-structure and LF are enriched with traces. b. D-structure is a pure representation of the pairing between grammatical functions and theta-roles. c. Movement is to non-theta positions only.

INTRODUCTION

9

1.1.4. The Categorial Component The remainder of this chapter deals with case. Before addressing case theory, however, it is necessary to outline the properties of the phrase structure component, which plays such a vital role in determining case relationships. GB assumes the framework of "X-bar syntax" to constrain the categorial component. This theory, as originally put forward in Chomsky (1970), has two essential aspects. First of all, part-of-speech categorial labels are actually abbreviations for feature complexes. The major parts-of-speech-—noun (N), verb (V), adjective (A), and preposition (P)—can be defined in terms of the features [ ± N(ominal)] (or "substantive") and [±V(erbal)] (or "predicative"), as follows: (8)

—V P +N N

-N

+V V A

Breaking words down into features in this manner facilitates the expression of crosscategorial generalizations (cf. e.g. Stowell, 1981, for discussion). The second basic property of X-bar syntax is that phrasal nodes are viewed as projections of terminal nodes, so that relationships between constituents can be hierarchically represented. The degree of projection of a node X is represented by the number of prime nodes, and the maximal projection of a node X is the phrasal node XP (or X"). A typical phrase will permit a specifier (subject) and complement (object), as shown in (9). X"

(9)

X'

Specx

X

Compx

Specx and Compx are not themselves category symbols; instead, they simply stand for the sets of categories that can serve as the specifier or complement of X. X-bar syntax in and of itself does not specify the order of constituents in the schema in (9). Rather, it separates dominance relationships from precedence ones, restricting the former but leaving the latter to be independently determined.6 Whether phrases are left-headed or right-headed in a particular language is subject to parametric variation. The child must therefore establish this, presumably on the basis of verb-object order. It has been proposed (cf. e.g. Koopman, 1984; Travis, 1984,1989) that direction of headedness is not actually the appropriate concept, and instead follows from more basic parametric choices such as direction of case assignment and/or government. The range of X has recently been extended from the lexical categories in (8) to a host of functional (i.e. purely grammatical) categories (cf. e.g. Pesetsky, 1982;

10

PARAMETERS OF SLAVIC MORPHOSYNTAX

Chomsky, 1986a; Pollock, 1989). Here I adopt the system of Chomsky (1986a). In it, both S and S' are seen as maximal projections of the functional categories I(nflection) and C(omplementizer), respectively, as in (10). (10)

CP (= S')

c

Specc

C

IP (= S)

r

Specj

i

VP

Specc is typically the target of w/z-movement and Spe^ is the position of the sentential subject. This uniform system assimilates S and S' to the other endocentric categories, allowing further cross-categorial generalizations and making it possible to restrict phrase structure representations to binary branching (cf. e.g. Kayne, 1984).

1.2. Case Theory The remaining module of GB theory that needs to be presented is case theory, which plays a fundamental role in many of the grammatical processes in the Slavic languages. Abstract case is divorced from the traditional notion of morphological case in that NPs in all languages have abstract case whether or not they manifest it morphologically. The abstract case system of standard GB theory is examined in this section and is then extended to accommodate the kinds of detailed morphological case phenomena found in Slavic in the next chapter. The two central questions concerning abstract case are (1) What receives abstract case and why? and (2) What assigns abstract case and how? The first is addressed in section 1.2.1, the second in 1.2.2. 1.2.1. Principles of Abstract Case The basic observation that drives case theory is that overt NPs are only licensed in positions to which case is assigned. Thus, the examples in (11) are ungrammatical because Bill lacks case. (11)

a. *John painted a picture Bill b. *John is proud Bill

INTRODUCTION

11

c. *John laughed Bill d. *John hopes Bill to leave The condition that requires every nominal with a phonetic matrix to have case is known as the Case Filter, which Chomsky (1981a, 49) states as follows: (12)

*NP if NP has phonetic content and has no Case.

The examples in (11) are thus ruled out, since in none of them does Bill have any source for case. This version of the Case Filter treats case as a property of NPs prerequisite to lexicalization and, as a principle of morphological well-formedness, pertains to the PF side of the grammar. In this sense, case can be construed as making NPs "visible" at PF. In more recent work within the GB paradigm, stemming from suggestions made in chapter 6 of Chomsky (1981a), abstract case is regarded as a property of wellformedness at LF. This effect is accomplished by requiring that arguments have case at LF in order for their theta-roles to be visible, thus tying case theory into the ThetaCriterion. This version of the Case Filter, following Chomsky (1981a, 334—335), is roughly as in (13). (13)

A theta-role can be assigned to an argument chain only if that chain has case or is headed by PRO.

Although (13) is stated as a condition on theta-role assignment, it should properly be understood as a condition on the visibility of theta-roles at LF. An argument chain must have case at LF in order for its theta-role to be visible. Note that this technically allows case assignment (or checking) to be postponed until LF, a point that will play an important role in the analysis of null subjects in chapter 7. Also, the disjunction must be stipulated because PRO necessarily lacks case but nonetheless always bears a theta-role. PRO must therefore be independently licensed at LF. Presumably this is accomplished by virtue of the fact that PRO is an obligatory element, as suggested in Hornstein and Lightfoot (1987), in the sense that its existence follows from the requirement that all clauses have subjects. The Case Filter thus appears to have two sides, one pertaining to PF-visibility, as in (12), and the other to LF-visibility, as in (13). Although Chomsky's intention in proposing (13) was to subsume the filter in (12), there is a residue of effects that cannot be brought under the theta-theoretic Case Filter. Recent work by Milsark (1988), Belletti (1988), Raposo and Uriagereka (1990), and Lasnik (1992a) demonstrates that the PF Case Filter must have independent status. These issues will be discussed in chapter 7, where it is argued that both versions of the Case Filter are necessary. The question posed above of what receives case can now be addressed. Since (12) is rooted in nominal morphology, it should extend to all NPs present at PF, whether or not they bear a theta-role or are even present at LF.7 The LF-visibility Case Filter (13), on the other hand, should pertain to all argument chains, regardless of their morphological category and hence whether or not they are actually able to bear morphological case. This means that argument chains headed by clauses and PPs must also in some way be associated with case features if these are to bear theta-

12

PARAMETERS OF SLAVIC MORPHOSYNTAX

roles. This reasoning drives much of Stowell's (1981) theory of phrase structure, where he attacks the problem of what determines the relative ordering of non-heads within a phrase. Relevant aspects of his theory will be presented in chapter 7, when the case properties of clausal arguments are discussed. It can thus tentatively be concluded that case is motivated both by PF-visibility of nominals and by LFvisibility of arguments.

1.2.2. Case Assignment According to standard GB theory, case is assigned to an NP by a lexical head under the configurational relationship of government. The canonical instance of case assignment occurs when an NP is governed by a V or a P—that is, by a [ —N] category, as in (14). (14)

a. [v- read [NP the book]] b. [P. in [NP the book]]

The [ + N] categories N and A, on the other hand, do not assign case to their complements, as shown in (15). (15)

a. *[N. reader [NP the book]] b. *[A. proud [NP the book]]

The structure in (15) can be salvaged through the mechanism of o/-insertion, however, as suggested by Chomsky (1981a, 50-51), whereby the minimal preposition of, arguably devoid here of semantic content, is inserted to mark case on the book, as in (16).« (16)

a. [N. reader [of [NP the book]]] b. [A, proud [of [NP the book]]]

Case in English is also assigned to subjects of finite clauses, as in (17). (17)

tip [NP tne book] [r is on the table]]

The Agr(eement) element heading the Inflection Phrase (IP) governs its specifier NP and is thus able to assign it nominative case. Alternatively, it may be that Agr is itself nominative and that the subject receives nominative case through "SPEC-head agreement" rather than through government. This approach seems particularly plausible under the extended X-bar system discussed in chapter 8, in which IP is broken down into its components and AgrP constitutes a separate functional projection. Finally, subjects of NPs are themselves assigned genitive, as in (18). (18)

[NP CNP tne book's] [N. cover]]

Once again, one might imagine that case assignment is licensed here because the head N governs its specifier. There is reason to suspect, however, that genitive assignment in English is not actually case assignment under government by N. Chomsky (1981a, 170) states that genitive is assigned to the left sister of X' within

INTRODUCTION

13

NP, that is, under sisterhood to N'. Perhaps, then, it could be assimilated to nominative, with P and V assigning case to the right and N' and F to the left. This might allow case assignment to be restricted to sisterhood, rather than government. Alternatively, it may be that the 's is a case assignor rather than a manifestation of case itself, on a par with Agr, so that SPEC-head agreement can again be invoked.9 This is supported by the fact that 's cliticizes onto the entire NP. (19)

[NP [NP the king of England]'s [N. hat]]

This analysis has given rise to the Determiner Phrase (DP) hypothesis, according to which NPs are in fact complements of determiners, as, for example, in Fukui and Speas (1986), Abney (1987), and Stowell (1989). The DP is headed by functional elements such as determiners, demonstratives, and (in some analyses, but not Abney's) the possessive morpheme. The structure of (19) could thus more accurately be represented as in (20). (20)

DP

D1

DP

D'

D

NP

the

King of England

D

NP

's

hat

This idea further helps to increase the symmetry between NPs and clauses. However the subject DP the king of England receives case in (20), one can claim that subjects of IP are in a comparable configuration, whether government by D/I, sisterhood to D'/I', or SPEC-head agreement with D/I is the relevant factor. Since most of the Slavic languages (other than Macedonian, Bulgarian, and North Russian dialects) display little overt evidence for a separate DP projection, however, I will continue to refer to NPs and use DP only when specifically warranted. There is, nonetheless, one unifying aspect of the DP/IP theory that will play an important role in this study. Since the subject/specifier of DP or IP is actually an argument of the NP or VP complement to D or I, respectively, it seems natural to assume that this subject originates as the subject of NP/VP and subsequently raises to SPEC-DP/IP. This movement is motivated by considerations of case—the subject of NP/VP requires case but is not generated in a case-marked position, hence it must move to SPEC-DP/IP, where it can be assigned case. This kind of derivation is completely parallel to other standard instances of NP-movement, such as raising to derive Johnt

14

PARAMETERS OF SLAVIC MORPHOSYNTAX

appears \e1 to be tired]. Under this view, then, both NP and VP contain all their arguments at D-structure and D and I are raising categories.10 Abstract nominative and genitive in GB theory seem to correspond to their morphological manifestations. I shall argue that there is in principle no difference between specific abstract cases and morphological ones. Thus, while Chomsky (1981a, 170) calls the abstract cases assigned by V and N objective and oblique, respectively, there is no real reason to distinguish the two in English. The English pronominal system has only objective, nominative, and genitive, with objective case arising in default situations. By "default" I mean that whenever an NP is assigned case, and that case is not specified as nominative or genitive, it appears in the objective. This explains the complete syntactic identity of any cases assigned by any V or P. Kayne (1981b) uses the observation that V and P both assign objective in English to account for a host of differences between French and English. For example, he claims that reanalysis of V-P groups, permitting subsequent preposition stranding, is made possible by the identical case-marking properties of these two categories in English.11 This completes the survey of the Principles and Parameters model of syntax and the modules proposed within GB theory. These will be elaborated upon in the chapters to follow to explain specific problems in the syntax of the Slavic languages. Notes 1. The language faculty is thus an "input system" in the sense of Fodor (1983). Properties of such systems include that they are domain specific, their operation is fast and mandatory, they are informationally encapsulated and allow only limited access to the representations they compute, and they are associated with fixed neural architecture and exhibit characteristic breakdown patterns. 2. There are numerous proposals in the literature to revise the binding theory, which is given in (2) in its simplest form. 3. This raises many interesting and open questions, such as (i) what feature complexes each principle pertains to, (ii) what level(s) of representation the principles apply at, and (iii) whether or not the principles apply as a block. It is reasonable to suppose, for example, that (2b) should be formulated in terms of [ — anaphoric] rather than [ + pronominal] and that (2a) and (2b) should be divorced from (2c). I return to this issue in chapter 6. 4. This principle also rules out things like vacuous quantification, free variables and expletive elements at LF. 5. Pesetsky's formulation, instead of the original one in Chomsky (1981a, 29), is chosen in that it makes specific reference to the projection of theta-roles rather than subcategorization in general. An empirical difference between the two will be discussed in chapter 4. 6. Other frameworks, notably Generalized Phrase Structure Grammar (cf. Gazdar, Klein, Pullum, and Sag, 1985), also factor phrase structure rules into precedence and dominance statements. GB, however, does this by eliminating the categorial component altogether and ascribing its properties to distinct modules, rather than embellishing it. 7. Chomsky (1986b) argues that the principle of full interpretation forces expletive elements to be eliminated at LF and proposes a rule of "expletive replacement." 8. Whether the complement in the derived structure should be regarded as an NP, with of adjoined, or a PP, with o/the head of a phrase, is unclear. 9. One problem with this idea is that postnominal genitive NPs still require o/-insertion,

INTRODUCTION

15

despite the presence of the putative case assigner 's, as in a friend *(of) John's. See Anderson (1984) and Abney (1987) for discussion. 10. This analysis is developed in Abney (1987) for DP/NP and in Koopman and Sportiche (1988) for IP/VP. 11. Objective as a "default" case might also explain its use with topics (i), inside conjunctions (ii), and in copular constructions (iii). (i) Them, they always come late, (ii) Me and John want to go first, (iii) It is me. Certain potentially serious problems with this notion nonetheless remain. For one thing, in chapter 2 I shall argue for a feature system that regards accusative as the default case in Russian (at least when governed by V orP), although nominative is still employed in contexts (i) and (iii). More importantly, the appearance of spurious nominative NPs in contexts where only PRO is admitted must be prevented.

2 Matrices, Indices, and Morphosyntactic Features

This chapter develops a system of case representation and assignment originally proposed in Franks (1985, 1986). Section 2.1 presents the basic framework and the mechanisms involved. Section 2.2 examines the issue of what kinds of morphosyntactic case features are appropriate for describing Russian, with special reference to the set of features put forward in Jakobson (1936, 1958). 2.1. A Model of Morphological Case Most of the work on case theory within the Government and Binding (GB) paradigm focuses on the factors motivating and regulating abstract case assignment, with the concomitant assumption that this will be instantiated morphologically whenever possible. In this study, however, I argue that morphological case is not fundamentally distinct from abstract case, but rather reflects its language-particular realization. The problem of relating abstract case to morphological case should therefore be attacked from the opposite direction. Case theory must first and foremost be able to account for morphological case properties such as are found in Slavic. It needs to be based upon explicit and detailed morphological analysis of nominals, connecting that analysis to abstract and general properties of case assignment. In this section, just such a theory is proposed. 2.1.1. Case Submatrices Most theories of phrase structure, including X-bar syntax, assume that syntactic nodes are actually bundles of morphosyntactic features, whether or not they are generally explicitly represented as such. In chapter 1, for example, it was argued that category symbols such as V and A are really abbreviations for the feature complexes [ + V, - N] and [ + V, + N] and that the referential status of NPs can be described in 16

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

17

terms of the features [±a(naphoric), ±p(ronominal]. This breakdown, however, only supplies limited information about a node's features. Clearly, part-of-speech does not exhaust the set of morphosyntactic properties that must be formally represented. If the idea that all nodes are exhaustively feature complexes is taken seriously, other kinds of information should also be reflected, depending on what the grammatical processes of the language require. In Russian, this might include gender, number, case, and sometimes animacy for nouns; gender, number, person, case, and animacy for pronouns; gender, number, and case for adjectives; person, number, tense, aspect, and sometimes gender for finite verbs, and so on. Whether these are supplied syntactically or are inherent to the lexical item is technically immaterial—they must be represented in the eventual feature matrix. A node may accrue features through agreement or movement, as in recent GB analyses stemming from Chomsky (1986a) and expanded in Pollock (1989), who show that verbs acquire tense and agreement features through movement to tense and agreement nodes. This has been extended to other verbal properties such as mood and aspect, and proposals also exist in the literature for treating nominals similarly through the proliferation of functional categories such as case, gender, and number phrases. More recently, Chomsky (1992) adopts the idea that words are inserted complete with morphosyntactic features, which are then checked against licensing functional categories. In this book, however, I assume a more traditional derivational approach. Whatever system ultimately proves to be correct, the result is that by S-structure all terminal nodes must be fully specified as clusters of morphosyntactic features. I take this kind of representation as prerequisite to the successful operation of lexical insertion, which I regard as the replacement of morphosyntactic feature matrices with phonological ones drawn from the lexicon. More precisely, the first rule of the phonology is one pairing the morphosyntactic feature matrices of X° elements—terminal nodes in phrase structure trees—with non-distinct ones in lexical entries, and then replacing these with their appropriate phonological representations.1 I thus envision a two-stage process of lexical insertion.2 At D-structure, words are inserted into phrase structure trees on the basis of non-distinctness from X° feature complexes, that is, so long as they do not contrast in terms of part-of-speech. Lexical insertion presumably introduces the word itself, that is, a stem with the intended meaning, together with all the word's inherent features. For nouns this probably includes all morphosyntactic information except case, which is syntagmatically determined. For adjectives, on the other hand, gender and number would also remain unspecified, since these features are induced through agreement. In the course of the derivation, all unspecified features become fixed (in a manner to made explicit in the next section), eventually enabling insertion of the word in its paradigmatic phonological form at or by PF.3 There are several interesting consequences of this general system worth noting. It allows both derivational and inflectional morphology to be done in the lexicon, distinguishing the two in that inflectional features are syntactically relevant. Lexical entries must be fully formed and contain or be able to access entire paradigms, consisting of complete morphosyntactic matrices and associated phonological representations. Moreover, the two-stage model of lexical insertion resulting from the

PARAMETERS OF SLAVIC MORPHOSYNTAX

18

adoption of the traditional distinction between "words" and "wordforms" meshes nicely with the standard T-model presented in chapter 1. Morphosyntactic representations are visible to the rules of LF, but are no longer available to this component once they have been mapped into phonological representations. LF rules cannot operate on phonological representations, just as PF rules cannot operate on morphosyntactic ones.4 This approach induces the bifurcation of the grammar, obviating any need to stipulate that the rules of the PF and LF components interpret S-structures, neither being able to depend on information exclusively germane to the other component. PF and LF are thus mutually opaque. Two important empirical issues arise if, just as part-of-speech can be broken down into features, so also must all other morphological categories be reduced to their distinctive components. The first and more general one concerns how these feature complexes might be represented. The second, more specific issue deals with the problem of what features are appropriate for expressing each morphological category. Here I will address the first issue, postponing the feature question until section 2.2, when the Russian case features are examined in detail. Taking features as discrete bipolar categories, and abstracting away from the content of the case features Fj, F2, and F3, the Russian form knigu 'book(acc sg)' might simply be represented as a list of all its features.5 (1)

+N -V

PART-OF-SPEECH REFERENCE

-P +fem -masc -pi -1 -2

GENDER NUMBER PERSON

?F!

?F2 ?F3

CASE

I do not regard the feature matrix as a seamless whole, however. Additional structure must be introduced in order to allow each category to function to some degree independently.61 suggest that each morphological category be regarded as a bundle of features. These will be represented as '' submatrices'' internal to the larger matrix that constitutes the node itself, as in (2).

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

19

Representing each set of features as a submatrix facilitates the statement of processes like agreement and government, which involve entire categories. It is now possible to refer to "case" relationships in general and directly, with succinct formulations of these relationships. That is, case assignment does not selectively ignore certain features—it either obtains, or it does not. All features of a given morphological category behave uniformly in that they are subject to morphosyntactic processes as a set. The submatrix approach is thus amenable to traditional pronouncements about Russian grammar, such as "predicate adjectives (and past tense verbs) agree with their subjects in gender and number, but not in person, whereas (non-past tense) verbs agree with their subjects in person and number, but not in gender." It also accommodates processes such as case assignment and feature propagation in a straightforward way. This is discussed in the next section, where several applications of the above system are demonstrated. 2.1.2. Coindexation Although morphosyntactic matrices are composites of inherent and syntactically dependent information, all the feature values therein do not arise in a uniform manner. Thus part-of-speech, from which phrases are projected, is specified at D-structure, whereas case is determined in the syntax (and in some instances also at D-structure). One way to accomplish this is to assume that feature values may be either fixed or variable, in the sense that their values are open and in need of determination. A feminine singular noun might therefore be more accurately represented upon lexical insertion at D-structure as in (3), an adjective as in (4).

20

PARAMETERS OF SLAVIC MORPHOSYNTAX

The variable feature values become fixed in the course of the derivation. Clearly, this fixing comes about through the operation of government and agreement processes. I suggest that the various processes with this effect can be conceptually unified if they are all formally regarded as involving coindexation. Assume, as a matter of definition, that all matrices may bear indices, hence all syntactic nodes may bear them (cf. e.g. Williams, 1982, 294). Next, assume that these indices are distinct unless principles of grammar specify otherwise. In particular, the grammar contains various principles that require the identification of the values of indices on distinct nodes. Perhaps the most familiar coindexation rule is the one binding anaphors in their governing categories. Others to be investigated in this section include projection coindexation, agreement coindexation, and theta-role assignment coindexation. I will argue that case assignment is a concomitant result of the application of coindexation in all these instances. The system works by taking advantage of the idea that submatrices, as matrices themselves, can also be endowed with indices. There is then one overarching principle that sets the variable values of the features in a submatrix in one node equal to the specified ones in the submatrix of another node when these two bear the same index. This Agreement Principle can be stated as follows: (5) Everything else being equal, the values of the features in coindexed matrices are the same. What I mean by "everything else being equal" is that if the feature values have already been fixed, either syntactically or inherently, then they cannot be altered by the Agreement Principle. Principle (5) does not change feature values; it only sets them. This idea that, once established, feature values remain fixed, is known as the Principle of Inertness (cf. e.g. Babby, 1985, 108). (6) No rule of grammar may alter morphosyntactic feature values. Binding, for example, has no effect on feature values, since it applies at S-structure and/or LF. Thus, the cases of each argument are necessarily different in (7). (7)

Ivan uvazaet sebja 'lvan(nom) respects himself (ace)'

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

21

The reflexive object sebja 'self is marked accusative and the subject Ivan nominative before binding principle A coindexes them. That binding theory does not apply at D-structure can be shown by the fact that it is fed by both NP-movement and scrambling, as shown by the following examples:7 (8) a. Joh^ seems to himself [ej to be upset] b. ja sebja! s&taju [dto on uvazaet ej I self(acc) consider that he respects c. ja studentovi drug drugu poobes'cal [PRO sprosit' Cj] I students(acc) each other promised to-ask In English (8a), John can bind himself only after it has raised. Similarly, in Russian (8b), scrambling the reflexive sebja results in ambiguity, since moving it out of the domain of the lower tensed Inflection node (INFL) allows it to be bound by ja. In Russian (8c), studentov 'students' has been scrambled to a position from which it can bind the reciprocal drug drugu, rendering the sentence grammatical. Returning then to example (7), it is clear that the coindexation induced by binding theory occurs too late to cause sebja to be assigned nominative by the Agreement Principle. Note, incidentally, that the reflexive still must match its antecedent in pronominal features (although this is only evident in the English gloss). This, however, is an independent requirement on proper binding (cf. e.g. Lasnik and Uriagereka, 1988), ruling out, for example, (8c) with studentov in situ. Before going on to illustrate how the Agreement Principle effects case assignment, there are several mechanical details of the system that need to be made explicit. Since (5) really applies to coindexed submatrices, the problem of how these acquire indices must be addressed. It seems to me there are two subcases. For submatrices whose features are inherently specified, I claim that their indices are the same as that of the superordinate node, since inherent properties arise by virtue of the word itself. Submatrices whose feature values are variable, on the other hand, can be understood either as lacking indices at D-structure or as having indices distinct from all others in the tree. I will adopt the former alternative in order to preserve insights of the Aoun, Hornstein, Lightfoot, and Weinberg (1987) approach to the ECP, although the difference is irrelevant at this point. Either way, it is necessary to ensure that submatrices of coindexed nodes be coindexed as well, if their contents are of the same morphosyntactic category, in order for the Agreement Principle to work. This is stated in (9), where "compatible" means "of the same morphological type." (9) Coindex all compatible submatrices of coindexed matrices. The noun in (3) and the adjective in (4) thus have the indexical structures in (10) and (ID- 8

22

PARAMETERS OF SLAVIC MORPHOSYNTAX

The 1 index of the N node is identified with those of the part-of-speech, reference, gender, number, and person submatrices; the 2 index of the A node is identifed only with the part-of-speech submatrix. The case feature submatrices of the N and the gender, number, and case feature submatrices of the A, however, remain unindexed but will become fixed by S-structure, eventually enabling insertion of the correct wordform. In this section I have outlined a general system for representing and manipulating morphosyntactic features. All nodes are matrices bearing indices, some of which are required to be identical in accordance with principles that remain to be explicated. This implies that the sole function of indices cannot be to indicate coreferentiality, otherwise the inventory of coindexation principles would be too small to be of benefit. Instead, indices have multiple functions, the most transparent of which is referential. In other words, indices do not "mean" reference, or anything else for that matter. They are merely a formalism, and as such are free to be interpreted in any appropriate manner. This view of indices is, I believe, in harmony with the spirit of generative grammar, which eschews any one-to-one correspondence between form and function. Grammars are formal constructs, their abstract properties therefore existing independently of any of language's many functions. 2.1.2.1. Projection The clearest illustration of the Agreement Principle can be found in the phenomenon of percolation. This is the mechanism by which morphosyntactic features propagate throughout a phrase and, in particular, up and down the projectional spine of a category. Everything else being equal, the values of the features of a node X and its phrasal projection XP are equivalent. This is necessitated by the observation that some of the properties of X are inherent to X, whereas others depend on the position of XP. So, for example, an NP might be feminine because its head N is feminine, but the N might be dative because the NP is the complement of a preposition assigning dative. This effect can be simply achieved by assuming the following principle of coindexation:9

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

23

(12) All members of a projection bear the same index. Thus, at D-structure the N in (10) would project up to NP, forcing agreement by virtue of the Agreement Principle, as in (13), simplifying somewhat for ease of exposition.

The maximal projection NP will subsequently be assigned case, which in turn will percolate down to the head N. This percolation system is thus bidirectional, a property that follows from the transitive nature of indices. 2.7.2.2. Agreement Adjective-noun agreement can be handled in a similar manner, if it is assumed that modification is also a coindexation relationship (as, for example, suggested in Bouchard, 1984). If so, the adjective in (11) and the noun in (10) can be combined as follows:10

24

PARAMETERS OF SLAVIC MORPHOSYNTAX

Once the index 2 of the AP is identified with index 1 of the N, agreement in pronominal features occurs and percolates down the A projection, as in (15).n

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

25

Eventually, when NP is assigned case, agreement will also take place with respect to this submatrix as well. I leave open the issue of just what principle requires this coindexation. Possibilities include simply stipulating that modifiers are coindexed with what they mod-

26

PARAMETERS OF SLAVIC MORPHOSYNTAX

ify, perhaps in order to facilitate interpretation at LF, including it under the percolation projection of Kayne (1984), or utilizing the Strict Opacity Condition of Williams (1982), which roughly states that nothing can be free in anything. Of significantly more interest will be the relationship between predicate APs and the NPs they predicate, since these involve neither string adjacency nor constituency. Agreement nonetheless does occur between predicate adjectives and NPs, a fact that will be treated at length in chapter 6. 2.1.2.3. Theta-roles and Case In this section I develop a mechanism that allows case assignment to be a consequence of theta-role assignment. In order to accomplish this, theta-role assignment must also be regarded as a coindexation process. This is done by adapting to the submatrix model StowelFs (1981) idea that theta-role assignment involves the entering of an argument's index into the grid of theta-roles (his "theta-grid") of a thetarole-assigning element. Assume first of all that the property of assigning a theta-role is on a par with the morphosyntactic categories in that it is directly represented as a part of a syntactic node. Under this view, syntactic nodes may actually contain a theta-role submatrix alongside submatrices for the other, more familiar features. I shall refer to this submatrix as a theta-slot in order to differentiate it from Stowell's theta-grid. Each node may assign at most one theta-role, drawn from the inventory of available thetaroles in the theta-grid in the lexical representation of the head's predicate-argument structure.12 Although theta-slots presumably can also be decomposed into bundles of features, I will refer to their contents using the traditional theta-role labels. Thus, for example, a verb or preposition can be partially represented as follows:

Note that I have endowed the theta-slots with the indices of their containing nodes, since they are inherent lexical properties of the items in question. Next I assume that case-assigning nodes contain submatrices, just as case-bearing ones do. That is, any node may in principle contain a case submatrix or not, regardless of whether it itself can bear case. The [ — N] categories V and P do not themselves show case morphologically, but rather realize it indirectly on their [ + N] arguments. Providing (16) and (17) with case submatrices gives rise to the following structures:

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

27

Note that, although I am temporarily postponing determination of the precise case features for Russian, this theory permits two possibilities—the feature values are either specified or variable. If the verb in (18) requires some specific case (' 'quirky'' in the GB parlance), it will be represented as in (20a) with set case feature values, and if it assigns the default accusative it will be represented as in (20b) with open ones.

For the purposes of discussion I assume the former option, turning to (20b) in the next section. It is now a relatively straightforward matter to motivate case assignment to arguments through theta-role assignment. Theta-role assignment is the identification of the index of a theta-role assignor's theta-slot and the index of some argument. This presumably occurs under sisterhood and, in English at least, strict adjacency.13 The process of theta-slot coindexation, stated in (21), thus transforms an object into a complement, where the term ' 'object'' refers to the position of sister to a head and the term "complement" refers specifically to a phrase that receives its theta-role from that head.14 (21) Assign the index of a theta-slot to an argument. Notice that this process is intimately related to the Theta-Criterion. If this principle is reformulated as in (22), and moreover holds at all syntactic levels (given the Projection Principle), then the standard effects of theta-theory can be achieved. (22) Every theta-slot must bind a unique argument and every argument must bind a unique theta-slot.

28

PARAMETERS OF SLAVIC MORPHOSYNTAX

The mutual binding of (22), in fact, forces the sisterhood locality condition under Reinhart's (1983) "first branching node" definition of c-command, hence of binding.15 Before theta-role assignment, therefore, a V plus its object NP can be represented as follows:

Principle (22) requires index 1 to equal index 2. Consequently, index 1 is also assigned to the case submatrix of the complement NP, and, by the Agreement Principle, the variable case features of the NP are set as those of the governing verb. The result is shown in (24).

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

29

These are, in sum, the mechanisms whereby case assignment to an argument takes place. Before concluding this section, it is worth noting that the proposed system works precisely because the case submatrices of case assignors and case bearers are identical formal entities.16 This is roughly in keeping with Stowell's (1981) idea that elements may assign or bear case, but not both, although once formalized as above it has somewhat different consequences than his Case Resistance Principle. It thus allows case transmission processes whereby an element assigns the case it receives to something that is not its argument. A noun, for example, may assign its case to an AP via agreement or an NP to a predicate adjective via predication coindexation, as will be discussed in chapter 6. Even prepositions also occasionally have the property of transmitting case, as with Russian za 'for' in (25).17 (25)

a. dto what 'what b. c"to what 'what

za celovek kupil etu knigu? for person(nom) bought this book(acc) kind of person bought this book?' za knigu on cital? for book(acc) he(nom) read kind of book was he reading?'

Presumably za in (25) neither assigns a theta-role to its object nor governs any particular case. Its case features can thus be fixed externally, by INFL in (25a) and by the verb in (25b). This account of the cto za construction is somewhat problematic in that, unlike the German was fur ein, it is somewhat less acceptable with verbs that take oblique case complements, as in (26), although pseudo-clef ting, as in (27), results in perfect acceptability.18 (26) a. ?dto za knige on obradovalsja? what for book(dat) he enjoyed 'what kind of book did he enjoy?' b. ?&o za knigoj on uvleksja? what for book(inst) he(nom) get-carried-away-with ' what kind of book did he get carried away with?' (27) a. dto what 'what b. cto what 'what

eto za knige on obradovalsja? this for book(dat) he enjoyed kind of book was it that he enjoyed?' eto za knigoj on uvleksja? this for book(inst) he(nom) was-carried-away-with kind of book was it that he got carried away with?'

The preposition za thus seems in principle to be able to transmit to its object whatever case it receives. Another preposition that exhibits this sort of behavior is the distributive preposition po in Serbo-Croatian; compare chapter 5 and Franks (1994). In formal terms, such prepositions have variable case feature values that are set by the same mechanisms that determine the case of an NP. This phenomenon thus appears to contradict the Case Resistance Principle, since here a case-assigning

30

PARAMETERS OF SLAVIC MORPHOSYNTAX

category is able to receive case, and raises the question of whether other categories, such as verbs and INFL, can enter into this process. Although there is some evidence to think that they can, I do not pursue this matter here. Agreement, predication, and the cto za construction are all examples of an element assigning case to something it does not theta-mark. The Case Resistance Principle, however, states that case assigners may not themselves be assigned case, motivating extraposition of the finite clause in the following example from Stowell (1981): (28) a. *Paul mentioned [CP that his shirt was dirty] [PP to Bill] b. Paul mentioned [NP e]: [PP to Bill] [CP that his shut was dirtylj Extraposition makes the trace available for case-marking, assuming it can be determined to be an NP.19 In the CP/IP framework of extended X-bar syntax, however, CP is not in fact headed by a case assignor. Instead, I claim that the clause must extrapose simply because CP lacks a case submatrix. In this respect it contrasts to IP, which explains why Exceptional Case Marking (ECM) takes place into IP but not CP, as in (29). (29)

a. I believe [IP him to have left] b. *I believe [CP that [IP him to have/has left]]

In (28b), then, extraposition occurs because the clause must be in a case-marked chain in order for its theta-role to be visible at LF. Stowell is therefore correct that the finite clause moves because it cannot itself bear case, but the reason is not that it already has case or is headed by a case assigner, as his theory would claim. Moreover, Stowell observes that PPs and infinitival clauses, although they are themselves headed by case assigners, do not need to extrapose. This is because they contain case submatrices and can thereby satisfy the theta-theory visibility requirement intrinsically. It also allows them actually to receive and transmit case, as discussed above. 2.1.3. Phrase Structural Case The previous section concentrated on instances in which a verb or preposition assigned a particular case to its complement. Since this case was to some extent idiosyncratic, it was explicitly represented in the case assignor's case submatrix. There are numerous instances, however, in which an NP receives case solely by virtue of its syntactic position. The configuration in which the NP finds itself determines the case it bears, independently of specific properties of any potential case assigner. I begin by considering the most obvious such instance—the default assignment of accusative to objects of verbs. Despite its superficial simplicity, the default accusative turns out to be one of the most conceptually complex illustrations of phrase structural case, interacting as it does with the Slavic quantificational genitive and the genitive of negation, as we shall see in chapters 4 and 5. A further problem is that the default accusative appears to instantiate all three possible case assignment configura-

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

31

tions. It was argued that a node may either have a case submatrix or not, and that if it does, the features of that matrix could be lexically specified or not. The default accusative, however, arises on arguments of transitive verbs with open case features, and on non-argument time and distance phrases with potentially all classes of verbs. Consider first of all an ordinary transitive verb, as in (30). (30)

Ivan cital etu knigu 'Ivan read this book(acc)'

The verb citat' 'to read' can be represented as in (20b), with unspecified feature values. The verb thus combines with its object NP as follows:

Theta-role assignment at D-structure equates the 2 index of the complement etu knigu with the 1 index of the theta-role submatrix [theme]. Theta-role assignment should also then cause the case submatrices of the V and the NP to be equated, but no case assignment can take place since neither case submatrix contains anything to agree with. This results in the structure in (32).

32

PARAMETERS OF SLAVIC MORPHOSYNTAX

I have placed the indices of the case submatrices in parentheses because technically speaking they lack them until their features become fixed. This only happens at S-structure, however, when the variable case features of objects of verbs (and prepositions) are set as accusative. Assuming accusative to be [ —Fj, — F2, — F3], for reasons to be made clear in section 2.2.2, the eventual structure is as in (33).

The proposed system has several curious properties worth noting. For one thing, it has the counterintuitive result that the case features of the verb are actually assigned to it by the complement NP, although the fact that the NP has (default accusative) case is licensed by the V. This is necessary in order to assimilate the default

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

33

accusative when assigned to arguments to the default accusative when assigned to non-arguments, and yet to differentiate it from other phrase structural case. For another, it implies that rule (9), which Comdexes compatible submatrices of coindexed nodes, is not entirely accurate, but should be revised as in (34). (34)

Assign the indices of all submatrices of coindexed matrices to compatible unindexed submatrices.

This change has no effect on the functioning of the Agreement Principle. Predictably, if a verb lacks case features or they are specified for some oblique case, then it cannot assign accusative to any argument.20 It appears, however, still to be able to assign accusative to non-argument time and distance phrases, as in the Russian examples in (35). (35)

a. Ivan spal vsju no£' 'Ivan slept all night(acc)' b. Boris celuju nedelju dumal *Ma§u/o Mase/5to Masa ego ne ljubit 'Boris the whole week(acc) thought Masha(acc)/about Masha/that Masha doesn't love him' c. etot direktor upravljal *fabriku/fabrikoj vsego odin god 'this director managed the factory(acc)/(instr) only one year(acc)'

The verbs spat' 'to sleep' in (35a) and dumat' 'to think' in (35b) are intransitive in the sense that they contain no case submatrices, hence cannot case-mark any complement, but differ in that dumat' contains a theta-slot whereas spat' does not. The verb upravljat' 'to manage' in (35c), on the other hand, assigns both theta-role and case, but that case is inherently specified as instrumental. Regardless of these differences, they are all free to assign accusative to non-argument objects. I tentatively conclude that accusative is assigned purely under the configuration "sister to V." This conclusion, however, raises an important conceptual concern: adjuncts are generally assumed to occupy a higher position in the tree, with the object "sister to a head" position reserved for true complements; compare, for example, Chomsky (1981a, 47), who comments "complements of X' are always theta-positions." This is usually taken to mean, I believe, that sisters of heads are necessarily complements.21 The Russian facts in (35), as well as those regarding the genitive of negation to be cited in (38), would seem to contradict this assumption. Otherwise, it becomes difficult to explain why the same structural case-marking rules appear to apply to both arguments and adjuncts. This conclusion also follows from observations about structural case-marking in Finnish and Korean presented in Maling (1993) and Kim and Maling (1993), respectively. In section 5.3.2, however, an alternative approach to apparent structural case assignment to Slavic time phrase adjuncts is considered that preserves the traditional system (cf. also Franks and Dziwirek, 1993). That Russian verbs may indeed assign quirky instrumental or genitive to their complements can be seen in the fact that they are able to passivize, so long as this is compatible with their semantics. Fowler (1987b) amply demonstrates this, citing examples such as the following:

34

PARAMETERS OF SLAVIC MORPHOSYNTAX

(36) a. russkaja armija upravljalas' Kutuzovym 'the Russian army was managed by Kutuzov' b. prenebregaemyj toboj molodoj £elovek okazalsja otlicinym pomosc'nikom 'the young man scorned by you turned out to be an excellent assistant' c. eel' byla dostignuta uc"enym 'the goal was achieved by the scientist' d. v na§em dome izbegalos' vse, cto moglo davat' pis"cu voobrazeniju 'in our house everything was avoided that could give food for the imagination' The verbs upravljat' 'to manage' andprenebregat' 'to scorn' govern the instrumental and the verbs dostignut' 'to achieve' and izbegat' 'to avoid' govern the genitive exclusively, but nonetheless passivize with nominative subjects. Fowler shows that the conditions that prevent a quirky case-assigning verb from passivizing are the same as those that operate to restrict the passivization of accusative-assigning verbs. As will be discussed in chapter 8, passive morphology typically nullifies the ability of a verb to assign case, canceling or "absorbing" its case submatrix regardless of the precise contents. As expected, this nonetheless has no effect on the verb's ability to assign accusative to a non-argument, as the passive of (35c) shows: (37)

fabrika upravljalas' etim direktorom vsego odin god 'the factory was managed by this director only one year'

The idea that non-arguments may receive phrase structural case is supported by the fact that they also participate in the "genitive of negation" construction; compare Franks (1990a) for further discussion. This construction, in which objects of negated verbs that do not assign quirky case may appear in the genitive, will be considered more carefully in chapter 5. At this point, simply note that the genitive of negation rule in Russian appears to pertain with roughly equal force to nonarguments, as shown in (38). (38)

a. Ivan ne cital etoj knigi ni minuty 'Ivan didn't read that book(gen) even a minute(gen)' b. Boris ne spal ni odnoj minuty 'Boris didn't sleep even one minute(gen)' c. general ne pravil stranoj ni odnogo goda 'the general didn't run the country(inst) even one year(gen)'

These examples support the conclusion that genitive is structurally assigned to sisters of negated verbs, and that this rule applies regardless of the case features of the negated verb—it could have an open case submatrix, as in (38a), no case submatrix, as in (38b), or one with inherent feature values, as in (38c). The only restriction is, just as with non-negated verbs, if the verb assigns quirky case then that case must be discharged to its argument. Significantly, this has no bearing on its ability to assign structural accusative or genitive to a non-argument. One reason the accusative is generally regarded as having default status, in con-

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

35

trast to the oblique cases, is that it is overridden by the genitive of quantification. Compare the examples in (39). (39)

a. Boris imel pjat' domov 'Boris had five(acc) houses(gen pi)' b. Boris vladel pjat'ju domami 'Boris possessed five(inst) houses(inst pi)'

The proper analysis of this striking pattern, which has been the focus of much debate, will be the subject of chapter 4. For the time being, it suffices to assume that the contrast in (39) in some way reduces to the fact that vladet' 'to possess' has inherent case features specified as instrumental, whereas imef 'to have' does not. That is, lexically fixed case submatrices interact with the genitive of quantification differently than open ones. Now consider in this light the behavior of complements of prepositions. A preposition assigning accusative behaves like a verb assigning accusative, as in (40a), but a preposition requiring an oblique case behaves just like a verb requiring an oblique case, as in (40b). (40)

a. pro pjat' knig 'about five(acc) books(gen pi)' b. o pjati knigax 'about five(loc) books(loc pi)'

Like the verbs in (39), the prepositions pro 'about' and o 'about' are virtually indistinguishable in meaning, but govern different cases. Once again, the accusative in (40a) is overridden by the genitive of quantification, but the locative in (40b) is not. I thus conclude that any statement about the default accusative should be generalized as in (41). (41) Assign accusative to sisters of [ —N]° categories. I have not included the directionality of case assignment in (41) for several reasons. First of all, the relatively free word order of Russian makes its rightward application difficult to prove. Moreover, once established, it is still reasonable to assume that directionality can be factored out, either as a consequence of the rightward assignment of theta-roles in Slavic or as a generalization of headedness or the direction of case assignment. That is, whatever mechanism(s) set underlying word order (cf. e.g. Koopman, 1984; Travis, 1984,1989) will also regulate the directionality of (41) and similar rules. I have also ignored the fact that (41) applies equally to verbal participles, which are adjectival in form and hence categorially [ + N]. Since no other adjectives exhibit this possibility and participial phrases show the complete internal structure of VPs, however, I assume their heads are verbs at the point their case assignment properties are determined. An additional question raised by (41) is why non-theme complements cannot similarly be assigned accusative. For example, why can't the locative argument a house be assigned accusative in *John lived a house? The problem here, it seems to me, must have to do with theta-theory rather than case, since non-arguments can

36

PARAMETERS OF SLAVIC MORPHOSYNTAX

receive case, such as accusative time phrases, as in John lived 80 years (cf. also example [35]), or cognate objects, as in John lived a long life. Locatives, on the other hand, require a preposition to assign the appropriate theta-role, perhaps in combination with the verb. Another configuration in which case assignment is clearly driven by phrase structure is that of object of N. Since the overwhelming majority of NP objects of nouns are marked genitive, it has often been suggested that this is also a configurational case (cf. e.g. Babby, 1987). While I agree that the genitive on objects of N is indeed configurationally determined, I will argue that it is not default in the same sense as is the accusative on objects of Vs and Ps. There are two important ways in which the genitive on objects of nouns differs from the accusative on objects of verbs. The first is that every noun is in principle able to take a genitive complement, but not all verbs can take an accusative one. This property goes hand in hand with the observation that there is a far greater degree of semantic independence between a noun and its (genitive) object than between a verb and its (accusative) object. For one thing, a genitive object may be ambiguous as to whether its semantics correspond to those of the subject or object of the related verb, as in (42).22 (42)

a. vyzov vracSa 'the summons of the doctor' b. zaScita tovarisca 'the defense of the comrade' c. prie'm posla 'the reception of the ambassador' d. ctenie PuSkina 'the reading of Pushkin'

Unlike objects of the associated verbs, the sense of such expressions can only be understood in context. The semantic range of genitive objects of Ns is of course much freer than this, since it admits an extremely broad range of interpretations.23 The 'textbook' in (43), for example, may be the one Professor Ivanov wrote, uses, owns, recommends, or whatever. (43)

ucebnik Professora Ivanova 'Professor Ivanov's textbook'

It suffices that there be some relationship between the professor and the book, which the construction leaves vague. There is, moreover, no absolutely direct correspondence between the case governed by a verb and that of the noun derived from it, although the opposite has sometimes been claimed; compare, for example, Babby (1987) and Freidin and Sprouse (1991). Thus the noun upravlenie 'management', based on the verb upravljat' 'to manage', which governs the instrumental, may but need not also govern that case, as shown in (44).24 (44)

a. upravlenie etim zavodom bylo poruceno Petrovu 'the management of this factory(inst) was assigned to Petrov'

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

37

b. upravlenie zavoda sozvalo raboclx na railing 'the management of the factory(gen) called the workers to a meeting' And of course upravlenie, unlike the associated verb but like all other nouns, can appeal- without any complement at all. It thus seems that although case is configurationally licensed on sisters of N, the case is not assigned by the N itself. Instead, it may be whatever case is appropriate to the semantics of the argument, or the genitive if these are left vague, as in the familiar example (45). (45)

a. pamjatnik Puskina 'a monument of Pushkin(gen)' b. pamjatnik PuSkinu 'a monument to Pushkin(dat)

Rather than say that the noun pamjatnik 'monument' governs the dative, I claim that if the role of the object is one that calls for a particular case, then that case is selected, otherwise the genitive is structurally assigned. This admittedly leaves open the complex issue of how theta-roles and morphological cases are associated, but I assume that, for example, goals will be dative, everything else being equal. The noun pamjatnik thus assigns no theta-role (i.e. lacks a theta-slot), and the case any argument appears in will be the semantically justified one or the genitive in the default. Thus, when it combines with a goal argument, as in (45b), the dative is used, but otherwise, as in (45a), the object appears in the genitive. Similarly, there must be some connection between being in some loose sense "controlled" and the instrumental to explain (44a), as well as why the related verb also assigns instrumental. Since nouns, unlike verbs, do not actually assign case, however, the fact that complements to both upravljat' 'to manage' and upravlenie 'management' appear in the same case must to some extent be accidental. The verb assigns instrumental because there is an LF requirement that a ' 'controllee" must be associated with this case,25 and the complement of the noun appears in the instrumental for the same reason, even though the case is not assigned by the N directly. It is thus predicted that when a verb assigns a quirky case, complements to related nouns may also appear in that case, or in the genitive, depending on their semantic roles. There are instances, however, where the object of a verb is assigned accusative, but comparable objects of related nouns appear in some semantically motivated quirky case. Consider the following V-N pairs:26 (46)

a. uprekat' Ol'gu/*Ol'ge 'to reproach Ol'ga(acc)/(dat)' b. uprek *Ol'gi/Ol'ge 'a reproach *of/to Ol'ga(gen)/(dat)'

(47)

a. xvalit' Ol'gu/*Ol'ge 'to praise Ol'ga(acc)/(dat)' b. xvala *Ol'gi/Ol'ge 'praise *of/for Ol'ga(gen)/(dat)'

38

PARAMETERS OF SLAVIC MORPHOSYNTAX

The goal is expressed using the accusative with the verbs in (46a) and (47a), but using the dative with the nouns in (46b) and (47b). I take this contrast to mean that it is not an absolute requirement that the goal argument appear in the dative, and the difference lies in the fact that the verbs assign a theta-role to their objects, whereas the nouns do not. That is, the verbs in these examples have theta-slots, but the nouns do not, hence their objects must appear in the semantically appropriate case in order to determine that they bear the goal role. Note, however, that if one attempts to derive nouns from these verbs using productive morphological processes, as in (48) and (49), the noun literally inherits the theta-slot of the base verb and the goal thus reappears in the thematically neutral genitive.27 (48) a. uprekatel' Ol'gi/*Ol'ge 'a reproacher of/*to Ol'ga(gen)/(dat)' b. uprekanie Ol'gi/*Ol'ge 'reproaching of/*to Ol'ga(gen)/(dat)' (49) a. xvalitel' Ol'gi/*Ol'ge 'a praiser of/*for Ol'ga(gen)/(dat)' b. xvalenie Ol'gi/*Ol'ge 'praising of/*for Ol'ga(gen)/(dat)' The other way that the "default" genitive on objects of nouns differs fundamentally from the ' 'default'' accusative on objects of verbs is that it cannot be overridden by the genitive of quantification. Because both involve genitive, however, this can only be demonstrated by considering what happens to NPs quantified by oba 'both', dva 'two', tri 'three', or cetyre 'four', since these require a "paucal" form rather than the genitive plural.28 Compare (50) with (51). (50) a. opisat' dva galstuka 'to describe two(acc) ties(paucal)' b. upravljat' dvumja fabrikami 'to manage two(inst) factories(inst pi)' (51)

a. opisanie dvux galstukov 'a description of two(gen) ties(gen pi)' b. upravlenie dvumja fabrikami 'the management of two(inst) factories(inst)'

The form galstukov 'ties(gen pi)' in (51a) reveals that genitive complements to nouns pattern like other oblique complements and thus that the genitive is not default in the same sense the accusative is. That is, in all relevant aspects the genitive after nouns behaves like a lexically specified (i.e. inherent) oblique case and unlike the default accusative after verbs, although it is clearly structural and not inherent. The genitive rule can be stated roughly as follows: (52) Assign genitive to sisters of N°.

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

39

The problem is now somehow to differentiate this from the accusative rule in (41), in order that (52) and its ilk take effect at D-structure and (41) and its ilk take effect at S-structure. This result can be achieved by eliminating specific mention of accusative case in rule (41) and then letting the appropriate feature values be filled in at S-structure, as indeed has been assumed all along. Rule (41) should thus be revised as follows: (41)' Assign case to sisters of [ —N]° categories. The nominative must also be assigned at S-structure, since it too is overridden by the genitive of quantification, as in (53). (53)

pjat' studentov kupili etu knigu 'five students(gen pi) bought(pl) this(acc) book(acc)'

This can be instantiated in any of several ways: here I adopt the standard approach that nominative is assigned under SPEC-head agreement with I. In any event, it must be assigned at S-structure in order not to percolate throughout the quantified NP as oblique cases do. Note that, for this system to work, the genitive of quantification must be construed as an S-structure case in Russian. Oblique cases override it since they are assigned at D-structure, nominative and accusative ("direct" cases in the terminology of Jakobson, 1958) do not, under the assumption that a minimality condition prevails when two cases are assigned to nested domains at the same level of representation.29 There are, finally, several other reasonable candidates for configurationally determined cases in Russian. These are the dative under sisterhood to V and the instrumental under sisterhood to VP, as stated in (54) and (55).30 (54) Assign dative to sisters of V. (55) Assign instrumental to sisters of VP. As argued in Fowler (1987b), their consistent failure to passivize reveals that dative arguments are never true complements to V, but are rather formally indirect objects.31 Consider his example (56). (56)

a.

ego svedenija sootvetstvujut dejstvitel'nosti his information(nompl) corresponds(pl) reality(dat) ' his information corresponds to reality' b. *dejstvitel'nost' sootvetstvuetsja ego svedenijami reality(nom sg) correspond(sg)-REFL his information(inst) ' reality is corresponded to by his information'

This contrasts strikingly to the behavior of verbs that take genitive or instrumental complements. I maintain that datives are sisters of V, whether the verb is otherwise transitive, as in (57), or intransitive, as in (58). (57) Lena [VP [v. podarila [NP:ACC knigu]] [NP;DAT svoemu drugu]] Lena gave book(acc) her friend(dat)

PARAMETERS OF SLAVIC MORPHOSYNTAX

40

(58)

Boris [VP [v- pomog] [NP:DAT studentam]] Boris helped students(dat)

In contrast to these datives, instrumental sisters of VP are non-argument adjuncts. They thus range freely in interpretation, as in the following example, drawn from Jakobson (1958, 157): (59)

on el ikru rebenkom/pudami/lozkoj/ he(nom) ate caviar(acc) child(inst)/pood(inst)/spoon(inst)/ dorogoj/utrom/gresnym delom road(inst)/morning(inst)/sinful business(inst) 'he ate caviar as a child/by the pood/with a spoon/on the road/in the morning/ to our regret'

These phrases receive case solely by virtue of their configurational status, as in (60). (60)

[VP [VP fv

el

[NP:ACC ikfu]]] INP-.INST rebenkom/pudami/ . . . ]]

Unlike the theme ikru 'caviar(acc)' , they may be associated with any adjunct semantic role compatible with the verb est' 'to eat' . The configurational case assignment principles in (54) and (55) leave open the possibility of quirky case assignment by dint of a lexically specified case submatrix. There is good evidence that V at least exhibits this alternative, since there are verbs whose second argument must be in some case other than dative, such as liSit' 'to deprive'. (61)

rebenok lisll roditelej sna 'child deprived parents(acc pi) sleep(gen)'

Accusative is assigned to roditelej 'parents' under sisterhood to V, but genitive is presumably assigned to sna 'sleep' by V, which in this instance must therefore contain a genitive case submatrix (as well as a theta-slot). Note that when such verbs passivize only the accusative case-assigning ability is affected, as in (62). (62)

roditeli lisilis' sna 'parents(nom pi) deprived-REFL sleep(gen)'

The difference between dative case assignment rule (54) and instrumental case assignment rule (55) can be described in terms of case submatrices—V can contain a case submatrix, whereas VP cannot. This in turn is probably related to the fact that V can take arguments (i.e. contain a theta-slot), whereas VP cannot. Adjuncts to VP in Russian are thus instrumental, and no variation is admitted. Before concluding this section, it is worth taking stock of the different types of phrase structural case assignment encountered and the problems this typology engenders. At one extreme there is quirky case assignment by verbs with lexically specified case requirements, and at the other purely configurational case assignment to objects of nouns. Both of these appear to apply at D-structure, since they cannot be superseded by the genitive of quantification. Only the former, however, is associated with a case submatrix and a theta-slot. I see this as a problem for the inherent/ structural dichotomy often maintained in GB (cf. e.g. Chomsky, 1981a, 292), where

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

41

inherent case is assigned at D-structure and is associated with a particular theta-role, while structural is assigned at S-structure and is semantically unrestricted. Consider, furthermore, the accusative assigned to objects of verbs, which applies at S-structure whether the target NP is an argument or not. Thus, although it is clearly necessary to relegate the effects of different case assignment rules to different levels of representation, this does not seem to correlate in any obvious way with the case assignment mechanism involved. Rather, it is the morphological case per se that appears to be all that is relevant, and the level at which it takes effect must at this point simply be stipulated. That is, claiming accusative is assigned at D-structure but the default feature values are not filled in until S-structure is tantamount to saying it actually applies at S-structure. For Russian at least, then, the default values of nominative, accusative, and genitive of quantification are filled in at S-structure, and all other cases are assigned at D-structure, whether purely configurational or driven by the quirky case features of a theta-role assignor. Following GB practice, I will refer to the D-structure cases as inherent and the rest as structural, although keeping in mind that the way they are standardly depicted (as involving theta-theoretic versus configurational considerations) cannot be maintained. It is naturally expected that quirky case should be inherent in the GB sense, since this involves a lexically specified case submatrix. The problem is thus to explain why some configurationally determined cases are structural and others are inherent.

2.2. Russian Case Features This dichotomy may be made to follow from the properties of a carefully constructed model of case features. It is therefore important to explore specific proposals about the nature of the case features themselves. My point of departure will be Jakobson's (1936, 1958) enlightening and influential work on case.32 I discuss Jakobson's feature system from the perspective of morphosyntactic desiderata. I show how it fails to display the kinds of characteristics necessary for dealing with appropriate problems in section 2.1, and consider alternative proposals that are more morphosyntactically relevant in section 2.2. The study of case relations is central to a proper understanding of Russian morphosyntax. In this regard, Jakobson's work constitutes the single most important contribution to classic case theory. His articles on case features have inspired more research and debate in the field of Slavic linguistics than any other set of ideas. They are standardly discussed in introductory graduate courses and regularly treated in scholarly discussions of case. Here too, Jakobson's ideas will form the intellectual backbone of my model of case. It is therefore essential that I present Jakobson's theory before turning to my own system, which relies on Jakobson's work but also departs from it conceptually in important ways. Although the specific features I will adopt in this section will for the most part play only a marginal role in the analysis of the individual problems addressed in this book, the general system is tacitly assumed throughout, and occasionally explicit reference to the properties of the case submatrix will be required. The way the case system that underlies Russian morphosyntax operates is best demonstrated in chapter

42

PARAMETERS OF SLAVIC MORPHOSYNTAX

3, where the role of feature value non-distinctness and case syncretism is explored in relation to across-the-board w/z-extraction. In chapters 4 and 5, the feature [ ± oblique] will play a crucial part in accounting for the mysterious behavior of Slavic numeral phrases. In the remaining chapters, however, the specific features proposed in section 2.2.2 will be less important than the coindexation mechanisms that are developed to implement case assignment. 2.2.1. Excursus on Jakobson's Features In this section I first present Jakobson's system, then discuss certain problems with extending it to handle morphosyntactic phenomena. Jakobson sought to elucidate the many semantic aspects of case by positing for each case a Gesamtbedeutung,ot "general meaning." He attributed an abstract invariant meaning to each case, arguing that every Sonderbedeutung, or "particular meaning," predictably derives from this general meaning. In order to do this, Jakobson needed to put forward a case theory abstract and flexible enough to capture the general meanings of individual cases. His model therefore assumes a restricted set of features that combine to form a basic invariant meaning for each case. The three necessary and sufficient features proposed in Jakobson (1958) for describing the Russian case system were [± quantified] (ob"emnyj), [ ± directional] (napravlennyj), and [± marginal] (periferijnyj). Jakobson (1958, 179) defines these "semantic marks" as "focusing upon the extent to which the entity takes part in the message," "signalizing the goal of an event," and "assigning to the entity an accessory place in the message," respectively. Accordingly, the following table of feature complexes for the six Russian cases can be constructed:33 (63) nominative accusative genitive instrumental dative locative

= = = = = =

[-marg, [-marg, [-marg, [ + marg, [ + marg, [ + marg,

-quant, -quant, + quant, —quant, —quant, + quant,

-dir] +dir] —dir] —dir] +dir] —dir]

One obvious question about this feature arrangement is how it should be interpreted in terms of markedness.34 One possibility would simply be to consider the feature complexes as unordered sets and count the number of pluses for each case. This straightforward approach gives the markedness hierarchy in (64). (64) nom < ace, gen, inst < dat, loc 0 < 1 <2 An alternative, due to Robertson (1983), is to regard the individual features as ordered, counting each plus as one and minus as zero, and then to treat the result as a binary number. This has the advantage of inducing a markedness relationship between any two given cases. The output of this system, of course, depends directly on the order in which the features are considered. Applying Robertson's principle to (63), in that order, produces the following result:

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

(65)

43

nom < ace < gen < inst < dat < loc 0 <1 <2 <4 <5 <6

Note that two possibilities—[ — marg, + quant, +dir] and [4-marg, + quant, + dir]—have not been realized. Jakobson (1958) remedies this situation by incorporating the so-called second genitive (genitive2) and second locative (Iocative2) into his system as fully fledged cases. Genitive2 is an alternative genitive ending in -u that some masculine singular nouns take in certain set expressions and optionally in constructions indicating quantity. Locative2 is an alternative ending that some monosyllabic35 nouns may take in certain locational constructions. It has the form stressed -u for masculines (and two neuters) and -( for third declension feminines. Both genitive2 and Iocative2 typically occur only on Ns without any kind of modification or determiner.36 Jakobson deemed these examples of Gestaltungskasus ("cases of shaping") and claimed that the regular genitive and locative differ in that they ' 'ascribe to a given entity a property or an action undergone and thus they may be unified with the directional cases into a wider class of ascriptive cases" (1958, 180-181), as follows: (66) genitive2 genitivel Iocative2 locativel

= = = =

[ — marg, [ — marg, [ + marg, [ + marg,

+ quant, + quant, + quant, + quant,

-dir] +dir] —dir] +dir]

This analysis would assign genitive2, genitivel, Iocative2, and locativel markedness values of 2, 3, 6, and 7, respectively. It has the advantage of fully exploiting the three case features, thus allowing Jakobson to represent his system geometrically as a cube, but is in several respects artificial. These secondary cases have a peripheral status and certainly should not be less marked than the primary ones, which can substitute for them. It also strikes me as suspect that Jakobson (1958) apparently switched a supposedly invariant feature in reworking his 1936 system in order to make it more symmetrical. That Iocative2 should be [ +directional] is especially surprising since the very difference between some central uses of the accusative versus the locative is one of directionality—for example, v Leningrad 'to Leningrad(acc)' but v Leningrade 'in Leningrad(loc)'—although of course this is one reason why he proposes the new term "ascriptive." Jakobson adopted a two-pronged assault on morphological case in justifying his analysis of the Russian case system. First, he examined diverse usages of the various cases, attempting to demonstrate that they all share the appropriate features. Second, he showed that case syncretism corresponds to the neutralization of case oppositions. I take up each of these issues in turn. The first problem with Jakobson's system has to do with getting all and only the functions of each case to line up with its "meaning," this being understood solely in terms of its feature content. In order to see the difficulties involved, one might compare these semantic marks with phonological distinctive features. Taking language to be a sign system mediating between sound and meaning, Jakobson grounded phonological features in the former and case features in the latter. Both

44

PARAMETERS OF SLAVIC MORPHOSYNTAX

thus have some kind of real (and potentially confirmable) basis, although this is considerably more transparent for phonological features, which are acoustically or articulatorily based. Similarly, both kinds of features must have explanatory value in that if one thinks in terms of the features themselves rather than the entities they describe (sounds or cases, respectively), then it should be easier to understand just why the kinds of linguistic patterning encountered exists. For example, in order to make linguistically significant generalizations, we need phonological features that pick out the relevant sets of sounds and properties—a feature that never played any role in phonological patterning would be impossible to motivate, for it would be neither inducible nor testable. Likewise, the existence of a phonological process that is not comprehended by whatever feature set one is assuming might mean that this set is deficient. The question now is, What happens if one applies these kinds of criteria to Jakobson's case features? These features are intended to explain all the different ways the cases are used. There are, however, well-known difficulties in getting from the general meanings of the cases to each of the particular ones, and vice versa. This is a serious prima facie conceptual stumbling block in that two different cases can be used in conjunction with virtually identical meanings, such as the accusative in pro knigu but the locative in o knige, although both mean 'about (the) book'. The other side of the coin is that the same case can be used with very different meanings, as in (59). It thus seems extremely doubtful that morphological case constitutes a direct pairing between form and meaning.37 Consider, for example, the instrumental utrom 'in (the) morning' in (59), which refers to the time of eating. It could be replaced by any number of expressions, such as a preposition followed by the accusative (vo vtornik 'on Tuesday'), the locative (v detstve 'in childhood'), the instrumental (pered ekzamenom 'before [the] exam'), the genitive (posle poludnja 'after noon'), or the dative (po celym dnjam 'for whole days'), as well as the bare (prepositionless) genitive (pervogo maja '[on] May first') or accusative (yse vremja 'all [the] time'). It is far from clear how the subtleties of meaning here can be made to relate to the proposed features. I think instead that at some level the grammar simply has to state that, for example, posle 'after' governs the genitive and pered 'before' the instrumental. The use of the instrumental in (59), on the other hand, is probably best understood in terms of some shared syntactic property—earlier I suggested that they can all be configurationally described as VP adjuncts.38 Semantically, however, they are about as diverse as conceivable for a non-subcategorized element, their meaning depending solely on the range of likely interpretations for the NPs in question, so that, for example, lozkoj is interpreted as 'with a spoon' because lozka 'spoon' is an implement rather than a time or place, but other adjunct roles are not inconceivable—it could mean 'when he was a spoon' or 'like a spoon eats' in a fairy tale.39 Jakobson's formal arguments for his particular features were both phonological and morphological. In keeping with his general theory of language as a semiotic system, he argued for an isomorphic relationship between morphological categories and their expression. That is, just as the features have semantic correlates, they must also have phonological ones. Jakobson examined the non-syllabic phonemes that occur in Russian declension and noted that of the thirty-three that are possible, only

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

45

four, /j v m x/, are actually present. Of these, he maintained that the fricatives l\l and /x/ mark the quantified cases (with the exception of animate masculine accusative, which uses genitive form) and /m/ marks the marginal cases. He also associated the syllabicity and number of phonemes in a desinence with the morphological category of the ending. The putative "generalizations" adduced in section IV of Jakobson (1958), however, amount more to a collection of random and curious observations than to any system for relating sound and meaning. For example, he noted that (1) the zero (i.e. phonologically null) ending occurs in only one of the nominative singular or genitive plural, (2) /v/ (and /f/?) signifies the genitive, (3) /e/ signifies the locative, (4) an independent accusative form implies the nominative ending is -a, (5) a polyphonemic ending will contain /j/ in the nominative or accusative (the "direct" cases) and in oblique feminine forms, and so on. There are, however, disconcerting problems with many of these observations, on both the empirical and conceptual level. Empirically, his statements often seem to arise with little rhyme or reason (Why is the fact that /v/ implies animacy in Czech relevant?), and, more seriously, virtually every one of Jakobson's "generalizations" admits exceptions of a stipulative nature. That marginality implies that the presence of/m/, for example, is overridden by the possibility of/j/, and /j/ in turn is overridden in the locative plural by the possibility of /x/, which is understood as a sign of quantification. It is true that instrumental is most marked in that it generally has more phonemes than any other case, but this too can be overridden—when the instrumental is shorter, as with the masculine singular instrumental ending -ym versus the dative -omu, Jakobson observed that it is still most marked in that it is the only ending not beginning in -o. The problem is, however, that Jakobson's case features actually assign the highest markedness value to locativel, with three pluses, and instrumental, which is only marked for marginality, ends up being relatively unmarked. Note that even if, as I will indeed argue, instrumental is the most marked case, this is a case-based rather than feature-based generalization. This points out the major conceptual problem with Jakobson's observations—to the extent that the correlations he noted are correct, they hold between sound patterns and cases, not features. Moreover, the sets of cases he referred to generally do not correspond to some straightforward feature complex and can often only be expressed using complex boolean combinations of features. In order to overcome this deficiency, Jakobson (1958) proliferated his terminology, introducing several new oppositions, such as direct versus oblique and indefinite versus definite, as in (67). (67) a. direct oblique b. indefinite definite

= = = =

nominative, accusative all others nominative, instrumental all others

These were presumably new primitives of the system, since to express [ + definite] one would have to make use of a statement like "either [ +quant] or [ + dir]," although [ +direct] would be simply "[ — quant, — marg]." Indeed, it seems as if these features are more appropriate to characterizing the phonological properties of

46

PARAMETERS OF SLAVIC MORPHOSYNTAX

cases than are the core ones. This is even more true of the morphological syncretism facts discussed by Jakobson, to which I now turn. The other phalanx of Jakobson's attack on case features finds its strength in morphological arguments. Using the proposed feature system, Jakobson (1936, 65) suggested that the following table of distinctive feature oppositions can be constructed: (68) (nom ~ ace) ~ (gen ~ gen2)

I

!

!

!

(inst ~ dat) ~ (loc ~ Ioc2) In (68), the marked member of each case opposition is to the right of or below the unmarked one. Jakobson then went on to show how in the various Russian declensional paradigms one or more of these oppositions is neutralized. He did this by setting up a grid (for the six "primary" cases) as in (69).

(69)

nom ace inst dat

gen loc

In every pattern discussed by Jakobson, morphological syncretism involves the coalescence of adjacent case boxes. For example, consider his presentation of the class I "feminine" type of nominal declension in (70). (?0)

I a I u | j oj(u) I e

The dative and locative fall together here, exhibiting a single ending -e. This system appears to be very elegant; one almost never finds two non-adjacent boxes coalescing.40 The various syncretisms Jakobson concentrated on are summarized as follows: (71) nom ace ace dat dat dat gen

= = = = = = =

ace: gen: gen gen loc: gen loc:

= =

loc: loc:

=

loc

=

inst:

okno syna nas no£i zene sta zlyx

'window' 'son' 'us' 'night' 'wife' 'hundred' 'mean (pi inan)'

The problem is that, although these syncretisms can indeed be expressed by erasing the lines separating adjacent boxes in (69), this is not an especially illuminating fact with respect to Jakobson's particular case features. Erasing lines between adjacent boxes reflects case neutralizations rather than case feature neutralizations. Although Jakobson's arrangement of the boxes is interesting in its own right, it is not apparent how or if this should be connected to the feature system he was trying to defend. Reference to cases rather than case features shows that syncretisms provide no grist for Jakobson's mill and, in fact, lead to the conclusion that these case features are morphosyntactically irrelevant.

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

47

If syncretism is instead regarded as the neutralization of feature oppositions, as in phonology, then one might ask what syncretisms Jakobson's features actually predict. For example, consider how genl and gen2, loci and Ioc2 might collapse. Since (in the 1958 system) they differ with respect to directionality (ascriptiveness), it must be this feature that is neutralized for the many words that do not distinguish the primary and secondary genitive and locative. Thus, the form torta 'cake' would be listed in the lexicon with the features [ —marg, + quant], allowing it to fit into contexts where either [ + dir] genl or [-dir] gen2 is called for, assuming nondistinctness as the appropriate criterion for wordform insertion. Other syncretisms, however, do not correspond so readily to underspecified or "reduced" feature matrices. In (70), for example, there is no feature neutralization that can account for a syncretism of dative ([-quant, +marg, + dir]) and locative ([ +quant, -fmarg, — dir]), since singling out the marginal cases would fail to exclude instrumental. Similarly, there is no underspecified matrix that would refer exclusively to accusative and genitive, although this is a frequently encountered syncretism. The complete set of potential syncretisms picked out by the 1936 feature system is as follows: (72)

[-quant] [-dir] [ — marg] [ +quant] [ + dir] [ + marg] [-quant, -dir] [-quant, -marg] [ — quant, +dir] [ — quant, +marg] [ +quant, —dir] [ +quant, -marg] [-I-quant, +dir] [ +quant, +marg] [ — dir, —marg] [-dir, +marg] [ + dir, —marg] [ + dir, +marg]

= = = = = = = = = = = = = = = = = =

nom, ace, dat, inst nom, gen, inst, loc nom, ace, gen gen, loc ace, dat dat, inst, loc nom, inst nom, ace ace, dat dat, inst gen, loc gen 0 loc nom, gen inst, loc ace dat

Of these, only two projected syncretisms (nom = ace, gen = loc) actually correspond to existing Russian case syncretisms, leaving the other five in (71) unaccounted for. I therefore conclude that Jakobson's features do not find support in (at least this aspect of) inflectional morphology. In general, then, Jakobson's distinctive features for Russian do not appear to be morphologically active. I believe that the error lies in his having taken semantic invariance, rather than morphosyntax, as a point of departure. Nonetheless, syntacticians in need of some feature set for cases have often taken these features as being applicable to morphosyntactic properties of Russian. Neidle, for example, in her 1982 dissertation, which develops a Lexical-Functional Grammar (LFG) analysis of Russian case, observes that if one considers the plural adjectival endings, the

48

PARAMETERS OF SLAVIC MORPHOSYNTAX

nominative-accusative syncretic form -ye can be characterized as [ - quant, - marg] and the genitive-locative one of -yx as [ + quant, — dir] (ignoring the complicating factor of animacy). The problem with this is that, although Neidle reasonably cites these two syncretisms to argue that such decomposition is realistic and would produce great economy and generalization throughout the lexicon, these two are, as shown above, the only ones that actually work using Jakobson's three features. One cannot therefore help but wonder what sort of feature system would arise if a systematic investigation of morphosyntactic properties were undertaken. It turns out, in fact, that a modified version of Jakobson's system appears to be appropriate in that it is better able to address the kinds of morphological issues I have raised.

2.2.2. Revising Jakobson's System In this section I discuss desiderata for an appropriate set of morphosyntactic case features, presenting the systems described in Franks (1985) and Neidle (1988). In doing so I shall be more concerned with the kinds of considerations involved and the properties a viable feature system ought to possess than with putting forward a definitive system, since the precise features will be immaterial for much of the argumentation in this book. It should be noted, however, that other scholars have recently come to recognize the difficulties in applying Jakobson's features to morphological syncretism, including Viel (1984), Chvany (1986), Neidle (1988), and McCreight and Chvany (1991). Neidle's (1988, 2-6) reconsideration of Jakobson's features in revising her 1982 dissertation is particularly illustrative, I think, and most of her objections echo my own. She modifies Jakobson's system as follows, where the feature [±partial] identifies the two secondary cases, [ ± ascriptive] is broken down into [ ± locational] and [±directional], and dative is classified with genitive and locative as [ +quantifying]: (73) nominative accusative genitivel genitive2 Iocative2 locativel dative instrumental

= = = = = = = =

[ — locational, [ — locational, [-locational, [-locational, [ +locational, [ +locational, [-locational, [-(-locational,

—quantifying, —quantifying, + quantifying, -(-quantifying, -(-quantifying, -(-quantifying, -(-quantifying, -quantifying,

—directional, + directional, + directional, + directional, —directional, —directional, —directional, -directional,

—partial] —partial] —partial] + partial] -(-partial] —partial] -partial] -partial]

Neidle states that "partial" is intended to "capture the physical/quantificational part:whole relationship emphasized" by genitive2 and Iocative2, and the "directional" versus "locational" opposition reflects the fact that "a preposition may select different cases according to whether it is used to express motion or location," with location construed "in a broad sense to include not only spatial location, but time intervals and other restrictions." The interested reader is referred to Neidle (1988) for discussion of how these particular features operate within the LFG framework to derive the necessary ef-

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

49

fects. It suffices here to point out that the syncretisms in (71) can be accommodated as follows, using feature neutralization: (74) nom ace ace dat dat dat gen

= = = = = = =

ace: gen: gen = loc: gen = loc: loc: gen = loc = inst: loc:

[ — loc, -quant] [ — loc, +dir] not expressible [ +quant] [ +quant, -dir] not expressible not expressible

It should be evident that, although Neidle's 1988 system is an improvement over Jakobson's with regard to its ability to express syncretisms in Russian declension, it remains empirically inadequate. The problem is that it still essentially relies on three features, [± partial] only playing a peripheral role. A system with at least four features, however, is neccessary to handle the observed sycretisms. I now turn to such a system, as outlined in Franks (1985). As an initial attempt at a reasonably promising feature system, I propose taking advantage of the auxiliary features [± indefinite] and [±oblique], which Jakobson (1958) demonstrated were more phonologically active than the three primary ones. If [±indefinite] replaces [±directional], and [±oblique], whose relevance to morphosyntax is emphasized in Chvany (1986), replaces [±quantified], the following system emerges: (75)

nominative accusative genitive dative instrumental locative

= = = = = =

[ + indef, [ —indef, [ — indef, [ — indef, [ +indef, [ — indef,

-obi, -obi, +obl, +obl, +obl, +obl,

-marg] —marg] —marg] +marg] +marg] +marg]

This system can accommodate most, but not quite all, of the syncretisms. It is still incomplete in that dative and locative remain undifferentiated.411 therefore assume that a fourth feature will be required, tentatively [±F 4 ]. Two related questions may now be posed about [±F4]: 1. What is its morphosyntactic significance? 2. What values does it have for the various cases? I begin by attacking the second question, whose resolution must logically precede any attempt to determine a morphosyntactic correlate for [ ± F4]. First of all, notice that in (75) accusative has a minus value for all three features and instrumental a plus. This roughly accords with the markedness considerations discussed in the last section, if accusative is regarded as the least marked case on the basis of its being the default case assigned by V and P, and instrumental as the most marked on the basis of morphological criteria, to be presented shortly, and assuming that degree of markedness corresponds directly to the number of positive feature values. Assum-

50

PARAMETERS OF SLAVIC MORPHOSYNTAX

ing, therefore, that accusative is [ —F 4 ] and instrumental is [ + F4], genitive and locative must also be [ - F4] because the only syncretism (75) fails to handle with a reduced matrix is "ace = gen = loc," hence [±F 4 ] must bear the same value for these three cases. Dative, which must contrast with locative, is consequently [ + F4]. The matrices in (75) can thus be augmented as in (76), with the appropriate value of [±F4] for nominative still unresolved. (76) nominative accusative genitive dative instrumental locative

= = = = = =

[?F4] [ - F4] [ - F4] [ + F4] [ + F4] [ - F4]

The issue of whether nominative is [±F4] is not easy to resolve. Evidence from syncretisms makes no specific indication either way. On the one hand, if number of pluses determines relative markedness, then perhaps it would be better for nominative to be [ - F4], so that it only has a positive value for one feature, [ + indef]. On the other hand, speculations about the possible significance of [ ± F4] would seem to lead to the opposite conclusion. Consider how dative and instrumental might be opposed to accusative, genitive, and locative. One idea that comes to mind is that dative and instrumental are configurationally assigned under sisterhood to X' and XP, respectively, rather than X°. Of course, dative and instrumental may be idiosyncratically assigned by heads, such as the preposition k 'toward' for dative or the preposition nod 'above' and the verb upravljat' 'to manage' for instrumental. The point is that there is no purely configurational context for assigning accusative, genitive, or locative that makes reference to projections of a head X°. Let us express the fact that indirect objects and adjuncts are assigned case not by terminal nodes but rather by higher projections of these nodes ,42 by the mnemonic feature [ ± phrasal]. The [ + phrasal] cases are assigned by being associated with (not necessarily maximal) phrasal projections. Returning, then, to the issue of the nominative, it becomes clear that, since nominative is assigned to the specifier of IP (perhaps under sisterhood to INFL'), it should be t + phrasal]. I therefore tentatively ascribe nominative the value [ +phrasal], completing the four-feature system. Although this decision has no immediate consequences, in chapters 4 and 5 the analysis of the genitive of quantification as [ — oblique] will force it to be differentiated from the nominative in some other way. Since four features potentially delimit sixteen cases, there are several redundancies inherent in the feature system as it now stands, which can be stated as redundancy rules such as [ - obi] > [ — marg] and [ + marg] > [ + obi]. However, it will turn out that when the existence of a [ — oblique] genitive and dative is postulated in chapters 4 and 5, all implicational redundancies in the proposed system disappear. This suggests that the case features are completely independent of one another. Of more interest is the observation that [ — indef] is unnecessary in identifying the genitive and locative. This suggests that [± indef] could be used to distinguish Ioc2 and gen2 from loci and genl. Although Jakobson did not indicate whether these cases differed in terms of definiteness, one might expect Ioc2 and gen2 to be

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

51

definite on the basis of their meanings. This translates into [-indef], however, which runs counter to the desideratum of opposing the secondary cases to their primary variants as more marked. I therefore propose as an alternative replacing (in)definiteness with the (non)ascriptiveness feature of Jakobson (1958), which lumps together nominative, instrumental, Iocative2, and genitive2 as nonascriptive. However, in keeping with markedness considerations, I take [ +nonascriptive] as the marked value, which gives rise to the system in (77). (77) accusative nominative genitivel genitive2 locativel dative Iocative2 instrumental

= = = = = = = =

[ — obi, [ — obi, [ + obl, [ + obl, [ + obl, [ + obl, [ + obl, [ + obl,

— marg, —marg, -marg, —marg, +marg, +marg, +marg, +marg,

— nonascr, + nonascr, -nonascr, + nonascr, -nonascr, -nonascr, + nonascr, + nonascr,

— phras] + phras] —phras] —phras] —phras] + phras] —phras] + phras]

The cases and their feature correlates in (77) are arranged in order of increasing markedness, exploiting the binary number system of Robertson (1983) and ranking the features as follows: oblique = 8, marginal = 4, nonascriptive = 2, and phrasal = 1. This results in the following relative values of markedness relations among the cases: ace (0) < nom (3) < genl (8) < gen2 (10) < loci (12) < dat (13) < Ioc2 (14) < inst (15). This ranking raises various questions of naturalness and the senses in which markedness is intended. Other rankings are of course possible, depending on the features selected and their relative markedness strengths.43 I adopt the system in (77) both for its ability to explain the morphosyntactic problems treated in this book and for its minimal departure from Jakobson's features. Although (77) is adequate for current purposes, I fully expect that a better system can be devised, particularly on the basis of markedness, learnability, and comparative considerations. Proper treatment of these issues is beyond the scope of this study, however, and I will henceforth assume (77) when reference to specific case features is required. Turning, then, to the problem of syncretism that initiated (77), note that the syncretisms discussed above can now be expressed as follows:44 (78) nom ace ace dat dat dat gen

= = = = = = =

ace: gen: gen = loc: gen = loc: loc: gen = loc = inst: loc:

[-obi] [-nonascr, -marg] [-nonascr, -phras] [-nonascr, +obl] [-nonascr, +marg] [ + obl] [ + obl, -phras]

To this list one might add that neutralizing [ ± nonascriptive] for the genitive and locative allows these cases to be represented as [ + obl, —marg, —phras] and t + obl, +marg, -phras], respectively. At this juncture, let me briefly point out some of the paradigmatic concerns leading to my conclusion that instrumental should be regarded as the most marked case. These arguments for relative markedness have to do with mixed declensions.

52

PARAMETERS OF SLAVIC MORPHOSYNTAX

These are declensions of a single lexical item that belong to some combination of paradigms, typically distinguished by declension class or syntactic category. Their significance is that they serve to establish a dividing line or opposition within the declension, a division that presumably falls along markedness lines. Consider, for example, the role of singular versus plural in characterizing the mixed verbal paradigms of, for example, esf 'to eat', dat' 'to give', and xotet' 'to want', or in describing Russian nominal stress paradigms. I contend that arguments from mixed declensions also show that the instrumental is the most marked case. The instrumental form of put' 'path', for example, is opposed to all the other forms in terms of declension class, since it alone exhibits the appropriate masculine ending: putem versus genitive, dative and locative puti, which looks like the feminine 0-stem declension, and nominative and accusative put', which could represent either type. Similarly, the instrumental of surnames like Puskin is adjectival in form (Puskinym), although all the other case forms are nominal. The morphology of Polish numerals also suggests a special status for the instrumental. For one thing, the ending -oma is opposed to that of all other obliques, but more striking is the fact that with collectives the instrumental is the only form that governs, like a noun, rather than agreeing, like an adjective.45 Thus, for example, one finds z dwojgiem dzieci 'with two(inst) children(gen pi)', contrary to the agreement pattern found with collectives in any other case. These facts support the idea that the instrumental is at the marked end of the case spectrum, being most resistant to morphological change and generally standing out against the other forms, both morphologically and, as observed by Jakobson himself, phonologically. Another interesting conclusion revealed by the "mixed declension" approach is that the paucal genitive after Russian dva 'two', tri 'three', and cetyre 'four' is less marked than the oblique cases and, in particular, than the regular genitive. This can be seen in the declension of female surnames such as Puskina, where the direct cases are nominal and the oblique ones adjectival—itself a good argument for the direct/ oblique opposition. Curiously, and so far as I am aware not previously observed,46 the paucal patterns with the direct cases rather than the oblique ones: dve Puskiny 'two (female) Pushkins' displays the nominal genitive ending, in contrast to the regular genitive form Puskinoj, which is adjectival. This opposes the paucal to the oblique cases in a surprising way. Let us return, finally, to the problem of the mechanics of default accusative assignment, which arose in the previous section. Recall the following dilemma: principles of case assignment dictate that accusative is assigned to arguments at D-structure and to adjuncts at S-structure, but in both instances it behaves as though assigned at S-structure. This was in particular demonstrated by its interaction with the so-called genitives of quantification and sentential negation. Moreover, since the accusative assigned by prepositions behaved the same as that assigned by verbs, it was claimed that both are default cases, although how to formalize this notion remained obscure. It may now be possible to accomplish this in terms of the accusative's minimal markedness, along lines suggested in Franks (1985,1986) and Neidle (1988). Rather than endow the case submatrix of a verb or preposition that assigns accusative with minus specifications, I propose leaving these empty until S-structure, where they are filled in as minuses in keeping with their unmarked nature. A

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

53

verb or preposition that assigns accusative, then, really just assigns case, and this takes effect at S-structure, even though theta-role assignment/coindexation still takes place at D-structure. There are several other reasons to think accusative is the least marked case and that this is true whether assigned by a verb or preposition. Briefly, the morphology of the accusative suggests this in that in general in Russian, with the exception of the feminine a-stem declension, there is not even any specific form for the accusative. That is, masculine, neuter, and plural accusative forms are either equivalent to the genitive if animate or the nominative if inanimate. Moreover, even in the feminine declension the accusative is opposed to all other cases, including the nominative, in that the accusative singular -u desinence has no inherent stress, giving rise to the "M-retraction" paradigm, which exhibits desinential stress in all cases except the accusative. Following the analysis of Hamilton (1980) or Melvold (1987), this means that the accusative is unmarked for stress and thus receives stress by a default rule. Another area in which the accusative seems to be the maximally unmarked form is in the morphology of quantified expressions in various Slavic languages. Quantifiers often appear as fixed, frozen forms resembling the accusative. This might be explained if these frozen quantifiers are really caseless, under the assumption that being unmarked for all case features, that is, accusative, is tantamount to caselessness. For example, as will be treated in chapter 5, Russian and Polish classifier quantifiers such as masu, masq 'a mass of sometimes appear in the accusative even in nominative contexts, and in Polish entire quantified NPs are for some reason marked accusative as subjects. In Serbo-Croatian, the frozen accusative has been even more generalized, so that accusative forms such as hiljadu 'thousand' appear even in oblique contexts. Finally, one important reason why it is claimed that prepositions, like verbs, assign accusative in the absence of any case feature specifications is that whenever a P may assign one of two cases In related meanings, the accusative participates in the alternation.47 To demonstrate this point, consider the following two sets of Russian prepositions and the cases they may govern: (79) a. b. c. d.

v gorod na vokzal pod stol za stol

'to city(acc)' 'to station(acc)' 'to-under table(acc)' 'to-behind table(acc)'

(80) a. b. c. d.

v gorode na vokzale pod stolom za stolom

'in city(loc)' 'at station(loc)' 'under table(inst)' 'behind table(inst)'

The prepositions v, na, pod, and za all govern the accusative when they imply direction, that is, take a goal argument. I take this to mean that in this usage the feature values of their case submatrices are open. When they take locational arguments, on the other hand, the case submatrices are specified as locative ([ + obl, + marg, -phras]) for v, na48 and instrumental ([ + obl, + marg, +nonascr,

54

PARAMETERS OF SLAVIC MORPHOSYNTAX

+ phras]) for pod, za. I thus conclude that when these prepositions are locational they contain fixed feature values and when they are directional they contain open ones, with the result that their complements appear in the default accusative regardless of the case governed in the locational usage. Note that the two prepositions complementing pod and za, namely, nad 'above' and pered 'in front of, take instrumental complements in both senses. However, this was not always so, and in Old Church Slavonic (cf. Vaillant, 1948,189) nad" andpraf' behaved like modern Russian pod and za, assigning accusative or instrumental. Old Russian pered" and nad" allowed for a similar flexibility, depending on the role of the complement (cf. Borkovskij and Kuznecov, 1965, 474; Bulaxovskij, 1958, 311). This loss of flexibility can be analyzed as the generalization of the instrumental to the directional construction, so that the case features of nad and pered became fixed as instrumental in Russian, making the unmarked accusative—which arises only in the absence of lexical specification—no longer possible. It is important to realize that the proposed system, whereby unspecified (unmarked) feature values are filled in with minuses at S-structure, does not necessarily hinge on the accusative being analyzed as the absolutely least marked case, although this is the simplest assumption. Alternatively, one might contend that the accusative is Mnderspecified, rather than completely unspecified, so long as its realization is delayed until S-structure. This is roughly the approach taken by Neidle (1988), translating her representational LFG analysis into a derivational GB one. Recall that for her the accusative is [ - locational, — quantifying, + directional] and the genitive is [ - locational, + quantifying, + directional]. She thus maintains that verbs assign the feature complex [ — locational, + directional] with the value of [ ± quantifying] independently determined on the basis of whether or not the object is in the scope of negation. This allows Neidle to make the traditional Jakobsonian assumption that accusative is marked at least for directionality, so long as any case it alternates with (namely, genitive) is similarly marked. Notice, however, that it puts the onus of the alternation on the assignment rules rather than the case itself, so that it is, correctly, only the accusative assigned by verbs that participates in the genitive of negation construction. Nominative subjects are also unavailable for this alternation, since nominative is [ - directional] and the addition of [ + quantifying] would therefore produce dative rather than genitive. However, Neidle is forced to adopt a completely different analysis for the quantificational genitive, which does apply to nominative subjects, and develops a stipulative account in terms of defective paradigms. Perhaps this too might be assimilated to the underspecification approach given appropriate case features, but I have been unable to devise an adequate system that satisfies the other morphosyntactic criteria discussed in this chapter. This issue nonetheless raises again the problem of how the structural cases— nominative, accusative, and, in Russian at least, the quantificational genitive—are to be distinguished from the inherent ones. Recall that their realization somehow needs to be delayed until S-structure. Although I have no definitive solution to offer, the underspecification approach seems the most promising, short of simply designating the [-oblique] cases as assigned at S-structure. In this scheme, regardless of when the relevant plus values are assigned, the minus feature values are not filled in until S-structure, although only for these structural cases. This, of course, is the

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

55

most troublesome aspect of the structural/inherent dichotomy. It is relatively easy to differentiate nominative and accusative consistently from the other cases; the problem ultimately hinges on the quantificational genitive, which, I shall argue in chapter 4, is parametrically determined to be either structural or inherent, with little or no apparent substantive difference. This reveals the somewhat arbitrary character of the opposition, which I will argue, should be assimilated to the [± oblique] feature, analyzing the Russian quantificational genitive as [ — oblique], in contrast to the other genitives.49 I leave this recalcitrant problem unresolved, in the hope that a more satisfying feature-based analysis will eventually be forthcoming, since the primary point of this chapter—that the individual cases are best expressed as bundles of morphosyntactically driven features—is independent of any definitive statement of those features. I reiterate my belief, however, that some modified version of Jakobson's features that pays less heed to semantics and more to morphology will ultimately prove to be correct. In concluding this chapter, I think it is important to stress the fundamental theoretical thrust of Jakobson's separation of morphology from syntax. Although Jakobson (1936, 28) argued that "die Kasus morphologischer, nicht syntaktischer Natur sind'' and quoted Br0ndal as writing ' 'es gibt kein notwendiges Verha'ltnis zwischen einer Kasusfunktion und Satzfunktion; Kasuslehre und Morphologic sind nicht Syntax," I think his decision to overlook the qualification "notwendiges" and outrightly reject case as syntactic was misguided.50 This left him unable to recognize that sometimes case may do no more than mark syntactic or grammatical relations. The problem is that morphology is a place where syntax, semantics, and phonology all meet. In part, Jakobson was protecting morphology from syntax, as suggested by the following warning (1958, 182-182; translation from Jakobson, 1984): It goes without saying that bridges must be built between the two grammatical disciplines, morphology and syntax; but at the same time it does not follow that we should ignore the relative autonomy of these two subdomains and reduce the morphological problem of word inflection—in particular, the questions of the external and internal structure of cases—to the syntactic problem of phrase structure.

In claiming that morphological entities are endowed with general meanings that somehow become particular in a syntactic environment, Jakobson was burning those very bridges he deemed necessary to construct. Case can and indeed must sometimes be reduced to phrase structure. By trying to characterize case purely in semantic terms, Jakobson in my opinion was forced to distort and/or ignore as much as he was able to explain. I have argued in this chapter instead that the correct features for case, which can largely be chosen from the extended set proffered by Jakobson, must be selected on the basis of morphosyntactic rather than semantic criteria. Notes 1. See Pranka (1983) for somewhat similar proposals. 2. Something like this is assumed in the model of language production proposed in Garrett (1988). See also Halle and Marantz (1992) for an explicit model of "Vocabulary Insertion" that makes some similar claims.

56

PARAMETERS OF SLAVIC MORPHOSYNTAX

3. Translating this system into a checking model would involve differentiating features that need to be checked from those that do not. I take this to be an interesting mechanical problem. 4. Actually, although this conception of the restricted visibility of different modes of representation does force a bifurcation of the grammar, it does not necessarily force the T-model, since it also allows LF rules to "see" any level prior to wordform insertion. 5. Other more elaborate systems, such as multivalued or analog features, are conceivable. Also, the schema in (1) is by no means meant to eliminate all redundancy or be exhaustive. In particular, it ignores the obvious need to indicate that the word inserted at D-structure is knig'book.' Note, however, that once this is done specification for person becomes superfluous, since third person presumably follows from the semantics of knig-. Alternatively, person is deducible from the feature set [ - a, - p], although this itself is probably predictable from the meaning of knig-. 6. While further internal structure is probably necessary, for example, to reflect the division between derivation and inflection or the fact that the pronominal features generally behave as a unit, this issue is ignored for present purposes. 7. Another theory-internal reason for believing the binding theory cannot apply at D-structure is that at this level of representation PRO can be governed, as in John hopes [e to be admired PRO by all]. It is standardly assumed that PRO cannot be governed, and that this distributional property can be derived by virtue of its need to satisfy both principles A and B of the binding theory simultaneously. See chapter 6, as well as Franks and Hornstein (1992), for discussion and an alternative proposal. 8. Under a system in which Ns acquire gender and/or number features from dominating gender and/or number phrases, these submatrices would also be unindexed at D-structure. 9. Williams (1982, 279) also assumes such a coindexation process, which he calls "the inheritance of indices through heads." He suggests that this idea derives from the theory of morphology in Williams (1981), where features percolate up to words from their morphological heads. 10. The extra N' node is necessitated by the traditional assumption that APs are usually sisters of N', which can iterate. In more recent models, such non-branching intermediate nodes are unnecessary (cf. e.g. Speas, 1990, for discussion). 11. I assume that the indices on the part-of-speech submatrix of the AP remain unchanged, whereas those on A' and A must all be consistent with the new index of AP. The reason for this is that the part-of-speech is determined when the item is drawn from the lexicon, whereas the properties of X-bar syntax, including the projection of phrases from heads, hold at all levels of syntax. This assumption has no empirical consequences, however, since only submatrices with fixed feature values are involved. 12. Since the uniqueness assumption is not crucial to any of the analyses in this book, I do not justify it here. See Franks (1985, 1988), however, for detailed arguments that a theta-role assignor only assigns one theta-role through syntactic coindexation, as well as discussion of the principles according to which nodes select a role for their syntactic theta-slot from the lexical theta-grid. See also Grimshaw (1990) and Speas (1990) for general discussion of these and related issues. 13. The reason for divorcing the locality condition from case assignment per se and associating it with theta-role assignment, contra the Case Adjacency Principle of Stowell (1981), will become clear when long-distance predicate adjective agreement phenomena are considered in chapter 6, since here case assignment is not contingent on theta-role assignment.

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

57

14. Thus all complements are also objects, but there may be objects that are not true complements. 15. If it turns out that this is too strict a requirement, then one might relax the bidirectionality such that "every argument must be bound by a unique theta-slot." 16. See Battistella (1981) and Fabb (1984) for conceptually similar proposals. 17. This construction is probably a caique on the comparable German wasfitr ein, literally 'what for a' (cf. Sandfeld Jensen, 1912). 18. Not all speakers report these judgments. 19. This assumption follows if all aspects of trace theory and all properties of traces, including their category, are determined by independent principles. As will be discussed in chapter 4, Pesetsky (1982) argues that traces realize the canonical category of the argument type associated with the position. Unfortunately, in (28) the verb mention takes a proposition whose Canonical Structural Realization (cf. Chomsky, 1986b) is presumably CP. Note, however, that there might be other reasons to select a particular category for a trace and that in this instance case and theta-theories have precedence—the trace must be an NP precisely so that it can be assigned case. 20. This approach leaves open the possibility of inherent accusative. Since I have found no evidence that accusative is ever not purely phrase structural in Slavic, I ignore this possibility here, although noting that there is reason to assume its existence in other languages (cf. e.g. Freidin and Sprouse, 1991; van Riemsdijk, 1983). 21. Chomsky does note that idioms are a "possible exception," to which one would probably want to add cognate objects. Other linguists, such as Jackendoff (1990a), argue against such a rigid correspondence between syntactic and semantic structure. 22. See Comrie (1976), Revzin (1973), and Zimmermann (1967), as well as Prokopovic et al. (1981), for general discussion. 23. In this sense, it is just as free as are genitive possessives in English, which, it has been argued (cf. the "Del Rule" of Williams, 1982; and Franks, 1988), is in principle completely unrestricted. 24. See Vlaxov and Muckov (1974, 292), from which (44b) is taken. 25. I say "associated with" rather than "in" because of examples like (37), where/izfcrifaz is still in a chain with instrumental in that the verb contains an instrumental case submatrix, but these features have been "absorbed" by the reflexive morpheme -sja. 26. As with all objects of N, the genitive in (46b) and (47b) is of course admissible, but not with the goal reading. 27. Although the speakers who provided these judgments regarded the items in (48) and (49) as innovations, these forms are not totally hypothetical. The words in (48) are cited in Dal' (1880-1882), although not in any of the four dictionaries (both the four- and seventeenvolume Academy dictionaries, as well as Ozegov and Usakov) listed in the Obratnyj slovar' ('reverse dictionary'), and those in (49) are in some dictionaries but not in others (e.g. Ozegov). 28. Although the paucal form is generally identical to the genitive singular, it occasionally differs in stress and is also distinct for female last names of the pronominal declension class, as discussed in the next section. 29. The details of case assignment within quantified NPs in Slavic is the subject of chapters 4 and 5. 30. Both, it will be argued in chapter 6, should be extended to other X' and XP categories. For further discussion of the configurational nature of dative and instrumental, see Fowler (1987a), Franks (1990a), Greenberg and Franks (1991), and Franks and Hornstein (1992). 31. One exception may be podrazat' 'to imitate'. Some speakers accept the dative argument of the active as the nominative of the passive, as follows:

58

PARAMETERS OF SLAVIC MORPHOSYNTAX

(i) peredovye rabotniki podrazajut inostrannym metodam forefront(nom) workers(nom) imitate foreign(dat) methods(dat) (ii)

inostrannye melody podrazajut-sja peredovymi rabotnikami foreign(nom) methods(nom) imitate-REFL forefront(inst) workers(inst)

Oddly enough, this verb is in fact cited in Freidin (1992, 206-207) in support of his erroneous claim that quirky case—assigning verbs in Russian never passivize. 32. Some of the discussion in this section may not be useful for readers unfamiliar with Jakobson's work on Russian case. The exposition and criticism of phonological and semantic aspects of Jakobson's model is not integral to the analyses in this book. There has been considerable recent interest in the morphosyntactic applicability of Jakobson's features, however, and much of my argumentation recapitulates issues discussed in, for example, Viel (1984), Franks (1985), Chvany (1986), Neidle (1988), and McCreight and Chvany (1991). 33. I have taken the liberty of replacing unmarked feature values with minuses; in Jakobson's theory this actually does not mean that a feature is absent, but rather that it is not explicitly present. 34. For detailed discussion of this issue, see Chvany (1986), who contrasts a variety of ways of representing the information in (63) and the various algorithms for determining the markedness value of the various cases. 35. See Franks (1990b) for discussion of the monosyllabicity requirement on Iocative2. See Fowler (1987a) for an exhaustive listing of those Russian nouns that exhibit special genitive2 and Iocative2 forms. 36. There are two principled classes of exceptions that allow Iocative2 on modified NPs: (i) set phrases like v rodnom kraju 'in (one's) homeland', and (ii) place names like v Glavnom botaniceskom sadu 'in the Main Botanical Garden' or v Leningradskom morskom torgovom portu 'in the Leningrad Sea Trade Port'. One additional exception is the word god 'year', which has the Iocative2 form godu in all temporal uses, regardless of modification: v etom godu 'in this year', v kakom godu 'in which year?', v 1944-om godu 'in 1944', and so on. 37. This is not to deny that there has been much interesting work by scholars who accept this proposition, arguing that a single general idea can capture all and only the forms of its expression. Certainly, the more general the idea put forward, the more persuasive this approach seems. In fact, what has inevitably happened is that, in trying to develop Jakobson's system, linguists (cf. especially the work of van Schooneveld and his students) have had to propose more and more abstract' 'meanings'' for the case features to unify more and more (in their view only superficially) disparate phenomena. This has the unfortunate result, however, that any conceivable procedure for getting from the general meaning to the particular ones, let alone going in the opposite direction, becomes progressively more obscure. 38. In other frameworks, such as Lexical-Functional Grammar, they might be functionally unified, perhaps in terms of the ADJUNCT function. 39. Forfurther discussion of the shortcomings of Jakobson's system, see Kilby (1977, 2-7) and Wierzbicka (1980). 40. One technical exception is that masculine singular nouns with a distinct gen2 have an identical dative form (snegu 'snow'), and the second locative differs only in place of stress (snegu). 41. Such a system might be appropriate for Serbo-Croatian, however, where these two cases generally fall together, although they are accentually differentiated in the literary language. 42. This is by no means a new idea. See, for example, "Le probleme du classement des cas," in Kuryfowicz (1973).

MATRICES, INDICES, AND MORPHOSYNTACTIC FEATURES

59

43. For example, similar results to (77) can be achieved by retaining Jakobson's [±oblique], [±indefinite], and [±marginal] and adding a fourth feature [±non-term], where the "term" cases are nominative, accusative, and dative, since they are canonically used to mark what is known in Relational Grammar as the "term" grammatical relations of Subject, Direct Object, and Indirect Object. In fact, [±oblique] could even be replaced by [ ± nuclear], since in RG ' 'nuclear'' includes Subject and Direct Object only. This alternative system is amenable to roughly the same markedness rankings and has the advantage that nominative is only marked as [ +indefinite], which may explain its use as the citation and default topic form. The sycretisms and the [ — oblique] genitive and dative of chapters 4 and 5 will also fit in, but it is unclear how gen2 and Ioc2 could be assimilated to such a system. Fortunately, little of what I will have to say hinges on this choice. 44. Although the nom=acc and ace = gen syncretisms are central to the Russian case system, it is not obvious that they should be handled in terms of feature neutralization. There is a sense in which paradigms that lack an independent accusative use the nominative or genitive form instead, the selection process depending on animacy (and on other features in the history of Russian and in other Slavic languages, particularly Polish). Replacement of one case by another could take the form of an accusative prediction rule that simply equates the accusative to nominative or genitive, or a readjustment rule in the theory of distributed morphology of Halle and Marantz (1992) might literally change the features from accusative to nominative or genitive. Assuming such rule-based approaches to be more appropriate than using a syncretic lexical entry, the only relevant syncretisms for feature determination become gen = loc, dat = gen = loc, dat = loc, and dat=gen = loc = inst; compare (71). However, such a move would not simplify the case system, since the same number of cases will still need to be differentiated. 45. This is reminiscent of the behavior of Russian tysjada 'thousand', which, like other numerals, has mixed adjectival and nominal properties, but appears to lie right on the dividing line in that it can either govern or agree with the following nominal material; compare Corbett (1978) for discussion. Interestingly, it has two instrumental forms, tysjaf'ju being used as a numeral (agreeing) or noun (governing) and tysjadej only as a noun, as observed in, for example, Borras and Christian (1971, 393). This bifurcation of forms does not exist for the other cases, although they participate in both usages. Lindseth (1993a) also points out that in Czech tisic 'thousand' can also function as a noun or an adjectival modifier: (i) a. o about b. o about

poslednim tisici lidi last(loc) thousand(loc) people(gen pi) poslednfch tisicfch lidech last(loc pl)thousand(loc) peopleQoc pi)

In instrumental contexts, however, the adjectival option is unavailable: (ii) a. s with b. *s with

poslednim last(inst) poslednimi last(inst pi)

tisicem thousand(inst) tisici thousand(inst)

lidi people(gen pi) lidmi people(inst pi)

46. Jakobson ignored the paucal in his case scheme. That the paucal is distinct from the regular genitive can be seen from the PuSkina paradigm, as well as in the fact that this form has a special stress for some words: dva Saga 'two steps', dva casd 'two hours' (versus the regular genitives saga, cdsa). 47. One exception is the preposition po, which generally takes the dative but in some uses

60

PARAMETERS OF SLAVIC MORPHOSYNTAX

also assigns locative to pronominal objects. See Franks (1994, n. 39) and chapter 5 for discussion. Note that these two cases differ in the current system only in terms of the feature [±phras]. 48. Although specification for [±nonascr] is vacuous for the locative, and generally assumed to be [ — nonascr], one might want to designate the case governed here as [ + nonascr], since it is precisely in these contexts that the second locative is assigned. 49. Another possibility for interpreting this feature might be to exploit the fact that the structural cases are consistently the least marked, with accusative and nominative being weighted 0 and 3 in the system suggested above. Even the [ — oblique] genitive and dative to be proposed in the analysis of Russian quantification will be less marked than the lowest [ +oblique] case (genitive, with a markedness value of 8), having values of 2 and 5, respectively. 50. It is not at all transparent what a Jakobsonian conception of syntax might be, other than something that provides "die reellen und formellen Bedeutungen der umgebenden Worte" (Jakobson, 1936, 51), in order that a particular meaning of a case may be determined.

3 Across-the-Board Dependencies

In this chapter I investigate the phenomenon of across-the-board (ATB) dependencies in Polish and Russian in order to illustrate how the theory of syncretism developed in the preceding chapter might be used to shed light on certain syntactic problems.1 ATB constructions are those involving syntactic processes that apply with equal force and in a parallel fashion across conjoined phrases. Here I will be concerned primarily with ATB dependencies involving the rule of w/z-movement, which takes a w/z-word and moves it to some clause-initial operator position, presumably COMP or, in the CP/IP system, the specifier of CP. From there it binds the trace that is left behind in its underlying argument position, as required by the Projection Principle, and that therefore functions as a variable. This process applies in both question and relative clause formation, as in the following Polish example:2 (1)

dziewczyna, ktora Janek lubi e girl who(acc) likes 'the girl who Janek likes'

In (1) the operator ktorq binds a single variable. It is also possible, however, for ktorq to have scope over parallel multiple variables in conjoined phrases, as in (2). (2)

dziewczyna, ktora Janek lubi e a Jerzy kocha e girl who(acc) likes and loves 'the girl who Janek likes and Jerzy loves'

In this ATB dependency the accusative w/i-word ktorq appears to be binding both gaps simultaneously. Both gaps are in accusative positions bearing the theme thetarole, and the construction is well-formed. Two related questions then arise: What happens when the multiple gaps are not parallel in morphological case? What happens when the gaps are not parallel in terms of thematic role? The first question is addressed in section 3.1, the second in section 3.2. Finally, on the basis of conclusions reached in the first two sections, I argue that 61

62

PARAMETERS OF SLAVIC MORPHOSYNTAX

ATB dependencies may involve a null operator in the second conjunct, thereby assimilating them to the well-studied parasitic gap construction. My point of departure in considering superficially non-parallel ATB dependencies is Dyta (1984). Dyf a argues that such dependencies can in fact be regarded as parallel if appropriate parallelism constraints are adopted. He formulates these in terms of overt morphological form and GB' s " abstract Case," so that neither requirement makes direct reference to morphological case per se. I will show how the identity in form requirement is really morphosyntactic in nature and thus directly dependent on case, and how the abstract Case requirement fails to account for the data, leading to an analysis in terms of relative thematic prominence. These facts support one of the central theses of this study: that there is no need for any concept of abstract Case distinct from the traditional understanding of morphological case (beyond the minimal assumption that NPs have case even in languages where this is not paradigmatically realized).

3.1. Case-Matching Dyfa, following Borsley (1983), observes that ATB dependencies involving conjuncts demanding different cases generally result in ungrammaticality, as in the Polish example (3).3 (3)

*dziewczyna, ktora/ktorej Janek lubi e a Jerzy nienawidze e girl who(acc)/(gen) likes and hates ' the girl who Janek likes and Jerzy hates'

Neither the accusative form ktorq nor the genitive form ktorej can on its own simultaneously satisfy the subcategorization of both lubi 'likes', which governs the accusative, and nienawidzi 'hates', which governs the genitive. However, if the relative or interrogative pronoun happens to be morphologically compatible with both required cases, as in (4), the ATB dependency becomes grammatical: (4)

chlopiec,ktorego Maria lubi e a Ewa nienawidzi e boy who(acc-gen) likes and hates 'the boy who Maria likes and Ewa hates'

The syncretic form ktorego in (4) is able to satisfy both conjuncts at the same time, serving as accusative with respect to lubi and genitive with respect to nienawidzi. Similarly, (5) is acceptable because the feminine singular relative pronoun ktorej is ambiguously genitive or dative. (5)

dziewczyna, ktorej Janek nigdy przedtem nie girl who(gen-dat) never before not widzial ea dzisiaj pozyczytpieni^dzy e saw and today lent money 'the girl who Janek never saw before and lent money to today'

ACROSS-THE-BOARD DEPENDENCIES

63

The negated verb nie widzial 'not saw' requires its object to be in the genitive, whereas the indirect object of pozyczyl 'lent' must be in the dative. From these facts I conclude, along with DyJa, that it is its morphological realization, rather than the case per se, that is relevant in licensing ATB dependencies.4 It is thus necessary somehow to distinguish an item's "case form" from the particular morphological case it is assigned. The case form of a word is determined on the basis of its lexical entry, whereas its morphological case depends on licensing factors in its syntactic environment. This suggests that the morphological identity requirement can be formalized by exploiting the feature neutralization approach to case syncretism. The case that the relative pronoun bears is not accusative, genitive, dative, or instrumental, but rather some feature complex that is non-distinct from the feature complexes expressing any of these specific cases. Thus, Polish ktorego in (4) and ktorej in (5) are [ —marg, — nonascr] and [ + obl], respectively.5 They can for this reason simultaneously satisfy the more specific case requirements imposed on each relevant gap. That is, in (5) for example, ktorej literally functions as [ + obl, +marg, -nonascr, +phras] dative with respect to pozyczyl pieniqdzy 'lent money' and [ + obl, -marg, -nonascr, -phras] genitive with respect to nie widzial 'did not see'. In this way both subcategorization frames are satisfied without taking recourse to the PF notion of "actual morphological form," as Dyla is forced to do. Thus, assuming that the relevant criterion for lexical insertion is non-distinctness, a syncretic form can be inserted into any position that subsumes its set of feature values. Whatever mechanism is then adopted for deriving ATB dependencies will simply have to check that the form of the relative or interrogative pronoun is non-distinct in case features from the case assigned to each of the gaps. The effects of morphological syncretism are similarly displayed by other languages. Consider the following Russian examples:6 (6)

mal'cik, kotorogo MaSa ljubit e i Vera boitsja e boy who(acc-gen) (nom)loves and (nom) fears 'the boy who Masha loves and Vera fears'

(7)

devuSka, kotoroj Vanja daval den'gi e no Borja izbegal e girl who(dat-gen) (nom) gave money but (nom) avoided 'the girl who Vanja gave money to but Borja avoided'

(8)

devuSka, kotoroj ja byl uvleden e girl, who(inst-dat) I was carried-away-with i daval den'gi e and gave money 'the girl who I was carried away with and gave money to'

These dependencies are acceptable even though the verbs in the two conjuncts call for different cases, as expressed in the case annotations following who in each gloss. The reason is, as before, that morphological syncretism enables the relative pronoun to serve a dual case function.

64

PARAMETERS OF SLAVIC MORPHOSYNTAX

3.2. Parallelism in Prominence Treating syncretism as the neutralization of morphosyntactic case features does not, however, solve all the problems associated with ostensibly non-parallel ATB dependencies. As DyJa points out, mere morphological case identity is not sufficient to license all ATB dependencies in Polish. Instead, a further parallelism requirement is called for. In this section I discuss the facts adduced by DyJa and develop an alternative analysis in terms of thematic prominence. 3.2.1. The Nature of the Parallelism Dyta demonstrates the inadequacy of morphological case identity by citing the following unacceptable examples: (9) *dziewczyna, ktorej Janek chcial ztozycwizyte. e girl who(dat-gen) wanted to-pay visit a e nie by to w domu and not was at home (10)

*dziewczyna, ktorej Janek daJ swoj^ marynarke/ e girl who(dat-dat) gave self's jacket a mimo tego e byto zimno and moreover was cold

In (9), ktorej must be simultaneously interpreted as a dative indirect object and a genitive under existential negation, while in (10) it is a dative indirect object and a dative experiencer. The fact that both conjuncts in example (10) involve the dative clearly shows that the traditional notion of morphological case has no bearing on ATB dependencies. Rather, Dyta argues, it is identity of abstract Case role, as well as actual case form, that must be taken into consideration in deciding whether a particular ATB dependency is licit. The problem remains, however, of determining what abstract Case any given occurrence of a morphological case instantiates. Using the ATB construction as a diagnostic, Dyta suggests that (5) is well-formed because both genitive and dative are surface realizations of abstract Objective in these clauses. In (9) and (10), on the other hand, he claims that this is not so, although offering no specific proposal as to the nature of the discrepancy. He simply states that the genitive in (9) "cannot instantiate Objective."7 One might speculate that the genitive and dative on the second conjuncts of (9) and (10), respectively, are realizations of an abstract Nominative, in some sense reflecting logical "subjecthood." Note that this assumption leads one to expect that the two might be conjoinable. Interestingly, this prediction is borne out, as shown by the grammaticality of (11). (11) dziewczyna, ktorej e by to zimno girl who(dat-gen) was cold i z powodu tego e nie by Jo na zaj^ciach and therefore not was at class

ACROSS-THE-BOARD DEPENDENCIES

65

These facts, however, have serious implications for the relationship between abstract Case and morphological case. They essentially lead to the awkward conclusion that there is no consistent relationship at all. Abstract Nominative can be realized by at least nominative, genitive, or dative, and abstract Objective by accusative, genitive, or dative. Instrumental also would have to be able to instantiate both abstract Nominative and Objective, in order to account for the ungrammaticality of Polish (12). (12)

*cztowiek, ktorym zatrzeslo e man who(inst-inst) shook i z tego powodu (my) zainteresowalismy sie e and therefore we were-interested-in REFL

The relative pronoun ktorym in (12) cannot simultaneously serve as the object of the impersonal verb zatrzqsto 'shook', which has a null expletive subject,8 and the object of the transitive verb zainteresowalismy si$ 'were-interested-in', even though both govern the instrumental. The abstract Cases one would have to assign various arguments to make Dyta's system viable thus correspond neither to morphological case nor to semantic role. This suggests that abstract Case is not the appropriate mechanism in which to couch the parallelism restriction on ATB dependencies. As with the syncretism effect, similar facts can be adduced from other languages. Contrast the following ungrammatical Russian examples to the acceptable ones in (6)-(8): (13)

*mal'cik, kotorogo e ne bylo na uroke i my izbegali e na ulice boy who(gen-gen) not was in class and we avoided on street

(14)

*cto vy skazete tem, kogo vse what you say to-those who(acc/gen-gen) everyone zdali e i e ne pojavilos' waited-for and not appeared

(15)

*mal'cik, kotoromu e bylo xolodno i ja dal svoj pidzak e boy who(dat-dat) was cold and I gave my jacket

Recall that the dependency in as unusual an example as (8) was acceptable, even though the first conjunct contained an instrumental gap and the second a dative one. Thus, according to Dyta's hypothesis, both would have to be Objective.9 This would be a patently circular conclusion, however, motivated solely by the need to explain the ATB facts. Russian (13) and (14) are parallel to Polish (9), and (15) is parallel to (10), again suggesting that "subjective" genitives and datives should be regarded as abstract Nominative. There is little to support this conjecture, however, nor is there any sufficiently pervasive definition of abstract Nominative that would accommodate all relevant constructions. Moreover, I shall soon present further examples of bizarre yet viable ATB dependencies that make Dyla's abstract Case analysis seem even more untenable. There is considerable evidence, for example, that "subjective" genitives under existential negation in Russian are really underlyingly objects of "unaccusa-

66

PARAMETERS OF SLAVIC MORPHOSYNTAX

live" verbs—that is, verbs that assign no case to their objects and no theta-role to their subjects (cf. e.g. Pesetsky, 1982; Neidle, 1988). Chvany (1975, 134), for example, cites the following examples: (16)

a. Ivana ne bylo v ego/*svoej komnate Ivan(gen) NEG was in his/self's room b. Ivana ne bylo u nego/*sebja v komnate Ivan(gen) NEG was at him/self in room

Since she argues that the genitive Ivana is a surface subject, although originating as an object, Chvany is forced to maintain that some unknown aspect of the construction must be blocking reflexivization in (16). If we assume that the word order properties of negated existential constructions are determined along the lines of Babby (1980b, 1980c), however, Ivana is free to be analyzed as an S-structure object as well. As such, it cannot bind the reflexive pronouns in these examples. Arguments that negated "subjects" of intransitive verbs are in fact VP-internal objects are even more telling in Polish, where the only intransitive verb to assign genitive under negation is the copula bye.10 Standard arguments against the subject status of such genitives in Russian are that they fail to display expected subject properties, such as anteceding reflexives and controlling gerunds (adverbial participial adjuncts). Working within a Relational Grammar framework, Dziwirek (1991) summarizes these for Russian and offers a variety of further tests demonstrating that the genitive with negated bye in Polish is not a subject. (17)

Janka nie byJo w jego/*swoim pokoju (gen) not was in his/self's office

(18)

*Skonczywszy prace. Janka nie byto w biurze having-finished work (gen) not was in office

(19)

*Mimo powaznej choroby Janka nie bylo w domu despite serious illness (gen) not was at home

In (17), just as the Russian example (16), the reflexive anaphor swolm, which must be bound by some subject within its binding domain (in this instance the entire clause), cannot be used to refer to Janka. Instead, the only admissible form is the pronoun jego, which must be free from all subjects within its binding domain. The fact that jego can be coreferential with Janka reveals that Janka is not a syntactic subject, hence is assigned genitive under negation just like the object of any other negated verb. In (18) the non-subject status of Janka renders it unavailable to control the gerund, and similarly for adverbial phrases such as in (19). Since such genitives under existential negation are neither syntactic subjects nor morphological nominatives, any sense in which they might be abstract Nominative remains obscure. For reasons such as these, I now consider an alternative way of stating the operative restriction on ATB dependencies. The data presented so far indicate that the factors affecting the acceptability of ATB extractions are complex. Beyond the obvious need for morphological compatibility between the relative/interrogative pronoun and the multiple gaps, the nature of any further relevant licensing condi-

ACROSS-THE-BOARD DEPENDENCIES

67

tion(s) is far from clear. I have so far argued that a solution formulated in terms of abstract Case is not tenable. In this section I propose instead that the appropriate generalization involves relative argument prominence. The superiority of the prominence account over the abstract Case one will be further demonstrated by its predictions about the acceptability of a variety of unusual ATB dependencies. There are essentially two kinds of approaches to the problem of what rules out certain morphologically well-formed but nonetheless unacceptable ATB dependencies. The first is to invoke some independently motivated structural licensing condition on gaps, and then to argue that one of the gaps in illicit ATB constructions violates that condition.11 Analyses involving structural constraints include those of Gazdar (1981) in terms of the Generalized Left Branch Condition, Pesetsky (1982) in terms of the Path Containment Condition, Goodall (1983) and Woolford (1987) in terms of the Empty Category Principle, and Goodall (1987) in terms of Condition C of the Binding Theory. The second kind of aipproach is to impose some kind of parallelism constraint on ATB dependencies in general, and then to argue that the unacceptable constructions are asymmetric with respect to that constraint. DyJa's account belongs to this latter class of strategies for ruling out ATB asymmetries, of course, as does Williams's (1978) seminal work, which argues that the "simultaneous factors" in English ATB dependencies must be either both leftmost or not leftmost. While it may be that both types of constraint will ultimately prove necessary, here I focus on the correct formulation of the parallelism condition. A survey of the Slavic examples considered so far reveals the following simple descriptive generalization: (20) In any ATB construction, the gaps must pertain either to most prominent or to not most prominent arguments, consistently across all the conjuncts. The essence of this proposal is that, within each conjunct that enters into an ATB dependency, each of the verb's arguments can be characterized in terms of some sort of relative prominence scale with respect to all of the verb's other arguments. It is thus always possible to designate a unique argument as "most prominent" in any given clause. The parallelism requirement on ATB dependencies can then be stated in terms of whether the gaps in all of the conjoined clauses correspond to arguments that are either (1) most prominent or (2) not most prominent, exclusively. If the gaps are not consistently of one type or the other, the ATB dependency will be infelicitous. It should be clear that this condition will hold regardless of the abstract Case associated with the various gaps, however "prominence" is ultimately defined. The reader is invited to apply the criterion in (20) to the examples thus far considered: in (4)-(8) neither gap is most prominent and in (11) both are; while in (9), (10), and (12)-(15) a conflict arises because one gap is most prominent and the other is not. There are at least two distinct ways in which one might conceive of this notion of relative prominence. Prominence could be instantiated as a thematic relation among arguments or as a structural one. Since discriminating between these two possibilities is no mean task, and my basic claims hold regardless of one's conception of the true nature of prominence, at this point I simply present both possibilities. Either way, it should be clear that some kind of hierarchical notion of prominence rather

68

PARAMETERS OF SLAVIC MORPHOSYNTAX

than a linear one of precedence is called for. Although the parallelism requirement in (20) is essentially Williams's (1978) proposal with "most prominent" substituted for "leftmost," a word order approach is obviously inappropriate for Slavic. At D-structure, some (leftmost) subjects and (not leftmost) objects can enter into ATB dependencies, whereas some object-object pairs cannot, depending on whether or not a theta-marked subject is present. By S-structure, scrambling—which presumably may apply at this level since it enters into LF processes such as scope determination and cancellation of weak crossover effects—renders word order considerations irrelevant for the ATB parallelism requirement. Note, in particular, that for none of the examples discussed does ATB grammaticality directly depend on surface word order. The thematic prominence account relies on the kind of thematic prominence hierarchy stated in (21): (21)

Agent > Experiencer > Theme > Goal/Source/Location > Manner/Time

This is the hierarchy assumed in Speas (1990), based on proposals due to Belletti and Rizzi (1988) and Carrier-Duncan (1985). It is also very close in all relevant respects to the hierarchy argued for in the theory of lexical-conceptual structure put forward in Jackendoff (1990a).12 Although many other thematic hierarchies exist in the literature, of course, I tentatively adopt (21) since it seems to work best for the ATB data to be presented in this chapter, which for the most part show that agents and experiencers outrank themes and that themes, in turn, outrank all other arguments.13 In any event, for most of the argumentation herein, specifics of the hierarchy are irrelevant, since in constructing examples I generally employ the technique of contrasting verbs that take a single argument (which I take to be de facto most prominent) to those that take multiple arguments. This highlights the fact that the only opposition actually needed is [ ± maximally prominent] and suggests that an alternative characterization of the ATB restriction might be couched in terms of some such feature. This variant of the thematic prominence account would capitalize on the fact that all Condition (20) has to do is single out the highest argument on a hierarchy of prominence; arguments lower than this highest one never need to be differentiated.14 A structural prominence account might also be constructed along similar lines. Under such an approach, "most prominent" in (20) would be defined as the argument in the highest A-position in the phrase structure tree. All gaps in an ATB dependency would then need to be in the highest A-position associated with a thetarole (to render expletives transparent) or not in the highest one, consistently across the conjuncts. Note that this procedure does not simply pick out subjects, since NPs that by the ATB criterion are most prominent may still fail to satisfy standard diagnostics of subjecthood. For example, although it was demonstrated that Janka in (17)-(19) clearly does not display subject properties, ATB extraction shows that it is still the most prominent argument. The viability of the structural prominence proposal depends on whether or not the necessary structures are available. It succeeds to the extent that it can be shown that any argument that the ATB test indicates to be most prominent c-commands (taking c-command to be the operative definition of "higher") all the verb's other arguments and any argument that the ATB test

ACROSS-THE-BOARD DEPENDENCIES

69

indicates not to be most prominent is c-commanded by some other argument of the verb. Of course, once c-command is demonstrated, appropriate structures must be posited. For example, Barss and Lasnik's (1986) tests that in the English double object construction the indirect object is more prominent than the direct object led Larson (1988) to develop a configurational analysis in which Mary at S-structure c-commands a letter in John sent Mary a letter, although whether Larson's abstract structure is motivated is questioned by Jackendoff (1990b).15 I do not pursue the issue here, however, since my overall claims hold regardless of which of these approaches to prominence ultimately turns out to be correct. Moreover, distinguishing among them is no simple matter for several reasons. For one thing, scrambling can subsequently obscure structural relations. More importantly, if Larson's claim is correct that the theta-roles in a verb's theta-grid are mapped onto syntactic argument positions in accordance with a thematic hierarchy such as (21), then any potentially discriminating evidence will be very obscure. This uniform mapping of thematic prominence into structural prominence is argued for at length by Speas (1990), who maintains that the Uniformity of Theta Assignment Hypothesis of Baker (1988) should be construed to mean that there is a necessarily direct correspondence between prominence on the thematic hierarchy and syntactic prominence. Interestingly, she reaches conclusions that are completely compatible with my analysis of ATB dependencies in Slavic, contending that "the principles of grammar must be able to pick out the most prominent argument in the syntactic structure, but that they do not show that the most prominent argument is necessarily external to the maximal projection of the predicator" (Speas, 1990,101).16 There is thus a significant convergence among these various conceptions of prominence. 3.2.2. The Scope of Prominence Effects I turn now to further examples that offer support for some version of Condition (20) and the particular thematic hierarchy in (21). Consider the following impersonal constructions, (22)-(24) from Polish and (25)-(26) from Russian: (22) a.

czlowiek, ktorego swedzi e man who(acc-gen) itch i ztegopowodu e nie bytona zajeciach and therefore not was in class

b. *czlowiek, ktorego swe.dzi e i z tego powodu man who(acc-gen) itch and therefore (my) nie zapraszatysmy e na przyj^cie we not ask to reception (23) a.

portfel, ktory ukradziono e i e jest drogi briefcase which(acc-nom) was-stolen and is valuable

b. *portfel, ktory ukradziono e i (ja) lubiteme briefcase which(acc-acc) was-stolen and I liked

70

(24)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a.

droga, ktora w zesztymroku zbudowano e road which(acc-acc) in last year was-built i zasypalo e sniegiem and was-covered by-snow

b. *droga, ktor^ zasypalo e sniegiem road which(acc-acc) was-covered by-snow i (my) sprzatnelismy e and we cleared (25)

a.

celovek, kotorogo vsegda tosnilo e man who(acc-gen) always was-nauseous i poetomu e nikogda ne bylo na zanjatijax and therefore never not was in class

b. *celovek, kotorogo vsegda tosnilo e man who(acc-gen) always was-nauseous i poetomu vse izbegali e and therefore everyone avoided (26)

a.

okno, kotoroe razbilo e vetrom i e lezalo na zemle window, which(acc-nom) broke wind(inst) and lay on ground ' the window which was broken by the wind and lay on the ground'

b. *okno, kotoroe razbilo e vetrom window, which(acc-acc) broke wind(inst) i my uvideli e na zemle and we saw on ground 'the window which was broken by the wind and we saw on the ground' The first verb in each conjunct is impersonal in the traditional sense that it occurs with an accusative VP-internal argument and an expletive subject. Thus, even though it is the theme of the first verb, the relativized constituent is most prominent, being either the sole argument or higher than other obliques such as the time phrase in (24) or the "natural force" instrumental in (26). The themes are clearly most prominent despite being neither the syntactic subject nor an abstract Nominative. Such arguments may only enter into ATB dependencies with other most prominent arguments, as in (22a)-(26a). In Polish (22a) and Russian (25a) this is a "subjective" genitive, in Polish (19a) and Russian (26a) a nominative intransitive subject, and in Polish (24a) an accusative object of an impersonal verb. However, as soon as a more prominent argument is added to the second conjunct, as in the transitive (22b)-(26b), the result becomes ungrammatical. Notice that the participle ukradziono 'stolen' in Polish (23) takes an accusative object (and genitive under negation); see chapter 8 for discussion of this construction. The acceptability of (23a), where the gaps do not match in terms of being both leftmost or not leftmost, coupled with the unacceptability of (23b), where they do, demonstrates the inadequacy of Williams's (1978) account for Slavic ATB asymmetries. Consider next what happens with a predicate such as Polish zal 'sorry', which takes accusative and dative arguments. The judgments are given in (27).

ACROSS-THE-BOARD DEPENDENCIES

71

(27) a. cztowiek, ktoremu e by to zal Jana i ebytozimno man who(dat-dat) was sorry (ace) and was cold 'the man who was sorry for Jan and was cold' b. cztowiek, ktorego Maria lubitae a Janowi by to zal e man who(acc-acc) (nom) liked and (dat) was sorry 'the man who Maria liked and Jan was sorry for' c. "cztowiek, ktorego e nie bytona zajeciach man who(gen-acc) not was at class a Janowi jest zal e and (dat) is sorry 'the man who was not in class and Jan is sorry for' These data indicate that the dative experiencer is more prominent than the accusative theme, in keeping with the hierarchy in (21). Thus, the dative argument can enter into an ATB dependency with another most prominent dative, as in (27a), and the accusative argument can with another not most prominent accusative, as in (27b). Example (27c) violates Condition (20), however, since the accusative argument of zal 'sorry', unlike the dative one in (27a), is incompatible with the most prominent genitive argument of nie bylo na zajeciach 'was not in class'.17 The grammaticality of (27b) is of particular interest, as a personal construction is being conjoined with an impersonal one, showing that the traditional classification of Slavic sentence types into "personal" and "impersonal" is not relevant to ATB acceptability. It is worth observing that reflexive binding displays the same asymmetry between the accusative and dative arguments of ml 'sorry', as shown by a comparison of (28a) and (28b). (28) a. Janowi by to zal (samego) siebie (dat) was sorry own self(acc) ' Jan was sorry for himself b. *Jana by to zal (samemu) sobie (ace) was sorry own self(dat) These data follow straightforwardly from Jackendoff's (1972) claim that antecedent NPs may not be less thematically prominent than the anaphors that they bind. The fact that the dative experiencer binds the accusative theme, but not vice versa, corroborates my assumption that it is higher (whether structurally, thematically, or both). Dziwirek (1991,121) cites other relevant examples of datives binding accusatives in Polish, such as the following "inversion" structures: (29) a. Jankowi najlepiej czyta si$ swoje ksiazki (dat) best reads REEL self's books(acc) 'Janek enjoys reading his own books the most' b. Ewie mitooglada sie. swoje zdjecia (dat) nice looks-at REFL self's pictures(acc) c. swoj pokoj sprzatato sie. Ewie szybko self's room(acc) cleaned REFL (dat) quickly 'Ewa quickly cleaned her own room'

72

PARAMETERS OF SLAVIC MORPHOSYNTAX

In (29), the dative NPs Jankowi and Ewie are experiences, hence more prominent than the accusative themes. Crucially, as will be demonstrated shortly, dative goals do not pattern in the same manner. A consideration of weak crossover in Polish reveals similar judgments. For example, in (30a) the pronoun jego 'his' can be interpreted as a variable bound by the dative kazdemu 'each (person)', but in (3Qb)jego cannot be bound by the accusative kazdego.ls (30) a. kazdemu by to zal jego sasiada each(dat) was sorry his neighbor(acc) ' each person was sorry for his neighbor' b. *kazdego by Jo zal jego sasiadowi each(acc) was sorry his neighbor(dat) The only way (30b) can possibly be understood is with jego 'his' referring to some specific individual. The ATB test of prominence thus seems consistent with other more familar metrics. Given such tests, as well as the broad range of constructions examined, let us turn to the issue of the relative prominence of dative indirect object goals and accusative direct object themes. It is possible to devise a Polish clause containing just these two arguments if the participial construction in (23) and (24) is employed. Subjecting this construction to the ATB test reveals that the theme is consistently higher than the goal, as shown by the following contrasts using dac 'to give' and przedstawic 'to introduce':19 (31) a. *ksiazka, ktora dano e Jankowi a ja przeczytatem e book which(acc) is-given (dat) but I read ' the book which was given to Janek but I read' b. ksiazka, ktora dano e Jankowi a ukradziono e book which(acc) is-given (dat) but is-stolen ' the book which was given to Janek but was stolen' (32) a. ?czfowiek, ktoremu dano marynarke e i japomogteme man who(dat) is-given jacket(acc) and I helped ' the man who was given a jacket and I helped' b. "tziowiek, ktoremu dano marynarke e i bytozimno e man who(dat) is-given jacket(acc) and was cold ' the man who was given a jacket and was cold' (33) a. *czfowiek, ktorego przedstawiono Jankowi e i ja zapomniatem e man who(acc) is-introduced (dat) and I forgot ' the man who was introduced to Janek and I forgot' b. czlowiek, ktorego przedstawiono Jankowi e i zapomnieno e man who(acc) is-introduced (dat) and is-forgotten ' the man who was introduced to Janek and was forgotten'

ACROSS-THE-BOARD DEPENDENCIES

73

(34) a. ?cztowiek, ktoremu przedstawiono Janka e i ja dalempapierosa e man who(dat)is-introduced (ace) and I gave cigarette(acc/gen) ' the man who Janek was introduced to and I gave a cigarette to' b. *cztowiek, ktoremu przedstawiono Janka e i bytonudnoe man who(dat)is-introduced (ace) and was bored ' the man who Janek was introduced to and was bored' The assumption that the dative goal is less prominent than the accusative theme in the first conjunct in the above examples explains why it can enter into an ATB dependency with another less prominent dative, as in (32a) and (34a), but not with a most prominent dative, as in (32b) and (34b). Similarly, a most prominent theme cannot enter into an ATB dependency with another theme that is not most prominent, as in (31a) and (33a), but can with another most prominent theme, as in (31b) and (33b). The examples in (31)-(34) thus provide compelling evidence that dative experiencers and dative goals should be ranked separately on the thematic hierarchy, with themes necessarily lower than experiencers but higher than goals, as in the hierarchy in (21). The binding test provides the same results, although this conclusion is somewhat obscured by the fact that reflexives are subject-oriented in Polish. Instead, as Reinders-Machowska (1991) points out, when bound by a non-subject, anaphors that may otherwise be reflexive admit only a reciprocal interpretation, reciprocity (apparently universally) lacking any kind of subject-orientation restriction. Thus, (35a) is well-formed, whereas (35b) is unacceptable.20 (35) a.

(ja) przedstawitemmoich przyjaciof sobie I introduced my friends(acc) self(dat) ' I introduced my friends to each other'

b. *(ja) przedstawiJemmoim przyjaciotom siebie I introduced my friends(dat) self(acc) Only the more prominent theme moichprzyjaciol 'my friends(acc)' can bind the less prominent goal sobie 'self(dat)'. Since there is no reason to think that binding in (35b) is blocked for structural reasons, these examples also demonstrate that local scrambling does not affect binding possibilities. Interestingly, the crossover test shows no comparable contrast. Instead, a possessive pronoun in either argument can be translated into a variable, so long as it follows the quantified expression, as shown in (36). (36) a. Ewa przedstawiJakazdemu gosciowi jego sasiada (nom) introduced each guest(dat) his neighbor(acc) 'Ewa introduced to each guest his neighbor' b. Ewa przedstawitakazdego goscia jego sasiadowi (nom) introduced each guest(acc) his neighbor(dat) 'Ewa introduced each guest to his neighbor' In both examples, the bound variable interpretation is unavailable if the order of the

74

PARAMETERS OF SLAVIC MORPHOSYNTAX

two NPs is reversed, but is fine so long as the quantified one precedes the possessive pronoun. Similarly, (37a) and (37b) have equally acceptable status with jego 'his' interpreted as a bound variable. (37)

a. kogo przedstawiles jego sasiadowi who(acc) you-introduced his neighbor(dat) 'who did you introduce to his neighbor?' b. komu przedstawiJes jego sasiada who(dat) you-introduced his neighbor(acc) 'to whom did you introduce his neighbor?'

The same is true of comparable passive constructions, with the verb agreeing with the nominative subject and the dative indirect object scrambled to initial position in (38b) and (39b) in order to precede the pronoun. (38)

a. kazdy gosc byt przedstawiony jego sasiadowi each guest(nom) was introduced his neighbor(dat) 'each guest was introduced to his neighbor' b. kazdemu gosciowi byJ przedstawiony jego sasiad each guest(dat) was introduced his neighbor(nom) 'to each guest was introduced his neighbor'

(39) a. kto byl przedstawiony jego sasiadowi who(nom) was introduced his neighbor(dat) 'who was introduced to his neighbor?' b. komu byl przedstawiony jego sasiad who(dat) was introduced his neighbor(nom) 'to whom was introduced his neighbor?' As in the active, the bound variable interpretation is only possible if the quantified NP precedes the possessive pronoun. Russian speakers give similar judgments for all these constructions. These results suggest to me that the ATB and binding tests are better indicators of thematic prominence than weak crossover is, although how this discrepancy should be understood is unclear. One possibility is that scrambling may also be implicated in (36)-(39), and it is the adjustment of c-command relationships by virtue of scrambling that allows these various possibilities in Polish and Russian. Such an account implies, following Speas (1990), that Baker's Uniformity of Theta Assignment Hypothesis determines D-structures but does not constrain post-scrambling structures. As Speas (1990, 146) observes, "scrambled NPs behave as though they are structurally superior to unscrambled NPs" regardless of what grammatical functions these NPs bear. If so, any principle of grammar that is concerned with relative thematic prominence can in general be stated either thematically or structurally, but once scrambling has occurred, the thematic hierarchy needs to be consulted. Under this kind of account, ATB dependencies do indeed require parallelism in thematic prominence, and binding requires that the antecedent not be thematically lower than the anaphor, whereas other standard c-command tests, such as those reflected in (36)-(39), are purely structural. I thus conclude that all arguments that are most prominent or not most prominent

ACROSS-THE-BOARD DEPENDENCIES

75

within their respective predicate-argument structures behave in a parallel fashion with respect to ATB extraction, even if they bear markedly different roles, and that even arguments bearing the same morphological case and similar semantic roles are not compatible if they are not parallel in thematic prominence. Such examples defy alternative treatments, such as the abstract Case analysis of Dyla (1984), but are consistent with the prominence analysis. The results of sections 3.1 and 3.2 can be summarized in light of four competing notions of "case." From least to most abstract, this term has been applied to refer to (1) the morphological'' wordform'' of a particular nominal, (2) the traditional morphosyntactic categories called for by a particular syntactic position or semantic role, (3) the abstract syntactic licensing property of NPs employed by GB principles, and (4) a semantic role label for grammatical relations, as in the Case Grammar of Fillmore (1968) and known in GB as theta-roles. In this chapter I have argued that the only senses in which "case" is relevant to ATB dependencies are the morphological notion, which relies on the atomization of cases into the features required for lexical entries, and the semantic notion, which relies on predicate-argument structure and thematic prominence. 3.2.3. Some Possible Extensions I now briefly explore some of the ramifications of the proposal to handle asymmetric ATB dependencies in terms of thematic prominence, raising some potential problems and discussing some possible solutions. There are, for example, numerous instances cited in the literature of more or less acceptable ATB extractions involving arguments that prima facie do not seem to be parallel in terms of prominence (cf. e.g. Anderson, 1983). A typical example is given in (40). (40) ?the dress which Mary bought e and e cost $500 I should note, as does Goodall (1987) on comparing his analysis with others, that all structural accounts fail completely to accommodate such examples. However, because I am using a semantic criterion of parallelism, it may be possible to understand the relative acceptability of (40) in terms of the amount argument $500 being more prominent than the theme argument the dress of the predicate cost. This conjecture is supported by the observation that acceptability degrades strikingly if an adverb, arguably not an argument, is used for the amount, as in (41). (41) *the dress which Mary bought e and e cost a lot The behavior of verbs with multiple predicate-argument structures also accords with the prominence analysis. Consider, for example, the ambiguity of the verb frighten, which may be volitional or not (cf. Grimshaw, 1990, for detailed discussion of its properties). When the subject of frighten is an agent, it is more prominent than the object experiencer, and when the subject is non-volitional, it is a source and less prominent than the object. With this in mind, observe that the ambiguity inherent in (42) disappears under conjunction, as in (43). (42)

the boy who e frightened Mary

(43) a. the boy who e frightened Mary and e hit her

76

PARAMETERS OF SLAVIC MORPHOSYNTAX

b. ?the boy who e frightened Mary and she hit e The ATB construction in (43a) only admits the agentive reading of frighten, and (43b) only the non-agentive one, exactly as predicted.21 Most studies of ATB extraction asymmetries have taken the following paradigm as a central puzzle: (44)

a. *the man who John saw e and e kissed Mary b.

the man who John saw e and Sue thinks e kissed Mary

The problem is to explain why whatever is wrong with (44a) disappears when the subject gap is contained in a complement clause, as in (44b). Under my account, (44a) is infelicitous because the gaps are not parallel in thematic prominence. Other analyses, such as Woolford's (1987) explanation in terms of the ECP, focus on structural issues relating to the gap in the second conjunct. For Woolford, for example, the gap is not antecedent-governed in (44a), but can be in (44b) assuming a trace in COMP. The prominence account also automatically accommodates (44), since the introduction of the more prominent argument Sue in the second conjunct of (44b) prevents the subject gap of kissed Mary from being most prominent, under the assumption, following Speas (1990, 88), that "any embedded theta roles are lower on the thematic hierarchy than nonembedded ones." On the other hand, this analysis raises interesting questions about the felicity of variations on English (44b). One would expect, for example, that (45) should be worse than it actually is. (45)

the man who e saw John and Sue thinks e kissed Mary

I do not find any significant difference between (45) and (44b), however; both are perfectly acceptable. This suggests that whether or not the superimposed matrix argument counts in diminishing the prominence of the embedded subject is optional. Thus, both (44b) and (46) have acceptable alternatives: when the higher argument Sue counts in (44b) the structure is saved, and when it does not count in (45) the structure is saved. Note also that (46) is considerably worse than (44b), presumably because it is an expletive, hence it was thought does not introduce a more prominent argument. (46)

*the man who John saw e and it was thought e kissed Mary

Similarly, (47) is far better than (46), as expected. (47)

the man who it was thought e saw John and it was said e kissed Mary

Since neither embedding even has the option of lowering the prominence of the gap, e remains most prominent in both conjuncts. Purely structural accounts, which we have seen also cannot handle problems such as that posed by (40), do not predict any of these differences. While, as Neijt (1980) and Goodall (1987) note, this curious state of affairs is not universal,22 it does seem to carry over to Polish, as a comparison of (48a) with (48b) shows. (48)

a. *dziecko, ktore lubisz e i e bylodobre boy who(acc-nom) you-like and was good

ACROSS-THE-BOARD DEPENDENCIES

77

b. dziecko, ktore lubisz e i chciesz, zeby e bylodobre boy who(acc-nom) you-like and you-want that was good Crucially, there is no way that Dyla's approach can accommodate these facts, since here we find an internally contradictory paradigm—it is possible to create an ATB dependency in (48b), but not (48a), although the abstract Cases are held constant. The problem with (48a) can have nothing to do with any mismatch in abstract Case, since the discrepancy between the abstract Objective and Nominative gaps is irrelevant in (48b). Dyla's characterization of the source of the parallelism in ATB constructions is thus both too strong and too weak. In conclusion, the evidence reveals that the felicity of ATB dependencies involves not only structural and morphological factors, but semantic ones as well. The Slavic data show that if the sole NP argument of an impersonal verb is questioned or relativized, then it can enter into an ATB dependency with (1) the sole argument of another impersonal verb, or (2) the subject argument of a personal verb, regardless of the abstract Case or thematic role of any of these NPs. I have argued that what is crucial is the relative thematic prominence of each argument within its own conjunct, proposing that the operative constraint is that all gaps in any ATB construction must pertain to either most prominent or not most prominent arguments, exclusively. This condition on coordinate structures appropriately handles the facts originally observed by Dyfa for Polish and extends to others in both Slavic and English for which his system is unable to account.

3.3. The Parasitic Gap Analysis In this section I discuss the possibility that ATB constructions may involve movement of a null operator in the second (or dependent) conjunct and should therefore be assimilated to parasitic gap constructions. Crucially, both the morphological compatibility and thematic prominence constraints also apply in parasitic gap constructions. This striking similarity suggests that the relevant constraints should be formulated in terms of the properties of null operators. 3.3.1. Similarities Parasitic gap constructions are those involving an overt relative or interrogative w/i-word binding its trace and, in addition, a secondary gap also understood as bound by the same wft-word, but typically not located in a position from which the w/z-word could have moved and also not c-commanded by the trace of the moved w/z-word.23 A standard example is given in (49a), which is interpreted roughly as in (49b). (49)

a. which articles did you file e [without [PRO [reading e]] b. For which x, x some articles, did you file x without reading x?

Although which articles appears simultaneously to bind both variables, the gap in the adjunct phrase is not a viable source for w/z-movement, as shown by (50).

78

(50)

PARAMETERS OF SLAVIC MORPHOSYNTAX

*which articles did you file the letter [without PRO reading e]

Beginning with Contreras (1984) and as developed in Chomsky (1986a), these constructions have been analyzed in terms of a null operator binding the variable in the adjunct phrase, roughly as in (51). (51)

which articles did you file e [without [0 [PRO [reading e]]]

The null operator "0," just like an overt w/z-phrase or quantifier raised at LF, moves to some pre-sentential position from which it has scope over the adjunct and can bind its trace as a variable. As with any non-overt element, its exact position is somewhat debatable, although it presumably either moves by substitution into SpecCP (COMP) or adjunction to IP or conceivably to CP. One important reason for favoring this kind of analysis is that, as Contreras (1984) observed, the parasitic gap shows movement effects, as in (52). (52)

a.

*which articles did you file e [after [0 [[PRO discussing e] upset John]]]

b. ?*which articles did you file e [after [0 [PRO hearing [stories about [claims in e]]]]] Movement of the null operator out of the subject of the after-phrase in (52a) would violate Huang's (1982) Condition on Extraction Domains (CED), which prohibits movement out of subjects and adjuncts, since they are not properly governed. Movement of the null operator out of the complex NP in (52b) would violate Subjacency, since two bounding nodes are crossed.24 That is, the ungrammaticality of these examples has nothing to do with the parasitic gaps per se, but rather with general conditions on movement, comparable to those operative in (53) with overt H'/i-movement. (53)

a. *which articles did [PRO discussing e] upset John b. *?which articles did you hear [stories about [claims in e]]

Parasitic gaps thus involve movement of a null operator, and the one-to-one relationship between operators and variables required by Koopman and Sportiche's (1982) Bijection Principle is preserved. This analysis raises, of course, a host of questions about the status and properties of the null operator. In broad terms, although as expected it derives its case and thematic properties from its source position, the null operator appears to need to be licensed by a "real" overt operator.25 Not only can the null operator exist in the scope of an overt operator, but it is restricted by the case and thematic properties of the overt operator in exactly the same way as observed for ATB constructions. This implies that these conditions actually involve the licensing of null operators and that ATB constructions should be analyzed in these terms. Consider in this light the contrast between the acceptable Russian (54), (55) and the unacceptable (56), (57).26 (54)

devuska, kotoroj Ivan daval den'gi e do logo, kak girl who(dat-gen) gave money until (on) stal izbegat' e (he) started to-avoid 'the girl who Ivan gave money to until he started to avoid'

ACROSS-THE-BOARD DEPENDENCIES

79

(55)

okno, kotoroe razbilo e vetrom do logo, kak window, which(acc-nom) broke wind(inst) before e lezalo na zemle lay on ground 'the window which was broken by the wind before it lay on the ground'

(56)

*devuska, kotoroj e bylo veselo do logo, kak girl who(dat-gen) was merry until Ivan stal izbegat' e started to-avoid ' the girl who was happy until Ivan started to avoid'

(57)

*okno, kotoroe razbilo e vetrom do logo, kak window, which(acc-acc) broke wind(inst) before my uvideli e na zemle we saw on ground ' the window which was broken by the wind before we saw it on the ground'

In all these examples, the morphological form of the relative pronoun is appropriate to both gaps, but only in (54) and (55) is the thematic prominence parallelism requirement met. That is, in (54) both gaps are thematically not most prominent while in (55) both are most prominent, even though the first corresponds to an object of a subjectless verb and the second to a subject. In both (56) and (57), on the other hand, the gaps are most prominent in the first conjunct and not most prominent in the second, leading to a violation of thematic prominence parallelism. It is worth noting that, just as with asymmetrical ATB constructions, this has nothing to do with identity in case, either morphological or abstract—indeed, (55) is superior to (57), even though the former involves an accusative object (presumably abstract Objective for Dyfa) and a nominative subject (presumably abstract Nominative for Dyta), whereas the latter involves two accusative objects. What is crucial about the construction in (55) and (57) is, instead, that it is subjectless, so that the theme argument must be most prominent.27 In order to express this restriction in a unified manner, then, ATB and parasitic gap constructions must receive similar analyses. Thus, adopting the null operator account of parasitic gaps implies its extension to ATB phenomena.28 Just as the null operator is restricted by the overt one in terms of thematic prominence, it is also limited by the case of the overt operator. However, as with the ATB constructions examined in section 3.1, this parallelism requirement must be stated in terms of case features rather than case per se. Thus (54) and (55) were acceptable despite the disagreement in case, since the form of the relative pronoun was compatible with the case requirements imposed on each gap. If, on the other hand, there is no compatible syncretic form, then the construction remains ungrammatical. Compare (58) with (54). (58)

*mal'cik, kotoromu/kotorogo Masa davala den'gi e boy who(dat)/(gen) (nom)gave money do togo, [0 [kak (ona) stala izbegat' e]] until (she) started to-avoid ' the boy who Masha gave money to until she started to avoid'

80

PARAMETERS OF SLAVIC MORPHOSYNTAX

There is no viable form of the relative pronoun that can simultaneously satisfy the dative indirect object of davala 'gave' (under the sister to V configuration) and the quirky genitive object of izbegat' 'to avoid' (lexically specified in its case submatrix). This result is not completely expected given the null operator analysis, since the postulation of two independent operator-variable pairs might suggest that they should have independent case properties. That is, the relative pronoun should be dative kotoromu since it originates in indirect object position, and the null operator 0 adjoined to the adjunct CP should be genitive since it is the D-structure complement of izbegat', which idiosyncratically assigns genitive. Assuming that the null operator has pronominal and case features, the source of the conflict is unclear. In any event, the syncretism must be the crucial factor in licensing such asymmetric dependencies, with the case features of the overt relative or interrogative pronoun somehow being imposed on the null operator. There are two ways that this might be formally realized. The most straightforward approach is simply to require that the case features of the null operator be nondistinct from those of the relative pronoun. This stipulation, which is essentially a translation of the case-matching requirement from section 3.1 into the null operator theory, works because the licensing syncretic relative pronoun will be underspecified for its case features and the null operator—by virtue of the fact that it lacks an actual wordform—will be fully specified for all case features. For example, the neuter singular relative pronoun wordform kotoroe is lexically associated with the degenerate case feature subset [ — obi], allowing it to replace fully specified nominative or accusative neuter singular relative pronouns in syntactic trees, from which it is non-distinct. But note that it is also non-distinct from a null operator in either of these cases, regardless of whether it has been inserted into a nominative or accusative position. Wordform insertion thus obscures all morphosyntactic features except those actually present on the wordform itself. Under this view, one might consider the case and thematic parallelism requirements respectively to reflect PF and LF sides of null operator visibility. That is, in order for the null operator to be licensed at PF, it must be identified by an overt operator, a process that depends on non-distinctness of morphosyntactic features. The null operator presumably has its pronominal features set by those of the overt operator under coindexation, which in turn receives them from the head of the relative clause, but its case features are set within its own clause. So long as these are non-distinct from those of the overt operator, however, the null operator is identified at PF. At LF, on the other hand, identification requires a minimal kind of thematic compatibility that can be expressed in terms of the absolute thematic prominence of the null operator and the licensing overt operator. This might in some sense be understood as a non-distinctness in thematic prominence requirement, assuming non-distinctness to be satisfied by all operators being equally most prominent.29 The second way of realizing feature matching is literally to let the overt operator impose its case features on the null operator. Although technically complex, there are a few reasons why such a process might be favored. Notice that according to the non-distinctness analysis just put forward the features of the null operator cannot be directly contrasted to those of any other coindexed null operator, since in a multiple

ACROSS-THE-BOARD DEPENDENCIES

81

ATB or parasitic gap construction involving two null operators of different cases a mismatch would be inevitable, although these are acceptable so long as the overt wh-word is morphologically non-distinct from each of the null operators. Under the licensing proposal, therefore, it is crucial that only the overt operator be a valid licenser. If, on the other hand, the case features of null operators were not set within their respective clauses, but rather only through coindexation with the overt operator, then this potential inconsistency would not arise. I do not adopt the feature-matching approach, however, since I see no simple way to prevent the null operator from receiving case in its underlying position. It is tempting, nonetheless, to suppose that null operators are in fact caseless, since they generally do not have the option of being overt. This is at least true of parasitic gap constructions, although null operators do alternate with overt relative pronouns in relative clauses of both free relative and ATB varieties—use of a second overt relative pronoun is indeed the way to save an asymmetric construction in which no syncretic form is available, since no licensing requirements will be imposed. That is, any of the unacceptable ATB dependencies discussed in this chapter can obviously be remedied by introducing the appropriate relative pronoun in the second conjunct, as in (59), based on (3). (59)

dziewczyna, ktora Janek lubi e a ktorej Jerzy nienawidzi e girl who(acc) likes and who(gen) hates 'the girl who Janek likes and who Jerzy hates'

Suppose, however, that this strategy can be formally distinguished from the null operator one. It may then be possible to maintain the argument that null operators lack case features altogether. They would in this sense be comparable to PRO, harking back to proposals in Epstein (1984) and Lebeaux (1984).30 That is, a wh-pbrase can have a case submatrix or it can lack one and be a wh-PRO. As PRO, it would presumably need to move by S-structure in order to be ungoverned31 and perhaps to find an antecedent/controller. More importantly, as a w/i-operator it would have to move by LF in order to create an operator-variable structure. Now the wh-PRO, although itself without a case submatrix, would bind a trace that has one. There would thus be a chain of indices from the overt relative pronoun to the wh-PRO to the variable, so that the non-distinctness requirement could be stated on coindexed maximal projections.32 Thus, the relative pronoun will be partially specified (if syncretic) and the trace will be fully specified, so non-distinctness works as before. Since PRO is totally unspecified for case, it is always non-distinct from every case submatrix. Finally, it is conceivable that whether the w/z-operator is PRO or not might correspond to the position it eventually moves to, in the sense that overt tWz-phrases substitute into Spec-CP, while null ones adjoin, either to CP or IP.33 This then could be the difference between overt and null operator relative pronoun constructions needed above. The strategy of introducing a second relative pronoun is viable in ATB constructions since these have an available Spec-CP position, but this will not work for parasitic gap clauses, which, according to the Lasnik and Saito (1992) analysis, can only be adjoined to. Before turning to that analysis and a few further similarities between ATB and parasitic gap constructions, it is worth noting that many symmetrical ATB depen-

82

PARAMETERS OF SLAVIC MORPHOSYNTAX

dencies probably do not involve a null operator, but rather instantiate simple conjunction with a single operator having scope over two conjuncts that share a single variable, perhaps along lines developed in Goodall (1987). ATB dependencies are thus fairly normal, whereas the parasitic gap construction is highly marked and unavailable in many languages, such as Polish and Serbo-Croatian. Although one might be inclined to think that this variation has to do with the lack of null operators in such languages, the licensing effects exhibited by Polish ATB constructions show that this cannot be true. The distribution of the two constructions is thus not as parallel as might be expected, although the difference may lie in the range of positions in which null operators are licensed in various grammars. Notice, however, that when either an ATB or parasitic gap dependency fails, languages tend to employ the same resumptive pronoun strategy to save the construction. In Serbo-Croatian, for example, resumptive pronouns are used instead of parasitic gaps (although Zeljko Boskovic [personal communication] points out that sometimes parasitic gaps are possible). Resumptive pronouns similarly appear in illegal ATB constructions, such as (60). (60)

devojka, kojoj je Marko dao svoj kaputic e girl who(dat-dat) AUX (nom) gave(m) self's jacket a usprkos toga *(njoj) je bilo zima but despite that her(dat) AUX was(n) cold 'the girl who Marko gave his jacket to but was nonetheless cold'

A gap in the second conjunct would violate the thematic prominence parallelism restriction; therefore, the overt dative pronoun njoj 'her' must be used instead. Even with valid ATB dependencies, the use of a resumptive pronoun is admissible in Serbo-Croatian, and if anything more natural: (61)

ucitelj, koga Lucija voli e a Rada (ga) mrzi e teacher, who(acc-acc) (nom) loves but (nom) him(acc) hates 'the teacher who Lucija loves but Rada hates'

Since the use of resumptive pronouns is traditionally assumed to be a last resort strategy, as argued in Shlonsky (1992), for example, the existence of both possibilities in (61) suggests that there are two alternative structures. Notice that this pattern parallels that of non-conjoined relatives in Serbo-Croatian, where the koji-type disallows a resumptive pronoun and the sto-type requires one:34 (62)

a. u£itelj, teacher, b. ucltelj, teacher,

koga (*ga) Lucija voli e who(acc) him(acc) (nom) loves sto *(ga) Lucija voli e who him(acc) (nom) loves

Since koga inflects for the pronominal features of the head of the relative clause and the case features appropriate to the gap position, as well as obeying island constraints on movement, it is clear that (62a) involves w/i-movement. In (62b), on the other hand, sto does not inflect and the resumptive pronoun is obligatory, suggesting that no movement is involved. Assuming the w/z-phrase koga is in Spec-CP and the complementizer sto is a head in C, there must then be an empty operator in (62b).

ACROSS-THE-BOARD DEPENDENCIES

83

However, the obligatoriness of the resumptive pronoun implies that movement has not occurred, thus the null operator and the resumptive pronoun are both basegenerated. Returning then to (61), the variant with the resumptive pronoun represents a base-generated null operator and the one without the pronoun reflects a simple conjunction.

3.3.2. Differences In concluding this chapter, I draw attention to some important differences between these two constructions. My initial realization that they may deserve a common analysis was due to the observation in section 4.1.2.2 of Lasnik and Saito (1992) that subject parasitic gaps can be saved by embedding, as a comparison of (63a) and (63b) reveals.35 (63)

a. *who did John see e [CP 0 [CP after [e kissed Mary]]] b. who did John see e [CP after [Sue thinks [CP 0 [e kissed Mary]]]]

This is precisely the same pattern observed for ATB constructions, which I analyzed in terms of the addition of a higher argument Sue optionally lowering the relative prominence of the subject gap in the embedded clause, thereby rendering both gaps not most prominent.36 The alternative configurational analysis presented in Lasnik and Saito (1992) is not conceptually incompatible with my proposals, however. It relies on the ECP to rule out examples such as (63a). The idea is roughly that the presence of after in COMP in (63a) prevents the null operator from moving to head position and hence it must adjoin to the CP. This in turn prevents it from governing its trace, under the assumption that only X° categories can be proper governors. The problem is circumvented in (63b), however, because another COMP position is available from which to antecedent-govern, as well as in object gap constructions, where the trace is lexically governed instead:37 (64)

who did John see e [CP 0 [CP after [Mary kissed e]]]

However, since examples like (63a) are independently ruled out by the thematic parallelism constraint, the need for Lasnik and Saito's ECP account is not clear on the basis of these examples alone. On the other hand, there do exist acceptable asymmetric ATB constructions, such as (40) above. Consider the following typical example, cited in Anderson (1983): (65)

?a book which I haven't read e [but e was recommended by several professors]

It is unclear why certain asymmetric ATB constructions should be valid and others not.38 Although I suspect that this may follow from a more carefully articulated theory of prominence, such as those of Jackendoff (1990a) or Grimshaw (1990), it is interesting to note that making the second conjunct active, as in (66), does not lead to unacceptability. (66)

a book which I haven't read e [but several professors recommended e]

84

PARAMETERS OF SLAVIC MORPHOSYNTAX

Of course, (66) should be perfectly fine according to all the ATB criteria discussed in this chapter. The point is that, since passivization usually lowers the prominence of the external argument, active and passive variants of the same ATB dependency should not have equal felicity.39 Under both Jackendoff and Grimshaw's theories, the passivization should render the external argument of the active no longer most prominent. For Jackendoff, passivization makes the external argument into an adjunct and the thematic hierarchy only pertains to arguments; for Grimshaw, the external argument becomes something called an argument-adjunct. Either way, it seems as though passivization need not render the erstwhile thematic prominence of the demoted subject completely inaccessible.40 Nonetheless, this phenomenon does not extend to parasitic gap constructions. Compare ATB (65) with the comparable parasitic gaps in (67). (67) a. *a book which I read e [after e was recommended by several professors] b. *a book which I haven't read e [although e was recommended by several professors] Whatever it is that renders (65) acceptable is inadequate to save (67). It may thus be that the ECP is applicable in (67) but not in (65). Although it is unclear then how to formalize this difference in Lasnik and Saito's system, it appears that conjunctions do not inhibit proper government as complementizers do. This suggests that conjunctions are not heads, or at least do not fill COMP position.41 I return in this light to Lasnik and Saito's ECP analysis of parasitic gaps, which at first blush seemed to extend to ATB dependencies. Although the similarities displayed are extremely suggestive, there are other comparable ways in which ATB and parasitic gap constructions differ. For example, while it is impossible to form a parasitic gap construction on an adjunct position, this is not generally true of ATB constructions, as demonstrated by (68). (68) a. *how did Bill fix the car e [after [0 [PRO ruining the truck e]]] b. how did Bill fix the car e [after [Sue said [0 [Paul ruined the truck e]]]] c. tell me how Bill fixed the car e [and [0 [Paul ruined the track e]]] Lasnik and Saito would rale out (68a) because the adjunct parasitic gap is neither lexically nor antecedent-governed, just like the subject gap in (63a), and the presence of after in C prevents the null operator from moving to head position, hence it must adjoin to the CP. This in turn prevents it from antecedent-governing its trace, under the assumption that only X° categories can be proper governors. This treatment is supported by the fact that (68a) improves similarly under embedding, as shown by (68b). However, the ATB adjunct extraction is perfectly fine even without the embedding, as shown by (68c). Some other constructions in which Lasnik and Saito make use of the ECP to rale out parasitic gap constructions are given in (69). (69) a. *to whom did you speak e [without [PRO sending a paper e]]

ACROSS-THE-BOARD DEPENDENCIES

85

b. *how angry can John appear e [without [PRO becoming e]] c. *what did Mary become e [without [PRO considering John e]] They argue that the PP trace in (69a), the AP trace in (69b), and the predicate nominal trace in (69c) are not lexically governed and that without blocks antecedent government. However, it seems to me that the parallel ATB constructions in (70) are much better.42 (70) a. ?the girl to whom you spoke e [and Bill sent a paper e] b. how angry did John appear e [and [(Bill) become e]] c. what did Mary become e [and [(Sue) consider John e]] I therefore conclude that the ECP applies in parasitic gap constructions but not in ATB constructions. This can be implemented in Lasnik and Saito's system by claiming that conjunctions do not inhibit antecedent government as the prepositional complementizers involved in parasitic gaps do. This follows directly if conjunctions do not fill the C position, but rather are analyzed either as adjoined elements or as heads of something other than CP.431 thus conclude that it is only the discrepancy in thematic prominence that rules out ATB (44a), so that introducing a more prominent argument in (44b) saves the construction, whereas the ECP is also violated in parasitic gap (63a), although avoided in (63b). I turn, finally, to some interpretive parallelisms between these two structures.44 Essentially, reconstruction effects do not obtain for either parasitic gap or ATB constructions. Consider the following examples: (71) a. b. c. d. e. f.

which picture of himself; j did Bill; say Johnj likes? which picture of himself; *j does Bill; like [and [0 [Johnj hate e]]] which picture of himself; *j does Bill; like [because [0 [John^ bought e]]] whose*; picture did he; see? whose; *j picture does Billj like [and [0 [he; hate e]]] whose; picture is tough [0 [for him; to hate e]]

As shown by (71a), a reflexive contained in a w/t-phrase can be interpreted in its D-structure position or in that of the intermediate COMP. The null operator constructions in (71b, c), on the other hand, are unambiguous. It looks as though the antecedent of the reflexive is determined solely on the basis of the position to which the overt w/z-phrase reconstructs, with the properties of the null operator, as always, deriving from the overt one. Similarly, note that reconstruction ordinarily feeds Condition C, so that in (71d) whose must be disjoint from he. Null operators, on the other hand, do not display this reconstruction effect, as illustrated by the examples in (71e, f). The pronoun may be coreferent with whose in (71e, f), demonstrating once again that the referential properties of the null operator are determined by those of its overt antecedent and that, whatever their correct anaysis, ATB and parasitic gap constructions behave similarly.

86

PARAMETERS OF SLAVIC MORPHOSYNTAX

3.4. Conclusion The evidence reveals that the felicity of ATB dependencies involves not only structural and morphological factors, but semantic ones as well. The Slavic data show that if the sole NP argument of an impersonal verb is questioned or relativized, then it can enter into an ATB dependency with (1) the sole argument of another impersonal verb or (2) the subject argument of a personal verb, regardless of the abstract Case or thematic role of any of these NPs. I have argued that what is crucial is the relative thematic prominence of each argument within its own conjunct, proposing that the operative constraint is that all gaps in any ATB construction must pertain to either most prominent or not most prominent arguments, exclusively. This condition on coordinate structures appropriately handles the facts originally observed by Dyla for Polish and extends to others for which his system was unable to account, including parasitic gap constructions. I therefore suggested that the thematic parallelism constraint should be regarded as a licensing condition on null operators in general, and that ATB constructions may be assimilated to parasitic gaps and analyzed as involving binding by a null operator. Interestingly, the case of this operator is determined by the overt operator in the sense that it agrees in case features, rather than in Case per se.45 Null operators thus have an abstract morphological form, which they inherit from the overt element that licenses them, and they must match this overt operator in terms of the thematic prominence of the arguments that they bind, although they need neither head chains with the same Case nor theta-role. Notes 1. Some of the observations in this chapter were originally made in Franks (1985) and reported at the 1990 Midwest Slavic Conference in Urbana-Champaign, as a joint project with my Indiana University S535 class. The participation and assistance of the students in that class—Paul Heck, William McKernan, Annie Joly Sperling, and Curt Woolhiser—is gratefully acknowledged. Preliminary results of the present analysis have also been presented at colloquia at Bucknell University and the University of Pennsylvania, as well as the 1991 meeting of the Formal Linguistics Society of Mid-America at Ann Arbor. Versions of the material in this chapter have appeared as Franks (1992) and Franks (1993). 2. Gaps are indicated by boldface e. In the ATB examples, the two cases required by each conjunct are indicated in the gloss of the wh-word as follows: 'who (gen-dat)' means that the syncretic form of who must be construed as genitive with respect to the first conjunct and dative with respect to the second, 'who (acc-acc)' means it is accusative for both. 3. Examples (3)-(5), (9), and (10) are taken from Dy-ta (1984); the rest have been either supplied or confirmed by native speakers. It should be noted, however, that some Russian speakers do not readily accept some of the more bizarre examples. This might be explained in terms of variation in how individuals represent specific syncretisms, in that they may not be capturing all possible redundancies by using available underspecified lexical wordform entries.

ACROSS-THE-BOARD DEPENDENCIES

87

4. The same is true of other constructions involving multiple dependencies, such as free relatives (cf. Franks, 1981) and parasitic gaps, as discussed in section 3.3. 5. Clearly, some modification of the proposed Russian feature system will be necessary for the other languages, but there is no principled reason why the syncretism phenomena in Polish, for example, should not be handled in exactly the same way as for Russian. 6. The phenomena treated in this chapter reflect one area that, in contrast to the bulk of constructions analyzed in this book, does not display theoretically significant distinctions across the Slavic languages. I do not discuss other Slavic languages here chiefly because many of them avoid asymmetric ATB dependencies altogether, making use of an alternative resumptive pronoun strategy instead. Relevant aspects of this issue will be briefly addressed in section 3.3. 7. Dy-fa makes a similar statement concerning (10), but since neither conjunct involves the genitive, I assume he intended to deny that the dative experiencer could be an instance of Objective, although claiming that the dative benefactive must be. 8. The status of expletive (i.e. dummy or pleonastic) subjects is treated in depth in chapter 7. 9. They cannot both be Oblique, since examples such as (5) show that indirect object datives are compatible with Objective genitives. 10. For arguments that dative experiencers in Slavic are not true syntactic subjects, see the discussion in chapter 6, as well as Greenberg and Franks (1991). 11. This is in addition, of course, to the kinds of parallelism constraints on coordinate structures in general proposed in, for example, Ross (1967/1986), Schachter (1977), and Williams (1978). 12. One point of discrepancy has to do with the status of "Experiencer" as a role. Interestingly, Thomas Stroik (personal communication) points out to me that the hierarchy in (21) fails to explain the ungrammatically of examples with psych-verbs, such as the following: (i)

*the kind of person who politics bothers e and e writes letters to newspapers

(ii)

*the sort of literature which e bothers Bill and Mary writes e

The problem here is that the thematically more prominent Experiencer occupies a structurally less prominent position. The status of psych-verbs is under considerable debate (cf. e.g. Belletti and Rizzi, 1988; Grimshaw, 1990; Pesetsky, 1992). It may be that the problem posed by (i) and (ii) can be solved in terms of suggestions by both Jackendoff and Grimshaw that the thematic hierarchy interacts with other hierarchies ("causation" and "aspectual," respectively), so that predicates such as bother involves a mismatch in prominence. Notice, incidentally, that these examples greatly improve when the second conjunct contains an Experiencer that is both thematically and structurally most prominent (i.e. is highest on all relevant hierarchies, so that there is no mismatch): (iii)

?the kind of person who politics bothers e and e loves to debate

(iv)

?the sort of literature which e bothers Bill and Mary despises e

13. For example, one popular alternative is given in (i), based on seminal work by Jackendoff (1972) and more recently developed in chapter 2 of Grimshaw (1990): (i)

(Agent (> Experiencer (> Goal/Source/Location (> Theme))))

One way in which the hierarchy in (i) crucially differs from (21) is that it ranks location above theme. Since in the ATB examples considered, such as (9) and (11), location never patterns as more prominent than theme, the formulation in (21) is preferable. A second respect in which

00

PARAMETERS OF SLAVIC MORPHOSYNTAX

(21) and (i) differ is in the relative prominence of goal and theme; some facts supporting my claim that goal datives should be lower on the hierarchy than themes, in clear contrast to experiencer datives, are discussed shortly. 14. The role of thematic prominence in the analysis of ATB dependencies in this chapter accords well with the kind of linking strategy (for connecting semantic arguments and syntactic positions) employed for example in Carrier-Duncan (1985), Grimshaw (1990), and perhaps best explicated in Jackendoff (1990a, ch. 11). A related alternative within categorial frameworks might be formulated in terms of the idea that the most prominent argument is the last one to be added in building the clause (cf. e.g. Dowty's, 1982, treatment of grammatical relations within Montague grammar). 15. See Bailyn (1991) for an analysis of Russian that conforms to Larson's system. 16. The gist of her argument is that Williams's (1981) reasons for positing a distinct external argument are based on tests which show that every head X has a unique most prominent argument that is "visible to rules of grammar in a way that other arguments are not" (Speas, 1990, 102), but that in fact fail to show, contra Williams, that this argument is syntactically external to Xmax. 17. Recall that the genitive form here is syncretic with the accusative, so that morphological compatibility is satisfied. Note also that this example provides further evidence for placing locative lower than the theme. 18. These examples present several extraneous problems that reduce their effectiveness. For one thing, scrambling to place the accusative NP before the dative one is highly marked; for another, speakers much prefer to use the reflexive possessive form swojego 'self's' instead ofjego in (30a) when the bound reading is intended. 19. Judgments about the ATB dependencies in (32a) and (34a) are admittedly strained, given the stylistic infelicity of such examples as well as the awkward contrast between the tenseless participial form and the tensed second conjunct. 20. Note that example (35b) is ungrammatical only under the reciprocal reading; it can of course mean 'I introduced myself to my friends.' 21. The awkwardness of (43b) may be an effect of conjoining a stative predicate with a non-stative one, as suggested to me by Rich Campbell. 22. The example they discuss comes from Dutch. Niejt and Goodall have no explanation for this variation. I believe it may either relate to how different languages interpret relative thematic prominence when one functional complex is embedded within another or it may follow from some structural constraint in addition to the parallelism requirement in (20). This latter possibility seems reasonable if the parasitic gap analysis presented in the next section is adopted. 23. One potential exception is the marginal type Who did you give pictures ofe to e?, since Who did you give pictures of Bill to e? is also a valid question. In his discussion of this, Stowell (1985) notes that this type can be understood as multiple interrogation (it can be answered "I gave pictures of Bill to Fred") and proposes an explanation in terms of null operator licensing. There is a copious literature on parasitic gaps, beginning with Engdahl (1983) and treated in detail in Chomsky (1982, 1986a); for a comprehensive study of the properties and distribution of null operators, see Browning (1987). Other similarities between the two constructions beyond the fairly obscure ones discussed here have also been recently noted in the literature. Pesetsky (1982), Hai'k (1985), Huybregts and van Riemsdjik (1985), Cowper (1985), Williams (1986/1987), and Munn (1992) all argue for unifying ATB and PG in one way or another. Haik and Williams maintain that some or all PG constructions should be treated as coordination and assimilated to ATB, whereas Munn proposes that ATB is actually an adjunction structure and should thus be assimilated to the PG construction.

ACROSS-THE-BOARD DEPENDENCIES

89

Other similarities between the two constructions than those discussed here have also been noted, as in Huybregts and van Riemdjik (1985), Cowper (1985), and Williams (1986/1987), although the proposal that ATB constructions may involve null operators has not to my knowledge previously been made. 24. I refer to the CED and Subjacency descriptively, without concern for their precise formulations, since both have been revised in terms of the barriers model of Chomsky (1986a). 25. See Stowell (1985) for an interesting discussion of licensing conditions on null operators. One probable counterexample to this generalization is the towg/i-movement construction, as in John is tough [0 [PRO to please e]], where the null operator seems to be licensed by the matrix subject John. The presence of the null operator is revealed by the fact that it licenses parasitic gaps, since only A'-movement does this: John is tough to please without liking. This construction is also unusual in that it is reasonable to suppose that tough assigns no theta-role to subject position; compare the grammaticality of it is tough to please John, with no movement (hence *it is tough to please John without liking). It would appear that in this case the null operator somehow shares its theta-role with the matrix subject John and is in this way licensed, although the mechanics of this process remain mysterious. 26. I use Russian examples here because Polish for some reason does not tolerate parasitic gap constructions, employing the resumptive pronoun strategy instead. This is presumably not due to a prohibition against null operators in general since, as we have seen, Polish does exhibit ATB constructions that show the parallelism restrictions ascribed to null operators. 27. More precisely, it has a null expletive subject, as discussed in chapter 7. The theme argument is more prominent than the instrument because the latter, as a demoted or "suppressed' ' subject, no longer counts as an argument for the purposes of the thematic hierarchy (cf. Jackendoff, 1990a). An alternative way of conceptualizing this is that the demoted subject becomes an argument-adjunct in the sense of Grimshaw (1990), and is thus less prominent than any true argument (cf. also Pollard and Sag, 1987). For a comprehensive study of this unusual construction, see Green (1980). 28. One might, of course, follow Williams (1986/1987) and take the opposite approach of extending the ATB theory to parasitic gaps by analyzing the latter in terms of an optional and covert coordinating conjunction. I will not pursue this possibility here, adopting the more standard null operator account of parasitic gaps instead. 29. These speculations suggest that the licensing requirements on null operators might be reducible to the ECP that, as [ — pronominal] empty categories, they should be subject to, or to generalized binding theory (cf. Aoun, 1986; Aoun, Hornstein, Lightfoot, and Weinberg, 1987), since proper antecedency requires feature non-distinctness. I do not explore this possibility here, although the fact that null operators must be licensed by coindexed and c-commanding overt operators that are, furthermore, morphosyntactically and thematically compatible is tantalizing. 30. The feature content of null operators is a subject of much speculation; compare Contreras (1993) for the interesting proposal that their features are freely assigned but are then subject to licensing conditions, which allows for different kinds of null operators. See also Browning (1987), who argues that null operators should be analyzed as A'-pro elements that eventually delete at LF. The behavior of PRO with respect to its ability to participate in case agreement processes despite its lack of a case submatrix (and concomitant inability to bear case and be phonetically realized) is discussed in detail in chapter 6. 31. PRO must not be governed under standard assumptions; compare, however, Franks and Hornstein (1992) and the many references therein for alternative approaches.

90

PARAMETERS OF SLAVIC MORPHOSYNTAX

32. It probably cannot be extended to all coindexed nodes due to the existence of case conflicts in quantified NPs, as discussed in the next two chapters. 33. See, however, Deprez (1989) for a theory in which all w/z-phrases adjoin to CP (except for locally moved subjects, which substitute into Spec-CP). Although whether null operators adjoin to CP or to IP is unclear, as observed in Lasnik and Saito (1992), for example, the choice is essentially immaterial to their analysis. 34. This is not true of subject gaps, as in ucitelj, sto je doSao 'the teacher who came'. Following Shlonsky (1992), I believe the absence of an overt subject resumptive pronoun here is due to the fact that Serbo-Croatian is a null-subject language, as treated in chapter 7. Hence, there actually is a resumptive pronoun, but it is phonologically null. See Browne (1986) for a comprehensive discussion of relative clause structures in Serbo-Croatian. 35. They in turn attribute this observation to Juan Uriagereka. Although the constrast is clear, (63b) and (64) are not entirely felicitous, since parasitic gaps in finite clauses are in general somewhat degraded. 36. Note that embedding a subject parasitic gap in Dutch fails to improve it, just as with non-parallel subject ATB gaps. Thus, *Wie beldeje op nadatMarie zei (daf) aankwam 'who did you telephone after Marie said (that) arrived' is no better than *Wie belde je op nadat aankwam 'who did you telephone after arrived' (examples provided by Alice ter Meulen). This parallelism further suggests that the same effect is involved. 37. Although this account relies on the traditional bifurcated ECP (as in e.g. Chomsky, 1981a), according to which [ — pronominal] empty categories must be either antecedentgoverned or lexically governed, it also translates into more recent versions, such as that of Aoun, Hornstein, Lightfoot, and Weinberg (1987), in which empty categories must satisfy both requirements (at different levels). 38. Another confounding factor that I have ignored in this chapter is the choice of conjunction; (65), for example, sounds considerably better than (i) with and: (i)

??a book which I recently read e [and e was recommended by several professors] . . .

There is a curious effect in which use of the conjunction but, especially when coupled with negation in the first clause, overrides all other considerations to vastly improve the construction. 39. For example, (i) is much better than (ii): (i) (ii)

the letter which John wrote e and Bill mailed e *the letter which John wrote e and e was mailed by Bill

Once again, (ii) seems to me to be improved by substituting but for and: (ii)

??the letter which John wrote e but e was mailed by Bill

40. Interestingly, the same is true for binding, which is also sensitive to relative thematic prominence. Thus, when passivization demotes the external argument, perhaps as a lexical process that simply changes it into an adjunct, the next highest argument is promoted. This process causes the theme to become the subject (either by movement or direct linking), thereby turning it into a proper antecedent for subject-oriented anaphors, as in Polish and Russian. However, demoting an agentive subject does not necessarily prevent it from binding, as shown by Polish (i), from Dziwirek (1991, 120), or Russian (ii): (i)

Anna byfaodesJanaprzez Janka do swoich rodzicow (nom)was sent by (ace) to self's parents 'Anna was sent by Janek to his/her parents'

ACROSS-THE-BOARD DEPENDENCIES

(ii)

91

stat'i drug o druge byli opublikovany nami v raznyx 2urnalax articles each about other were published us(inst) in different journals' 'articles about each other were published by us in different journals'

In (i) Janka can bind swoich even though it neither is a subject nor c-commands the reflexive. In (ii) nami can bind drug o druge even though it neither precedes nor c-commands the reciprocal (subjecthood not being relevant). Dziwirek (1991) points out, however, that the range of possibilities in the passive is more restricted than in the active; these and related facts do not seem amenable to a purely structural account. 41. A further major issue Lasnik and Saito's approach raises is how their system, which relies on the traditional exocentric S' -> COMP S phrase structure, translates into the current one in which S' is analyzed as CP, the maximal projection of a functional category C. 42. On the other hand, CP parasitic and ATB gaps are both possible since, as Lasnik and Saito note, the trace is assigned a theta-role and hence is lexically governed: (i)

that John is here, Mary claimed e [without really believing e]

(ii) I know that John is here, which Mary claimed e [but didn't really believe e] 43. Such a conclusion is inevitable, I think, given the fact that (i) below contrasts markedly with ATB (40), repeated as (ii). The same is true for the parasitic gap construction (iii) compared with ATB (65), from Anderson (1983) and repeated as (iv): (i)

*which dress did Mary buy e [because [0 [e cost $500]]]

(ii)

?the dress which Mary bought e [and [0 [e cost $500]]]

(iii)

*a book which I read e [after [0 [e was recommended by several professors]]]

(iv)

?a book which I haven't read e [but [0 [e was recommended by several professors]]]

Although the ECP rules out (i) and (iii), it has no impact on (ii) on (iv). 44. The significance of these facts was originally drawn to my attention by Rich Campbell. 45. One potentially serious problem for the present theory of ATB dependencies, which has been based solely on a consideration of w/j-movement, is whether and how the analysis should extend to ATB NP-movement, as in (i). (i) the article was copied e and filed e Positing a null operator to bind the second trace raises a host of questions, and one might seek an alternative solution, perhaps along the lines of Goodall (1987), although this would of course diminish the credibility of the null operator analysis for ATB w/z-movement. On the other hand, it may also be possible to develop a null operator analysis of ATB NPmovement, since questions of the status of the null operator do not differ substantially from those raised by other putative null operator constructions. Thus, there is also no overt operator licensing the null operator in the towg/i-movement construction, and null operators do not need to be in Spec-CP, but rather may also bind traces from adjoined positions. The only significant difference is that null operator NP-movement would involve movement from a caseless position. The trace would be caseless, hence not functionally determined to be a variable; however, if the null operator is PRO, then this problem might disappear, since PRO is not subject to case requirements. Still, one wonders why, if (i) can involve a null operator, the comparable parasitic gap construction in (ii) is unacceptable. (ii)

*the article was copied e [without [0 [PRO reading e]]]

92

PARAMETERS OF SLAVIC MORPHOSYNTAX

Whatever the reason for this, it does seem to me that conjunction considerably improves the parasitic gap adjunct, as in (iii). (iii)

?the article was copied e [[and [0 [filed e]]] [without [0 [reading e]]]]

It thus may be that the null operator analysis, developed on the basis of ATB wto-movement, is salvageable for NP-movement as well.

4 Quantified Structures: Russian versus Serbo-Croatian

Quantified phrases in Russian and the other Slavic languages have long been a subject of linguistic investigation in both traditional and generative studies. Major contributions by linguists in the Russian tradition include Suprun (1959) and Mel'cuk (1985), and among important recent Western work is Pesetsky (1982), Corbett (1978, 1983a), Babby (1987), and Neidle (1988). In this chapter, specific proposals by Babby and Pesetsky about the syntax of Russian quantified phrases are examined and extended to Serbo-Croatian.1 It is argued that if aspects of each of these researchers' models of Russian are adopted and modified slightly in order to accommodate minor parametric variation, then conflicting phenomena in SerboCroatian can receive an explanatorily satisfactory and conceptually unified account. The interaction of case, agreement, and numeric quantification is one of the most recalcitrant problems of Slavic grammar. To quote Chvany (1975,134), "the grammar of Russian quantifiers is of harrowing complexity."2 Numeral phrases display many unusual and mysterious morphosyntactic properties, central among which are that (1) the numeral sometimes governs the nominal material following it and sometimes agrees in case with it, and (2) the numeral phrase sometimes induces subjectverb agreement and sometimes does not. Solutions to these puzzles proposed by Babby and Pesetsky both involve the assumption of a formal dichotomy: Babby (in a variety of works, including 1980a, 1984, 1985, 1986, and especially 1987) relies on an opposition between what he calls lexical and configurational case to explain the case government/agreement contrast in Russian; and Pesetsky (1982) argues that Russian numeral phrases may be either Quantifier Phrases (QPs) or NPs to explain the predicate agreement/non-agreement contrast. In what follows, I will show how both of these dichotomies can be parameterized to handle comparable phenomena in other Slavic languages. By extending to Serbo-Croatian Babby's insights into issues of how case is assigned to numerically quantified phrases and spread among them, and Pesetsky's into the syntactic distribution of numeral phrases, I intend to demonstrate that the range of variation encountered follows from general principles. 93

94

PARAMETERS OF SLAVIC MORPHOSYNTAX

Although Babby and Pesetsky are attacking distinct sets of problems and do not substantially address what the other considers to be the core puzzle posed by the Russian facts, I believe that the two analyses are essentially compatible. Moreover, although neither explores how his account might accommodate comparable yet inconsistent data in other Slavic languages, I believe that these too can be assimilated to their approaches. The broader applicability of these analyses is a crucial measure of their validity since, if the properties of Russian numeral phrases derive (as both Babby and Pesetsky claim) from principles of UG, then theoretically consistent accounts of related phenomena in the other Slavic languages should follow suit. After establishing appropriate parameterizations of Babby's and Pesetsky's models, I recast Pesetsky's analysis in terms of the hypothesis that subjects are canonically D-structure VP-specifiers that move to IP-specifier position in search of case, and that QPs differ from NPs in being exempt from this requirement. I thereby remedy several factual inadequacies of Pesetsky's account and explain why certain distinctions between QP and NP subjects in Russian are absent in Serbo-Croatian.

4.1. Case Properties of Numeral Phrases This section discusses the internal properties of numeral phrases in Russian and Serbo-Croatian. Babby (1987) offers an in-depth treatment of case distribution in such NPs in Russian. He argues that the complex pattern of existing possibilities can be accounted for by assuming an extremely hierarchical structure for these NPs, and then by letting the level at which a given modifier is adjoined determine the case it is eventually assigned. Crucial to the operation of Babby's system is the idea that case is first assigned (by an external governor) to NP and is only subsequently ' 'percolated down to all available lexical and phrasal categories in the phrase" (Babby 1987, 91). This view, which I regard as essentially correct, allows Babby to handle certain problems in the distribution of case within numerically quantified NPs in terms of the mechanics of NP-internal case assignment, mediated by general principles for resolving case conflicts. It is a relatively straightforward matter, however, to show that the facts of SerboCroatian are incompatible with the analysis put forward by Babby for Russian. My attempt to reconcile these facts with Babby's account requires the assumption that the quantificational genitive is a structural case in Russian but an inherent case in Serbo-Croatian.3 This relatively trivial aspect of variation enables me to account for the entire range of case and agreement facts.

4.1.1. Babby's Analysis of Russian Here, as elsewhere, Babby is concerned with explaining the striking contrast between what he calls heterogenous and homogenous internal case distribution in quantified NPs.4 When such phrases occur in a nominative or accusative position, the quantifier (Q) induces genitive in the following material. This phenomenon is known as the genitive of quantification, henceforth GEN(Q). In (1) the quantified NPs follow the verb citat' 'to read' and the preposition pro 'about'; such f — N]

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

95

categories in Russian assign accusative to their complements, unless otherwise specified. Following Babby and others, I assume throughout this chapter that the quantifier projects up to a QP. (1) a. citat' [NP [QP pjat'] [N. interesnyx knig]] to-read five(acc) interesting(gen pi) books(gen pi) b. pro [NP [QP about

pjat'] [N. interesnyx knig]] five(acc)interesting(gen pi) books(gen pi)

In contrast to this heterogenous pattern, a quantified NP in an oblique case position appears entirely in the appropriate case. The homogenous pattern is illustrated in (2), where the verb vladet' 'to possess' and the preposition s 'with' both govern the instrumental. (2)

a. vladet' [NP [QP pjat'ju] [N. interesnymi knigami]] to-possess five(inst)interesting(inst pi) books(inst pi) b- s [NP [QP pjat'ju] [N. interesnymi knigami]] with five(inst)interesting(inst pi) books(inst pi)

Because it is an idiosyncratic lexical property of these words that they take instrumental complements, Babby refers to this phenomenon as lexical case. This curious asymmetry between how quantified phrases behave in NOM/ACC and lexical case contexts constitutes a fundamental problem for any account of Russian case. This issue is treated in section 3 of Babby (1987). Having argued in the preceding section that heterogenous and homogenous constructions do not differ in X-bar structures, he must seek an external explanation. Babby's solution requires the adoption of a hierarchy for determining which case has precedence in conflict situations. In order to handle the facts in (1) and (2), Babby (1987,116) proposes the Syntactic Case Hierarchy in (3), which places GEN(Q) between lexical case and NOM/ACC. (3)

Syntactic Case Hierarchy: LEXICAL CASE > GEN(Q) > NOM/ACC

While the Syntactic Case Hierarchy indeed gets the required results, merely stipulating such a hierarchy lacks explanatory force. Babby therefore suggests that the three-place hierarchy in (3) can be reduced to two-places, as in (4), if GEN(Q) is taken to be a purely structural case.5 (4)

Syntactic Case Hierarchy (revised): LEXICAL CASE > CONFIGURATIONAL CASE

The operative principle here is one of locality—NOM/ACC cannot percolate down to N' since it is already genitive by virtue of being in the domain of a quantifier, but oblique case can since it is assigned earlier, before N' has had a chance to receive GEN(Q). In fact, it is clear that even this two-place hierarchy is an artifact of the dichotomy between inherent and structural case of Chomsky (1981a, 1986b), under the assumption that inherent case is assigned at D-structure and structural case at S-stracture. Indeed, in other work, Freidin and Babby (1984) and Babby (1985) account for the facts in (1) and (2) in terms of an alternative principle of UG—the Principle of Lexical Satisfaction in (5).6

96

PARAMETERS OF SLAVIC MORPHOSYNTAX

(5) Principle of Lexical Satisfaction: Lexical properties must be satisified. The Principle of Lexical Satisfaction holds at all levels of representation and, in particular, at D-structure. This more standard kind of solution exploits the observation that it is an idiosyncratic lexical property of the words in (2) that they take instrumental complements. In terms of the model developed in chapter 2, these lexical items contain case submatrices fully specified for instrumental. If lexical case is in this sense subcategorized for, then it must take precedence over any other casemarking strategy. This approach seems to me to be indistinguishable from the standard theta-theoretic account of so-called quirky case, as espoused by Pesetsky (1982) and others, whereby certain arguments must bear a particular oblique case in order for their theta-roles to be visible. Consequently, if a verb or preposition requires its complement to bear some specific case, then this consideration overrides (or blocks) any other. In what follows, I too shall assume that some such requirement is in effect. Babby's lexical case can be equated with the inherent case of standard GB, and his configurational case with structural case, so that the former will be assigned obligatorily at D-structure and the latter not until S-structure.7 Notice, as Babby argues at length and as discussed in Franks (1985, 1986), that this account requires the reinterpretation of some conventional GB assumptions about case assignment. In particular, since case appears to be being assigned by the QP to N', we must allow for case assignment directly to non-maximal projections. Although there is no reason why this should be impossible if N' has a case submatrix, one might instead regard the phrase over which the QP has scope as itself a maximal projection NP. Foreshadowing the analysis in section 4.2, I will accordingly analyze the QP as adjoined to NP in these examples, so that GEN(Q) is assigned to an NP under sisterhood to a QP. Alternatively, following proposals in Abney (1987), one might analyze the Russian quantifier as a kind of Determiner taking an NP complement. There are, however, several prima facie difficulties with this type of approach. First, as discussed in Franks (1986), Russian generally lacks the kinds of overt evidence for a functional category DP adduced by Abney. Second, this approach would only be applicable to those structures in which the numeral phrase is a QP, and would not extend to the NP ones, leaving the central dichotomy of section 4.2 unexplained. Third, again as argued for in Franks (1986) and Babby (1985,1987), there is substantial evidence that an entire QP is restricting the following nominal material. That quantifiers project a phrase follows both theoretically from the principles of X'-syntax and empirically from the fact that they admit various types of adverbial modifiers, as in Russianpriblizitel'nopjaf knig 'approximately five books' or the distributive po pjaf knig 'five books each' construction treated in the next chapter. Most of these problems, however, can be solved within an enriched theory of functional categories, as discussed in detail in chapter 5 and Franks (1994). Next, the analysis relies on the model of morphological case assumed in chapter 2, in which projections of [ + N] categories, crucially here at least both N and NP, may bear case features, the value of which is set either under coindexation with other cased categories or configurationally under sisterhood. Since all members of a projection are by definition coindexed, assignment of case to NP results in percolation down the projection. The spreading of case down the projection is thus an

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

97

automatic consequence of case assignment to NP, which takes place whenever possible.

4.1.2. Extending the Analysis: GEN(Q) Is Inherent in Serbo-Croatian I now turn to some curious differences between Russian and Serbo-Croatian. In Serbo-Croatian, most quantified phrases only exhibit the heterogenous case pattern, regardless of syntactic context. This is illustrated in (6), with the adjustment in phrase structure just proposed made explicit: (6) a. kupili smo [NPma* [QP pet] [NP knjiga]] bought(m pi) AUX(1 pi) five books(gen pi)

b. za [NPmax [QP osam] [^ dana]]

in eight days(gen pi) max c. sa [NP [QP pet] [NP djevojaka]] five girls(gen pi) with d. Turska se nalazi na [NPmax [QP oba] [NP ova kontinenta]]8 Turkey REFL locates on both these(pauc)continents(pauc) e. izmedu [NPmax [QP dva] [NP zla]] between two evils(pauc) f.

bojao sam se [NPmax [QP pet] [NP ljudi]] feared(m sg) AUX(1 sg) REFL five people(gen pi)

g. u toku [HP""*" [QP tri] [NP poslednje godine]] in course three last(pauc) years(pauc) h. vlasnik [NP [QP pet] [N. malih kuca] owner five small(gen) houses(gen pi) 'the owner of five small houses' Quantified NPs in Serbo-Croatian therefore appear to flout Babby's would-be universal principles, since the QP assigns case not only after items that assign ACC, as in (6a, b), but also after those that require specific oblique cases, as in (6c-h). A closer consideration reveals, however, that this problem can be resolved if the hierarchical account of NP-internal case distribution is rejected and the thetatheoretic one properly understood. Since the prepositions sa 'with', izmedu 'between', and na 'on' in (6) require instrumental, genitive, and dative/locative,9 respectively, the verb bojati se 'to fear' requires genitive, and virtually all nouns take genitive objects, I contend that the quantified phrases must be in these cases. This follows from the theta-theoretic view that such prepositions impose as an absolute requirement on their complements that they bear specific inherent cases. Moreover, the fact that these quantified phrases are cased—even if no element within them actually shows it morphologically—implies that they must be NPs. This point will become relevant when Pesetsky's analysis is examined in section 4.2. Even so, it is worth noting that these data are complicated by the fact that undeclined numeral phrase complements to verbs that subcategorize for an oblique

98

PARAMETERS OF SLAVIC MORPHOSYNTAX

case vary in acceptability. Speakers readily accept the genitive in such contexts, as in (6f) or the following similarly felicitous examples: (7)

a. cuvao sam se pet ljudi 'I guarded myself against five people' b. domogao sam se pet knjiga 'I obtained five books'

Since the reflexive verbs cuvati se 'guard oneself against' and domoci se 'obtain' call for the genitive, and the numeral pet also assigns the genitive, it may be that this is why verbs that assign quirky genitive so readily accept such complements. For other cases the facts are not as clear and various peripheral strategies for making the oblique case transparent on quantified complements exist, such as vacuously inserting the preposition s(a) 'with' after verbs governing the instrumental. Leko (1987) cites examples such as the following: (8) a. predsjednik vlada zemljom president rules country(inst) b. komandujem bataljonima I-command batallions(inst) (9) a. predsjednik vlada sa nekoliko zemalja president rules with several countries(gen pi) b. komandujem sa tri bataljona I-command with three batallions(pauc) Although, as (8) shows, the verbs vladati and komandovati govern the instrumental, when they take a quantified complement the preposition s(a) is required, as in (9). In a sense, then, this preposition can be regarded as an instrumental marker.10 Interestingly, the dative in such contexts is for some reason strongly disfavored, with Serbo-Croatian speakers reporting the following judgments: (10)

a.

*dao je knjigu pet ljudi he-gave AUX book(acc) five people(gen pi) b. ?*koracao je prema pet ljudi he-stepped AUX toward five people(gen pi)

The indirect object dative in (lOa) is unacceptable and the dative governed by the preposition in (lOb) is marginal, curiously contrasting to the locative context in (6b).n These facts are complex and worthy of further investigation, but nonetheless lead to the conclusion that quantified NPs in Serbo-Croatian can in principle occur in an oblique case position without marking that oblique case overtly. A second, important conclusion is that the quantifier in Serbo-Croatian must be inherently assigning GEN(Q) to the NP to its right, otherwise it would be overridden by percolation of the inherent oblique case on NP"13*. Thus, in (6c-h) there is a conflict of inherent case assigners, with the external governor requiring one case and the internal one another.12 According to the Principle of Lexical Satisfaction, both requirements must be met at D-structure.I conclude therefore that at D-structure

QUANTIFIED STRUCTURES! RUSSIAN VERSUS SERBO-CROATIAN

99

Npmax is iNST and the lower NP is GEN(Q). It is only in this way that Principle (5) can be satisfied. This parametric difference between the two languages, stated descriptively in (11), thus constitutes a fundamental source of variation between quantificational structures in Russian and Serbo-Croatian. (11)

a. GEN(Q) is a structural case in Russian. b. GEN(Q) is an inherent case in Serbo-Croatian.

In other words, quantifiers in Russian are structural case assignors on a par with verbs and prepositions taking accusative complements; quantifiers in SerboCroatian, on the other hand, are inherent case assignors on a par with verbs and prepositions taking oblique complements. Russian (Ib) and (2b) thus have D-structures, as in (12): (12)

a. [PP pro [NPmax [QP pjat'] [NP interesnyx knig]]] b. [PP s [Npmax:iNST [QP PJat'ju] [NP interesnymi knigami]]]

The only inherent case is instrumental in (12b), which percolates throughout NPmax onto all of its parts. In (12a), on the other hand, neither case is assigned at D-structure since neither is lexically required. Only at S-structure, where structural case is assigned, does the preposition in (12a) mark NPmax ACC and the QP mark NP GEN(Q), as in (13). (13)

[PP pro [NPmax:ACc [QP PJat'] [NP:GEN(Q) interesnyx knig]]]

In Serbo-Croatian, on the other hand, the QP assigns inherent case. Hence, SerboCroatian (6b, c) will have roughly the D-structures in (14), with ACC being assigned in (14a) only at S-structure. (14)

a. [PP za [NPmax [QP osam] [NP:GEN(Q) dana]] b. [PP sa [Npmax:iNsT [QP pet] [NP:GEN(Q) devojaka]]]

Either way, percolation into NP is impossible. Recall that percolation is the result of coindexation among members of a projection, so that case percolates down as an automatic and immediate consequence of case assignment. Note that percolation must be an obligatory process, otherwise one might still expect the material in the scope of the QP to be able to receive GEN(Q) in Russian examples such as (12b). A node already assigned case by a more local governor, however, will prevent further percolation. Thus, it is invariably blocked by the D-structure presence of GEN(Q) on NP in Serbo-Croatian. In Russian, on the other hand, the heterogenous/homogenous pattern arises because NP is not assigned case until S-structure and consequently only blocks other less locally assigned structural cases (i.e. nominative and accusative). Notice that this account is not readily compatible with the three-place hierarchical model in (3), where GEN(Q) is ascribed special status. Thus, the Serbo-Croatian facts lead to the strengthening of Babby's idea that Russian may be handled with a two-place hierarchy, as in (4), to the conclusion that it must. The proposal that we are only dealing with a two-way opposition also strongly suggests that the case hierarchy per se is epiphenomenal, deriving from the primitive distinction between

100

PARAMETERS OF SLAVIC MORPHOSYNTAX

inherent and structural case. This has the additional implication that all analyses employing hierarchical strategies for resolving case conflicts may be misguided, in that the hierarchies themselves may follow from more fundamental principles. The opposition in (11) is reflected in other contrasts between Russian and SerboCroatian. It explains, for example, the different forms of demonstratives standing before the QP in the two languages. In Russian, most adjectival modifiers appear in the NOM/ACC form if they precede the quantifier and genitive plural if they follow it. In Serbo-Croatian, on the other hand, all modifiers must be in the genitive regardless of position.13 This is shown in Russian (15) and Serbo-Croatian (16).14 (15) eti pjat' krasivyx devusek prisli/*prislo these(nompl) five beautiful(gen pi) girls(genpl) arrived(pl/n) (16) ovih pet lepih devojaka je these(genpl) five beautiful(gen pi) girls(genpl) AUX(3sg) doslo /?su dosle arrived(nsg) / AUX(3pl) arrived(fpl) This dichotomy clearly follows from the assumption that Russian eti 'these' is nominative because pjat' 'five' is a structural case assigner, but Serbo-Croatian ovih 'these' is genitive because pet 'five' is an inherent case assigner. Of course, for ovih to be assigned GEN(Q) by pet it must be governed by it. Various methods of realizing this configurationally are conceivable. In analyzing Russian, Babby (1987) contends that the determiner is not c-commanded by the quantifier (under the "first branching node" definition) and is therefore unable to receive GEN(Q). He opposes the situation of determiners and other NOM/ACC modifiers in prequantifier position, as in (17), to that of prequantifiers such aspolnyx '(a) full', dobryx '(a) good', dolgix '(a) long', and celyx '(a) whole', as in (18). (17) poslednie sem' let last(nom/acc pi) seven years(gen pi) (18)

a. polnyx sem' let full(gen pi) seven years(gen pi) b. dobryx pjat' butylok good(gen pi) five bottles(gen pi)

Since these are invariably genitive, Babby claims that they must be sisters of the quantifier in order to be c-commanded by it. Unfortunately, this analysis fails to extend to Serbo-Croatian, where the form of modifiers is irrespective of their position within the NP or the semantic class to which they belong. If, as Babby would need to maintain, Serbo-Croatian pet governs ovih in (16), then it would also be reasonable to assume that Russian pjat' also governs eti in (15). Pjat' must therefore be prevented from assigning case to determiners and most adjectival prequantifiers in Russian by some other means. One simple method for accomplishing this, yet allowing GEN(Q) to be assigned to prequantifiers in Serbo-Croatian, comes to mind. Although not compatible with Babby's structural analysis, it relies on his inherent/structural case dichotomy. This

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

101

approach is based on proposals in Corbett (1979), and I shall not review his arguments here. Corbett argues that Serbo-Croatian prequantifiers are actually basegenerated within N', where they receive the case appropriate to government by the quantifier, and subsequently move to the NP-specifier position. Assuming such an analysis not only for Serbo-Croatian but for Russian as well, and adapting it to the adjunction phrase structure, the D-structures would be roughly as in (19) for the quantified phrases in Russian (15) and Serbo-Croatian (16). (19)

a. [NPmax [QP pjat'] [NP et- krasiv- devusVk- ] b. [Npmax [QP P^] [NP:GEN ovih lepih devojaka]

The determiner would adjoin to NP1"3* by a local movement rule that applies more or less obligatorily. Since GEN(Q) is assigned to NP at D-structure in Serbo-Croatian, this case is retained under movement. In Russian, on the other hand, movement of the determiner puts it outside the scope of the genitive of quantification rule by S-structure, where GEN(Q) is assigned. In support of this, note that it is also possible to leave the demonstrative in its D-structure position (cf. e.g Melcuk, 1985, 95-96), depending on the relative scope of the numeral and demonstrative, in which case it must appear in the genitive even in Russian, as in the S-structure representation in (20). (20) [NPmax [QP pjat'] [NP etix krasivyx devu§ek]] prisli/pris'lo five these(genpl) beautiful(gen pi) girls(genpl) arrived(pl/n) Interestingly, this option allows for both agreement possibilities, for reasons to be explained in the next section. Note further that, in contrast to Russian, the demonstrative in Serbo-Croatian is in the case required by the quantifier, even in oblique positions. This is illustrated in (21). (21) a. [PP sa [Npmax:INST UP:GEN ovih] [NPmax [QP pet] [NP:GEN devojaka]]]] b. [PP s [NPmax:INST [AP:INST etimi] [NPma* [QP:,NST PJat'ju] [NP:INST knigami]]]] As before, the demonstrative moves from inside the NP following the numeral, adjoining to NPmax, and thus retains inherent GEN(Q) in Serbo-Croatian in (21a) but, moving before structural GEN(Q) is assigned in Russian, is free to receive instrumental in (21b). The movement approach to prequantifiers nonetheless runs counter to certain current assumptions about movement and therefore raises the following important conceptual questions: 1. What are the positions involved in this movement? 2. What factors motivate it? Chomsky's (1986b) framework assumes that movement is relativized, such that heads only adjoin to heads and phrases to phrases. Therefore, if the prequantifiers indeed adjoin to NPmax, they must themselves be APs. However, since adjunction is only permitted to non-arguments, the movement itself is suspect; but compare Miiller and Sternefeld (1993) for arguments against such a restriction, both universally

102

PARAMETERS OF SLAVIC MORPHOSYNTAX

and in Russian in particular. More problematic is the question of why this movement occurs in the first place. Presumably, movement occurs out of scope considerations: in order for the determiner to c-command the numeral at LF, it must move, and since it can move in the syntax, it does. However, the Principle of Economy of Derivations of Chomsky (1991) requires the shortest derivation possible, hence we must conclude that the determiner could not have been generated in prequantifier position in the first place. I will argue that Slavic determiners, with the exception of the postpositive articles found in Macedonian, Bulgarian, and North Russian dialects, are basically adjectival, hence cannot be generated in determiner position per se, but can only move there to have wide scope. Nonetheless, this idea does not solve the technical problems mentioned above. In section 5.2.1, I shall therefore revise the analysis of the structure of quantified NPs to conform to the system in Abney (1987), in which NPs are embedded in D(eterminer) Phrases. Slavic adjectival determiners will be generated as As and subsequently raise to D position, which is underlyingly empty in all Slavic languages (other than the exceptions noted above). It is also conceivable that D, although phonologically empty, bears features for defmiteness, and it is the presence of [ + DEF] that drives the raising from A to D. Putting these concerns temporarily aside, it is clear that opting for the movement approach leads to the conclusion that the quantifier cannot in fact directly assign case to prequantifier position. This entails the rejection of Babby's configurational explanation of the difference between the behavior of the vast majority of prequantifiers, which are NOM/ACC, and those few that bear GEN(Q). Following Mel'cuk (1985), one might regard these special prequantifiers as "frozen" adverbial forms internal to the quantifier phrase.15 Since they have scope only over the quantifier, I assume an adjoined structure as in (22). (22)

[NP™* [QPm» dobryx [QP pjat']] [NP butylok]] good(gen pi) five bottles(gen pi)

Note that this constituent structure can be used to explain why both dobryx '(a) good' and butylok 'bottles' are assigned GEN(Q)—both are sisters to a QP. The analysis in (22) is of course better motivated semantically than is Babby's, assuming a sisterhood restriction on modification (cf. Koopman and Sportiche, 1988, 1990). Babby (1987, 126-128) in fact actually considers a suggestion by Gil Rappaport that such prequantifiers form a constituent with the numeral, but rejects it since he is unable then to explain why GEN(Q) is assigned and why these forms are plural. The former property follows from our adjunction analysis, assuming sisterhood to QP to be the relevant factor. Since Babby employs the minimal branching node definition of c-command, allowing QP to branch would cause problems for him. There are, moreover, several additional conceptual problems with his approach to GEN(Q) assignment. For one thing, QP can indeed branch to admit adverbial as well as prepositional material (cf. Babby, 1985) with no ill effect on its ability to assign case. For another, Babby's contention that the head Q can assign GEN(Q) outside its maximal projection QP simply because QP does not branch contradicts general properties of case assignment and relies on accidental consequences of assuming the "first branching node" definition. I also do not regard the plurality of dobryx as insurmountable, although it is certainly true that if it does not mod-

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

103

ify the plural head N this feature cannot arise by virtue of agreement with that head.16

4.1.3. Characterizing the Structural/Inherent Dichotomy In the preceding sections I have argued that certain differences between Russian and Serbo-Croatian can best be understood in terms of the status of GEN(Q) in the two languages, as structural in the former and inherent in the latter. This raises the question, however, of how this dichotomy is to be formally expressed within the feature model of case espoused in chapter 2. There it was claimed that syntactic nodes may differ in terms of whether or not they contain case submatrices and, if they do, whether these submatrices have fixed or unspecified feature values. Those with specified case submatrices always assign inherent case, in the sense that their case features must be discharged and this process must take place at D-structure, along with theta-role assignment. This is how lexical case—the quirky case of GB— is formally represented. Underspecified or unspecified feature matrices, on the other hand, behave like structural cases. Nominative and accusative are instances of this type. It was further shown, however, that there is no straightforward correspondence between structural case, defined as case assignment that only takes effect at S-structure (or later; cf. chapter 7), and purely configurational determination of case assignment. In fact, if anything, the opposite of what might be expected is usually true: purely configurational case assignment, such as genitive on sisters of N, dative on sisters of V, and instrumental on sisters of VP, must take place at D-structure and cannot be overridden by the genitive of quantification. The only apparent exception to this generalization was the accusative on time and distance phrases, which appears to be configurational since it applies to sisters of V regardless of the case submatrix of the verb. I therefore suggested that this case too can be understood as assigned whenever the configuration is met, hence at D-structure, but with the accusative feature values left unspecified until S-structure. This system is admittedly unwieldy and, moreover, obscures the generally accepted nature of GB's inherent/structural dichotomy as rooted in theta-theory. Since I have relied on this opposition to characterize the difference between GEN(Q) in Russian and Serbo-Croatian, however, it is necessary to consider ways of instantiating it. Serbo-Croatian GEN(Q) might either be lexically specified or configurational in order to translate into an inherent case. Russian GEN(Q), on the other hand, would have to be underspecified in case features in order to translate into a structural case. Unfortunately, none of these solutions is particularly satisfying. For one thing, GEN(Q) certainly appears to be configurationally rather than lexically determined in both languages (modulo the variation induced by the paucal numerals two, three, and four). For another, since it is sisterhood to a maximal projection QP rather than to its head Q that is relevant, and I know of no other situation in which an XP must be endowed with case assignment features, the proposal that Serbo-Croatian QPs have fully specified GEN(Q) features whereas Russian ones do not would be otherwise unmotivated. An alternative approach to the structural/inherent dichotomy thus seems to be called for. The bulk of the evidence points to the conclusion that it is the particular cases

104

PARAMETERS OF SLAVIC MORPHOSYNTAX

themselves, rather than any difference in the manner in which they are assigned, that determines their status as structural or inherent.17 That is, nominative and accusative always behave as if structural, while genitive, dative, locative, and instrumental always behave as if inherent, and the genitive of quantification is able to pattern either way. Moreover, rules that cause an otherwise nominative subject or accusative object to appear in some oblique case also cause it to make the concomitant switch from structural to inherent. Thus, dative subjects of non-agreeing predicates, as analyzed in chapter 6, behave as if inherently case-marked. The same is true of genitive objects of negated verbs, as discussed in section 5.2.2. Most telling is the fact that when the accusative of animate objects is realized using the genitive form it switches from a structural to an inherent case. This can be deduced from the form of the animate phrases quantified by Russian dva 'two', tri 'three', and cetyre 'four', since these numerals assign the paucal (glossed as genitive singular). Either the animacy of the head noun causes the NP to be genitive plural and the numeral agrees, or the animacy does not percolate up and the numeral governs. This is illustrated in (23).18 (23)

a. ja soscital soldat I(nom) counted soldiers(acc/gen pi) b. ja soscital cetyrex soldat I(nom) counted four(gen) soldiers(acc/gen pi) c. ja soscital cetyre soldata I(nom) counted four(acc) soldiers(gen sg)

Example (23a) shows that the accusative form of animate nouns is identical to the genitive, and (23b) shows that this may also be true of an adjectival quantifier.19 The genitive plural form of the noun thus depends on the fact that it is animate and accusative, rather than on sisterhood to a QP. This can be contrasted to the alternative in (23c), in which the QP fails to reflect animacy but instead determines the form of the noun. What is impossible is for the adjectival quantifier to agree and govern simultaneously: (24)

*ja soscital cetyrex soldata I(nom) counted four(gen) soldiers(gen sg)

This paradigm demonstrates that the quantified noun phrase is either accusative and the QP assigns GEN(Q), or it is genitive and the QP agrees. Even though this genitive is due to the animacy rule and is unrelated to case properties of the governing verb, it behaves just like any other genitive with respect to GEN(Q), namely, as an inherent case. This all suggests that one might most perspicuously locate the structural/inherent property in the case features themselves. If this opposition is really intrinsic to the individual cases, the simplest hypothesis is that it reflects one of the feature oppositions already assumed. A cursory inspection of the features proposed in (77) of chapter 2 reveals that f ± oblique] differentiates the cases precisely as needed—[ — oblique] (or "direct") cases are structural and [ +oblique] ones are inherent. The oblique cases are fully specified at D-structure; the direct cases are not. What then of GEN(Q), the feature content of which I

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

105

have so far left vague? Clearly, the idea that Jakobson's oblique should be equated with GB's inherent, coupled with my analysis of variation in Slavic quantified phrases, implies that GEN(Q) is [-oblique] in Russian and [ +oblique] in SerboCroatian. Notice, however, that changing genitivel to [ — oblique] would render it identical to the accusative. This is an important alternation that plays a role in marking animacy and the genitive of negation. I therefore assume that GEN(Q) is otherwise identical to genitive2 in feature content. This is, I believe, a reasonable move since genitive2 is only used in quantificational contexts; as we shall see in section 5.3.1.1, the genitive2 involves a phonologically null quantifier and induces a partitive or large-quantity interpretation. Indeed, it may be that the existence of a distinct genitive2 correlates to some extent to the existence of a [ — oblique] genitive.20 The following system of genitives thus arises for Russian: (25)

genitivel genitive2 GEN(Q)

= = =

[ + obl, -marg, -nonascr, -phras] [ + obl, -marg, + nonascr, -phras] [ — obi, -marg, + nonascr, —phras]

None of these oppositions exists in Serbo-Croatian. One place where the proposed feature system seems to diverge from the criteria developed in chapter 2 has to do with the status of the feature [±phrasal]. If GEN(Q) is always assigned configurationally on the basis of sisterhood to QP, it should be analyzed as [ +phrasal]. This is of course impossible in the system adopted in chapter 2, since then GEN(Q) would be identical to the nominative. However, in chapter 51 will argue that GEN(Q) is in fact assigned by a head Q to its complement NP. This has the desired result with respect to the feature [ ±phrasal]. In addition, it opens the possibility of different Qs assigning different cases. I will argue, for example, that the distributive preposition po assigns a [ — oblique] dative in Russian and the null quantifier of partitive expressions assigns genitive2. This leaves open the question of what the paucal numerals assign. To my mind, the fundamental issue here is whether the paucal is a case or a number. If it is a case, as I have been assuming, the only realistic option with the current system would be as follows: (26)

paucal

=

[ — obi, +marg, + nonascr, -phras]

This has the advantage of exploiting existing features and placing the paucal appropriately in terms of markedness. Recall that the paucal dve Puskiny 'two (female) Pushkins' is less marked than the regular genitive Puskinoj, since it exhibits a nominal ending (like NOM and ACC female surnames of this pronominal declension type), and not an adjectival one (as genitive and other oblique case forms do). Following the binary number markedness system based on (77) of chapter 2, the paucal would have a markedness value of 6; GEN(Q) in (25) has a markedness value of 2. Both are thus higher in markedness than ACC, and crucially lower in markedness than any of the oblique cases, the least marked of which was genitivel with a value of 8. On the other hand, making the paucal [ + marginal] leads to the obvious question of what Jakobson's [ ± marginal] really means, and what the paucal might share with the other [ +marginal] cases, dative, locative, and instrumental. Although I do not see this as a particularly serious problem, since Jakobson's own

106

PARAMETERS OF SLAVIC MORPHOSYNTAX

definitions of the features strike me as vague enough to admit this kind of flexibility, one might try to sidestep the issue by regarding the paucal as a singular version of GEN(Q). This would accord with the fact that the paucal endings are virtually identical to the genitive singular, whereas the regular GEN(Q) endings are identical to the genitive plural. However, making such a move would lead to the troublesome questions of why paucal modifiers agreeing with paucal nouns are always plural and why the verb, when it agrees, shows up in the plural; see Halle (1990) for a sketch of how this discrepancy between form and agreement might be handled within a dynamic theory of morphology.21 Given these concerns, I leave the problem unresolved, as well as the more general issue of the interaction between number and case features, a complete investigation of which would take us beyond the scope this study.

4.2. The Category of Numeral Phrases I now turn to the issues of the categorial status and distribution of quantified phrases in the two languages under consideration. These questions are intimately related to the agreement of verbs predicated of quantified subjects. I first review the account of Pesetsky (1982), who argues convincingly that an expression such as pjat' zenscin 'five women(gen pi)' in Russian is ambiguous between being either a QP or an NP. I next show how his arguments do not carry over to Serbo-Croatian, where all quantified phrases appear to have the categorial status of NPs. This provides the background for subsequent modification of Pesetsky's model in terms of the theory that the canonical position of subject is the specifier of Verb Phrase (VP) which will serve to rectify several factual inadequacies.

4.2.1. Pesetsky's Analysis of Russian For Pesetsky, the puzzling problem of subject-verb agreement constitutes the core mystery posed by Russian quantificational structures. It is well known that both "syntactic" and "semantic" subject-verb agreement can take place with quantified subjects, as in (27).22 (27)

a. pjat' krasivyx devusek prisli five beautiful(genpl) girls(gen pi) arrived(pl) b. prislo pjat' krasivyx devusek arrived(nsg) five beautiful(genpl) girls(gen pi)

Pesetsky contends that when plural agreement obtains, as (27a), the quantified phrase is a subject NP, but when the default neuter singular form appears, the quantified phrase is actually a QP internal to the verb phrase. The S-structures of (27) would thus roughly be as in (28). (28)

a. [CP [NP:NOM PJat' krasivyx devusek] [VP prisli [NP e]]] b- tcp [NP el fvp prislo [QP pjat' krasivyx devusek]]]

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

107

Assuming that the verb prijti 'to arrive' is unaccusative, the surface subject originates in object position.23 Unaccusative verbs are superficially intransitive verbs whose surface subjects are D-structure objects. More generally, they have a VPinternal argument but lack an external one. If the internal argument is an NP but the verb does not assign case, then movement to subject position ensues so that the NP can receive nominative case in order to circumvent the Case Filter. This renders unaccusative verbs identical at S-structure to ordinary intransitive (i.e. "unergative") verbs. Thus, whenpjat' krasivyx devusek 'five beautiful girls' is an NP and requires case, as in (28a), it must move to subject position, but when it is a QP, which does not bear case, it remains in situ within VP. According to Pesetsky, this explains why the unmarked word order is subject-verb in (28a), but verb-subject in (28b). This dichotomy does not exist in English or, it will be argued, in Slavic outside of East Slavic. In most languages all nominal expressions are maximally NPs (or DPs, in the analysis of chapter 5), hence raising to subject position in search of case is obligatory. Note that this implies that other factors, presumably rooted in principles of functional sentence perspective, induce similar word-order effects in South and West Slavic. This diminishes Pesetsky's reliance on word-order facts to motivate his unaccusativity analysis. Given the relative freedom of Russian word order, unaccusativity clearly cannot be based on the fact that the subject is post-verbal at S-structure. In neutral contexts, even non-quantified unaccusative subjects are more natural after the verb and unergative ones before it. For example, in answer to the question Cto slucilos'? 'What happened?', the following replies are expected: (29) a. prisel Vanja arrived(m) John(nom) b. Vanja umer John(nom) died(m) This correlation between word order and unaccusativity is clearly indirect, however, since Vanja presumably first moves to SPEC-IP position for NOM case in both (29a) and (29b), then scrambles to follow the verb in (29a) for stylistic reasons. Thus, apparent unaccusative behavior could instead be due to concomitant factors such as functional sentence perspective, thematic relations, or existentiality, so that the need to posit unaccusative predicates in Slavic becomes moot.24 One reason why the verb prijti 'to arrive' might be analyzed as unaccusative, however, is that under negation it is possible for the subject to appear in the genitive case, as in (30): (30)

ne prislo ni odnogo celoveka NEG arrived(n) not-even one(gen) person(gen) 'not a single person came'

Unlike the genitive of quantification, the genitive of negation only applies VPinternally, as demonstrated, for example, by Chvany (1975), Pesetsky (1982), and Neidle (1988), and as discussed in chapter 5. It is thus a clear-cut test for unaccusativity in Russian. Pesetsky then argues that QPs can only be underlyingly VP-internal, which he calls "the D-structure [XP, VP] restriction." While this is indeed true for the

108

PARAMETERS OF SLAVIC MORPHOSYNTAX

genitive of negation, as discussed in chapter 5, the facts are hardly conclusive for the numerically quantified QPs I am concerned with here, since, contrary to Pesetsky's claims, speakers do fairly readily accept non-agreeing verbs with quantified subjects of both unergative and transitive verbs. Moreover, as we shall see in section 4.3, his analysis will fail to carry over when the internal subject hypothesis of Koopman and Sportiche (1988, 1990) is adopted. There I will argue that although QPs—like all arguments—are base-generated inside the VP, they may be either underlying subjects in the VP-specifier position or underlying objects. Nevertheless, in this section I adhere to Pesetsky's presentation, since other aspects of his theory will remain relevant to my eventual analysis. He claims that, except for the few verbs that actually select for QPs as external arguments, QPs cannot be true subjects. This (falsely) predicts the following paradigms: (31)

a. dvadcat' samoletov pereleteli/(*)pereletelo granicu twenty planes(gen pi) flew-across(pl)/(n) border(acc) b. neskol'ko studentov procitali/(*)procitalo etu kneigu several students(gen pi) read(pl)/(n) this book(acc) c. v etom restorane obedali/(*)obedalo desjat' celovek in this restaurant ate-lunch(pl)/(n) ten people(gen pi) d. na ulice guljali/(*)guljalo pjat' studentov on street walked(pl)/(n five students(gen pi)

I have placed the asterisks in parentheses since speakers do not actually reject nonagreement in such constructions. However, the alleged impossibility of nonagreement with transitive verbs (31a, b) and unergative verbs (31c, d)25 leads Pesetsky to conclude that QPs cannot be subjects.26 He rules this option out by means of a complex interaction between the Empty Category Principle (ECP) and categorial selection, which Pesetsky claims must be satisfied at Logical Form (LF). Pesetsky's reasoning proceeds roughly as follows. If (most) verbs select for NPs, then Quantifier Raising (QR) is forced whenever the quantified phrase is a QP. Its trace will then be determined to be an NP, thereby satisfying categorial selection at LF. The trace left behind by QR must be properly governed, however, in accordance with the version of the ECP in (32). (32)

Empty Category Principle: a non-pronominal empty category must be either (1) lexically governed or (2) locally bound.

Any empty category inside the verb phrase can fulfill this requirement by being lexically governed by the head V. Crucially, this option is not available for subjects, which are generally not governed by lexical categories. Traces of subjects can satisfy the ECP, however, by being locally bound by a c-commanding operator. The trick then is to render this latter option unavailable to the trace of a QP. Pesetsky accomplishes this by relying on the assumption that the trace has the categorial status of an NP since it originates in a position canonically occupied by an NP. He argues that the categorial mismatch between the QP potential binder and the NP trace inhibits proper government. In sum, the QP must undergo QR since its D-structure

QUANTIFIED STRUCTURES! RUSSIAN VERSUS SERBO-CROATIAN

109

position must canonically be occupied by an NP at LF, but since its trace will be an NP the QP cannot bind it, leading to a violation of the ECP. Note at this juncture that Pesetsky is assuming that all properties of a trace are determined by independently motivated principles of grammar—its index by binding theory, its category by categorial selection, and its very existence by thetatheory's Projection Principle (and/or by Koopman and Sportiche's, 1982, Bijection Principle, which requires a one-to-one relationship between operators and variables). While I agree that all aspects of trace-theory are in fact epiphenomenal, I think his conception of categorial selection is in need of revision. As I see it, the issue is how the process of Canonical Structural Realization (CSR) proceeds. The basic idea, as proposed in Chomsky (1986b), for example, is that explicit subcategorization statements can be eliminated in favor of semantic selection plus some set of principles delimiting the possible categorial realizations of arguments of any given semantic type. According to the theory of CSR, each semantic type of argument, where the semantic types are presumably derivative from Lexical-Conceptual Structure, is canonically realized by means of a particular syntactic category. The question, to my mind, is how the term "canonically" is construed. For Pesetsky, there is a list of phrase types that can realize each semantic type, and crucially QPs cannot realize entity roles, resulting in the analysis above in which QPs undergo obligatory QR at LF. A reasonable alternative, however, is that the CSR of a semantic type is simply its default (or least marked) categorial realization; everything else being equal, an entity will be an NP, a proposition a CP, a location a PP, and so forth.27 This is not an absolute requirement, however, so that QPs and CPs can occupy a position canonically realized by an NP so long as the QP or CP is semantically compatible with the role assigned to that position, and additionally satisfies all relevant syntactic restrictions. Similarly, an NP can appear in a CP or PP position if it is semantically and syntactically viable. For example, an NP can serve as a hidden proposition or question, as shown in (33a) and (34a), respectively. (33) a. I told Bill [NP the answer to your question] b. I told Bill [CP what the answer to your question was] c. I told Bill [CP that the answer to your question was "five"] (34)

a. John asked (Bill) [NP the time] b. John asked (Bill) [CP what the time was] c. *John asked (Bill) [CP that the time was 3 o'clock]

A [ + WH] CP can function as a proposition, as in (33b), but a [ —WH] CP cannot function as a question, as shown by the unacceptability of (34c). Similarly, a CP can serve as an entity or even as a temporal or locative adjunct, as in (35), and, as observed by Larson (1985), an NP can also serve as a time or location adverbial, as in (36). (35) a. Elisabeth always eats [CP what you eat] b. David will go home [CP when you go home] c. I saw Julia [CP where I least expected her to be]

110

(36)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a. Elisabeth will visit you [NP next Thursday] b. David was relaxing [NP someplace warm]

Larson (1985, 595) argues that the bare NP adverbs in (36) are able to receive case "through the lexical properties of their own heads," and thereby satisfy the Case Filter. Whatever the mechanics of this process may be, the crucial fact is that these NPs not only must be semantically appropriate to function as temporal or locative adjuncts, but also must satisfy syntactic criteria. In a similar vein, Pesetsky (1982) observed that case plays a critical role in ruling out examples such as (37b), discussed in Grimshaw (1979). (37)

a. John wondered f cp what the time was] b. *John wondered [NP the time]

Pesetsky's idea was that the examples in (37) contrast with those in (34) in that wonder, unlike ask, is not a case-assigning verb, hence no NP complement is admissible in (37b). In light of these facts, he argued that the CSR of questions should be either NP or CP, and that extraneous factors such as Case Theory may interfere to limit the actual range of categories instantiating questions in any given context. Pesetsky then claimed that the reason why Russian QPs necessarily undergo QR at LF and their traces are obligatorily NPs is that they cannot satisfy categorial selection at LF otherwise, QP not being a CSR of entities. Notice, however, that under this view no primacy is assigned to a [ + WH] CP as realizing a question, a PP as realizing a location (assuming NPs and [ + WH] CPs can also do this), or an NP as realizing an entity (assuming a [ +WH] CP can also do this). Under the alternative approach to CSR suggested above, it is certainly possible for another non-canonical phrase to bear the required semantic role if (1) it has appropriate semantics in order to express the required role indirectly, and (2) it is independently able to satisfy all relevant syntactic conditions. If so, Pesetsky loses any motivation for obligatory QR of QPs and, concomitantly, loses his account of why they cannot appear in subject position. This is actually an advantage, however, since, as I mentioned above in putting the asterisks in (31) in parentheses, QPs can indeed appear in subject position in Russian. Notice, however, that it still might make sense to claim that the traces of QPs must be NPs, since if all properties of traces are determined by independent principles, it may be that the theory of CSR is exactly what is implicated. However, assuming his category-matching requirement on antecedent government (i.e. local binding) as a reasonable restriction, the view that CSR is just "canonical," not absolute, achieves exactly the opposite result than Pesetsky intended. Rather than forcing all QPs in canonical NP positions to undergo QR, it in fact predicts that QPs in positions that are not properly governed (i.e. subject and adjunct QPs) should be unable to raise at LF, for the NP traces resulting from this movement would not meet the ECP. I have intentionally not addressed Pesetsky's semantic arguments for QR, since his predictions are inconsistent with speakers' judgments. Essentially, he claims that QR of a numeral phrase leads to an individuated (as opposed to group) reading. Since Pesetsky's conception of CSR forces QR whenever the numeral phrase is a QP, it should be obligatory in (28b) but not in (28a). If anything, however, (28b)

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

111

strongly favors the group reading and (28a) the individuated one. Interestingly, this is precisely what might be expected if QR of a QP is blocked by virtue of its trace being determined by the principles of CSR to be an NP and if QR induces individuation. Notice, however, that this correlation between category and interpretation holds irrespective of the position of the QP. The group reading obtains whenever the subject is a QP—as indicated by the form of verb—regardless of whether that QP is the "subject" of an unaccusative, intransitive, or transitive verb.There is no asymmetry between lexically governed and ungoverned positions, suggesting that QR of a QP from a canonical NP position must be invariably impossible. This state of affairs could follow if the traditional ECP in (32) were replaced by a more recent version requiring antecedent government at LF (cf. e.g. Chomsky, 1986b; Aoun, Hornstein, Lightfoot, and Weinberg, 1987). For object positions it is impossible to tell whether the numeral phrase is an NP or a QP; there is no morphosyntactic difference, and both group and individuated readings are equally possible. I assume that both NPs and QPs can appear in structural case positions and that, whereas NPs are cased, QPs are not. That is, structural case need not be assigned, since things such as clauses and prepositional phrases, which do not bear case, can be objects of transitive verbs, as in (38), where the object of Russian znaju 'I know' can be realized by a case-marked NP or a caseless clause. (38)

a. ja znaju [NP otvet na vas vopros] 'I know the answer to your question' b. ja znaju, [CP cto net otveta na va§ vopros] 'I know that there is no answer to your question'

On the other hand, only NPs can appear in oblique positions, theta-theory requiring oblique case to be discharged. In keeping with this observation, note that if a clause occurs in an oblique position, it must be embedded in a nominal phrase headed by to 'it', as in(39b). (39)

a. ja dumaju ob [NP:LOc otvete na vas vopros] 'I am thinking about the answer to your question' b. ja dumaju o [NP:Loc tom> tcp &° net otveta na vas vopros]] 'I am thinking about it-LOC that there is no answer to your question'

Since the preposition o(b) 'about' assigns locative, it requires an NP after it to bear this case. Any numeral phrase complement to o(b) must thus be an NP rather than a QP: (40)

ja dumaju o [NP:LOc PJati km'gax] 'I am thinking about five books'

Here, the locative numeral pjati modifies and agrees with the locative head noun knigax. One problem posed by the QP-hypothesis that Pesetsky does not address is that of the internal structure of quantified phrases; he simply represents the two possibilities as in (41).

112

(41)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a.

NP

Q

b. N

QP

Q

N

This inexplicitness leaves unexplained just how GEN(Q) is assigned and, more importantly, raises the question of why quantified NPs and QPs exhibit identical internal case properties. Indeed, the observation that GEN(Q) is assigned in both is a compelling reason for rejecting the NP/QP dichotomy, as Babby does. There are, good arguments that Russian countenances two kinds of quantified phrases, however, those that are headed by a noun and are fundamentally NPs, and those that are headed by a quantifier and are fundamentally QPs. There is a host of factors distinguishing these two as subjects, including that NP subjects, but crucially not QP subjects, (1) induce plural subject-verb agreement, (2) control infinitives, (3) control gerunds, (4) antecede reflexives, and (5) cannot long-distance move. Examples of these contrasts are given in section 4.3.2; reasons for these contrasts are complex and may have as much to do with the position of NP versus QP "subjects" as with their respective categories. It is nonetheless clear that an analysis as in (41) is suggestive at best, since it avoids the question of the details of the internal structure of these phrases. In particular, it does not respect the principles of X-bar syntax, which require that each head X project up a phrasal maximal projection XP. Thus, in (36a) the noun rublej should project an NP, and in (41b) the numeral pjat' should project a QP. A system with just these properties will be developed in chapter 5. Under the adjunction approach I have assumed in this chapter, however, quantified NPs and QPs can be more precisely represented as in (42). (42)

a.

QP

NPmax

NP:GEN(Q)

b. QPmax

QP

NP:GEN(Q)

This scheme accurately accounts for the fact that both display the same case properties and, in addition, allows for the material following the quantifier to constitute a full NP. Recall that this was necessary in order to accommodate the behavior of demonstratives. Note, furthermore, that all we need assume to explain why subjectverb agreement is obligatory in Russian (15) is that the nominative demonstrative can adjoin only to NP, not QP. I therefore assume that the demonstrative is required to adjoin to NPmax rather than to QPmax. This follows if, as a local movement, the demonstrative cannot move outside the maximal projection that immediately dominates it, that is, the phrase in which it is generated. When, on the other hand, it remains in situ as in Russian (20), both NPmax and QPmax are possible. That is, given

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

113

the existence of the string "QP NP," these can conceivably be combined in two distinct ways: either QP is adjoined to NP, or NP is adjoined to QP. Before turning to Serbo-Croatian, let us consider the mechanism by which the verb appears in the third neuter singular with QP "subjects." One possibility is that the agreement features of the verb, whether intrinsic to V or mediated through an abstract AGR node, are simply filled in as third person neuter singular in the absence of a nominative subject with pronominal features. However, since QP "subjects" are hi fact VP-internal, subject position must actually be occupied by a null expletive element.28 This raises the alternative possibility that in Russian the null expletive pronoun itself bears third neuter singular features, either inherently or filled in as such by default. Therefore, given an S-structure such as (28b), the verb may actually be agreeing with the empty NP subject. If so, the neuter third person singular is technically not a non-agreeing form but rather the result of syntactic agreement with an empty subject. Under this approach, verbs may only agree with NP subjects and the third singular neuter is simply the verb form one always finds in Russian with empty expletive subjects. 4.2.2. Extending the Analysis: Serbo-Croatian Numeral Phrases Are NPs In this section I address the import of Serbo-Croatian for Pesetsky's account of Russian. The fact that numeral phrases in Serbo-Croatian may appear in oblique NP positions and still exhibit GEN(Q) case-marking, as discussed in section 4.1.2, indicates that they must themselves be NPs in this language. As such, they should potentially be able to bear any oblique case, although this need not be reflected morphologically, since percolation of the oblique case down from NPmax is blocked by GEN(Q), which I have argued is an inherent case in Serbo-Croatian. The only necessary modification to Babby's system is to allow both the inherent oblique case and GEN(Q) to be satisfied in the same structure, the one on NPmax, the other on NP. Although both are [ +oblique], they do not conflict since all government requirements are met simultaneously. I thus reject the QP hypothesis for SerboCroatian. What significance does the claim that numeral phrases are invariably NPs in Serbo-Croatian have for subject-verb agreement? It turns out that in this language both agreement patterns are in principle acceptable, as shown by the following examples: (43)

a. dvadeset "migova" preslo je/ twenty MIGs(genpl) crossed(nsg) AUX(3 sg)/ ?pre§li su granicu crossed(mpl) AUX(3pl) border(acc) b. 70 miliona 70 million ?su AUX(3pl)

lica je napustilo/ people(gen) AUX(3sg) left(n sg)/ napustili ovaj kontinent left(mpl) this continent(acc)

114

PARAMETERS OF SLAVIC MORPHOSYNTAX

c. nekoliko ljudi je kupilo/ several people(gen pi) AUX(3sg) bought(n sg)/ ?su kupili imanja u Tetovu AUX(3 pi) bought(m pi) homes(acc pi) in Tetovo(loc) The neuter singular is considered standard and far preferred by most speakers, however, with the plural option having the status of a "performance error." I therefore consider the singular with the non-paucal numerals to reflect true syntactic agreement with the subject NP and the plural to instantiate "semantic agreement" with the cardinality of the head N. For clarity of presentation, however, I abstract away from the marginal status of the plural in subsequent presentation of SerboCroatian examples.29 In light of these Serbo-Croatian facts, recall that in Russian the third person singular neuter form is in fact possible in comparable constructions, which was taken to reflect the failure of syntactic agreement with a QP "subject," which is really VP-internal, and actual agreement with a null expletive NP subject. Numeral phrases that induce plural agreement, on the other hand, were analyzed as simple plural NP subjects. From the Russian perspective, the surprising thing is that the neuter singular form is the norm in Serbo-Croatian. If numeral phrases in SerboCroatian differ from those in Russian in that they are able to bear case and are consequently always categorially NPs, then we must conclude that the neuter singular verb form in Serbo-Croatian actually represents agreement with a quantified NP subject. However, the claim that the "agreeing" form is different in these two languages raises the interesting problem of why quantified NPs behave as if they were neuter singular in Serbo-Croatian but plural in Russian. A possible solution to this problem can be found in the mechanics of number percolation and its interaction with the structural/inherent case dichotomy assumed above. Recall that case is assigned to NP and percolates downward throughout the N projection. Pronominal features, however, are lexical properties of heads, and must therefore percolate upward. Ordinarily, nothing prevents percolation of pronominal features all the way up to NPmax, but something appears to be inhibiting this percolation in quantified NPs in Serbo-Croatian, although not in Russian. Now, the essential difference between these categories in the two languages is in the level at which GEN(Q) is assigned. Recall the D-structure contrast between Russian (44a) and Serbo-Croatian (44b). (44)

a. [NPmax pjat' [NP krasiv- devus/k- ]] b. [NPraax pet [NP:GEN lepih devojaka]]

Presumably, the fact that NP is GEN(Q) in Serbo-Croatian inhibits percolation of pronominal features up to NPmax. Even though the phrase pet lepih devojaka 'five beautiful girls' in (44b) is semantically plural, upward percolation of the plural feature from the head N to NPmax is blocked by the oblique status of NP. Although [ + plural] can percolate up to NP, since both NP and N are the same case, it cannot proceed up to NPmax, since this is not genitive as well. Consequently, the pronominal features of NPmax in Serbo-Croatian (44b) are set as neuter singular in the absence of any further specification. Here this constitutes the truly default option, in that the NP must bear pronominal features, and these must be endowed with some

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

115

value. In Russian (44a), on the other hand, no such case discrepancy arises, enabling the plural feature to travel up the entire projection. GEN(Q) is not assigned until S-structure, so it does not block percolation, which is induced at D-structure by virtue of all members of the projection bearing the same index. The plural option in Serbo-Croatian is thus a marked variant, in which the verb appears to exhibit semantic agreement. This might be understood as agreement with the head N, rather than with its maximal projection, so that verb is able to show number (and gender) features of the subject noun. There is one obvious problem with my claim that Serbo-Croatian numerically quantified NP subjects are, by default, endowed with neuter singular feature values. When ordinary neuter NP subjects conjoin in Serbo-Croatian, the result is a plural NP, as in (45). (45)

tele i dete su skakali calf(n sg) and child(n sg) AUX(3pl) jumped(m pi) 'the calf and the child jumped'

The conjunction of numerically quantified subjects, however, remains neuter singular.30 (46)

pet devojaka i nekoliko momaka je skakalo five girls(gen pi) and several boys(gen pi) AUX(3sg) jumped(n sg) 'five girls and several boys jumped'

Note that the conjunction of two neuter singular nouns in (45) results in masculine plural agreement, the expected neuter plural verb form being skakala. The process whereby pronominal features are passed up the tree within a projection must be complex to cause the sum of two neuters to be a masculine. Although no similar problem exists for the conjunction of two feminines or masculines, resolution rule complexities also arise when the conjoined NPs do not match in gender; see Corbett (1983a) for discussion of these and related issues. With respect to (46), then, it is clear that pet devojaka cannot be neuter in the same sense tele is, since the conjunction of the two numeral phrases appears to be receiving its gender/number specification by the same default mechanism its constituents do, rather than somehow deriving its pronominal features from those of the conjuncts. The pronominal features of the numeral phrases are somehow inaccessible to their dominating NP. Since these features are not intrinsic to the numeral phrases, unlike true neuter NPs, it is reasonable to suppose that they are all fixed as neuter singular at once, preventing from applying whatever percolation process produces a masculine plural from neuter singulars and similar resolution rules. 4.2.3. Some Semantic Issues Much of Pesetsky's argumentation that QPs undergo obligatory Quantifier Raising is semantic in nature. I have so far ignored this aspect of his analysis because certain of his predictions are inconsistent with many speakers' judgments. He claims that applying QR in LF leads to an individuated reading, while leaving the numeral phrase in its S-structure position results in a group reading. This analysis, which

116

PARAMETERS OF SLAVIC MORPHOSYNTAX

Pesetsky attributes to Higginbotham, is based on the idea that a numeral phrase will have a distributive interpretation if it raises at LF and a collective one if it does not. Consider in this light the structures in (28), repeated as (47). (47)

a. [CP [Np:NOM PJat' krasivyx devusek] [VP prisli [NP e]]] b- [CP [NP e] [VP prislo [QP pjat' krasivyx devustek]]]

According to Pesetsky, the QP in (47b) must undergo QR so that its trace can satisfy categorial selection at LF, whereas the NP in (47a) is free to remain in situ. Consequently, (47b) should differ from (47a) in only admitting the individuated reading 'beautiful girl #1 arrived, beautiful girl #2 arrived, . . .'. If anything, however, the opposite is true, in that (47b) favors a group reading and (47a) an individuated one. It is therefore difficult to accept Pesetsky's arguments that QPs undergo obligatory QR and this process necessarily induces an individuated interpretation. It is nonetheless possible to support the categorial contrast between Russian and Serbo-Croatian numeral phrases without taking a definitive stand on the precise interpretive ramifications of the NP/QP dichotomy. Essentially, if Serbo-Croatian numeral phrases are only NPs, and the two forms of the verb therefore do not reflect any inherent properties of the numeral phrase, then there should be no corresponding semantic contrast comparable to that found in Russian. This is borne out by (48), where both group and individuated readings are possible with either agreement pattern. (48)

a. [CP [NP pet ljudi] [VP su dosli na miting]] five people(gen pi) AUX(3pl) arrived(m pi) at meeting b. [CP [NP pet ljudi] [VP je doslo na miting]] five people(gen pi) AUX(3sg) arrived(n sg) at meeting

Although in Russian, contra Pesetsky, the group reading generally correlates with the neuter/third person singular predicate,31 this is not true of Serbo-Croatian. My account, which divorces the range of possible interpretations from the form of the verb in Serbo-Croatian, correctly predicts that whether a quantified NP in this language exhibits a group or individuated reading will not depend on the form of the verb. In other words, either semantic or syntactic agreement is possible in (48), and this is formally independent of whether the subject NP pet ljudi undergoes QR or not. Thus, whatever the correct analysis of the Russian semantics, the approach I have adopted makes no prediction about any correlation between these two readings and verbal morphology for Serbo-Croatian. Another interesting result of the proposed categorial contrast between Russian and Serbo-Croatian has to do with predicates that require group subjects. As Pesetsky (1982, 84-85) observes, a Russian verb such as rasstat'sja 'to disperse' can appear in the plural only with a quantified subject, as in (49). (49)

a.

[CP [NP pjat'zenscin] [VP rasstalis' [NP£] na mostu]] five women(gen pi) dispersed(pl) on bridge

b- *tcp [NP e] [VP rasstalos' [QP pjat' zenscin] na mostu]] dispersed(n sg) five women(gen pi) on bridge

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

117

Pesetsky explains this by arguing that since the QP must undergo QR, only an individuated reading is possible. Since individuation is incompatible with the semantics of this predicate, however, only the NP numeral phrase in (49a) is viable. Although this contrast does indeed follow from Pesetsky's analysis, it is worth noting that with large numerals the felicity of the neuter form improves, so that (50) is perfectly acceptable. (50)

§est' tysjac agitatorov raz"exalos' po raznym derevnjam six thousand agitators(gen pi) dispersed(n) to various villages

This fact implies that imposing obligatory QR on QPs is not the appropriate mechanism to explain the data in (49) and that some alternative solution should be sought. This is especially true since Pesetsky's contentions about the semantics of QPs are extremely controversial, as is his analysis of obligatory QR as inducing a distributed reading. If anything, the problem with (49b) seems to have to do with the predicate requiring a semantically composite NP subject that can be broken down into its components (i.e. individuals). This suggests that the opposite problem than the one for which Pesetsky argues might in fact be at issue—if the QP cannot undergo QR and this is necessary for the compositionality of the numeral phrase to be accessible to the predicate, then QP "subjects" should not be admissible for such verbs. After all, for any group to disperse, each of its members must be understood as separating from the others, rather than the group as a seamless whole. It thus seems to me that Pesetsky's claim that rasstat'sja requires that its subject be a "group" is backward. This verb really means that each individual in some non-singleton set went off in a different direction. It thus requires that it be possible to "look inside" of the semantically plural subject in order to make a statement about each of its constituents. Be that as it may, the point remains that no such contrast arises in Serbo-Croatian, supporting my view that the neuter form in this language does not indicate that the numeral phrase is a QP. In the identical Serbo-Croatian construction, therefore, both variants are acceptable (with the neuter singular form preferred, as always). (51)

a. [CP [NP pet zena] [VP su se razisle]] five women(gen pi) AUX(3pl) REFL dispersed(f pi) b. [CP [NP pet zena] [VP se razislo]]32 five women(gen pi) REFL dispersed(n sg)

Both agreement patterns are viable because they both reflect a semantically plural NP subject, which, I have argued, is grammatically neuter singular. The agreement of the verb, as before, has no effect on interpretation comparable to that found in Russian, and the correct reading obtains regardless of the form of the verb. Since the quantified phrase subject is always a nominative NP in Serbo-Croatian, both semantic agreement, as in (51a), and syntactic agreement, as in (51b), are possible.

118

PARAMETERS OF SLAVIC MORPHOSYNTAX

4.3. The Distribution of Numeral Phrases In this section, I consider what import recent proposals about the underlying position of subjects might have for the analysis of QPs in Pesetsky (1982). Adopting the hypothesis that subjects are canonically D-structure VP-specifiers that move to IPspecifier position in search of case, I will argue that QPs are exempt from this requirement, hence remain in their underlying positions. This set of assumptions can be used to account for a range of seemingly unrelated facts about the behavior of numeral phrases in the Slavic languages. Recall that Pesetsky's arguments for obligatory QR were largely based on the putative impossibility of VP-external QPs. It was thus crucial that he reject nonagreeing transitive and unergative verbs, although these are in fact acceptable in Russian. Interestingly, when the analysis is recast in terms of the idea that subjects invariably originate internal to VP, the possibility of true QP subjects reemerges. Moreover, whether a quantified subject is an NP or a QP turns out to correspond to a host of formal differences that follow directly from this new model of phrase structure, explaining several distinctions between NP and QP subjects within Russian and accounting for their absence in Serbo-Croatian. 4.3.1. The Internal Subject Hypothesis A number of linguists, including Koopman and Sportiche (1988, 1990) and Kuroda (1988), have recently argued that the canonical underlying position of subjects is internal to the VP. Here, for the sake of explicitness, I adopt the proposal of Kuroda (1988) that at D-structure the subject is the specifier of VP, and at S-structure it is the specifier of IP; see Huang (1993) for discussion. These two positions are designated NP* and NPA, respectively, in (52).

IP

(52)

r

NPA e

ICNFL)

VP

NP*

V'

SUBJECT

V

Following Koopman and Sportiche, I shall refer to this analysis, in which I(NFL) is treated as a raising category, as the Internal Subject Hypothesis (ISH). In many

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

119

languages, including English and the Slavic ones under consideration here, the subject undergoes NP movement from NP* to NP" position in order to receive nominative case at S-structure. This kind of approach raises an interesting dilemma: if quantified "subjects" may be QPs in Russian, then they do not need case and hence nothing prevents them from remaining in NP* position. 4.3.2. The NP/QP Dichotomy Revisited With this idea in mind, let us return to the Slavic agreement facts and see how they might follow. The ISH has little effect on the analysis of Serbo-Croatian. Quantified phrases are always NPs, hence they must always undergo NP movement from NP* to NP" position. Whether or not they subsequently undergo QR is not dictated by any principles of UG, beyond those deriving the intended reading in accordance with the semantic requirements of the predicate. In Russian, on the other hand, the situation is not so straightforward. First of all, it seems that nothing in principle prevents a QP either from occupying the VP-specifier position at D-structure or from remaining in that position at S-structure. Indeed, since QPs lack case features, raising to the VP-specifier position should never be motivated by case considerations. The result is that in Russian the S-structure possibilities are more explicitly represented as in (53) and (54). (53)) a.

IP

b.

NPA e

NPA

r I

IP

VP

QP* pjat' Celovek

pjat' Celoveki

V V—o....

r I

VP

NP*

«i

V V—i....

120

(54)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a.

IP

b.

NPA e

NPA

r I

VP

NP*

e

IP

pjaf Celoveki

V

V—o

r I

VP

V

NP*

QP s\ pjat' celovek

e

V—i

NP

ej

Structure (53a) contains a QP with an unergative verb and (53b) an NP with an unergative verb, while structure (54a) contains a QP with an unaccusative verb and (54b) an NP with an unaccusative verb. When NP* is occupied by the plural NP/y'af' celovek 'five people', the verb takes the (past tense) plural ending -i, and when NP' contains a null expletive the verb takes the neuter singular ending -o. Both variants thus reflect true subject-verb agreement. Notice, however, that the possibility of (53a) does not depend on the transitivity of the verb, and in particular one ought to encounter QP subjects so long as they are not actually in NP' position. Indeed, as pointed out earlier in connection with the examples in (31), non-agreement is in fact acceptable even with unergative and transitive verbs. They will thus have the schematic structure in (53a), with a QP appearing and remaining as a VP-specifier, and the verb agreeing with the expletive subject NP. Whether Pesetsky's analysis of Russian should carry over to this model is unclear. For one thing, as discussed above, Pesetsky's conception of categorial selection as forcing QR requires several peculiar assumptions, such as that categorial selection need not be satisfied until LF and that the Canonical Structural Realization of a semantic type is an absolute specification. Indeed, I know of no subsequent work that relies on the complex theoretical apparatus assumed in Pesetsky (1982). For another, the motivating data themselves are extremely indecisive, with the status of non-agreement with quantified subjects by transitive and unergative verbs ranging from preferred to marginal, depending on the semantics of the VP and the choice of quantifier. Nonetheless, the facts presented in this chapter argue that Russian has both QPs and NPs, whereas Serbo-Croatian has only NPs, and that whereas QPs are S-structure VP-specifiers, NPs are S-structure IP-specifiers. Pesetsky's account of why QPs cannot appear as IP-specifiers is of course possible, but it seems simpler just to maintain that they do not raise because they cannot bear case.33 Let us turn now to some further discrepancies between Russian and Serbo-

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

121

Croatian that support this analysis. I have argued that QPs do not induce agreement because they are not IP-specifiers. They also fail to bind reflexives and control gerunds, two other important subject-oriented diagnostics (cf. Pesetsky, 1982; Neidle, 1988). Consider the following examples: (55)

a. pjat' zenSftn smotreli/smotrelo na Ivana five women looked(pl/n) at Ivan b. pjat' zenScin smotreli/?*smotrelo na sebja five women looked(pl/n) at self

(56)

a. po doroge domoj, pjat' mal'c'ikov zasli/zas'lo v magazin on way home, five boys dropped-in(pl/n) to store b. vozvras'c'ajas' domoj, pjat' mal'cikov za§li/*zaslo v magazin returning home, five boys dropped-in(pl/n) to store

The presence of the reflexive pronoun in (55b) or the gerund clause in (56b) forces plural agreement. Otherwise, both options are viable. I conclude that in (55a) and (56a) the quantified phrases are either NPs in NP* position or QPs in NP* position, but in (55b) and (56b) they can only be NPs in NPA position. This follows under the assumption that only IP-specifiers can bind reflexives or control gerunds in Russian.34 Interestingly, Serbo-Croatian displays none of these contrasts, as illustrated in (57) and (58). (57)

pet zena je kupilo/ su kupile five women AUX(3sg) bought(n sg)/AUX(3pl) bought(f pi) ovu knjigu za sebe this book(acc) for themselves

(58)

pet zena je to diskutovalo/ su to five women AUX(3sg) that(acc) discussed(n sg)/AUX(3pl) that diskutovale, iduci kudi discussed(f pi) going home

Either agreement option is possible, with the plural option, as always, somewhat marginal, despite the presence of the reflexive pronoun in (57) or the gerundive clause in (58). The failure of forced subject orientation to have any impact on agreement demonstrates that agreement is not a function of the status of the quantified phrase in Serbo-Croatian. Unlike Russian, it is always an NP and as such must invariably raise to IP-specifier position. Obligatory control constitutes another potential diagnostic, and, indeed, in unequivocal structures of obligatory control, as defined in Williams (1980) and Franks and Hornstein (1992), for example, the plural is required. This is shown in Russian (59). (59)

pjat' zenscin staralis'/*staralos' kupit' etu knigu five women(gen pi) tried(pl/*n) to-buy this book(acc)

As before, this restriction does not hold for Serbo-Croatian.

122

PARAMETERS OF SLAVIC MORPHOSYNTAX

It should be pointed out, however, that these facts do not necessarily demonstrate that what is crucial is the position of the quantified phrase, rather than its category (cf. also n. 34). One potential test that might distinguish between these two possibilities would be to see whether the effect in (38b) disappears when the Russian reciprocal drug druga 'each other' is considered, since the reciprocal, unlike the reflexive, is not subject-oriented. Interestingly, the same result obtains, as shown in (60), suggesting that categorial mismatch may indeed be relevant. (60)

pjat' studentov pomogali/*pomogalo drug drugu na ekzamene five students(gen pi) helped(pl/n) each other(dat) on exam

Whether this consitutes the sole factor or not is unclear, and I leave the issue open.35 An alternative explanation for (60) might be to maintain that the reciprocal requires its antecedent to raise at LF. Then, just as with reciprocal verbs such as rasstat'sja 'to disperse', selecting the QP in (60) would lead to a violation of the antecedent government requirement on traces at LF, assuming as Pesetsky did that CSR determines the trace to be an NP and that antecedent government requires categorial nondistinctness. Consider one final consequence of the NP/QP dichotomy. According to Koopman and Sportiche, movement from NPA should show ECP effects and movement from NP* should not, since the VP-specifier position is lexically governed by V. Hence, there should be a contrast in long-distance movement, corresponding to the agreement morphology on the verb. This prediction appears to be borne out (for those speakers of Russian who show ECP effects at all): (61)

skol'ko celovek; [Ivan dumaet [cto [e; procitalo/ how-many people(gen pi) Ivan thinks that read(n)/ ?*procitali etu knigu]]] read(pl) this book(acc)

The conclusion once again is that when skol'ko celovek 'how many people' is a QP it is a VP-specifier, and when it is an NP it is an IP-specifier. Notice, however, that this fact runs counter to my suggestion in sections 4.2.1 and 4.2.3 that perhaps Pesetsky's analysis should be turned on its head to disallow movement of QPs at LF. Although the semantics of QR and the NP/QP dichotomy are somewhat tangential to my main claims in this chapter, let me briefly explore the implications of the ISH for this possibility. Assume that Pesetsky is correct in letting the principles of Canonical Structural Realization determine the syntactic category of null elements and that antecedent government is indeed blocked by categorial mismatch. If so, any QP moved from a canonical NP position will be unable to antecedent-govern its trace, hence subject QPs should never undergo QR. However, if these are actually in VP-specifier position, their traces will be lexically governed by the verb and thereby satisfy the ECP, as was demonstrated by overt wh-movement of a QP subject in (61). On the other hand, many versions of the ECP since Chomsky (1981a) require antecedent government of all traces, either exclusively, as in Chomsky (1986a), or in addition to lexical government, as in the system of Aoun, Hornstein, Lightfoot, and Weinberg (1987) to be adopted in chapter 6. Under this

QUANTIFIED STRUCTURES! RUSSIAN VERSUS SERBO-CROATIAN

123

approach, empty categories must be antecendent-governed at LF and lexically governed at PP. If so, and assuming a categorial matching requirement, then no QP should be able to move from any canonical NP position. Since this would leave the facts in (61) unexplained, however, it seems to me that the feature-matching and antecedent government requirements are incompatible.36 The basic problem with trying to determine whether QPs and quantified NPs raise at LF is the uncertainty about just what semantic implications QR should have. According to Pesetsky, QR should impose an individuated reading. As observed in connection with (28), there is indeed a contrast along these lines: (62)

a. [QP pjat' krasivyx devusek] prislo na vecerinku five beautiful women came(n) to party b. [NP pjat' krasivyx devusek] prisli na vecerinku five beautiful women came(pl) to party

Contra Pesetsky, (62a) strongly favors a group reading, while (62b) is more ambiguous. Thus, if QR correlates with individuation, the QP ought not to be expected to undergo it, whereas the NP may but need not. It seems to me, however, that alternative and more standard interpretations of QR as resulting in wide scope (and existential quantification) lead to the opposite conclusion. That is, the QP subject in (63a) may have wide or narrow scope with respect to the quantifier in the matrix clause, while with the NP subject in (63b) the narrow scope reading is greatly preferred.37 (63)

a. kazdyj every b. kazdyj every

rezisser director rezisser director

dumaet, thinks dumaet, thinks

cto [QP pjat' that five cto [QP pjat' that five

akterov] actors akterov] actors

ego uvazaet him respect(3sg) ego uvazajut him respect(3pl)

This fact, coupled with the demonstration in (61) that long-distance syntactic movement of a QP subject circumvents the ECP effect displayed when an NP subject is similarly moved, suggests that the proper interpretation of QR is one of scope. The QP in (63a) is free to raise at LF since it is in VP-specifier position, whereas the NP in (63b) may only do so in violation of the ECP since it is in IP-specifier position. If this conclusion is correct, some factor other than QR must be involved in the group/ individuated contrast exhibited by (62) that presumably correlates with whether the numeral phrase subject is in VP-specifier or IP-specifier position.38 The conclusion that QR applies with equal force to QPs and NPs, but only the latter are constrained by the ECP, predicts that quantified NPs in simple sentences should allow for either interpretation, since local movement does not create ECP violations. This is difficult to test, however, since there is no independent way of establishing a NP/QP contrast in object position. The interaction of sentential negation with NP/QP subjects might provide an answer, but here a surprising result obtains. Consider the examples in (64). (64)

a. esce ne dopisalo ekzamen [QP mnogo/pjat' studentov] still NEG finished-writing(n) exam many/five students

124

PARAMETERS OF SLAVIC MORPHOSYNTAX

b. e§ce ne dopisali ekzamen [NP mnogo/pjat' studentov] still NEG finished-writing(pl) exam many/five students The QP in (64a) is ambiguous between the readings 'there are many/five students who have still not finished the exam' and 'it is not the case that many/five students have finished the exam', whereas the NP in (64b) strongly favors the wide scope reading. This fact perhaps follows from my analysis of QPs as in VP-specifier position, assuming that the negation element c-commands the entire VP but is itself internal to IP. At S-structure the QP subject will thus be inside the scope of negation, whereas the NP subject will be outside its scope. Raising the QP subject at LF introduces the wide scope reading, but no similar effect obtains for the NP subject (which is already outside the negation). For these reasons, I conclude that both NPs and QPs may but need not undergo QR, with the only restriction being that longdistance movement (both in the syntax and LF) of NP subjects only is ruled out by

the ECP. Notes 1. Versions of much of the material in this and the following chapter have been presented at various colloquia and meetings, including AATSEEL 1988 in Washington, D.C., the Colloquium on Recent Advances in Russian Linguistics in 1989, and the Workshop on Functional Categories in Slavic Syntax in 1992, both in Ann Arbor, the Mid-West Slavic Conference in 1990 in Urbana, and AATSEEL 1991 in San Francisco. Major portions have also been published as Franks (1990d, 1994). 2. By "numeric quantification" I intend to restrict the focus of inquiry to those quantifiers that have some sort of numeric value, including Russian neskol' ko 'a few' andmnogo 'many', and that participate in the kinds of alternations described in the text. Logical quantifiers that are always adjectival, such as Russian kazdyj 'each' and vse 'all', are excluded from the discussion. 3. See Lasnik (1992a) for the idea that partitive (by which he means the "genitive of negation") may be structural in Russian but inherent in Polish. There are problems, however, with his data, which suggest that the conclusion that Russian and Polish differ in this respect may be misguided or, at least, that the status of the genitive of negation varies among Polish speakers, some treating it like a structural case and others like an inherent one. The status of the genitive of negation will be taken up in section 5.3.2. 4. I defer until section 4.2 any discussion of whether these phrases are indeed NPs. 5. He argues for this partially on the grounds that the adnominal genitive is also structural. This characterization seems appropriate in the sense that it is optional and not lexically driven. There are, however, several serious discrepancies between the structural accusative assigned by [ - N] categories and the adnominal genitive that make it impossible to sustain this claim. For one thing, virtually any noun can take a genitive object, whereas only certain verbs and prepositions can take an accusative one. The adnominal genitive thus patterns like the adjunct instrumental, in that it is inherent on NPs in a particular configuration. More importantly, it takes precedence over GEN(Q), just as any other inherent case would. 6. It is unclear to me why Babby (1986,1987) employs the hierarchical account rather than the theta-theoretic one. As observed by Freidin and Babby (1984, 87), working with a threeplace hierarchy as in (3), the "precedence of lexical case over other case types follows from the Principle of Lexical Satisfaction." Babby's insight that GEN(Q) is actually structural in Russian should therefore have led him to reject the Syntactic Case Hierarchy altogether.

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

125

7. In chapter 7 and Franks (1990c), I argue that structural case can also be assigned at LF, and, in particular, that whether nominative is assigned at S-structure or at either S-structure or LF constitutes an aspect of parametric variation giving rise to certain null-subject phenomena. 8. I use the term "paucal" rather than "genitive" here for several reasons. The form ova is not the regular genitive, which is ovog(a). Ova resembles an indefinite (short form) genitive adjective, except that (i) demonstrative (and other) adjectives lack such a form in other contexts and, as Wayles Browne (personal communication) points out, (ii) some adjectives distinguish the paucal form from the indefinite genitive singular: dva cfna ovna 'two black rams', with long rising, is indefinite and dva cfna ovna, with long falling, is definite. Although most reference grammars simply state that the paucal form is genitive singular, there is clearly no consistent analysis of the morphology of adjectives and nouns in the scope of these numerals. The confusion lies in the fact that masculine and neuter As occur in what may be either regarded as the neuter nominative plural or genitive singular, and masculine and neuter Ns unambiguously occur in the genitive singular, whereas feminine As and Ns are in the nominative plural. However, one must rely on subtle and frequently absent accentual and/ or length distinctions to tell that the neuter nouns are genitive singular and the feminine ones are nominative plural. The issue of subject-verb agreement, discussed in section 4.2, also bears heavily on the resolution of this matter. There are at least two schools of thought about how to treat this problem: either the paucal numerals assign nominative plural but an exceptional morphological readjustment rule, as in Corbett (1983a, 89-91) or Halle (1990), changes masculine As to neuter plural and masculine Ns to genitive singular, or the paucal numerals govern a special paucal form. Following Browne's suggestion, I shall assume the latter approach and tentatively gloss these endings as "pauc," focusing on the less idiosyncratic numerals 'five' and above instead. I will also refer to paucal elements as GEN(Q) throughout the text in order to facilitate the discussion. Note that localizing these idiosyncracies in the case-governing properties of the paucal numerals is consistent with the analysis adopted in chapter 5 that Q takes (APs and) NPs as complements; if the Serbo-Croatian paucal numerals are heads, then one can claim that [ + PAUC, +Q] heads take paucal complements. 9. Except for a minor accentual opposition in some dialects, these two cases are syncretic in all forms in Serbo-Croatian. 10. Leko (1987) also notes some curious asymmetries between complements to Vs and Ns. For example, although the noun komandovanje 'commanding' can take a bare overtly instrumental paucal NP (e.g. trima bataljonima), the verb komandovati cannot. This may have to do with the highly restricted nature of declined paucal forms in the modern language; compare Leko (1987) for details and examples. 11. This seems to be one place where the dative and locative must be distinguished. According to one speaker, (6d) with tri or pet replacing oba would be equally acceptable; (lOb) with tri or oba replacing per is still marginal, although possibly slightly less so with oba. 12. According to Babby, such conflicts between lexical case assignors are theoretically impossible, although I see no reason why his Principle of Lexical Satisfaction cannot be interpreted as proposed in the text. 13. This is not technically correct, given that in Serbo-Croatian the paucal numerals may also decline in oblique positions in more literary styles (although this is quite rare in the modern language) and that oba 'both', dva 'two', tri 'three', and cetyre 'four' induce a special paucal ending. Even in Russian, the form required by these numerals is not always identical to the genitive singular. Moreover, in Russian, adjectives after the paucal numerals may optionally be in the nominative instead of genitive plural, depending on the form of the following noun—to the extent that the genitive singular (governed by the QP) is syncretic with the nominative plural, adjectives may show false nominative agreement. This fact can probably be handled using feature neutralization as in chapter 3, provided some way of incorporating

126

PARAMETERS OF SLAVIC MORPHOSYNTAX

number into the system can be developed. My focus in this chapter, however, remains on canonical quantifiers such as five and above, and within this context my main point here is that in Serbo-Croatian the form of the demonstrative invariably depends on the quantifier, whereas in Russian it depends on case of NPma*. 14. The issue of subject-verb agreement is taken up in section 4.2. 15. Babby (1987, 124, n. 27) discards this possibility on the basis of putative agreeing oblique prequantifiers. Speakers I have consulted, however, reject the example he cites. 16. Note a similar problem with English these/*this kind of books, where the demonstrative presumably restricts the classifier kind (of), but nonetheless for many speakers must agree in number with the head noun. 17. Neidle (1988) reaches similar conclusions, that the morphosyntactic behavior of a case depends solely on the particular morphological case. Indeed, I know of no convincing arguments that reference need ever be made to the "abstract Case" of GB independent of morphological case, beyond simply ascertaining its presence in order to satisfy the Case Filter. Recall the analysis of ATB extraction in chapter 3, where it was demonstrated that the appropriate restriction is unrelated to abstract Case and should be formulated in terms of thematic prominence instead. 18. These data are mentioned in Borras and Christian (1971, 391), who note that although both variants are possible, agreement, as in (23b), is the norm for simple numerals while government, as in (23c), is the norm for compound numerals ending in two, three, or four. 19. The accusative form of inanimates is identical to the nominative. Note that the animacy rule does not apply to cz-declension nouns in the singular, since these have an independent accusative form; nor does it apply to the singular of feminine 0-declension nouns such as mat' 'mother' or doc' 'daughter', as well as numerals historically in this class, such aspjat' 'five'. 20. Although Russian has both and Serbo-Croatian has neither, in chapter 5 we shall see that West Slavic exhibits mixed systems. I have been assuming, as Jakobson apparently did, that roughly the same feature set is able to characterize the case systems of the other Slavic languages.Thus, where dative and locative fall together in the singular in Serbo-Croatian the result is [-nonascr, +obl, +marg], and their complete syncretism with instrumental in the plural can be expressed as [ + obi, + marg]. Even if this is so, however, it is clear that the various languages need not maximize the feature oppositions as suggested for Russian, since the very number of distinct forms varies greatly. The details of Serbo-Croatian and other feature systems are tangential to the main claim in the text, however: oblique can be identified with inherent and the value of this particular feature in expressing GEN(Q) is not stable, but rather exhibits some variation. 21. One could of course claim that the paucal is neither singular nor plural, but simply paucal in number, and agreeing forms attempt to do the best they can. This would probably require adding [ ± singular] to the necessary [ ± plural] number feature. Based on its meaning, paucal would then be [ — sg, - pi], although [ + sg, + pi] might conceivably be a better option (if not ruled out by UG), since the paucal has properties of both numbers, rather than of neither. 22. See Corbett (1983a) for general discussion of this issue in various Slavic languages. 23. The term "unaccusative" is usually attributed to Perlmutter (1978), although there is much debate as to the original ownership of the idea that some intransitive verbs have underlying (or initial) objects and no subject; compare Pullum (1988) for discussion. There also seems to be considerable vacillation in the class of unaccusative verbs, both across languages and within a single language. One complicating factor is that the concept of unaccusativity is closely linked to existentiality, in that when a non-unaccusative (i.e. "unergative") verb is used in an existential sense it sometimes exhibits hallmark unaccusative behavior. This is particularly true of the genitive-of-negation rule in Russian, which in

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

127

existential contexts for many speakers extends well beyond the paradigm set of unaccusative verbs. Emphatic and scope markers, such as (n)i '(not-) even', also greatly improve the felicity of the genitive of negation. Here, however, my aim is merely to sketch out Pesetsky's analysis for comparison. Since I will eventually argue that unaccusativity is irrelevant to the distribution of QPs, there is no need for me to establish a definitive relationship between unaccusativity and the genitive of negation. For purposes of discussion, I follow Pesetsky in assuming the genitive of negation only applies to VP-internal NPs; see Babby (1980b) for an alternative treatment of genitive "subjects" that catalogs a variety of nonstructural licensing factors. 24. Interestingly, unmarked word order is the sole test that comes to mind of unaccusativity in Polish, since in that language the genitive of negation applies only in instances that have the accusative when affirmative, there being no Polish correlate to Russian (30). 25. The unergative status ofobedat' 'to eat lunch' and guljat' 'to walk' is demonstrated by the failure of sentential negation to induce genitive case: (i)

*v etom restorane ne obedalo ni odnogo deloveka in this restaurant NEG ate-lunch(n) not-even one(gen) person(gen)

(ii)

*na ulice ne guljalo ni odnogo studenta on street NEG walked(n) not-even one(gen) student(gen)

Compare these judgments with (30). However, some speakers report that even (i) and (ii) improve somewhat in purely existential contexts. 26. For the LFG analysis of Neidle (1988, 109) as well, non-agreement with quantified subjects of transitive verbs is erroneously regarded as impossible. In her system, homogenous numeral phrases are NPs and heterogenous ones are QPs. Like Pesetsky, she relates the genitive of negation to GEN(Q) in that both involve the "scope-marking feature" [ + Q], although they differ in that genitive phrases under negation are NPs assigned [ + Q] syntactically whereas homogenous numeral phrases are QPs that receive [ + Q] from their Q heads. She argues—contra Pesetsky—that genitive-of-negation NPs are invariably objects but QPs need not be. For her, however, "subjective" [ + Q] phrases become objects by a rule of demotion, which is blocked in transitive clauses by the LFG "principle of function-argument biuniqueness." Thus, it is only when QPs demote that non-agreement obtains. While I agree that the two phenomena should be distinguished, the fact that inter alia numeral phrases may occur as subjects of transitive verbs whereas genitive-of-negation phrases cannot leads me to a different analysis. Genitive-of-negation phrases are, following Neidle, NPs not QPs, but, following Pesetsky, they are both D- and S-structure objects. Non-agreeing numeral phrases, it will be argued in section 4.3, may on the other hand also be subjects. In my account they differ in category—and, consequently, in S-structure position—from heterogenous numeral phrases that induce agreement. This move is unavailable to Neidle since (i) for her all heterogenous numeral phrases are QPs, (ii) QPs have case and pronominal features, leading to potential subject-verb agreement, and (iii) grammatical functions are primitives rather than structurally defined, so that no correlate of the internal subject hypothesis is conceivable. 27. Pesetsky (personal communcation) apparently sees CSR somewhat differently, more as a function mapping semantic types into one or more syntactic categories. Hence the CSR of a question would have to be either CP or NP, to explain (34), but the CSR of an entity would be only NP, to force QPs to raise at LF. 28. Once "subject position" is taken to mean SPEC-IP, as argued in section 4.3.1, the same problems and solutions persist: the verb cannot agree with a QP subject since this remains in SPEC-VP position, but might be agreeing with a null expletive in SPEC-IP. Once again, the issue of unaccusativity is irrelevant. This state of affairs is possible because

128

PARAMETERS OF SLAVIC MORPHOSYNTAX

Russian, like all the Slavic languages, admits null expletive subjects. This property is technically independent of whether or not the language allows morphologically null theta-marked subjects. See chapter 7 and Franks (1990c) for a parametric account of these and related phenomena. 29. A further problem here is that the acceptability of the plural depends partly on the cardinality of the numeral. Corbett (1978, 1983a) summarizes data arguing for a numeral "squish," both across Slavic and universally, such that the lower a number is, the more adjectival properties it displays. Within this squish, the paucal numerals constitute a subclass of their own, being by far the most adjectival after 'one'. To the extent that the Serbo-Croatian paucal numerals are pure modifiers of a nominative head noun—albeit in a special paucal rather than the expected plural form (cf. the discussion in n. 8)—they generally take plural syntactic subject-verb agreement. Sand (1981) found that with 'two' agreement occurred in 97 percent of her examples, with 'three' 89 percent, and with 'four' 83 percent. A further complication with the paucals is that masculine nouns typically induce a special agreement option that resembles the neuter plural ending, as in (i): (i)

dva muskarca su dosla/ ?je doslo/ ?su dosli two men(pauc) AUX(3pl) came(n pl)/AUX(3sg) came(n sg)/AUX(3pl) came(m pi)

Although speakers report "fuzzy" judgments, especially where "three" and "four" are concerned, the first pattern clearly represents syntactic agreement. Compare this with pet 'five', for which the neuter is standard: (ii)

pet muSkaraca *su dosla/ je doSlo/ ?su do§li five men(gen pi) AUX(3pl) came(n pl/AUX(3sg) came(n sg)/AUX(3pl) came(m pi)

In both (i) and (ii), I take the su dosli forms as reflecting plural semantic "agreement," although the expected true syntactic agreements are different. The problem is that even if the paucal numerals technically occur with nominative paucal nouns, there is no paucal auxiliary, so that the plural su must be used instead, and nominative paucal forms such as muSkarca in (i) are easily taken as genitive singular, leading to the je doslo option found with the nonpaucal numerals. 30. Note that similar problems exist whenever categories defective in features are conjoined, such as clausal subjects. 31. Russian verbs, unlike Serbo-Croatian ones, do not mark person in the past tense since there is no auxiliary. This fact will be relevant to the analysis of null-subject phenomena presented in chapter 7. Also, Slavic verbs in general do not mark gender in non-past tenses. 32. Note that the third singular auxiliary ye disappears after the reflexive clitic se. 33. Although I assume that QPs, unlike NPs, do not raise to IP-specifier position since they do not need case, the possibility remains that they may raise anyway, although this would be an unmotivated movement, contra the "least effort" principle of Chomsky (1991). One might maintain, however, that nothing prevents raising of a QP into IP-specifier position, but that once there the construction is ruled ungrammatical along Pesetsky's lines, with the category of the trace being determined by the CSR theory. A further and more general question—one that in fact arises for all non-cased arguments—is how the theta-role of a QP is identified if it is not in a case-marked chain. Perhaps Q undergoes LF incorporation into V as an alternative to satisfying the Case Filter, as proposed in Baker (1988) and Johnson (1991). This would independently restrict QP to VP-internal positions, since otherwise its trace would not be antecedent-governed. 34. There are some possible counterexamples to both claims (cf. Greenberg and Franks, 1991, for discussion). It may be that small clause subjects also exhibit these properties, in which case the relevant fact is not the location of the QP but rather merely its categorial status.

QUANTIFIED STRUCTURES: RUSSIAN VERSUS SERBO-CROATIAN

129

35. For a recent treatment of reciprocals within GB theory, see Heim, Lasnik, and May (1991). 36. Although the effect might be circumvented by movement in the syntax (vs. LF) by allowing the intermediate trace in COMP to be an NP and antecendent-govern the original trace, after which it deletes at LF, following Lasnik and Saito (1984, 1992). 37. The judgments are not absolute, which may be related to the absence of ECP effects in many varieties of colloquial Russian. 38. The difference in interpretations does not have to do with the fact that the QP in (62a) is actually in object position (prijti 'to arrive' being an unaccusative verb), whereas the NP in (62b) is an S-structure subject, since the same contrast exists for transitive and unergative verbs.

5 Quantified Structures: Polish and Other Puzzles

In chapter 4 I argued for two intersecting parametric contrasts between quantified phrases in Russian and Serbo-Croatian. These are summarized in (1). (1) a. 1. GEN(Q) is a structural case in Russian. 2. GEN(Q) is an inherent case in Serbo-Croatian. b. 1. Quantified phrases are either NPs or QPs in Russian. 2. Quantified phrases are only NPs in Serbo-Croatian. NPs obligatorily raise to IP-specifier position in order to receive nominative case; QPs do not. These oppositions extend proposals due to Babby (1987) and Pesetsky (1982), thereby accounting for a range of differences between the two languages. The parametric approach thus provides support for their original analyses of Russian. The discussion in chapter 4 was fairly self-contained, however, and a variety of potentially problematic issues were ignored. For one thing, the important question of whether the parameterization in (1) can accommodate quantified phrases in the other Slavic languages remains unexplored. Strong support for the analysis of the differences between East Slavic Russian and South Slavic Serbo-Croatian might be drawn from these languages, provided they can be treated as variations on the above theme. In this chapter the analysis is extended to West Slavic, and several striking inconsistencies are resolved by invoking a third option that cuts across the parameters in (1). Next, I treat three residual types of constructions that are formally related to numeric quantification in Slavic and that raise a host of new problems. The analysis of distributive /Jo-phrases in Russian extends the idea that [ ± oblique] is the appropriate mechanism to handle GEN(Q) and leads to a new account of the internal structure of numerically quantified expressions. The approximative inversion construction and status of frozen quantifiers, on the other hand, do not lend themselves to such straightforward resolution. Finally, the relationship of bare genitive constructions to numeral phrases in general is discussed, and it is argued that whereas the partitive is 130

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

131

properly quantificational and involves a null quantifier, the genitive of negation construction is independent of GEN(Q) per se.

5.1. West Slavic and the Accusative Restriction This section attempts to reconcile the facts of West Slavic numeral expressions with the analysis of variation developed in chapter 4. Polish is examined in detail, and it is argued that Polish numeral phrases can be assimilated to the proposed system so long as GEN(Q) only applies in accusative contexts in that language. This mysterious restriction serves to explain a host of superficially unrelated phenomena in the morphosyntax of Polish numeral phrases. It is then argued that the facts of Czech, Upper Sorbian, and Lower Serbian are consistent with the claim that languages may restrict application of GEN(Q) to accusative NPs. 5.1.1. Polish Numeral Phrases The behavior of quantified phrases in Polish presents several prima facie problems for the account in (1). Concentrating as before on the type instantiated by piqc 'five' that assigns GEN(Q), we see that such quantified phrases do not conform to either pattern described so far.1 In particular, most relevant tests indicate that quantified phrases in Polish are NPs, not QPs. With respect to tests for whether GEN(Q) is structural or inherent, on the other hand, the data provided by speakers and reference grammars are sometimes in conflict, suggesting that the system is in flux and that whether GEN(Q) is structural or inherent is to some extent a lexical property of specific items. After presenting the intricacies of the Polish system, I shall discuss how they can be understood within the range of the two parametric choices argued for in the previous chapter. So far as I can tell, there is never any choice as to the verb form. With numerals that assign GEN(Q), the verb must always be in the neuter singular, as shown in (2).2 (2) a. piec kobiet five(nom) women(gen pi) b. pieciu studentow five(gen) students(gen pi)

gfosowaJo przeciwko Walesie voted(n sg) against Walesa(dat) gtosowaJoprzeciwkoWaJesie voted(n sg) against Walesa(dat)

Unlike in Russian, and even Serbo-Croatian, the plural option is virtually never available.3 This is a surprising fact, given that "semantic" agreement with semantically plural subjects but syntactically singular (or numberless) subjects is generally possible as a marked option in other languages. This might nonetheless suggest that Polish quantified phrases of thepiqc type are invariably QPs. However, they pass all the other subject/NP tests catalogued in section 4.3, such as controlling gerunds and infinitives and anteceding reflexives. Consider the following example of gerund control:4 (3)

pi^c kobiet weszlo do pokoju spiewajac five women(gen pi) entered(n sg) to room(gen) singing

132

PARAMETERS OF SLAVIC MORPHOSYNTAX

This shows that the NP option must therefore be available, drawing into question my initial contention that these are QPs. Furthermore, in oblique positions the material following the numeral appears in the appropriate oblique case and the numeral declines and agrees, as in (4). (4)

a. z pi^cioma stolami with five(inst) tables(inst pi) b. o pie.ciu kobietach about five(loc) women(loc pi)

The fact that the oblique case spreads throughout the entire phrase implies that these must be NPs. I therefore conclude that Polish quantified phrases are only NPs and that some other explanation for the impossibility of plural agreement in (2) must be sought. Turning now to the issue of whether GEN(Q) is structural or inherent in Polish, a similar conflict arises. Although the examples in (4) appear to argue that it is structural, as in Russian, the fact that demonstratives appear in the genitive before the numeral, as in Serbo-Croatian, shows that it must really be inherent: (5)

tych pie.c kobiet czyta ksiazke. these(gen pi) five women(gen pi) reads(3sg) book(acc)

Assuming the same structure as was posited in chapter 4 for the other languages, I conclude that the QP headed by piqc 'five' is assigning inherent case to its sister NP and the demonstrative adjoins to NPma* after GEN(Q) has been assigned. We are thus in a quandary, since the evidence suggests that GEN(Q) may be both structural and inherent in Polish. As we shall soon see, this conclusion is corroborated by the fact that genitive tych in (5) is not the only option. Any approach to Polish that treats it essentially like Serbo-Croatian with respect to the choices in (1) leaves several empirical inconsistencies to be addressed. It turns out that a single stipulation can account for all the divergencies from the SerboCroatian pattern. My solution is that GEN(Q), be it structural or inherent, is only assigned in accusative NPs in Polish. Although I am unable to explain the reason for this stipulation, it is morphologically motivated and makes all the correct predictions about Polish's exceptionality. In order to see this, consider again the form of the numeral in (2)— nominative piqc with a feminine noun and genitive piqciu with a "virile" (i.e. masculine human) one. Why should this discrepancy exist? It turns out that the only way to analyze these forms consistently is to treat them both as accusative.5 The reason is that the accusative is syncretic with either the genitive or the nominative, depending on whether or not the noun is virile. Consequently, quantified subjects in Polish must be analyzed as accusative rather than nominative, as follows: (6)

W^ACC [QP Q] tNP:GEN(Q) • • • 11

Although surprising, given that subjects are otherwise nominative in Polish, the assumption that GEN(Q) is only assigned in accusative NPs readily explains all of

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

133

the above phenomena, in addition to the morphological idiosyncracies of quantified subjects. In particular, the reason that verbs never show plural agreement becomes trivial and straightforward: subject-verb agreement occurs with NPs only in the nominative, never the accusative.6 The neuter singular is thus the only viable option; even the possibility of marked "semantic" agreement, as in Serbo-Croatian, is ruled out. Strong support for this approach can be drawn from further data not yet mentioned. Although, according to most speakers consulted, prequantifiers in Polish only appear in the genitive, several reference grammars cite the alternative possibility of the nominative.7 I interpret this to mean that in Polish GEN(Q) may optionally be structural, so that the prequantifier may move out of the domain of the QP before GEN(Q) is assigned. Surprisingly, even those speakers who accept this option only allow the verb to appear in the neuter singular, as in (7). (7) tych/te pie.c kobiet pojechaiodo Warszawy these(gen/nom-acc) five women(gen pi) went(nsg) to Warsaw(gen) This is in striking contrast to the pattern found in Russian, where the (plural) nominative demonstrative always forces (plural) subject-verb agreement. If, however, the quantified phrase is analyzed as accusative rather than nominative, then the failure of agreement follows immediately. Note here that in making this claim I am crucially interpreting the te 'these' option in (7) as an instance of accusative and not nominative case, contrary to traditional views but completely consistent with other properties of Polish numeral phrases. Thus in the modern standard language GEN(Q) is inherent, although in older styles it was apparently structural, and this relatively recent change is still in some flux. These assumptions have no other effect on the proposed analysis. In oblique positions the accusative NP is disallowed, since the Theta-Criterion calls for the appropriate oblique case. Hence in (4), for example, the numeral agrees and is purely adjectival, as in Russian. The structure in (6) cannot be used, since the accusative on NP1"3* will not satisfy the lexical requirements of the oblique case assigner and, by assumption, in Polish QPs occur only in accusative contexts. The Serbo-Croatian pattern is therefore disrupted. Although the source of this "intrinsic" accusative is unclear, it is worth noting that quantified subject NPs are in IP-specifier rather than VP-specifier position. This can be shown by their inability to support long-distance extraction, unlike Russian QP subjects, which I argued are VP-specifiers. Polish exhibits the standard subjectobject asymmetry governed by the ECP, as shown in (8). (8) a. jaka ksiazke.; chcesz, zeby Janek przeczytal e; which book(acc) you-want that Janek(nom) read(m sg) b. *jaki czJowieki chcesz, zeby Cj przeczytat te. ksiazke^ which person(nom) you-want that read(m sg) this book(acc) Interestingly, quantified subjects pattern like other subjects, so that (9) has the status of (8b) rather than (8a). (9)

*ilu studentow; chcesz, zeby Cj przeczytato te. ksiazke. how-many students(gen pi) you-want that read(n sg) this book(acc)

134

PARAMETERS OF SLAVIC MORPHOSYNTAX

This indicates that the quantified NP must raise to IP-specifier position regardless of the fact that it inherently bears accusative case.8 Like all aspects of morphosyntax, this intrinsic accusative also needs to be licensed. Since it is generated in a caseless position, it must move to one in which case is licensed: Spec-IP. Note that a structural case position, such as the nominative subject position, can license oblique case NPs. Obliques can similarly occur in accusative object position, as we shall see when the genitive of negation is examined in 5.3.2.1. Crucially, a subject NP may be accusative or an object NP genitive only when otherwise motivated, in these instances by containing a QP and being in the scope of negation, respectively.9 This is in keeping with the "principles of economy" outlined in Chomsky (1991, 1992): an NP does not have case unless it has to, but the ' 'accusative restriction'' in Polish forces an NP to be accusative if it contains a QP. If this restriction is a matter of categorial selection, then such quantified NPs must bear accusative case at D-structure, that is, intrinsically, but other NPs in structural case positions wait (Chomsky's "Procrastinate") until they need case for PF (Phonetic Form) or LF (Logical Form) visibility (as discussed in chapter 7). I have argued that there is some vacillation in Polish as to whether GEN(Q) is structural or inherent, although it is only assigned in accusative NPs and quantified phrases are consistently NPs regardless. The view that the status of GEN(Q) fluctuates is also supported by the behavior of collective numerals in Polish, used for mixed gender groups of people, children, and some animals. According to Schenker (1966, 241-243), these items function like other cardinals except that they still govern GEN(Q) in instrumental contexts, just as in nominative and accusative (i.e. [ — oblique]) positions.10 He thus contrasts accusative and instrumental (10) to dative and locative (11). (10)

a. mam piecioro dzieci I-have five(acc) children(gen pi) b. jade z pieciorgiem dzieci I-go with five(inst) children(gen pi)

(11)

a. pieciorgu dzieciom sie nudzi five(dat) children(dat pi) REFL bored b. siedze^ przy pieciorgu chorych dzieciach I-sit by five(loc) sick(loc pi) children(loc pi)

This can be explained by assuming that of the various oblique forms only the instrumental pieciorgiem in (lOb) assigns inherent GEN(Q). This can be accounted for in terms of the idea that instrumental is the most marked case, bearing positive values for all case features, and is thus most resistant to change.11 Interestingly, not all speakers share these judgments, and there is considerable vacillation as to the the proper behavior of collectives, with either agreement extended to the instrumental or government to the other oblique cases. Alongside (lOb) the agreeing instrumental is also found, as in (12) (12)

z pieciorgiem dziecmi with five(inst) children(inst pi)

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

135

Note that demonstratives appear in the genitive in (lOb) and the instrumental in (12), in accordance with their analysis as originating inside NP. This is shown in (13). (13)

a. z with b. z with

tych these(gen pi) tymi these(inst pi)

pieciorgiem five(inst) pieciorgiem five(inst)

dzieci children(gen pi) dziecmi children(inst pi)

Finally, some speakers prefer government in examples like (11), replacing them by (14).

(14)

a. pieciorgu dzieci sie nudzi five(dat) children(gen pi) REFL bored b. siedze przy pieciorgu chorych dzieci I-sit by five(loc) sick(gen pi) children(gen pi)

For such speakers, all forms of collective numerals presumably govern inherent GEN(Q).12 They have thus regularized the previously mixed system that characterizes collective numerals in the literary standard.

5.1.2. Numeral Phrases in Other Languages It is probably no accident that GEN(Q) in Polish is restricted to NPs of a particular case and that this case should be accusative. Since accusative is the default case, this is the least marked environment in which QPs might be expected to occur. Comparing Polish to Russian to Serbo-Croatian, we see that the distribution of QPs— measured by the range of positions in which GEN(Q) occurs—progressively expands from accusative to nominative and accusative to all cases. In case feature terms, this means QPs can occur in NPs that are [ — oblique, -marginal, -nonascriptive, —phrasal], [-obi, amarg, (Snonascr, •vphras], and [aobl, pmarg, jnonascr, 8phras], respectively. This seems not to be a hierarchy of choices, however, but rather the interaction of two independent parameters. Recall that the latter opposition is handled by varying the [± oblique] feature of GEN(Q), as in (la). The accusative restriction, on the other hand, suggests that subcategorization might be implicated, such that only the least marked NPs, those with all minus features, subcategorize for QPs. There are two obvious problems with this approach. One is that the structure for quantified NPs assumed in chapter 4 does not involve a head—complement relation, so that categorial selection would be impossible to implement. This problem will be rectified in the next section, however, where it will be argued that QPs are complements of Ds. The other problem is that, since the accusative restriction essentially imposes a third parameter, it is expected to interact with the other two numeral phrase options. In this section I argue that it indeed does so. How might the accusative restriction interact with the two parameters in (1)? The very fact that GEN(Q) in Polish can be [ ± oblique] reveals that it can cut across (la). In this section, relying on data reported in Lindseth (1993a), I show that West Slavic languages exhibit the full range of possibilities. With respect to (Ib), in which the

136

PARAMETERS OF SLAVIC MORPHOSYNTAX

maximal projection of a quantified phrase varies between NP and QP or NP, the accusative restriction is necessarily much more limited. If the restriction is formalized in terms of subcategori/ation, then it should be impossible for a bare QP to be restricted to accusative contexts. The reason is simply that the containing NP (or DP, in the analysis of section 5.2.1), on which the accusative restriction is imposed, is absent. Of course, we might expect that, even in a language with both QP and NP quantified phrases, when the NP option is chosen it could be limited to accusative contexts. This is clearly not the case in Russian, however: (15)

a. *[NP etix pjat' devusek] prislo these(genpl) five girls(genpl) arrived(n) b. *[NP tysjacu devusek] prislo thousand(acc) girls(genpl) arrived(n)

The unambiguously accusative subject NPs in (15) are ungrammatical. Their ungrammaticality follows from the assumption that GEN(Q) in Russian is not restricted to accusative contexts. Quantified subjects in Russian are thus either nominative NPs or bare QPs. It is interesting to recall that sentences like those in (15) do occur in SerboCroatian, but for other reasons: (16)

a. [NP ovih pet devojaka] je doslo these(gen pi) five girls(gen pi) AUX(3 sg) arrived(n sg) b. [NP hiljadu devojaka] je do§lo thousand(acc) girls(genpl) AUX(3sg) arrived(n sg)

I have argued that ovih in (16a) is genitive because it orginates within the scope of pet, which assigns [ +oblique] GEN(Q) in Serbo-Croatian. In section 5.2.3,1 shall discuss hiljadu in (16b) as a frozen, caseless form that resembles the accusative. Although it is significant that the accusative restriction and frozen forms both involve the accusative, I believe the two phenomena are independent consequences of accusative being the least marked case. The subject NPs in Serbo-Croatian (16) cannot be analyzed along the same lines as Polish. The reason is that ovih in (16) must be genitive, since the accusative form of the feminine plural demonstrative is ove, and hiljadu in (16b) must be frozen, since it appears in oblique contexts as well. Lindseth (1993a), on the other hand, convincingly argues that the accusative restriction cuts across the West Slavic languages. She shows that the facts of Czech and Slovak are consistent with the claim that GEN(Q) in these languages is inherent but only assigned in accusative NPs. Thus we find that although Czech is like Polish and Russian in exhibiting the heterogenous/homogenous case pattern, it differs from these languages (and is like Serbo-Croatian) in requiring demonstratives and other prequantificational modifiers to appear in the genitive: (17)

a. s peti pany with five(inst) men(inst pi) b. pet pami prislo five men(genpl) arrived(n sg)

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

137

c. te"ch/*ty pSt hezkych dfvek jelo these(gen pl)/(nom pi) five beautiful(gen pi) girls(gen pi) traveled(n sg) d. tech p£t panii pfiSlo these(gen pi) five men(gen pi) arrived(n sg) e. v§ech pet chlapcii demonstrovalo all(gen pi) five boys(gen pi) demonstrated(n sg) The genitive prequantifiers in (17c-e) suggest that GEN(Q) is only inherent in Czech. Of course, as with Polish, one is then left to explain any differences between Czech and Serbo-Croatian by some other means. For most of the relevant facts, Czech numeral phrases behave like Polish ones, dva 'two', tfi 'three', and ctyfi 'four' restricting nominative plural NPs and requiring plural subject-verb agreement, pet 'five' and above governing the genitive plural in direct case contexts, requiring neuter singular agreement in (17b-e),13 and—in contrast to SerboCroatian—declining throughout in oblique case contexts, as in (17a).14 This array of facts suggests that the solution proposed for Polish that GEN(Q) is only assigned in accusative NPs—will extend also to Czech. However, the morphological evidence supporting my crucial claim that numeral phrase subjects in Polish are for some reason intrinsically accusative is absent in Czech. That is, a form such as pet 'five' in (17b-e) could be nominative or accusative, unlike Polish piqciu, which in comparable sentences is clearly accusative. Nonetheless, since the subject NPs in (17b-e) are equally viable in accusative positions, Czech is consistent with my analysis of Polish. I therefore conclude that the only significant difference between Czech (and tentatively also Slovak and Slovenian) and Polish is that the [ — oblique] GEN(Q) option found in Polish does not exist in Czech. Lindseth (1993a) claims that Lower Serbian, on the other hand, instantiates only this option. Working with texts and reference grammars such as Starosta (1991), she cites the following examples to argue that GEN(Q) is structural in this language: (18) a. nase sedym krowow jo podojte our(nom-acc) seven cows(gen pi) AUX(3sg) milked(n sg) b. wsykne sto hektarow trajdy jo posecone all(nom-acc) hundred hectares of-grain AUX(3sg) sewn(n sg) The obligatoriness of the neuter verb, even with the "nominative" prequantifiers nase and wsykne, reveals that Lower Serbian differs crucially from Russian. If these forms are analyzed as accusative, as proposed for Polish, however, the failure of the verb to show plural agreement is expected. Note that (18a) is the expected accusative form since only animate masculine nouns substitute the genitive for the accusative, and krowa 'cow' is feminine. Moreover, Lindseth points out that in the plural animacy marking is restricted to quantified (masculine) direct objects. Compare (19a) with (19b): (19) a. mam styroch konjow I-have four(acc-gen) horses(acc-gen) b. wizis nase konje you-see our(acc-nom) horses(acc-nom)

138

PARAMETERS OF SLAVIC MORPHOSYNTAX

The true test of the accusative proposal would thus be to find an unambiguously animacy-marked quantified masculine singular subject, such as the hypothetical (20). (20)

nasych sescoch golcow jo serbski nawuknulo our(acc-gen) six(acc-gen) boys AUX(3sg) Sorbian learned(n sg)

Unfortunately, Lindseth does not report having found such an example. The problem, she suspects, is that Lower Sorbian typically employs a special adjectival form of the numeral with masculine animates. This form agrees with the noun, which is nominative plural, and the predicate displays similar plural subject-verb agreement: (21) wsykne sedmyj bratsy su psisli all(nompl) seven(anim) brothers(nom pi) AUX(3pl) arrived(pl) Contrast the verb forms of (21) and (18b). The fact that the former is plural and the latter neuter singular supports my claim that wsykne 'all', which is clearly nominative in (21), must be analyzed as accusative in (18b). I thus conclude, along with Lindseth, that the Lower Sorbian data are consistent with my contention that the accusative restriction constitutes a valid parametric option in the morphosyntax of Slavic quantified expressions. Finally, Upper Sorbian, primarily as described in the work of FaBke (1981,1987), seems to exhibit a range of patterns. Lindseth interprets these patterns to mean that different varieties of Upper Sorbian pattern like Polish (with both Lower Sorbian and Czech/Slovak/Slovenian options instantiated) and Russian (although probably without maximal QPs). Data that express these three complex possibilities, drawn from FaBke (1981, 499-527), are summarized in (22). (22)

a. zaste dwaceci let stworichu . . . past(nompl) twenty years(genpl) created(3pl) b. dalsich dzewjec kolektiwow wojuje . . . further(gen pi) nine teams(genpl) compete(3sg) c. te piec fenkow je falowafo these five pfennigs(gen pi) AUX(3sg) lacking(n sg)

Although a proper analysis of such facts would require careful field work with native speakers, one can make the following tentative conclusions: the plural verb in (22a) indicates that the subject is a nominative NP and the nominative plural form te 'these' indicates that GEN(Q) is [-oblique] here; the obligatory singular verb in (22b), coupled with the genitive form dalsich 'further', indicates that the GEN(Q) is [ +oblique] here; the singular verb in (22c), coupled with the syncretic form te, indicates that the accusative restriction option is in force here. Let us examine more closely the form of te in (22c), which could be taken as nominative plural, accusative plural (as I claim), or neuter nominative singular. Because the verb is neuter singular, FaBke (1981, 518) maintains that te in such examples must be neuter singular as well. The nominative plural option can be rejected, since this would give a plural verb, as in (22a). Further evidence against

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

139

the nominative plural hypothesis lies in the fact that the virile nominative plural demonstrative is ci, not te, but te nonetheless occurs with virile quantified nouns, as in (23). (23) a.

te pjec dzefacerow these(acc pi) five workers(gen pi) b. *ci pjec dzelacerow these(nom pi virile) five workers(gen pi)

The accusative plural option of (23a), on the other hand, is consistent with my analysis of Polish and is no more unmotivated than FaBke's neuter singular proposal.15 Furthermore, it is worth noting that FaBke cites no examples of quantified NPs with the increasingly standard alternative neuter demonstrative to 'this' (or other similar members of the pronominal declension, such as wso 'all', samo 'only'). He only cites forms in -e, which—unlike forms in -o—can also be regarded as accusative plural. Finally, as Lindseth notes, the neuter singular approach will not extend to Lower Serbian, since prequantifiers such as nase 'our' and wsykne 'all' cannot be analyzed as neuter singular, the only ending for which is -o.

5.2. More Quantified Expressions In this section I consider the following three Slavic quantificational constructions not yet discussed: (1) distributive /?o-phrases, (2) approximative inversion, and (3) frozen quantifiers. These constructions seem to be in some flux in modern Russian, with speakers occasionally providing conflicting or inconsistent judgments. While approximative inversion does not exist outside of East Slavic, the other two phenomena are more general, but have distinct aspects in the different languages. Their properties serve both to shed light on the analysis developed in chapter 4 and to raise new questions about its applicability. 5.2.1. Distributive Po-Phrases and the Structure of DP One particularly problematic quantified expression in Russian is the distributive /w-phrase. This is discussed in several recent treatments of Russian, including Mel'cuk (1985), Babby (1985), Crockett (1976a), and Neidle (1988), and is also incorporated into Pesetsky's (1982) analysis. This section examines the curious government paradigm of distributive po in Russian. The behavior of this element poses a host of problems for standard views of case assignment. It is argued that the properties of po follow immediately if po is treated as a preposition assigning a structural dative case DAT(Q), comparable to the structural GEN(Q), and if Russian numerals are structurally assimilated to other more familiar types of quantificational elements. Once QPs are regarded as functional categories, with obligatory specifierhead agreement, and treated like other phrases headed by operators, their unusual interaction with the special preposition po becomes clear.

140

PARAMETERS OF SLAVIC MORPHOSYNTAX

5.2.1.1. Government Patterns and Initial Proposals Russian po applies to a numerically quantified phrase to impose a distributive (DIST) meaning roughly corresponding to 'each'. This element displays several distinct government patterns that are in some flux in the contemporary language and that make it difficult to justify a unified analysis. It typically appears in ' 'accusative'' direct object and adjunct time/distance phrase positions, although it can appear in "nominative" subject positions as well. The range of government possibilities is illustrated in (24): (24)

a. kazdyj ucenik polucil each student received b. kazdyj ucienik polucil each student received c. kazdyj ucenik polucil each student received d. kazdyj ucenik polucil each student received e. kazdyj ucenik polucil each student received

po odnomu rublju DIST one(dat sg) ruble(dat sg) po rublju DIST ruble(dat sg) po dva rublja DIST two ruble(gen sg) po pjat' rublej DIST five ruble(gen pi) po pjati rublej DIST five(dat) ruble(gen pi)

The argument NP following po is distributed over some other individuated argument NP in the sentence; this argument is often indicated by an explicit quantifier, typically kazdyj 'each'. Po implies an iteration of the action but does not affect the predicate-argument structure of the clause. The NP in the po-phrase thus receives whatever theta-role the verb assigns to the position occupied by the po-phrase, po itself assigning no theta-role. In this sense, po is different from other prepositions in Russian, since it bears no thematic properties of its own.16 Its government properties are also baffling, since, as a consideration of the examples in (24) reveals, distributive po appears to be able to assign several different cases.17 Existing accounts ofpo, such as Crockett (1976a), Mel'cuk (1985), and Babby (1985), generally assume a mixed analysis of its case government properties, such that the particular cases it governs depend to some extent on the cardinality of its object NP. Cases proposed ordinarily include the dative and accusative to handle examples like (24a, b) and (24c, d), respectively. Some scholars, such as Mel'cuk (1985) and Neidle (1988), add the genitive to this list in discussing examples such as (24e), although the form of the numeral here could also be taken to be dative (as I have glossed it in anticipation of the analysis), or even locative. The basic puzzle posed by po is thus that it defies a uniform analysis as a simple preposition. As a point of departure, consider the fact that when po quantifies over singular NPs, whether modified by odin 'one' or not, it clearly acts like a preposition governing the dative case, as in (24a, b) or the following examples cited in Babby (1985):18 (25)

a. otec dal detjam po (odnoj) gruse father(nom) gave(m) children(datpl) DIST one(dat) pear(dat)

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

141

b. otcom dano detjam po (odnoj) gruse father(inst) given(n) children(datpl) DIST one(dat) pear(dat) Babby (1985,1986) analyzes these/jo-phrases as NPs, arguing that in (25b) passivization moves them from object to subject position, concomitantly assigning nominative case. It seems simpler to me, however, to follow Franks (1986) and Neidle (1988) and regard them as PP objects. This allows the/Jo-phrase in (25b) to remain in situ under passivization, since as a PP it does not bear case and hence need not move to subject position. Based on this, the null hypothesis is to claim thatpo governs the dative and is thus analogous to other prepositions that do so, such as k 'to'. (26)

[PP k [NP:DAT odnomu rublju]] to one(dat) ruble(dat)

The structure in (26) represents a preposition with an NP complement, that is, an ordinary PP. The most reasonable conclusion is that the same structure should be ascribed also to /w-phrases. That is, whatever else it may be, distributive po must at some level be analyzed as a preposition assigning the dative case—po is necessarily transitive and there is no other available source for the dative. The problem with this conclusion is that, unlike other prepositions that govern the dative, when the distributed NP contains a numeral higher than one, this NP does not similarly appear in the dative. Compare (27) with (28):19 (27) a. po DIST b. po DIST

dva rublja two ruble(gen sg) pjat' rublej five ruble(gen pi)

(28) a. k dvum rubljam to two(dat) ruble(dat pi) b. k pjati rubljam to five(dat) ruble(dat pi) Crucially, even though po assigns the dative case to singular NPs, it is somehow prevented from assigning this same case to the quantified NPs in (27). In this respect, it contrasts markedly to other prepositions that govern the dative, such as k 'to' in (28). In (27), the dative cannot be realized, whereas in (28) it must percolate throughout the entire numeral phrase. Converse application is clearly ungrammatical in both instances: (29) a. *po dvum rubljam DIST two(dat) ruble(dat pi) b. *po pjati rublam DIST five(dat)ruble(dat pi) (30) a. *k dva rublja to two ruble(gen sg) b. *k pjat' rublej to five ruble(gen pi)

142

PARAMETERS OF SLAVIC MORPHOSYNTAX

Since examples like (24a, b) and (25) demonstrate that distributive po is able to govern the dative, blocking its assignment in (30) is a serious problem, one that has not been fully appreciated in the existing literature on po. To my mind, this is the fundamental mystery posed by /w-phrases. A further complexity is that po sometimes exhibits two government options with non-paucal quantified complements: either po induces dative on the numeral and the material following the numeral appears in GEN(Q), or else po combines with the numeral without imposing any apparent case, the material following still in GEN(Q). Thus both (24d) and (24e) are acceptable, with (23d) being the more colloquial standard. Mel'cuk (1985, 235-250), who offers copious examples and discussion of the various possibilities with different numerals, contends that po is actually assigning the genitive in examples like (24e).20 There are serious problems with this idea, however, and little evidence to support it. For one thing, it forces careful stipulation of the contexts in which genitive rather than dative is assigned;21 for example, the genitive is not a viable option in (24a-c). Moreover, whenever it is possible to distinguish the dative from the genitive, the quantifier following po is unambiguously dative. Mel'cuk himself cites examples such as those in (31).22 (31)

a. po tysjace/*tysjaci rublej DIST thousand(dat)/(gen) rubles(gen pi) b. po mnogu/*mnogogo raz DIST many(dat)/(gen) times(gen pi) c. po DIST d. po DIST

millionu/*milliona rublej million(dat)/(gen) rubles(gen pi) kilogrammu/*kilogramma xleba kilogram(dat)/(gen) bread(gen)

Crucially, the genitive form is absolutely impossible, demonstrating thatpo does not in fact assign this case. Note, however, that mnogu 'many' in (31b) is a special dative form used after po, the normal dative being mnogomu. As observed by Mel'cuk (1985, 244), other similar special datives are skol'ku 'how many' and neskol'ku 'several'. This fact supports my claim, to be developed in the next section, that distributive po assigns a special [-oblique] dative. In all fairness, Mel'cuk does devote some attention to the dative versus genitive issue, concluding that the traditional analysis ofpopjati as dative is wrong. He does so on the basis of problematic examples such as (32). (32)

po pjatisot/*pjatistam rublej DIST five-hundred(gen)/(dat) rubles(gen pi)

Here, -sot (the genitive form of 'hundred') supersedes the dative form -stam found after other prepositions that assign the dative: (33)

k pjatistam rubljam to five-hundred(dat) rubles(dat pi)

The evidence is thus far from consistent, although the bulk of it seems to me to favor the dative solution. Mel'cuk remarks that this unexpected pjatisot has been pointed

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

143

out in the literature, where it is considered some sort of "idiosyncratic" dative. I do not think that this claim is as outlandish as Mel'cuk would have us believe, however. I believe that only the pjati portion is dative, with the sot component either assigned genitive by the numeral pjat' or frozen as such. When the structure of quantified phrases is elaborated in section 5.2.1.3, (32) will be analyzed as follows: (34)

tPP po [QP pjati [Q, sot [NP rublej]]]]

Only the pjati portion, which occupies the specifier position of QP, will be able to receive (structural) dative frompo. I shall argue that this process takes place under Exceptional Case Marking. Putting this eventual analysis temporarily aside, let us return to the problem posed by (24e), repeated as (35). (35)

po pjati rublej DIST five(dat) rubles(gen pi)

This property of Russian distributive po, which is now preserved only in more literary styles of the language, is usually analyzed as idiosyncratic, since (35) looks quite unlike any other case phenomenon in Russian. In this kind of example, po appears to be assigning dative to pjati, with the quantifier nonetheless still assigning genitive to the nominal material following it. This is in fact how I believe the construction in (35) should be analyzed; the problem lies in figuring out an appropriate structure that will have the effect of allowing po to assign one case to the numeral and simultaneously allow the numeral to assign another case to rublej. One fairly standard kind of approach to this problem, following arguments in Babby (1985) and Franks (1986), for example, is to claim that what is involved in examples like (35) is a "prepositional quantifier," in the sense thatpo and pjati form a quantificational PP that itself assigns the genitive of quantification. In other words, (35) could be given a structure roughly as in (36). (36)

[NP [PP po [NP:DAT pjati]] [N.:GEN(Q) rublej]]

Although this account seems reasonable on both morphosyntactic and semantic grounds, in the next section I will argue that a more insightful analysis can be constructed by taking advantage of my claim that po in fact assigns a [ - oblique] dative case, DAT(Q). Before doing so, however, it is worth pointing out that I differ from Babby (1985) with regard to the possibility of extending the structure in (36) to other putative "prepositional quantifiers." Consider, for example, his treatment of okolo 'about', which, in addition to meaning 'near', can mean 'approximately'. Babby argues that a structure roughly comparable to the one I suggested in (36) is also appropriate for examples like (37). (37)

okolo desjati rublej about ten(gen) rubles(gen pi)

Babby's reason for connecting the two is that both distributive po and approximative okolo have a quantificational function and, in that function, semantically restrict the numeral only, rather than the entire NP. While this is indeed true, okolo displays

144

PARAMETERS OF SLAVIC MORPHOSYNTAX

none of the formal peculiarities associated with po. Instead, as Neidle (1988) observes, the two display strikingly different case government patterns, with okolo behaving like a garden variety preposition. For one thing, neither of the alternatives in (38) is available. (38) a. *okolo desjat' rublej about ten rubles(gen pi) b. *okolo dva rublja about two ruble(gen) That is, okolo obligatorily assigns the regular [ + oblique] genitive to the entire NP. This removes much of the morphosyntactic motivation for any special treatment.23 More importantly, po never assigns its dative case to the paucal numerals, although okolo does assign its genitive to them. This contrast is shown in (39): (39) a. po dva/*dvum rublja DIST two/two(dat) ruble(gen sg) b. okolo dvux rublej/*rublja about two(gen) ruble(gen pl)/(gen sg) Clearly, after okolo the form of the noun depends on the governing preposition, not the numeral. The unacceptability of the genitive singular (paucal) rublja, which should be governed by dvux 'two' in (39b), demonstrates that the genitive plural on rublej in (37) is due to the preposition okolo 'about' rather than the numeral desjat' 'ten'. These facts illustrate two related points: 1. Distributive po is not able to assign case to the paucal numerals, only to pjat' and higher. 2. Other prepositions with quantificational force, such as okolo, invariably behave as ordinary prepositions taking an NP object, regardless of their interpretation. The first observation follows from the fact that the paucal numerals are morphologically opposed to the higher numerals in being essentially adjectival rather than nominal. Assuming that case can only be directly assigned to NPs, and not APs, the impossibility of assigning dative to the adjectival numeral dvum in (39a) immediately follows.24 Note that this conclusion holds regardless of the structure of po-phrases, so long as the case of the numeral is ascribed to government by the preposition. The second observation—that po is the sole realistic candidate for a "prepositional quantifier"—suggests that maybe even/?o can be assimilated to the standard structure of a preposition simply taking an NP complement, if its case properties are properly understood. That is, although the structure in (36) is credible in that it captures the fact thaipo exclusively governs the numeral, it is not otherwise motivated. One wonders, therefore, whether there may be a simpler analysis that makes use of independent properties of po, one that conforms to the general PP schema used to analyze po so far. In the remainder of this chapter, just such an analysis is explored.

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

145

Finally, it is also standardly assumed that po is able to assign accusative in examples like (24c, d). In the interests of unifying the case government properties of po, I reject this approach. A reasonable initial proposal is, however, that the quantificational material following po is in a frozen caseless form, po in effect having been adverbialized into a sort of prequantificational prefix. This is roughly the analysis put forward in Crockett (1976a). Although I concur with Crockett thatpo is not assigning accusative here, I shall develop an alternative analysis in which po assigns a [ — oblique] dative instead. Evidence against the idea that po assigns accusative can be found in the behavior of tysjaca 'thousand', which is the one numeral that exhibits a distinct accusative. After po the only acceptable variant is as in (3la), with the dative. Unambiguously nominative or accusative forms, as in (40), are ungrammatical. (40)

*po tysjaca/tysjacu rublej DIST thousand(nom)/(acc) rubles(gen pi)

Unlike numerals in thepjat' class, which includes also the mnogo 'many' type, there is no comparable caseless form for tysjaca.25 1 thus tentatively conclude that as a distributive preposition, po only assigns the dative case. Further evidence for the status of po as a true preposition can be drawn from its behavior before the adjectival quantifiers dva 'two', tri 'three', and cetyre 'four'. Unlike the higher numerals, which have a nominal paradigm, these cannot appear in the dative after po, as was seen in (39a). This curious fact can be understood in terms of the analysis of po as a preposition since, as such, it can only assign case to a nominal object. This option is unavailable when the quantifier is not nominal, however, thereby preventing the adjectival quantifiers dva, tri, and cetyre from appearing in the dative. Note, incidentally, that the same is true of compound numerals beginning with these numerals: (41) a. *po DIST b. po DIST

tremstam rubljam/rublej26 three hundred(dat) rubles(dat pl)/(gen pi) trista rublej three-hundred rubles(gen pi)

This suggests that it is only the first part of the numeral that can count as governed by the preposition po, as in the structure (34). However, since tri is an adjectival quantifier, it cannot be assigned case bypo. Moreover, if Mel'cuk were correct that po can assign the genitive, we might expect the equally ungrammatical (42). (42)

*po trexsot rublej DIST three-hundred(gen) rubles(gen pi)

I claim that distributive po never assigns the genitive, and the reason (32) is acceptable whereas (41a) and (42) are not is that it assigns (structural) dative case only to the first part of compound numerals, but the adjectival numerals are ineligible to receive case from po since they cannot function as NPs.

146

PARAMETERS OF SLAVIC MORPHOSYNTAX

5.2.1.2. The Structural Dative Analysis How does this morass of data fit into the structural analysis of QPs developed in chapter 4? For one thing, it demands more explicitness about the categorial feature content of QPs. Like other categories, QP must be defined in terms of part-of-speech features. I propose, following ideas originally presented in Franks (1986), that what distinguishes QPs from other categories is the presence of the feature [ + Q]. Although this feature is essential to QPs and is what determines the context for the assignment of GEN(Q), it must be augmented by the other, more universal, features [±N, ±V]. Thus, nominal quantifiers likepjaf 'five'are [ + Q, +N, -V], adjectival ones like tri 'three' are [ + Q, +N, +V], and prepositional ones like po are [ + Q, -N, -V].27 The feature [ + Q] can be understood as having arisen in the history of Russian and, as such, was added to the part-of-speech feature submatrices of existing quantifiers; detailed arguments for this change in categorial status can be found in Babby (1987). In the DP analysis of the next section, we can then assume that GEN(Q) is assigned to NP by [ + Q] heads; when a numeral agrees and declines for case, however, I speculate that it is [ —Q]. 28 Within such a system, Crockett's (1976a) adverbialization approach can be formalized as follows: po can also adjoin to the left of any [ + Q] element without imposing any case requirements. This is the colloquial variant for 'five' and above, and the only option for 'two', 'three', and 'four', since these are adjectival. The only remaining possibility concerns examples where the distributive phrase contains no quantifier. Clearly, in such cases neither alternative is available—po is simply a preposition assigning dative case, with the structure in (43). (43)

[PP po [NP odnomu mal'ciku]] DIST one(dat) boy(dat)

Note that here it becomes crucial that prefixalpo only adjoin to [ + Q] elements; odin 'one' is presumably [ —Q], since it never imposes GEN(Q). There is, however, a much simpler solution to the paradox of howpo manages not to govern the same case on phrases containing numerals higher than 'one' as on those containing (an explicit or implicit) 'one'. The answer is that this discrepancy is only apparent and that it indeed does govern a single case in both instances. The observed pattern is a consequence of the now familiar kind of interaction between structural case and GEN(Q). In order to see that this is so, compare distributive po with a preposition governing the accusative. Such a preposition, in fact, exhibits the exact same government pattern as doespo—it assigns case to its object NP, but this is blocked when the NP contains a numeric quantifier greater than 'one'. Compare the examples in (44) with those in (45): (44)

a. cerez in b. cerez in

odnu minutu one(acc sg) minute(acc sg) dve minuty two minutes(gen sg)

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

147

c. cerez pjat' minut five minutes(gen pl/) in (45)

a. po DIST b. po DIST c. po DIST

odnomu rublju two(dat) ruble(dat sg) dva rublja two rubles(gen sg) pjat' rublej five rubles(gen pi)

The difference between (44a) and (44b, c) with the accusative-assigning preposition cerez is parallel to that between (45a) and (45b, c) with the "dative"-assigning po. Thus, whatever mechanisms were invoked to explain the latter contrast should be equally applicable to the former one. In both instances, the case assigned by the preposition is unable to percolate into the numeral phrase. The reason for this kind of pattern, it was argued in chapter 4, is that the quantifier provides a more local governor at the same level of representation. Since the accusative assigned by cerez and Russian GEN(Q) both apply at S-structure, minimality blocks ACC from percolating into the GEN(Q) domain of the quantifier. It is easy to see that the proper solution to the po puzzle should capitalize on this case conflict mechanism. The mixed government pattern of distributive po will result if it is analyzed on a par with accusative-assigning prepositions such as cerez. In particular, following my account of the structural/inherent dichotomy, assume that po assigns a [ — oblique] dative case. That is, the dative case assigned by po is not the regular [ +oblique] dative, but rather differs from it precisely in being [ - oblique]. This case, which for the sake of concreteness I shall call the dative of quantification, shares its nonobliqueness with GEN(Q). The following statement thus characterizes the feature content of DAT(Q): (46)

Russian DAT(Q) is [ — obi, +marg, -nonascr, +phras]

The crucial point here is that the [-oblique] DAT(Q) differs from the regular [ + oblique] DAT in that it is assigned at S-structure. Therefore, just like the accusative, it is blocked by another closer [-oblique] case assignor, such as GEN(Q), which is similarly structural in Russian. This accounts for the ungrammaticality of (47a) and (47b) in a parallel manner.29 (47)

a. *po [NP:DAT pjati rubljam] DIST five(dat)rubles(dat pi) b. *cerez [NP:ACC pjat' minuty] in five(acc)minutes(acc pi)

Distributive po is no more able to assign dative uniformly to a quantified object than cerez is to assign accusative. Both are similarly blocked by the genitive of quantification, under minimality of government at S-structure. The ungrammaticality of the examples in (47) is due to the unmotivated case on the nouns rubljam and minuty, which have no source for DAT and ACC, respectively. They cannot be assigned these cases since the quantifier assigns GEN(Q)

148

PARAMETERS OF SLAVIC MORPHOSYNTAX

more locally, requiring the nouns to appear in their genitive plural forms rublej and minut. This raises an interesting question: Why can't the numeral appear in the case governed by the preposition and the noun in the case governed by the numeral? I have up to this point not indicated the case of the numeral when it itself governs, taking it to be a caseless, frozen form, since in non-oblique contexts syncretism generally makes it impossible to determine whether the numeral is nominative, accusative, or caseless. This is not true of the structural dative assigned by po, however, which is in fact directly assigned to the numeral in examples such as (24e). The problem is how this special type of case assignment can be effected. The solution, as before, is to see that the proper analogy to make is not with other prepositions that semantically apply to the numeral, such as okolo 'about', but rather with other prepositions that assign a structural case, such as cerez 'in, across'. Once such a move is made, it becomes possible to treat po exactly like any other preposition that assigns a [ - oblique] case. That is, po simply heads a PP and assigns case to its NP complement, just like any garden variety preposition. This makes sense for canonical instances of distributive po, where I have argued that if po assigns a structural dative, then po odnomu rublju 'one ruble each' and po pjat' ruble] 'five rubles each' can be represented as follows: (48)

a. [PP [P po] [NP:DAT(Q) odnomu rublju]] b. [PP [P po] [Npmax:DAT(Q) [QP PJat'] [NP:GEN(Q) rublej]]]

In (48a), po assigns structural dative to NP, which percolates down the phrase to the head N rublju and, eventually, by agreement, to the modifer odnomu. In (48b), on the other hand, although po again assigns structural dative to NPmax, it cannot percolate down to NP, since this is marked GEN(Q) under sisterhood to the numeral phrase headed by pjat'. 30 Now, the question is whether the type in (24e) and (35), repeated once more as (49), can be assimilated to this standard structure, instead of invoking a construction specific analysis along the lines of (36). (49)

po pjati rublej DIST five(dat) rubles(gen pi)

I will claim that it can if we allow po directly to assign its dative to the numeral phrase, rather than to the phrase containing the numeral, and if the structure of numeral phrases is modified accordingly. In short, (49) results if po is able exceptionally to assign its case to the specifier of its complement rather than to the complement itself. This phenomenon is comparable to the mechanism of Exceptional Case Marking (ECM) standardly employed to explain what happens after believe-type verbs in English, as in (50). (50)

John believes [IP [NP me [r to have written the letter]]]

The complement clause is an IP, since it contains no COMP material. Believe assigns the appropriate theta-role to this complement IP—the role of the proposition which is 'believed'—but cannot assign it case since it is a clause rather than an NP. Instead, the verb believe exceptionally assigns its objective case to the specifier of the IP, namely, to the subject NP me of that complement clause. Of particular

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

149

relevance for my analysis ofpo is the fact that such ECM occurs only with structural cases, never with inherent ones (cf. e.g. Chomsky, 1981a; Speas, 1990). The reason for this is presumably that only structural cases can be divorced from assignment of semantic roles, so that in (50) believe is assigning its theta-role to one thing (the IP complement) but its case to another (the NP specifier of that IP). Be that as it may, the important observation is that what is going on in Russian /w-phrases is entirely parallel—po is assigning its case to what looks like the specifier of its complement rather than the complement per se, this case assignment occurs independently of assignment of a semantic role, and this possibility arises precisely because the case po assigns is a structural one. The existence of the type of po-phrase in (49) thus provides striking support for my claim that distributive po assigns a structural dative case, since the possibility of ECM exists only for structural (i.e. [ - oblique]) cases. Other motivation for ECM within Russian is, admittedly, not overwhelming, although one reasonably likely candidate is the verb scitat' 'to consider', as in (51). (51)

ja scitaju [sc Veru krasavicej] I consider Vera(acc) beauty(inst) *I consider Vera a beauty'

In this example, the proposition Veru krasavicej is a kind of small clause (SC), corresponding as it does to the full clause (cto) Vera krasavica '(that) Vera (is a) beauty'.31 It is this small clause that is the object of the verb scitaju 'I consider', which takes two arguments—a 'believer' entity and a 'believed' proposition. Veru is thus interpreted as the subject of the predicate NP krasavicej, but nonetheless receives its case from the verb, even though it is not assigned a theta-role by this verb. Note that this possibility once again is connected to the fact that ACC is a [-oblique] case. For example, if (51) were negated, the genitive would not be acceptable, as shown in (52). (52)

ja ne scitaju [sc Veru/*Very krasavicej] I NEG consider Vera(acc)/(gen) beauty(inst) 'I don't consider Vera a beauty'

The reason is simply that the genitive of negation is [ + oblique] and so cannot be directly assigned to the specifier of the verb's complement. In this respect, as will be argued in section 5.3.2.1, the genitive of negation is clearly an instance of the regular [ +oblique] GEN rather than [ — oblique] GEN(Q). Correspondingly, it can override GEN(Q), as illustrated by (53).32 (53)

ja ne ponjal etix pjati zadac I NEG understood these(gen pi) five(gen) problems(gen pi) 'I didn't understand these five problems'

The ECM hypothesis thus extends to po-phrases to accommodate the otherwise mysterious case pattern in (49). 5.2.1.3. The DP Analysis I have argued in the preceding section that the Russian distributive preposition po assigns a special [ - oblique] dative case. This makes sense in terms of my claim that

150

PARAMETERS OF SLAVIC MORPHOSYNTAX

GEN(Q) is also structural in Russian, since it is similarly quantificational. The difference between these two datives can be invoked in addition to explain the special dative forms that occur after distributive po, such as mnogu 'many' in (31b).33 More importantly, I have shown how this single assumption accounts without further embellishment for the seemingly idiosycratic phenomena associated with the distributive preposition po and summarized in (24). The fact that/w assigns structural dative captures the complete range of government possibilities within a unified analysis. However, my claim that (49) is an instance of ECM still leaves several important questions unresolved. One might for example wonder what it is about the object of po that allows the preposition to assign case to the specifier of that object rather than to the object itself. The answer to this question can be found in a proper treatment of the ECM phenomenon in general. That is, by exploiting the parallelism with English ECM constructions, we may better be able to understand the case assignment mechanisms involved in Russian /w-phrases. The hallmarks of English ECM, as typified in (50), are listed in (54): (54)

a. The case assigned by V is structural, rather than inherent. b. The complement is an IP, rather than an NP, and so cannot be assigned case. c. The specifier of the complement is an NP that would otherwise have no source for case.

Now, is it possible to recreate all these characteristics for the Russian construction? I have already shown that the first claim is necessary in order to explain the impossibility of dative homogeneously percolating throughout the quantified NP, as in the infelicitous (47a). That the second and third claims also apply to Russian is somewhat more difficult to see. I have up to now been referring to the phrase after po as a quantified NP, but there is no reason to suppress the QP option, since in Russian QPs are freely admissible in structural case positions. This satisfies the second requirement of ECM, QP being roughly comparable to IP in terms of its case properties. In fact, making QPs completely parallel to other phrases leads to an interesting solution to the problem of the internal structure of numeral phrases. Case is standardly argued to be assigned by the heads of various categories to the noun phrases that they govern. Assimilating numerals to this model, one would ideally like the numeral to be a head Q that takes an NP complement, as in (55): (55)

[QP [Q, [Q pjat'] [NP rublej]]]

The relationship between the Q and the NP is thus identical to that between any verb or preposition and its object. In other words, I am claiming that QP is a functional category along the lines of much similar current work with the theory of phrase structure,34 and that quantified NPs are properly regarded as complements to Qs.There are, however, two reasons why the precise structure in (55) is inadequate for Russian numeral phrases, although both have solutions that follow straight forwardly from recent proposals about phrase structure. First, notice that structure (55) does not help very much in assimilating po to ECM constructions. It sheds no light

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

151

on why po is able to govern the numeral as the specifier of its complement, since pjat' is not only not a specifier, but not even a phrase. To remedy this situation, (55) must be revised so that pjaf is in fact the specifier of the QP, as in (56). (56)

[QP pjat' [Q. [Q e] [NP rublej]]]

In other words, the QP is headed by an empty quantifier and the numeral is actually its specifier.35 Hence, putting (56) after po, as in (57), results in po being able to assign its structural DAT(Q) to pjati and the null quantifier [Q e] in turn to assign its structural GEN(Q) to the NP ruble).

PP

(57)

P1

p po

QP NP:DAT(Q)

pjati

Q'

g

NP:GEN(Q)

e

rublej

The idea that the numeral could be the specifier rather than the head of the QP is not particularly radical. A QP is a kind of operator phrase, and much work since Chomsky (1986b) contends that it is generally true that the lexical material in an operator phrase can be in either the specifier or head position, or sometimes both, with obligatory SPEC-head agreement. For example, interrogative sentences, analyzed as CPs, typically have the overt [ + WH] material in the CP-specifier, as in (58). (58) a. [CP when; [c, [ C[+WH] willj] [,P John tj leave tj]] b. I wonder [CP when; f c , [Q+WHJ e] tip J°hn kft ^H In English, WH-movement is movement to the specifier of CP, but in order for a clause to be interpreted as interrogative its head must be [ + WH]. In (58b), for example, wonder selects for a [ + WH] complement, but CP will be [ + WH] only if its head C is also [ + WH], even if that head is lexically empty, This is a standard example of "SPEC-head agreement." More recently, Ouhalla (1990, 1991) has argued that in negation phrases the negation element can be either the head or the specifier of the negation phrase, with the other position being lexically empty. He uses this to account for variation in the position of the negation element in different

152

PARAMETERS OF SLAVIC MORPHOSYNTAX

languages.36 Adapting these ideas to the analysis of Russian QPs in (56) and (57) thus makes perfect sense.37 The other problem with (55) is that it still obscures Pesetsky's contrast between numeral phrases that are maximally QPs versus those that are NPs. Since we want the relation between Q and its NP complement to be constant, regardless of whether maximal QP or NP behavior is exhibited, the solution must lie in building up some additional structure above the QP. This can be easily accomplished, however, within the current conception of a nominal phrase as projecting up higher functional categories. Following Abney (1987), it is now widely accepted that NPs are actually embedded in determiner phrases (DPs), with the head D taking an NP complement.38 In line with this hypothesis, I propose that QPs may be embedded in DPs. That is, in addition to the structure in (56), the structure in (59) also exists.

(59)

[DP to- to e] [QP pjat' [Q. [Q e] [NP ruble)]]]]]

Numeral phrases that I have been analyzing as QPs have the structure in (59), but those that were analyzed in chapter 4 as NPs are actually DPs with the structure in (59). Assuming this distinction, placing a DP rather than a QP after the preposition po protects the numeral from ECM by the preposition, since the numeral is no longer the specifier of the complement. This structure is given in (60).

PP

(60)

p.

P

DP

po

D' QP

D

e

Q1

pjat'

Q

NP:GEN(Q)

e

rublej

One might then ask what case pjat' 'five' in (60) is, if this involves a DP complement to a preposition. Given standard assumptions about case assignment, pjat' should in fact have no source for case. We are therefore led to the not unreasonable conclusion that it is caseless, that it is a frozen form. Note that this runs contrary

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

153

to the traditional wisdom that it is accusative, although caseless quantifiers have occasionally been argued for (cf. e.g. Fowler, 1987a). However, if this were an accusative position, there would be no way to explain why unambiguously accusative numerals cannot appear here. Consider the behavior of tysjaca 'thousand', repeated as (61).39 (61)

po tysjace/*tysjacu rublej DIST thousand(dat)/(acc) rubles(gen pi) 'a thousand rubles each'

The dative is the only viable form in (61), suggesting that hereto can only take a QP complement, never a DP one. The reason, I suggest, is simply that Russian does not countenance a caseless form of tysjaca, so that the DP option is necessarily suppressed. It is worth noting that Russian tysjaca elsewhere always appears in the required case, with NOM tysjaca as a subject and ACC tysjacu as an object. Both of these cases can be assigned under ECM into the specifier of the QP, which is where I claim tysjaca is, the QP and its empty head being caseless. I now turn to the subject-verb agreement properties of Russian/w-phrases. Some examples of /w-phrase subjects of transitive and unergative verbs are given in (62); (62a) is cited in Babby (1985, 92). (62) a. na kazdom stule sidelo/*sideli po odnomu mal'ciku on each chair sat(n)/(pl) DIST one(dat) boy(dat) b. kazduju knigu procitalo/*procitali po pjat' studentov each book(acc) read(n)/(pl) DIST five students(gen pi) c. na kazdom zavode rabotalo/*rabotali po sto celovek at each factory(loc) worked(n)/(pl) DIST hundred people(gen pi) In line with his general theory, Pesetsky (1982) considers po-phrases to be invariably QPs, and thus erroneously assumes the neuter to be impossible in such examples for the same ECP reason as with other QP subjects. He argues for this on the dubious grounds that /w-phrases (1) never induce plural agreement on the verb, (2) cannot appear as subjects of transitive and unergative verbs, and (3) must undergo QR. As with other numeral phrases, as discussed in chapter 4, however, these claims are somewhat misleading. Most significantly, although the plural verb form is indeed unacceptable in these examples, the neuter singular is a viable option. This fact follows if we assume that /w-phrases in Russian are never DPs, since only DPs induce subject-verb agreement. However, Pesetsky's restriction of pophrases to QPs is merely a stipulation designed to accommodate the data; a more principled explanation is clearly called for. Under my account the agreement data follow straightforwardly: po-phrases in Russian are only PPs, never DPs or QPs. They are therefore expected to behave just as PP subjects do in general. The lack of a clear-cut unaccusative/unergative-transitive asymmetry and the general failure of subject—verb agreement immediately follows. Note that the actual S-structure position occupied by PP subjects is immaterial to the analysis; wherever they are, the point remains that />o-phrases can function as semantic subjects and, as such, they exhibit the subject-verb agreement behavior expected of PP subjects generally.

154

PARAMETERS OF SLAVIC MORPHOSYNTAX

The facts are somewhat more complicated than indicated by the data in (62), however. Ostensibly agreeing (plural) verbs, it turns out, are acceptable under certain circumstances. The two relevant factors are the cardinality of the numeral and the animacy of the noun. Consider the following examples: (63)

a. za behind b. kazdyj each

kazdym stolom sidelo/sideli po dva each table sat(n)/(pl) DIST two ucebnik procitalo/procitali po tri text(acc) read(n)/(pl) DIST three

ucenika pupils(gen) studenta students(gen)

With paucal numerals and animate subjects the plural is also possible.40 This effect is somewhat mitigated by inanimate paucal subjects, as shown by (64). (64)

na kazdom stole stojalo/?stojali po dva stakana on each table stood(n)/(pl) DIST two glasses(gen)

These examples are in clear contrast to otherwise identical ones involving/yaf' 'five' and above, in which only the neuter singular is acceptable, whether the numeral is oblique or not. I do not know why this should be so if these are all PPs, although the possibility of plural agreement per se is not especially problematic, since examples with plural agreement with other PP subjects is attested (cf. e.g. Babby, 1985).41 Agreement with PP subjects is a complex issue that I have so far suppressed since it introduces an unwanted complexity into the data. The problem is that under certain circumstances plural agreement with PP subjects is admissible, as discussed, for example, by Chvany (1975), Crockett (1976b), Babby (1980b, 1985), and Neidle (1988). Babby (1980b, 34), for example, shows that sometimes the plural (alongside the expected neuter singular) is acceptable with quantificational PP subjects, citing examples with subjects such as okolo cetyrexsot predstavitelej 'about 400(gen) representatives(gen pi)' and do trexsot oficerov 'up to 300(gen) officers(gen pi)', in addition to occasional /w-phrases. I take this to be an instance of semantic agreement, just as in Serbo-Croatian, so that the form of the verb reflects the plurality of the lexical head noun, despite the syntactic fact that it is technically an oblique complement to a functional Q head. I thus maintain that, whatever the factors may be for licensing optional plural semantic agreement with quantified PP subjects, pophrase subjects behave similarly to other PP subjects in this regard, supporting my claim that they are simply PPs. The functional analysis of QPs suggests certain further revisions to the analysis of chapter 4. For one thing, the DP theory of Abney (1987) treats adjectives as heads taking NP complements, resulting in a schematic structure as follows:

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

(65)

155

DP

QP

D

AP

Q

A

NP

Note that this analysis treats Qs, Ds, and As in a structurally parallel fashion.42 This makes some sense in that Russian numerals are essentially adjectival in oblique positions. That is, they do not belong to a syntactically uniform class: when they agree they are pure As and when they govern they are Qs. Determiners are also more or less adjectival in Russian, there being no morphologically distinct or uniform class of Ds. For example, we have seen that determiner-like elements can occur internal to QP, as in (66a), or can iterate up to semantic acceptability, as in (66b). (66)

a. [DP [QP pjat' [AP etix [AP krasivyx [NP devusek]]]]] five these(gen pi) beautiful(gen pi) girls(gen pi) b- top [APetot [AP m°J [AP ocurl this(nom sg) my(nom sg) one(nom sg) [AP staryj [NP drug]]]]]] old(nom sg) friend(nom sg)

For the sake of discussion, I have labeled the phrases headed by demonstratives, possessives, and odin 'one' APs, although they are technically distinct functional categories that overlap in features with adjectives but also presumably bear additional grammatical features; these might alternatively all be represented as iterated DPs. As argued by Neidle (1988), numbers in homogenous numeral phrases must be modifiers since they decline. I agree with her that as modifiers they lack a NOMACC form, although I differ in that as Qs they are caseless rather than NOM-ACC.43 This is reasonable both in that a few other adjectival quantifiers, such as skol'kixl*skol'kie 'how-many(gen)/(nom)' and neskol'kixl*neskol'kie 'several(gen)/(nom)', also lack a NOM-ACC, and in that this fact can be stated in terms of there being no [-oblique] adjectival forms. QPs (and other functional categories associated with nouns) are thus part of an "extended" nominal projection, roughly in the sense of Grimshaw (1991). Note that the problem of headedness in numeral phrases described in Babby (1987) disappears from this perspective: it is perfectly consistent to understand the noun always as a semantic head (i.e. "lexical," in Grimshaw's terms) but the numeral always as a purely syntactic head (i.e. "functional").44 This idea requires some clarification of the percolation mechanism employed in chapter 4 to explain the Russian heter-

PARAMETERS OF SLAVIC MORPHOSYNTAX

156

ogenous/homogenous case pattern, however. Assume not only that maximal projections are coindexed with their heads but also that functional categories dominating a substantive maximal projection also bear the same index. This is necessary for case features to percolate down from DP and pronominal features up from N, assuming that coindexation is a prerequisite for percolation. However, since I have argued that QPs lack case features, it must be possible for them to "transmit" these features down the tree without actually acquiring them. The existence of such case transmission follows from the nature of agreement and a literal interpretation of the case feature model. Given the Case Filter, DPs universally contain case features (even if just [ ± case]). Whether other nodes contain case features is a matter of great linguistic variation, however. Nothing in UG forces Ns, let alone As or Ds, to have case features. Since it is perfectly reasonable for a language to mark Ns with case but not As, it must be possible for AP also to transmit case features in the same manner QP does. A similar approach is required to handle case transmission by obligatorily controlled PRO from its controller to an agreeing predicate adjective in Slavic, to be discussed in the next chapter.451 conclude that the assumption that QP is caseless does not present any special problems for percolation. In this light, let us now return to the problem of pre- and postquantifier adjectives treated by Babby (1987) and the solution I proposed in chapter 4 based on the idea of adjective raising due to Corbett (1979). All that need be assumed is that adjectives with wide scope can move up to D. In (66a), for example, the demonstrative can remain in place and be marked genitive in the scope of the quantifier, or it can raise to D and be marked NOM/ACC. Genitive prequantifiers, on the other hand, originate inside QP to the left of numeral, possibly as in (67). (67)

[DP [QP dobryx [[QP pjat' ] [NP butylok]]]] good(gen pi) five bottles(gen pi) 'a good five bottles'

Dobryx is presumably adjoined to QP (or, if it is an X°, to Q), rather than in specifier position, if my analysis of pjat' as a specifier is correct. The point is that since the entire phrase could be either a DP, as in (67), or a QP,46 either agreement pattern is fine. (68)

dobryx pjat' butylok stojalo/stojali na stole good(gen pi) five bottles(gen pi) stood(n/pl) on table

Consider also the interaction of the paucal numerals with the rule that animate accusative 0-declension and plural nouns appear in the genitive, as discussed, for example, in Neidle (1988). Although obligatory for non-quantified DPs, the animacy rule is in principle optional for numeral phrases quantified by 'two', 'three', and 'four'.47 The examples in (69) therefore have the following structures: (69)

a. ja I b. ja I

videl [QP [Q cetyre] [NP soldata]] saw four soldiers(pauc) videl [DP [D [AP [A cetyrex [NP soldat]]]]] saw four(gen) soldiers(gen pi)

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

157

Neidle's system requires a similar contrast between Q and A. Note that whenever a demonstrative is present, regardless of the cardinality of the numeral, the animacy rule necessarily applies, since in order for the demonstrative to move to prequantifier position the numeral phrase must be a DP. To my mind, the behavior of paucals thus strongly suggests that they are heads rather than specifiers of QP. Finally, the analysis in this section raises a further potential aspect of variation that remains to be explored. The possibility exists that numerals in different languages might vary in whether they occupy the specifier or head position of QP. I can think of no compelling argument why numerals should not be head Qs in SerboCroatian and Polish; this is also suggested by the far greater incidence of frozen, accusative-like forms in these languages. Even within Russian, I think there is good reason to believe that the paucal numerals at least are invariably heads. For one thing, they have many more adjectival properties than do the higher numerals (cf. Corbett, 1978, 1983a, for discussion). For another, the very fact that they govern a special form suggests that they are heads, assuming that government is a property of heads. Moreover, in compound numerals ending in paucals, it is the paucal numeral that determines the form of the following material, suggesting that it alone is relevant. In fact, the clearest reason for the non-paucal numerals ever to be QP-specifiers is to account for the ECM possibility in terms of the structure in (57). Since this phenomenon exists only in East Slavic, and even there is regarded as archaic by many speakers, it is conceivable that Slavic numerals are always heads of QPs, except in the Russian ECM construction, and that the ongoing loss of this phenomenon reflects the reanalysis of Qs in this construction as heads. The only place where this reanalysis would never apply is with tysjaca 'thousand', as in (61).48 5.2.1.4. Serbo-Croatian Po-Phrases The fact that distributive po-phrases appear freely in the other Slavic languages, regardless of whether their numeral phrases are QPs or not, also argues against Pesetsky's analysis. In the remainder of this section, I briefly discuss SerboCroatian, Polish, Czech, and Slovak. Of these languages, the most interesting case is probably that of Serbo-Croatian. As Dickey (1992) observes, the Serbo-Croatian distributive preposition po does not assign any specific case.49 Consider the following examples: (70)

a. svako razgovara sa po jednim kandidatom everyone(nom) speaks with DIST one(inst) candidate(inst) 'everyone is speaking with one candidate each' b. kupio sam tri knjige po uceniku bought AUX(lsg) three books DIST student(dat) 'I bought three books for each student' c. dobijali smo municiju od po jednog vojnika got AUX(lpl) ammunition from DIST one(gen) soldier(gen) od prilike svakih pola sata about every half hour 'we received ammunition from each soldier about every half hour'

158

PARAMETERS OF SLAVIC MORPHOSYNTAX

d. secam se po jednog dogadaja iz svakog grada I-remember REFL DIST one(gen) event(gen) from each town u kome sam zivio in which AUX(lsg) lived 'I remember an event from every town I have lived in' The case after po is invariably the case independently required of a DP in the position in question, which can in principle be any case at all. In (70a) this case is INST because of the preposition sa 'with', in (70b) DAT because the NP is an indirect object, in (70c) GEN because of the preposition od 'from', and in (70d) also GEN, as lexically required by the verb secati se 'to remember'. This situation is reminiscent of the Russian cto za/German wasfilr 'what kind of construction. It is reasonable to suppose that this state of affairs exists precisely because distributive po assigns no theta-role of its own. In this respect it differs from all other prepositions (except za in its 'what kind of function). Thus the fact that the distributive preposition po has unique case assignment properties in Russian, resulting from its quantificational rather than thematic force, is consistent with its more general behavior. In Russian, I have argued that po assigns a special structural dative. For Serbo-Croatian, it seems that case and thematic properties are even more divorced, in thatpo literally inherits the case-assigning properties of its configuration. That is, Serbo-Croatian PPs headed bypo can be generated in a wide variety of structural positions. As to be discussed in chapter 7, other PPs are intrinsically cased in the sense that their heads have specified case features and so satisfy the thetarole visibility aspect of the Case Filter automatically. For this reason, they do not receive case through government. Serbo-Croatian /w-phrases are unique, however, in that their head po has open case features, the value of which can be fixed only by context. They are thus "assigned" case in accordance with their position, which percolates down to po and is "transmitted" to the DP complement of po. It is this mechanism that renders Serbo-Croatian /w-phrases "transparent" to case assignment, producing the effect that the DP appears to be assigned case from outside the PP projection. As in Russian, Serbo-Croatian /w-phrases may also serve as PP subjects of all types of verbs—unaccusative, unergative, and transitive. Consider the following nominative examples with singular /w-phrases, supplied by Wayles Browne (personal communication): (71)

a. . . . u kojima se bira in which REFL elects(3sg) b. na polici stajala je po on shelf stood AUX(3sg) DIST

po jedan zastupnik DIST one(nom) representative(nom) jedna glavica crnoga luka one(nom) bulb(nom) onion(gen)

Unlike quantified DP subjects, however, the plural agreement option is not viable, as shown by (72b). (72)

a.

svaku each b. *svaku each

knjigu je procitalo book(acc) AUX(3sg) read(n sg) knjigu su procitali book(acc) AUX(3pl) read(m pi)

po pet ljudi DIST five people(gen pi) po pet ljudi DIST five people(gen pi)

QUANTIFIED STRUCTURES! POLISH AND OTHER PUZZLES

159

The marginal possibility of plural agreement without po was argued to follow from the assumption that numeral phrases are DPs in Serbo-Croatian. With po, however, they are PPs, which blocks agreement much more effectively. It is worth noting that, just as in Russian, other PP subjects behave similarly, as in (73). (73)

ovu knjigu je kupilo oko pet zena this book(acc) AUX(3sg) bought(n sg) about five women(gen pi)

Here, however, speakers find plural agreement slightly more felicitous, especially with SVO word order, as in (74). (74)

?*oko pet zena su kupile ovu knjigu about five women(genpl) AUX(3sg) bought(fpl) this book(acc)

I have no explanation for this contrast, although it may have to do with the distributive semantics of /w-phrases, if (following Pesetsky) this forces them to undergo QR. It is also worth observing that the paucal numerals in Serbo-Croatian also behave somewhat differently with respect to subject-verb agreement, just as in Russian. Judgments obtained for agreement with paucal /w-phrases are roughly as follows, although there is considerable variation across speakers: (75)

a.

za svakim stolom je sedelo po dva studenta at each table AUX(3sg) sat(n sg) DIST two student(gen)

b. za svakim stolom je sedelo po dve zene at each table AUX(3sg) sat(n sg) DIST two woman(gen) c. na svakom stolu je stajalo po dva casopisa on each table AUX(3sg) stood(n sg) DIST two journal(gen) d. ?na svakom stolu je stajalo po dve case on each table AUX(3sg) stood(n sg) DIST two glass(gen) (76)

a. za svakim stolom su sedeli po dva studenta at each table AUX(3pl) sat(m pi) DIST two student(gen) b. za svakim stolom su sedele po dve zene at each table AUX(3pl) sat(f pi) DIST two woman(gen) c. na svakom on each d. na svakom on each

stolu table stolu table

su AUX(3pl) su AUX(3pl)

stajala50 stood(n pi) stajale stood(f pi)

po DIST po DIST

dva casopisa two journals(gen) dve case two glass(gen)

These examples show masculine animate, feminine animate, masculine inanimate, and feminine inanimate subjects, respectively, with neuter singular "nonagreement" in (75) and plural agreement in (76). The singular is much preferred in (75a), the plural somewhat preferred in (76b) and (76c), and greatly preferred in (76d). The marginality of (75d) may have to do with the fact that case in (75d) is also the nominative plural form, although why this effect is diminished in the (b) examples is unclear. Finally, note that without po the neuter singular option in -o as in (75) is extremely marginal throughout, agreement as in (76) being the norm with

160

PARAMETERS OF SLAVIC MORPHOSYNTAX

paucal subjects. This shows thatpo has a clear effect in rendering non-agreement viable, in keeping with the idea that these are PPs, although the factors differentiating the paucal numerals from 'five' and above remain to be thoroughly investigated. 5.2.1.5. Polish Po-Phrases Polish/w-phrases, although not as complex as /w-phrases in Russian, are considerably more idiosyncratic than those in Serbo-Croatian. In its distributive function Polish po assigns the locative case to singular DPs, as in (77). (77)

a. po [DP:LOC (jednym) jabiku] DIST one(loc) apple(loc) b. po [DP:LOC (Jednej) gruszce] DIST one(loc) pear(loc)

With quantified DPs, on the other hand, distributive po appears to be assigning the accusative, as in (78). (78)

a. po [DP:AccPi?c sliwek] DIST five(nom/acc)plums(gen pi) b. po [DP:ACC pi?ciu studentow] DIST five(acc/gen)male-students(acc/gen pi) c. po [DP:ACC pi?c studentek] DIST five(nonVacc)female-students(gen pi) d. PO bpiAccdwa jabJka] DIST two(nom/acc) apple(nom/acc pi) e. po [DP:ACC dwoch studentow] DIST two(acc/gen) male-students(acc/gen pi)

The contrast between the singular NPs in (77) and these numerically quantified DPs is thus comparable to that found in Russian, except that in Polish distributive po assigns the locative rather than the dative case to singular NPs. Polish po, however, does not exhibit the added complexity found in Russian of optionally assigning its case only to the quantifier. Collocations such as (79) are not grammatical in Polish. (79)

*po pi^ciu ksiazek DIST five(loc) books(gen pi)

That is, po cannot assign locative to a numeral that itself assigns GEN(Q) to the following nominal material. This follows if, as I have argued, po-phrases are true PPs and, unlike in Russian, Polish numeral phrases are always maximally DPs. The ECM possibility discussed for Russian po-phrases is thus unavailable, so that the options in (77) and (78) are thus the only viable ones in Polish. Notice that the DPs following po in (78) are unambiguously accusative. In the glosses I have indicated a range of possibilities, but the only consistent treatment is to take the quantified DPs as accusative to explain the apparent nominative-genitive alternation. When the head noun is virile, as in (78b, e), the genitive form is used,

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

161

and when it is non-virile, as in (78a, c, d), the nominative form is used. In keeping with the analysis developed for Polish in section 5.1.1, this means the quantified DPs must be intrinsically accusative. They are thus completely compatible with the traditional assumption that po assigns locative or accusative, depending on the cardinality of the following DP and, although I have found little mention of distributive po-phrases in the Polish literature, this seems to be the generally assumed analysis (cf. Lojasiewicz, 1979). It is possible to circumvent this stipulation, however, by invoking the idea that Polish po actually assigns a structural locative case, comparable to the [ — oblique] DAT proposed for Russian. If so, the fact that this stractural LOG is obligatory in (77) but is necessarily overridden by GEN(Q) in (78a-c) follows immediately, just as with the structural dative for Russian. The need to block ACC assignment in (77) and LOG in (78) by stipulation for the most part disappears.51 This analysis conforms to established properties of structural cases in that a DP bearing some other case may still appear in the structural case position and satisfy all grammatical requirements. It was shown, for example, that intrinsically accusative quantified DPs in Polish move to nominative subject position, even though they clearly remain accusative, as demonstrated by their morphology and their inability to induce subject-verb agreement. They cannot appear in oblique case positions, since inherent case must be syntactically realized, presumably in order to satisfy the Theta-Criterion. This, then, is the elusive thematic contrast between structural and inherent case assignors: stractural case can be satisfied by any otherwise well-formed argument, whereas inherent case can be satisfied only by a DP in the appropriate inherent case. Returning in this light to the po-phrases in (77), it becomes clear that if Polish quantified DPs are invariably accusative, then the only way for them to follow po is for po to assign a stractural case. Therefore, either it consistently assigns structural locative, or else it assigns inherent locative or stractural accusative depending on the cardinality of its complement. I opt for the first alternative since it seems conceptually simpler and accords with the analysis of Russian, concluding in favor of the idea that distributive po assigns stractural LOG in Polish and stractural DAT in Russian.52 The distribution of po-phrases in Polish is the same as in the other Slavic languages considered in that they freely appear in subject and object positions. (80)

a. kazdy student miaf po (jednej) dolarze each student had DIST one(loc) dollar(loc) b. z drzew pad-to po (jednym) jaWku from trees fell(n sg) DIST one(loc) apple(loc) c. po (jednym) studencie przyszlo do biura DIST one(loc) student(loc) arrived(n sg) to office d. po (jednym) studencie siedzialow biurze DIST one(loc) student(loc) sat(n sg) in office e. po (jednym) studencie czytaJo te ksiazki DIST one(loc) student(loc) read(n sg) these(acc) books(acc)

162

PARAMETERS OF SLAVIC MORPHOSYNTAX

In addition to the typical object function in (80a), Polish po-phrases may in principle also serve as subject of unaccusative (80b, c),53 unergative (80d), and transitive (80e) verbs. Since these are always PPs, when po-phrases are subjects the verb appears in the third person neuter singular (non-agreeing) form. Similarly, the singular DP in (80e) could be replaced with a higher virile DP, as in (81a, b) or a non-virile one, as in (81c, d), without incurring any other change. (81)

a. po dwoch studentow czytalo te ksiazki DIST two(acc) student(acc pi) read(n sg) these(acc) books(acc) b. po pieciu studentow czytalo te ksiazki DIST one(acc) students(acc pi) read(n sg) these(acc) books(acc) c. po dwie studentki c/ytalo te ksiazki DIST two(acc) female-students(acc pi) read(n sg) these(acc) books(acc) d. po piec studentek czytaio te ksiazki DIST five(acc) female-students(gen pi) read(n sg) these(acc) books(acc)

The verb remains in its non-agreeing form throughout. As in the other languages, with the paucal numerals the situation is slightly more complicated than with 'five' and above. Although the neuter is always preferred, as in (81), indicating that Polish po-phrases are indeed PPs, with dwa 'two', trzy 'three', and cztery 'four' the plural is also sometimes possible. Lojasiewicz (1979, 154) cites the following examples with plural agreement: (82)

a. w pokojach be.da po dwa fotele in rooms will-be(3pl) DIST two armchairs(nom/acc pi) b. na lawkach lezaly po trzy arkusze papieru on shelves lay(nonvir pi) DIST three sheets(nom/acc pi) paper(gen)

Why should this option be available? It seems to me to have to do with the fact that Polish dwa 'two', trzy 'three', and cztery 'four' are agreeing adjectival forms that occur with nominative plural noun forms. We saw in section 5.1.1 that when they stand on their own as subjects they require plural subject-verb agreement. Now note that it is precisely in those forms where the numeral phrase following po is nondistinct from the nominative plural that agreement becomes a viable option. It is thus conceivable that the verb is somehow agreeing with the apparent nominative plural subject DP, ignoring the fact that it is embedded in a PP. Moreover, according to speakers I have consulted, although the examples in (82) are indeed acceptable, a neuter singular verb is far preferable, especially in (82b), so that judgments might be "?" for (82a) and "?*" for (82b). This difference appears to be attributable to tense, since replacing lezaty 'lay(nonvirile pi)' with lezq 'lies(Spl)' improves (82b) somewhat. Here I believe that the relevant distinction has to do with the salience of plurality in the endings for the two tenses, in that in the past there are distinct tense and gender/number suffixes, whereas in the present a single ending expresses both tense and person/number. The formal conflict between a PP subject and plural verb form is thus more salient in (82b) than in (82a). Word order is also an important factor here, since when the po-phrase precedes the verb only the

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

163

neuter third person singular is acceptable. That is, when the PP remains in canonical subject position it must induce the non-agreeing verb form, but when it moves from that position (or never moves into it, if the verbs in [82] are unaccusative) plural agreement may arise. This suggests that the potential nominative status of the NP following po is somewhat more accessible the further the/w-phrase actually is from (nominative) subject position. Curiously, this is the opposite pattern from the general relationship between the position of the subject and the form of the verb found in Russian. This is because whether the Russian verb agrees or not depends on the category of the "subject" numeral phrase, which for unaccusative verbs may even remain after the verb when a QP. In Polish, on the other hand, there is no choice: the category of quantified subjects is always invariable. They are ordinary DPs, so must move from VP-specifier to IP-specifier position, although when quantified by 'five' and above they are accusative, not nominative. After po they are PPs, hence do not induce subject-verb agreement.54 There is, finally, one curious divergence from the general paradigm treated above. Lojasiewicz (1979) describes a mysterious quirk in the government pattern of Polish po. Essentially, when a single po takes a series of conjoined numerals or DPs as its complement, it must govern consistently. Hence, one finds the locative even on plural quantified DPs, as shown in (83a).55 (83)

a.

dostaniecie you-will-get b. *dostaniecie you-will-get

po jednym, DIST one(loc) po jednym, DIST one(loc)

dwoch jabfkach two(loc) apples(loc pi) dwa jablka two(nom/acc) apples(nom/acc pi)

Once po assigns locative this case must percolate throughout its complement, as in (50a); mixed case realization, as in (83b), is impossible. With conjoined quantified DPs, the form is determined by the first one, as shown in (84). (84)

a. dostaniecie po jednym jab-tku, dwoch gruszkach you-will-get DIST one(loc) apple(loc) two(loc) pears(loc pi) i pieciu sliwek and five(loc) plums(loc pi) b. dostaniecie po dwie gruszki, jedno jablko you-will-get DIST two(acc) pears(accpl) one(acc) apple(acc) i piec sliwek and five(acc) plums(gen pi)

It is worth noting that Russian behaves quite differently, as the following contrast cited by Mel'cuk (1985, 239) attests: (85)

a.

po DIST b. *po DIST

odnomu, one(dat) odnomu, one(dat)

dva posetitelja two(nom/acc) visitor(gen) dvum posetiteljam two(dat) visitors(dat pi)

Although I do not understand the source of this pattern in Polish, its absence in Russian suggests that it cannot have to do purely with the nature of conjunction, and

164

PARAMETERS OF SLAVIC MORPHOSYNTAX

the difference between the two languages may provide the key to an analysis. I suspect that the difference between Polish (83b) and Russian (85a) has to do with the fact that Polish QPs only appear in accusative DPs, so that once the locative is instantiated onjednym, a case conflict of some sort arises. 5.2.1.6. Other West Slavic Languages In this chapter I have analyzed po-phrases as PPs, accounting for the unusual government patterns of Russian and Polish in terms of the idea that the ' 'quantificational'' preposition po assigns a special [ — oblique] case and explaining the case transparency of Serbo-Croatian in terms of the idea that po is unspecified for case feature values in that language—a possibility admitted precisely because po is quantificational rather than thematic. In light of this variation, it is worth noting yet another possibility: that/w simply assigns a garden variety oblique case. It turns out that this possibility is instantiated by Czech, in which po always governs the locative, as in (86). (86)

a. dali gave b. dali gave c. dali gave

nam po jednom novem kapesniku us DIST one(loc) new(loc) handkerchief(loc) nam po dvou(ch)56 novych kapesmcich us DIST two(loc) new(loc) handkerchiefs(loc) nam po peti novych kapesnfcich us DIST five(loc) new(loc) handkerchiefs(loc)

The contrast with Polish indicates that distributive po in Czech is assigning the ordinary [ +oblique] locative. Notice that this aspect of distributive po in Czech accords with our conclusion that GEN(Q) is similarly unequivocally inherent in that language and suggests that there may be a broader correlation between the [ ± oblique] status of the cases assigned by numerals and po in any given Slavic language. In crude terms, numerals and po assign [ — oblique] case in Russian and [ +oblique] case in Czech. In Polish, however, the situation is not as clear, since both numerals andpo may assign [ — oblique] case, although numerals generally do not and po perhaps always does. SerboCroatian, in which numerals assign [ +oblique] case and/70 assigns no case of its own, stands somewhat apart but still seems unproblematic for any potential correlation. Turning to the other West Slavic languages, Lower Serbian po should assign a [ — oblique] case and, given the variation found in Upper Serbian, one would expect similar complications in the syntax of po. Lindseth (1993a), however, claims that in both Sorbian languages po always assigns the ordinary (i.e. inherent) LOG case, citing the following examples from Upper Sorbian: (87)

a. tfi kroc po styrjoch tydzenjach w jastwje sedzal three times DIST four(loc) weeks(loc) in jail he-sat b. po tfoch hriwnach dostac DIST three(loc) marks(loc) to-get

This demonstrates that the status of the case assigned in the two quantificational situations is not necessarily the same. Instead, the case assigned by numerals and by

QUANTIFIED STRUCTURES! POLISH AND OTHER PUZZLES

165

prepositions—including po—is a lexical property that seems to vary freely and independently between [±oblique]. Finally, Slovak, in which GEN(Q) is oblique, might be expected to pattern like Czech (86). Unfortunately, the data are unclear, and the sources I have consulted cite multiple possibilities. Mistrik (1988, 69) offers examples with the locative for ordinary numerals,57 as in (88), but the prescriptive Oravec, Bajzikova, and Furdfk (1984,102) is careful to point out that one should not use the locative with numerals greater than 'one', as in (89). (88)

a. po jednom DIST one(loc) b. po dvoch DIST two(loc)

(89)

a. po (jednej) korune DIST one(loc) crown(loc) b. po DIST c. po DIST

tri koruny three crowns(nom-acc pi) pat' korun five crowns(gen pi)

Mistrik offers (88b) and Lindseth (1993a) cites (90a), while Oravec, Bajzikova, and Furdflc warn against (88b) and (90b) as not prescribed by the "contemporary codification." (90)

a. po DIST b. po DIST

piatich jablkoch five(loc) apples(loc pi) piatich korunach five(loc) crowns(loc pi)

I therefore think it is fair to conclude that Slovak po assigns both [ ± oblique] LOG, with the current situation in flux. Since both possibilities exist for GEN(Q) in closely related languages, this state of affairs for Slovak LOC(Q) is hardly surprising.

5.2.2. Approximative Inversion Another place where Russian differs from Serbo-Croatian and Polish is in what may be called the approximative inversion construction. This construction is described in considerable detail in Mel'cuk (1985, 146-161), from which I have freely drawn in this section. The reader is referred to Mel'cuk for further discussion of restrictions on approximative inversion, as well as copious examples. In Russian quantified expressions, the noun may appear before the numeral, which results in an approximative reading, as in (91).58 (91)

a. my tarn zili goda dva we there lived year(gen) two 'we lived there about two years'

166

PARAMETERS OF SLAVIC MORPHOSYNTAX

b. Ivan s"el buterbrodov pjat' Ivan ate sandwiches(gen pi) five 'Ivan ate about five sandwiches' In (91a) approximative inversion applies to the adjunct phrase dva goda 'two years(gen)' and in (91b) to the argument phrase pjat' buterbrodov 'five sandwiches(gen pi)'. Note that the form of the noun always depends directly on the government requirements of the numeral. The examples in (92) involving numeral phrase subjects indicate that the possibility of inversion is much clearer for QPs than for quantified DPs, and, as before, it applies equally to subjects of unaccusative, unergative, and transitive verbs.59 (92)

a. celovek pjat' prislo/?prisli na zasedanie people(gen pi) five arrived(n)/(pl) at meeting 'about five people came to the meeting' b. rabotalo/?*rabotali v etom magazine celovek pjat' worked(n)/(pl) in this store people(gen pi) five' 'about five people worked in this store' c. studentov pjat' sdalo/??sdali ek/amen students(gen pi) five passed(n/pl) exam 'about five students passed the exam'

Since the neuter singular is far more acceptable than the agreeing plural in (92), I tentatively conclude that approximative inversion does not take place in numeral phrases that are maximally DPs. Further evidence for this restriction can be found in the fact that inversion is impossible in (93), since phrases modified by 'one' are only DPs and higher numbers like milliard 'billion' (as well as million 'million' and sometimes tysjaca 'thousand', especially in compound numerals) are head nouns. (93)

a. *oni privezli tonnu odnu they brought ton(acc) one(acc) ' they brought about one ton' b. *u menja dollarov dva milliarda vosem'sot by me dollars(gen pi) two billion(gen) eight-hundred ' I have about two billion eight hundred dollars'

Consider also the fact that an approximative inversion phrase cannot antecede drug druga: (94)

studentov pjat' *pomogali/*pomogalo drug drugu na ekzamene students(genpl) five helped(pl)/(n) each other(dat) on exam 'about five students were helping each other on the exam'

Although there is nothing semantically wrong with (94), both options are equally unacceptable, since adjunction of N requires a QP but the reciprocal requires a DP. Further corroboration for my claim that approximative inversion constructions are QPs is that they similarly cannot appear in any of the other obligatory DP positions discussed in chapter 4. Consider for example (95), the gloss again indicating semantic well-formedness.

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

(95)

167

zenScin pjat' *staralis'/*staralos' kupit' etu knigu women(gen pi) five tried(pl)/(n) to-buy this book 'about five women tried to buy this book'

Approximative inversion can also be used to support my analysis of the two possibilities in (69) in terms of the contrast between QP and DP. The reason is that approximative inversion forces taking the QP option when a paucal numeral is followed by a masculine animate noun; compare (96) with (69): (96)

a. ja videl soldata cetyre I saw soldiers(pauc) four b. *ja videl soldat cetyrex I saw soldiers(gen pi) four(gen)

Since the animacy rule only applies to DPs but adjunction does not take place in argument DPs, (96b) cannot be derived, as desired. Approximative inversion is also generally not viable when the quantified expression is oblique, which presumably follows from the fact that these are invariably DPs, as shown by the examples in (97). (97)

a. *direktor upravljaet fabrikami PJat'Ju director manages factories(inst pi) five(inst) ' the director manages about five factories' b. *on vladeet jazykami pjat'ju he possesses languages(inst pi) five(inst) ' he speaks about five languages' c. *Ivan boitsja c"elovek pjati Ivan fears people(gen pi) five(gen) ' Ivan fears about five people' d. *on dal podarki devuskam pjati60 he gave presents girls(dat pi) five(dat) ' he gave presents to about five girls'

The verbs upravljat' 'to manage' and vladet' 'to possess' take instrumental complements, bojat'sja 'to fear' assigns the genitive, and dat' 'to give' takes a dative indirect object. Although it is clear that approximative inversion moves nominal material to the left, its precise analysis is far from obvious. Moreover, as we shall soon see, the two open questions of What moves? and Where does it move to? lead to theoretically inconsistent answers. Fortunately, it is fairly easy to demonstrate that it is the head N° that adjoins to some category to its left. It is clearly not N' or the entire NP that is moving, since it impossible to bring along with the head a modifying AP. (98)

a. *interesnyx celovek dvadcat' interesting(gen pi) people(gen pi) twenty ' about twenty interesting people'

168

PARAMETERS OF SLAVIC MORPHOSYNTAX

b. *tjazelyx let pjat' difficult(gen pi) years(genpl) five ' about five difficult years' The same is true for adnominal complements to N, as shown in (99), although these do not seem quite as bad. (99)

?*ucebnika pofizike tri textbook(gen) in physics(dat) three ' about three physics textbooks'

In order to derive (98) and (99), the entire NP would have to move. Note that since po fizike is a complement in (99), appeal to Abney's structure in (65) will not suffice. On the contrary, the approximative inversion facts of Russian argue against (65) and in favor of keeping (non-functional) APs within the NP projection. Moreover, it is sometimes possible to leave a modifying AP or complement PP to N behind, as in (100). This possibility is improved by contrastive stress on the adjective and implies that a conjunction will follow with contrasting information. (100)

a. ?knig dvadcat' interesnyx books(gen pi) twenty interesting(gen pi) 'about twenty interesting books' b. ucebnika tri po fizike textbook(gen) three in physics(dat) 'about three physics textbooks'

Finally, although PP adjuncts cannot front along with the N, they can remain after the numeral under contrast: (101)

a. *devusek s karimi glazami pjat' girls(gen pi) with brown(inst) eyes(inst) five ' about five girls with brown eyes' b. devusek pjat' s karimi glazami girls(gen pi) five with brown(inst) eyes(inst) ' about five girls with brown eyes'

While the facts in (98)-(101) all lead to the conclusion that approximative inversion is an instance of head movement, reconciling this movement with the functional structure adopted earlier in this chapter is no straightforward matter. The problem is that there is a standard claim in the literature, adopted in Chomsky (1986a) and stemming from Baltin's (1982) Like-Attracts Like Constraint, that adjunction must respect X-bar structure in the sense that heads and phrasal categories are required to adjoin to other heads and phrasal categories, respectively. This assumption forces the structure in (102), in which N raises to adjoin to the left of Q.

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES (102)

169

QP

Q' NP

Q

Nj

ei

Q

Recall, however, that I argued that Russian numerals appear as specifiers rather than heads of QP. Adjoining to the left of Q, as in (102), does not therefore actually place the noun before the numeral. Consider next the alternative structure in (103). QP

(103)

Ni

QP

XP

[+Q] Q

Q1 NP e;

Here the relative order of elements is correct, but the structure does not respect the Like-Attracts Like Constraint. There are two ways of resolving this dilemma: either adjust the "numeral as specifier" analysis developed to accommodate Russian pophrases, or modify the theory of adjunction to admit (103).61 Given the inconclusiveness of the Russian data combined with the lack of a definitive theory of adjunction, I leave the choice between them as an open puzzle. Although one should keep both structures in mind throughout this section, in what follows I will assume for ease of exposition the head movement analysis in (102), together with its implication that numerals are heads of QP. It is interesting to note that the possibility of approximative inversion also exists for PPs, with the noun adjoining directly to the P or PP. Consider the following examples, where the fronted nouns appear in whatever form they would have if they had not been fronted:

170

(104)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a. ja vernus' 6asa derez dva I return hour(gen) in two 'I shall return in about two hours' b. my pridem casov v pjat' we arrive hours(genpl) at five' 'we shall return at about five o'clock'

(105)

a. on zivet metrax v sta ot nas he lives meters(loc pi) in hundred(loc) from us 'he lives about a hundred meters from us' b. ja vernus' casam k dvurn I return hours(datpl) toward two(dat) 'I shall return by about two o'clock'

In (104) the prepositions govern (structural) accusative so the fronted nouns bear GEN(Q), while in (105) the prepositions govern (inherent) oblique cases so the nouns bear the appropriate oblique case. These can be analyzed as adjunction to P (or PP). Assuming the Head Movement Constraint of Travis (1984) and Baker (1988), movement presumably proceeds in steps, with the noun first adjoining to Q and then to D before finally adjoining to P. This derivation is supported by the colloquial alternative in which the fronted noun follows the preposition, as in (106). (106)

cerez c"asov pjat' in hours(genpl) five 'in about five hours'

It thus seems that the N must raise successive cyclically (in the literary language, to the highest available head), "excorporating" from intermediate adjunction sites. As Roberts (1991) shows, such excorporation is allowed precisely where incorporation does not involve genuine affixation. Since remaining adjoined to D for some reason leads to unacceptability, the examples in (92) with DPs are degraded in acceptability. Note that the colloquial variant in which the noun remains between the preposition and the numeral is not possible if the preposition assigns oblique case, as in (107a, b), or if the preposition indirectly assigns the theta-role of the verb, as in (107c, d). Compare these with (105b), (111), and (114a, b), respectively. (107)

a.

b.

c.

*ja vernus' k casam dvum I return toward hours(datpl) two(dat) ' I shall return by about two o'clock' *on zil v gorodax pjati he lived in cities(loc pi) five(loc) ' he lived in about five cities' *ja vsegda nadejalsja na celovek pjat' I always relied on people(gen pi) five ' I have always relied on about five people'

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

171

d. ?*ja verju v bogov pjat' I believe in gods(gen pi) five ' I believe in about five gods' Interestingly, doubling of the preposition is also possible in colloquial speech, and this process can be used to "save" the examples in (107). (108)

a. cerez casov cerez pjat' in hours(genpl) in five 'in about five hours' b. ja vernus' k casam k dvum I return toward hours(datpl) toward two(dat) 'I shall return by about two o'clock'

This suggests that the option of doubling the preposition is actually base-generated. Although the head movement analysis is conceptually superior to one in which movement adjoins the noun to an XP, it is difficult to construct examples that clearly discriminate between these two types of analysis. If we suppose that adverbs are in Spec-PP, however, then it is worth noting that (109a), although redundantly approximative, is somewhat preferable to (109b). (109)

a. ja vernus' priblizitel'no casov I return approximately hours(genpl) ' I shall return in about five hours' b. ?ja vernus' casov priblizitel'no I return hours(genpl) approximately

cerez pjat' in five cerez pjat' in five

In (109a) casov adjoins to the left of the P cerez, and thus appears immediately to the right of priblizitel'no; (109b) can only have a completely different (and presumably scrambled) structure.62 Like approximative inversion in general, approximative inversion out of PPs is subject to some variability and certain restrictions. In rough terms, this process is inhibited when the PP is an argument of the verb, so that the examples in (110) are degraded in acceptability and some speakers find them completely unacceptable. (110) a. ??on rabotal zadacami nad pjat'ju he worked problems(inst pi) on five(inst) 'he worked on about five problems' b. ?ja nuzdajus' dollarax v pjati I need dollars(loc pi) in five(loc) 'I am in need of about five dollars' Here, the verb rabotaf 'to work' strongly governs the preposition nad 'on' and the verb nuzdat'sja 'to need' the preposition v 'in', in the sense that they require these prepositions to assign their theta-roles. On the other hand, constructions such as (111) are perfectly acceptable for most speakers.63 (Ill)

on rabotal magazinax v pjati he worked stores(loc pi) in five(loc) 'he worked in about five stores'

172

PARAMETERS OF SLAVIC MORPHOSYNTAX

I attribute this contrast to the fact that the location argument in (111) is not part of the obligatory predicate-argument structure of the verb rabotat' 'to work'. One problem is thus why approximative inversion should be more restricted for PP arguments, since it applies to QP arguments freely, as demonstrated at the beginning of this section. There is, nonetheless, a clear adjunct/argument asymmetry for DPs. This is revealed by a consideration of oblique quantified adjuncts, which, unlike the oblique arguments in (97), allow approximative inversion. Compare the example in (112) with (97a). (112)

ubili ego vystrelami pjat'ju killed him shots(inst pi) five(inst) 'they killed him with about five shots'

The felicity of inversion in (112) demonstrates that the noun can adjoin to an adjunct D (or DP), although not to an argument one. Another productive type of approximative inversion with oblique DPs occurs on genitive adjuncts to nouns, as in (113). (113)

zenscina let dvadcati pjati women years(genpl) twenty five(gen) 'a women of approximately twenty-five years of age'

It is interesting to note that this adjunct/argument asymmetry is reminiscent of the standard GB assumption, as expressed in Chomsky (1986a, 6), for example, that adjunction is only possible to non-argument maximal projections.64 However, this prohibition is normally understood as applying to XP, not X°, adjunction. Nonetheless, I suspect that the approximative inversion facts can be made to follow from theta-theoretic restrictions on adjunction, suitably extended to heads.65 Even so, there is a residue of instances where approximative inversion adjunction appears also to apply to arguments. This is the problem posed by QP arguments, which, we have seen, freely admit inversion. Given the overall import of the prohibition against adjunction to argument XPs, it would be desirable to seek a way of assimilating adjunction to Q (or QP) arguments to the general pattern. Unfortunately, I do not see any simple way of treating these as adjuncts so as to allow them to participate in the approximate inversion construction. However, one possibility— under the assumption of section 5.2.1 that the numeral is actually in Spec-QP— might be to argue that the noun is adjoining to the head of the numeral phrase in Spec-QP. This would be a way of reconciling the Like-Attracts Like Constraint with the structure proposed for Russian QPs. Conceivably, it is the "adjoined to Q" structure that gives rise to the approximative meaning, perhaps because it involves a kind of "quantificational circularity'' in that the noun over which the numeral ranges is adjoined to the numeral itself. A further puzzle is why approximative inversion onto argument PPs is fully grammatical if the preposition assigns structural rather than inherent case, as in (114). (114)

a. ja vsegda nadejalsja celovek na pjat' 1 always relied people(gen pi) on five 'I have always relied on about five people'

QUANTIFIED STRUCTURES! POLISH AND OTHER PUZZLES

173

b. ja verju bogov v pjat' I believe gods(gen pi) in five 'I believe in about five gods' c. deti polucili jablok po pjat'/pjati children received apples(gen pi) DIST five(acc/dat) 'the children received about five apples each' The felicity of (114c) with dative pjati is particularly interesting in view of the ECM analysis adopted earlier in this chapter, since it shows that approximative inversion is compatible with the QP having a filled specifier. This supports my conclusion that, whatever else may be unclear about this mysterious construction, approximative inversion is an instance of head movement. Be that as it may, the examples in (114) are in marked contrast to argument PPs in which the preposition assigns an (inherent) oblique case, as was seen in (110), and it is far from obvious how to treat these consistently. Perhaps PPs are never true arguments, so that even when the preposition is strongly governed by the verb it is the verb together with the preposition that assigns the theta-role. Thus nothing in principle ever prevents adjunction to a P (or PP), and the difference between (110) and (114) must be sought in terms of the structural/inherent dichotomy. This idea might be combined with the suggestion in the previous paragraph that approximative inversion out of QPs proceeds via adjunction to the head of a numeral phrase in specifier position. However, I leave the feasibility of this kind of approach open, since the full implications of motivating and extending the prohibition against adjunction remain to be investigated. Finally, as Mel'cuk (1985, 156) notes, the approximative inversion phenomenon only exists in the East Slavic languages. Ukrainian examples, such as (115), are comparable to those discussed for Russian. (115)

a. pryjslo colovik visim arrived people(gen pi) eight 'about eight people arrived' b. vin procytav storinok desjat' he read pages(genpl) ten 'he read about ten pages'

Wayles Browne (personal communication) informs me that Ukrainian 'two', 'three', and 'four' also provide interesting confirmation of this account. In this language the paucal numerals are agreeing forms and thus occur with nominative plural nouns, as in (116a); in this respect it is unlike Russian but like Polish. However, when approximative inversion applies, the genitive re-emerges, as in (116b). (116)

a. dva dolary two dollars(nom pi) b. dolariv dva dollars(gen pi) two

174

PARAMETERS OF SLAVIC MORPHOSYNTAX

Assuming that the absence of GEN(Q) in (116a) indicates that the phrase is a pure DP, the contrast in (116) is a welcome result, supporting my contention that approximative inversion applies only to argument QPs, not DPs.66 In contrast to East Slavic, this construction is completely nonexistent in South Slavic and West Slavic languages. Apparently, the possibility of approximative inversion in a language somehow hinges on the existence of QPs in that language. It remains mysterious, however, why QPs should only serve as an escape hatch for approximative inversion if they can also serve as maximal extended (functional) projections of N. 5.2.3. Frozen Quantifiers I have left intentionally vague the question of whether numerals that do not explicitly show a case ending are actually cased (as nominative or accusative) or are frozen, caseless forms, since the Russian facts are far from definitive in this regard. Other Slavic languages, however, display clear signs that certain quantifiers are frozen forms. In this section I briefly survey the relevant facts and some of their implications. Frozen quantifier forms are in general identical to the accusative, which is in keeping with my treatment of this case as the least marked, that is, with minus values for all case features. While most quantifiers belong to declensional types with -0 nominative/accusative endings, a few belong to the -a declension, which has distinct and unambiguous nominative and accusative endings. The case status of quantifiers can thus be assessed by considering the distribution of such items. Although the Russian facts are inconsistent, they suggest that frozen quantifiers are becoming increasingly viable. Here I will consider two such items, the numeral tysjaca 'a thousand' and the numeric classifier para 'a couple'. The behavior of the numeral tysjaca 'a thousand' indicates that Russian numbers do indeed bear case. Recall that in nominative contexts it is unequivocally nominative and in accusative contexts unequivocally accusative: (117) a. ja cital tysjacu/*tysjaca knig I read thousand(acc)/(nom) books(gen pi) b. tysjaca/*tysjacu knig lezali/?*lezalo/?lezala na stole thousand(nom)/(acc) books(gen pi) lay(n sg)/(pl)/(f sg) on table The unacceptability of the inappropriate form shows that tysjaca must be receiving case. Presumably, as a noun, case is necessary for its morphological realization. If QPs are caseless, as argued in chapter 4, then tysjaca is contained within a DP. However, this would fail to explain the marginal acceptability of non-agreement in (117b), as well as the possibility of feminine singular agreement.67 Femine singular agreement with tysjaca is possible only when it is treated as a head noun, having the meaning 'one thousand'. In this respect, note that tysjaca differs from true nominal quantifiers such as million, which always induces masculine singular agreement. Consequently, it may but need not be a head noun. This is corroborated by the contrast in (118).

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

(118)

a. v in b. v in

175

tysjace knig/knigax thousand books(gen pl)/(loc pi) millione knig/*knigax million books(gen pl)/(loc pi)

In the first variant in (118a) tysjaia is a noun taking an adnominal genitive, in the second it is an adjectival quantifier. Returning to (117b), a quandary therefore arises. The feminine agreement possibility follows straightforwardly from tysjaca as a head noun, but in order to explain either plural agreement or neuter singular non-agreement (to the extent this is indeed acceptable), we must assume the numeral phrase may be either DP or QP, as with other numerals. But if the entire phrase can also be a QP, why is nominative tysjaca the only viable option and, in particular, how does it get marked nominative? There are two conceivable ways of resolving this puzzle: either numerals must be allowed to receive case even though contained in phrases that are maximally QPs, or the QP hypothesis should be rejected for tysjaca. On the basis of the analysis of /w-phrases, I have argued for the first alternative, which depends on allowing case assignment directly into a QP. The second possibility would require us to treat all numeral phrases quantified by tysjaca as DPs, although they would presumably differ from the normal situation in that tysjaca would be the head. One way of checking this idea might be to see whether the QP tests adduced for pjat' in section 4.3.2 hold for tysjaca as well. It turns out, however, that as far as I can tell judgments are consistent with respect to these two tests, although this is complicated by the fact that speakers vary in accepting non-agreement with subject tysjaca at all. Note two other potential problems for this second alternative. First, some other way of getting neuter subject-verb agreement would have to be invoked if tysjaca never heads a QP, which would shed doubt on one important motivating factor for the DP/QP dichotomy itself. The variation in this regard might be accommodated by claiming that some speakers allow tysjaca to project a QP, perhaps under specific conditions, whereas others do not. Second, demonstratives work as expected in numeral phrases involving tysjaca. An initial demonstrative generally agrees in number with the head noun and forces a plural verb; the alternative in which tysjaca is a head noun is also viable, with the demonstrative then agreeing and appearing in the feminine singular.68 (119)

a. eti tysjaca knig lezali na stole these(nom pi) thousand(nom) books(gen pi) lay(pl) on table b. eta tysjaca knig lezala na stole this(nom f ) thousand(nom f) books(gen pi) lay(f) on table

These alternatives are the only ones possible. The subject phrases contrast in that in (119a) knig functions as the head in determining the plural form of the demonstrative and verb, whereas in (119b) tysjaca is clearly the head, requiring these to be feminine singular. In both, nonetheless, the case of tysjaca is necessarily nominative. It thus must be that tysjaca can somehow receive this case directly, even when embedded inside a QP, which leads us back to the ECM analysis. Locating tysjaca in SPEC-QP, however, leads to a problem: Why should non-

176

PARAMETERS OF SLAVIC MORPHOSYNTAX

agreement be marginal with tysjaca subjects when it is perfectly fine with pjat' subjects? Clearly, the solution to this puzzle should rely on the fact that tysjaca is a distinctly nominative form, while pjat' is not. Recall in this light the curious fact that accusative tysjacu cannot be used in />o-phrases; only the dative tysjace is acceptable. Sincepo assigns structural dative and the form tysjacu is only accusative, never frozen, it should not be acceptable after po because it would receive either no case or the wrong case. However, the claim that the unacceptability of tysjacu after distributive po follows from its being accusative implies that pjat' 'five' should be taken as caseless in constructions likepo pjat' rublej 'DIST five rubles', since these are perfectly normal. Consistent with this contrast is the fact that non-agreement with subjects quantified by pjat' is far superior to non-agreement with subjects quantified by tysjaca. One potential problem with the ECM analysis of tysjaca has to do with Russian para 'a couple', which is more reasonably regarded as having a caseless formparw, since this can (for at least some speakers) appear as such in subject position: (120)

%paru celovek prislo/*prisli couple(acc) people(gen pi) arrived(n)/(pl)

Unlike tysjacu, the accusative-like paru for such speakers can be used in subject QPs, indicating that it is a frozen form. In fact, the impossibility of plural agreement shows that the frozen form paru only appears in QPs, never DPs. Since QPs are viable in structural case positions, one would expect paru to be able to appear after distributive po. Nonetheless, speakers who accept (120) still do not necessarily find paru acceptable in/w-phrases, as shown in (121). (121) po pare/*paru rublej DIST couple(dat)/(acc) rubles(gen pi) There may be some variation on this point, since at least one (generally very liberal speaker) did accept paru here, and more research is needed to establish the exact status and correlation of the various possibilities before adopting a definitive analysis. If, however, just as with tysjaca, only the dative is allowed in (121), some way of limiting the distribution of QPs may be required. I have argued that QPs are not assigned case, and the impossibility of QP appearing in a cased position was implicit in the analysis developed in chapter 4, in that I argued that QP never raises to IP-specifier position.69 That these are to some extent idiosyncratic properties of individual lexical items is further supported by the behavior of comparable words in Serbo-Croatian and Polish. As seen in chapter 4, Serbo-Croatian numerals are regularly frozen, invariable forms. Since, with the marginal exception of the paucal numerals, they govern inherent GEN(Q) and also do not themselves decline for case, the form in which they appear is fixed, as in the following examples: (122)

a. u [Dp:DAT-Loc P^ gradova] in fivecities(gen pi) b. od [DP:QEN stotinu Beogradana] from hundred(acc) Belgraders(gen pi)

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

177

c. [DP:NOM hiljadu ljudi] je procitalo ovu knigu thousand(acc) people(gen pi) AUX(3sg) read(n sg) this book d. [Dp:NOM tisucu ljudi] je doSlo thousand(acc) people(gen pi) AUX(3sg) arrived(n sg) e. u roku od [DP;QEN nedelju dana] in period of week(acc) day(gen pi) As in Russian, the frozen "accusative-like" nature of Serbo-Croatian numeric quantifiers is most visible in the few forms that historically belong to the a-declension, since the fixed form in which they appear is maximally distinct. In (I22a)pet 'five' is formally nominative/accusative, whereas stotinu 'hundred' in (122b), hiljadu 'thousand' in (122c), along with its eastern variant tisucu in (122d), and nedelju 'week' in (122e) are formally accusative, all regardless of the position of the DP containing them. I thus conclude that Serbo-Croatian numerals are frozen forms and that this form is the historical accusative.70 This loss of declension is a relatively old phenomenon, stabilized by about the end of the sixteenth century.71 Clear instances of numeric classifier nouns in this function are difficult to find in Serbo-Croatian. Par 'a couple' , for example, is of the 0-declension, and so resembles pet: (123)

a. [DP:NOMpar ljudi] Je doslo couple people(gen pi) AUX(3sg) arrived(n sg) b. od [DP:GEN P*«" ljudi] from couple people(gen pi) c- po [DPrAccpar Uudi] DIST couple people(gen pi)

It is thus similarly frozen in its NOM/ACC form in the examples in (123). Further evidence for this claim follows from the form of demonstratives, which do not agree with par, but rather behave as they do with numerals in general. I thus ascribe par the same structure as a numeral: (124) [DP:NOM ovih [QP par [NP:GEN(Q) ljudi]]] je do§lo these(genpl) couple people(gen pi) AUX(3sg) arrived(n sg) Of course, par can also serve as a noun rather than a quantifier. When it does, as in the following examples, it declines as expected: (125)

a. torn paru ljudi this(dat) pair(dat) people(gen pi) b. sa tim parom ljudi with this(inst) pair(inst) people(gen pi)

Reasonable a-declension candidates, such as masa 'a lot' , do not behave as frozen quantifiers for most speakers, functioning instead as head nouns with adnominal genitive complements.

178

PARAMETERS OF SLAVIC MORPHOSYNTAX

(126) a. ?*[Dp:NOM masu

ljudi]

je

doslo72

lot(acc) people(gen pi) AUX(3sg) arrived(n sg)

b-

*[DP:NOM masu ljudi] je kupilo ovu knigu lot(acc) people(gen pi) AUX(3sg) bought(n sg) this book

c.

*od [DP:GEN masu ljudi] from lot(acc) people(gen pi)

d. *ka[DP;GEN masu ljudi] to lot(acc) people(gen pi) (127)

a. [DP:NOM

masa ljudi] je dosla lot(nom) people(gen pi) AUX(3sg) arrived(f sg)

b- [DPINOM

masa ljudi] je kupila ovu knigu lot(nom) people(gen pi) AUX(3sg) bought(f sg) this book

c. od [DP:GEN mase ljudi] from lot(gen) people(gen pi) d. ka[DP:GEN masi ljudi] to lot(dat) people(gen pi) I thus conclude that Serbo-Croatian quantifiers are frozen caseless forms, and whether any given numeric classifier noun is analyzed as a quantifier or not is partly an idiosyncratic lexical property. Polish also displays the frozen quantifier phenomenon, in some ways more clearly than Serbo-Croatian. The bulk of the evidence suggests, however, that such items are accusative, rather than frozen in a form that looks like the accusative. It was already shown at length in section 5.1 that GEN(Q) is only assigned in accusative DPs in Polish, and I subsequently argued that such DPs may—in fact, must-—appear in some structural case position. With numerals proper the clearest argument for their accusative status was the fact that numerals alternate between nominative and genitive forms depending on the virility of the head noun. Yet the very fact that they exhibit this alternation seems to me to undermine any claims that they are themselves frozen forms. The behavior of specific lexical items is of some interest, however. Numerals proper, unfortunately, do not provide much evidence, since Polish tysiqc 'thousand' is formally masculine and so lacks an independent accusative. Numeric classifier nouns, on the other hand, show a much more robust unambiguous frozen accusative, always appearing as such in structural case positions. Items with this behavior include parq 'a couple', trochq 'a little', and masq 'a lot'.73 In addition to appearing in (accusative) direct object position, they may also serve as (nominative) subjects, as in (128), or (structural locative) objects of po, as in (129). (128)

a. tu bylo troche. pieniedzy here was(nsg) little(acc) money(gen) b. mase. kobiet przyszfo do biura lot(acc) women(genpl) arrived(nsg) to office

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

179

c. pare ludzi przeczytato te. ksiazke. couple people(gen pi) read(n sg) this book (129)

dostaniecie po par^ dolarow you-will-get DIST couple(acc) dollars(gen pi)

The formparq even appears in compound numerals (cf. Schenker, 1966, 241), such asparqnascie ('a couple on ten' = 11-19) and parqdziesiqt ('a couple tens' = 2099).74 Like true numerals in Polish, and unlike numerals in Serbo-Croatian, these forms cannot appear in oblique case positions. Instead, they decline and either agree, as in (130a), or govern the adnominal genitive, as in (130b). (130)

a. z paroma dolarami with couple(inst) dollars(inst pi) b. z masa dolarow with lot(inst) dollars(gen pi)

This contrast shows that numeric classifiers may function as (modifying) quantifiers or (head) nouns. Similarly, compare the two corresponding possibilities in (131). (131)

a. mase kobiet przeczytafo lot(acc) women(gen pi) read(n sg) b. masa kobiet przeczytala lot(nom) women(gen pl)read(f sg)

te. this 1$ this

ksiazke. book ksiazke. book

Final support for my claim that parq is technically a quantifier can be found in the agreement possibilities for prequantifiers. The demonstrative may either be GEN(Q) or NOM/ACC, as follows: (132)

a. te this(acc pi) b. tych this(genpl)

pare couple(acc) par$ couple(acc)

kobiet women(gen pi) kobiet women(gen pi)

As before, this result follows from the assumption that GEN(Q) may be either structural or inherent in Polish.

5.3. Bare Genitives In the final section of this chapter, I consider two sets of phenomona conceptually related to numerically quantified constructions: (1) empty quantifier structures, and (2) the genitive of negation. There is a vast literature on both of these constructions in the various Slavic languages, which I shall not attempt to survey here. Instead, I focus on their major properties, contrasting these construction types to the GEN(Q) proper and discussing differences between Russian and Polish in the genitive of negation. Since these constructions both involve genitive case, they have sometimes been analyzed as resulting from the same general process as GEN(Q). I shall argue,

180

PARAMETERS OF SLAVIC MORPHOSYNTAX

however, that whereas empty quantifier structures can be assimilated to GEN(Q), the genitive of negation cannot and thus represents a distinct phenomenon.

5.3.1. Empty Quantifier Structures There are two types of nominal constructions in Russian that exhibit genitive casemarking and have a clearly quantincational interpretation, yet lack an overt quantifier to assign the genitive case.75 The more familiar of these is the so-called partitive construction, in which a direct object appears in the genitive to impart a meaning of 'some' with mass nouns and, less commonly, 'many' with count nouns. The other construction involves a bare genitive in a presentential focus position and some quantificational element in object position that serves to restrict that quantity of the item expressed in the genitive. Following House (1982), I shall refer to this sentence type as a genitive initial sentence (GIS). I will argue that both the partitive and GIS construction should be analyzed in terms of a null quantifier that assigns GEN(Q) just as an overt quantifier would. 5.3.1.1. Partitives The partitive genitive typically appears on mass noun objects of certain verbs, as in the following representative Russian examples: (133)

a. dajte mne saxar/saxaru give me sugar(acc)/(gen2) b. Ivan vypil caj/caju Ivan drank tea(acc)/(gen2) c. ja nalil emu moloko/moloka I poured him milk(acc)/(gen) d. Masa vzjala xleb/xleba Masha took bread(acc)/(gen)

Both the accusative and partitive are possible here, but the latter has the additional implication 'some amount/portion of X'. Note that nouns which possess a genitive2 form in -u, as discussed in chapter 2, may (but need not) employ this form in partitive usages, as in (133a, b). Since such genitive2 forms are generally inadmissible in other contexts, I employ them as a test for partitivity.76 As Babby (1980b, 79-83) and Franks (1986) observe, although the partitive reading for such bare genitives is predominant, an alternative "large quantity" interpretation additionally exists. This possibility is illustrated in (134). (134)

a. on knig kupil he books(genpl) bought(m) 'he sure bought a lot of books' b. Ijudej sobralos' people(gen pi) gathered-together(n) 'a lot of people sure got together'

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

181

c. i zabot bylo and troubles(gen pi) was(n) 'there sure were enough troubles' d. urokov zadajut assignments(gen pi) assign(3pl) 'they sure give a lot of assignments' The use of the genitive here indicates that an inordinate number of the object in question was involved and is thus typically accompanied by expressive intonation and often focusing of the genitive phrase. This usage generally involves nouns in the genitive plural, where there is no special genitive! ending. Finally, as (134b, c) demonstrate and as treated in Babby (1980b), the partitive genitive applies with equal force to underlying objects of unaccusative verbs. It cannot be used, however, on subjects of transitives or unergatives. The factors governing the felicity of the partitive are complex and even differ among speakers. Klenin (1978,173), for example, states both conditions on the type of noun likely to appear in the partitive and, more importantly, restrictions on the class of appropriate verbs, namely, ' 'those that affect or effect a change in the status of their objects." For example, in contrast to (133b), the partitive genitive! form is impossible in (135), cited by Klenin (1978, 177), who based her discussion on observations in Paduceva (1974): (135)

on s"el sup/*supu he ate-up soup(acc)/(gen2)

The semantic similarity between the verbs vypit' 'to drink(perf)' and s"est' 'to eat(perf)' suggests that the relevant factor for licensing partitive complements is some reasonably subtle lexical property. Following Neidle (1988), I shall refer to this semantic property as [ + Q], meaning that the verb may have intrinsic quantificational force.77 Klenin (1978) further notes—again, relying on Paduceva (1974)—that perfective verbs often accept partitive objects while their imperfective counterparts may not, although this limitation sometimes appears to be overridden by sentential negation. Compare (136a) with (136b). (136)

a. Sa§a pil caj/*caju Sasha drank(impf) tea(acc)/(gen2) b. Sasa ne pil caj/caju Sasha NEG drank(impf) tea(acc)/(gen2)

In (133b) we saw that the perfective form of this verb allows a partitive complement, hence the verb is [ + Q]. Although the affirmative imperfective in (136a) does not admit the partitive caju, the negated imperfective in (136b) does. These examples show that imperfective verbs typically suppress f + Q], but the effect disappears under negation. Note that the acceptability of the partitive phrase in (136b) is not due solely to the presence of negation. This is demonstrated by (137), also from Klenin, in which negation fails to introduce the possibility of the partitive into (135).

182

(137)

PARAMETERS OF SLAVIC MORPHOSYNTAX

on ne s"el sup/supa/*supu he NEG ate-up soup(acc)/(gen)/(gen2)

Here, only the regular genitive of negation is acceptable, not the distinctively partitive genitive2 form in -u. The reason for the contrast between (136b) and (137) is simply that the verb (vy)pit' 'to drink' contains the [ + Q] feature whereas the verb (s")est' 'to eat' does not. How does the [ + Q] property interact with the partitive genitive to determine its interpretation and distribution? Following widespread practice and argument, as in Pesetsky (1982), Franks (1986), Neidle (1988), and Franks and Dziwirek (1993), I assume that the partitive phrase consists of a null Q assigning genitive to its sister NP, schematically as in (138). (138) [QP [Q e] [NP:GEN(Q) • • • 11 It is thus completely parallel to numeral phrases containing overt quantifiers. There are, however, two substantive differences that affect its form and distribution. For one thing, the special genitive2 -u ending on a subset of masculine singular nouns can be understood as assigned by a null Q.78 For another, unlike overt quantifiers, the null Q lacks a specific value and must, therefore, look elsewhere for its interpretation. Although the null Q is [ + Q], it must be coindexed with something else that can specify its quantificational force. As discussed in chapter 7, this is the identification requirement that all null elements must meet. In partitive constructions, I maintain that it is precisely the [ + Q] feature of the governing verb that satisfies this requirement. This analysis explains the observed distributional restrictions on partitive phrases. They can only appear with a limited class of licensing verbs and must be complements to those verbs in order for the [ + Q] feature to be accessible.79 This approach also explains the variation in interpretation, since the Q assigns essentially what Babby (1980b, 81), citing Bulaxovskij (1949, 262), terms the genitive of indefinite quantity. Thus the 'some' and 'many' readings are in principle equally conceivable. In this sense, then, such bare genitives can be regarded as Incomplete Quantified Expressions (IQEs), as proposed in Franks and House (1982), for the genitive initial sentences to be treated in section 5.3.1.2. They are inherently quantificational, but that quantification is technically incomplete and needs somehow to be specified. It is perhaps worth noting that the fact that [ + Q] can only identify a null Q in the verb's object requires a tighter formal restriction than mere government, since we have seen that for purposes of the ECP a QP in VP-specifier position is able to move long-distance (both in the syntax and LF) and so must be regarded as properly governed. If the relationship between the [ + Q] verb and the null Q is one of binding, however, then the stricter locality requirement follows straightforwardly, since the first-branching node definition of minimal c-command is the one relevant for binding theory (cf. e.g. Chomsky, 1986a, 8). Thus, a null Q inside the subject would not be eligible to be bound by [ 4- Q] on the verb since, although underlyingly within the verb's maximal projection VP, it lies outside V. Only if properly governed by V can the null quantifier be bound and identified by the verb's [ + Q] feature.

QUANTIFIED STRUCTURES! POLISH AND OTHER PUZZLES

183

The relationship between this [ + Q] feature and the null Q is examined in Franks and Dziwirek (1993). In this paper we concluded that even though a particular verb might bear the feature [ + Q], that feature must itself be activated by being in the scope of perfective aspect. To formalize this, we adopted the idea that the verb phrase is itself embedded in an aspect phrase (AspP); a [ + Q] V must then be minimally c-commanded by a [ + Perf] Asp. That is, the next higher head up the tree must be perfective aspect. The structure of (133b), with the partitive complement caju, can thus be represented as in (139).

(139)

IP

r

NP

i

AspP

[+past] [+3.sg.m]

Asp r+Perfl

VP QP

V

Ivan

vypil [+Q]

Q

NP

e

Caju

To accommodate the effect of negation in (136b), we proposed that the negation morpheme ne, just like perfective aspect, is also able to activate the [ + Q] feature on a verb. Following standard claims, as proposed for example in Ouhalla (1991), negation is also treated as a functional head, projecting its own phrase NegP. In general terms, then, the [ + Q] that identifies the null Q of the partitive phrase must be in the scope of some kind of quantificational operator, and both negation and perfective aspect can serve this purpose; see Franks and Dziwirek (1993) for details. Pesetsky (1982, 201-203) briefly considers Russian partitives and, in line with his QP hypothesis, tentatively suggests that the partitive phrase is a QP and that verbs that allow partitive complements select for a semantic type that he calls K (for Russian kvantor), as opposed to the majority of verbs, which select entity or T(erm) complements. Recall that Pesetsky argues that verbs which semantically select for T must have categorially DP (or, in his 1982 terms, NP) objects at LF.80 Verbs that semantically select for K, on the other hand, will have to be treated as taking either QP or DP complements, since both partitive and accusative objects are viable. While conceptually elegant in that it fills an otherwise mysterious gap in his paradigm, there are several serious problems with Pesetsky's ideas that (1) partitives

184

PARAMETERS OF SLAVIC MORPHOSYNTAX

are invariably QPs, (2) verbs that occur with partitives select for the semantic type K, and (3) the semantic type K can be categorially realized by a DP or QP.81 In fact, although the evidence is inconsistent, the bulk of it suggests that partitives not only can but must be DPs. Moreover, as Pesetsky (1982, 203) himself notes, his ECP account falsely predicts that partitives should be able to appear as subjects of verbs that semantically select K subjects, because then they could remain in this position at LF without violating categorial selection, or raise at LF without violating the ECP. The absence of this possibility suggests, as I have maintained following Neidle (1988), that the partitive crucially depends on some semantic feature of the verb that licenses and identifies the null Q in its object. If the possibility of the partitive phrase were the result of the verb semantically selecting complements of the type K, and K could be syntactically realized either as QP or DP, then we would expect DP objects of such verbs to have a hidden quantificational interpretation, just as DP objects of verbs that take question-type complements (e.g. ask the time) have a hidden interrogative interpretation. This is false for partitive-taking verbs, otherwise there would be no difference between accusative and partitive usages. We would also expect partitives to be ambiguously DPs or QPs, although there is no way of distinguishing them in object position. Moreover, partitives do not appear in subject (and other not properly governed) positions, a problem already noted for Pesetsky. As an illustration, consider Pesetsky's (1982, 202) arguments that a verb like dobavit' 'to add' may take QP complements, hence that it semantically selects for K. His example, with associated structure, is given in (140). (140)

povar dobavil [QP [Q e] [NP saxaru]] v smes' cook(nom) added sugar(gen2) into mixture 'the cook added some sugar to the mixture'

Although maintaining that the phrase is a QP, Pesetsky notes that dobavit' can also take DP complements, as in (141). (141)

povar dobavil [DP stakan saxaru] v smes' cook(nom) added glass sugar(gen2) into mixture 'the cook added a glass of sugar to the mixture'

Example (141) is ambiguous in the sense that stakan 'glass' could either delimit the amount of sugar added or itself be the object introduced into the mixture; I assume this ambiguity derives from stakan being either a Q or a head N. Pesetsky states, however, that the latter reading results from the fact that dobavit' also semantically selects for T complements. He is thus claiming that dobavit' selects for either K and T semantic types, both of which can be realized as a DP syntactic category, and that K additionally can be realized as QP—a range of possibilities that, to my mind, seriously undermines the explanatory force of his model. I have argued instead that partitives may be DPs and that their distribution results not from semantic selection but rather from being in a position where the value of the null quantifier can be identified. Example (140) is thus grammatical and unambiguous because the quantifier is bound by the [ + Q] feature of the verb, whereas (141) is ambiguous because stakan 'glass' can be taken as quantifying over saxaru 'sugar' or not.

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

185

Unlike QPs with overt numerals, partitive phrases never occur in VP-specifier position, whether of a transitive or intransitive verb. This follows from the assumption that partitives contain a null quantifier that must be identified, hence lexically governed by the verb. A consideration of partitives in passive constructions suggests that they can remain in caseless positions, however, in this respect resembling true QPs. For example, the following passive of (142) is acceptable: (142)

bylo dobavleno [QP [Q e] [NP saxaru]] v smes' was(n) added(n) sugar(gen2) into mixture 'some sugar was added to the mixture'

In this sentence the partitive certainly looks like a QP, since it appears to be in object position, with subject position occupied by a null expletive to induce neuter singular agreement. Alternatively, the partitive phrase could be a DP that then moves into SPEC-IP position82 in order to receive nominative case, just like any other DP object under passivization. The obligatory neuter singular subject-verb agreement would, then have to be forced independently of category, however. Either way, (142) shows that what is crucial about the distribution of partitives and other IQEs is that they must be underlying direct objects, that is, lexically governed by the verb. This conclusion is also supported by the fact that they can also serve as "subjects" of unaccusative verbs, as we saw in (134b), although partitives with unaccusatives give rise to the same range of questions partitives with passives do. It should nonetheless be clear that IQEs are not necessarily QPs, since they appear equally in other Slavic languages in which numeral phrases have been argued to be only DPs. For example, in both Serbo-Croatian and Polish the partitive genitive is used productively.83 If these are simply null quantifier numeral phrases, then they must be maximally DPs. One can then ask whether partitives in these languages are nonetheless consistent with passive verbs. It turns out that they are in SerboCroatian, but not in Polish. (143) a. ?dodala sam secera u smesu I-added(fsg) AUX(lsg) sugar(gen) in mixture 'I added some sugar to the mixture' b. u smesu je dodato secera in mixture AUX(3sg) added(n sg) sugar(gen) ' some sugar was added to the mixture' c. ?mesa se jelo meat(gen) REFL eat(n sg) ' some meat was being eaten' (144) a.

chleb zostal kupiony bread(nom) was(m sg) bought(m sg) b. *chleba zostaJo kupione bread(gen) was(n sg) bought(n sg)

The compatibility of Serbo-Croatian participial passive (143b) and reflexive passive (143c) with partitives is problematic, since if partitive phrases must be analyzed as DPs in this language, then they should not be expected to appear in caseless posi-

186

PARAMETERS OF SLAVIC MORPHOSYNTAX

tions such as object of a participial or reflexive passive verb. I therefore suggest that these are really surface subjects, having moved under passivization to subject position in search of nominative case. That is, the alternative treatment of Russian (142) proposed earlier is the only one viable for Serbo-Croatian and if also adopted for Russian would concomitantly eliminate any motivation for analyzing partitives as QPs. The absence of Polish (144b) further corroborates the idea that the partitives in Serbo-Croatian (143b, c) undergo raising, since there is considerable evidence to suggest that this passive construction in Polish is lexical and no NP-movement is involved in its derivation, as discussed in chapter 8. Partitives are therefore NPs (and conceivably, although not necessarily, also QPs in Russian) that contain null quantifiers and that must be D-structure objects of verbs able to license the null quantifier. 5.3.1.2. Genitive Initial Sentences I now turn to the second type of empty quantifier construction under consideration, the genitive initial sentence (GIS). This sentence type is often mentioned in the traditional literature and is treated in depth in Crockett (1976b, 318-335) and House (1982), both of whom survey existing discussion. Pesetsky (1982, 233-236) devotes an entire three-page footnote to GIS structures, cataloguing most of their puzzling properties. Here I will adopt the analysis of Franks and House (1982), recasting it in the current framework. Some typical examples are given in (145). (145)

a. mal'cikov ostalos'/*ostalis' sem' boys(genpl) remained(n)/(pl) seven 'the number of boys that remained was seven' b. interesnyx knig ja procital pjat' interesting(gen pi) books(gen pi) I(nom) read(m) five 'the number of interesting books I read was five'

In these sentences, as Crockett (1976b, 318) observes, "the quantifier is separated from the quantified noun and constitutes the focus of the sentence from the communicative standpoint." The initial bare genitive serves as a sort of quantificational theme, the cardinality of which must be satisfied by some element in an argument position in the sentence. It is thus an IQE with the rough meaning "talking about some number of X." In (145a) this corresponds to the "subject" of an unaccusative verb and in (145b) to the object of a transitive one. Of primary significance is the fact that these bare genitives differ from partitives in that they do not occupy an argument position. They must, nonetheless, be related to some element that is in an argument position and that addresses the quantity of the item expressed by the bare genitive. Although it might seem plausible to propose a movement analysis, Crockett (1976b) and Franks and House (1982) adduce a variety of types of evidence against the proposal that the genitives in (145) are generated after the numerals and subsequently fronted. The inadequacy of movement is demonstrated by pairs of items such as the following:

QUANTIFIED STRUCTURES'. POLISH AND OTHER PUZZLES

(146)

187

a. knig ja procital tol'ko odnu books(gen pi) I(nom) read(m) only one(acc) 'the number of books I read was only one' b. ja procital odnu knigu/*knig I(nom) read(m) one(acc) book(acc)/(gen pi) 'I read only one book'

(147)

a. sardinok v banke ostalos'/*ostalis' dve sardines(gen pi) in tin remained(n)/(pl) two 'the number of sardines that remained in the tin was two' b. v banke ostalos'/ostalis' dve sardinki/*sardinok in tin remained(n)/(pl) two sardines(gen)/(gen pi) 'two sardines remained in the tin'

(148)

a.

publiki na ploscadi bylo jabloku negde upast' public(gen) on square was(n) apple(dat) nowhere to-fall ' the public in the square was wall-to-wall' b. *na ploscadi bylo jabloku negde upast' publiki on square was(n) apple(dat) nowhere to-fall public(gen) ' in the square it was wall-to-wall public'

In each of these examples the putative source position does not provide the appropriate form; in (146) odnu 'one' does not govern GEN(Q), in (147) dve 'two' governs the (paucal) genitive singular, and in (148) the idiomatic phrase jabloku negde upast' 'not even room for an apple to fall' does not even accept a complement. Instead, all that is necessary is that the domain specified by the bare genitive be quantificationally delimited by some phrase in an argument position in the sentence. In other words, the bare genitive is an IQE of the now familiar form (149). (149) QP [ [Q e] [NP:GEN(Q) • • • I!

It is thus formally comparable to a partitive phrase, although the range of interpretations allowed and positions occupied differs. In particular, the null quantifier in GIS structures is not identified by the verb, but rather has its value determined by an appropriate phrase in some argument position. There is thus no limit to its variation; it can be satisfied not only by numerals or quantifiers per se, but also by anything that conceivably delimits its cardinality. Consider the following examples from Crockett (1976b): (150)

a. ljudej bylo devjat' celovek people(gen pi) was(n) nine persons(gen pi) 'the number of people was nine persons' b. vremeni proslo dve nedeli time(gen) passed(n) two weeks(gen) 'the amount of time that passed was two weeks' c. druzej u menja bylo vsego odna podruga friends(gen pi) by me was(n) only one(nom) girlfriend(nom)

188

PARAMETERS OF SLAVIC MORPHOSYNTAX

So long as celovek 'persons' ranges over ljudej 'people', nedeli 'weeks' over vremeni 'time', andpodruga 'girlfriend' over druzej 'friends', the sentences in (150) are well-formed. GIS structures, however, require more than just a minimal degree of contextual relatedness between the bare genitive and the quantificational element, as shown by the following contrast from Franks and House (1982); compare House (1982) for a comprehensive discussion of the possibilities: (151)

a. *milicionerov na stole lezalo dve furazki policemen(gen pi) on table lay(n) two service-caps ' the number of police that lay on the table was two caps' b.

milicionerov za stolom sidelo dva serzanta policemen(gen pi) at table sat(n) two sergeants ' the number of police that sat at the table was two sergeants'

Example (151a) fails because the bare genitive implies an ensuing quantification, but furazka 'service cap' does not denote policemen, even though dve furazki could be taken to imply that there were two policemen. Example (151b), on the other hand, succeeds because serzant 'sergeant' is a subset of milicionery 'police', that is, sergeants are a kind of policeman. In contrast, note that if (151a) were presented as a topic-comment structure, as in (152), it would be grammatical. (152)

milicionery, na stole lezalo dve furazki policemen(nom pi) on table lay(n) two service-caps '(talking about) police, on the table lay two service caps'

Since furazka 'service cap' is an article of clothing worn by police, it is easy to construct a suitable context for (152) in which the presence of the caps gives information about policemen. Moreover, the felicity of (152) shows that these bare genitives are formally not topics, which are taken to be base-generated CP-adjoined phrases. This conclusion is further supported by the fact that nominative topics are compatible with bare genitives, which they necessarily precede, as in the following example from Franks and House (1982): (153)

knigi /*knig, romanov ja prodital v obrez books(nom pl)/(gen pi) novels(gen pi) I read in edge '(speaking of) books, the number of novels I read was just enough'

Nor are they in either specifier or head position of CP, as shown by (154). (154)

a. eto student, kotoryj knig procital tol'ko odnu this student, who(nom) books(genpl) read(m) only one(acc) 'this is the student who read only one book' b. ja znaju, cto deneg u nego v obrez I know that money(gen pi) by him in edge 'I know that he has just enough money'

Instead, these non-argument bare genitives follow complementizers and interrogative and relative pronouns. They thus appear to be base-generated as adjoined to the

QUANTIFIED STRUCTURES! POLISH AND OTHER PUZZLES

189

left of IP. This positioning may account for their topic-like interpretation, leading us in Franks and House (1982) in fact to dub such IQEs genitive themes. The questions then arise of how these IQEs become associated with an argument position and how the null quantifiers they contain are supplied with an interpretation. The answer to the second question, I believe, leads to a natural account also for the first one. Unlike partitives, these IQEs are not properly governed by the verb, so some independent strategy must be adopted for interpreting the null Q. I suggest that an appropriate phrase in argument position must be able to bind the quantifier to give it some value. In order to do this, the phrase undergoes Quantifier Raising at LF to a position c-commanding the bare genitive, so that coindexing will result in proper binding. Consequently, the relationship between the moved phrase and the null quantifier is roughly as in (155) at LF for the sentences in (146a) and (147a). (155)

a. [IP [tol'ko odnu]; [IP [QP Cj knig]j [IP ja [VP procital BJ]]]] b. [IP [dve]j [IP [QP e; sardinok]j [IP e [VP v banke ostalos' Cj]]]]

The idea captured by the representation in (118) is that once the phrase in argument position adjoins to IP to bind the null quantifier, it itself leaves a trace that must be bound. The antecedent of the trace, however, is not the phrase that vacated that position, but rather the IQE, as reflected in the indices provided. This somewhat unorthodox analysis, according to which the trace of a moved phrase is bound not by what moved—which binds the null quantifier—but by the IQE instead, is nonetheless in keeping with by now standard assumptions about trace theory, as discussed in chapter 4. That is, if all properties of trace theory are epiphenomenal and, in particular, if coindexing is done freely by general rale rather than automatically in tandem with movement, then there is no compelling reason why the trace should not be able to be bound by some element other than the moved phrase, so long as all other principles of grammar are satisfied. The proposed analysis is also consistent with Pesetsky's idea that categorial selection does not take place until LF. This allows phrases of any type to occupy D-structure object position, provided that they vacate this position by LF. The only crucial factor is that the phrase have adequate quantificational force to be able to identify the null quantifier in the IQE. Thus, the examples cited include DP, AP, PP, and even clausal objects, all of which raise at LF to bind the null Q. If the trace is determined to be a DP, assuming this to be the Canonical Structural Realization of entities (or "terms"), and if proper binding requires that an antecedent and trace must be of the same category, then IQEs are DPs, not QPs.84 On the other hand, these bare genitives can only be associated with non-oblique object positions—a distribution that led Pesetsky (1982) to suggest tentatively that they were QPs. We have already seen in (147) and (148), for example, that they occur with "subjects" of unaccusative verbs, but not with subjects of unergative or transitive ones. Compare (119a) with (156b) and (156c). (156)

a. devusek prislo navecerinku kotnaplakal girls(gen pi) arrived(n) at party cat cried 'the number of girls who came to the party was paltry'

190

PARAMETERS OF SLAVIC MORPHOSYNTAX

b. *ucenyx rabotalo nad etoj zadacej kotnaplakal scientist(gen pi) worked(n) on this problem cat cried ' the number of scientists working on this problem was paltry' c. *devusek pocelovalo Vanju vsegoodna girls(gen pi) kissed(n) Vanya(acc) in-all one(nom) ' the number of girls who kissed Vanya was only one' The phrase that raises at LF must be a D-structure object, as in (156a). Subjects, as in (156b, c) are unavailable for IQE binding; the same is true of adjuncts, even when accusative, as shown by (157). (157)

*let ja zil v etom gorode vsego dva years(genpl) I lived(m) in this city in-all two ' the number of years I lived in this city is only two'

This pattern certainly seems to implicate the ECP, provided the effect of antecedent government is removed, since the IQE can only be interpreted as binding lexically governed positions. Note that I have derived this distribution by being particularly selective about the examples used. The reason for this selectivity is that one must be very cautious in evaluating putative GIS structures, since it is often possible to construct comparable sentences that do involve movement. The way to control for this, of course, is to use quantifiers that do not allow genitive plural complements, as I have done in (156) and (157). Had I usedpjat' 'five' as the quantifier in (156b, c) and (157) instead, the sentences may have been deemed felicitous, although with scrambled GEN(Q) NPs rather than base-generated IQEs.85 The fact that these IQEs cannot be associated with subject position must receive a different explanation than the one employed for partitives, since their null Qs are not licensed by properties of the specific verb. As QPs, in fact, one might expect them to be able to be associated with a quantificational phrase in subject position, as in the examples of non-agreement with transitive and unergative verbs discussed in chapter 4. Recall, however, that these QPs, which were argued to be in VP-specifier position, remained in situ. The IQEs under consideration, on the other hand, require LF movement. So if a quantificational subject phrase raised to bind the null quantifier of the IQE, its trace would violate the ECP if the IQE were somehow prevented from properly governing. Categorial mismatch, along the lines of Pesetsky (1982), is one credible candidate for the intervening factor; another might be absence of an appropriate lexical head, assuming with Lasnik and Saito (1992) that head government is always necessary for the ECP, as discussed in chapter 3. Although I have no strong cause to reject either sort of approach, there are at least two reasons to think that IQEs in genitive initial sentences should not be uniformly analyzed as QPs. The first is that, just as with partitives, there is no obvious way of preventing them from being DPs, so that the question of why this option should be excluded remains. Second, again as with partitives, other Slavic languages that I have argued lack QPs exhibit the GIS construction. However, as will be briefly discussed at the end of this section, GIS structures in Serbo-Croatian and Polish seem to be more restricted than their Russian counterparts. One particularly curious aspect of the GIS structure, noted by Crockett, House,

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

191

and Pesetsky, is that even when the "subject" appears in the nominative case, as in (150c) or (158), cited by Pesetsky (1982, 234), the verb still fails to display agreement. (158)

druzej prislo odna Masa friends(genpl) arrived(n) one(nom) Masha(nom) 'of my friends Masha alone came'

This discrepancy, unique in Russian syntax, coupled with the limitation of GIS structures to unaccusative verb frames, strongly suggests that these nominative "subjects" are not actually in IP-specifier position, and that some indirect mechanism for effecting nominative case assignment must be in operation. The need for some case is forced by morphological requirements—essentially, the PF side of the Case Filter—and conceivably nominative is assigned by virtue of coindexation with an expletive subject, although the details of this marked process are unclear. It should be noted, moreover, that some speakers find examples like (158) marginal, although accepting nominatives freely with purely existential verbs such as bylo 'was(n)' and ostalos' 'remained(n)'. This may be related to the ability of these verbs to take a second nominative in their copular functions (although the predicative instrumental is much preferred, especially with ostat'sja 'to remain'). It is also worth noting that Serbo-Croatian and Polish speakers offer similar judgments for comparable constructions in their languages. Despite the fact that the base-generated IQE is unable to antecedent-govern the trace of the moved element, whether that is due to categorial mismatch or to some other restriction on antecedent-government, the IQE clearly binds the trace in the sense of functioning for all semantic purposes as if it itself were the underlying direct object. So, for example, the IQE must satisfy all selectional restrictions imposed by the verb and may enter into idiomatic expressions, as in (159), from Franks and House (1982). (159)

fil'mov idet vsego tri lilms(gen pi) goes in-all three 'the number of movies showing is only three'

In short, the IQE is interpreted as the verb's direct object at LF, regardless of what occupies this position at D-structure. This fact is captured by the indexing assumed in (155). It also explains the prohibition against IQEs binding oblique positions, as in the ungrammatical (160), also drawn from Franks and House (1982).86 (160)

a. *druzej on dal pomosc' dvum friends(gen pi) he(nom) gave(m) help(acc) two(dat) ' the number of friends he helped was two' b. *fabrik on upravljaet dvumja factories(gen pi) he(nom) manages(3sg) two(inst) ' the number of factories he manages is two'

Under the present analysis, the oblique numeral would raise at LF, its trace being bound by the IQE. However, this representation would violate the Theta-Criterion at LF, assuming that the appropriate oblique case is prerequisite for theta-role visi-

192

PARAMETERS OF SLAVIC MORPHOSYNTAX

bility, as discussed in chapter 7. As an interesting aside, note that since the quirky case requirement is in fact met at both D-structure and S-structure in (160), the ungrammatically of these examples is crucial in demonstrating the significance of case for theta-role visibility at LF. Before turning to GIS structures in Serbo-Croatian and Polish, it should be pointed out that this term is somewhat of a misnomer in that the genitive IQE is not necessarily initial in its clause. For one thing, as already noted, topics, w/z-phrases, and complementizers precede it. For another, as observed by Crockett (1976b, 332334), functional considerations dictate not that the genitive phrase be initial but rather that the element that identifies its cardinality be final (which Crockett calls the quantifier phrase). As she puts it, "the order of the constituents in the surface sentence . . . is determined by their relative communicative weight. The quantifier phrase is the most informative element in the sentence and it must therefore be rightmost in linear order," although she then discusses several exceptions. Clearly, the IQE is generated left-adjoined to IP and, since thematic, remains there, while the quantificational object tends to scramble rightward. Crockett suggests, however, that scrambling may introduce other possibilities; two of her examples follow: (161)

a. zagotovleno silosa odna tonna is-ready(n) silage(gen) one(nom) ton(nom) 'the amount of silage that is ready is one ton' b. celovek dvadcat' ix sobralos' people(gen pi) twenty them(gen) gathered(n) 'there are about twenty of them gathered'

Note that silosa 'silage(gen)' in (161a) could not be a scrambled adnominal complement to odna tonna 'one ton(nom)', since the verb, which presumably scrambles to the left, lacks the expected feminine singular agreement that tonna (as a nominative subject) would impose. In (161b), the IQE is the pronominal form ix 'them(gen)', celovek dvadcat' 'about twenty people', which also presumably scrambles to the left, being an instance of approximative inversion. Of course, for a pronoun to be felicitous in this construction, in which the IQE is functionally thematic, it must be contextually determined as the topic of discussion. It is for this reason that Crockett cites example (161b) as occurring after the sentence ne odna ved' '(she) is not alone, after all'. The referent of ix is old information, even if thematic in the discourse, and celovek dvadcat' has been moved to the front by virtue of the functional sentence perspective process of "rhematic proposing," since the rheme often precedes the theme in colloquial Russian. Finally, I have much less information about this construction in other Slavic languages. Although judgments are less consistent and somewhat muddled, it does apparently exist outside of East Slavic. For Serbo-Croatian, the following were all deemed felicitous, although with slight reservations when the GIS is based on a transitive verb, in (162c, d). (162)

a. knjiga ima samo/svega jedna book(genpl) there-is(3sg) only/in-all one(nom) 'of books there is only one'

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

193

b. knjiga ima samo/svega dve/tri/pet book(genpl) there-is(Ssg) only/in-all two/three/five 'of books there are only two/three/five' c. ?knjiga sam kupila samo/svega jednu book(genpl) AUX(lsg) bought(fsg) only/in-all one(acc) ' of books I bought only one' d. Tknjiga sam kupila samo/svega dve/tri/pet book(genpl) AUX(lsg) bought(f sg) only/in-all two/three/five ' of books I bought only two/three/five' Without samo 'only' or svega 'in-all', the examples with the singular were judged unacceptable; no similarly striking contrast was displayed for numerals higher than jedan 'one'. (163) a. *knjiga ima jedna book(genpl) there-is(3sg) one(nom) ' of books there is one' b. knjiga ima dve/tri/pet book(gen pi) there-is(3sg) two/three/five ' of books there are two/three/five' c. ?*knjiga sam kupila jednu book(genpl) AUX(lsg) bought(f sg) one(acc) ' of books I bought one' d. ??knjiga sam kupila dve/tri/pet book(genpl) AUX(lsg) bought(f sg) two/three/five ' of books I bought two/three/five' This effect is perhaps pragmatic, since a comparison of (162d) and (163d) suggests that it also exists for 'two' and above, albeit considerably diminished. One might expect that the lower the cardinality, the more necessary an emphatic element such as samo 'only' or svega 'in-all' becomes. This does not explain why (163b) is completely acceptable, however. Perhaps this is related to the ability of ima 'thereis' to govern the genitive in general. Interestingly, (164a) was judged unacceptable, although as Zeljko Boskovic (personal communication) notes, adding the demonstrative tih in (164b) renders it much improved: (164) a. *knjiga smo kupili svega dva primerka books(gen pi) AUX(lpl) bought(m pi) in-all two items(gen) b. tih knjiga smo kupili svega dva primerka these books(gen pi) AUX(lpl) bought(m pi) in-all two items(gen) This suggests to me that movement should be implicated, since tih knjiga in (164b) is a much more credible complement to primerka than bare knjiga is in (164a). It is thus unclear whether or not such Serbo-Croatian examples are comparable to the Russian IQEs discussed above. Consider also the following judgments for passives with count noun IQEs:

194

(165)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a. ??knjiga je kupljeno dva primerka books(genpl) AUX(3sg) bought(nsg) two items ' the number of books that was bought is two' b.

?knjiga je kupljeno svega dva primerka books(gen pi) AUX(3sg) bought(n sg) in-all two items ' the number of books that was bought is in all two'

Svega 'in all' has the same amelioratory effect as before. As pointed out to me by Zeljko Boskovic, movement has a similar "focusing" effect; compare (166) with (165a): (166)

?knjiga je dva primerka kupljeno books(gen pi) AUX(3sg) two items bought(n sg)

Note finally that this construction is more readily acceptable for mass nouns, supporting the analysis that these genitives are IQEs formally identical to partitives, (167a) with an active verb and (167b) with a passive. (167)

a. mesa je kupila dva kilograma meat(gen) AUX(3sg) bought(f sg) two kilograms 'the amount of meat (she) bought was two kilograms' b. secera je potroseno dva kilograma sugar(gen) AUX(3sg) used-up(nsg) two kilograms 'the amount of sugar used up was two kilograms'

The superiority of (167) over (163d) is presumably related to the fact that SerboCroatian partitives generally apply to mass nouns, the 'large quantity' reading being very restricted, although it may be that the examples in (167) are perfectly fine because the possibility exists for the genitive noun to have simply been scrambled frontward from inside the numeral phrase. Serbo-Croatian thus appears to exhibit the GIS construction, although on a far more limited basis than in Russian, and further research into its distribution and semantics is called for. The situation in Polish is comparable. One speaker consulted provided the following judgments: (168)

a. *ksiazek kupilam jedna books(genpl) bought(lsgf) one(acc) ' the number of books I bought was one' b. ksiazek kupilam dwie/pi^c books(genpl) bought(lsgf) two/five ' the number of books I bought was two/five' c. mie.sa zjadfam dwie banki meat(gen) ate(lsgf) two tins ' the amount of meat I ate was two tins'

The same contrast observed for 'one' versus higher numerals as in Serbo-Croatian, and to a lesser extent among some speakers of Russian, is manifested by Polish. Here, too, the introduction of tylko 'only' leads to considerable improvement, although the result is not completely acceptable.

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

(169)

195

a. ?ksiazek kupifam tylko jedna books(gen pi) bought(lsg f) only one(acc) ' the number of books I bought was only one' b. ?zeszytow kupiJam tylko jeden notebooks(gen pi) bought(lsgf) only one(acc) ' the number of notebooks I bought was only one'

Whether there are any significant differences between Serbo-Croatian and Polish in this regard is unclear, both in the interpretation and acceptability of the construction.87 The relative status of GIS structures in Russian versus Serbo-Croatian and Polish is also debatable. Its existence outside of East Slavic implies that these IQEs need not be QPs, although, given that the QP hypothesis offered one way of accounting for the distribution of these bare genitives in Russian, it is conceivable that the analysis assumed in chapters 4 and 5 might need to be relaxed to allow for (nonargument only?) QPs in the other languages as well. This seems doubtful, however, for several reasons above and beyond the immediate problem of either restricting QPs in Serbo-Croatian and Polish to the appropriate position(s) or explaining why they lack the effects found in Russian. For one thing, the distributional and lexical parallelism with partitives in Serbo-Croatian and Polish would be lost, unless these too were analyzed as QPs. This would be a difficult move, as discussed in the previous section, since partitives do occupy argument positions. For another, the problem still remains of preventing these IQEs from being DPs, without which the ECP account of their distribution fails. I therefore conclude that IQEs are free to be of whatever category overt numeral phrases in a given language are (i.e. the minimal hypothesis), and that they cannot antecedent-govern for some independent reason. Whether this prohibition has to do with some intrinsic property of the IQE—perhaps that the quantifier is null—or with the fact that the trace in question is derivationally not actually that of the IQE is unclear. This assumption may shed some light, however, on the more restricted distribution of IQEs, especially GIS ones, outside of East Slavic—it is reasonable to suppose that this might follow from their limitation to DPs in Serbo-Croatian and Polish, but not in Russian. Ultimately, the solutions to this and the many other puzzles raised in this chapter await more detailed comparative study.

5.3.2. The Genitive of Negation A proper treatment of the genitive of negation in Slavic would require a book in and of itself, if not several volumes. In this final section on quantification, my more modest goal is merely to compare the major typological properties of this phenomenon in the various languages and to address the issue of whether or not the genitive of negation is fundamentally a variety of the quantificational genitive. I will argue that the two are independent processes and formally unrelated, despite certain syntactic and semantic similarities. I also present an interesting correlation between the productivity of genitive negated adjuncts in a language and the productivity of the partitive.

196

PARAMETERS OF SLAVIC MORPHOSYNTAX

5.3.2.1. Russian versus Polish Before comparing the genitive of negation construction among Slavic languages, I first take up general issues in its analysis within Russian. We have already had several occasions in this book to make reference to this phenomenon, which applies in Russian to assign genitive case to otherwise accusative direct objects of negated transitive verbs, as in (170). (170)

a. on ne videl etogo fil'ma he NEG saw this(gen) movie(gen) b. on svoix vescej ne nasel he selfs(genpl) things(gen pi) NEG found

In Russian the application of the genitive of negation rule is optional, so that the direct object in (170) could just as well have appeared in the accusative. The factors affecting this choice are extremely complex and have engendered a great amount of investigation and debate. Corbett (1986), in fact, has compiled an entire bibliography that catalogs works on this subject up to 1985, and many more are sure to appear; probably the best study of the specific factors determining the appropriateness of genitive or accusative objects of negated transitive verbs can be found in Timberlake (1975). Although the details of this problem are tangential to the central concerns of this section, it will be important to keep in mind the following two facts: 1. The genitive of negation applies only under sentential negation to NP sisters of V that are specifically within the scope of negation. 2. The optionality of the genitive of negation rule is not universal across Slavic. As noted in chapter 2, the genitive of negation rule also seems to apply to nonargument objects of V. That is, time and distance phrases that would be accusative in affirmative sentences may appear in the genitive under negation, just as do true complements, as follows:88 (171) Ivan ne cital ni minuty (nom) NEG read not-even minute(gen) What is most striking about this possibility is that it persists even when (1) the verb already has an object, (2) the verb is otherwise intransitive, or (3) the verb assigns a particular oblique case. This is illustrated in the examples in (172), from Franks (1990a). (172)

a. Ivan ne cital etoj knigi ni minuty (nom) NEG read this book(gen) not-even minute(gen) b. Boris ne rabotal nad etoj problemoj ni minuty (nom) NEG worked on this problem not-even minute(gen) c. general ne pravil stranoj ni odnogo goda general NEG ruled country(inst) not-even one year(gcn)

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

197

I thus tentatively conclude that genitive case under negation is assigned in Russian to any NP that meets the structural requirement "sister to neg-V," and that any V may in principle satisfy this. That is, the principles of structural case assignment apply with equal force to arguments and non-arguments.89 The only difference is that for non-arguments the proper structure is sufficient, since no theta-role assignment is involved, whereas for arguments other factors may be relevant, since theta-role visibility is required. For example, rabotat' 'to work' in (172b) cannot assign the theme theta-role directly, but requires instead the mediation of the preposition nod 'on', and pravit' 'to rule' in (172c) requires the instrumental in order to assign its theta-role. Thus, non-argument "objects" appear to get case more or less accidentally. They are blind to specific lexical properties of the verb, such as whether or not it assigns some special case or even no case at all, presumably because as adjuncts they are not theta-marked by the verb. This interesting observation will also serve as a point of departure for the analysis of the second dative phenomenon to be presented in the next chapter. Another interesting facet of the the genitive of negation rule in Russian is that it applies to subjects of unaccusatives and object of passives, as in (173). (173) a. ni priSlo ni odnogo Celoveka NEG arrived(n) not-even one person(gen) b. moroza ne cuvstvovalo-s' frost(gen) NEG was-felt(n) -REFL c. ne bylo opublikovano ni odnoj knigi NEG was(n) published(n) not-even one book(gen) These facts are discussed in great detail in the literature (cf. Chvany, 1975; Pesetsky, 1982; Neidle, 1982, 1988; and especially Babby, 1980b). Crucially, the genitive of negation never applies to subjects of transitive or unergative verbs. While Babby carefully treats the factors necessary for the application of the genitive-of-negation rule to "subjects," those that he eventually comes up with are characteristic of unaccusative contexts, and this is indeed the conclusion reached by other researchers. Pesetsky (1982) and Neidle (1982, 1988) demonstrate that the genitive of negation rule only applies VP-internally. The object of any verb in the scope of negation, even an intransitive one, may surface in the genitive, and in all the examples considered the genitive NPs are VP-internal. The genitive of negation rule in Russian is thus properly understood as dependent not on the case-assigning features of the verb, but rather solely on configuration. Their arguments that the genitive of negation pertains exclusively to objects are incontrovertible, I believe, but do not extend to quantified phrases, contrary to Pesetsky's account. Pesetsky claims that both are QPs and therefore cannot be D-structure subjects, given the ECP argument discussed in chapter 4. However, although genitive "subjects" under negation only appear with the highly restricted set of unaccusative verbs, we have seen that quantificational QPs (as evidenced by the lack of subject-verb agreement) may serve as D-structure subjects. This lack of correspondence in distribution leads me to reject Pesetsky's assimilation of the two phenernona.

198

PARAMETERS OF SLAVIC MORPHOSYNTAX

In addition to this rather basic distributional fact, there are a host of other arguments for treating the genitive of negation as formally distinct from GEN(Q). In doing so, I depart radically from Pesetsky's approach and thus will have to explain any similarities between the two constructions independently; I am, however, in keeping with Chvany (1975) and, in particular, Neidle (1988, 41^-7), who offers several reasons for contrasting the genitives of quantification and negation. Here I review some of her more compelling arguments, mostly based on differences in behavior between partitives and genitives of negation. If these two are formally identical, as Pesetsky contends, and if the partitive is simply an instance of GEN(Q) that contains a null quantifier, as I have argued (in general agreement with both Neidle and Pesetsky), then the appropriate conclusion to be drawn is that the genitive of negation is not after all assigned by a null quantifier.90 On semantic grounds, it is patently obvious that genitives under negation need have no particular partitive interpretation; the examples in (170) are adequate to illustrate this. Moreover, the partitive is incompatible with singular count nouns, whereas this is by no means true of the genitive of negation. In fact, use of the partitive is independent of sentential negation, whereas the genitive of negation of course requires it. Since the partitive is viable in accusative object positions and the genitive of negation is optional, the partitive is expected to appear in both affirmative and negative contexts if appropriate to the verb. Thus, the partitive, as evidenced by the genitive2 ending -u and as observed by Klenin (1978), is acceptable in (174) regardless of polarity. (174)

a. SaSa Sasha 'Sasha b. SaSa Sasha 'Sasha

vypil caj/caju drank tea(acc)/(gen2) drank the/some tea' ne vypil caj/caju NEG drank tea(acc)/(gen2) didn't drink the/any tea'

On the other hand, if a verb does not accept partitive objects, as demonstrated in (135) and (137) and repeated in (175), then negation does not necessarily improve the felicity of the partitive genitive. (175)

a. on s"el sup/*supu he ate-up soup(acc)/(gen2) 'he ate the soup up' b. on ne s"el sup/*supu he NEG ate-up soup(acc)/(gen2) 'he didn't eat the soup up'

In these same contexts, however, the regular genitive is acceptable: (176)

a. SaSa ne vypil c"aja Sasha NEG drank tea(gen) 'Sasha didn't drink the tea'

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

199

b. on ne s"el supa he NEG ate-up soup(gen) 'he didn't eat the soup up' The felicity of the partitive thus appears to be independent of negation per se. Another kind of argument that the genitive of negation is not quantificational adduced by Neidle (1988,156-157) is that structural factors beyond those internal to the genitive NP are relevant (cf. also Greenberg, 1985, for discussion). She observes that although the genitive of negation can apply into embedded subject control complement clauses, as in (177a), it cannot apply into object control clauses, as in (177b). (177) a. Natasa ne xotela '[PRO citat' knig] (nom) NEG wanted to-read books(gen pi) ' Natasha didn't want to read any books' b. *ja ne ugovorila Nata§u '[PRO citat' knig] I(nom) NEG persuaded (ace) to-read books(gen pi) ' I didn't persuade Natasha to read any books' The felicity of the English gloss shows that purely semantic factors are not at work in ruling out (177b). It is also of interest that obescat' 'to promise' behaves like xotet' 'to want', as in (178). (178)

ja ne poobeSdala (Natase) '[PRO citat' knig] I(nom) NEG promised (dat) to-read books(gen pi) 'I didn't promise Natasha to read any books'

This is true regardless of the presence of the indirect object Natase; these facts parallel restrictions on predicate adjective agreement into embedded clauses to be discussed in the next chapter. Finally, Neidle (1988, 46-47) points out that not only are the partitive genitive and genitive of negation not coextensive in Russian, but they have also evolved differently in the Slavic languages.91 We return to comparative issues shortly. I now consider the complex question of whether the Russian genitive of negation is structural or inherent. It turns out that it is inherent in the sense that the genitive is in general [ +oblique] but, like other configurationally determined oblique cases, it applies optionally. First of all, like the accusative, it too is overridden by lexical case, as shown in (179). (179)

on ne vladeet ni odnim (nom) NEG speaks not-even one inostrannym jazykom/*odnogo inostrannogo jazyka foreign language(inst/gen) 'he doesn't even speak a single foreign language'

That is, the general genitive of negation rule competes with the specific lexical case governed by verb, and the latter—given that instrumental is necessary for theta-role visibility—wins. One might on the basis of this fact conclude that the genitive of

200

PARAMETERS OF SLAVIC MORPHOSYNTAX

negation is structural, and so at D-structure the quirky instrumental required by vladet' must be assigned, before the genitive of negation rule becomes applicable. I do not think this is the correct conclusion, however, since if the genitive of negation were structural it should never be able to override GEN(Q), but it can.92 Consider the following examples: (180)

a. on ne kupil [DP:GEN etix dvux knig] 'he didn't buy these two books' b. on ne kupil [DP:ACC eti [QP:Acc [Q dve] [NP:GEN(Q) knigi]]] 'he didn't buy these two books'

Either the genitive of negation applies to the entire quantified DP, as in (180a), or else it does not apply and the DP is marked accusative and the Q assigns (paucal) GEN(Q) to the NP that follows, as in (180b). Crucially, (181) is not a possible variant, where the external putatively structural genitive of negation would casemark everything except the innermost NP, which is assigned structural GEN(Q): (181)

*on ne kupil etix dvux knigi he NEG bought these(gen pi) two(gen) book(pauc)

If both processes applied at S-structure, a locality principle such as the minimality condition would predict the ungrammatical (181).93 I thus conclude that the genitive is optionally assigned to any sister of a negated verb. Since the genitive is [ + oblique], that process takes place at D-structure; however, since it is not lexically driven it is optional. Thus it may apply to all objects of transitive and unaccusative verbs, including passive ones, but is blocked if the transitive verb contains a specified case submatrix. This blocking effect does not obtain for non-argument objects, however, since these are not theta-marked by the verb, hence no quirky case is required. Here are a few further examples of the interaction of the genitive of negation with GEN(Q):94 (182)

a. on (etix) pjati knig ne cital he these(genpl) five(gen) books(genpl) NEG read 'he didn't even read (these) five books' b. on sta celovek ne znaet he hundred(gen) people(gen pi) NEG knows 'he doesn't even know a hundred people' c. on ne znaet i sta celovek he NEG knows even hundred(gen) people(gen pi) 'he doesn't even know a hundred people' d. on ne znaet sto celovek he NEG knows hundred(acc) people(gen pi) 'he doesn't know a hundred people'

(183)

a. on dvux slov svjazat' ne mozet he two(gen) words(gen pi) to-tie-together NEG can 'he can't even put two words together'

QUANTIFIED STRUCTURES! POLISH AND OTHER PUZZLES

201

b. on vam dvux slov ne skazet he you(dat) two(gen) words(gen pi) NEG will-say 'he won't even say two words to you' (184) a. sto celovek ne prisli na sobranie hundred(nom) people(gen pi) NEG came(pl) to meeting 'a hundred people did not come to the meeting' b. i sta celovek ne prislo na sobranie even hundred(gen) people(gen pi) NEG came(n) to meeting 'not even a hundred people came to the meeting' The genitive in (182a-c) implies that not even the stated number was reached; that is, in (182b) the implication is not that there are 100 people who he does not know, but rather that he knows some number of people, but fewer than 100. The examples in (183) are idiomatic expressions, so that the accusative would be inappropriate. The scope facts are somewhat clearer in the unaccusative (184)—in (184a) the preferred reading is for the number to have wide scope over the negation, and in (184b) the preferred reading is for the negation to have the wider scope. The semantics of the genitive of negation are extremely complex, and the reader is referred to works such as Babby (1980b), Klenin (1978), Neidle (1988), Paduceva (1974), Pesetsky (1982), and Timberlake (1975) for detailed discussion. Here I simply wish to point out the contrast just observed with respect to quantified objects of negated verbs: An accusative DP after a negated verb generally undergoes Quantifier Raising (QR) at LF and so has wide scope, whereas a genitive DP generally does not. That is, according to Paduceva (1974) and as also cited by Kleinin (1978) and Neidle (1988), (185a) is normally interpreted to mean 'he did not solve all of the problems', whereas (185b) is normally interpreted to mean 'he did not solve any of the problems'. (185)

a. on he b. on he

ne resil vsex NEG solved all(gen pi) ne reSil vse NEG solved all(acc pi)

zadac problems(gen pi) zadaci problems(acc pi)

A similar contrast exists for numeral phrase objects, as in the following minimal pairs, where negation has wide scope in (186a) and narrow scope in (186b):95 (186) a. on ne resil pjati/ mnogix zadac: he NEG solved five(gen)/many(gen pi) problems(gen pi) b. on ne resil pjat'/ mnogo zadad he NEG solved five(acc)/many(acc) problems(gen pi) Although these seem to be relative rather than absolute judgments, one would like to explain this contrast.96 Presumably, one alternative can be treated as formally driven and the other as pragmatically motivated, in that failure to select the determined option implies, in a sort of Gricean fashion, that the other reading should be dominant. But it is unclear which variant is which, or why this should this be so. However, if such an approach is correct, we must either force QR with accusative

202

PARAMETERS OF SLAVIC MORPHOSYNTAX

objects or inhibit it for genitive ones. It is unlikely that movement should be forced with accusative objects, since this does not generally obtain and no principles come to mind to cause obligatory QR, which suggests that QR of the genitive object under negation must somehow be inhibited. If so, the accusative object will be free to undergo QR or not, but the assumption is that if accusative has been selected then the DP moves, since if narrow scope had been intended the speaker would have used the genitive. The accusative can after all have narrow scope in examples such as (187), from Klenin (1978, 166).

(187) a. ne es' ves' syr NEG eat all cheese(acc) 'don't eat all the cheese' b. smotri, ne vypej vsju butylku watch-out NEG drink-up all bottle(acc) 'watch out, don't drink the whole bottle' Unfortunately, I do not know why it should be hard (or impossible) for a genitive object of a negated verb to raise at LF, although perhaps it is required to be in the scope (defined in terms of c-command) of negation at all levels of representation, including LF. I now turn to a comparison of the Russian genitive of negation facts with those available for other Slavic languages, primarily Polish. The genitive of negation in Polish applies obligatorily to complements of verbs that would be accusative in affirmative sentences. 1'his is shown by the following examples, drawn from Dziwirek (1990) (cf. also Franks and Dziwirek, 1993): (188) a. Janek nie czytalksiazki/*ksiazke (nom) NEG read book(gen)/(acc) 'Janek didn't read the book' b. Ewa nie lubi piwa/*piwo (nom) NEG likes beer(gen)/(acc) 'Ewa doesn't like beer' c. nie przeczyta-fam szesciu ksiazek/*szesc ksiazek NEG read(f Isg) six(gen) books(gen pl)/six(acc) books(gen pi) 'I didn't read six books' The accusative is never acceptable under sentential negation. Note also that there can thus be no semantic scope contrast in (188c) comparable to that observed for Russian. Example (189a) is thus ambiguous, although the first reading is with negation having wide scope. This follows stylistically from the surface word order of the two operators and pragmatically from the fact that there is an unambiguous way to express (189a) in the other reading, using zadnych 'not-any', as in (189b). (189) a. ja nie rozwiazatam wszystkich zadari I NEG solved(f Isg) all(genpl) problems(gen pi) 'I didn't solve all the problems'

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

203

b, ja nie rozwiazafam zadnych zadafi I NEG solved(f Isg) all(gen pi) problems(gen pi) 'I didn't solve any of the problems' The genitive of negation in Polish also cannot apply to objects of unaccusative and passive verbs, unlike in Russian.97 (190)

a. *nie NEG b. *nie NEG

przysto zadnej/ (ani) jednej kobiety arrived not-any(gen)/ not-even one(gen) woman(gen) by to budowano katedry was(nsg) built(nsg) cathedral(gen)

c. *nie zostato przeczytano ksiazki NEG became(n sg) read(n sg) book(gen) The genitive of negation in Polish thus differs from its Russian counterpart in two crucial ways: 1. Its application is obligatory. 2. It does not affect objects of unaccusative/passive verbs. Any parametric approach must at least explain these differences. One possibility that might be considered is that the genitive of negation is structural in Russian but inherent in Polish. This would perhaps explain why it is optional in Russian but obligatory in Polish, and such a contrast would fit with our idea that the GEN(Q) is structural in Russian but (can be) inherent in Polish. However, this would be a false advantage, since I have argued that the two sets of phenomena are not really related. Moreover, I have argued above that the genitive of negation rule assigns the regular genitive, hence it is always [ +oblique] (i.e. inherent), in the sense that when it applies this invariably takes place at D-structure. Finally, if the genitive of negation were truly inherent in Polish, we would expect negated quirky case assigning verbs to be ungrammatical in Polish, since it is impossible simultaneously to satisfy both requirements. However, such verbs behave just like in Russian: (191)

Janek nie kierowat szescioma sklepami/*szesciu sklepow (nom) NEG managed six(inst) stores(inst pl)/six(gen) stores(gen pi) 'Janek didn't manage six stores'

Crucially, the verb kierowac 'to manage' is lexically specified as assigning instrumental case, and it is this property that renders the otherwise obligatory genitive of negation rule irrelevant. I therefore consider an alternative analysis of the differences between these two languages in terms of the case feature submatrix theory developed in this book. I have argued that although the genitive of negation is configurational in the sense that it pertains to the structure "sister to a negated V," it is inherent in the sense that is the ordinary [ + oblique] genitive, as opposed to the [ — oblique] GEN(Q). It is thus much like the adnominal genitive, always able to apply, everything else being equal, but never needing to. In Polish, on the other hand, the genitive of negation is

204

PARAMETERS OF SLAVIC MORPHOSYNTAX

obligatory and only applies with transitive verbs. I conclude from this that in Polish an accusative-assigning verb under negation is literally transformed into a genitiveassigning one. By this I mean that the features in the verb's case submatrix are actually altered from those necessary for accusative case assignment to those appropriate for genitive case assignment. Assuming the same feature system for Polish as for Russian, [-obi, -marg, -nonascr, -phras] must be "changed" under negation into [ + obi, -marg, -nonascr, -phras]; that is, the feature value [-oblique] becomes [ + oblique]. Note that this is the minimal change required to make a verb into a quirky case assignor—it is simply designated as [ + oblique]. The verb is thus literally changed into an inherent (or lexical) case assigner in Polish. Now observe that this operation has exactly the desired effects: 1. If it becomes [ +oblique]), then it must be assigned at D-structure. 2. If the rule operates on case feature submatrices, then the verb must assign case in first place. It is for this reason that unaccusative and passive verbs are ineligible; they are not otherwise case assignors, either lacking a case submatrix or having their caseassigning features "absorbed" in the sense of chapter 8. The fact that the genitive of negation rule cannot affect verbs that assign quirky case in Polish also follows, since these have their case submatrix features set in the lexicon and, in particular, already assign a [ +oblique] case. There are two ways in which the Polish genitive of negation rule might be conceived: either (1) it changes [ - oblique] to [ + oblique] or (2) it changes [aoblique] to [ +oblique]. Although for present purposes either approach is adequate, since if a verb assigns inherent case it is already specified as [ + oblique], there are several reasons to prefer the latter formulation. For one thing, we needed to assume unspecified feature values for the accusative to obtain the proper interaction between accusative case assignment and GEN(Q), since accusative case assignment is a consequence of theta-role assignment and thus should pertain to D-structure, but its effects—the filling in of unspecified feature values by minuses—must be delayed until S-structure. For another, it is probably theoretically undesirable to allow feature-changing rules of morphosyntax,98 but by just filling in an unspecified [a oblique] as [ + oblique] this problem is circumvented. Be that as it may, consider as further support for this analysis the fact that in passive-like constructions in Polish in which the accusative is indeed assigned to objects of "participles" (192a) and "reflexives" (192b), the genitive of negation is also required, as shown in (193). (192)

a. przeczytano ksiazke read book(acc) 'the book was read' b. sprzedaje si§ ksiazki sell(3sg) REFL books(acc pi) 'the books are being sold' (193) a. nie przeczytano ksiazki NEG read book(gen) 'no book was read'

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

205

b. nie sprzedaje sie. ksiazek NEG sell(3sg) REFL books(gen pi) 'no books are being sold' This evidence corroborates the feature changing analysis for Polish: verbs in passive siq and -no—a special frozen form discussed in chapter 8—assign accusative case to their objects and so undergo the regular genitive of negation rule. 5.3.2.2. Conclusion: A Curious Correlation I conclude this chapter with a consideration of the true nature of adjunct genitive of negation phrases, based on work reported in Franks and Dziwirek (1993). Before doing so, however, it is worth mentioning the status of the genitive of negation rule in some of the other Slavic languages whose numeral phrase properties have been discussed. In both Serbo-Croatian and Czech, the genitive of negation existed roughly as in Russian but is now no longer productive, and its use has a decidedly archaic flavor. In both instances this loss is relatively recent, so that in literature from the nineteenth century one frequently encounters genitives under sentential negation. It is thus presumably independent of GEN(Q), which remained unaffected by the loss of the genitive of negation and, as we have seen, survives and functions normally in Serbo-Croatian and Czech. Let us now examine the genitive of negation applying to adjunct phrases in Slavic. Interestingly, although such phrases do appear in the genitive in Polish, as in (194) from Dziwirek (1990), this process applies optionally rather than obligatorily. (194)

a. nie spalam godzine7godziny NEG slept(f Isg) hour(acc)/(gen) b. Jan nie przeszedf kilometr/kilometra (nom) NEG walked(m 3sg) kilometer(acc)/(gen) c. ten dom nie kosztowaf majatek/majatku this house NEG cost(m 3sg) fortune(acc)/(gen) d. ta ryba nie wazy kilogram/kilograma this fish NEG weighs(3sg) kilogram(acc)/(gen)

All of these phrases are non-arguments and adverbial in function, (194a) specifying time, (194b) distance, (194c) price, and (194d) weight; none of them admit passivization. In all these instances the genitive is optional, with certain semantic differences between the accusative and genitive choices discussed by Dziwirek (1990) and Franks and Dziwirek (1993). This optionality raises the question of whether (194) should really be assimilated to the genitive of negation rule applied to arguments, as discussed in the previous section. For example, Dziwirek (1990) remarks that (194a) with the genitive "suggests that the speaker did sleep but for less than/part of an hour," whereas (194a) with the accusative ' 'may have the 'less than' reading, or can be interpreted as a negation of the entire proposition." On this basis she concludes that the genitive here is really an instance of the partitive and that the genitive-ofnegation rule does not in fact apply to non-argument phrases in Polish." Clearly, since the Relational Grammar framework within which Dziwirek's analysis is

206

PARAMETERS OF SLAVIC MORPHOSYNTAX

couched states rules such as the genitive of negation in grammatical relation rather than structural terms, it is central to her approach to divorce these phrases, which have adverbial functions, from true complements. Within the structural approach adopted in this book, however, one might expect phrases accidentally appearing in configurationally determined case positions to receive that case, so long as thetatheoretic considerations are respected. Curiously, within the GB framework, Lasnik (1992a) has argued that the genitive of negation in Polish is inherent on the grounds that it does not apply to nonarguments and in this respect should be contrasted to the genitive of negation in Russian, which he concludes is structural. There are two serious problems with this conclusion, however. First of all, the data in (194) show that the genitive of negation can indeed apply to non-arguments in Polish as well as Russian.100 Second, the Russian data otherwise indicate that the genitive of negation is inherent, as I argued in section 5.3.2.1 and Lasnik (1992a, n. 17) himself implies. It is nonetheless worth pointing out that, although I reject Lasnik's structural/inherent dichotomy for the genitive of negation, I have in these two chapters made considerable use of a similar conceptual contrast for GEN(Q) and distributivepo in the various Slavic languages. Dziwirek's idea that these "genitives" are in fact partitives and thus independent of the regular genitive of negation rule makes certain cross-linguistic predictions that, surprisingly, are more or less borne out. As shown in Franks and Dziwirek (1993), a comparison of the status of the adjunct genitive on time/distance phrases and the partitive construction reveals an apparently close correspondence between these two phenomena, rather than between the adjunct genitive and the genitive of negation. In Czech, for example, all three processes are moribund: the genitive of negation is archaic and the partitive is no longer productive, so that the accusative is far preferred in (195).1M (195)

dej mi cukr/?*cukru i vodu/?*vody give me sugar(acc)/sugar(gen) and water(acc)/water(gen) 'give me some sugar and water'

As expected, time phrases also cannot appear in the genitive under negation, as shown by (196). (196)

a. pracoval celou noc worked all(acc) night(acc) 'he worked all night' b. nepracoval celou noc/*cele noci NEG-worked all night(acc)/(gen) 'he didn't work all night' c. nepromluvil cely tyden/*celeho tydne NEG-spoke all week(acc)/(gen) 'he didn't utter a word all week' d. nepracoval ani jednu minutu NEG-worked not-even one minute(acc) 'he didn't work for even a minute'

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

207

Example (196a) shows the accusative in the affirmative; (196b) shows that only the accusative is viable in the negative as well. The same is true under sentential negation in (196c) and even in (196d), where the particle ani 'not even' might be expected to induce the genitive if this were possible. In Serbo-Croatian, on the other hand, the partitive continues to be productive, as shown in (197). (197)

daj mi sira, mesa i vode give me(dat) cheese(gen) meat(gen) and water(gen) 'give me some cheese, meat, and water'

Interestingly, even though in Serbo-Croatian the genitive of negation is even more archaic than in Czech, as a comparison of (198a) and (198b) reveals, genitive time phrases are perfectly viable under negation, as indicated by the judgment in (198c). (198) a.

nisam Citao nijedan casopis NEG-AUX(lsg) read(m sg) not-even-one(acc) magazine(acc) ' I didn't read even one magazine' b. *nisam citao nijednog casopisa NEG-AUX(lsg) read(m sg) not-even-one(gen) magazine(gen) c. nisam spavao nijednog trenutka NEG-AUX(lsg) sleep(msg) not-even-one(gen) moment(gen) ' I didn't sleep even for a moment'

The sharp contrast in grammaticality between (198b) and (198c) is extremely suggestive, although further comparative and historical research is necessary to corroborate the correspondence completely. Slovenian, which is essentially the opposite of Serbo-Croatian in all relevant respects, provides similar evidence in support of this curious relationship between partitive and the adjunct genitive. In Slovenian, the genitive of negation is virtually obligatory, as in Polish, although the partitive in this language appears to much more restricted, as in Czech. Correspondingly, adjunct phrases under sentential negation appear in the accusative rather than genitive. Compare the following examples: (199) a. *nisem prebral eno knjigo NEG-AUX(lsg) read one(acc) book(acc) b. niti ene knjige nisem prebral not-even one(gen) book(gen) NEG-AUX(lsg) read ' I didn't read even a single book' c. niti en trenutek / *enega trenutka not-even one(acc) moment(acc)/one(gen) moment(gen) nisem spal NEG-AUX(lsg) slept ' I didn't sleep even for a moment' The accusative is required on the adjunct in (199c), but is impossible for the argument object in (199a). This is consistent with the facts drawn from the other languages, which suggest that the genitive on negated adjuncts is independent of the

208

PARAMETERS OF SLAVIC MORPHOSYNTAX

general Slavic genitive of negation rule and should instead be analyzed together with the partitive construction. The data for the various languages are summarized in the following table:

(200) Language

Genitive complements

Genitive adjuncts

+

+

"Genitive" partitives +

+ (oblig) _

+ (opt) + _

+ (opt) + -

Russian Polish Czech Serbo-Croatian Slovenian

4-

There is one serious problem with Dziwirek's proposal that makes me reluctant to adopt it, however: this genitive is still dependent on negation, the accusative appearing in the corresponding affirmatives. Since the partitive is otherwise independent of sentential negation, it is somewhat surprising that just in this instance it should be triggered by negation. One possible way of handling the role of negation in licensing partitive adjuncts is suggested in Franks and Dziwirek (1993). There we maintained that adjuncts, unlike arguments, are not c-commanded by V but rather assumed that they were adjoined to VP, as in the following diagram of sentence (171):102

IP

(201)

X

NP

r

I

NegP s

x

ne

VP

VP V

Ivan

Cital

QP ni

Q' Q

NP

e

minuty

Since the QP ni minuty 'not a minute' is outside the lower maximal projection VP, the verb cital 'read' does not c-command it. We would therefore not expect the adjunct to be sensitive to particular features on the verb. Hence, partitive time

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

209

phrases may occur regardless of whether the verb is one that otherwise allows partitives or not. Indeed, such adjuncts are totally blind to any features of the verb, including whether or not the verb is transitive (assigns case) or assigns quirky case. As it is too high up, the adjunct is necessarily unaffected by the presence or absence of [ + Q] on V. However, being a partitive, the adjunct QP still has a null Q that needs to be licensed. I propose instead that it is licensed by the negation element, which, unlike the verb, does c-command the null quantifier. In this way, the relationship between negation as a licensing element and the distribution of genitive adjuncts can be explained without necessarily unifying the phenomenon with the genitive of negation. Nonetheless, there are several remaining conceptual problems. In divorcing the genitive from configurational factors here, one should also reject a configurational explanation for the accusative on time/distance phrases in affirmative sentences, thereby necessitating an independent source for this use of the accusative. Possibly, they receive accusative case by some kind of default mechanism, assuming accusative to be the least marked case in the Slavic languages. Finally, any attempt to assimilate these non-argument genitives to the partitive genitive runs into all the same difficulties discussed above for regarding the genitive of negation in general as an instance of GEN(Q) with a null quantifier, that is, as technically partitive. Still, the facts summarized in (200) are particularly problematic for any attempt to assimilate the genitive on negated adjuncts to the standard genitive-of-negation rule. They demonstrate that the two processes are to some extent independent. In fact, my analysis of Polish in terms of manipulating the feature [ ± oblique] in the verb's case submatrix should lead one to expect that non-arguments can never receive genitive under negation, since it was essentially claimed that the verb in Polish is transformed into a quirky case assigner and inherent case is only assignable to true complements. Consequently, some other mechanism is necessary to obtain the optional genitive in (194). This mechanism may be the partitive, as described above and in Franks and Dziwirek (1993), or it may be the result of a configurational rule assigning genitive to NPs governed by a negated verb. Note, incidentally, that just as in Russian this process is not dependent on the transitivity of the verb. The difference between Russian and Polish, then, would be that the verb's case feature submatrix is also affected in Polish. One potential problem with this kind of approach would be to prevent the configurational rule in Polish from applying to complements of verbs that do not otherwise assign (accusative) case, in essence undermining the explanatory advantages of the case feature manipulation account. The way to resolve this issue is to similarly restrict the Russian genitive-of-negation rule that applies to NPs governed by a negated V to non-arguments of the verb, so that true complements (i.e. NPs that are theta-marked/L-marked by the verb) are only able to receive the genitive of negation under sisterhood to V. In Russian this applies to all complements, in Polish only to complements of verbs with case submatrices, since the Polish genitive of negation rule proper actually affects the verb's case submatrix rather than the object directly. There are, in conclusion, reasons for treating these non-argument genitives either as partitives or as genitives of negation, and each approach has its advantages and complications.

210

PARAMETERS OF SLAVIC MORPHOSYNTAX

This chapter has been a lengthy excursus on various problems associated with the parametric analysis of numeral phrases proposed in chapter 4. In some sections promising solutions to these problems were developed, while in others a variety of (equally credible and problematic) proposals were compared. A consideration of Polish led to the idea that the distribution of QPs can be limited to accusative contexts, formalizable in terms of only ACC Ds selecting for QPs. A study of the distributive preposition po allowed for further fine-tuning of the structure of numeral phrases, in keeping with the tenets of X-bar syntax. Finally, an array of diverse puzzles lacking definitive resolutions was examined. It is my hope that the discussion in this chapter will help to shed light on these puzzles and to encourage more careful scrutiny than I have provided here. Notes 1. The numerals jeden, dwa, trzy, cztery 'one, two, three, four' are adjectival in form and function. They thus agree with the head noun in case and, where possible, gender. As subjects, they also take agreeing verbs—the plural for dwa, trzy, cztery 'two, three, four' and the singular for jeden 'one'. Higher numerals are similar to piec 'five', as described in the text. Interestingly, and unlike in Russian and Serbo-Croatian, this includes those ending in jeden, which remains frozen in the "masculine" form regardless of the gender of the GEN(Q) noun. 2. Note that the formpieciu in (2b) reflects the fact that the head noun is masculine human or "virile." I return to the significance of this shortly. 3. Suprun (1963, 140-141) found two Polish examples (out of a written corpus of 650 examples) with plural agreement with numeral phrases containing numbers 'five' and greater. He regarded these as deviations from the general rule of non-agreement, in which the remote possibility of "semantic" agreement occurs due to extraneous factors such as unusual distance between the subject and the verb. Speakers consulted judged both examples as unnatural and stilted in the modern spoken language; one was nineteenth-century prose. 4. This example is drawn from Dziwirek (1990), to which the reader is referred for further evidence that Polish quantified phrases display subject properties. 5. See Schenker (1971) for arguments in support of the claim that quantified subjects in Polish are actually accusative. This fact is also noted by Pesetsky (1982) and Neidle (1982, 1988). 6. See Dziwirek (1990, 1991) for detailed arguments that agreement is only with nominative subjects in Polish. 7. See Corbett (1983a, 218, 240, n. 3) and references therein. 8. By "inherently" I do not mean that this accusative is any different from the regular [ — oblique] ACC, but rather that it has no obvious structural motivation. Freidin and Sprouse (1991) similarly propose that inherently case-marked NPs in Icelandic still must move to subject position in order to be licensed; see also Zaenen, Maling, and Thrainsson (1985) for an analysis of quirky case constructions in Icelandic. 9. Nominative objects in North Russian dialects, as discussed in Timberlake (1974), may be amenable to a similar analysis. 10. Schenker also classifies the genitive as agreeing rather than governing, as in od pieciorga dzieci 'from five(gen) children(gen pi)'. Since it is impossible to tell the source of the genitive on the noun on the basis of the morphology, I include this with the other regular obliques. 11. Government of genitive by collective numerals in the instrumental also reflects the adnominal origin of GEN(Q), as discussed by Babby (1987).

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

211

12. Alternatively, one might consider these to be nouns taking the adnominal genitive, like nouns derived from numerals with the -ka suffix, for example, piatka 'five'. Notice that they are distinguished from the nominative/accusative form by containing the -org- suffix (with the variant -ojg in the collectives dwoje 'two', troje 'three', and oboje 'both'). However, the nominative/accusative form should probably be analyzed as assigning GEN(Q), since it may occur with genitive prequantifiers, as in tych piecioro dzieci 'these(gen pi) five children(gen pi)' alongside te piefioro dzieci 'these(nom/acc pi) five children(gen pi)'. 13. According to Charles Townsend (personal communication), some speakers find the plural jely 'traveled' marginally acceptable with numeral phrase subjects, although the grammar books only cite neuter singular. I do not know whether Czech truly differs in this respect from Polish, but if so I can offer no explanation for this possibility, since the status of agreement should be identical in the two languages if subject numeral phrases receive a similar analysis (as intrinsically accusative). 14. One curious fact is that in oblique contexts malo 'few' neither declines nor governs, remaining transparent to the oblique case. This situation is even more widespread in other West Slavic languages. For example, as pointed out to me by Curt Woolhiser and discussed in FaBke (1987), in Upper Serbian in oblique contexts quantifiers in general neither agree nor fail to block percolation. This gives rise to such constructions as z tfi tysac wojakami 'with three thousand soldiers(inst pi)' and z tymipjed zraiymijabtukami 'with these(inst pi) five ripe(inst pi) apples(inst pi)'. Interestingly, this property also extends to classifier nouns, as in z husk papjeru 'with piece paper(inst)', demonstrating that such nouns are true quantifiers. Such classifiers may of course also function nominally, assigning adnominal genitive like any other noun: z kuskom papjery 'with piece(inst) paper(gen)'. 15. The only potential problem is that the demonstrative in (23a) does not reflect animacy. Lindseth (personal communication) explains this in terms of the number also not reflecting animacy, since otherwise the special virile nominative numeral formpjeco would be used. The ungrammatical (23b), on the other hand, involves an animacy mismatch, the only acceptable form with ci being (i). (i)

ci PJeco dzS-facero these(nom pi virile) five(nom virile) workers(nom pi)

Corroboration for this account depends on forms such as (ii) arising only as a combination of [ +oblique] GEN(Q) and the accusative restriction, comparable to (22b). (ii) tych pjec dzSlacerow these(gen pi) five workers(gen pi) 16. One exception is za, in the (to za 'what for' construction, which is a caique on German was fur. In addition to assigning no theta-role to its object NP, za in this construction does not assign any specific case; see Franks (1985) for discussion of the range of possibilities. The NP appears instead in whatever case is independently called for. Interestingly, as discussed in section 5.2.1.4, Serbo-Croatian po has precisely this property of being transparent to case assignment. 17. To be fair, po is exceptional in this regard in other usages as well—even though it necessarily assigns DAT to a complement NP, as in (i), it also admits LOC pronominal complements, as in (ii) and (iii): (i)

Vera skufiaet po otcu/*otce Vera longs for father(dat/loc)

(ii)

Vera skucaet po nemu/?nem Vera longs for him(dat/loc)

212

(iii)

PARAMETERS OF SLAVIC MORPHOSYNTAX

Vera skuc"aet po ?vam/vas Vera longs for you(dat/loc)

Speakers seem to prefer the LOG form for first and second person plural pronouns, although judgments vary. The DAT/LOC first and second singular forms mne and tebe are syncretic. 18. This is also true of compound numerals ending in 'one': po tysjace dvesti odnomu rublju 'DIST thousand(dat) two hundred one(dat) ruble(dat)'. 19. The numerals in (27) are traditionally regarded as accusative, although I shall argue that they are caseless Qs. 20. Recall that the ioimpjati is syncretic and may be any of genitive, dative, or locative. In case feature terms it is [ + obl, — nonascr]. 21. This problem is also noted by Neidle (1988, 171-172, n. 6), who nonetheless adopts Mel'cuk's suggestion. 22. Millionu and kilogrammu in (31c) and (3 Id) could be regarded as nominal complements to po, with the NPs following them in the adnominal genitive. 23. One could conceivably accommodate this difference by distinguishing okolo frompo by stipulating that okolo lacks either an adverbialized variant, following Crockett's analysis discussed immediately below, or the ability to assign accusative, following Mel'fiuk. 24. By "directly" I mean by virtue of government rather than agreement. See chapter 6 for discussion of another instance where a morphologically nominal phrase is directly assigned case even though it is not an NP. 25. The issue of frozen quantifiers is discussed in section 5.2.3. 26. Some speakers assign collocations like po tremstam rublej 'DIST three-hundred(dat) rubles(gen pi)' marginal status. 27. Further possibilities are considered in Franks (1986). See also Fowler (1987a), who argues that other parts-of-speech in Russian can assign GEN by virtue of their being quantificational. He claims, for example, that there are quantificational [ + Q, — N, + V] verbs, such as those prefixed by na-, such as nakupit' 'buy a lot of, which requires a quantity object, and possibly also [ + Q, - N, — V] prepositions. Interestingly—and unlike Fowler's putative [ + Q] prepositions—verbs in na- take QP objects with either a null or overt quantifier (cf. section 5.3.1). Unlike verbs that assign quirky GEN, verbs in na- do not take genitive DP objects, as evidenced by the fact that numeral phrase complements are not genitive. This suggests to me that they semantically select for quantified expressions rather than for entities, with the CSR of a quantified expression being a QP. 28. Note that numerals which appear both to govern GEN(Q) and decline for case, such as tysjaca 'thousand', are in my analysis not actually Qs. I will argue that they are instead specifiers of QP. It is only frozen numerals, which are [ + Q], that appear in the head position Q. 29. Pjaf 'five' in (47b) is glossed as accusative to facilitate comparison with (47a), the essential point being that the nominal head of the object of the preposition fails to receive the case governed by that preposition. 30. This discussion is couched in the analysis of chapter 4, to be revised in the next section. 31. Such small clauses are only viable in Russian when the predicate is an AP, NP, or PP, but never a full VP. This suggests that this construction necessarily lacks tense and agreement features. Regardless of these restrictions, however, the point remains that the complement of scitat' is a constituent, the subject of which is externally accessible for the purposes of caseassignment. See chapter 6 for further discussion of small clauses in Russian. 32. The genitive of negation does not interact with the structural dative assigned by po since this heads a PP rather than an NP.

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

213

33. This form also occurs after po in other, related uses, such as the idiom po mnogu let 'for many years on end'. 34. Compare, for example, Abney (1987) or Ouhalla (1990, 1991). 35. But recall the suggestion in (34) that compound numerals may make use of the head position for overt material. This sort of analysis should perhaps be extended to other more obvious compound numerals, such as dvadcat' dva 'twenty-two' and dvadcat' pjat' 'twentyfive', which—just like dva and pjat'—govern paucal and genitive plural, respectively. Similarly, dvadcat' odin 'twenty-one' agrees, just like odin (although in Polish such numerals end in frozen jeden and govern GEN[Q]). 36. In principle, both positions may be occupied, as in the French ne pas construction. 37. Shlonsky (1991), extending ideas due to Abney (1987), proposes that the Hebrew quantifier kol 'all' heads its own functional category QP. Ritter (1991) similarly claims that numerals and other quantifiers in Hebrew head NumPs. Giusti (1991) also argues for Q as a functional head in a variety of languages. From this it is a small step to assimilate variation in QPs to other types of functional phrases involving operators, so that the lexical item may be analyzed as either the specifier or the head of the phrase. 38. There are, I believe, independent reasons for assuming DPs in Russian. Although it is true that determiner-like elements can iterate, as in (i), from Avrutin (1992), this does not mean that the phrase is not maximally a DP, or that they are not independent heads: (i) ja cital etu ego poemu I read this his poem As Avrutin points out, although possessives in Russian can otherwise extract (in apparent violation of the Left-Branch Condition), this extraction is blocked by the demonstrative, both in the syntax as in (ii) and at LF as in (iii), assuming the reflexive svoj raises to I at LF. (ii)

*C'ju ja cital etu poemu whose I read this poemu

(iii)

*Ivan slomal etot svoj velosiped Ivan broke this self's bicycle

If each of these determiner-like elements heads a separate functional category projection, then the trace left after extraction will not be antecedent-governed under the relativized minimality theory of Rizzi (1990). For more detailed arguments that Russian requires something akin to a DP, see Paduceva (1985, 83-107), the relevance of which was pointed out to me by Michael Yadroff. 39. The structure of compound numerals in tysjaca 'thousand' is somewhat more complex, since the tysjaia part always shows up dative: (i)

po tysjace odnomu rublju DIST thousand(dat) one(dat) ruble(dat sg)

(ii)

po tysjace dva rublja DIST thousand(dat) two ruble(gen sg)

(iii)

po tysjace pjat' rublej DIST thousand(dat) five rubles(gen pi)

Tysjaca can also be a true (i.e. [ - Q]) noun; see Corbett (1978,1983a) for details. The proper analysis of these and related structures within a fully articulated theory of Slavic DPs is left for future research. 40. For discussion of the role of animacy in inducing plural agreement with paucal subjects in Russian in general, see Neidle (1988, 112-115).

214

PARAMETERS OF SLAVIC MORPHOSYNTAX

41. Crockett's approach might seem to offer a possible solution, but fails to explain the special status of the paucal numerals. 42. Possible support for the structure in (65) might be found in Bulgarian and Macedonian, which have post-positive articles. These can be regarded as enclitics in D. The article appears to attach to the right of the first head beneath it—the N if there are no modifiers, the first A if one is present, and so on. This suggests an analysis in terms of head raising to D, subject to the Head Movement Constraint of Travis (1984) and Baker (1988). One problem for this analysis, however, is that As do not raise over intervening adverbs; instead, D appears to the right of the entire AP: mnogo xubavata kniga 'very nice-DET book'. This suggests that maybe only functional categories (i.e. QP and DP in [65]) are projected up from lexical ones, and that AP is internal to NP, as was standardly assumed prior to Abney and contra (65). Under this structure, the highest XP would move to SPEC-DP, so that the article in Bulgarian cliticizes onto the phrase in its specifier. 43. She explains the lack of expected agreement by claiming that Qs are plural, having lost their historical gender; see the discussion of the evolution of numerals in Babby (1987). 44. Babby's dilemma stemmed from the fact that the N satisfies selection requirements but the Q exhibits certain properties of a syntactic head. This is exactly what Grimshaw's view of functional projections is designed to accommodate, however. Note that Babby's problem that subject-verb agreement is never with the numeral disappears when one realizes that Qs bear neither nominative case nor pronominal features. He thus points out the pjaf 'five' was originally a feminine noun and induced feminine agreement, but this never occurs in the modern language, as shown in (i): (i)

pjat' zenScin prislo/*prisla five women arrived(n)/(f)

Notice that even when pjat arguably functions as a noun, the verb shows third person neuter singular rather than feminine agreement, as in (ii): (ii)

pjat' delilos' na tri five divided(n) by three

I would claim here that pjat' (and tri) are still Qs and that N is simply empty. 45. In some languages, most notably Icelandic, as argued by Sigurdsson (1991), for example, PRO has case features, but this situation seems to be relatively rare. 46. This is true of Russian; in Serbo-Croatian they can only be maximally DPs. 47. Borras and Christian (1971, 391) mention this fact, although the particular examples they cite were deemed infelicitous by native speakers for extraneous reasons. They note that marking animacy with the genitive, as in (69b), is standard. Interestingly, they observe that with compound numerals ending in paucals the situation is reversed, with failure of the animacy rule considered the norm. The non-paucal numbers show no alternation, since as Qs they are caseless and even as nouns they did not belong to the appropriate declensional class, being /-stem substantives, which never undergo the animacy rule. 48. Martina Lindseth (personal communication) points out that explaining the variation between ECM and no ECM in terms of whether the numeral is a head or specifier of the QP would eradicate any observable distinction between numeral phrase DPs and QPs after po, so that one could then maintain that allpo-phrases are QPs, thus obviating the structure in (60). I think this move is unwarranted for several reasons. First of all, note that the same problem of ambiguous structure exists in all structural case positions, with the exception of subject. Second, there is no obvious way to prevent DPs from appearing in this position, since they clearly do (i) in Russian when the numeral is 'one' and (ii) in the other Slavic languages for which I have argued nouns always project maximally to DPs and never to QPs.

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

215

49. Dickey comments that example (70b) has a fairly "mathematical" flavor. He also notes similar case behavior for Serbo-Croatian po in examples such as the following, which should be compared to (70a) and (70c), respectively: (i)

razgovaramo sa jednim po jednim kandidatom we-speak with one(inst) DIST one(inst) candidate(inst) 'we are speaking with each candidate, one after the other'

(ii)

dobijali smo municiju od jednog po jednog vojnika got AUX(lpl) ammunition from one(gen) DIST one(gen) soldier(gen) 'we received ammunition from each soldier, one at a time'

These examples illustrate that it is also possible in Serbo-Croatian for the po-phrase to distribute over the verb's event argument, resulting in the jedan po jedan 'one by one' reading. The fact that the twopo constructions display the same case properties and comparable distributive interpretation shows that they are intimately connected. 50. The -a ending is a special paucal form of the verb that is identical to the neuter plural but is also used with paucal masculine subjects. For at least one speaker, the regular masculine plural sedeli 'sat' was inadmissible here, in puzzling contrast to (76a), where both agreements were viable. For at least one other, the paucal sedela was the only correct form in examples involving dva, or po dva as subjects. See Corbett (1978) for discussion. 51. One troublesome aspect of this analysis remains—it is not clear how structural dative is to be blocked with the paucal numerals in Polish since these, unlike in Russian, agree rather than govern. 52. It is worth noting an interesting possibility for partially eliminating the stipulative character of the second alternative. Imagine that distributive po freely assigned (inherent) locative or (structural) accusative—the problem would then be to reduce the realizations of these two options to the pattern in (77) and (78). Now observe that assignment of locative to the quantified DPs in (78) would independently be blocked under the assumption that such NPs are invariably accusative. However, there is no comparable way to block accusative assignment in (77), nor would this kind of blocking extend to Russian. 53. Although I assume these verbs are unaccusative on the basis of their meaning, I know of no reliable test for unaccusativity in Polish. 54. In all probability, PP subjects are like QP subjects (in Russian) in remaining in VPspecifier position, so that the "default" agreement on the verb actually reflects agreement with an empty expletive in IP-specifier position. 55. Some speakers do not accept either variant. 56. The form dvouch is the colloquial variant. Either way, the locative is obligatory throughout. 57. Mistrik opposes "ordinary" numerals to "group" and "multiplicative" numerals. Group numerals are frozen forms and multiplicative numerals are agreeing modifiers. 58. Just why inversion should induce an approximative reading remains a mystery. Mel'duk (1985,156) cites a similar process operative in the Abkhaz language, although with a completely different interpretation. 59. Although approximative inversion does strongly favor non-agreement, judgments vary considerably and the plural option is sometimes also perfectly acceptable. Relevant factors that to varying degrees improve the status of the plural include (i) focusing the subject, either by intonation or word-order, (ii) using certain nouns, such as celovek 'people' in (92c), instead of studentov, and (iii) increasing the cardinality of the numeral. These all suggest to me that semantic agreement with the plural noun inside the QP—just as was noted for PP subjects—may be at work here, rather than syntactic agreement with a nominative plural DP.

216

PARAMETERS OF SLAVIC MORPHOSYNTAX

60. Some speakers report the following examples of approximative inversion applying to dative argument DPs as marginally acceptable: (i)

(ii)

?on predstavil -sja gostjam pjati he introduced-REFL guests(dat pi) flve(dat) ' he introduced himself to about five guests' ja pokazal svoju knigu I showed self's book

?Movekam/*ljudjam/*druz'jam pjati persons(dat pl)/people(dat pl)/friends(dat pi) five

Note that gostjam is a typical complement forpredstavilsja 'introduced' in (i), and that only the special "counted" form celovekam is admissible in (ii). There is thus some variation, depending on subtle judgments and complex factors, as discussed in the preceding note. 61. Note that since Baltin (1991) argues that the Like-Attracts Like Constraint does not apply at LF, it is conceivable that it is also somehow circumvented in the structure in (103). I think the answer will lie in the ultimate motivation for this constraint, however, which surely goes beyond a general appeal to elegance and symmetry. 62. If, on the other hand, one opts for the adunction to XP analysis to accommodate (109b), then it becomes more difficult to explain (109a). 63. There are other complicating factors, such as the effect of the particular preposition. For example, (111) would be far worse with fabrikax na pjati 'factories(loc pi) in five(loc)', although the only difference is the choice of preposition. 64. A non-argument or "adjunct" can be understood as an XP that does not receive its theta-role by virtue of coindexation with the verb's theta-slot, under the formalization proposed in chapter 2. This process is roughly what Chomsky (1986a) refers to as L-marking. 65. There are two scenarios to consider: N is adjoining either to some X (Q or P) or to some XP (QP or PP). Adoption of the latter analysis only requires that the category to which adjunction occurs be an XP for the argument/adjunct asymmetry to arise. 66. However, Browne adds that Byelorussian, in which the paucal numerals behave similarly and which also has approximative inversion, does not change the noun from NOM to GEN(Q), citingpraz dni dva 'in days(nom pi) two'. Notice here that, unlike in Russian, the numeral does not adjoin to the PP. 67. The status of these two options is unclear, in that many speakers find (non-agreeing) neuter singular -o with subject numeral phrases containing tysjaca very marginal, suggesting that these are always DPs, with the choice being whether the numeral or noun is taken to be the head N (cf. Corbett, 1978, for discussion). Interestingly, non-agreement is slightly better with unaccusatives, as inpri$li/?prislo tysjada delovek 4arrived(pl)/(n) thousand people'. Here the agreement is as expected if tysjaca Selovek is an object NP and there is an expletive subject, but the problem still remains of associating the numeral phrase with nominative case. 68. See Corbett (1978, 1983a) for discussion of the various possibilities. Note that in this respect tysjaca again differs from million, which only allows an agreeing masculine singular demonstrative (but cf. Corbett, 1983a, 240, n. 6). 69. It may be that the same results can be achieved by just not requiring QP to raise in search of case, since there was some evidence that the asymmetries followed from the categorial dichotomy independently and the possibility of long-distance movement follows so long as raising of QP is not obligatory. 70. As always, dva 'two', tri 'three', and cetyre 'four' can decline in more literary styles, but then, instead of governing GEN(Q), they agree with the head noun. 71. According to Vaillant (1948, 181), numerals were already invariable forms in the language of the late-sixteenth-century Dubrovnik poet Zlataric, although in slightly earlier documents, such as the lectionaries of Bernardin Splitski (1495) and N. Ranjina (1500s), both

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

217

fixed and agreeing forms were found. Thanks are due to Wayles Browne for bringing this information to my attention. 72. Some speakers accept this, assigning it marginal stuatus, but nonetheless consistently rejecting masu + non-agreement with unergative and transitive verbs. I have no explanation for this contrast in Serbo-Croatian; no comparable unaccusative pattern was found, for example, for Russian paru or Polish paref subjects. 73. The Polish facts are actually somewhat more complicated. Speakers consulted report that troche is used in NOM, ACC, and possibly GEN positions, but are not sure what to do in other oblique contexts. Pare, on the other hand, has a special oblique fonaparu, which is used in virile NOM/ACC, GEN, DAT, and LOG contexts, so that only pare more realistically seems like the frozen form. What is clear is that the forms in -« are frozen Qs and thus can only appear in accusative DPs, in accordance with the analysis of Polish presented in section 5.1. Masq may decline like a feminine singular noun in oblique positions, although it is used alongside the expected form masy in GEN contexts. 74. Once again, these frozen Q forms only appear in accusative DPs; in oblique contexts they decline, that is, parunastu '11-19'. 75. The discussion in this section is based in part on Franks (1986), Franks and House (1982), and Franks and Dziwirek (1993), although details of the analysis differ considerably. 76. They also appear in several other extremely limited contexts (cf. Borras and Christian, 1971, 20-23, for examples). Klenin (1978,177-178) argues, however, that although -u genitive forms casu Ihour(gen2)' and sagu 'step(gen2)' exist, these are not really partitives since they only appear under sentential negation and lack partitive force; but see section 5.3.2.2. 77. There is, of course, a sense in which [ + Q] is a diacritic feature, since the range of verbs that admit partitive complements varies among speakers and across languages. See Franks and Dziwirek (1993) for discussion. 78. Note that, even assuming that Q is generally null, as proposed in section 5.2, thenullQ of partitives differs from null Q associated with numeral specifiers in that it has no features through SPEC-head agreement. 79. An additional concern is that partitives cannot generally serve as the object of a preposition. There are three subcases to consider: prepositions assigning accusative, genitive, and other oblique case. Since numeral phrases are otherwise viable after prepositions assigning accusative, I conclude that prepositions do not contain the [ + Q] feature necessary for licensing the null quantifier. This factor is presumably also relevant for prepositions that assign oblique cases, with the added factor that failure to assign the required case results in a theta-theory violation anyway. With prepositions assigning genitive there are a few collocations in which the partitive genitive is acceptable; Borras and Christian (1971, 22-23) cite so smexu 'fromlaughter(gen2)', izlesu 'from the wood(gen2)', bezrazboru 'without discrimination(gen2)'as examples. Although this process is not productive and these are typically adverbial or set phrases, one might contend that in these constructions the preposition is quantificational, thereby licensing the empty quantifier. See also Fowler (1987a) for ideas along these lines. 80. More precisely, Pesetsky argues that category selection actually takes place at the level of LF', which is one step removed from LF, but this distinction is immaterial to the points at hand. 81. A reasonable candidate for selecting QP objects, however, are verbs prefixed by quantificational na-. Such verbs take either bare genitive or overtly quantified complements: (i)

on nakupil knig/ mnogo knig/ sto knig he bought books(gen pl)/many books(gen pl)/hundred books(gen pi) 'he bought a lot of/many/a hundred books'

218

PARAMETERS OF SLAVIC MORPHOSYNTAX

The verb nakupif 'to buy a lot of presumably subcategorizes for QP objects (perhaps via CSR of the semantically selected type K), and Q can be either null or overt. The impossibility of mnogix knig in (ii) shows, however, that nakupit' differs from verbs such as bojat'sja 'to fear' that truly govern GEN: (ii)

on bojalsja/*nakupil mnogix knig/ sta knig he feared/ bought many(gen pi) books(gen pi) /hundred(gen) books(gen pi)

See Fowler (1987a) for further discussion and examples, as well as Pinon (forthcoming) for a treatment of na- verbs in Polish. 82. Scrambling makes surface word-order considerations irrelevant in determining underlying grammatical relations here, so that (142) could have just as well been expressed as saxaru bylo dobavleno v smes' 'some sugar was added to the mixture'. 83. The "large quantity" reading also exists in Serbo-Croatian and Polish, although it is even more marked than in Russian. Speakers judge Polish ludzi siet zebra-to 'people(gen pi) REFL gathered(n sg)' as odd and Serbo-Croatian ljudi je doslo 'people(gen pi) AUX(3sg) arrived(n sg)' as archaic. In Czech, partitives are disappearing, a phenomenon that, as discussed in section 5.3.2.2, seems to correlate with the lack of genitive time/distance phrases under negation. 84. This actually seems doubtful for binding, since DPs can bind reflexive possessives that may be formally APs. Pesetsky's use of categorial mismatch pertains to antecedent binding for purposes of satisfying the ECP only. 85. Crockett (1976b, 325) does cite as acceptable an example with the unergative verb rabotat' 'to work' that could not plausibly be derived through movement: ucitelej v nix rabotalo nemnogim bolee 12,4 tysjac celovek 'teachers(gen pi) in them worked(n) a-little over 12.4 thousand people(gen pi)'. Speakers consulted strongly prefer this without celovek 'people', suggesting movement, and clearly contrast ucitelej tarn rabotalopjat' 'teachers(gen pi) there worked(n) five', where scrambling might have occurred, with *ucitelej tarn rabotalo tri 'teachers(gen pi) there worked(n) three', where the form ucitelej could only have been basegenerated as such. 86. These examples would also be unacceptable with initial dative druz'jam 'friends' or instrumental fabrikami 'factories' (except as scrambled structures, since then there would be no null Q to identify). 87. It may be that the contrastive rather than thematic reading of the IQE is much more dominant in these languages than in Russian, one Polish speaker suggesting that (169b), for example, could only be followed by something like 'and (of) magazines I bought six'. 88. An alternative possibility to be considered in section 5.3.2.2 is that these nonargument genitives might be treated as partitives. 89. Maling (1993) and Kim and Maling (1993) argue convincingly that comparable adjunct phenomena in Finnish and Korean, respectively, should be analyzed purely in terms of structural case assignment. 90. Neidle (1988, 62, n. 9) mistakenly claims that in Franks (1986) I assume a null quantifier in genitive NPs under negation. In that article I actually only discuss partitive and GIS structures, arguing for a null quantifier in these structures, but make no similar implications for the genitive of negation. I have thus always been in agreement with Neidle on distinguishing the two types of bare genitives analyzed in section 5.3. 91. Although her basic point is correct, Neidle mistakenly deems the status of the partitive genitive in Polish and Czech as equivalent. In fact, as discussed later, this process is of extremely limited productivity in spoken Czech. 92. Similar observations are made by Neidle (1988, 101).

QUANTIFIED STRUCTURES: POLISH AND OTHER PUZZLES

219

93. Note that this conflict is only visible because the paucal numeral dva 'two' assigns the paucal rather than genitive plural. 94. I stress this point because it has occasionally been claimed that the genitive of negation in Russian can never override GEN(Q), but examples of this phenomenon are relatively common (cf. e.g. Dziwerek, 1991). Those cited in (182)-(184) have been provided by Natasha Kondrashova. 95. Judgments provided by Katya Zubritskaya. 96. This alone might account for the tendency for numeral phrases under negation to appear in the accusative rather than genitive, if they typically undergo QR. 97. It should be noted that there is no reliable test for unaccusativity in Polish; an analysis of (190b, c) and related constructions will be presented in chapter 8. The sole exception to the transitive generalization is the copula bye 'to be', which requires the genitive under negation. 98. This assumption, which is in keeping with Babby's Principle of Inertness, has recently been challenged by the framework of Distributed Morphology espoused in Halle (1990) and Halle and Marantz (1992). 99. Since of the languages considered here only Russian ever has a distinct genitive2 form for the partitive, it is impossible morphologically to distinguish the partitive genitive from the genitive of negation in Polish, Czech, Slovenian, or Serbo-Croatian. 100. In making his claims, Lasnik (1992a) was relying on data provided by Willim (1990), who contended that the genitive of negation could not apply to bare NP adverbials. Unfortunately, in her sole example, deszcz nie padai trzy godzinyl*trzech godzin 'rain NEG fell three hours(acc)/(gen)', the genitive is pragmatically unlikely, and she did not compare Polish to Russian at all. Lasnik drew his Russian examples from Pesetsky (1982) and would in closer investigation have found that comparable examples of genitive non-arguments under negation are grammatical in Polish as well, as the data in (194) amply demonstrate. Of course, had he been aware of the facts, Lasnik could have availed himself of Dziwirek's idea that these are really partitives, but then there would be no substantive difference between the two languages for him to report. 101. These data were provided by Jindfich Toman, and the Serbo-Croatian judgments in (196) are due to Ljiljana Progovac. 102. Other reasonable possibilities are that they are adjoined to or in the specifier of AspP, which I have not indicated in (201) but presumably intervenes between NegP and VP, as in (139).

6 Secondary Predication

This chapter deals with specific issues related to the syntax of secondary predicates in Slavic.1 Its primary goals are (1) to examine the conditions affecting case agreement between secondary predicates and their antecedents, and (2) to develop an account of the case that arises when agreement is blocked. Section 6.1 introduces the various types of predicate adjectives that exist in Russian and that will figure in subsequent discussion in the chapter. These differ in terms of the case they bear when they for some reason fail to agree; one particular problem to be addressed in detail is the form of the Russian semipredicatives odin 'one' and sam 'alone', whose non-agreement case is the so-called second dative, a term due to Comrie (1974). In section 6.2 it will be proposed that an appropriate solution should be formulated in terms of control theory, and that agreement is possible only when the understood PRO subject of an infinitival clause is lexically governed. Section 6.3 attempts to explain the second dative in the same way as dative subjects of infinitives, that is, by invoking a marked phrase structural case assignment strategy. This configurational analysis is integrated into the internal subject hypothesis adopted in chapter 4 and is then shown to provide an insightful account of certain mysterious gaps in the second dative paradigm. Finally, in section 6.4, comparative Slavic arguments are adduced to motivate the phrase structural source for the second dative, as well as to support the connection between case agreement and obligatory control proposed in section 6.2.

6.1. Predicate Adjectives This section presents familiar data on the possible forms of predicate adjectives in Russian in various constructions. The distinct behaviors of regular (long form) predicate adjectives, (pronominal) semipredicative adjectives, and short form adjectives are contrasted. These differ in their distribution, ability to agree with the NP2 they predicate, and default "non-agreement" case. Explicit proposals will be made to accommodate these differences in configurational and morphological terms. 220

SECONDARY PREDICATION

221

Although the source of the default instrumental will receive an explanation here, discussion of the second dative is deferred to section 6.3. 6.1.1. Three Types of Predicate Adjective Various categories can serve as predicates. In this chapter we will be concerned with the form and function of predicate adjectives.3 Predicate adjectives may be either primary or secondary predicates, where a secondary predicate makes an adjunct predication of some item in the sentence, auxiliary to the main subject-predicate relation. Although ordinary attributive adjectives agree with the nouns they modify in gender/number and case, this is not necessarily true of predicate adjectives. Predicate adjectives in Russian exhibit three options with respect to how they reflect the relevant features4 of the NP they predicate: they may (1) agree with this NP in all pronominal features, including case, (2) agree with this NP in all features except case, receiving instead a default case by some other mechanism, or (3) agree with this NP in all features except case but not receive a default case, appearing instead in the caseless '' short form.'' The fact that there are certain constructions in which case agreement does not take place although gender/number agreement does suggests that these are fundamentally distinct processes. In order to distinguish case from person, number, and gender, I shall refer to the latter set as pronominal features. The behavior of secondary predicates also supports the idea that case constitutes its own feature submatrix, independent of the pronominal features, although whether the pronominal features should simply be combined together or some further organization will need to be imposed is unclear. Short form adjectives are never used attributively. They cannot appear in NPs because they are formally caseless, bearing only the gender and number features of nouns and /-participle verb forms. They only occur in predicative position and are extremely restricted in usage in the modern language. As secondary predicates they are generally marginal and/or stylistically marked; usage also depends on the particular adjective involved. As primary predicates they are most common after the copula byt' 'to be', and also appear with a small list of other equative and resultative verbs (cf. Nichols, 1981, 298). Opposed to this type of predicate adjective are those that bear case features. Ordinary predicate adjectives exhibit either (1) the nominative, which appears under agreement with a nominative antecedent, although this is possible only in a limited range of constructions, or (2) the instrumental, which is always available (if there is a verb) and is consequently the predicate adjective's most prevalent and stylistically preferred realization. The third type of predicate adjective, the semipredicatives odin 'one' and sum 'alone', will be introduced after we have examined the agreement properties of ordinary predicate adjectives. In simple clauses the following three variants are all conceivable when the adjective is predicated of a subject NP:5 (1) Ivan prise'l domoj p'janyj/p'janym/?p'jan (m nom) came home drunk(m nom)/(m inst)/(m) Here, the adjective serves as a secondary predication on the subject Ivan, which is masculine singular. It must therefore agree in pronominal features with Ivan; since

222

PARAMETERS OF SLAVIC MORPHOSYNTAX

gender in Russian is not distinguished in the plural, only gender (m, f, or n) or number (pi) is specified in the glosses, not both. Matching in these features is taken to be a requirement for predication, in that predicates that bear pronominal features distinct from their subjects cannot be appropriately construed. This seems to be a general property of binding, which is presumably checked at LF. With respect to case, however, the predicate adjective is free to agree in case, appear in the default instrumental, or receive no case at all. Secondary predication on object and oblique NPs, on the other hand, allows only the long form, although the instrumental is strongly favored, as shown in (2). (2) a. ja vstretil svoego druga p'janogo/p'janym/*p'jan I met my friend(m ace) drunk(m acc)/(m inst)/(m) b. Vera boitsja Ivana p'janogo/p'janym/*p'jan Vera fears (m gen) drunk(m gen)/(m inst)/(m) c. Vera emu dala den'gi p'janomu/p'janym/*p'jan Vera him(m dat) gave money drunk(m dat)/(m inst)/(m) In (2a) the AP is predicated of an accusative direct object NP, in (2b) of a genitive direct object NP, and in (2c) of a dative indirect object NP. In all three, agreement with this NP is possible, albeit somewhat forced.6 Of greater concern is the perplexing range of agreement possibilities exhibited by predicate adjectives in embedded infinitival clauses. Essentially, only a PRO that is obligatorily controlled by a higher subject is able to license long-distance agreement, although the instrumental is considerably preferred. Otherwise, as with object controlled PRO, the instrumental is obligatory. Contrast the examples in (3) and (4). (3)

Ivan ne xocet [PRO prijti domoj p'janyj/p'janym] (m nom) NEG wants to-come home drunk(m nom)/(m inst) 'Ivan doesn't want to come home drunk'

(4)

ja poprosil Ivana [PRO ne prixodit' *p'janogo/p'janym] I asked (m ace) NEG to-come drunk(m acc)/(m inst) 'I asked Ivan not to come drunk'

In the next section instances of non-obligatorily controlled PRO, which pattern with (4), will also be considered. In short, agreement of the secondary predicate in case with the controller of PRO is optional when this is a nominative subject but impossible under all other conditions. Furthermore, the non-agreeing adjective always appears in the instrumental. The caseless short form is extremely marginal in these embedded structures in the modern language, and its distribution is somewhat tangential to the main concerns of this chapter, so judgments will be provided only when necessary. In addition to the secondary predicate adjectives described above, there is also a pair of semipredicative pronouns in Russian, odin 'alone' and mm 'oneself, which can be used to form a secondary predication. The unusual behavior of these items has been subjected to much scrutiny since the publication of Comrie's seminal 1974 article (cf. e.g. Greenberg, 1983; Neidle, 1988; Schein, 1982). These semipredicatives, however, have quite different case properties from ordinary predicate adjec-

SECONDARY PREDICATION

223

tives. Consider the examples in (5)-(8), where the case of the semipredicative is the only one possible. (5)

ja vernulsja odin I(mnom) returned alone(m nom) 'I returned alone'

(6)

ja na§el ego odnogo I found him(m ace) alone(m ace) 'I found him alone'

(7)

ja xoc"u [PRO prijti odna] I(f nom) want to-come alone(f nom) 'I want to come alone'

(8)

ja poprosil Veru [PRO pojti odnoj] I asked (face) to-go alone(f dat) 'I asked Vera to go alone'

As can be seen from (5)-(8), the semipredicatives must always agree in case with an NP in their own clause, regardless of whether that NP is a subject or an object. In embedded clauses, the semipredicatives also agree with the controller of PRO in exactly the same environments as ordinary predicate adjectives do, as demonstrated by examples (7) and (8). Once again, in contrast to predicate adjectives, there is never any choice as to whether or not they agree. Moreover, semipredicatives differ from ordinary predicate adjectives in that their non-agreeing case is the dative rather than the expected instrumental. To summarize, semipredicatives always agree in case with the NPs they predicate in a simple clause, whereas ordinary secondary predicates may agree in case or not. In embedded clauses, both types of adjective are able to agree with the controller of PRO under conditions to be explored below; as in simple clauses, this agreement is obligatory for semipredicatives but optional for ordinary secondary predicate adjectives. That is, semipredicatives may appear in their default case only when agreement is otherwise impossible, but ordinary secondary predicates always have this option. Instrumental is the non-agreeing case of ordinary predicate adjectives, but dative is the non-agreeing case of semipredicatives. The semipredicatives never appear in the default instrumental available to all other secondary predicates. On the other hand, ordinary predicate adjectives never appear in the default dative.7 Several other characteristics of these two semipredicatives are worth observing as well. For one thing, they actually belong to a distinct morphological class, the socalled pronominal declension, and are thus opposed to ordinary adjectives. This declension class exhibits nominal endings in the direct cases and adjectival ones in the oblique cases. Other adjectival elements that display this mixed pattern are demonstratives, such as etot 'this', and possessive adjectives, such as nas 'our' and mamin 'mother's'. The semipredicatives are thus in morphological terms more nominal than other adjectives are. Note that because the nominative forms of odin and sam are nominal, they are formally identical to caseless short form adjectives. However, since short forms are rare except in copular constructions, in clear contrast

224

PARAMETERS OF SLAVIC MORPHOSYNTAX

to the semipredicatives, and agreement for them is obligatory whenever possible, I assume that these are actually agreeing in case.8 On the other hand, in semantic terms they seem to be more adverbial than adjectival. Indeed, we shall see in section 6.4 that sam 'alone' tends to become adverbialized, although differently in the various Slavic languages, by which I mean that instead of showing pronominal and case features it is replaced by a fixed form.

6.1.2. On Agreement In this section, structures are posited to explain the range of predicate adjective agreement phenomena surveyed above. Here I consider simple clauses, examining predicate adjective agreement in complex and control constructions in more detail in section 6.2. I contend, first of all, that the impossibility of the short form in (2) follows from a mutual c-command requirement on predication, roughly as proposed in Williams (1980) and adopted in Hornstein and Lightfoot (1987) and Franks and Hornstein (1992), for example (cf. also Rothstein, 1991). The assumption that the short form is exclusively VP-external, perhaps due to its semi-adverbial status, means that it can only predicate a subject NP, since all other arguments of the verb are dominated by VP and thus would fail to c-command the short form. This assumption also explains its predominance and greater felicity in equative constructions with the copula byt', as in (9). (9) a. moj drug p'jan/p'janyj/*p'janym my friend(mnom) drunk(m)/(m nom)/(m inst) b. moj drug byl p'jan/p'janyj/p'janym my friend(m nom) was(m) drunk(m)/(m nom)/(m inst) The short form is perfectly felicitous in both present (9a) and past (9b), as is the agreeing nominative long form. Note that the instrumental is only viable when there is a lexical copula, that is, in the past and future tenses but crucially not in the present. An analysis of these facts will be developed in the next section. Recall that according to the VP-internal subject hypothesis adopted in chapter 4, the subject is underlyingly inside the maximal VP in NP* position. There are several competing versions of the ISH, depending on the exact locating of the subject (cf. Huang, 1993, for comparisons). Imagine, for example, that the subject is generated in NP* position in a kind of "small clause" (SC) structure, as in (10) and following Koopman and Sportiche (1988), from which it subsequently raises to NP' position. (10)

[Cp [IP [NP- e] [r KNFL) [VP-na* [Np. SUBJECT] [VP . . . ]]]]]

Now observe that adopting this particular model of phrase structure leads to several possible interpretations of the c-command requirement. First of all, it might mean that the short form adjective is "VP-external" in the sense that it is external to the lower VP, although still, like the subject, under VPmax. On the other hand, if the mutual c-command requirement is not a D-structure requirement but rather can be met after NP-movement (at S-structure or conceivably LF), then the short form must indeed be external to VPmax. Of course, under this set of assumptions it, like the subject, might raise from a VP-internal position (presumably to adjoin to IP). Alter-

SECONDARY PREDICATION

225

natively, one might relax the mutual c-command requirement so that only subjects are required to c-command their predicates, but not vice versa. If so, all that needs to be stipulated is that short form adjectives cannot be generated under V, assuming a first-branching node definition of c-command. Second, I maintain that ordinary long form predicate adjectives are actually part of small clause APs with PRO subjects, as in Hornstein and Lightfoot (1987).9 They are therefore actually predicated of these PRO subjects, whose reference is derived from the overt controller. The representations of (1) and (2a), with long form agreeing adjectives, will thus be roughly as follows: (11) a. tcp tip Ivan [VP [VP prisel domoj] [AP«">* PRO [AP p'janyj]]]]] b. [Cp tip Ja tvp tvp vstretil svoego druga] [APmax PRO [AP p'janogo]]]]] Although I am intentionally inexplicit about the exact position of the small clause AP, as an adjunct it is presumably in some kind of adjoined position from which the PRO is controlled. Following Hornstein and Lightfoot (1987), who argue that the small clause is syntactically comparable to an adverb, I assume that it is adjoined to VP, hence governed by the verb. As the subject of the SC, the PRO is also governed by V, which is what renders it anaphoric hence obligatorily controlled. This is what causes the secondary predicate to agree with the controller of the SC PRO, according to the theory of control to be adopted in section 6.2.3. This analysis means that there are technically two processes at work in inducing or preventing agreement—predication of PRO by the adjective and control of PRO by an antecedent. Both processes are formally represented in terms of coindexation. That is, the adjective is coindexed with PRO by predication, following Williams (1980), who argues that a rule of predication coindexation applies to coindex a subject with its predicate, and PRO is coindexed with its antecedent by control, as proposed in Franks (1983, 1985) and Franks and Hornstein (1992). This chain of identical indices is what allows for agreement in case, according to the theory of case assignment by coindexation developed in chapter 2. The idea is essentially that the PRO subject of the adjectival SC is coindexed both with its controller and with the predicate AP. Even though PRO in Russian lacks case, it is able to participate in the process of "transmitting" case features from its controller to the predicate AP. The possibility of such a mechanism follows from the case submatrix model. It is more or less accidental that PRO in Russian has no case submatrix.10 UG requires NPs to be case-marked, but the Case Filter, however construed, does not extend to non-theta-marked categories such as APs. Nothing in principle prevents a language from also endowing adjectives with morphological case, however; it is just that nothing in UG requires this. On the other hand, according to Stowell (1981), certain theta-role-bearing categories, such as infinitive phrases and PPs, differ from NPs in that they do not need to be assigned case. Similarly, it is standardly assumed that PRO is somehow exceptional in being able to satisfy the Case Filter intrinsically, although the reason for its exceptionality is far from clear. Be that as it may, this means PRO need not have a case submatrix, hence, following general considerations of parsimony, in most languages it does not. PRO nonetheless bears the index of its controller through control and, in turn, shares this index with its predicate. Consequently, the antecedent NP and the predi-

226

PARAMETERS OF SLAVIC MORPHOSYNTAX

cate AP—both of which themselves do have case submatrices—are coindexed, and the values of the case features of the AP are set accordingly. Of course, the adjective also has the option of appearing in the instrumental as a default case. This instrumental appears to be comparable to the instrumental on NPs adjoined to VP. Since this was argued in chapter 2 to be a consquence of the ' 'sister of VP" configuration, I suggest that the instrumental in these SCs can be similarly handled. Two possibilities for the default instrumental come to mind. The most straightforward is simply to adjoin the AP to VP, as in (12). (12) a. [CP [IP Ivan [VP [VP prise'l domoj] [AP p'janym]]]] b- [Cp tip Ja [Vp [VP vstretil svoego druga] [AP p'janym]]]] The instrumental thus arises under the standard "sister of VP" configuration and nothing more need be said. A more complicated approach might be to make use of the same structure as in (11), and then to claim that the instrumental option in the CAP CNP PRO] UP:INST • • • ]] structure is a generalization of the "sister of VP" configuration to sisters of Xmax, specifically NP. Either way, this would explain why the instrumental is always available yet optional—the structure allows it, but the fact that these are adjuncts means that no particular case is required for theta-role visibility, hence other case-marking strategies are equally viable when available. I shall adopt the former approach, that instrumental predicate adjectives are simply APs adjoined to VP, which is more direct and requires fewer special assumptions. Finally, the small clause structure does not seem appropriate for the semipredicatives sam and odin. We have seen that, like ordinary predicate adjectives, they may agree with their antecedents or appear in a default form, but unlike them the default form is dative, not instrumental, and this alternative is only available when agreement fails. In order to explain this, I propose that odin and sam differ from ordinary predicate adjectives in that they never appear in small clause APs. Since no PRO mediates between the semipredicative and its antecedent NP, control is not invoked, just agreement. Of course, although this accounts for the unavailability of the instrumental and the consequent obligatory agreement, it says nothing about the cause of the second dative. I will address this important issue in section 6.3, where it will be argued that this phenomenon can be understood as direct case assignment to the semipredicative, suitably positioned.

6.1.3. Clausal Functional Projections In this section I consider in more detail the interplay between tense and agreement features in Russian, as well as their possible structural manifestations. It is argued that within tense there exists a [±past] opposition, and within agreement a [ ± person] opposition. These are related in that a plus value for one of these morphosyntactic categories demands a minus value for the other. In this light, I then return to copular constructions, proposing a specific analysis of them in terms of byt' 'be' as a raising verb taking an underlying SC complement. Recent approaches to clausal phrase structure, stemming from ideas in Pollock (1989) and Chomsky (1991), treat functional categories such as tense, agreement, mood, and aspect as formally comparable to lexical categories such as V and N.11

SECONDARY PREDICATION

227

Each of these functional categories thus serves as an independent head and projects up its own phrase. These heads, however, consist of purely grammatical features that are typically realized in combination with some kind of verbal element. This might occur either through incorporation (understood as head adjunction; "incorporation' ' in the sense of Baker, 1988) or morphological unification (understood as a feature-driven process), although it is reasonable to suppose that adjunction must first take place regardless of whether the particular features are combined synthetically or analytically. This kind of approach endows the clause with an extremely hierarchical binarybranching structure. It is by now standard to separate out from I(NFL) at least the clausal properties of T(ense) and Agr(eement), which had been treated in various ways under the CP-IP system of Chomsky (1986a), and to allow them to project Tense and Agreement Phrases, respectively. These new phrases, which are extended functional projections of the verb in the sense of Grimshaw (1991), were previously typically represented as features on I and/or C. Of course, as discussed in chapter 2, node labels are nothing more than abbreviations for feature complexes, hence the claim that a node such as I bears agreement features really means that it is those features. There is thus no a priori reason either why features for tense and agreement or case and gender, for example, should be combined in the same node, or why they should not be able to serve as heads of phrases. Keeping these types of features syntactically separate seems especially reasonable in view of the argument that they constitute independent submatrices, within the feature theory developed in this book. In fact, one might regard submatrices themselves as S-structure artifacts of combining distinct underlying functional heads with the lexical categories that determine and eventually host them. Within such a model, at D-structure all feature complexes would be pure representations of a single morphosyntactic category— either a lexical one if part-of-speech features are involved or a functional one if grammatical features are involved. There would be a one-to-one pairing of S-structure submatrices and D-structure heads, with functional feature complexes accruing to lexical heads in the course of the derivation. That is, all submatrices would arise through syntactic incorporation by adjoining lexical heads to functional ones, or vice versa if lowering is permitted. Of course, such a system would proliferate functional categories and would necessitate treating all submatrices as independent heads, in both nominal and verbal morphosyntax. For Russian nouns, for example, at least case and gender would have to constitute distinct heads; number could be regarded as a fourth gender and person could be disregarded except for pronouns. For Russian verbs, at least tense, agreement, and aspect would all have to be separate projections. In this book I remain agnostic about whether such a radical analysis is motivated, since I feel that unbridled proliferation of functional categories leads to considerable derivational uncertainty. I will only separate out functional heads to make specific points of analysis.12 The problem is especially compounded by the standard assumption that heads may either raise or lower to incorporate, lowering being essentially undone at LF to satisfy proper binding (cf. Chomsky, 1991). All this creates what is to my mind a particularly severe learnability problem. Once the CP-IP system of

228

PARAMETERS OF SLAVIC MORPHOSYNTAX

"Extended X-bar Syntax" is extended to a host of potential functional heads, so long as they display some sort of morphological or sometimes simply semantic reality, the abstractness and diversity of possible analyses multiplies drastically. Conceivable parameters of variation for any given grammatical system include (1) the range of operative functional categories, (2) the relative scope of these categories, and (3) the direction/order in which these combine with each other and/or lexical heads. As the discussion of Slavic clauses will reveal, evidence for independent TPs and AgrPs is not consistent across languages. Although they are both extended projections of the verb, it is unclear which should have wide scope and whether head raising or lowering is required; rather different structures and analyses have been proposed for different languages. What is needed in order to make the functional projection system work, of course, is a way to contain the explosion of possible analyses by developing a more restrictive conception of functional categories within UG.13 Let us nonetheless reconsider the specific properties of the Russian system without resolving these more general issues relating to the theory of functional categories. I briefly sketch the most.relevant facts about clausal functional categories in Russian. First of all, Russian verbs reflect grammatical aspect, with imperfective productively marked as a suffix and perfective somewhat more idiosyncratically marked as a prefix. Overt marking of tense only occurs in the past, where the suffix -/ is added to the verbal stem (of either aspect). This tense marker is followed only by the nominal gender/number endings -0 for masculine, -a for feminine, -o for neuter, and -i for plural, never by fully fledged person agreement, which distinguishes first, second, and third person in both singular and plural. Instead, the present-future opposition is established by means of the following interaction between agreement and aspect: imperfective conjugated verbs have a present tense meaning; perfective conjugated verbs have a future tense meaning. Additionally, there is a periphrastic imperfective future formed by the conjugated verb byt' 'to be' plus the imperfective infinitive; there is no present perfective. These facts are summarized in (13), as exemplified by the verb (s)delaf 'to do'. (13) a. PAST

IMPERFECTIVE PERFECTIVE b. PRESENT IMPERFECTIVE c. FUTURE PERFECTIVE IMPERFECTIVE

MASCULINE

FEMININE

delal delala sdelal sdelala ISO 2SG 3SG delaju delaes' delaet 1SG 2SG 3SG sdelaju sdelaes' sdelaet 1SG budu delat' 1PL 2SG budes' delat' 2PL 3SG budet delat' 3PL

NEUTER

PLURAL

delalo delali sdelalo sdelali 1PL 2PL delaem delaete 1PL 2PL sdelaem sdelaete budem delat' budete delat' budut delat'

3PL delajut 3PL sdelajut

The verbal stem is delaj-, with the prefix s- in the perfective. Suffixes beginning in consonants, such as past tense -/ and infinitival -t', cause the stem-final jot to delete, suffixes beginning in vowels, such as the conjugational endings, do not.14 These facts suggest a certain complementary distributionality of tense and agree-

SECONDARY PREDICATION

229

ment in Russian, in that only one of these can be overtly marked on a verb at a time. They also suggest that within tense the only feature opposition necessary is [ ± past] and within agreement [ ± person], assuming aspect to determine present or future reference and gender agreement to result from some independent process, either through SPEC-head agreement or predication (subject-predicate coindexation). That is, the clause may be marked [ ± tense], and if [ + tense] then it is additionally specified for [±past], and the clause may also be marked [±agreement], and if [ + agreement] then it is additionally specified for [ ± person]. However, the following bidirectional implications seem to hold: (14) a. [ + past] <—> [ — person] b. [ — past] <—> [ +person] In other words, tense and agreement, as well as the features dependent on them, vary independently, although in accordance with (14). There are essentially two ways of instantiating these dependencies. One might literally contend that tense and agreement are in complementary distribution in Russian. This would entail, however, that tense and agreement be generated under the same node, presumably INFL, so that either could occupy INFL, but not both simultaneously. This is conceptually undesirable given the current theory of functional categories, although it is certainly possible that such a conflation is necessary in order to describe Russian properly.15 Perhaps more significant is the fact that it would treat -/ as the only marker of tense and the person/number endings as the only markers of agreement. That is, tense would be interpreted as "past," agreement on imperfective verbs would be interpreted as "present," and agreement on perfective verbs would be interpreted as "future." Under the complementary distribution analysis, there would be no [±past] and [± person] features per se, instead (13a) would bear only tense and (13b, c) only agreement. This set of assumptions raises questions about the role of tense and agreement in UG, however, since these are generally argued to be present even if not overt. Furthermore, it is difficult to imagine how nominative case assignment would work if agreement and tense were mutually exclusive, since subjects would be equally marked nominative whether INFL contained tense or agreement. Finally, the existence of dative subjects of tensed non-agreeing verbs, as discussed later in this chapter, would pose a prima facie difficulty for this system. For these reasons, I reject the complementary distribution approach in favor of one that allows for abstract tense and agreement features in all clauses. Under this alternative, the dependencies in (14) will have to be selectional in nature, or perhaps determined by available morphology. Tense and agreement either could arise under separate nodes, as in the Pollock (1989) system, or they could both be generated under INFL, as in the Chomsky (1986a) system. I opt here for the latter approach, mostly for reasons of simplicity. Similarly, rather than allowing aspect to head its own phrase and take a VP complement, as in section 5.3.2, I will follow more traditional treatments in which aspect is represented as a feature on the verb.16 Finally, following Baker (1991), I have represented sentential negation as simply left-adjoined to V, rather than as in the last chapter as a head of a phrase.17 The schematic D-structure of a Russian sentence is thus as follows:

230

(15)

PARAMETERS OF SLAVIC MORPHOSYNTAX

CP x

c

C

IP

r \. VP

i TNS

AGR

NP

V

(ne) (t±past]>

([±person])

V V ([+perf])

As before, the subject NP is generated under VP; here I have employed Kuroda's analysis, in which the subject originates as SPEC-VP. From there it raises to SPECIP in search of case.18 Whether TNS expands to [ ± past] and AGR to [ ± person] is optional, as is, of course, the presence of negation. However, as stated in (14), only one of TNS or AGR can—in fact, must—actually contain specific feature values, even though both are ordinarily present in I.19 This set of possibilities and the apparent dependency between TNS and AGR turns out to follow from rather natural assumptions about phrase structure. Specifically, by base-generating a split INFL node, I am claiming that it is possible for an X° to have a bipartite head. Note that this is comparable to the kind of structure that arises through incorporation, as in Baker (1988), except that with incorporation each head projects its own phrase at D-stracture so that under head adjunction the adjoined element remains the head of its underlying phrase, although transferring some of its properties (in particular its government domain) to the X° dominating it. There is, nonetheless, a kind of competition between TNS and AGR for head status of I in that only one can percolate its features up the projection. Thus, either TNS can contain [±past] features or AGR can contain [ ±person] features, but not both.20 The fact that past and person cannot coexist in Russian presumably derives from the assumption that these features must percolate up the tree at least to I in order to c-command the VP. They are incompatible within a single node, however, so only past or person can be expressed on I at any one time. Notice that I have been somewhat indiscriminate about referring to these elements

SECONDARY PREDICATION

231

using privative features such as TNS or polar ones such as [±past]. I believe that both kinds of features are probably necessary in syntactic representations, importing an essentially Jakobsonian conception of features into X-bar syntax.21 TNS and AGR are reasonable privative features, since they designate presence of general categories. The situation with past and person is not so clear; on the one hand TNS [ - past] and AGR [ - person] might be taken specifically to mean ' 'non-past tense'' and "non-person agreement," respectively, but on the other hand there is no obvious reason why PAST and PERSON could not be taken themselves as privative features, the latter containing valued (probably polar) features for the six different possibilities. TNS without PAST would simply be interpreted as non-past, with verbal aspect differentiating present from future time reference, and AGR without PERSON would simply be interpreted as gender agreement. These two conceptions of how past and person are represented entail somewhat different approaches to the interaction between headedness and presence of features. With [±past], for example, we would need to claim that [ + past] must be a head feature and thus percolates up to I in order to c-command V, whereas [ — past] does not. That is, both TNS and AGR as X°s can have features, but only a feature that percolates up to the c-commanding X° can be fully realized morphologically, and positive feature values must be realized. This view is a bit problematic in the past tense, however, as in (13a), since here the past marker is followed by gender agreement. If the latter is actually in some way an instantiation of AGR, then both past and AGR can be simultaneously realized. In fact, my motivation for splitting TNS and AGR in (15) was to no mean degree to produce precisely this linear order of morphemes in the past tense. But then, assuming gender agreement to be a SPEChead relationship, with the subject NP raising to IP-specifier position, AGR must still in some sense be a head even when overt TNS is indicated. If so, I must be a composite of TNS and AGR, with the relevant restriction being that AGR cannot contain person features if TNS contains tense features. Alternatively, if PAST and PERSON are regarded as privative features, then I can have the following structures:

(16) a.

TNS

I

b.

AGR

TNS

PERSON

PAST

I

AGR

Under either approach, only one of the bipartite heads TNS and AGR can tolerate further specification. There are several conceptual advantages to adopting such a compromise solution to the representation of tense and agreement in Russian, I believe, even if formal details of the analysis must still be worked out. Both functional categories have abstract representation even when not fully realized morphologically, their apparent

232

PARAMETERS OF SLAVIC MORPHOSYNTAX

complementary distribution deriving from the fact that they jointly head a single projection. One reason for representing TNS and AGR in this way is that it allows for a consistent relationship between I and the subject NP. This is important if we want the case of subjects to arise in a uniform manner. Nominative will be assigned to the specifier of IP if I contains TNS and AGR, regardless of their further specification in terms of past and person. Which of these is of primary significance is not readily evident, since the two generally exist in tandem. Presumably, since AGR usually depends on the presence of TNS, it is actually the AGR component of I that marks subjects nominative. This assumption is further justified by data from other languages in which non-tensed phrases (e.g. Portuguese infinitives, Hungarian NPs and PPs) that exhibit agreement in pronominal features also allow nominative subjects.22 These contain AGR but crucially lack TNS, yet nominative is nonetheless assigned. Of course, since Slavic lacks this option, AGR only existing in the scope of TNS, the point does not bear directly on the analysis at hand. However, treating TNS and AGR as potentially independent and claiming that nominative is a function of AGR rather than a function of TNS will allow for important insights into the origin of Russian dative subjects. I will argue that dative is assigned to the subject NP (in IP-specifier position) when I contains TNS but not AGR. These two categories in principle thus exist independently, although in Slavic AGR requires TNS but not vice versa. The analysis in section 6.3 will corroborate the status of AGR as distinct from tense, showing that although AGR in Russian can only appear in the presence of TNS, it is not required to do so. Furthermore, whether or not AGR is present will be argued to determine the case assigned to the subject NP of a tensed clause—nominative when abstract AGR is present (regardless of whether it exhibits person) and dative when it is absent. A second conceptual advantage of allowing the abstract representation of TNS and AGR, but localizing the idiosyncracies of Russian verbal morphology in the relationship between the two categories, is that this offers a natural way of accommodating cross-linguistic variation within the confines of UG. That is, TNS and AGR are universal categories, but languages vary in their instantiation. One familiar option is whether TNS selects for AGR or vice versa; Ouhalla (1991), for example, proposes this as a parameter. Russian appears to instantiate a different option, where both share a single X°. In the next chapter I will show that this restriction does not pertain to South and West Slavic languages, in which person may be marked even in the past, that is, in the presence of overt tense, as well as on the copula in the present. This latter fact is of course related to the properties of the verb byf 'to be'. In particular, let us assume that this verb is exclusively perfective.23 This assumption is consistent with its morphology and readily explains, moreover, why the copula only has past and future forms, as in (17).24 (17)

3. PAST

MASCULINE

FEMININE NEUTER PLURAL

byl

byla

b. PRESENT

0 throughout

c. FUTURE

ISO budu

2SG budes'

bylo

3SG 1PL budet budem

byli 2PL 3PL budete budut

SECONDARY PREDICATION

233

Compare this paradigm with the full one in (13). Like any verb, byf either shows TNS in the form of past or AGR in the form of person. However, since it is perfective, the conjugated forms have future meaning. In the present tense there is simply nothing for AGR specified for person to attach to. This option presumably reflects a structure where neither TNS nor AGR contains any subordinate features or, conceivably, AGR is specified for person, but these features delete since they are bound morphemes and there is no appropriate V available.25 This view of byf as formally perfective also explains the structure of the Russian periphrastic future, in which the copula serves as an auxiliary verb, followed by the imperfective infinitive. The future meaning is derived from the perfective character of the auxiliary, while the verb itself remains imperfective in aspect. In this way, even though there is a temporal opposition between present and future for imperfective verbs, there is no need for any tense features beyond [ ± past], since the future sense of the periphrastic future technically results from the perfective aspect of the verb's auxiliary component. Further support for its perfective status is the impossibility of *budu byt' as a periphrastic future, since byf is a perfective infinitive. Before turning to the issues of secondary predication and control central to the present chapter, I conclude this discussion with a brief consideration of equative constructions in Russian. Recall from the previous section that in the past and future predicate adjectives can be (agreeing) nominative, (non-agreeing) instrumental, or caseless short forms, while predicate nouns can be either nominative or instrumental. In the present, however, the instrumental option is lacking. It is important to recognize that the possibility of instrumental is directly related to the presence of the verb byt' 'to be'. The simplest hypothesis is thus that the instrumental here is a quirky case and that the verb byt' is lexically specified to assign instrumental case to its NP complement. There exists an emphatic and invariant form est', incidentally, which can be used in present tense constructions but never assigns instrumental case. It is no longer part of the conjugational paradigm of byt' and is thus presumably a realization of I rather than V; see Chvany (1975) for details of its usage. The point of view that instrumental is a quirky case here is corroborated by the fact that in Polish, in which bye 'to be' has a complete paradigm, the instrumental is also assigned in the present tense, as well as by the existence of a sizable class of verbs in Russian, many with the general meaning of 'being, becoming', that assign instrumental to their complements. The alternative to treating the instrumental after byt' as lexically specified would be to claim that it is purely configurational and should be explained as an instance of instrumental under sisterhood/adjunction to VP/XP, that is, the "adjunct" instrumental. I think, however, that the close association with the actual presence of a particular verb and the semantic similarity of this verb to others in the instrumental class argue for assimilating this use of instrumental to quirky rather than configurational case. The only potential problem is its apparent optionality—if this instrumental is treated as a quirky case, it should of course be obligatory. The fact that it is indeed obligatory on NP objects in Polish suggests that this is the correct approach and that in Russian two distinct structures are possible. That is, I assume that the nominative variant arises when byt' is not actually a transitive verb, so that agreement rather than direct case assignment is involved. Specifically, as a verb the following structure is assumed to be standard:

234

(18)

PARAMETERS OF SLAVIC MORPHOSYNTAX

[„, [, TNS AGR] [VP NP [v. be[+perq NPINST/AP ]]]

The verb assigns inherent instrumental to its complement, and the subject NP, as always, raises to SPEC-IP in search of case. Since in Russian byt' only exists in the perfective, this structure is unavailable in the present. The other structure involves a small clause, roughly as in (19). (19)

[IP [j TNS AGR] tsc NP NP/AP]]

The first NP again raises to SPEC-IP for case, but here the second NP has no choice but to agree. Notice that this analysis is only consistent with the "sister of VP" approach to the default instrumental, since we crucially do not want instrumental to be an option in small clauses. The structure in (19) thus provides further support for rejecting the alternative account of default instrumental suggested above. The TNS + AGR complex is realized in the past and future using the copula byt', and as 0 in the present. Note that this SC differs from the secondary predication ones introduced above in that it has an overt rather than PRO subject. In both, the subject of the SC is governed.26 Hence, by the theory of Hornstein and Lightfoot (1987) to be discussed in the next section, when PRO occupies this position it will be anaphoric and so must be bound within its governing category. For secondary predicates there will be a primary subject to serve this function, but in the SC structure in (19) there is no other subject argument to c-command and bind. Hence, only an overt subject is possible in SCs in raising contexts. Notice, finally, that I have indicated in both these representations that the predicate may be either nominal or adjectival. This is motivated by cross-linguistic considerations, although it introduces a certain redundancy into the system. In particular, I assume that APs do not bear theta-roles and are not assigned case by Vs, so that predicate APs can only receive case through predication, that is, agreement. The instrumental APs in structure (18) are therefore actually embedded within NPs with null heads. Next, since in Polish the nominative option for predicative NPs is unavailable, I must assume that structure (19) is absent in this language. There is, in general, no evidence for small clauses in Polish; not even the verb uwazac 'to consider' takes an SC complement of this type, using the preposition za 'as, for' instead. If so, the nominative on APs must have a different possible origin, as in (18), with agreement inducing nominative rather than instrumental.

6.2. Secondary Predication and Control I now leave these somewhat murky issues of clausal structure to concentrate on the interaction between secondary predication and control. Agreement of the predicate adjective with the ultimate antecedent of PRO is only possible in certain instances. We first survey the range of constructions where this agreement is possible or blocked and then attempt to correlate the phenomenon with obligatory control, understood in very restrictive terms.

SECONDARY PREDICATION

235

6.2.1. Case Transmission We have seen that the phenomenon of long-distance case agreement may take place with both ordinary predicate adjectives and the semipredicatives sam and odin. As shown in (3)-(4) and (7)-(8), such long-distance agreement sometimes obtains and sometimes is blocked. Basically, it occurs when the predicate adjective is in a subject control infinitival clause lacking a complementizer, but not when it is in any other type of infinitival phrase. Compare the agreeing examples in (20) with the nonagreeing ones in (21); here I focus on the semipredicatives for ease of exposition: (20)

a. Ivan xocet [PRO pojti na vecerinku sam] (m nom) want to-go to party alone(m nom) 'Ivan wants to go to the party alone' b. Nadja ljubit [PRO gotovit' sama]27 (f nom) likes to-cook alone(f nom) 'Nadya likes to cook by herself c. Ljuba priexala [PRO pokupat' maslo sama] (f nom) came to-buy butter alone(f nom) 'Lyuba came to buy the butter herself (21) a. Masa poprosila Vanju [PRO prijti odnomu] (fnom) asked (mace) to-come alone(m dat) 'Masha asked Vanya to come alone' b. Masa ugovorila Vanju [PRO prigotovit' obed odnomu] (f nom) persuaded (m ace) to-cook lunch alone(m dat) 'Masha persuaded Vanya to cook lunch by himself c. Ljuba priexala [ctoby [PRO pokupat' maslo samoj]] (f nom) came in-order to-buy butter alone(f dat) 'Lyuba came in order to buy the butter herself

Obligatory agreement in (20) contrasts to the obligatory "default" second dative in (21). It thus appears that agreement is viable only with a proper subset of subject control constructions. Much recent research, stemming from the "successive cyclic" movement theory of Chomsky (1977), has demonstrated that apparent long-distance phenomena are generally best conceived of as the concatenation of multiple local processes. Even so-called long-distance binding has been profitably analyzed in terms of successive local movements at LF by, for example, Pica (1987) and Cole, Hermon, and Sung (1990). I believe that the kind of long-distance case agreement displayed by secondary predicates can also be understood as the result of multiple mutually dependent local agreement processes. Whether this "chain" of local agreements ultimately succeeds or fails depends on each of its links. These links are essentially created by the PRO subject of which the AP is predicated and which, in turn, is itself anteceded by an overt NP or another PRO. The control status of each PRO will thus be crucial in allowing or blocking agreement. Let us therefore employ the descriptive term "case transmission" to refer to the general process according to which one or more

236

PARAMETERS OF SLAVIC MORPHOSYNTAX

PRO elements intervenes between a predicate adjective and the NP with which it agrees in case. PRO is able, in a way to be made explicit in section 6.2.3, to mediate between its controller and the predicate adjective so as to cause the AP to agree with this controller. 6.2.2. Factors Blocking Case Transmission There are two primary factors that block case transmission. One concerns the antecedent of PRO: if PRO is controlled by a non-subject, has more than one acceptable antecedent, or is arbitrary in reference, then case transmission is blocked. The other concerns the relative position of PRO with respect to its antecedent: if an overt complementizer intervenes, then case transmission is also blocked. As a result, not even all PROs anteceded by subjects are able to participate in case transmission. It will eventually be desirable to try to reduce these two factors to a single explanation and to couch this explanation in terms of control theory. The following are further representative examples of constructions in which case transmission is blocked; indicated bracketing is intended to be expository rather than exhaustive: (22) a. Ivan pose'l domoj [ctoby [PRO ne obedat' odnomu]] (m nom) went home in-order NEG to-eat alone(m dat) 'Ivan went home in order not to eat dinner alone' b. Ivan ne imeetpredstavlenija [kak [PROzif samomu]] (m nom) NEG has idea how to-live alone(m dat) 'Ivan doesn't have any idea how to live alone' c. Ivan dumaet [cto [PRO pojti domoj odnomu]] vazno (m nom) thinks that to-go home alone(m dat) important 'Ivan thinks that it is important to go home alone' d. Ivan ne znaet [kak [PRO tuda dobrat'sja odnomu]] (m nom) NEG knows how there to-reach alone(m dat) 'Ivan doesn't know how to get there by himself (23) a. [PRO prijti odnomu] ocen' trudno to-come alone(dat) very hard 'to arrive alone is very difficult b. [PRO pojti tuda odnomu] rasstroilo by menja28 to-go there alone(m dat) upset MOD me(acc) 'to go there alone would upset me' c. [zelanie Igorja [PRO pojti odnomu]] nas ocen' rasstroilo desire(nom) (mgen) to-go alone(dat) us(acc) very upset 'Igor's desire to go alone upset us very much' d. dlja nas utomitel'no [PRO delat' eto samim] for us exhausting to-do this alone(pl dat) 'it's exhausting for us to this on our own' e. nevozmozno [PRO perejti etot most samomu] impossible to-cross this bridge alone(dat) 'it is impossible to cross this bridge by oneself

SECONDARY PREDICATION

237

In the examples in (22), PRO is protected by an overt complementizer, while in those in (23) the controller is either absent or a non-subject. It is important to observe that there is considerable overlap between these two conditions on case transmission. Thus, at least in examples (22b-d), it is relatively easy to understand PRO as being arbitrary in reference instead of referring to the overt subject antecedent. Similarly in (23b-d), although the available non-subject antecedent is obligatorily understood as the controller of PRO, this property is not an inherent aspect of these constructions. That is, the antecedent could mutatis mutandis be expressed in some other terms, absent, arbitrary, or recoverable from the discourse without inducing ungrammaticality. Following Williams (1980), I take such potential variation in control properties to indicate that these are in fact not structures of obligatory control. This suggests that a proper characterization of secondary predicate agreement should follow from an appropriate theory of control. Before turning to such a theory, it is worth observing that comparable nonagreement facts can be adduced also on the basis of ordinary predicate adjectives, as in (24). (24) a. [PRO spat' golym/*golomu/*golyj] nine nravitsja to-sleep naked(m inst/dat/nom) me(dat) pleases 'I like to sleep naked' b. vazno dlja menja [PRO spat' golym/*gologo/*golyj] important for me(gen) to-sleep naked(m inst/gen/nom) 'it is important for me to sleep naked' c. Lena ne znaet [pocemu [nel'zja [PRO spat' goloj/*golaja]]] (fnom) NEG knows why must-not to-sleep naked(f inst/nom) 'Lena doesn't know why she shouldn't sleep naked' Once again, agreement is impossible whenever PRO's antecedent is a non-subject or is not obligatorily present. Note that while the controller in (24c) is most likely Lena, this is induced by the choice of a feminine singular adjective; the arbitrary reading would be forced by the masculine form golym, however. Of course, ordinary predicate adjectives always allow the default instrumental, differing in this respect from the semipredicatives sam and odin, which never exhibit any alternative form. I will for this reason continue in this chapter to concentrate on the more arcane semipredicatives. These examples confirm my claim in chapter 2 that case must be treated separately from the pronominal features. Formally, this was effected by locating the case features in their own submatrix. The agreement pattern of the predicate adjectives demonstrates that the pronominal features are accessible even when case is not. The reason for this is that PRO contains pronominal features but not case features. Thus, the AP predicated of a PRO must agree in pronominal features with that PRO. PRO can conceivably be generated with any set of pronominal features, depending on its understood referent; in true arbitrary contexts the masculine singular is used in Russian.29 The AP can only agree in case, on the other hand, when PRO is in a configuration of obligatory control.30 This system conforms to my general model, in which case is syntagmatically determined and percolates down from DP, whereas

238

PARAMETERS OF SLAVIC MORPHOSYNTAX

the pronominal features are paradigmatically determined and intrinsic to N. Thus, it is no accident that even though PRO lacks case features in Russian, it still bears pronominal features, as shown by predicate adjective agreement in examples such as (23d) and (24c).

6.2.3. Control Theory and Agreement Control theory is that module of grammar that deals with how the distribution and reference of PRO is derived. It addresses the questions of where PRO can appear, how it is interpreted, and how it is distinguished from other empty categories. In this section I first discuss two general approaches to these problems—one that regards PRO as an element that is simultaneously anaphoric and pronominal, and another that regards it as ambiguously anaphoric or pronominal—and then argue that a version of the latter approach allows for important insights into case transmission phenomena in Russian. The analysis is then extended to account for a greater range of secondary predicate agreement facts, and several potentially problematic constructions are considered. 6.2.3.1. Two Theories The "classical" theory of PRO accounts for its distribution by treating it as a pronoun and an anaphor. Since it bears both [ + anaphoric, + pronominal] features, PRO is argued to be simultaneously subject to both principles A and B of the binding theory (cf. e.g. Chomsky, 1981a, 1982, 1986b). These principles require that an anaphor be bound and a pronoun free, respectively, within a specific domain. This domain, the element's "governing category," is the minimal maximal projection (i.e. phrase) that crucially contains that element together with its governor. It has additionally been proposed that the binding domain contain a SUBJECT, where this concept includes not only clausal subjects but also "subjects" of NPs, as well as AGR (or the AGR element of INFL) (cf. Chomsky, 198la).31 More recently, as in Chomsky (1986b), this notion has been replaced by the alternative formulation that the binding domain be a "Complete Functional Complex," where this represents the set of arguments of the predicate containing the element, although not necessarily a SUBJECT. These details and innovations aside, the standard argument about the distribution of PRO runs roughly as follows: if PRO is [ +anaphoric] and [ +pronominal], it will have to be both bound and free within its governing category—a logical impossibility unless it has no governing category, which follows only if PRO is necessarily ungoverned. This line of reasoning, due to Chomsky (1981a), has come to be known as the PRO theorem. The essential claim is that the binding features of PRO interact with the binding theory to limit its distribution to the subject position of infinitivals, gerunds, and possibly COMP. I see at least four potentially serious problems with this account of the distribution of PRO. First of all, it is not clear what it should mean for an element to be simultaneously an anaphor and a pronoun. However, even if we grant this possibility, there are several other factors that might disrupt such a derivation of the PRO theorem. For one thing, the binding domain for anaphors and pronouns might not be

SECONDARY PREDICATION

239

the same. Indeed, it has been claimed that, whereas anaphors need accessible SUBJECTs, pronouns do not. One might argue, for example, either that SUBJECTS for pronouns do not need to be accessible or that the domain need not contain them at all. It may be possible, nonetheless, to derive the PRO theorem even if the domains are not identical, as in Huang (1983). Even so, it may be that something other than absence of a governor is what prevents PRO from having a binding domain, although not having a governor seems to be the most workable approach. Finally, it is worth noting the PRO theorem depends on principles A and B referring specifically to [ +anaphoric] and [ +pronominal] elements respectively, and applying at the same level. If problems are shown to exist for either of these assumptions, proving the PRO theorem in terms of the binding theory becomes unlikely. Consider, for example, the contrast in (25). (25) a. Sue expected Bill to kiss Mary before he did b. Sue expected that Bill would kiss Mary before he did Sentence (25b) is ambiguous in a way that (25a) is not, in that only it can mean 'Sue expected that Bill would kiss Mary before Bill himself expected that he would kiss her'. How can this result be derived? Consider the following rough reconstructions of the matrix VP at LF: (26) a. Sue expected Bill to kiss Mary before he expected Bill to kiss Mary b. Sue expected that Bill would kiss Mary before he expected that Bill would kiss Mary Replacing the adjunct pro-VP did with the matrix VP apparently creates a binding problem in (26a) but not in (26b). This fact implies that principle C, which states that R-expressions such as the name Bill must be free everywhere, is not implicated, since in both sentences Bill is c-commanded by a coindexed NP. Principle B, on the other hand, might be able to distinguish the two and thereby account for the contrast, since only in LF representation (26a) is Bill coindexed with the c-commanding NP he in its governing category.32 This analysis is only possible, however, if principle B is understood as applying to [ - anaphoric] elements instead of [ + pronominal] ones. This will allow it to apply to both pronouns and R-expressions. Such a revision does not alter its empirical coverage, of course, except in this kind of example. However, it does have serious ramifications for the PRO theorem, which crucially relies on the relevance of the feature [ ± pronominal] for the binding theory. The second approach to PRO essentially treats it as an ambiguous element, either anaphoric or pronominal, but not both at once. This kind of approach was first suggested by Bouchard (1982/1984) and has subsequently been developed in various ways by Sportiche (1983), Koster (1984), Hornstein and Lightfoot (1987), and Franks and Hornstein (1992). It is designed to account straightforwardly for the observation that in some contexts PRO receives an anaphoric interpretation, while in others it receives a pronominal interpretation. That is, it assimilates the interpretive properties of PROs in various positions to those of comparable overt anaphors and pronouns. In such a "split" PRO system, of course, its distribution cannot be derived as it is done in the classical theory. Instead, the usual solution is to invoke

240

PARAMETERS OF SLAVIC MORPHOSYNTAX

case theory: if PRO cannot be case-marked, then it cannot appear in the subject position of a finite clause or the object position of a case assignor.33 Although the various approaches differ in how they execute this idea, they share the insight that the distribution of PRO should be dealt with in the context of case theory rather than binding theory. The issue of how PRO derives its reference is to some extent dependent upon whether one adopts the classical or split analysis. Under the former point of view, two possibilities have been entertained: either the theory requires a special control module, as in Chomsky (1981a) or Huang's (1984) Generalized Control Rule, which contains specific principles for determining the controller of PRO, or else control can be assimilated to some existing module, typically binding theory, as in Manzini (1983). This kind of approach requires, of course, that PRO be endowed with some kind of binding domain larger than the one appropriate to ordinary anaphors and pronouns, without PRO itself actually being governed. In Manzini (1983) this is PRO's "domain-governing category," defined as the minimal domain containing the governor of the domain of PRO (rather than PRO itself) and an accessible subject. Only for obligatorily controlled PROs will there be such a domain, if the definitions are properly constructed; see Manzini (1983) for details. Although the insight that control should somehow involve the anaphoric status of PRO is correct, I believe, in order for PRO to avail itself of the binding theory and still respect the PRO theorem, far-reaching and potentially undesirable revisions of the standard binding theory are required. It seems to me a much simpler approach to reject the PRO theorem instead and then to subject anaphoric PRO to some version of the standard binding principle A, as discussed in the next section. 6.2.3.2. Governed PRO PRO is a base-generated empty category. It may be either [ +anaphoric] or [ +pronominal], but not both, assuming the two to be conceptually incompatible. It differs from the other [ +anaphoric] empty category, NP-trace,34 however, and the other [ +pronominal] empty category, "small" pro,35 in that it necessarily lacks case. Aoun, Hornstein, Lightfoot, and Weinberg (1987) argue for a correlation between anaphoric PRO and proper government that we shall use in analyzing case transmission in Russian. The basic claim is that only those PROs that are governed can be anaphoric, hence enter into agreement processes such as case transmission, and have PF ramifications. The theoretical underpinnings of this approach are presented in detail in Franks and Hornstein (1992); here I simply sketch the basic idea. Aoun, Hornstein, Lightfoot, and Weinberg (1987) split the ECP into a PF half that requires empty indexed categories to be lexically properly governed at PF and an LF half that requires them to be antecedent-governed at LF. They further propose that a necessary condition for visibility is bearing an index. This means that only those PROs that at S-structure are subject to binding condition A (as anaphors) will induce case transmission, under the assumption that binding endows PRO with an index and that these PROs will be all and only those that occur in properly governed positions. Obligatory control, in the very restrictive sense of Williams (1980) adopted above, is thus only a property of governed PROs, which are anaphoric.

SECONDARY PREDICATION

241

On the other hand, pronominal PROs, which are not obligatorily controlled, are precisely those which occur in positions that are not lexically governed, such as in (27). (27) a. b. c. d.

[PRO to leave now] would bother Fred John said that [[PRO leaving now] would be silly] [any decision [PRO to leave now]] could be unfortunate Bill doesn't know [how [PRO to make lasagna]]

None of these PROs is lexically governed, hence they are not obligatorily controlled. In general, assuming PRO to originate as the specifier of VP just like any other subject, it is protected from proper government whenever this VP is further embedded in a CP/IP complex. PRO clearly does not need to raise to IP-specifier position in search of case. Assuming it remains in situ, PRO can be governed from outside the VP only if there are no higher functional projections intervening between the VP and the potential lexical governor.36 Similarly, even if it raises to SPEC-IP, PRO will still be protected from outside government if the IP is contained in a CP. In example (27d) how provides direct evidence of CP, hence PRO cannot be governed by the matrix verb. In (27a), with PRO in a subject sentence, there is no lexical head available that could govern PRO; (27b) is similar, with PRO both in a subject sentence and protected by the CP from external government. Finally, (27c) also contains an ungoverned and non-anaphoric PRO. Its pronominal status is clear from both its interpretation and the fact that it does not have a syntactic binder (but may have; replace any with his). The PRO is thus not in a position of obligatory control; furthermore, following Grimshaw (1990), sentential "complements" within NP are technically adjuncts rather than arguments, hence PRO cannot be properly governed by the N° head.37 The governed PRO approach predicts that only properly governed PROs are truly anaphoric and syntactically bound by their controllers. Thus, only these PROs are coindexed with their antecedents at S-structure and can be PF active. In particular, by virtue of coindexation, governed/anaphoric PROs should be able to transmit the case of their controller to a predicate adjective, whereas ungoverned/pronominal PROs should not. In this light, let us return to the Russian examples discussed above. In the examples of subject control, as in (20), PRO is governed by the matrix V and agreement takes place as expected. On the other hand, in (22), where PRO is protected from government by CP, case transmission is blocked. Similarly, the PROs in (23) are not properly governed, so that they are pronominal and fail to induce case transmission. The analysis of all these examples follows straightforwardly. In short, predicate adjective agreement is always local in the sense of being clause-bounded. This means that, in cases of apparent long-distance agreement, the agreement is in fact mediated by the intervening embedded PRO, which locally provides the necessary information. We can further assume, following Stowell (1981), that case features are checked on the PF side of the grammar.38 Given this, PRO can only participate in this PF process if it is visible at PF, where, in order to be visible, PRO must bear an index. However, according to Aoun, Hornstein, Light-

242

PARAMETERS OF SLAVIC MORPHOSYNTAX

foot, and Weinberg (1987) and Franks and Hornstein (1992), if an empty category bears an index at PF, it is subject to the ECP and so must be properly governed by a lexical head. This is impossible for the sentences in (22) and (23), so we predict that in these examples case agreement will not take place because this would lead to an indexed PRO at PF that is not lexically properly governed, in violation of the ECP. The examples in (20) do not encounter this problem, since the mediating PRO is in a position in which it can be properly governed. To illustrate, consider (20c), which has the structure in (28). (28)

Ljuba [v priexala [IP/Vp PRO pokupat' maslo sama]]

In contrast, when a COMP appears there must be a CP present, which blocks proper government by the verb, as in (21c), which has the structure in (29). (29)

Ljuba [v priexala f cp ctoby [IP/Vp PRO pokupat' maslo samoj]]]

Corroboration for the claim that only properly governed PROs can mediate this agreement process is found in some more complex instances of the same phenomenon. Comrie (1974) observed that after right dislocation the default case is required: (30)

a. ja bojalsja odnogo: byt' ostavlenu na doroge I feared one(gen) to-be left(dat) on road 'I feared one thing: to be left on the road' b. ja bojalsja odnogo: ostavat'sjav boju odnomu I feared one(gen) to-remain in battle alone(dat) 'I feared one thing: to remain in battle alone'

Right dislocation of the embedded clause results in a non-properly governed PRO at PF, so that the default dative case is required.39 Note that example (30a), taken from Pushkin, employs the archaic "short form" dative, which was still attested in earlynineteenth-century literature. In modern Russian, of course, the short form is caseless and purely predicational, with the instrumental ostavlennym required here instead. The assignment of the dative rather than instrumental to adjectives other than the semipredicatives when they could still appear with short form endings will follow from the morphologically motivated account of the special behavior of the semipredicatives to be developed in section 6.3.2. In contrast, the examples in (21)—which seem to involve obligatory control but nonetheless cannot transmit the case of their controller to the semipredicative—are at first glance problematic for the analysis. The solution to this dilemma lies in distinguishing obligatory controllees (anaphoric PROs) from obligatory controllers (unambiguous antecedents of pronominal PROs). That is, in a given configuration it may well be that PRO can only be understood as controlled by a particular NP, even if that PRO is technically not anaphoric. A case in point is (21c), in which the PRO of the purpose clause is protected from government by the complementizer ctoby, as was seen in (29), despite the fact that it can only reasonably be understood as referring to Ljuba. Thus, in contrast to the PRO in (20c) with the structure (28), this PRO is actually neither anaphoric nor bound. Instances of object control, as in (21a) and (21b), are open to the same interpretation. In both examples, proper government

SECONDARY PREDICATION

243

of PRO by V is blocked by the presence of the direct object Vanju, despite the fact that the controllers of these PROs are obligatorily determined.40 Assimilating object control to non-obligatory control raises the questions of how the antecedent is established and why, if PRO is not anaphoric, the antecedent must be the object NP. Franks and Hornstein (1992) propose that constructions in which pronominal PRO has an obligatory antecedent can be treated as instances of predication, on a par with relative clauses and perhaps topic-comment (left-dislocation) structures. (31)

a. [DP devu§ka [CP kotorujUj [Ivan ljubit e;]]] 'the girl who Ivan loves' b. [CP Masa, [[P Ivan ee4 ljubit]] Masha, Ivan loves her'

Predication establishes an obligatory coreference relationship between kotoruju 'which' and devuska 'girl' in (31a) and between Masa and ee 'her' in (31b), even though kotoruju and ee are technically not anaphors. Of course, object control structures differ slightly from those in (31) in that under the traditional analysis they involve two distinct sisters of the V, the object NP and the clause, related by predication. However, it may be possible to render the parallelism between (31) and object control constructions more complete by analyzing object control as involving a small clause complement, roughly as in (32). (32)

Masa poprosila [sc Vanju [PRO prijti odnomu]] 'Masha asked Vanya to come alone'

An analysis along these lines is developed in Franks and Homstein (1992). There we cite several types of independent evidence to suggest that this structure is in fact correct for 'persuade'-type verbs with infinitival complements. One piece of support comes from Sportiche's (1988) observation that quantifiers can be floated from subject positions but not from object positions:41 (33)

a.

all the men have eaten a piece of cake

b. the men all have eaten a piece of cake c. John has seen all the men of late d. *John has seen the men all of late This observation leads to the conclusion that in (34) the logical "object" is really a syntactic "subject" of some kind: (34)

John persuaded [sc the men all [PRO to leave]]

This structure is appropriate for the analysis of object control in Russian and conforms to the widely adopted idea that only binary branching is permitted. Although standard cases of object control display non-agreement, this analysis really divorces the phenomenon from control, except tangentially. Agreement hinges instead on whether the embedded PRO can be properly governed at PF, not on whether it is linked to the matrix subject or object. A potential problem for this approach comes from a consideration of the Russian verb obescat' 'to promise',

244

PARAMETERS OF SLAVIC MORPHOSYNTAX

which, like its English counterpart, is a subject control verb and which can optionally realize an object NP in the dative. On the account outlined above, constructions without an object should pattern with standard cases of object control because, regardless of how the PRO is interpreted, it should not be properly governed if a direct object intervenes. This is indeed what the data in Comrie (1974) suggest: (35)

a. Kostja obesc"al prijti odin/*odnomu (nom) promised to-come (nom/dat) 'Kostya promised to come alone' b. Volodja obescal materi vernut'sja ??odin/odnomu (nom) promised mother(dat) to-return (nom/dat) 'Voldya promised his mother to return alone'

Comrie notes, however, that the nominative form odin in (35b) "was felt to be more acceptable than a nominative would be" in the other examples discussed. What is special about the obescat' construction? For one thing, the infinitival itself is awkward, speakers uniformly preferring to replace it with a finite clause. More seriously, subsequent investigation reveals that for most (if not all) speakers the dative is unacceptable in (35b), just as in (35a), with the nominative being the only viable (if for some speakers rather infelicitous) alternative. Greenberg (1983), for example, claims that his informants only accept the nominative, the second dative being impossible, and the speakers I have consulted confirm this.42 Why might this be so, contrary to the prima facie predictions made by both Comrie's approach and ours? For some reason, dative materi 'mother' in (35b) does not interfere with government relations between the verb and its clausal complement. A reasonable explanation lies in the fact that materi is actually not a direct object but an indirect one. This is why it appears in the dative case and is optional. Furthermore, applying the Q-float test to English promise demonstrates that its object lacks the subject properties exhibited by object control verbs. Compare the following ungrammatical sentence with (34). (36)

*John promised the men all [PRO to leave]

Therefore, promise cannot be taking a small clause complement. Moreover, the putative "subject" of such a SC, in addition to getting somehow assigned dative in Russian, would have to have the unusual option of being able to be PRO. Such a possibility is ruled out under the traditional approach since PRO would be governed, and under the governed PRO approach such a PRO would be anaphoric, so that (35a) would necessarily mean "Kostya promised himself that he would come alone." Since in Franks and Hornstein (1992) we analyze object control in terms of predication (the relevant relation in the examples in [31]), the small clause structure literally forces object control. For these reasons, I suggest that the complement structure of obescat' may be ambiguous (if there indeed are speakers who display the pattern in [35]), depending on whether a speaker analyzes the postverbal indirect object NP as able to block government by the verb or not; see Franks and Hornstein (1992), as well as Greenberg (1983), for details.

SECONDARY PREDICATION

245

6.2.3.3. A Subject-Orientation Approach The account of case transmission just put forward relies on structural properties of PRO. For the most part, it correlates very nicely with intuitive notions of the difference between obligatory and non-obligatory control. Extra assumptions are required, however, to handle the curious asymmetry betweeen subject and object control. Let us consider, therefore, a conceptually very different kind of alternative to this asymmetry. I have maintained that PRO may be either anaphoric or pronominal; contrary to the theory elaborated in Chomsky (1981a), it is never simultaneously both. Control can therefore be regarded as a generalization of binding, in that anaphoric PRO is subject to Principle A of the binding theory and must be bound within its governing category. It thus becomes possible to assimilate the subject orientation of PRO to the properties of other anaphors. In particular, with a few controversial potential exceptions (cf. Greenberg and Franks, 1991), reflexives in Russian can only be bound by subjects.43 This explains the lack of ambiguity in (37). (37)

Masaj rasskazala Tanej o sebet *j (nom) told (dat) about self 'Masha told Tanya about herself

It may be possible, therefore, to derive the lack of case transmission in object control structures from the subject orientation of PRO. That is, if anaphoric PRO resembles reflexives in Russian in being subject-oriented, then its control and agreement properties follow automatically. Notice, however, that this implies that so-called object control PRO is not anaphoric and that the relationship between it and its antecedent cannot be one of binding. Instead, even though the antecedent is obligatorily determined, we have seen that with respect to predicate adjective agreement it patterns with other clearer instances of pronominal PRO. Although taking advantage in this way of PRO as an anaphoric element presents an interesting possibility, there are several difficulties with the execution of this idea. As discussed in Franks (1983,1985), the correlation between whether a reflexive in a given language is subject-oriented and whether PRO can transmit case only from subjects is far from perfect. Both classical Greek, according to Andrews (1971), and Icelandic—at least for those dialects in which sig 'self is always subjectoriented—are problematic. Even within Slavic, one can cite Czech and Slovak as languages that allow case transmission from object controllers even though reflexives in these languages cannot be bound by non-subjects (cf. Toman, 1991, for binding in Czech). The following Czech example is from Comrie (1974): (38)

donutil jsem ho pfijit samotneho/*samotnemu forced AUX(lsg) him(m ace) to come alone(m acc)/(m dat) 'I forced him to come alone'

This shows that a blind correlation, with simple stipulation that PRO be subjectoriented, is inadequate. Although I do not pursue the matter here, there are a variety of ways out of this dilemma. For one thing, there is no a priori reason why all

246

PARAMETERS OF SLAVIC MORPHOSYNTAX

anaphors in general, and PRO and the morphologically simplex reflexive in any given language in particular, should behave alike. It is indeed well known that anaphors need not be consistent. As observed by Yang (1983), morphologically complex reflexives such as English himself are generally not subject-oriented and must be bound locally, whereas morphologically simplex reflexives such as Russian sebja are generally subject-oriented and can be bound in certain long-distance contexts. However, most analyses of diversity in anaphoric systems propose principled reasons why each anaphor should behave the way it does (cf. e.g. Pica, 1987; Cole, Hermon, and Sung, 1990; Progovac, 1992; Progovac and Franks, 1992). On the other hand, nothing about PRO in Russian versus Czech, for example, would lead us to expect the observed differences in behavior. The fact that predicate adjective agreement takes place in Czech (38) may be related to the regular adjectival morphology of samotneho, as opposed to Russian sam. Although samotneho in Comrie's example actually means 'lonely' more than it does 'alone', (38) is equally possible with sameho 'alone(acc)'. Note also that the non-agreeing form in Czech is sdm, not samemu, as in (39). (39)

je dulezite prijit sam is important to-come alone 'it is important to come alone'

Although this looks like the nominative, I will argue later in this chapter that such elements are caseless and have an adverbialized function. As such, Czech sdm can appear in any construction that also admits agreement, such as (38). The existence of (38) also raises the issue of how local agreement with nonsubject antecedents takes place, as in examples (2b, c), which I had ascribed a small clause structure as in (lib). The presence of PRO, however, leads to a problem under both the subject orientation and governed PRO accounts. If PRO can only be bound by a subject, then why does it discriminate between a local antecedent and one in the matrix clause? If PRO is anaphoric only when governed, and presence of an object blocks government, then why can PRO be PF active when it is the subject of a small clause after an object NP? Note that a similar problem presents itself for the semipredicatives, except that there agreement is obligatory; compare (6). In previous work, including Franks (1990a) and Greenberg and Franks (1991), I argued that local agreement with non-subjects was indicative of a simple adjoined AP analysis rather than a SC with PRO subject. Since I claimed above that the default instrumental was precisely this, the issue is reduced to restricting the distribution of predicate APs; they can generally agree in case with their c-commanding local antecedents, and, when adjoined to VP, they can also appear in the default instrumental. However, if both ordinary adjectives and semipredicatives can be simple APs (as in fact has been assumed all along), I see no way to prevent the latter from appearing in the instrumental other than stipulating that they cannot be VPadjuncts. It seems to me that such an approach could be made to work under either the governed PRO or subject orientation analysis. The question then is how an account of local agreement with non-subjects might extend to examples like Czech (38). Unlike with APs, however, there is no alternative but to assume a PRO subject in front of the infinitive prijit. Why then should this

SECONDARY PREDICATION

247

PRO be bindable by an apparent non-subject? Under the binding approach, one would have to argue that ho 'him' was actually a subject (in Czech but not Russian), or, as suggested above, PRO in Czech happens not to be subject-oriented. Under the government approach one would have to make the unlikely claim that object control in Czech involves a different structure than Russian (32), presumably the traditional one with the NP and clause forming separate constituents. Then string adjacency would not be necessary for lexical government. This in fact seems to be true for Slavic, although not English, as the obescat' facts discussed above suggest, since the optional dative argument does not appear to block government of PRO by V. Furthermore, adjuncts of various sorts clearly can intervene without affecting government relations. In (20c), for example, interpolation of an adjunct does not affect case transmission: (40)

Ljuba priexala vcera vecerom/na masine pokupat' maslo sama 'Lyuba came yesterday evening/by car to buy the butter herself

These elements clearly do not impede lexical government of PRO. The claim that adjacency is not required for government in languages with relatively rich nominal morphology, such as Russian, has recently been made by Chomsky (1986b, 88), who remarks that the invocation of Stowell's case adjacency principle in English is ' 'presumably a reflex of the poverty of the morphological Case system.'' In sum, I conclude that connecting the case transmission behavior of PRO to the binding orientation of reflexives is a promising possibility, but many details of the approach remain to be worked out. 6.2.3.4. Further Predictions Before turning to the second dative and the range of related phenomena across Slavic, let us briefly consider how the analysis extends to some further predicate adjective constructions in Russian. I have shown that, although it is necessary for PRO's controller to be a subject for case transmission to take place, this is by no means a sufficient condition. Instead, following arguments in Franks and Hornstein (1992), another way to consider the subject control/object control asymmetry is in terms of the government status of PRO. In subject control constructions PRO is properly governed by the verb, whereas in object control ones it is not. We thus proposed that anaphoric PROs are precisely those that are properly governed, which eliminates any need to refer explicitly to the case or grammatical function of the antecedent. Thus, agreement should be blocked whenever PRO is contained in a CP, since the verb cannot govern into the complement IP of C, or it is in a bare VP/IP that is not properly governed by the verb. Therefore, in addition to object control constructions, agreement is expected in general to fail when PRO is protected by CP, as in the following: (41)

Ivan ne ponimaet [DP moe zelanie [CP PRO idti na vecer (nom) NEG understand my desire to-go to party *odin/odnomu]] alone(nom)/(dat) ' Ivan doesn't understand my desire to go to the party alone'

248

PARAMETERS OF SLAVIC MORPHOSYNTAX

Case transmission is impossible in (41) since I have argued that N° can never govern into its complement clause and the DP blocks government of PRO by V. There is, however, a curious effect manifested by certain verb-noun collocations. As illustrated in (41), the PRO subject of an infinitival complement to N is ordinarily not properly governed and therefore pronominal. This appears to be contradicted, however, by the examples in (42). (42)

a. Ivan prinjal resenie [PRO prijti na vecer odin] (nom) took decision to-come to party (nom) 'Ivan made a decision to come to the party alone' b. Ivan vyrazil zelanie [PRO prijti na vecer odin] (nom) expressed desire to-come to party (nom) 'Ivan expressed a desire to come to the party alone' c. Ivan dal obeSdanie [PRO prijti na veSer odin] (nom) gave promise to-come to party (nom) 'Ivan made a promise to come to the party alone' d. Ivan imel vozmoznost' [PRO prijti na vecer odin] (nom) had opportunity to-come to party (nom) 'Ivan had a chance to come to the party alone'

I suggest that what is going on here is that the verb-noun sequence is being restructured into a complex verb.44 The result of this process of complex verb formation is that PRO becomes properly governed by the complex verbs [v prinjal resenie] 'made a decision', [v vyrazil zelanie] 'expressed a desire', and so on, with the familiar result that agreement occurs. Although the precise factors allowing complex verb formation are poorly understood, it is clear that the verb—noun combination must be highly idiomatic, perhaps restricted to verbs that are minimal lexical functions of the nouns in question. Notice also that if the noun is possessed, restructuring is blocked. Hence, agreement fails in (43). (43) a. Ivan poprosil razresenija [PRO prijti na vecer *odin/odnomu] (nom) asked permission to-come to party alone(nom/dat) 'Ivan asked permission to come to the party alone' b. soldat polucil prikaz [PRO poexat' v gorod *odin/odnomu] soldier(nom) received order to-go to city alone(nom/dat) 'the soldier received an order to go to the city alone' c. Ivan vyrazil svoe zelanie [PRO prijti na veder *odin/odnomu] (nom) expressed self's desire to-come to party alone(nom/dat) 'Ivan expressed his desire to come to the party alone' In (43a, b) the noun-verb collocation is not sufficiently idiomatic to permit restructuring, whereas in (43c) the possessive pronoun svoe 'self's' intervenes. Corroboration for this analysis can be found in related extraction phenomena. Extraction from within a clause embedded in a DP is ordinarily blocked by Subjacency, since two barriers must be crossed, following Chomsky (1986a), for example. Restructuring eliminates the DP barrier, however, concomitantly facilitating extraction. Compare (44a) with (44b).

SECONDARY PREDICATION

(44)

249

a. *ctoj Ivan vyrazil [DP svoe zelanie [PRO kupit' ej] ' what did Ivan express his desire to buy' b. ctoj Ivan [v vyrazil zelanie] [PRO kupit' ej ' what did Ivan express a desire to buy'

Note that the English gloss of (44a) is ungrammatical for similar reasons. The acceptability of case transmission in (42), assuming this analysis, supports the idea that infinitival phrases are not necessarily CPs. Since government of the embedded PRO by V (but crucially not by N) is possible, I conclude that (under restructuring at least) these infinitival complements are not CPs, but rather bare VPs (or IPs or TPs, depending on one's analysis of clauses). In sum, then, PRO may be either pronominal or anaphoric, depending upon whether or not it is properly governed. Only anaphoric PRO is grammatically controlled by its antecedent, hence only anaphoric PRO induces case transmission to a predicate adjective. This analysis renders the subject orientation stipulation superfluous and extends to numerous situations where PRO is coreferential with a matrix subject yet nonetheless fails to transmit nominative case (because it is ungoverned and thus pronominal rather than anaphoric).

6.3. The Second Dative In this section I develop an analysis of the second dative phenomenon, laying the necessary groundwork for the comparison of these and related constructions in section 6.4. Two conceivable approaches to the agreement pattern of the second dative immediately come to mind: the second dative could either (1) result from some kind of case agreement rule, or (2) arise through direct case assignment. The former ' 'agreement'' sort of analysis is widespread in the literature; variants of it are espoused in Bouchard (1982/1984), Comrie (1974), Greenberg (1983), Neidle (1982, 1988), and Schein (1982). In these analyses, the second dative is related to the "subjective" dative found elsewhere in Russian. If, the reasoning goes, overt dative subjects can occur with infinitives (as discussed shortly), then why not let the second dative result from agreement with an unrealized dative subject? In GB terms, the default dative semipredicatives are actually somehow agreeing with a dative PRO. The latter "direct assignment'' sort of approach was originally proposed in Franks (1985). I have subsequently developed it in various forms in Franks and Greenberg (1988), Franks (1991), and Franks and Hornstein (1992). Here, the onus of the analysis is to uncover a mechanism by which the semipredicative might be the direct target of an otherwise motivated dative assignment rule, such as that assigning dative to indirect objects of verbs or subjects of infinitives. In what follows, I elaborate such an account, discussing certain technical and conceptual pitfalls of the agreement approach as well as the advantages to the direct assignment model.

6.3.1. The Agreement Analysis What is the origin of this curious default second dative on the semipredicate adjectives? As mentioned above, two reasonable structurally based solutions are that

250

PARAMETERS OF SLAVIC MORPHOSYNTAX

either (1) dative arises through agreement with an implicitly dative PRO subject, or (2) dative is assigned directly to the semipredicatives odin and sam. Either way, it seems to me that the second dative phenomenon is viewed as a function—whether an indirect one via agreement or a direct one via government—of the structural ability of F to assign dative case to subject position in infinitival clauses. The agreement approach capitalizes on the fact that under certain circumstances Russian seems to exhibit dative subjects. Although the subject NP of a finite clause in Russian typically appears in the nominative case, there are sentences in Russian that do not contain a canonical NP subject in the nominative. Instead, they contain a dative NP that appears to bear the same kind of semantic role canonical subjects do, although it is not in an agreement relation with the predicate. In such sentences, the predicate may be expressed by a so-called impersonal or category-of-state predicate, as in (45), or by an infinitive, as in (46). (45)

a. emu zal' etu devusku him(dat) sorry that(acc) girl(acc) 'he feels sorry for that girl' b. mne legko govorit' po-russki me(dat) easy to-speak in Russian 'it is easy for me to speak Russian'

(46)

a. gde mne spat' where me(dat) to-sleep 'where am I to sleep' b. mne uxodit' me(dat) to-leave 'I have to leave'

Of course, it is clear that sentences like these actually contain a zero-form of the copula, since in other tenses the copula appears in a default neuter (singular) form. Compare the following past tense examples, which show the copula bylo 'was', with (45) and (46): (47)

a. emu bylo zal' etu devusku him(dat) was(n) sorry that girl(acc) 'he felt sorry for that girl' b. mne bylo legko govorit' po-russki me(dat) was(n) easy to-speak in Russian 'it was easy for me to speak Russian'

(48)

a. gde mne bylo spat' where me(dat) was(n sg) to-sleep 'where was there for me to sleep' b. mne bylo uxodit' me(dat) was(n) to-leave 'I had to leave'

SECONDARY PREDICATION

251

The status of these "dative subjects" has been widely debated in the Russian linguistic literature.45 This debate has continued into the generative sphere, where, despite the explicitness of the analyses, there has been much theoretical disagreement about these constructions. On the one hand, proponents of GB theory have relied on the principles of case and government theory to argue that these dative NPs do not receive case from within their own clause (cf. Schein, 1982). Proponents of other theories, on the other hand, have argued that the Russian data lend support to their particular theory and at the same time call into question basic principles of GB (cf. the LFG analysis in Neidle, 1982, 1988). In this chapter I argue not only that Russian does indeed have dative subjects that are assigned case from within their own clause, but that these data can be explained in a principled fashion within GB theory. The explanation involves the introduction of a peripheral parametric casemarking strategy based on properties of the I(nflection) and AGR(eement) system of Russian. This account will receive empirical support from two independent areas: 1. Within Russian grammar, the existence of the non-agreeing second dative construction is directly related to the possibility of dative subjects. 2. As demonstrated in section 6.4, this correlation also holds throughout the remaining Slavic languages, even though the productivity of these constructions varies from language to language. Let us concentrate first on developing a structural account of how dative subjects of infinitives might arise, since that account will provide valuable insight into the direct assignment model to be proposed in section 6.3.2.46 Recall that sentences are headed by an I(inflection) node consisting of features for tense and agreement. In finite clauses, which have the structure [CP [jP NP [r I VP ]]], the subject NP is assigned nominative under SPEC-head agreement with I. That is, only when a clause is finite and I contains both a TNS and AGR component does its subject NP get marked nominative. In the absence of tense and agreement, on the other hand, the subject receives no case and therefore appears as PRO. This is the typical scenario for infinitival clauses. How is it then that subjects of infinitives can sometimes be assigned dative case? Consider the representative examples in (49). (49) a. mne bol'se tarn ne rabotat' me(dat) more there NEG to-work 'I can't work there anymore' b. kuda nam postavit' etotjascik where us(dat) to-put this box 'where are we to put this box?' c. gruzovikune proexat' truck(dat) NEG to-drive-through 'the truck can't get through' Since these dative subjects of infinitives seem to bear the same theta-roles as would comparable subjects of conjugated verbs, I suggest that they are in the same struc-

252

PARAMETERS OF SLAVIC MORPHOSYNTAX

tural position. These examples can thus be assigned the standard clausal structure in (50), with the dative subjects occupying the specifier position of IP. (50)

a. [CP [IP mne [,, I [VP bol'se tarn ne rabotat']]]] b. [CP kuda [IP nam [r I [VP postavit' etot jasclk]]]] c. [CP [IP gruzoviku [r I [VP ne proexat']]]]

Two aspects of this proposed structure deserve special attention: 1. Dative subjects are sisters of I'. 2. Dative subjects occur embedded in complete CP clauses. The first point is true of all case-marked subjects, under the assumption that raising from SPEC-VP takes place. The second admits considerable variation, however, in that the X-bar theory of phrase structure allows for the possibility of clauses being less than maximal CPs. In Russian, however, the subject of a bare IP or VP has no source for case and is thus necessarily a phonologically null PRO. All case-marked subjects, whether dative or nominative, are on the other hand IP-specifiers in full CP clauses. This can be shown in several ways. The fact that dative subjects may be preceded by question words, as in (50b), indicates that they are part of CPs. Furthermore, any sentence containing a dative subject is inherently tensed, as demonstrated by the possibility of realizing tense in non-present forms, as was shown in (47) and (48). The examples in (49) and (50) are similarly tensed. On the other hand, in infinitival constructions where no finite copula is possible, neither is a dative subject. Subject sentences (5la), object control constructions (51b), and infinitival complements to nouns (51c), adjectives (51d), and impersonal predicates (51e) are all examples of this.47 (51)

a. [(*Ivanu) pljunut' v lico tovarisca] sostavjalo potrebnost' ego dusi (dat) to-spit in face comrade comprised necessity his soul 'for Ivan to spit in the face of his comrade was a necessity of his soul' b. ja poprosil Volodju [(*Mase) polucit' premiju] I(nom) asked (ace) (dat) to-receive prize 'I asked Volodya for Masha to get the prize' c. moe zelanie [(*MaSe) poludit' premiju] my desire (dat) to-receive prize 'my desire for Masha to get the prize' d. gotovyj/zelajuscij [(*Mase) polucit' premiju] ready/wishing (dat) to-receive prize 'ready/wishing for Masha to get the prize' e. bogatym vazno [(*bednym) mnogo rabotat'] rich(dat) important poor(dat) much to-work 'it is important for the rich for the poor to work hard'

SECONDARY PREDICATION

253

Clearly, all these embedded clauses are tenseless and thus differ from those that accept dative subjects in that they do not admit the copula. Hence, I conclude that dative subjects are licensed only in tensed clauses. Given that dative subjects in Russian may occur in tensed infinitival clauses, where these are presumably different from finite clauses in that they lack AGR, the question remains of why the dative is assigned. Notice that the configuration under which this marked instance of dative case assignment occurs is "sister to F." This particular configuration can be regarded as a generalization of the canonical "sister to V" indirect object structure. Recall that dative case assignment can be accounted for configurationally by assuming that the position of indirect objects is "sister to V" in a phrase structure tree. Dative subjects, I believe, can be regarded as a generalization of this schema to "sister to I'." 4 ® In Russian, then, there is a special process whereby subjects of tensed infinitival clauses are marked dative by virtue of appearing in the "sister of I'" position. This structural account will be important in the next section, in which a mechanism for directly assigning the second dative is presented. First, however, let us consider the evidence in favor of treating both infinitival and impersonal types of dative NPs as subjects.49 It turns out that these dative NPs exhibit certain subjectlike syntactic properties, the most salient being that they may antecede reflexives. As discussed in chapter 4 and section 6.2.3.3, this property is normally associated only with subjects in Russian. In (52), these dative NPs control both possessive adjective svoj 'self's' and pronominal sebja 'self reflexivization. (52) a. nine zal' sebja me(dat) sorry self(acc) 'I feel sorry for myself b. mne legko govorit' o sebe me(dat) easy to-speak about self(loc) 'it is easy for me to speak about myself c. mne svoix nagrad nedego stydit'sja me(dat) self's awards nothing to-be-ashamed-of 'there is no reason for me to be ashamed of my awards' A second subject-oriented diagnostic displayed by these dative NPs is that they may control subject reference into gerund and ctoby purpose clauses, as in (53) and (54), respectively. (53) a. sluSaja etot rasskaz, mne bylo strasno listening-to that story me(dat) was scary 'while listening to that story, I was scared' b. znakomja menja s kollekciej, ej acquainting me with collection her(dat) trudno bylo skryt' vosxiScenie difficult was to-hide delight 'while acquainting me with her collection, it was difficult for her to hide her delight'

254

PARAMETERS OF SLAVIC MORPHOSYNTAX

(54) a. dostatocno bylo emu tol'ko posevelit'sja enough was him(dat) only to-move-a-little ili kak-nibud' nenarokom mignut' brov'ju, or somehow unintentionally to-blink eyebrow ctoby podpast' vdrug pod gnev in-order to-fall-under suddenly under anger 'it was enough for him just to move a little, or somehow unintentionally blink, to attract his anger' b. . . . mne dovol'no bylo etix dnej,ctoby . . . me(dat) sufficient was these days in-order ponjat' i pocuvstvovat' materinskuju ljubov'. . . to-understand and to-feel maternal love '. . . these days were sufficient for me to understand and to feel maternal love . . .' While these facts might initially seem to provide compelling evidence that such dative NPs are indeed subjects, they are actually far from definitive. The problem is that although the properties of controlling reflexivization and the implicit subjects of gerund and purpose clauses are most frequently associated with canonical subject NPs, they also pertain to probable non-subjects. As indicated by the examples in (55) and (56), reflexivization can potentially be controlled by non-subject NPs. In (55) the reflexive pronoun's antecedent is an object of a preposition: (55)

a. u nego ne ostavilos' vremeni na sebja athim(gen) NEG remained time for self 'he didn't have any time left for himself

b. malo u menja svoix boljacek few at me(gen) self's sores 'I don't have enough of my own pains as it is' c. dlja nego vse okoncatel'no stalo jasno v sebe for him(gen) allfinally became clear in self 'for him everything finally became clear in himself While much rarer than the examples just cited, the sentences in (56) contain reflexive pronouns anteceded by direct objects in the accusative: (56)

a. i est' esceinoe tjagotenie, ono vozvrascaet and is still other force it(nom) return menja k samomu sebe me(acc) to EMPH self 'and there is still another force, it returns me to myself b. on zastal menja v svoej komnate he(nom) found me(acc) in self's room 'he found me in my room'

Note that these are technically ambiguous, the local nominative NP also being a conceivable—if in (56a) very unlikely—antecedent.

SECONDARY PREDICATION

255

The ability to control subject reference into gerund and purpose clauses is also exhibited by clear non-subject NPs, as shown in (57). (57) a. proxodja mimo krasivogo cvetnika, na ix crossing by beautiful garden on their ustalyx licax pojavljaetsja dobraja blagodarstvennaja ulybka tired faces appears kind thankful smile 'while crossing the beautiful garden, on their faces a kind, thankful smile appears' b. projdja pervyj obzig, izdelie pokryvajut glazur'ju passing-through first firing article(acc) they-cover glaze(inst) 'when going through the first firing, they cover the article with glaze' These examples show that even possessors inside PPs and direct objects can control the subject reference of gerunds, given an appropriate context. In (58a) an indirect object controls the implicit subject of a ctoby purpose clause, and in (58b) the implicit subject is left with arbitrary reference. (58) a. i prjamo k nam ide't djadja . . . on stal and straight to us comes uncle . . . he began babuske ruku podavat', ctoby vylezti grandmother hand to-give in-order to-climb-out 'and straight at us comes uncle . . . he began to offer his hand to grandmother, so that she could climb out' b. askameecka ne prosto skameec'ka, a eto but bench NEG simply bench but that kozly, ctoby sidet' i pravit' seat in-order to-sit and to-drive 'the bench is not simply a bench, but a coach-box, on which one sits and drives' I conclude that the evidence that datives with impersonal verbs and infinitives are indeed subjects is far from definitive. In section 6.4 I shall argue, on the basis of comparative data, that the two types of dative NPs must be structurally distinguished, so that only infinitives really have dative subjects. Assuming, then, that Russian does countenance overt dative subjects, one might contend that the second dative forms are actually agreeing with a dative PRO. There are serious conceptual problems with the direct assignment approach, however, and so in this chapter I shall opt for the latter alternative. In brief, the primary stumbling block this kind of account encounters is that it countenances case-marked PRO in Slavic. In particular, whenever the second dative appears one must assume that a dative PRO subject is also present. While this makes some sense in constructions where an overt dative subject is also possible, the range of distribution of the second dative semipredicative is clearly far greater than that of overt dative subjects. Thus, it is impossible to replace PRO with an overt dative in the object control and other

256

PARAMETERS OF SLAVIC MORPHOSYNTAX

constructions in (51), even though second datives are viable. Instead, the subject of all these infinitivals must be PRO. Consider also the following examples: (59)

a. nine vazno [(*vam) zit' odnomu] me(dat) important you(dat) to-live alone(dat) 'it is important for me (for you) to live alone' b. resenie [(*vam) zit' odnomu] . . . decision you(dat) to-live alone(dat) 'the decision (for you) to live alone . . . "

The reason these embedded infinitive phrases do not tolerate dative subjects is that such subjects would have no source for case. If all overt NPs require case, only PRO, which is caseless, can occupy the subject position of the embedded infinitivals in (59). The absence of a mechanism to assign case to this position, I argued in section 6.1.3, correlates with the absence of tense in these clauses. Note that this is in sharp contrast to the grammatical English glosses; the mechanism assigning dative to subjects in Russian is therefore not comparable to English for, since the latter option generally exists in embedded infinitival clauses. It is thus unlikely that the Russian second dative results from agreement with a dative PRO, since the potential within any given construction of having a dative subject is not coextensive with that of having a dative semipredicative. The conceptual problem with the agreement analysis is thus that the failure of these PROs to alternate with a dative NP implies that they are necessarily caseless and, if so, they cannot serve as the source of agreement for second datives. I thus conclude that, appealing as it may be to assume the existence of a dative PRO with which the semipredicatives agree, there are potentially insurmountable conceptual difficulties with this approach. Once the theory admits case-marked PRO, there is no obvious mechanism for preventing overt NPs from appearing in the same position.50 That is, we expect the possibility of assigning case to a position to mean that an overt NP can in principle occupy it. In many constructions that require default dative semipredicatives, thereby implying the presence of a dative PRO, no overt dative NP is however possible.

6.3.2. A Phrase Structural Analysis Although I have argued that the agreement approach to second datives is problematic, it would nonetheless seem to garner support from the existence of a special rale assigning dative to subjects of infinitives in Russian. Moreover, as I will show in section 6.4, the dative subject rule applies in precisely those Slavic languages that also exhibit the second dative phenomenon. This fact does not in and of itself argue for the agreement strategy over its alternative, however, since direct assignment might equally well be contingent on the same special rule. This is indeed what I shall argue. If my rejection of case-marked PRO is correct, then the absence of any potential antecedent for the second dative to agree with (which is, after all, what is meant by

SECONDARY PREDICATION

257

"default") implies that it is receiving case directly. We must therefore ask what configuration the semipredicatives might be in that allows them to be assigned dative. The crux of the proposal is to capitalize on the independently motivated configuration for dative subjects. It was hypothesized that "sister to I'" is the position that accounts for dative subjects of infinitives. Let us simply extend this approach to handle the semipredicatives, assigning them the following schematic structure:

IP

(60)

NP

r

PRO

r r

AP (dative)

VP (infinitive)

Given this configuration, it is possible that dative case is assigned directly to the semipredicatives sam and odin. Since dative is assignable to subjects under the "sister of F " configuration, I contend that the same process is at work here. That is, the assumption that the semipredicatives can appear as structural sisters of F allows them to be directly assigned dative case just as argument sisters of F are. There are several reasons why I opted for the particular phrase structure in (60). For one thing, if PRO is in SPEC-IP (or the specifier position of some other functional projection of VP) rather than in SPEC-VP, then the semipredicative cannot be internal to VP since it is predicated of PRO, assuming mutual c-command. For another, the second dative AP must be sister to F if it is to receive its case by dint of being the accidental, non-argument target of a configurational case assignment rule. Finally, adjoining the semipredicative to F in (60), rather than making it a daughter of IP, allows it to be differentiated formally from the subject and is consistent with binary rather than ternary branching. Next, consider why these two words might be able to receive case directly, even though they are adjectives. First, my proposal relies on the semipredicatives not entering into small clauses as other predicate adjectives do, hence they can receive case directly rather than only through agreement. Second, these adjectives are defective in that they belong to the "mixed" pronominal declension. Their non-oblique forms are thus those of nouns, so it comes as no surprise that they should be able to be assigned case directly, even if by accident of configuration, just as true NPs are. Another way to look at it is that the forms sam and odin are ambiguous between being short form (i.e. caseless) adverbial "adjectives" and long form (i.e. cased) pronominal adjectives. My claim is thus that the option of directly assigning case to

258

PARAMETERS OF SLAVIC MORPHOSYNTAX

odin and sam is based on the fact that they are sufficiently "noun-like" in that they display partial nominal morphology. Perhaps such adjectives are non-distinct from nouns at D-structure, and the configurational "sister of F " case rule blindly treats them as NP targets on the basis of this non-distinctness. Recall in this light the fact discussed in connection with (30) that in an earlier stage of Russian, when the short form (nominal) endings were generally productive, the second dative could occur with other adjectives. That is, precisely because the short form existed for ordinary adjectives and truly was a cased option, the second dative could be found on other predicate adjectives. This is attested by the following examples, which Comrie (1974) cites from Pushkin; (30a) is repeated as (61a): (61)

a. ja bojalsja odnogo: byt' ostavlenu na doroge 1 feared one(gen) to-be left(dat) on road 'I feared one thing: to be left on the road' b. prisudili ego byt' posazenomu na kol condemned him(acc) to-be impaled(dat) on stake 'they condemned him to be impaled at the stake'

The opposition between long form, as in (61b), and short form, as in (61a), was not one of cased versus caselessness: both ostavlenu 'left' and posazenomu 'impaled' were equally dative. They presumably received dative case by the same "sister of F " mechanism at work in the modern second dative construction. Thus, the limitation of the second dative appears to be intimately connected with the loss of the short form adjective. The proposal that semipredicatives receive case by virtue of their structural position, however, raises the same issue encountered by the agreement hypothesis: Why is it that in some clauses second datives are licensed although dative subjects are not? Both are sister to I'—and by this criterion both should be able to be assigned dative—but they are not treated equally. In particular, it seems that tense is necessary to activate structural case assignment by I' to its sister subject, but not to a semipredicative. Let us therefore ask what difference between subjects and semipredicatives might explain why the tense condition is irrelevant for F assigning dative to semipredicatives. The answer seems to me to lie in the relative argument status of these two targets: subjects are arguments of the verb, whereas the semipredicatives are not arguments. The relevant observation is that configurational principles of case assignment seem to apply with equal force to arguments and non-arguments. Thus, for example, it was suggested earlier that adjunct time/distance phrases appear in Russian in the accusative within VP, just like true arguments of the verb, even when the verb is intransitive. In other words, verbs that do not assign accusative case to their complements nonetheless may take accusative adjunct phrases. Moreover, as discussed in section 5.3.2, the same may be true for the genitive of negation. We thus conclude that licensing conditions such as (1) whether the clause is [ + tense] (for ' 'sister to F = dative"), or (2) whether the verb is a case-assigner (for "sister to V = accusative") pertain exclusively to arguments. A phrase not represented in the verb's predicate-argument structure, such as a semipredicative or time/distance phrase, may accidentally be assigned the appropriate structural case. However, since it is not

SECONDARY PREDICATION

259

assigned a semantic role by the verb, it is immune to any factors beyond pure configuration. It is precisely for this reason that, although dative subjects are sensitive to the feature [ +tense], second datives are not. In this section I have argued that a second dative clause such (21b) has the following structure: (62)

[IP PRO/NP(dat) [r [r I [VP prigotovit' obed] odnomu]]]

The semipredicatives are adjoined to F, making them configurationally similar to subjects. The difference between subject and adjunct is that the argument PRO cannot receive dative in the absence of tense, whereas this condition is irrelevant for the non-argument semipredicatives. Just as with temporal non-argument accusatives, the sensitivity to licensing conditions exhibited by arguments does not extend to non-arguments. The second dative is contingent, nonetheless, upon the existence of the possibility of assigning dative in the marked context of "sister to F." That is, I predict that only languages with dative subjects should also be able to have dative semipredicatives. This prediction is supported in detail by the cross-Slavic facts adduced in section 6.4, drawn from Greenberg and Franks (1991). This structural account also predicts that the second dative should be impossible in bare VP constructions, since an IP projection system is required for the second dative form to arise. I suggest that this is just what happens in gerundive and participial constructions, to which we now turn. 6.3.3. Gerunds and Participles Although gerundive and participial clauses are "functionally complete complexes" in the sense of Chomsky (1986b) that they contain the full range of arguments possible in any clause, predicate adjectives within these clauses display some striking properties. There are two related factors that essentially conspire to restrict the distribution of predicate adjectives in gerundive and participial clauses: 1. They always have a PRO subject. 2. They are reduced clauses in the sense of lacking (at least some) sentential functional projections of VP. In connection with these issues, the government status of the PRO subject should determine its ability to transmit case under control and the presence of an IP projection should correlate with the availability of the second dative. Hence, both the agreement and non-agreement options raise structural questions.51 The data regarding the second dative are fairly clear—this form does not occur precisely where we expect it to be impossible. With respect to case transmission (of nominative) by a PRO subject of a gerund or participle, the data are not as clear. Here I shall describe the results as reported in Franks (1990a), considering alternative scenarios as we proceed. Predicate adjectives in gerundive clauses fail to agree in case with PRO's antecedent, even when that antecedent is a nominative subject.52

260

(63)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a. [PRO vernuvsis' domoj ?p'jan/%p'janyj/p'janym], Ivan having-returned home drunk(m)/(nom)/(inst) (nom) srazu leg spat' at-once lay to-sleep 'returning home drunk, Ivan went at once to bed' b. [PRO vernuvSis' domoj ?goloden/*golodnyj/golodnym], Ivanu having-returned home hungry(m)/(nom)/(inst) (dat) zaxotelos' est' wanted to-eat 'returning home hungry, Ivan felt like eating'

The behavior of predicate adjectives in gerund clauses follows directly from our analysis, under the assumption that since gerunds are always adjuncts, never arguments , they can never be properly governed by the matrix verb. If the PRO of gerund clauses is never grammatically controlled, agreement should always be impossible.53 The issue of whether PRO in gerunds is governed or not is somewhat murky, since two Russian speakers consulted readily accepted long form agreeing adjectives in (63a). There are several possible explanations for this, which I briefly discuss before developing the pronominal PRO analysis. One obvious possibility is that (for such speakers at least) PRO is lexically governed in such contexts. This turns out to be difficult to establish, since it depends on the one hand on how proper government is defined and on the other hand on how anaphoric PRO subjects of gerunds really are. On the formal side, it is conceivable that these PROs are governed either internally or externally. The latter option seems doubtful, especially for "detached" gerundive clauses (cf. Rappaport, 1984). I did have to allow an unprotected PRO in infinitival purpose clauses, as in (28), to be governed by the matrix V, however, even though it was also inside an adjunct phrase. Alternatively, if gerundive clauses are bare VPs with PRO in VP-specifier position, then they will be governed by V under traditional definitions of government. Nonetheless, it is reasonable to suppose that gerunds involve some functional projection beyond VP (if only an Aspect Phrase) and, crucially, an Adverb Phrase headed by the gerundive suffix, into which the V incorporates, adjoining to the left of the suffix. PRO presumably raises from VPspecifier to the specifier position of this AdvP, thereby escaping lexical government by V. Just as suggested above for CPs, it moves precisely in order not to be governed by the head V, since if it were governed it would be an anaphor and need to be locally bound. The issue thus boils down to whether PRO in gerundive clauses exhibits the binding characteristics of anaphoric or pronominal PRO. I will assume the latter, although this conclusion remains open I think to considerable debate. It is also conceivable that the acceptability of the (long form) nominative adjective in examples like (63a) is not actually indicative of case transmission from PRO's controller and mediation by PRO because one of the speakers who accepted (63a) with long fonnp'janyj also allowed the nominative in (63b), at least to the extent that he accepted non-nominative controllers of gerundive PROs at all. Perhaps, then, for some speakers of Russian there is an alternative mechanism for marking predicate adjectives as nominative (or "nominative-like"). It could truly be a default case, along the lines of similar phenomena in Icelandic—maybe even with agree-

SECONDARY PREDICATION

261

ment with a nominative PRO in gerunds. It is finally possible that the nominative predicate adjective is agreeing with a small pro nominative subject, although this is unlikely given that not even this speaker freely allows overt nominative subjects of gerunds. Returning, then, to the scenario under which agreement is blocked, it is important to note that this holds despite the fact that gerund clauses are categorially not CPs, as demonstrated by their lack of a COMP position. This is evidenced by the unavailability of w/z-movement within gerund clauses, the impossibility of extraction from gerund clauses, and the absence of tense or agreement features on gerunds (cf. also Greenberg, 1988, for discussion). (64)

a. *[NP kniga[7? kotoruju; [VPPRO citaja e{ . . . book which(acc) reading b. *[VP sprosil [?? kudaj [VPPRO pobezav e; . . . asked where having-run

(65)

a. *[CP ctoj [IP Ivan vose'l v komnatu [VPPRO citaja ej]] what(acc) (nom) entered into room reading b. *[NP kniga [CP kotoruju; [IP on umer [VP PRO napisav ej]] book which(acc) he died having-written

Structures such as (64) are unacceptable (in the literary language) because gerunds lack a CP projection, consequently there is no COMP (the specifier of CP) to serve as the target position for w/z-movement. Extraction from a gerundive clause, as in (65), is similarly impossible because (1) there is no COMP available as an escape hatch within the gerund clause, and (2) the gerund clause is not an argument of the higher verb, so it is a barrier for movement. Note that in these examples I have labeled the gerundive phrase a VP, although as mentioned above they are probably maximally Adverb Phrases, with an underlying structure roughly as in (66): (66)

|AdvP [Mv-v(si)/-ja]

[VP V NP . . . ]]

V incorporates to the left of Adv, which is a bound morpheme, giving the surface word order and explaining why gerundive clauses have the internal structure of VPs but the external distribution of AdvPs. Of course, as discussed by Babby (1974) and Greenberg (1988), for example, the entire gerund clause can move to the COMP of the relative clause, as shown in (67). (67)

[NP kniga [CP [AdvP PRO napisav kotorujUj [IP on umer ej]] book having-written which(acc) he died

This possibility arises precisely because gerundive clauses are actually Adverb Phrases, since other AdvPs—but crucially not VPs—can undergo question or relative clause formation. Note in this regard that although collocations such as in (64) are actually permissible in colloquial speech, these can now be analyzed as scrambled versions of acceptable constructions with the verb pied-piped into COMP as in (67); for example, citaja kotoruju must scramble to produce kotoruju citaja. Further evidence that gerundive clauses are smaller than CPs comes from the fact that gerunds have no independent tense reference, their temporal properties deriving solely from their own aspect and the tense of the main clause, as shown in (68).54

262

PARAMETERS OF SLAVIC MORPHOSYNTAX

(68) a. on vosel v komnatu, citaja/procitav gazetu 'he walked into the room, reading/having read the paper' b. on vojdet v komnatu, citaja/procitav gazetu 'he will walk into the room, reading/having read the paper' There are several possibilities as to what the maximal functional projection of V is in a gerundive clause, assuming they have aspect but no other verbal categories. Since these lack tense and agreement, I assume gerunds lack not only CP but also IP. Aspect is therefore not part of IP, but rather either heads its own projection, which takes VP as a complement, or is intrinsic to V. Although the former may well be correct—assuming a clause to consist at least of the functional categories CP, AgrP, TP, AspP, VP, hierarchically arranged in that order—here I shall simply represent gerundive phrases as bare VPs. Even so, they are embedded in AdvPs, as in (66). If so, PRO subjects of gerunds are never truly anaphoric, even though they appear in bare VPs, since these VPs are never properly governed. This explains their inability to transmit case. Therefore, as shown in (63), an adjective predicated of the PRO subject of a gerund appears in the instrumental or (as always, marginally) in the short form. This structure makes one further surprising prediction: if gerundive clauses are not IPs, they should not admit second dative semipredicatives. Indeed, it turns out that the default dative is universally unavailable to the semipredicatives sam and odin in gerundive clauses. That is, even when the antecedent of PRO is unable to transmit its case, the semipredicatives nonetheless cannot appear in the "default" dative. This state of affairs holds regardless of the reason case transmission is blocked. In (69), for example, the antecedent is a PRO itself, so there is no case for it to transmit. (69) a.

[PRO vernuvsis' domoj], neobxodimo [PRO srazu lee' spat'] having-returned home necessary at-once to-lie to-sleep ' after coming home, one must go right to bed'

b. *[PRO vernuvsis' domoj odnomu], neobxodimo [PRO srazu having-returned home (dat) necessary at-once lee' spat'] to-lie to-sleep ' after coming home alone, one must go right to bed' c. [PRO podbezav k stancii], nel'zja [PRO srazu prygat' v having-run-up to station should-not at-once to-jump into poezd] train ' after running up to the station, one should not jump right into the train' d. *[PRO podbezav k stancii odnomu], nel'zja [PRO srazu having-run-up to station (dat) should-not at-once prygat' v poezd] to-jump into train ' after running up to the station alone, one should not jump right into the train'

SECONDARY PREDICATION

263

Examples (69a, c) are normative, since the PRO subject of the gerund is controlled by a higher subject, even though that subject is also a PRO. It is impossible, however, to insert a dative semipredicative, as shown in (69b, d). This demonstrates that the second dative cannot be the result of agreement with a covert dative PRO, since then there would be no way to prevent (69b, d). Moreover, a matrix second dative would be perfectly acceptable: (70)

a. neobxodimo lee' spat' odnomu 'it is necessary to go to sleep alone(dat)' b. nel'zja prygat' v poezd odnomu 'it is not allowed to jump into a train alone(dat)'

This clearly shows that it must be a property peculiar to the gerundive clause itself that renders the second dative unavailable. A consideration of (non-normative) non-subject controlled PRO leads to similar conclusions. In (71b) the PRO subject of the gerund refers to the dative experiencer emu and in (72b) its reference is determined outside the sentence. (71)

a.

[PRO slusaja etu muzyku], emu stalo grustno listening-to this music, he(dat) became(n) sad b. *[PRO slusaja etu muzyku odnomu], emu stalo grustno listening-to this music alone(dat), he(dat) became(n) sad

(72)

a.

[PRO podbezav k having-run-up to b. *[PRO podbezav k having-run-up to

stancii], poezd uze otosel station, train(nom) already left stancii odnomu], poezd uze otosel station alone(dat), train(nom) already left

Although (71a) and (72a) are not permissible in the literary language, they are colloquially acceptable, and, significantly, speakers who acknowledge them still do not admit (71b) and (72b) with the second dative. Note that the colloquial acceptability of examples like (7la), and especially ones like (72a), where there is no syntactic antecedent, provides further evidence that the PRO subject of gerunds cannot be a true anaphor, but is pronominal instead. In all other instances of pronominal PRO considered, the "default" second dative is viable. This mysterious gap in the second dative paradigm proves, however, that the second dative does not arise automatically by default, but must in fact be configurationally licensed. If it requires sisterhood to I' (or to the intermediate projection of some other functional category outside VP), then we have a ready explanation for why gerundive clauses have no source for second dative assignment. Before turning to some further arguments in favor of this analysis, there is one controversial aspect of the behavior of gerund clauses that needs to be addressed. If the PRO subject of gerunds is never properly governed, hence always pronominal, it should never induce predicate adjective agreement. Although we have seen that this is indeed so for ordinary predicate adjectives, as in (63a),55 the semipredicatives odin and sam systematically appear to flout this restriction. Compare (73) with (63a).

264

(73)

PARAMETERS OF SLAVIC MORPHOSYNTAX

[PRO vernuvsis' domoj odin], Ivan srazu leg spat' having-returned home alone(nom) (nom) at-once lay to-sleep 'returning home alone, Ivan at once went to bed'

The semipredicative odin here seems to be agreeing with Ivan, presumably by virtue of Ivan controlling PRO. However, the failure of long form adjectives to agree under the same circumstances suggests that, whatever it is, odin cannot actually be an agreeing nominative form in (73). Further reason for not treating odin as nominative in (73) is that these putatively nominative semipredicates are marginally acceptable even in gerundive constructions where PRO has no nominative antecedent, as in (74), cited in Neidle (1988) and Schein (1982). (74)

[PRO podbezav k stancii odin], poezd uze otosel having-run-up to station alone(nom), train(nom) already left

Interestingly, the only conceivable form of the semipredicative in this context is odin not odnomu; compare (72b). The fact that constructions such as (74) are generally awkward I attribute to the strong normative prohibition against non-subject controlled gerunds. This conclusion is supported by (75), which provides a similar contrast to (69b) and, unlike (74), does not suffer from normative prohibitions since the PRO subject of the gerund is anteceded by the matrix PRO subject. (75)

[PRO vernuvsis' domoj odin], neobxodimo srazu lee' spat' having-returned home (nom) necessary at-once to-lie to-sleep 'after coming home alone, one must go right to bed'

These contrasts I believe offer quite remarkable proof that predicate adjectives never truly agree in gerundive clauses. More support for this contention can be drawn from Polish, which is treated in more detail in section 6.4.3. In Franks (1983, 1985) it was observed that in Polish, like Russian, an agreeing predicate adjective is never acceptable in a gerundive clause:56 (76)

a. *b^dac mlody, Jan zwiedziX wiele krajow being young(nom), Jan(nom) visited many countries b. *idac sam, Jan przybyi na czas going alone(nom), Jan(nom) arrived on time

The instrumental mtodym must appear in (76a), and in (76b) it is colloquially possible to use "dative-like" samemu, but only as a frozen, adverbialized form, unchanging in gender and number regardless of antecedent (cf. Comrie, 1974, 140; Brodowska-Honowska, 1964). Comparing, then, Polish (76b) with Russian (74) and (75), we find that Polish allows "dative" but not nominative, while Russian allows "nominative" but not dative. How can this curious state of affairs be reconciled in the two languages? As argued above, both agreeing nominative and non-agreeing dative should technically be impossible in gerund clauses—hence my use of quotation marks in the preceding paragraph. Agreement is ruled out because PRO is not anaphoric and

SECONDARY PREDICATION

265

the second dative is ruled out because gerunds are not IPs. Since Polish samemu, to the extent that this is acceptable in (76b), is not really dative, but rather a caseless adverbial, I propose explaining the possibility of Russian odinlsam in gerund clauses in a similar fashion. Recall that odin and sam are ambiguous between being nominative pronominal declension adjectives or caseless short forms. This suggests that whenever odin or sam appears in a context that is otherwise expected to be caseless (as independently determined, for example, by the impossibility of inserting unequivocally nominative long forms or by the lack of any clear source for case), it is really a caseless short form. That is, odin in (73)—(75) should be glossed as a "masculine (singular)" short form rather than a "nominative" long form. Note that the absence of this contrast in Polish precludes a similar interpretation for Polish sam in (76b), which can only be analyzed as cased and is therefore unequivocally ungrammatical. It thus becomes obvious why Polish and Russian make use of superficially opposite strategies: Polish samemu (but not Russian samomu) and Russian sam (but not Polish sam) have alternative analyses as caseless forms. The final body of evidence supporting this account is based on the behavior of predicate adjectives in infinitival complements to adjectives and participles. These infinitivals function similarly to gerunds, suggesting a similar bare VP analysis. In brief, long form agreeing adjectives are inadmissible in (77), demonstrating that odin in (78) cannot be the result of agreement and must, instead, be regarded as a short form.57 (77)

a.

[DP celovek [ AP PRO zelajuscij [VPPRO vernut'sja dornoj person(m) wishing(m nom) to-return home golodnym/*golodnyj/?goloden]]]— Cudak hungry(inst)(m nom)/(m) oddball ' a person who wants to come home hungry is strange' b. [DP celovek [APPRO gotovyj [VPPRO borot'sja person(m) ready(m nom) to-fight p'janym/*p'janyj/?p'jan]]]— glup drunk(inst)/(m nom)/(m) stupid ' a person who is ready to fight drunk is stupid'

(78)

a. [DP celovek [APPRO zelajugcij [VP PRO vernut'sja domoj person(m) wishing(m nom) to-return home odin/*odnomu]]]— £udak alone(m)/(dat) oddball 'a person who wants to come home alone is strange' b. [DP celovek [AP PRO gotovyj [VP PRO borot'sja person(m) ready(m nom) to-fight odin/*odnomu]J]— glup alone(m)/(dat) stupid 'a person who is ready to fight alone is stupid'

The failure of case transmission in these examples follows from the pronominal status of the PRO subject of the AP.58 On the other hand, the PRO subject of a

PARAMETERS OF SLAVIC MORPHOSYNTAX

266

comparable infinitive, which I have represented in (79) as a VP but could equally well be regarded as a [ — tense] TP, is presumably anaphoric and lexically governed by the adjective/participle, since the long form nominative is acceptable. (79)

a. Ivan zelaet [VPPRO vernut'sja domoj golodnyj/golodnym/?goloden] (nom) wishes to-retum home hungry(nom)/(inst)/(m) b. Ivan gotov[VPPRO borot'sja p'janyj/p'janym/?p'jan] (nom) ready to-fight drunk(nom)/(inst)/(m)

The PRO subject of the APs in (77) and (78) is not properly governed, however, and thus lacks anaphoric status; its reference is determined through predication on celovek 'person' rather than control. Thus, even though this PRO itself controls the PRO subject of the infinitive complement to the adjective, case transmission is blocked. Note that the second dative is not a viable option, even when the participial phrase is oblique: (80)

a. *ja vstretil celoveka, zelajuscego vernut'sja domoj odnomu I met man(acc) wishing(acc) to-return home alone(dat) b. *ja vstretil celoveka, gotovogo borot'sja odnomu I met man(acc) ready(acc) to-fight alone(dat)

Since the semipredicative is inside the infinitival clause, the impossibility of the second dative in (80) supports my claim that obligatorily controlled infinitival clauses must be bare VPs, rather than IPs. Otherwise, the second dative option should be available under sisterhood to I'. The reason why the second dative cannot appear is thus the same as with gerundive clauses—controlled PRO infinitivals are not complete IPs. Observe, finally, that if this analysis is correct, the admissibility of the second dative after participles has nothing to do with participial constructions per se, but merely the categorial status of the participle's complement. We therefore predict that second datives should reappear whenever the infinitival clause is a CP. Compare (81a) and (81b). (81)

a. *. . . celoveka, [PRO zelajuscego [VP PRO zit' samomu]] person, wishing to live alone(dat) b.

. . .celoveka, [PRO ne znajuscego [CP kak [ IP PRO tuda person NEG knowing how there dobrat'sja samomu]]] to-get alone(dat)

In (81a), the lower PRO is in VP, properly governed, hence anaphoric and controlled by the higher pronominal one. In (81b), the presence of the complementizer kak reveals that the complement clause must be a CP, hence there is also an IP and the second dative can be assigned. Moreover, the lower PRO in (81b) is not properly governed, hence it is pronominal and potentially arbitrary in reference—it may refer to the higher PRO, which is necessarily predicated of celovek, or not. In this section I have examined a curious and somewhat marginal body of data that bears on the issues of predicate adjective agreement and the source of the second

SECONDARY PREDICATION

267

dative. The surprising behavior of secondary predicates in gerundive and certain adjectival constructions was investigated in detail. Two significant observations were made: the second dative is not available in these clauses and, for many but not all speakers, case -transmission is also blocked. These facts, for the most part previously unobserved, provide new support for the above analysis. The first observation follows from my idea that the second dative is structurally assigned, since it can only arise when the semipredicative appears in a full CP (= S') clause. The second one suggests that the PRO subjects of such clauses are pronominal rather than anaphoric, although variation and the subtlety of the data make this conclusion more tenuous. In sum, these marginal constructions provide novel confirmation for the phrase structural account of the second dative.

6.4. Parametric Variation In this section,59 I describe the distribution of apparent dative subjects and the second dative in various Slavic languages.60 While there are significant differences among the languages with respect to the extent to which they manifest these constructions, the investigation reveals a surprising consistency in the range of phenomena that exist in each language: for any given language, the possibility of having dative subjects with infinitives (but not with impersonal predicatives) seems to correlate with the existence of the second dative. This relation sheds light on the analysis of the second dative and provides strong evidence for the existence of dative subjects. It is shown how, in the closely related languages of the Slavic family, clusters of differing surface properties are the direct result of relatively minor abstract oppositions. In this way, the analysis corroborates the Principles and Parameters approach to Universal Grammar. 6.4.1. Non-Agreeing Semipredicatives As we shall see from the data presented below, the second dative does not occur in all of the languages. In this respect, it seems similar in distribution to the dative + infinitive construction. In Polish, which will be discussed in more detail in section 6.4.3, the behavior of sam (and jederi) is actually very similar to that of sam (and odiri) in Russian.61 These semipredicatives agree in gender, number, and case with their antecedents within the same finite clause, as shown in (82)-(83). (82)

Janek przyszedt sam/jeden (nom) came himself(nom)/alone(nom) 'Janek came himself/alone'

(83)

zostawmy Jerzego samego/jednego let's-leave (ace) himself(acc)/alone(acc) 'let's leave Jerzy on his own'

In infinitival phrases, the distribution of the semipredicatives is also very similar to their distribution in Russian. If sam (or jederi) refers to the subject of the matrix

268

PARAMETERS OF SLAVIC MORPHOSYNTAX

verb, and there is no intervening complementizer, then it agrees with that subject NP in case, as in (84). (84)

Maria lubi gotowac sama (nom) likes to-cook herself(nom) 'Maria likes to cook by herself

In other instances, the semipredicative fails to agree with its antecedent and appears instead in the default second dative. In (85), the presence of a complementizer intervening between the semipredicative in the purpose clause and its matrix antecedent yields the non-agreeing second dative, just as in Russian. In (86), the antecedent is not the subject of the sentence, and the semipredicative also appears in the second dative. Finally, in (87) there is no visible antecedent, hence the default form is used once again. (85)

dziecko jest za male, zeby zapiac plaszcz samemu child(nom) is too small COMP to-button coat himself(dat) 'the child is too small to button his coat himself

(86)

wazne jest dlja mnie isc samemu important is for me(gen) to-go myself(dat) 'it is important for me to go myself

(87)

dobrze jest isc samemu good is to-go himself(dat) 'it is good to go alone'

All three of these examples can be treated just as the comparable Russian ones discussed earlier in this chapter, with PRO not being properly governed, hence pronominal, and consequently not able to transmit case to the predicate adjective. The remaining West and South Slavic languages completely lack the second dative phenomenon. In South Slavic Slovenian and Serbo-Croatian and West Slavic Slovak, the default case of the semipredicatives is not the second dative, but rather appears to be the second nominative. In other words, when a semipredicative occurs in an infinitival phrase where agreement is impossible, then a nominative form is used. This is indicated by (88) for Slovenian and (89) for Serbo-Croatian, where there is no overt antecedent with which the semipredicate sam might agree. (88)

a. ni dobro ziveti sam NEG good to-live oneself(nom) 'it is not good to live alone' b. pomembno je uciti se sam important is to-study oneself(nom) 'it is important to study alone'

(89)

a. vazno je ziveti sam important is to-live alone(nom) 'it is important to live alone'

SECONDARY PREDICATION

269

b. meni bi smetalo raditi sam me(dat) SUBJ bother to-work alone(nom) 'it would bother me to work alone' In Slovak, the semipredicatives sdm and samotny, as well as all adjectival secondary predicates, always agree in case with their overt antecedent. This is true regardless of whether they and their antecedent are in the same finite clause, as in (90), or they are in an infinitival phrase and their antecedent is in the next higher clause, as in (91). The second dative as well can appear only as the result of agreement with an overt dative NP, as in (92). There is thus no non-agreeing second dative in Slovak.62 (90)

mat' ju nasla vyplakanu mother her(acc) found in-tears(acc) 'her mother found her in tears'

(91)

nechava ju starat' sa o domacnost' samu he-leaves her(acc) to-take-care-of about housework herself(acc) 'he leaves her to look after the housework herself

(92)

. . .kde mu prichodilo bludit' samemu where him(dat) it-came to-wander himself(dat) ' . . . where it was necessary for him to wander on his own'

Once again, however, when there is no antecedent, the default case is the nominative.63 (93)

nie je dobre jest' samo NEG is good to-eat alone(nom) 'it isn't good to eat alone'

This section has revealed the following facts concerning the case of the semipredicatives: in some languages the default (non-agreeing) case is the dative, while in others it is the nominative. This information may be summarized for the individual languages as follows: (94)

Language

Default case of semipredicative

Russian Polish Slovenian Serbo-Croatian

DATIVE DATIVE NOMINATIVE NOMINATIVE

Slovak

NOMINATIVE

6.4.2. Dative Subjects This section surveys the range of potential dative subjects in the Slavic languages considered in section 6.4.1. As mentioned above, one can distinguish two types of

270

PARAMETERS OF SLAVIC MORPHOSYNTAX

constructions with possible dative subjects: category-of-state clauses and infinitive clauses, as was seen in (45) and (46), respectively. There are two logically possible ways of accounting for the dative case on the NPs in these types of clauses. On the one hand, one could assume that the dative arises in the same way that it arises in other constructions, that is, it might turn out that the NP is sister to V. In this instance, the NP would not be a true subject. On the other hand, it might turn out that for some reason a special rule marks the otherwise nominative subject NP with the dative, yielding a dative subject. It is also possible that the dative NPs in the category-of-state phrases are subjects while the dative NPs in the infinitive phrases are not, or vice versa. I will in fact argue against a uniform account, adducing comparative evidence that only dative NPs with infinitive phrases are true subjects. 6.4.2.1. Infinitives Slavic languages exhibit a striking degree of variation with respect to the possibility of marking subjects of infinitives dative. Some languages allow dative subjects in infinitive phrases and others do not. Although much more restrictive than Russian, Polish does permit subjects of infinitives to appear in the dative, as can be seen in (95)-(97), from Lewandowska (1976, 311): (95)

tobie isc do klasztoru, nie wychodzic zamaz you(dat) to-go to nunnery NEG to-many 'you should go to a nunnery, not get married'

(96)

po co ci sobie zawracac tym gtowe. for what you(dat) self to-turn that head 'why fall in love with that one?'

(97)

chocbys byl Waligora nie mierzyc ci sie z ta gora even-if-you were (a giant) NEG to-contend you(dat) with that mountain 'even if you were very strong, you still couldn't contend with that mountain'

Furthermore, as Lewandowska explains, this type of sentence occurred considerably more frequently in older stages of Polish than in the contemporary language, and the use of a dative NP as the subject of an infinitival phrase is much more limited in Polish than in Russian. Slovenian, on the other hand, has virtually eliminated the use of the dative + infinitive construction. While examples such as those in (98), from Lencek (1982, 51) may be found in the written language, they are no longer used by native speakers and are immediately recognized as archaic style. (98)

a. kaj nam je napraviti? what we(dat) is to-do 'what are we to do?' b. otrokom je molcati children(dat) is to-be-quiet 'children should be quiet'

SECONDARY PREDICATION

271

The use of such dative subjects is extremely restricted, such that (98a) has the flavor of a formulaic expression and (98b) can only be understood as strongly elliptical. In the contemporary language, Slovenian speakers instead use tensed verb forms with agreement to express these meanings. This is shown in (99) and (100), where (99a) and (lOOa) are contemporary expressions and (99b) and (lOOb) are archaic variants, which were judged infelicitous in the modern language. (99)

a. moramo iti domov we-have to-go home b. iti nam je domov to-go us(dat) is home 'we have to go home'

(100)

a. kam naj gremo? to-where shall we-go b. kam nam je oditi? to-where us(dat) is to-go 'where shall we go?'

Serbo-Croatian and Slovak also differ from Russian in this regard. As can be seen from the Serbo-Croatian examples in (lOla) and (102a) and the Slovak examples in (103a) and (104a), compared with the standard Russian variants in (101b)-(104b), these languages differ from Russian in that they do not allow the use of the dative with the infinitive. Instead, as in Slovenian, they utilize a finite, agreeing form of the verb.64 (101)

a. nemamo sta ovde da radimo we-don't-have what here COMP we-do b. nam zdes' necego delat' us(dat) here there-is-nothing to-do

(102)

a. nemas kome da se zalis you-don't-have who COMP REFL you-complain b. tebe nekomu pozalovat'sja you(dat) there-is-no one to-cornplain-to

(103)

a. co what b. cto what

(104)

a. ja spevaka nepochopim/nemozem pochopit' I(nom) singer NEG-understand/NEG-can to-understand b. mne pevca ne ponjat' me(dat) singer NEG to-understand

mam robit' I-have to-do mne delat'? me(dat) to-do

A summary of these data is provided in (105):

272

PARAMETERS OF SLAVIC MORPHOSYNTAX

(105) Language Russian Polish Slovenian Serbo-Croatian Slovak

Dative NP with infinitive yes yes, but more restricted than Russian no no no

We thus find languages with dative subjects in infinitival clauses (i.e. Russian and Polish) and languages without them (i.e. Slovenian, Serbo-Croatian, and Slovak). Furthermore, it should be borne in mind that Polish, while still showing the dative subject construction, seems to be in the process of losing it. 6.4.2.2. Impersonate I now turn to the other type of construction that is often described in terms of taking a dative subject. Finite clauses in Slavic standardly occur with nominative subjects and agreeing verbs. There is a class of "impersonal" predicates, however, referred to in the Russian linguistic tradition as ' 'category-of-state'' (kategorija sostojanija) predicates, that also seem to admit dative subjects. Some typical examples were given in (45); these are repeated as (106): (106)

a. emu zal' etu devusku him(dat) sorry that(acc) girl(acc) 'he feels sorry for that girl' b. mne legko govorit' po-russki me(dat) easy to-speak in Russian 'it is easy for me to speak Russian'

The logical subject in these examples is expressed in the dative case and there can be no overt nominative subject, hence no subject-verb agreement. It is in this sense that such sentences are "impersonal." Impersonal sentences differ from infinitival sentences in terms of the acceptability of a "subjective" dative experiencer NP. Contrary to the results of the preceding section regarding dative subjects of infinitives, it turns out that all the Slavic languages allow dative experiencers with impersonal predicates. There thus seems to be no direct connection between this phenomenon and the possibility of dative subjects with infinitives. The viability of dative experiencers with impersonal predicates is shown in (107) for Polish, in (108) for Slovenian, in (109) for Serbo-Croatian, and in (110) for Slovak.65 (107) a. zal mi ciebie sorry me(dat) you 'I feel sorry for you' b. cie.zko Marii mowic Janowi o sobie hard Mary(dat) to-tell John about self 'it's hard for Mary to tell John about herself

SECONDARY PREDICATION

(108)

273

a. zal mu je za to dekle sorry him(dat) is for that girl 'he feels sorry for that girl' b. tezko mi je govoriti rusko hard me(dat) is to-speak Russian 'it is hard for me to speak Russian'

(109) a. meni je tesko govoriti ruski me(dat) is hard to-speak Russian 'it's hard for me to speak Russian' b. zao mi je sto niste do§li sorry me(dat) is that NEG-AUX came Tm sorry that you didn't come' (110)

a. a nie je mi 1'uto minulosti ani trosku and NEG is me(dat) sorry past even little 'and I am not sorry about the past even a bit' b. treba vam uz ist' must you(dat) already to-go 'you have to go already'

The appearance of the dative NPs in impersonal clauses in all of these languages—as opposed to their appearance in only some of the languages in infinitive clauses— suggests that these are in fact different types of dative NPs, resulting from distinct case-marking strategies. I thus conclude that these two "subjective" datives have different structural sources; the dative subject of infinitives is assigned under sisterhood to F, whereas the experiencer dative is perhaps either structurally comparable to an indirect object or arises by virtue of some kind of covert modal.66 A survey of the data presented in this section leads to a conclusion concerning the status of the various dative NPs and points toward an interesting correlation that exists throughout the languages. Since the dative + impersonal construction appears in all the Slavic languages, whereas the dative + infinitive construction appears only in selected languages, the two types of dative NPs must be analyzed differently. More interestingly, it turns out that the languages that have the second dative phenomenon are precisely those languages that have dative subjects in infinitival sentences. A comparison of the charts in (94) and (105) yields the following correlation: (111) Languages with dative in infinitive sentence

Languages without dative in infinitive sentence

Russian Polish

Default case of semipredicative DATIVE DATIVE

Slovenian Serbo-Croatian Slovak

NOMINATIVE NOMINATIVE NOMINATIVE

274

PARAMETERS OF SLAVIC MORPHOSYNTAX

This account explains why the distribution of the second dative throughout the Slavic languages parallels the distribution of dative subjects in infinitive clauses— they are both consequences of a marked rule assigning dative to the sister of a F. This approach solves the problem of relating the second dative to the existence of dative subjects without actually endowing PRO with case. The two constructions are instead connected indirectly, via the mediation of a special structural case-assigning property parametrically present in certain Slavic languages. Therefore, the presence of a default dative semipredicative does not imply the existence of a dative subject in any given construction; it implies rather the existence of dative subjects in the language. Before turning to a discussion of certain idiosyncracies of secondary predication in Polish, it is worth digressing to consider a curious fact about experiencer datives in Russian which suggests that they cannot literally be in indirect object position. These facts are discussed in Chvany (1970), Babby (1975a), and Corbett (1975, 1983), as well as in Franks and Greenberg (1988). There we argued that they implied that experiencer datives were structurally identical to dative subjects of infinitives. The cross-linguistic data show, however, that this cannot be so and that an alternative account is required. This insight demonstrates the power of the comparative method for syntactic analysis, since a consideration of the way the two types of "subjective" datives line up across languages forces a different conclusion than the one reached on the basis of Russian data alone. Modal category-of~state words like nuzno appear in two superficially distinct types of construction. In the first type, nuzno appears in the default neuter singular form along with an infinitival phrase and an optional dative NP, which serves as the understood subject of the infinitive: (112)

(nine) nulno procitat' etu knigu me(dat) necessary(n) to-read that book 'it is necessary (for me) to read that book'

In the second type of construction, nuzno appears with a nominative NP, with which it agrees in gender and number according to the regular subject-verb agreement rule, as well as with a dative NP. (113)

ona nuzna obscestvu she(nom) necessary(f) society(dat) 'she is necessary to society'

The central issue here is whether or not these two types of dative with nuzno 'necessary' should be identified. One might suppose, for example, that in (111) the infinitival clause is the subject, comparable to the nominative NP ona 'she' in (113), thereby allowing the two datives mne 'me' and obscestvu 'society' to be analyzed in a parallel fashion as complements to nuzno, as in Matual (1977), for example. This sort of approach would seem to make sense on semantic grounds, since it would render the themes ('that which is necessary') and goals ('who/what it is necessary for') parallel in the two constructions, as well as on structural grounds, since the two datives are incompatible with each other, just as the nominative NP and infinitival clause are. Nonetheless, it is far from clear that such an analysis is justified on either

SECONDARY PREDICATION

275

count, since one could equally well argue that in (112) the dative NP bears the experiencer role, whereas in (113) the dative NP bears the goal role. There would thus be two different predicate-argument structures for nuzno, one with an experiencer and clausal theme, the other with a goal and nominal theme.67 Alternatively, then, one might suppose that in (112) the dative NP is the subject of nuzno, whereas in (113) it is an object. It turns out that there is a test that helps to distinguish these two dative NPs. When nuzno phrases such as (112) and (113) are contained in small clause complements, the acceptability of the two dative NPs differs in a surprising way. As can be seen from (114) and (115), nuzno phrases are able to serve as small clauses after the verb scitat' 'to consider'. The modal adjective appears in the instrumental, just like any predicate AP or NP after this verb: (114)

on s&taet [sc (*mne) nuznym procitat' etu knigu] he considers me(dat) necessary to-read this book 'he considers it necessary (for me) to read this book'

(115)

my scitaem [sc ee nuznoj obscestvu] we consider her(acc) necessary society(dat) 'we consider her necessary to society'

It is unexpected under the first analysis suggested above that the dative NPs should be treated differently in (114) and (115). However, in (114) expressing the dative NP of (112) is absolutely prohibited, although in (115) the dative NP freely remains. Additionally, the nominative NP of (113) appears in the accusative in (115), presumably because as the subject of the small clause it is assigned case from the matrix verb (under the mechanism of Exceptional Case Marking, as discussed in chapter 5). If both dative NPs were objects, there would be no explanation for why they should behave differently when embedded in a small clause. On the basis of these data I thus conclude that only the dative NP in (113/115) is an object of nuzno, and that the experiencer dative in (112/114) should be analyzed in some other way. In Franks and Greenberg (1988) we maintained that these facts would follow if the experiencer were a subject, assuming that dative subjects arise under sisterhood to a [ + tense, -agreement] F and small clauses lack tense, if not an IP completely. This precise solution is no longer open, since the comparative data show that the distribution of dative experiencers is different from that of true dative subjects (of infinitives). Why then is (114) ungrammatical whenever the dative NP is present? Although I cannot give a definitive answer to this question, there are several reasonable possibilities. Whatever the ultimate explanation for (114), it seems to me that just because the dative NP cannot be analyzed as a canonical indirect object, this does not mean that the only alternative is for it to be a canonical subject in IPspecifier position. There are two related avenues for resolving the dilemma posed by (114) versus (115). Following Corbett (1975, 1983b), one might contend that nuzno itself represents two distinct kinds of items in the two constructions. He argues convincingly that in (114) and its ilk the instrumental is really a predicate NP (with an empty N head) rather than a simple AP. Reasons for treating predicate instrumentals such as nuznym 'necessary', neobxodimym 'unavoidable', anddolznym, 'obligatory' as substantivized adjectives are numerous. Most importantly, Corbett cites examples

276

PARAMETERS OF SLAVIC MORPHOSYNTAX

such as (116), which demonstrate that the form is neuter singular even when predicated of plural or feminine NPs and also shows that the nominal analysis is independently required for precisely those elements that enter into the small clause construction. (116)

ja scitaju neobxodimym cistotu vo vse'm 'I consider necessary(m-n inst) cleanliness(facc) in everything'

Since the masculine-neuter form neobxodimym fails to agree in gender with feminine cistotu, it must be analyzed as the instrumental of a predicate (neuter) NP neobxodimoe 'necessity' rather than an agreeing AP. In other words, contra Babby (1974), for instance, these small clause instrumentals are not predicative (adverbial) modals with case imposed on them, but rather substantivized adjectives derived from these modals. Given this, Corbett maintains that the reason why (114) differs from (115) has to do with the fact that in the former nuznym is a noun whereas in the latter nuznoj is an adjective—although the adjective takes a dative complement, the noun does not. The dative in the unembedded variants (112) and (113) are similarly distinct, since in (112) the dative arises for whatever reason subjects of impersonal predicates are dative, while in (113) the dative is the complement to a personal predicate, the agreeing short form adjective nuzna. Along similar lines, it is absolutely impossible to have an experiencer dative in a small clause since Russian small clauses only allow AP, NP, or PP predicates, never verbal ones (of any type). Russian scitat' 'to consider' thus differs from its English counterpart in not taking an infinitival complement. Small clause complements to scitat' are therefore never really clausal in that they necessarily lack not only a complete IP projection, but any rudiments of one. No projections of V whatsoever are tolerated. Since it is just as impossible to generate a small clause with an impersonal predicate in Russian as a personal one, there can be no experiencer dative in a small clause complement to scitat'. Hence, although nuznoj in (115) is indeed an instrumental long form adjective comparable to the short form nuzna in (113), nuznym in (114) cannot be related to the impersonal predicate nuzno in (112). It must instead be more or less accidental that there is a substantivized version that can, as an NP, serve as a small clause predicate. Crucially, this NP nuznoe differs from the impersonal predicate nuzno in not taking a dative experiencer as its external argument. In sum, our comparative study of dative "subjects" of infinitives and impersonals across Slavic led to the conclusion that only the former are canonical subjects in the sense of being IP-specifiers. Although adopting a similar analysis for dative experiencers with impersonal predicates would have explained the ungrammaticality of (114) with the dative NP, the need to differentiate datives with infinitives and impersonals forced us to seek an alternative account. It turned out that the nominal analysis of this construction in Corbett (1983b) has all the required properties. 6.4.3. Secondary Predication in Polish I conclude this chapter with a more careful consideration of secondary predication phenomena in Polish. Although on the whole the facts replicate those displayed by

SECONDARY PREDICATION

277

Russian, there are several curious idiosyncracies which suggest that a more carefully articulated connection between control and case transmission is required. As in Russian, adjectives in Polish agree in case with the NP of which they are predicated. Some typical examples are as follows: (117)

a. Jan jest przystojny Jan(nom) is handsome(nom) b. Jan idzie sam Jan(nom) goes alone(nom)

(118)

a. zastatemgo chorego I-found him(acc) sick (ace) b. zostawilem go samego I-left him(acc) alone(acc)

Examples (117a) and (118a) contain an ordinary adjective predicated of a subject and an object, and (117b) and (118b) contain the semipredicate sam predicated of a subject and an object. As in Russian, they agree in similar contexts, and so I represent them in a parallel fashion.68 The non-agreeing forms are instrumental and dative, as expected, and appear in the same range of contexts as in Russian:69 (119)

a. dobrze jest [PRO bye *m*ody/mlodym] good is to-be young(nom/inst) 'it is good to be young' b. poprosi-femJana [PRO bye *szczery/szczerym/*szczerego] I-asked Jan(acc) to-be sincere(nom/inst/acc) c. kazatemJanowi [PRO bye *szcz$sliwy/szcze.sliwym/*szcze:sliwemu] I-told Jan(dat) to-be happy(nom/inst/dat) d. dobrze mi jest [PRO bye *ml'ody/m}odym] good me(dat) is to-be young(nom/inst) 'it is good for me to be young' e. wazne jest dla mnie [PRO bye *dobry/dobrym/*dobrego] important is for me(gen) to-be good(nom/inst/gen) 'it is important for me to be good'

(120)

a. dobrze jest [PRO isc *sam/samemu] good is to-go alone(nom/dat) 'it is good to go alone' b. poprosifemJana [PRO isc *sam/samemu/*samego] I-asked Jan(acc) to-go alone(nom/dat/acc) c. kazaJemJanowi [PRO isc *sam/samemu] I-told Jan(dat) to-go alone(nom/dat)' d. dobrze mi jest [PRO isc *sam/samemu] good me(dat) is to-go alone(nom/dat) 'it is good for me to go alone'

278

PARAMETERS OF SLAVIC MORPHOSYNTAX

e. wazne jest dla mnie [PRO isc *sam/samemu/*samego] important is for me(gen) to-go alone(nom/dat/gen) 'it is important for me to go alone Also as in Russian, case transmission occurs in standard contexts of obligatory control, such as (121).70 (121)

a. Jan Jan(nom) b. Jan Jan(nom)

chce [PRO bye przystojny/*przystojnym] wants to-be handsome(nom/inst) chce [PRO isc sam/*samemu] wants to-go alone(nom/dat)

As these data indicate, the governed PRO account proposed for Russian can be carried over directly to Polish. It is somewhat easier to demonstrate using Polish rather than Russian, however, that, contra Neidle (1988), the presence of surface nominative is not the correct criterion for case transmission. On the one hand, as illustrated by the examples in (122) and (123), there are nominative controllers that do not transmit their case to the predicate adjective: (122)

a. [PRO bycie *szczery/szczerym] to twqj obowiazek being sincere(nom/inst) that your(nom) obligation(nom) 'being sincere is your obligation' b. [PRO wykonanie tego przedsiewziecia *sam/samemu] completion this enterprise(gen) alone(nom/dat) to twqj obowiazek that your(nom) obligation(nom) 'completion of this enterprise on your own is your obligation'

(123)

a. Jan sadzi ze wazne jest [PRO bye *przystojny/przystojnym] (nom) thinks that important is to-be handsome(nom/inst) 'Jan thinks that it is important to be handsome' b. Jan sadzi ze wazne jest [PRO isc *sam/samemu] (nom) thinks that important is to-go alone(nom/dat) 'John thinks that it is important to go alone'

Since PRO in these examples is not properly governed, case transmission is blocked as expected, even though PRO can refer to the nominative two] 'your' in (122) or Jan in (123).71 As before, control, rather than simple construal, is the relevant factor. On the other hand, Polish exhibits clear evidence that agreement and case transmission can indeed take place from a non-nominative subject. Recall from chapter 5 that quantified DP subjects in Polish are accusative, with the NP complement to Q assigned GEN(Q). It turns out that predicate adjectives agree with this genitive NP in case, as shown in (124).72 (124)

a. wielu studentow jest mtodych/*ml'odymi many students(gen pi) is young(gen pl/inst pi) 'many students are young'

SECONDARY PREDICATION

279

b. wielu studentow idzie samych/*samym many students(gen pi) goes alone(gen pl/dat pi) 'many students go alone' In structures of obligatory control, such quantified subjects freely transmit the genitive to the predicate adjective through the intermediary of PRO: (125)

a. wielu studentow 'many students(gen pi) b. wielu studentow 'many students(gen pi)

chce [PRO bye mlodycn/*mtodymi] want to-be young(gen pl/inst pi) chce [PRO isc samych/*samym] want to-go alone(gen pl/dat pi)

I now turn to the question of how case transmission works with more deeply embedded PROs. As with Russian, and following Comrie's original "cohesion principle,'' it is generally true that the greater the syntactic distance from the controller, the less likely speakers are to accept case transmission. However, there does appear to be some variation in the ability to transmit case into embedded clauses of various types. Compare, for example, (126) and (127): (126)

a. Jan zlozyt obietnice. [PRO bye *szczery/szczerym] 'Jan made a promise to be sincere(nom/inst)' b. Jan zJozyl obietnic^ [PRO isc *sam/samemu] 'Jan made a promise to go alone(nom/dat)'

(127)

a. Jan ma zamiar [PRO 'John has an intent to b. Jan ma zamiar [PRO 'John has an intent to

bye szczery/*szczerym] be sincere(nom/inst)' isc sam/*samemu] go alone(nom/dat)'

Case transmission does not take place in (126), although it freely does in (127). These are comparable to the Russian examples discussed in section 6.2.3.4, and the same kind of analysis is assumed, where ma zamiar 'has an intent' in (127) is a complex V taking the infinitival clause as a complement, but obietnicq isc samemu 'promise to go alone' in (126) is just an NP direct object. This analysis is supported by the extraction facts in (128). (128)

a. *co Jan zlozyt obietnice. czytac 'what did John make a promise to read' b. co Jan ma zamiar czytac 'what did John have an intent to read'

A comparison of (128a) and (128b) indicates that the clause must be inside the object DP in (126) but outside it in (127), since otherwise movement of co 'what' to SpecCP (COMP) would violate Subjacency. When a filled COMP intervenes between PRO and its controller, the facts of case transmission become exceedingly complex, and Polish speakers show much more vacillation than Russian speakers do for comparable Russian examples. As expected, a filled COMP does consistently make case transmission harder, but the problem for the purely structural "governed PRO" approach is that (pragmatic or

280

PARAMETERS OF SLAVIC MORPHOSYNTAX

semantic) control still appears to be relevant. Compare, for example, (129) and (130):73 (129)

a. Jan nie wie [CP czy [IP PRO bye ?szczery/szczerym]] 'Jan doesn't know whether to be sincere(nom/inst)' b. Jan nie wie [CP czy [IP PRO isc sam/samemu]] 'Jan doesn't know whether to go alone(nom/dat)'

(130)

a. Jan nie wie [CP jak [IP PRO bye *szczery/szczerym]] 'John doesn't know how to be sincere(nom/inst)' b. Jan nie wie [CP jak [IP PRO isc *sam/samemu]] 'Jan doesn't know how to go alone(nom/dat)'

Although these sentences are presumably structurally identical, control in (129) is obligatory, whereas in (130) it is optional. Correspondingly, it turns out that case transmission can take place in (129), but not in (130). Why this should be so is unclear under a structural account. Purpose clauses provide similar examples, as in (131), which differ from the Russian ctoby purpose clauses considered earlier. (131)

a. Jan pracuje [CP zeby [IP PRO bye ?bogaty/bogatym]] 'Jan works in-order to-be rich(nom/inst)' b. Jan pracuje [CP zeby [IP PRO mieszkac sam/samemu]] 'Jan works in-order to-live alone(nom/dat)'

It therefore looks like a COMP that exclusively contains grammatical material, for example, czy 'whether' or zeby 'in-order', does not really count as a barrier to control or case transmission. It is significant, I believe, that these two phenomena once again go hand in hand, but it remains mysterious why Polish should be more liberal about case transmission than Russian is, since the structures and their control properties seem identical. After all, w/z-movement out of a czy clause is impossible: (132)

*co Jan nie wie czy robic ' what does John not know whether to do'

In Polish, movement over a filled COMP, regardless of its contents (except for subjunctive ze&y-clauses), is ungrammatical. Finally, there are some even stranger and more idiosyncratic factors at work here. Although case transmission in sentences like (133) can take place from a neuter, feminine, or plural subject, it is for some reason (for many speakers) impossible from a masculine subject, as shown in (134). (133)

a. dziecko jest za male [CP zeby [jP PRO bye child(n) is too little(n) in-order to-be inteligentne/inteligentnym] ] intelligent(nom/inst) 'the child is too little to be intelligent' b. dziecko jest za mafe [CP zeby [IP PRO isc samo/samemu]] 'the child is too little to go alone(nom/dat)'

SECONDARY PREDICATION

(134)

281

a. chlopiecjest za maJy [CP zeby [IP PRO bye boy(m) is too little(m) in-order to-be ?* inteligentny/inteligentny m] ] intelligent(nom/inst) 'the boy is too little to be intelligent' b. cMopiec jest za maiy [CP zeby [IP PRO isc ?*sam/samemu]] 'the boy is too little to go alone(nom/dat)'

I have no explanation for this puzzling phenomenon. In summary, on the basis mostly of Russian, I have argued in this chapter that a PRO is able to transmit the case of its controller to a coindexed predicate adjective only if that PRO is anaphoric. I adopted the structural "governed PRO" account of Franks and Hornstein (1992), in which a PRO is syntactically determined to be anaphoric if it is lexically governed; only such PROs are ' 'obligatorily'' controlled. The Polish facts discussed in this section suggest, however, that any purely formal account of case transmission may need to be weakened in tandem with a more finely tuned theory of control. Notes 1. Major aspects of the analysis in this chapter, as well as the specific data discussed, are based on work previously reported in Franks and Greenberg (1988), Franks (1990a), Franks and Hornstein (1992), and Greenberg and Franks (1991). Thanks are due to my coauthors Gerald Greenberg and Norbert Hornstein, as well as to AATSEEL and the editors ofSEEJ for permission to use material in Greenberg and Franks (1991). Additionally, portions of Franks (1990a), which appeared in Russian Linguistics 14: 231-254, and Franks and Hornstein (1992), which appeared in Control and Grammar, are reprinted by permission of Kluwer Academic Publishers. 2. In this and subsequent chapters I use the term "NP" rather than "DP", unless specific reference to the DP-hypothesis is required. 3. For comprehensive treatments of predicate adjectives of various types in Russian, see Gustavsson (1976) and especially Nichols (1981). Note that secondary predicate resultatives of the type in English (i) and (ii) do not exist in Slavic. (i) John wiped the floor clean (ii) Mary laughed herself silly 4. By "relevant features" I mean gender/number and case, where the latter may also reflect animacy, but specifically not person, which nouns and adjectives do not distinguish. Russian displays two kinds of predicate agreement: verbal and nominal. The former type is basically person agreement and the latter type is basically gender agreement. Although each additionally expresses number, this is not an independent parameter of agreement but rather is incorporated directly into the person or gender categories. See Franks and Greenberg (1994) for discussion. 5. The short form is generally marginal as a secondary predicate, except in conservative literary style, although acceptability improves, as Nichols (1981, 299, n. 3) observes, with "adjectives denoting internal states." Speakers consulted foundp'jan in constructions like (1) somewhat archaic, but preferred her examples on sidel bleden/vesel 'he(m nom) sat pale(m)/ happy(m)'. Regardless of these restrictions, the point remains that the short form reflects only the gender/number of the subject NP.

282

PARAMETERS OF SLAVIC MORPHOSYNTAX

6. See Nichols (1981) for further examples and discussion. Note that in Franks and Greenberg (1988) and Franks (1990a), predicative adjective agreement with non-nominative NPs was erroneously treated as ungrammatical, with the failure of agreement attributed to PRO's putative subject orientation. However, the small clause analysis (which was adopted in these articles) actually requires agreement to be possible, since PRO necessarily has anaphoric status, as to be discussed in section 6.2.3. 7. One exception is pervyj 'first', which exhibits mixed semipredicative and secondary predicate behavior for many speakers, apparently belonging to both classes. 8. In section 6.3.4 the possibility of short form semipredicatives is explored to explain their appearance in gerundive constructions. 9. Hornstein and Lightfoot (1987) actually use the representation [PRO INFL0 drunk]s for the SC, treating the phrase as a sentence with a reduced INFL. To avoid confusion and because a small clause INFL never shows tense or agreement features, I shall refer to these as APs throughout, in the spirit of Stowell (1981). 10. As convincingly argued by Sigurdsson (1991), for example, PRO in Icelandic should be analyzed as bearing case features and so, in my model, must be endowed with a case submatrix. 11. As mentioned in chapters 4 and 5, similar arguments have been made about the internal structure of NPs, which, in addition to being analyzed as maximally DPs, have also been embedded in constituents such as Kase Phrases, Gender Phrases, and Number Phrases. 12. See Franks and Greenberg (1994), however, for a possible analysis of Slavic clauses within this kind of system. 13. Along these lines, see Ouhalla (1991) for arguments against lowering and Speas (1991) for arguments that there is a universal hierarchical arrangement of functional categories. 14. The standard transliteration system sometimes obscures the true morphophonemics, so that the spelling ae stands for /aje/: 2sg delaj + es', and so on. See Townsend (1975) for discussion. 15. A reviewer suggests that a similar approach might be appropriate for English as well. 16. In addition to simplifying the representation of Russian clauses, this approach may be more true to the morphological properties of aspect. There is no unique marker of perfective or imperfective, nor are the various morphological indicators of Russian aspect uniquely prefixal or suffixal, and some verbs are biaspectual, while others may lack either one or the other aspectual form. 17. Baker proposes an idiosyncratic rule of English grammar that moves V over not; this language-specific rule is absent in Russian. Although it may eventually prove possible or even necessary to allow each of these categories to serve as separate functional heads, as is the current trend in X-bar syntax, there is no particular reason to posit such an abstract analysis for Russian for the purposes of the present chapter. 18. Of course, for unaccusative verbs the subject would originate within V and only subsequently move from there to SPEC-VP or, conceivably, directly to SPEC-IP. 19. One might alternatively extract TNS at D-structure but require it to adjoin to AGR in some way, perhaps by claiming that AGR is too weak to move down to V (or ASP, if aspect too heads its own projection) without TNS. This "weakness" of AGR (cf. Pollock, 1989) might be related to the fact that Russian is not morphologically uniform in representing person, differing in this respect from South and West Slavic; see chapter 7 for discussion. If such an approach turns out to be feasible, the analysis in (15) can be regarded as beginning immediately after this initial adjunction. 20. The idea here is basically that only the true head of I can have further articulation of features (cf. Lieber, 1992). 21. Information-based theories of syntax, such as Generalized Phrase Structure Grammar

SECONDARY PREDICATION

283

(GPSG) and its descendants, similarly distinguish "atom valued" and "category valued" features (cf. Gazdar, Klein, Pullum, and Sag, 1985). It is worth observing, I think, that this kind of representation of features also finds a parallel in contemporary phonology, where tree geometries consist of planes and dependent features, roughly akin to the broader privative features and subordinate polar features I assume. 22. As Wayles Browne (personal communication) points out, the parallelism is disrupted by the fact that Hungarian infinitives, even though they agree, require dative subjects. 23. There is a derived imperfective form byvat', but this has a-more iterative meaning and does not enter into a synchronic opposition with byf. 24. Wayles Browne (personal communication) points out that treating byt' as perfective has two potentially serious drawbacks: (i) Byt' can follow verbs such as na&inat 'to begin' andperestaf 'to cease' that otherwise only precede imperfective infinitives. (ii) Byt' has the stative semantics of an imperfective. I believe that these properties can be resolved with a proper understanding of the relationship between grammatical aspect and extensional aspect. Roughly speaking, the imperfective extensional properties of byf derive from its intensional stativity, despite the fact that it is grammatically perfective. 25. Alternatively, it is possible that features which find no overt host are abstractly supported at LF. 26. However, in the structure in (19) the governor is I rather than V. If it turns out that government by I is not sufficient to induce anaphoric status on PRO, then one might revise (19) so that I takes a VP complement and V takes an SC complement. Thus, the two structures would arise by virtue of Russian byt' having a dual subcategorization, taking either an NP/AP or SC complement. Polish bye would then lack the latter possibility. 27. This and several other examples cited in this chapter are originally due to Comrie (1974). However, since they have subsequently been cited in numerous other works, I will not regularly make specific reference to Comrie as the source. 28. While intonation and punctuation indicate that this "subject sentence" has been topicalized (moved to COMP/SPEC-CP), the point remains that case agreement is impossible. Note that clausal subjects are generally infelicitous in Russian, which presumably follows from Stowell's Case Resistance Principle, as discussed in chapter 7. 29. The masculine singular default option is far from universal, however. See Franks and Schwartz (1992) for discussion. 30. Interestingly, this is not true of Icelandic, for the reason that PRO itself bears case features in that language (cf. e.g. Sigurdsson, 1991). 31. This SUBJECT must additionally respect an accessibility requirement, which need not concern us here. 32. These facts also suggest that principles A and B apply at LF in a way that Principle C does not. 33. The case solution will probably not work for Icelandic. See Sigurdsson (1991) for an alternative approach to the distribution of PRO. 34. In the Generalized Binding Theory of Aoun (1986), wft-trace is also anaphoric, but since it is subject to principle C, it can (and must) only be A'-bound rather than A-bound. It is thus free within S but bound (by something in COMP) within S'. 35. Pro is the null subject of finite verbs in languages such as Spanish and Italian. Its general nature and specific status within Slavic will be a central concern of chapter 7. 36. Under this analysis, lexical heads do not properly govern their specifiers, otherwise the PRO in SPEC-VP position would be governed by V. Instead, government by a head X will

284

PARAMETERS OF SLAVIC MORPHOSYNTAX

have to be restricted to its complement YP and the specifier of YP. Note that although adopting this interpretation of government simplifies the analysis, it is not vital to the governed PRO account. Under the approach in which PRO raises to SPEC-IP position, one might in fact suppose that raising is forced through the interaction of control and government theory. Assume—as we in fact did in chapter 4 to account for apparent ECP violations—that V does properly govern its specifier. PRO will therefore be properly lexically governed by V in (i):

(0 [CP C [IP I [VP PRO V]]] This PRO should thus necessarily be anaphoric. Perhaps, then, as an anaphor in this position, it has no possibility of satisfying Principle A; getting this precise result depends of course on a proper formulation of the binding domain to handle anaphoric PRO. If so, it is reasonable to suppose that PRO raises to SPEC-IP in order to be pronominal, since it is able to satisfy Principle B. Conceivably, in a system with AgrP and TP, infinitival phrases are basegenerated as bare TPs headed by [-tense], so PRO could raise to spec-TP, out of the government domain of V, with the same effect: PRO in spec-TP would be protected from external lexical government whenever TP is inside CP. 37. Alternatively, clausal complements to N are necessarily CPs, so that government of the PRO is blocked. See Hornstein and Lightfoot (1987) for further discussion. 38. This is in line with the theory developed in chapter 7 that there are two aspects of the visibility effect of case, one having to do with phonological overtness and operating at PF, the other having to do with visibility of theta-roles and operating at LF. 39. As Hornstein and Lightfoot (1987) note, right dislocation is expected to destroy proper government configurations. In English, for example, rightward extraposition of a relative clause prevents deletion of the complementizer, since the residue of deletion will not be properly governed. Compare the following two sentences: (i) a man (who) John likes is in the kitchen (ii) a man is in the kitchen *(who) John likes 40. In making this distinction, I am following the spirit of Williams (1980). By this analysis I do not mean to imply that object-controlled PROs are never PF active. In Icelandic, for example, they can mediate in case transmission. See Hornstein (1990) for an analysis of Icelandic in terms of verb raising internal to the embedded clause, leading to governed (hence PF active) PRO regardless of its control situation. 41. These examples were first discussed by Postal (1976) and are cited in Chomsky (1977). In Sportiche's analysis the Q actually fails to raise to SPEC-IP with the rest of the subject and is thus actually stranded in SPEC-VP position rather than floated from off of the subject. This approach will require some modification in order to carry over to a structure like that of (33); here I merely use Q-float as a diagnostic of subjecthood. In addition, although Sportiche's account works nicely for English and the French Q-float data he discusses, it unfortunately does not carry over to Russian object control structures such as (32), raising serious questions about the appropriateness of the SC structure. 42. According to Greenberg (1983), Comrie has since conceded in personal communication that the "construction is too artificial to evoke reliable judgment." 43. As discussed in Rappaport (1986), the Russian reciprocal drug druga is not subjectoriented. It has been frequently observed that this is generally true of morphologically complex anaphors—compare Russian sebja with English himself. See Progovac (1992) or Progovac and Franks (1992) for discussion and analysis. 44. See Franks and Hornstein (1992) for details and supporting arguments. 45. See Bachman (1980) for a thorough overview of this debate. The definition of subjecthood is further clouded by a plethora of Russian terms that refer to "subject-like" entities,

SECONDARY PREDICATION

285

including podlezascee, sub"ekt, logiceskij sub"ekt, psixologiCeskij sub"ekt, and kosvennyj sub"ekt; the notion of "theme" is also frequently confused with the notion of subject. See Greenberg (1985) for discussion. 46. I focus on the infinitival rather than impersonal examples since the comparative evidence examined in section 6.4 will suggest that only infinitives have true dative subjects. 47. Some speakers allow dative resumptive pronoun subjects in some examples of this type. 48. Indeed, one also finds in South Slavic that the dative is similarly extended to possessors within NPs, which may be treated as sisters of N', as in Serbo-Croatian majka mi 'mother me(dat)' or Macedonian zenata mu 'wife him(dat)'. 49. This discussion is largely from Greenberg and Franks (1991), to which the reader is referred for sources of the citations. These and several other examples in this chapter involving gerunds were originally cited in Rappaport (1984). The examples involving nonnominative antecedents of reflexives in (55) and (56) are due to Timberlake (1979, 1980). 50. I have only been considering one way in which the agreement approach might be implemented within the GB framework, of course. It is certainly possible that further conditions on the distribution of PRO versus lexical NPs, such as those outlined in Sigurdsson (1991), might be adopted to make some version of the agreement hypothesis more viable. 51. The instrumental is correctly predicted to be always available for ordinary predicate adjectives, since these are simply APs adjoined to VP. 52. Although, as in English, gerunds can only refer prescriptively to nominative subjects, this norm is frequently violated in colloquial Russian. See Rappaport (1984) for discussion and examples of this phenomenon. Since such non-canonical gerunds are crucial to the arguments in this section, I abstract away from the effects of the dichotomy between subject and non-subject controllers throughout. 53. Note in addition that the caseless short form is more or less fine in gerund clauses, especially after buduci 'being', although this is generally not so in controlled PRO contexts, as follows: (i)

*Ivan ne mog [PRO sidet' p'jan] (nom) NEG was-able to-sit drunk(m) ' Ivan could not sit drunk'

(ii)

*Ivan ne xotel [PRO prixodit' p'jan] (nom) NEG wanted to-arrive drunk(m) ' Ivan did not want to arrive drunk'

54. See also Stowell (1982) for arguments that English gerunds lack tense features. 55. However, as Neidle (1988) points out, Svedova (1970, 637) does cite some examples of nominative long forms within gerund clauses. The presentation is somewhat confusing, however, since Svedova clearly states that the only two possibilities in gerundive constructions are the short form and the instrumental, but subsequently remarks that the long form is elsewhere colloquially acceptable. The speakers I have consulted, however, generally reject agreeing long forms after gerunds. Note that they also do not admit overt nominative subjects of gerunds, although this was possible in older varieties of Russian, so it may be that Svedova's judgments reflect relics of no longer productive processes. 56. There is one caveat: I have assumed that the empty subject of the gerund clauses in (76) is PRO, but some speakers allow nominative subjects of gerunds (as in older Russian), introducing the possibility of predicate adjective agreement. When, for these speakers, the subject of the gerund is the phonologically empty but case-marked "pro" (not PRO), (76) becomes viable.

286

PARAMETERS OF SLAVIC MORPHOSYNTAX

57. The same speakers who allowed apparent long form nominatives with gerunds, as in (63), also allowed them in these examples with participles. 58. The adjective/participle receives the appropriate case (here, nominative) through agreement, not control. The AP agrees in gender, number, and case with the noun it modifies, and this percolates down to the head A. 59. The bulk of this section is a slightly revised version of Greenberg and Franks (1991); section 6.4.3 is based on observations that originally appeared in Franks (1983). 60. The particular Slavic languages examined in this section were selected on the basis of their potential to display the relevant constructions. Thus, for example, Bulgarian and Macedonian are not included because they lack infinitives, and Czech is not mentioned because it patterns like Slovak. 61. The Polish examples in (82)-(85) are drawn from Comrie (1974). Comrie refers to jeden 'alone' as if it were on a par with sum in Polish, but Oscar Swan (personal communication) indicates otherwise. According to Swan, (82) and (83) with jeden andjednego are not acceptable. 62. The Slovak examples are from Kacala (1971) and cited in Comrie (1974). 63. The "neuter" form samo suggests that adverbialization may have actually taken place here; the number/gender neutralization of samemu indicates the same for colloquial Polish. Perhaps, then, the second nominative option is available only in South Slavic. Here, too, it is extremely restricted, typically failing to apply beyond the degree of embedding in (88) and (89). 64. Examples (101M102) are from Zencuk et al. (1986, 243), and (103)-(104) are from Svetlik (1966, 178). 65. The number of these impersonal predicates in Slovenian seems to be smaller than in Russian, Polish, and even Slovak, but they are perfectly acceptable in the contemporary language. The Slovak examples in (110) are from Svetlik (1966). 66. Some version of the latter approach is presumably correct, as suggested by the nuzno facts considered immediately below and as argued for in depth by Schoorlemmer (1991). 67. Similar conclusions were reached on the basis of the ATB phenomena considered in chapter 3; see also Franks (1992, 1993). 68. Unlike Russian, these are the only forms possible, the default instrumental generally being used in Polish only when agreement fails. 69. Note that in (120c) and (120d) the dative is the default case, not the transmitted one. This is indicated by the obligatory nature of the instrumental in the comparable examples in (119). Also, some speakers reject (120b) and (121b) altogether. 70. Recall that, in Polish, predicate NPs differ from APs in that they are obligatorily instrumental. Therefore, Poles often accept or even prefer instrumental predicate adjectives, construing them to be periphrastic for an entire NP. Because of this, my focus is on whether or not agreement is acceptable at all. 71. The dative on samemu in (122b) must be the frozen adverbial one, since it appears inside an NP rather than IP. This claim is supported by the fact that parallel constructions are impossible in Russian, since Russian lacks the phenomenon of adverbial dative. Indeed, it turns out that examples like (122b) are impossible in Russian, supporting my claim about Polish. Unfortunately, although sentences like (122a) should be acceptable in Russian as well as in Polish, I have been unable to construct any relevant examples. 72. Although some of my informants showed vacillation in judging (124a) and (125a), transmission in (124b) and (125b) proves that any problem in the former sentences lies in the agreement rule itself. 73. A few speakers find agreement in (129a) and (131a) slightly odd.

7 Null Subject Phenomena

This chapter surveys the range of problems associated with the analysis of null subjects in the Slavic languages.1 In it are explored some of the ramifications of recent proposals within GB theory about the factors motivating case assignment to subject position. Section 7.1 introduces the issue of null subjects in general and presents a range of related Slavic problems. Section 7.2 treats null theta-marked subjects, and section 7.3 treats null expletive subjects. It is argued that case plays a crucial role in both morphological visibility at the level of Phonetic Form and semantic visibility at the level of Logical Form. The issue of which null and overt subjects are assigned case is addressed, and it is concluded that whereas all overt nominals require case, only those null nominals that must be semantically visible do. The implications of this model are then applied to the Slavic languages in section 7.4, where the problem of whether null expletives receive case is examined in detail. The existence of certain marginally overt expletive subjects in Slavic is explained in terms of the LF-visibility requirement. It is proposed that precisely those expletive positions that are filled by arguments requiring case at LF are case-marked, although the data are obscured by several intersecting parametric differences among the languages.

7.1. The "Pro-Drop" Parameter A central concern of syntactic theory is the formal status of unexpressed elements, otherwise known as empty categories. An empty category is a syntactic unit that has no phonetic feature matrix associated with it—a silent or "null" element. The study of empty categories is one of the most productive areas of ongoing research in GB theory. Although empty categories are phonologically null, their existence is nonetheless demanded by principles of Universal Grammar. Practitioners of GB maintain that this framework can account for linguistic phenomena more insightfully than can 287

288

PARAMETERS OF SLAVIC MORPHOSYNTAX

comparable frameworks that avoid positing inaudible elements. More importantly, the fact that speakers acquire grammatical systems countenancing empty categories argues not only for the cognitive reality of these categories but also for the GB model as a whole. That is, the absence of direct evidence for empty categories implies that speakers develop particular grammars utilizing them because UG leaves no alternative. Empty categories exist where they must, and nothing more need be said (provided the rest of the theory is properly formulated).2 Because they could not plausibly be deduced from the primary linguistic data, empty categories offer crucial evidence for the deductive structure of the theory. They thus serve as a unique window into the workings of UG. Empty categories are far from contentless, and at least four types have been proposed to play a major role in several subtheories of GB. These are NP-trace, whtrace, pro, and PRO, each with its own distinct syntactic behavior and features, which it will be recalled are standardly given as in (1). (1) wh-trace

=

[-pronominal, -anaphoric]

=

NP-trace pro PRO

= = =

[-pronominal, + anaphoric] [ +pronominal, -anaphoric] [ +pronominal, + anaphoric]

= = =

R-expression anaphor pronominal "pronominal anaphor"

The traces are created by Move a; they occupy the position vacated by the moved element, with which they are coindexed. The remaining two empty categories, pro and PRO, differ from traces in that they are thematically independent null pronouns. Since they are not generated through movement, they need not be coindexed with another element in the sentence, although they may be. PRO, as discussed extensively in the previous chapter, is typically the subject of an infinitival or gerundive clause. In this chapter we concentrate on so-called small pro. Pro is just like an overt pronoun except that it lacks phonetic content. It occurs as the unemphatic null subject of a finite clause, as in Polish (2a) and Bulgarian (2b), structurally parallel to (3a, b) with the overt pronouns ja T and toj 'he'. (2)

a. pro nie rozumiem NEG understand(lsg) 'I don't understand' b. pro ste

uspee

AUX(3sg) succeed(Ssg) 'he will succeed' (3) a. ja nie rozumiem I NEG understand(lsg) b. toj ste uspee he AUX(3sg) succeed(3sg) One general term for the omission of otherwise expressible subjects, stemming from the important transformational study of Perlmutter (1971) and developed within GB in Rizzi (1982), is "pro-drop." The theoretical problem posed by the phenomenon of pro-drop is that languages

NULL SUBJECT PHENOMENA

289

differ in whether or not they permit pro subjects: English does not, but many other languages do, including at least some Slavic ones. Although it is not entirely clear what makes it possible for a given language to have pro, it is generally agreed that richness of verbal agreement morphology is probably involved. This corresponds to the traditional, intuitively satisfying generalization that subjects can "drop" when verbal inflection is rich enough to identify at least some features of the missing subject. There is a large class of languages, on the other hand, including Japanese and Chinese, that drop subjects despite the complete absence of any subject-verb agreement morphology whatsoever. This unexpected similarity between rich and impoverished agreement systems poses a challenge for purely "feature identification" based approaches to the phenomenon of empty subjects. One might well ask why syntactic theory posits empty categories at all, and why there is a need to assume four different kinds of silent elements. This clearly raises the issue of how a child learning a language could project the existence and nature of empty categories. The usual response to this question is that the theory of empty catgories follows automatically from basic GB principles, particularly the Projection Principle and the Theta-Criterion. In short, according to GB's Extended Projection Principle (cf. Chomsky 1981a, 1982), all clauses must contain subjects at all syntactic levels of representation. Crucially, it is claimed that this state of affairs holds regardless of the particular nature of the subject, so that subject position is argued to be present whether it has phonological or semantic content or not. Thus, a subject may be phonologically null or overt, and it may be a meaningful argument of the verb or be semantically vacuous.

7.1.1. Types of Null Subjects The pros in (2) correspond to theta-marked pronouns. Since the Extended Projection Principle forces the postulation of a subject position even when no theta-role is assigned, however, UG also must admit null expletive pronouns. An expletive is a semantically null (or "pleonastic") element in a non-theta-marked position, such as the null pro in Serbo-Croatian (4) or overt it in its English translation. (4)

pro Sini mi se da Jovan nije dosao '(it) seems to me that John didn't come'

Note that since English lacks null expletives (as well as null theta-marked pro subjects), it requires the subject in this example to be phonetically realized. In Serbo-Croatian, on the other hand, null expletives generally lack overt counterparts (but cf. section 7.4). Expletive pro contrasts to thematic pro in this regard, since the latter alternates with an overt NP. As we shall see in this chapter, the two also differ in distribution—null expletives occur in all the Slavic languages, regardless of the status of theta-marked pro, whereas the factors licensing null thematic pro seem to differ among the languages. Although the GB requirement that all clauses consist of subjects and predicates forces us to posit phonologically null subjects in any syntactic context where an overt subject is absent, there is an entire range of constructions that fit this description.3 In ihis chapter it is argued that the simplistic division into [± Pro-Drop] is

290

PARAMETERS OF SLAVIC MORPHOSYNTAX

clearly inadequate and that null subject phenomena must be decomposed into at least three distinct parameters, the NOM-Drop Parameter ([ ± NDP]), which I will eventually recast as the Overt Subject Parameter ([±OSP]), the Free-Inversion Parameter ([ ±FIP]) and the Overt Expletive Subject Parameter ([ ± OESP]), in addition to null subjects resulting from a general rule of discourse ellipsis. We can then classify the various Slavic languages according to this typology. One important result is that, whereas all Slavic languages allow empty subjects when the subject bears no semantic role (i.e. are [ — OESP]), only West and South Slavic languages do when some role is assigned to that position (i.e. are [ —OSP]). 7.1.2. Parametric Approaches In this section I briefly survey approaches to null subject phenomena within GB, based primarily on comparative studies of Romance languages. I then sketch the typology of empty subjects found in Slavic, attempting to understand the Slavic situation in GB terms. It has long been recognized that the grammaticality of empty subject pronouns varies across languages. Early GB studies of this problem, especially Jaeggli (1982) and Rizzi (1982), relying on observations due to Perlmutter (1971), argued for a uniform Null Subject (or "Pro-Drop") Parameter that would divide languages into two groups: (1) those, like English and French, that normally express non-emphatic subject pronouns, and (2) those, like Spanish and Italian, that do not. This property was initially correlated with a host of other missing-subject phenomena ultimately having to do with whether or not the language had a mechanism for circumventing the ECP prohibition against empty categories lacking local lexical governors. Descriptively, the inventory of these phenomena included missing subject pronouns, free subject postposing, apparent violation of the that-trace filter, subject extraction out of wh-islands, and null resumptive pronouns.4 Consider the fact that Spanish and Italian, for example, in addition to allowing empty pronominal subjects, also permit long-distance extraction of subjects, as in w/z-movement. Without a local governor, the w/z-trace in subject position should violate the ECP, since, unlike the object of a verb, it is not governed by a lexical category. The proposed Pro-Drop Parameter would therefore have to be formulated along the following lines: in Spanish and Italian the subject is somehow able to be lexically governed, whereas in English and French it is not. The initial mechanism for accomplishing this was to allow the nominativeassigning "inflection" element (which, as with the Affix Hopping rule of Chomsky, 1957, eventually adjoins onto the verb) to move in the syntax as well as the phonology. Since case assignment takes place in the syntax, this meant that the subject NP would move into (or adjoin to) the VP in order to be assigned nominative case. This "inversion" operation also caused subject position to be lexically governed by the verb, however. The subject could then subsequently vacate that position freely, its trace being lexically governed. An alternative solution was to stipulate that in certain languages subjects are lexically governed in situ. Under extended X-bar syntax, the subject is governed by I, since it occupies the IP-specifier position. The matter of whether government of subject position by I is sufficient to satisfy the ECP can then be made to depend upon whether or not I counts as lexical in a particular

NULL SUBJECT PHENOMENA

291

language. Under this approach, the choice between I being lexical (and consequently a proper governor) or not is regarded as a parameter of core grammar. Both systems obtained the desired correlation between null pronominal subjects of finite clauses and apparent lack of ECP effects. It was later realized, however, that the range of possibilities is more complex than this and that at least two distinct parameters are involved. In particular, Safir (1985) maintains that there exist two independent parameters, denning four language types. His NOM-Drop Parameter (NDP) allows optional phonetic realization of nominative subjects, and his Free-Inversion Parameter (FIP) allows optional movement of subjects to inside the VP, with consequent apparent failure to show ECP violations. These parameters are conceptually close to the two approaches just described, except that for Safir the two pro-drop phenomena are divorced from each other. Safir (1985, 242) uses his system to characterize different Romance languages as follows: (5) Language

NDP

FIP

Standard Italian

+

+

French

—

—

Trentino Italian

—

+

Modenese Italian

—

+

Standard Portuguese

+

—

Although in this chapter I will only be concerned with the NOM-Drop Parameter and related effects in Slavic, it is worth noting that Slavic in general corroborates Safir's partitioning of the facts. For example, Russian, Ukrainian, Bulgarian, and Polish exhibit ECP effects, indicating that they are [ — FIP]. This can be seen in Polish (6) and Russian (7), where extraction from the subject position of an embedded finite clause is markedly worse than from an object postion.5 (6) a.

co Maria chce [CP zeby [IP Janek kupil e]] 'what does Maria want Janek to buy' b. *kto Maria chce [CP zeby [jp e kupit chleb]] 'who does Maria want to buy bread'

(7) a. b.

cto Mas'a xofiet [CP Ctoby [JP Ivan kupil e]] 'what does Masha want Ivan to buy' %kto Masa xocet [CP dtoby [IP e kupil xleb]] 'who does Masha want to buy bread'

Note that Russian speakers often find examples such as (7b) colloquially acceptable. It thus appears that although the literary language shows ECP effects and is thus presumably [ — FTP], some spoken styles of Russian are best treated as [ + FIP]. Serbo-Croatian, on the other hand, freely allows subject extraction and is therefore also [ + FIP], as indicated by (8). (8) a. ko Mite [CP da [IP vam e kupi knjigu]] 'who do you want to buy you a book'

292

PARAMETERS OF SLAVIC MORPHOSYNTAX

b. sto zelite [CP da [IP vam Jovan kupi e ]] 'what do you want Jovan to buy you' However, as shown below, whereas Polish, Serbo-Croatian, and Bulgarian normally omit subject pronouns and thus are [ + NDP], Ukrainian and Russian retain them and, according to Safir's system, should be [-NDP]. This results in the following schema: (9) Language

NDP JTP

Serbo-Croatian

+

+

Ukrainian

-

-

Standard Russian

—

—

Colloquial Russian

—

+

Polish

+

Closer investigation of null subject phenomena reveals that even this bipartite approach is inadequate to handle the range of null subject possibilities that in fact exist. Consider for example how Polish (lOa) and Russian (lOb) behave with respect to the phonetic realization of pronominal theta-marked subject NPs. Polish, as pointed out above, typically omits non-emphatic pronominal subjects, but Russian retains them: (10)

a. ja/*e I(nom) b. ja/e I(nom)

poexal v gorod went(m) to town pojechalem do miasta went(lsgm) to town

Russian (lOa) is of course grammatical without the subject pronoun as an elliptical construction. As discussed in section 7.2.2, however, such discourse ellipsis applies with equal force to virtually any element, regardless of syntactic position, suggesting that this is a PF-deletion process. Phrases deleted under ellipsis are contextually recoverable and need be neither subjects nor pronouns (although of course they may be and often are). In Polish, however, the subject is grammatically privileged in that its expression, if pronominal and unemphatic, is marked. The point is thus that Russian differs fundamentally from Polish in the syntactic status of empty thetamarked subjects.6 Non-theta-marked subjects, on the other hand, behave in more or less the same way in all the Slavic languages in the sense that they all employ null expletives.7 The following are some typical Russian examples: (11)

a. e temneet8 '(it) is getting dark' b. e bylo ocen' prijatno u vas '(it) was very nice at your place' c. mne e kazetsja, cto my zabludilis' '(it) seems to me that we are lost'

NULL SUBJECT PHENOMENA

293

d. e ne stoit daze i govorit' ob etom '(it) isn't even worth talking about' e. v lesu e okazalos' teplo '(it) turned out to be warm in the woods' f. bez detej v dome e kazalos' pusto 'without the children (it) seemed empty in the house' g. ob etom nigde e ne napisano '(it) isn't written about that anywhere' Polish, although otherwise different from Russian in being [ + NDP], behaves similarly with regard to all the same types of non-thematic subjects: (12) a. e pachnie '(it) smells' b. e przesta-fo grzmiec (it) has stopped thundering' c. w parku e jest przyjemnie 'in the park (it) is nice' d. e zdawaio si$ jej, ze . . . '(it) seemed to her that . . . ' The same is true of the other Slavic languages, whether clearly [ + NDP], like Polish, or probably not pro-drop, like Russian. Ukrainian resembles its East Slavic neighbor Russian in retaining thematic pronominal subjects, whereas the South and West Slavic languages resemble Polish. They all nonetheless can drop expletive pronouns, as indicated by the following examples, representing Ukrainian (13), Serbo-Croatian (14), Czech (15), and Bulgarian (16): (13) a. e ne vadylo b s'ohodni pohuljaty '(it) would not harm to go for a walk today' b. e smerkaje '(it) is growing dark' c. nadvori e xolodno '(it) is cold outside' (14) a. e zima je '(it) is cold' b. e treba da ... '(it) is necessary that . . . ' c. e cini mi se da . . . '(it) seems to me that . . . " (15)

a. e pr§i '(it) is raining' b. e svitalo '(it) dawned'

294

PARAMETERS OF SLAVIC MORPHOSYNTAX

c. e je tfeba, ze . . . '(it) is necessary that . . . ' (16)

a. e vali '(it) is raining' b. e mrakva se '(it) is getting dark' c. e izglezda, ce . . . '(it) seems that . . . '

There are arguably many other types of null expletive constructions, such as verbs that occur with accusative or oblique arguments but no nominative subject, as in Czech (17), or the "dispositional se" construction, as in Bulgarian (18).9 (17)

a. e pichalo ho v boku pricked him(acc) in side 'he had a twinge in his side' b. e ovanulo ho chladem seized him(acc) cold(inst) 'he was seized by cold' c. e Petrem trhlo (inst) shook 'Peter started'

(18) a. e ne mi se pie NEG me(dat) REFL drinks 'I don't feel like drinking' b. e bori mi se wrestles me(dat) REFL 'I feel like wrestling' Whatever the analysis of these and similar structures, the point is, as discussed in chapter 6, that the range of impersonal constructions is more or less constant across the entire Slavic family. What is presumably significant about these predicates is that no theta-role is assigned to subject position. This for some reason allows a deviation from our general conclusion that Russian and Ukrainian are [-NDP]. In other more "standard" [ —NDP] languages, such as English and French, even subjects lacking a theta-role must be realized phonetically (as the pleonastic pronouns it, il). Since all the Slavic languages can be characterized as having null expletives, we must invoke a new parameter to differentiate languages along these lines. This constitutes yet a third aspect of the now far from monolithic Pro-Drop Parameter. Since this parameter licenses null expletive subject pronouns independently of whether theta-marked ones can be omitted, I will refer to it descriptively as the Overt Expletive Subject Parameter ([±OESP]). Early work tended to view the omission of expletive subjects as intimately connected with the general possibility of having null subjects, however formulated.10 It

NULL SUBJECT PHENOMENA

295

is indeed true that any language that allows null thematic subjects will also allow null non-thematic (i.e. expletive) subjects. It is also generally believed, however, that pro-drop languages require expletive subjects to be null. While the evidence presented in section 7.4 will demonstrate that this is in fact not so, it is worth considering some implications of this claim. It may be that the general avoidance of expletives in otherwise pro-drop languages is simply the result of a pragmatic principle of economy, along the general lines of Chomsky (1991). Such a principle would eschew unmotivated derivations and representations: if expletives never need to be overt in a pro-drop language, then they won't be. This aspect of the principle of economy was originally characterized in Chomsky (1981a) as the Avoid Pronoun Principle. If a pronoun is only used where the grammar offers no viable alternative and, unlike ordinary pronouns, an expletive can never be emphatic, then in a language that does not require them they should simply never occur. The converse of this argument has been applied to conclude that any language in which overt expletive pronouns occur does not allow null subject pronouns, otherwise there would be no call for these dummy elements. In fact, Hyams (1986,1987) proposed that it is precisely the presence of lexical expletives in a language that provides the crucial input in allowing the child to deduce that his/her language is not pro-drop. She adduces a host of arguments, based on comparative research into first language acquisition, suggesting that having null subjects is the unmarked (or initial) setting of her version of the Pro-Drop Parameter (the "AG/PRO Parameter"). Hyams therefore reasoned that some salient non-PD property must be available to the child to trigger the resetting of this parameter to non-PD, and developed the idea that lexical expletives can be made to serve this purpose. The problem with this proposal is that, since she was working under the assumption that there is a single parameter that accounts for all the diverse phenomena originally characterized as pro-drop, Hyams failed to consider the fact that languages need not have the same values for [ ± NDP] and [ ± OESP], let alone FIP. Furthermore, there is no absolute predictive value between null expletive and null thematic subjects. The observations about Russian and Ukrainian above demonstrate that if a language happens to have null expletives, then no conclusion about the status of null theta-marked subjects can be made. In section 7.4 it will be shown that the (optional) presence of overt expletives in a language also does not mean that the language is not pro-drop. About the only definitive correlation is that if expletive subjects are obligatorily overt, theta-marked subjects will also be obligatory. Hyams (1989, 234, n. 17) mentions some of these difficulties, citing Modern Hebrew as a null subject language that optionally employs the overt expletive ze in extraposition constructions, and suggesting that "the presence of a lexical expletive is most probably related to the fact that Hebrew is in some sense not a 'pure' NS language,'' since it only allows phonologically null thematic subjects if they are first or second person and only in the past or future tense (i.e. precisely where agreement is explicitly marked). Our discussion of related phenomena in Slavic, and especially colloquial Czech, will show that the use of Hebrew ze 'it' should be treated independently of the language's restricted pro-drop capabilities. On the flip side, Hyams (1989, 235, n. 21) also notes that there are languages like German that are "nonthematic pro-drop" in the sense that "the null subject phenomenon is restricted to

296

PARAMETERS OF SLAVIC MORPHOSYNTAX

expletives," much like the Russian and Ukrainian situation. She concludes that both types of language are problematic for her analysis of overt expletives as triggering a developmental shift from pro-drop to non-pro-drop. In the case of the German type, Hyams states that her account implies that "German children might take longer to restructure their grammars given that the triggering data are less robust."11 The Hebrew type is far more damning, since reliance on expletives would lead children to the wrong solution. As we shall see below, this problem is quite widespread, since there are also limited cases of overt expletives in Irish, Galician Portuguese, and Slavic. In short, the presence of overt expletives in a language cannot be the trigger that allows children to determine that their language is not pro-drop; instead, an analysis that incorporates separate parameters for null thematic and non-thematic subjects is called for. There has been much recent inquiry into the theoretical status and empirical properties of null elements in syntax. Jaeggli and Safir (1989a) present an interesting overview of the major issues; the other articles in Jaeggli and Safir (1989b) are also representative of the range of problems and analyses in the GB literature. They cite the traditional wisdom that optionality of subject pronouns is somehow related to ' 'richness'' of verbal inflection, but point out that any correlation between these two properties must be far from direct. For example, Jaeggli and Safir (1989a, 27-29) compare conjugations in Spanish, German, and Irish to observe that even though German AGR(eement) appears to be richer than that of Irish, and almost as rich as Spanish, it differs from the other two languages in not allowing phonologically null thematic subjects. Japanese and Chinese, on the other hand, although both pro-drop, show no subject-verb agreement inflection at all. Clearly, either "richness of inflection" cannot be taken simply to mean "number of distinct person/number/gender endings" or some other criterion is in fact relevant to pro-drop; probably both conclusions are correct. One standard method of attacking the complexity of pro-drop is to break down the factors that allow pro into two distinct components, "licensing" and "identification. '' Licensing is a formal requirement which states that empty categories can only appear in specifically sanctioned positions. Identification is a substantive requirement which states that the content of empty categories must in some specific way be recoverable. With respect to pro subjects, these two requirements have been characterized by Rizzi (1986) roughly as follows:12 (19)

a. LICENSING: pro is licensed under head government (typically by INFL/ AGR) b. IDENTIFICATION: the content of pro is recoverable through rich agreement specification (typically with INFL/AGR)

The usual interpretation of (19b) is that the licensing AGR head provides pro with grammatical specification of the necessary pronominal features under coindexation. Jaeggli and Safir (1989a) propose two important revisions to Rizzi's model. First, they argue that although all empty categories must meet the licensing requirement, only contentful empty categories need to meet the identification requirement. The result is that in a context where null subjects are licensed but not identified, null

NULL SUBJECT PHENOMENA

297

expletives, but crucially not null thematic pronouns, should be allowed. Such an approach is appropriate for German and perhaps the East Slavic languages as well, although even this division of labor seems inadequate, in that there is a definite sense in which the East Slavic languages are "more pro-drop" than is German. The other significant contribution of Jaeggli and Safir (1989a) is the proposal that the notions of "pro-drop" and "identification" need to be broadened. They note that an obvious problem with (19b) is its failure to extend to apparent null subject languages with defective or completely lacking AGR-type subject—verb agreement. Instead, Jaeggli and Safir (1989a, 29-31) argue that the null subject parameter should be reformulated in terms of "morphological uniformity," as stated in (20). (20) Null subjects are permitted in all and only languages with morphologically uniform inflectional paradigms. For a paradigm to be "morphologically uniform" it must have either only derived inflectional forms or underived ones, where by "derived" Jaeggli and Safir mean formed from a stem (or root) via affixation and by "underived" they mean morphologically identical to the stem (or root). This approach allows a consistent treatment of the Chinese/Japanese and Spanish/German language types as morphologically uniform, although in opposite ways, and opposes both types to English and French, in which verbs are only inflected in some forms. In addition to this modification, the concept of "identification" needs some adjustment. After all, there are some morphologically uniform languages, like Chinese, that do not exhibit identification by local agreement yet are fully pro-drop and others, like German, that appear to be morphologically uniform yet are not fully pro-drop. Here Jaeggli and Safir suggest on the one hand that Chinese countenances a process of ' 'non-local identification by agreement," essentially adopting the null topic and controlled pro analysis of Huang (1984), and on the other that in German identification is blocked because in verb second languages tense is in COMP rather than INFL. Their analysis is extremely tentative, with many open questions remaining. Be that is it may, this kind of approach to pro-drop has the appeal of more universal coverage of pro-drop phenomena, uniting the Chinese and Italian types through the concept of morphological uniformity and distinguishing null expletives from null thematic subjects through the separation of identification from licensing. 7.1.3. Summary of Slavic Facts Let us now briefly consider null subject phenomena in the Slavic languages in light of these ideas. How direct is the correlation between pro-drop and agreement in Slavic? To what extent is the verbal morphology of the various Slavic languages morphologically uniform? A quick survey of Slavic reveals that subject pronouns are standardly omitted in exactly those languages that have full-fledged copular agreement systems. That is, whereas the copula and past tense auxiliary typically show no person/number agreement morphology in East Slavic, they do so in South and West Slavic. Compare East Slavic Russian (21a), (22a), and (23a) with South Slavic Serbo-Croatian (21b), (22b), and (23b) and West Slavic Polish (21c), (22c), and (23c):

298

PARAMETERS OF SLAVIC MORPHOSYNTAX

(21) a. my ditali knigu 'we read the book' b. ecitali smo knjigu AUX(lpl) c. e czytaiy-smy ksiazk$ AUX(lpl) (22)

a. ja student/ustal 'I am a student/tired' b. e student/umoran sam be(lsg) c. e jest-em studentem/zmeczony be(lsg)

(23)

a. ty byla studentka/ustala 'you were a (female) student/tired' b. e bila si studentica/umorna AUX(2sg) c. e byla-s studentka/zm^czona AUX(2sg)

The past tense /-participles in (21) only show adjectival (gender/number) agreement, but in South Slavic (21b) and West Slavic (21c) this is supported by verbal (person/ number) agreement in the form of an auxiliary clitic. No such clitic is possible in (21a), since, as indicated by (22a) and (23a), Russian lacks a morphological copula auxiliary, which can be regarded as the independent realization of AGR. In South and West Slavic, however, AGR can always be lexicalized, regardless of whether there is an available host verb or not. In (21)-(23) I have lumped the present copula and past auxiliary together because in broad terms they pattern in the same way, in both form and distribution. One minor difference is that in West Slavic (and Macedonian and sometimes Bulgarian, as opposed to the other South Slavic languages) the person auxiliary differs from the copula in that it is a morphological zero. As Ron Feldstein (personal communication) observes, such a situation is in fact functionally optimal, number (but not person) being redundantly specified on the /-participle verb form with which the AGR auxiliary would co-ocur. The question therefore remains of whether and how the copula and auxiliary should be distinguished. Although the parallelism, as well as their general correlation with null theta-marked subjects, is intriguing, it is clear that the copula is essentially a bearer of tense, whereas the auxiliary is essentially a bearer of agreement. Hence, I suspect that formally we can regard the copula as a realization of TNS + AGR and the auxiliary as a pure realization of AGR, and variation among the languages depends on how these are instantiated. It thus seems that the presence of null theta-marked subjects roughly correlates with richness of inflection in Slavic, in that the person/number features of the subject can be realized on the copula. If AGR is sufficiently rich to serve as an independent word, then it surfaces as a "subject" clitic, as discussed in section 7.2.3. This clitic

NULL SUBJECT PHENOMENA

299

is coindexed with the pro in subject position, with which it shares pronominal features. Since such an AGR also governs this pro, it binds it. It is presumably the fact that the features of the subject are accessible on AGR and can thus be identified that allows the subject to escape phonetic realization. Here it becomes clear why, as Safir (1985, 185) observes, this option only pertains to non-emphatic pronominal subjects: the features of a full NP extend beyond the pronominal ones found in AGR (person, number, gender, and case), since the full NP has semantic (versus purely grammatical) content. A rich AGR, however, is sufficient to identify the pronominal features of a null subject. Note that whether or not AGR is "rich enough" to identify the subject pronoun depends on its features rather than on their surface realization. Thus, in Polish there is no segmentable third person morpheme parallel to the other clitic forms -m (Isg), -I (2sg), -smy (Ipl), -scie (2pl), no third person singular present tense AGR ending, and the third person copula is often omitted (e.g. in presentational sentences with to), but this does not force the subject pronoun to be present. The subject pronoun is no more required in by la studentkq '(she) was a student' than it is in second person (23c). I therefore follow traditional analyses in positing a zero ending to reflect third person AGR, which is in fact unique in its opposition to those of first and second persons. Similarly, Russian conjugated (present and future) verb forms show person, while past tense ones do not, but this does not necessarily imply that subject pronouns should be more easily dropped in the present/future than the past.13 Instead, the parametric approach implies that Russian is a non-pro-drop language and Polish is a pro-drop language throughout, in the sense that thematic pronominal subjects are regularly dropped only in the latter. Before considering the theoretical implications of treating Russian as non-prodrop, let us review the typological reasons for making such a claim. Recall that in a language like English both theta-marked and non-theta-marked subjects must be lexically filled, as illustrated in (24) and (25), respectively. (24)

a. John/he/*e bought a newspaper b. I/*e introduced John to Mary

(25)

a. b. c. d.

it/*e it/*e it/*e it/*e

is getting dark was very nice at your place isn't even worth [PRO talking about that] turned out [PRO to be warm in the woods]

In (25) no theta-role is assigned to subject position, hence it is occupied by the nonreferential expletive element it. This pleonastic pronoun is necessary to fill subject position but plays no meaningful role in the interpretation of the sentence. Languages like English that require overt subjects for all finite verbs can be described as "canonical" non-null subject languages. In contrast, "canonical" null subject languages have both phonologically null theta-marked subjects when these are non-emphatic and pronominal, as well as null non-theta-marked subjects. Some more examples are provided in Serbo-Croatian (26) and Polish (27).

300

(26)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a. Jovan/on/e cita knjigu 'Jovan/he(nom) is reading a book' b. e smrkava se '(it) is getting dark' c. e bilo je prijatno u parku '(it) was nice in the park' d. e zanimljivo je [PRO citati ovu knjigu] '(it) is interesting to read this book' e. e ne vrijedi [PRO ici kuci] '(it) is not worth (it) to go home'

(27)

a. Jan/on/e czyta ksiazke/ 'Jan/he(nom) is reading a book' b. e grzmi '(it) is thundering' c. e bedzie miJo w parku '(it) will be nice in the park' d. e byto bardzo przyjemnie [PRO pojsc do domu] '(it) was very nice to go home' e. e nie da sie [PRO pojsc do parku] '(there) is no way to go to the park'

In the (a) examples, the subject bears a thematic role, but need not be expressed unless it has non-pronominal semantic content or receives special emphasis. In the remaining examples of (26) and (27), on the other hand, subject position cannot be lexically filled and is therefore in these items occupied by a null expletive. As discussed above, it has recently been recognized that in addition to the English and Serbo-Croatian types there are also "mixed" null subject languages that exhibit neither canonical pattern.14 Russian is a prime example of a language that occupies the middle ground between these two extremes, since it is more or less like English in requiring theta-marked subjects to be lexically filled in stylistically neutral contexts,15 but resembles Serbo-Croatian and Polish in its use of null expletives. The English sentences in (24)-(25) thus have the following forms in Russian: (28)

a. Ivan/on/*e kupil gazetu 'Ivan/he(nom) bought a newspaper' b. ja/*e predstavil Ivana Mase 'l(nom) introduced Ivan to Masha'

(29)

a. e temneet '(it) is getting dark' b. e bylo ocen' prijatno u vas '(it) was very pleasant at your place'

NULL SUBJECT PHENOMENA

301

c. e ne stoit daze i govorit' ob etom '(it) is not worth it even to speak about that' d. v lesu e okazalos' teplo '(it) turned out warm in the woods' Russian thus differs fundamentally from canonical null subject languages in that overt pronominal subjects are the norm in Russian (and East Slavic in general) but not in Serbo-Croatian (i.e. South Slavic) or Polish (i.e. West Slavic). These facts support Jaeggli and Safir's conclusion that whatever allows null expletive subjects in a language does not necessarily extend to theta-marked subjects. What options are available in their system for the analysis of pro-drop in Russian? Recall that for Jaeggli and Safir null subjects are licensed if inflection is morphologically uniform, as stated in (20), and that this licensing is augmented by an identification requirement for theta-marked pro but not for expletive pro. Notice that this results in a three-way contrast: null subjects can be (1) neither licensed nor identified (no pro-drop), (2) licensed but not identified (expletive pro-drop only), or (3) both licensed and identified (full pro-drop). At first glance it would appear that Russian pro is licensed but not identified, since the facts cited so far show that Russian has expletive pro-drop only. The problem with this conclusion is that for Jaeggli and Safir licensing requires morphological uniformity, but the examples in (21)-(23) suggest that Russian differs from the South and West Slavic languages in precisely this regard. That is, it would appear that the relevant factor in distinguishing East Slavic from South and West Slavic is the former's defective verbal morphology in that person is not consistently marked. It is not absolutely clear that this fact renders Russian not morphologically uniform, since Jaeggli and Safir (1989a, 31) discuss the superficially similar case of Hebrew and seem at the same time to be arguing that it is morphologically uniform. Recall that in Hebrew only first and second person pronouns are dropped, and only in the past and future tenses, although null expletives are always allowed. They claim that this is because, although null subjects are licensed in Hebrew, they are only identified when agreement morphology is rich enough. On the one hand Russian exhibits a comparable subuniformity, with rich agreement only in the non-past, but on the other hand even when a thematic subject is fully identified in Russian, this does not lead to its unmarked omission. This suggests that Russian really is not morphologically uniform, hence not prodrop in the sense that null subjects are not licensed. This leaves open the question of why Russian should have null expletives, since in the Jaeggli and Safir system the same parameter licenses both thematic and non-thematic pro. It is also not clear whether they regard identification without licensing as a viable option, although some of the facts from Chvany (1973) might lead one to imagine that this is exactly what goes on in Russian in non-past conjugations. She cites several contexts in which theta-marked subjects can be stylistically omitted when contextually determined, and in personal communication points out that in certain inversion constructions there is a contrast in subject pronoun omissibility depending on whether the verb is non-past (i.e. fully inflected) or not. Here are some Russian examples and judgments; (30)-(32) are drawn from Chvany (1973).16

302

PARAMETERS OF SLAVIC MORPHOSYNTAX

(30)

a. Ivan dumaet, cto ja/?e pojdu (nom) thinks(3sg) that I(nom) will-come(lsg) 'Ivan thinks that I will come' b. Ja ne pojdu, no Ivan dumaet, I(nom) NEG will-come(lsg) but (nom) thinks(3sg) cto ja/e pojdu that I(nom) will-come(lsg) 'I am not coming, but Ivan thinks that I will come'

(31)

a. Ja ne znaju, pojde't li on/*e I(nom) NEG know will-come(3sg) if he(nom) 'I don't know whether he will come' b. Ja ne znaju, pojdu li ja/?e I(nom) NEG know will-come(lsg) if I(nom) 'I don't know whether I will come' c. On ne znaet, pojdet li on/?e he(nom) NEG knows will-come(3sg) if he(nom) 'I don't know whether he will come'

(32)

a. Vosel Ivan. On/e skazal "Rad vas videt'" entered (nom) he(nom) said(m) glad you to-see 'Ivan entered. He said "Glad to see you"' b. Vosel Ivan. "Rad vas videt'," skazal on/*e entered (nom) glad you to-see said(m) he(nom) 'Ivan entered. "Glad to see you," said he'

(33)

a.

"Rad vas videt'," skazal ja/*e glad you to-see said(m) I(nom) ' "Glad to see you," said F

b.

"Rad vas videt'," govorju ja/e glad you to-see say(lsg) I(nom) ' "Glad to see you," say F

(34)

a. Ja ne pomnju, posel li on/*e I(nom) NEG remember(lsg) left(m) if he(nom) 'I don't remember whether he left' b. Ja ne pomnju, posel li ja/*e I(nom) NEG remember(lsg) left(m) if I(nom) 'I don't remember whether I left' c. On ne pomnit, posel li on/*e he(nom) NEG remember(3sg) left(m) if he(nom) 'he doesn't remember whether he left'

Chvany (1973) states that in casual speech it is possible to delete subject pronouns if their referent is contextually determined. I shall refer to this process as discourse ellipsis; its full import will be considered in section 7.2.2. In rough terms, it seems

NULL SUBJECT PHENOMENA

303

as though Russian pro-drop is of the Chinese "non-local identification" type, despite its verbal morphology. That is, an empty subject agrees with a discourse topic, whether null or overt, rather than with rich agreement inflection. The examples in (30) show how a preceding topic facilitates the omission of the subject pronoun. Although the verb pojdu 'I-will-come' inflects for person, it is clear that discourse rather than grammatical factors are actually relevant. On the one hand, it is the introduction of the discourse topic ja T in (30b) that licenses the pro-drop. In the absence of this overt topic, as in (30a), the null subject is awkward since it presupposes the embedded subject as having been a recently highlighted topic. Crucially, the rich inflection of the verb pojdu cannot in and of itself license a null subject, even though it would appear to identify pro. On the other hand, the same discourse effect is seen in (32b), despite the relatively impoverished past tense inflection. In short, it looks like identification by inflection is irrelevant to Russian pro-drop. It seems to me that this is exactly as expected, if Russian pro-drop is really discourse-driven ellipsis. Now notice that in the YES/NO inversion construction in (31), the felicity of prodrop is similarly contingent on discourse licensing. However, this effect is mitigated by the absence of rich agreement, as shown by the judgment in (34). Here is one place, then, where identification of pro in Russian seems to rely on the local agreement properties of verbal inflection. The same phenomenon can be found in the direct speech inversion construction in (33), where ja T can only be dropped when fully identified by the verb. Since the examples in (31b, c) and (33b) only differ from those in (34b, c) and (33a) in terms of richness of AGR, discourse factors cannot be at work here. It is as if the inversion itself somehow blocks discourse licensing, so that only identification through local agreement is viable. However, even though identified, the null subject still cannot be licensed (by virtue of some special consequence of inversion), since otherwise (31a) should be acceptable with a null subject. These facts thus remain mysterious and a potential problem for standard models of pro-drop. The facts discussed thus far in this chapter demonstrate that multiple interrelated factors are involved in null subject phenomena. The diversity of pro-drop types suggests that notions of both licensing and identification are required, and the thematic versus expletive typology suggests that at least two distinct (although perhaps related) parameters are involved in licensing null subjects. I tentatively assumed a special [±OESP] parameter, in addition to Safir's [±NDP], to handle whether or not expletive subjects had to be overt, postponing its formalization until section 7.3. This move divorced the phonological status of expletive pronouns from that of thematic ones, allowing Russian to fit into the pro-drop paradigm. Note now that adopting such a parameter also implies that the existence of overt expletives in a language does not mean it is necessarily a non-pro-drop language. Although the typical situation, as in English, is for the two to go hand in hand, Raposo and Uriagereka (1990), for example, describe Galician Portuguese as a clear counterexample. Galician differs from standard Portuguese in that both theta-marked and expletive subjects are optionally overt. Raposo and Uriagereka (1990) cite the following examples:

304

(35)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a. nos/e comimos o caldo '(we) ate(lpl) the soup' b. el/e chovia '(it) rained(3sg)' c. el/e parecfa que o patron andaba canso '(it) seemed(3sg) that the boss went around tired'

Although empty subjects are both licensed and identified by rich AGR, the expletive pronouns in (35b, c) are not required to be null. Instead, they behave just as their thematic counterparts do. This state of affairs suggests that Galician Portuguese is a pro-drop ([ + NDP]) but non-null expletive ([-OESP]) language. Although a relatively uncommon situation, we shall soon see that even in Slavic this combination of properties can be found.

7.1.4. Two Sides of Visibility There are two independent aspects of the GB idea that NPs require case, known as the Case Filter. On the one hand, according to the Case Filter as originally proposed in Rouveret and Vergnaud (1980) and Chomsky (1981a), for example, NPs require case in order to be assigned a phonological representation at the level of P(honetic) F(orm). Consequently, in the absence of case the subject cannot be overt. We can therefore understand this aspect of the Case Filter as stipulating that case is necessary for PF-visibility. Case, in this sense, renders an NP visible at PF. On the other hand, an alternative conception of the Case Filter is that in L(ogical) F(orm) an NP must have case in order to be semantically visible. This aspect of the Case Filter was also suggested in Chomsky (1981a) and has since become fairly widely accepted. Under this view, an argument requires case in order for its theta-role to be visible at LF—a universal necessity independently motivated by the Theta-Criterion. For this reason, a theta-marked subject must bear case to be semantically well-formed. This aspect of the Case Filter can be regarded as stipulating that case is necessary for LF-visibility. Recent work by Belletti (1988), Lasnik (1992a), and Raposo and Uriagereka (1990) demonstrates that, contrary to standard assumptions, the Case Filter cannot be wholly reduced to the LF-visibility hypothesis. Following their reasoning, I too shall assume both principles are active in UG.17 Since the PF requirement is rooted in nominal morphology and extends to non-theta-marked elements, it appears to exist independently of the LF requirement. Moreoever, the two are actually statements about different types of entities: PF-visibility is a property of words, and LFvisibility is a property of argument chains. More precisely, then, it is a chain that must have case for its theta-role to be visible at LF: the chain is theta-marked if one of its elements is in a position to which a theta-role is assigned, and it is case-marked if one of its elements is in a position to which case is assigned. Notice, furthermore, that LF-visibility merely requires arguments to have case by LF, whereas for PFvisibility a nominal must have case by PF in order for the position that it occupies to be lexically filled. If case is assigned at or by S-structure, both sides of visibility are satisfied. Let us approach in this light the problem of why English and French differ from

NULL SUBJECT PHENOMENA

305

the other languages discussed with respect to the treatment of non-theta-marked subjects. Note that if case is regarded as licensing LF-visibility of an NP's semantic role, then the NOM-Drop Parameter cannot mean that for some languages nominative assignment is optional, since the subject NP must receive case for its theta-role to be visible. The issue, instead, is whether a nominative NP must be realized. Following the theory developed in chapter 2, realization can be understood as the replacement of a morphosyntactic matrix (a "syntactic word") with a phonological one, where only the latter can be phonetically instantiated.18 This issue leads back to the proper way to understand the Overt Expletive Subject Parameter. We have seen that the existence of this option in UG is hardly surprising, given the visibility version of the Case Filter: if case-marking is necessitated by the Theta-Criterion alone, we might expect non-thematic arguments not to receive case. Languages like English and French are unusual in requiring nominative assignment in all finite clauses, whether motivated by theta-theory or not. The question thus becomes why any language should make use of overt expletives at all. While the answer probably lies in the case assignment mechanisms of certain languages with impoverished systems, I can offer no explanation for why the [ + OESP] option should exist. Presumably, the phonological realization of expletives has to do with whether or not this position receives case. Since it is the subject of a finite clause containing AGR, an expletive may be assigned nominative. Although this is clearly not motivated by a need for LF-visibility, it is necessary under the PF-visibility hypothesis in order for the subject position to be occupied by an overt nominal. We can conclude, therefore, that canonical non-null subject languages are characterized by the property that (nominative) case features must be discharged; whenever AGR contains case features, these must be realized on its specifier. It is a positive specification for this parameter, stated in (36), that distinguishes languages like English from the others discussed. (36) Nominative case features must be discharged. Since universal principles require nominative to be phonologically realized when associated with a theta-marked argument, (36) can be regarded as a kind of generalization of this principle of UG to semantically vacuous expletive subjects. The parameter in (36) is a formalization of what I have been referring to as the Overt Expletive Subject Parameter.19 A [ + OESP] language, such as English, will always have overt subjects in finite clauses, whether theta-marked or not. A language with a negative specification, on the other hand, will not assign nominative case to subject position unless it is required for LF-visibility; such a language will therefore have null expletive subjects. Authier (1989) proposes a similar parameter for accusative case to differentiate English, which has expletive objects, from French, which lacks them. He points out the conceptual similarity between this approach and the Case Realization Conditions of Safir (1985, 185), whose NornDrop Parameter states that "Nominative Case need not be phonetically realized." The parameter in (36) differs crucially from the NDP, however, in that it (1) is stated in terms of the obligatory discharging of case features, and (2) does not refer explicitly to phonetic realization. I shall argue that this latter property allows it to

306

PARAMETERS OF SLAVIC MORPHOSYNTAX

handle the optionally overt expletives of languages like colloquial Czech and Galician Portuguese, as in (35). It is thus not a formulation of a Null Subject Parameter per se, since its effect is to license overt expletive subjects independently of the status of theta-marked ones. As it stands, the OESP is specifically designed to accommodate the fact that some languages have obligatorily overt expletive subject pronouns and others do not. The true situation is, of course, far more complex. We have already seen that in Galician expletive pronouns behave similarly to theta-marked ones in that both are omissible; in section 7.3.2 I show how the OESP interacts with the null subject parameter to derive this effect. In section 7.4 I discuss the marginal possibility of having overt expletives in Slavic even though they are [ — OESP] languages, showing how this follows from the LF-visibility hypothesis.

7.2. Null Thematic Subjects In this section I address the question of whether null theta-marked pronominal subjects are cased. The prediction of the LF-visibility hypothesis is that they are, since they bear theta-roles that must be visible at LF. This raises the problem of why South and West Slavic languages allow this cased pronominal to remain lexically unfilled, whereas Russian does not. 7.2.1. Case and Agreement The morphosyntactic properties of empty categories can only be observed indirectly, since they lack phonological realization. As we saw in chapter 6, however, it is possible to use agreement as a test for the presence of specific features. In particular, when an adjective is predicated of some NP, it must agree in as many features as possible. For the purposes of illustration, let us take secondary predication to involve the coindexation of a predicative AP with an NP antecedent.20 Recall that this coindexation induces agreement in all open features. It is therefore possible to identify the features of an empty subject on the basis of the form of agreeing predicate adjectives. For example, an application of this test to PRO reveals that this empty category must be endowed with the pronominal features of person, number, and gender, but crucially not case. When obligatorily controlled, we saw that agreement obtains in gender and number as well as case, as in the Polish example (37): (37)

Jan; chce [IP PROS bye przystojny;] 'John(m sg nom) wants to be handsome(m sg nom)'

This agreement is possible because the chain of indices is established by S-structure and is thus able to feed into PF. However, when PRO is not obligatorily controlled, we saw that the predicate adjective still shows number and gender features appropriate to the understood antecedent of PRO: (38) a. e wazne jest dla mnie [CP [IP PRO; bye dobryms]] 'it is important for me(gen) to be good(m sg inst)'

NULL SUBJECT PHENOMENA

307

b. Jan sadzi [CP ze [IP e wazne jest [CP [IP PRO; bye dobryttij]]]] 'John(nom) thinks that it is important to be good(m sg inst)' Nonetheless, the adjective appears in the default instrumental. Assuming that only grammatical control is assigned at S-structure, the coreference between PRO and mnie in (38a) or PRO and Jan in (38b) does not lead to agreement in case. These facts indicate that PRO must have intrinsic pronominal features but cannot have case. Indeed, it is well known that even when PRO has arbitrary reference, which is a possible interpretation of (38b), it still must have specific pronominal features, which may vary from language to language. What is of interest here is that pro in Polish differs from PRO in that it always behaves as if it had case features with respect to this diagnostic. Consider the following examples: (39)

a. pro; jest zme/czona; '(she) is tired(f sg nom)' b. pro; jestesmy zmeczeni; '(we) are tired(m pi nom)'

The predicate adjective agrees with its antecedent not only in pronominal features but also in case. The presence (or accessibility) of case therefore appears to be one of the basic differences between PRO and pro. The failure of PRO to bear case is a serious problem for the LF-visibility hypothesis. Since PRO is always theta-marked,21 we would expect it to also be casemarked. If, however, case is seen as but one of several conceivable ways of licensing an argument at LF,22 it is possible that some independent property of PRO satisfies the visibility requirement. One suggestion tentatively made in Lasnik (1992b) is that control provides the necessary licensing for PRO. Another due to Hornstein and Lightfoot (1987) is that PRO is privileged because it appears in subject position, which is always obligatory. This problem remains an open issue, however—one that has, in fact, received surprisingly little attention.

7.2.2. Ellipsis The reason I qualified my characterization of Russian (28) as being "more or less" like English is that Russian (and Slavic in general) is considerably more discourseoriented than English. I am using this term roughly in the sense of Huang (1984) and the references therein. Huang analyzes zero topics in Chinese as binding variables in argument positions; it is not clear whether his analysis of Chinese should be extended to handle comparable ellipsis phenomena in Slavic. For the purposes of this chapter, it suffices to take discourse ellipsis to mean that items recoverable from the context are frequently omitted on the surface. I take no stand as to whether this is specifically a PF process of deletion or should be analyzed in terms of null elements at all levels of representation, licensed by null topics adjoined to S. The point is simply that discourse ellipsis is available only (or at least far more extensively) in discourse-oriented languages such as Russian. Assuming some such mechanism, then, it seems reasonable to let ellipsis account

308

PARAMETERS OF SLAVIC MORPHOSYNTAX

for all omission of thematic subjects in Russian. In this regard, and in contrast to the South and West Slavic languages, there is nothing privileged about subject position. That is, although it is true that ellipsis of theta-marked subjects does in fact occur freely in Russian when recoverable from the context, this is only part of a more general phenomenon encompassing all arguments. Thus, it is possible to answer questions like 'Did Ivan buy a newspaper?' or 'Did you introduce Ivan to Masha?' in Russian as in (40). (40)

a. net, ne kupil 'no, (he) didn't buy (it)' b. da, predstavil yes,(I) introduced (him) (to her)'

This process also extends in colloquial Russian to verbs, as in (41a) and (41b), and to the contents of COMP (Spec-CP) in both relative clauses, as in (41c) and (41d), and in sentential complements, as in (41e) and (41f).23 (41)

a. tebja k telefonu 'you (are being called) to the phone' b. pocemu ty tak pozdno ? 'why have you (come) so late?' c. ja emu pokazala odnu ikonu u Soni visit v komnate 'I showed him an icon (which) is hanging in Sonya's room' d. daj-ka nine rubasku ja v teatr xozu 'give me the shirt (in which) I go to theater' e. Dima skazal zavtra budet xorosaja pogoda 'Dima said (that) tomorrow the weather will be fine' f. Dima prisel ja esce lezala 'Dima arrived (when) I was still in bed'

In (41a) the verbprosjat '(they) ask' has been ellided; in (41b) a verb of motion such as prisel 'arrived' is understood; and in (41c-f) the complementizers kotoraja 'which', v kotoroj 'in which', cto 'that', and kogda 'when' have been elided, respectively. I make no claims about just what discourse factors allow an element to be omitted. For present purposes, I simply characterize the phenomenon as deletion of contextually recoverable elements, as in (42), without attempting to make precise either the nature of discourse orientation or the formal mechanisms involved. (42)

Discourse-recoverable elements may be freely elided.

It is important to emphasize that this ellipsis process pertains not just to unemphatic pronominal subjects but to all elements, including verbs and embedded COMPs. Since (42) is operative in Slavic but not in English, it can be regarded as another parameter related (albeit indirectly) to null subject phenonema.

NULL SUBJECT PHENOMENA

309

7.2.3. Why Russian Is Different Why is it that Russian, although superficially similar to the other Slavic languages cited, does not omit pronominal subjects outside of discourse sanctioned contexts? Clearly, one would like to claim that Russian is not a true null subject language, although it appears to be of the mixed type since it patterns like the other Slavic languages with respect to the OESP. It must therefore be distinguished from the other languages in terms of its pro-drop capabilities. As discussed above, numerous attempts to instantiate a canonical Null Subject Parameter exist in the literature; what is crucial for our account is not exactly how it be formulated but rather that this parameter be independent of the OESP in order to interact with it. The answer to the question of how East Slavic differs from South and West hinges on the familiar observation that rich subject-verb agreement morphology tends to correlate with optionality of the subject itself. For Slavic this was illustrated in the examples in (21)~(23), where it was shown that Russian differs from Polish and Serbo-Croatian in that there is no independent realization of AGR: in equational constructions the copula is null, and in the /-participle past tense there is no personnumber auxiliary.24 It thus appears that in East Slavic AGR must become amalgamated with the verb in order to be realized. In South and West Slavic, on the other hand, AGR can be lexicalized as a clitic if no appropriate verb form is available, as in the participial and equational structures in (21)-(23). We can therefore generalize the traditional statement that Russian exhibits a zero copula in the present tense, as in (22a), to the observation that AGR in Russian has no independent lexicalization in any tense. It can only be morphologically realized as a verbal affix. Since the Russian verb byt' 'to be' is defective in that it has no present imperfective form, there will be nothing for AGR to attach to. This state of affairs is reminiscent of the pronominal clitic situation in the two language types in general. The South and West Slavic languages are similarly opposed to the East Slavic ones in that only the former have pronominal clitics. That is, whereas Serbo-Croatian uses the clitics ga *him(acc/gen)', mu 'him(dat)', te 'you(acc/gen)', ti 'you(dat)', etcetera, and Polish uses go, mu, tie., ci, etcetera, Russian has only the full forms ego, emu, tebja, tebe, etcetera. Let us therefore also regard person-number AGR as a kind of subject clitic, as is now standard in analyses of pro-drop in Romance (cf. e.g. Safir, 1985). This hypothesis amounts to the claim that all types of pronominal agreement—extending this notion to non-subjects, along the lines of Chomsky (1991)—are rich enough to able to identify null pronouns in the South and West Slavic languages, but not in East Slavic. It essentially unifies case and agreement, so that in Serbo-Croatian (43), for example, mu (as the masculine third person singular dative clitic pronoun) identifies the indirect object as njemu 'him' just as much as sam (as the first person auxiliary) identifies the subject asja T; in a sense, sam in (43) is the realization of nominative case. (43)

dao sam mu knjigu gave(m sg) AUX(lsg) him(dat) book(acc) 'I gave him a book'

310

PARAMETERS OF SLAVIC MORPHOSYNTAX

Both sam and mu are clitics and must appear in second position in the order given. In Russian (44), on the other hand, neither of these cases can be realized as clitics. (44)

ja dal emu knigu I(nom) gave(m) him(dat) book(acc) 'I gave him a book'

The analysis of clitic pronouns as realizations of agreement helps to explain why the South and West Slavic languages do not freely elide pronominal arguments. I do not develop this approach here, however, as it has far-reaching and rather complex implications for phrase structure. For example, if sam identifies a pro subject, then we would also expect mu to identify a pro object. It also raises the question of why, when the subject in (43) is expressed as ja for emphasis, the agreement element sam remains, but when the object is similarly emphatically expressed as njemu, no clitic mu appears. This kind of problem does not seem to me to be insurmountable. The agreement approach allows us to view it as clitic doubling, so that the question can be restated as, Why do only subject clitics in Serbo-Croatian (and Polish) allow doubling? In fact, in the East Balkan languages of Bulgarian and Macedonian, other pronominal clitics also participate in doubling. In Bulgarian, as in most other Balkan languages, this process is connected with emphasis, whereas in Macedonian it sometimes marks specificity or definiteness (see Berent, 1980, for details). What this would mean is that Serbo-Croatian njemu is 'him' + indirect object/dative AGR, just as Russian ja is T + subject/nominative AGR, but Serbo-Croatian ja (and East Balkan Slavic full pronouns and NPs) do not contain AGR, which is instead realized as the distinct clitic morpheme -m. Note finally that this method of connecting agreement, pronominal clitics, and case perhaps relates to my conclusion in chapter 6 that Russian had no need of the so-called split INFL hypothesis. That is, although in the South and West Slavic languages AGR and TNS nodes constitute independent projections, it was argued that in East Slavic they both appear under I. In the former languages, AGR is an independent head and can be realized as such, whereas in Russian it requires amalgamation with the verb. The approach is workable, I believe, although it suggests a much more complex phrase structure system than envisioned to handle most of the problems treated in this book. Disregarding the exact phrase structural nature of the AGR element, then, let us return to the question of how to characterize the Null Subject Parameter in a way that cuts across the language types properly. An initial statement of the NSP along these lines is given in (45). (45)

AGR is able to case-identify null subjects at S-structure.

Recall that, following proposals in Jaeggli and Safir (1989a), identification is necessary for all theta-marked null categories. This is achieved solely through case assignment in non-null subject languages, but in null subject languages it may also be accomplished via government by the element that otherwise assigns case to the empty position, that is, AGR. This is what I mean by "case-identify" in (45). The idea is that AGR is itself pronominal enough to instantiate nominative case. This suggests another way to think of (45): in null subject languages, nominative case assignment to subject position is optional. Optimally, of course, all rules of

NULL SUBJECT PHENOMENA

311

grammar are optional, including those for case assignment, as pointed out in Lasnik (1992a). Hence I claim, contrary to the tentative conclusion of section 7.2.1, that pro is in fact caseless at S-structure. This proposal helps to strengthen the PF-visibility aspect of the Case Filter into a biconditional, as argued in Milsark (1988). That is, any nominal with case at S-structure will be realized morphologically, and any nominal lacking case will not, where—as in chapter 2—the process of morphological realization is taken to be the replacement of a morphosyntactic feature matrix by the phonological one associated in the lexicon with a non-distinct set of morphosyntactic features. Note that the view that in null subject languages nominative case assignment is optional forces us to reject the earlier assumption, based on the agreement facts presented in (39), that pro bears case at S-structure. One possibility instead is that the predicate adjective is nominative through agreement, where the AP is coindexed with the pro subject (perhaps via a controlled small clause PRO subject), which is in turn coindexed with AGR. As discussed at length in chapter 6, this kind of case transmission mechanism for predicate adjective agreement is of course independently necessary to handle the ordinary subject control instance of (37). Another possibility is that, although caseless at S-structure, pro receives case at LF and it is this that causes the predicate adjective to agree. Indeed, the approach represented by (45) leaves open the question of whether pro needs case for its theta-role to be visible at LF. The minimal assumption that case is necessary for LF-visibility for all arguments (except PRO) leads to the idea that in null subject languages case assignment may take place at LF. That is, I suggest that case is still necessary for LF-visibility, but in null subject languages case assignment takes place by LF, whereas in non-null subject languages it must take place by S-structure. Hence, case assignment can in principle occur at any syntactic level, including D-structure for oblique (inherent) case. It is an aspect of UG that nominative case assignment is free to take place at S-structure or LF, although this is restricted to S-structure in non-null subject languages. This analysis raises the question of whether or not we ought to regard the level at which nominative case is assigned as an independent parameter, perhaps subsuming the NSP in (45). On the one hand, if case assignment were to fail to take place by S-structure in a [ — NSP] language, then the construction would be ungrammatical because the pro subject would not be identified. However, according to Jaeggli and Safir (1989a), expletive subjects do not need to be identified; hence, we ought to expect English to avail itself of the option of assigning nominative case to expletive subjects at LF. Intuitively, in order to prevent this we need to force nominative case assignment in English at S-structure. It thus seems that whether nominative case assignment is limited to S-structure or may also apply at LF cannot be completely derived from something like (45) and must instead be taken as a parametric property of individual grammars. I therefore conclude that the Null Subject Parameter should in fact be stated in terms of the level at which nominative case assignment takes place. Moreover, just as with the OESP—since the unmarked effect of such a parameter should be that pronominal subjects remain silent (cf. Hyams, 1986) and its default value consequently negative—I dub this new version of the parameter the Overt Subject Parameter (OSP).

312

PARAMETERS OF SLAVIC MORPHOSYNTAX

(46) Nominative case assignment cannot take place at LF. The OSP in (46) replaces the NSP in (45). When a [ - OSP] language has an overt subject, case has been assigned by S-structure, enabling the form to feed into PF, and when it has a null subject it is identified at S-structure for the pronominal features of person, number, and gender by AGR, but only receives case at LF.

7.3. Expletives and Visibility In this section I consider whether expletive pronouns need case. Given the dual conception of the Case Filter, I contend, following Lasnik (1992a), that overt expletives require case in order to PF-visible, but silent ones do not. For LF-visibility, on the other hand, an expletive pronoun should not need case since it is not associated with any theta-role. However, I will distinguish expletive pronouns that are in a CHAIN with a theta-role from those that are not, where a CHAIN is an S-structure chain headed by an expletive element, following Chomsky (1986b).

7.3.1. Null Expletives Do Not Need Case There has been much recent discussion of the syntax of argument CHAINs headed by expletives (cf. e.g. Chomsky, 1986b; Lasnik, 1992a; Raposo and Uriagereka, 1990). A standard example is given as in (47). (47) there; are [NP three men]; in the room. Earlier it was argued that three men receives nominative case by transmission from the coindexed subject position. Lasnik (1992a) has convincingly shown, however, that both elements receive case independently in situ—there from AGR and three men from be. This accords with my claim that all nominals must have case at S-structure in order to be overt. At LF, according to Chomsky (1986b), the NP three men raises to fill the subject position, replacing the expletive. This follows from his Principle of Full Interpretation, which states that all elements must receive an appropriate interpretation at PF and LF. The result of expletive replacement is that all expletives are eliminated at LF and all S-structure CHAINs are transformed into LF chains. Consider in this light what should happen in a null expletive language that has an inversion structure such as (47). In the present system these languages are [ —OESP]. The expletive does not receive case at S-structure, since it is not PFvisible, although the NP does. At LF various scenarios are possible. If the NP still moves to subject position, it is not clear that it needs to be assigned case there, since it originates in a cased position. Alternatively, since the expletive lacks case and is therefore for all intents and purposes invisible, perhaps no Principle of Full Interpretation violation obtains, obviating the need for expletive replacement. It thus seems unlikely that such an expletive subject should be assigned case at any level of representation.

NULL SUBJECT PHENOMENA

313

7.3.2. Overt Expletives Need Case at S-Structure I have argued that, so far as LF-visibility is concerned, case is not necessary for overt expletive NPs, although for PF-visibility it is. Hence, languages that do not require nominative case features to be discharged ought to lack overt expletives, since neither PF- nor LF-visibility motivates case assignment. Languages that are [ + OESP], on the other hand, should have obligatory expletive subject pronouns. As we have seen above, although this is true of English, Galician Portuguese poses a potential problem for the parametric account. The relevant examples from (35) are repeated in (48): (48) a. nos/e comimos o caldo '(we) ate(lpl) the soup' b. el/e chovia '(it) rained(3sg)' c. el/e parecfa que o patron andaba canso '(it) seemed(3sg) that the boss went around tired' Even though Galician is clearly a [ - OSP] language, it freely allows overt expletive subjects. In order to accommodate this fact, I propose that Galician Portuguese differs from standard Portuguese in that it is positively specified for the OESP. Nominative case must therefore be discharged to subject position. However, since it is also f — OSP], the option always exists for nominative case to be assigned at LF rather than at S-structure. Thus, the e in (48a) is pro and receives case at LF for reasons of LFvisibility, but those in (48b, c) are expletives and are assigned case at LF only because nominative case must be discharged at some point.25 Several Slavic languages also exhibit a pro-drop pattern that includes optionally overt expletives. Both Upper Serbian (49) and Lower Serbian (50), for example, are null subject languages that allow overt expletives of various kinds.26 (49) a. wono hrima 'it is thundering' b. wone be wetrokojte 'it was windy' c. won so descuje 'it is raining' d. to so lije 'boy, it is really raining' (50)

a. wono by wSo lepjej bylo 'it all would have been better' b. wono se scerka 'it is being secretly talked about' c. to se zasej swetli 'it's getting light out again'

314

PARAMETERS OF SLAVIC MORPHOSYNTAX

d. to se pjersy 'it is drizzling' Lindseth (1991, 1993b) reports that overt expletives are found in written and spoken Lower Serbian but are now restricted to dialects of Upper Sorbian, although they used to be a common feature of the written language. Both languages thus admit the expletive wono 'it' in impersonal constructions. The demonstrative to serves a similar function in Lower Sorbian, as in (50c, d). In Upper Sorbian, as in (49d), to adds an emphatic nuance, as in the translation. Despite this range of overt expletive phenomena, Upper and Lower Sorbian, like the other West Slavic languages, are otherwise pro-drop. Examples (5la, b) are from Upper Sorbian, and (51c, d) are from Lower Sorbian: (51)

a. wonka mjerzuje '(it) is freezing outside' b. smoj piloj '(we both) have drunk' c. pojedu bjez tebje k wefowanju '(I) will go without you to the wedding' d. som jogo plasii do pfednego apryla '(I) paid him until April first'

In (51b-d) a thematic subject pronoun is omitted, and in (51a) there is a null expletive subject. Both variants of Sorbian are apparently thus [ —OSP] but t + OESP]. Lindseth (1993b), however, raises the interesting question of whether these expletive pronouns really occupy SPEC-IP. For Upper Sorbian wane at least, she marshalls convincing evidence that the overt expletive is actually in SPEC-CP position. For one thing, wane is incompatible with a w/z-word: (52)

hdyz (*wone) taje, nejmozemy so smykac 'when it thaws, we can't skate'

It similarly cannot follow a complementizer, such as dokelz 'because', but can follow a conjunction such as ale 'but': (53)

a. nejridu won dokelz so (*wone) descuje 'I don't go outside because it is raining' b. chcyla bych won hie, ale wone so descuje 'I would like to go outside, but it is raining'

Note that the positioning of the reflexive clitic so in these examples clearly indicates a difference between complementizers and conjunctions: complementizers, but not conjunctions, count in determining '' second position.'' This suggests that the former are in CP but the latter are not. Colloquial Czech also countenances overt expletives. According to Koenitz (1988), colloquial Czech allows the expletive (v)ono 'it' in examples such as the following:27

NULL SUBJECT PHENOMENA

315

(54) a. (v)ono je chladno 'it is cold' b. (v)ono se tarn nepracuje it REFL there NEC-work 'one doesn't work there' (55) a. (v)ono je mozno, ze . . . 'it is possible that . . . ' b. (v)ono je pfece nutno, abychom . . . 'it is after all necessary that I . . . ' In these examples the expletive alternates with zero, as expected; compare the Czech examples in (15). Just as with thematic subjects, the presence of the expletive is marked and indicates some kind of discourse focus or emphasis; (56) for example is pragmatically infelicitous.28 (56)

*(v)ono pr§i, ale sviti slunce ' it is raining, but the sun is shining'

These can thus be treated the same way as in Galician and Serbian, with colloquial Czech also being analyzed as [-OSP] and [ + OESP]. There is an interesting variation on this construction that requires some comment, however. An expletive also appears in certain emphatic contexts with an inverted thematic nominative subject. This is illustrated in (57).29 (57) a. (v)ona se Mafenka stara o vSechno it(fnom) REFL (f nom) takes-care of everything 'Mafenka really takes care of everything' b. (v)on je to nSkdy problem it(mnom) is that sometimes problem(m nom) 'sometimes it is really a problem' c. myslil jsem, ze to byl Vaclav, thought AUX(lsg) that that was Vaclav a (v)on to byl nelcdo jiny but it(mnom) that was someone other(m nom) 'I thought it was Vaclav, but really it was someone else' d. (v)ono tarn bylo moc lidi it(n nom) there was many people(gen pi) 'there were many people there' In the modern language, the form of the expletive pronoun generally agrees in gender with the thematic subject, feminine ona for Mafenka in (57a), masculine on for problem in (57b) and nekdo jiny in (57c), and neuter ono for the quantified subject moc lidi in (57d). These are presumably subjects, since the verb agrees with them and they appear in the nominative case. Interestingly, according to Lamprecht, Slosar, and Bauer (1986) and as drawn to my attention by Martina Lindseth, in Old Czech ono translated Latin ecce and the agreeing forms are a later development. It is not clear how this more complex effect should be derived within the current

316

PARAMETERS OF SLAVIC MORPHOSYNTAX

model. One possibility is that the expletive occupies SPEC-IP/AgrP position at S-structure. If so, the thematic subject for some reason either remains in its D-structure SPEC-VP position or raises and then extraposes. Although in section 7.4 we shall see that extraposition sometimes gives rise to an overt expletive in subject position, word order considerations do not seem to implicate extraposition. Scrambling, as always, is a complicating factor; Mafenka in (57a), for example, remains pre-verbal. It is nonetheless clear that the expletive pronoun and the subject NP are in a CHAIN at S-structure, however this is to be established. Case and pronominal features are apparently shared by both members of the CHAIN, just as in English (47), where the verb are shows appropriate plural agreement. Another possibility here is that the expletive occupies SPEC-CP position, as Lindseth (1993b) proposes for Upper Serbian. This would not free up SPEC-IP/AgrP for the full NP, however, since the presence of agreement indicates that the expletive must at least pass through this position. This possibility also seems unlikely because, as Lindseth points out, Czech and Upper Sorbian expletives also differ in distribution; compare Czech (58) with Upper Sorbian (53a).30 (58)

nemuzu jit ven, protoze (v)ono tarn prsi 'I can't to go outside, because it is raining there'

Finally, the historical change mentioned above suggests that Czech ono went from being extrasentential to SPEC-CP to SPEC-IP, with these last two options still perhaps competing in the modem language.31 The phenomenon of overt expletives in otherwise pro-drop languages is thus much more widespread than has been generally thought. Indeed, one can find potential examples from a broad spectrum of East Slavic, such as Ukrainian (59), Belarusian (60), and dialects of Russian (61).32 (59)

a. ja ne mozu ne plakat', I NEG can NEG to-cry bo vono placet'-sja proty mojeji voli for it cry -REFL against my will 'I can't help crying, for I cry against my will' b. vono taky pravda, sco . . . 'it is true that . . . ' c. vono b ne vadylo s'ohodny pohuljaty (cf. 13a) 'it would not be worthwhile to take a walk today' d. vono tut scos je 'there is something here'

(60)

a. a jano j moza j ryzdazdzycca 'and it might even rain' b. jano dobra, a navat treba znac' susedzkuju movu 'it is good, and even necessary to know your neighbor's language' c. jono bez korovy trudno 'it is difficult without a cow'

NULL SUBJECT PHENOMENA

317

d. jano wse dobra it everything good 'everything is fine' (61) a. ono teplo v izbe 'it is warm in the house' b. xuot' ono sniegu i niet at-least it snow(gen) even is-not 'at least there isn't even any snow' c. a ona xotelo -s' mated pogljadet' and it wanted-REFL mother(dat) to-look 'and mother felt like taking a look' d. da ana govorit' nel'zja but it to-speak impossible 'but one isn't able to talk' e. ono ran'se to bylo u nas tri zerdi it before that was by us three poles 'we used to have three poles' f. net, ana nyne nebal'Soj snex byl 'no, there was just a little snow(nom)' g. xos' ana nadvigali tucki vse ravno do§s ne budet' although it moved-over clouds(nom) all same rain(nom) NEG will-be 'although some small clouds have moved in, there won't be any rain' These examples represent the three types of overt expletive subject pronouns already encountered. There are the quasi-argumental weather expletives, such as (61a), expletives that fill SPEC-IP position when the logical subject is non-nominative, such as an oblique NP as in (61b, c), or a non-cased argument as in (61d), and expletives that are in a CHAIN with an overt nominative subject, as in (61f, g). Note that (61e) is presumably of this last type, assuming the analysis in chapter 4 of nonagreeing QPs in East Slavic as occupying SPEC-VP position. This suggests that in expletive argument CHAINs the overt nominal subject actually remains in SPEC-VP position, whether inverted, as in (61g), or not, as in (61f). The construction with an overt subject is of the English presentational there type, rather than the colloquial Czech type in (57), in that the expletive does not agree in pronominal features with the subject. In sum, the language types we have considered are parametrically opposed in the following way: (62) a. b. c. d.

Spanish, Serbo-Croatian, Polish Standard Russian Galician, colloquial Czech, etc. French, English

[-OESP] [-OESP] [ + OESP] [ + OESP]

[-OSP] [ + OSP] [-OSP] [ + OSP]

At S-structure, then, type (62a) has optionally case-marked (hence overt) thematic subjects only, type (62b) has necessarily case-marked thematic subjects only,

318

PARAMETERS OF SLAVIC MORPHOSYNTAX

type (62c) has optionally case-marked both thematic and expletive subjects, and type (62d) has necessarily case-marked both thematic and expletive subjects. The result is that the only configuration predicted to be absolutely impossible is for thematic subjects to be necessarily cased, but expletive ones to be only optionally cased.

7.3.3. Different Kinds of Null Expletives I have argued that in at least Galician Portuguese and colloquial Czech null expletives must receive case at LF. It is not obvious, however, that even in [-OESP] languages all null expletives are equally irrelevant for the LF-visibility of arguments and consequently impervious to case assignment. In particular, one might ask whether argument chains that are not headed by NPs also require case for their thetaroles to be visible, since if they are, then the expletive position might be its only viable source. Here, the two sides of the requirement contrast. So far as PF-visibility is concerned, case should not be necessary, but for LF-visibility it should so long as the chain is theta-marked. This raises the problem of whether null expletive-CP or null expletive-PP CHAINs are case-marked. So far, we have been treating examples such as (48b) and (48c) as comparable. If the clause in (48c) requires case for thetarole visibility, however, then we might expect some difference between them. Since LF is relatively inaccessible and probably tolerates little (if any) crosslinguistic variation, I assume that clausal and prepositional arguments also require case for their theta-roles to be visible at LF. Stowell (1981) argues that while this is indeed so, other factors complicate the picture and make the case status of non-NP arguments extremely difficult to verify. Citing examples such as those in (63) and (64), he points out that whereas finite CPs do behave as though they need to be in a chain with case, infinitival CPs and PPs do not. (63)

a.

it; seems [PP to us] [CP that John is an excellent student];

b. *it; seems [CP that John is an excellent student]; [PP to us] (64)

a. John seems [Pp to us] [CP to be an excellent student] b. John seems [CP to be an excellent student] [PP to us]

According to Stowell's analysis, based on proposals in Emonds (1976), finite clauses move into subject position to receive nominative case and then obligatorily extrapose or topicalize because, by his Case Resistance Principle, they cannot remain in a cased position at S-structure.33 This accounts for the contrast in (60), where the clause moves to subject position and subsequently extraposes, with the result that the finite clause appears sentence finally.34 Infinitivals and PPs, on the other hand, have intrinsic case and thus do not need to move into subject position to get case. Consequently, the infinitival clause in (61) may remain in its D-structure VP-internal position and still satisfy the Case Filter at LF. Assuming such an analysis, I conclude that clauses too must be in a chain with case, although finite clauses and infinitivals meet this requirement in different ways.

NULL SUBJECT PHENOMENA

319

7.4. Overt Expletives in Slavic The preceding discussion leads to the conclusion that there are, in fact, two reasons for the appearance of expletive it in English when it is in a CHAIN with a finite CP but only one when associated with an infinitival CP or a PP. Consider, for example, the following sentences: (65)

a. it; is nice [CP that we are walking in the park], b. it is nice [CP PRO to walk in the park] c. it is nice [PP in the park]

The expletive it in (65a) is in a theta-marked CHAIN and it is the subject of a finite verb; it in (65b) and (65c), on the other hand, is only the subject of a finite verb. Because English is [ + OESP], nominative case must be realized regardless, and so there is no reason in English to resolve the motivation for its lexicalization. However, in [ — OESP] languages the situation is somewhat different. As already observed, although the English PF requirement that nominative case must be realized is absent, the LF requirement that a theta-marked chain also be cased remains. Consequently, only expletive subjects coindexed with finite clauses ought to be assigned case. That is, the expletive in a chain with a finite clause needs case for its theta-role to be visible at LF, while other expletives would only need case in order to be phonologically visible at PF. Hence, we expect that expletive subjects coindexed with finite clauses should in principle be lexicalizable in the Slavic languages, even though they are all otherwise [-OESP].35 This expectation is supported by the Hebrew fact mentioned above that ze 'it' is optionally present in subject position when finite clauses are extraposed, as in (66), cited by Hyams (1989) from Borer (1984). (66)

(ze) margiz ?oti she Itamar tamid me ?axer '(it) annoys me that Itamar is always late'

The reason for the optional appearance of ze, even though Hebrew is [ — OESP], is that the clause that is in a CHAIN with the expletive requires case in order for its theta-role to be visible at LF. In the remainder of this section we investigate the status of comparable examples in the Slavic languages.

7.4.1. Russian Interestingly, the relevant constructions do contrast with respect to the viability of an overt expletive subject. Compare the Russian versions of English (65): (67)

a. (eto) prijatno [CP cto my guljaem v parke] 'it is nice that we are walking in the park' b. (*eto) prijatno [CP PRO guljat' v parke] 'it is nice to walk in the park' c. (*eto) prijatno [PP v parke] 'it is nice in the park'

320

PARAMETERS OF SLAVIC MORPHOSYNTAX

In all three examples the bracketed phrase originates in the VP. According to Stowell's (1981) analysis, in (67a) the finite clause moves into subject position and then extraposes, while in (67b, c) the infinitival clause and PP remain in situ. Consequently, the overt expletive eto is possible only in (67a).36 Russian thus bears out the prediction based on the theoretical opposition between the two kinds of expletive subjects, a prediction that is difficult to test relying solely on data drawn from either canonical null subject or non-null subject languages. These Russian facts are reminiscent of a similar phenomenon in Irish, as discussed, for example, in McCloskey (1985) and Stowell (1989). Essentially, raising verbs in Irish typically have overt expletives when they take tensed CP complements and null ones when they take infinitival CP complements. Compare the examples in (68) with those in (69), from Stowell (1989, 333-335): (68)

a. caithfidh (se) [CP go bhfuil si breoite] must it COMP is she ill 'it must be that she is ill' b. ni feidir (do) [CPgo bhfuil si breoite] NEG possible to-it COMP is she ill 'it is not possible that she is ill'

(69)

a. caithfidh (*se) [CP gan f a bheith breoite] must it NEG her to be ill 'it must be that she is not ill' b. ni feidir (*do) [CP f a bheith breoite] NEG possibleto-it her to be ill 'she can't be ill'

In (68), when in a CHAIN with a finite clause, the expletive is optional, whereas in (69), when in a CHAIN with an infinitival clause, the expletive is unacceptable. Irish thus otherwise appears to be like Spanish and Polish, that is, [ —OESP] and [-OSP], but expletives can appear precisely in those contexts that require case assignment for theta-theoretic reasons. Returning to Russian, it is in general felicitous to lexicalize an expletive subject only when coindexed with a finite clause. Otherwise, as in (67b, c), the presence of an overt expletive is virtually unacceptable simply because in other constructions there is no LF motivation for assigning case to subject position.37 Infinitival clauses and prepositional phrases thus contrast to finite clauses, providing novel empirical support for the account in Stowell (1981). Looking back at the examples in (29), we observe that in none of them is it possible to replace the null subject with an overt eto 'it'. These can be contrasted to the Russian examples in (70), in which eto is completely optional. (70)

a. (eto) xoroso/stranno, eto Ivan poSe'l domoj 'it is good/odd that Ivan went home' b. (eto) interesno, eto ona kupila etu knigu 'it is remarkable that she bought this book'

NULL SUBJECT PHENOMENA

321

c. (eto) vam tol'ko kazetsja, cto ray zabludilis' 'it only seems to you that we lost our way' This approach, according to which case-marking is motivated by LF considerations but in turn drives PF-visibility, leads us to expect the observed contrast between (29) and (70). Concomitantly, it forces us to reappraise the status of the examples in (70) when the expletive pronoun eto is omitted. This empty subject differs from the null expletives of (29) in that it must be case-marked for LF-visibility. The question is thus not why eto is possible in (70), but why it is omissible. Recall that the sole typological prohibition predicted by the two parameters proposed above is that no language may require thematic subjects to be assigned case at S-structure but allow expletive ones to be assigned case at either S-structure or LF. The subject positions in (70) must be case-marked because Russian is a [ -I- OSP] language, hence we expect nominative case to be assigned to such expletives at S-structure. I contend, however, that their cased status is obscured by the general PF ellipsis rule. Since pleonastic elements carry no informational burden and are entirely recoverable, they should be freely deletable regardless of discourse factors. For this reason, the expletive eto in (67a) and (70) is frequently omitted even though it is syntactically present. Notice that this problem does not arise in (29). No case need ever be assigned to these expletive subjects, since the infinitival CP and PP satisfy LF-visibility by virtue of being intrinsically cased. An alternative analysis might be to reject (46) in favor of (45) and claim that the subjects in (70) are able to wait to receive case until LF, after expletive replacement transforms the CHAIN into a chain. However, there are several serious difficulties with this alternative that lead me to adopt the ellipsis account. First, if identification is a property of all argument CHAINs, as seems reasonable, then the clause still needs to be identified at S-structure. Since Russian is not a null subject language, the only way that this can be achieved is by assigning nominative case to subject position. Second, the assumption that case can be assigned either at S-structure or LF even in non-null subject languages raises the question of why this option should be absent in examples like English (65b, c), as well as in languages that lack expletives altogether. Third, assuming (45) forces one to ask the opposite question of why nominative case cannot be assigned to expletive subjects at LF in English, whereas (46) rules this option out in principle for [ + OESP, +OSP] languages. Finally, the presence of Case Resistance Principle effects in Russian (albeit weakened by the relatively free surface word order of this language) supports the conclusion that the expletive subjects in (67) must receive case by S-structure. For example, (71a) represents a far more natural word order than the scrambled (71b). (71) a. Boris objasnil [NP podruge] [CP Cto Ivan uze usel] ' Boris explained to his girl friend that Ivan had left' b. ?Boris objasnil [CP 6to Ivan uze usel] [NP podruge] ' Boris explained that Ivan had left to his girl friend' Consequently, I opt for the analysis of Russian according to which the OSP in (46) i a parameter of UG, an expletive subject in a CHAIN with a finite clause is obliga-

322

PARAMETERS OF SLAVIC MORPHOSYNTAX

torily cased at S-structure, and its PF optionality is due to an ellipsis rule such as (42). 7.4.2. South and West Slavic Returning now to the other Slavic languages, we see that overt expletives are prohibited in counterparts to Russian (29) and marginal in constructions comparable to those in Russian (70), where subject position is coindexed with an extraposed finite clause. Just as in Russian, it is impossible to replace the null expletive subjects in Serbo-Croatian (26) or Polish (27) with the overt form to 'it'. If, however, subject position enters into an argument CHAIN, then a to subject becomes marginally acceptable. Consider Serbo-Croatian (72) and Polish (73).38 (72)

a. (?to) je dobro/zanimljivo [sto Jovan ide kudi] 'it is good/interesting that John is going home' b. (?to) izgleda dobro [sto ste dosli] 'it looks good that you arrived' c. (?to) vam se samo cini [da smo se izgubili] 'it only seems to you that we are lost'

(73)

a. (?to) jest dobrze/milo [ze wszyscy poszli do domu] 'it is good/nice that everyone went home' b. (?to) Jana smuci [ze wszyscy poszli do domu] 'it upsets John that everyone went home' c. (?to) sie. tylko zdaje [ze wszyscy poszli do domu] 'it only seems that everyone went home'

Note that although there is considerable variation as to the viability of to depending on complex factors pertaining to semantic properties of the predicate selected, it is striking that the judgments for Serbo-Croatian and Polish are consistently slightly worse than those for Russian.39 And once again, any attempt to insert to when the expletive subject is not in an argument CHAIN results in ungrammaticality. Just as in (standard) Russian, to cannot replace the expletive e in any of the examples in Serbo-Croatian (26) or Polish (27). This follows, as before, from the assumption that nominative case is unassigned since the subject position is not associated with any argument bearing a theta-role. It remains to be explained why the overt expletive to is one notch less admissible in South and West Slavic than in East Slavic even when coindexed with a finite clause. The former are true null subject languages, which I have characterized as [ - OSP]. This means that pronominal subjects in general need not be phonologically realized. Considered from the perspective of this property, such an expletive to is just like any other subject pronoun in that it has the option of remaining phonologically empty since case may be assigned at LF. Moreover, there is rarely any functional reason for expletives to be emphatic and thus to be expressed overtly (although factors such as being within the scope of 'only', as in [70c], [72c], and

NULL SUBJECT PHENOMENA

323

[73c], certainly help). For these reasons, even to coindexed with a finite CP preferably remains silent. Finally, it is somewhat mysterious why it is at all possible to lexicalize this to in South and West Slavic, given the impossibility of having a comparable overt expletive in other canonical [-OESP, — OSP] languages, such as Spanish and Italian. That is, if expletive subjects in argument CHAINs with finite clauses do not need case until expletive replacement takes place at LF, 1. Why does case assignment at S-structure ever occur in South and West Slavic? 2. Why is this option apparently absent in otherwise similar languages? This kind of problem was also noted by Stowell (1989, n. 10) in discussing Irish, who comments that "the fact that expletive pronouns may occur overtly in Irish, despite the fact that it is a null subject language, is surprising; this is not possible in Romance languages such as Spanish and Italian. This raises an interesting problem for most theories of null subjects in the GB literature." The solution to this problem awaits a clearer understanding of the true range and nature of overt expletives in otherwise null subject languages.

7.5. Conclusion: Arbitrary Third Plural Subjects In this chapter I have compared constructions in English and the three major branches of Slavic in support of a parametric approach to the lexicalization of expletive and thematic subject pronouns. Relevant data were also introduced from Galician Portuguese, Irish, and Hebrew. It was argued that both conceptions of the Case Filter are necessary and valid: 1. A theta-marked element needs case to be LF-visible. 2. An NP needs case to be PF-visible. In general, whereas LF-visibility is universally required, PF-visibility admits some parametric variation. If case-marking were necessitated by the Theta-Criterion alone, we should never expect expletive subjects to receive case (unless they are part of an argument chain that has no other source for case). While this is indeed true for canonical null subject languages, other languages, such as English, colloquial Czech, and Galician Portuguese, have an additional parametric property that nominative case must be realized, regardless of theta-theoretic concerns. This situation contrasts, however, to that of Russian, Serbo-Croatian, and Polish, where nominative case must be assigned to subject position only if that position is theta-marked. This led to the postulation of the Overt Expletive Subject Parameter in (36). Next, languages like English and Russian were contrasted to languages like Serbo-Croatian and Polish with respect to the admissibility of null theta-marked subjects. It was argued that a treatment of this opposition is best formulated in terms of the level at which nominative case assignment may take place, stated as the Overt

324

PARAMETERS OF SLAVIC MORPHOSYNTAX

Subject Parameter in (46), which in the present framework corresponds most closely to the traditional non-null subject/null subject opposition. This led to a straightforward understanding of the existence of optionally overt expletive subjects in Galician and Czech. Finally, it was observed that in Slavic there is a PF process whereby contextually recoverable arguments may be deleted, stated in (42), and that this relatively superficial property obscures the fact that Russian is not a true null subject language. These various parameters are summarized in (74). (74)

a. Nominative case features must be realized. b. Discourse-recoverable elements may be freely elided. c. Nominative case assignment cannot take place at LF.

English, Galician, and colloquial Czech have property (74a), all the Slavic languages have property (74b), and only Russian and English have property (74c). Note, finally, that (74a) appears to be a PF generalization of the universal requirement that argument chains must have case at LF. These parametric differences among the languages discussed illuminate the observed hierarchy of admissibility of overt expletive pronouns. By focusing on the status of expletive subjects in thetamarked argument chains, it becomes clear that (74a) allows them to be lexicalized in Galician Portuguese and colloquial Czech, the combination of (74a) and (74c) forces their phoneticization in English and French, (74b) allows their PF elimination in East Slavic, and, in addition to the effect of (74b), (74c) allows them to be null at S-structure in South and West Slavic. I have in this chapter addressed standard pro-drop issues from a Slavic perspective. This discussion is not intended to be an exhaustive treatment of all types of null subjects found in the Slavic languages, however. There are a variety of important null subject personal constructions, that is, with an agreeing verb but typically (if not necessarily) a null subject pronoun. In Slavic, as in English, these include generic and proverbial expressions with second person singular and first person plural subjects, and to a lesser extent second plural subjects, as in Russian (75) or Serbo-Croatian (76).40 (75)

a. slezami delu ne e pomozes' tears(inst) matter(dat) NEG help(2sg) '(you) won't fix it by crying' b. tise cedes', dal'se e budes' quieter go(2sg) further be(2sg) 'the quieter (you) go, the further (you) get' c. e ljubis' katat'sja—ljubi i sanocki vozit' love(2sg) to-ride love(2sg imper) also sleigh to-drive '(you) that dance must pay the fiddler'

(76)

a. e zivimo da ucimo live(lpl) COMP learn(lpl) '(we) live in order to learn'

NULL SUBJECT PHENOMENA

325

b. kad e zelite napraviti kolac . . . when want(2pl) to-make cake 'when (you) want to bake a cake . . . ' Probably the most pervasive null subject personal construction, however, is the arbitrary third person plural type. Again as in English, such sentences exist with varying degrees of productivity throughout Slavic and are particularly common in East Slavic. Consider the Russian examples in (77) and the Polish examples in (78), with arbitrary third person plural pro. (77) a. v Amerike e govorjat po-anglijski 'in America (they) speak English' b. otec znaet, cto e syna ne primut v institut 'father knows that (they) won't accept his son into the institute' c. vo Francii e edjat ulitok 'in France (they) eat snails' d. na benzokolonkax e prodajut sigarety 'at gas stations (they) sell cigarettes' (78)

a. e mowia, ze Bush wygra wybory '(they) say that Bush will win the elections' b. we Francji e jedza slimaki 'in France (they) eat snails' c. w Holandii e hoduja tulipany 'in Holland (they) grow tulips' d. w tym miescie e sprzedaja dobra kietbase 'in this town (they) sell good sausage'

Given the overall significance of this construction, I conclude this chapter by examining more carefully this type of null subject in Slavic. Such arbitrary plural pronominals have been analyzed in English and Spanish in Jaeggli (1986b), and Sperling (1990) offers an in-depth discussion of the complete range of arbitrary subject constructions in Slavic. As she points out, this construction exists in all the languages but is far more pervasive in East Slavic. The reason for this is presumably that South and West Slavic use reflexive constructions to express arbitrary subjecthood much more extensively than does East Slavic. Polish (79a) with a third plural verb and (79b) with a reflexive neuter third singular verb are thus (very roughly) equivalent.41 (79)

a. na ulicy e spiewaja in street sing(3pl) '(they) are singing in the street' b. na ulicy e sie. spiewa in street REFL sing(3sg) '(one) sings in the street'

326

PARAMETERS OF SLAVIC MORPHOSYNTAX

Whereas the pro subject in (79a) is theta-marked, in (79b) it is presumably a null expletive; neither, nonetheless, can be overt. Serbo-Croatian (80a) and (80b) form a much closer minimal pair, since in the middle voice (80b) subject position is occupied by kava 'coffee'. (80)

a. ovdje e prodaju kavu here sell(3pl) coffee(acc) '(they) sell coffee here' b. ovdje se prodaje kava here REFL sell(3sg) coffee(nom) 'coffee is sold here'

Interestingly, when a third plural pronominal is overtly expressed in Slavic, it generally loses its arbitrary character and becomes referential. Serbo-Croatian (80a) with the overt subject oni, for example, has only a referential reading. (81)

oni ovdje prodaju kavu 'they sell coffee here'

This fact is clearly related in some way to the pro-drop status of these languages. Notice in this regard that English they is used in a similarly arbitrary way: (82)

a. they sell cigarettes in gas stations b. they speak Czech in Prague

In English, of course, they cannot be omitted since English is not a null subject language in any of the senses of the term recognized in this chapter. The question, then, is why overt they can have an arbitrary sense in English, whereas overt oni 'they' in Serbo-Croatian (81) cannot. Jaeggli (1986b, 65), in comparing English with Spanish, argues that "in a language where the overt/empty alternation occurs for pronominals, a pronominal chain containing an overt pronoun may not have arbitrary reference." It is unclear, however, how this generalization should be extended to Slavic, if the same prohibition against overt arbitrary third person plural subjects also holds in Russian, since this language is not a canonical null subject language in that it is [ + OSP]. Considering this phenomenon from the perspective of (62), in fact, it would seem that the only conclusion one could make is that if a language drops subjects for any reason, then it will have necessarily null third person plural arbitrary subjects. Even if this generalization were correct, the status of pro in examples such as Russian (77) would still be mysterious from the perspective of the parameters in (74). These arbitrary pros are presumably nominative at S-structure, since they are theta-marked and Russian does not countenance LF case-marking. The only option within the present theory is thus that, in Russian, dropping pro in the arbitrary third plural construction is some kind of extension of discourse-ellipsis. In the South and West Slavic languages, on the other hand, it is more reasonable to treat third plural arbitrary pro as a true null subject, since these are [-OSP] languages. Assuming this kind of approach is correct, one would expect to find differences not only in the distribution of arbitrary third plural pro but also in its ability to be overt. That is, further typological investigation should show that the generalization

NULL SUBJECT PHENOMENA

327

of the preceding paragraph is too strong as it stands. One might in particular expect that whereas in Russian it should be possible in some styles to replace this pro with an overt pronoun and still retain its arbitrary reading, in the South and West Slavic languages this should be no more possible than in Spanish. Although I have not explored this avenue of research in any systematic way, informal discussion with speakers indicates that this is indeed correct. Russian speakers consulted state that the presence of overt oni 'they' is still consistent with the arbitrary interpretation. That is, contrary to the way the Russian arbitrary third plural construction is usually described, an overt pronominal subject does not force a referential or specific reading.42 In South and West Slavic, however, the situation is quite different—SerboCroatian and Polish speakers were unequivocal in their conviction that overt oni necessarily gives rise to the referential interpretation. This previously unobserved contrast between the Slavic language types supports the approach to pro-drop adopted in this chapter. Another related point where there is a difference between these two language types has to do with Montalbetti's (1984) Overt Pronoun Constraint, which is adopted by Jaeggli (1986b, 64) in his account of arbitrary plural pronominals: (83)

Overt pronouns cannot be linked to formal variables iff they occur in a position where the alternation overt/empty is possible.

This constraint is used to explain why he in English (84) can have a bound variable interpretation, whereas its Spanish counterpart requires the phonologically null pro under this reading. (84)

every student thinks that he will get an A

Jaeggli (1986b, 66) formally unifies Montalbetti's (83) with his own characterization of arbitrary plural pronominals by means of the bidirectional implication in (85). (85)

Overt pronouns may not be arbitrary in reference iff the overt/empty alternation obtains.

We have seen that whether this is observationally correct for Slavic depends both upon how broadly the expression "the overt/empty alternation" is interpreted and upon how the Russian arbitrary third plural construction is treated. Crucially, in order for (81) to apply properly, the relevant pro-drop parameter must be the OSP and overt oni in Russian must be understood as not forcing a referential reading. The generalization in (83), however, clearly gives the correct results for the relevant Slavic examples. As Sperling (1990) points out, whereas Russian (86a) can admit the bound variable reading, the otherwise identical Czech (86b) and Serbo-Croatian (86c) cannot. (86)

a. kazdyj student dumaet, cto on polucit pjaterku 'every student thinks that he will get an A' b. kazdy student mysli, ze on dostane jednicku c. svaki student misli da ce on dobiti desetku

Without the pronoun on 'he', of course, the bound variable reading obtains in all three languages. Here, then, Russian behaves like English, presumably because

328

PARAMETERS OF SLAVIC MORPHOSYNTAX

neither is pro-drop (i.e. both are [ + OSP]), whereas Czech and Serbo-Croatian behave like Spanish, since all three languages typically drop unemphatic pronominal subjects (i.e. are [-OSP]).43 In terms of the OSP defended in this chapter, it appears that a principle along the lines of Chomsky's (1991) Principle of Least Effort may be at work here: in a [ - OSP] language, case is not assigned to subject position until it has to be—at LF— unless there is some specific reason to assign it earlier. Emphasis, with concomitant phonological repercussions, is one instance of S-structure nominative assignment. Referentiality (or perhaps definiteness or specificity) is apparently another, at least to the extent that failure to assign nominative by S-structure in the bound variable and arbitrary third plural constructions necessarily leads to a non-referential interpretation. This suggests a conceptually satisfying way of unifying the two pro-drop phenomena described by Jaeggli (1986b). Notice, finally, that one would expect Galician Portuguese, colloquial Czech, and Serbian not to allow overt 'they' to be arbitrary, since these languages are [ - OSP] and—if I am correct—their [ + OESP] status should be irrelevant. Although I have not been able to check this prediction for Serbian, speakers of the other languages confirm that despite the presence of overt expletives, there are no overt arbitrary third person pronominals.44 This state of affairs strongly suggests that Jaeggli's attempt to tie the omission of arbitrary plural pronominals with the omission of bound variable pronominals requires a highly specific sense of what constitutes ' 'the overt/empty alternation." In conclusion, the behavior of bound variable and arbitrary plural pronominals provides yet another dimension of support for the partitioning of Slavic null subject phenomena adopted in this chapter and, in particular, for not treating Russian as pro-drop in the sense of the f ± OSP] parameter. Notes 1. This chapter is a revised and considerably expanded version of Franks (1990c). I am indebted to Elsevier Science Publishers for permission to reprint this material. 2. Bouchard (1982/1984) takes this idea to its logical limit, arguing that all aspects of empty categories must follow from principles based on overt categories. Thus, nothing in the grammar may refer exclusively to empty categories, their distribution and properties deriving entirely from independently motivated principles. 3. See Ovsjaniko-Kulikovskij (1900) for an insightful early discussion of the different types of null subject constructions found in various languages. His division of the possibilities into two major classes, whether a subject is psychologically ' 'perceived" or not, is very much like the conclusion reached in GB that there are thematic and non-thematic null subjects. See also Galkina-Fedoruk (1958) for the classic treatment of Russian "impersonal sentences" (bezlicnye predloienijd). 4. See Chomsky (1981a) or van Riemsdijk and Williams (1986, 300) for Italian examples of these phenomena and discussion of how the apparent lack of ECP effects might follow from Rizzi's Pro-Drop Parameter. 5. This example is drawn from Lasnik and Saito (1984). See also Rudin (1988a, 1988b) for a detailed presentation and analysis of the Slavic data relevant to multiple and long-distance w/i-movement. Boldface e is genetically used to represent any empty category, including traces, pro, and PRO. 6. See also Chvany (1973, 272), who comments that "Russian personal pronouns nor-

NULL SUBJECT PHENOMENA

329

mally appear in surface structures" except under certain contextually dependent optional deletions, but adds (n. 11) that "Russian is unlike Polish or Latin, where a personal pronoun in surface structure carries emphasis." 7. The differing status of overt expletives among the various languages is the focus of the latter part of this chapter. 8. Chomsky (1981a) proposes that subjects of weather-verbs are "quasi-arguments" and do in fact bear a kind of semantically vacuous theta-role. Catherine Chvany (personal communication) also points out that in poetic styles certain verbs may take overt subjects that do not ordinarily allow them, for example, dal' svetaet 'the distance is dawning' or sneg znobit nogi 'snow shivers the feet'. In the first instance the location dal', which would ordinarily be expressed prepositionally (v dali 'in the distance'), appears as the subject. This can be understood as involving a marked association of lexically available theta-roles and syntactic positions (cf. Grimshaw, 1990; Jackendoff, 1990a). In the second instance, a verb that ordinarily only takes an object is used transitively. Here the inventory of theta-roles is in fact manipulated, so that this verb znobit' has its own (derived) semantic structure and is really a distinct lexical item—which probably explains why speakers disagree about the acceptability of such constructions. 9. Recall that a variety of Russian and Polish impersonal constructions were cited in chapter 3. See chapter 8 for further discussion of the dispositional construction in Slavic, as well as Riizicka (1986, 1988). 10. But see Travis (1984) for an interesting cross-linguistic study of null subject possibilities, in which she proposes that the extent of a language's ability to omit subjects depends on the richness of inflection in that language; this would mean that a [ - OESP] language has moderately rich inflection and a [ + NDP] very rich. 11. This problem of how Russian and Ukrainian children are able to fix their grammars to require overt subject pronouns is more serious, since expletives are far rarer in these languages and, unlike in (verb-second) German, never required. See Lebeaux (1987) for further objections to Hyams's system and alternative solutions in terms of the possibilities that (i) subject case-marking strategies differ for adults and children, and (ii) apparent' 'subjects'' for "pro-drop" children are really topics. 12. See also Jaeggli and Safir (1989a) for discussion. 13. Feldstein (1987) suggests that whereas Polish and Serbo-Croatian conjugational agreement morphology submit to phonological segmentalization with relative ease, it is impossible to analyze Russian conjugational forms in any similarly transparent fashion. 14. Polish is in fact also mixed in the sense that although it freely drops subjects, making it [ + NDP] in the Safir system presented earlier, it shows ECP effects and does not allow their long-distance extraction, making it [ - FIP]. See Lasnik and Saito (1984), Giejgo (1981), and Rudin (1988a, 1988b) for discussion of ECP and w/!-movement in Polish. 15. The non-pro-drop status of Russian is partially obscured by the effect of ellipsis, as discussed in section 7.2.2. The asterisks in (28) must therefore be understood to mean that pro-drop is ungrammatical unless contextually sanctioned. 16. Kosta (1990), which presents a broad overview of null subject phenomena in Russian and includes a comprehensive bibliography of works on subject omission in Slavic, also discusses some of the examples in (30)-(32). Note that the judgments are Chvany's; other, presumably less conservative, speakers have judged (31b, c) as perfectly acceptable and (32b) as questionable. 17. Several other linguists have argued for the independent status of a morphologically motivated Case Filter. See, for example, Milsark (1988) and the references therein. 18. Recall that lexical insertion is constrained by non-distinctness rather than identity in

330

PARAMETERS OF SLAVIC MORPHOSYNTAX

morphosyntactic features. See Lumsden (1992) for a similar theory that makes crucial use of underspecification. 19. See also Franks (in press) for discussion of this parameter, where it is called the Null Expletive Subject Parameter. Here I have recast the parameter in terms of overtness out of markedness considerations. 20. The arguments in this section hold even if, as in chapter 6, the secondary predicate is actually predicated of a PRO small clause subject, which itself is bound by an empty subject antecedent, since this merely extends the chain of identical indices. 21. See Lasnik (1992b) for arguments that PRO can never be an expletive. 22. See, for example, Baker (1988), who argues that incorporation also serves this purpose. 23. Examples (41a-c, e-f) are based on the texts and commentary in Nahimovsky and Leed (1987); example (41d) is drawn from Woolhiser (1991). 24. As discussed in the last chapter, the imperfective future in Russian is periphrastic, consisting of byt' 'to be' plus the imperfective infinitive. In morphological terms, the forms budu, budes', etcetera, are simply perfective conjugated verbs (i.e. V + AGR), and I therefore analyze them as such syntactically. 25. In section 7.4 I discuss another possible reason for case assignment in constructions such as (48c). 26. Thanks are due to Martina Lindseth for collecting the Sorbian examples, described in Lindseth (1991,1993b). Lindseth (1993b, 4) comments that "the different forms wono, wane, won are distributed according to dialect areas. Wono is typical for the North-Eastern dialect of Upper Sorbian (and for Lower Sorbian), won for the South-Eastern area and wane for the other Upper Sorbian dialects." See also Schuster-Sewc (1974) for discussion. 27. The literary form is ono; in the spoken language there is a prothetic glide. Several of the colloquial Czech examples were collected by Curt Woolhiser; see also Lindseth (1991, 1993b). 28. Lindseth (1991), following Schuster-Sewc (1974), notes that this is comparable to the use of to (and also wono) in Lower Sorbian. 29. I have glossed to as 'that', but this little word has a variety of subtle functions that complicate the analysis in this chapter. It seems to be some kind of emphatic adverb that often co-occurs with the overt expletive, as in (57b) and the last clause of (57c). The first occurrence of to in (57c), on the other hand, is demonstrative and has a deictic function, corresponding to English that or Russian eto. 30. Lindseth (1993b) states that comparable examples with expletive (v)ono after interrogatives are pragmatically inappropriate: (i)

*kde je ono chladno 'where, oh boy, is it cold?'

31. Lindseth (1991) cites Schuster-Sewc (1974) as claiming that Upper Sorbian also displays the same behavior as colloquial Czech with respect to theta-marked subjects. She states that some grammars point out that Upper Sorbian wono should not be used if there is another subject in the clause; this is why the examples in (49) were all of the meteorological variety. Lower Sorbian, on the other hand, appears to be just like Galician instead. Lindseth (1993b) claims that Upper Sorbian expletives of the agreeing type are actually no longer acceptable. Her informants uniformly rejected Schuster-Sewc's (1974) example: (i)

woni su to Mupi ludzo, hizom wescili it(pl nom) are that stupid(nom pi) people(nom pi) already predict 'stupid people have predicted that before'

NULL SUBJECT PHENOMENA

331

Even acceptability of non-agreeing expletives with nominative subjects was limited to some speakers: (ii)

won tarn sedzi zona na iawce 'there is a woman sitting on the bench'

All this suggests that there are two typologically very distinct kinds of overt expletive prodrop languages, with Czech and Upper Serbian developing in different directions. Unfortunately, the implications of these data are too complex to be pursued here. 32. Some of these examples were collected by Curt Woolhiser and Jonathan Ludwig in 1990 and 1991 course work; the Russian dialect examples are from Sapiro (1953). 33. In a sentence like that John went home upset Mary the D-structure clausal subject is regarded as an S-structure topic, binding a trace in subject position. This raises the question of why expletive it does not appear, since it is obligatory when the subject sentence is extraposed. This difference between the two constructions suggests an ECP account: under extraposition the case-marked trace is not properly governed, but under topicalization it is. 34. Of course, given the Principle of Full Interpretation, these clauses would still undergo LF expletive replacement. 35. It is possible that East Slavic dialects that allow overt expletives in all contexts, and are otherwise not pro-drop (i.e. [ + OSP]), are in fact [ + OESPJ and only delete expletive subjects for the same discourse-ellipis reasons that thematic subjects can be dropped. The general viability of overt expletives is not clear on the basis of the textual examples I have found, however; the exact status of expletives in these colloquial languages is in need of much further study. 36. I have nothing substantive to saying regarding the fact that it is the expletive eto rather than ono that appears in these examples. The former seems more comparable to English it and the latter to English there in expletive contexts, but it is similarly unclear what differentiates these two expletives in English. It is worth noting, in addition, that ono sometimes appears even in colloquial Russian in such examples as: (i)

ono i po roze bylo vidno, cto ne p'es' it and by snout was(n) visible that NEG drink 'it was even visible from your face that you don't drink'

37. This raises the question, however, of why overt expletives can appear in CHAINs with infinitivals in the East Slavic examples in (59}-(61). Since these dialects seem to allow overt expletives in any context, they are presumably [ + OESP]. 38. In these examples I have abstracted away from word order effects for ease of exposition; since the clitic cluster cannot appear first, in the absence of to some other full element must appear to the left of the clitics. Also, there is another construction without the copula in (72a) and (73a), although for some reason to in Polish (73a) becomes completely acceptable if jest is left out and to in Serbo-Croatian (72a) becomes completely unacceptable without the copula je. 39. Another complicating factor, pointed out to me by Ljiljana Progovac (personal communciation), is the existence of referential adverbial to, as in Serbo-Croatian da li to Milan dolazi? 'Is that Milan coming?' or to Milan samo blefira 'Milan is only bluffing'. 40. These and several other examples cited in this section are based on Sperling (1990). 41. Katarzyna Dziwirek (personal communication) draws my attention to several important differences between the third plural and reflexive types of arbitrary reference construction. Whereas the former implies necessarily more than one person involved, the latter is primarily modal in meaning and in no way implies plurality of the human subject involved. In addition, as Wayles Browne (personal communication) draws to my attention, the reflexive type (79b)

332

PARAMETERS OF SLAVIC MORPHOSYNTAX

can include the speaker, but the third plural type (79a) does not. Yet a third type is the -nol-to construction to be discussed in chapter 8, when the structure and range of various "passive" and "reflexive" constructions in Slavic is treated. These thus seem to require different kinds of unspecified human subjects. Note that speakers made no similar objections to the equivalence of the items in Serbo-Croatian (80). 42. The same is true for the other arbitrary personal constructions mentioned at the beginning of this section. 43. It is interesting also to observe that the overt pronoun/bound variable pro contrast correlates with a host of other domain-extending properties in Serbo-Croatian. Contrast, for example, the acceptability of topicalization of to 'this' in (i)-(iii). (i)

to Cj ne zelim da e; potpisem this NEG want(lsg) that to-write(lsg) 'this, I do not want to write'

(ii)

?*to Cj ne zelim da Cj potpise this NEG want(lsg) that to-write(3sg) ' this, I do not want (him/her) to write'

(iii)

*to e; ne zelim da oHj potpise this NEG want(lsg) that he to-write(3sg) ' this, I do not want him to write'

See Progovac (1993), who presents comparable examples involving phenomena such as clitic doubling and negative polarity. 44. Thanks are due to Juan Uriagereka and Jindfich Toman for discussion of Portuguese and Czech, respectively. Uriagereka notes that in some Brazilian dialects eles 'they' may be able to have the arbitrary reading, but suspects that this correlates with the ongoing loss of pro-drop in Brazilian Portuguese. One would similarly expect that overt ele 'he' in these dialects should be able to function as a bound variable.

8 Voice Alternations

This chapter investigates voice-altering morphemes in the various Slavic languages and their interaction with null subjects in finite clauses.1 After a brief consideration of the general nature of passive and middle voice, the bulk of the chapter treats the properties and distribution of various Slavic constructions involving "passive" participial and reflexive morphology. The canonical passive pattern is for there to be a nominative subject bearing the identical semantic role that the accusative object would have in the active counterpart. We shall see, however, that various Slavic languages and dialects also permit the object to remain in the accusative, while still adding the same passive morphology and prohibiting the agent from appearing in subject position. A typical example is given in Ukrainian (1). (1)

e bulo prodytano knyzku AUX(n) read(n) book(acc) 'a book was read'

For convenience, I shall refer to these unusual constructions as a whole by the cover term "passive + accusative," regardless of whether the morpheme involved is reflexive or participial and even of whether the interpretation is truly passive or not. The existence of this ' 'passive + accusative'' construction raises the questions of what kind of empty subject is involved and why the option is only present in certain languages. I shall argue that it is potentially available in all the Slavic languages, since as we saw in chapter 7 they all allow null expletive subjects. However, various factors having to do with specific properties of the "passive" (i.e. reflexive or participial) morphemes in the individual languages conspire either to admit or preclude its instantiation. An initial proposal is that the cases a morpheme can "absorb" should be parameterized, following the Case-Absorption Parameter originally proposed in Franks (1982). It is then argued that this parameter must be generalized so as not to refer to specific cases. This approach is augmented by the idea that 333

334

PARAMETERS OF SLAVIC MORPHOSYNTAX

morphological rules may apply in the syntax as well as the lexicon, so that participial and reflexive morphemes are in some instances able to function either as (lexical) word-formatives or as (syntactic) arguments. The analysis is then adapted to take into account the VP-internal subject hypothesis and extended X-bar models that have proved useful in explaining other phenomena treated in this book. Finally, the mysterious "dispositional reflexive construction" illustrated by Russian (2) is discussed. (2)

mne ne rabotaet-sja me(dat) NEG work -REFL 'I don't feel like working'

8.1. Voice This section develops those aspects of theta-theory that will be necessary to understand the properties of the voice-altering morphemes discussed in the remainder of the chapter. Their behavior will be interpreted in terms of an interaction between the range of empty subjects discussed in chapter 7 and the manipulation of the verb's lexical predicate-argument structure.

8.1.1. Predicate-Argument Structure Theta-theory consists essentially of two general principles, requiring (1) that there be a one-to-one correspondence between (syntactic) arguments and (lexically specified) theta-roles, and (2) that this correspondence be respected at all syntactic levels of representation. These requirements are known as the Theta-Criterion and the Projection Principle, respectively, and together conspire to restrict derivations greatly, since categorial structure must be consistent at all syntactic levels in order to reflect the underlying mapping of theta-roles onto arguments. A verb can be regarded as a semantic predicate that occurs with a set of arguments bearing specific theta-roles. The nature of these roles is presumably derivable from the semantics of the verb; for example, Russian dat' 'to give' takes of necessity a GIVER, a GIVEN, and a GIVEE, which can be generalized to the roles agent, theme, and goal, respectively. I therefore assume that knowledge of a word's meaning allows the deduction of a set of appropriate theta-roles, which are given for each entry in the lexicon, roughly as follows: (3) dat' (Agent, Theme, Goal) Notice that this highly schematic predicate-argument structure ignores the fundamental problem of how these roles come to be associated with specific arguments. There are a number of conceivable approaches to the pairing of semantic roles and grammatical relations. Some, such as the model espoused in Babby (1989, in press), posit lexically stipulated pairings for individual items. These entail an onerous acquisitional burden, in that the child must learn much idiosyncratic information about each word (independent of its meaning) in order to use it correctly. Since the

VOICE ALTERNATIONS

335

simplest hypothesis is that there is little in early lexical learning that transcends meaning, we explore here a different kind of solution that avoids this pitfall and attempts to generalize the strategies for associating lexical representations as in (3) with syntactic ones. One currently popular approach to the problem of mapping lexical predicateargument structures onto syntactic phrase structure trees, based on work by Larson (1988) and developed in Speas (1990), for example, maintains that this process is mediated by a universal thematic prominence hierarchy of the type employed in chapter 3. The theta-roles are associated with syntactic positions such that the lower a role is on the hierarchy the lower it is in the tree. If so, no special pairings between grammatical functions and thematic roles need to be stipulated. I do not explore the ramifications of this kind of model here, as it requires considerable elaboration of the phrase structure representation; see Franks (1993) and the references therein for discussion. Even within such systems, it is often argued that a minimal enrichment of the kind of representation in (3) is independently necessary.2 Two theta-roles are singled out for special treatment. First of all, as argued in Williams (1981), it is necessary to distinguish theta-roles assigned external to the VP and those assigned internal to it. A verb may assign at most one theta-role externally, which is associated with the subject position, traditionally regarded as SPEC-IP.3 In (3) this would be the agent, which we can indicate (as Williams does) by underlining: (4) dot' 'to give' (Agent, Theme, Goal) I shall continue to use the term "external theta-role," although it should be kept in mind that the model adopted in this book analyzes the subject as originating in SPEC-VP. In section 8.3.4, the analysis of voice will be modified to take this adjustment into account. Second, as discussed in chapter 2, a verb can assign at most one theta-role syntactically to some argument.4 That is, the verb selects from the lexical set of internal theta-roles one privileged role to be literally incorporated into the syntactic representation of that verb and assigned to some complement via coindexation. For daf this will be the theme role, which becomes part of the syntactic node that is the verb. As argued in chapter 2, a node (such as V) is actually an abbreviation for a matrix of morphosyntactic features. This matrix contains various kinds of morphosyntactic information, including at least the word's part-of-speech, morphological form (determined either lexically or syntactically through government or agreement), and, it was proposed, a theta-role. The form dast 'will-give(3sg)', for example, might have features as in (5), leaving out superfluous detail.

336

PARAMETERS OF SLAVIC MORPHOSYNTAX

Recall that features pertaining to each category are grouped together into submatrices to allow processes that involve them, such as agreement and government, to be stated directly on the appropriate category. This system is extended to include the syntactically assigned theta-role. This role is assigned to the verb's direct object through a process of coindexation, such that the index on the theta-role submatrix is identified with that of the complement. In the lexical representation in (5), the theme theta-role can thus be enclosed in angled brackets to reflect the fact that it is this role that becomes part of the verb's morphosyntactic feature matrix, as follows: (6) dot' 'to give' (Agent, , Goal) I shall refer to these two special theta-roles as the external role and the internal role, respectively, bearing in mind that (1) in point of fact the "external" role is really assigned to the argument in VP-specifier position, (2) it is the positions rather than the roles themselves that are "external" or "internal," and (3) only one of the "internal" roles is directly assigned by the verb. While this model obviously leaves unanswered many residual issues, it provides the prerequisite groundwork for our consideration of how voice-altering morphemes affect the association of roles with positions, to which I now turn.5 8.1.2. Standard Passive Constructions Traditional transformational grammar holds that passive constructions are derived through syntactic movement. A transformational process moves the D-structure object to subject position, leaving a coindexed trace behind, as in (7). (7)

the book; was sold 6; (by the store) (to make money)

The position occupied by the trace is required to be present even in the derived S-structure representation, since if the verb assigns the theme theta-role to direct object position, then by the Projection Principle it does so at all levels of representation. Under standard assumptions, the passive morpheme "absorbs" objective case from the verb, thereby rendering it incapable of assigning case. Hence, the book must move to subject position to receive nominative in order both to be overt and to have its theta-role licensed, assuming the dual function of case for both PF- and LFvisibility argued for in chapter 7. Consequently, the book and its trace e form a kind of discontinuous element known as a chain, sharing a single case and theta-role. Notice, crucially, that the lexical predicate-argument structure of the verb is essentially preserved in that the internal theta-role at all points in the derivation remains as such. This property is one of the hallmarks of the syntactic passive construction. Moreover, the sentence remains semantically agentive, as demonstrated by the possibility of appending an optional fey-phrase agent or purpose clause. Other voice-altering operations, however, literally rearrange lexical predicateargument structure. Consider, first of all, the middle construction in (8), which can be analyzed as involving a lexical voice-altering operation. (8) the book sold easily

VOICE ALTERNATIONS

337

No agent phrase or purpose clause may be appended. (9) the book sold easily (*by the store)/(*to make money) Pagan (1988) argues that in middles the internal (theme) theta-role is lexically externalized, so that it is directly assigned to subject position, and, concomitantly, the external (agent) theta-role is of course no longer associated with this position. Adapting ideas in Rizzi (1986) about implicit object constructions, such as (10), in which the object can be freely omitted without affecting the interpretation of the sentence, Pagan proposes that the external theta-role is lexically assigned arbitrary reference in middles, "saturating" it (using it up) and thereby rendering it unavailable for syntactic purposes. (10)

this sign cautions (people) against avalanches

This would express the fact that the agent is understood, but nonetheless syntactically inert. Schematically, under this analysis middle formation is a process that takes place in the lexicon to manipulate predicate-argument structure as follows: (11)

(Agent )

> (Theme Agent^)

If the lexical operation is as in (11), then middles, such as (8), do not involve any syntactic movement. Alternatively, middle formation might consist of saturation of the agent role without externalization. The relevant lexical operation would then be as in (12). (12)

(Agent )

> ( Agent^)

If so, although middles still differ from passives in that they involve manipulation of predicate-argument structure in the lexicon, the two might still be similar in that the theme raises to subject position in the syntax. Deciding between these two options is no straightforward task, since the evidence to be considered in this chapter seems compatible with either approach. This externalization analysis raises the issue of why ordinary passives are not analyzed in a similar fashion—they could conceivably arise, like middles, through lexical externalization of the internal theta-role. Indeed, adjectival passives do directly assign the theme role to subject position, as in (13), where no agent phrase or purpose clause is possible.5 Compare (13) with (9). (13)

the book was unsold (*by the store) (*to make money)

The passive participle sold is regarded as a defective verb with only the feature [ + V]. However, since un- attaches to adjectives (there is no verb *to unsell), there must also be a true adjective sold with the features [ + V, +N]. And since this adjectival passive form acts like a predicate, assigning the theme theta-role to its subject, its creation presumably involves externalization. A form such as sold is therefore ambiguous as to whether it is lexically or syntactically derived; compare the book was sold by the store. How then can we tell that the standard passive in (7) is actually syntactic? The best argument is that it contrasts to lexical passives in that the verb's predicate-argument structure is unchanged, so that the internal theta-role is still assigned by the participle

338

PARAMETERS OF SLAVIC MORPHOSYNTAX

and the external theta-role is still syntactically active. Middles and adjectival passives must be lexical. Otherwise, they would lead to a Projection Principle violation, since the D-structure projected from a lexical representation in which the agent is associated with subject position clearly would not be maintained. However, when this reasoning is applied to syntactic passives a potentially serious problem arises: Why does movement from object to subject position not give rise to a Projection Principle violation as well, since this position is specified in the lexicon as receiving the external theta-role? In other words, what happens to the external theta-role associated with subject position in (7)? Some sort of lexical "rearrangement" is apparently called for here as well. Intuitively, syntactic passive is a way of altering voice relations by suppressing the external theta-role without saturating or eliminating it. If passive involved some kind of demotion or internalization of the agent role, then it would violate the Projection Principle. One method of preserving the integrity of this principle and still capturing the spirit of the difference between passive and middle constructions has been developed in recent Government and Binding approaches to the passive. Jaeggli (1986d) and Baker, Johnson, and Roberts (1989) argue that passive morphology "absorbs" the external theta-role, where this does not preclude the role's expression by an adjunct phrase. They claim that the passive morpheme shares the argument status of a /ry-phrase if present, somewhat akin to clitic doubling. This analysis is intended to satisfy the objection that an operation that internalizes the external theta-role must be lexical, otherwise it would violate the Projection Principle. Let us thus tentatively adopt the idea that syntactic passive morphology absorbs both external theta-role and objective case, where this can be construed to mean that it blocks their regular assignment. Baker, Johnson, and Roberts (1989) offer interesting motivation for the observation that passive morphology absorbs not only the case the verb assigns but also its external theta-role. In their theory, the passive morpheme is a true syntactic argument, receiving both case and theta-role. That is, like any other argument, it requires a theta-role to satisfy the Theta-Criterion and case to render this theta-role visible. Voice-altering morphemes thus seem to have a life of their own, participating in syntactic processes and satisfying syntactic requirements. Borer and Wexler (1987) have argued that this state of affairs exists because morphological rules are not constrained to apply only in the lexicon. Instead, they may also apply in the syntax, up to Projection Principle violations. They therefore maintain that the passive morpheme will have different properties depending on the component in which affixation takes place. If it occurs in the lexicon, according to Borer and Wexler, "passivization" can (1) change the category of the verb (f-N]) into an adjective ([ + N]), (2) eliminate the subject theta-role, and (3) externalize the internal theta-role. True syntactic passivization can do none of these things without violating the Projection Principle. Nonetheless, there remain several disturbing aspects of the Baker, Johnson, and Roberts analysis. For one thing, it seems unlikely on typological grounds that the passive morpheme in English (and other languages) could be a clitic doubling the nominal argument in the fry-phrase. This is especially true since passive morphology is realized by various kinds of morphemes, such as participial suffixes and reflexive

VOICE ALTERNATIONS

339

clitics in Slavic, and the demoted subject argument is also expressed in various ways, such as a bare instrumental or object of a preposition in Slavic. It is of course also typically left unexpressed. There is no realistic morphological sense in which the participial or reflexive morpheme might be consistently taken as a doubled pronominal clitic. For another thing, even this approach does not completely resolve the Projection Principle problem discussed above, since the external theta-role is now for some reason assigned to the -en—by-phrase chain rather than to the argument in subject position. This raises the important question of what enables the passive morpheme to absorb the subject theta-role. One cannot simply claim, for example, that it is itself the subject, since the only truly subject property of the passive morpheme is its hypothetical theta-role: it is by most criteria neither nominative nor in subject position.7 It is additionally unclear why subject theta-role and object case should be associated via absorption by the same morpheme. Perhaps these issues could be clarified through a more explicit characterization of the position(s) in which the passive morpheme is generated and the processes by which it (1) absorbs/receives case, (2) absorbs/receives a theta-role, (3) becomes identified with the by-phrase, and (4) attaches to the verb.8 Despite these caveats, I will continue to assume some version of the standard case absorption model in the discussion of passive until section 8.3.4, where the analysis is modified in light of the VP-internal subject hypothesis.

8.2. Null Subjects and Passive This section expands on arguments developed in chapter 7 against treating so-called pro-drop phenomena as indicative of a unitary parameter. Data from passive and middle constructions in various Slavic languages are presented that contradict the GB analysis of parallel phenomena from Italian, which I first briefly recapitulate.9 8.2.1. Si-Constructions in Italian Chomsky (1981a, 270-271) discusses the following three impersonal sentences in Italian, all glossed by him identically: (14) a. si mangia le mele REFL eats(3sg) the apples 'the apples are being eaten' b. le mele si mangiano the apples REFL eat(3pl) c. si mangiano le mele REFL eats(3pl) the apples All presumably have the underlying structure in (15). (15)

[NP ej si mangia [NP le mele]

Chomsky (1981a) relates the possibilities in (14) to the pro-drop status of Italian. According to him, the canonical (14b), with a nominative VP-external subject and

340

PARAMETERS OF SLAVIC MORPHOSYNTAX

the expected plural number agreement, would be the only option in a non-pro-drop language such as English. This is so because such a language requires that subject position in finite clauses be lexically filled. Both (14a) and (14c), on the other hand, have empty subjects. The three S-structures are thus roughly as in (16). (16) a. [NP e] [VP si-mangia le mele] b. [NP le mele] [VP si-mangiano e] c. [NP el [VP [VP si-mangiano] le mele] Following Rizzi (1982), Chomsky claims that si is coindexed with the subject position, whose theta-role it thereby absorbs.10 Similarly, si (like passive morphemes in general) is able to absorb case-marking. In (16b, c) si absorbs objective case and in (16a) nominative, preventing V and AGR from case-marking object and subject position, respectively. This means that le mele 'the apples' is nominative in (14b, c), but objective in (14a)—a reasonable result, assuming that subject-verb agreement is only possible with nominative subjects. Chomsky's reasoning about the inadmissibility of (14a, c) in non-pro-drop languages seems to be as follows: si can in principle absorb either case; if it absorbed nominative in a language like English, however, an illicit empty subject would result; hence, it must absorb objective, producing (14b). Notice, however, that this assumption alone does not suffice to rule out (14c), since here objective rather than nominative is absorbed. Making (14a) and (14c) viable in a pro-drop language requires two distinct sets of assumptions: in (14a), si absorbs nominative and null subjects are permitted; in (14c), si absorbs objective, null subjects are permitted, and nominative assignment to VP-internal arguments is possible. As argued in chapter 7, however, the second and third properties are independent, with the one requiring [ + NDP] and the other [ + FIP], in the system of Safir (1985). In addition, we have seen that [ ± NDP] is itself not a unitary parameter and must be broken down at least into [±OESP] for expletive subjects and [±OSP] for theta-marked ones. Presumably it is [ + OESP] that is relevant here, since a language must allow expletive subjects in order to generate constructions such as in (14a) and (14c). For these reasons, we do not expect that the only two pro-drop possibilities should be as in Italian or English. A survey of the way these facts manifest themselves in Slavic bears this prediction out and, moreover, demonstrates the need to allow for some variability in the auxiliary properties of voice-altering morphemes. 8.2.2. Some Slavic Variations In this section the idea of relating the "passive + accusative" construction to the status of pro-drop in a language is discussed. Data from various Slavic languages are presented to argue that the only relevant aspect of pro-drop is whether or not the language is [ + OESP]. A cursory comparison of Russian and Polish would seem to reveal the English versus Italian pattern. Recall that in general terms Russian can be regarded as a nonnull subject language and Polish can be regarded as a null subject language; in terms of the parameters of chapter 7 Russian is f + OSP] but Polish is [ —OSP]. If there were a monolithic Pro-Drop Parameter on which the range of passive-like construe-

VOICE ALTERNATIONS

341

tions in a language depended, as in the Rizzi-Chomsky proposal, then we might expect Russian to have only -sja constructions parallel to (14b), but Polish to have siq constructions like (14a) also.11 The examples in Russian (17) and Polish (18) would appear to corroborate this. (17)

a. Ivan prodaet knigi (nom) sells(3sg) books(acc) 'Ivan sells books' b. *e prodaet -sja knigi sell(3sg)-REFL books(acc) c. knigi prodajut-sja books(nom) sell(3pl)-REFL 'books are being sold/are selling'

(18)

a. Jan sprzedaje ksiazki (nom) sells(Ssg) books(acc) 'Jan sells books' b. e sprzedaje si? ksiazki sells(3sg) REFL books(acc) 'books are being sold' c. ksiazki sie sprzedaja books(nom) REFL sell(3pl) 'books are selling'

The Rizzi-Chomsky analysis would claim the ungrammaticality of Russian (18b) is due to the empty subject position. There are, unfortunately, at least three major problems with this type of approach. First of all, as shown in chapter 7, the theoretical assumptions that motivate it are misguided, since it is impossible to accommodate all the phenomena originally regarded as ' 'pro-drop'' within a single parameter. Second, the purely syntactic account falsely predicts that Polish (18b) and (18c) should be synonymous, since the Projection Principle prohibits movement from altering predicate-argument structure. In fact, however, they differ significantly in meaning: (18b) is essentially passive, with an implicit agent, while (18c) is middle. Since they have different semantic structures, and D-structure is a pure representation of the association between arguments and theta-roles, (18b) and (18c) must have different D-structures.12 The third and most damning reason why the null subject account of the illformedness of Russian (17b) is inadequate for Slavic is simply that there is no consistent correlation between pro-drop and the existence of the passive + accusative construction in any given language. As observed in Franks (1982,143), standard (stokavian) Serbo-Croatian patterns like Russian in only allowing (19a), but many kajkavian dialects also allow the passive + accusative (19b), superficially like Polish.13 (19)

a. kolaci se jedu cakes(nom) REFL eat(3pl) 'cakes are being eaten'

342

PARAMETERS OF SLAVIC MORPHOSYNTAX

b. e jede se kolace eats(3sg) REFL cakes(acc) The construction in (19b) is acceptable only in kajkavian Serbo-Croatian, even though the two dialects are identical in normally omitting both theta-marked and expletive subject pronouns, hence are [ - OSP] and [ - OESP]. In this respect kajkavian Serbo-Croatian resembles its close neighbor Slovenian. Toporisic (1984, 294— 295) cites examples such as (20). (20)

a. isce se mlajsa zenska seek(3sg) REFL younger(nom) woman(nom) 'a younger woman is being sought' b. isce se mlajso zensko14 seek(3sg) REFL younger(acc) woman(acc)

The availability of the passive type in (19b) and (20b) thus does not automatically follow from the absence of subject pronouns, contra claims made by the RizziChomsky model. Similarly, the fact that a language is [ + OSP] does not mean that the passive + accusative option cannot be exercised. In her survey of passive in Slavic, for example, Siewierska (1988) observes that a number of Slavic languages, including Ukrainian and Belarusian, resemble Polish in expressing the underlying object in the accusative in the passive voice. In this chapter I concentrate on Ukrainian. As Sobin (1985) points out and Shevelov (1963,139-146; 1969) discusses at length, the -nol-to construction in Ukrainian allows accusative objects with passive verbs, as follows: (21)

a. knyzku bulo procytano book(acc) AUX(n) read(n) 'the book was read' b. koly zbudovano cju cerkvu when built this(acc) church(acc) 'when was this church built?' c. v teatri bude zaprovadzeno suvoru dyscyplinu in theater FUT(3sg) instituted(n) strict(acc) discipline(acc) 'a strict discipline will be instituted in the theatre'

The verb shows the default form in -o (the old neuter singular ending, as opposed to the modern -e), no overt subject is present, and the object remains in the accusative. However, in all other null subject respects Ukrainian patterns just like Russian— both are [ + OSP], [ - FIP], and [ - OESP]. Thus, the difference between the grammatical Ukrainian (21) and the ungrammatical standard Russian (22b) cannot be due to any difference in the setting of any parameter related to pro-drop. (22)

a. kniga byla procitana book(f nom) AUX(f) read(f) 'the book was read' b. *knigu bylo procitano book(acc) AUX(n) read(n)

VOICE ALTERNATIONS

343

North Russian dialects, however, do allow constructions like (22b).15 Such dialects use passive participial morphology to create intransitive statives, as illustrated in (23). (23) a. u menja uze vstato bylo by me already arisen(n) was(n) 'I had already gotten up' b. u menja vyspano -s' -to teper' by me rested(n)-REFL-EMPH now 'now I am rested' c. ego ubito na vojne he(acc) killed(n) at war 'he was killed in the war' These "passives" differ from their counterparts in Ukrainian in two significant ways. First, the subject is normally expressed using the "possessive" u + genitive construction (although some dialects also employ ot + genitive and/or bare instrumental). More importantly, as indicated by (23a) and (23b), passive participles in such dialects are also freely formed from intransitive verbs. This means that -no/-to in these dialects differs from standard Russian (or -en in English) in that it does not absorb accusative case: vstat' 'to stand' in (23a) is clearly intransitive and not a caseassigner, vyspat'sja 'to get enough sleep' in (23b) cannot (under traditional assumptions) assign accusative since it is a -sja verb, and ubit' 'to kill' in (23c) has no available accusative to be absorbed since it retains its accusative object. There are several other mysterious idiosyncracies of the North Russian passive that suggest additional complexity in the construction. Timberlake (1974,1976) cites examples in which the D-structure object is marked nominative but appears to remain in its underlying position and fails to induce subject-verb agreement, as in (24). (24)

pereexano bylo doroga tut crossed(n) AUX(n) road(f nom) here 'there's been crossing over the road there'

There are even mixed examples with the auxiliary showing gender/number agreement but the participle appearing with the default -nol-to ending; Timberlake (1976) regards these as "a historically transitional type." Since his examples are drawn from diverse dialects, it is difficult to tell what the range of coherent systems is. It would seem, however, that the nominative object in such dialects receives case in its D-structure position, as in Italian, although I have no independent evidence (such as extraction data) that these North Russian dialects are [ + FIP]. Even so, the failure of the participle (and auxiliary, except in the mixed type) to agree remains problematic. I suggest that the -nol-to participle (and auxiliary in some North Russian dialects) is a fixed non-agreeing predicative form. It is frozen as such hence never agrees with a subject, nominative or otherwise. It thus must be distinguished from the agreeing participial passive in the various languages. The reader is referred to Siewierska (1988, 269-275) for more detail on this distinction; compare also Doros (1975).

344

PARAMETERS OF SLAVIC MORPHOSYNTAX

Returning then to the passive + accusative participial construction in Ukrainian, note that agents are optionally expressed as instrumental adjuncts: (25)

a. z Vinnyci pryslano vistku sestryceju from Vinnycja brought(n) news(acc) sister(inst) 'news was brought from Vinnycja by the sister' b. mojim mylym mene zradzeno my(inst) beloved(inst) me(acc) betrayed(n) 'I was betrayed by my beloved' c. dim bulo zbudovano robitnykamy house(acc) AUX(n) built(n) workers(inst) 'the house was built by the workers'

In this regard the Ukrainian -nol-to construction in (25) does not differ from either the standard Russian participial passive or its Ukrainian passive counterpart with a nominative subject and agreeing passive participle. As expected, these also freely admit instrumental agents. (26)

a. list napisanyj studentamy letter(mnom) written(m) students(inst) 'the letter was written by students' b. cerkva bula zbudovana robitnykamy church(f nom) AUX(f) built(f) workers(inst) 'the church was built by the workers'

The examples in (25) and (26) show that in Ukrainian both accusative and nominative passive participle constructions allow optional instrumental agents. This is one respect in which Ukrainian differs from Polish, since no overt agent phrase (przez + accusative) is possible either in Polish "reflexive" passive (18b) or "reflexive" middle (18c). It can easily be shown, however, that an agent is nonetheless implicit in the Polish reflexive passive, since Polish passives and middles in siq differ with respect to the admissibility of purpose clauses: (27)

a. otwiera sie. kase^ (zeby sprzedawac bilety) 'the ticket window is being opened (to sell tickets)' b. kasa otwiera sie (*zeby sprzedawac bilety) 'the ticket window is opening (*to sell tickets)'

This difference is precisely what is expected if only (27a) is a true passive in the sense that it is syntactically created and the verb's lexical predicate-argument structure is preserved. The verb in (27b) is a middle voice formed lexically from a transitive verb by assignment of the "arb" index to the external (agent) theta-role of the active. In describing these phenomena I have alternated freely between reflexive and participial constructions, depending on what was most productive in the language at hand. This reflects my contention that the voice-altering properties of reflexive and

VOICE ALTERNATIONS

345

participial morphemes should be analyzed in a similar fashion. To complete the picture, however, let me summarize the relevant facts with respect to reflexive passives in Ukrainian and participial ones in Polish. It turns out that in addition to reflexive passives with nominative subjects, as in (28), some dialects of Ukrainian also admit them with accusatives, as in (29). (28)

a. rusnyky vze tkaly -s' towels(nom pi) already weave(pl)-REFL 'the towels were already being woven' b. cerkva budujet -sja robitnykamy church(nom sg) build(3sg)-REFL workers(inst) 'the church is being built by the workers'

(29)

a. e hotuvalo -sja nicnu ekspedyciju prepare(n)-REFL night(acc) expedition(acc) 'a night expedition was being prepared' b. e cerkvu budujet -sja robitnykamy church(acc) build(3sg)-REFL workers(inst) 'the church is being built by the workers'

These examples are formally exactly parallel to the participial constructions presented above, although they differ considerably stylistically. Shevelov (1963) comments, for example, that the type in (28) is "characteristic of the written language," while (29) is ' 'characteristic of writers from the territories in southwestern dialects" and is "infrequently used in literary Ukrainian." Once again, both constructions admit agents, in clear contrast to the Polish siq constructions, which they otherwise resemble. Turning now to true participial passives in Polish, we see that they require nominative subjects and subject-verb agreement and allow optional agents. They thus resemble the periphrastic passive familiar from English. (30)

a. katedra byta budowana (przez nas) cathedral (f sg nom) AUX(f sg) built(f sg) by us 'the cathedral was being built (by us)' b. ksiazka zostaia przeczytana (przez Jana) book(fsgnom) AUX(fsg) read(fsg) by Jan 'the book was read (by Jan)'

The zostac 'become' passive is used with perfective verbs and the bye 'be' passive is used with both perfective and imperfective verbs; see Siewierska (1988, 250-251) and the references therein for discussion of the differences in interpretation. There is thus a selection relationship between the auxiliary chosen and the aspect. This should follow from a functional analysis in which the zostac auxiliary selects a perfective Aspect Phrase complement and bye is simply inserted to support tense, regardless of aspect. As mentioned above, however, there is in addition one curious participial construction in which the object (if there is one) remains in the accusative:

346

(31)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a. katedre zbudowano cathedral(acc) built 'the cathedral (has been) built' b. przec/ytano ksiazke^ read book(acc) 'the book (has been) read'

This impersonal construction is structurally parallel to the siq + accusative passive in that no overt agent is possible. Note that in this respect Polish (31) differs from Ukrainian (25); compare Shevelov (1969, 177) for further comparisons. Another important difference is that this Polish construction does not tolerate the copula by to 'was', jest 'is', or bqdzie 'will-be', and thus only exists in the timeless form of (31). On the other hand, as with Ukrainian -nol-to, the participle appears in a special nonagreeing form (distinct from neuter in -e ). As Siewierska (1988) notes, in older stages of Polish both overt (instrumental) agents and overt auxiliaries were possible, which suggests a correlation between these two phenomena.16 She goes on to argue that this construction is no longer passive in contemporary Polish, in that speakers regard it, like the comparable reflexive + accusative construction, as "active impersonal"—or, in her terms, as "active indefinite." The reason cited for this terminological move is that the "implied human agent behaves like a subject" with respect to control, as in the following roughly synonymous examples with -nol-to participles from Siewierska (1988, 272): (32)

a. kiedy when mala small 'when

analizowano szczegolowo zdjecia satelitarne odkryto analyzed in-detail pictures(acc) satellite discovered wyspe island(acc) satellite pictures were scrutinized a small island was discovered'

b. analizujac szczegolowo zdjecia satelitarne odkryto mala analyzing in-detail pictures(acc) satellite discovered small wyspe island(acc) The understood (presumably PRO) subjects of both participles in (32a) and of the gerund and participle in (32b) are identical. Siewierska therefore concludes that in Polish this participial construction in Polish involves a "fixed non-declinable form with a specialized impersonal function." This is untrue of Russian participials, which are still clearly passive by her criteria. As argued by Dy-ia (1983), the Polish -nol-to construction has an implicit third plural arbitrary human subject.17 As Doros (1975) points out, this makes it virtually synonymous with the Russian third person arbitrary construction discussed in section 7.5. Siewierska (1988, 274) offers the following telling contrast between Polish (33a) and Russian (33b) to demonstrate this point: (33)

a. kupiono mi jeszcze tornister bought me(dat) still school-bag 'they bought me a school bag as well'

VOICE ALTERNATIONS

347

b. mne kupili jesce ranee me(dat) bought(3pl) still school-bag In our analysis, then, a covert pronominal element will be required in both. In this section, data from a variety of Slavic languages have been adduced to demonstrate that it cannot be solely the availability of null pronominal subjects that licenses the passive + accusative construction in its various instantiations. In the next section I therefore pursue an alternative solution in terms of the augmented parametric model of pro-drop phenomena.

8.3. Passive Morphology, Case, and Theta-Theory In this section a different kind analysis is developed of the various Slavic constructions utilizing voice-altering morphemes. After reconsidering the relationship between pro-drop and the passive + accusative construction, an approach in terms of parameterizing the cases a voice-altering morpheme can absorb will be discussed.18 Several proposals about the observed range of syntactic and interpretive variation in these structures across Slavic will be made, taking advantage of theta-theory and the VP-internal subject hypothesis.

8.3.1. Expletive Subjects and Passive We have seen that the existence of the passive + accusative construction in a language does not correlate with any standard pro-drop pattern. It appears, instead, that this construction is potentially possible in all the Slavic languages. It must be asked, therefore, what property of these languages (and Italian) makes the passive + accusative construction feasible at all. The answer, it turns out, does have to do with the possibility of null subjects. Crucially, regardless of the status of null thetamarked pronominal subjects or ECP-violations (i.e. whether the language is [-OSP] or [ + NDP], [-FIP] or [ + FIP]), all Slavic languages (and Italian) allow null expletive subject pronouns—in parametric terms they are all [ + OESP]. Similarly, all passive + accusative constructions must contain a null expletive subject, since (1) the subject position is universally required in clauses, (2) no overt element may occupy it (whence "null"), and (3) no theta-role is assigned to this position (whence ' 'expletive''). To put it simply, if the option of leaving a non-theta-marked subject position lexically unfilled were not available, then the passive + accusative construction would not be viable. In principle, therefore, these constructions could exist in any of the Slavic languages, since they are all [ + OESP].

8.3.2. Case Absorption Issues The account sketched in the preceding paragraph raises several important questions, the most obvious being, Why do some Slavic languages only have the canonical passive construction? The solution I put forward in Franks (1982) relied on the observation (essentially the same as in Babby's, 1989, in press, proposal; cf. n. 18)

348

PARAMETERS OF SLAVIC MORPHOSYNTAX

that in passive + accusative constructions no nominative subject is possible. I take this fact to mean that nominative case is simply unavailable for assignment to IPspecifier position. Since this absence of nominative with passive morphology parallels the absence of accusative in the canonical passive construction, a similar account seems justified. If, following standard GB assumptions, accusative is absorbed in the canonical passive construction, then in the passive + accusative construction it must be nominative that is being absorbed. In other words, this extension of the case absorption idea implies that in Polish, kajkavian SerboCroatian, Slovenian, North Russian dialects, Belarusian, and Ukrainian certain passive morphemes are able to absorb accusative or nominative case, whereas in the other languages they can only absorb accusative. Absorption of accusative necessitates movement of the D-structure object to the S-structure subject position in the familiar way. Absorption of nominative does not, since (1) the verb still assigns accusative, and (2) nominative is in any event no longer available. That proposal can roughly be summarized as follows for the data presented thus far: (34)

Russian North Russian Ukrainian Polish

-sja absorbs ace only -n/-t absorbs ace only -sja absorbs ace, nom -n/-t absorbs ace, nom -sja absorbs ace, nom [some dialects] -n/-t absorbs ace, nom sic absorbs ace, nom -n/-t absorbs ace only [except in (31)]

stokavian Serbo-Croatian

se absorbs ace only

kajkavian Serbo-Croatian

se absorbs ace, nom

Slovenian

se absorbs ace, nom

I shall refer to this choice between (1) absorbing only accusative or (2) absorbing either nominative or accusative as the Case Absorption Parameter. It will turn out that this parameter is too specific in that no reference need be made to actual morphological cases. Instead, it is the source of the case—hence the position(s) in which case can be absorbed—that is the relevant factor. Nonetheless, it should be emphasized that this sort of solution to the passive dilemma properly relegates the parameter involved to a relatively minor fact about individual morphemes, one that might be expected to vary independently of core null subject properties. I now turn to some specific problems with the Case Absorption Parameter as formulated in (34). The study of reflexive morphology in Slavic has had a long history, both traditionally and within the transformational-generative paradigm. While early treatments of -sja, such as Babby and Brecht (1975) or Chvany (1975), called for a unitary account, even its former proponents now believe this is neither possible nor desirable (Chvany, personal communication). Most of the reasons for rejecting this kind of treatment are familiar, and the literature is replete with detailed discussion of the problems engendered by lexical and syntactic idiosyncracy; (cf. e.g. Ruzicka, 1986, 1988; Schenker, 1986). Not surprisingly, an approach in terms of parameterizing the cases a morpheme can absorb also encounters many of these

VOICE ALTERNATIONS

349

same problems and leaves unexplained certain more subtle facts about the constructions being discussed. In Franks (1982) certain asymmetries between Polish passives in siq and Russian ones in -sja were explained in terms of -sja's inability to absorb any case but accusative. Compare the perfectly acceptable Polish (35a) with the impossible Rus sian (35b): (35) a. e pomaga sie slabym studentom help(3sg) REFL weak(dat) students(dat) 'one helps weak students' b. *pomagaet-sja slabym studentam help(3sg)-REFL weak(dat) students(dat) These examples support my contention that Polish siq can absorb nominative, but Russian -sja cannot.19 There are several reasons, however, why this claim might need to be weakened. Most important is the fact that other cases can in fact be absorbed by morphemes that supposedly absorb only accusative. Certain Russian verbs assign oblique cases to their direct objects, yet still undergo passivization; compare Fowler (1987b) for further argumentation that these NPs are indeed direct objects. For example, the Russian verb upravljaf 'to manage' assigns its complement instrumental case, as shown in (36a), but is nonetheless able to form a passive in -sja, as in (36b), or a present passive participle, as in (36c). (36) a. nasi druz'ja upravljajut fabrikoj our(nom) friends(nom) manage(3pl) factory(inst) 'our friends manage a factory' b. fabrika upravljaet -sja nasimi druz'jami factory(nom) manages(3sg)-REFL our(inst) friends(inst) 'the factory is managed by our friends' c. fabrika, upravljaemaja nasimi druz'jami, . . . 'the factory, (which is) managed by our friends, . . . If, as Fowler convincingly demonstrates, Russian verbs can in principle assign their complements any of accusative, genitive, or instrumental,20 and they are free to form passives regardless of the particular case assigned, then we must revise our claim about the primacy of accusative. We must conclude, instead, that -sja can absorb whatever case the verb assigns to the argument that it theta-marks. There are various ways of conceiving of this generalization of case absorption. Cinque (1988) argues, for example, that Italian si necessarily blocks the verb from assigning case because it absorbs the external theta-role, under the assumption that any verb that assigns case to its object must also assign a theta-role to its subject. This is an inventive way of connecting the suppression of the external theta-role with the suppression of the internal case, without literally invoking case absorption. Nonetheless, I do not share this assumption, due to Burzio (1986) and known in the literature as (half of) "Burzio's generalization," since there are many verbs in Slavic (and other) languages that appear to defy it, assigning a VP-internal case and

350

PARAMETERS OF SLAVIC MORPHOSYNTAX

no external theta-role. Many of the verbs used in chapter 3 to argue for the thematic prominence effect in across-the-board constructions are of this type. Within the structural kind of approach adopted here, we want to say that, however a verb assigns its case to its complement, -sja must be in a position to absorb this case without literally being in the direct object position (since otherwise it would be expected to absorb the object's theta-role as well). It must therefore occupy some functional head position within an extended X-bar system. That is, -sja must be the head of some functional projection outside of VP. Within such a model, the precise location of -sja depends on one's inventory of functional categories. Various methods of formalization are considered in section 8.3.4. A second problem with stipulating that Russian -sja is literally required to absorb accusative case is that it is empirically false that -sja must always absorb case at all. That is, a consideration of lexical idiosyncracy reveals that it is impossible to maintain as an absolute requirement that the morpheme -sja needs to absorb some case. The Case Absorption Parameter is thus too rigid in the sense that the specific morphemes implicated need neither to absorb the specific case(s) mandated by the parameter nor even to absorb any case at all. I have in mind the existence of numerous Russian verbs in -sja that lack a corresponding transitive form, which makes it unlikely that -sja could be absorbing case in any syntactic sense. Instead, these primary -sja verbs are lexically specified as requiring -sja, so that no syntactic operation is necessary. One might of course retreat to the traditional generalization that -sja precludes an accusative object, but even this characterization of -sja is not completely accurate. In the colloquial language there is a tendency for some verbs that inherently end in -sja to take ordinary accusative objects, as opposed to the normative genitive. Consider the Russian examples in (37) of -sja verbs with accusative objects; while (37b) is considered non-standard and colloquial, (37a) has entered the literary norm. (37)

a. syn slusaet-sja mamu/mamy son(nom) 'listen' -REFL mother(acc)/(gen) 'the son obeys his mother' b. Ivan boit -sja sobaku/sobaki (nom) 'fear'-REFL dog(acc)/(gen) 'Ivan is afraid of the dog'

While slusat' 'to listen' does exist as a transitive verb, the meaning of slusat'sja 'to obey' is clearly not based on it in any compositional way. With regard to bojat'sja 'to fear', there is no verb *bojat' from which it could arguably be derived. The -sja morpheme is thus inherent to these verbs, in the sense that they are represented in the lexicon with -sja. Since both nominative and accusative arguments are present in (37), we conclude that -sja in these examples does not need to absorb either case. Even the claim that Polish siq absorbs nominative is not without problems. Recall from chapter 6 that predicate adjectives in Polish agree with their ultimate antecedent if possible; otherwise they appear in the instrumental case. Nominative (and quantificational genitive) subjects thus induce more or less obligatory predicate adjective agreement. Consider in this light the fact that no agreement is possible in the Polish impersonal construction in (38).

VOICE ALTERNATIONS

351

(38) e jest sie. tutaj dobrym/*dobry is REFL here good(inst)/(nom) ' one is good here' Since the (S-structure) subject position is necessarily empty, I assume that the nominative case that would otherwise be assigned to this position has been absorbed by the siq morpheme. The puzzle posed by (38) is thus why case agreement should be blocked if siq is indeed nominative. The subject is presumably an expletive pro, its case and perhaps also argument status having been transferred to the reflexive clitic siq. This process for some reason results in nominative case becoming inaccessible to the predicate adjective for purposes of agreement. While a formal account consistent with the idea that siq absorbs nominative is certainly possible, one might recast "case absorption" in terms of saturating but not actually bearing case or, alternatively, simply reject the case absorption paradigm. To summarize, the proper generalization about Russian -sja seems to be that in passive constructions it can and must absorb whatever case the verb assigns, but that no comparable generalization is possible for inherent -sja. This array of facts suggests that a solution to the variation in passive-like constructions across Slavic cannot be couched solely in terms of the case(s) that a particular morpheme is in principle able to absorb. Instead, the case aborption properties of a morpheme depend crucially on the semantic import of attaching that morpheme: when -sja has a syntactic function, as in passives, it is subject to stricter case absorption requirements than when it has a lexical function. It is only when a morpheme functions in a voice-altering capacity that it is needs to be associated with case. If case is, in turn, a necessary property of arguments, then this indicates that voice-altering morphemes need to be accorded some kind of argumental status. In the remainder of this section, therefore, let us re-examine the various voice-altering morphemes we have been discussing from the perspectives of theta-theory and phrase structure. 8.3.3. Theta-Theoretic Issues Affixing a voice-altering morpheme to a verb typically alters the active pattern of association between the theta-roles lexically specified for that verb and the argument positions in the syntactic representation of the clause ultimately headed by the verb. In this section, I consider how the Slavic case absorption facts might be understood in terms of whether and how the various voice-altering morphemes rearrange predicate-argument structure. In the next section, I consider how the different constructions might be syntactically represented within various functional category phrase structure systems. Each of the voice-altering morphemes, it turns out, displays characteristic properties that must, to some extent, be lexically specified, but whose instantiation might also follow from the component in which passive affixation takes place. Two qualifications are in order, however. Since any readjustment of the realization of theta-roles is only evident at S-structure, it is not prima facie clear whether the relevant manipulation occurs in the lexicon or in the syntax. Moreover, if the feature-checking model of Chomsky (1992) is correct, the distinction between syntactic and lexical affixation will need to be reevaluated.

352

PARAMETERS OF SLAVIC MORPHOSYNTAX

As mentioned in section 8.1.2, one common view of morphological operations is that they can in principle apply either syntactically or in the lexicon. With respect to the addition of what I have been loosely calling voice-altering morphemes, middle formation must be a lexical process, since it rearranges predicate-argument structure,21 but passive formation need not be. According to Borer and Wexler (1987), passive may apply syntactically so long as it preserves semantic relations. One standard method of testing this is to see whether or not the external theta-role of the "original" active is available syntactically. By comparing constructions in terms of what passivization does to the active external theta-role, it is possible to establish just what concomitant lexical process affixation of each morpheme might involve. We first focus on those constructions in which passive morphology co-occurs with an accusative object, as in Polish (18b) with siq, (31) with -nol-to, kajkavian SerboCroatian (19b) and Slovenian (20b) with se, North Russian (23) with sja and -nol-to, Ukrainian (21) and (25) with -nol-to, and (29) with -sja. Since in all these constructions the ability of the verb to assign case and theta-role is unaffected, as evidenced by the fact that the active object continues to be accusative and the active subject continues to be syntactically active, I conclude that in all these constructions passive morphology has been added syntactically. This entails that the verb's predicateargument structure is intact when the item is drawn from the lexicon. If the agent is specified as the external argument at D-structure, it must then be available throughout the derivation. That is, all these sentences are predicted to be underlyingly transitive and syntactically personal, with syntactically active understood agents.22 The phrase structure of these constructions, to be discussed in the next section, should reflect this fact by representing the subject argument in an appropriate position. To see that the syntactic accessibility of the active external argument is unaffected in Ukrainian, for example, recall that the external theta-role is blocked from appearing overtly in subject position, but crucially it is not eliminated. This was demonstrated by the possibility of instrumental agents in (25), (28b), and (29b), as well as by the acceptability of purpose clauses, as follows: (39)

a. xatu zrujnovano tomu, scob zbuduvaty tut park 'the house(acc) was destroyed to build a park here' b. dvery bulo zaperto, scob sobaka ne vybih na dvir 'the doors(acc) were closed so that the dog would not run outside' c. cerkvu budujetsja, s£ob ljudy maly de molytysja 'the church(acc) is being built so that people will have somewhere to pray'

The agent is implicitly present in these Ukrainian passive -I- accusative examples, for both the participial and reflexive passive construction. It thus appears that the passive morpheme "absorbs" both nominative case and the subject theta-role, behaving, in the sense of Baker, Johnson, and Roberts (1989), just like a syntactic argument. Why it should do both at once remains problematic, of course; we return to the more explicit phrase structural analysis of the various passive-like constructions in section 8.3.4. Similar evidence is provided by Polish, since purpose clauses freely occur in both reflexive and participial types of passive + accusative constructions:

VOICE ALTERNATIONS

(40)

353

a. buduje sie. katednj zeby uczic boga 'the cathedral (ace) is being built to worship God' b. przeczytano ksiazke. zeby dowiedziec si$ o Afryce 'the book(acc) was read to learn about Africa'

Control into the purpose clauses in (40) once again implies the existence of an agentive antecedent. Recall that in the Polish constructions illustrated in (40) the agent can never be overtly expressed; it is only implicit. As noted above, it is curious that in this respect Polish differs from superficially comparable Ukrainian examples. Since the agent is syntactically active in the complete range of passive + accusative constructions in both languages, however, I assume that the external argument of the active must be associated with some argument position, although not necessarily the same one in all relevant constructions. Various possibilities are presented in the next section. We have seen that Ukrainian -sja and Polish siq can both be regarded as syntactic affixes, in that they absorb (rather than eliminate) the external theta-role and, concomitantly, absorb nominative case to give rise to a passive + accusative construction. We now ask whether simply parameterizing the sources of the cases that these morphemes can absorb is adequate to accommodate the parallel passive + nominative constructions. It turns out that for Ukrainian the answer is affirmative, as demonstrated by the acceptability of agent phrases in examples such as (28b); purpose clauses are similarly possible in such constructions. All available evidence leads to the conclusion that in Ukrainian whether the active object is expressed in the passive voice as a nominative subject, inducing subject—verb agreement, or as an accusative object, with default (or no) subject-verb agreement, is purely stylistic or regional. Since there is no substantive difference in interpretation, no restriction need be imposed on the Ukrainian morphemes, which are free to absorb the case assigned either by V or I. In Polish, on the other hand, no purpose clause is possible when siq coocurs with a nominative subject, as was illustrated by (27b).23 Thus, although Polish siq + accusative constructions are (syntactic) passives, siq + nominative constructions are really (lexical) middles in that the agent theta-role is eliminated (i.e. lexically saturated by being assigned arbitrary reference). In middle formation siq lexically derives from a transitive verb either an intransitive verb (assuming externalization of the theme argument) or an unaccusative one (assuming the theme raises to subject position in search of case). If the verb no longer thetamarks an object position, then there is not even any case available to be absorbed and the issue of what case is absorbed becomes moot. If it continues to theta-mark the object, we must still assume that no case can be assigned in order to force raising. The siq morpheme thus does seem to "absorb'' the verb's case in the Polish middle, although as part of a lexical operation. Unfortunately, in the absence of any model of what case absorption really is, this new distinction between syntactic and lexical case absorption is not particularly enlightening. This, then, is another problem that a proper understanding of the difference between lexical and syntactic affixation might help to elucidate. In the "minimalist" model put forward in Chomsky (1992), the only two levels of linguistic representation are PF and LF, which interface with what he calls the

354

PARAMETERS OF SLAVIC MORPHOSYNTAX

articulatory-perceptual and conceptual-intentional systems. Chomsky argues that lexical items are drawn from the lexicon complete with all morphosyntactic features and inflectional affixes. These are checked against what, in more traditional approaches, would be assigning those features or adjoining those affixes in the first place. Chomsky's characterization of the checking procedure is extremely sketchy, however. A fully formed element, such as a noun complete with case and pronominal features or a verb complete with tense and agreement inflection, has some or all of these checked against the appropriate functional head at some point in the derivation up to LF. The idea is essentially that the element raises into the functional projection covertly in LF (or overtly on the way to LF, depending on the "strength" of the functional head), and if the inflectional features match they are considered checked and the functional head deletes. To my mind, this scenario leaves a number of important questions unaddressed. Two of these were mentioned above in connection with the analysis of voice. First of all, it is not clear how the system can identify which features need to be checked and which do not. A second and related problem is how the traditional distinction between lexically and syntactically introduced morphology can be made. I deal with these in turn. The first question boils down to how checking might be able to distinguish between inflectional and derivational morphology. Inflectional morphemes are those that need to be syntactically sanctioned and derivational ones are those that do not. It seems to me that the best way to approach this distinction is to assume that derivational morphemes do not have features that are visible to the syntax. Even so, only a subset of the inflectional features must be checked. For example, the case of a noun needs to be syntactically licensed but its number and gender do not (although coindexing factors such as predication and binding might require matching). Furthermore, there must be a pairing between functional projections and morphosyntactic feature submatrices, in the sense that a functional projection exists only to check features of lexical heads. The morphological properties of words thus determine the functional structure of the clause or phrase. Turning now to the second question of the difference between lexically and syntactically added morphology, it is obvious that in the minimalist system there can be no derivational distinction. Let us assume, then, that the relevant fact is that lexically introduced morphology is derivational in the sense that it does not need to be checked. However, in order to prevent a lexical passive or reflexive, for example, from being checked, we need somehow to differentiate the morpheme in question from its syntactically active counterpart. The most straightforward way of doing this is to dissociate the morpheme from its morphosyntactic features, thereby making it invisible at LF and thus depriving it of the need to be checked. A passive -sja, for example, has morphosyntactic features (reflecting case and perhaps also voice), whereas a lexical -sja, as in bojat'sja 'to fear', has none. Only the former will therefore need to be checked against some functional projection. Notice that under this view the only morphology that needs to be checked is that which has morphosyntactic features and corresponds to some functional category. Finally, it is interesting in this light to consider the implications of this model for the evolution of inflectional morphology. Initially, as part of a paradigm, a morpheme is necessarily associated with a morphosyntactic feature matrix, which in turn

VOICE ALTERNATIONS

355

needs to be checked against some functional projection. Redundantly, the morpheme will appear in the lexicon on those words with which it composes. However, once a word exists in the lexicon it is free to begin a life of its own. In particular, the inflectional morphology may lose its inflectional status, essentially by jettisoning the morphosyntactic features that once defined it, perhaps gradually. Once this occurs, the word no longer needs to be checked for the features previously contributed by the erstwhile inflectional morpheme, hence ceases to require the projection of the appropriate functional category above it, and the word's interpretation diverges from its earlier purely compositional status. With respect to the voice-altering morphemes under consideration in this chapter, this means that the very existence of -sja and -n-l-t- as syntactic entities implies the creation of syntactically opaque lexical items that contain these same morphemes but display diminished syntactic properties. 8.3.4. Phrase Structure Issues I now return to the problem of how the various structures might be represented syntactically. The primary issue that needs to be addressed is that of identifying the functional categories associated with voice-altering morphology and the positions occupied by the morphemes in question. In this section, I speculate about various possible ways of approaching this issue. One general strategy for dealing with the syntactically active implicit agent that we have observed in many of these constructions is to represent the agent as a null element in the phrase structure. Assuming, following arguments in chapter 5, that the subject originates in SPEC-VP and, following arguments in chapter 6, that PRO can remain in this position at S-structure, the most straightforward approach is simply to claim that the implicit agent is a PRO in the specifier position of VP.24 Consider, for example, the Polish -nol-to construction, which might be ascribed the minimal structure in (41). (41)

typ PRO [v. przeczytano ksiazkej]

Since forms such as przeczytano are fixed as such, there is no need to project any kind of INFL shell. The structure in (41) is thus a kind of sentential small clause. This assumption is supported by the fact that no copula is admissible. Moreover, the presence of PRO can be used to preclude the overt expression of the agent in adjunct przez-phrases, since otherwise the agent role would be associated with two formally unrelated positions. Its expression in a przez-phrase would therefore be a ThetaCriterion violation, although PRO still freely controls into a purpose clause. Note that this use of PRO to block the fey-phrase is the opposite tack from that adopted in Baker, Johnson, and Roberts (1989), who, it will be recalled, argue that the passive morpheme is an argument that doubles the by-phrase. In (41), however, there is no case absorption, since the verb form is essentially opaque to the syntax, hence does not require the projection of any higher functional categories. There is no subject AGR to assign nominative, and accusative is assigned in the same way it would be in a full clause with active verbal morphology. PRO, since it is not properly governed, is interpreted as a pronominal with arbitrary reference. This account, however, leaves unresolved the structure of the superficially similar

356

PARAMETERS OF SLAVIC MORPHOSYNTAX

Ukrainian passive + accusative construction. Recall that in Ukrainian the past and future auxiliaries bulo and bude, as well as instrumental agent phrases, freely cooccur with the -no/-to form. This was illustrated by the examples in (25), which is repeated in (42). (42)

a. z Vinnyci pryslano vistku sestryceju from Vinnycja brought(n) news(acc) sister(inst) 'news was brought from Vinnycja by the sister' b. mojim mylym mene zradzeno my(inst) beloved(inst) me(acc) betrayed(n) 'I was betrayed by my beloved' c. dim bulo zbudovano robitnykamy house(acc) AUX(n) built(n) workers(inst) 'the house was built by the workers'

It is clear that (42) cannot be given the same kind of analysis as (41). If morphology is to drive the projection of functional categories, as argued in the preceding section, then the difference must be localized in the participle. The Ukrainian participle cannot be similarly opaque; instead, the affix must have visible features that need to be checked. INFL is therefore projected, and with it the concomitant tense features. Here, then, the passive morpheme does absorb nominative case, making it impossible to realize the underlying subject or object in the nominative. Where might the passive morpheme—or, in the checking model, the element that checks its features on the verb—be located? In some functional category systems, such as that of Ouhalla (1991), there is a Passive Phrase whose head can contain the necessary morpheme. The PassP is in turn embedded in higher functional projections such as AgrP and TP, roughly as in (43). (43)

[AgrP AGR [TP TNS [PassP -n- [VP PRO [v, V NP]]]]]

V moves up to the head of PassP, adjoining to -n- or, in the minimalist model, having its features checked, and then continues to raise. However, since it is AGR that assigns/checks nominative on its specifier, it is not obvious how nominative case becomes absorbed. A more reasonable approach might therefore be to locate the passive morpheme (or features) in a functional projection implicated in nominative case assignment instead. This is shown in (44a) for the system that just has INFL, and in (44b) for the system with separate tense and agreement projections.

(44)

a. [IP -n- [VP PRO [v. V NP]]] b- [Agrp -»- [TP TNS [vp PRO [v. V NP]]]]

Under either structure, the verb still eventually moves up to the passive morpheme/ feature cluster and it is possible to conceive of nominative case absorption as the failure of SPEC-head agreement. Nominative case features are presumably just retained by the head -n- or, in the checking model, checked against the passive morphology of the verb after it has adjoined to AgrP/IP. Notice, however, that this account says nothing about the alternative expression

VOICE ALTERNATIONS

357

of the agent in an instrumental fey-phrase. I see two ways of handling this issue, neither unfortunately particularly elegant. The first is to claim that the agent PRO in (44) should not actually be located in SPEC-VP. Instead, it is adjoined to VP, in essence as a covert counterpart to the overt adjunct. This is shown in (45), adopting IP over the split INFL for ease of exposition.25

(45)

[IP -n- [VP, PRO [VP V NP]]]

The agent can thus be realized either as an instrumental adjunct, which it was argued in chapter 6 arises in precisely this configuration, or as PRO if no case is assigned. Under the checking alternative an instrumental NP will have its case checked by virtue of being sister to VP, whereas PRO, because it lacks case, will not. Since this kind of case checking does not involve agreement with a functional head,26 there is no case that needs to be discharged, hence checking (or assignment) will be optional. Even so, there are two possibilities with respect to how PRO comes to be in the adjoined position—either it is base-generated there, or it moves there from an initial VP-speciner position. Both possibilities raise interesting questions, and I will not attempt to resolve this issue here. If generated as an adjunct, then passive will have to be a lexical operation since it manipulates predicate-argument structure. If PRO moves, then it will leave a trace in VP-specifier position, as in (46). (46)

[IP -n- [VP* PRO; [VP e; [v, V NP]]]]

This analysis leads to the obvious question of what motivates the movement since, following considerations of economy laid out in Chomsky (1991,1992), PRO should not raise unless it is required to do so. It is possible that PRO adjoins to VP in order to avoid being lexically governed, assuming that if the lexical material in INFL governs PRO this would force PRO to be anaphoric. PRO would thus be a free anaphor, in violation of the binding theory, unless it vacates the VP-specifier position. Notice that no such problem arises in Polish (41), since projections above VP are lacking. Alternatively, one might argue for an IP in (41) whose head lacks feature content, so that the PRO subject will not be lexically governed. The second way of handling the alternation between an adjunct and PRO agent would be to claim that there is always a PRO in SPEC-VP and that, when the adjunct phrase is present, the overt agent controls the PRO. This is represented schematically in (47), where the fey-phrase is intended to cover the range of instantiations of the overt adjunct agent. (47)

[IP -n- [VP* by NP [VP PRO [v, V NP]]]]

In addition to requiring different assumptions about the distribution and interpretation of PRO than those adopted in chapter 6, this approach comes up against the same theta-theoretic objections as does the Baker-Jaeggli analysis, in which the passive morpheme shares a theta-role with the optional fey-phrase. Another possibility worth exploring, discussed in Franks (in press), relates the difference between Polish and Ukrainian in realization of the copula to the fact that the former is a canonical null subject language whereas the latter is not. Under some accounts the [ + OSP] property of the East Slavic languages can be explained in terms of AGR not being rich enough to constitute an independent lexical item, so

358

PARAMETERS OF SLAVIC MORPHOSYNTAX

that it always lowers onto the verb—even when the verb is the copula. This approach has the benefit that at S-structure the copula governs the subject in Polish, but not in Ukrainian. This would force the omission of the copula in Polish whenever there is a PRO subject, since otherwise it would be a free anaphor. There are several problems with this approach beyond the obvious conceptual ones of relating it to the checking model, however. It seems to me unlikely to give the necessary typological and historical coverage, since other [ + OSP] languages retain the copula, including earlier stages of Polish. In addition, one wonders why the NP of the przez-phrase will not bind PRO, satisfying its anaphoric status if lexically governed. Ukrainian aside, there is another interesting correlation one might want to treat as a significant generalization. A comparison of the range of structures in Polish and Russian reveals that the passive + accusative construction is only available when no Zry-phrase is possible. For example, in Polish the agent can be expressed overtly only in the zostac and bye participial passive constructions. If the D-structure subject can be represented either as a PRO in SPEC-VP position or as a phrase adjoined to VP, and if passive involves movement of the D-structure object through SPEC-VP eventually to SPEC-IP, then PRO will block this movement, although an adjunct phrase will not. This approach might provide a way of connecting the passive + nominative possibility with the potential for an adjunct agent and obviate the need for explicit statements about case absorption. For a variety of reasons, it is unfortunately difficult to see how the blocking approach could be executed. For one thing, it would need to allow the object to remain in situ and receive accusative case as a regular option, but then, if movement takes place only when necessary, it is unclear why the grammar should ever avail itself of the passive + nominative strategy. This suggests that it will be still be necessary to parameterize in some way the case absorption properties of the passive morphemes. Ukrainian, on the other hand, presents an empirical problem in that passive + accusative is compatible with an adjunct instrumental agent, as in the examples in (42). Hence, some other factor must be invoked in blocking movement. However, this factor would have to be optional, since Ukrainian participial passives are of both nominative and accusative types. This again presumably relates to the case absorption properties of the passive morpheme itself, which seems to me to render the blocking effect of PRO superfluous. In short, any solution to the problem of variation in case assigned to the D-structure object will necessarily involve variation in the behavior of the passive morpheme. I conclude, therefore, that whether the passive morpheme absorbs the case assigned by the verb or by subject agreement needs to be parameterized. In the phrase structure system with just IP this must be stipulated. That is, assuming for the sake of argumentation that V-raising applies to a structure as in (44b) to produce (48), there are two options.

(48)

[IP [, V + «-] [VP PRO [v. e NP]]]

When -n- absorbs the verb's case, the object NP will move up to SPEC-IP and be assigned nominative. When -n- absorbs the nominative case of the AGR element of INFL, the object NP remains in its D-structure position and is assigned case by the verb (or its trace). This analysis follows, as before, regardless of whether V-raising obtains in the syntax proper or in LF (if a case-checking model is adopted).

VOICE ALTERNATIONS

359

Another method of approaching the problem of what drives this variation in the case absorption properties of the passive morpheme is literally to locate it in different phrase structure positions under the two scenarios. Consider, for example, the system of Chomsky (1992), in which a transitive clause contains (at least) two AgrPs, one for subject case and the other for object case. The structure of a clause is thus roughly as in (49), where AGR-S and AGR-O represent the two agreement projections.27 (49)

[Agr_SP AGR-S [TP TNS [Agr.OP AGR-O [VP NPsub [V NPobj]]]]]

Both nominative and accusative are structural cases assigned under SPEC-head agreement with AGR. The subject NP moves to SPEC-Agr-SP and the object NP moves to SPEC-Agr-OP; again, these movements can take place at LF, assuming the checking model. Given this kind of phrase structure, a reasonable account of parametric variation in the realization of passive might be to allow the voice-altering morpheme to be potentially located in either AGR position. This leads to the two options indicated below: (50)

a. [Agr.SP AGR-S [TP TNS [Agr.OP -«- [Vp NPsub [V NPobj]]]]] b- Ugr-sP -n- [TP TNS [Ag,OP AGR-O [VP NPsub [V NPobj]]]]]

When the passive morpheme—or features, in a checking model—occupies the head position AGR-O, it absorbs accusative case, and when it occupies the head position AGR-S, it absorbs nominative case. Which position it appears in might be handled as a matter of selection, with the passive morpheme selecting either TP, VP, or, in the case of Ukrainian, both; compare Ouhalla (1991) for a discussion of functional heads and selection. Under both (50a) and (50b) the subject will presumably be PRO. When the object NP raises it will have to cross the subject NP, which under traditional views of minimality constitutes a movement violation. Notice, however, that crossing is required in the derivation of an active sentence anyway. Consider what happens in structure (49): the subject NP raises to SPEC-Agr-SP to receive nominative case, and the object NP raises to SPEC-Agr-OP to receive accusative case; in English, which lacks overt object-shift, this latter movement takes place in LF. Chomsky (1992) discusses these issues and sketches out a solution in terms of V raising to AGR-O forcing the subject NP not to move to SPEC-Agr-OP. The analysis of parametric variation in passive thus appears to be completely consistent with the account of clause structure and case-checking adopted in Chomsky's minimalist program. One interesting consequence of any of the structural analyses considered in this section is that in all passive + accusative constructions, nominative case becomes dissociated from IP-specifier position. This means that in passive + accusative constructions SPEC-IP cannot be regarded as a cased position, hence it is not expected to display the kinds of properties expletive pro subjects do. In particular, the overt option discussed in chapter 7 should never be available, since nominative case has been absorbed by the passive morpheme. Compare this prediction with the alternative possibility, as proposed, for example, in Sobin (1985), according to which the voice-altering morpheme absorbs no case and the subject is null simply

360

PARAMETERS OF SLAVIC MORPHOSYNTAX

because Ukrainian has null expletive subjects, that is, it is [-OESP]. Aside from theoretical considerations, there is some empirical evidence against this hypothesis. As far as I can tell, the Ukrainian expletive vono, which it was shown in chapter 7 is marginally possible in true expletive subject contexts, is totally unacceptable in passive + accusative constructions. Since nominative is absorbed in these constructions and case is necessary for a nominal to be phonetically realized, no overt expletive is conceivable. I have in this section focused mostly on the status of the participial rather than the reflexive morpheme. As shown above and made especially clear in (34), the two elements exhibit the same kinds of syntactic behavior in the various dialects and languages, the difference being which displays what properties rather than the nature of the properties themselves. I therefore assume that any analysis that successfully accounts for the range of constructions associated with the participial passive will also account with minimal modifications for the range of constructions associated with the reflexive passive. The two primary potential distinctions to be drawn have to do with the morphology of the elements in question and the positions that they might occupy in the IP (or AgrP) projections. The reflexive morphemes are historically pronominals and, to varying degrees in the different languages, still exhibit paradigmatic and syntagmatic properties of pronouns. The participial morphemes, on the other hand, are derivational suffixes that change part-of-speech to create adjectives. One possibility therefore is that the reflexive morphemes are maximal projections and should be properly situated in specifier rather than head positions. Pursuing this possibility, one might claim that in passive constructions, Polish siq, for example, originates as the specifier of Agr-SP or IP, depending on which phrase structure system is adopted, and subsequently cliticizes onto the verb. Since movement of the reflexive morpheme would have to be rightward lowering, the cliticization rules could not be formulated, however, in a structure-preserving manner. The relative freedom of positioning of the reflexive morphemes with respect to the verb stem in South and West Slavic render a checking approach—with the reflexive clitic drawn from the lexicon together with its "host" verb—virtually unworkable. Furthermore, since Russian only has passive + nominative, -sja would have to be generated as the specifier of Agr-OP; it is not clear how a system without Agr-OP would express this contrast.28 Finally, given arguments in Progovac and Franks (1992) and Progovac (1992) that simplex reflexives, such as Russian sebja 'self, function as heads for the purpose of the binding theory, clitics surely also need to be treated as X°s rather than as XPs. Turning to the issue of the categorial properties of these morphemes, the clitics are probably best analyzed as [ + N, — V] and do not affect the [ — N, + V] status of the associated verb. The reflexive clitics have no impact on the internal morphology of the verb, implying that the verb does not adjoin to them, since otherwise the verb + clitic complex would then be headed by the clitic rather than the verb. This suggests either that the clitics lower on to the verb29 or, as discussed above, that the clitics are actually specifiers of functional projections. The participial morphemes, on the other hand, appear to alter the categorial status of the verb. Sometimes, as in Russian and Polish for the passive constructions in (30), a [ -N] verb is changed into a [ + N] adjective, so that accusative case can no longer be assigned to the object. Of course,

VOICE ALTERNATIONS

361

under the AGR-O and checking analysis, the reason the verb can no longer assign case is not that it changes category—since it is drawn from the lexicon as a participle—but rather that the accusative case that AGR-O would assign is needed by the participial morpheme, generated in the AGR-O position. Note that it cannot be the participial suffix per se that causes a category change, otherwise it would be impossible to have accusative objects in Ukrainian, Polish, or North Russian dialects. This fact is perhaps best seen in Russian, where short form (syntactic) participles are spelled with one "n," and long-form (adjectival) ones with two. The internal theta-role is necessarily externalized only when a verb ([-N, +V]) becomes an adjective ([ + N, + V]), since adjectives must be predicated of something external to themselves. If we assume that adding -n-l-t- in Russian neutralizes the feature [ ± N] to create a defective [ + V] category non-distinct from both A and V, then Russian participial passives will still be syntactic. Since accusative case in Russian is assigned by [ —N] categories (i.e. V and P), their objects are caseless in Russian, necessitating movement. Whether the internal argument is also externalized or not is unclear. In passive + accusative constructions the verb surely still assigns theta-role and case to its object. In passive + nominative constructions, however, it is difficult to decide whether there is manipulation of predicate argument structure in addition to loss of accusative case. Although no externalization is required under the analysis sketched above, with a PRO subject and the passive morpheme in INFL or AGR-O absorbing case, there are a number of treatments of Polish participial passives that argue against NPmovement (cf. e.g. Zabrocki, 1981; Giejgo, 1981; Kardela, 1986). However, since the main thrust of these studies is that passive involves lexical attachment of the voice-altering morpheme, and under a checking analysis this is true regardless, I see no compelling reason to reject the traditional transformational account in terms of NP-movement. As discussed in chapter 5, one way of testing whether an argument is VP-internal in Russian, Ukrainian, and, to a lesser extent, Polish is whether it may surface in the genitive under negation; compare Chvany (1975), Pesetsky (1982), and Neidle (1982, 1988) for detailed arguments that the Russian genitive of negation applies exclusively to VP-internal noun phrases.30 Babby's (1980b) observation that, for example, Russian rasti 'to grow' andplavat' 'to float' (although crucially not in the meaning 'to swim', which has an agent) may take genitive "subjects" under negation, in fact reveals that the non-agentive "subjects" of these verbs are really underlying objects: (51) a. nikakix gribov ne raste't v etom lesu not-any(gen) mushrooms(gen) NEG grow(3sg) in this wood 'no mushrooms grow in these woods' b. nikakogo mjasa ne plavalo v supe not-any(gen) meat(gen) NEG floated(n) in soup 'no meat floated in the soup' Although the verbs do not themselves assign case, which in the affirmative necessitates movement of the object NP to subject position and eventually to SPEC-IP, under negation the objects can be assigned genitive case directly. Extending this idea

362

PARAMETERS OF SLAVIC MORPHOSYNTAX

to passives, we see that some Russian constructions with passive morphology are clearly created in the syntax: (52)

a. nikakogo moroza ne cuvstvovalo-s' not-any(gen) frost(gen) NEG felt(n) -REFL 'no frost was felt' b. v etoni godu ni odnoj knigi ne bylo opublikovano in this year not one(gen) book(gen) NEG AUX(n) published(n) 'not one book was published this year'

In (52) the genitive phrases nikakogo moroza 'no frost' and ni odnoj knigi 'not one book' are assigned genitive in situ and thus remain in their D-structure object positions. Crucially, as discussed in chapter 5, the genitive of negation can be assigned to complements of V regardless of whether that V is a case assigner or not. It therefore appears both in unaccusative constructions of the type in (51) and passive/middle constructions of the type in (52). While I take the possibility of genitive under negation to be a reasonably reliable diagnostic that the genitive NP is an object, unfortunately the failure of genitivemarking does not imply that the NP is not an object. In particular, whereas negated participial passives fairly commonly admit genitive objects, negated reflexive passives are much more resistant to this process. It is thus difficult to use the genitive of negation as a reliable test of whether reflexive middles in Russian involve syntactic NP-movement or lexical externalization, and I leave the matter unresolved. Another interesting question about the genitive of negation arises in connection with passive forms of transitive verbs that assign some quirky case. If being a complement of a negated verb is the minimal requirement for the operation of the genitive-of-negation rule, regardless of the case-assigning properties of the verb, then one would expect that even quirky case assigning verbs might be able to participate in this process when they are passive because voice-altering morphology can in principle absorb any case the verb assigns. Recall example (36) in connection with the argument from Fowler (1987b) that, so long as they meet the same requirements all verbs must, verbs that assign a quirky case can passivize, with the D-structure object appearing as a nominative subject. Another example for many speakers is given in (53), where zloupotrebljat' 'to misuse' obligatorily assigns its complement instrumental rather than accusative. (53)

a. Ivan zloupotrebil moimi sovetami Ivan(nom) misused my(inst pi) advice(inst pi) b. %moi sovety byli zloupotrebleny (Ivanom) my(nompl) advice(nompl) was(pl) misused(pl) (by-Ivan)

In the passive (53b) moi sovety 'my advice' is marked nominative. I thus conclude that the passive morpheme absorbs instrumental from the verb. Crucially, (53b) is well-formed despite the surface fact that the underlying object does not bear instrumental case. Assuming instrumental to be a lexical requirement, perhaps implemented in terms of visibility of the complement's theta-role, it must be satisfied by virtue of the passive morpheme. Under the IP analysis the passive morpheme is

VOICE ALTERNATIONS

363

assigned instrumental in INFL by the verb, and INFL and its specifier—which is occupied by moi sovety at S-structure—are coindexed hence agree. Under the AgrOP analysis the passive morpheme absorbs instrumental in AGR-O, still mediated by the verb since it raises to AGR-O at LF. If the object NP passes through SPECAgr-OP on its way to SPEC-Agr-SP, then it will be coindexed with the instrumental AGR-O; otherwise some more complicated agreement mechanism will be required. Either way, the point is that passive (53b) is able to satisfy the verb's lexical requirements. One therefore wonders whether genitive of negation can also be assigned in such contexts. Once again, it is difficult to tell since the number of verbs that assign quirky case and exhibit participial passives is extremely limited; Fowler (1987b) lists only two that assign instrumental and have past passive participles. Speakers who accept (53b), however, still find (54) impossible. (54)

*e ne bylo zloupotrebleno nikakix sovetov NEG was(n) misused(n) not-any(gen) advice(gen)

If this fact is interpreted to mean that genitive of negation does not apply to quirky-cased objects of passives, then an account must be constructed that differentiates these from examples like (53b). This account will probably involve NPmovement creating the necessary chain with the passive morpheme that absorbs the quirky object case in examples such as (53b), so that in (54) the quirky case requirement will not be met since the object NP does not raise.31 Applying the genitive-of-negation test to passive structures in the other languages that exhibit it is even less informative. In general, passive constructions that leave the verb's ability to assign accusative intact also allow genitive under negation. Here is an example from Ukrainian: (55)

tobi ne bude zrobleno nijakoho lyxa 'no(gen) harm(gen) will be done to you'

The problem posed by Ukrainian is that since (55) can be regarded as the negated version of a passive + accusative structure, no conclusion can be made about whether or not passive + nominative structures involve NP-movement. This is true despite the fact that Ukrainian resembles Russian in the scope and operation of its genitive-of-negation rule. Genitive of negation in Polish, on the other hand, is obligatory and restricted to contexts in which the accusative would be assigned in the affirmative. Recall from chapter 5 that it was proposed that the Polish genitive-ofnegation rule literally changes a (default) accusative-assigning verb into a (quirky) genitive-assigning one, so that the genitive of negation is not expected ever to apply in passive, unaccusative, or middle contexts. It is therefore impossible to use the genitive as a diagnostic for underlying object status in Polish, so that the ungrammaticality of (56) says nothing about whether the surface subjects of true participial passives originate in object position or not. (56) a. *nie bylo budowano katedry NEG AUX(nsg) built(n sg) cathedral(gen) b. *nie zostato przeczytano ksiazki NEG AUX(nsg) read(n sg) book(gen)

364

PARAMETERS OF SLAVIC MORPHOSYNTAX

Instead, it is only passive contexts in which the accusative is assigned in the affirmative that are predicted to exhibit the genitive of negation. Indeed, as shown by the examples in (57), it is precisely these that have genitive direct objects under negation. (57)

a. nie przeczytano ksiazki NEG read book(gen) 'no book has been read' b. nie sprzedaje sie; ksiazek NEG sell(3sg) REFL books(gen) 'no books are being sold'32

This fact does not, provide additional information about the structure of such clauses, however, since I have already taken the assignment of accusative in the affirmative versions of these examples as incontrovertible evidence that there is no lexical externalization of the theme.

8.4. Epilogue: The Dispositional Reflexive Construction I conclude this chapter with some speculations about the "dispositional reflexive" construction. A typical Russian example was cited in (2) and is repeated in (58). (58)

mne ne rabotaet-sja me(dat) NEG work -REFL 'I don't feel like working'

There is an implicit modality about this construction, so that it means either 'I just can't get down to work,' 'I just don't seem able to work', or 'I don't feel like working'. Since this construction involves the addition of the reflexive morpheme, the optional expression of the logical subject argument in the dative case, and the presence of some adverbial element, including negation,33 that specifies the disposition of the logical subject to perform the action of the verb, I follow Ruzica (1988) in adopting Bauer and Grepl's (1980) term "dispozicnf typy" (dispositional types) (cf. also Schoorlemmer, 1991). Recall that the dispositional construction is inconsistent with the generalization that Russian -sja absorbs accusative case, since here it clearly accompanies loss of overt nominative. Instead, the generalization appears to be that Russian -sja absorbs accusative case where available; otherwise, as in (58), it absorbs nominative. As discussed by Ruzica (1988, 175—176), the impersonal sentence type illustrated in (58) is found in all the Slavic languages. He cites the following Czech examples: (59)

a. ne -hraje se mi dobfe NEG-play(3sg) REFL me(dat) well 'I can't seem to play well' b. sestfe se tarn pracuje vyborne sister(dat) REFL there work(3sg) excellently '(my) sister is working excellently there'

VOICE ALTERNATIONS

365

Here are some simple examples from Bulgarian (60) and Serbo-Croatian (61): (60) a. ne mi se jade NEG me(dat) REFL eat(3sg) 'I don't feel like eating' b. place mi se cry(3sg) me(dat) REFL 'I feel like crying' (61)

a. ne radi mi se danas NEG work(3sg) me(dat) REFL today 'I don't feel like working today' b. meni se dobro pracuje me(dat) REFL well work(3sg) 'I am working well'

So far as I can tell, all the Slavic languages exhibit the impersonal version of the disposition construction in roughly the same way. There is, however, a personal variant of this construction that Russian lacks but freely exists in the South and West Slavic languages. For example, the following are not acceptable in Russian:34 (62)

a. *mne me(dat) b. *mne me(dat)

ves' den' citaet -sja eta kniga all day read(3sg)-REFL this(nom) book(nom) ne citaet -sja ni odna kniga NEG read(3sg)-REFL not one(nom) book(nom)

Expression of an underlying object as a nominative subject is apparently precluded. Furthermore, as shown by (63), even if the complement receives genitive under negation and thereby is able to remain in object position, the dispositional reflexive construction is not viable. (63)

*mne ne citaet -sja ni odnoj knigi me(dat) NEG read(3sg)-REFL not one(gen) book(gen)

The ungrammaticality of (63) suggests that considerations of case are not involved, since -sja does not otherwise necessarily block the genitive of negation. More generally, a Russian verb may not enter into the dispositional reflexive construction if it has any internal arguments, regardless of the case-assigning properties of the verb. This observation is corroborated by the fact that verbs with prepositional arguments, as in (64), do not participate in this construction. (64)

a. *mne ne rabotaet-sja nad etoj zadacej me(dat) NEG work -REFL on this problem ' I don't feel like working on this problem' b. *mne ne idet-sja na rabotu me(dat) NEG go -REFL to work ' I don't feel like going to work'

366

PARAMETERS OF SLAVIC MORPHOSYNTAX

Compare (64a) with (58): the only difference is the presence of the prepositional phrase argument.35 Ruzicka (1988) observes that the personal variant of the dispositional construction is perfectly acceptable in Czech, however, citing (65). (65)

matematika se mi studuje lehce math(nom) REFL me(dat) study(3sg) easily 'math studies easily for me'

More telling are the following well-formed Serbo-Croatian and Polish sentences, which sharply contrast with the impossible Russian examples (62)-(64). (66)

a. ova knjiga mi se ne cita this(nom) book(nom) me(dat) REFL NEG read(3sg) 'I don't feel like reading this book' b. ne ide mi se na posao NEG go(3sg) me(dat) REFL to work 'I don't feel like going to work'

(67)

a. ten artykuJ nie pisze this(nom) article(nom) NEG write(3sg) 'I just can't write this article well' b. nie pracuje mi sie. dobrze NEG write(3sg) me(dat) REFL well 'I just can't work well on this article'

mi sie. dobrze36 me(dat) REFL well nad tym artykutem on this(inst) article(inst)

Data such as these indicate that there is nothing wrong in principle with deriving a dispositional reflexive construction from any clause; it is just that this process is for some reason blocked in Russian under certain circumstances. The operative restriction seems to me to be that in Russian no internal arguments are allowed in the dispositional construction, whereas no such limitation exists in South and West Slavic. Thus, all the examples in Russian (62)-(64) are ruled out since they all involve some kind of internal argument of V. Notice, on the other hand, that nonargument phrases, as in (68), are perfectly admissible. (68)

a. mne ne rabotaet-sja pri takix uslovijax me(dat) NEG work -REFL under such conditions 'I don't feel like working under such conditions' b. mne ne citaet -sja v etom zale me(dat) NEG read(3sg)-REFL in this hall 'I just can't read in this hall'

It is unclear just how this difference between the language groups should be handled. Clearly, the existence of Serbo-Croatian (66) and Polish (67), as well as the expected English translations of the infelicitous Russian examples, shows that the relevant Russian restriction is not semantic in nature. In descriptive terms, one might claim that, if creation of the dispositional construction is a lexical process, in Russian it operates on non-branching V's exclusively. That is, it takes something of the form [v, V°] and changes it into [v, V°-sja]. Rich Campbell (personal communi-

VOICE ALTERNATIONS

367

cation) suggests that this restriction might follow from general principles of word formation. Assuming that Russian dispositional verbs are created in the lexicon from non-dispositional verbs, and that the rule applies to V, it is reasonable to suppose that a target V cannot be construed as a word unless it is non-branching. Even so, it will need to be stipulated that the rule applies to V's rather than Vs. Perhaps, then, the problem should be approached from the perspective of predicate-argument structure. The relevant rule might take a predicate with the structure in (69) and change it into a predicate with the structure in (70).

(69) verb (Agent, <0>) (70)

verb-sja (Experiencer,

<0>)

It can thus be made explicit that the verb must have no internal arguments in order to undergo the necessary lexical process. Note that in (70) "Agent" was literally replaced by "Experiencer." This accords with the new interpretation of the logical subject, as well as its case. It is also possible, following suggestions in Schoorlemmer (1991), that the dispositional construction introduces a null modal into the clause. Accordingly, the dative experiencer will be the external argument of that null modal, and the agent will remain as PRO, now controlled by the experiencer NP. However the details of this process work, let us assume that the dispositional construction is created in the lexicon in Russian, and that the restriction that it apply only to verbs with no internal arguments is a result of the fact that it is a lexical word formation rule. Even so, it remains to be explained why no such restriction holds in the other languages. If the Russian restriction is a consequence of lexical principles (or stipulation), then it is reasonable to suppose that its absence in Serbo-Croatian (66) and Polish (67) implies that the dispositional reflexive construction is syntactic in these languages. That is, I suggest that the lack of restriction outside of East Slavic follows from the dispositional construction arising syntactically rather than lexically. The question then reduces to the following: Why do some languages have a syntactic dispositional construction and others a lexical one? I speculate that the reason Russian has a lexical dispositional reflexive construction is that the syntactic one would be incompatible with the properties of -sja in that language. We have seen that Russian -sja only absorbs the case assigned by the verb, perhaps because it can only be in AGR-O, never in AGR-S. Imagine, then, that in the syntactic dispositional construction the reflexive morpheme either must be in AGR-S or cannot be in AGR-O. If so, deriving this construction syntactically will be impossible, hence Russian must avail itself of the lexical option. Notice, however, that the existence of the personal dispositional reflexive construction in a language does not necessarily correlate with the presence of the reflexive + accusative construction, which we took to mean that the reflexive morpheme need not be in AGR-O/absorb the verb's case. Czech, Slovak, and standard Serbo-Croatian are cases in point. However, in these languages it is still clear that the reflexive morpheme has much greater freedom of distribution than in Russian. Aside from its clitic status, the reflexive morpheme appears in active impersonal intransitive sentences of the type illustrated in (71) for Czech, (72) for Slovak, and (73) for Serbo-Croatian:37

368

(71)

PARAMETERS OF SLAVIC MORPHOSYNTAX

a. bydlelo se ve stanech lived(n sg) REFL in tents 'people lived in tents' b. tancovalo se az do rana danced(n sg) REFL even to morning 'people were dancing until morning' c. slo se domii went(n sg) REFL to-home 'people were going home'

(72)

a. zajtra sa pojde na vylet tomorrow REFL will-go(3sg) on excursion 'people are going on an excursion tomorrow' b. najviac sa dnes umiera na srdcove choroby mostly REFL today die(3sg) from heart diseases 'today people mostly die of heart diseases'

(73)

u klubu se pevalo i igralo in club REFL sang(n sg) and danced(n sg) 'people were singing and dancing in the club'

I thus conclude that in these languages the reflexive morpheme can appear in AGR-S and absorb nominative case. It is just that when the accusative is available, as in the passive construction, the reflexive morpheme must absorb it. In terms of the minimalist phrase structure schema, this can be stated as follows: the reflexive morpheme appears in AGR-O if present, but is not restricted to this position. For this kind of approach to work, we must still explain why the reflexive morpheme cannot be in AGR-O in the syntactic dispositional construction. I speculate that this has to do with its relation to the covert modal. The Modal Phrase (MP) is presumably outside the other functional categories. If MP is simply added on top of the standard structure, the result will be roughly as in (74a), taking the reflexive morpheme se to head Agr-SP, or as in (74b), taking it to head MP, and if it replaces the AGR-S projection, the result will be roughly as in (74c). (74)

a. [MP NPexper [M [Agr_SP [se [TP [T [Agr.OP [AGR-O [VP NPagent [V (NP)]]]]]]]]] b. [MP NPexper [se [Agr.SP [AGR-S [TP [T [Agr.OP [AGR-O [VP NPagent [V (NP)]]]]]]]]] c. [MP NPexper [se [TP [T [AGR.OP [AGR-O [VP NPagent [V (NP)]]]]]]]]

Each of these structures raises technical questions. Under structure (74a), the problems are (1) how se, as head of Agr-SP, is connected with MP, and, more seriously, (2) if se absorbs nominative, then why is this case still available in Serbo-Croatian (66a) and Polish (67a), for example. Structure (74b) solves both these problems by locating se in M itself, thereby also freeing up nominative case, but then raises the issue of why the subject (indicated here as NPagent) cannot then raise to SPEC-AgrSP to receive nominative case. Structure (74c) solves this problem by eliminating

VOICE ALTERNATIONS

369

Agr-SP (or, if MP is identified with Agr-SP, by filling the head position and thereby rendering nominative unavailable for NPagent) but at the same time unfortunately also makes this case unavailable for the object NP, as required for Serbo-Croatian (66a) and Polish (67a). This suggests that some kind of mixed approach is still called for, where the reflexive morpheme can appear in various positions. Even so, the existence of the type in Serbo-Croatian (66a) and Polish (67a) indicates that the reflexive morpheme is absorbing accusative case, since otherwise the object NP should not raise for nominative.38 Hence, the idea that it cannot appear in AGR-O may not be viable, and a different kind of solution needs to be sought to the problem of variation in the dispositional reflexive construction. The range of examples found outside Russian actually suggests that the reflexive morpheme in its dispositional function has the same syntactic properties as the reflexive morpheme in passive and middle constructions, being generated either in AGR-O/absorbing the verb's case or in AGR-S/absorbing INFL's case. If so, then it is only the dispositional construction in Russian, with its mysterious prohibition against internal arguments, that remains, as before, to be accounted for. At this point I can see no nonstipulative solution, the question mostly being how best to state the stipulation. If the relevant distinction is lexical versus syntactic, as suggested above, then the fact that this construction is lexical in Russian may have to be stipulated. If it is a matter of whether the target is a V or a V, with V + REFL giving the expected range of possibilities and V + REFL requiring a non-branching V, then this too may have to be stipulated. Ultimately, of course, any differences among the Slavic languages in terms of the dispositional reflexive construction should be made to follow from independent properties of the reflexive morpheme in each language, but just how this is to be accomplished remains unclear. Notes 1. The discussion is largely based on Franks (1982) and Franks (in press). I am indebted to IJSLP for permission to reprint material from the latter article. I also wish to thank Gerald Greenberg, George Fowler, and Oleh Ilnytskyj for advice on the Ukrainian examples and Jerzy Kolodziej and Piotr Drozdowski for their help with Polish. 2. See Rizzi (1986) and Belletti and Rizzi (1988) for discussion of the principles projecting lexical predicate-argument structures onto syntactic representations. Their system is conceptually similar to the one proposed below. 3. See Cinque (1988) for one specific proposal about how this may be formally accomplished in terms of VP assigning the external theta-role to its sister I, which in turn discharges it onto the subject under coindexation. 4. See Rizzi (1986, 509) for a similar distinction, where he uses the term "direct thetarole" to indicate the one role the verb directly assigns. 5. Franks (1985) addresses some of these issues having to do with (i) the source of roles such as time and place that seem to derive from the category meaning of verb and in this sense should not form part of the verb's core predicate-argument structure, and (ii) the choice of particular roles for assignment to particular positions, which seems to follow consistent principles for determining which roles will be selected to be assigned externally and internally. 6. See Levin and Rappaport (1986) for discussion of adjectival passives. 7. The passive morpheme is clearly not nominative when it co-occurs with a nominative

370

PARAMETERS OF SLAVIC MORPHOSYNTAX

subject and prohibits an accusative object, as in the canonical passive construction. Even in the Slavic "passive + accusative" constructions discussed in this chapter, for which I argue that the passive morpheme absorbs nominative, this case is not accessible for syntactic agreement processes; compare the Polish example in (38). The passive morpheme also cannot be in subject position underlyingly, since then this would be a theta-position and the object could not move to it without violating the Projection Principle. Similarly, the passive morpheme cannot be in S-structure subject position in those cases where this position is occupied by the D-structure object, aside from the obvious surface fact that it generally attaches to the verb in some way. 8. See Cinque (1988) for some interesting proposals about Italian si directed toward resolving some of these problems. 9. The basic argument about the significance of the Italian facts for pro-drop and caseabsorption in Slavic was orginally made in Franks (1982). See Ruzicka (1986, 1988) for insightful discussion of these and similar data within a GB treatment of diathesis in Slavic. See also Siewierska (1988) and the references therein for a comprehensive description of the range of passive and passive-like phenomena in Slavic. 10. See also Burzio (1986) for further discussion and a similar analysis. 11. Given the problems engendered by free word order in the analysis of the Slavic constructions, as well as the fact that most Slavic languages are probably [ — FIP] in that they disallow long-distance subject extraction, I ignore throughout the discussion the issue of whether type (14c) is instantiated in Slavic. 12. See Siewierska (1988) for further discussion of this distinction and a generous assortment of relevant examples. 13. Thanks are due to Wayles Browne for drawing my attention to the kajkavian examples in (19) and the Slovenian ones in (20). 14. Browne (personal communication) suggests that this form, which may be translated "one is looking for a younger woman'' or ' 'younger woman wanted,'' is typical of a want ad, perhaps one seeking domestic help. The active impersonal translation implies the presence of a covert agent, the structural location of which is explored in section 8.3.4. I treat (20a) and (20b) as stylistic variants, since Toporisic comments that in many dialects of Slovenian, as well as normative varieties of the literary language, the reflexive + accusative type of expression in (20b) is not customary at all, being replaced by the reflexive + nominative in (20a) or even the active in (i). (i)

iscejo mlajso zensko seek(3pl) younger(acc) woman(acc) 'they are looking for a younger woman'

This is presumably an instance of the alternation between reflexive and third plural arbitrary constructions discussed in section 7.5. 15. See Sapiro (1953) and Kuz'mina and Nemcenko (1971) for copious examples, as well as Timberlake (1976) and Siewierska (1988) for discussion. 16. Although she does not mention anything about the chronology of the loss of overt agents, Siewierska (1988, 271) states that the use of the auxiliary in this construction began to diminish noticeably in the second half of the sixteenth century. See also Doros (1975) for discussion. 17. Thanks are due to Loren Billings and Katarzyna Dziwirek for independently bringing Dyfa (1983) to my attention. 18. Several attempts exist in the literature to account for the passive + accusative phenomenon, although none adequately cover the complete range of facts discussed. Sobin (1985), Babby (1989, in press), Baker (1988), and Baker, Johnson, and Roberts (1989) all observe that

VOICE ALTERNATIONS

371

Ukrainian provides counterevidence to claims made in Chomsky (1981a) that passive morphology universally absorbs accusative case. Sobin suggests that case absorption (or blocking) is optional in Ukrainian passives; when it fails to occur, the object NP has no need to move to subject position to receive case. Babby contends that the behavior of Ukrainian derives from the optionality of subject position in principle and in Ukrainian passives in particular; movement thus cannot take place when the subject position is absent and must when it is present. Baker, Johnson, and Roberts argue that the passive morpheme is an argument, and as such its theta-role can be licensed either by receiving case or by morphological incorporation; the Ukrainian pattern results from its passive morpheme allowing both options. I mention these analyses mostly for the sake of completeness. Each of these has its insights and pitfalls, but a thorough treatment of these approaches cannot be undertaken within the confines of this chapter. 19. One construction that is inconsistent with this generalization is the dispositional reflexive construction of the type Russian mne ne spitsja 'l(dat) can't sleep/don't feel sleepy', in which -sja clearly accompanies loss of nominative. Section 8.4 offers an in-depth cross-Slavic treatment of this interesting phenomenon. 20. Interestingly, Fowler's passive test indicates that dative is never assigned to true complements of V, being reserved for indirect objects instead. One Russian speaker, however, accepted passive forms ofpodrazat' 'to imitate', with the dative argument of the active appearing in the nominative in the passive, as follows: (i) (ii)

peredovye rabotniki podrazajut inostrannym metodam forefront workers(nompl) imitate(3pl) foreign methods(dat pi) inostrannye melody podrazajut -sja peredovymi rabotnikami foreign methods(nom pi) imitate(3pl)-REFL forefront workers(inst pi)

This example is in fact cited in Freidin (1992, 206-207) in support of his erroneous claim that quirky case—assigning verbs in Russian never passivize. 21. At the very least, middle formation saturates the agent theta-role, rendering it syntactically inaccessible; compare Fagan (1988). Whether or not the active internal argument is lexically externalized in the middle is unclear. For Russian, the genitive-of-negation test suggests that the middle subject may be a D-structure unaccusative object, although the data are far from conclusive. 22. Applying the traditional term "impersonal," as in Babby (1989), is thus somewhat misleading, since passive + accusative constructions are inherently agentive (cf. Siewierska, 1988, for discussion). Of course, use of this misnomer probably relates to the observation that there can be neither person-number agreement nor an overt subject here. 23. As expected, expressing the agent with an adjunct przez-phrase is similarly impossible, but this must be independently blocked, since these are also ruled out in the passive + accusative construction. One way of explaining this restriction is suggested in section 8.3.4. 24. Ouhalla (1991) similarly argues for the presence of PRO in passives. 25. There is of course no compelling reason to assume that PRO is adjoined to the left rather than the right. Pesetsky (1982) in fact argues that PRO subjects are to the right in order to explain the possibility of wanna-contraction. 26. In the system espoused by Bailyn and Rubin (1991), instrumental arises within a Predicate Phrase and is thus assimilated to the SPEC-head agreement approach employed for structural cases. 27. Chomsky (1992) explicitly states that the "-S" and "-O" extensions are simply mnemonics, the distinction apparently reducing to whether AGR is located above or below TP. I am not sure that this reduction is warranted, in that the cases associated with each are

372

PARAMETERS OF SLAVIC MORPHOSYNTAX

distinct, as are the agreement morphemes in languages that show both subject and object agreement. In addition, Chomsky will probably need (at least) an AGR-IO as well, since this too, in his system, should be a structural case. Perhaps the phonological realization of AGR could be modulated by the verbal (or other) head that adjoins to it; I leave these issues unresolved. 28. One could eschew functional categories altogether, of course, and take the traditional transformational grammar route, as proposed in Babby (1975), for example, and treat -sja basically as an object, which would lead to analyzing Polish sie as a D-structure subject in passive constructions, assigned theta-role and nominative case as any other subject is. Generalizing this kind of non-functional approach seems to me to lead to serious problems for the analysis of middles and unaccusatives, however, and is not compatible with theta-theory. Consider, for example, the problem posed by Ukrainian dialects (as well as Slovenian and kajkavian Serbo-Croatian) in which the reflexive element can absorb either case with no nonstylistic impact on meaning. 29. The verb might raise to the clitic at LF. 30. It may be that this generalization is breaking down in the colloquial language, in that several speakers consulted report genitive subjects of unergative and even transitive subjects as acceptable, as in (i) and (ii), respectively: (i)

%nad etoj zada&j ne rabotalo ni odnogo studenta on this problem NEG worked(n) not one(gen) student(gen)

(ii)

%etu knigu ne citalo ni odnogo studenta this(acc) book(acc) NEG read(n) not one(gen) student(gen)

This process is greatly facilitated by—and perhaps even requires—the negative scope particle ni. Furthermore, it is impossible for both arguments in examples such as (ii) to be marked genitive simultaneously. This suggests that scrambling somehow restructures the clause so that the object is removed from the VP—or at least from direct object position—and the subject is scrambled to a position inside VP. Since placing the subject literally in object position would violate virtually any version of the Projection Principle, I suggest that the subject must instead be scrambled to adjoin to V. This might explain its genitive case but runs up against two problems of improper movement: an XP is not supposed to be able adjoin to a head, assuming Baltin's "like-attracts-like" principle," and lowering operations are generally disfavored, assuming that traces must be c-commanded by their antecedents. Conceivably, however, raising at LF undoes both of these illicit aspects of the proposed representation. 31. Somewhat more frequent are verbs that assign quirky genitive and have past passive participles. Some speakers consulted find the negated passive of dostic' 'to achieve' in (i) marginally acceptable and in clear contrast to the affirmative passive in (ii). (i)

%e ne bylo dostignuto ni odnoj iz na§ix celej NEG was(n) achieved(n) not one(gen) of our goals 'not one of our goals was achieved'

(ii)

*e bylo dostignuto odnoj iz nasix celej was(n) achieved(n) one(gen) of our goals

It is possible that here the genitive assigned under negation accidentally satisfies the lexical requirements of the verb dostic'. 32. Notice that we have to rely on the semantics to tell that Polish (57b) is the negation of passive (18b), rather than middle (18c). 33. Exactly how choosy languages and speakers are about the adverbial and/or negation

VOICE ALTERNATIONS

373

seems to vary. Polish, for example, is more restrictive than Russian; Serbo-Croatian less so. 34. The examples in (62) and (63) were erroneously presented as grammatical in Franks (1982). 35. Despite superficial similarities, I do not regard examples such as Russian (i) and (ii) as exceptions to this paradigm: (i)

mne ne idet-sja me(dat) NEG go -REFL 'things are not going well for me'

(ii)

mne ne xodet-sja rabotat' me(dat) NEG want -REFL to-work 'I don't feel like working'

Instead, I take these as idiosyncratic lexical items, not derived by the process that creates the dispositional construction. After all, (i) is an idiomatic expression and does not mean 'I don't feel like going,' and (ii) does not mean 'I don't feel like wanting (to work)'. 36. The acceptability of this construction in Polish with the object NP remaining in situ, as in (i) with the accusative or even (ii) with the genitive of negation, seems to vary among speakers: (i)

%dobrze mi sie pisze te prace, well me(dat) REFL write(Ssg) this(acc) work(acc) ' I can write this work easily'

(ii)

%nie pisze mi sie dobrze tej pracy NEG write(3sg) me(dat) REFL well this(gen) work(gen) ' I can't write this work easily'

37. These examples are all cited in Ruzicka (1988), to which the reader is referred for references and further discussion. 38. Although this is required in Serbo-Croatian, the Polish examples in n. 36 suggest that either nominative or accusative can be absorbed.

9 Summary and Conclusions

In this book, I have surveyed a comprehensive range of major aspects of Slavic syntax from the general perspective of Principles and Parameters theory. I would like in this final chapter briefly to summarize the results of this survey. The core topics covered the case-theoretic properties of numeral phrases, secondary predication, null subjects, and voice. Treatment of these problems led inevitably to discussion of a host of related issues and phenomena, as well as to the proposal of diverse mechanisms for expressing the differences among Slavic languages. The specific areas were of course selective and are by no means meant to provide an exhaustive account of Slavic syntax. My intent instead has been to demonstrate that by reconsidering familiar puzzles from the parametric perspective one can arrive at novel insights. In investigating, cataloguing, and analyzing related syntactic constructions across the Slavic languages, I have come to understand that on the one hand fundamental differences often lie beneath superficially similar constructions, and on the other that simple oppositions often turn out to explain a host of superficially unrelated phenomena. The areas studied are representative of the kinds of problems that can be addressed and the solutions Government and Binding has to offer. I have not attempted to integrate the discrete parametric properties encountered into a general typology of parametric variation, approaching the problem of variation as a practical rather than theoretical one. At our present state of knowledge, many questions remain open about the form and function of parameters. Aside from familiar issues of learnability, the range of possible dimensions of parametric variation and the possible forms of individual parameters need to be determined, and the kinds of interactions between different parameters need to be investigated in more detail. By focusing on contrastive Slavic syntax in this light, I hoped that the results would provide fruitful insight into both the Principles and Parameters model and the kinds of parametric variation operative in the Slavic languages. In the first two chapters I sketched out aspects of GB theory as presented in Chomsky (198 la, 1986b) and argued for a theory of case that breaks case down into 374

SUMMARY AND CONCLUSIONS

375

morphosyntactic features and assigns case under coindexation. In chapter 8 the possibility of couching this in terms of case-checking was considered, but I believe this reformulation is mostly a matter of execution. Chapter 3 discussed across-theboard dependencies in Polish and Russian within the context of case theory and theta-theory. Chapter 4 and subsequent chapters dealt with parametric issues in Slavic syntax. In these chapters, the following kinds of distinctions were identified as playing a possible role in accounting for the observed range of variation: (1) a. b. c. d. e. f. g. h. i.

the features of a particular morphological case the maximal category of a particular extended projection complement selection the phrase structural contexts for particular configurational case assignment rules the level by which a particular configurational case must be assigned whether or not a particular configurational case must be discharged the functional positions in which a particular morpheme can be generated the argument positions in which a particular morpheme can be generated the level at which a particular morphological operation applies

Although I take the types of choices in (1) as representative of the factors that are typically invoked to account for parametric variation, it would be advantageous to delimit these in some explanatory way. For the most part, the sources of variation considered in this book reflect lexical properties and, as a consequence, are idiosyncratic and somewhat stipulative in character. This is desirable, I believe, in the sense that variation should be couched in terms of specific properties of individual lexical items as much as possible. Languages differ in their lexical items, after all; language learning must therefore largely consist of identifying and representing idiosyncratic properties of all elements of the lexicon. Of the distinctions in (1), (Ic, g, h) are lexical properties of heads, and (Id, e, f, i) can also probably be stated in terms of head properties, especially in a well-articulated feature licensing/checking theory. If cases are feature bundles, the variation expressed in (la) is the expected analogy to how phonological features define phonemes. Lastly, I suspect that (Ib) can be accommodated within a careful theory of Canonical Structural Realization, although how this is to incorporate variation is a completely unexplored area. Of course, even if checking and CSR theories prove adequate, the important question remains of how children acquiring the language are able to identify the necessary lexical factors. This question is best addressed by (1) narrowing the range of admissible choices, and (2) identifying easily observable properties of the relevant lexical items. Unfortunately, it is far from clear that the particular factors I have suggested to account for parametric variation in the Slavic languages can be made to conform to these desiderata. Let us review the major proposals from this perspective in turn. It was argued that the structural/inherent dichotomy of GB is best regarded in terms of the Jakobsonian feature [ ± oblique], and that the genitive case assigned under quantification varies along this dimension. In order to delimit this kind of variation, a restrictive theory of case features will be required. I assume that there is a universal inventory, along the

376

PARAMETERS OF SLAVIC MORPHOSYNTAX

lines of phonological distinctive features. Focusing, then, on the [ ± oblique] distinction, there must be some general way of identifying cases as [ +oblique] or [ - oblique]. I suggest that the primary way of characterizing [ + oblique] cases is in terms of theta-theory in that only [ + oblique] cases can be theta-related, in the sense that a particular case is required by a particular lexical item for LF-visibility of a particular theta-role. Although it was shown that by no means all instances of [ + oblique] NPs are theta-related, the point is that in principle they can be. That is, it suffices for the child to identify some use of the instrumental, genitive, dative, or locative in Russian as lexically required to know this case is [ + oblique] regardless of its source. In Russian, nominative and accusative are never lexically required, hence these case are [ — oblique] by default. The trigger for [ + oblique] is thus that it must be stipulated for some lexical item, in association with that item's thematic properties. This is why the status of the genitive of quantification is able to vacillate—it looks on the one hand as though it is assigned by specific items, but on the other hand this case is never associated with a particular theta-role. Languages therefore resolve this dilemma in two ways, treating GEN(Q) as either structural or inherent. As shown in chapter 5, Polish even exercises both options. It was also argued that in Russian numeral phrases can be QPs, whereas in other languages they are always DPs. Presumably, children assume that a phrase instantiates the CSR of an element with its particular semantic role, everything else being equal. Entities that are agents, patients, and so forth will thus be analyzed as DPs, unless there is some evidence to the contrary. For this reason, I would claim that Russian exhibits the marked system, where numeral phrases can be bare QPs, as evidenced by a host of factors, most notably the optional lack of subject-verb agreement with quantified subjects. In chapter 61 claimed that the second dative phenomenon is related to the fact that some Slavic languages tolerate dative subjects and others do not. It was suggested that dative is assigned to sisters of tensed, non-agreeing INFL by generalizing the indirect object phrase structure schema. While this is not a matter of lexical idiosyncrasy, I think there must also be room for variation in the range of application of specific configurational case assignment rules. That is, the child must determine which cases are assigned under which configurations. Let us assume that the existence of dative indirect objects leads to the conclusion that sisters of a [ + V, - N]' node can be marked dative. Note that the conclusion that this case is [ +oblique] dative yet structurally determined is not inconsistent, since it can be independently established that dative is sometimes required as a quirky lexical requirement of specific verbs and prepositions. Now, in order to extend this to some functional projection of V, I propose that slight modification in the features required of the relevant X' is required. Although I have been vague both about which functional category should be implicated in the dative subject rule, as well as the features of the various possible categories, it is reasonable to suppose that functional projections of a lexical category are minimally distinct in part-of-speech feature terms from the lexical category itself. Grimshaw (1991), for example, argues for the idea that functional projections constitute "extended projections" of lexical heads, so that they have the same [ ± N, ± V] features but are also marked for degree of functional projection above the lexical head.

SUMMARY AND CONCLUSIONS

377

Chapter 7 proposed two core parameters of Slavic pro-drop phenomena. The basic null subject parameter, dubbed out of markedness considerations the Overt Subject Parameter, restricts nominative case assignment to the syntax proper in certain languages. It is argued that case plays a visibility role at both PF and LF, the latter being its essential property. Under a checking approach, this analysis could conceivably be stated in terms of the level by which checking must take place, perhaps exploiting Chomsky's (1992) idea that it will be procrastinated until LF if at all possible. It is far from clear how the OSP could be implemented beyond relegating particular rules to particular levels, but one method in keeping with the minimalist system might be in terms of Chomsky's approach to AGR. He claims that weak AGR is invisible at PF, whereas strong AGR is visible, and since AGR is not a legitimate PF object it must delete by PF, but only when visible. Hence, if [ + OSP] languages had strong AGR, and if nominative case assignment/checking results in AGR's deletion, then one might contend that it is strong AGR that implies [ + OSP]. I see two troublesome aspects of this explanation. For one thing, the argumentation seems somewhat circular, in that there is no independent criterion of "strong" AGR. If anything, the traditional view is that in null subject languages AGR is strong in the sense that it is more transparent morphologically and can often stand on its own, rather than the other way around. The other issue is that, as mentioned in Chomsky (1992), one must distinguish two senses of AGR: it plays the dual roles of (1) checking agreement on the verb and (2) checking nominative case, typically on an NP in its specifier. It is the latter role that is relevant for the OSP, hence it must be in this sense that AGR is strong in East Slavic, even though it is clearly weaker in East Slavic than in South and West Slavic in the more familiar sense of verbal morphology. The other core parameter proposed in chapter 7, again out of markedness considerations dubbed the Overt Expletive Subject Parameter, stated that in some languages nominative case must be discharged, whether or not motivated at LF. Although here it is much easier to see what might trigger setting this parameter to [ + OESP]—simply the presence of overt subject expletives when not otherwise sanctioned—the question of why this should be a parameter of core grammar remains. It seems to me that an account in terms of AGR's role in assigning/checking nominative case might again be possible. Following Chomsky (1992), there are two sets of features associated with AgrP: case features directed toward the specifier of AgrP and instantiated through SPEC-head agreement; and agreement features on the head node AGR°. The former set of features are nominal features, the latter set are verbal features. Let us assume that the two, although clearly related somehow, are technically independent and that the existence and properties of functional categories are determined by lexical categories. Furthermore, if AGR has case features to discharge/check, then they must at some point be discharged/checked. Now the [ ± OESP] can be couched in terms of whether or not AGR actually has case features when their presence is not motivated by theta-theoretic concerns. In other words, in [ — OESP] languages AGR can have agreement features without also having case features, whereas in [ + OESP] languages the presence of agreement features necessarily implies the presence of case features. Finally, chapter 8 considered a variety of ways of conceiving of variation in the

378

PARAMETERS OF SLAVIC MORPHOSYNTAX

behavior of voice morphemes that in some way adjust the relationship between lexically represented predicate-argument structures and the surface positions in which they are realized. The chief observational problem was that these morphemes seem to be in complementary distribution with either the subject's case or the object's case. That is, the morphemes vary in terms of whether they absorb case from INFL (subject AGR) or V (perhaps mediated through object AGR). Although far from conclusive, various possible ways of formalizing this observation were discussed. These included (1) generating the morpheme in question in INFL and allowing it to absorb either the case of V or the case of INFL/AGR, (2) generating the morpheme in either AGR-S or AGR-O with similar effect, (3) generating the morpheme either in INFL/AGR or in subject position (SPEC-VP, an argument position), and (4) introducing the morpheme either lexically or syntactically. This last possibility, combined with the checking theory, amounts to distinguishing morphemes that do not need to be checked in the syntax from those that do. This last chapter was the most speculative and inconclusive. The variety of constructions across the languages examined, coupled with the range of conceivable analyses of these constructions, draws attention, I believe, to the urgent need to delimit possible parameters of variation. Until a sufficiently restrictive theory of parametric variation is developed, I see no simple way to decide among alternative analyses. Clearly, the goal of such a theory should be to account for how children are able to project appropriate grammars from the primary data. This requires principled limitations on the kinds of variation in grammar that they might entertain, however, probably combined with explicit selection mechanisms. There are two distinct aspects to this notion of selection mechanisms: 1. The notion of possible triggers for parameter (re)setting needs to be refined. 2. An evaluation metric for determining which among competing analyses is least costly needs to be adopted. Considerable headway has been made in both these directions. Lightfoot (1989) is among important attempts to restrict the nature of accessible triggering data, and Chomsky (1991, 1992) puts forward a promising program for defining the relative cost, hence accessibility, of particular analyses. Chomsky's basic idea is that the grammar only invokes what it absolutely needs to in order to lead to legitimate representations. While the spirit of his minimalist program is admirable, the larger problem remains of delimiting the range of specific parametric choices and the analyses they lead to. Until this issue is resolved, it will be difficult to decide among alternative solutions to the Slavic problems considered in this book. It is my hope that the speculations about Slavic parametric syntax offered in this volume will promote and encourage further research into the mechanisms of syntactic diversity, with the ultimate goal of uncovering an explanatory theory of Universal Grammar.

References

Abney, Steven. 1987. 'The English Noun Phrase in Its Sentential Aspect". Ph.D. dissertation, MIT. Anderson, Carol. 1983. 'Generating Coordinate Structures with Asymmetric Gaps'. Chicago Linguistic Society 19:3-14. Anderson, Mona. 1984. 'Prenominal Genitive NPs'. Linguistic Review 3:1-24. Andrews, Avery. 1971. 'Case Agreement of Predicate Modifiers in Ancient Greek'. Linguistic Inquiry 2:127-151. . 1976. "The VP Complement Analysis in Modern Icelandic'. Papers from the Sixth Meeting of the North Eastern Linguistic Society (Montreal), Working Papers in Linguistics 6:1-21. Aoun, Joseph. 1986. Generalized Binding. Foris, Dordrecht. Aoun, Joseph, and Dominique Sportiche. 1983. 'On the Formal Theory of Government'. Linguistic Review 2—3:19-37. Aoun, Joseph, Norbert Hornstein, David Lightfoot, and Amy Weinberg. 1987. 'Two Types of Locality'. Linguistic Inquiry 18:537-578. Authier, J.-Marc: 1989. 'V-governed Expletives, Case Theory, and the Projection Principle'. Paper presented at the Eighth West Coast Conference on Formal Linguistics. Avrutin, Sergey. 1992. 'Movement of Bound Variables'. Third Leiden Conference for Junior Linguists. Leiden, Netherlands, pp. 1-16. Babby, Leonard. 1974. 'Towards a Formal Theory of "Part of Speech" '. In Richard Brecht and Catherine Chvany (eds.), Slavic Transformational Syntax, Michigan Slavic Materials, No. 10, University of Michigan, Ann Arbor, pp. 151-181. . 1975a. A Transformational Grammar of Russian Adjectives. Mouton, The Hague. . 1975b. 'A Transformational Analysis of Transitive -sja Verbs in Russian'. Lingua 35:297-332. . 1980a. "The Syntax of Surface Case Marking'. In Wayne Harbert and Julia Herschensohn (eds.), Cornell Working Papers in Linguistics 1. Department of Modern Languages and Linguistics, Cornell University, Ithaca, NY, pp. 1-32. . 1980b. Existential Sentences and Negation in Russian. Karoma, Ann Arbor, MI. . 1980c. 'Word Order, Case, and Negation in Russian Existential Sentences'.

379

380

REFERENCES

In Catherine Chvany and Richard Brecht (eds.), Morphosyntax in Slavic. Slavica, Columbus, OH, pp. 221-234. . 1984. 'Case Conflicts and Their Resolution'. In Wayne Harbert (ed.), Cornell Working Papers in Linguistics 6. Department of Modern Languages and Linguistics, Cornell University, Ithaca, NY, 1-21. . 1985. 'Prepositional Quantifiers and the Direct Case Condition'. In Michael Flier and Richard Brecht (eds.), Issues in Russian Morphosyntax. Slavica, Columbus, OH, pp. 91-117. . 1986, 'The Locus of Case Assignment and the Direction of Percolation'. In Richard Brecht and James Levine (eds.), Case in Slavic. Slavica, Columbus, OH, pp. 170219. . 1987. 'Case, Pre-Quantifiers, and Discontinuous Agreement in Russian'. Natural Language and Linguistic Theory 5:91-138. . 1989. 'Subjectlessness, External Subcategorization, and the Projection Principle'. Zbornik Matice Srpske zafilologiju i lingvistiku 37:7-40. . In press. 'Morpholexical Aspects of Russian Syntax: Verbal Paradigms and Voice'. International Journal of Slavic Linguistics and Poetics. Babby, Leonard, and Richard Brecht. 1975. 'The Syntax of Voice in Russian'. Language 51:342-367. Bachman, Richard. 1980. 'The Subject of the Dative Case in Modern Russian'. Ph.D. dissertation, Ohio State University. Bailyn, John. 1991. 'The Configurationality of Case Assignment in Russian'. In Almeida Toribio and Wayne Harbert (eds.), Cornell Working Papers in Linguistics 9. Department of Modern Languages and Linguistics, Cornell University, Ithaca, NY, pp. 5798. Bailyn, John, and Edward Rubin. 1991. 'The Unification of Instrumental Case Assignment in Russian'. In Almeida Toribio and Wayne Harbert (eds.), Cornell Working Papers in Linguistics 9. Department of Modern Languages and Linguistics, Cornell University, Ithaca, NY, pp. 99-126. Baker, Carl Lee. 1979. 'Syntactic Theory and the Projection Problem'. Linguistic Inquiry 10:533-581. . 1991. 'The Syntax of English Not: The Limits of Core Grammar'. Linguistic Inquiry 22:387-429. Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing. University of Chicago Press, Chicago. Baker, Mark, Kyle Johnson, and Ian Roberts. 1989. 'Passive Arguments Raised'. Linguistic Inquiry 20:219-251. Baltin, Mark. 1982. 'A Landing Site Theory of Movement Rules'. Linguistic Inquiry 13:1-38. . 1991. 'Head Movement in Logical Form'. Linguistic Inquiry 22:225-249. Barss, Andrew, and Howard Lasnik. 1986. 'A Note on Anaphora and Double Objects'. Linguistic Inquiry 17:347-354. Battistella, Edwin. 1981. 'Topics in the Theory of Inflection'. Ph.D. dissertation, City University of New York. Bauer, Jaroslav, and Miroslav Grepl. 1980. Skladba spisovne destiny. Statni Pedagogicke Nakladelstvi, Prague. Beard, Robert. 1991. 'Lexeme-Morpheme Base Morphology: A General Theory of Inflection and Word Formation'. Manuscript, Bucknell University. Belletti, Adriana. 1988. 'The Case of Unaccusatives'. Linguistic Inquiry 19:1-34. Belletti, Adriana, and Luigi Rizzi. 1988. 'Psych-verbs and G-theory'. Natural Language and Linguistic Theory 6:291-352.

REFERENCES

381

Berent, Gerald. 1980. 'On the Realization of Trace: Macedonian Clitic Pronouns'. In Catherine Chvany and Richard Brecht (eds.), Morphosyntax in Slavic. Slavica, Columbus, OH, pp. 150-186. Borer, Hagit. 1984. Parametric Syntax. Foris, Dordrecht. Borer, Hagit, and Kenneth Wexler. 1987. "The Maturation of Syntax". In Thomas Roeper and Edwin Williams (eds.), Parameter Setting. Reidel, Dordrecht, pp. 123-172. Borkovskij, V. I., and P. S. Kuznecov. 1965. Istoriceskaja gratnmatika russkogo jazyka. Nauka, Moscow. Borras, F. M., and R. F. Christian. 1971. Russian Syntax. Oxford University Press, Oxford. Borsley, Robert. 1983. 'A Note on the Generalized Left Branch Condition'. Linguistic Inquiry 14:169-174. Bouchard, Denis. 1982/1984. 'On the Content of Empty Categories'. Ph.D. dissertation, MIT; Foris, Dordrecht. Brodowska-Honowska, Maria. 1964. 'Przysiowkowe samemu'. Prace Filologiczne 18:277— 293. Browne, Wayles. 1986. Relative Clauses in Serbo-Croatian in Comparison with English. New Studies 4. Institute of Linguistics, University of Zagreb. Browning, Marguerite. 1987. 'Null Operator Constructions'. Ph.D. dissertation, MIT. Bulaxovskij, Leonid Arsenevich. 1949. Kurs russkogo literaturnogo jazyka. Radjans'ka Skola, Kiev. . 1958. Istoriceskij kommentarij k russkomu literaturnomu jazyku. 5th ed. Radjans'ka Skola, Kiev. Burzio, Luigi. 1986. Italian Syntax. Reidel, Dordrecht. Carrier-Duncan, Jill. 1985. 'Linking of Thematic Roles in Derivational Word Formation'. Linguistic Inquiry 16:1-34. Chomsky, Noam. 1957. Syntactic Structures. Mouton, The Hague. -.1970. 'Remarks on Nominalization'. In Roderick Jacobs and Peter Rosenbaum (eds.), Readings in English Transformational Grammar. Ginn, Waltham, MA, pp. 184-221. . 1975. Reflections on Language. Random House, New York. . 1977. 'On Wh-movement'. In Peter Culicover, Thomas Wasow, and Adrian Akmajian (eds.), Formal Syntax. Academic Press, New York, pp. 71-132. . 1981a. Lectures on Government and Binding. Foris, Dordrecht. . 1981b. 'Principles and Parameters in Syntactic Theory'. In Norbert Hornstein and David Lightfoot (eds.), Explanation in Linguistics. Peter Longmans, London, pp. 32-75. . 1982. Some Concepts and Consequences of the Theory of Government and Binding. MIT Press, Cambridge, MA. . 1986a. Barriers. MIT Press, Cambridge, MA. . 1986b. Knowledge of Language: Its Nature, Origin, and Use. Praeger, New York. . 1991.' Some Notes on Economy of Derivation and Representation'. In Robert Friedin (ed.), Comparative Grammar. MIT Press, Cambridge, MA. . 1992. A Minimalist Program for Lingustic Theory. MIT Occasional Papers in Linguistics. Chomsky, Noam, and Howard Lasnik. 1977. 'Filters and Control'. Linguistic Inquiry 8:425504. Chvany, Catherine. 1970. 'Deep Structures with dolzen and moc'. Ph.D. dissertation, Harvard University. . 1973. 'Notes on "Root" and "Structure-Preserving" in Russian'. In You Take the High Node and I'll Take the Low Node, CLS Parassesion 9, pp. 252-290.

382

REFERENCES

. 1975. BE-Sentences in Russian. Slavica, Columbus, OH. . 1986. 'Jakobson's Fourth and Fifth Dimensions: On Reconciling the Cube Model of Case Meanings with the Two-Dimensional Matrices for Case Forms'. In Richard Brecht and James Levine (eds.), Case in Slavic. Slavica, Columbus, OH, pp. 107129. Cinque, Guglielmo. 1988. 'On si Constructions and the Theory of arb'. Linguistic Inquiry 19:521-581. Cole, Peter, Gabriella Hermon, and Li-May Sung. 1990. 'Principles and Parameters of Longdistance Reflexives'. Linguistic Inquiry 21:1-22. Comrie, Bernard. 1974. 'The Second Dative: A Transformational Approach'. In Richard Brecht and Catherine Chvany (eds.), Slavic Transformational Syntax. Michigan Slavic Materials, No. 10, University of Michigan, Ann Arbor, pp. 123-150. . 1976. 'The Syntax of Action Nominals: A Cross Linguistic Study'. Lingua 40:177201. Contreras, Heles. 1984. 'A Note on Parasitic Gaps'. Linguistic Inquiry 15:704—713. . 1993. 'On Null Operator Structures'. Natural Languages and Linguistic Theory 11:1— 30. Corbett, Greville. 1975. 'Constructions with scitat' nuznym'. Russian Linguistics 2:11-21. . 1978. 'Numerous Squishes and Squishy Numerals in Slavonic'. International Review of Slavic Linguistics 3:43-73. . 1979. 'Adjective Movement'. Nottingham Linguistic Circular 8:1-10. . 1983a. Hierarchies, Targets and Controllers: Agreement Patterns in Slavic. Pennsylvania University Press, University Park. . 1983b. 'Raising Revisited'. In Willy Birkenmaier (ed.), Studies in Descriptive Russian Grammar. Julius Groos Verlag, Heidelberg, pp. 110-120. . 1986. 'The Use of the Genitive or Accusative for the Direct Object of Negated Verbs in Russian: A Bibliography'. In Richard Brecht and James Levine (eds.), Case in Slavic. Slavica, Columbus, OH, pp. 361-372. Cowper, Elizabeth. 1985. 'Parasitic Gaps, Co-ordinate Structures and the Subjacency Condition'. Proceedings of the North Eastern Linguistic Society 15. Graduate Students Linguistic Association, University of Massachusetts, Amherst, pp. 75-86. Crockett, Dina. 1976a. 'The Distributivepo in Contemporary Russian'. In Thomas Magner (ed.), Slavic Linguistics and Language Teaching. Slavica, Columbus, OH, pp. 153163. . 1976b. Agreement in Contemporary Standard Russian. Slavica, Columbus, OH. Deprez, Viviane. 1989. 'On the Typology of Syntactic Positions and the Nature of Chains'. Ph.D. dissertation, MIT. Dickey, Stephen. 1992. 'Serbo-Croation Distributive po'. Paper presented at the Conference on Balkan and South Slavic Linguistics, Literature, and Folklore, Chicago. Di Sciullo, Anna Maria, and Edwin Williams. 1987. On the Definition of Word. MIT Press, Cambridge, MA. Doros, Aleksander. 1975. Werbalne konstrukcje bezosobowe wjezyku rosyjskim ipolskim na tie innych jezykow stowianskich. Polskiej Akademii Nauk, Krakow. Dowty, David. 1982. 'Grammatical Relations and Montague Grammar'. In Pauline Jakobson and Geoffrey Pullum (eds.), The Nature of Syntactic Representation. Reidel, Dordrecht. Dyia, Stefan. 1983. 'Some Further Evidence Against an Impersonal Passive Analysis of Polish Impersonal Constructions'. Studia anglicaposnaniensa 15:123-128. . 1984. 'Across the Board Dependencies and Case in Polish'. Linguistic Inquiry 15:701-705.

REFERENCES

383

Dziwirek, Katarzyna. 1990. 'Default Agreement in Polish'. In Katarzyna Dziwirek, Patrick Farrell, and Errapel Mejias-Bikadi (eds.), Grammatical Relations: A Cross-theoretical Perspective. Center for the Study of Language and Information, Stanford University, Stanford, CA. . 1991. 'Aspects of Polish Syntax'. Ph.D. dissertation, University of California, San Diego. Emonds, Joseph. 1976. A Transformational Approach to English Syntax. Academic Press, New York. Engdahl, Elisabet. 1983. 'Parasitic Gaps'. Linguistics and Philosophy 6:5-34. Epstein, Sam. 1984. 'Quantifier-pro and the LF Representation of PRO^'. Linguistic Inquiry 15:499-504. Fabb, Nigel. 1984. 'Syntactic Affixation'. Ph.D. dissertation, MIT. Pagan, Sarah. 1988. 'The English Middle'. Linguistic Inquiry 19:181-203. Fafike, Helmut. 1981. Grammatik der Obersorbischen Schriftsprache der Gegenwart: Morphologic. Domowina, Bautzen. . 1987. 'Bemerkungen zur Syntax quantifizierender Lexeme (am Beispiel des Sorbischen)'. Zeitschrift fur Phonetik, Sprachwissenschaft und Kommunikationsforschung 40:82-91. Feldstein, Ronald. 1987. 'Czyta kontra czytaj i ustalenie polskich koricowek koniugacyjnych'. Poradnik Jezykowy 2:91-99. Fillmore, Charles. 1968. 'The Case for Case'. In Emmon Bach and Robert Harms (eds.), Universals in Linguistic Theory. Holt, Rinehart and Winston, New York, pp. 1-88. Fodor, Jerry. 1983. The Modularity of Mind. MIT Press, Cambridge, MA. Fowler, George. 1987a. "The Syntax of the Genitive Case in Russian'. Ph.D. dissertation, University of Chicago. . 1987b. 'Oblique Passivization in Russian'. Paper presented at American Association of Teachers of Slavic and East European Languages Annual Meeting, Washington, D.C. Franks, Steven. 1981. 'Deep and Surface Case'. Proceedings of the North Eastern Linguistic Society 11. Graduate Students Linguistic Association, University of Massachusetts, Amherst, pp. 79-95. . 1982. 'Is There a Pro-drop Parameter for Slavic?' Chicago Linguistic Society 18:140154. . 1983. 'Case and Control in Polish'. In Wayne Harbert (ed.), Cornell Working Papers in Linguistics 4. Department of Modern Languages and Linguistics, Cornell University, Ithaca, NY, pp. 23-49. . 1985. 'Matrices and Indices: Some Problems in the Syntax of Case'. Ph.D. dissertation, Cornell University. 1986. 'Case and the Structure of NP'. In Richard Brecht and James Levine (eds.), Case in Slavic. Slavica, Columbus, OH, pp. 220-243. . 1988. 'On Theta-theory and Case Theory'. Proceedings of the Eastern States Conference on Linguistics 5:151-162. . 1990a. 'Case, Configuration and Argumenthood: Reflections on the Second Dative'. Russian Linguistics 14:231-254. . 1990b. 'Vowel-zero Alternations and Syllable-Counting Morphology'. Indiana Slavic Studies 5:81-98. . 1990c. 'On the Status of Null Expletives'. Lingua 81:1-24. . 1990d. 'The Position of Subjects and the QP Hypothesis'. In Denis Meyer, Satoshi Tomioka, and Leyla Zidani-Eroglu (eds.), Proceedings of the Formal Linguistics Society of Midamerica 1:114—128.

384

REFERENCES

. 1992. 'A Prominence Constraint on Null Operator Constructions'. Lingua 87:35-54. . 1993. 'On Parallelism in Across-the-Board Dependencies'. Linguistic Inquiry 24:509-529. . 1994. 'Parametric Properties of Numeral Phrases in Slavic'. Natural Language and Linguistic Theory 12:570-649. . In press. 'Empty Subjects and Voice-altering Morphemes in Slavic'. International Journal of Slavic Linguistics and Poetics. Franks, Steven, and Katarzyna Dziwirek. 1993. 'Negated Adjunct Phrases Are Really Partitive'. Journal of Slavic Linguistics 1:280-305. Franks, Steven, and Gerald Greenberg. 1988. 'Agreement, Tense, and the Case of Subjects in Slavic'. Chicago Linguistic Society, Parasession on Agreement 24:71-86. . 1994. 'The Functional Structure of Slavic Clauses'. In Jindfich Toman (ed.), Proceedings of the Workshop on Functional Categories in Slavic Syntax. Michigan Slavic Publications, University of Michigan, Ann Arbor, pp. 77-108. Franks, Steven, and Norbert Hornstein. 1989. 'Governed PRO'. McGill Working Papers in Linguistics 6:167-191. . 1992. 'Secondary Predication in Russian and Proper Government of PRO'. In Richard Larson, Sabine latridou, Utpal Lahiri, and James Higginbotham (eds.), Control and Grammar. Kluwer, Dordrecht, pp. 1-50. Franks, Steven, and Richard House. 1982. 'Genitive Themes in Russian'. Chicago Linguistic Society 18:156-168. Franks, Steven, and Linda Schwartz. 1992. 'Non-distinctness, Pseudo-Agreement and Binding'. Manuscript, Indiana University. Freidin, Robert. 1992. Foundations of Generative Syntax. MIT Press, Cambridge, MA. Freidin, Robert, and Leonard Babby. 1984. 'On the Interaction of Lexical and Syntactic Properties: Case Structure in Russian'. In Wayne Harbert (ed.), Cornell Working Papers in Linguistics 6. Department of Modern Languages and Linguistics, Cornell University, Ithaca, NY, pp. 71-103. Freidin, Robert, and Rex Sprouse. 1991. 'Lexical Case Phenomena'. In Robert Friedin (ed.), Principles and Parameters in Comparative Grammar. MIT Press, Cambridge, MA, pp. 392-416. Fukui, Naoki. 1986. 'A Theory of Category Projection and Its Implications'. Ph.D. dissertation, MIT. Fukui, Naoki, and Margaret Speas. 1986. 'Specifiers and Projection'. MIT Working Papers in Linguistics 8:128-172. Galkina-Fedoruk, Evdokija Mixailovna. 1958. Bezlicnye predloienija v sovremennom russkom jazyke. MGU, Moscow. Garrett, Merrill. 1988. 'Processes in Language Production'. In Frederick Newmeyer (ed.), Linguistics: The Cambridge Survey. Vol. 3: Language: Psychological and Biological Aspects. Cambridge University Press, Cambridge, pp. 69-96. Gazdar, Gerald. 1981. 'Unbounded Dependencies and Coordinate Structure'. Linguistic Inquiry 12:155-184. Gazdar, Gerald, Ewan Klein, Geoffrey Pullum, and Ivan Sag. 1985. Generalized Phrase Structure Grammar. Harvard University Press, Cambridge, MA. Giejgo, Jagnina. 1981. 'Movement Rules in Polish Syntax'. Ph.D. dissertation, University College, London. Giusti, Giuliana. 1991. 'The Categorial Status of Quantified Nominals'. Linguistische Berichte 136:438-454. Goodall, Grant. 1983. 'A Three-Dimensional Analysis of Coordination'. Chicago Linguistic Society 19:146-154.

REFERENCES

385

. 1987. Parallel Structures in Syntax: Coordination, Causatives and Restructuring. Cambridge University Press, New York. Green, Mark. 1980. 'On the Syntax and Semantics of Impersonal Sentences in Russian: A Study of the Sentence Type 'vetrom uneslo lodku'.' Ph.D. dissertation, Cornell University. Greenberg, Gerald. 1983. 'Another Look at the Second Dative and Dative Subjects'. Linguistic Analysis 11:167-218. . 1985. 'The Syntax and Semantics of the Russian Infinitive'. Ph.D. dissertation, Cornell University. . 1986. 'Dative Subjects and Objects With nuzln-'. Proceedings of the Kentucky Foreign Language Conference 1986: Slavic Section 4.1:114-124. . 1987. 'Concerning Infinitives Within Slavic'. Paper presented at American Association of Teachers of Slavic and East European Languages Annual Meeting. -. 1988. 'Relativization and Subjacency in Russian'. Proceedings of the Eastern States Conference on Linguistics 5:187-197. Greenberg, Gerald, and Steven Franks. 1991. 'A Parametric Approach to Dative Subjects and Second Datives in Slavic'. Slavic and East European Journal 35:71—97. Grimshaw, Jane. 1979. 'Complement Selection and the Lexicon'. Linguistic Inquiry 10:279326. . 1990. Argument Structure. MIT Press, Cambridge, MA. . 1991. 'Extended Projection'. Manuscript, Brandeis University. Gustavsson, Sven. 1976. Predicate Adjectives with the Copula byt' in Modern Russian. Almqwist & Wiksell International, Stockholm. Ha'ik, Isabelle. 1985. "The Syntax of Operators'. Ph.D. dissertation, MIT. Halle, Morrris. 1990. 'An Approach to Morphology'. Proceedings of the North Eastern Linguistic Society 20. Graduate Students Linguistic Association, University of Massachusetts, Amherst, pp. 150-184. Halle, Morris, and Alec Marantz. 1992. 'Distributed Morphology and the Pieces of Inflection'. Manuscript, MIT. Hamilton, William. 1980. Introduction to Russian Phonology and Word Structure. Slavica, Columbus, OH. Heim, Irene, Howard Lasnik, and Robert May. 1991. 'Reciprocality and Plurality'. Linguistic Inquiry 22:63-101. Hornstein, Norbert. 1990. 'Verb Raising in Icelandic Infinitives'. Proceedings of the North Eastern Linguistic Society 20. Graduate Students Linguistic Association, University of Massachusetts, Amherst, pp. 215-229. Hornstein, Norbert, and David Lightfoot. 1981. 'Introduction'. In Norbert Hornstein and David Lightfoot (eds.), Explanation in Linguistics. Peter Longmans, London, pp. 9-31. . 1987. 'Predication and PRO'. Language 63:23-52. House, Richard. 1982. 'The Use of Genitive Initial Sentences for the Specification of Quantity in Russian'. Ph.D. dissertation, Cornell University. Huang, James. 1982. 'Logical Relations in Chinese and the Theory of Grammar'. Ph.D. dissertation, MIT. . 1983. 'A Note on the Binding Theory'. Linguistic Inquiry 14:554-561. -. 1984. 'On the Distribution and Reference of Empty Pronouns'. Linguistic Inquiry 15:531-574. . 1993. 'Reconstruction and the Structure of VP: Some Theoretical Consequences'. Linguistic Inquiry 24:103-138. Huybregts, Riny, and Henk van Riemsdijk. 1985. 'Parasitic Gaps and ATB'. Proceedings of

386

REFERENCES

the North Eastern Linguistic Society 15. Graduate Students Linguistics Association, University of Massachusetts, Amherst, pp. 168-187. Hyams, Nina. 1986. Language Acquisition and the Theory of Parameters. Reidel, Dordrecht. . 1987. 'The Theory of Parameters and Syntactic Development'. In Thomas Roeper and Edwin Williams (eds.), Parameter Setting. Reidel, Dordrecht, pp. 1-22. 1989. 'The Null Subject Parameter in Language Acquisition'. In Oswaldo Jaeggli and Ken Safir (eds.), The Null Subject Parameter. Kluwer, Dordrecht, pp. 215-238. Ickovic, V. 1982. Ocerki sintaksiceskoj normy. Nauka, Moscow. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. MIT Press, Cambridge, MA. . 1990a. Semantic Structures. MIT Press, Cambridge, MA. . 1990b. 'On Larson's Treatment of the Double Object Construction'. Linguistic Inquiry 19:427-456. Jaeggli, Oswaldo. 1982. Topics in Romance Syntax. Foris, Dordrecht. . 1986a. 'Passive'. Linguistic Inquiry 17:587-622. . 1986b. 'Arbitrary Plural Pronominals'. Natural Language and Linguistic Theory 4:43-76. Jaeggli, Oswaldo, and Ken Safir. 1989a. 'The Null Subject Parameter and Parametric Theory'. In Oswaldo Jaeggli and Ken Safir (eds.), The Null Subject Parameter. Kluwer, Dordrecht. . 1989b. The Null Subject Parameter. Kluwer, Dordrecht. Jakobson, Roman. 1936/1962. 'Beitrag zur allgemeinen Kasuslehre: Gesamtbedeutungen der russischen Kasus'. In Selected Writings 2. Mouton, The Hague, pp. 23-71. . 1958/1962. 'MorfologiCeskienabljudenijanadslavjanskimskloneniem'. In Selected Writings 2. Mouton, The Hague, pp. 154-183. . 1984. Russian and Slavic Grammar: Studies 1931-1981. Linda Waugh and Morris Halle (eds.). Mouton, Berlin. Johnson, Kyle. 1991. 'Object Positions'. Natural Language and Linguistic Theory 9:577636. Kacala, Jan. 1971. Doplnok v sloveniine. Slovenskej Akademie Vied, Bratislava. Kardela, Henryk. 1986. 'Wh-movement in English and Polish: Theoretical Implications.' Ph.D. dissertation, Uniwersytet Marii Curie-Sk-fodowskiej. Kayne, Richard. 1981a. 'ECP Extensions'. Linguistic Inquiry 12:93-133. . 1981b. 'On Certain Differences Between French and English'. Linguistic Inquiry 12:349-371. . 1984. Connectedness and Binary Branching. Foris, Dordrecht. Keenan, Edward. 1976. 'Towards a Formal Definition of "Subject"'. In Charles Li (ed.), Subject and Topic. Academic Press, New York, pp. 305-333. Kilby, David. 1977. Deep and Surface Cases in Russian. Kubon and Sagner, Munich. Kim, Soowon, and Joan Maling. 1993. 'Syntactic Case and Frequency Adverbials in Korean'. Harvard Studies in Korean Linguistics 5:368-378. Klenin, Emily. 1978. 'Quantification, Partitivity, and the Genitive of Negation in Russian'. In Bernard Comrie (ed.), Classification of Grammatical Categories. Linguistic Research, Edmonton, pp. 163-182. Koenitz, Bernd. 1988. 'Zur Charakteristik der Satzformen mil "expletivem" Personalpronomen im Tschechischen und Deutschen'. Linguistische Arbeitsberichte 66:8-14. Koopman, Hilda. 1984. The Syntax of Verbs: From Verb Movement Rules in the Kru Languages to Universal Grammar. Foris, Dordrecht. Koopman, Hilda, and Dominique Sportiche. 1982. 'Variables and the Bijection Principle'. Linguistic Review 2:139-160.

REFERENCES

387

. 1988. 'Subjects'. Manuscript, UCLA. . 1990. 'The Position of Subjects'. Manuscript, UCLA. Kosta, Peter. 1990. 'Zur formalen Lizensierung und inhaltlichen Determination leerer Subjekte im Russischen'. Slavistische Beitrage 260:117-165. Koster, Jan. 1984. 'On Binding and Control'. Linguistic Inquiry 15:417-459. Kuroda, S.-Y. 1988. 'Whether We Agree or Not'. Lingvisticae Investigationes 12:1-47. Kury-towicz, Jerzy. 1973. Esquisses linguistiques. Fink, Munich. Kuz'mina, I. B., and E. B. Nemc'enko. 1971. Sintaksis prifastnyx form v russkix govorax. Nauka, Moscow. Lamprecht, Arnost, Du§an Slosar, and Jaroslav Bauer. 1986. Historicka mluvnice festiny. Statni Pedagogicke Nakladatelstvi, Prague. Larson, Richard. 1985. 'Bare NP Adverbs'. Linguistic Inquiry 16:595-621. . 1988. 'On the Double Object Construction'. Linguistic Inquiry 19:335-391. Lasnik, Howard. 1992a. 'Case and Expletives: Notes Towards a Parametric Account". Linguistic Inquiry 23:381-405. . 1992b. 'Two Notes on Control and Binding'. In Richard Larson, Sabine latridou, Utpal Lahiri, and James Higginbotham (eds.), Control and Grammar. Kluwer, Dordrecht, pp. 235-251. Lasnik, Howard, and Mamoru Saito. 1984. 'On the Nature of Proper Government'. Linguistic Inquiry 15:235-289. . 1992. Move Alpha. MIT Press, Cambridge, MA. Lasnik, Howard, and Juan Uriagereka. 1988. A Course in GB Syntax. MIT Press, Cambridge, MA. Lebeaux, David. 1984. 'Anaphoric Binding and the Definition of PRO'. Proceedings of the North Eastern Linguistic Society 14. Graduate Students Linguistic Association, University of Massachusetts, Amherst, pp. 253—274. . 1987. 'Comments on Hyams'. In Thomas Roeper and Edwin Williams (eds.), Parameter Setting. Reidel, Dordrecht, pp. 23-39. Leko, Nedzad. 1987. 'Syntax of Noun Headed Structures in Serbo-Croatian and Corresponding Phrasal Structures in English'. Ph.D. dissertation, Indiana University. Lencek, Rado. 1982. The History and Structure of the Slovene Language. Slavica, Columbus, OH. Levin, Beth, and Malka Rappaport. 1986. 'The Formation of Adjectival Passives'. Linguistic Inquiry 17:623-^61. Lewandowska, Barbara. 1976. 'Derivation of Infinitives in English and Polish'. Papers and Studies in Contrastive Linguistics 4:303-317. Lieber, Rochelle. 1992. Deconstructing Morphology: Word Formation in Syntactic Theory. University of Chicago Press, Chicago. Lightfoot, David. 1989. "The Child's Trigger Experience: Degree 0 Leamability'. Behavioral and Brain Sciences 12:321-375. Lindseth, Martina. 1991. 'Overt and Non-overt Pronominal Subjects in Czech, Lusatian and German'. Manuscript, Indiana University. . 1993a. 'Slavic Quantifier Phrases—with Special Reference to West Slavic'. Manuscript, Indiana University. . 1993b. 'The Syntactic Status of Overt Non-referential Pronouns in Upper Serbian'. Manuscript, Indiana University. Lumsden, John. 1992. 'Underspecification in Grammatical and Natural Gender'. Linguistic Inquiry 23:469^86. -Lojasiewicz, Anna. 1979. 'O budowie wyrazen z przyimkiempo dystrybutywnym'. Polonica 5:153-160.

388

REFERENCES

Macjusovic1, Ja, V. 1969. 'O dvux sintaktideskix konstrukcijax sovremennogo pol'skogo jazyka'. In Z. N. Strelakova (ed.). Issledovanija po polskomu jazyku. Nauka, Moscow, pp. 114-133. Maling, Joan. 1993. 'Of Nominative and Accusative: The Hierarchical Assignment of Grammatical Case in Finnish'. In Anders Holmberg and Urpo Nikanne (eds.), Case and Other Functional Categories in Finnish Syntax. Studies in Generative Grammar. Mouton de Gruyter: Dordrecht, pp. 51-76. Manzini, Rita. 1983. 'On Control and Control Theory'. Linguistic Inquiry 14:421-446. Matual, David. 1977. 'O prepodavanii predikativnyx konstrukcij—predikativy neobxodimosti'. Russian Language Journal 31:15-20. McCloskey, James. 1985. 'Case Movement and Raising in Modern Irish'. In J. Goldberg, S. MacKaye, and M. Wescoat (eds.), Proceedings of the West Coast Conference on Formal Linguistics 4. Stanford Linguistics Association, Stanford, CA, pp. 190-205. McCreight, Katherine, and Catherine Chvany. 1991. 'Geometric Representation of Paradigms in a Modular Theory of Grammar'. In Frans Plank (ed.), Paradigms: The Economy of Inflection. Mouton de Gruyter, Berlin, pp. 91-111. Mel'Cuk, Igor'. 1985. Poverxnostnyj sintaksis russkix (islovyx vyrazenij. Wiener Slawischer Almanach Sonderband 16. Institut fur Slawistik der Universitat Wien, Vienna. Melvold, Janis. 1987. 'Cyclicity and Russian Stress'. Proceedings of the North Eastern Linguistic Society 17. Graduate Student Linguistic Association, University of Massachusetts, Arnherst, pp. 467—481. Milsark, Gary. 1988. 'Singl-ing'. Linguistic Inquiry 19:611-634. Mistrik, Jozef. 1988. A Grammar of Contemporary Slovak. Slovenske Pedagogicke Nakladatel'stvo, Bratislava. Montalbetti, Mario. 1984. 'After Binding', Ph.D. dissertation, MIT. Muller, Gereon, and Wolfgang Sternefeld. 1993. 'Improper Movement and Unambiguous Binding'. Linguistic Inquiry 24:461-507. Munn, Alan. 1992. 'A Null Operator Analysis of ATB Gaps'. Linguistic Review 9:1-26. Nahimovsky, Alexander, and Richard Leed. 1987. Advanced Russian. 2nd ed., rev. Slavica, Columbus, OH. Neidle, Carol. 1982a. 'The Role of Case in Russian Syntax'. Ph.D. dissertation, MIT. . 1982b, 'Case Agreement in Russian'. In Joan Bresnan (ed.), Mental Representation of Grammatical Relations. MIT Press, Cambridge, MA, pp. 391-426. . 1988. The Role of Case in Russian Syntax. Kluwer, Dordrecht. Neijt, Anneke. 1980. Gapping. Foris, Dordrecht. Nichols, Johanna. 1981. Predicate Nominals: A Partial Surface Syntax of Russian. University of California Press, Berkeley. Oravec, Jan, Eugenia Bajzfkova, and Juraj Furdik. 1984. Sucasny' slovensky spisovny jazyk: Morfologia. Slovenske Pedagogicke Nakladatel'stvo, Bratislava. Ouhalla, Jamal. 1990. 'Sentential Negation, Relativised Minimality and the Aspectual Status of Auxiliaries'. Linguistic Review 7:183-231. . 1991. Functional Categories and Parametric Variation. Routledge, London. Ovsjaniko-Kulikovskij, D. N. 1900. 'Iz sintaktic'eskix nabljudenij: k" voprosu o klassifikacii bessub"etnyx" predlozenij'. Izvestija otdelenija russkago jazyka i slovesnosti imperatorskoj akademii nauk 5:1146-1186. Ozegov, Sergej. 1972. Slovar' russkogo jazyka. 9th ed. Sovetskaja Enciklopedija, Moscow. Paduceva, Elena. 1974. O semantike sintaksisa. Nauka, Moscow. . 1985. Vyskazyvanie i ego sootnesennost' s dejstvitel'nost'ju. Nauka, Moscow. Perlmutter, David. 1971. Deep and Surface Structure Constraints in Syntax. Holt, Rinehart and Winston, New York.

REFERENCES

389

. 1978. 'Impersonal Passives and the Unaccusative Hypothesis'. Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society, pp. 157-189. Pesetsky, David. 1982. 'Paths and Categories'. Ph.D. dissertation, MIT. . 1992. 'Zero Syntax'. Manuscript, MIT. Peskovskij, Aleksander. 1956. Russkij sintaksis v nauinom osvescenii. Gosudarstvennoe U&bno-Pedagogiceskoe Izdatel'stvo Ministerstva ProsveSCenija RSFSR, Moscow. Phinney, Marianne. 1984. 'Subjectless Sentences in the Acquisition of English and Spanish'. Proceedings of the North Eastern Linguistic Society 14. Graduate Students Linguistic Association, University of Massachusetts, Amherst, pp. 326-342. . 1987. 'The Pro-drop Parameter'. In Thomas Roeper and Edwin Williams (eds.), Parameter Setting. Reidel, Dordrecht, pp. 221-238. Pica, Pierre. 1987. 'On the nature of the Reflexivization Cycle'. Proceedings of the North Eastern Linguistic Society 17. Graduate Students Linguistic Association, University of Massachusetts, Amherst, pp. 483-499. Pin6n, Christopher. Forthcoming. 'Accumulation and Aspectuality in Polish'. Proceedings of the North Eastern Linguistic Society 24. Graduate Students Linguistic Association, University of Massachusetts, Amherst. Pollard, Carl, and Ivan Sag. 1987. Information-based Syntax and Semantics. Vol. 1: Funda mentals. CSLI Lecture Notes, Center for the Study of Language and Information, Stanford University, Stanford, CA. Pollock, Jean-Yves. 1989. 'Verb Movement, Universal Grammar and the Structure of IP'. Linguistic Inquiry 20:365-424. Postal, Paul. 1976. 'Avoiding Reference to Subject'. Linguistic Inquiry 7:151-182. Pranka, Paula. 1983. 'Syntax and Word Formation'. Ph.D. dissertation, MIT. Progovac, Ljiljana. 1992. 'Relativized SUBJECT: Long-Distance Reflexives Without Movement'. Linguistic Inquiry 23:671-680. . 1993. 'Locality and Subjunctive-like Complements in Serbo-Croatian'. Journal of Slavic Linguistics 1:116-144. Progovac, Ljiljana, and Steven Franks. 1992. 'Relativized SUBJECTS for Reflexives'. Pro ceedings of the North Eastern Linguistic Society 22. Graduate Students Linguistic Association, University of Massachusetts, Amherst, pp. 34-363. Prokopovic, Nikolaj, Lija Deribas, and Elena ProkopoviC. 1981. Imennoe i glagol'noe upravlenie v sovremennom russkom jazyke. 2nd ed. Russkij Jazyk, Moscow. Pullum, Geoffrey. 1988. 'Citation Etiquette Beyond Thunderdome'. Natural Language and Linguistic Theory 6:579-588. Raposo, Eduardo. 1987. 'Case Theory and Infl-to-Comp: The Inflected Infinitive in European Portuguese'. Linguistic Inquiry 18:85-110. Raposo, Eduardo, and Juan Uriagereka. 1990. 'Long-Distance Case Assignment'. Linguistic Inquiry 21:505-538. Rappaport, Gilbert. 1984. Grammatical Function and Syntactic Structure: The Adverbial Participle in Russian. Slavica, Columbus, OH. . 1986. 'On Anaphor Binding in Russian'. Natural Language and Linguistic Theory 4:97-120. Reinders-Machowska, Ewa. 1991. 'Binding in Polish'. In Eric Reuland and Jan Koster (eds.), Long-Distance Anaphora. Cambridge University Press, New York, pp. 137-150. Reinhart, Tanya. 1983. Anaphora and Semantic Interpretation. University of Chicago Press, Chicago. Revzin, Isaak losifovii. 1973. 'Transformacionnoe issledovanie konstrukcij s sub"jektivnym i ob"jektivnym prjamym dopolneniem'. In Andrej AnatoloviC Zaliznjak (ed.), Problemy grammatiCeskogo modelirovanija. Nauka, Moscow, pp. 88-96.

390

REFERENCES

van Riemsdijk, Henk. 1978. A Case Study in Syntactic Markedness. Foris, Dordrecht. . 1983. 'The Case of German Adjectives'. In Frank Heny and Barry Richards (eds.), Linguistic Categories: Auxiliaries and Related Puzzles. Reidel, Dordrecht, pp. 223252. van Riemsdijk, Henk, and Edwin Williams. 1981. 'NP-Structure'. Linguistic Review 1:171— 217. : 1986. Introduction to the Theory of Grammar. MIT Press, Cambridge, MA. Ritter, Elizabeth. 1991. 'Two Functional Categories in Noun Phrases: Evidence from Modern Hebrew'. In Susan Rothstein (ed.), Perspectives on Phrase Structure: Heads and Licensing, Syntax and Semantics 25. Academic Press, New York, pp. 37-62. Rizzi, Luigi. 1982. Issues in Italian Syntax. Foris, Dordrecht. . 1986. 'Null Objects in Italian and the Theory of pro'. Linguistic Inquiry 17:501-557. . 1989. 'On the Format of Parameter'. Comments on David Lightfoot, 'The Child's Trigger Experience: Degree-0 Learnability'. Brain and Behavioral Sciences 12:355356. . 1990. Relativized Minimality. MIT Press, Cambridge, MA. Roberts, Ian. 1991. 'Excorporation and Minimality'. Linguistic Inquiry 22:209-218. Robertson, John. 1983. 'From Symbol to Icon'. Language 59:529-540. Ross, John Robert. 1967/1986. 'Infinite Syntax'. Ph.D. dissertation, MIT; Ablex, Norwood, NJ. Rothstein, Susan. 1991. 'Binding, C-Command and Predication'. Linguistic Inquiry 22:572578. Rouveret, Alain, and Jean-Roger Vergnaud. 1980. 'Specifying Reference to the Subject: French Causatives and Conditions on Representations'. Linguistic Inquiry 11:97-202. Rubinstein, George. 1986. 'Subjective Dative in Russian Infinitival Clauses of Purpose'. Slavic and East European Journal 30:367-379. . 1988. 'On Subject Expression in Russian Infinitive Clauses of Comparison'. Slavic and East European Journal 32:428-444. Rudin, Catherine. 1988a. 'Multiple Questions in South Slavic, West Slavic, and Romanian'. Slavic and East European Journal 32:1-24. . 1988b. 'On Multiple Questions and Multiple WH Fronting'. Natural Language and Linguistic Theory 6:445-501. Rusanovskij, Vitalij Markorovich. 1986. Ukrainskaja grammatika. Naukova Dumka, Kiev. Ruzicka, Rudolf. 1986. 'Typologie der Diathese slavischer Sprachen in parametrischen Variationen'. Die Welt der Sloven 21:225-274. . 1988. 'On the Array of Arguments in Slavic Languages'. Zeitschrift fur Sprachwissenschaft und Kommunikationsforschung 41:155-178. Safir, Ken. 1985. Syntactic Chains. Cambridge University Press, New York. Sand, Diane Ellen Zubrow. 1981. 'Agreement of the Predicate with Quantitative Subjects in Serbo-Croatian'. Ph.D. dissertation, University of Pennsylvania. Sandfeld Jensen, Kristian. 1912. 'Notes sur les caiques linguistiques'. Festschrift Vilhelm Thomsen zur Vollendung des siebsigsten Lebensjahres, edited by his friends and students. Otto Harrassowitz, Leipzig, pp. 166-173. Schachter, Paul. 1977. 'Constraints on Coordination'. Language 53:86-103. Schein, Barry. 1982. 'Non-finite Complements in Russian'. MIT Working Papers in Linguistics 4:217-244. Schenker, Alexander. 1966. Beginning Polish. Yale University Press, New Haven, CT. . 1971. 'Some Remarks on Polish Quantifiers'. Slavic and East European Journal 15:54-60.

REFERENCES

391

1986. 'On the Reflexive Verbs in Russian'. International Journal of Slavic Linguistics and Poetics 33:27—41. Schoorlemmer, Maikke. 1991. 'Dative Subjects and Modality in Russian'. Manuscript, University of Utrecht. Schuster-Sewc, Heinz. 1974. 'Satze mit fiktiven Subjekt vom Typ os. "Wono so descuje/ ns. to se pada" und ihre Stellung in der slawischen Syntax'. Zeitschrift fur Slawistik 19:340-352. Shevelov, George. 1963. The Syntax of Modern Literary Ukrainian. Mouton, The Hague. . 1969. 'The Vicissititudes of a Syntactic Construction in Eastern Slavic'. ScandoSlavica 15:171-86. Shlonsky, Ur. 1991. 'Quantifiers as Functional Heads: A Study of Quantifier Float in Hebrew'. Lingua 84:159-180. . 1992. 'Resumptive Pronouns as a Last Resort'. Linguistic Inquiry 23:443-468. Siewierska, Anna. 1988. 'The Passive in Slavic'. In Masayoshi Shibatani (ed.), Passive and Voice. Benjamins, Amsterdam, pp. 243-289. Sigurdsson, Haldor. 1991. 'Icelandic Case-marked PRO and the Licensing of Lexical Arguments'. Natural Language and Linguistic Theory 9:327-363. Sobin, Nicholas. 1985. 'On Case Assignment in Ukrainian Morphologically Passive Constructions' . Linguistic Inquiry 16:649-662. SokaT, N. 1972. 'BezliCnye infinitivnye predlozenija v sovremennom serboxorvatskom jazyke'. In R. Bulatova (ed.), Issledovanija po serboxorvatskomu jazyku. Nauka, Moscow, pp. 324-338. Speas, Margaret. 1990. Phrase Structure in Natural Language. Kluwer, Dordrecht. . 1991. 'Functional Heads and the Mirror Principle'. Lingua 84:181-214. Sperling, Annie Joly. 1990. 'Arbitrary Pronominals in Slavic'. Manuscript, Indiana University. Sportiche, Dominique. 1981. 'Bounding Nodes in French'. Linguistic Review 1-2:219-246. . 1983. 'Structural Invariance and Symmetry'. Ph.D. dissertation, MIT. . 1988. 'A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure' . Linguistic Inquiry 19:425-450. Starosta, Manfred. 1991. Niedersorbisch. Domowina, Bautzen. Stowell, Timothy. 1981. 'Origins of Phrase Structure'. Ph.D. dissertation, MIT. . 1982. 'The Tense of Infinitives'. Linguistic Inquiry 13:561-570. . 1985. 'Licensing Conditions on Null Operators'. In J. Goldberg, S. Mackaye, and M. Wescoat (eds.), Proceedings of the West Coast Conference on Formal Linguistics, pp. 314-326. . 1989. 'Raising in Irish and the Projection Principle'. Natural Language andLinguistic Theory 7:317-359. Suprun, Adam. 1959. O russkix lislitel'nix. Kirgizskij Gosudarstvennyj Universitet, Frunze. . 1963. 'Zametki po sintaksisu pol'skix fiislitel'nyx'. Pytannja slov'jansk'oho movoznavstva 7-8:135-145. Svetlik, Ja. N. 1966. Russkij sintaksis v sopostavlenii s slovackim. Slovackoe Pedagogiceskoe Izdatel'stvo, Bratislava. Sapiro, A. B. 1953. Oderkipo sintaksisu russkix narodnyx govorov. Akademija Nauk SSSR, Moscow. Svedova, Natalia Julevna. 1970. Grammatika sovremennogo russkogo literaturnogo jazyka, Nauka, Moscow. Timberlake, Alan. 1974. 'The Nominative Object in North Russian'. In Richard Brecht and Catherine Chvany (eds.), Slavic Transformational Syntax. Michigan Slavic Materials, No. 10, University of Michigan, Ann Arbor, pp. 219-243.

392

REFERENCES

. 1975. 'Hierarchies in the Genitive of Negation'. Slavic and East European Journal 19:123-138. . 1976. 'Subject Properties in the North Russian Passive'. In Charles Li (ed.), Subject and Topic. Academic Press, New York, pp. 545-567. . 1979. 'Reflexivization and the Cycle in Russian'. Linguistic Inquiry 10:109-141. . 1980. 'Oblique Control of Russian Reflexivization'. In Catherine Chvany and Richard Brecht (eds.), Morphosyntax in Slavic. Slavica, Columbus, OH, pp. 235-259. Toman, JindHch. 1991. 'Anaphors in Binary Trees: An Analysis of Czech Reflexives'. In Eric Reuland and Jan Koster (eds.), Long-Distance Anaphora. Cambridge University Press, New York, pp. 151-170. ToporiSic, Joze. 1984. Slovenska slovnica. Obzorja, Maribor. Townsend, Charles. 1975. Russian Word Formation. Rev. ed. Slavica, Columbus, OH. . 1981. Czech Through Russian. Slavica, Columbus, OH. Travis, Lisa. 1984. 'Parameters and Effects of Word Order Variation'. Ph.D. dissertation, MIT. . 1989. 'Parameters of Phrase Structure'. In Mark Baltin and Anthony Kroch (eds.), Alternative Conceptions of Phrase Structure. University of Chicago Press, Chicago. Tulapina, E. 1969. 'Infinitivnye predlozenija s "bylo/budet"'. Russkaja rec' 1:162-168. Usakov, Dmitrij Nikolaevich. 1935. Tolkovyj slovar' russkogo jazyka. 4 vols. OGIZ, Moscow. Vaillant, Andr6. 1948. Manuel du vieux slave. Institut d'Etudes Slaves, Paris. . 1979. La Langue de Dominko Zlataric, poete ragusain de la fin du XVI" siecle. Srpska Akademija Nauka i Umetnosti, Belgrade. Vanek, Anthony. 1977. Aspects of Subject-Verb Agreement. Department of Slavic Languages, University of Alberta, Edmonton. Veyrenc, Jacques. 1979. Les Propositions infinitives en russe. Institut d'Etudes Slaves, Paris. Viel, Michel. 1984. La Notion de 'markue' chez Trubetzkoy et Jakobson. Didier, Paris. Vlaxov, Sergej, and Ljubomir Muckov. 1974. Neglagol'naja leksiko-sintakticeskaja socetaemost'. Narodna Prosveta, Sofia. White, Lydia. 1985. 'The "Pro-drop" Parameter in Adult Second Language Acquisition'. Language Learning 35:47-62. Wierzbicka, Anna. 1980. The Case for Surface Case. Karoma, Ann Arbor, MI. Williams, Edwin. 1978. 'Across-the-Board Rule Application'. Linguistic Inquiry 9:31—43. . 1980. 'Predication'. Linguistic Inquiry 11:203-238. . 1981. 'Argument Structure and Morphology'. Linguistic Review 1:81-114. . 1982. "The NP Cycle'. Linguistic Inquiry 13:277-295. . 1986/1987. 'The ATB Theory of Parasitic Gaps'. Linguistic Review 6:265-279. . 1989. "The Anaphoric Nature of Theta-Roles'. Linguistic Inquiry 20:425^*56. Willim, Ewa. 1990. 'On Case-marking in Polish'. Papers and Studies in Contrastive Linguistics 25:203-220. Woolford, Ellen. 1987. 'An ECP Account of Constraints on Across-the-Board Extraction'. Linguistic Inquiry 18:166-171. Woolhiser, Curt. 1991. "'Missing Prepositions" in Colloquial Russian Relative Clauses'. In Margaret Mills (ed.), Topics in Colloquial Russian. Peter Long, New York, pp. 1730. Yang, Dong-Whee. 1983. 'The Extended Binding Theory of Anaphors'. Language Research 19:169-192. Zabrocki, Tadeusz. 1981. 'Lexical Rules of Semantic Interpretation: Control and NPmovement in English and Polish'. Ph.D. dissertation, Uniwersytet im. Adama Mickiewicza.

REFERENCES

393

Zaenen, Annie, Joan Maling, and Hoskuldur Thrainsson. 1985. 'Case and Grammatical Functions: The Icelandic Passive'. Natural Language and Linguistic Theory 3:441-

483.

Zimmerman, 1.1967. 'Der Parallelismus verbaler und substantivischer Konstruktionen in der russischen Sprache der Gegenwart'. Zeitschrift fur Slawistik 12:744-755. Zenduk, V. N. et al. 1986. Udebnik serbo-xorvatskogo jazyka. Izdatel'stvo Moskovskogo Universiteta, Moscow.

This page intentionally left blank

Name Index

Chvany, C., 49, 66, 93, 301-302, 328n.6, 329n.8, 348 Cinque, G., 349 Comrie, B., 242-243, 244, 245-246, 258, 284n.42,286n.61 Contreras, H., 78 Corbett, G., 101, 125n.8, 128n.29, 156, 275-276 Crockett, D., 145, 186-188, 190-191, 192, 218n.85

Abney, S., 96, 154-155, 168 Anderson, C., 83, 91n.43 Andrews, A., 245 Aoun, J., 6, 240-242, 283n.34 Authier, J.-M., 305 Avrutin, S., 213n.38 Babby, L., 66, 93-96, 100-101, 102, 124nn.5, 6, 125n.l2, 126n.l5, 140141, 143, 153, 154, 156, 180-181, 197, 214n.44, 361, 370n.l8 Bailyn, J., 371n.26 Baker, M., 69, 338-339, 352, 370n.l8 Belletti, A., 304 Borer, H., 319, 338, 352 Borras, F. M., 126n.l8, 214n.47, 217n.79 Borsley, R., 62 Boskovic.Z., 82, 193, 194 Bouchard, D.,328n.2 Browne, W., 125n.8, 158, 173, 216n.66, 283nn.22, 24, 332n.41, 370n.l4 Burzio, L., 349

Dickey, S., 157, 215n.49 Dy*a, S., 62-65, 75, 77, 86n.3, 87n.7, 346 Dziwirek, K., 66, 71, 90n.40, 183, 203, 205-206, 208-209, 331n.41

Campbell, R., 88n.21, 91n.44 Chomsky, N., vii, ix, 4, 5, 6, 7, 9, 10, 14, 14n.7, 17, 57n.21, 78, 95, 101, 134, 168-169, 235, 238, 247, 259, 295, 304, 309, 312, 339-340, 353-354, 359, 371n.27, 377, 378 Christian, R. P., 126n.l8, 214n.47, 217n.79

Pagan, S., 337 FaBke, H., 138-139, 211n.l4 Feldstein, R., 298, 329n.l3 Fowler, G., 33-34, 39, 212n.27, 349, 369n.l Franks, S., 48, 49, 52, 86n.l, 96, 124n.l, 125n.7, 141, 146, 180-181, 183, 186188, 191, 196, 205-206, 208-209, 240-246, 259-260, 264, 274-275, 281, 282n.6, 341, 347-348, 349, 369n.5 Freidin, R., 57n.31, 95, 124n.6, 210n.8, 371n.20

395

396

NAME INDEX

Garrett, M., 55n.2 Goodall, G., 88n.22 Greenberg, G., 246, 274-275, 282n.6, 284n.42, 369n.l Grimshaw, J., 83-84, 87nn.l2, 13, 89n.27, 241, 376 Halle, M., 59n.44, 125n.8 Hamilton, W., 53 Hornstein, N., 3-4, 11, 225, 240-242, 243, 244, 281, 282n.9, 284n.39, 307 House, R., 186-188, 190-191 Huang, J., 78, 239, 307-308 Hyams, N., 295-296, 319 Jackendoff, R., 68, 69, 71, 83-84, 87n.l2 Jaeggli, O., 290, 296-297, 299-301, 310, 311, 325, 326, 327, 338 Jakobson, R., 39-43, 45-47, 49, 51, 60n.50, 105-106, 126n.20 Johnson, K., 338-339, 352, 370n.l8

Kim, S., 33 Klenin, E., 181-182, 198, 202, 217n.76 Koenitz, B., 314-315 Kondrashova, N., 219n.94 Koopman, H., 108, 118, 122 Kuroda, S.-Y., 118 Larson, R., 69, 109-110 Lasnik, H., 4-5, 69, 81, 83-85, 91nn.41, 42, 124n.3, 206, 219n.lOO, 304, 307, 312, 328n.5 Lebeaux, D., 329n.ll Leko, N., 98, 125n.lO Lencek, R., 270-271 Lewandowska, B., 270 Lightfoot, D., 3-4, 11, 225, 240-242, 282n.9, 284n.39, 307, 378 Lindseth, M., 59n.45, 136-139, 164-165, 211n.l5, 214n.48, 314-315, 316, 330nn.26, 31 Lojasiewicz, A., 162-163 Ludwig, J., 331n.32 Maling, J., 33 Manzini, R., 240 Mel'cuk, I., 142-143, 163, 165, 173 Melvold, J., 53

Mistrik, J., 165 Montalbetti, M., 327 Muckov, L., 57n.24 Munn, A., 88n.23 Neidle, C., 47-49, 52, 54, 126n.l7, 127n.26, 141, 144, 155, 197-199, 218nn.90, 91, 285n.55 Neijt, A., 88n.22 Nichols, J., 281n.5 Ouhalla, J., 151, 371n.24 Paduceva, E., 181 Perlmutter, D., 288, 290 Pesetsky, D., 8, 57n.l9, 93-94, 96, 106110, 111-112, 116-118, 120, 122, 123, 153,197-198, 217n.80, 371n.25 Pollock, J.-Y., 17 Progovac, Lj., 219n.l01, 246, 284n.43, 331n.39, 332n.43, 360 Raposo, E., 303-304 Rappaport, G., 284n.43 Reinders-Machowska, E., 73 Reinhart, T., 28 Rizzi, L., 6, 68, 288, 290, 296, 340-341 Roberts, L, 338-339, 352, 370n.l8 Robertson, J., 42, 51 Rudin, C., 328n.5, 371n.26 Ruzicka, R., 364, 366 Safir, K., 291-292, 296-297, 299, 299301, 305, 310, 311 Saito, M., 81, 83-85, 91nn.41, 42, 328n.5 Sand, D. E. Z., 128n.29 Schenker, A., 134 Shevelov, G., 342, 345 Shlonsky, U., 90n.34 Siewierska, A., 342, 346, 370n.l6 Sigurdsson, H., 282n.lO Sobin, N., 359-360, 370n.l8 Speas, M., 68, 69, 74, 76, 88n.l6 Sperling, A. J., 325-326, 327, 331n.40 Sportiche, D., 6, 108, 118, 122, 243, 284n.41 Sprouse, R., 210n.l8 Stowell, T., 12, 26, 29, 30, 56n.l3, 88n.23, 225, 318, 320, 323 Stroik, T., 87n.l2

NAME INDEX

Suprun, A., 210n.3 Svedova, N. Ju., 285n.55 Swan, O., 286n.61 terMeulen, A., 90n.36 Timberlake, A., 343 Toman, J., 219n.l01, 332n.44 Topori§i5, J., 342 Townsend, C., 211n.l3 Travis, L. , 329n.lO

397

van Schoonefeld, C., 58n.37 Vlaxov, S., 57n.24 Weinberg, A., 240-242 Wexler, K., 338, 352 Williams, E., 26, 56n.9, 225, 335 Willim, E., 219n.lOO Woolford, E., 76 Woolhiser, C., 211n.l4, 330n.23, 331n.32

Uriagereka, J., 90n.35, 303-304, 332n.44

Yadroff, M., 213n.38 Yang, D.-W., 246

Vaillant, A.,216n.71

Zubritskaya, K., 219n.95

This page intentionally left blank

Subject Index

Abstract case, 10-14, 64-67, 77, 126n.l7 Accusative. See also Passive and AGR-O, 359-361 as default case, 15n.ll, 30-36, 52, 135 as frozen form, 174-179 interaction with genitive of quantification, 34-35, 201-202 intrinsic, 133-134, 161 as least marked case, 49, 51-54, 136, 174 restriction in West Slavic, 132-139, 160161, 164, 178-179, 211n.l5 revision of rule, 39 statement of rule, 35 as a structural case, 32-33, 95, 104 on time/distance phrases, 33, 209, 258 Across-the-board (ATB) dependencies, 6192 Adjacency and government, 247 and theta-role assignment, 27 Adjunction, 101-102, 112-113, 156, 168169, 173, 188-189, 227, 257. See also Approximative inversion; Head movement; Incorporation Adjuncts. See Non-arguments Adnominal genitive, 172, 175, 177-178 as configurationally determined, 36-39 as inherent case, 38, 124n.5, 203 statement of rule, 38

versus verbal case government, 36-38 Adverb phrase See Gerunds Affix Hopping, 290 AGR, AGR-S, and AGR-O. See Agreement: as a functional category Agreement adjectival (gender/number) versus verbal (person/number), 298-299 adjective noun, 23-26, 106 in complementary distribution with tense in Russian, 228-233 correlation with subject diagnostics, 121124, 131-132, 253-255 and expletive subjects, 113, 120 as functional category, 226-234, 298, 309-310, 356, 359-361, 363, 367369, 371n.27, 28, 377-378. See also Verbal morphology of predicate adjectives. See Predicate adjectives: and agreement; Secondary predication; Semipredicatives: and agreement with PRO, 237-238, 249-250, 255-256, 306-307. See also Case transmission: and control with pro, 307, 311 in pronominal versus case features, 17, 80-81, 82, 114-115, 237-238, 306307, 377 with quantified subjects in Czech, 137, 211n.l3

399

400

SUBJECT INDEX

Agreement (Cont.) with quantified subjects in Lower Serbian, 137-138 with quantified subjects in Polish, 131133, 161-163, 210nn.l, 3 with quantified subjects in Russian, 101, 106-109, 112, 113, 153-154, 156, 163, 166, 174-176, 191, 215n.50 with quantified subjects in SerboCroatian, 113-115, 158-160, 215n.50 with quantified subjects in Upper Sorbian, 138 realized as a clitic, 309-310 and word order, 159, 162-163, 215n.59, 343 Agreement Principle, 20-22 and agreement, 23-26 and case assignment, 26-28 and projection, 22-23 revised formulation, 33 Animacy, 45, 59n.44, 104, 126n.l9, 137138, 154, 156-157, 167, 211n.l5. See also Virile Antecedent government. See Empty Category Principle Approximative inversion, 165-174, 192, 216nn.60, 66 Arbitrary third plural subjects, 325-328, 346-347 and Overt Pronoun Constraint, 327 and Principle of Least Effort, 328 relation to pro-drop, 326-327, 332n.44 in Russian versus other Slavic languages, 326-327 versus reflexive construction, 325-326, 331n.41 Aspect, 181, 183, 228, 233, 260, 282n.l6, 283n.24, 345 Avoid Pronoun Principle, 295 Barriers. See Government; Subjacency Belarusian (Byelorussian), 216n.66, 316, 342 Bijection Principle, 78, 109 Binding, 6, 20-21, 27-28, 56n.7, 66, 67, 71, 73, 109, 182, 189, 222, 239, 245246, 358. See also Control theory; Subject orientation: of PRO Bound variable reading, 72, 73-74, 327328

Bulgarian, 13, 214n.42, 286n.60, 288, 291, 292, 294, 310, 334, 365 Burzio's generalization, 349—450 Canonical Structural Realization (CSR), 57n.l9, 109-111, 120, 122, 184, 189, 212n.27, 375 Case absorption, 204, 336, 340, 347-351, 356 and arguments, 351 and functional heads, 356, 358-361, 367-369 and manipulation of predicate-argument structure, 351-355, 361. of oblique/quirky case, 349, 362-363 as parameter, 333, 348-351, 358-359. properties not consistent, 350-351 in Russian versus Polish, 349 and theta-role suppression, 349-350, 352-353 Case features, 41-55, 104-105, 135, 147, 156, 377. See also Checking; Visibility of expletives, 312-318 and implicational redundancies, 50 Jakobson's system of, 42-48, 105-106 and markedness, 42-43, 49, 51-53, 60n.49, 105, 134 Neidle's system of, 48-49 of null operators, 80-81 of pro, 307, 311-312 revised system of, 49-51 and Serbo-Croatian po, 158 and structural/inherent dichotomy, 4041, 54-55, 103-106, 147 and syncretism, 14, 46-47, 59n.44, 6263, 126n.20, 212n.20 and underspecification, 54, 63, 80, 204 of verbs, 30-36 and visibility, 304-306, 307, 311-313, 318-323 Case Filter, 11, 107, 110, 128n.33, 191, 225, 304-305, 311 Case Grammar, 75 Case-identification, 310, 321 Case matching, 62-63, 79-81 Case Resistance Principle, 283n.28 and case transmission, 29-30 and extraposition, 30

SUBJECT INDEX

and finite clauses versus infinitivals and PPs, 318 and Russian word order, 321 Case transmission blocking of, 236-237, 247-249, 268, 278, 280-281 and control, 225-226, 235-238, 240242, 279-280 into gerundive and participial clauses, 259-267 in Polish, 278-281 and prepositions, 29-30, 158 by pro, 311 and QPs, 156 C-command, 6, 68-69, 74, 100, 102, 182183, 208-209, 224-225, 231 CHAINS and chains, 312, 316, 317, 318, 319-322, 336, 339 Checking, 17, 56n.3, 222, 241, 351, 354, 356-361, 375, 377-378 Chinese, 289, 296-297, 303, 307 Clitics, 298-299, 309-310, 314, 338-339, 360 Cognate objects, 36 Cohesion principle, 279 Coindexation. See Agreement Principle Collective numerals in Polish, 52, 134-135 Complement, 27, 183, 209. See also Approximative inversion government into specifier of, 148, 150151. See also Exceptional Case Marking of N and proper government. See PRO: inside noun phrase of nuzno as modal and as adjective, 274-276 and quirky case, 33-34, 95, 349, 362363 Complex verb formation, 248-249, 279 Compound numerals, 142-143, 145, 157, 166, 179, 212n.l8, 213nn.35, 39, 214n.47 Condition on Extraction Domains (CED), 78 Configurational case, 95-96. See also Structural case Conjunctions, 84-85, 90n.38, 163-164 Control. See also Case transmission: and control; PRO and genitive of negation, 199

401

into infinitival complements of adjectives and participles, 265-266 by object, 242-244. See also Small clause: complement analysis of object control into Polish -no-/-to~ construction, 346 and predicate adjective agreement. See Predicate adjectives: and agreement and 'promise', 243-244 and semipredicative agreement. See Semipredicatives: and agreement as subject diagnostic, 112, 121-122, 166167 Control theory, 238-244 Copula, 191, 232-234, 250, 297-299, 309, 331n.38, 346, 355-356, 357-358 Czech, 45, 59n.45, 136-137, 164, 206207, 245-247, 314-316, 364, 366, 368 Dative, 39-40, 64, 71, 98, 104, 242. See also Control: and 'promise'; Dispositional reflexive construction; Distributive po-phrases; Semipredicatives correlation between types of, 251, 255256, 267-275 direct assignment to Semipredicatives, 256-259 experiencers and goals in nuzno constructions, 274-276 and passivization, 39, 57n.31, 371n.20 statement of rule, 39, 253, 376 subjects of impersonals and infinitives in Russian, 232, 249-256, 263, 272 subjects of impersonals in Polish, SerboCroatian, Slovak, and Slovenian, 272273 subjects of infinitives in Polish, 270 unavailable for subjects of infinitives in Serbo-Croatian, 271 unavailable for subjects of infinitives in Slovak, 271 unavailable for subjects of infinitives in Slovenian, 270-271 Dative of quantification (DAT[QJ). See also Distributive /w-phrases case features of, 147 parallelism with accusative, 146-147 and paucal numerals, 144, 145 special dative forms of, 142, 150

402

SUBJECT INDEX

Dative of quantification (DAT[Q]> (Cont.) versus accusative, 145 versus genitive, 142-143, 145 Demonstratives, 100-101, 112, 125n.8, 126n.l5, 132, 135, 139, 154, 156-157, 175, 177, 179, 193, 314. See also Determiner Phrase Determiner Phrase (DP), 13, 96, 102, 152164, 166-167, 172, 174, 189, 213n.38. See also PRO: inside noun phrase Direct case, 45-46, 104. See also Oblique case; Structural case Discourse ellipsis in Russian, 290, 292, 302-303, 307-308, 321-322, 324, 326 Dispositional reflexive construction, 364369 in Bulgarian, 334, 365 and case absorption, 364, 365, 367 created in lexicon in Russian, 366-367 in Czech, 364, 368 and internal argument restriction, 366367 interpretation of, 364 lexical versus syntactic, 367—369 personal variant in South and West Slavic, 365-369 in Russian, 364, 365, 366-367 in Serbo-Croatian, 365 in Slovak, 368 Distributed Morphology, 55n.2, 59n.44, 106, 219n.98 Distributive /w-phrases, 139-154, 157-165, 173, 176. See also Dative of quantification blocking of percolation, 141-142, 147 in Czech, Serbian, and Slovak, 164-165 and dative numerals, 142-143 government patterns in Russian, 140 in Polish, 160-164 as prepositional, 140-141, 146-148 in Serbo-Croatian, 157-160, 215n.49 and subject-verb agreement, 153-154, 158-160 D-structure, 4-5, 17, 74, 95-96, 104, 200, 230, 338, 341 Economy, 102, 134, 295, 328, 357 Empty categories, 287-288, 296-297. See

also Null subjects; PRO; Small pro; Traces Empty Category Principle (ECP), 7, 67, 76, 83-85, 108, 110, 111, 122-124, 133, 184, 190, 240, 242, 290-292 Equative constructions in Russian, 233-234 Exceptional Case Marking (ECM), 30, 143, 148-153, 157, 175-176, 275 Excorporation, 170 Expletive replacement, 312, 321, 331n.34 Expletive subjects, 68, 70, 113, 191, 292296, 299-301, 305-306, 312-323, 351, 359-360. See also Null subjects; Overt expletive subjects Extended projection, 155-156, 174, 227228, 376. See also Functional categories Extended Projection Principle, 289 External argument/theta-role, 84, 335-336, 337-339, 344, 352-353. See also Dispositional reflexive construction: and internal argument restriction; Implicit: agent as PRO in SPEC-VP; Saturation absorption of, 338, 340, 351-355 Externalization of internal argument/theta-role, 337-338, 361-364 Extraposition, 30, 295, 316, 318-322, 331n.33 Feature neutralization, 125n.l3. See also Case features: and syncretism Feature submatrices, 18-28 and coindexation, 21-22, 225-226 and default accusative, 31-33 and functional heads, 227, 354-355, 377 of null operator wh-PRO, 81 and Polish genitive-of-negation-rule, 204-205 and PRO, 225-226, 237-238, 282n.lO, 306-307 and quirky case, 27, 34, 40, 103, 200 and theta-role assignment, 26-28, 335336 Free-Inversion Parameter (FIP), 290-292, 329n.l4 French, 290, 291 Frozen forms, 53, 102, 136, 145, 152153, 157, 174-179, 264-265. See

SUBJECT INDEX

also Passive: and -no-l-to- constructions Full interpretation, 7, 14n.7, 312 Functional categories. See also Aspect; Negation and determiners, 155, 213n.38. See also Determiner Phrase as driven by morphology, 354-355, 356, 377 and feature checking, 17, 354—361, 377378 and passive. See Passive: and functional heads Quantifier Phrases as, 150, 152, 154, 155-156, 168-169, 174 and reflexive clitics, 360, 367-369 of Russian clause, 226-234, 262 Galician Portuguese, 296, 303-304, 313 Generalized Phrase Structure Grammar (GPSG), 14n.6, 282n.21 Genitive, 12-14, 104-105. See also Adnominal genitive; Genitive of negation; Genitive of quantification; Genitive2; Partitive; Paucal Genitive initial sentence (GIS), 186-195 inadequacy of movement analysis for, 186-188 in Polish, 194-195 restrictions on distribution of, 189-190, 191-192 in Serbo-Croatian, 192-194 structure and interpretation of, 188-189, 191 and subject-verb agreement, 191 and topic-comment structure, 188, 192 versus partitive, 186-187, 189, 190 Genitive of negation, 195-209, 361-364, 365 applied to non-arguments, 34, 196-197, 205-209 and case feature submatrices in Polish, 204-205, 209 and control, 199 in Czech and Slovenian, 206-207 differences between Russian and Polish, 66, 196, 200, 202-205, 219n.lOO, 363-364 as inherent case, 104, 149, 199-201, 203-204

403

interaction with GEN(Q), 200-201 optional in Russian, 196, 200 and scope, 201, 372n.30 in Ukrainian, 363 and unaccusativity, 66, 197, 203, 361364, 372n.30 versus partitive GEN(Q), 197-199, 205209 and word order, 202 Genitive of quantification (GEN[Q]), 94117,131-139 and approximative inversion in Ukrainian, 173-174 assigned at S-structure, 39 case features of, 104—105 and empty quantifiers, 180-195 and frozen quantifiers in Polish, 178-179 and frozen quantifiers in Serbo-Croatian, 176-178 heterogenous versus homogenous case distribution, 94-97, 99, 136, 155 and number percolation, 114-115 and predicate adjective agreement in Polish, 275-276 and prequantifier modifiers, 100-103, 133, 136, 156 as structural or inherent case, 99, 104105, 124n.3, 132, 164 and word order, 107 Genitive2, 43, 48, 50-51, 105, 180-182, 198, 217n.76. See also Partitive German, 29, 295-297 Gerunds, 66, 121, 131, 259-265 as adjuncts, 260-261 controlled by non-subjects in colloquial Russian, 253-255, 263-264 failure of case transmission into, 259261 impossibility of second dative in, 262263 lack of COMP position in, 261 and predicate adjectives, 259-262 as reduced clauses, 259—262 temporal interpretation of, 261-262 Government, 7, 100, 157, 182, 240-244, 247, 283n.36, 290, 358. See also C-command; Proper government Greek, 245 Group reading, 110-111, 115-117, 123. See also Quantifier Raising

404

SUBJECT INDEX

Head movement, 102, 167-173, 356-360. See also Incorporation Head Movement Constraint, 170, 214n.42 Hebrew, 295-296, 301, 319 Hungarian, 232, 283n.22

and overt expletive subjects, 312, 315316 Irish, 296, 320 ISH. See VP-internal subject hypothesis Italian, 290, 291, 339-340, 349

Icelandic, 210n.8, 214n.45, 245, 260, 282n.lO, 284n.40 Identification of null operators, 80 of null quantifier, 182-185, 187-189 of null subjects, 296-297, 299, 301, 303-304, 309-312 Impersonal constructions, 69-73, 250, 252-256, 272-273, 313, 346, 350351, 368. See also Dispositional reflexive construction; Expletive subjects; Passive: and -no-l-toconstructions; Passive and null subjects Implicit, agent as PRO in SPEC-VP, 355-358 object, 337 subject/agent, 344, 346, 352-353 Incomplete quantified expression (IQE). See Genitive initial sentence; Partitive Incorporation, 227, 230, 260-261 Individuation, 110-111, 115-117, 123, 140. See also Quantifier Raising INFL. See Inflection Inflection (I), 10, 12, 227, 229-234, 290291, 297-299, 301, 303, 354-355, 356-358. See also Agreement: as functional category; Tense: as functional category Inherent case, 95-99, 103-104, 132, 137, 161, 203-204 Innertness, 20, 219n.98 Instrumental, 44, 125n.lO as case of adjuncts, 40, 44, 246 complement, 33-34, 349 as most marked case, 45, 49, 51-52, 59n.45, 134 predicative, 191, 221-222, 226, 233234, 275-276, 286n.70 statement of rule, 39 versus PRO in passive, 356-357 Inversion as blocking discourse licensing of null subjects in Russian, 302-303

Japanese, 289, 296-297 Language faculty, viii, 3-4, 14n.l Learnability, viii-ix, 3-4, 9, 44, 227-228, 334-335, 374-378 and null subject parameters, 288, 289, 294-297, 329n.ll Lexical case. See Inherent case; Quirky case Lexical-Functional Grammar, 47^9, 54, 58n.38, 127n.26, 251. See also Neidle, C. (Name Index) Lexical government. See Empty Category Principle; PRO: and lexical government; Proper government Lexical insertion, 17-18, 63, 80, 329n.l8 Licensing. See also Visibility conditions for configurational case on non-arguments, 258-259 of intrinsic accusative, 134 of null operators, 78, 80-81, 89n.25 of null quantifiers, 181, 190, 208-209 of null subjects, 296-297, 301, 303304 of PRO, 11, 307 Like-Attracts-Like Constraint, 168-169, 172, 216n.61 Locality, 95, 99, 235, 241. See also Minimality Locative. See Distributive /w-phrases: in Czech, Serbian, and Slovak; Distributive po-phrases: in Polish Locative2, 43, 48, 50-51, 58n.36 Logical Form (LF), 4-5, 7, 81, 108-111, 120, 122-124, 184, 189-192, 239, 240, 304-305, 318, 353-354, 359 Long-distance agreement. See Predicate adjectives: and agreement; Semipredicatives: and agreement Long-distance extraction, 122-123, 133134, 182, 248-249, 290, 291-292, 332n.43 Lower Sorbian, 137-138, 164, 313-314

SUBJECT INDEX

Macedonian, 13, 214n.42, 285n.48, 298, 310 Markedness, 107, 115, 127n.24, 131, 133, 135, 295, 311, 315, 329n.8. See also Case features: and markedness Middle voice, 326 incompatibility of agent, 337, 344, 371n.21 as lexical operation, 336-337, 344, 352 versus passive in Polish, 341, 344, 353 Minimalism, 353-354, 356-361, 368-369, 377-378 Minimality, 39, 147, 213n.38, 359. See also Locality Mixed declension. See Pronominal declension Modal adjectives, 274-275 covert, 273, 367, 368 phrase, 368-369 Morphological uniformity, 297, 301 Negation. See also Dispositional reflexive construction; Genitive of negation as functional category, 151-152, 183 as licensing element for adjunct genitives, 208-209 and partitivity, 181-182, 198, 209 NOM-drop Parameter (NDP), 290-294, 305. See also Overt Subject Parameter Nominative. See also Case absorption AGR clitic 309-310 assigned at S-structure, 39, 119, 290, 321 assigned by tense or agreement, 229, 232, 251 in genitive initial sentences, 191 objects of passive participles in North Russian, 343 obligatory discharging of, 305, 313. See also Overt Expletive Subject Parameter optionally assigned at LF, 311-313. See also Overt Subject Parameter in passive + accusative constructions, 356-360 as structural case, 95, 104 and tysjafa, 174-176, 216n.67 Non-arguments as adjoined to VP, 40, 208-209, 226

405

and approximative inversion, 166, 168173 and case of semipredicatives, 257-259 and default accusative, 33, 35-36 in genitive initial sentences. See Genitive initial sentence and genitive of negation, 34, 196-197, 205-209 prepositional phrase, 365-367 Non-distinctness, 17, 63, 80, 81, 258, 361. See also Lexical insertion North Russian dialects, 13, 210n.9, 343 Null operator, 78-86, 89nn.25, 30, 91n.45 Null quantifiers. See Genitive initial sentence; Genitive of negation; Genitive2; Partitive Null Subject Parameters. See FreeInversion Parameter; Morphological uniformity; NOM-drop Parameter; Overt Expletive Subject Parameter; Overt Subject Parameter; Pro-drop Parameter Null subjects, 287-332. See also Dispositional reflexive construction; Impersonal constructions; Small pro as bound variables, 327-328 and clitics, 309-310 discourse. See Discourse ellipsis in Russian and ECP effects in Slavic, 291-292 expletive, 113, 289, 292-297, 300-301, 304-305, 340. See also Expletive subjects; Overt expletive subjects generic and proverbial, 324—325. See also Arbitrary third plural subjects and passive. See Passive: and null subjects in personal constructions, 324-328 and Russian as "mixed" null subject language, 300-303, 309-310 theta-marked (thematic), 288, 289, 292, 296-297, 298-303, 306-307, 310312 and topics, 297, 302-303, 307 types of, 289-290, 295-297, 328n.3 and verbal morphology, 289, 295-299, 301-303, 309, 329n.lO Numeral phrases, 93-195, 200-202 Numeral "squish", 128n.29 Numeric classifiers. See Frozen forms

406

SUBJECT INDEX

Object control. See Control: by object Objective absorption of, 336, 340 as default case, 14, 15n.ll Oblique case, 104-105, 111, 113, 164-165, 167, 203-204, 294, 349. See also Inherent case Old Church Slavonic (OCS), 54 Old Russian, 54 Overt Expletive Subject Parameter (OESP), 290, 294, 305-306, 313-318, 340, 347, 357-358, 377. See also Learnability: and null subject parameters Overt expletive subjects agreeing, 315-316, 330n.31 and case, 305-306, 359-360 in chain with finite clause, 318-323 discourse function and semantic properties of, 315, 321-323, 330nn.29, 30 in East Slavic, 316-317, 319-322, 331n.35, 360 in English, 299, 319 impossibility in Romance, 323 in Irish, 320 in null subject languages, 303-304, SISSIS in Polish and Serbo-Croatian, 322-323 position of, 314—316 as possible non-pro-drop triggers, 295296, 303 types of, 317-323 Overt Pronoun Constraint, 327 Overt Subject Parameter (OSP), 290, 310312, 377. See also NOM-drop Parameter Parasitic gaps, 77-85, 88n.23 Participial passive. See Passive Partitive, 180-186 genitive2 as test for, 180, 182 interaction with aspect and negation, 181-183 and "large quantity" interpretation, 180181, 194, 218n.8S and non-argument genitives under negation, 205-209 objects of prepositions, 217n.79 and passive, 185-186 as QP or DP, 183-185

and selection, 183-184 in Serbo-Croatian and Polish, 185-186, 194-195, 205-207 and unaccusativity, 181 Passive, 74, 84, 141, 185-186, 194. See also Case absorption; Complement: and quirky case; Dative: and passivization; Voice + accusative in Polish, 341, 352, 358 + accusative in Serbo-Croatian dialects, 341-342, 352 + accusative in Slovenian, 342, 352, 370n.l4 + accusative in Ukrainian, 333, 342, 344-345, 352-353, 356-360 adjectival (lexical), 337-338, 360-361 and binding, 90n.40 and expression of agent in Slavic, 344346, 352-353, 355-358 and functional heads, 350, 355-361, 363 and genitive of negation, 362-364 lexical versus syntactic, 336-338, 344, 351-355, 357, 360-364 and middle voice, 336-337, 352 morphology and theta-role/case absorption, 34, 338-340, 343, 347-355, 358-359 and -no-l-to- constructions, 343-346, 352-353, 355-360, 363-364 and null subjects, 339-347, 357-357 participles in North Russian dialects, 343 participles versus reflexive passives, 333, 344-345, 360-361, 362 in Polish, 204-205, 340-341, 345, 349, 353, 363-364 and pro-drop parameters, 339-343 in Russian, 34, 341, 342, 349 in Ukrainian, 342, 344, 355-360, 363 in Ukrainian dialects, 345 Paucal, 52, 57n.28, 59n.46, 105-106, 125nn.8, 13, 126n.21, 128n.29, 154, 156-157, 159-160, 162-163, 167, 173. See also Dative of quantification: and paucal numerals Percolation, 22-23, 96-97, 98, 99, 104, 114, 147, 148, 155-156, 163-164, 230231 Periphrastic future in Russian, 228, 233, 330n.24 passive in Polish, 345

SUBJECT INDEX

Person. See Null subjects: and verbal morphology Phonetic Form (PF), 4-5, 11, 17, 191, 240, 241-242, 304-305, 353-354. See also Discourse ellipsis in Russian Portuguese, 232, 291. See also Galician Portuguese Predicate adjectives and agreement, 221-226, 237, 241-242, 246, 278-279, 306-307, 350-351 c-command and short form, 221-222, 224-225 and gerunds, 259-262, 285n.55 instrumental as default case of, 222-223, 234, 277 in Polish, 276-281, 306-307, 350-351 as small clause APs, 225-226, 234, 246 substantivized, 275-276 types of, 220-224 Predication coindexation, 225-226, 234 and mutual c-command, 224—225, 257 and obligatory coreference, 243, 266 Prepositional quantifiers, 143-144, 146, 154 Prepositions and accusative, 35, 53-54 and approximative inversion, 169-173 doubling of, 171 and inherent case, 97 partitive objects of, 217n.79 quantificational, 143-144, 146, 148, 154, 212n.27. See also Distributive pophrases and reflexivization in Russian, 254 s(a) as instrumental marker in SerboCroatian, 98 and subject-verb agreement, 153-154, 158-160 and theta-role assignment, 158, 171, 173 Principle A, 6, 21, 240, 245. See also Binding; Reflexives Principle B, 6, 239. See also Binding Principle of Lexical Satisfaction, 95-96, 98-99, 124n.6 Privative features, 231 PRO, 6, 11, 56n.7, 81, 222, 225-226, 234-249, 306-307, 371n.25. See also Case transmission: and control; Control; Subject orientation: of PRO

407

as anaphor or pronoun, 239-241, 246, 283n.36, 357 distribution of, 238-240, 241, 357 as implicit agent, 355-358, 367 and lexical government, 240-243, 247249, 260, 266, 357 in noun phrase, 241, 247-249 subject of gerunds and participles, 259267 Pro drop. See Null subjects Pro-drop Parameter, 290-291, 294-295, 297, 340-342. See also Passive and pro-drop parameters Projection Principle, 8, 27, 61, 109, 289, 334, 336, 338-339, 372n.30 Pronominal declension, 52, 139, 222, 257258. See also Demonstratives; Semipredicatives Proper government, 78, 83, 84, 108, 110, 182, 260, 290, 355. See also Control theory; Empty Category Principle Purpose clauses, 253-255, 280-281, 337, 344, 352-353 Quantifier float, 243-244, 284n.41 Quantifier Phrase (QP), 95, 102, 104, 106113, 115-124, 131-139, 150-157, 189, 195, 217n.81 and approximative inversion, 166-169, 172-174 and case, 111, 118-119 as distinct from genitive of negation, 197-199 as distinct from partitive, 183-185 features of, 146 internal structure of, 111-113, 150-152, 155-157,175-176 and interpretation, 115-119, 122-124, 172 not maximal in Polish, 131-32, 160-164 not maximal in Serbo-Croatian, 113, 117 as operator phrase, 151 position of, 119-124, 317 and word order, 107 Quantifier Raising (QR), 78, 108-111, 115117, 122-124, 189, 201-202 Quasi-arguments, 329n.8 Quirky case, 27, 33-34, 37, 40, 95-98, 199-200, 203-204, 233-234, 362-

408

SUBJECT INDEX

Quirky case (Cont.) 363. See also Complement; Inherent case Reciprocals, 73, 122, 166, 284n.43. See also Binding Reconstruction, 85, 239 Reflexive passive. See Passive Reflexives, 20-21, 66, 71, 73, 112, 121, 131, 213n.38, 245-246, 253-254, 360. See also Binding Relational Grammar (RG), 59n. 43, 66, 205-206. See also Dziwirek, K.; Perlmutter, D. (Name Index) Relative clauses, 61-92, 261 Resolution rules, 115 Restructuring. See Complex verb formation Resultatives, 281n.3 Resumptive pronouns, 82-83, 90n.34, 290 Right dislocation, 242, 284n.39 Russian dialects, 317. See also North Russian dialects Saturation, 337, 344 Scope, 101, 123-124, 156, 183, 196, 200202. See also Genitive of negation; Quantifier Raising Scrambling, 21, 68, 69, 73, 74, 192, 194, 261, 372n.30 Secondary predication, 220-286. See also Predicate adjectives; Semipredicatives Second dative. See Semipredicatives Semipredicatives as adverbials in colloquial Polish and Slovak, 264-265, 286n.63, 286nn.63, 71 and agreement, 222-223, 226, 235, 247-248, 263-265, 279 as belonging to pronominal declensional, 223-224 configuration of, 257 dative as default case of, 223, 242, 249259, 262-263, 268, 277 and gerunds in Russian and Polish, 262265 in infinitival complements to adjectives and participles, 265-266 nominative as default case in Slovenian, Serbo-Croatian, and Slovak, 268-269 in Polish, 267-268, 276-281

and short form adjectives, 265 Short form adjectives in Russian, 221-222, 223-225, 242, 257-258, 265, 276, 281n.5, 285n.53, 361 Sisterhood and adjunct instrumental, 39, 50, 226 and case assignment in English, 12— 13 and dative case schema, 39, 50, 253, 257 and GEN(Q) assignment, 100, 102 requirement on genitive of negation, 196-197 restriction on modification, 100, 102 and theta-role assignment, 27-28, 33 Slovak, 136, 137, 165, 245, 269, 271, 273, 368 Slovenian, 137, 207, 268, 270-271, 273, 342, 370n.l4 Small clause, 212n.31, 224-226, 234 complement analysis of object control, 243-244 complements with nuzno, 275-276 Polish -no-l-to- construction as, 355 subject of, 128n.34, 149, 244 Small pro, 261, 285n.56, 288-289, 307, 311. See also Arbitrary third plural subjects; Null subjects Spanish, 290, 296-297 SPEC-head agreement, 12-13, 39, 151, 217n.78, 231, 251, 356, 359, 377 Split INFL in Russian, 229-233, 310 S-structure, 4-5, 32, 41, 95-96, 147, 204, 306, 310-312, 317-318 Structural case, 54-55, 95-96, 99, 103104, 111, 124n.3, 147, 149, 161, 172173, 197. See also Dative of quantification; Exceptional Case Marking; Genitive of quantification Subcategorization, 62-63, 96, 135-136, 283n.26, 345 Subjacency, 6, 78, 248-249, 279 Subjecthood, 64, 66, 68, 175, 253-255. See also VP-internal subject hypothesis Subject orientation, 66, 73, 121, 253-255 of PRO, 245-247, 249 Submatrices. See Feature submatrices

SUBJECT INDEX

409

Syntactic Case Hierarchy, 95, 99-100, 124n.6

Triggers and parameter (re)setting, 3, 295296, 376, 377, 378

Tense (TNS) as assigning dative in absence of agreement, 251-253, 256 in complementary distribution with agreement in Russian, 228-233 as functional category, 226-234, 297299, 345, 356, 359 irrelevance for dative semipredicatives, 258-259 missing in gerunds, 261-262 and subject-verb agreement, 162, 297299, 309 Thematic prominence, 67-86 and binding, 71, 73 of dative experiencer, 71, 73, 75-76, 87n.l2, 275 and embedding, 76-77, 84 hierarchy, 68, 87n.l3, 335 and passivization, 84 relative prominence of theme and goal, 72-73 and structural prominence, 67-69, 335 and volitionality, 75-76 Theta-Criterion, 7, 133, 161, 191-192, 289, 304-305, 323, 334, 355 and movement, 8 reformulated as theta-slot coindexation, 27, 335-336 Theta-slot coindexation, 27, 40, 216n.64, 335336 definition, 26 nouns versus verbs, 37-38 Traces, 5, 8, 57n.l9, 77, 81, 91nn.42, 45, 108-110, 122, 189, 288, 331n.33, 336. See also Empty Category Principle

Ukrainian, 173-174, 291, 292, 293, 316317, 333, 342, 344-345, 352-353, 356-358 Unaccusativity, 66, 107, 120, 126n.23, 181, 189, 197, 362. See also Genitive of negation Uniformity of Theta Assignment Hypothesis (UTAH), 69, 74 Upper Sorbian, 138-139, 164-165, 211n.l5, 313-314 Verbal morphology and null subjects. See Null subjects: and verbal morphology and Russian copula, 232-234, 250, 297298, 309 and functional categories, 228-233, 298, 309-310, 356-361 and voice. See Passive: morphology and theta-role/case absorption Virile, 132, 138-139, 160-161, 178 Visibility, 11-12, 30, 80, 96, 134, 191-192, 197, 226, 240, 241-242, 304-306, 307, 311-313, 318-319, 323, 354, 356, 362, 377 Voice, 333-373. See also Middle voice; Passive and predicate-argument structure, 334339 VP-internal subject hypothesis (ISH), 108, 118-124, 224-225, 241, 316, 355359 Weak crossover, 72, 73-74 X-bar syntax, 9-10, 16, 22, 30, 56n.ll, 112, 230-231. See also Adjunction; Functional categories